wuggy/Dactyloidae
wuggy/Dactyloidae
1
0
Fork 0
mirror of https://repo.dactyloidae.xyz/Dactyloidae/UXP.git synced 2026-10-06 15:27:30 +09:00
Code Issues Projects Releases Packages Wiki Activity

Import aom library

Browse source 
This is the reference implementation for the Alliance for Open Media's av1 video code.

The commit used was 4d668d7feb1f8abd809d1bca0418570a7f142a36.
This commit is contained in:
trav90 2018-10-15 21:45:30 -05:00 • committed by Roy Tam
commit edc8d83307
989 changed files with 470949 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

518
third_party/aom/av1/av1.cmake vendored Normal file
View file

@ -0,0 +1,518 @@
##
## Copyright (c) 2017, Alliance for Open Media. All rights reserved
##
## This source code is subject to the terms of the BSD 2 Clause License and
## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
## was not distributed with this source code in the LICENSE file, you can
## obtain it at www.aomedia.org/license/software. If the Alliance for Open
## Media Patent License 1.0 was not distributed with this source code in the
## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
##
set(AOM_AV1_COMMON_SOURCES
"${AOM_ROOT}/av1/av1_iface_common.h"
"${AOM_ROOT}/av1/common/alloccommon.c"
"${AOM_ROOT}/av1/common/alloccommon.h"
# TODO(tomfinegan): Foward transform belongs in encoder.
"${AOM_ROOT}/av1/common/av1_fwd_txfm1d.c"
"${AOM_ROOT}/av1/common/av1_fwd_txfm1d.h"
"${AOM_ROOT}/av1/common/av1_fwd_txfm2d.c"
"${AOM_ROOT}/av1/common/av1_fwd_txfm2d_cfg.h"
"${AOM_ROOT}/av1/common/av1_inv_txfm1d.c"
"${AOM_ROOT}/av1/common/av1_inv_txfm1d.h"
"${AOM_ROOT}/av1/common/av1_inv_txfm2d.c"
"${AOM_ROOT}/av1/common/av1_inv_txfm2d_cfg.h"
"${AOM_ROOT}/av1/common/av1_loopfilter.c"
"${AOM_ROOT}/av1/common/av1_loopfilter.h"
"${AOM_ROOT}/av1/common/av1_txfm.h"
"${AOM_ROOT}/av1/common/blockd.c"
"${AOM_ROOT}/av1/common/blockd.h"
"${AOM_ROOT}/av1/common/common.h"
"${AOM_ROOT}/av1/common/common_data.h"
"${AOM_ROOT}/av1/common/convolve.c"
"${AOM_ROOT}/av1/common/convolve.h"
"${AOM_ROOT}/av1/common/debugmodes.c"
"${AOM_ROOT}/av1/common/entropy.c"
"${AOM_ROOT}/av1/common/entropy.h"
"${AOM_ROOT}/av1/common/entropymode.c"
"${AOM_ROOT}/av1/common/entropymode.h"
"${AOM_ROOT}/av1/common/entropymv.c"
"${AOM_ROOT}/av1/common/entropymv.h"
"${AOM_ROOT}/av1/common/enums.h"
"${AOM_ROOT}/av1/common/filter.c"
"${AOM_ROOT}/av1/common/filter.h"
"${AOM_ROOT}/av1/common/frame_buffers.c"
"${AOM_ROOT}/av1/common/frame_buffers.h"
"${AOM_ROOT}/av1/common/idct.c"
"${AOM_ROOT}/av1/common/idct.h"
"${AOM_ROOT}/av1/common/mv.h"
"${AOM_ROOT}/av1/common/mvref_common.c"
"${AOM_ROOT}/av1/common/mvref_common.h"
"${AOM_ROOT}/av1/common/odintrin.c"
"${AOM_ROOT}/av1/common/odintrin.h"
"${AOM_ROOT}/av1/common/onyxc_int.h"
"${AOM_ROOT}/av1/common/pred_common.c"
"${AOM_ROOT}/av1/common/pred_common.h"
"${AOM_ROOT}/av1/common/quant_common.c"
"${AOM_ROOT}/av1/common/quant_common.h"
"${AOM_ROOT}/av1/common/reconinter.c"
"${AOM_ROOT}/av1/common/reconinter.h"
"${AOM_ROOT}/av1/common/reconintra.c"
"${AOM_ROOT}/av1/common/reconintra.h"
"${AOM_ROOT}/av1/common/resize.c"
"${AOM_ROOT}/av1/common/resize.h"
"${AOM_ROOT}/av1/common/restoration.h"
"${AOM_ROOT}/av1/common/scale.c"
"${AOM_ROOT}/av1/common/scale.h"
"${AOM_ROOT}/av1/common/scan.c"
"${AOM_ROOT}/av1/common/scan.h"
"${AOM_ROOT}/av1/common/seg_common.c"
"${AOM_ROOT}/av1/common/seg_common.h"
"${AOM_ROOT}/av1/common/thread_common.c"
"${AOM_ROOT}/av1/common/thread_common.h"
"${AOM_ROOT}/av1/common/tile_common.c"
"${AOM_ROOT}/av1/common/tile_common.h")
set(AOM_AV1_DECODER_SOURCES
"${AOM_ROOT}/av1/av1_dx_iface.c"
"${AOM_ROOT}/av1/decoder/decodeframe.c"
"${AOM_ROOT}/av1/decoder/decodeframe.h"
"${AOM_ROOT}/av1/decoder/decodemv.c"
"${AOM_ROOT}/av1/decoder/decodemv.h"
"${AOM_ROOT}/av1/decoder/decoder.c"
"${AOM_ROOT}/av1/decoder/decoder.h"
"${AOM_ROOT}/av1/decoder/detokenize.c"
"${AOM_ROOT}/av1/decoder/detokenize.h"
"${AOM_ROOT}/av1/decoder/dsubexp.c"
"${AOM_ROOT}/av1/decoder/dsubexp.h"
"${AOM_ROOT}/av1/decoder/dthread.c"
"${AOM_ROOT}/av1/decoder/dthread.h")
set(AOM_AV1_ENCODER_SOURCES
"${AOM_ROOT}/av1/av1_cx_iface.c"
"${AOM_ROOT}/av1/encoder/aq_complexity.c"
"${AOM_ROOT}/av1/encoder/aq_complexity.h"
"${AOM_ROOT}/av1/encoder/aq_cyclicrefresh.c"
"${AOM_ROOT}/av1/encoder/aq_cyclicrefresh.h"
"${AOM_ROOT}/av1/encoder/aq_variance.c"
"${AOM_ROOT}/av1/encoder/aq_variance.h"
"${AOM_ROOT}/av1/encoder/av1_quantize.c"
"${AOM_ROOT}/av1/encoder/av1_quantize.h"
"${AOM_ROOT}/av1/encoder/bitstream.c"
"${AOM_ROOT}/av1/encoder/bitstream.h"
"${AOM_ROOT}/av1/encoder/block.h"
"${AOM_ROOT}/av1/encoder/context_tree.c"
"${AOM_ROOT}/av1/encoder/context_tree.h"
"${AOM_ROOT}/av1/encoder/cost.c"
"${AOM_ROOT}/av1/encoder/cost.h"
"${AOM_ROOT}/av1/encoder/dct.c"
"${AOM_ROOT}/av1/encoder/encodeframe.c"
"${AOM_ROOT}/av1/encoder/encodeframe.h"
"${AOM_ROOT}/av1/encoder/encodemb.c"
"${AOM_ROOT}/av1/encoder/encodemb.h"
"${AOM_ROOT}/av1/encoder/encodemv.c"
"${AOM_ROOT}/av1/encoder/encodemv.h"
"${AOM_ROOT}/av1/encoder/encoder.c"
"${AOM_ROOT}/av1/encoder/encoder.h"
"${AOM_ROOT}/av1/encoder/ethread.c"
"${AOM_ROOT}/av1/encoder/ethread.h"
"${AOM_ROOT}/av1/encoder/extend.c"
"${AOM_ROOT}/av1/encoder/extend.h"
"${AOM_ROOT}/av1/encoder/firstpass.c"
"${AOM_ROOT}/av1/encoder/firstpass.h"
"${AOM_ROOT}/av1/encoder/hybrid_fwd_txfm.c"
"${AOM_ROOT}/av1/encoder/hybrid_fwd_txfm.h"
"${AOM_ROOT}/av1/encoder/lookahead.c"
"${AOM_ROOT}/av1/encoder/lookahead.h"
"${AOM_ROOT}/av1/encoder/mbgraph.c"
"${AOM_ROOT}/av1/encoder/mbgraph.h"
"${AOM_ROOT}/av1/encoder/mcomp.c"
"${AOM_ROOT}/av1/encoder/mcomp.h"
"${AOM_ROOT}/av1/encoder/picklpf.c"
"${AOM_ROOT}/av1/encoder/picklpf.h"
"${AOM_ROOT}/av1/encoder/ratectrl.c"
"${AOM_ROOT}/av1/encoder/ratectrl.h"
"${AOM_ROOT}/av1/encoder/rd.c"
"${AOM_ROOT}/av1/encoder/rd.h"
"${AOM_ROOT}/av1/encoder/rdopt.c"
"${AOM_ROOT}/av1/encoder/rdopt.h"
"${AOM_ROOT}/av1/encoder/segmentation.c"
"${AOM_ROOT}/av1/encoder/segmentation.h"
"${AOM_ROOT}/av1/encoder/speed_features.c"
"${AOM_ROOT}/av1/encoder/speed_features.h"
"${AOM_ROOT}/av1/encoder/subexp.c"
"${AOM_ROOT}/av1/encoder/subexp.h"
"${AOM_ROOT}/av1/encoder/temporal_filter.c"
"${AOM_ROOT}/av1/encoder/temporal_filter.h"
"${AOM_ROOT}/av1/encoder/tokenize.c"
"${AOM_ROOT}/av1/encoder/tokenize.h"
"${AOM_ROOT}/av1/encoder/treewriter.c"
"${AOM_ROOT}/av1/encoder/treewriter.h"
"${AOM_ROOT}/av1/encoder/variance_tree.c"
"${AOM_ROOT}/av1/encoder/variance_tree.h")
set(AOM_AV1_COMMON_INTRIN_SSE2
# Requires CONFIG_GLOBAL_MOTION or CONFIG_WARPED_MOTION
#"${AOM_ROOT}/av1/common/x86/warp_plane_sse2.c"
"${AOM_ROOT}/av1/common/x86/idct_intrin_sse2.c")
set(AOM_AV1_COMMON_INTRIN_SSSE3
"${AOM_ROOT}/av1/common/x86/av1_convolve_ssse3.c")
set(AOM_AV1_COMMON_INTRIN_SSE4_1
"${AOM_ROOT}/av1/common/x86/av1_fwd_txfm1d_sse4.c"
"${AOM_ROOT}/av1/common/x86/av1_fwd_txfm2d_sse4.c")
set(AOM_AV1_COMMON_INTRIN_AVX2
"${AOM_ROOT}/av1/common/x86/hybrid_inv_txfm_avx2.c")
set(AOM_AV1_COMMON_INTRIN_DSPR2
"${AOM_ROOT}/av1/common/mips/dspr2/av1_itrans16_dspr2.c"
"${AOM_ROOT}/av1/common/mips/dspr2/av1_itrans4_dspr2.c"
"${AOM_ROOT}/av1/common/mips/dspr2/av1_itrans8_dspr2.c")
set(AOM_AV1_COMMON_INTRIN_MSA
"${AOM_ROOT}/av1/common/mips/msa/av1_idct16x16_msa.c"
"${AOM_ROOT}/av1/common/mips/msa/av1_idct4x4_msa.c"
"${AOM_ROOT}/av1/common/mips/msa/av1_idct8x8_msa.c")
set(AOM_AV1_ENCODER_ASM_SSE2
"${AOM_ROOT}/av1/encoder/x86/dct_sse2.asm"
"${AOM_ROOT}/av1/encoder/x86/error_sse2.asm"
"${AOM_ROOT}/av1/encoder/x86/temporal_filter_apply_sse2.asm")
set(AOM_AV1_ENCODER_INTRIN_SSE2
"${AOM_ROOT}/av1/encoder/x86/dct_intrin_sse2.c"
"${AOM_ROOT}/av1/encoder/x86/highbd_block_error_intrin_sse2.c"
"${AOM_ROOT}/av1/encoder/x86/av1_quantize_sse2.c")
set(AOM_AV1_ENCODER_ASM_SSSE3_X86_64
"${AOM_ROOT}/av1/encoder/x86/av1_quantize_ssse3_x86_64.asm")
set(AOM_AV1_ENCODER_INTRIN_SSSE3
"${AOM_ROOT}/av1/encoder/x86/dct_ssse3.c")
set(AOM_AV1_ENCODER_INTRIN_AVX2
"${AOM_ROOT}/av1/encoder/x86/error_intrin_avx2.c"
"${AOM_ROOT}/av1/encoder/x86/hybrid_fwd_txfm_avx2.c")
set(AOM_AV1_ENCODER_INTRIN_NEON
"${AOM_ROOT}/av1/encoder/arm/neon/quantize_neon.c")
set(AOM_AV1_ENCODER_INTRIN_MSA
"${AOM_ROOT}/av1/encoder/mips/msa/error_msa.c"
"${AOM_ROOT}/av1/encoder/mips/msa/fdct16x16_msa.c"
"${AOM_ROOT}/av1/encoder/mips/msa/fdct4x4_msa.c"
"${AOM_ROOT}/av1/encoder/mips/msa/fdct8x8_msa.c"
"${AOM_ROOT}/av1/encoder/mips/msa/fdct_msa.h"
"${AOM_ROOT}/av1/encoder/mips/msa/temporal_filter_msa.c")
if (CONFIG_HIGHBITDEPTH)
set(AOM_AV1_COMMON_INTRIN_SSE4_1
${AOM_AV1_COMMON_INTRIN_SSE4_1}
"${AOM_ROOT}/av1/common/x86/av1_highbd_convolve_sse4.c"
"${AOM_ROOT}/av1/common/x86/highbd_inv_txfm_sse4.c")
set(AOM_AV1_COMMON_INTRIN_AVX2
${AOM_AV1_COMMON_INTRIN_AVX2}
"${AOM_ROOT}/av1/common/x86/highbd_inv_txfm_avx2.c")
set(AOM_AV1_ENCODER_INTRIN_SSE4_1
${AOM_AV1_ENCODER_INTRIN_SSE4_1}
"${AOM_ROOT}/av1/encoder/x86/av1_highbd_quantize_sse4.c"
"${AOM_ROOT}/av1/encoder/x86/highbd_fwd_txfm_sse4.c")
else ()
set(AOM_AV1_COMMON_INTRIN_NEON
${AOM_AV1_COMMON_INTRIN_NEON}
"${AOM_ROOT}/av1/encoder/arm/neon/dct_neon.c"
"${AOM_ROOT}/av1/common/arm/neon/iht4x4_add_neon.c"
"${AOM_ROOT}/av1/common/arm/neon/iht8x8_add_neon.c")
set(AOM_AV1_ENCODER_INTRIN_NEON
${AOM_AV1_ENCODER_INTRIN_NEON}
"${AOM_ROOT}/av1/encoder/arm/neon/error_neon.c")
endif ()
if (CONFIG_CDEF)
set(AOM_AV1_COMMON_SOURCES
${AOM_AV1_COMMON_SOURCES}
"${AOM_ROOT}/av1/common/clpf.c"
"${AOM_ROOT}/av1/common/clpf.h"
"${AOM_ROOT}/av1/common/clpf_simd.h"
"${AOM_ROOT}/av1/common/cdef_simd.h"
"${AOM_ROOT}/av1/common/cdef.c"
"${AOM_ROOT}/av1/common/cdef.h"
"${AOM_ROOT}/av1/common/od_dering.c"
"${AOM_ROOT}/av1/common/od_dering.h"
"${AOM_ROOT}/av1/common/od_dering_simd.h")
set(AOM_AV1_ENCODER_SOURCES
${AOM_AV1_ENCODER_SOURCES}
"${AOM_ROOT}/av1/encoder/pickcdef.c")
set(AOM_AV1_COMMON_INTRIN_SSE2
${AOM_AV1_COMMON_INTRIN_SSE2}
"${AOM_ROOT}/av1/common/clpf_sse2.c"
"${AOM_ROOT}/av1/common/od_dering_sse2.c")
set(AOM_AV1_COMMON_INTRIN_SSSE3
${AOM_AV1_COMMON_INTRIN_SSSE3}
"${AOM_ROOT}/av1/common/clpf_ssse3.c"
"${AOM_ROOT}/av1/common/od_dering_ssse3.c")
set(AOM_AV1_COMMON_INTRIN_SSE4_1
${AOM_AV1_COMMON_INTRIN_SSE4_1}
"${AOM_ROOT}/av1/common/clpf_sse4.c"
"${AOM_ROOT}/av1/common/od_dering_sse4.c")
set(AOM_AV1_COMMON_INTRIN_NEON
${AOM_AV1_COMMON_INTRIN_NEON}
"${AOM_ROOT}/av1/common/clpf_neon.c"
"${AOM_ROOT}/av1/common/od_dering_neon.c")
endif ()
if (CONFIG_EXT_INTER)
set(AOM_AV1_ENCODER_SOURCES
${AOM_AV1_ENCODER_SOURCES}
"${AOM_ROOT}/av1/encoder/wedge_utils.c")
set(AOM_AV1_ENCODER_INTRIN_SSE2
${AOM_AV1_ENCODER_INTRIN_SSE2}
"${AOM_ROOT}/av1/encoder/x86/wedge_utils_sse2.c")
endif ()
if (CONFIG_FILTER_INTRA)
set(AOM_AV1_COMMON_INTRIN_SSE4_1
${AOM_AV1_COMMON_INTRIN_SSE4_1}
"${AOM_ROOT}/av1/common/x86/filterintra_sse4.c")
endif ()
if (CONFIG_ACCOUNTING)
set(AOM_AV1_DECODER_SOURCES
${AOM_AV1_DECODER_SOURCES}
"${AOM_ROOT}/av1/decoder/accounting.c"
"${AOM_ROOT}/av1/decoder/accounting.h")
endif ()
if (CONFIG_INSPECTION)
set(AOM_AV1_DECODER_SOURCES
${AOM_AV1_DECODER_SOURCES}
"${AOM_ROOT}/av1/decoder/inspection.c"
"${AOM_ROOT}/av1/decoder/inspection.h")
endif ()
if (CONFIG_INTERNAL_STATS)
set(AOM_AV1_ENCODER_SOURCES
${AOM_AV1_ENCODER_SOURCES}
"${AOM_ROOT}/av1/encoder/blockiness.c")
endif ()
if (CONFIG_PALETTE)
set(AOM_AV1_ENCODER_SOURCES
${AOM_AV1_ENCODER_SOURCES}
"${AOM_ROOT}/av1/encoder/palette.c"
"${AOM_ROOT}/av1/encoder/palette.h")
endif ()
if (CONFIG_CFL)
set(AOM_AV1_COMMON_SOURCES
${AOM_AV1_COMMON_SOURCES}
"${AOM_ROOT}/av1/common/cfl.c"
"${AOM_ROOT}/av1/common/cfl.h")
endif ()
if (CONFIG_PVQ)
set(AOM_AV1_COMMON_SOURCES
${AOM_AV1_COMMON_SOURCES}
"${AOM_ROOT}/av1/common/laplace_tables.c"
"${AOM_ROOT}/av1/common/pvq.c"
"${AOM_ROOT}/av1/common/pvq.h"
"${AOM_ROOT}/av1/common/pvq_state.c"
"${AOM_ROOT}/av1/common/pvq_state.h"
"${AOM_ROOT}/av1/common/partition.c"
"${AOM_ROOT}/av1/common/partition.h"
"${AOM_ROOT}/av1/common/generic_code.c"
"${AOM_ROOT}/av1/common/generic_code.h"
"${AOM_ROOT}/av1/common/zigzag4.c"
"${AOM_ROOT}/av1/common/zigzag8.c"
"${AOM_ROOT}/av1/common/zigzag16.c"
"${AOM_ROOT}/av1/common/zigzag32.c")
set(AOM_AV1_DECODER_SOURCES
${AOM_AV1_DECODER_SOURCES}
"${AOM_ROOT}/av1/decoder/decint.h"
"${AOM_ROOT}/av1/decoder/pvq_decoder.c"
"${AOM_ROOT}/av1/decoder/pvq_decoder.h"
"${AOM_ROOT}/av1/decoder/generic_decoder.c"
"${AOM_ROOT}/av1/decoder/laplace_decoder.c")
set(AOM_AV1_ENCODER_SOURCES
${AOM_AV1_ENCODER_SOURCES}
"${AOM_ROOT}/av1/encoder/daala_compat_enc.c"
"${AOM_ROOT}/av1/encoder/encint.h"
"${AOM_ROOT}/av1/encoder/pvq_encoder.c"
"${AOM_ROOT}/av1/encoder/pvq_encoder.h"
"${AOM_ROOT}/av1/encoder/generic_encoder.c"
"${AOM_ROOT}/av1/encoder/laplace_encoder.c")
set(AOM_AV1_COMMON_INTRIN_SSE4_1
${AOM_AV1_COMMON_INTRIN_SSE4_1}
"${AOM_ROOT}/av1/common/x86/pvq_sse4.c"
"${AOM_ROOT}/av1/common/x86/pvq_sse4.h")
if (NOT CONFIG_AV1_ENCODER)
# TODO(tomfinegan): These should probably be in av1/common, and in a
# common source list. For now this mirrors the original build system.
set(AOM_AV1_DECODER_SOURCES
${AOM_AV1_DECODER_SOURCES}
"${AOM_ROOT}/av1/encoder/dct.c"
"${AOM_ROOT}/av1/encoder/hybrid_fwd_txfm.c"
"${AOM_ROOT}/av1/encoder/hybrid_fwd_txfm.h")
set(AOM_AV1_DECODER_ASM_SSE2
${AOM_AV1_DECODER_ASM_SSE2}
"${AOM_ROOT}/av1/encoder/x86/dct_sse2.asm")
set(AOM_AV1_DECODER_INTRIN_SSE2
${AOM_AV1_DECODER_INTRIN_SSE2}
"${AOM_ROOT}/av1/encoder/x86/dct_intrin_sse2.c")
set(AOM_AV1_DECODER_INTRIN_SSSE3
${AOM_AV1_DECODER_INTRIN_SSSE3}
"${AOM_ROOT}/av1/encoder/x86/dct_ssse3.c")
endif ()
endif ()
if (CONFIG_WARPED_MOTION)
set(AOM_AV1_COMMON_SOURCES
${AOM_AV1_COMMON_SOURCES}
"${AOM_ROOT}/av1/common/warped_motion.c"
"${AOM_ROOT}/av1/common/warped_motion.h")
set(AOM_AV1_COMMON_INTRIN_SSE2
${AOM_AV1_COMMON_INTRIN_SSE2}
"${AOM_ROOT}/av1/common/x86/warp_plane_sse2.c")
endif ()
# Setup AV1 common/decoder/encoder targets. The libaom target must exist before
# this function is called.
function (setup_av1_targets)
add_library(aom_av1_common OBJECT ${AOM_AV1_COMMON_SOURCES})
set(AOM_LIB_TARGETS ${AOM_LIB_TARGETS} aom_av1_common)
target_sources(aom PUBLIC $<TARGET_OBJECTS:aom_av1_common>)
if (CONFIG_AV1_DECODER)
add_library(aom_av1_decoder OBJECT ${AOM_AV1_DECODER_SOURCES})
set(AOM_LIB_TARGETS ${AOM_LIB_TARGETS} aom_av1_decoder)
target_sources(aom PUBLIC $<TARGET_OBJECTS:aom_av1_decoder>)
endif ()
if (CONFIG_AV1_ENCODER)
add_library(aom_av1_encoder OBJECT ${AOM_AV1_ENCODER_SOURCES})
set(AOM_LIB_TARGETS ${AOM_LIB_TARGETS} aom_av1_encoder)
target_sources(aom PUBLIC $<TARGET_OBJECTS:aom_av1_encoder>)
endif ()
if (HAVE_SSE2)
require_flag_nomsvc("-msse2" NO)
add_intrinsics_object_library("-msse2" "sse2" "aom_av1_common"
"AOM_AV1_COMMON_INTRIN_SSE2")
if (CONFIG_AV1_DECODER)
if (AOM_AV1_DECODER_ASM_SSE2)
add_asm_library("aom_av1_decoder_sse2" "AOM_AV1_DECODER_ASM_SSE2" "aom")
endif ()
if (AOM_AV1_DECODER_INTRIN_SSE2)
add_intrinsics_object_library("-msse2" "sse2" "aom_av1_decoder"
"AOM_AV1_DECODER_INTRIN_SSE2")
endif ()
endif ()
if (CONFIG_AV1_ENCODER)
add_asm_library("aom_av1_encoder_sse2" "AOM_AV1_ENCODER_ASM_SSE2" "aom")
add_intrinsics_object_library("-msse2" "sse2" "aom_av1_encoder"
"AOM_AV1_ENCODER_INTRIN_SSE2")
endif ()
endif ()
if (HAVE_SSSE3)
require_flag_nomsvc("-mssse3" NO)
add_intrinsics_object_library("-mssse3" "ssse3" "aom_av1_common"
"AOM_AV1_COMMON_INTRIN_SSSE3")
if (CONFIG_AV1_DECODER)
if (AOM_AV1_DECODER_INTRIN_SSSE3)
add_intrinsics_object_library("-mssse3" "ssse3" "aom_av1_decoder"
"AOM_AV1_DECODER_INTRIN_SSSE3")
endif ()
endif ()
if (CONFIG_AV1_ENCODER)
add_intrinsics_object_library("-mssse3" "ssse3" "aom_av1_encoder"
"AOM_AV1_ENCODER_INTRIN_SSSE3")
endif ()
endif ()
if (HAVE_SSE4_1)
require_flag_nomsvc("-msse4.1" NO)
add_intrinsics_object_library("-msse4.1" "sse4" "aom_av1_common"
"AOM_AV1_COMMON_INTRIN_SSE4_1")
if (CONFIG_AV1_ENCODER)
if ("${AOM_TARGET_CPU}" STREQUAL "x86_64")
add_asm_library("aom_av1_encoder_ssse3"
"AOM_AV1_ENCODER_ASM_SSSE3_X86_64" "aom")
endif ()
if (AOM_AV1_ENCODER_INTRIN_SSE4_1)
add_intrinsics_object_library("-msse4.1" "sse4" "aom_av1_encoder"
"AOM_AV1_ENCODER_INTRIN_SSE4_1")
endif ()
endif ()
endif ()
if (HAVE_AVX2)
require_flag_nomsvc("-mavx2" NO)
add_intrinsics_object_library("-mavx2" "avx2" "aom_av1_common"
"AOM_AV1_COMMON_INTRIN_AVX2")
if (CONFIG_AV1_ENCODER)
add_intrinsics_object_library("-mavx2" "avx2" "aom_av1_encoder"
"AOM_AV1_ENCODER_INTRIN_AVX2")
endif ()
endif ()
if (HAVE_NEON)
if (AOM_AV1_COMMON_INTRIN_NEON)
add_intrinsics_object_library("${AOM_INTRIN_NEON_FLAG}"
"neon"
"aom_av1_common"
"AOM_AV1_COMMON_INTRIN_NEON")
endif ()
if (AOM_AV1_ENCODER_INTRIN_NEON)
add_intrinsics_object_library("${AOM_INTRIN_NEON_FLAG}"
"neon"
"aom_av1_encoder"
"AOM_AV1_ENCODER_INTRIN_NEON")
endif ()
endif ()
if (HAVE_DSPR2)
add_intrinsics_object_library("" "dspr2" "aom_av1_common"
"AOM_AV1_COMMON_INTRIN_DSPR2")
endif ()
if (HAVE_MSA)
add_intrinsics_object_library("" "msa" "aom_av1_common"
"AOM_AV1_COMMON_INTRIN_MSA")
add_intrinsics_object_library("" "msa" "aom_av1_encoder"
"AOM_AV1_ENCODER_INTRIN_MSA")
endif ()
# Pass the new lib targets up to the parent scope instance of
# $AOM_LIB_TARGETS.
set(AOM_LIB_TARGETS ${AOM_LIB_TARGETS} PARENT_SCOPE)
endfunction ()
function (setup_av1_test_targets)
endfunction ()

180
third_party/aom/av1/av1_common.mk vendored Normal file
View file

@ -0,0 +1,180 @@
##
## Copyright (c) 2016, Alliance for Open Media. All rights reserved
##
## This source code is subject to the terms of the BSD 2 Clause License and
## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
## was not distributed with this source code in the LICENSE file, you can
## obtain it at www.aomedia.org/license/software. If the Alliance for Open
## Media Patent License 1.0 was not distributed with this source code in the
## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
##
AV1_COMMON_SRCS-yes += av1_common.mk
AV1_COMMON_SRCS-yes += av1_iface_common.h
AV1_COMMON_SRCS-yes += common/alloccommon.c
AV1_COMMON_SRCS-yes += common/av1_loopfilter.c
AV1_COMMON_SRCS-yes += common/av1_loopfilter.h
AV1_COMMON_SRCS-yes += common/blockd.c
AV1_COMMON_SRCS-yes += common/debugmodes.c
AV1_COMMON_SRCS-yes += common/entropy.c
AV1_COMMON_SRCS-yes += common/entropymode.c
AV1_COMMON_SRCS-yes += common/entropymv.c
AV1_COMMON_SRCS-yes += common/frame_buffers.c
AV1_COMMON_SRCS-yes += common/frame_buffers.h
AV1_COMMON_SRCS-yes += common/alloccommon.h
AV1_COMMON_SRCS-yes += common/blockd.h
AV1_COMMON_SRCS-yes += common/common.h
AV1_COMMON_SRCS-yes += common/entropy.h
AV1_COMMON_SRCS-yes += common/entropymode.h
AV1_COMMON_SRCS-yes += common/entropymv.h
AV1_COMMON_SRCS-yes += common/enums.h
AV1_COMMON_SRCS-yes += common/filter.h
AV1_COMMON_SRCS-yes += common/filter.c
AV1_COMMON_SRCS-yes += common/idct.h
AV1_COMMON_SRCS-yes += common/idct.c
AV1_COMMON_SRCS-yes += common/thread_common.h
AV1_COMMON_SRCS-$(CONFIG_LV_MAP) += common/txb_common.h
AV1_COMMON_SRCS-$(CONFIG_LV_MAP) += common/txb_common.c
AV1_COMMON_SRCS-yes += common/mv.h
AV1_COMMON_SRCS-yes += common/onyxc_int.h
AV1_COMMON_SRCS-yes += common/pred_common.h
AV1_COMMON_SRCS-yes += common/pred_common.c
AV1_COMMON_SRCS-yes += common/quant_common.h
AV1_COMMON_SRCS-yes += common/reconinter.h
AV1_COMMON_SRCS-yes += common/reconintra.h
AV1_COMMON_SRCS-yes += common/av1_rtcd.c
AV1_COMMON_SRCS-yes += common/av1_rtcd_defs.pl
AV1_COMMON_SRCS-yes += common/scale.h
AV1_COMMON_SRCS-yes += common/scale.c
AV1_COMMON_SRCS-yes += common/seg_common.h
AV1_COMMON_SRCS-yes += common/seg_common.c
AV1_COMMON_SRCS-yes += common/tile_common.h
AV1_COMMON_SRCS-yes += common/tile_common.c
AV1_COMMON_SRCS-yes += common/thread_common.c
AV1_COMMON_SRCS-yes += common/mvref_common.c
AV1_COMMON_SRCS-yes += common/mvref_common.h
AV1_COMMON_SRCS-yes += common/quant_common.c
AV1_COMMON_SRCS-yes += common/reconinter.c
AV1_COMMON_SRCS-yes += common/reconintra.c
AV1_COMMON_SRCS-yes += common/resize.c
AV1_COMMON_SRCS-yes += common/resize.h
AV1_COMMON_SRCS-yes += common/restoration.h
AV1_COMMON_SRCS-yes += common/common_data.h
AV1_COMMON_SRCS-yes += common/scan.c
AV1_COMMON_SRCS-yes += common/scan.h
# TODO(angiebird) the forward transform belongs under encoder/
AV1_COMMON_SRCS-yes += common/av1_txfm.h
AV1_COMMON_SRCS-yes += common/av1_fwd_txfm1d.h
AV1_COMMON_SRCS-yes += common/av1_fwd_txfm1d.c
AV1_COMMON_SRCS-yes += common/av1_inv_txfm1d.h
AV1_COMMON_SRCS-yes += common/av1_inv_txfm1d.c
AV1_COMMON_SRCS-yes += common/av1_fwd_txfm2d.c
AV1_COMMON_SRCS-yes += common/av1_fwd_txfm2d_cfg.h
AV1_COMMON_SRCS-yes += common/av1_inv_txfm2d.c
AV1_COMMON_SRCS-yes += common/av1_inv_txfm2d_cfg.h
AV1_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/av1_convolve_ssse3.c
ifeq ($(CONFIG_HIGHBITDEPTH),yes)
AV1_COMMON_SRCS-$(HAVE_SSE4_1) += common/x86/av1_highbd_convolve_sse4.c
endif
AV1_COMMON_SRCS-yes += common/convolve.c
AV1_COMMON_SRCS-yes += common/convolve.h
ifeq ($(CONFIG_LOOP_RESTORATION),yes)
AV1_COMMON_SRCS-yes += common/restoration.h
AV1_COMMON_SRCS-yes += common/restoration.c
AV1_COMMON_SRCS-$(HAVE_SSE4_1) += common/x86/selfguided_sse4.c
endif
ifeq (yes,$(filter $(CONFIG_GLOBAL_MOTION) $(CONFIG_WARPED_MOTION),yes))
AV1_COMMON_SRCS-yes += common/warped_motion.h
AV1_COMMON_SRCS-yes += common/warped_motion.c
endif
ifeq ($(CONFIG_CDEF),yes)
AV1_COMMON_SRCS-yes += common/clpf.c
AV1_COMMON_SRCS-yes += common/clpf.h
AV1_COMMON_SRCS-yes += common/clpf_simd.h
AV1_COMMON_SRCS-yes += common/cdef_simd.h
AV1_COMMON_SRCS-$(HAVE_SSE2) += common/clpf_sse2.c
AV1_COMMON_SRCS-$(HAVE_SSSE3) += common/clpf_ssse3.c
AV1_COMMON_SRCS-$(HAVE_SSE4_1) += common/clpf_sse4.c
AV1_COMMON_SRCS-$(HAVE_NEON) += common/clpf_neon.c
AV1_COMMON_SRCS-$(HAVE_SSE2) += common/od_dering_sse2.c
AV1_COMMON_SRCS-$(HAVE_SSSE3) += common/od_dering_ssse3.c
AV1_COMMON_SRCS-$(HAVE_SSE4_1) += common/od_dering_sse4.c
AV1_COMMON_SRCS-$(HAVE_NEON) += common/od_dering_neon.c
AV1_COMMON_SRCS-yes += common/od_dering.c
AV1_COMMON_SRCS-yes += common/od_dering.h
AV1_COMMON_SRCS-yes += common/od_dering_simd.h
AV1_COMMON_SRCS-yes += common/cdef.c
AV1_COMMON_SRCS-yes += common/cdef.h
endif
AV1_COMMON_SRCS-yes += common/odintrin.c
AV1_COMMON_SRCS-yes += common/odintrin.h
ifeq ($(CONFIG_CFL),yes)
AV1_COMMON_SRCS-yes += common/cfl.h
AV1_COMMON_SRCS-yes += common/cfl.c
endif
ifeq ($(CONFIG_PVQ),yes)
# PVQ from daala
AV1_COMMON_SRCS-yes += common/pvq.c
AV1_COMMON_SRCS-yes += common/partition.c
AV1_COMMON_SRCS-yes += common/partition.h
AV1_COMMON_SRCS-yes += common/zigzag4.c
AV1_COMMON_SRCS-yes += common/zigzag8.c
AV1_COMMON_SRCS-yes += common/zigzag16.c
AV1_COMMON_SRCS-yes += common/zigzag32.c
AV1_COMMON_SRCS-yes += common/zigzag.h
AV1_COMMON_SRCS-yes += common/generic_code.c
AV1_COMMON_SRCS-yes += common/pvq_state.c
AV1_COMMON_SRCS-yes += common/laplace_tables.c
AV1_COMMON_SRCS-$(HAVE_SSE4_1) += common/x86/pvq_sse4.c
AV1_COMMON_SRCS-$(HAVE_SSE4_1) += common/x86/pvq_sse4.h
endif
ifneq ($(findstring yes,$(CONFIG_PVQ)$(CONFIG_DAALA_DIST)$(CONFIG_XIPHRC)),)
AV1_COMMON_SRCS-yes += common/pvq.h
AV1_COMMON_SRCS-yes += common/pvq_state.h
AV1_COMMON_SRCS-yes += common/generic_code.h
endif
ifneq ($(CONFIG_HIGHBITDEPTH),yes)
AV1_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/av1_itrans4_dspr2.c
AV1_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/av1_itrans8_dspr2.c
AV1_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/av1_itrans16_dspr2.c
endif
# common (msa)
AV1_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/av1_idct4x4_msa.c
AV1_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/av1_idct8x8_msa.c
AV1_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/av1_idct16x16_msa.c
AV1_COMMON_SRCS-$(HAVE_SSE2) += common/x86/idct_intrin_sse2.c
AV1_COMMON_SRCS-$(HAVE_AVX2) += common/x86/hybrid_inv_txfm_avx2.c
ifeq ($(CONFIG_AV1_ENCODER),yes)
AV1_COMMON_SRCS-$(HAVE_SSE4_1) += common/x86/av1_txfm1d_sse4.h
AV1_COMMON_SRCS-$(HAVE_SSE4_1) += common/x86/av1_fwd_txfm1d_sse4.c
AV1_COMMON_SRCS-$(HAVE_SSE4_1) += common/x86/av1_fwd_txfm2d_sse4.c
endif
ifeq ($(CONFIG_HIGHBITDEPTH),yes)
AV1_COMMON_SRCS-$(HAVE_SSE4_1) += common/x86/highbd_txfm_utility_sse4.h
AV1_COMMON_SRCS-$(HAVE_SSE4_1) += common/x86/highbd_inv_txfm_sse4.c
AV1_COMMON_SRCS-$(HAVE_AVX2) += common/x86/highbd_inv_txfm_avx2.c
endif
ifneq ($(CONFIG_HIGHBITDEPTH),yes)
AV1_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/iht4x4_add_neon.c
AV1_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/iht8x8_add_neon.c
endif
ifeq ($(CONFIG_FILTER_INTRA),yes)
AV1_COMMON_SRCS-$(HAVE_SSE4_1) += common/x86/filterintra_sse4.c
endif
ifneq ($(findstring yes,$(CONFIG_GLOBAL_MOTION) $(CONFIG_WARPED_MOTION)),)
AV1_COMMON_SRCS-$(HAVE_SSE2) += common/x86/warp_plane_sse2.c
ifeq ($(CONFIG_HIGHBITDEPTH),yes)
AV1_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/highbd_warp_plane_ssse3.c
endif
endif
$(eval $(call rtcd_h_template,av1_rtcd,av1/common/av1_rtcd_defs.pl))

165
third_party/aom/av1/av1_cx.mk vendored Normal file
View file

@ -0,0 +1,165 @@
##
## Copyright (c) 2016, Alliance for Open Media. All rights reserved
##
## This source code is subject to the terms of the BSD 2 Clause License and
## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
## was not distributed with this source code in the LICENSE file, you can
## obtain it at www.aomedia.org/license/software. If the Alliance for Open
## Media Patent License 1.0 was not distributed with this source code in the
## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
##
AV1_CX_EXPORTS += exports_enc
AV1_CX_SRCS-yes += $(AV1_COMMON_SRCS-yes)
AV1_CX_SRCS-no += $(AV1_COMMON_SRCS-no)
AV1_CX_SRCS_REMOVE-yes += $(AV1_COMMON_SRCS_REMOVE-yes)
AV1_CX_SRCS_REMOVE-no += $(AV1_COMMON_SRCS_REMOVE-no)
AV1_CX_SRCS-yes += av1_cx_iface.c
AV1_CX_SRCS-yes += encoder/av1_quantize.c
AV1_CX_SRCS-yes += encoder/av1_quantize.h
AV1_CX_SRCS-yes += encoder/bitstream.c
AV1_CX_SRCS-yes += encoder/context_tree.c
AV1_CX_SRCS-yes += encoder/context_tree.h
AV1_CX_SRCS-yes += encoder/variance_tree.c
AV1_CX_SRCS-yes += encoder/variance_tree.h
AV1_CX_SRCS-yes += encoder/cost.h
AV1_CX_SRCS-yes += encoder/cost.c
AV1_CX_SRCS-yes += encoder/dct.c
AV1_CX_SRCS-yes += encoder/hybrid_fwd_txfm.c
AV1_CX_SRCS-yes += encoder/hybrid_fwd_txfm.h
AV1_CX_SRCS-yes += encoder/encodeframe.c
AV1_CX_SRCS-yes += encoder/encodeframe.h
AV1_CX_SRCS-yes += encoder/encodemb.c
AV1_CX_SRCS-yes += encoder/encodemv.c
AV1_CX_SRCS-yes += encoder/ethread.h
AV1_CX_SRCS-yes += encoder/ethread.c
AV1_CX_SRCS-yes += encoder/extend.c
AV1_CX_SRCS-yes += encoder/firstpass.c
AV1_CX_SRCS-$(CONFIG_GLOBAL_MOTION) += ../third_party/fastfeat/fast.h
AV1_CX_SRCS-$(CONFIG_GLOBAL_MOTION) += ../third_party/fastfeat/nonmax.c
AV1_CX_SRCS-$(CONFIG_GLOBAL_MOTION) += ../third_party/fastfeat/fast_9.c
AV1_CX_SRCS-$(CONFIG_GLOBAL_MOTION) += ../third_party/fastfeat/fast.c
AV1_CX_SRCS-$(CONFIG_GLOBAL_MOTION) += encoder/corner_match.c
AV1_CX_SRCS-$(CONFIG_GLOBAL_MOTION) += encoder/corner_match.h
AV1_CX_SRCS-$(CONFIG_GLOBAL_MOTION) += encoder/corner_detect.c
AV1_CX_SRCS-$(CONFIG_GLOBAL_MOTION) += encoder/corner_detect.h
AV1_CX_SRCS-$(CONFIG_GLOBAL_MOTION) += encoder/global_motion.c
AV1_CX_SRCS-$(CONFIG_GLOBAL_MOTION) += encoder/global_motion.h
AV1_CX_SRCS-$(CONFIG_GLOBAL_MOTION) += encoder/ransac.c
AV1_CX_SRCS-$(CONFIG_GLOBAL_MOTION) += encoder/ransac.h
AV1_CX_SRCS-yes += encoder/block.h
AV1_CX_SRCS-yes += encoder/bitstream.h
AV1_CX_SRCS-yes += encoder/encodemb.h
AV1_CX_SRCS-yes += encoder/encodemv.h
AV1_CX_SRCS-$(CONFIG_LV_MAP) += encoder/encodetxb.c
AV1_CX_SRCS-$(CONFIG_LV_MAP) += encoder/encodetxb.h
AV1_CX_SRCS-yes += encoder/extend.h
AV1_CX_SRCS-yes += encoder/firstpass.h
AV1_CX_SRCS-yes += encoder/lookahead.c
AV1_CX_SRCS-yes += encoder/lookahead.h
AV1_CX_SRCS-yes += encoder/mcomp.h
AV1_CX_SRCS-yes += encoder/encoder.h
AV1_CX_SRCS-yes += encoder/ratectrl.h
ifeq ($(CONFIG_XIPHRC),yes)
AV1_CX_SRCS-yes += encoder/ratectrl_xiph.h
endif
AV1_CX_SRCS-yes += encoder/rd.h
AV1_CX_SRCS-yes += encoder/rdopt.h
AV1_CX_SRCS-yes += encoder/tokenize.h
AV1_CX_SRCS-yes += encoder/treewriter.h
AV1_CX_SRCS-yes += encoder/mcomp.c
AV1_CX_SRCS-yes += encoder/encoder.c
ifeq ($(CONFIG_PALETTE),yes)
AV1_CX_SRCS-yes += encoder/palette.h
AV1_CX_SRCS-yes += encoder/palette.c
endif
AV1_CX_SRCS-yes += encoder/picklpf.c
AV1_CX_SRCS-yes += encoder/picklpf.h
AV1_CX_SRCS-$(CONFIG_LOOP_RESTORATION) += encoder/pickrst.c
AV1_CX_SRCS-$(CONFIG_LOOP_RESTORATION) += encoder/pickrst.h
AV1_CX_SRCS-yes += encoder/ratectrl.c
ifeq ($(CONFIG_XIPHRC),yes)
AV1_CX_SRCS-yes += encoder/ratectrl_xiph.c
endif
AV1_CX_SRCS-yes += encoder/rd.c
AV1_CX_SRCS-yes += encoder/rdopt.c
AV1_CX_SRCS-yes += encoder/segmentation.c
AV1_CX_SRCS-yes += encoder/segmentation.h
AV1_CX_SRCS-yes += encoder/speed_features.c
AV1_CX_SRCS-yes += encoder/speed_features.h
AV1_CX_SRCS-yes += encoder/subexp.c
AV1_CX_SRCS-yes += encoder/subexp.h
AV1_CX_SRCS-$(CONFIG_INTERNAL_STATS) += encoder/blockiness.c
AV1_CX_SRCS-yes += encoder/tokenize.c
AV1_CX_SRCS-yes += encoder/treewriter.c
AV1_CX_SRCS-yes += encoder/aq_variance.c
AV1_CX_SRCS-yes += encoder/aq_variance.h
AV1_CX_SRCS-yes += encoder/aq_cyclicrefresh.c
AV1_CX_SRCS-yes += encoder/aq_cyclicrefresh.h
AV1_CX_SRCS-yes += encoder/aq_complexity.c
AV1_CX_SRCS-yes += encoder/aq_complexity.h
AV1_CX_SRCS-yes += encoder/temporal_filter.c
AV1_CX_SRCS-yes += encoder/temporal_filter.h
AV1_CX_SRCS-yes += encoder/mbgraph.c
AV1_CX_SRCS-yes += encoder/mbgraph.h
ifeq ($(CONFIG_CDEF),yes)
AV1_CX_SRCS-yes += encoder/pickcdef.c
endif
ifeq ($(CONFIG_PVQ),yes)
# PVQ from daala
AV1_CX_SRCS-yes += encoder/daala_compat_enc.c
AV1_CX_SRCS-yes += encoder/pvq_encoder.c
AV1_CX_SRCS-yes += encoder/pvq_encoder.h
AV1_CX_SRCS-yes += encoder/generic_encoder.c
AV1_CX_SRCS-yes += encoder/laplace_encoder.c
endif
ifneq ($(findstring yes,$(CONFIG_XIPHRC)$(CONFIG_PVQ)),)
AV1_CX_SRCS-yes += encoder/encint.h
endif
AV1_CX_SRCS-$(HAVE_SSE2) += encoder/x86/av1_quantize_sse2.c
AV1_CX_SRCS-$(HAVE_SSE2) += encoder/x86/temporal_filter_apply_sse2.asm
ifeq ($(CONFIG_HIGHBITDEPTH),yes)
AV1_CX_SRCS-$(HAVE_SSE2) += encoder/x86/highbd_block_error_intrin_sse2.c
endif
AV1_CX_SRCS-$(HAVE_SSE2) += encoder/x86/dct_sse2.asm
AV1_CX_SRCS-$(HAVE_SSE2) += encoder/x86/error_sse2.asm
ifeq ($(ARCH_X86_64),yes)
AV1_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/av1_quantize_ssse3_x86_64.asm
endif
AV1_CX_SRCS-$(HAVE_SSE2) += encoder/x86/dct_intrin_sse2.c
AV1_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/dct_ssse3.c
AV1_CX_SRCS-$(HAVE_AVX2) += encoder/x86/hybrid_fwd_txfm_avx2.c
ifeq ($(CONFIG_HIGHBITDEPTH),yes)
AV1_CX_SRCS-$(HAVE_SSE4_1) += encoder/x86/av1_highbd_quantize_sse4.c
AV1_CX_SRCS-$(HAVE_SSE4_1) += encoder/x86/highbd_fwd_txfm_sse4.c
endif
ifeq ($(CONFIG_EXT_INTER),yes)
AV1_CX_SRCS-yes += encoder/wedge_utils.c
AV1_CX_SRCS-$(HAVE_SSE2) += encoder/x86/wedge_utils_sse2.c
endif
AV1_CX_SRCS-$(HAVE_AVX2) += encoder/x86/error_intrin_avx2.c
ifneq ($(CONFIG_HIGHBITDEPTH),yes)
AV1_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/dct_neon.c
AV1_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/error_neon.c
endif
AV1_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/quantize_neon.c
AV1_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/error_msa.c
AV1_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/fdct4x4_msa.c
AV1_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/fdct8x8_msa.c
AV1_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/fdct16x16_msa.c
AV1_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/fdct_msa.h
AV1_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/temporal_filter_msa.c
AV1_CX_SRCS-yes := $(filter-out $(AV1_CX_SRCS_REMOVE-yes),$(AV1_CX_SRCS-yes))

1605
third_party/aom/av1/av1_cx_iface.c vendored Normal file
View file

File diff suppressed because it is too large Load diff

71
third_party/aom/av1/av1_dx.mk vendored Normal file
View file

@ -0,0 +1,71 @@
##
## Copyright (c) 2016, Alliance for Open Media. All rights reserved
##
## This source code is subject to the terms of the BSD 2 Clause License and
## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
## was not distributed with this source code in the LICENSE file, you can
## obtain it at www.aomedia.org/license/software. If the Alliance for Open
## Media Patent License 1.0 was not distributed with this source code in the
## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
##
AV1_DX_EXPORTS += exports_dec
AV1_DX_SRCS-yes += $(AV1_COMMON_SRCS-yes)
AV1_DX_SRCS-no += $(AV1_COMMON_SRCS-no)
AV1_DX_SRCS_REMOVE-yes += $(AV1_COMMON_SRCS_REMOVE-yes)
AV1_DX_SRCS_REMOVE-no += $(AV1_COMMON_SRCS_REMOVE-no)
AV1_DX_SRCS-yes += av1_dx_iface.c
AV1_DX_SRCS-yes += decoder/decodemv.c
AV1_DX_SRCS-yes += decoder/decodeframe.c
AV1_DX_SRCS-yes += decoder/decodeframe.h
AV1_DX_SRCS-yes += decoder/detokenize.c
AV1_DX_SRCS-yes += decoder/decodemv.h
AV1_DX_SRCS-$(CONFIG_LV_MAP) += decoder/decodetxb.c
AV1_DX_SRCS-$(CONFIG_LV_MAP) += decoder/decodetxb.h
AV1_DX_SRCS-yes += decoder/detokenize.h
AV1_DX_SRCS-yes += decoder/dthread.c
AV1_DX_SRCS-yes += decoder/dthread.h
AV1_DX_SRCS-yes += decoder/decoder.c
AV1_DX_SRCS-yes += decoder/decoder.h
AV1_DX_SRCS-yes += decoder/dsubexp.c
AV1_DX_SRCS-yes += decoder/dsubexp.h
ifeq ($(CONFIG_ACCOUNTING),yes)
AV1_DX_SRCS-yes += decoder/accounting.h
AV1_DX_SRCS-yes += decoder/accounting.c
endif
ifeq ($(CONFIG_INSPECTION),yes)
AV1_DX_SRCS-yes += decoder/inspection.c
AV1_DX_SRCS-yes += decoder/inspection.h
endif
ifeq ($(CONFIG_PVQ),yes)
# PVQ from daala
AV1_DX_SRCS-yes += decoder/pvq_decoder.c
AV1_DX_SRCS-yes += decoder/pvq_decoder.h
AV1_DX_SRCS-yes += decoder/decint.h
AV1_DX_SRCS-yes += decoder/generic_decoder.c
AV1_DX_SRCS-yes += decoder/laplace_decoder.c
AV1_DX_SRCS-yes += encoder/hybrid_fwd_txfm.c
AV1_DX_SRCS-yes += encoder/hybrid_fwd_txfm.h
AV1_DX_SRCS-yes += encoder/dct.c
AV1_DX_SRCS-$(HAVE_SSE2) += encoder/x86/dct_sse2.asm
AV1_DX_SRCS-$(HAVE_SSE2) += encoder/x86/dct_intrin_sse2.c
AV1_DX_SRCS-$(HAVE_SSSE3) += encoder/x86/dct_ssse3.c
ifneq ($(CONFIG_HIGHBITDEPTH),yes)
AV1_DX_SRCS-$(HAVE_NEON) += encoder/arm/neon/dct_neon.c
endif
AV1_DX_SRCS-$(HAVE_MSA) += encoder/mips/msa/fdct4x4_msa.c
AV1_DX_SRCS-$(HAVE_MSA) += encoder/mips/msa/fdct8x8_msa.c
AV1_DX_SRCS-$(HAVE_MSA) += encoder/mips/msa/fdct16x16_msa.c
AV1_DX_SRCS-$(HAVE_MSA) += encoder/mips/msa/fdct_msa.h
endif
AV1_DX_SRCS-yes := $(filter-out $(AV1_DX_SRCS_REMOVE-yes),$(AV1_DX_SRCS-yes))

1223
third_party/aom/av1/av1_dx_iface.c vendored Normal file
View file

File diff suppressed because it is too large Load diff

146
third_party/aom/av1/av1_iface_common.h vendored Normal file
View file

@ -0,0 +1,146 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_AV1_IFACE_COMMON_H_
#define AV1_AV1_IFACE_COMMON_H_
#include "aom_ports/mem.h"
static void yuvconfig2image(aom_image_t *img, const YV12_BUFFER_CONFIG *yv12,
void *user_priv) {
/** aom_img_wrap() doesn't allow specifying independent strides for
* the Y, U, and V planes, nor other alignment adjustments that
* might be representable by a YV12_BUFFER_CONFIG, so we just
* initialize all the fields.*/
int bps;
if (!yv12->subsampling_y) {
if (!yv12->subsampling_x) {
img->fmt = AOM_IMG_FMT_I444;
bps = 24;
} else {
img->fmt = AOM_IMG_FMT_I422;
bps = 16;
}
} else {
if (!yv12->subsampling_x) {
img->fmt = AOM_IMG_FMT_I440;
bps = 16;
} else {
img->fmt = AOM_IMG_FMT_I420;
bps = 12;
}
}
img->cs = yv12->color_space;
img->range = yv12->color_range;
img->bit_depth = 8;
img->w = yv12->y_stride;
img->h = ALIGN_POWER_OF_TWO(yv12->y_height + 2 * AOM_BORDER_IN_PIXELS, 3);
img->d_w = yv12->y_crop_width;
img->d_h = yv12->y_crop_height;
img->r_w = yv12->render_width;
img->r_h = yv12->render_height;
img->x_chroma_shift = yv12->subsampling_x;
img->y_chroma_shift = yv12->subsampling_y;
img->planes[AOM_PLANE_Y] = yv12->y_buffer;
img->planes[AOM_PLANE_U] = yv12->u_buffer;
img->planes[AOM_PLANE_V] = yv12->v_buffer;
img->planes[AOM_PLANE_ALPHA] = NULL;
img->stride[AOM_PLANE_Y] = yv12->y_stride;
img->stride[AOM_PLANE_U] = yv12->uv_stride;
img->stride[AOM_PLANE_V] = yv12->uv_stride;
img->stride[AOM_PLANE_ALPHA] = yv12->y_stride;
#if CONFIG_HIGHBITDEPTH
if (yv12->flags & YV12_FLAG_HIGHBITDEPTH) {
// aom_image_t uses byte strides and a pointer to the first byte
// of the image.
img->fmt = (aom_img_fmt_t)(img->fmt | AOM_IMG_FMT_HIGHBITDEPTH);
img->bit_depth = yv12->bit_depth;
img->planes[AOM_PLANE_Y] = (uint8_t *)CONVERT_TO_SHORTPTR(yv12->y_buffer);
img->planes[AOM_PLANE_U] = (uint8_t *)CONVERT_TO_SHORTPTR(yv12->u_buffer);
img->planes[AOM_PLANE_V] = (uint8_t *)CONVERT_TO_SHORTPTR(yv12->v_buffer);
img->planes[AOM_PLANE_ALPHA] = NULL;
img->stride[AOM_PLANE_Y] = 2 * yv12->y_stride;
img->stride[AOM_PLANE_U] = 2 * yv12->uv_stride;
img->stride[AOM_PLANE_V] = 2 * yv12->uv_stride;
img->stride[AOM_PLANE_ALPHA] = 2 * yv12->y_stride;
}
#endif // CONFIG_HIGHBITDEPTH
img->bps = bps;
img->user_priv = user_priv;
img->img_data = yv12->buffer_alloc;
img->img_data_owner = 0;
img->self_allocd = 0;
}
static aom_codec_err_t image2yuvconfig(const aom_image_t *img,
YV12_BUFFER_CONFIG *yv12) {
yv12->y_buffer = img->planes[AOM_PLANE_Y];
yv12->u_buffer = img->planes[AOM_PLANE_U];
yv12->v_buffer = img->planes[AOM_PLANE_V];
yv12->y_crop_width = img->d_w;
yv12->y_crop_height = img->d_h;
yv12->render_width = img->r_w;
yv12->render_height = img->r_h;
yv12->y_width = img->d_w;
yv12->y_height = img->d_h;
yv12->uv_width =
img->x_chroma_shift == 1 ? (1 + yv12->y_width) / 2 : yv12->y_width;
yv12->uv_height =
img->y_chroma_shift == 1 ? (1 + yv12->y_height) / 2 : yv12->y_height;
yv12->uv_crop_width = yv12->uv_width;
yv12->uv_crop_height = yv12->uv_height;
yv12->y_stride = img->stride[AOM_PLANE_Y];
yv12->uv_stride = img->stride[AOM_PLANE_U];
yv12->color_space = img->cs;
yv12->color_range = img->range;
#if CONFIG_HIGHBITDEPTH
if (img->fmt & AOM_IMG_FMT_HIGHBITDEPTH) {
// In aom_image_t
// planes point to uint8 address of start of data
// stride counts uint8s to reach next row
// In YV12_BUFFER_CONFIG
// y_buffer, u_buffer, v_buffer point to uint16 address of data
// stride and border counts in uint16s
// This means that all the address calculations in the main body of code
// should work correctly.
// However, before we do any pixel operations we need to cast the address
// to a uint16 ponter and double its value.
yv12->y_buffer = CONVERT_TO_BYTEPTR(yv12->y_buffer);
yv12->u_buffer = CONVERT_TO_BYTEPTR(yv12->u_buffer);
yv12->v_buffer = CONVERT_TO_BYTEPTR(yv12->v_buffer);
yv12->y_stride >>= 1;
yv12->uv_stride >>= 1;
yv12->flags = YV12_FLAG_HIGHBITDEPTH;
} else {
yv12->flags = 0;
}
yv12->border = (yv12->y_stride - img->w) / 2;
#else
yv12->border = (img->stride[AOM_PLANE_Y] - img->w) / 2;
#endif // CONFIG_HIGHBITDEPTH
yv12->subsampling_x = img->x_chroma_shift;
yv12->subsampling_y = img->y_chroma_shift;
return AOM_CODEC_OK;
}
static AOM_REFFRAME ref_frame_to_av1_reframe(aom_ref_frame_type_t frame) {
switch (frame) {
case AOM_LAST_FRAME: return AOM_LAST_FLAG;
case AOM_GOLD_FRAME: return AOM_GOLD_FLAG;
case AOM_ALTR_FRAME: return AOM_ALT_FLAG;
}
assert(0 && "Invalid Reference Frame");
return AOM_LAST_FLAG;
}
#endif // AV1_AV1_IFACE_COMMON_H_

209
third_party/aom/av1/common/alloccommon.c vendored Normal file
View file

@ -0,0 +1,209 @@
/*
*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "./aom_config.h"
#include "aom_mem/aom_mem.h"
#include "av1/common/alloccommon.h"
#include "av1/common/blockd.h"
#include "av1/common/entropymode.h"
#include "av1/common/entropymv.h"
#include "av1/common/onyxc_int.h"
void av1_set_mb_mi(AV1_COMMON *cm, int width, int height) {
// TODO(jingning): Fine tune the loop filter operations and bring this
// back to integer multiple of 4 for cb4x4.
const int aligned_width = ALIGN_POWER_OF_TWO(width, 3);
const int aligned_height = ALIGN_POWER_OF_TWO(height, 3);
cm->mi_cols = aligned_width >> MI_SIZE_LOG2;
cm->mi_rows = aligned_height >> MI_SIZE_LOG2;
cm->mi_stride = calc_mi_size(cm->mi_cols);
#if CONFIG_CB4X4
cm->mb_cols = (cm->mi_cols + 2) >> 2;
cm->mb_rows = (cm->mi_rows + 2) >> 2;
#else
cm->mb_cols = (cm->mi_cols + 1) >> 1;
cm->mb_rows = (cm->mi_rows + 1) >> 1;
#endif
cm->MBs = cm->mb_rows * cm->mb_cols;
}
static int alloc_seg_map(AV1_COMMON *cm, int seg_map_size) {
int i;
for (i = 0; i < NUM_PING_PONG_BUFFERS; ++i) {
cm->seg_map_array[i] = (uint8_t *)aom_calloc(seg_map_size, 1);
if (cm->seg_map_array[i] == NULL) return 1;
}
cm->seg_map_alloc_size = seg_map_size;
// Init the index.
cm->seg_map_idx = 0;
cm->prev_seg_map_idx = 1;
cm->current_frame_seg_map = cm->seg_map_array[cm->seg_map_idx];
if (!cm->frame_parallel_decode)
cm->last_frame_seg_map = cm->seg_map_array[cm->prev_seg_map_idx];
return 0;
}
static void free_seg_map(AV1_COMMON *cm) {
int i;
for (i = 0; i < NUM_PING_PONG_BUFFERS; ++i) {
aom_free(cm->seg_map_array[i]);
cm->seg_map_array[i] = NULL;
}
cm->current_frame_seg_map = NULL;
if (!cm->frame_parallel_decode) {
cm->last_frame_seg_map = NULL;
}
}
void av1_free_ref_frame_buffers(BufferPool *pool) {
int i;
for (i = 0; i < FRAME_BUFFERS; ++i) {
if (pool->frame_bufs[i].ref_count > 0 &&
pool->frame_bufs[i].raw_frame_buffer.data != NULL) {
pool->release_fb_cb(pool->cb_priv, &pool->frame_bufs[i].raw_frame_buffer);
pool->frame_bufs[i].ref_count = 0;
}
aom_free(pool->frame_bufs[i].mvs);
pool->frame_bufs[i].mvs = NULL;
aom_free_frame_buffer(&pool->frame_bufs[i].buf);
}
}
#if CONFIG_LOOP_RESTORATION
// Assumes cm->rst_info[p].restoration_tilesize is already initialized
void av1_alloc_restoration_buffers(AV1_COMMON *cm) {
int p;
av1_alloc_restoration_struct(cm, &cm->rst_info[0], cm->width, cm->height);
for (p = 1; p < MAX_MB_PLANE; ++p)
av1_alloc_restoration_struct(
cm, &cm->rst_info[p], ROUND_POWER_OF_TWO(cm->width, cm->subsampling_x),
ROUND_POWER_OF_TWO(cm->height, cm->subsampling_y));
aom_free(cm->rst_internal.tmpbuf);
CHECK_MEM_ERROR(cm, cm->rst_internal.tmpbuf,
(int32_t *)aom_memalign(16, RESTORATION_TMPBUF_SIZE));
}
void av1_free_restoration_buffers(AV1_COMMON *cm) {
int p;
for (p = 0; p < MAX_MB_PLANE; ++p)
av1_free_restoration_struct(&cm->rst_info[p]);
aom_free(cm->rst_internal.tmpbuf);
cm->rst_internal.tmpbuf = NULL;
}
#endif // CONFIG_LOOP_RESTORATION
void av1_free_context_buffers(AV1_COMMON *cm) {
int i;
cm->free_mi(cm);
free_seg_map(cm);
for (i = 0; i < MAX_MB_PLANE; i++) {
aom_free(cm->above_context[i]);
cm->above_context[i] = NULL;
}
aom_free(cm->above_seg_context);
cm->above_seg_context = NULL;
#if CONFIG_VAR_TX
aom_free(cm->above_txfm_context);
cm->above_txfm_context = NULL;
#endif
}
int av1_alloc_context_buffers(AV1_COMMON *cm, int width, int height) {
int new_mi_size;
av1_set_mb_mi(cm, width, height);
new_mi_size = cm->mi_stride * calc_mi_size(cm->mi_rows);
if (cm->mi_alloc_size < new_mi_size) {
cm->free_mi(cm);
if (cm->alloc_mi(cm, new_mi_size)) goto fail;
}
if (cm->seg_map_alloc_size < cm->mi_rows * cm->mi_cols) {
// Create the segmentation map structure and set to 0.
free_seg_map(cm);
if (alloc_seg_map(cm, cm->mi_rows * cm->mi_cols)) goto fail;
}
if (cm->above_context_alloc_cols < cm->mi_cols) {
// TODO(geza.lore): These are bigger than they need to be.
// cm->tile_width would be enough but it complicates indexing a
// little elsewhere.
const int aligned_mi_cols =
ALIGN_POWER_OF_TWO(cm->mi_cols, MAX_MIB_SIZE_LOG2);
int i;
for (i = 0; i < MAX_MB_PLANE; i++) {
aom_free(cm->above_context[i]);
cm->above_context[i] = (ENTROPY_CONTEXT *)aom_calloc(
2 * aligned_mi_cols, sizeof(*cm->above_context[0]));
if (!cm->above_context[i]) goto fail;
}
aom_free(cm->above_seg_context);
cm->above_seg_context = (PARTITION_CONTEXT *)aom_calloc(
aligned_mi_cols, sizeof(*cm->above_seg_context));
if (!cm->above_seg_context) goto fail;
#if CONFIG_VAR_TX
aom_free(cm->above_txfm_context);
cm->above_txfm_context = (TXFM_CONTEXT *)aom_calloc(
aligned_mi_cols, sizeof(*cm->above_txfm_context));
if (!cm->above_txfm_context) goto fail;
#endif
cm->above_context_alloc_cols = aligned_mi_cols;
}
return 0;
fail:
// clear the mi_* values to force a realloc on resync
av1_set_mb_mi(cm, 0, 0);
av1_free_context_buffers(cm);
return 1;
}
void av1_remove_common(AV1_COMMON *cm) {
av1_free_context_buffers(cm);
aom_free(cm->fc);
cm->fc = NULL;
aom_free(cm->frame_contexts);
cm->frame_contexts = NULL;
}
void av1_init_context_buffers(AV1_COMMON *cm) {
cm->setup_mi(cm);
if (cm->last_frame_seg_map && !cm->frame_parallel_decode)
memset(cm->last_frame_seg_map, 0, cm->mi_rows * cm->mi_cols);
}
void av1_swap_current_and_last_seg_map(AV1_COMMON *cm) {
// Swap indices.
const int tmp = cm->seg_map_idx;
cm->seg_map_idx = cm->prev_seg_map_idx;
cm->prev_seg_map_idx = tmp;
cm->current_frame_seg_map = cm->seg_map_array[cm->seg_map_idx];
cm->last_frame_seg_map = cm->seg_map_array[cm->prev_seg_map_idx];
}

47
third_party/aom/av1/common/alloccommon.h vendored Normal file
View file

@ -0,0 +1,47 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_ALLOCCOMMON_H_
#define AV1_COMMON_ALLOCCOMMON_H_
#define INVALID_IDX -1 // Invalid buffer index.
#ifdef __cplusplus
extern "C" {
#endif
struct AV1Common;
struct BufferPool;
void av1_remove_common(struct AV1Common *cm);
int av1_alloc_context_buffers(struct AV1Common *cm, int width, int height);
void av1_init_context_buffers(struct AV1Common *cm);
void av1_free_context_buffers(struct AV1Common *cm);
void av1_free_ref_frame_buffers(struct BufferPool *pool);
#if CONFIG_LOOP_RESTORATION
void av1_alloc_restoration_buffers(struct AV1Common *cm);
void av1_free_restoration_buffers(struct AV1Common *cm);
#endif // CONFIG_LOOP_RESTORATION
int av1_alloc_state_buffers(struct AV1Common *cm, int width, int height);
void av1_free_state_buffers(struct AV1Common *cm);
void av1_set_mb_mi(struct AV1Common *cm, int width, int height);
void av1_swap_current_and_last_seg_map(struct AV1Common *cm);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_ALLOCCOMMON_H_

227
third_party/aom/av1/common/arm/neon/iht4x4_add_neon.c vendored Normal file
View file

@ -0,0 +1,227 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <arm_neon.h>
#include <assert.h>
#include "./aom_config.h"
#include "./av1_rtcd.h"
#include "aom_dsp/txfm_common.h"
#include "av1/common/common.h"
static INLINE void TRANSPOSE4X4(int16x8_t *q8s16, int16x8_t *q9s16) {
int32x4_t q8s32, q9s32;
int16x4x2_t d0x2s16, d1x2s16;
int32x4x2_t q0x2s32;
d0x2s16 = vtrn_s16(vget_low_s16(*q8s16), vget_high_s16(*q8s16));
d1x2s16 = vtrn_s16(vget_low_s16(*q9s16), vget_high_s16(*q9s16));
q8s32 = vreinterpretq_s32_s16(vcombine_s16(d0x2s16.val[0], d0x2s16.val[1]));
q9s32 = vreinterpretq_s32_s16(vcombine_s16(d1x2s16.val[0], d1x2s16.val[1]));
q0x2s32 = vtrnq_s32(q8s32, q9s32);
*q8s16 = vreinterpretq_s16_s32(q0x2s32.val[0]);
*q9s16 = vreinterpretq_s16_s32(q0x2s32.val[1]);
return;
}
static INLINE void GENERATE_COSINE_CONSTANTS(int16x4_t *d0s16, int16x4_t *d1s16,
int16x4_t *d2s16) {
*d0s16 = vdup_n_s16((int16_t)cospi_8_64);
*d1s16 = vdup_n_s16((int16_t)cospi_16_64);
*d2s16 = vdup_n_s16((int16_t)cospi_24_64);
return;
}
static INLINE void GENERATE_SINE_CONSTANTS(int16x4_t *d3s16, int16x4_t *d4s16,
int16x4_t *d5s16, int16x8_t *q3s16) {
*d3s16 = vdup_n_s16((int16_t)sinpi_1_9);
*d4s16 = vdup_n_s16((int16_t)sinpi_2_9);
*q3s16 = vdupq_n_s16((int16_t)sinpi_3_9);
*d5s16 = vdup_n_s16((int16_t)sinpi_4_9);
return;
}
static INLINE void IDCT4x4_1D(int16x4_t *d0s16, int16x4_t *d1s16,
int16x4_t *d2s16, int16x8_t *q8s16,
int16x8_t *q9s16) {
int16x4_t d16s16, d17s16, d18s16, d19s16, d23s16, d24s16;
int16x4_t d26s16, d27s16, d28s16, d29s16;
int32x4_t q10s32, q13s32, q14s32, q15s32;
int16x8_t q13s16, q14s16;
d16s16 = vget_low_s16(*q8s16);
d17s16 = vget_high_s16(*q8s16);
d18s16 = vget_low_s16(*q9s16);
d19s16 = vget_high_s16(*q9s16);
d23s16 = vadd_s16(d16s16, d18s16);
d24s16 = vsub_s16(d16s16, d18s16);
q15s32 = vmull_s16(d17s16, *d2s16);
q10s32 = vmull_s16(d17s16, *d0s16);
q13s32 = vmull_s16(d23s16, *d1s16);
q14s32 = vmull_s16(d24s16, *d1s16);
q15s32 = vmlsl_s16(q15s32, d19s16, *d0s16);
q10s32 = vmlal_s16(q10s32, d19s16, *d2s16);
d26s16 = vqrshrn_n_s32(q13s32, 14);
d27s16 = vqrshrn_n_s32(q14s32, 14);
d29s16 = vqrshrn_n_s32(q15s32, 14);
d28s16 = vqrshrn_n_s32(q10s32, 14);
q13s16 = vcombine_s16(d26s16, d27s16);
q14s16 = vcombine_s16(d28s16, d29s16);
*q8s16 = vaddq_s16(q13s16, q14s16);
*q9s16 = vsubq_s16(q13s16, q14s16);
*q9s16 = vcombine_s16(vget_high_s16(*q9s16), vget_low_s16(*q9s16)); // vswp
return;
}
static INLINE void IADST4x4_1D(int16x4_t *d3s16, int16x4_t *d4s16,
int16x4_t *d5s16, int16x8_t *q3s16,
int16x8_t *q8s16, int16x8_t *q9s16) {
int16x4_t d6s16, d16s16, d17s16, d18s16, d19s16;
int32x4_t q8s32, q9s32, q10s32, q11s32, q12s32, q13s32, q14s32, q15s32;
d6s16 = vget_low_s16(*q3s16);
d16s16 = vget_low_s16(*q8s16);
d17s16 = vget_high_s16(*q8s16);
d18s16 = vget_low_s16(*q9s16);
d19s16 = vget_high_s16(*q9s16);
q10s32 = vmull_s16(*d3s16, d16s16);
q11s32 = vmull_s16(*d4s16, d16s16);
q12s32 = vmull_s16(d6s16, d17s16);
q13s32 = vmull_s16(*d5s16, d18s16);
q14s32 = vmull_s16(*d3s16, d18s16);
q15s32 = vmovl_s16(d16s16);
q15s32 = vaddw_s16(q15s32, d19s16);
q8s32 = vmull_s16(*d4s16, d19s16);
q15s32 = vsubw_s16(q15s32, d18s16);
q9s32 = vmull_s16(*d5s16, d19s16);
q10s32 = vaddq_s32(q10s32, q13s32);
q10s32 = vaddq_s32(q10s32, q8s32);
q11s32 = vsubq_s32(q11s32, q14s32);
q8s32 = vdupq_n_s32((int32_t)sinpi_3_9);
q11s32 = vsubq_s32(q11s32, q9s32);
q15s32 = vmulq_s32(q15s32, q8s32);
q13s32 = vaddq_s32(q10s32, q12s32);
q10s32 = vaddq_s32(q10s32, q11s32);
q14s32 = vaddq_s32(q11s32, q12s32);
q10s32 = vsubq_s32(q10s32, q12s32);
d16s16 = vqrshrn_n_s32(q13s32, 14);
d17s16 = vqrshrn_n_s32(q14s32, 14);
d18s16 = vqrshrn_n_s32(q15s32, 14);
d19s16 = vqrshrn_n_s32(q10s32, 14);
*q8s16 = vcombine_s16(d16s16, d17s16);
*q9s16 = vcombine_s16(d18s16, d19s16);
return;
}
void av1_iht4x4_16_add_neon(const tran_low_t *input, uint8_t *dest,
int dest_stride, int tx_type) {
uint8x8_t d26u8, d27u8;
int16x4_t d0s16, d1s16, d2s16, d3s16, d4s16, d5s16;
uint32x2_t d26u32, d27u32;
int16x8_t q3s16, q8s16, q9s16;
uint16x8_t q8u16, q9u16;
d26u32 = d27u32 = vdup_n_u32(0);
q8s16 = vld1q_s16(input);
q9s16 = vld1q_s16(input + 8);
TRANSPOSE4X4(&q8s16, &q9s16);
switch (tx_type) {
case 0: // idct_idct is not supported. Fall back to C
av1_iht4x4_16_add_c(input, dest, dest_stride, tx_type);
return;
break;
case 1: // iadst_idct
// generate constants
GENERATE_COSINE_CONSTANTS(&d0s16, &d1s16, &d2s16);
GENERATE_SINE_CONSTANTS(&d3s16, &d4s16, &d5s16, &q3s16);
// first transform rows
IDCT4x4_1D(&d0s16, &d1s16, &d2s16, &q8s16, &q9s16);
// transpose the matrix
TRANSPOSE4X4(&q8s16, &q9s16);
// then transform columns
IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16);
break;
case 2: // idct_iadst
// generate constantsyy
GENERATE_COSINE_CONSTANTS(&d0s16, &d1s16, &d2s16);
GENERATE_SINE_CONSTANTS(&d3s16, &d4s16, &d5s16, &q3s16);
// first transform rows
IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16);
// transpose the matrix
TRANSPOSE4X4(&q8s16, &q9s16);
// then transform columns
IDCT4x4_1D(&d0s16, &d1s16, &d2s16, &q8s16, &q9s16);
break;
case 3: // iadst_iadst
// generate constants
GENERATE_SINE_CONSTANTS(&d3s16, &d4s16, &d5s16, &q3s16);
// first transform rows
IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16);
// transpose the matrix
TRANSPOSE4X4(&q8s16, &q9s16);
// then transform columns
IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16);
break;
default: // iadst_idct
assert(0);
break;
}
q8s16 = vrshrq_n_s16(q8s16, 4);
q9s16 = vrshrq_n_s16(q9s16, 4);
d26u32 = vld1_lane_u32((const uint32_t *)dest, d26u32, 0);
dest += dest_stride;
d26u32 = vld1_lane_u32((const uint32_t *)dest, d26u32, 1);
dest += dest_stride;
d27u32 = vld1_lane_u32((const uint32_t *)dest, d27u32, 0);
dest += dest_stride;
d27u32 = vld1_lane_u32((const uint32_t *)dest, d27u32, 1);
q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), vreinterpret_u8_u32(d26u32));
q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), vreinterpret_u8_u32(d27u32));
d26u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16));
d27u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16));
vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d27u8), 1);
dest -= dest_stride;
vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d27u8), 0);
dest -= dest_stride;
vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d26u8), 1);
dest -= dest_stride;
vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d26u8), 0);
return;
}

593
third_party/aom/av1/common/arm/neon/iht8x8_add_neon.c vendored Normal file
View file

@ -0,0 +1,593 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <arm_neon.h>
#include <assert.h>
#include "./aom_config.h"
#include "./av1_rtcd.h"
#include "aom_dsp/txfm_common.h"
#include "av1/common/common.h"
static INLINE void TRANSPOSE8X8(int16x8_t *q8s16, int16x8_t *q9s16,
int16x8_t *q10s16, int16x8_t *q11s16,
int16x8_t *q12s16, int16x8_t *q13s16,
int16x8_t *q14s16, int16x8_t *q15s16) {
int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16;
int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16;
int32x4x2_t q0x2s32, q1x2s32, q2x2s32, q3x2s32;
int16x8x2_t q0x2s16, q1x2s16, q2x2s16, q3x2s16;
d16s16 = vget_low_s16(*q8s16);
d17s16 = vget_high_s16(*q8s16);
d18s16 = vget_low_s16(*q9s16);
d19s16 = vget_high_s16(*q9s16);
d20s16 = vget_low_s16(*q10s16);
d21s16 = vget_high_s16(*q10s16);
d22s16 = vget_low_s16(*q11s16);
d23s16 = vget_high_s16(*q11s16);
d24s16 = vget_low_s16(*q12s16);
d25s16 = vget_high_s16(*q12s16);
d26s16 = vget_low_s16(*q13s16);
d27s16 = vget_high_s16(*q13s16);
d28s16 = vget_low_s16(*q14s16);
d29s16 = vget_high_s16(*q14s16);
d30s16 = vget_low_s16(*q15s16);
d31s16 = vget_high_s16(*q15s16);
*q8s16 = vcombine_s16(d16s16, d24s16); // vswp d17, d24
*q9s16 = vcombine_s16(d18s16, d26s16); // vswp d19, d26
*q10s16 = vcombine_s16(d20s16, d28s16); // vswp d21, d28
*q11s16 = vcombine_s16(d22s16, d30s16); // vswp d23, d30
*q12s16 = vcombine_s16(d17s16, d25s16);
*q13s16 = vcombine_s16(d19s16, d27s16);
*q14s16 = vcombine_s16(d21s16, d29s16);
*q15s16 = vcombine_s16(d23s16, d31s16);
q0x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(*q8s16), vreinterpretq_s32_s16(*q10s16));
q1x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(*q9s16), vreinterpretq_s32_s16(*q11s16));
q2x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(*q12s16), vreinterpretq_s32_s16(*q14s16));
q3x2s32 =
vtrnq_s32(vreinterpretq_s32_s16(*q13s16), vreinterpretq_s32_s16(*q15s16));
q0x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[0]), // q8
vreinterpretq_s16_s32(q1x2s32.val[0])); // q9
q1x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[1]), // q10
vreinterpretq_s16_s32(q1x2s32.val[1])); // q11
q2x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[0]), // q12
vreinterpretq_s16_s32(q3x2s32.val[0])); // q13
q3x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[1]), // q14
vreinterpretq_s16_s32(q3x2s32.val[1])); // q15
*q8s16 = q0x2s16.val[0];
*q9s16 = q0x2s16.val[1];
*q10s16 = q1x2s16.val[0];
*q11s16 = q1x2s16.val[1];
*q12s16 = q2x2s16.val[0];
*q13s16 = q2x2s16.val[1];
*q14s16 = q3x2s16.val[0];
*q15s16 = q3x2s16.val[1];
return;
}
static INLINE void IDCT8x8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
int16x8_t *q10s16, int16x8_t *q11s16,
int16x8_t *q12s16, int16x8_t *q13s16,
int16x8_t *q14s16, int16x8_t *q15s16) {
int16x4_t d0s16, d1s16, d2s16, d3s16;
int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16;
int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16;
int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16;
int16x8_t q0s16, q1s16, q2s16, q3s16, q4s16, q5s16, q6s16, q7s16;
int32x4_t q2s32, q3s32, q5s32, q6s32, q8s32, q9s32;
int32x4_t q10s32, q11s32, q12s32, q13s32, q15s32;
d0s16 = vdup_n_s16((int16_t)cospi_28_64);
d1s16 = vdup_n_s16((int16_t)cospi_4_64);
d2s16 = vdup_n_s16((int16_t)cospi_12_64);
d3s16 = vdup_n_s16((int16_t)cospi_20_64);
d16s16 = vget_low_s16(*q8s16);
d17s16 = vget_high_s16(*q8s16);
d18s16 = vget_low_s16(*q9s16);
d19s16 = vget_high_s16(*q9s16);
d20s16 = vget_low_s16(*q10s16);
d21s16 = vget_high_s16(*q10s16);
d22s16 = vget_low_s16(*q11s16);
d23s16 = vget_high_s16(*q11s16);
d24s16 = vget_low_s16(*q12s16);
d25s16 = vget_high_s16(*q12s16);
d26s16 = vget_low_s16(*q13s16);
d27s16 = vget_high_s16(*q13s16);
d28s16 = vget_low_s16(*q14s16);
d29s16 = vget_high_s16(*q14s16);
d30s16 = vget_low_s16(*q15s16);
d31s16 = vget_high_s16(*q15s16);
q2s32 = vmull_s16(d18s16, d0s16);
q3s32 = vmull_s16(d19s16, d0s16);
q5s32 = vmull_s16(d26s16, d2s16);
q6s32 = vmull_s16(d27s16, d2s16);
q2s32 = vmlsl_s16(q2s32, d30s16, d1s16);
q3s32 = vmlsl_s16(q3s32, d31s16, d1s16);
q5s32 = vmlsl_s16(q5s32, d22s16, d3s16);
q6s32 = vmlsl_s16(q6s32, d23s16, d3s16);
d8s16 = vqrshrn_n_s32(q2s32, 14);
d9s16 = vqrshrn_n_s32(q3s32, 14);
d10s16 = vqrshrn_n_s32(q5s32, 14);
d11s16 = vqrshrn_n_s32(q6s32, 14);
q4s16 = vcombine_s16(d8s16, d9s16);
q5s16 = vcombine_s16(d10s16, d11s16);
q2s32 = vmull_s16(d18s16, d1s16);
q3s32 = vmull_s16(d19s16, d1s16);
q9s32 = vmull_s16(d26s16, d3s16);
q13s32 = vmull_s16(d27s16, d3s16);
q2s32 = vmlal_s16(q2s32, d30s16, d0s16);
q3s32 = vmlal_s16(q3s32, d31s16, d0s16);
q9s32 = vmlal_s16(q9s32, d22s16, d2s16);
q13s32 = vmlal_s16(q13s32, d23s16, d2s16);
d14s16 = vqrshrn_n_s32(q2s32, 14);
d15s16 = vqrshrn_n_s32(q3s32, 14);
d12s16 = vqrshrn_n_s32(q9s32, 14);
d13s16 = vqrshrn_n_s32(q13s32, 14);
q6s16 = vcombine_s16(d12s16, d13s16);
q7s16 = vcombine_s16(d14s16, d15s16);
d0s16 = vdup_n_s16((int16_t)cospi_16_64);
q2s32 = vmull_s16(d16s16, d0s16);
q3s32 = vmull_s16(d17s16, d0s16);
q13s32 = vmull_s16(d16s16, d0s16);
q15s32 = vmull_s16(d17s16, d0s16);
q2s32 = vmlal_s16(q2s32, d24s16, d0s16);
q3s32 = vmlal_s16(q3s32, d25s16, d0s16);
q13s32 = vmlsl_s16(q13s32, d24s16, d0s16);
q15s32 = vmlsl_s16(q15s32, d25s16, d0s16);
d0s16 = vdup_n_s16((int16_t)cospi_24_64);
d1s16 = vdup_n_s16((int16_t)cospi_8_64);
d18s16 = vqrshrn_n_s32(q2s32, 14);
d19s16 = vqrshrn_n_s32(q3s32, 14);
d22s16 = vqrshrn_n_s32(q13s32, 14);
d23s16 = vqrshrn_n_s32(q15s32, 14);
*q9s16 = vcombine_s16(d18s16, d19s16);
*q11s16 = vcombine_s16(d22s16, d23s16);
q2s32 = vmull_s16(d20s16, d0s16);
q3s32 = vmull_s16(d21s16, d0s16);
q8s32 = vmull_s16(d20s16, d1s16);
q12s32 = vmull_s16(d21s16, d1s16);
q2s32 = vmlsl_s16(q2s32, d28s16, d1s16);
q3s32 = vmlsl_s16(q3s32, d29s16, d1s16);
q8s32 = vmlal_s16(q8s32, d28s16, d0s16);
q12s32 = vmlal_s16(q12s32, d29s16, d0s16);
d26s16 = vqrshrn_n_s32(q2s32, 14);
d27s16 = vqrshrn_n_s32(q3s32, 14);
d30s16 = vqrshrn_n_s32(q8s32, 14);
d31s16 = vqrshrn_n_s32(q12s32, 14);
*q13s16 = vcombine_s16(d26s16, d27s16);
*q15s16 = vcombine_s16(d30s16, d31s16);
q0s16 = vaddq_s16(*q9s16, *q15s16);
q1s16 = vaddq_s16(*q11s16, *q13s16);
q2s16 = vsubq_s16(*q11s16, *q13s16);
q3s16 = vsubq_s16(*q9s16, *q15s16);
*q13s16 = vsubq_s16(q4s16, q5s16);
q4s16 = vaddq_s16(q4s16, q5s16);
*q14s16 = vsubq_s16(q7s16, q6s16);
q7s16 = vaddq_s16(q7s16, q6s16);
d26s16 = vget_low_s16(*q13s16);
d27s16 = vget_high_s16(*q13s16);
d28s16 = vget_low_s16(*q14s16);
d29s16 = vget_high_s16(*q14s16);
d16s16 = vdup_n_s16((int16_t)cospi_16_64);
q9s32 = vmull_s16(d28s16, d16s16);
q10s32 = vmull_s16(d29s16, d16s16);
q11s32 = vmull_s16(d28s16, d16s16);
q12s32 = vmull_s16(d29s16, d16s16);
q9s32 = vmlsl_s16(q9s32, d26s16, d16s16);
q10s32 = vmlsl_s16(q10s32, d27s16, d16s16);
q11s32 = vmlal_s16(q11s32, d26s16, d16s16);
q12s32 = vmlal_s16(q12s32, d27s16, d16s16);
d10s16 = vqrshrn_n_s32(q9s32, 14);
d11s16 = vqrshrn_n_s32(q10s32, 14);
d12s16 = vqrshrn_n_s32(q11s32, 14);
d13s16 = vqrshrn_n_s32(q12s32, 14);
q5s16 = vcombine_s16(d10s16, d11s16);
q6s16 = vcombine_s16(d12s16, d13s16);
*q8s16 = vaddq_s16(q0s16, q7s16);
*q9s16 = vaddq_s16(q1s16, q6s16);
*q10s16 = vaddq_s16(q2s16, q5s16);
*q11s16 = vaddq_s16(q3s16, q4s16);
*q12s16 = vsubq_s16(q3s16, q4s16);
*q13s16 = vsubq_s16(q2s16, q5s16);
*q14s16 = vsubq_s16(q1s16, q6s16);
*q15s16 = vsubq_s16(q0s16, q7s16);
return;
}
static INLINE void IADST8X8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
int16x8_t *q10s16, int16x8_t *q11s16,
int16x8_t *q12s16, int16x8_t *q13s16,
int16x8_t *q14s16, int16x8_t *q15s16) {
int16x4_t d0s16, d1s16, d2s16, d3s16, d4s16, d5s16, d6s16, d7s16;
int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16;
int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16;
int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16;
int16x8_t q2s16, q4s16, q5s16, q6s16;
int32x4_t q0s32, q1s32, q2s32, q3s32, q4s32, q5s32, q6s32, q7s32, q8s32;
int32x4_t q9s32, q10s32, q11s32, q12s32, q13s32, q14s32, q15s32;
d16s16 = vget_low_s16(*q8s16);
d17s16 = vget_high_s16(*q8s16);
d18s16 = vget_low_s16(*q9s16);
d19s16 = vget_high_s16(*q9s16);
d20s16 = vget_low_s16(*q10s16);
d21s16 = vget_high_s16(*q10s16);
d22s16 = vget_low_s16(*q11s16);
d23s16 = vget_high_s16(*q11s16);
d24s16 = vget_low_s16(*q12s16);
d25s16 = vget_high_s16(*q12s16);
d26s16 = vget_low_s16(*q13s16);
d27s16 = vget_high_s16(*q13s16);
d28s16 = vget_low_s16(*q14s16);
d29s16 = vget_high_s16(*q14s16);
d30s16 = vget_low_s16(*q15s16);
d31s16 = vget_high_s16(*q15s16);
d14s16 = vdup_n_s16((int16_t)cospi_2_64);
d15s16 = vdup_n_s16((int16_t)cospi_30_64);
q1s32 = vmull_s16(d30s16, d14s16);
q2s32 = vmull_s16(d31s16, d14s16);
q3s32 = vmull_s16(d30s16, d15s16);
q4s32 = vmull_s16(d31s16, d15s16);
d30s16 = vdup_n_s16((int16_t)cospi_18_64);
d31s16 = vdup_n_s16((int16_t)cospi_14_64);
q1s32 = vmlal_s16(q1s32, d16s16, d15s16);
q2s32 = vmlal_s16(q2s32, d17s16, d15s16);
q3s32 = vmlsl_s16(q3s32, d16s16, d14s16);
q4s32 = vmlsl_s16(q4s32, d17s16, d14s16);
q5s32 = vmull_s16(d22s16, d30s16);
q6s32 = vmull_s16(d23s16, d30s16);
q7s32 = vmull_s16(d22s16, d31s16);
q8s32 = vmull_s16(d23s16, d31s16);
q5s32 = vmlal_s16(q5s32, d24s16, d31s16);
q6s32 = vmlal_s16(q6s32, d25s16, d31s16);
q7s32 = vmlsl_s16(q7s32, d24s16, d30s16);
q8s32 = vmlsl_s16(q8s32, d25s16, d30s16);
q11s32 = vaddq_s32(q1s32, q5s32);
q12s32 = vaddq_s32(q2s32, q6s32);
q1s32 = vsubq_s32(q1s32, q5s32);
q2s32 = vsubq_s32(q2s32, q6s32);
d22s16 = vqrshrn_n_s32(q11s32, 14);
d23s16 = vqrshrn_n_s32(q12s32, 14);
*q11s16 = vcombine_s16(d22s16, d23s16);
q12s32 = vaddq_s32(q3s32, q7s32);
q15s32 = vaddq_s32(q4s32, q8s32);
q3s32 = vsubq_s32(q3s32, q7s32);
q4s32 = vsubq_s32(q4s32, q8s32);
d2s16 = vqrshrn_n_s32(q1s32, 14);
d3s16 = vqrshrn_n_s32(q2s32, 14);
d24s16 = vqrshrn_n_s32(q12s32, 14);
d25s16 = vqrshrn_n_s32(q15s32, 14);
d6s16 = vqrshrn_n_s32(q3s32, 14);
d7s16 = vqrshrn_n_s32(q4s32, 14);
*q12s16 = vcombine_s16(d24s16, d25s16);
d0s16 = vdup_n_s16((int16_t)cospi_10_64);
d1s16 = vdup_n_s16((int16_t)cospi_22_64);
q4s32 = vmull_s16(d26s16, d0s16);
q5s32 = vmull_s16(d27s16, d0s16);
q2s32 = vmull_s16(d26s16, d1s16);
q6s32 = vmull_s16(d27s16, d1s16);
d30s16 = vdup_n_s16((int16_t)cospi_26_64);
d31s16 = vdup_n_s16((int16_t)cospi_6_64);
q4s32 = vmlal_s16(q4s32, d20s16, d1s16);
q5s32 = vmlal_s16(q5s32, d21s16, d1s16);
q2s32 = vmlsl_s16(q2s32, d20s16, d0s16);
q6s32 = vmlsl_s16(q6s32, d21s16, d0s16);
q0s32 = vmull_s16(d18s16, d30s16);
q13s32 = vmull_s16(d19s16, d30s16);
q0s32 = vmlal_s16(q0s32, d28s16, d31s16);
q13s32 = vmlal_s16(q13s32, d29s16, d31s16);
q10s32 = vmull_s16(d18s16, d31s16);
q9s32 = vmull_s16(d19s16, d31s16);
q10s32 = vmlsl_s16(q10s32, d28s16, d30s16);
q9s32 = vmlsl_s16(q9s32, d29s16, d30s16);
q14s32 = vaddq_s32(q2s32, q10s32);
q15s32 = vaddq_s32(q6s32, q9s32);
q2s32 = vsubq_s32(q2s32, q10s32);
q6s32 = vsubq_s32(q6s32, q9s32);
d28s16 = vqrshrn_n_s32(q14s32, 14);
d29s16 = vqrshrn_n_s32(q15s32, 14);
d4s16 = vqrshrn_n_s32(q2s32, 14);
d5s16 = vqrshrn_n_s32(q6s32, 14);
*q14s16 = vcombine_s16(d28s16, d29s16);
q9s32 = vaddq_s32(q4s32, q0s32);
q10s32 = vaddq_s32(q5s32, q13s32);
q4s32 = vsubq_s32(q4s32, q0s32);
q5s32 = vsubq_s32(q5s32, q13s32);
d30s16 = vdup_n_s16((int16_t)cospi_8_64);
d31s16 = vdup_n_s16((int16_t)cospi_24_64);
d18s16 = vqrshrn_n_s32(q9s32, 14);
d19s16 = vqrshrn_n_s32(q10s32, 14);
d8s16 = vqrshrn_n_s32(q4s32, 14);
d9s16 = vqrshrn_n_s32(q5s32, 14);
*q9s16 = vcombine_s16(d18s16, d19s16);
q5s32 = vmull_s16(d2s16, d30s16);
q6s32 = vmull_s16(d3s16, d30s16);
q7s32 = vmull_s16(d2s16, d31s16);
q0s32 = vmull_s16(d3s16, d31s16);
q5s32 = vmlal_s16(q5s32, d6s16, d31s16);
q6s32 = vmlal_s16(q6s32, d7s16, d31s16);
q7s32 = vmlsl_s16(q7s32, d6s16, d30s16);
q0s32 = vmlsl_s16(q0s32, d7s16, d30s16);
q1s32 = vmull_s16(d4s16, d30s16);
q3s32 = vmull_s16(d5s16, d30s16);
q10s32 = vmull_s16(d4s16, d31s16);
q2s32 = vmull_s16(d5s16, d31s16);
q1s32 = vmlsl_s16(q1s32, d8s16, d31s16);
q3s32 = vmlsl_s16(q3s32, d9s16, d31s16);
q10s32 = vmlal_s16(q10s32, d8s16, d30s16);
q2s32 = vmlal_s16(q2s32, d9s16, d30s16);
*q8s16 = vaddq_s16(*q11s16, *q9s16);
*q11s16 = vsubq_s16(*q11s16, *q9s16);
q4s16 = vaddq_s16(*q12s16, *q14s16);
*q12s16 = vsubq_s16(*q12s16, *q14s16);
q14s32 = vaddq_s32(q5s32, q1s32);
q15s32 = vaddq_s32(q6s32, q3s32);
q5s32 = vsubq_s32(q5s32, q1s32);
q6s32 = vsubq_s32(q6s32, q3s32);
d18s16 = vqrshrn_n_s32(q14s32, 14);
d19s16 = vqrshrn_n_s32(q15s32, 14);
d10s16 = vqrshrn_n_s32(q5s32, 14);
d11s16 = vqrshrn_n_s32(q6s32, 14);
*q9s16 = vcombine_s16(d18s16, d19s16);
q1s32 = vaddq_s32(q7s32, q10s32);
q3s32 = vaddq_s32(q0s32, q2s32);
q7s32 = vsubq_s32(q7s32, q10s32);
q0s32 = vsubq_s32(q0s32, q2s32);
d28s16 = vqrshrn_n_s32(q1s32, 14);
d29s16 = vqrshrn_n_s32(q3s32, 14);
d14s16 = vqrshrn_n_s32(q7s32, 14);
d15s16 = vqrshrn_n_s32(q0s32, 14);
*q14s16 = vcombine_s16(d28s16, d29s16);
d30s16 = vdup_n_s16((int16_t)cospi_16_64);
d22s16 = vget_low_s16(*q11s16);
d23s16 = vget_high_s16(*q11s16);
q2s32 = vmull_s16(d22s16, d30s16);
q3s32 = vmull_s16(d23s16, d30s16);
q13s32 = vmull_s16(d22s16, d30s16);
q1s32 = vmull_s16(d23s16, d30s16);
d24s16 = vget_low_s16(*q12s16);
d25s16 = vget_high_s16(*q12s16);
q2s32 = vmlal_s16(q2s32, d24s16, d30s16);
q3s32 = vmlal_s16(q3s32, d25s16, d30s16);
q13s32 = vmlsl_s16(q13s32, d24s16, d30s16);
q1s32 = vmlsl_s16(q1s32, d25s16, d30s16);
d4s16 = vqrshrn_n_s32(q2s32, 14);
d5s16 = vqrshrn_n_s32(q3s32, 14);
d24s16 = vqrshrn_n_s32(q13s32, 14);
d25s16 = vqrshrn_n_s32(q1s32, 14);
q2s16 = vcombine_s16(d4s16, d5s16);
*q12s16 = vcombine_s16(d24s16, d25s16);
q13s32 = vmull_s16(d10s16, d30s16);
q1s32 = vmull_s16(d11s16, d30s16);
q11s32 = vmull_s16(d10s16, d30s16);
q0s32 = vmull_s16(d11s16, d30s16);
q13s32 = vmlal_s16(q13s32, d14s16, d30s16);
q1s32 = vmlal_s16(q1s32, d15s16, d30s16);
q11s32 = vmlsl_s16(q11s32, d14s16, d30s16);
q0s32 = vmlsl_s16(q0s32, d15s16, d30s16);
d20s16 = vqrshrn_n_s32(q13s32, 14);
d21s16 = vqrshrn_n_s32(q1s32, 14);
d12s16 = vqrshrn_n_s32(q11s32, 14);
d13s16 = vqrshrn_n_s32(q0s32, 14);
*q10s16 = vcombine_s16(d20s16, d21s16);
q6s16 = vcombine_s16(d12s16, d13s16);
q5s16 = vdupq_n_s16(0);
*q9s16 = vsubq_s16(q5s16, *q9s16);
*q11s16 = vsubq_s16(q5s16, q2s16);
*q13s16 = vsubq_s16(q5s16, q6s16);
*q15s16 = vsubq_s16(q5s16, q4s16);
return;
}
void av1_iht8x8_64_add_neon(const tran_low_t *input, uint8_t *dest,
int dest_stride, int tx_type) {
int i;
uint8_t *d1, *d2;
uint8x8_t d0u8, d1u8, d2u8, d3u8;
uint64x1_t d0u64, d1u64, d2u64, d3u64;
int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16, q13s16, q14s16, q15s16;
uint16x8_t q8u16, q9u16, q10u16, q11u16;
q8s16 = vld1q_s16(input);
q9s16 = vld1q_s16(input + 8);
q10s16 = vld1q_s16(input + 8 * 2);
q11s16 = vld1q_s16(input + 8 * 3);
q12s16 = vld1q_s16(input + 8 * 4);
q13s16 = vld1q_s16(input + 8 * 5);
q14s16 = vld1q_s16(input + 8 * 6);
q15s16 = vld1q_s16(input + 8 * 7);
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
switch (tx_type) {
case 0: // idct_idct is not supported. Fall back to C
av1_iht8x8_64_add_c(input, dest, dest_stride, tx_type);
return;
break;
case 1: // iadst_idct
// generate IDCT constants
// GENERATE_IDCT_CONSTANTS
// first transform rows
IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
// transpose the matrix
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
// generate IADST constants
// GENERATE_IADST_CONSTANTS
// then transform columns
IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
break;
case 2: // idct_iadst
// generate IADST constants
// GENERATE_IADST_CONSTANTS
// first transform rows
IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
// transpose the matrix
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
// generate IDCT constants
// GENERATE_IDCT_CONSTANTS
// then transform columns
IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
break;
case 3: // iadst_iadst
// generate IADST constants
// GENERATE_IADST_CONSTANTS
// first transform rows
IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
// transpose the matrix
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
// then transform columns
IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16, &q12s16, &q13s16, &q14s16,
&q15s16);
break;
default: // iadst_idct
assert(0);
break;
}
q8s16 = vrshrq_n_s16(q8s16, 5);
q9s16 = vrshrq_n_s16(q9s16, 5);
q10s16 = vrshrq_n_s16(q10s16, 5);
q11s16 = vrshrq_n_s16(q11s16, 5);
q12s16 = vrshrq_n_s16(q12s16, 5);
q13s16 = vrshrq_n_s16(q13s16, 5);
q14s16 = vrshrq_n_s16(q14s16, 5);
q15s16 = vrshrq_n_s16(q15s16, 5);
for (d1 = d2 = dest, i = 0; i < 2; i++) {
if (i != 0) {
q8s16 = q12s16;
q9s16 = q13s16;
q10s16 = q14s16;
q11s16 = q15s16;
}
d0u64 = vld1_u64((uint64_t *)d1);
d1 += dest_stride;
d1u64 = vld1_u64((uint64_t *)d1);
d1 += dest_stride;
d2u64 = vld1_u64((uint64_t *)d1);
d1 += dest_stride;
d3u64 = vld1_u64((uint64_t *)d1);
d1 += dest_stride;
q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), vreinterpret_u8_u64(d0u64));
q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), vreinterpret_u8_u64(d1u64));
q10u16 =
vaddw_u8(vreinterpretq_u16_s16(q10s16), vreinterpret_u8_u64(d2u64));
q11u16 =
vaddw_u8(vreinterpretq_u16_s16(q11s16), vreinterpret_u8_u64(d3u64));
d0u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16));
d1u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16));
d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q10u16));
d3u8 = vqmovun_s16(vreinterpretq_s16_u16(q11u16));
vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d0u8));
d2 += dest_stride;
vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d1u8));
d2 += dest_stride;
vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d2u8));
d2 += dest_stride;
vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d3u8));
d2 += dest_stride;
}
return;
}

2312
third_party/aom/av1/common/av1_fwd_txfm1d.c vendored Normal file
View file

File diff suppressed because it is too large Load diff

45
third_party/aom/av1/common/av1_fwd_txfm1d.h vendored Normal file
View file

@ -0,0 +1,45 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_FWD_TXFM1D_H_
#define AV1_FWD_TXFM1D_H_
#include "av1/common/av1_txfm.h"
#ifdef __cplusplus
extern "C" {
#endif
void av1_fdct4_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
const int8_t *stage_range);
void av1_fdct8_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
const int8_t *stage_range);
void av1_fdct16_new(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range);
void av1_fdct32_new(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range);
void av1_fdct64_new(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range);
void av1_fadst4_new(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range);
void av1_fadst8_new(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range);
void av1_fadst16_new(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range);
void av1_fadst32_new(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range);
#ifdef __cplusplus
}
#endif
#endif // AV1_FWD_TXFM1D_H_

230
third_party/aom/av1/common/av1_fwd_txfm2d.c vendored Normal file
View file

@ -0,0 +1,230 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <assert.h>
#include "./av1_rtcd.h"
#include "av1/common/enums.h"
#include "av1/common/av1_fwd_txfm1d.h"
#include "av1/common/av1_fwd_txfm2d_cfg.h"
#include "av1/common/av1_txfm.h"
static INLINE TxfmFunc fwd_txfm_type_to_func(TXFM_TYPE txfm_type) {
switch (txfm_type) {
case TXFM_TYPE_DCT4: return av1_fdct4_new;
case TXFM_TYPE_DCT8: return av1_fdct8_new;
case TXFM_TYPE_DCT16: return av1_fdct16_new;
case TXFM_TYPE_DCT32: return av1_fdct32_new;
case TXFM_TYPE_ADST4: return av1_fadst4_new;
case TXFM_TYPE_ADST8: return av1_fadst8_new;
case TXFM_TYPE_ADST16: return av1_fadst16_new;
case TXFM_TYPE_ADST32: return av1_fadst32_new;
default: assert(0); return NULL;
}
}
static INLINE void fwd_txfm2d_c(const int16_t *input, int32_t *output,
const int stride, const TXFM_2D_FLIP_CFG *cfg,
int32_t *buf) {
int c, r;
const int txfm_size = cfg->cfg->txfm_size;
const int8_t *shift = cfg->cfg->shift;
const int8_t *stage_range_col = cfg->cfg->stage_range_col;
const int8_t *stage_range_row = cfg->cfg->stage_range_row;
const int8_t *cos_bit_col = cfg->cfg->cos_bit_col;
const int8_t *cos_bit_row = cfg->cfg->cos_bit_row;
const TxfmFunc txfm_func_col = fwd_txfm_type_to_func(cfg->cfg->txfm_type_col);
const TxfmFunc txfm_func_row = fwd_txfm_type_to_func(cfg->cfg->txfm_type_row);
// use output buffer as temp buffer
int32_t *temp_in = output;
int32_t *temp_out = output + txfm_size;
// Columns
for (c = 0; c < txfm_size; ++c) {
if (cfg->ud_flip == 0) {
for (r = 0; r < txfm_size; ++r) temp_in[r] = input[r * stride + c];
} else {
for (r = 0; r < txfm_size; ++r)
// flip upside down
temp_in[r] = input[(txfm_size - r - 1) * stride + c];
}
round_shift_array(temp_in, txfm_size, -shift[0]);
txfm_func_col(temp_in, temp_out, cos_bit_col, stage_range_col);
round_shift_array(temp_out, txfm_size, -shift[1]);
if (cfg->lr_flip == 0) {
for (r = 0; r < txfm_size; ++r) buf[r * txfm_size + c] = temp_out[r];
} else {
for (r = 0; r < txfm_size; ++r)
// flip from left to right
buf[r * txfm_size + (txfm_size - c - 1)] = temp_out[r];
}
}
// Rows
for (r = 0; r < txfm_size; ++r) {
txfm_func_row(buf + r * txfm_size, output + r * txfm_size, cos_bit_row,
stage_range_row);
round_shift_array(output + r * txfm_size, txfm_size, -shift[2]);
}
}
void av1_fwd_txfm2d_4x4_c(const int16_t *input, int32_t *output, int stride,
int tx_type, int bd) {
int32_t txfm_buf[4 * 4];
TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(tx_type, TX_4X4);
(void)bd;
fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf);
}
void av1_fwd_txfm2d_8x8_c(const int16_t *input, int32_t *output, int stride,
int tx_type, int bd) {
int32_t txfm_buf[8 * 8];
TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(tx_type, TX_8X8);
(void)bd;
fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf);
}
void av1_fwd_txfm2d_16x16_c(const int16_t *input, int32_t *output, int stride,
int tx_type, int bd) {
int32_t txfm_buf[16 * 16];
TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(tx_type, TX_16X16);
(void)bd;
fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf);
}
void av1_fwd_txfm2d_32x32_c(const int16_t *input, int32_t *output, int stride,
int tx_type, int bd) {
int32_t txfm_buf[32 * 32];
TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_cfg(tx_type, TX_32X32);
(void)bd;
fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf);
}
void av1_fwd_txfm2d_64x64_c(const int16_t *input, int32_t *output, int stride,
int tx_type, int bd) {
int32_t txfm_buf[64 * 64];
TXFM_2D_FLIP_CFG cfg = av1_get_fwd_txfm_64x64_cfg(tx_type);
(void)bd;
fwd_txfm2d_c(input, output, stride, &cfg, txfm_buf);
}
#if CONFIG_EXT_TX
static const TXFM_2D_CFG *fwd_txfm_cfg_ls[FLIPADST_ADST + 1][TX_SIZES] = {
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_dct_dct_4, &fwd_txfm_2d_cfg_dct_dct_8,
&fwd_txfm_2d_cfg_dct_dct_16, &fwd_txfm_2d_cfg_dct_dct_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_adst_dct_4, &fwd_txfm_2d_cfg_adst_dct_8,
&fwd_txfm_2d_cfg_adst_dct_16, &fwd_txfm_2d_cfg_adst_dct_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_dct_adst_4, &fwd_txfm_2d_cfg_dct_adst_8,
&fwd_txfm_2d_cfg_dct_adst_16, &fwd_txfm_2d_cfg_dct_adst_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_adst_adst_4, &fwd_txfm_2d_cfg_adst_adst_8,
&fwd_txfm_2d_cfg_adst_adst_16, &fwd_txfm_2d_cfg_adst_adst_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_adst_dct_4, &fwd_txfm_2d_cfg_adst_dct_8,
&fwd_txfm_2d_cfg_adst_dct_16, &fwd_txfm_2d_cfg_adst_dct_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_dct_adst_4, &fwd_txfm_2d_cfg_dct_adst_8,
&fwd_txfm_2d_cfg_dct_adst_16, &fwd_txfm_2d_cfg_dct_adst_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_adst_adst_4, &fwd_txfm_2d_cfg_adst_adst_8,
&fwd_txfm_2d_cfg_adst_adst_16, &fwd_txfm_2d_cfg_adst_adst_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_adst_adst_4, &fwd_txfm_2d_cfg_adst_adst_8,
&fwd_txfm_2d_cfg_adst_adst_16, &fwd_txfm_2d_cfg_adst_adst_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_adst_adst_4, &fwd_txfm_2d_cfg_adst_adst_8,
&fwd_txfm_2d_cfg_adst_adst_16, &fwd_txfm_2d_cfg_adst_adst_32 },
};
#else // CONFIG_EXT_TX
static const TXFM_2D_CFG *fwd_txfm_cfg_ls[TX_TYPES][TX_SIZES] = {
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_dct_dct_4, &fwd_txfm_2d_cfg_dct_dct_8,
&fwd_txfm_2d_cfg_dct_dct_16, &fwd_txfm_2d_cfg_dct_dct_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_adst_dct_4, &fwd_txfm_2d_cfg_adst_dct_8,
&fwd_txfm_2d_cfg_adst_dct_16, &fwd_txfm_2d_cfg_adst_dct_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_dct_adst_4, &fwd_txfm_2d_cfg_dct_adst_8,
&fwd_txfm_2d_cfg_dct_adst_16, &fwd_txfm_2d_cfg_dct_adst_32 },
{
#if CONFIG_CB4X4
NULL,
#endif
&fwd_txfm_2d_cfg_adst_adst_4, &fwd_txfm_2d_cfg_adst_adst_8,
&fwd_txfm_2d_cfg_adst_adst_16, &fwd_txfm_2d_cfg_adst_adst_32 },
};
#endif // CONFIG_EXT_TX
TXFM_2D_FLIP_CFG av1_get_fwd_txfm_cfg(int tx_type, int tx_size) {
TXFM_2D_FLIP_CFG cfg;
set_flip_cfg(tx_type, &cfg);
cfg.cfg = fwd_txfm_cfg_ls[tx_type][tx_size];
return cfg;
}
TXFM_2D_FLIP_CFG av1_get_fwd_txfm_64x64_cfg(int tx_type) {
TXFM_2D_FLIP_CFG cfg;
switch (tx_type) {
case DCT_DCT:
cfg.cfg = &fwd_txfm_2d_cfg_dct_dct_64;
cfg.ud_flip = 0;
cfg.lr_flip = 0;
break;
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
default:
cfg.ud_flip = 0;
cfg.lr_flip = 0;
assert(0);
}
return cfg;
}

444
third_party/aom/av1/common/av1_fwd_txfm2d_cfg.h vendored Normal file
View file

@ -0,0 +1,444 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_FWD_TXFM2D_CFG_H_
#define AV1_FWD_TXFM2D_CFG_H_
#include "av1/common/enums.h"
#include "av1/common/av1_fwd_txfm1d.h"
// ---------------- config fwd_dct_dct_4 ----------------
static const int8_t fwd_shift_dct_dct_4[3] = { 2, 0, 0 };
static const int8_t fwd_stage_range_col_dct_dct_4[4] = { 15, 16, 17, 17 };
static const int8_t fwd_stage_range_row_dct_dct_4[4] = { 17, 18, 18, 18 };
static const int8_t fwd_cos_bit_col_dct_dct_4[4] = { 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_dct_dct_4[4] = { 13, 13, 13, 13 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_dct_4 = {
4, // .txfm_size
4, // .stage_num_col
4, // .stage_num_row
// 0, // .log_scale
fwd_shift_dct_dct_4, // .shift
fwd_stage_range_col_dct_dct_4, // .stage_range_col
fwd_stage_range_row_dct_dct_4, // .stage_range_row
fwd_cos_bit_col_dct_dct_4, // .cos_bit_col
fwd_cos_bit_row_dct_dct_4, // .cos_bit_row
TXFM_TYPE_DCT4, // .txfm_type_col
TXFM_TYPE_DCT4
}; // .txfm_type_row
// ---------------- config fwd_dct_dct_8 ----------------
static const int8_t fwd_shift_dct_dct_8[3] = { 2, -1, 0 };
static const int8_t fwd_stage_range_col_dct_dct_8[6] = {
15, 16, 17, 18, 18, 18
};
static const int8_t fwd_stage_range_row_dct_dct_8[6] = {
17, 18, 19, 19, 19, 19
};
static const int8_t fwd_cos_bit_col_dct_dct_8[6] = { 13, 13, 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_dct_dct_8[6] = { 13, 13, 13, 13, 13, 13 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_dct_8 = {
8, // .txfm_size
6, // .stage_num_col
6, // .stage_num_row
// 0, // .log_scale
fwd_shift_dct_dct_8, // .shift
fwd_stage_range_col_dct_dct_8, // .stage_range_col
fwd_stage_range_row_dct_dct_8, // .stage_range_row
fwd_cos_bit_col_dct_dct_8, // .cos_bit_col
fwd_cos_bit_row_dct_dct_8, // .cos_bit_row
TXFM_TYPE_DCT8, // .txfm_type_col
TXFM_TYPE_DCT8
}; // .txfm_type_row
// ---------------- config fwd_dct_dct_16 ----------------
static const int8_t fwd_shift_dct_dct_16[3] = { 2, -2, 0 };
static const int8_t fwd_stage_range_col_dct_dct_16[8] = { 15, 16, 17, 18,
19, 19, 19, 19 };
static const int8_t fwd_stage_range_row_dct_dct_16[8] = { 17, 18, 19, 20,
20, 20, 20, 20 };
static const int8_t fwd_cos_bit_col_dct_dct_16[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_dct_dct_16[8] = { 12, 12, 12, 12,
12, 12, 12, 12 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_dct_16 = {
16, // .txfm_size
8, // .stage_num_col
8, // .stage_num_row
// 0, // .log_scale
fwd_shift_dct_dct_16, // .shift
fwd_stage_range_col_dct_dct_16, // .stage_range_col
fwd_stage_range_row_dct_dct_16, // .stage_range_row
fwd_cos_bit_col_dct_dct_16, // .cos_bit_col
fwd_cos_bit_row_dct_dct_16, // .cos_bit_row
TXFM_TYPE_DCT16, // .txfm_type_col
TXFM_TYPE_DCT16
}; // .txfm_type_row
// ---------------- config fwd_dct_dct_32 ----------------
static const int8_t fwd_shift_dct_dct_32[3] = { 2, -4, 0 };
static const int8_t fwd_stage_range_col_dct_dct_32[10] = { 15, 16, 17, 18, 19,
20, 20, 20, 20, 20 };
static const int8_t fwd_stage_range_row_dct_dct_32[10] = { 16, 17, 18, 19, 20,
20, 20, 20, 20, 20 };
static const int8_t fwd_cos_bit_col_dct_dct_32[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const int8_t fwd_cos_bit_row_dct_dct_32[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_dct_32 = {
32, // .txfm_size
10, // .stage_num_col
10, // .stage_num_row
// 1, // .log_scale
fwd_shift_dct_dct_32, // .shift
fwd_stage_range_col_dct_dct_32, // .stage_range_col
fwd_stage_range_row_dct_dct_32, // .stage_range_row
fwd_cos_bit_col_dct_dct_32, // .cos_bit_col
fwd_cos_bit_row_dct_dct_32, // .cos_bit_row
TXFM_TYPE_DCT32, // .txfm_type_col
TXFM_TYPE_DCT32
}; // .txfm_type_row
// ---------------- config fwd_dct_dct_64 ----------------
static const int8_t fwd_shift_dct_dct_64[3] = { 0, -2, -2 };
static const int8_t fwd_stage_range_col_dct_dct_64[12] = {
13, 14, 15, 16, 17, 18, 19, 19, 19, 19, 19, 19
};
static const int8_t fwd_stage_range_row_dct_dct_64[12] = {
17, 18, 19, 20, 21, 22, 22, 22, 22, 22, 22, 22
};
static const int8_t fwd_cos_bit_col_dct_dct_64[12] = { 15, 15, 15, 15, 15, 14,
13, 13, 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_dct_dct_64[12] = { 15, 14, 13, 12, 11, 10,
10, 10, 10, 10, 10, 10 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_dct_64 = {
64, // .txfm_size
12, // .stage_num_col
12, // .stage_num_row
fwd_shift_dct_dct_64, // .shift
fwd_stage_range_col_dct_dct_64, // .stage_range_col
fwd_stage_range_row_dct_dct_64, // .stage_range_row
fwd_cos_bit_col_dct_dct_64, // .cos_bit_col
fwd_cos_bit_row_dct_dct_64, // .cos_bit_row
TXFM_TYPE_DCT64, // .txfm_type_col
TXFM_TYPE_DCT64
}; // .txfm_type_row
// ---------------- config fwd_dct_adst_4 ----------------
static const int8_t fwd_shift_dct_adst_4[3] = { 2, 0, 0 };
static const int8_t fwd_stage_range_col_dct_adst_4[4] = { 15, 16, 17, 17 };
static const int8_t fwd_stage_range_row_dct_adst_4[6] = {
17, 17, 17, 18, 18, 18
};
static const int8_t fwd_cos_bit_col_dct_adst_4[4] = { 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_dct_adst_4[6] = { 13, 13, 13, 13, 13, 13 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_adst_4 = {
4, // .txfm_size
4, // .stage_num_col
6, // .stage_num_row
// 0, // .log_scale
fwd_shift_dct_adst_4, // .shift
fwd_stage_range_col_dct_adst_4, // .stage_range_col
fwd_stage_range_row_dct_adst_4, // .stage_range_row
fwd_cos_bit_col_dct_adst_4, // .cos_bit_col
fwd_cos_bit_row_dct_adst_4, // .cos_bit_row
TXFM_TYPE_DCT4, // .txfm_type_col
TXFM_TYPE_ADST4
}; // .txfm_type_row
// ---------------- config fwd_dct_adst_8 ----------------
static const int8_t fwd_shift_dct_adst_8[3] = { 2, -1, 0 };
static const int8_t fwd_stage_range_col_dct_adst_8[6] = {
15, 16, 17, 18, 18, 18
};
static const int8_t fwd_stage_range_row_dct_adst_8[8] = { 17, 17, 17, 18,
18, 19, 19, 19 };
static const int8_t fwd_cos_bit_col_dct_adst_8[6] = { 13, 13, 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_dct_adst_8[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_adst_8 = {
8, // .txfm_size
6, // .stage_num_col
8, // .stage_num_row
// 0, // .log_scale
fwd_shift_dct_adst_8, // .shift
fwd_stage_range_col_dct_adst_8, // .stage_range_col
fwd_stage_range_row_dct_adst_8, // .stage_range_row
fwd_cos_bit_col_dct_adst_8, // .cos_bit_col
fwd_cos_bit_row_dct_adst_8, // .cos_bit_row
TXFM_TYPE_DCT8, // .txfm_type_col
TXFM_TYPE_ADST8
}; // .txfm_type_row
// ---------------- config fwd_dct_adst_16 ----------------
static const int8_t fwd_shift_dct_adst_16[3] = { 2, -2, 0 };
static const int8_t fwd_stage_range_col_dct_adst_16[8] = { 15, 16, 17, 18,
19, 19, 19, 19 };
static const int8_t fwd_stage_range_row_dct_adst_16[10] = {
17, 17, 17, 18, 18, 19, 19, 20, 20, 20
};
static const int8_t fwd_cos_bit_col_dct_adst_16[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_dct_adst_16[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_adst_16 = {
16, // .txfm_size
8, // .stage_num_col
10, // .stage_num_row
// 0, // .log_scale
fwd_shift_dct_adst_16, // .shift
fwd_stage_range_col_dct_adst_16, // .stage_range_col
fwd_stage_range_row_dct_adst_16, // .stage_range_row
fwd_cos_bit_col_dct_adst_16, // .cos_bit_col
fwd_cos_bit_row_dct_adst_16, // .cos_bit_row
TXFM_TYPE_DCT16, // .txfm_type_col
TXFM_TYPE_ADST16
}; // .txfm_type_row
// ---------------- config fwd_dct_adst_32 ----------------
static const int8_t fwd_shift_dct_adst_32[3] = { 2, -4, 0 };
static const int8_t fwd_stage_range_col_dct_adst_32[10] = {
15, 16, 17, 18, 19, 20, 20, 20, 20, 20
};
static const int8_t fwd_stage_range_row_dct_adst_32[12] = {
16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 20
};
static const int8_t fwd_cos_bit_col_dct_adst_32[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const int8_t fwd_cos_bit_row_dct_adst_32[12] = {
12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12
};
static const TXFM_2D_CFG fwd_txfm_2d_cfg_dct_adst_32 = {
32, // .txfm_size
10, // .stage_num_col
12, // .stage_num_row
// 1, // .log_scale
fwd_shift_dct_adst_32, // .shift
fwd_stage_range_col_dct_adst_32, // .stage_range_col
fwd_stage_range_row_dct_adst_32, // .stage_range_row
fwd_cos_bit_col_dct_adst_32, // .cos_bit_col
fwd_cos_bit_row_dct_adst_32, // .cos_bit_row
TXFM_TYPE_DCT32, // .txfm_type_col
TXFM_TYPE_ADST32
}; // .txfm_type_row
// ---------------- config fwd_adst_adst_4 ----------------
static const int8_t fwd_shift_adst_adst_4[3] = { 2, 0, 0 };
static const int8_t fwd_stage_range_col_adst_adst_4[6] = { 15, 15, 16,
17, 17, 17 };
static const int8_t fwd_stage_range_row_adst_adst_4[6] = { 17, 17, 17,
18, 18, 18 };
static const int8_t fwd_cos_bit_col_adst_adst_4[6] = { 13, 13, 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_adst_adst_4[6] = { 13, 13, 13, 13, 13, 13 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_adst_adst_4 = {
4, // .txfm_size
6, // .stage_num_col
6, // .stage_num_row
// 0, // .log_scale
fwd_shift_adst_adst_4, // .shift
fwd_stage_range_col_adst_adst_4, // .stage_range_col
fwd_stage_range_row_adst_adst_4, // .stage_range_row
fwd_cos_bit_col_adst_adst_4, // .cos_bit_col
fwd_cos_bit_row_adst_adst_4, // .cos_bit_row
TXFM_TYPE_ADST4, // .txfm_type_col
TXFM_TYPE_ADST4
}; // .txfm_type_row
// ---------------- config fwd_adst_adst_8 ----------------
static const int8_t fwd_shift_adst_adst_8[3] = { 2, -1, 0 };
static const int8_t fwd_stage_range_col_adst_adst_8[8] = { 15, 15, 16, 17,
17, 18, 18, 18 };
static const int8_t fwd_stage_range_row_adst_adst_8[8] = { 17, 17, 17, 18,
18, 19, 19, 19 };
static const int8_t fwd_cos_bit_col_adst_adst_8[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_adst_adst_8[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_adst_adst_8 = {
8, // .txfm_size
8, // .stage_num_col
8, // .stage_num_row
// 0, // .log_scale
fwd_shift_adst_adst_8, // .shift
fwd_stage_range_col_adst_adst_8, // .stage_range_col
fwd_stage_range_row_adst_adst_8, // .stage_range_row
fwd_cos_bit_col_adst_adst_8, // .cos_bit_col
fwd_cos_bit_row_adst_adst_8, // .cos_bit_row
TXFM_TYPE_ADST8, // .txfm_type_col
TXFM_TYPE_ADST8
}; // .txfm_type_row
// ---------------- config fwd_adst_adst_16 ----------------
static const int8_t fwd_shift_adst_adst_16[3] = { 2, -2, 0 };
static const int8_t fwd_stage_range_col_adst_adst_16[10] = {
15, 15, 16, 17, 17, 18, 18, 19, 19, 19
};
static const int8_t fwd_stage_range_row_adst_adst_16[10] = {
17, 17, 17, 18, 18, 19, 19, 20, 20, 20
};
static const int8_t fwd_cos_bit_col_adst_adst_16[10] = { 13, 13, 13, 13, 13,
13, 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_adst_adst_16[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_adst_adst_16 = {
16, // .txfm_size
10, // .stage_num_col
10, // .stage_num_row
// 0, // .log_scale
fwd_shift_adst_adst_16, // .shift
fwd_stage_range_col_adst_adst_16, // .stage_range_col
fwd_stage_range_row_adst_adst_16, // .stage_range_row
fwd_cos_bit_col_adst_adst_16, // .cos_bit_col
fwd_cos_bit_row_adst_adst_16, // .cos_bit_row
TXFM_TYPE_ADST16, // .txfm_type_col
TXFM_TYPE_ADST16
}; // .txfm_type_row
// ---------------- config fwd_adst_adst_32 ----------------
static const int8_t fwd_shift_adst_adst_32[3] = { 2, -4, 0 };
static const int8_t fwd_stage_range_col_adst_adst_32[12] = {
15, 15, 16, 17, 17, 18, 18, 19, 19, 20, 20, 20
};
static const int8_t fwd_stage_range_row_adst_adst_32[12] = {
16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 20
};
static const int8_t fwd_cos_bit_col_adst_adst_32[12] = {
12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12
};
static const int8_t fwd_cos_bit_row_adst_adst_32[12] = {
12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12
};
static const TXFM_2D_CFG fwd_txfm_2d_cfg_adst_adst_32 = {
32, // .txfm_size
12, // .stage_num_col
12, // .stage_num_row
// 1, // .log_scale
fwd_shift_adst_adst_32, // .shift
fwd_stage_range_col_adst_adst_32, // .stage_range_col
fwd_stage_range_row_adst_adst_32, // .stage_range_row
fwd_cos_bit_col_adst_adst_32, // .cos_bit_col
fwd_cos_bit_row_adst_adst_32, // .cos_bit_row
TXFM_TYPE_ADST32, // .txfm_type_col
TXFM_TYPE_ADST32
}; // .txfm_type_row
// ---------------- config fwd_adst_dct_4 ----------------
static const int8_t fwd_shift_adst_dct_4[3] = { 2, 0, 0 };
static const int8_t fwd_stage_range_col_adst_dct_4[6] = {
15, 15, 16, 17, 17, 17
};
static const int8_t fwd_stage_range_row_adst_dct_4[4] = { 17, 18, 18, 18 };
static const int8_t fwd_cos_bit_col_adst_dct_4[6] = { 13, 13, 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_adst_dct_4[4] = { 13, 13, 13, 13 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_adst_dct_4 = {
4, // .txfm_size
6, // .stage_num_col
4, // .stage_num_row
// 0, // .log_scale
fwd_shift_adst_dct_4, // .shift
fwd_stage_range_col_adst_dct_4, // .stage_range_col
fwd_stage_range_row_adst_dct_4, // .stage_range_row
fwd_cos_bit_col_adst_dct_4, // .cos_bit_col
fwd_cos_bit_row_adst_dct_4, // .cos_bit_row
TXFM_TYPE_ADST4, // .txfm_type_col
TXFM_TYPE_DCT4
}; // .txfm_type_row
// ---------------- config fwd_adst_dct_8 ----------------
static const int8_t fwd_shift_adst_dct_8[3] = { 2, -1, 0 };
static const int8_t fwd_stage_range_col_adst_dct_8[8] = { 15, 15, 16, 17,
17, 18, 18, 18 };
static const int8_t fwd_stage_range_row_adst_dct_8[6] = {
17, 18, 19, 19, 19, 19
};
static const int8_t fwd_cos_bit_col_adst_dct_8[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_adst_dct_8[6] = { 13, 13, 13, 13, 13, 13 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_adst_dct_8 = {
8, // .txfm_size
8, // .stage_num_col
6, // .stage_num_row
// 0, // .log_scale
fwd_shift_adst_dct_8, // .shift
fwd_stage_range_col_adst_dct_8, // .stage_range_col
fwd_stage_range_row_adst_dct_8, // .stage_range_row
fwd_cos_bit_col_adst_dct_8, // .cos_bit_col
fwd_cos_bit_row_adst_dct_8, // .cos_bit_row
TXFM_TYPE_ADST8, // .txfm_type_col
TXFM_TYPE_DCT8
}; // .txfm_type_row
// ---------------- config fwd_adst_dct_16 ----------------
static const int8_t fwd_shift_adst_dct_16[3] = { 2, -2, 0 };
static const int8_t fwd_stage_range_col_adst_dct_16[10] = {
15, 15, 16, 17, 17, 18, 18, 19, 19, 19
};
static const int8_t fwd_stage_range_row_adst_dct_16[8] = { 17, 18, 19, 20,
20, 20, 20, 20 };
static const int8_t fwd_cos_bit_col_adst_dct_16[10] = { 13, 13, 13, 13, 13,
13, 13, 13, 13, 13 };
static const int8_t fwd_cos_bit_row_adst_dct_16[8] = { 12, 12, 12, 12,
12, 12, 12, 12 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_adst_dct_16 = {
16, // .txfm_size
10, // .stage_num_col
8, // .stage_num_row
// 0, // .log_scale
fwd_shift_adst_dct_16, // .shift
fwd_stage_range_col_adst_dct_16, // .stage_range_col
fwd_stage_range_row_adst_dct_16, // .stage_range_row
fwd_cos_bit_col_adst_dct_16, // .cos_bit_col
fwd_cos_bit_row_adst_dct_16, // .cos_bit_row
TXFM_TYPE_ADST16, // .txfm_type_col
TXFM_TYPE_DCT16
}; // .txfm_type_row
// ---------------- config fwd_adst_dct_32 ----------------
static const int8_t fwd_shift_adst_dct_32[3] = { 2, -4, 0 };
static const int8_t fwd_stage_range_col_adst_dct_32[12] = {
15, 15, 16, 17, 17, 18, 18, 19, 19, 20, 20, 20
};
static const int8_t fwd_stage_range_row_adst_dct_32[10] = {
16, 17, 18, 19, 20, 20, 20, 20, 20, 20
};
static const int8_t fwd_cos_bit_col_adst_dct_32[12] = {
12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12
};
static const int8_t fwd_cos_bit_row_adst_dct_32[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const TXFM_2D_CFG fwd_txfm_2d_cfg_adst_dct_32 = {
32, // .txfm_size
12, // .stage_num_col
10, // .stage_num_row
// 1, // .log_scale
fwd_shift_adst_dct_32, // .shift
fwd_stage_range_col_adst_dct_32, // .stage_range_col
fwd_stage_range_row_adst_dct_32, // .stage_range_row
fwd_cos_bit_col_adst_dct_32, // .cos_bit_col
fwd_cos_bit_row_adst_dct_32, // .cos_bit_row
TXFM_TYPE_ADST32, // .txfm_type_col
TXFM_TYPE_DCT32
}; // .txfm_type_row
#endif // AV1_FWD_TXFM2D_CFG_H_

2334
third_party/aom/av1/common/av1_inv_txfm1d.c vendored Normal file
View file

File diff suppressed because it is too large Load diff

45
third_party/aom/av1/common/av1_inv_txfm1d.h vendored Normal file
View file

@ -0,0 +1,45 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_INV_TXFM1D_H_
#define AV1_INV_TXFM1D_H_
#include "av1/common/av1_txfm.h"
#ifdef __cplusplus
extern "C" {
#endif
void av1_idct4_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
const int8_t *stage_range);
void av1_idct8_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
const int8_t *stage_range);
void av1_idct16_new(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range);
void av1_idct32_new(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range);
void av1_idct64_new(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range);
void av1_iadst4_new(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range);
void av1_iadst8_new(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range);
void av1_iadst16_new(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range);
void av1_iadst32_new(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range);
#ifdef __cplusplus
}
#endif
#endif // AV1_INV_TXFM1D_H_

256
third_party/aom/av1/common/av1_inv_txfm2d.c vendored Normal file
View file

@ -0,0 +1,256 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "./av1_rtcd.h"
#include "av1/common/enums.h"
#include "av1/common/av1_txfm.h"
#include "av1/common/av1_inv_txfm1d.h"
#include "av1/common/av1_inv_txfm2d_cfg.h"
static INLINE TxfmFunc inv_txfm_type_to_func(TXFM_TYPE txfm_type) {
switch (txfm_type) {
case TXFM_TYPE_DCT4: return av1_idct4_new;
case TXFM_TYPE_DCT8: return av1_idct8_new;
case TXFM_TYPE_DCT16: return av1_idct16_new;
case TXFM_TYPE_DCT32: return av1_idct32_new;
case TXFM_TYPE_ADST4: return av1_iadst4_new;
case TXFM_TYPE_ADST8: return av1_iadst8_new;
case TXFM_TYPE_ADST16: return av1_iadst16_new;
case TXFM_TYPE_ADST32: return av1_iadst32_new;
default: assert(0); return NULL;
}
}
const TXFM_2D_CFG *inv_txfm_cfg_ls[TX_TYPES][TX_SIZES] = {
// DCT_DCT
{
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_dct_dct_4, &inv_txfm_2d_cfg_dct_dct_8,
&inv_txfm_2d_cfg_dct_dct_16, &inv_txfm_2d_cfg_dct_dct_32 },
// ADST_DCT
{
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_dct_4, &inv_txfm_2d_cfg_adst_dct_8,
&inv_txfm_2d_cfg_adst_dct_16, &inv_txfm_2d_cfg_adst_dct_32 },
// DCT_ADST
{
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_dct_adst_4, &inv_txfm_2d_cfg_dct_adst_8,
&inv_txfm_2d_cfg_dct_adst_16, &inv_txfm_2d_cfg_dct_adst_32 },
// ADST_ADST
{
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
#if CONFIG_EXT_TX
// FLIPADST_DCT
{
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_dct_4, &inv_txfm_2d_cfg_adst_dct_8,
&inv_txfm_2d_cfg_adst_dct_16, &inv_txfm_2d_cfg_adst_dct_32 },
// DCT_FLIPADST
{
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_dct_adst_4, &inv_txfm_2d_cfg_dct_adst_8,
&inv_txfm_2d_cfg_dct_adst_16, &inv_txfm_2d_cfg_dct_adst_32 },
// FLIPADST_FLIPADST
{
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
// ADST_FLIPADST
{
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
// FLIPADST_ADST
{
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
{ // IDTX
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
{ // V_DCT
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_dct_adst_4, &inv_txfm_2d_cfg_dct_adst_8,
&inv_txfm_2d_cfg_dct_adst_16, &inv_txfm_2d_cfg_dct_adst_32 },
{ // H_DCT
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_dct_4, &inv_txfm_2d_cfg_adst_dct_8,
&inv_txfm_2d_cfg_adst_dct_16, &inv_txfm_2d_cfg_adst_dct_32 },
{ // V_ADST
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
{ // H_ADST
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
{ // V_FLIP_ADST
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
{ // H_FLIP_ADST
#if CONFIG_CB4X4
NULL,
#endif
&inv_txfm_2d_cfg_adst_adst_4, &inv_txfm_2d_cfg_adst_adst_8,
&inv_txfm_2d_cfg_adst_adst_16, &inv_txfm_2d_cfg_adst_adst_32 },
#endif // CONFIG_EXT_TX
};
TXFM_2D_FLIP_CFG av1_get_inv_txfm_cfg(int tx_type, int tx_size) {
TXFM_2D_FLIP_CFG cfg;
set_flip_cfg(tx_type, &cfg);
cfg.cfg = inv_txfm_cfg_ls[tx_type][tx_size];
return cfg;
}
TXFM_2D_FLIP_CFG av1_get_inv_txfm_64x64_cfg(int tx_type) {
TXFM_2D_FLIP_CFG cfg = { 0, 0, NULL };
switch (tx_type) {
case DCT_DCT:
cfg.cfg = &inv_txfm_2d_cfg_dct_dct_64;
set_flip_cfg(tx_type, &cfg);
break;
default: assert(0);
}
return cfg;
}
static INLINE void inv_txfm2d_add_c(const int32_t *input, int16_t *output,
int stride, TXFM_2D_FLIP_CFG *cfg,
int32_t *txfm_buf) {
const int txfm_size = cfg->cfg->txfm_size;
const int8_t *shift = cfg->cfg->shift;
const int8_t *stage_range_col = cfg->cfg->stage_range_col;
const int8_t *stage_range_row = cfg->cfg->stage_range_row;
const int8_t *cos_bit_col = cfg->cfg->cos_bit_col;
const int8_t *cos_bit_row = cfg->cfg->cos_bit_row;
const TxfmFunc txfm_func_col = inv_txfm_type_to_func(cfg->cfg->txfm_type_col);
const TxfmFunc txfm_func_row = inv_txfm_type_to_func(cfg->cfg->txfm_type_row);
// txfm_buf's length is txfm_size * txfm_size + 2 * txfm_size
// it is used for intermediate data buffering
int32_t *temp_in = txfm_buf;
int32_t *temp_out = temp_in + txfm_size;
int32_t *buf = temp_out + txfm_size;
int32_t *buf_ptr = buf;
int c, r;
// Rows
for (r = 0; r < txfm_size; ++r) {
txfm_func_row(input, buf_ptr, cos_bit_row, stage_range_row);
round_shift_array(buf_ptr, txfm_size, -shift[0]);
input += txfm_size;
buf_ptr += txfm_size;
}
// Columns
for (c = 0; c < txfm_size; ++c) {
if (cfg->lr_flip == 0) {
for (r = 0; r < txfm_size; ++r) temp_in[r] = buf[r * txfm_size + c];
} else {
// flip left right
for (r = 0; r < txfm_size; ++r)
temp_in[r] = buf[r * txfm_size + (txfm_size - c - 1)];
}
txfm_func_col(temp_in, temp_out, cos_bit_col, stage_range_col);
round_shift_array(temp_out, txfm_size, -shift[1]);
if (cfg->ud_flip == 0) {
for (r = 0; r < txfm_size; ++r) output[r * stride + c] += temp_out[r];
} else {
// flip upside down
for (r = 0; r < txfm_size; ++r)
output[r * stride + c] += temp_out[txfm_size - r - 1];
}
}
}
static INLINE void inv_txfm2d_add_facade(const int32_t *input, uint16_t *output,
int stride, int32_t *txfm_buf,
int tx_type, int tx_size, int bd) {
// output contains the prediction signal which is always positive and smaller
// than (1 << bd) - 1
// since bd < 16-1, therefore we can treat the uint16_t* output buffer as an
// int16_t*
TXFM_2D_FLIP_CFG cfg = av1_get_inv_txfm_cfg(tx_type, tx_size);
inv_txfm2d_add_c(input, (int16_t *)output, stride, &cfg, txfm_buf);
clamp_block((int16_t *)output, cfg.cfg->txfm_size, stride, 0, (1 << bd) - 1);
}
void av1_inv_txfm2d_add_4x4_c(const int32_t *input, uint16_t *output,
int stride, int tx_type, int bd) {
int txfm_buf[4 * 4 + 4 + 4];
inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_4X4, bd);
}
void av1_inv_txfm2d_add_8x8_c(const int32_t *input, uint16_t *output,
int stride, int tx_type, int bd) {
int txfm_buf[8 * 8 + 8 + 8];
inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X8, bd);
}
void av1_inv_txfm2d_add_16x16_c(const int32_t *input, uint16_t *output,
int stride, int tx_type, int bd) {
int txfm_buf[16 * 16 + 16 + 16];
inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X16, bd);
}
void av1_inv_txfm2d_add_32x32_c(const int32_t *input, uint16_t *output,
int stride, int tx_type, int bd) {
int txfm_buf[32 * 32 + 32 + 32];
inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_32X32, bd);
}
void av1_inv_txfm2d_add_64x64_c(const int32_t *input, uint16_t *output,
int stride, int tx_type, int bd) {
int txfm_buf[64 * 64 + 64 + 64];
// output contains the prediction signal which is always positive and smaller
// than (1 << bd) - 1
// since bd < 16-1, therefore we can treat the uint16_t* output buffer as an
// int16_t*
TXFM_2D_FLIP_CFG cfg = av1_get_inv_txfm_64x64_cfg(tx_type);
inv_txfm2d_add_c(input, (int16_t *)output, stride, &cfg, txfm_buf);
clamp_block((int16_t *)output, 64, stride, 0, (1 << bd) - 1);
}

447
third_party/aom/av1/common/av1_inv_txfm2d_cfg.h vendored Normal file
View file

@ -0,0 +1,447 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_INV_TXFM2D_CFG_H_
#define AV1_INV_TXFM2D_CFG_H_
#include "av1/common/av1_inv_txfm1d.h"
// ---------------- config inv_dct_dct_4 ----------------
static const int8_t inv_shift_dct_dct_4[2] = { 0, -4 };
static const int8_t inv_stage_range_col_dct_dct_4[4] = { 18, 18, 17, 17 };
static const int8_t inv_stage_range_row_dct_dct_4[4] = { 18, 18, 18, 18 };
static const int8_t inv_cos_bit_col_dct_dct_4[4] = { 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_dct_4[4] = { 13, 13, 13, 13 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_dct_4 = {
4, // .txfm_size
4, // .stage_num_col
4, // .stage_num_row
// 0, // .log_scale
inv_shift_dct_dct_4, // .shift
inv_stage_range_col_dct_dct_4, // .stage_range_col
inv_stage_range_row_dct_dct_4, // .stage_range_row
inv_cos_bit_col_dct_dct_4, // .cos_bit_col
inv_cos_bit_row_dct_dct_4, // .cos_bit_row
TXFM_TYPE_DCT4, // .txfm_type_col
TXFM_TYPE_DCT4
}; // .txfm_type_row
// ---------------- config inv_dct_dct_8 ----------------
static const int8_t inv_shift_dct_dct_8[2] = { 0, -5 };
static const int8_t inv_stage_range_col_dct_dct_8[6] = {
19, 19, 19, 19, 18, 18
};
static const int8_t inv_stage_range_row_dct_dct_8[6] = {
19, 19, 19, 19, 19, 19
};
static const int8_t inv_cos_bit_col_dct_dct_8[6] = { 13, 13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_dct_8[6] = { 13, 13, 13, 13, 13, 13 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_dct_8 = {
8, // .txfm_size
6, // .stage_num_col
6, // .stage_num_row
// 0, // .log_scale
inv_shift_dct_dct_8, // .shift
inv_stage_range_col_dct_dct_8, // .stage_range_col
inv_stage_range_row_dct_dct_8, // .stage_range_row
inv_cos_bit_col_dct_dct_8, // .cos_bit_col
inv_cos_bit_row_dct_dct_8, // .cos_bit_row
TXFM_TYPE_DCT8, // .txfm_type_col
TXFM_TYPE_DCT8
}; // .txfm_type_row
// ---------------- config inv_dct_dct_16 ----------------
static const int8_t inv_shift_dct_dct_16[2] = { -1, -5 };
static const int8_t inv_stage_range_col_dct_dct_16[8] = { 19, 19, 19, 19,
19, 19, 18, 18 };
static const int8_t inv_stage_range_row_dct_dct_16[8] = { 20, 20, 20, 20,
20, 20, 20, 20 };
static const int8_t inv_cos_bit_col_dct_dct_16[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_dct_16[8] = { 12, 12, 12, 12,
12, 12, 12, 12 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_dct_16 = {
16, // .txfm_size
8, // .stage_num_col
8, // .stage_num_row
// 0, // .log_scale
inv_shift_dct_dct_16, // .shift
inv_stage_range_col_dct_dct_16, // .stage_range_col
inv_stage_range_row_dct_dct_16, // .stage_range_row
inv_cos_bit_col_dct_dct_16, // .cos_bit_col
inv_cos_bit_row_dct_dct_16, // .cos_bit_row
TXFM_TYPE_DCT16, // .txfm_type_col
TXFM_TYPE_DCT16
}; // .txfm_type_row
// ---------------- config inv_dct_dct_32 ----------------
static const int8_t inv_shift_dct_dct_32[2] = { -1, -5 };
static const int8_t inv_stage_range_col_dct_dct_32[10] = { 19, 19, 19, 19, 19,
19, 19, 19, 18, 18 };
static const int8_t inv_stage_range_row_dct_dct_32[10] = { 20, 20, 20, 20, 20,
20, 20, 20, 20, 20 };
static const int8_t inv_cos_bit_col_dct_dct_32[10] = { 13, 13, 13, 13, 13,
13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_dct_32[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_dct_32 = {
32, // .txfm_size
10, // .stage_num_col
10, // .stage_num_row
// 1, // .log_scale
inv_shift_dct_dct_32, // .shift
inv_stage_range_col_dct_dct_32, // .stage_range_col
inv_stage_range_row_dct_dct_32, // .stage_range_row
inv_cos_bit_col_dct_dct_32, // .cos_bit_col
inv_cos_bit_row_dct_dct_32, // .cos_bit_row
TXFM_TYPE_DCT32, // .txfm_type_col
TXFM_TYPE_DCT32
}; // .txfm_type_row
// ---------------- config inv_dct_dct_64 ----------------
static const int8_t inv_shift_dct_dct_64[2] = { -1, -7 };
static const int8_t inv_stage_range_col_dct_dct_64[12] = {
19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 18, 18
};
static const int8_t inv_stage_range_row_dct_dct_64[12] = {
20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20
};
static const int8_t inv_cos_bit_col_dct_dct_64[12] = { 13, 13, 13, 13, 13, 13,
13, 13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_dct_64[12] = { 12, 12, 12, 12, 12, 12,
12, 12, 12, 12, 12, 12 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_dct_64 = {
64, // .txfm_size
12, // .stage_num_col
12, // .stage_num_row
inv_shift_dct_dct_64, // .shift
inv_stage_range_col_dct_dct_64, // .stage_range_col
inv_stage_range_row_dct_dct_64, // .stage_range_row
inv_cos_bit_col_dct_dct_64, // .cos_bit_col
inv_cos_bit_row_dct_dct_64, // .cos_bit_row
TXFM_TYPE_DCT64, // .txfm_type_col
TXFM_TYPE_DCT64
}; // .txfm_type_row
// ---------------- config inv_dct_adst_4 ----------------
static const int8_t inv_shift_dct_adst_4[2] = { 0, -4 };
static const int8_t inv_stage_range_col_dct_adst_4[4] = { 18, 18, 17, 17 };
static const int8_t inv_stage_range_row_dct_adst_4[6] = {
18, 18, 18, 18, 18, 18
};
static const int8_t inv_cos_bit_col_dct_adst_4[4] = { 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_adst_4[6] = { 13, 13, 13, 13, 13, 13 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_adst_4 = {
4, // .txfm_size
4, // .stage_num_col
6, // .stage_num_row
// 0, // .log_scale
inv_shift_dct_adst_4, // .shift
inv_stage_range_col_dct_adst_4, // .stage_range_col
inv_stage_range_row_dct_adst_4, // .stage_range_row
inv_cos_bit_col_dct_adst_4, // .cos_bit_col
inv_cos_bit_row_dct_adst_4, // .cos_bit_row
TXFM_TYPE_DCT4, // .txfm_type_col
TXFM_TYPE_ADST4
}; // .txfm_type_row
// ---------------- config inv_dct_adst_8 ----------------
static const int8_t inv_shift_dct_adst_8[2] = { 0, -5 };
static const int8_t inv_stage_range_col_dct_adst_8[6] = {
19, 19, 19, 19, 18, 18
};
static const int8_t inv_stage_range_row_dct_adst_8[8] = { 19, 19, 19, 19,
19, 19, 19, 19 };
static const int8_t inv_cos_bit_col_dct_adst_8[6] = { 13, 13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_adst_8[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_adst_8 = {
8, // .txfm_size
6, // .stage_num_col
8, // .stage_num_row
// 0, // .log_scale
inv_shift_dct_adst_8, // .shift
inv_stage_range_col_dct_adst_8, // .stage_range_col
inv_stage_range_row_dct_adst_8, // .stage_range_row
inv_cos_bit_col_dct_adst_8, // .cos_bit_col
inv_cos_bit_row_dct_adst_8, // .cos_bit_row
TXFM_TYPE_DCT8, // .txfm_type_col
TXFM_TYPE_ADST8
}; // .txfm_type_row
// ---------------- config inv_dct_adst_16 ----------------
static const int8_t inv_shift_dct_adst_16[2] = { -1, -5 };
static const int8_t inv_stage_range_col_dct_adst_16[8] = { 19, 19, 19, 19,
19, 19, 18, 18 };
static const int8_t inv_stage_range_row_dct_adst_16[10] = {
20, 20, 20, 20, 20, 20, 20, 20, 20, 20
};
static const int8_t inv_cos_bit_col_dct_adst_16[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_adst_16[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_adst_16 = {
16, // .txfm_size
8, // .stage_num_col
10, // .stage_num_row
// 0, // .log_scale
inv_shift_dct_adst_16, // .shift
inv_stage_range_col_dct_adst_16, // .stage_range_col
inv_stage_range_row_dct_adst_16, // .stage_range_row
inv_cos_bit_col_dct_adst_16, // .cos_bit_col
inv_cos_bit_row_dct_adst_16, // .cos_bit_row
TXFM_TYPE_DCT16, // .txfm_type_col
TXFM_TYPE_ADST16
}; // .txfm_type_row
// ---------------- config inv_dct_adst_32 ----------------
static const int8_t inv_shift_dct_adst_32[2] = { -1, -5 };
static const int8_t inv_stage_range_col_dct_adst_32[10] = {
19, 19, 19, 19, 19, 19, 19, 19, 18, 18
};
static const int8_t inv_stage_range_row_dct_adst_32[12] = {
20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20
};
static const int8_t inv_cos_bit_col_dct_adst_32[10] = { 13, 13, 13, 13, 13,
13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_dct_adst_32[12] = {
12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12
};
static const TXFM_2D_CFG inv_txfm_2d_cfg_dct_adst_32 = {
32, // .txfm_size
10, // .stage_num_col
12, // .stage_num_row
// 1, // .log_scale
inv_shift_dct_adst_32, // .shift
inv_stage_range_col_dct_adst_32, // .stage_range_col
inv_stage_range_row_dct_adst_32, // .stage_range_row
inv_cos_bit_col_dct_adst_32, // .cos_bit_col
inv_cos_bit_row_dct_adst_32, // .cos_bit_row
TXFM_TYPE_DCT32, // .txfm_type_col
TXFM_TYPE_ADST32
}; // .txfm_type_row
// ---------------- config inv_adst_adst_4 ----------------
static const int8_t inv_shift_adst_adst_4[2] = { 0, -4 };
static const int8_t inv_stage_range_col_adst_adst_4[6] = { 18, 18, 18,
18, 17, 17 };
static const int8_t inv_stage_range_row_adst_adst_4[6] = { 18, 18, 18,
18, 18, 18 };
static const int8_t inv_cos_bit_col_adst_adst_4[6] = { 13, 13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_adst_adst_4[6] = { 13, 13, 13, 13, 13, 13 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_adst_adst_4 = {
4, // .txfm_size
6, // .stage_num_col
6, // .stage_num_row
// 0, // .log_scale
inv_shift_adst_adst_4, // .shift
inv_stage_range_col_adst_adst_4, // .stage_range_col
inv_stage_range_row_adst_adst_4, // .stage_range_row
inv_cos_bit_col_adst_adst_4, // .cos_bit_col
inv_cos_bit_row_adst_adst_4, // .cos_bit_row
TXFM_TYPE_ADST4, // .txfm_type_col
TXFM_TYPE_ADST4
}; // .txfm_type_row
// ---------------- config inv_adst_adst_8 ----------------
static const int8_t inv_shift_adst_adst_8[2] = { 0, -5 };
static const int8_t inv_stage_range_col_adst_adst_8[8] = { 19, 19, 19, 19,
19, 19, 18, 18 };
static const int8_t inv_stage_range_row_adst_adst_8[8] = { 19, 19, 19, 19,
19, 19, 19, 19 };
static const int8_t inv_cos_bit_col_adst_adst_8[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_adst_adst_8[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_adst_adst_8 = {
8, // .txfm_size
8, // .stage_num_col
8, // .stage_num_row
// 0, // .log_scale
inv_shift_adst_adst_8, // .shift
inv_stage_range_col_adst_adst_8, // .stage_range_col
inv_stage_range_row_adst_adst_8, // .stage_range_row
inv_cos_bit_col_adst_adst_8, // .cos_bit_col
inv_cos_bit_row_adst_adst_8, // .cos_bit_row
TXFM_TYPE_ADST8, // .txfm_type_col
TXFM_TYPE_ADST8
}; // .txfm_type_row
// ---------------- config inv_adst_adst_16 ----------------
static const int8_t inv_shift_adst_adst_16[2] = { -1, -5 };
static const int8_t inv_stage_range_col_adst_adst_16[10] = {
19, 19, 19, 19, 19, 19, 19, 19, 18, 18
};
static const int8_t inv_stage_range_row_adst_adst_16[10] = {
20, 20, 20, 20, 20, 20, 20, 20, 20, 20
};
static const int8_t inv_cos_bit_col_adst_adst_16[10] = { 13, 13, 13, 13, 13,
13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_adst_adst_16[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_adst_adst_16 = {
16, // .txfm_size
10, // .stage_num_col
10, // .stage_num_row
// 0, // .log_scale
inv_shift_adst_adst_16, // .shift
inv_stage_range_col_adst_adst_16, // .stage_range_col
inv_stage_range_row_adst_adst_16, // .stage_range_row
inv_cos_bit_col_adst_adst_16, // .cos_bit_col
inv_cos_bit_row_adst_adst_16, // .cos_bit_row
TXFM_TYPE_ADST16, // .txfm_type_col
TXFM_TYPE_ADST16
}; // .txfm_type_row
// ---------------- config inv_adst_adst_32 ----------------
static const int8_t inv_shift_adst_adst_32[2] = { -1, -5 };
static const int8_t inv_stage_range_col_adst_adst_32[12] = {
19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 18, 18
};
static const int8_t inv_stage_range_row_adst_adst_32[12] = {
20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20
};
static const int8_t inv_cos_bit_col_adst_adst_32[12] = {
13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13
};
static const int8_t inv_cos_bit_row_adst_adst_32[12] = {
12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12
};
static const TXFM_2D_CFG inv_txfm_2d_cfg_adst_adst_32 = {
32, // .txfm_size
12, // .stage_num_col
12, // .stage_num_row
// 1, // .log_scale
inv_shift_adst_adst_32, // .shift
inv_stage_range_col_adst_adst_32, // .stage_range_col
inv_stage_range_row_adst_adst_32, // .stage_range_row
inv_cos_bit_col_adst_adst_32, // .cos_bit_col
inv_cos_bit_row_adst_adst_32, // .cos_bit_row
TXFM_TYPE_ADST32, // .txfm_type_col
TXFM_TYPE_ADST32
}; // .txfm_type_row
// ---------------- config inv_adst_dct_4 ----------------
static const int8_t inv_shift_adst_dct_4[2] = { 0, -4 };
static const int8_t inv_stage_range_col_adst_dct_4[6] = {
18, 18, 18, 18, 17, 17
};
static const int8_t inv_stage_range_row_adst_dct_4[4] = { 18, 18, 18, 18 };
static const int8_t inv_cos_bit_col_adst_dct_4[6] = { 13, 13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_adst_dct_4[4] = { 13, 13, 13, 13 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_adst_dct_4 = {
4, // .txfm_size
6, // .stage_num_col
4, // .stage_num_row
// 0, // .log_scale
inv_shift_adst_dct_4, // .shift
inv_stage_range_col_adst_dct_4, // .stage_range_col
inv_stage_range_row_adst_dct_4, // .stage_range_row
inv_cos_bit_col_adst_dct_4, // .cos_bit_col
inv_cos_bit_row_adst_dct_4, // .cos_bit_row
TXFM_TYPE_ADST4, // .txfm_type_col
TXFM_TYPE_DCT4
}; // .txfm_type_row
// ---------------- config inv_adst_dct_8 ----------------
static const int8_t inv_shift_adst_dct_8[2] = { 0, -5 };
static const int8_t inv_stage_range_col_adst_dct_8[8] = { 19, 19, 19, 19,
19, 19, 18, 18 };
static const int8_t inv_stage_range_row_adst_dct_8[6] = {
19, 19, 19, 19, 19, 19
};
static const int8_t inv_cos_bit_col_adst_dct_8[8] = { 13, 13, 13, 13,
13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_adst_dct_8[6] = { 13, 13, 13, 13, 13, 13 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_adst_dct_8 = {
8, // .txfm_size
8, // .stage_num_col
6, // .stage_num_row
// 0, // .log_scale
inv_shift_adst_dct_8, // .shift
inv_stage_range_col_adst_dct_8, // .stage_range_col
inv_stage_range_row_adst_dct_8, // .stage_range_row
inv_cos_bit_col_adst_dct_8, // .cos_bit_col
inv_cos_bit_row_adst_dct_8, // .cos_bit_row
TXFM_TYPE_ADST8, // .txfm_type_col
TXFM_TYPE_DCT8
}; // .txfm_type_row
// ---------------- config inv_adst_dct_16 ----------------
static const int8_t inv_shift_adst_dct_16[2] = { -1, -5 };
static const int8_t inv_stage_range_col_adst_dct_16[10] = {
19, 19, 19, 19, 19, 19, 19, 19, 18, 18
};
static const int8_t inv_stage_range_row_adst_dct_16[8] = { 20, 20, 20, 20,
20, 20, 20, 20 };
static const int8_t inv_cos_bit_col_adst_dct_16[10] = { 13, 13, 13, 13, 13,
13, 13, 13, 13, 13 };
static const int8_t inv_cos_bit_row_adst_dct_16[8] = { 12, 12, 12, 12,
12, 12, 12, 12 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_adst_dct_16 = {
16, // .txfm_size
10, // .stage_num_col
8, // .stage_num_row
// 0, // .log_scale
inv_shift_adst_dct_16, // .shift
inv_stage_range_col_adst_dct_16, // .stage_range_col
inv_stage_range_row_adst_dct_16, // .stage_range_row
inv_cos_bit_col_adst_dct_16, // .cos_bit_col
inv_cos_bit_row_adst_dct_16, // .cos_bit_row
TXFM_TYPE_ADST16, // .txfm_type_col
TXFM_TYPE_DCT16
}; // .txfm_type_row
// ---------------- config inv_adst_dct_32 ----------------
static const int8_t inv_shift_adst_dct_32[2] = { -1, -5 };
static const int8_t inv_stage_range_col_adst_dct_32[12] = {
19, 19, 19, 19, 19, 19, 19, 19, 19, 19, 18, 18
};
static const int8_t inv_stage_range_row_adst_dct_32[10] = {
20, 20, 20, 20, 20, 20, 20, 20, 20, 20
};
static const int8_t inv_cos_bit_col_adst_dct_32[12] = {
13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13
};
static const int8_t inv_cos_bit_row_adst_dct_32[10] = { 12, 12, 12, 12, 12,
12, 12, 12, 12, 12 };
static const TXFM_2D_CFG inv_txfm_2d_cfg_adst_dct_32 = {
32, // .txfm_size
12, // .stage_num_col
10, // .stage_num_row
// 1, // .log_scale
inv_shift_adst_dct_32, // .shift
inv_stage_range_col_adst_dct_32, // .stage_range_col
inv_stage_range_row_adst_dct_32, // .stage_range_row
inv_cos_bit_col_adst_dct_32, // .cos_bit_col
inv_cos_bit_row_adst_dct_32, // .cos_bit_row
TXFM_TYPE_ADST32, // .txfm_type_col
TXFM_TYPE_DCT32
}; // .txfm_type_row
extern const TXFM_2D_CFG *inv_txfm_cfg_ls[TX_TYPES][TX_SIZES];
#endif // AV1_INV_TXFM2D_CFG_H_

2336
third_party/aom/av1/common/av1_loopfilter.c vendored Normal file
View file

File diff suppressed because it is too large Load diff

162
third_party/aom/av1/common/av1_loopfilter.h vendored Normal file
View file

@ -0,0 +1,162 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_LOOPFILTER_H_
#define AV1_COMMON_LOOPFILTER_H_
#include "aom_ports/mem.h"
#include "./aom_config.h"
#include "av1/common/blockd.h"
#include "av1/common/seg_common.h"
#ifdef __cplusplus
extern "C" {
#endif
#define MAX_LOOP_FILTER 63
#define MAX_SHARPNESS 7
#define SIMD_WIDTH 16
#define MAX_MODE_LF_DELTAS 2
enum lf_path {
LF_PATH_420,
LF_PATH_444,
LF_PATH_SLOW,
};
struct loopfilter {
int filter_level;
int sharpness_level;
int last_sharpness_level;
uint8_t mode_ref_delta_enabled;
uint8_t mode_ref_delta_update;
// 0 = Intra, Last, Last2+Last3(CONFIG_EXT_REFS),
// GF, BRF(CONFIG_EXT_REFS), ARF
signed char ref_deltas[TOTAL_REFS_PER_FRAME];
signed char last_ref_deltas[TOTAL_REFS_PER_FRAME];
// 0 = ZERO_MV, MV
signed char mode_deltas[MAX_MODE_LF_DELTAS];
signed char last_mode_deltas[MAX_MODE_LF_DELTAS];
};
// Need to align this structure so when it is declared and
// passed it can be loaded into vector registers.
typedef struct {
DECLARE_ALIGNED(SIMD_WIDTH, uint8_t, mblim[SIMD_WIDTH]);
DECLARE_ALIGNED(SIMD_WIDTH, uint8_t, lim[SIMD_WIDTH]);
DECLARE_ALIGNED(SIMD_WIDTH, uint8_t, hev_thr[SIMD_WIDTH]);
} loop_filter_thresh;
typedef struct {
loop_filter_thresh lfthr[MAX_LOOP_FILTER + 1];
uint8_t lvl[MAX_SEGMENTS][TOTAL_REFS_PER_FRAME][MAX_MODE_LF_DELTAS];
} loop_filter_info_n;
// This structure holds bit masks for all 8x8 blocks in a 64x64 region.
// Each 1 bit represents a position in which we want to apply the loop filter.
// Left_ entries refer to whether we apply a filter on the border to the
// left of the block. Above_ entries refer to whether or not to apply a
// filter on the above border. Int_ entries refer to whether or not to
// apply borders on the 4x4 edges within the 8x8 block that each bit
// represents.
// Since each transform is accompanied by a potentially different type of
// loop filter there is a different entry in the array for each transform size.
typedef struct {
uint64_t left_y[TX_SIZES];
uint64_t above_y[TX_SIZES];
uint64_t int_4x4_y;
uint16_t left_uv[TX_SIZES];
uint16_t above_uv[TX_SIZES];
uint16_t left_int_4x4_uv;
uint16_t above_int_4x4_uv;
uint8_t lfl_y[MAX_MIB_SIZE][MAX_MIB_SIZE];
uint8_t lfl_uv[MAX_MIB_SIZE / 2][MAX_MIB_SIZE / 2];
} LOOP_FILTER_MASK;
/* assorted loopfilter functions which get used elsewhere */
struct AV1Common;
struct macroblockd;
struct AV1LfSyncData;
// This function sets up the bit masks for the entire 64x64 region represented
// by mi_row, mi_col.
void av1_setup_mask(struct AV1Common *const cm, const int mi_row,
const int mi_col, MODE_INFO **mi_8x8,
const int mode_info_stride, LOOP_FILTER_MASK *lfm);
void av1_filter_block_plane_ss00_ver(struct AV1Common *const cm,
struct macroblockd_plane *const plane,
int mi_row, LOOP_FILTER_MASK *lfm);
void av1_filter_block_plane_ss00_hor(struct AV1Common *const cm,
struct macroblockd_plane *const plane,
int mi_row, LOOP_FILTER_MASK *lfm);
void av1_filter_block_plane_ss11_ver(struct AV1Common *const cm,
struct macroblockd_plane *const plane,
int mi_row, LOOP_FILTER_MASK *lfm);
void av1_filter_block_plane_ss11_hor(struct AV1Common *const cm,
struct macroblockd_plane *const plane,
int mi_row, LOOP_FILTER_MASK *lfm);
void av1_filter_block_plane_non420_ver(struct AV1Common *const cm,
struct macroblockd_plane *plane,
MODE_INFO **mi_8x8, int mi_row,
int mi_col);
void av1_filter_block_plane_non420_hor(struct AV1Common *const cm,
struct macroblockd_plane *plane,
MODE_INFO **mi_8x8, int mi_row,
int mi_col);
void av1_loop_filter_init(struct AV1Common *cm);
// Update the loop filter for the current frame.
// This should be called before av1_loop_filter_rows(),
// av1_loop_filter_frame()
// calls this function directly.
void av1_loop_filter_frame_init(struct AV1Common *cm, int default_filt_lvl);
void av1_loop_filter_frame(YV12_BUFFER_CONFIG *frame, struct AV1Common *cm,
struct macroblockd *mbd, int filter_level,
int y_only, int partial_frame);
// Apply the loop filter to [start, stop) macro block rows in frame_buffer.
void av1_loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer,
struct AV1Common *cm,
struct macroblockd_plane planes[MAX_MB_PLANE],
int start, int stop, int y_only);
typedef struct LoopFilterWorkerData {
YV12_BUFFER_CONFIG *frame_buffer;
struct AV1Common *cm;
struct macroblockd_plane planes[MAX_MB_PLANE];
int start;
int stop;
int y_only;
} LFWorkerData;
void av1_loop_filter_data_reset(
LFWorkerData *lf_data, YV12_BUFFER_CONFIG *frame_buffer,
struct AV1Common *cm, const struct macroblockd_plane planes[MAX_MB_PLANE]);
// Operates on the rows described by 'lf_data'.
int av1_loop_filter_worker(LFWorkerData *const lf_data, void *unused);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_LOOPFILTER_H_

20
third_party/aom/av1/common/av1_rtcd.c vendored Normal file
View file

@ -0,0 +1,20 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "./aom_config.h"
#define RTCD_C
#include "./av1_rtcd.h"
#include "aom_ports/aom_once.h"
void av1_rtcd() {
// TODO(JBB): Remove this once, by insuring that both the encoder and
// decoder setup functions are protected by once();
once(setup_rtcd_internal);
}

644
third_party/aom/av1/common/av1_rtcd_defs.pl vendored Executable file
View file

@ -0,0 +1,644 @@
sub av1_common_forward_decls() {
print <<EOF
/*
* AV1
*/
#include "aom/aom_integer.h"
#include "av1/common/common.h"
#include "av1/common/enums.h"
#include "av1/common/quant_common.h"
#include "av1/common/filter.h"
#include "av1/common/convolve.h"
#include "av1/common/av1_txfm.h"
#include "av1/common/odintrin.h"
struct macroblockd;
/* Encoder forward decls */
struct macroblock;
struct aom_variance_vtable;
struct search_site_config;
struct mv;
union int_mv;
struct yv12_buffer_config;
typedef uint16_t od_dering_in;
EOF
}
forward_decls qw/av1_common_forward_decls/;
# functions that are 64 bit only.
$mmx_x86_64 = $sse2_x86_64 = $ssse3_x86_64 = $avx_x86_64 = $avx2_x86_64 = '';
if ($opts{arch} eq "x86_64") {
$mmx_x86_64 = 'mmx';
$sse2_x86_64 = 'sse2';
$ssse3_x86_64 = 'ssse3';
$avx_x86_64 = 'avx';
$avx2_x86_64 = 'avx2';
}
#
# 10/12-tap convolution filters
#
add_proto qw/void av1_lowbd_convolve_init/, "void";
specialize qw/av1_lowbd_convolve_init ssse3/;
add_proto qw/void av1_convolve_horiz/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams fp, const int subpel_x_q4, int x_step_q4, ConvolveParams *conv_params";
specialize qw/av1_convolve_horiz ssse3/;
add_proto qw/void av1_convolve_vert/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams fp, const int subpel_x_q4, int x_step_q4, ConvolveParams *conv_params";
specialize qw/av1_convolve_vert ssse3/;
if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
add_proto qw/void av1_highbd_convolve_init/, "void";
specialize qw/av1_highbd_convolve_init sse4_1/;
add_proto qw/void av1_highbd_convolve_horiz/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams fp, const int subpel_x_q4, int x_step_q4, int avg, int bd";
specialize qw/av1_highbd_convolve_horiz sse4_1/;
add_proto qw/void av1_highbd_convolve_vert/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams fp, const int subpel_x_q4, int x_step_q4, int avg, int bd";
specialize qw/av1_highbd_convolve_vert sse4_1/;
}
#
# Inverse dct
#
if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
{
add_proto qw/void av1_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/av1_iht4x4_16_add sse2/;
add_proto qw/void av1_iht4x8_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/av1_iht4x8_32_add sse2/;
add_proto qw/void av1_iht8x4_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/av1_iht8x4_32_add sse2/;
add_proto qw/void av1_iht8x16_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/av1_iht8x16_128_add sse2/;
add_proto qw/void av1_iht16x8_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/av1_iht16x8_128_add sse2/;
add_proto qw/void av1_iht16x32_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/av1_iht16x32_512_add sse2/;
add_proto qw/void av1_iht32x16_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/av1_iht32x16_512_add sse2/;
add_proto qw/void av1_iht4x16_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
add_proto qw/void av1_iht16x4_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
add_proto qw/void av1_iht8x32_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
add_proto qw/void av1_iht32x8_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
add_proto qw/void av1_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/av1_iht8x8_64_add sse2/;
add_proto qw/void av1_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
specialize qw/av1_iht16x16_256_add sse2 avx2/;
add_proto qw/void av1_iht32x32_1024_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
}
} else {
{
add_proto qw/void av1_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/av1_iht4x4_16_add sse2 neon dspr2/;
add_proto qw/void av1_iht4x8_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/av1_iht4x8_32_add sse2/;
add_proto qw/void av1_iht8x4_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/av1_iht8x4_32_add sse2/;
add_proto qw/void av1_iht8x16_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/av1_iht8x16_128_add sse2/;
add_proto qw/void av1_iht16x8_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/av1_iht16x8_128_add sse2/;
add_proto qw/void av1_iht16x32_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/av1_iht16x32_512_add sse2/;
add_proto qw/void av1_iht32x16_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/av1_iht32x16_512_add sse2/;
add_proto qw/void av1_iht4x16_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
add_proto qw/void av1_iht16x4_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
add_proto qw/void av1_iht8x32_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
add_proto qw/void av1_iht32x8_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
add_proto qw/void av1_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/av1_iht8x8_64_add sse2 neon dspr2/;
add_proto qw/void av1_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
specialize qw/av1_iht16x16_256_add sse2 avx2 dspr2/;
add_proto qw/void av1_iht32x32_1024_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
if (aom_config("CONFIG_EXT_TX") ne "yes") {
specialize qw/av1_iht4x4_16_add msa/;
specialize qw/av1_iht8x8_64_add msa/;
specialize qw/av1_iht16x16_256_add msa/;
}
}
}
add_proto qw/void av1_iht32x32_1024_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
if (aom_config("CONFIG_TX64X64") eq "yes") {
add_proto qw/void av1_iht64x64_4096_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
}
if (aom_config("CONFIG_NEW_QUANT") eq "yes") {
add_proto qw/void quantize_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
add_proto qw/void quantize_fp_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
add_proto qw/void quantize_32x32_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
add_proto qw/void quantize_32x32_fp_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
if (aom_config("CONFIG_TX64X64") eq "yes") {
add_proto qw/void quantize_64x64_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
add_proto qw/void quantize_64x64_fp_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
}
}
# FILTER_INTRA predictor functions
if (aom_config("CONFIG_FILTER_INTRA") eq "yes") {
add_proto qw/void av1_dc_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, int bs, const uint8_t *above, const uint8_t *left";
add_proto qw/void av1_v_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, int bs, const uint8_t *above, const uint8_t *left";
add_proto qw/void av1_h_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, int bs, const uint8_t *above, const uint8_t *left";
add_proto qw/void av1_d45_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, int bs, const uint8_t *above, const uint8_t *left";
add_proto qw/void av1_d135_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, int bs, const uint8_t *above, const uint8_t *left";
add_proto qw/void av1_d117_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, int bs, const uint8_t *above, const uint8_t *left";
add_proto qw/void av1_d153_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, int bs, const uint8_t *above, const uint8_t *left";
add_proto qw/void av1_d207_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, int bs, const uint8_t *above, const uint8_t *left";
add_proto qw/void av1_d63_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, int bs, const uint8_t *above, const uint8_t *left";
add_proto qw/void av1_tm_filter_predictor/, "uint8_t *dst, ptrdiff_t stride, int bs, const uint8_t *above, const uint8_t *left";
# High bitdepth functions
if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
add_proto qw/void av1_highbd_dc_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, int bs, const uint16_t *above, const uint16_t *left, int bd";
add_proto qw/void av1_highbd_v_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, int bs, const uint16_t *above, const uint16_t *left, int bd";
add_proto qw/void av1_highbd_h_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, int bs, const uint16_t *above, const uint16_t *left, int bd";
add_proto qw/void av1_highbd_d45_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, int bs, const uint16_t *above, const uint16_t *left, int bd";
add_proto qw/void av1_highbd_d135_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, int bs, const uint16_t *above, const uint16_t *left, int bd";
add_proto qw/void av1_highbd_d117_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, int bs, const uint16_t *above, const uint16_t *left, int bd";
add_proto qw/void av1_highbd_d153_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, int bs, const uint16_t *above, const uint16_t *left, int bd";
add_proto qw/void av1_highbd_d207_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, int bs, const uint16_t *above, const uint16_t *left, int bd";
add_proto qw/void av1_highbd_d63_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, int bs, const uint16_t *above, const uint16_t *left, int bd";
add_proto qw/void av1_highbd_tm_filter_predictor/, "uint16_t *dst, ptrdiff_t stride, int bs, const uint16_t *above, const uint16_t *left, int bd";
}
}
# High bitdepth functions
if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
#
# Sub Pixel Filters
#
add_proto qw/void av1_highbd_convolve_copy/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
add_proto qw/void av1_highbd_convolve_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
add_proto qw/void av1_highbd_convolve8/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
specialize qw/av1_highbd_convolve8/, "$sse2_x86_64";
add_proto qw/void av1_highbd_convolve8_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
specialize qw/av1_highbd_convolve8_horiz/, "$sse2_x86_64";
add_proto qw/void av1_highbd_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
specialize qw/av1_highbd_convolve8_vert/, "$sse2_x86_64";
add_proto qw/void av1_highbd_convolve8_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
specialize qw/av1_highbd_convolve8_avg/, "$sse2_x86_64";
add_proto qw/void av1_highbd_convolve8_avg_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
specialize qw/av1_highbd_convolve8_avg_horiz/, "$sse2_x86_64";
add_proto qw/void av1_highbd_convolve8_avg_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
specialize qw/av1_highbd_convolve8_avg_vert/, "$sse2_x86_64";
#
# dct
#
add_proto qw/void av1_highbd_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
add_proto qw/void av1_highbd_iht4x8_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
add_proto qw/void av1_highbd_iht8x4_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
add_proto qw/void av1_highbd_iht8x16_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
add_proto qw/void av1_highbd_iht16x8_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
add_proto qw/void av1_highbd_iht16x32_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
add_proto qw/void av1_highbd_iht32x16_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
add_proto qw/void av1_highbd_iht4x16_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
add_proto qw/void av1_highbd_iht16x4_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
add_proto qw/void av1_highbd_iht8x32_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
add_proto qw/void av1_highbd_iht32x8_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
add_proto qw/void av1_highbd_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
add_proto qw/void av1_highbd_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type, int bd";
}
if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
#inv txfm
add_proto qw/void av1_inv_txfm2d_add_4x4/, "const int32_t *input, uint16_t *output, int stride, int tx_type, int bd";
specialize qw/av1_inv_txfm2d_add_4x4 sse4_1/;
add_proto qw/void av1_inv_txfm2d_add_8x8/, "const int32_t *input, uint16_t *output, int stride, int tx_type, int bd";
specialize qw/av1_inv_txfm2d_add_8x8 sse4_1/;
add_proto qw/void av1_inv_txfm2d_add_16x16/, "const int32_t *input, uint16_t *output, int stride, int tx_type, int bd";
specialize qw/av1_inv_txfm2d_add_16x16 sse4_1/;
add_proto qw/void av1_inv_txfm2d_add_32x32/, "const int32_t *input, uint16_t *output, int stride, int tx_type, int bd";
specialize qw/av1_inv_txfm2d_add_32x32 avx2/;
add_proto qw/void av1_inv_txfm2d_add_64x64/, "const int32_t *input, uint16_t *output, int stride, int tx_type, int bd";
}
#
# Encoder functions below this point.
#
if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
# ENCODEMB INVOKE
if (aom_config("CONFIG_AOM_QM") eq "yes") {
if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
# the transform coefficients are held in 32-bit
# values, so the assembler code for av1_block_error can no longer be used.
add_proto qw/int64_t av1_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz";
add_proto qw/void av1_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t *iqm_ptr";
add_proto qw/void av1_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t *iqm_ptr";
if (aom_config("CONFIG_TX64X64") eq "yes") {
add_proto qw/void av1_quantize_fp_64x64/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t *iqm_ptr";
}
add_proto qw/void av1_fdct8x8_quant/, "const int16_t *input, int stride, tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t *iqm_ptr";
} else {
add_proto qw/int64_t av1_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz";
specialize qw/av1_block_error avx2 msa/, "$sse2_x86inc";
add_proto qw/int64_t av1_block_error_fp/, "const int16_t *coeff, const int16_t *dqcoeff, int block_size";
specialize qw/av1_block_error_fp neon/, "$sse2_x86inc";
add_proto qw/void av1_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t *iqm_ptr";
add_proto qw/void av1_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t *iqm_ptr";
if (aom_config("CONFIG_TX64X64") eq "yes") {
add_proto qw/void av1_quantize_fp_64x64/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t *iqm_ptr";
}
add_proto qw/void av1_fdct8x8_quant/, "const int16_t *input, int stride, tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t *iqm_ptr";
}
} else {
if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
# the transform coefficients are held in 32-bit
# values, so the assembler code for av1_block_error can no longer be used.
add_proto qw/int64_t av1_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz";
add_proto qw/void av1_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
add_proto qw/void av1_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
if (aom_config("CONFIG_TX64X64") eq "yes") {
add_proto qw/void av1_quantize_fp_64x64/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
}
add_proto qw/void av1_fdct8x8_quant/, "const int16_t *input, int stride, tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
} else {
add_proto qw/int64_t av1_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz";
specialize qw/av1_block_error sse2 avx2 msa/;
add_proto qw/int64_t av1_block_error_fp/, "const int16_t *coeff, const int16_t *dqcoeff, int block_size";
specialize qw/av1_block_error_fp neon sse2/;
add_proto qw/void av1_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/av1_quantize_fp neon sse2/, "$ssse3_x86_64";
add_proto qw/void av1_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/av1_quantize_fp_32x32/, "$ssse3_x86_64";
if (aom_config("CONFIG_TX64X64") eq "yes") {
add_proto qw/void av1_quantize_fp_64x64/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
}
add_proto qw/void av1_fdct8x8_quant/, "const int16_t *input, int stride, tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/av1_fdct8x8_quant sse2 ssse3 neon/;
}
}
# fdct functions
add_proto qw/void av1_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/av1_fht4x4 sse2/;
add_proto qw/void av1_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
add_proto qw/void av1_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/av1_fht8x8 sse2/;
add_proto qw/void av1_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/av1_fht16x16 sse2 avx2/;
add_proto qw/void av1_fht32x32/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/av1_fht32x32 sse2 avx2/;
if (aom_config("CONFIG_TX64X64") eq "yes") {
add_proto qw/void av1_fht64x64/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
}
add_proto qw/void av1_fht4x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/av1_fht4x8 sse2/;
add_proto qw/void av1_fht8x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/av1_fht8x4 sse2/;
add_proto qw/void av1_fht8x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/av1_fht8x16 sse2/;
add_proto qw/void av1_fht16x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/av1_fht16x8 sse2/;
add_proto qw/void av1_fht16x32/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/av1_fht16x32 sse2/;
add_proto qw/void av1_fht32x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/av1_fht32x16 sse2/;
add_proto qw/void av1_fht4x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
add_proto qw/void av1_fht16x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
add_proto qw/void av1_fht8x32/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
add_proto qw/void av1_fht32x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
if (aom_config("CONFIG_HIGHBITDEPTH") ne "yes") {
if (aom_config("CONFIG_EXT_TX") ne "yes") {
specialize qw/av1_fht4x4 msa/;
specialize qw/av1_fht8x8 msa/;
specialize qw/av1_fht16x16 msa/;
}
}
add_proto qw/void av1_fwd_idtx/, "const int16_t *src_diff, tran_low_t *coeff, int stride, int bs, int tx_type";
if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
#fwd txfm
add_proto qw/void av1_fwd_txfm2d_4x4/, "const int16_t *input, int32_t *output, int stride, int tx_type, int bd";
specialize qw/av1_fwd_txfm2d_4x4 sse4_1/;
add_proto qw/void av1_fwd_txfm2d_8x8/, "const int16_t *input, int32_t *output, int stride, int tx_type, int bd";
specialize qw/av1_fwd_txfm2d_8x8 sse4_1/;
add_proto qw/void av1_fwd_txfm2d_16x16/, "const int16_t *input, int32_t *output, int stride, int tx_type, int bd";
specialize qw/av1_fwd_txfm2d_16x16 sse4_1/;
add_proto qw/void av1_fwd_txfm2d_32x32/, "const int16_t *input, int32_t *output, int stride, int tx_type, int bd";
specialize qw/av1_fwd_txfm2d_32x32 sse4_1/;
add_proto qw/void av1_fwd_txfm2d_64x64/, "const int16_t *input, int32_t *output, int stride, int tx_type, int bd";
specialize qw/av1_fwd_txfm2d_64x64 sse4_1/;
}
#
# Motion search
#
add_proto qw/int av1_full_search_sad/, "const struct macroblock *x, const struct mv *ref_mv, int sad_per_bit, int distance, const struct aom_variance_vtable *fn_ptr, const struct mv *center_mv, struct mv *best_mv";
specialize qw/av1_full_search_sad sse3 sse4_1/;
$av1_full_search_sad_sse3=av1_full_search_sadx3;
$av1_full_search_sad_sse4_1=av1_full_search_sadx8;
add_proto qw/int av1_diamond_search_sad/, "struct macroblock *x, const struct search_site_config *cfg, struct mv *ref_mv, struct mv *best_mv, int search_param, int sad_per_bit, int *num00, const struct aom_variance_vtable *fn_ptr, const struct mv *center_mv";
add_proto qw/int av1_full_range_search/, "const struct macroblock *x, const struct search_site_config *cfg, struct mv *ref_mv, struct mv *best_mv, int search_param, int sad_per_bit, int *num00, const struct aom_variance_vtable *fn_ptr, const struct mv *center_mv";
add_proto qw/void av1_temporal_filter_apply/, "uint8_t *frame1, unsigned int stride, uint8_t *frame2, unsigned int block_width, unsigned int block_height, int strength, int filter_weight, unsigned int *accumulator, uint16_t *count";
specialize qw/av1_temporal_filter_apply sse2 msa/;
if (aom_config("CONFIG_AOM_QM") eq "yes") {
add_proto qw/void av1_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr, int log_scale";
} else {
add_proto qw/void av1_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, int log_scale";
}
if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
# ENCODEMB INVOKE
if (aom_config("CONFIG_NEW_QUANT") eq "yes") {
add_proto qw/void highbd_quantize_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
add_proto qw/void highbd_quantize_fp_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
add_proto qw/void highbd_quantize_32x32_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
add_proto qw/void highbd_quantize_32x32_fp_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
if (aom_config("CONFIG_TX64X64") eq "yes") {
add_proto qw/void highbd_quantize_64x64_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
add_proto qw/void highbd_quantize_64x64_fp_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
}
}
add_proto qw/int64_t av1_highbd_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz, int bd";
specialize qw/av1_highbd_block_error sse2/;
if (aom_config("CONFIG_AOM_QM") eq "yes") {
add_proto qw/void av1_highbd_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr, int log_scale";
add_proto qw/void av1_highbd_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr, int log_scale";
if (aom_config("CONFIG_TX64X64") eq "yes") {
add_proto qw/void av1_highbd_quantize_fp_64x64/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr, int log_scale";
}
add_proto qw/void av1_highbd_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr, int log_scale";
} else {
add_proto qw/void av1_highbd_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, int log_scale";
specialize qw/av1_highbd_quantize_fp sse4_1/;
add_proto qw/void av1_highbd_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, int log_scale";
}
# fdct functions
add_proto qw/void av1_highbd_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/av1_highbd_fht4x4 sse4_1/;
add_proto qw/void av1_highbd_fht4x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
add_proto qw/void av1_highbd_fht8x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
add_proto qw/void av1_highbd_fht8x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
add_proto qw/void av1_highbd_fht16x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
add_proto qw/void av1_highbd_fht16x32/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
add_proto qw/void av1_highbd_fht32x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
add_proto qw/void av1_highbd_fht4x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
add_proto qw/void av1_highbd_fht16x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
add_proto qw/void av1_highbd_fht8x32/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
add_proto qw/void av1_highbd_fht32x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
add_proto qw/void av1_highbd_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
add_proto qw/void av1_highbd_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
add_proto qw/void av1_highbd_fht32x32/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
if (aom_config("CONFIG_TX64X64") eq "yes") {
add_proto qw/void av1_highbd_fht64x64/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
}
add_proto qw/void av1_highbd_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
add_proto qw/void av1_highbd_temporal_filter_apply/, "uint8_t *frame1, unsigned int stride, uint8_t *frame2, unsigned int block_width, unsigned int block_height, int strength, int filter_weight, unsigned int *accumulator, uint16_t *count";
}
# End av1_high encoder functions
if (aom_config("CONFIG_EXT_INTER") eq "yes") {
add_proto qw/uint64_t av1_wedge_sse_from_residuals/, "const int16_t *r1, const int16_t *d, const uint8_t *m, int N";
specialize qw/av1_wedge_sse_from_residuals sse2/;
add_proto qw/int av1_wedge_sign_from_residuals/, "const int16_t *ds, const uint8_t *m, int N, int64_t limit";
specialize qw/av1_wedge_sign_from_residuals sse2/;
add_proto qw/void av1_wedge_compute_delta_squares/, "int16_t *d, const int16_t *a, const int16_t *b, int N";
specialize qw/av1_wedge_compute_delta_squares sse2/;
}
}
# end encoder functions
# If PVQ is enabled, fwd transforms are required by decoder
if (aom_config("CONFIG_PVQ") eq "yes") {
# fdct functions
if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
add_proto qw/void av1_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/av1_fht4x4 sse2/;
add_proto qw/void av1_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/av1_fht8x8 sse2/;
add_proto qw/void av1_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/av1_fht16x16 sse2/;
add_proto qw/void av1_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
specialize qw/av1_fwht4x4 sse2/;
} else {
add_proto qw/void av1_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/av1_fht4x4 sse2 msa/;
add_proto qw/void av1_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/av1_fht8x8 sse2 msa/;
add_proto qw/void av1_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/av1_fht16x16 sse2 msa/;
add_proto qw/void av1_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
specialize qw/av1_fwht4x4 msa sse2/;
}
}
# Deringing Functions
if (aom_config("CONFIG_CDEF") eq "yes") {
add_proto qw/void aom_clpf_block_hbd/, "uint16_t *dst, const uint16_t *src, int dstride, int sstride, int sizex, int sizey, unsigned int strength, unsigned int bd";
add_proto qw/void aom_clpf_hblock_hbd/, "uint16_t *dst, const uint16_t *src, int dstride, int sstride, int sizex, int sizey, unsigned int strength, unsigned int bd";
add_proto qw/void aom_clpf_block/, "uint8_t *dst, const uint16_t *src, int dstride, int sstride, int sizex, int sizey, unsigned int strength, unsigned int bd";
add_proto qw/void aom_clpf_hblock/, "uint8_t *dst, const uint16_t *src, int dstride, int sstride, int sizex, int sizey, unsigned int strength, unsigned int bd";
add_proto qw/int od_dir_find8/, "const od_dering_in *img, int stride, int32_t *var, int coeff_shift";
add_proto qw/void od_filter_dering_direction_4x4/, "uint16_t *y, int ystride, const uint16_t *in, int threshold, int dir, int damping";
add_proto qw/void od_filter_dering_direction_8x8/, "uint16_t *y, int ystride, const uint16_t *in, int threshold, int dir, int damping";
add_proto qw/void copy_8x8_16bit_to_8bit/, "uint8_t *dst, int dstride, const uint16_t *src, int sstride";
add_proto qw/void copy_4x4_16bit_to_8bit/, "uint8_t *dst, int dstride, const uint16_t *src, int sstride";
add_proto qw/void copy_8x8_16bit_to_16bit/, "uint16_t *dst, int dstride, const uint16_t *src, int sstride";
add_proto qw/void copy_4x4_16bit_to_16bit/, "uint16_t *dst, int dstride, const uint16_t *src, int sstride";
add_proto qw/void copy_rect8_8bit_to_16bit/, "uint16_t *dst, int dstride, const uint8_t *src, int sstride, int v, int h";
add_proto qw/void copy_rect8_16bit_to_16bit/, "uint16_t *dst, int dstride, const uint16_t *src, int sstride, int v, int h";
# VS compiling for 32 bit targets does not support vector types in
# structs as arguments, which makes the v256 type of the intrinsics
# hard to support, so optimizations for this target are disabled.
if ($opts{config} !~ /libs-x86-win32-vs.*/) {
specialize qw/aom_clpf_block_hbd sse2 ssse3 sse4_1 neon/;
specialize qw/aom_clpf_hblock_hbd sse2 ssse3 sse4_1 neon/;
specialize qw/aom_clpf_block sse2 ssse3 sse4_1 neon/;
specialize qw/aom_clpf_hblock sse2 ssse3 sse4_1 neon/;
specialize qw/od_dir_find8 sse2 ssse3 sse4_1 neon/;
specialize qw/od_filter_dering_direction_4x4 sse2 ssse3 sse4_1 neon/;
specialize qw/od_filter_dering_direction_8x8 sse2 ssse3 sse4_1 neon/;
specialize qw/copy_8x8_16bit_to_8bit sse2 ssse3 sse4_1 neon/;
specialize qw/copy_4x4_16bit_to_8bit sse2 ssse3 sse4_1 neon/;
specialize qw/copy_8x8_16bit_to_16bit sse2 ssse3 sse4_1 neon/;
specialize qw/copy_4x4_16bit_to_16bit sse2 ssse3 sse4_1 neon/;
specialize qw/copy_rect8_8bit_to_16bit sse2 ssse3 sse4_1 neon/;
specialize qw/copy_rect8_16bit_to_16bit sse2 ssse3 sse4_1 neon/;
}
}
# PVQ Functions
if (aom_config("CONFIG_PVQ") eq "yes") {
add_proto qw/double pvq_search_rdo_double/, "const od_val16 *xcoeff, int n, int k, int *ypulse, double g2, double pvq_norm_lambda, int prev_k";
specialize qw/pvq_search_rdo_double sse4_1/;
}
# WARPED_MOTION / GLOBAL_MOTION functions
if ((aom_config("CONFIG_WARPED_MOTION") eq "yes") ||
(aom_config("CONFIG_GLOBAL_MOTION") eq "yes")) {
add_proto qw/void av1_warp_affine/, "int32_t *mat, uint8_t *ref, int width, int height, int stride, uint8_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int ref_frm, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta";
specialize qw/av1_warp_affine sse2/;
if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
add_proto qw/void av1_highbd_warp_affine/, "int32_t *mat, uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, int ref_frm, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta";
specialize qw/av1_highbd_warp_affine ssse3/;
}
}
# LOOP_RESTORATION functions
if (aom_config("CONFIG_LOOP_RESTORATION") eq "yes") {
add_proto qw/void apply_selfguided_restoration/, "uint8_t *dat, int width, int height, int stride, int eps, int *xqd, uint8_t *dst, int dst_stride, int32_t *tmpbuf";
specialize qw/apply_selfguided_restoration sse4_1/;
add_proto qw/void av1_selfguided_restoration/, "uint8_t *dgd, int width, int height, int stride, int32_t *dst, int dst_stride, int r, int eps, int32_t *tmpbuf";
specialize qw/av1_selfguided_restoration sse4_1/;
add_proto qw/void av1_highpass_filter/, "uint8_t *dgd, int width, int height, int stride, int32_t *dst, int dst_stride, int r, int eps";
specialize qw/av1_highpass_filter sse4_1/;
if (aom_config("CONFIG_HIGHBITDEPTH") eq "yes") {
add_proto qw/void apply_selfguided_restoration_highbd/, "uint16_t *dat, int width, int height, int stride, int bit_depth, int eps, int *xqd, uint16_t *dst, int dst_stride, int32_t *tmpbuf";
specialize qw/apply_selfguided_restoration_highbd sse4_1/;
add_proto qw/void av1_selfguided_restoration_highbd/, "uint16_t *dgd, int width, int height, int stride, int32_t *dst, int dst_stride, int bit_depth, int r, int eps, int32_t *tmpbuf";
specialize qw/av1_selfguided_restoration_highbd sse4_1/;
add_proto qw/void av1_highpass_filter_highbd/, "uint16_t *dgd, int width, int height, int stride, int32_t *dst, int dst_stride, int r, int eps";
specialize qw/av1_highpass_filter_highbd sse4_1/;
}
}
1;

211
third_party/aom/av1/common/av1_txfm.h vendored Normal file
View file

@ -0,0 +1,211 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_TXFM_H_
#define AV1_TXFM_H_
#include <assert.h>
#include <math.h>
#include <stdio.h>
#include "av1/common/enums.h"
#include "aom/aom_integer.h"
#include "aom_dsp/aom_dsp_common.h"
static const int cos_bit_min = 10;
static const int cos_bit_max = 16;
// cospi_arr[i][j] = (int)round(cos(M_PI*j/128) * (1<<(cos_bit_min+i)));
static const int32_t cospi_arr[7][64] = {
{ 1024, 1024, 1023, 1021, 1019, 1016, 1013, 1009, 1004, 999, 993, 987, 980,
972, 964, 955, 946, 936, 926, 915, 903, 891, 878, 865, 851, 837,
822, 807, 792, 775, 759, 742, 724, 706, 688, 669, 650, 630, 610,
590, 569, 548, 526, 505, 483, 460, 438, 415, 392, 369, 345, 321,
297, 273, 249, 224, 200, 175, 150, 125, 100, 75, 50, 25 },
{ 2048, 2047, 2046, 2042, 2038, 2033, 2026, 2018, 2009, 1998, 1987,
1974, 1960, 1945, 1928, 1911, 1892, 1872, 1851, 1829, 1806, 1782,
1757, 1730, 1703, 1674, 1645, 1615, 1583, 1551, 1517, 1483, 1448,
1412, 1375, 1338, 1299, 1260, 1220, 1179, 1138, 1096, 1053, 1009,
965, 921, 876, 830, 784, 737, 690, 642, 595, 546, 498,
449, 400, 350, 301, 251, 201, 151, 100, 50 },
{ 4096, 4095, 4091, 4085, 4076, 4065, 4052, 4036, 4017, 3996, 3973,
3948, 3920, 3889, 3857, 3822, 3784, 3745, 3703, 3659, 3612, 3564,
3513, 3461, 3406, 3349, 3290, 3229, 3166, 3102, 3035, 2967, 2896,
2824, 2751, 2675, 2598, 2520, 2440, 2359, 2276, 2191, 2106, 2019,
1931, 1842, 1751, 1660, 1567, 1474, 1380, 1285, 1189, 1092, 995,
897, 799, 700, 601, 501, 401, 301, 201, 101 },
{ 8192, 8190, 8182, 8170, 8153, 8130, 8103, 8071, 8035, 7993, 7946,
7895, 7839, 7779, 7713, 7643, 7568, 7489, 7405, 7317, 7225, 7128,
7027, 6921, 6811, 6698, 6580, 6458, 6333, 6203, 6070, 5933, 5793,
5649, 5501, 5351, 5197, 5040, 4880, 4717, 4551, 4383, 4212, 4038,
3862, 3683, 3503, 3320, 3135, 2948, 2760, 2570, 2378, 2185, 1990,
1795, 1598, 1401, 1202, 1003, 803, 603, 402, 201 },
{ 16384, 16379, 16364, 16340, 16305, 16261, 16207, 16143, 16069, 15986, 15893,
15791, 15679, 15557, 15426, 15286, 15137, 14978, 14811, 14635, 14449, 14256,
14053, 13842, 13623, 13395, 13160, 12916, 12665, 12406, 12140, 11866, 11585,
11297, 11003, 10702, 10394, 10080, 9760, 9434, 9102, 8765, 8423, 8076,
7723, 7366, 7005, 6639, 6270, 5897, 5520, 5139, 4756, 4370, 3981,
3590, 3196, 2801, 2404, 2006, 1606, 1205, 804, 402 },
{ 32768, 32758, 32729, 32679, 32610, 32522, 32413, 32286, 32138, 31972, 31786,
31581, 31357, 31114, 30853, 30572, 30274, 29957, 29622, 29269, 28899, 28511,
28106, 27684, 27246, 26791, 26320, 25833, 25330, 24812, 24279, 23732, 23170,
22595, 22006, 21403, 20788, 20160, 19520, 18868, 18205, 17531, 16846, 16151,
15447, 14733, 14010, 13279, 12540, 11793, 11039, 10279, 9512, 8740, 7962,
7180, 6393, 5602, 4808, 4011, 3212, 2411, 1608, 804 },
{ 65536, 65516, 65457, 65358, 65220, 65043, 64827, 64571, 64277, 63944, 63572,
63162, 62714, 62228, 61705, 61145, 60547, 59914, 59244, 58538, 57798, 57022,
56212, 55368, 54491, 53581, 52639, 51665, 50660, 49624, 48559, 47464, 46341,
45190, 44011, 42806, 41576, 40320, 39040, 37736, 36410, 35062, 33692, 32303,
30893, 29466, 28020, 26558, 25080, 23586, 22078, 20557, 19024, 17479, 15924,
14359, 12785, 11204, 9616, 8022, 6424, 4821, 3216, 1608 }
};
static INLINE int32_t round_shift(int32_t value, int bit) {
assert(bit >= 1);
return (value + (1 << (bit - 1))) >> bit;
}
static INLINE void round_shift_array(int32_t *arr, int size, int bit) {
int i;
if (bit == 0) {
return;
} else {
if (bit > 0) {
for (i = 0; i < size; i++) {
arr[i] = round_shift(arr[i], bit);
}
} else {
for (i = 0; i < size; i++) {
arr[i] = arr[i] * (1 << (-bit));
}
}
}
}
static INLINE int32_t half_btf(int32_t w0, int32_t in0, int32_t w1, int32_t in1,
int bit) {
int32_t result_32 = w0 * in0 + w1 * in1;
#if CONFIG_COEFFICIENT_RANGE_CHECKING
int64_t result_64 = (int64_t)w0 * (int64_t)in0 + (int64_t)w1 * (int64_t)in1;
if (result_64 < INT32_MIN || result_64 > INT32_MAX) {
printf("%s %d overflow result_32: %d result_64: %" PRId64
" w0: %d in0: %d w1: %d in1: "
"%d\n",
__FILE__, __LINE__, result_32, result_64, w0, in0, w1, in1);
assert(0 && "half_btf overflow");
}
#endif
return round_shift(result_32, bit);
}
static INLINE int get_max_bit(int x) {
int max_bit = -1;
while (x) {
x = x >> 1;
max_bit++;
}
return max_bit;
}
// TODO(angiebird): implement SSE
static INLINE void clamp_block(int16_t *block, int block_size, int stride,
int low, int high) {
int i, j;
for (i = 0; i < block_size; ++i) {
for (j = 0; j < block_size; ++j) {
block[i * stride + j] = clamp(block[i * stride + j], low, high);
}
}
}
typedef void (*TxfmFunc)(const int32_t *input, int32_t *output,
const int8_t *cos_bit, const int8_t *stage_range);
typedef enum TXFM_TYPE {
TXFM_TYPE_DCT4,
TXFM_TYPE_DCT8,
TXFM_TYPE_DCT16,
TXFM_TYPE_DCT32,
TXFM_TYPE_DCT64,
TXFM_TYPE_ADST4,
TXFM_TYPE_ADST8,
TXFM_TYPE_ADST16,
TXFM_TYPE_ADST32,
} TXFM_TYPE;
typedef struct TXFM_2D_CFG {
const int txfm_size;
const int stage_num_col;
const int stage_num_row;
const int8_t *shift;
const int8_t *stage_range_col;
const int8_t *stage_range_row;
const int8_t *cos_bit_col;
const int8_t *cos_bit_row;
const TXFM_TYPE txfm_type_col;
const TXFM_TYPE txfm_type_row;
} TXFM_2D_CFG;
typedef struct TXFM_2D_FLIP_CFG {
int ud_flip; // flip upside down
int lr_flip; // flip left to right
const TXFM_2D_CFG *cfg;
} TXFM_2D_FLIP_CFG;
static INLINE void set_flip_cfg(int tx_type, TXFM_2D_FLIP_CFG *cfg) {
switch (tx_type) {
case DCT_DCT:
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
cfg->ud_flip = 0;
cfg->lr_flip = 0;
break;
#if CONFIG_EXT_TX
case FLIPADST_DCT:
cfg->ud_flip = 1;
cfg->lr_flip = 0;
break;
case DCT_FLIPADST:
cfg->ud_flip = 0;
cfg->lr_flip = 1;
break;
case FLIPADST_FLIPADST:
cfg->ud_flip = 1;
cfg->lr_flip = 1;
break;
case ADST_FLIPADST:
cfg->ud_flip = 0;
cfg->lr_flip = 1;
break;
case FLIPADST_ADST:
cfg->ud_flip = 1;
cfg->lr_flip = 0;
break;
#endif // CONFIG_EXT_TX
default:
cfg->ud_flip = 0;
cfg->lr_flip = 0;
assert(0);
}
}
#ifdef __cplusplus
extern "C" {
#endif
TXFM_2D_FLIP_CFG av1_get_fwd_txfm_cfg(int tx_type, int tx_size);
TXFM_2D_FLIP_CFG av1_get_fwd_txfm_64x64_cfg(int tx_type);
#ifdef __cplusplus
}
#endif // __cplusplus
#endif // AV1_TXFM_H_

296
third_party/aom/av1/common/blockd.c vendored Normal file
View file

@ -0,0 +1,296 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <math.h>
#include "aom_ports/system_state.h"
#include "av1/common/blockd.h"
#include "av1/common/onyxc_int.h"
PREDICTION_MODE av1_left_block_mode(const MODE_INFO *cur_mi,
const MODE_INFO *left_mi, int b) {
if (b == 0 || b == 2) {
if (!left_mi || is_inter_block(&left_mi->mbmi)) return DC_PRED;
return get_y_mode(left_mi, b + 1);
} else {
assert(b == 1 || b == 3);
return cur_mi->bmi[b - 1].as_mode;
}
}
PREDICTION_MODE av1_above_block_mode(const MODE_INFO *cur_mi,
const MODE_INFO *above_mi, int b) {
if (b == 0 || b == 1) {
if (!above_mi || is_inter_block(&above_mi->mbmi)) return DC_PRED;
return get_y_mode(above_mi, b + 2);
} else {
assert(b == 2 || b == 3);
return cur_mi->bmi[b - 2].as_mode;
}
}
#if CONFIG_COEF_INTERLEAVE
void av1_foreach_transformed_block_interleave(
const MACROBLOCKD *const xd, BLOCK_SIZE bsize,
foreach_transformed_block_visitor visit, void *arg) {
const struct macroblockd_plane *const pd_y = &xd->plane[0];
const struct macroblockd_plane *const pd_c = &xd->plane[1];
const MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
const TX_SIZE tx_log2_y = mbmi->tx_size;
const TX_SIZE tx_log2_c = get_uv_tx_size(mbmi, pd_c);
const int tx_sz_y = (1 << tx_log2_y);
const int tx_sz_c = (1 << tx_log2_c);
const BLOCK_SIZE plane_bsize_y = get_plane_block_size(bsize, pd_y);
const BLOCK_SIZE plane_bsize_c = get_plane_block_size(bsize, pd_c);
const int num_4x4_w_y = num_4x4_blocks_wide_lookup[plane_bsize_y];
const int num_4x4_w_c = num_4x4_blocks_wide_lookup[plane_bsize_c];
const int num_4x4_h_y = num_4x4_blocks_high_lookup[plane_bsize_y];
const int num_4x4_h_c = num_4x4_blocks_high_lookup[plane_bsize_c];
const int step_y = 1 << (tx_log2_y << 1);
const int step_c = 1 << (tx_log2_c << 1);
const int max_4x4_w_y =
get_max_4x4_size(num_4x4_w_y, xd->mb_to_right_edge, pd_y->subsampling_x);
const int max_4x4_h_y =
get_max_4x4_size(num_4x4_h_y, xd->mb_to_bottom_edge, pd_y->subsampling_y);
const int extra_step_y = ((num_4x4_w_y - max_4x4_w_y) >> tx_log2_y) * step_y;
const int max_4x4_w_c =
get_max_4x4_size(num_4x4_w_c, xd->mb_to_right_edge, pd_c->subsampling_x);
const int max_4x4_h_c =
get_max_4x4_size(num_4x4_h_c, xd->mb_to_bottom_edge, pd_c->subsampling_y);
const int extra_step_c = ((num_4x4_w_c - max_4x4_w_c) >> tx_log2_c) * step_c;
// The max_4x4_w/h may be smaller than tx_sz under some corner cases,
// i.e. when the SB is splitted by tile boundaries.
const int tu_num_w_y = (max_4x4_w_y + tx_sz_y - 1) / tx_sz_y;
const int tu_num_h_y = (max_4x4_h_y + tx_sz_y - 1) / tx_sz_y;
const int tu_num_w_c = (max_4x4_w_c + tx_sz_c - 1) / tx_sz_c;
const int tu_num_h_c = (max_4x4_h_c + tx_sz_c - 1) / tx_sz_c;
const int tu_num_c = tu_num_w_c * tu_num_h_c;
int tu_idx_c = 0;
int offset_y, row_y, col_y;
int offset_c, row_c, col_c;
for (row_y = 0; row_y < tu_num_h_y; row_y++) {
for (col_y = 0; col_y < tu_num_w_y; col_y++) {
// luma
offset_y = (row_y * tu_num_w_y + col_y) * step_y + row_y * extra_step_y;
visit(0, offset_y, row_y * tx_sz_y, col_y * tx_sz_y, plane_bsize_y,
tx_log2_y, arg);
// chroma
if (tu_idx_c < tu_num_c) {
row_c = (tu_idx_c / tu_num_w_c) * tx_sz_c;
col_c = (tu_idx_c % tu_num_w_c) * tx_sz_c;
offset_c = tu_idx_c * step_c + (tu_idx_c / tu_num_w_c) * extra_step_c;
visit(1, offset_c, row_c, col_c, plane_bsize_c, tx_log2_c, arg);
visit(2, offset_c, row_c, col_c, plane_bsize_c, tx_log2_c, arg);
tu_idx_c++;
}
}
}
// In 422 case, it's possible that Chroma has more TUs than Luma
while (tu_idx_c < tu_num_c) {
row_c = (tu_idx_c / tu_num_w_c) * tx_sz_c;
col_c = (tu_idx_c % tu_num_w_c) * tx_sz_c;
offset_c = tu_idx_c * step_c + row_c * extra_step_c;
visit(1, offset_c, row_c, col_c, plane_bsize_c, tx_log2_c, arg);
visit(2, offset_c, row_c, col_c, plane_bsize_c, tx_log2_c, arg);
tu_idx_c++;
}
}
#endif
void av1_foreach_transformed_block_in_plane(
const MACROBLOCKD *const xd, BLOCK_SIZE bsize, int plane,
foreach_transformed_block_visitor visit, void *arg) {
const struct macroblockd_plane *const pd = &xd->plane[plane];
// block and transform sizes, in number of 4x4 blocks log 2 ("*_b")
// 4x4=0, 8x8=2, 16x16=4, 32x32=6, 64x64=8
// transform size varies per plane, look it up in a common way.
const TX_SIZE tx_size = get_tx_size(plane, xd);
#if CONFIG_CB4X4 && !CONFIG_CHROMA_2X2
const BLOCK_SIZE plane_bsize =
AOMMAX(BLOCK_4X4, get_plane_block_size(bsize, pd));
#else
const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
#endif
const uint8_t txw_unit = tx_size_wide_unit[tx_size];
const uint8_t txh_unit = tx_size_high_unit[tx_size];
const int step = txw_unit * txh_unit;
int i = 0, r, c;
// If mb_to_right_edge is < 0 we are in a situation in which
// the current block size extends into the UMV and we won't
// visit the sub blocks that are wholly within the UMV.
const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
// Keep track of the row and column of the blocks we use so that we know
// if we are in the unrestricted motion border.
for (r = 0; r < max_blocks_high; r += txh_unit) {
// Skip visiting the sub blocks that are wholly within the UMV.
for (c = 0; c < max_blocks_wide; c += txw_unit) {
visit(plane, i, r, c, plane_bsize, tx_size, arg);
i += step;
}
}
}
#if CONFIG_LV_MAP
void av1_foreach_transformed_block(const MACROBLOCKD *const xd,
BLOCK_SIZE bsize, int mi_row, int mi_col,
foreach_transformed_block_visitor visit,
void *arg) {
int plane;
for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
#if CONFIG_CB4X4
if (!is_chroma_reference(mi_row, mi_col, bsize,
xd->plane[plane].subsampling_x,
xd->plane[plane].subsampling_y))
continue;
#else
(void)mi_row;
(void)mi_col;
#endif
av1_foreach_transformed_block_in_plane(xd, bsize, plane, visit, arg);
}
}
#endif
#if CONFIG_DAALA_DIST
void av1_foreach_8x8_transformed_block_in_plane(
const MACROBLOCKD *const xd, BLOCK_SIZE bsize, int plane,
foreach_transformed_block_visitor visit,
foreach_transformed_block_visitor mi_visit, void *arg) {
const struct macroblockd_plane *const pd = &xd->plane[plane];
// block and transform sizes, in number of 4x4 blocks log 2 ("*_b")
// 4x4=0, 8x8=2, 16x16=4, 32x32=6, 64x64=8
// transform size varies per plane, look it up in a common way.
const TX_SIZE tx_size = get_tx_size(plane, xd);
const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
const uint8_t txw_unit = tx_size_wide_unit[tx_size];
const uint8_t txh_unit = tx_size_high_unit[tx_size];
const int step = txw_unit * txh_unit;
int i = 0, r, c;
// If mb_to_right_edge is < 0 we are in a situation in which
// the current block size extends into the UMV and we won't
// visit the sub blocks that are wholly within the UMV.
const int max_blocks_wide = max_block_wide(xd, plane_bsize, plane);
const int max_blocks_high = max_block_high(xd, plane_bsize, plane);
// Keep track of the row and column of the blocks we use so that we know
// if we are in the unrestricted motion border.
for (r = 0; r < max_blocks_high; r += txh_unit) {
// Skip visiting the sub blocks that are wholly within the UMV.
for (c = 0; c < max_blocks_wide; c += txw_unit) {
visit(plane, i, r, c, plane_bsize, tx_size, arg);
// Call whenever each 8x8 block is done
if ((r & 1) && (c & 1))
mi_visit(plane, i, r - 1, c - 1, plane_bsize, TX_8X8, arg);
i += step;
}
}
}
#endif
#if !CONFIG_PVQ || CONFIG_VAR_TX
void av1_set_contexts(const MACROBLOCKD *xd, struct macroblockd_plane *pd,
int plane, TX_SIZE tx_size, int has_eob, int aoff,
int loff) {
ENTROPY_CONTEXT *const a = pd->above_context + aoff;
ENTROPY_CONTEXT *const l = pd->left_context + loff;
const int txs_wide = tx_size_wide_unit[tx_size];
const int txs_high = tx_size_high_unit[tx_size];
#if CONFIG_CB4X4
const BLOCK_SIZE bsize = xd->mi[0]->mbmi.sb_type;
#else
const BLOCK_SIZE bsize = AOMMAX(xd->mi[0]->mbmi.sb_type, BLOCK_8X8);
#endif
const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
// above
if (has_eob && xd->mb_to_right_edge < 0) {
int i;
const int blocks_wide = max_block_wide(xd, plane_bsize, plane);
int above_contexts = txs_wide;
if (above_contexts + aoff > blocks_wide)
above_contexts = blocks_wide - aoff;
for (i = 0; i < above_contexts; ++i) a[i] = has_eob;
for (i = above_contexts; i < txs_wide; ++i) a[i] = 0;
} else {
memset(a, has_eob, sizeof(ENTROPY_CONTEXT) * txs_wide);
}
// left
if (has_eob && xd->mb_to_bottom_edge < 0) {
int i;
const int blocks_high = max_block_high(xd, plane_bsize, plane);
int left_contexts = txs_high;
if (left_contexts + loff > blocks_high) left_contexts = blocks_high - loff;
for (i = 0; i < left_contexts; ++i) l[i] = has_eob;
for (i = left_contexts; i < txs_high; ++i) l[i] = 0;
} else {
memset(l, has_eob, sizeof(ENTROPY_CONTEXT) * txs_high);
}
}
#endif
void av1_setup_block_planes(MACROBLOCKD *xd, int ss_x, int ss_y) {
int i;
for (i = 0; i < MAX_MB_PLANE; i++) {
xd->plane[i].plane_type = get_plane_type(i);
xd->plane[i].subsampling_x = i ? ss_x : 0;
xd->plane[i].subsampling_y = i ? ss_y : 0;
}
}
#if CONFIG_EXT_INTRA
const int16_t dr_intra_derivative[90] = {
1, 14666, 7330, 4884, 3660, 2926, 2435, 2084, 1821, 1616, 1451, 1317, 1204,
1108, 1026, 955, 892, 837, 787, 743, 703, 666, 633, 603, 574, 548,
524, 502, 481, 461, 443, 426, 409, 394, 379, 365, 352, 339, 327,
316, 305, 294, 284, 274, 265, 256, 247, 238, 230, 222, 214, 207,
200, 192, 185, 179, 172, 166, 159, 153, 147, 141, 136, 130, 124,
119, 113, 108, 103, 98, 93, 88, 83, 78, 73, 68, 63, 59,
54, 49, 45, 40, 35, 31, 26, 22, 17, 13, 8, 4,
};
#if CONFIG_INTRA_INTERP
int av1_is_intra_filter_switchable(int angle) {
assert(angle > 0 && angle < 270);
if (angle % 45 == 0) return 0;
if (angle > 90 && angle < 180) {
return 1;
} else {
return ((angle < 90 ? dr_intra_derivative[angle]
: dr_intra_derivative[270 - angle]) &
0xFF) > 0;
}
}
#endif // CONFIG_INTRA_INTERP
#endif // CONFIG_EXT_INTRA

1371
third_party/aom/av1/common/blockd.h vendored Normal file
View file

File diff suppressed because it is too large Load diff

445
third_party/aom/av1/common/cdef.c vendored Normal file
View file

@ -0,0 +1,445 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <assert.h>
#include <math.h>
#include <string.h>
#include "./aom_scale_rtcd.h"
#include "aom/aom_integer.h"
#include "av1/common/cdef.h"
#include "av1/common/od_dering.h"
#include "av1/common/onyxc_int.h"
#include "av1/common/reconinter.h"
int sb_all_skip(const AV1_COMMON *const cm, int mi_row, int mi_col) {
int r, c;
int maxc, maxr;
int skip = 1;
maxc = cm->mi_cols - mi_col;
maxr = cm->mi_rows - mi_row;
#if CONFIG_EXT_PARTITION
if (maxr > cm->mib_size_log2) maxr = cm->mib_size_log2;
if (maxc > cm->mib_size_log2) maxc = cm->mib_size_log2;
#else
if (maxr > MAX_MIB_SIZE) maxr = MAX_MIB_SIZE;
if (maxc > MAX_MIB_SIZE) maxc = MAX_MIB_SIZE;
#endif
for (r = 0; r < maxr; r++) {
for (c = 0; c < maxc; c++) {
skip = skip &&
cm->mi_grid_visible[(mi_row + r) * cm->mi_stride + mi_col + c]
->mbmi.skip;
}
}
return skip;
}
static int is_8x8_block_skip(MODE_INFO **grid, int mi_row, int mi_col,
int mi_stride) {
int is_skip = 1;
for (int r = 0; r < mi_size_high[BLOCK_8X8]; ++r)
for (int c = 0; c < mi_size_wide[BLOCK_8X8]; ++c)
is_skip &= grid[(mi_row + r) * mi_stride + (mi_col + c)]->mbmi.skip;
return is_skip;
}
int sb_compute_dering_list(const AV1_COMMON *const cm, int mi_row, int mi_col,
dering_list *dlist, int filter_skip) {
int r, c;
int maxc, maxr;
MODE_INFO **grid;
int count = 0;
grid = cm->mi_grid_visible;
maxc = cm->mi_cols - mi_col;
maxr = cm->mi_rows - mi_row;
#if CONFIG_EXT_PARTITION
if (maxr > cm->mib_size_log2) maxr = cm->mib_size_log2;
if (maxc > cm->mib_size_log2) maxc = cm->mib_size_log2;
#else
if (maxr > MAX_MIB_SIZE) maxr = MAX_MIB_SIZE;
if (maxc > MAX_MIB_SIZE) maxc = MAX_MIB_SIZE;
#endif
const int r_step = mi_size_high[BLOCK_8X8];
const int c_step = mi_size_wide[BLOCK_8X8];
const int r_shift = (r_step == 2);
const int c_shift = (c_step == 2);
assert(r_step == 1 || r_step == 2);
assert(c_step == 1 || c_step == 2);
if (filter_skip) {
for (r = 0; r < maxr; r += r_step) {
for (c = 0; c < maxc; c += c_step) {
dlist[count].by = r >> r_shift;
dlist[count].bx = c >> c_shift;
dlist[count].skip =
is_8x8_block_skip(grid, mi_row + r, mi_col + c, cm->mi_stride);
count++;
}
}
} else {
for (r = 0; r < maxr; r += r_step) {
for (c = 0; c < maxc; c += c_step) {
if (!is_8x8_block_skip(grid, mi_row + r, mi_col + c, cm->mi_stride)) {
dlist[count].by = r >> r_shift;
dlist[count].bx = c >> c_shift;
dlist[count].skip = 0;
count++;
}
}
}
}
return count;
}
void copy_rect8_8bit_to_16bit_c(uint16_t *dst, int dstride, const uint8_t *src,
int sstride, int v, int h) {
int i, j;
for (i = 0; i < v; i++) {
for (j = 0; j < h; j++) {
dst[i * dstride + j] = src[i * sstride + j];
}
}
}
void copy_rect8_16bit_to_16bit_c(uint16_t *dst, int dstride,
const uint16_t *src, int sstride, int v,
int h) {
int i, j;
for (i = 0; i < v; i++) {
for (j = 0; j < h; j++) {
dst[i * dstride + j] = src[i * sstride + j];
}
}
}
void copy_sb8_16(UNUSED AV1_COMMON *cm, uint16_t *dst, int dstride,
const uint8_t *src, int src_voffset, int src_hoffset,
int sstride, int vsize, int hsize) {
#if CONFIG_HIGHBITDEPTH
if (cm->use_highbitdepth) {
const uint16_t *base =
&CONVERT_TO_SHORTPTR(src)[src_voffset * sstride + src_hoffset];
copy_rect8_16bit_to_16bit(dst, dstride, base, sstride, vsize, hsize);
} else {
#endif
const uint8_t *base = &src[src_voffset * sstride + src_hoffset];
copy_rect8_8bit_to_16bit(dst, dstride, base, sstride, vsize, hsize);
#if CONFIG_HIGHBITDEPTH
}
#endif
}
static INLINE void fill_rect(uint16_t *dst, int dstride, int v, int h,
uint16_t x) {
int i, j;
for (i = 0; i < v; i++) {
for (j = 0; j < h; j++) {
dst[i * dstride + j] = x;
}
}
}
static INLINE void copy_rect(uint16_t *dst, int dstride, const uint16_t *src,
int sstride, int v, int h) {
int i, j;
for (i = 0; i < v; i++) {
for (j = 0; j < h; j++) {
dst[i * dstride + j] = src[i * sstride + j];
}
}
}
void av1_cdef_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
MACROBLOCKD *xd) {
int sbr, sbc;
int nhsb, nvsb;
uint16_t src[OD_DERING_INBUF_SIZE];
uint16_t *linebuf[3];
uint16_t *colbuf[3];
dering_list dlist[MAX_MIB_SIZE * MAX_MIB_SIZE];
unsigned char *row_dering, *prev_row_dering, *curr_row_dering;
int dering_count;
int dir[OD_DERING_NBLOCKS][OD_DERING_NBLOCKS] = { { 0 } };
int var[OD_DERING_NBLOCKS][OD_DERING_NBLOCKS] = { { 0 } };
int stride;
int mi_wide_l2[3];
int mi_high_l2[3];
int xdec[3];
int ydec[3];
int pli;
int dering_left;
int coeff_shift = AOMMAX(cm->bit_depth - 8, 0);
int nplanes = 3;
int chroma_dering =
xd->plane[1].subsampling_x == xd->plane[1].subsampling_y &&
xd->plane[2].subsampling_x == xd->plane[2].subsampling_y;
nvsb = (cm->mi_rows + MAX_MIB_SIZE - 1) / MAX_MIB_SIZE;
nhsb = (cm->mi_cols + MAX_MIB_SIZE - 1) / MAX_MIB_SIZE;
av1_setup_dst_planes(xd->plane, cm->sb_size, frame, 0, 0);
row_dering = aom_malloc(sizeof(*row_dering) * (nhsb + 2) * 2);
memset(row_dering, 1, sizeof(*row_dering) * (nhsb + 2) * 2);
prev_row_dering = row_dering + 1;
curr_row_dering = prev_row_dering + nhsb + 2;
for (pli = 0; pli < nplanes; pli++) {
xdec[pli] = xd->plane[pli].subsampling_x;
ydec[pli] = xd->plane[pli].subsampling_y;
mi_wide_l2[pli] = MI_SIZE_LOG2 - xd->plane[pli].subsampling_x;
mi_high_l2[pli] = MI_SIZE_LOG2 - xd->plane[pli].subsampling_y;
}
stride = (cm->mi_cols << MI_SIZE_LOG2) + 2 * OD_FILT_HBORDER;
for (pli = 0; pli < nplanes; pli++) {
linebuf[pli] = aom_malloc(sizeof(*linebuf) * OD_FILT_VBORDER * stride);
colbuf[pli] =
aom_malloc(sizeof(*colbuf) *
((MAX_SB_SIZE << mi_high_l2[pli]) + 2 * OD_FILT_VBORDER) *
OD_FILT_HBORDER);
}
for (sbr = 0; sbr < nvsb; sbr++) {
for (pli = 0; pli < nplanes; pli++) {
const int block_height =
(MAX_MIB_SIZE << mi_high_l2[pli]) + 2 * OD_FILT_VBORDER;
fill_rect(colbuf[pli], OD_FILT_HBORDER, block_height, OD_FILT_HBORDER,
OD_DERING_VERY_LARGE);
}
dering_left = 1;
for (sbc = 0; sbc < nhsb; sbc++) {
int level, clpf_strength;
int uv_level, uv_clpf_strength;
int nhb, nvb;
int cstart = 0;
curr_row_dering[sbc] = 0;
if (cm->mi_grid_visible[MAX_MIB_SIZE * sbr * cm->mi_stride +
MAX_MIB_SIZE * sbc] == NULL ||
cm->mi_grid_visible[MAX_MIB_SIZE * sbr * cm->mi_stride +
MAX_MIB_SIZE * sbc]
->mbmi.cdef_strength == -1) {
dering_left = 0;
continue;
}
if (!dering_left) cstart = -OD_FILT_HBORDER;
nhb = AOMMIN(MAX_MIB_SIZE, cm->mi_cols - MAX_MIB_SIZE * sbc);
nvb = AOMMIN(MAX_MIB_SIZE, cm->mi_rows - MAX_MIB_SIZE * sbr);
int tile_top, tile_left, tile_bottom, tile_right;
int mi_idx = MAX_MIB_SIZE * sbr * cm->mi_stride + MAX_MIB_SIZE * sbc;
BOUNDARY_TYPE boundary_tl =
cm->mi_grid_visible[MAX_MIB_SIZE * sbr * cm->mi_stride +
MAX_MIB_SIZE * sbc]
->mbmi.boundary_info;
tile_top = boundary_tl & TILE_ABOVE_BOUNDARY;
tile_left = boundary_tl & TILE_LEFT_BOUNDARY;
/* Right and bottom information appear unreliable, so we use the top
and left flags for the next superblocks. */
if (sbr != nvsb - 1 &&
cm->mi_grid_visible[mi_idx + MAX_MIB_SIZE * cm->mi_stride])
tile_bottom = cm->mi_grid_visible[mi_idx + MAX_MIB_SIZE * cm->mi_stride]
->mbmi.boundary_info &
TILE_ABOVE_BOUNDARY;
else
tile_bottom = 1;
if (sbc != nhsb - 1 && cm->mi_grid_visible[mi_idx + MAX_MIB_SIZE])
tile_right =
cm->mi_grid_visible[mi_idx + MAX_MIB_SIZE]->mbmi.boundary_info &
TILE_LEFT_BOUNDARY;
else
tile_right = 1;
const int mbmi_cdef_strength =
cm->mi_grid_visible[MAX_MIB_SIZE * sbr * cm->mi_stride +
MAX_MIB_SIZE * sbc]
->mbmi.cdef_strength;
level = cm->cdef_strengths[mbmi_cdef_strength] / CLPF_STRENGTHS;
clpf_strength = cm->cdef_strengths[mbmi_cdef_strength] % CLPF_STRENGTHS;
clpf_strength += clpf_strength == 3;
uv_level = cm->cdef_uv_strengths[mbmi_cdef_strength] / CLPF_STRENGTHS;
uv_clpf_strength =
cm->cdef_uv_strengths[mbmi_cdef_strength] % CLPF_STRENGTHS;
uv_clpf_strength += uv_clpf_strength == 3;
if ((level == 0 && clpf_strength == 0 && uv_level == 0 &&
uv_clpf_strength == 0) ||
(dering_count = sb_compute_dering_list(
cm, sbr * MAX_MIB_SIZE, sbc * MAX_MIB_SIZE, dlist,
get_filter_skip(level) || get_filter_skip(uv_level))) == 0) {
dering_left = 0;
continue;
}
curr_row_dering[sbc] = 1;
for (pli = 0; pli < nplanes; pli++) {
uint16_t dst[MAX_SB_SIZE * MAX_SB_SIZE];
int coffset;
int rend, cend;
int clpf_damping = cm->cdef_clpf_damping;
int dering_damping = cm->cdef_dering_damping;
int hsize = nhb << mi_wide_l2[pli];
int vsize = nvb << mi_high_l2[pli];
if (pli) {
if (chroma_dering)
level = uv_level;
else
level = 0;
clpf_strength = uv_clpf_strength;
}
if (sbc == nhsb - 1)
cend = hsize;
else
cend = hsize + OD_FILT_HBORDER;
if (sbr == nvsb - 1)
rend = vsize;
else
rend = vsize + OD_FILT_VBORDER;
coffset = sbc * MAX_MIB_SIZE << mi_wide_l2[pli];
if (sbc == nhsb - 1) {
/* On the last superblock column, fill in the right border with
OD_DERING_VERY_LARGE to avoid filtering with the outside. */
fill_rect(&src[cend + OD_FILT_HBORDER], OD_FILT_BSTRIDE,
rend + OD_FILT_VBORDER, hsize + OD_FILT_HBORDER - cend,
OD_DERING_VERY_LARGE);
}
if (sbr == nvsb - 1) {
/* On the last superblock row, fill in the bottom border with
OD_DERING_VERY_LARGE to avoid filtering with the outside. */
fill_rect(&src[(rend + OD_FILT_VBORDER) * OD_FILT_BSTRIDE],
OD_FILT_BSTRIDE, OD_FILT_VBORDER,
hsize + 2 * OD_FILT_HBORDER, OD_DERING_VERY_LARGE);
}
/* Copy in the pixels we need from the current superblock for
deringing.*/
copy_sb8_16(
cm,
&src[OD_FILT_VBORDER * OD_FILT_BSTRIDE + OD_FILT_HBORDER + cstart],
OD_FILT_BSTRIDE, xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << mi_high_l2[pli]) * sbr, coffset + cstart,
xd->plane[pli].dst.stride, rend, cend - cstart);
if (!prev_row_dering[sbc]) {
copy_sb8_16(cm, &src[OD_FILT_HBORDER], OD_FILT_BSTRIDE,
xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << mi_high_l2[pli]) * sbr - OD_FILT_VBORDER,
coffset, xd->plane[pli].dst.stride, OD_FILT_VBORDER,
hsize);
} else if (sbr > 0) {
copy_rect(&src[OD_FILT_HBORDER], OD_FILT_BSTRIDE,
&linebuf[pli][coffset], stride, OD_FILT_VBORDER, hsize);
} else {
fill_rect(&src[OD_FILT_HBORDER], OD_FILT_BSTRIDE, OD_FILT_VBORDER,
hsize, OD_DERING_VERY_LARGE);
}
if (!prev_row_dering[sbc - 1]) {
copy_sb8_16(cm, src, OD_FILT_BSTRIDE, xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << mi_high_l2[pli]) * sbr - OD_FILT_VBORDER,
coffset - OD_FILT_HBORDER, xd->plane[pli].dst.stride,
OD_FILT_VBORDER, OD_FILT_HBORDER);
} else if (sbr > 0 && sbc > 0) {
copy_rect(src, OD_FILT_BSTRIDE,
&linebuf[pli][coffset - OD_FILT_HBORDER], stride,
OD_FILT_VBORDER, OD_FILT_HBORDER);
} else {
fill_rect(src, OD_FILT_BSTRIDE, OD_FILT_VBORDER, OD_FILT_HBORDER,
OD_DERING_VERY_LARGE);
}
if (!prev_row_dering[sbc + 1]) {
copy_sb8_16(cm, &src[OD_FILT_HBORDER + (nhb << mi_wide_l2[pli])],
OD_FILT_BSTRIDE, xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << mi_high_l2[pli]) * sbr - OD_FILT_VBORDER,
coffset + hsize, xd->plane[pli].dst.stride,
OD_FILT_VBORDER, OD_FILT_HBORDER);
} else if (sbr > 0 && sbc < nhsb - 1) {
copy_rect(&src[hsize + OD_FILT_HBORDER], OD_FILT_BSTRIDE,
&linebuf[pli][coffset + hsize], stride, OD_FILT_VBORDER,
OD_FILT_HBORDER);
} else {
fill_rect(&src[hsize + OD_FILT_HBORDER], OD_FILT_BSTRIDE,
OD_FILT_VBORDER, OD_FILT_HBORDER, OD_DERING_VERY_LARGE);
}
if (dering_left) {
/* If we deringed the superblock on the left then we need to copy in
saved pixels. */
copy_rect(src, OD_FILT_BSTRIDE, colbuf[pli], OD_FILT_HBORDER,
rend + OD_FILT_VBORDER, OD_FILT_HBORDER);
}
/* Saving pixels in case we need to dering the superblock on the
right. */
copy_rect(colbuf[pli], OD_FILT_HBORDER, src + hsize, OD_FILT_BSTRIDE,
rend + OD_FILT_VBORDER, OD_FILT_HBORDER);
copy_sb8_16(
cm, &linebuf[pli][coffset], stride, xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << mi_high_l2[pli]) * (sbr + 1) - OD_FILT_VBORDER,
coffset, xd->plane[pli].dst.stride, OD_FILT_VBORDER, hsize);
if (tile_top) {
fill_rect(src, OD_FILT_BSTRIDE, OD_FILT_VBORDER,
hsize + 2 * OD_FILT_HBORDER, OD_DERING_VERY_LARGE);
}
if (tile_left) {
fill_rect(src, OD_FILT_BSTRIDE, vsize + 2 * OD_FILT_VBORDER,
OD_FILT_HBORDER, OD_DERING_VERY_LARGE);
}
if (tile_bottom) {
fill_rect(&src[(vsize + OD_FILT_VBORDER) * OD_FILT_BSTRIDE],
OD_FILT_BSTRIDE, OD_FILT_VBORDER,
hsize + 2 * OD_FILT_HBORDER, OD_DERING_VERY_LARGE);
}
if (tile_right) {
fill_rect(&src[hsize + OD_FILT_HBORDER], OD_FILT_BSTRIDE,
vsize + 2 * OD_FILT_VBORDER, OD_FILT_HBORDER,
OD_DERING_VERY_LARGE);
}
#if CONFIG_HIGHBITDEPTH
if (cm->use_highbitdepth) {
od_dering(
(uint8_t *)&CONVERT_TO_SHORTPTR(
xd->plane[pli]
.dst.buf)[xd->plane[pli].dst.stride *
(MAX_MIB_SIZE * sbr << mi_high_l2[pli]) +
(sbc * MAX_MIB_SIZE << mi_wide_l2[pli])],
xd->plane[pli].dst.stride, dst,
&src[OD_FILT_VBORDER * OD_FILT_BSTRIDE + OD_FILT_HBORDER],
xdec[pli], ydec[pli], dir, NULL, var, pli, dlist, dering_count,
level, clpf_strength, clpf_damping, dering_damping, coeff_shift,
0, 1);
} else {
#endif
od_dering(&xd->plane[pli]
.dst.buf[xd->plane[pli].dst.stride *
(MAX_MIB_SIZE * sbr << mi_high_l2[pli]) +
(sbc * MAX_MIB_SIZE << mi_wide_l2[pli])],
xd->plane[pli].dst.stride, dst,
&src[OD_FILT_VBORDER * OD_FILT_BSTRIDE + OD_FILT_HBORDER],
xdec[pli], ydec[pli], dir, NULL, var, pli, dlist,
dering_count, level, clpf_strength, clpf_damping,
dering_damping, coeff_shift, 0, 0);
#if CONFIG_HIGHBITDEPTH
}
#endif
}
dering_left = 1;
}
{
unsigned char *tmp;
tmp = prev_row_dering;
prev_row_dering = curr_row_dering;
curr_row_dering = tmp;
}
}
aom_free(row_dering);
for (pli = 0; pli < nplanes; pli++) {
aom_free(linebuf[pli]);
aom_free(colbuf[pli]);
}
}

53
third_party/aom/av1/common/cdef.h vendored Normal file
View file

@ -0,0 +1,53 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_DERING_H_
#define AV1_COMMON_DERING_H_
#define CDEF_STRENGTH_BITS 7
#define DERING_STRENGTHS 32
#define CLPF_STRENGTHS 4
#include "./aom_config.h"
#include "aom/aom_integer.h"
#include "aom_ports/mem.h"
#include "av1/common/od_dering.h"
#include "av1/common/onyxc_int.h"
#include "./od_dering.h"
static INLINE int sign(int i) { return i < 0 ? -1 : 1; }
static INLINE int constrain(int diff, int threshold, unsigned int damping) {
return threshold
? sign(diff) *
AOMMIN(
abs(diff),
AOMMAX(0, threshold - (abs(diff) >>
(damping - get_msb(threshold)))))
: 0;
}
#ifdef __cplusplus
extern "C" {
#endif
int sb_all_skip(const AV1_COMMON *const cm, int mi_row, int mi_col);
int sb_compute_dering_list(const AV1_COMMON *const cm, int mi_row, int mi_col,
dering_list *dlist, int filter_skip);
void av1_cdef_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm, MACROBLOCKD *xd);
void av1_cdef_search(YV12_BUFFER_CONFIG *frame, const YV12_BUFFER_CONFIG *ref,
AV1_COMMON *cm, MACROBLOCKD *xd);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_DERING_H_

27
third_party/aom/av1/common/cdef_simd.h vendored Normal file
View file

@ -0,0 +1,27 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_CDEF_SIMD_H_
#define AV1_COMMON_CDEF_SIMD_H_
#include "aom_dsp/aom_simd.h"
// sign(a-b) * min(abs(a-b), max(0, threshold - (abs(a-b) >> adjdamp)))
SIMD_INLINE v128 constrain16(v128 a, v128 b, unsigned int threshold,
unsigned int adjdamp) {
v128 diff = v128_sub_16(a, b);
const v128 sign = v128_shr_n_s16(diff, 15);
diff = v128_abs_s16(diff);
const v128 s =
v128_ssub_u16(v128_dup_16(threshold), v128_shr_u16(diff, adjdamp));
return v128_xor(v128_add_16(sign, v128_min_s16(diff, s)), sign);
}
#endif // AV1_COMMON_CDEF_SIMD_H_

240
third_party/aom/av1/common/cfl.c vendored Normal file
View file

@ -0,0 +1,240 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "av1/common/cfl.h"
#include "av1/common/common_data.h"
#include "av1/common/onyxc_int.h"
#include "aom/internal/aom_codec_internal.h"
void cfl_init(CFL_CTX *cfl, AV1_COMMON *cm, int subsampling_x,
int subsampling_y) {
if (!((subsampling_x == 0 && subsampling_y == 0) ||
(subsampling_x == 1 && subsampling_y == 1))) {
aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM,
"Only 4:4:4 and 4:2:0 are currently supported by CfL");
}
memset(&cfl->y_pix, 0, sizeof(uint8_t) * MAX_SB_SQUARE);
cfl->subsampling_x = subsampling_x;
cfl->subsampling_y = subsampling_y;
}
// CfL computes its own block-level DC_PRED. This is required to compute both
// alpha_cb and alpha_cr before the prediction are computed.
void cfl_dc_pred(MACROBLOCKD *xd, BLOCK_SIZE plane_bsize, TX_SIZE tx_size) {
const struct macroblockd_plane *const pd_u = &xd->plane[AOM_PLANE_U];
const struct macroblockd_plane *const pd_v = &xd->plane[AOM_PLANE_V];
const uint8_t *const dst_u = pd_u->dst.buf;
const uint8_t *const dst_v = pd_v->dst.buf;
const int dst_u_stride = pd_u->dst.stride;
const int dst_v_stride = pd_v->dst.stride;
const int block_width = (plane_bsize != BLOCK_INVALID)
? block_size_wide[plane_bsize]
: tx_size_wide[tx_size];
const int block_height = (plane_bsize != BLOCK_INVALID)
? block_size_high[plane_bsize]
: tx_size_high[tx_size];
// Number of pixel on the top and left borders.
const int num_pel = block_width + block_height;
int sum_u = 0;
int sum_v = 0;
// Match behavior of build_intra_predictors (reconintra.c) at superblock
// boundaries:
//
// 127 127 127 .. 127 127 127 127 127 127
// 129 A B .. Y Z
// 129 C D .. W X
// 129 E F .. U V
// 129 G H .. S T T T T T
// ..
// TODO(ltrudeau) replace this with DC_PRED assembly
if (xd->up_available && xd->mb_to_right_edge >= 0) {
for (int i = 0; i < block_width; i++) {
sum_u += dst_u[-dst_u_stride + i];
sum_v += dst_v[-dst_v_stride + i];
}
} else {
sum_u = block_width * 127;
sum_v = block_width * 127;
}
if (xd->left_available && xd->mb_to_bottom_edge >= 0) {
for (int i = 0; i < block_height; i++) {
sum_u += dst_u[i * dst_u_stride - 1];
sum_v += dst_v[i * dst_v_stride - 1];
}
} else {
sum_u += block_height * 129;
sum_v += block_height * 129;
}
xd->cfl->dc_pred[CFL_PRED_U] = (sum_u + (num_pel >> 1)) / num_pel;
xd->cfl->dc_pred[CFL_PRED_V] = (sum_v + (num_pel >> 1)) / num_pel;
}
// Predict the current transform block using CfL.
// it is assumed that dst points at the start of the transform block
void cfl_predict_block(const CFL_CTX *cfl, uint8_t *dst, int dst_stride,
int row, int col, TX_SIZE tx_size, int dc_pred) {
const int tx_block_width = tx_size_wide[tx_size];
const int tx_block_height = tx_size_high[tx_size];
// TODO(ltrudeau) implement alpha
// Place holder for alpha
const double alpha = 0;
const double y_avg = cfl_load(cfl, dst, dst_stride, row, col, tx_size);
for (int j = 0; j < tx_block_height; j++) {
for (int i = 0; i < tx_block_width; i++) {
dst[i] = (uint8_t)(alpha * y_avg + dc_pred + 0.5);
}
dst += dst_stride;
}
}
void cfl_store(CFL_CTX *cfl, const uint8_t *input, int input_stride, int row,
int col, TX_SIZE tx_size) {
const int tx_width = tx_size_wide[tx_size];
const int tx_height = tx_size_high[tx_size];
const int tx_off_log2 = tx_size_wide_log2[0];
// Store the input into the CfL pixel buffer
uint8_t *y_pix = &cfl->y_pix[(row * MAX_SB_SIZE + col) << tx_off_log2];
// Check that we remain inside the pixel buffer.
assert(MAX_SB_SIZE * (row + tx_height - 1) + col + tx_width - 1 <
MAX_SB_SQUARE);
for (int j = 0; j < tx_height; j++) {
for (int i = 0; i < tx_width; i++) {
y_pix[i] = input[i];
}
y_pix += MAX_SB_SIZE;
input += input_stride;
}
// Store the surface of the pixel buffer that was written to, this way we
// can manage chroma overrun (e.g. when the chroma surfaces goes beyond the
// frame boundary)
if (col == 0 && row == 0) {
cfl->y_width = tx_width;
cfl->y_height = tx_height;
} else {
cfl->y_width = OD_MAXI((col << tx_off_log2) + tx_width, cfl->y_width);
cfl->y_height = OD_MAXI((row << tx_off_log2) + tx_height, cfl->y_height);
}
}
// Load from the CfL pixel buffer into output
double cfl_load(const CFL_CTX *cfl, uint8_t *output, int output_stride, int row,
int col, TX_SIZE tx_size) {
const int tx_width = tx_size_wide[tx_size];
const int tx_height = tx_size_high[tx_size];
const int sub_x = cfl->subsampling_x;
const int sub_y = cfl->subsampling_y;
const int tx_off_log2 = tx_size_wide_log2[0];
const uint8_t *y_pix;
int diff_width = 0;
int diff_height = 0;
int pred_row_offset = 0;
int output_row_offset = 0;
int top_left, bot_left;
// TODO(ltrudeau) add support for 4:2:2
if (sub_y == 0 && sub_x == 0) {
y_pix = &cfl->y_pix[(row * MAX_SB_SIZE + col) << tx_off_log2];
int uv_width = (col << tx_off_log2) + tx_width;
diff_width = uv_width - cfl->y_width;
int uv_height = (row << tx_off_log2) + tx_width;
diff_height = uv_height - cfl->y_height;
for (int j = 0; j < tx_height; j++) {
for (int i = 0; i < tx_width; i++) {
// In 4:4:4, pixels match 1 to 1
output[output_row_offset + i] = y_pix[pred_row_offset + i];
}
pred_row_offset += MAX_SB_SIZE;
output_row_offset += output_stride;
}
} else if (sub_y == 1 && sub_x == 1) {
y_pix = &cfl->y_pix[(row * MAX_SB_SIZE + col) << (tx_off_log2 + sub_y)];
int uv_width = ((col << tx_off_log2) + tx_width) << sub_x;
diff_width = (uv_width - cfl->y_width) >> sub_x;
int uv_height = ((row << tx_off_log2) + tx_width) << sub_y;
diff_height = (uv_height - cfl->y_height) >> sub_y;
for (int j = 0; j < tx_height; j++) {
for (int i = 0; i < tx_width; i++) {
top_left = (pred_row_offset + i) << sub_y;
bot_left = top_left + MAX_SB_SIZE;
// In 4:2:0, average pixels in 2x2 grid
output[output_row_offset + i] = OD_SHR_ROUND(
y_pix[top_left] + y_pix[top_left + 1] // Top row
+ y_pix[bot_left] + y_pix[bot_left + 1] // Bottom row
,
2);
}
pred_row_offset += MAX_SB_SIZE;
output_row_offset += output_stride;
}
} else {
assert(0); // Unsupported chroma subsampling
}
// Due to frame boundary issues, it is possible that the total area of
// covered by Chroma exceeds that of Luma. When this happens, we write over
// the broken data by repeating the last columns and/or rows.
//
// Note that in order to manage the case where both rows and columns
// overrun,
// we apply rows first. This way, when the rows overrun the bottom of the
// frame, the columns will be copied over them.
if (diff_width > 0) {
int last_pixel;
output_row_offset = tx_width - diff_width;
for (int j = 0; j < tx_height; j++) {
last_pixel = output_row_offset - 1;
for (int i = 0; i < diff_width; i++) {
output[output_row_offset + i] = output[last_pixel];
}
output_row_offset += output_stride;
}
}
if (diff_height > 0) {
output_row_offset = diff_height * output_stride;
const int last_row_offset = output_row_offset - output_stride;
for (int j = 0; j < diff_height; j++) {
for (int i = 0; i < tx_width; i++) {
output[output_row_offset + i] = output[last_row_offset + i];
}
output_row_offset += output_stride;
}
}
int avg = 0;
output_row_offset = 0;
for (int j = 0; j < tx_height; j++) {
for (int i = 0; i < tx_width; i++) {
avg += output[output_row_offset + i];
}
output_row_offset += output_stride;
}
return avg / (double)(tx_width * tx_height);
}

52
third_party/aom/av1/common/cfl.h vendored Normal file
View file

@ -0,0 +1,52 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_CFL_H_
#define AV1_COMMON_CFL_H_
#include "av1/common/enums.h"
// Forward declaration of AV1_COMMON, in order to avoid creating a cyclic
// dependency by importing av1/common/onyxc_int.h
typedef struct AV1Common AV1_COMMON;
// Forward declaration of MACROBLOCK, in order to avoid creating a cyclic
// dependency by importing av1/common/blockd.h
typedef struct macroblockd MACROBLOCKD;
typedef struct {
// Pixel buffer containing the luma pixels used as prediction for chroma
uint8_t y_pix[MAX_SB_SQUARE];
// Height and width of the luma prediction block currently in the pixel buffer
int y_height, y_width;
// Chroma subsampling
int subsampling_x, subsampling_y;
// CfL Performs its own block level DC_PRED for each chromatic plane
int dc_pred[CFL_PRED_PLANES];
} CFL_CTX;
void cfl_init(CFL_CTX *cfl, AV1_COMMON *cm, int subsampling_x,
int subsampling_y);
void cfl_dc_pred(MACROBLOCKD *xd, BLOCK_SIZE plane_bsize, TX_SIZE tx_size);
void cfl_predict_block(const CFL_CTX *cfl, uint8_t *dst, int dst_stride,
int row, int col, TX_SIZE tx_size, int dc_pred);
void cfl_store(CFL_CTX *cfl, const uint8_t *input, int input_stride, int row,
int col, TX_SIZE tx_size);
double cfl_load(const CFL_CTX *cfl, uint8_t *output, int output_stride, int row,
int col, TX_SIZE tx_size);
#endif // AV1_COMMON_CFL_H_

116
third_party/aom/av1/common/clpf.c vendored Normal file
View file

@ -0,0 +1,116 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "./clpf.h"
#include "./av1_rtcd.h"
#include "./cdef.h"
#include "aom/aom_image.h"
#include "aom_dsp/aom_dsp_common.h"
int av1_clpf_sample(int X, int A, int B, int C, int D, int E, int F, int G,
int H, int s, unsigned int dmp) {
int delta = 1 * constrain(A - X, s, dmp) + 3 * constrain(B - X, s, dmp) +
1 * constrain(C - X, s, dmp) + 3 * constrain(D - X, s, dmp) +
3 * constrain(E - X, s, dmp) + 1 * constrain(F - X, s, dmp) +
3 * constrain(G - X, s, dmp) + 1 * constrain(H - X, s, dmp);
return (8 + delta - (delta < 0)) >> 4;
}
int av1_clpf_hsample(int X, int A, int B, int C, int D, int s,
unsigned int dmp) {
int delta = 1 * constrain(A - X, s, dmp) + 3 * constrain(B - X, s, dmp) +
3 * constrain(C - X, s, dmp) + 1 * constrain(D - X, s, dmp);
return (4 + delta - (delta < 0)) >> 3;
}
void aom_clpf_block_c(uint8_t *dst, const uint16_t *src, int dstride,
int sstride, int sizex, int sizey, unsigned int strength,
unsigned int damping) {
int x, y;
for (y = 0; y < sizey; y++) {
for (x = 0; x < sizex; x++) {
const int X = src[y * sstride + x];
const int A = src[(y - 2) * sstride + x];
const int B = src[(y - 1) * sstride + x];
const int C = src[y * sstride + x - 2];
const int D = src[y * sstride + x - 1];
const int E = src[y * sstride + x + 1];
const int F = src[y * sstride + x + 2];
const int G = src[(y + 1) * sstride + x];
const int H = src[(y + 2) * sstride + x];
const int delta =
av1_clpf_sample(X, A, B, C, D, E, F, G, H, strength, damping);
dst[y * dstride + x] = X + delta;
}
}
}
// Identical to aom_clpf_block_c() apart from "dst".
void aom_clpf_block_hbd_c(uint16_t *dst, const uint16_t *src, int dstride,
int sstride, int sizex, int sizey,
unsigned int strength, unsigned int damping) {
int x, y;
for (y = 0; y < sizey; y++) {
for (x = 0; x < sizex; x++) {
const int X = src[y * sstride + x];
const int A = src[(y - 2) * sstride + x];
const int B = src[(y - 1) * sstride + x];
const int C = src[y * sstride + x - 2];
const int D = src[y * sstride + x - 1];
const int E = src[y * sstride + x + 1];
const int F = src[y * sstride + x + 2];
const int G = src[(y + 1) * sstride + x];
const int H = src[(y + 2) * sstride + x];
const int delta =
av1_clpf_sample(X, A, B, C, D, E, F, G, H, strength, damping);
dst[y * dstride + x] = X + delta;
}
}
}
// Vertically restricted filter
void aom_clpf_hblock_c(uint8_t *dst, const uint16_t *src, int dstride,
int sstride, int sizex, int sizey, unsigned int strength,
unsigned int damping) {
int x, y;
for (y = 0; y < sizey; y++) {
for (x = 0; x < sizex; x++) {
const int X = src[y * sstride + x];
const int A = src[y * sstride + x - 2];
const int B = src[y * sstride + x - 1];
const int C = src[y * sstride + x + 1];
const int D = src[y * sstride + x + 2];
const int delta = av1_clpf_hsample(X, A, B, C, D, strength, damping);
dst[y * dstride + x] = X + delta;
}
}
}
void aom_clpf_hblock_hbd_c(uint16_t *dst, const uint16_t *src, int dstride,
int sstride, int sizex, int sizey,
unsigned int strength, unsigned int damping) {
int x, y;
for (y = 0; y < sizey; y++) {
for (x = 0; x < sizex; x++) {
const int X = src[y * sstride + x];
const int A = src[y * sstride + x - 2];
const int B = src[y * sstride + x - 1];
const int C = src[y * sstride + x + 1];
const int D = src[y * sstride + x + 2];
const int delta = av1_clpf_hsample(X, A, B, C, D, strength, damping);
dst[y * dstride + x] = X + delta;
}
}
}

18
third_party/aom/av1/common/clpf.h vendored Normal file
View file

@ -0,0 +1,18 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_CLPF_H_
#define AV1_COMMON_CLPF_H_
#include "av1/common/reconinter.h"
int av1_clpf_sample(int X, int A, int B, int C, int D, int E, int F, int G,
int H, int b, unsigned int dmp);
#endif

14
third_party/aom/av1/common/clpf_neon.c vendored Normal file
View file

@ -0,0 +1,14 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "aom_dsp/aom_simd.h"
#define SIMD_FUNC(name) name##_neon
#include "./clpf_simd.h"

446
third_party/aom/av1/common/clpf_simd.h vendored Normal file
View file

@ -0,0 +1,446 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "./av1_rtcd.h"
#include "./cdef_simd.h"
#include "aom_ports/bitops.h"
#include "aom_ports/mem.h"
// sign(a - b) * min(abs(a - b), max(0, strength - (abs(a - b) >> adjdamp)))
SIMD_INLINE v128 constrain(v256 a, v256 b, unsigned int strength,
unsigned int adjdamp) {
const v256 diff16 = v256_sub_16(a, b);
v128 diff = v128_pack_s16_s8(v256_high_v128(diff16), v256_low_v128(diff16));
const v128 sign = v128_cmplt_s8(diff, v128_zero());
diff = v128_abs_s8(diff);
return v128_xor(
v128_add_8(sign,
v128_min_u8(diff, v128_ssub_u8(v128_dup_8(strength),
v128_shr_u8(diff, adjdamp)))),
sign);
}
// delta = 1/16 * constrain(a, x, s, d) + 3/16 * constrain(b, x, s, d) +
// 1/16 * constrain(c, x, s, d) + 3/16 * constrain(d, x, s, d) +
// 3/16 * constrain(e, x, s, d) + 1/16 * constrain(f, x, s, d) +
// 3/16 * constrain(g, x, s, d) + 1/16 * constrain(h, x, s, d)
SIMD_INLINE v128 calc_delta(v256 x, v256 a, v256 b, v256 c, v256 d, v256 e,
v256 f, v256 g, v256 h, unsigned int s,
unsigned int dmp) {
const v128 bdeg =
v128_add_8(v128_add_8(constrain(b, x, s, dmp), constrain(d, x, s, dmp)),
v128_add_8(constrain(e, x, s, dmp), constrain(g, x, s, dmp)));
const v128 delta = v128_add_8(
v128_add_8(v128_add_8(constrain(a, x, s, dmp), constrain(c, x, s, dmp)),
v128_add_8(constrain(f, x, s, dmp), constrain(h, x, s, dmp))),
v128_add_8(v128_add_8(bdeg, bdeg), bdeg));
return v128_add_8(
v128_pack_s16_u8(v256_high_v128(x), v256_low_v128(x)),
v128_shr_s8(
v128_add_8(v128_dup_8(8),
v128_add_8(delta, v128_cmplt_s8(delta, v128_zero()))),
4));
}
// delta = 1/8 * constrain(a, x, s, d) + 3/8 * constrain(b, x, s, d) +
// 3/8 * constrain(c, x, s, d) + 1/8 * constrain(d, x, s, d) +
SIMD_INLINE v128 calc_hdelta(v256 x, v256 a, v256 b, v256 c, v256 d,
unsigned int s, unsigned int dmp) {
const v128 bc = v128_add_8(constrain(b, x, s, dmp), constrain(c, x, s, dmp));
const v128 delta =
v128_add_8(v128_add_8(constrain(a, x, s, dmp), constrain(d, x, s, dmp)),
v128_add_8(v128_add_8(bc, bc), bc));
return v128_add_8(
v128_pack_s16_u8(v256_high_v128(x), v256_low_v128(x)),
v128_shr_s8(
v128_add_8(v128_dup_8(4),
v128_add_8(delta, v128_cmplt_s8(delta, v128_zero()))),
3));
}
// Process blocks of width 8, two lines at a time, 8 bit.
static void SIMD_FUNC(clpf_block8)(uint8_t *dst, const uint16_t *src,
int dstride, int sstride, int sizey,
unsigned int strength,
unsigned int adjdamp) {
int y;
for (y = 0; y < sizey; y += 2) {
const v128 l1 = v128_load_aligned(src);
const v128 l2 = v128_load_aligned(src + sstride);
const v128 l3 = v128_load_aligned(src - sstride);
const v128 l4 = v128_load_aligned(src + 2 * sstride);
const v256 a = v256_from_v128(v128_load_aligned(src - 2 * sstride), l3);
const v256 b = v256_from_v128(l3, l1);
const v256 g = v256_from_v128(l2, l4);
const v256 h = v256_from_v128(l4, v128_load_aligned(src + 3 * sstride));
const v256 c = v256_from_v128(v128_load_unaligned(src - 2),
v128_load_unaligned(src - 2 + sstride));
const v256 d = v256_from_v128(v128_load_unaligned(src - 1),
v128_load_unaligned(src - 1 + sstride));
const v256 e = v256_from_v128(v128_load_unaligned(src + 1),
v128_load_unaligned(src + 1 + sstride));
const v256 f = v256_from_v128(v128_load_unaligned(src + 2),
v128_load_unaligned(src + 2 + sstride));
const v128 o = calc_delta(v256_from_v128(l1, l2), a, b, c, d, e, f, g, h,
strength, adjdamp);
v64_store_aligned(dst, v128_high_v64(o));
v64_store_aligned(dst + dstride, v128_low_v64(o));
src += sstride * 2;
dst += dstride * 2;
}
}
// Process blocks of width 4, four lines at a time, 8 bit.
static void SIMD_FUNC(clpf_block4)(uint8_t *dst, const uint16_t *src,
int dstride, int sstride, int sizey,
unsigned int strength,
unsigned int adjdamp) {
int y;
for (y = 0; y < sizey; y += 4) {
const v64 l0 = v64_load_aligned(src - 2 * sstride);
const v64 l1 = v64_load_aligned(src - sstride);
const v64 l2 = v64_load_aligned(src);
const v64 l3 = v64_load_aligned(src + sstride);
const v64 l4 = v64_load_aligned(src + 2 * sstride);
const v64 l5 = v64_load_aligned(src + 3 * sstride);
const v64 l6 = v64_load_aligned(src + 4 * sstride);
const v64 l7 = v64_load_aligned(src + 5 * sstride);
const v128 o =
calc_delta(v256_from_v64(l2, l3, l4, l5), v256_from_v64(l0, l1, l2, l3),
v256_from_v64(l1, l2, l3, l4),
v256_from_v64(v64_load_unaligned(src - 2),
v64_load_unaligned(src + sstride - 2),
v64_load_unaligned(src + 2 * sstride - 2),
v64_load_unaligned(src + 3 * sstride - 2)),
v256_from_v64(v64_load_unaligned(src - 1),
v64_load_unaligned(src + sstride - 1),
v64_load_unaligned(src + 2 * sstride - 1),
v64_load_unaligned(src + 3 * sstride - 1)),
v256_from_v64(v64_load_unaligned(src + 1),
v64_load_unaligned(src + sstride + 1),
v64_load_unaligned(src + 2 * sstride + 1),
v64_load_unaligned(src + 3 * sstride + 1)),
v256_from_v64(v64_load_unaligned(src + 2),
v64_load_unaligned(src + sstride + 2),
v64_load_unaligned(src + 2 * sstride + 2),
v64_load_unaligned(src + 3 * sstride + 2)),
v256_from_v64(l3, l4, l5, l6), v256_from_v64(l4, l5, l6, l7),
strength, adjdamp);
u32_store_aligned(dst, v128_low_u32(v128_shr_n_byte(o, 12)));
u32_store_aligned(dst + dstride, v128_low_u32(v128_shr_n_byte(o, 8)));
u32_store_aligned(dst + 2 * dstride, v128_low_u32(v128_shr_n_byte(o, 4)));
u32_store_aligned(dst + 3 * dstride, v128_low_u32(o));
dst += 4 * dstride;
src += 4 * sstride;
}
}
static void SIMD_FUNC(clpf_hblock8)(uint8_t *dst, const uint16_t *src,
int dstride, int sstride, int sizey,
unsigned int strength,
unsigned int adjdamp) {
int y;
for (y = 0; y < sizey; y += 2) {
const v256 x = v256_from_v128(v128_load_aligned(src),
v128_load_aligned(src + sstride));
const v256 a = v256_from_v128(v128_load_unaligned(src - 2),
v128_load_unaligned(src - 2 + sstride));
const v256 b = v256_from_v128(v128_load_unaligned(src - 1),
v128_load_unaligned(src - 1 + sstride));
const v256 c = v256_from_v128(v128_load_unaligned(src + 1),
v128_load_unaligned(src + 1 + sstride));
const v256 d = v256_from_v128(v128_load_unaligned(src + 2),
v128_load_unaligned(src + 2 + sstride));
const v128 o = calc_hdelta(x, a, b, c, d, strength, adjdamp);
v64_store_aligned(dst, v128_high_v64(o));
v64_store_aligned(dst + dstride, v128_low_v64(o));
src += sstride * 2;
dst += dstride * 2;
}
}
// Process blocks of width 4, four lines at a time, 8 bit.
static void SIMD_FUNC(clpf_hblock4)(uint8_t *dst, const uint16_t *src,
int dstride, int sstride, int sizey,
unsigned int strength,
unsigned int adjdamp) {
int y;
for (y = 0; y < sizey; y += 4) {
const v256 a = v256_from_v64(v64_load_unaligned(src - 2),
v64_load_unaligned(src + sstride - 2),
v64_load_unaligned(src + 2 * sstride - 2),
v64_load_unaligned(src + 3 * sstride - 2));
const v256 b = v256_from_v64(v64_load_unaligned(src - 1),
v64_load_unaligned(src + sstride - 1),
v64_load_unaligned(src + 2 * sstride - 1),
v64_load_unaligned(src + 3 * sstride - 1));
const v256 c = v256_from_v64(v64_load_unaligned(src + 1),
v64_load_unaligned(src + sstride + 1),
v64_load_unaligned(src + 2 * sstride + 1),
v64_load_unaligned(src + 3 * sstride + 1));
const v256 d = v256_from_v64(v64_load_unaligned(src + 2),
v64_load_unaligned(src + sstride + 2),
v64_load_unaligned(src + 2 * sstride + 2),
v64_load_unaligned(src + 3 * sstride + 2));
const v128 o = calc_hdelta(
v256_from_v64(v64_load_aligned(src), v64_load_aligned(src + sstride),
v64_load_aligned(src + 2 * sstride),
v64_load_aligned(src + 3 * sstride)),
a, b, c, d, strength, adjdamp);
u32_store_aligned(dst, v128_low_u32(v128_shr_n_byte(o, 12)));
u32_store_aligned(dst + dstride, v128_low_u32(v128_shr_n_byte(o, 8)));
u32_store_aligned(dst + 2 * dstride, v128_low_u32(v128_shr_n_byte(o, 4)));
u32_store_aligned(dst + 3 * dstride, v128_low_u32(o));
dst += 4 * dstride;
src += 4 * sstride;
}
}
void SIMD_FUNC(aom_clpf_block)(uint8_t *dst, const uint16_t *src, int dstride,
int sstride, int sizex, int sizey,
unsigned int strength, unsigned int dmp) {
if ((sizex != 4 && sizex != 8) || ((sizey & 3) && sizex == 4)) {
// Fallback to C for odd sizes:
// * block widths not 4 or 8
// * block heights not a multiple of 4 if the block width is 4
aom_clpf_block_c(dst, src, dstride, sstride, sizex, sizey, strength, dmp);
} else {
(sizex == 4 ? SIMD_FUNC(clpf_block4) : SIMD_FUNC(clpf_block8))(
dst, src, dstride, sstride, sizey, strength, dmp - get_msb(strength));
}
}
void SIMD_FUNC(aom_clpf_hblock)(uint8_t *dst, const uint16_t *src, int dstride,
int sstride, int sizex, int sizey,
unsigned int strength, unsigned int dmp) {
if ((sizex != 4 && sizex != 8) || ((sizey & 3) && sizex == 4)) {
// Fallback to C for odd sizes:
// * block widths not 4 or 8
// * block heights not a multiple of 4 if the block width is 4
aom_clpf_hblock_c(dst, src, dstride, sstride, sizex, sizey, strength, dmp);
} else {
(sizex == 4 ? SIMD_FUNC(clpf_hblock4) : SIMD_FUNC(clpf_hblock8))(
dst, src, dstride, sstride, sizey, strength, dmp - get_msb(strength));
}
}
// delta = 1/16 * constrain(a, x, s, d) + 3/16 * constrain(b, x, s, d) +
// 1/16 * constrain(c, x, s, d) + 3/16 * constrain(d, x, s, d) +
// 3/16 * constrain(e, x, s, d) + 1/16 * constrain(f, x, s, d) +
// 3/16 * constrain(g, x, s, d) + 1/16 * constrain(h, x, s, d)
SIMD_INLINE v128 calc_delta_hbd(v128 x, v128 a, v128 b, v128 c, v128 d, v128 e,
v128 f, v128 g, v128 h, unsigned int s,
unsigned int dmp) {
const v128 bdeg = v128_add_16(
v128_add_16(constrain16(b, x, s, dmp), constrain16(d, x, s, dmp)),
v128_add_16(constrain16(e, x, s, dmp), constrain16(g, x, s, dmp)));
const v128 delta = v128_add_16(
v128_add_16(
v128_add_16(constrain16(a, x, s, dmp), constrain16(c, x, s, dmp)),
v128_add_16(constrain16(f, x, s, dmp), constrain16(h, x, s, dmp))),
v128_add_16(v128_add_16(bdeg, bdeg), bdeg));
return v128_add_16(
x,
v128_shr_s16(
v128_add_16(v128_dup_16(8),
v128_add_16(delta, v128_cmplt_s16(delta, v128_zero()))),
4));
}
static void calc_delta_hbd4(v128 o, v128 a, v128 b, v128 c, v128 d, v128 e,
v128 f, v128 g, v128 h, uint16_t *dst,
unsigned int s, unsigned int dmp, int dstride) {
o = calc_delta_hbd(o, a, b, c, d, e, f, g, h, s, dmp);
v64_store_aligned(dst, v128_high_v64(o));
v64_store_aligned(dst + dstride, v128_low_v64(o));
}
static void calc_delta_hbd8(v128 o, v128 a, v128 b, v128 c, v128 d, v128 e,
v128 f, v128 g, v128 h, uint16_t *dst,
unsigned int s, unsigned int adjdamp) {
v128_store_aligned(dst,
calc_delta_hbd(o, a, b, c, d, e, f, g, h, s, adjdamp));
}
// delta = 1/16 * constrain(a, x, s, dmp) + 3/16 * constrain(b, x, s, dmp) +
// 3/16 * constrain(c, x, s, dmp) + 1/16 * constrain(d, x, s, dmp)
SIMD_INLINE v128 calc_hdelta_hbd(v128 x, v128 a, v128 b, v128 c, v128 d,
unsigned int s, unsigned int dmp) {
const v128 bc =
v128_add_16(constrain16(b, x, s, dmp), constrain16(c, x, s, dmp));
const v128 delta = v128_add_16(
v128_add_16(constrain16(a, x, s, dmp), constrain16(d, x, s, dmp)),
v128_add_16(v128_add_16(bc, bc), bc));
return v128_add_16(
x,
v128_shr_s16(
v128_add_16(v128_dup_16(4),
v128_add_16(delta, v128_cmplt_s16(delta, v128_zero()))),
3));
}
static void calc_hdelta_hbd4(v128 o, v128 a, v128 b, v128 c, v128 d,
uint16_t *dst, unsigned int s,
unsigned int adjdamp, int dstride) {
o = calc_hdelta_hbd(o, a, b, c, d, s, adjdamp);
v64_store_aligned(dst, v128_high_v64(o));
v64_store_aligned(dst + dstride, v128_low_v64(o));
}
static void calc_hdelta_hbd8(v128 o, v128 a, v128 b, v128 c, v128 d,
uint16_t *dst, unsigned int s,
unsigned int adjdamp) {
v128_store_aligned(dst, calc_hdelta_hbd(o, a, b, c, d, s, adjdamp));
}
// Process blocks of width 4, two lines at time.
static void SIMD_FUNC(clpf_block_hbd4)(uint16_t *dst, const uint16_t *src,
int dstride, int sstride, int sizey,
unsigned int strength,
unsigned int adjdamp) {
int y;
for (y = 0; y < sizey; y += 2) {
const v64 l1 = v64_load_aligned(src);
const v64 l2 = v64_load_aligned(src + sstride);
const v64 l3 = v64_load_aligned(src - sstride);
const v64 l4 = v64_load_aligned(src + 2 * sstride);
const v128 a = v128_from_v64(v64_load_aligned(src - 2 * sstride), l3);
const v128 b = v128_from_v64(l3, l1);
const v128 g = v128_from_v64(l2, l4);
const v128 h = v128_from_v64(l4, v64_load_aligned(src + 3 * sstride));
const v128 c = v128_from_v64(v64_load_unaligned(src - 2),
v64_load_unaligned(src - 2 + sstride));
const v128 d = v128_from_v64(v64_load_unaligned(src - 1),
v64_load_unaligned(src - 1 + sstride));
const v128 e = v128_from_v64(v64_load_unaligned(src + 1),
v64_load_unaligned(src + 1 + sstride));
const v128 f = v128_from_v64(v64_load_unaligned(src + 2),
v64_load_unaligned(src + 2 + sstride));
calc_delta_hbd4(v128_from_v64(l1, l2), a, b, c, d, e, f, g, h, dst,
strength, adjdamp, dstride);
src += sstride * 2;
dst += dstride * 2;
}
}
// The most simple case. Start here if you need to understand the functions.
static void SIMD_FUNC(clpf_block_hbd)(uint16_t *dst, const uint16_t *src,
int dstride, int sstride, int sizey,
unsigned int strength,
unsigned int adjdamp) {
int y;
for (y = 0; y < sizey; y++) {
const v128 o = v128_load_aligned(src);
const v128 a = v128_load_aligned(src - 2 * sstride);
const v128 b = v128_load_aligned(src - 1 * sstride);
const v128 g = v128_load_aligned(src + sstride);
const v128 h = v128_load_aligned(src + 2 * sstride);
const v128 c = v128_load_unaligned(src - 2);
const v128 d = v128_load_unaligned(src - 1);
const v128 e = v128_load_unaligned(src + 1);
const v128 f = v128_load_unaligned(src + 2);
calc_delta_hbd8(o, a, b, c, d, e, f, g, h, dst, strength, adjdamp);
src += sstride;
dst += dstride;
}
}
// Process blocks of width 4, horizontal filter, two lines at time.
static void SIMD_FUNC(clpf_hblock_hbd4)(uint16_t *dst, const uint16_t *src,
int dstride, int sstride, int sizey,
unsigned int strength,
unsigned int adjdamp) {
int y;
for (y = 0; y < sizey; y += 2) {
const v128 a = v128_from_v64(v64_load_unaligned(src - 2),
v64_load_unaligned(src - 2 + sstride));
const v128 b = v128_from_v64(v64_load_unaligned(src - 1),
v64_load_unaligned(src - 1 + sstride));
const v128 c = v128_from_v64(v64_load_unaligned(src + 1),
v64_load_unaligned(src + 1 + sstride));
const v128 d = v128_from_v64(v64_load_unaligned(src + 2),
v64_load_unaligned(src + 2 + sstride));
calc_hdelta_hbd4(v128_from_v64(v64_load_unaligned(src),
v64_load_unaligned(src + sstride)),
a, b, c, d, dst, strength, adjdamp, dstride);
src += sstride * 2;
dst += dstride * 2;
}
}
// Process blocks of width 8, horizontal filter, two lines at time.
static void SIMD_FUNC(clpf_hblock_hbd)(uint16_t *dst, const uint16_t *src,
int dstride, int sstride, int sizey,
unsigned int strength,
unsigned int adjdamp) {
int y;
for (y = 0; y < sizey; y++) {
const v128 o = v128_load_aligned(src);
const v128 a = v128_load_unaligned(src - 2);
const v128 b = v128_load_unaligned(src - 1);
const v128 c = v128_load_unaligned(src + 1);
const v128 d = v128_load_unaligned(src + 2);
calc_hdelta_hbd8(o, a, b, c, d, dst, strength, adjdamp);
src += sstride;
dst += dstride;
}
}
void SIMD_FUNC(aom_clpf_block_hbd)(uint16_t *dst, const uint16_t *src,
int dstride, int sstride, int sizex,
int sizey, unsigned int strength,
unsigned int dmp) {
if ((sizex != 4 && sizex != 8) || ((sizey & 1) && sizex == 4)) {
// Fallback to C for odd sizes:
// * block width not 4 or 8
// * block heights not a multiple of 2 if the block width is 4
aom_clpf_block_hbd_c(dst, src, dstride, sstride, sizex, sizey, strength,
dmp);
} else {
(sizex == 4 ? SIMD_FUNC(clpf_block_hbd4) : SIMD_FUNC(clpf_block_hbd))(
dst, src, dstride, sstride, sizey, strength, dmp - get_msb(strength));
}
}
void SIMD_FUNC(aom_clpf_hblock_hbd)(uint16_t *dst, const uint16_t *src,
int dstride, int sstride, int sizex,
int sizey, unsigned int strength,
unsigned int dmp) {
if ((sizex != 4 && sizex != 8) || ((sizey & 1) && sizex == 4)) {
// Fallback to C for odd sizes:
// * block width not 4 or 8
// * block heights not a multiple of 2 if the block width is 4
aom_clpf_hblock_hbd_c(dst, src, dstride, sstride, sizex, sizey, strength,
dmp);
} else {
(sizex == 4 ? SIMD_FUNC(clpf_hblock_hbd4) : SIMD_FUNC(clpf_hblock_hbd))(
dst, src, dstride, sstride, sizey, strength, dmp - get_msb(strength));
}
}

14
third_party/aom/av1/common/clpf_sse2.c vendored Normal file
View file

@ -0,0 +1,14 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "aom_dsp/aom_simd.h"
#define SIMD_FUNC(name) name##_sse2
#include "./clpf_simd.h"

14
third_party/aom/av1/common/clpf_sse4.c vendored Normal file
View file

@ -0,0 +1,14 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "aom_dsp/aom_simd.h"
#define SIMD_FUNC(name) name##_sse4_1
#include "./clpf_simd.h"

14
third_party/aom/av1/common/clpf_ssse3.c vendored Normal file
View file

@ -0,0 +1,14 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "aom_dsp/aom_simd.h"
#define SIMD_FUNC(name) name##_ssse3
#include "./clpf_simd.h"

64
third_party/aom/av1/common/common.h vendored Normal file
View file

@ -0,0 +1,64 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_COMMON_H_
#define AV1_COMMON_COMMON_H_
/* Interface header for common constant data structures and lookup tables */
#include <assert.h>
#include "aom_dsp/aom_dsp_common.h"
#include "aom_mem/aom_mem.h"
#include "aom/aom_integer.h"
#include "aom_ports/bitops.h"
#ifdef __cplusplus
extern "C" {
#endif
#define PI 3.141592653589793238462643383279502884
// Only need this for fixed-size arrays, for structs just assign.
#define av1_copy(dest, src) \
{ \
assert(sizeof(dest) == sizeof(src)); \
memcpy(dest, src, sizeof(src)); \
}
// Use this for variably-sized arrays.
#define av1_copy_array(dest, src, n) \
{ \
assert(sizeof(*(dest)) == sizeof(*(src))); \
memcpy(dest, src, n * sizeof(*(src))); \
}
#define av1_zero(dest) memset(&(dest), 0, sizeof(dest))
#define av1_zero_array(dest, n) memset(dest, 0, n * sizeof(*(dest)))
static INLINE int get_unsigned_bits(unsigned int num_values) {
return num_values > 0 ? get_msb(num_values) + 1 : 0;
}
#define CHECK_MEM_ERROR(cm, lval, expr) \
AOM_CHECK_MEM_ERROR(&cm->error, lval, expr)
// TODO(yaowu: validate the usage of these codes or develop new ones.)
#define AV1_SYNC_CODE_0 0x49
#define AV1_SYNC_CODE_1 0x83
#define AV1_SYNC_CODE_2 0x43
#define AOM_FRAME_MARKER 0x2
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_COMMON_H_

1405
third_party/aom/av1/common/common_data.h vendored Normal file
View file

File diff suppressed because it is too large Load diff

775
third_party/aom/av1/common/convolve.c vendored Normal file
View file

@ -0,0 +1,775 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <assert.h>
#include <string.h>
#include "./aom_dsp_rtcd.h"
#include "./av1_rtcd.h"
#include "av1/common/convolve.h"
#include "av1/common/filter.h"
#include "av1/common/onyxc_int.h"
#include "aom_dsp/aom_dsp_common.h"
#include "aom_ports/mem.h"
#define MAX_BLOCK_WIDTH (MAX_SB_SIZE)
#define MAX_BLOCK_HEIGHT (MAX_SB_SIZE)
#define MAX_STEP (32)
void av1_convolve_horiz_c(const uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, int w, int h,
const InterpFilterParams filter_params,
const int subpel_x_q4, int x_step_q4,
ConvolveParams *conv_params) {
int x, y;
int filter_size = filter_params.taps;
assert(conv_params->round == CONVOLVE_OPT_ROUND);
src -= filter_size / 2 - 1;
for (y = 0; y < h; ++y) {
int x_q4 = subpel_x_q4;
for (x = 0; x < w; ++x) {
const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
filter_params, x_q4 & SUBPEL_MASK);
int k, sum = 0;
for (k = 0; k < filter_size; ++k) sum += src_x[k] * x_filter[k];
sum = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
if (conv_params->ref)
dst[x] = ROUND_POWER_OF_TWO(dst[x] + sum, 1);
else
dst[x] = sum;
x_q4 += x_step_q4;
}
src += src_stride;
dst += dst_stride;
}
}
void av1_convolve_vert_c(const uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, int w, int h,
const InterpFilterParams filter_params,
const int subpel_y_q4, int y_step_q4,
ConvolveParams *conv_params) {
int x, y;
int filter_size = filter_params.taps;
assert(conv_params->round == CONVOLVE_OPT_ROUND);
src -= src_stride * (filter_size / 2 - 1);
for (x = 0; x < w; ++x) {
int y_q4 = subpel_y_q4;
for (y = 0; y < h; ++y) {
const uint8_t *const src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
filter_params, y_q4 & SUBPEL_MASK);
int k, sum = 0;
for (k = 0; k < filter_size; ++k)
sum += src_y[k * src_stride] * y_filter[k];
sum = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
if (conv_params->ref)
dst[y * dst_stride] = ROUND_POWER_OF_TWO(dst[y * dst_stride] + sum, 1);
else
dst[y * dst_stride] = sum;
y_q4 += y_step_q4;
}
++src;
++dst;
}
}
static void convolve_copy(const uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, int w, int h,
ConvolveParams *conv_params) {
assert(conv_params->round == CONVOLVE_OPT_ROUND);
if (conv_params->ref == 0) {
int r;
for (r = 0; r < h; ++r) {
memcpy(dst, src, w);
src += src_stride;
dst += dst_stride;
}
} else {
int r, c;
for (r = 0; r < h; ++r) {
for (c = 0; c < w; ++c) {
dst[c] = clip_pixel(ROUND_POWER_OF_TWO(dst[c] + src[c], 1));
}
src += src_stride;
dst += dst_stride;
}
}
}
void av1_convolve_horiz_facade(const uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, int w, int h,
const InterpFilterParams filter_params,
const int subpel_x_q4, int x_step_q4,
ConvolveParams *conv_params) {
assert(conv_params->round == CONVOLVE_OPT_ROUND);
if (filter_params.taps == SUBPEL_TAPS) {
const int16_t *filter_x =
av1_get_interp_filter_subpel_kernel(filter_params, subpel_x_q4);
if (conv_params->ref == 0)
aom_convolve8_horiz(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
NULL, -1, w, h);
else
aom_convolve8_avg_horiz(src, src_stride, dst, dst_stride, filter_x,
x_step_q4, NULL, -1, w, h);
} else {
av1_convolve_horiz(src, src_stride, dst, dst_stride, w, h, filter_params,
subpel_x_q4, x_step_q4, conv_params);
}
}
void av1_convolve_horiz_facade_c(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride, int w, int h,
const InterpFilterParams filter_params,
const int subpel_x_q4, int x_step_q4,
ConvolveParams *conv_params) {
assert(conv_params->round == CONVOLVE_OPT_ROUND);
if (filter_params.taps == SUBPEL_TAPS) {
const int16_t *filter_x =
av1_get_interp_filter_subpel_kernel(filter_params, subpel_x_q4);
if (conv_params->ref == 0)
aom_convolve8_horiz_c(src, src_stride, dst, dst_stride, filter_x,
x_step_q4, NULL, -1, w, h);
else
aom_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter_x,
x_step_q4, NULL, -1, w, h);
} else {
av1_convolve_horiz_c(src, src_stride, dst, dst_stride, w, h, filter_params,
subpel_x_q4, x_step_q4, conv_params);
}
}
void av1_convolve_vert_facade(const uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, int w, int h,
const InterpFilterParams filter_params,
const int subpel_y_q4, int y_step_q4,
ConvolveParams *conv_params) {
assert(conv_params->round == CONVOLVE_OPT_ROUND);
if (filter_params.taps == SUBPEL_TAPS) {
const int16_t *filter_y =
av1_get_interp_filter_subpel_kernel(filter_params, subpel_y_q4);
if (conv_params->ref == 0) {
aom_convolve8_vert(src, src_stride, dst, dst_stride, NULL, -1, filter_y,
y_step_q4, w, h);
} else {
aom_convolve8_avg_vert(src, src_stride, dst, dst_stride, NULL, -1,
filter_y, y_step_q4, w, h);
}
} else {
av1_convolve_vert(src, src_stride, dst, dst_stride, w, h, filter_params,
subpel_y_q4, y_step_q4, conv_params);
}
}
void av1_convolve_vert_facade_c(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride, int w, int h,
const InterpFilterParams filter_params,
const int subpel_y_q4, int y_step_q4,
ConvolveParams *conv_params) {
assert(conv_params->round == CONVOLVE_OPT_ROUND);
if (filter_params.taps == SUBPEL_TAPS) {
const int16_t *filter_y =
av1_get_interp_filter_subpel_kernel(filter_params, subpel_y_q4);
if (conv_params->ref == 0) {
aom_convolve8_vert_c(src, src_stride, dst, dst_stride, NULL, -1, filter_y,
y_step_q4, w, h);
} else {
aom_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, NULL, -1,
filter_y, y_step_q4, w, h);
}
} else {
av1_convolve_vert_c(src, src_stride, dst, dst_stride, w, h, filter_params,
subpel_y_q4, y_step_q4, conv_params);
}
}
#if CONFIG_CONVOLVE_ROUND
void av1_convolve_rounding(const int32_t *src, int src_stride, uint8_t *dst,
int dst_stride, int w, int h, int bits) {
int r, c;
for (r = 0; r < h; ++r) {
for (c = 0; c < w; ++c) {
dst[r * dst_stride + c] =
clip_pixel(ROUND_POWER_OF_TWO_SIGNED(src[r * src_stride + c], bits));
}
}
}
void av1_convolve_2d(const uint8_t *src, int src_stride, CONV_BUF_TYPE *dst,
int dst_stride, int w, int h,
InterpFilterParams *filter_params_x,
InterpFilterParams *filter_params_y, const int subpel_x_q4,
const int subpel_y_q4, ConvolveParams *conv_params) {
int x, y, k;
CONV_BUF_TYPE im_block[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) * MAX_SB_SIZE];
int im_h = h + filter_params_y->taps - 1;
int im_stride = w;
const int fo_vert = filter_params_y->taps / 2 - 1;
const int fo_horiz = filter_params_x->taps / 2 - 1;
(void)conv_params;
// horizontal filter
const uint8_t *src_horiz = src - fo_vert * src_stride;
const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
*filter_params_x, subpel_x_q4 & SUBPEL_MASK);
for (y = 0; y < im_h; ++y) {
for (x = 0; x < w; ++x) {
CONV_BUF_TYPE sum = 0;
for (k = 0; k < filter_params_x->taps; ++k) {
sum += x_filter[k] * src_horiz[y * src_stride + x - fo_horiz + k];
}
#if CONFIG_COMPOUND_ROUND
im_block[y * im_stride + x] =
clip_pixel(ROUND_POWER_OF_TWO_SIGNED(sum, conv_params->round_0));
#else
im_block[y * im_stride + x] =
ROUND_POWER_OF_TWO_SIGNED(sum, conv_params->round_0);
#endif
}
}
// vertical filter
CONV_BUF_TYPE *src_vert = im_block + fo_vert * im_stride;
const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
*filter_params_y, subpel_y_q4 & SUBPEL_MASK);
for (y = 0; y < h; ++y) {
for (x = 0; x < w; ++x) {
CONV_BUF_TYPE sum = 0;
for (k = 0; k < filter_params_y->taps; ++k) {
sum += y_filter[k] * src_vert[(y - fo_vert + k) * im_stride + x];
}
dst[y * dst_stride + x] +=
ROUND_POWER_OF_TWO_SIGNED(sum, conv_params->round_1);
}
}
}
static INLINE void transpose_uint8(uint8_t *dst, int dst_stride,
const uint8_t *src, int src_stride, int w,
int h) {
int r, c;
for (r = 0; r < h; ++r)
for (c = 0; c < w; ++c)
dst[c * (dst_stride) + r] = src[r * (src_stride) + c];
}
static INLINE void transpose_int32(int32_t *dst, int dst_stride,
const int32_t *src, int src_stride, int w,
int h) {
int r, c;
for (r = 0; r < h; ++r)
for (c = 0; c < w; ++c)
dst[c * (dst_stride) + r] = src[r * (src_stride) + c];
}
void av1_convolve_2d_facade(const uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, int w, int h,
const InterpFilter *interp_filter,
const int subpel_x_q4, int x_step_q4,
const int subpel_y_q4, int y_step_q4,
ConvolveParams *conv_params) {
(void)x_step_q4;
(void)y_step_q4;
(void)dst;
(void)dst_stride;
#if CONFIG_DUAL_FILTER
InterpFilterParams filter_params_x =
av1_get_interp_filter_params(interp_filter[1 + 2 * conv_params->ref]);
InterpFilterParams filter_params_y =
av1_get_interp_filter_params(interp_filter[0 + 2 * conv_params->ref]);
if (filter_params_x.interp_filter == MULTITAP_SHARP &&
filter_params_y.interp_filter == MULTITAP_SHARP) {
// Avoid two directions both using 12-tap filter.
// This will reduce hardware implementation cost.
filter_params_y = av1_get_interp_filter_params(EIGHTTAP_SHARP);
}
#else
InterpFilterParams filter_params_x =
av1_get_interp_filter_params(*interp_filter);
InterpFilterParams filter_params_y =
av1_get_interp_filter_params(*interp_filter);
#endif
if (filter_params_y.taps < filter_params_x.taps) {
uint8_t tr_src[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) *
(MAX_SB_SIZE + MAX_FILTER_TAP - 1)];
int tr_src_stride = MAX_SB_SIZE + MAX_FILTER_TAP - 1;
CONV_BUF_TYPE tr_dst[MAX_SB_SIZE * MAX_SB_SIZE];
int tr_dst_stride = MAX_SB_SIZE;
int fo_vert = filter_params_y.taps / 2 - 1;
int fo_horiz = filter_params_x.taps / 2 - 1;
transpose_uint8(tr_src, tr_src_stride,
src - fo_vert * src_stride - fo_horiz, src_stride,
w + filter_params_x.taps - 1, h + filter_params_y.taps - 1);
transpose_int32(tr_dst, tr_dst_stride, conv_params->dst,
conv_params->dst_stride, w, h);
// horizontal and vertical parameters are swapped because of the transpose
av1_convolve_2d(tr_src + fo_horiz * tr_src_stride + fo_vert, tr_src_stride,
tr_dst, tr_dst_stride, h, w, &filter_params_y,
&filter_params_x, subpel_y_q4, subpel_x_q4, conv_params);
transpose_int32(conv_params->dst, conv_params->dst_stride, tr_dst,
tr_dst_stride, h, w);
} else {
av1_convolve_2d(src, src_stride, conv_params->dst, conv_params->dst_stride,
w, h, &filter_params_x, &filter_params_y, subpel_x_q4,
subpel_y_q4, conv_params);
}
}
#endif // CONFIG_CONVOLVE_ROUND
typedef void (*ConvolveFunc)(const uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, int w, int h,
const InterpFilterParams filter_params,
const int subpel_q4, int step_q4,
ConvolveParams *conv_params);
static void convolve_helper(const uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, int w, int h,
#if CONFIG_DUAL_FILTER
const InterpFilter *interp_filter,
#else
const InterpFilter interp_filter,
#endif
const int subpel_x_q4, int x_step_q4,
const int subpel_y_q4, int y_step_q4,
ConvolveParams *conv_params,
ConvolveFunc convolve_horiz,
ConvolveFunc convolve_vert) {
int ignore_horiz = x_step_q4 == 16 && subpel_x_q4 == 0;
int ignore_vert = y_step_q4 == 16 && subpel_y_q4 == 0;
#if CONFIG_DUAL_FILTER
InterpFilterParams filter_params_x =
av1_get_interp_filter_params(interp_filter[1 + 2 * conv_params->ref]);
InterpFilterParams filter_params_y =
av1_get_interp_filter_params(interp_filter[0 + 2 * conv_params->ref]);
InterpFilterParams filter_params;
#else
InterpFilterParams filter_params =
av1_get_interp_filter_params(interp_filter);
#endif
assert(conv_params->round == CONVOLVE_OPT_ROUND);
assert(w <= MAX_BLOCK_WIDTH);
assert(h <= MAX_BLOCK_HEIGHT);
assert(y_step_q4 <= MAX_STEP);
assert(x_step_q4 <= MAX_STEP);
if (ignore_horiz && ignore_vert) {
convolve_copy(src, src_stride, dst, dst_stride, w, h, conv_params);
} else if (ignore_vert) {
#if CONFIG_DUAL_FILTER
filter_params = filter_params_x;
#endif
assert(filter_params.taps <= MAX_FILTER_TAP);
convolve_horiz(src, src_stride, dst, dst_stride, w, h, filter_params,
subpel_x_q4, x_step_q4, conv_params);
} else if (ignore_horiz) {
#if CONFIG_DUAL_FILTER
filter_params = filter_params_y;
#endif
assert(filter_params.taps <= MAX_FILTER_TAP);
convolve_vert(src, src_stride, dst, dst_stride, w, h, filter_params,
subpel_y_q4, y_step_q4, conv_params);
} else {
// temp's size is set to a 256 aligned value to facilitate SIMD
// implementation. The value is greater than (maximum possible intermediate
// height or width) * MAX_SB_SIZE
DECLARE_ALIGNED(16, uint8_t,
temp[((MAX_SB_SIZE * 2 + 16) + 16) * MAX_SB_SIZE]);
int max_intermediate_size = ((MAX_SB_SIZE * 2 + 16) + 16);
int filter_size;
#if CONFIG_DUAL_FILTER
if (interp_filter[0 + 2 * conv_params->ref] == MULTITAP_SHARP &&
interp_filter[1 + 2 * conv_params->ref] == MULTITAP_SHARP) {
// Avoid two directions both using 12-tap filter.
// This will reduce hardware implementation cost.
filter_params_y = av1_get_interp_filter_params(EIGHTTAP_SHARP);
}
// we do filter with fewer taps first to reduce hardware implementation
// complexity
if (filter_params_y.taps < filter_params_x.taps) {
int intermediate_width;
int temp_stride = max_intermediate_size;
ConvolveParams temp_conv_params;
temp_conv_params.ref = 0;
temp_conv_params.round = CONVOLVE_OPT_ROUND;
filter_params = filter_params_y;
filter_size = filter_params_x.taps;
intermediate_width =
(((w - 1) * x_step_q4 + subpel_x_q4) >> SUBPEL_BITS) + filter_size;
assert(intermediate_width <= max_intermediate_size);
assert(filter_params.taps <= MAX_FILTER_TAP);
convolve_vert(src - (filter_size / 2 - 1), src_stride, temp, temp_stride,
intermediate_width, h, filter_params, subpel_y_q4,
y_step_q4, &temp_conv_params);
filter_params = filter_params_x;
assert(filter_params.taps <= MAX_FILTER_TAP);
convolve_horiz(temp + (filter_size / 2 - 1), temp_stride, dst, dst_stride,
w, h, filter_params, subpel_x_q4, x_step_q4, conv_params);
} else
#endif // CONFIG_DUAL_FILTER
{
int intermediate_height;
int temp_stride = MAX_SB_SIZE;
ConvolveParams temp_conv_params;
temp_conv_params.ref = 0;
temp_conv_params.round = CONVOLVE_OPT_ROUND;
#if CONFIG_DUAL_FILTER
filter_params = filter_params_x;
filter_size = filter_params_y.taps;
#else
filter_size = filter_params.taps;
#endif
intermediate_height =
(((h - 1) * y_step_q4 + subpel_y_q4) >> SUBPEL_BITS) + filter_size;
assert(intermediate_height <= max_intermediate_size);
(void)max_intermediate_size;
assert(filter_params.taps <= MAX_FILTER_TAP);
convolve_horiz(src - src_stride * (filter_size / 2 - 1), src_stride, temp,
temp_stride, w, intermediate_height, filter_params,
subpel_x_q4, x_step_q4, &temp_conv_params);
#if CONFIG_DUAL_FILTER
filter_params = filter_params_y;
#endif
assert(filter_params.taps <= MAX_FILTER_TAP);
convolve_vert(temp + temp_stride * (filter_size / 2 - 1), temp_stride,
dst, dst_stride, w, h, filter_params, subpel_y_q4,
y_step_q4, conv_params);
}
}
}
void av1_convolve(const uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, int w, int h,
#if CONFIG_DUAL_FILTER
const InterpFilter *interp_filter,
#else
const InterpFilter interp_filter,
#endif
const int subpel_x_q4, int x_step_q4, const int subpel_y_q4,
int y_step_q4, ConvolveParams *conv_params) {
convolve_helper(src, src_stride, dst, dst_stride, w, h, interp_filter,
subpel_x_q4, x_step_q4, subpel_y_q4, y_step_q4, conv_params,
av1_convolve_horiz_facade, av1_convolve_vert_facade);
}
void av1_convolve_c(const uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, int w, int h,
#if CONFIG_DUAL_FILTER
const InterpFilter *interp_filter,
#else
const InterpFilter interp_filter,
#endif
const int subpel_x_q4, int x_step_q4, const int subpel_y_q4,
int y_step_q4, ConvolveParams *conv_params) {
convolve_helper(src, src_stride, dst, dst_stride, w, h, interp_filter,
subpel_x_q4, x_step_q4, subpel_y_q4, y_step_q4, conv_params,
av1_convolve_horiz_facade_c, av1_convolve_vert_facade_c);
}
void av1_lowbd_convolve_init_c(void) {
// A placeholder for SIMD initialization
return;
}
void av1_highbd_convolve_init_c(void) {
// A placeholder for SIMD initialization
return;
}
void av1_convolve_init(AV1_COMMON *cm) {
#if CONFIG_HIGHBITDEPTH
if (cm->use_highbitdepth)
av1_highbd_convolve_init();
else
av1_lowbd_convolve_init();
#else
(void)cm;
av1_lowbd_convolve_init();
#endif
return;
}
#if CONFIG_HIGHBITDEPTH
void av1_highbd_convolve_horiz_c(const uint16_t *src, int src_stride,
uint16_t *dst, int dst_stride, int w, int h,
const InterpFilterParams filter_params,
const int subpel_x_q4, int x_step_q4, int avg,
int bd) {
int x, y;
int filter_size = filter_params.taps;
src -= filter_size / 2 - 1;
for (y = 0; y < h; ++y) {
int x_q4 = subpel_x_q4;
for (x = 0; x < w; ++x) {
const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
filter_params, x_q4 & SUBPEL_MASK);
int k, sum = 0;
for (k = 0; k < filter_size; ++k) sum += src_x[k] * x_filter[k];
if (avg)
dst[x] = ROUND_POWER_OF_TWO(
dst[x] +
clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd),
1);
else
dst[x] = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
x_q4 += x_step_q4;
}
src += src_stride;
dst += dst_stride;
}
}
void av1_highbd_convolve_vert_c(const uint16_t *src, int src_stride,
uint16_t *dst, int dst_stride, int w, int h,
const InterpFilterParams filter_params,
const int subpel_y_q4, int y_step_q4, int avg,
int bd) {
int x, y;
int filter_size = filter_params.taps;
src -= src_stride * (filter_size / 2 - 1);
for (x = 0; x < w; ++x) {
int y_q4 = subpel_y_q4;
for (y = 0; y < h; ++y) {
const uint16_t *const src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
filter_params, y_q4 & SUBPEL_MASK);
int k, sum = 0;
for (k = 0; k < filter_size; ++k)
sum += src_y[k * src_stride] * y_filter[k];
if (avg) {
dst[y * dst_stride] = ROUND_POWER_OF_TWO(
dst[y * dst_stride] +
clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd),
1);
} else {
dst[y * dst_stride] =
clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
}
y_q4 += y_step_q4;
}
++src;
++dst;
}
}
static void highbd_convolve_copy(const uint16_t *src, int src_stride,
uint16_t *dst, int dst_stride, int w, int h,
int avg, int bd) {
if (avg == 0) {
int r;
for (r = 0; r < h; ++r) {
memcpy(dst, src, w * sizeof(*src));
src += src_stride;
dst += dst_stride;
}
} else {
int r, c;
for (r = 0; r < h; ++r) {
for (c = 0; c < w; ++c) {
dst[c] = clip_pixel_highbd(ROUND_POWER_OF_TWO(dst[c] + src[c], 1), bd);
}
src += src_stride;
dst += dst_stride;
}
}
}
void av1_highbd_convolve_horiz_facade(const uint8_t *src8, int src_stride,
uint8_t *dst8, int dst_stride, int w,
int h,
const InterpFilterParams filter_params,
const int subpel_x_q4, int x_step_q4,
int avg, int bd) {
uint16_t *src = CONVERT_TO_SHORTPTR(src8);
uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
if (filter_params.taps == SUBPEL_TAPS) {
const int16_t *filter_x =
av1_get_interp_filter_subpel_kernel(filter_params, subpel_x_q4);
if (avg == 0)
aom_highbd_convolve8_horiz(src8, src_stride, dst8, dst_stride, filter_x,
x_step_q4, NULL, -1, w, h, bd);
else
aom_highbd_convolve8_avg_horiz(src8, src_stride, dst8, dst_stride,
filter_x, x_step_q4, NULL, -1, w, h, bd);
} else {
av1_highbd_convolve_horiz(src, src_stride, dst, dst_stride, w, h,
filter_params, subpel_x_q4, x_step_q4, avg, bd);
}
}
void av1_highbd_convolve_vert_facade(const uint8_t *src8, int src_stride,
uint8_t *dst8, int dst_stride, int w,
int h,
const InterpFilterParams filter_params,
const int subpel_y_q4, int y_step_q4,
int avg, int bd) {
uint16_t *src = CONVERT_TO_SHORTPTR(src8);
uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
if (filter_params.taps == SUBPEL_TAPS) {
const int16_t *filter_y =
av1_get_interp_filter_subpel_kernel(filter_params, subpel_y_q4);
if (avg == 0) {
aom_highbd_convolve8_vert(src8, src_stride, dst8, dst_stride, NULL, -1,
filter_y, y_step_q4, w, h, bd);
} else {
aom_highbd_convolve8_avg_vert(src8, src_stride, dst8, dst_stride, NULL,
-1, filter_y, y_step_q4, w, h, bd);
}
} else {
av1_highbd_convolve_vert(src, src_stride, dst, dst_stride, w, h,
filter_params, subpel_y_q4, y_step_q4, avg, bd);
}
}
void av1_highbd_convolve(const uint8_t *src8, int src_stride, uint8_t *dst8,
int dst_stride, int w, int h,
#if CONFIG_DUAL_FILTER
const InterpFilter *interp_filter,
#else
const InterpFilter interp_filter,
#endif
const int subpel_x_q4, int x_step_q4,
const int subpel_y_q4, int y_step_q4, int ref_idx,
int bd) {
uint16_t *src = CONVERT_TO_SHORTPTR(src8);
uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
int ignore_horiz = x_step_q4 == 16 && subpel_x_q4 == 0;
int ignore_vert = y_step_q4 == 16 && subpel_y_q4 == 0;
assert(w <= MAX_BLOCK_WIDTH);
assert(h <= MAX_BLOCK_HEIGHT);
assert(y_step_q4 <= MAX_STEP);
assert(x_step_q4 <= MAX_STEP);
if (ignore_horiz && ignore_vert) {
highbd_convolve_copy(src, src_stride, dst, dst_stride, w, h, ref_idx, bd);
} else if (ignore_vert) {
#if CONFIG_DUAL_FILTER
InterpFilterParams filter_params =
av1_get_interp_filter_params(interp_filter[1 + 2 * ref_idx]);
#else
InterpFilterParams filter_params =
av1_get_interp_filter_params(interp_filter);
#endif
av1_highbd_convolve_horiz_facade(src8, src_stride, dst8, dst_stride, w, h,
filter_params, subpel_x_q4, x_step_q4,
ref_idx, bd);
} else if (ignore_horiz) {
#if CONFIG_DUAL_FILTER
InterpFilterParams filter_params =
av1_get_interp_filter_params(interp_filter[0 + 2 * ref_idx]);
#else
InterpFilterParams filter_params =
av1_get_interp_filter_params(interp_filter);
#endif
av1_highbd_convolve_vert_facade(src8, src_stride, dst8, dst_stride, w, h,
filter_params, subpel_y_q4, y_step_q4,
ref_idx, bd);
} else {
// temp's size is set to a 256 aligned value to facilitate SIMD
// implementation. The value is greater than (maximum possible intermediate
// height or width) * MAX_SB_SIZE
DECLARE_ALIGNED(16, uint16_t,
temp[((MAX_SB_SIZE * 2 + 16) + 16) * MAX_SB_SIZE]);
uint8_t *temp8 = CONVERT_TO_BYTEPTR(temp);
int max_intermediate_size = ((MAX_SB_SIZE * 2 + 16) + 16);
int filter_size;
InterpFilterParams filter_params;
#if CONFIG_DUAL_FILTER
InterpFilterParams filter_params_x =
av1_get_interp_filter_params(interp_filter[1 + 2 * ref_idx]);
InterpFilterParams filter_params_y =
av1_get_interp_filter_params(interp_filter[0 + 2 * ref_idx]);
if (interp_filter[0 + 2 * ref_idx] == MULTITAP_SHARP &&
interp_filter[1 + 2 * ref_idx] == MULTITAP_SHARP) {
// Avoid two directions both using 12-tap filter.
// This will reduce hardware implementation cost.
filter_params_y = av1_get_interp_filter_params(EIGHTTAP_SHARP);
}
#endif
#if CONFIG_DUAL_FILTER
if (filter_params_y.taps < filter_params_x.taps) {
int intermediate_width;
int temp_stride = max_intermediate_size;
filter_params = filter_params_y;
filter_size = filter_params_x.taps;
intermediate_width =
(((w - 1) * x_step_q4 + subpel_x_q4) >> SUBPEL_BITS) + filter_size;
assert(intermediate_width <= max_intermediate_size);
assert(filter_params.taps <= MAX_FILTER_TAP);
av1_highbd_convolve_vert_facade(
src8 - (filter_size / 2 - 1), src_stride, temp8, temp_stride,
intermediate_width, h, filter_params, subpel_y_q4, y_step_q4, 0, bd);
filter_params = filter_params_x;
assert(filter_params.taps <= MAX_FILTER_TAP);
av1_highbd_convolve_horiz_facade(
temp8 + (filter_size / 2 - 1), temp_stride, dst8, dst_stride, w, h,
filter_params, subpel_x_q4, x_step_q4, ref_idx, bd);
} else
#endif // CONFIG_DUAL_FILTER
{
int intermediate_height;
int temp_stride = MAX_SB_SIZE;
#if CONFIG_DUAL_FILTER
filter_params = filter_params_x;
filter_size = filter_params_y.taps;
#else
filter_params = av1_get_interp_filter_params(interp_filter);
filter_size = filter_params.taps;
#endif
intermediate_height =
(((h - 1) * y_step_q4 + subpel_y_q4) >> SUBPEL_BITS) + filter_size;
assert(intermediate_height <= max_intermediate_size);
(void)max_intermediate_size;
av1_highbd_convolve_horiz_facade(
src8 - src_stride * (filter_size / 2 - 1), src_stride, temp8,
temp_stride, w, intermediate_height, filter_params, subpel_x_q4,
x_step_q4, 0, bd);
#if CONFIG_DUAL_FILTER
filter_params = filter_params_y;
#endif
filter_size = filter_params.taps;
assert(filter_params.taps <= MAX_FILTER_TAP);
av1_highbd_convolve_vert_facade(
temp8 + temp_stride * (filter_size / 2 - 1), temp_stride, dst8,
dst_stride, w, h, filter_params, subpel_y_q4, y_step_q4, ref_idx, bd);
}
}
}
#endif // CONFIG_HIGHBITDEPTH

119
third_party/aom/av1/common/convolve.h vendored Normal file
View file

@ -0,0 +1,119 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_AV1_CONVOLVE_H_
#define AV1_COMMON_AV1_CONVOLVE_H_
#include "av1/common/filter.h"
#ifdef __cplusplus
extern "C" {
#endif
typedef enum CONVOLVE_OPT {
// indicate the results in dst buf is rounded by FILTER_BITS or not
CONVOLVE_OPT_ROUND,
CONVOLVE_OPT_NO_ROUND,
} CONVOLVE_OPT;
typedef int32_t CONV_BUF_TYPE;
typedef struct ConvolveParams {
int ref;
CONVOLVE_OPT round;
CONV_BUF_TYPE *dst;
int dst_stride;
int round_0;
int round_1;
int plane;
} ConvolveParams;
static INLINE ConvolveParams get_conv_params(int ref, int plane) {
ConvolveParams conv_params;
conv_params.ref = ref;
conv_params.round = CONVOLVE_OPT_ROUND;
conv_params.plane = plane;
return conv_params;
}
struct AV1Common;
void av1_convolve_init(struct AV1Common *cm);
#if CONFIG_CONVOLVE_ROUND
void av1_convolve_2d(const uint8_t *src, int src_stride, CONV_BUF_TYPE *dst,
int dst_stride, int w, int h,
InterpFilterParams *filter_params_x,
InterpFilterParams *filter_params_y, const int subpel_x_q4,
const int subpel_y_q4, ConvolveParams *conv_params);
void av1_convolve_2d_facade(const uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, int w, int h,
const InterpFilter *interp_filter,
const int subpel_x_q4, int x_step_q4,
const int subpel_y_q4, int y_step_q4,
ConvolveParams *conv_params);
static INLINE ConvolveParams get_conv_params_no_round(int ref, int plane,
int32_t *dst,
int dst_stride) {
ConvolveParams conv_params;
conv_params.ref = ref;
conv_params.round = CONVOLVE_OPT_NO_ROUND;
#if CONFIG_COMPOUND_ROUND
conv_params.round_0 = FILTER_BITS;
#else
conv_params.round_0 = 5;
#endif
conv_params.round_1 = 0;
conv_params.dst = dst;
conv_params.dst_stride = dst_stride;
conv_params.plane = plane;
return conv_params;
}
void av1_convolve_rounding(const int32_t *src, int src_stride, uint8_t *dst,
int dst_stride, int w, int h, int bits);
#endif // CONFIG_CONVOLVE_ROUND
void av1_convolve(const uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, int w, int h,
#if CONFIG_DUAL_FILTER
const InterpFilter *interp_filter,
#else
const InterpFilter interp_filter,
#endif
const int subpel_x, int xstep, const int subpel_y, int ystep,
ConvolveParams *conv_params);
void av1_convolve_c(const uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, int w, int h,
#if CONFIG_DUAL_FILTER
const InterpFilter *interp_filter,
#else
const InterpFilter interp_filter,
#endif
const int subpel_x, int xstep, const int subpel_y,
int ystep, ConvolveParams *conv_params);
#if CONFIG_HIGHBITDEPTH
void av1_highbd_convolve(const uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, int w, int h,
#if CONFIG_DUAL_FILTER
const InterpFilter *interp_filter,
#else
const InterpFilter interp_filter,
#endif
const int subpel_x, int xstep, const int subpel_y,
int ystep, int avg, int bd);
#endif // CONFIG_HIGHBITDEPTH
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_AV1_CONVOLVE_H_

90
third_party/aom/av1/common/debugmodes.c vendored Normal file
View file

@ -0,0 +1,90 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <stdio.h>
#include "av1/common/blockd.h"
#include "av1/common/onyxc_int.h"
static void log_frame_info(AV1_COMMON *cm, const char *str, FILE *f) {
fprintf(f, "%s", str);
fprintf(f, "(Frame %d, Show:%d, Q:%d): \n", cm->current_video_frame,
cm->show_frame, cm->base_qindex);
}
/* This function dereferences a pointer to the mbmi structure
* and uses the passed in member offset to print out the value of an integer
* for each mbmi member value in the mi structure.
*/
static void print_mi_data(AV1_COMMON *cm, FILE *file, const char *descriptor,
size_t member_offset) {
int mi_row, mi_col;
MODE_INFO **mi = cm->mi_grid_visible;
int rows = cm->mi_rows;
int cols = cm->mi_cols;
char prefix = descriptor[0];
log_frame_info(cm, descriptor, file);
for (mi_row = 0; mi_row < rows; mi_row++) {
fprintf(file, "%c ", prefix);
for (mi_col = 0; mi_col < cols; mi_col++) {
fprintf(file, "%2d ", *((int *)((char *)(&mi[0]->mbmi) + member_offset)));
mi++;
}
fprintf(file, "\n");
mi += 8;
}
fprintf(file, "\n");
}
void av1_print_modes_and_motion_vectors(AV1_COMMON *cm, const char *file) {
int mi_row;
int mi_col;
FILE *mvs = fopen(file, "a");
MODE_INFO **mi = cm->mi_grid_visible;
int rows = cm->mi_rows;
int cols = cm->mi_cols;
print_mi_data(cm, mvs, "Partitions:", offsetof(MB_MODE_INFO, sb_type));
print_mi_data(cm, mvs, "Modes:", offsetof(MB_MODE_INFO, mode));
print_mi_data(cm, mvs, "Ref frame:", offsetof(MB_MODE_INFO, ref_frame[0]));
print_mi_data(cm, mvs, "Transform:", offsetof(MB_MODE_INFO, tx_size));
print_mi_data(cm, mvs, "UV Modes:", offsetof(MB_MODE_INFO, uv_mode));
// output skip infomation.
log_frame_info(cm, "Skips:", mvs);
for (mi_row = 0; mi_row < rows; mi_row++) {
fprintf(mvs, "S ");
for (mi_col = 0; mi_col < cols; mi_col++) {
fprintf(mvs, "%2d ", mi[0]->mbmi.skip);
mi++;
}
fprintf(mvs, "\n");
mi += 8;
}
fprintf(mvs, "\n");
// output motion vectors.
log_frame_info(cm, "Vectors ", mvs);
mi = cm->mi_grid_visible;
for (mi_row = 0; mi_row < rows; mi_row++) {
fprintf(mvs, "V ");
for (mi_col = 0; mi_col < cols; mi_col++) {
fprintf(mvs, "%4d:%4d ", mi[0]->mbmi.mv[0].as_mv.row,
mi[0]->mbmi.mv[0].as_mv.col);
mi++;
}
fprintf(mvs, "\n");
mi += 8;
}
fprintf(mvs, "\n");
fclose(mvs);
}

6438
third_party/aom/av1/common/entropy.c vendored Normal file
View file

File diff suppressed because it is too large Load diff

428
third_party/aom/av1/common/entropy.h vendored Normal file
View file

@ -0,0 +1,428 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_ENTROPY_H_
#define AV1_COMMON_ENTROPY_H_
#include "./aom_config.h"
#include "aom/aom_integer.h"
#include "aom_dsp/prob.h"
#include "av1/common/common.h"
#include "av1/common/common_data.h"
#include "av1/common/enums.h"
#ifdef __cplusplus
extern "C" {
#endif
#define DIFF_UPDATE_PROB 252
#define GROUP_DIFF_UPDATE_PROB 252
#if CONFIG_Q_ADAPT_PROBS
#define QCTX_BIN_BITS 2
#define QCTX_BINS (1 << QCTX_BIN_BITS)
#endif // CONFIG_Q_ADAPT_PROBS
#if CONFIG_SUBFRAME_PROB_UPDATE
#define COEF_PROBS_BUFS 16
#endif // CONFIG_SUBFRAME_PROB_UPDATE
// Coefficient token alphabet
#define ZERO_TOKEN 0 // 0 Extra Bits 0+0
#define ONE_TOKEN 1 // 1 Extra Bits 0+1
#define TWO_TOKEN 2 // 2 Extra Bits 0+1
#define THREE_TOKEN 3 // 3 Extra Bits 0+1
#define FOUR_TOKEN 4 // 4 Extra Bits 0+1
#define CATEGORY1_TOKEN 5 // 5-6 Extra Bits 1+1
#define CATEGORY2_TOKEN 6 // 7-10 Extra Bits 2+1
#define CATEGORY3_TOKEN 7 // 11-18 Extra Bits 3+1
#define CATEGORY4_TOKEN 8 // 19-34 Extra Bits 4+1
#define CATEGORY5_TOKEN 9 // 35-66 Extra Bits 5+1
#define CATEGORY6_TOKEN 10 // 67+ Extra Bits 14+1
#define EOB_TOKEN 11 // EOB Extra Bits 0+0
#if CONFIG_NEW_TOKENSET
#define NO_EOB 0 // Not an end-of-block
#define EARLY_EOB 1 // End of block before the last position
#define LAST_EOB 2 // End of block in the last position (implicit)
#define BLOCK_Z_TOKEN 255 // block zero
#define HEAD_TOKENS 5
#define TAIL_TOKENS 9
#define ONE_TOKEN_EOB 1
#define ONE_TOKEN_NEOB 2
#define TWO_TOKEN_PLUS_EOB 3
#define TWO_TOKEN_PLUS_NEOB 4
#endif
#define ENTROPY_TOKENS 12
#define ENTROPY_NODES 11
#if CONFIG_LV_MAP
#define TXB_SKIP_CONTEXTS 13
#define SIG_COEF_CONTEXTS 20
#define EOB_COEF_CONTEXTS 25
#define COEFF_BASE_CONTEXTS 42
#define DC_SIGN_CONTEXTS 3
#define BR_TMP_OFFSET 12
#define BR_REF_CAT 4
#define LEVEL_CONTEXTS (BR_TMP_OFFSET * BR_REF_CAT)
#define NUM_BASE_LEVELS 2
#define COEFF_BASE_RANGE (15 - NUM_BASE_LEVELS)
#define COEFF_CONTEXT_BITS 6
#define COEFF_CONTEXT_MASK ((1 << COEFF_CONTEXT_BITS) - 1)
#endif
DECLARE_ALIGNED(16, extern const uint8_t, av1_pt_energy_class[ENTROPY_TOKENS]);
#define CAT1_MIN_VAL 5
#define CAT2_MIN_VAL 7
#define CAT3_MIN_VAL 11
#define CAT4_MIN_VAL 19
#define CAT5_MIN_VAL 35
#define CAT6_MIN_VAL 67
// Extra bit probabilities.
DECLARE_ALIGNED(16, extern const uint8_t, av1_cat1_prob[1]);
DECLARE_ALIGNED(16, extern const uint8_t, av1_cat2_prob[2]);
DECLARE_ALIGNED(16, extern const uint8_t, av1_cat3_prob[3]);
DECLARE_ALIGNED(16, extern const uint8_t, av1_cat4_prob[4]);
DECLARE_ALIGNED(16, extern const uint8_t, av1_cat5_prob[5]);
DECLARE_ALIGNED(16, extern const uint8_t, av1_cat6_prob[18]);
#if CONFIG_NEW_MULTISYMBOL
extern const aom_cdf_prob *av1_cat1_cdf[];
extern const aom_cdf_prob *av1_cat2_cdf[];
extern const aom_cdf_prob *av1_cat3_cdf[];
extern const aom_cdf_prob *av1_cat4_cdf[];
extern const aom_cdf_prob *av1_cat5_cdf[];
extern const aom_cdf_prob *av1_cat6_cdf[];
#endif
#define EOB_MODEL_TOKEN 3
typedef struct {
#if CONFIG_NEW_MULTISYMBOL
const aom_cdf_prob **cdf;
#else
const aom_prob *prob;
#endif
int len;
int base_val;
const int16_t *cost;
} av1_extra_bit;
// indexed by token value
extern const av1_extra_bit av1_extra_bits[ENTROPY_TOKENS];
static INLINE int av1_get_cat6_extrabits_size(TX_SIZE tx_size,
aom_bit_depth_t bit_depth) {
tx_size = txsize_sqr_up_map[tx_size];
#if CONFIG_TX64X64
// TODO(debargha): Does TX_64X64 require an additional extrabit?
if (tx_size > TX_32X32) tx_size = TX_32X32;
#endif
#if CONFIG_CB4X4
int tx_offset = (tx_size < TX_4X4) ? 0 : (int)(tx_size - TX_4X4);
#else
int tx_offset = (int)(tx_size - TX_4X4);
#endif
int bits = (int)bit_depth + 3 + tx_offset;
#if CONFIG_NEW_MULTISYMBOL
// Round up
bits = AOMMIN((int)sizeof(av1_cat6_prob), ((bits + 3) & ~3));
#endif
assert(bits <= (int)sizeof(av1_cat6_prob));
return bits;
}
#define DCT_MAX_VALUE 16384
#if CONFIG_HIGHBITDEPTH
#define DCT_MAX_VALUE_HIGH10 65536
#define DCT_MAX_VALUE_HIGH12 262144
#endif // CONFIG_HIGHBITDEPTH
/* Coefficients are predicted via a 3-dimensional probability table. */
#define REF_TYPES 2 // intra=0, inter=1
/* Middle dimension reflects the coefficient position within the transform. */
#define COEF_BANDS 6
/* Inside dimension is measure of nearby complexity, that reflects the energy
of nearby coefficients are nonzero. For the first coefficient (DC, unless
block type is 0), we look at the (already encoded) blocks above and to the
left of the current block. The context index is then the number (0,1,or 2)
of these blocks having nonzero coefficients.
After decoding a coefficient, the measure is determined by the size of the
most recently decoded coefficient.
Note that the intuitive meaning of this measure changes as coefficients
are decoded, e.g., prior to the first token, a zero means that my neighbors
are empty while, after the first token, because of the use of end-of-block,
a zero means we just decoded a zero and hence guarantees that a non-zero
coefficient will appear later in this block. However, this shift
in meaning is perfectly OK because our context depends also on the
coefficient band (and since zigzag positions 0, 1, and 2 are in
distinct bands). */
#define COEFF_CONTEXTS 6
#if CONFIG_EC_MULTISYMBOL
#define BLOCKZ_CONTEXTS 3
#endif
#define COEFF_CONTEXTS0 3 // for band 0
#define BAND_COEFF_CONTEXTS(band) \
((band) == 0 ? COEFF_CONTEXTS0 : COEFF_CONTEXTS)
// #define ENTROPY_STATS
typedef unsigned int av1_coeff_count[REF_TYPES][COEF_BANDS][COEFF_CONTEXTS]
[ENTROPY_TOKENS];
typedef unsigned int av1_coeff_stats[REF_TYPES][COEF_BANDS][COEFF_CONTEXTS]
[ENTROPY_NODES][2];
#define SUBEXP_PARAM 4 /* Subexponential code parameter */
#define MODULUS_PARAM 13 /* Modulus parameter */
struct AV1Common;
struct frame_contexts;
void av1_default_coef_probs(struct AV1Common *cm);
void av1_adapt_coef_probs(struct AV1Common *cm);
#if CONFIG_EC_ADAPT
void av1_adapt_coef_cdfs(struct AV1Common *cm, struct frame_contexts *pre_fc);
#endif
#if CONFIG_SUBFRAME_PROB_UPDATE
void av1_partial_adapt_probs(struct AV1Common *cm, int mi_row, int mi_col);
#endif // CONFIG_SUBFRAME_PROB_UPDATE
// This is the index in the scan order beyond which all coefficients for
// 8x8 transform and above are in the top band.
// This macro is currently unused but may be used by certain implementations
#define MAXBAND_INDEX 21
DECLARE_ALIGNED(16, extern const uint8_t,
av1_coefband_trans_8x8plus[MAX_TX_SQUARE]);
DECLARE_ALIGNED(16, extern const uint8_t, av1_coefband_trans_4x8_8x4[32]);
DECLARE_ALIGNED(16, extern const uint8_t, av1_coefband_trans_4x4[16]);
DECLARE_ALIGNED(16, extern const uint16_t, band_count_table[TX_SIZES_ALL][8]);
DECLARE_ALIGNED(16, extern const uint16_t,
band_cum_count_table[TX_SIZES_ALL][8]);
static INLINE const uint8_t *get_band_translate(TX_SIZE tx_size) {
switch (tx_size) {
case TX_4X4: return av1_coefband_trans_4x4;
case TX_8X4:
case TX_4X8: return av1_coefband_trans_4x8_8x4;
default: return av1_coefband_trans_8x8plus;
}
}
// 128 lists of probabilities are stored for the following ONE node probs:
// 1, 3, 5, 7, ..., 253, 255
// In between probabilities are interpolated linearly
#define COEFF_PROB_MODELS 255
#define UNCONSTRAINED_NODES 3
#define PIVOT_NODE 2 // which node is pivot
#define MODEL_NODES (ENTROPY_NODES - UNCONSTRAINED_NODES)
#define TAIL_NODES (MODEL_NODES + 1)
extern const aom_tree_index av1_coef_con_tree[TREE_SIZE(ENTROPY_TOKENS)];
extern const aom_prob av1_pareto8_full[COEFF_PROB_MODELS][MODEL_NODES];
typedef aom_prob av1_coeff_probs_model[REF_TYPES][COEF_BANDS][COEFF_CONTEXTS]
[UNCONSTRAINED_NODES];
typedef unsigned int av1_coeff_count_model[REF_TYPES][COEF_BANDS]
[COEFF_CONTEXTS]
[UNCONSTRAINED_NODES + 1];
void av1_model_to_full_probs(const aom_prob *model, aom_prob *full);
#if CONFIG_EC_MULTISYMBOL
typedef aom_cdf_prob coeff_cdf_model[REF_TYPES][COEF_BANDS][COEFF_CONTEXTS]
[CDF_SIZE(ENTROPY_TOKENS)];
typedef aom_prob av1_blockz_probs_model[REF_TYPES][BLOCKZ_CONTEXTS];
typedef unsigned int av1_blockz_count_model[REF_TYPES][BLOCKZ_CONTEXTS][2];
extern const aom_cdf_prob av1_pareto8_token_probs[COEFF_PROB_MODELS]
[ENTROPY_TOKENS - 2];
extern const aom_cdf_prob av1_pareto8_tail_probs[COEFF_PROB_MODELS]
[ENTROPY_TOKENS - 3];
struct frame_contexts;
#if CONFIG_NEW_TOKENSET
void av1_coef_head_cdfs(struct frame_contexts *fc);
#endif
void av1_coef_pareto_cdfs(struct frame_contexts *fc);
#endif // CONFIG_EC_MULTISYMBOL
typedef char ENTROPY_CONTEXT;
static INLINE int combine_entropy_contexts(ENTROPY_CONTEXT a,
ENTROPY_CONTEXT b) {
return (a != 0) + (b != 0);
}
static INLINE int get_entropy_context(TX_SIZE tx_size, const ENTROPY_CONTEXT *a,
const ENTROPY_CONTEXT *l) {
ENTROPY_CONTEXT above_ec = 0, left_ec = 0;
#if CONFIG_CB4X4
switch (tx_size) {
case TX_2X2:
above_ec = a[0] != 0;
left_ec = l[0] != 0;
break;
case TX_4X4:
above_ec = !!*(const uint16_t *)a;
left_ec = !!*(const uint16_t *)l;
break;
case TX_4X8:
above_ec = !!*(const uint16_t *)a;
left_ec = !!*(const uint32_t *)l;
break;
case TX_8X4:
above_ec = !!*(const uint32_t *)a;
left_ec = !!*(const uint16_t *)l;
break;
case TX_8X8:
above_ec = !!*(const uint32_t *)a;
left_ec = !!*(const uint32_t *)l;
break;
case TX_8X16:
above_ec = !!*(const uint32_t *)a;
left_ec = !!*(const uint64_t *)l;
break;
case TX_16X8:
above_ec = !!*(const uint64_t *)a;
left_ec = !!*(const uint32_t *)l;
break;
case TX_16X16:
above_ec = !!*(const uint64_t *)a;
left_ec = !!*(const uint64_t *)l;
break;
case TX_16X32:
above_ec = !!*(const uint64_t *)a;
left_ec = !!(*(const uint64_t *)l | *(const uint64_t *)(l + 8));
break;
case TX_32X16:
above_ec = !!(*(const uint64_t *)a | *(const uint64_t *)(a + 8));
left_ec = !!*(const uint64_t *)l;
break;
case TX_32X32:
above_ec = !!(*(const uint64_t *)a | *(const uint64_t *)(a + 8));
left_ec = !!(*(const uint64_t *)l | *(const uint64_t *)(l + 8));
break;
default: assert(0 && "Invalid transform size."); break;
}
return combine_entropy_contexts(above_ec, left_ec);
#endif
switch (tx_size) {
case TX_4X4:
above_ec = a[0] != 0;
left_ec = l[0] != 0;
break;
case TX_4X8:
above_ec = a[0] != 0;
left_ec = !!*(const uint16_t *)l;
break;
case TX_8X4:
above_ec = !!*(const uint16_t *)a;
left_ec = l[0] != 0;
break;
case TX_8X16:
above_ec = !!*(const uint16_t *)a;
left_ec = !!*(const uint32_t *)l;
break;
case TX_16X8:
above_ec = !!*(const uint32_t *)a;
left_ec = !!*(const uint16_t *)l;
break;
case TX_16X32:
above_ec = !!*(const uint32_t *)a;
left_ec = !!*(const uint64_t *)l;
break;
case TX_32X16:
above_ec = !!*(const uint64_t *)a;
left_ec = !!*(const uint32_t *)l;
break;
case TX_8X8:
above_ec = !!*(const uint16_t *)a;
left_ec = !!*(const uint16_t *)l;
break;
case TX_16X16:
above_ec = !!*(const uint32_t *)a;
left_ec = !!*(const uint32_t *)l;
break;
case TX_32X32:
above_ec = !!*(const uint64_t *)a;
left_ec = !!*(const uint64_t *)l;
break;
#if CONFIG_TX64X64
case TX_64X64:
above_ec = !!(*(const uint64_t *)a | *(const uint64_t *)(a + 8));
left_ec = !!(*(const uint64_t *)l | *(const uint64_t *)(l + 8));
break;
#endif // CONFIG_TX64X64
default: assert(0 && "Invalid transform size."); break;
}
return combine_entropy_contexts(above_ec, left_ec);
}
#define COEF_COUNT_SAT 24
#define COEF_MAX_UPDATE_FACTOR 112
#define COEF_COUNT_SAT_AFTER_KEY 24
#define COEF_MAX_UPDATE_FACTOR_AFTER_KEY 128
#if CONFIG_ADAPT_SCAN
#define ADAPT_SCAN_UPDATE_RATE_16 (1 << 13)
#endif
static INLINE aom_prob av1_merge_probs(aom_prob pre_prob,
const unsigned int ct[2],
unsigned int count_sat,
unsigned int max_update_factor) {
return merge_probs(pre_prob, ct, count_sat, max_update_factor);
}
static INLINE aom_prob av1_mode_mv_merge_probs(aom_prob pre_prob,
const unsigned int ct[2]) {
return mode_mv_merge_probs(pre_prob, ct);
}
#if CONFIG_EC_ADAPT
void av1_average_tile_coef_cdfs(struct frame_contexts *fc,
struct frame_contexts *ec_ctxs[],
aom_cdf_prob *cdf_ptrs[], int num_tiles);
void av1_average_tile_mv_cdfs(struct frame_contexts *fc,
struct frame_contexts *ec_ctxs[],
aom_cdf_prob *cdf_ptrs[], int num_tiles);
void av1_average_tile_intra_cdfs(struct frame_contexts *fc,
struct frame_contexts *ec_ctxs[],
aom_cdf_prob *cdf_ptrs[], int num_tiles);
void av1_average_tile_inter_cdfs(struct AV1Common *cm,
struct frame_contexts *fc,
struct frame_contexts *ec_ctxs[],
aom_cdf_prob *cdf_ptrs[], int num_tiles);
#if CONFIG_PVQ
void av1_default_pvq_probs(struct AV1Common *cm);
void av1_average_tile_pvq_cdfs(struct frame_contexts *fc,
struct frame_contexts *ec_ctxs[], int num_tiles);
#endif // CONFIG_PVQ
#endif // CONFIG_EC_ADAPT
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_ENTROPY_H_

3792
third_party/aom/av1/common/entropymode.c vendored Normal file
View file

File diff suppressed because it is too large Load diff

575
third_party/aom/av1/common/entropymode.h vendored Normal file
View file

@ -0,0 +1,575 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_ENTROPYMODE_H_
#define AV1_COMMON_ENTROPYMODE_H_
#include "av1/common/entropy.h"
#include "av1/common/entropymv.h"
#include "av1/common/filter.h"
#include "av1/common/seg_common.h"
#include "aom_dsp/aom_filter.h"
#if CONFIG_PVQ
#include "av1/common/pvq.h"
#include "av1/common/pvq_state.h"
#include "av1/common/generic_code.h"
#endif // CONFIG_PVQ
#ifdef __cplusplus
extern "C" {
#endif
#define BLOCK_SIZE_GROUPS 4
#define TX_SIZE_CONTEXTS 2
#define INTER_OFFSET(mode) ((mode)-NEARESTMV)
#if CONFIG_EXT_INTER
#if CONFIG_COMPOUND_SINGLEREF
#define INTER_SINGLEREF_COMP_OFFSET(mode) ((mode)-SR_NEAREST_NEARMV)
#endif // CONFIG_COMPOUND_SINGLEREF
#define INTER_COMPOUND_OFFSET(mode) ((mode)-NEAREST_NEARESTMV)
#endif // CONFIG_EXT_INTER
#if CONFIG_PALETTE
// Number of possible contexts for a color index.
// As can be seen from av1_get_palette_color_index_context(), the possible
// contexts are (2,0,0), (2,2,1), (3,2,0), (4,1,0), (5,0,0). These are mapped to
// a value from 0 to 4 using 'palette_color_index_context_lookup' table.
#define PALETTE_COLOR_INDEX_CONTEXTS 5
// Maximum number of colors in a palette.
#define PALETTE_MAX_SIZE 8
// Minimum number of colors in a palette.
#define PALETTE_MIN_SIZE 2
// Palette mode is available for block sizes >= 8x8.
#define PALETTE_BLOCK_SIZES (BLOCK_LARGEST - BLOCK_8X8 + 1)
// Palette Y mode context for a block is determined by number of neighboring
// blocks (top and/or left) using a palette for Y plane. So, possible Y mode'
// context values are:
// 0 if neither left nor top block uses palette for Y plane,
// 1 if exactly one of left or top block uses palette for Y plane, and
// 2 if both left and top blocks use palette for Y plane.
#define PALETTE_Y_MODE_CONTEXTS 3
// Palette UV mode context for a block is determined by whether this block uses
// palette for the Y plane. So, possible values are:
// 0 if this block doesn't use palette for Y plane.
// 1 if this block uses palette for Y plane (i.e. Y palette size > 0).
#define PALETTE_UV_MODE_CONTEXTS 2
#define PALETTE_MAX_BLOCK_SIZE (64 * 64)
#endif // CONFIG_PALETTE
#if CONFIG_INTRABC
#define INTRABC_PROB 192
#endif // CONFIG_INTRABC
struct AV1Common;
typedef struct {
const int16_t *scan;
const int16_t *iscan;
const int16_t *neighbors;
} SCAN_ORDER;
struct seg_counts {
unsigned int tree_total[MAX_SEGMENTS];
unsigned int tree_mispred[MAX_SEGMENTS];
unsigned int pred[PREDICTION_PROBS][2];
};
typedef struct frame_contexts {
aom_prob y_mode_prob[BLOCK_SIZE_GROUPS][INTRA_MODES - 1];
aom_prob uv_mode_prob[INTRA_MODES][INTRA_MODES - 1];
#if CONFIG_EXT_PARTITION_TYPES
aom_prob partition_prob[PARTITION_CONTEXTS][EXT_PARTITION_TYPES - 1];
#else
aom_prob partition_prob[PARTITION_CONTEXTS][PARTITION_TYPES - 1];
#endif
av1_coeff_probs_model coef_probs[TX_SIZES][PLANE_TYPES];
#if CONFIG_NEW_TOKENSET
coeff_cdf_model coef_tail_cdfs[TX_SIZES][PLANE_TYPES];
coeff_cdf_model coef_head_cdfs[TX_SIZES][PLANE_TYPES];
aom_prob blockzero_probs[TX_SIZES][PLANE_TYPES][REF_TYPES][BLOCKZ_CONTEXTS];
#elif CONFIG_EC_MULTISYMBOL
coeff_cdf_model coef_cdfs[TX_SIZES][PLANE_TYPES];
#endif // CONFIG_NEW_TOKENSET
aom_prob switchable_interp_prob[SWITCHABLE_FILTER_CONTEXTS]
[SWITCHABLE_FILTERS - 1];
#if CONFIG_ADAPT_SCAN
// TODO(angiebird): try aom_prob
#if CONFIG_CB4X4
uint32_t non_zero_prob_2x2[TX_TYPES][4];
#endif
uint32_t non_zero_prob_4X4[TX_TYPES][16];
uint32_t non_zero_prob_8X8[TX_TYPES][64];
uint32_t non_zero_prob_16X16[TX_TYPES][256];
uint32_t non_zero_prob_32X32[TX_TYPES][1024];
uint32_t non_zero_prob_4X8[TX_TYPES][32];
uint32_t non_zero_prob_8X4[TX_TYPES][32];
uint32_t non_zero_prob_16X8[TX_TYPES][128];
uint32_t non_zero_prob_8X16[TX_TYPES][128];
uint32_t non_zero_prob_32X16[TX_TYPES][512];
uint32_t non_zero_prob_16X32[TX_TYPES][512];
#if CONFIG_CB4X4
DECLARE_ALIGNED(16, int16_t, scan_2x2[TX_TYPES][4]);
#endif
DECLARE_ALIGNED(16, int16_t, scan_4X4[TX_TYPES][16]);
DECLARE_ALIGNED(16, int16_t, scan_8X8[TX_TYPES][64]);
DECLARE_ALIGNED(16, int16_t, scan_16X16[TX_TYPES][256]);
DECLARE_ALIGNED(16, int16_t, scan_32X32[TX_TYPES][1024]);
DECLARE_ALIGNED(16, int16_t, scan_4X8[TX_TYPES][32]);
DECLARE_ALIGNED(16, int16_t, scan_8X4[TX_TYPES][32]);
DECLARE_ALIGNED(16, int16_t, scan_8X16[TX_TYPES][128]);
DECLARE_ALIGNED(16, int16_t, scan_16X8[TX_TYPES][128]);
DECLARE_ALIGNED(16, int16_t, scan_16X32[TX_TYPES][512]);
DECLARE_ALIGNED(16, int16_t, scan_32X16[TX_TYPES][512]);
#if CONFIG_CB4X4
DECLARE_ALIGNED(16, int16_t, iscan_2x2[TX_TYPES][4]);
#endif
DECLARE_ALIGNED(16, int16_t, iscan_4X4[TX_TYPES][16]);
DECLARE_ALIGNED(16, int16_t, iscan_8X8[TX_TYPES][64]);
DECLARE_ALIGNED(16, int16_t, iscan_16X16[TX_TYPES][256]);
DECLARE_ALIGNED(16, int16_t, iscan_32X32[TX_TYPES][1024]);
DECLARE_ALIGNED(16, int16_t, iscan_4X8[TX_TYPES][32]);
DECLARE_ALIGNED(16, int16_t, iscan_8X4[TX_TYPES][32]);
DECLARE_ALIGNED(16, int16_t, iscan_8X16[TX_TYPES][128]);
DECLARE_ALIGNED(16, int16_t, iscan_16X8[TX_TYPES][128]);
DECLARE_ALIGNED(16, int16_t, iscan_16X32[TX_TYPES][512]);
DECLARE_ALIGNED(16, int16_t, iscan_32X16[TX_TYPES][512]);
#if CONFIG_CB4X4
int16_t nb_2x2[TX_TYPES][(4 + 1) * 2];
#endif
int16_t nb_4X4[TX_TYPES][(16 + 1) * 2];
int16_t nb_8X8[TX_TYPES][(64 + 1) * 2];
int16_t nb_16X16[TX_TYPES][(256 + 1) * 2];
int16_t nb_32X32[TX_TYPES][(1024 + 1) * 2];
int16_t nb_4X8[TX_TYPES][(32 + 1) * 2];
int16_t nb_8X4[TX_TYPES][(32 + 1) * 2];
int16_t nb_8X16[TX_TYPES][(128 + 1) * 2];
int16_t nb_16X8[TX_TYPES][(128 + 1) * 2];
int16_t nb_16X32[TX_TYPES][(512 + 1) * 2];
int16_t nb_32X16[TX_TYPES][(512 + 1) * 2];
SCAN_ORDER sc[TX_SIZES_ALL][TX_TYPES];
int16_t eob_threshold[TX_SIZES_ALL][TX_TYPES][EOB_THRESHOLD_NUM];
#endif // CONFIG_ADAPT_SCAN
#if CONFIG_LV_MAP
aom_prob txb_skip[TX_SIZES][TXB_SKIP_CONTEXTS];
aom_prob nz_map[TX_SIZES][PLANE_TYPES][SIG_COEF_CONTEXTS];
aom_prob eob_flag[TX_SIZES][PLANE_TYPES][EOB_COEF_CONTEXTS];
aom_prob dc_sign[PLANE_TYPES][DC_SIGN_CONTEXTS];
aom_prob coeff_base[TX_SIZES][PLANE_TYPES][NUM_BASE_LEVELS]
[COEFF_BASE_CONTEXTS];
aom_prob coeff_lps[TX_SIZES][PLANE_TYPES][LEVEL_CONTEXTS];
#endif
#if CONFIG_REF_MV
aom_prob newmv_prob[NEWMV_MODE_CONTEXTS];
aom_prob zeromv_prob[ZEROMV_MODE_CONTEXTS];
aom_prob refmv_prob[REFMV_MODE_CONTEXTS];
aom_prob drl_prob[DRL_MODE_CONTEXTS];
#endif // CONFIG_REF_MV
aom_prob inter_mode_probs[INTER_MODE_CONTEXTS][INTER_MODES - 1];
#if CONFIG_EXT_INTER
aom_prob inter_compound_mode_probs[INTER_MODE_CONTEXTS]
[INTER_COMPOUND_MODES - 1];
#if CONFIG_COMPOUND_SINGLEREF
aom_prob inter_singleref_comp_mode_probs[INTER_MODE_CONTEXTS]
[INTER_SINGLEREF_COMP_MODES - 1];
#endif // CONFIG_COMPOUND_SINGLEREF
aom_prob compound_type_prob[BLOCK_SIZES][COMPOUND_TYPES - 1];
aom_prob interintra_prob[BLOCK_SIZE_GROUPS];
aom_prob interintra_mode_prob[BLOCK_SIZE_GROUPS][INTERINTRA_MODES - 1];
aom_prob wedge_interintra_prob[BLOCK_SIZES];
#endif // CONFIG_EXT_INTER
#if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
aom_prob motion_mode_prob[BLOCK_SIZES][MOTION_MODES - 1];
#if CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
aom_prob obmc_prob[BLOCK_SIZES];
#endif // CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
#endif // CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
aom_prob intra_inter_prob[INTRA_INTER_CONTEXTS];
aom_prob comp_inter_prob[COMP_INTER_CONTEXTS];
aom_prob single_ref_prob[REF_CONTEXTS][SINGLE_REFS - 1];
#if CONFIG_EXT_REFS
aom_prob comp_ref_prob[REF_CONTEXTS][FWD_REFS - 1];
aom_prob comp_bwdref_prob[REF_CONTEXTS][BWD_REFS - 1];
#else
aom_prob comp_ref_prob[REF_CONTEXTS][COMP_REFS - 1];
#endif // CONFIG_EXT_REFS
#if CONFIG_EXT_INTER && CONFIG_COMPOUND_SINGLEREF
aom_prob comp_inter_mode_prob[COMP_INTER_MODE_CONTEXTS];
#endif // CONFIG_EXT_INTER && CONFIG_COMPOUND_SINGLEREF
aom_prob tx_size_probs[MAX_TX_DEPTH][TX_SIZE_CONTEXTS][MAX_TX_DEPTH];
#if CONFIG_VAR_TX
aom_prob txfm_partition_prob[TXFM_PARTITION_CONTEXTS];
#endif
aom_prob skip_probs[SKIP_CONTEXTS];
#if CONFIG_REF_MV
nmv_context nmvc[NMV_CONTEXTS];
#else
nmv_context nmvc;
#endif
#if CONFIG_INTRABC
nmv_context ndvc;
#endif
int initialized;
#if CONFIG_EXT_TX
aom_prob inter_ext_tx_prob[EXT_TX_SETS_INTER][EXT_TX_SIZES][TX_TYPES - 1];
aom_prob intra_ext_tx_prob[EXT_TX_SETS_INTRA][EXT_TX_SIZES][INTRA_MODES]
[TX_TYPES - 1];
#else
aom_prob intra_ext_tx_prob[EXT_TX_SIZES][TX_TYPES][TX_TYPES - 1];
aom_prob inter_ext_tx_prob[EXT_TX_SIZES][TX_TYPES - 1];
#endif // CONFIG_EXT_TX
#if CONFIG_SUPERTX
aom_prob supertx_prob[PARTITION_SUPERTX_CONTEXTS][TX_SIZES];
#endif // CONFIG_SUPERTX
struct segmentation_probs seg;
#if CONFIG_EXT_INTRA
#if CONFIG_INTRA_INTERP
aom_prob intra_filter_probs[INTRA_FILTERS + 1][INTRA_FILTERS - 1];
#endif // CONFIG_INTRA_INTERP
#endif // CONFIG_EXT_INTRA
#if CONFIG_FILTER_INTRA
aom_prob filter_intra_probs[PLANE_TYPES];
#endif // CONFIG_FILTER_INTRA
#if CONFIG_GLOBAL_MOTION
aom_prob global_motion_types_prob[GLOBAL_TRANS_TYPES - 1];
#endif // CONFIG_GLOBAL_MOTION
#if CONFIG_LOOP_RESTORATION
aom_prob switchable_restore_prob[RESTORE_SWITCHABLE_TYPES - 1];
#endif // CONFIG_LOOP_RESTORATION
#if CONFIG_EC_MULTISYMBOL
aom_cdf_prob y_mode_cdf[BLOCK_SIZE_GROUPS][CDF_SIZE(INTRA_MODES)];
aom_cdf_prob uv_mode_cdf[INTRA_MODES][CDF_SIZE(INTRA_MODES)];
#if CONFIG_EXT_PARTITION_TYPES
aom_cdf_prob partition_cdf[PARTITION_CONTEXTS][CDF_SIZE(EXT_PARTITION_TYPES)];
#else
aom_cdf_prob partition_cdf[PARTITION_CONTEXTS][CDF_SIZE(PARTITION_TYPES)];
#endif
aom_cdf_prob switchable_interp_cdf[SWITCHABLE_FILTER_CONTEXTS]
[CDF_SIZE(SWITCHABLE_FILTERS)];
aom_cdf_prob inter_mode_cdf[INTER_MODE_CONTEXTS][CDF_SIZE(INTER_MODES)];
/* Keep track of kf_y_cdf here, as this makes handling
multiple copies for adaptation in tiles easier */
aom_cdf_prob kf_y_cdf[INTRA_MODES][INTRA_MODES][CDF_SIZE(INTRA_MODES)];
aom_cdf_prob tx_size_cdf[MAX_TX_DEPTH][TX_SIZE_CONTEXTS]
[CDF_SIZE(MAX_TX_DEPTH + 1)];
#if CONFIG_DELTA_Q
aom_cdf_prob delta_q_cdf[CDF_SIZE(DELTA_Q_PROBS + 1)];
#if CONFIG_EXT_DELTA_Q
aom_cdf_prob delta_lf_cdf[CDF_SIZE(DELTA_LF_PROBS + 1)];
#endif
#endif // CONFIG_DELTA_Q
#if CONFIG_EXT_TX
aom_cdf_prob intra_ext_tx_cdf[EXT_TX_SETS_INTRA][EXT_TX_SIZES][INTRA_MODES]
[CDF_SIZE(TX_TYPES)];
aom_cdf_prob inter_ext_tx_cdf[EXT_TX_SETS_INTER][EXT_TX_SIZES]
[CDF_SIZE(TX_TYPES)];
#else
aom_cdf_prob intra_ext_tx_cdf[EXT_TX_SIZES][TX_TYPES][CDF_SIZE(TX_TYPES)];
aom_cdf_prob inter_ext_tx_cdf[EXT_TX_SIZES][CDF_SIZE(TX_TYPES)];
#endif // CONFIG_EXT_TX
#if CONFIG_EXT_INTRA && CONFIG_INTRA_INTERP
aom_cdf_prob intra_filter_cdf[INTRA_FILTERS + 1][CDF_SIZE(INTRA_FILTERS)];
#endif // CONFIG_EXT_INTRA && CONFIG_INTRA_INTERP
#endif // CONFIG_EC_MULTISYMBOL
#if CONFIG_DELTA_Q
aom_prob delta_q_prob[DELTA_Q_PROBS];
#if CONFIG_EXT_DELTA_Q
aom_prob delta_lf_prob[DELTA_LF_PROBS];
#endif
#endif
#if CONFIG_PVQ
// TODO(any): If PVQ is enabled, most of coefficient related cdf,
// such as coef_cdfs[], coef_tail_cdfs[], and coef_heaf_cdfs[] can be removed.
od_adapt_ctx pvq_context;
#endif // CONFIG_PVQ
} FRAME_CONTEXT;
typedef struct FRAME_COUNTS {
// Note: This structure should only contain 'unsigned int' fields, or
// aggregates built solely from 'unsigned int' fields/elements
unsigned int kf_y_mode[INTRA_MODES][INTRA_MODES][INTRA_MODES];
unsigned int y_mode[BLOCK_SIZE_GROUPS][INTRA_MODES];
unsigned int uv_mode[INTRA_MODES][INTRA_MODES];
#if CONFIG_EXT_PARTITION_TYPES
unsigned int partition[PARTITION_CONTEXTS][EXT_PARTITION_TYPES];
#else
unsigned int partition[PARTITION_CONTEXTS][PARTITION_TYPES];
#endif
av1_coeff_count_model coef[TX_SIZES][PLANE_TYPES];
unsigned int eob_branch[TX_SIZES][PLANE_TYPES][REF_TYPES][COEF_BANDS]
[COEFF_CONTEXTS];
unsigned int switchable_interp[SWITCHABLE_FILTER_CONTEXTS]
[SWITCHABLE_FILTERS];
#if CONFIG_ADAPT_SCAN
#if CONFIG_CB4X4
unsigned int non_zero_count_2x2[TX_TYPES][4];
#endif // CONFIG_CB4X4
unsigned int non_zero_count_4X4[TX_TYPES][16];
unsigned int non_zero_count_8X8[TX_TYPES][64];
unsigned int non_zero_count_16X16[TX_TYPES][256];
unsigned int non_zero_count_32X32[TX_TYPES][1024];
unsigned int non_zero_count_4x8[TX_TYPES][32];
unsigned int non_zero_count_8x4[TX_TYPES][32];
unsigned int non_zero_count_8x16[TX_TYPES][128];
unsigned int non_zero_count_16x8[TX_TYPES][128];
unsigned int non_zero_count_16x32[TX_TYPES][512];
unsigned int non_zero_count_32x16[TX_TYPES][512];
unsigned int txb_count[TX_SIZES_ALL][TX_TYPES];
#endif // CONFIG_ADAPT_SCAN
#if CONFIG_LV_MAP
unsigned int txb_skip[TX_SIZES][TXB_SKIP_CONTEXTS][2];
unsigned int nz_map[TX_SIZES][PLANE_TYPES][SIG_COEF_CONTEXTS][2];
unsigned int eob_flag[TX_SIZES][PLANE_TYPES][EOB_COEF_CONTEXTS][2];
unsigned int dc_sign[PLANE_TYPES][DC_SIGN_CONTEXTS][2];
unsigned int coeff_base[TX_SIZES][PLANE_TYPES][NUM_BASE_LEVELS]
[COEFF_BASE_CONTEXTS][2];
unsigned int coeff_lps[TX_SIZES][PLANE_TYPES][LEVEL_CONTEXTS][2];
#endif // CONFIG_LV_MAP
#if CONFIG_EC_MULTISYMBOL
av1_blockz_count_model blockz_count[TX_SIZES][PLANE_TYPES];
#endif
#if CONFIG_REF_MV
unsigned int newmv_mode[NEWMV_MODE_CONTEXTS][2];
unsigned int zeromv_mode[ZEROMV_MODE_CONTEXTS][2];
unsigned int refmv_mode[REFMV_MODE_CONTEXTS][2];
unsigned int drl_mode[DRL_MODE_CONTEXTS][2];
#endif
unsigned int inter_mode[INTER_MODE_CONTEXTS][INTER_MODES];
#if CONFIG_EXT_INTER
unsigned int inter_compound_mode[INTER_MODE_CONTEXTS][INTER_COMPOUND_MODES];
#if CONFIG_COMPOUND_SINGLEREF
unsigned int inter_singleref_comp_mode[INTER_MODE_CONTEXTS]
[INTER_SINGLEREF_COMP_MODES];
#endif // CONFIG_COMPOUND_SINGLEREF
unsigned int interintra[BLOCK_SIZE_GROUPS][2];
unsigned int interintra_mode[BLOCK_SIZE_GROUPS][INTERINTRA_MODES];
unsigned int wedge_interintra[BLOCK_SIZES][2];
unsigned int compound_interinter[BLOCK_SIZES][COMPOUND_TYPES];
#endif // CONFIG_EXT_INTER
#if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
unsigned int motion_mode[BLOCK_SIZES][MOTION_MODES];
#if CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
unsigned int obmc[BLOCK_SIZES][2];
#endif // CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
#endif // CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
unsigned int intra_inter[INTRA_INTER_CONTEXTS][2];
unsigned int comp_inter[COMP_INTER_CONTEXTS][2];
unsigned int single_ref[REF_CONTEXTS][SINGLE_REFS - 1][2];
#if CONFIG_EXT_REFS
unsigned int comp_ref[REF_CONTEXTS][FWD_REFS - 1][2];
unsigned int comp_bwdref[REF_CONTEXTS][BWD_REFS - 1][2];
#else
unsigned int comp_ref[REF_CONTEXTS][COMP_REFS - 1][2];
#endif // CONFIG_EXT_REFS
#if CONFIG_EXT_INTER && CONFIG_COMPOUND_SINGLEREF
unsigned int comp_inter_mode[COMP_INTER_MODE_CONTEXTS][2];
#endif // CONFIG_EXT_INTER && CONFIG_COMPOUND_SINGLEREF
// TODO(any): tx_size_totals is only used by the encoder to decide whether
// to use forward updates for the coeff probs, and as such it does not really
// belong into this structure.
unsigned int tx_size_totals[TX_SIZES];
unsigned int tx_size[MAX_TX_DEPTH][TX_SIZE_CONTEXTS][TX_SIZES];
#if CONFIG_VAR_TX
unsigned int txfm_partition[TXFM_PARTITION_CONTEXTS][2];
#endif
unsigned int skip[SKIP_CONTEXTS][2];
#if CONFIG_REF_MV
nmv_context_counts mv[NMV_CONTEXTS];
#else
nmv_context_counts mv;
#endif
#if CONFIG_INTRABC
nmv_context_counts dv;
#endif
#if CONFIG_DELTA_Q
unsigned int delta_q[DELTA_Q_PROBS][2];
#if CONFIG_EXT_DELTA_Q
unsigned int delta_lf[DELTA_LF_PROBS][2];
#endif
#endif
#if CONFIG_EXT_TX
#if CONFIG_RECT_TX
unsigned int tx_size_implied[TX_SIZES][TX_SIZES];
#endif // CONFIG_RECT_TX
unsigned int inter_ext_tx[EXT_TX_SETS_INTER][EXT_TX_SIZES][TX_TYPES];
unsigned int intra_ext_tx[EXT_TX_SETS_INTRA][EXT_TX_SIZES][INTRA_MODES]
[TX_TYPES];
#else
unsigned int intra_ext_tx[EXT_TX_SIZES][TX_TYPES][TX_TYPES];
unsigned int inter_ext_tx[EXT_TX_SIZES][TX_TYPES];
#endif // CONFIG_EXT_TX
#if CONFIG_SUPERTX
unsigned int supertx[PARTITION_SUPERTX_CONTEXTS][TX_SIZES][2];
unsigned int supertx_size[TX_SIZES];
#endif // CONFIG_SUPERTX
struct seg_counts seg;
#if CONFIG_EXT_INTRA
#if CONFIG_INTRA_INTERP
unsigned int intra_filter[INTRA_FILTERS + 1][INTRA_FILTERS];
#endif // CONFIG_INTRA_INTERP
#endif // CONFIG_EXT_INTRA
#if CONFIG_FILTER_INTRA
unsigned int filter_intra[PLANE_TYPES][2];
#endif // CONFIG_FILTER_INTRA
} FRAME_COUNTS;
// Default probabilities for signaling Intra mode for Y plane -- used only for
// intra-only frames. ('default_if_y_probs' is used for inter frames).
// Contexts used: Intra mode (Y plane) of 'above' and 'left' blocks.
extern const aom_prob av1_kf_y_mode_prob[INTRA_MODES][INTRA_MODES]
[INTRA_MODES - 1];
#if CONFIG_EC_MULTISYMBOL
// CDF version of 'av1_kf_y_mode_prob'.
extern const aom_cdf_prob av1_kf_y_mode_cdf[INTRA_MODES][INTRA_MODES]
[CDF_SIZE(INTRA_MODES)];
#endif
#if CONFIG_PALETTE
extern const aom_prob av1_default_palette_y_mode_prob[PALETTE_BLOCK_SIZES]
[PALETTE_Y_MODE_CONTEXTS];
extern const aom_prob
av1_default_palette_uv_mode_prob[PALETTE_UV_MODE_CONTEXTS];
extern const aom_prob av1_default_palette_y_size_prob[PALETTE_BLOCK_SIZES]
[PALETTE_SIZES - 1];
extern const aom_prob av1_default_palette_uv_size_prob[PALETTE_BLOCK_SIZES]
[PALETTE_SIZES - 1];
extern const aom_prob av1_default_palette_y_color_index_prob
[PALETTE_SIZES][PALETTE_COLOR_INDEX_CONTEXTS][PALETTE_COLORS - 1];
extern const aom_prob av1_default_palette_uv_color_index_prob
[PALETTE_SIZES][PALETTE_COLOR_INDEX_CONTEXTS][PALETTE_COLORS - 1];
#endif // CONFIG_PALETTE
extern const aom_tree_index av1_intra_mode_tree[TREE_SIZE(INTRA_MODES)];
extern const aom_tree_index av1_inter_mode_tree[TREE_SIZE(INTER_MODES)];
#if CONFIG_EC_MULTISYMBOL
extern int av1_intra_mode_ind[INTRA_MODES];
extern int av1_intra_mode_inv[INTRA_MODES];
extern int av1_inter_mode_ind[INTER_MODES];
extern int av1_inter_mode_inv[INTER_MODES];
#if CONFIG_EXT_TX
extern int av1_ext_tx_intra_ind[EXT_TX_SETS_INTRA][TX_TYPES];
extern int av1_ext_tx_intra_inv[EXT_TX_SETS_INTRA][TX_TYPES];
extern int av1_ext_tx_inter_ind[EXT_TX_SETS_INTER][TX_TYPES];
extern int av1_ext_tx_inter_inv[EXT_TX_SETS_INTER][TX_TYPES];
#endif
#endif
#if CONFIG_EXT_INTER
extern const aom_tree_index
av1_interintra_mode_tree[TREE_SIZE(INTERINTRA_MODES)];
extern const aom_tree_index
av1_inter_compound_mode_tree[TREE_SIZE(INTER_COMPOUND_MODES)];
#if CONFIG_COMPOUND_SINGLEREF
extern const aom_tree_index
av1_inter_singleref_comp_mode_tree[TREE_SIZE(INTER_SINGLEREF_COMP_MODES)];
#endif // CONFIG_COMPOUND_SINGLEREF
extern const aom_tree_index av1_compound_type_tree[TREE_SIZE(COMPOUND_TYPES)];
#endif // CONFIG_EXT_INTER
extern const aom_tree_index av1_partition_tree[TREE_SIZE(PARTITION_TYPES)];
#if CONFIG_EXT_PARTITION_TYPES
extern const aom_tree_index
av1_ext_partition_tree[TREE_SIZE(EXT_PARTITION_TYPES)];
#endif
extern const aom_tree_index
av1_switchable_interp_tree[TREE_SIZE(SWITCHABLE_FILTERS)];
#if CONFIG_PALETTE
extern const aom_tree_index av1_palette_size_tree[TREE_SIZE(PALETTE_SIZES)];
extern const aom_tree_index
av1_palette_color_index_tree[PALETTE_SIZES][TREE_SIZE(PALETTE_COLORS)];
#endif // CONFIG_PALETTE
extern const aom_tree_index av1_tx_size_tree[MAX_TX_DEPTH][TREE_SIZE(TX_SIZES)];
#if CONFIG_EXT_INTRA && CONFIG_INTRA_INTERP
extern const aom_tree_index av1_intra_filter_tree[TREE_SIZE(INTRA_FILTERS)];
#endif // CONFIG_EXT_INTRA && CONFIG_INTRA_INTERP
#if CONFIG_EXT_TX
extern const aom_tree_index av1_ext_tx_inter_tree[EXT_TX_SETS_INTER]
[TREE_SIZE(TX_TYPES)];
extern const aom_tree_index av1_ext_tx_intra_tree[EXT_TX_SETS_INTRA]
[TREE_SIZE(TX_TYPES)];
#else
extern const aom_tree_index av1_ext_tx_tree[TREE_SIZE(TX_TYPES)];
#endif // CONFIG_EXT_TX
#if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
extern const aom_tree_index av1_motion_mode_tree[TREE_SIZE(MOTION_MODES)];
#endif // CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
#if CONFIG_LOOP_RESTORATION
#define RESTORE_NONE_SGRPROJ_PROB 64
#define RESTORE_NONE_BILATERAL_PROB 16
#define RESTORE_NONE_WIENER_PROB 64
#define RESTORE_NONE_DOMAINTXFMRF_PROB 64
extern const aom_tree_index
av1_switchable_restore_tree[TREE_SIZE(RESTORE_SWITCHABLE_TYPES)];
#endif // CONFIG_LOOP_RESTORATION
#if CONFIG_EC_MULTISYMBOL
extern int av1_switchable_interp_ind[SWITCHABLE_FILTERS];
extern int av1_switchable_interp_inv[SWITCHABLE_FILTERS];
void av1_set_mode_cdfs(struct AV1Common *cm);
#endif
void av1_setup_past_independence(struct AV1Common *cm);
void av1_adapt_intra_frame_probs(struct AV1Common *cm);
void av1_adapt_inter_frame_probs(struct AV1Common *cm);
#if CONFIG_EC_MULTISYMBOL && !CONFIG_EXT_TX
extern int av1_ext_tx_ind[TX_TYPES];
extern int av1_ext_tx_inv[TX_TYPES];
#endif
static INLINE int av1_ceil_log2(int n) {
int i = 1, p = 2;
while (p < n) {
i++;
p = p << 1;
}
return i;
}
#if CONFIG_PALETTE
// Returns the context for palette color index at row 'r' and column 'c',
// along with the 'color_order' of neighbors and the 'color_idx'.
// The 'color_map' is a 2D array with the given 'stride'.
int av1_get_palette_color_index_context(const uint8_t *color_map, int stride,
int r, int c, int palette_size,
uint8_t *color_order, int *color_idx);
#endif // CONFIG_PALETTE
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_ENTROPYMODE_H_

315
third_party/aom/av1/common/entropymv.c vendored Normal file
View file

@ -0,0 +1,315 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "av1/common/onyxc_int.h"
#include "av1/common/entropymv.h"
// Integer pel reference mv threshold for use of high-precision 1/8 mv
#define COMPANDED_MVREF_THRESH 8
const aom_tree_index av1_mv_joint_tree[TREE_SIZE(MV_JOINTS)] = {
-MV_JOINT_ZERO, 2, -MV_JOINT_HNZVZ, 4, -MV_JOINT_HZVNZ, -MV_JOINT_HNZVNZ
};
/* clang-format off */
const aom_tree_index av1_mv_class_tree[TREE_SIZE(MV_CLASSES)] = {
-MV_CLASS_0, 2,
-MV_CLASS_1, 4,
6, 8,
-MV_CLASS_2, -MV_CLASS_3,
10, 12,
-MV_CLASS_4, -MV_CLASS_5,
-MV_CLASS_6, 14,
16, 18,
-MV_CLASS_7, -MV_CLASS_8,
-MV_CLASS_9, -MV_CLASS_10,
};
/* clang-format on */
const aom_tree_index av1_mv_class0_tree[TREE_SIZE(CLASS0_SIZE)] = {
-0, -1,
};
const aom_tree_index av1_mv_fp_tree[TREE_SIZE(MV_FP_SIZE)] = { -0, 2, -1,
4, -2, -3 };
static const nmv_context default_nmv_context = {
{ 32, 64, 96 }, // joints
#if CONFIG_EC_MULTISYMBOL
{ AOM_ICDF(4096), AOM_ICDF(11264), AOM_ICDF(19328), AOM_ICDF(32768),
0 }, // joint_cdf
#endif
{ {
// Vertical component
128, // sign
{ 224, 144, 192, 168, 192, 176, 192, 198, 198, 245 }, // class
#if CONFIG_EC_MULTISYMBOL
{ AOM_ICDF(28672), AOM_ICDF(30976), AOM_ICDF(31858), AOM_ICDF(32320),
AOM_ICDF(32551), AOM_ICDF(32656), AOM_ICDF(32740), AOM_ICDF(32757),
AOM_ICDF(32762), AOM_ICDF(32767), AOM_ICDF(32768), 0 }, // class_cdf
#endif
{ 216 }, // class0
{ 136, 140, 148, 160, 176, 192, 224, 234, 234, 240 }, // bits
{ { 128, 128, 64 }, { 96, 112, 64 } }, // class0_fp
{ 64, 96, 64 }, // fp
#if CONFIG_EC_MULTISYMBOL
{ { AOM_ICDF(16384), AOM_ICDF(24576), AOM_ICDF(26624), AOM_ICDF(32768),
0 },
{ AOM_ICDF(12288), AOM_ICDF(21248), AOM_ICDF(24128), AOM_ICDF(32768),
0 } }, // class0_fp_cdf
{ AOM_ICDF(8192), AOM_ICDF(17408), AOM_ICDF(21248), AOM_ICDF(32768),
0 }, // fp_cdf
#endif
160, // class0_hp bit
128, // hp
},
{
// Horizontal component
128, // sign
{ 216, 128, 176, 160, 176, 176, 192, 198, 198, 208 }, // class
#if CONFIG_EC_MULTISYMBOL
{ AOM_ICDF(28672), AOM_ICDF(30976), AOM_ICDF(31858), AOM_ICDF(32320),
AOM_ICDF(32551), AOM_ICDF(32656), AOM_ICDF(32740), AOM_ICDF(32757),
AOM_ICDF(32762), AOM_ICDF(32767), AOM_ICDF(32768), 0 }, // class_cdf
#endif
{ 208 }, // class0
{ 136, 140, 148, 160, 176, 192, 224, 234, 234, 240 }, // bits
{ { 128, 128, 64 }, { 96, 112, 64 } }, // class0_fp
{ 64, 96, 64 }, // fp
#if CONFIG_EC_MULTISYMBOL
{ { AOM_ICDF(16384), AOM_ICDF(24576), AOM_ICDF(26624), AOM_ICDF(32768),
0 },
{ AOM_ICDF(12288), AOM_ICDF(21248), AOM_ICDF(24128), AOM_ICDF(32768),
0 } }, // class0_fp_cdf
{ AOM_ICDF(8192), AOM_ICDF(17408), AOM_ICDF(21248), AOM_ICDF(32768),
0 }, // fp_cdf
#endif
160, // class0_hp bit
128, // hp
} },
};
static const uint8_t log_in_base_2[] = {
0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 7, 7,
7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 10
};
#if CONFIG_GLOBAL_MOTION
#if GLOBAL_TRANS_TYPES == 7 // All models
const aom_tree_index av1_global_motion_types_tree[TREE_SIZE(
GLOBAL_TRANS_TYPES)] = { -IDENTITY, 2, -TRANSLATION, 4,
-ROTZOOM, 6, -AFFINE, 8,
-HOMOGRAPHY, 10, -HORTRAPEZOID, -VERTRAPEZOID };
static const aom_prob default_global_motion_types_prob[GLOBAL_TRANS_TYPES - 1] =
{ 224, 128, 192, 192, 32, 128 };
#elif GLOBAL_TRANS_TYPES == 6 // Do not allow full homography
const aom_tree_index
av1_global_motion_types_tree[TREE_SIZE(GLOBAL_TRANS_TYPES)] = {
-IDENTITY, 2, -TRANSLATION, 4, -ROTZOOM, 6, -AFFINE, 8, -HORTRAPEZOID,
-VERTRAPEZOID
};
static const aom_prob default_global_motion_types_prob[GLOBAL_TRANS_TYPES - 1] =
{ 224, 128, 192, 192, 128 };
#elif GLOBAL_TRANS_TYPES == 4 // Upto Affine
const aom_tree_index av1_global_motion_types_tree[TREE_SIZE(
GLOBAL_TRANS_TYPES)] = { -IDENTITY, 2, -TRANSLATION, 4, -ROTZOOM, -AFFINE };
static const aom_prob default_global_motion_types_prob[GLOBAL_TRANS_TYPES - 1] =
{ 224, 128, 240 };
#elif GLOBAL_TRANS_TYPES == 3 // Upto rotation-zoom
const aom_tree_index av1_global_motion_types_tree[TREE_SIZE(
GLOBAL_TRANS_TYPES)] = { -IDENTITY, 2, -TRANSLATION, -ROTZOOM };
static const aom_prob default_global_motion_types_prob[GLOBAL_TRANS_TYPES - 1] =
{ 224, 128 };
#endif // GLOBAL_TRANS_TYPES
#endif // CONFIG_GLOBAL_MOTION
static INLINE int mv_class_base(MV_CLASS_TYPE c) {
return c ? CLASS0_SIZE << (c + 2) : 0;
}
MV_CLASS_TYPE av1_get_mv_class(int z, int *offset) {
const MV_CLASS_TYPE c = (z >= CLASS0_SIZE * 4096)
? MV_CLASS_10
: (MV_CLASS_TYPE)log_in_base_2[z >> 3];
if (offset) *offset = z - mv_class_base(c);
return c;
}
static void inc_mv_component(int v, nmv_component_counts *comp_counts, int incr,
int usehp) {
int s, z, c, o, d, e, f;
assert(v != 0); /* should not be zero */
s = v < 0;
comp_counts->sign[s] += incr;
z = (s ? -v : v) - 1; /* magnitude - 1 */
c = av1_get_mv_class(z, &o);
comp_counts->classes[c] += incr;
d = (o >> 3); /* int mv data */
f = (o >> 1) & 3; /* fractional pel mv data */
e = (o & 1); /* high precision mv data */
if (c == MV_CLASS_0) {
comp_counts->class0[d] += incr;
comp_counts->class0_fp[d][f] += incr;
if (usehp) comp_counts->class0_hp[e] += incr;
} else {
int i;
int b = c + CLASS0_BITS - 1; // number of bits
for (i = 0; i < b; ++i) comp_counts->bits[i][((d >> i) & 1)] += incr;
comp_counts->fp[f] += incr;
if (usehp) comp_counts->hp[e] += incr;
}
}
void av1_inc_mv(const MV *mv, nmv_context_counts *counts, const int usehp) {
if (counts != NULL) {
const MV_JOINT_TYPE j = av1_get_mv_joint(mv);
++counts->joints[j];
if (mv_joint_vertical(j))
inc_mv_component(mv->row, &counts->comps[0], 1, usehp);
if (mv_joint_horizontal(j))
inc_mv_component(mv->col, &counts->comps[1], 1, usehp);
}
}
void av1_adapt_mv_probs(AV1_COMMON *cm, int allow_hp) {
int i, j;
#if CONFIG_REF_MV
int idx;
for (idx = 0; idx < NMV_CONTEXTS; ++idx) {
nmv_context *fc = &cm->fc->nmvc[idx];
const nmv_context *pre_fc =
&cm->frame_contexts[cm->frame_context_idx].nmvc[idx];
const nmv_context_counts *counts = &cm->counts.mv[idx];
#else
nmv_context *fc = &cm->fc->nmvc;
const nmv_context *pre_fc = &cm->frame_contexts[cm->frame_context_idx].nmvc;
const nmv_context_counts *counts = &cm->counts.mv;
#endif // CONFIG_REF_MV
aom_tree_merge_probs(av1_mv_joint_tree, pre_fc->joints, counts->joints,
fc->joints);
for (i = 0; i < 2; ++i) {
nmv_component *comp = &fc->comps[i];
const nmv_component *pre_comp = &pre_fc->comps[i];
const nmv_component_counts *c = &counts->comps[i];
comp->sign = av1_mode_mv_merge_probs(pre_comp->sign, c->sign);
aom_tree_merge_probs(av1_mv_class_tree, pre_comp->classes, c->classes,
comp->classes);
aom_tree_merge_probs(av1_mv_class0_tree, pre_comp->class0, c->class0,
comp->class0);
for (j = 0; j < MV_OFFSET_BITS; ++j)
comp->bits[j] = av1_mode_mv_merge_probs(pre_comp->bits[j], c->bits[j]);
for (j = 0; j < CLASS0_SIZE; ++j)
aom_tree_merge_probs(av1_mv_fp_tree, pre_comp->class0_fp[j],
c->class0_fp[j], comp->class0_fp[j]);
aom_tree_merge_probs(av1_mv_fp_tree, pre_comp->fp, c->fp, comp->fp);
if (allow_hp) {
comp->class0_hp =
av1_mode_mv_merge_probs(pre_comp->class0_hp, c->class0_hp);
comp->hp = av1_mode_mv_merge_probs(pre_comp->hp, c->hp);
}
}
#if CONFIG_REF_MV
}
#endif // CONFIG_REF_MV
}
#if CONFIG_EC_MULTISYMBOL && !CONFIG_EC_ADAPT
void av1_set_mv_cdfs(nmv_context *ctx) {
int i;
int j;
av1_tree_to_cdf(av1_mv_joint_tree, ctx->joints, ctx->joint_cdf);
for (i = 0; i < 2; ++i) {
nmv_component *const comp_ctx = &ctx->comps[i];
av1_tree_to_cdf(av1_mv_class_tree, comp_ctx->classes, comp_ctx->class_cdf);
for (j = 0; j < CLASS0_SIZE; ++j) {
av1_tree_to_cdf(av1_mv_fp_tree, comp_ctx->class0_fp[j],
comp_ctx->class0_fp_cdf[j]);
}
av1_tree_to_cdf(av1_mv_fp_tree, comp_ctx->fp, comp_ctx->fp_cdf);
}
}
#endif
void av1_init_mv_probs(AV1_COMMON *cm) {
#if CONFIG_REF_MV
int i;
for (i = 0; i < NMV_CONTEXTS; ++i) {
// NB: this sets CDFs too
cm->fc->nmvc[i] = default_nmv_context;
}
#else
cm->fc->nmvc = default_nmv_context;
#endif // CONFIG_REF_MV
#if CONFIG_INTRABC
cm->fc->ndvc = default_nmv_context;
#endif // CONFIG_INTRABC
#if CONFIG_GLOBAL_MOTION
av1_copy(cm->fc->global_motion_types_prob, default_global_motion_types_prob);
#endif // CONFIG_GLOBAL_MOTION
}

149
third_party/aom/av1/common/entropymv.h vendored Normal file
View file

@ -0,0 +1,149 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_ENTROPYMV_H_
#define AV1_COMMON_ENTROPYMV_H_
#include "./aom_config.h"
#include "aom_dsp/prob.h"
#include "av1/common/mv.h"
#ifdef __cplusplus
extern "C" {
#endif
struct AV1Common;
void av1_init_mv_probs(struct AV1Common *cm);
void av1_adapt_mv_probs(struct AV1Common *cm, int usehp);
#define MV_UPDATE_PROB 252
/* Symbols for coding which components are zero jointly */
#define MV_JOINTS 4
typedef enum {
MV_JOINT_ZERO = 0, /* Zero vector */
MV_JOINT_HNZVZ = 1, /* Vert zero, hor nonzero */
MV_JOINT_HZVNZ = 2, /* Hor zero, vert nonzero */
MV_JOINT_HNZVNZ = 3, /* Both components nonzero */
} MV_JOINT_TYPE;
static INLINE int mv_joint_vertical(MV_JOINT_TYPE type) {
return type == MV_JOINT_HZVNZ || type == MV_JOINT_HNZVNZ;
}
static INLINE int mv_joint_horizontal(MV_JOINT_TYPE type) {
return type == MV_JOINT_HNZVZ || type == MV_JOINT_HNZVNZ;
}
/* Symbols for coding magnitude class of nonzero components */
#define MV_CLASSES 11
typedef enum {
MV_CLASS_0 = 0, /* (0, 2] integer pel */
MV_CLASS_1 = 1, /* (2, 4] integer pel */
MV_CLASS_2 = 2, /* (4, 8] integer pel */
MV_CLASS_3 = 3, /* (8, 16] integer pel */
MV_CLASS_4 = 4, /* (16, 32] integer pel */
MV_CLASS_5 = 5, /* (32, 64] integer pel */
MV_CLASS_6 = 6, /* (64, 128] integer pel */
MV_CLASS_7 = 7, /* (128, 256] integer pel */
MV_CLASS_8 = 8, /* (256, 512] integer pel */
MV_CLASS_9 = 9, /* (512, 1024] integer pel */
MV_CLASS_10 = 10, /* (1024,2048] integer pel */
} MV_CLASS_TYPE;
#define CLASS0_BITS 1 /* bits at integer precision for class 0 */
#define CLASS0_SIZE (1 << CLASS0_BITS)
#define MV_OFFSET_BITS (MV_CLASSES + CLASS0_BITS - 2)
#define MV_FP_SIZE 4
#define MV_MAX_BITS (MV_CLASSES + CLASS0_BITS + 2)
#define MV_MAX ((1 << MV_MAX_BITS) - 1)
#define MV_VALS ((MV_MAX << 1) + 1)
#define MV_IN_USE_BITS 14
#define MV_UPP ((1 << MV_IN_USE_BITS) - 1)
#define MV_LOW (-(1 << MV_IN_USE_BITS))
extern const aom_tree_index av1_mv_joint_tree[];
extern const aom_tree_index av1_mv_class_tree[];
extern const aom_tree_index av1_mv_class0_tree[];
extern const aom_tree_index av1_mv_fp_tree[];
typedef struct {
aom_prob sign;
aom_prob classes[MV_CLASSES - 1];
#if CONFIG_EC_MULTISYMBOL
aom_cdf_prob class_cdf[CDF_SIZE(MV_CLASSES)];
#endif
aom_prob class0[CLASS0_SIZE - 1];
aom_prob bits[MV_OFFSET_BITS];
aom_prob class0_fp[CLASS0_SIZE][MV_FP_SIZE - 1];
aom_prob fp[MV_FP_SIZE - 1];
#if CONFIG_EC_MULTISYMBOL
aom_cdf_prob class0_fp_cdf[CLASS0_SIZE][CDF_SIZE(MV_FP_SIZE)];
aom_cdf_prob fp_cdf[CDF_SIZE(MV_FP_SIZE)];
#endif
aom_prob class0_hp;
aom_prob hp;
} nmv_component;
typedef struct {
aom_prob joints[MV_JOINTS - 1];
#if CONFIG_EC_MULTISYMBOL
aom_cdf_prob joint_cdf[CDF_SIZE(MV_JOINTS)];
#endif
nmv_component comps[2];
} nmv_context;
static INLINE MV_JOINT_TYPE av1_get_mv_joint(const MV *mv) {
if (mv->row == 0) {
return mv->col == 0 ? MV_JOINT_ZERO : MV_JOINT_HNZVZ;
} else {
return mv->col == 0 ? MV_JOINT_HZVNZ : MV_JOINT_HNZVNZ;
}
}
MV_CLASS_TYPE av1_get_mv_class(int z, int *offset);
typedef struct {
unsigned int sign[2];
unsigned int classes[MV_CLASSES];
unsigned int class0[CLASS0_SIZE];
unsigned int bits[MV_OFFSET_BITS][2];
unsigned int class0_fp[CLASS0_SIZE][MV_FP_SIZE];
unsigned int fp[MV_FP_SIZE];
unsigned int class0_hp[2];
unsigned int hp[2];
} nmv_component_counts;
typedef struct {
unsigned int joints[MV_JOINTS];
nmv_component_counts comps[2];
} nmv_context_counts;
void av1_inc_mv(const MV *mv, nmv_context_counts *mvctx, const int usehp);
#if CONFIG_GLOBAL_MOTION
extern const aom_tree_index
av1_global_motion_types_tree[TREE_SIZE(GLOBAL_TRANS_TYPES)];
#endif // CONFIG_GLOBAL_MOTION
#if CONFIG_EC_MULTISYMBOL
void av1_set_mv_cdfs(nmv_context *ctx);
#endif
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_ENTROPYMV_H_

543
third_party/aom/av1/common/enums.h vendored Normal file
View file

@ -0,0 +1,543 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_ENUMS_H_
#define AV1_COMMON_ENUMS_H_
#include "./aom_config.h"
#include "aom/aom_codec.h"
#include "aom/aom_integer.h"
#ifdef __cplusplus
extern "C" {
#endif
#undef MAX_SB_SIZE
// Max superblock size
#if CONFIG_EXT_PARTITION
#define MAX_SB_SIZE_LOG2 7
#else
#define MAX_SB_SIZE_LOG2 6
#endif // CONFIG_EXT_PARTITION
#define MAX_SB_SIZE (1 << MAX_SB_SIZE_LOG2)
#define MAX_SB_SQUARE (MAX_SB_SIZE * MAX_SB_SIZE)
// Min superblock size
#define MIN_SB_SIZE_LOG2 6
// Pixels per Mode Info (MI) unit
#if CONFIG_CB4X4
#define MI_SIZE_LOG2 2
#else
#define MI_SIZE_LOG2 3
#endif
#define MI_SIZE (1 << MI_SIZE_LOG2)
// MI-units per max superblock (MI Block - MIB)
#define MAX_MIB_SIZE_LOG2 (MAX_SB_SIZE_LOG2 - MI_SIZE_LOG2)
#define MAX_MIB_SIZE (1 << MAX_MIB_SIZE_LOG2)
// MI-units per min superblock
#define MIN_MIB_SIZE_LOG2 (MIN_SB_SIZE_LOG2 - MI_SIZE_LOG2)
// Mask to extract MI offset within max MIB
#define MAX_MIB_MASK (MAX_MIB_SIZE - 1)
#define MAX_MIB_MASK_2 (MAX_MIB_SIZE * 2 - 1)
// Maximum number of tile rows and tile columns
#if CONFIG_EXT_TILE
#define TILE_NORMAL 0
#define TILE_VR 1
#define MAX_TILE_ROWS 1024
#define MAX_TILE_COLS 1024
#else
#define MAX_TILE_ROWS 4
#define MAX_TILE_COLS 64
#endif // CONFIG_EXT_TILE
#if CONFIG_VAR_TX
#define MAX_VARTX_DEPTH 2
#endif
// Bitstream profiles indicated by 2-3 bits in the uncompressed header.
// 00: Profile 0. 8-bit 4:2:0 only.
// 10: Profile 1. 8-bit 4:4:4, 4:2:2, and 4:4:0.
// 01: Profile 2. 10-bit and 12-bit color only, with 4:2:0 sampling.
// 110: Profile 3. 10-bit and 12-bit color only, with 4:2:2/4:4:4/4:4:0
// sampling.
// 111: Undefined profile.
typedef enum BITSTREAM_PROFILE {
PROFILE_0,
PROFILE_1,
PROFILE_2,
PROFILE_3,
MAX_PROFILES
} BITSTREAM_PROFILE;
// Note: Some enums use the attribute 'packed' to use smallest possible integer
// type, so that we can save memory when they are used in structs/arrays.
typedef enum ATTRIBUTE_PACKED {
#if CONFIG_CB4X4
BLOCK_2X2,
BLOCK_2X4,
BLOCK_4X2,
#endif
BLOCK_4X4,
BLOCK_4X8,
BLOCK_8X4,
BLOCK_8X8,
BLOCK_8X16,
BLOCK_16X8,
BLOCK_16X16,
BLOCK_16X32,
BLOCK_32X16,
BLOCK_32X32,
BLOCK_32X64,
BLOCK_64X32,
BLOCK_64X64,
#if CONFIG_EXT_PARTITION
BLOCK_64X128,
BLOCK_128X64,
BLOCK_128X128,
#endif // CONFIG_EXT_PARTITION
BLOCK_SIZES,
BLOCK_INVALID = BLOCK_SIZES,
BLOCK_LARGEST = (BLOCK_SIZES - 1)
} BLOCK_SIZE;
typedef enum {
PARTITION_NONE,
PARTITION_HORZ,
PARTITION_VERT,
PARTITION_SPLIT,
#if CONFIG_EXT_PARTITION_TYPES
PARTITION_HORZ_A, // HORZ split and the left partition is split again
PARTITION_HORZ_B, // HORZ split and the right partition is split again
PARTITION_VERT_A, // VERT split and the top partition is split again
PARTITION_VERT_B, // VERT split and the bottom partition is split again
EXT_PARTITION_TYPES,
#endif // CONFIG_EXT_PARTITION_TYPES
PARTITION_TYPES = PARTITION_SPLIT + 1,
PARTITION_INVALID = 255
} PARTITION_TYPE;
typedef char PARTITION_CONTEXT;
#define PARTITION_PLOFFSET 4 // number of probability models per block size
#define PARTITION_BLOCK_SIZES (4 + CONFIG_EXT_PARTITION)
#define PARTITION_CONTEXTS_PRIMARY (PARTITION_BLOCK_SIZES * PARTITION_PLOFFSET)
#if CONFIG_UNPOISON_PARTITION_CTX
#define PARTITION_CONTEXTS \
(PARTITION_CONTEXTS_PRIMARY + 2 * PARTITION_BLOCK_SIZES)
#else
#define PARTITION_CONTEXTS PARTITION_CONTEXTS_PRIMARY
#endif
// block transform size
typedef enum ATTRIBUTE_PACKED {
#if CONFIG_CB4X4
TX_2X2, // 2x2 transform
#endif
TX_4X4, // 4x4 transform
TX_8X8, // 8x8 transform
TX_16X16, // 16x16 transform
TX_32X32, // 32x32 transform
#if CONFIG_TX64X64
TX_64X64, // 64x64 transform
#endif // CONFIG_TX64X64
TX_4X8, // 4x8 transform
TX_8X4, // 8x4 transform
TX_8X16, // 8x16 transform
TX_16X8, // 16x8 transform
TX_16X32, // 16x32 transform
TX_32X16, // 32x16 transform
TX_4X16, // 4x16 transform
TX_16X4, // 16x4 transform
TX_8X32, // 8x32 transform
TX_32X8, // 32x8 transform
TX_SIZES_ALL, // Includes rectangular transforms
TX_SIZES = TX_4X8, // Does NOT include rectangular transforms
TX_INVALID = 255 // Invalid transform size
} TX_SIZE;
#define MAX_TX_DEPTH (TX_SIZES - 1 - TX_4X4)
#define MAX_TX_SIZE_LOG2 (5 + CONFIG_TX64X64)
#define MAX_TX_SIZE (1 << MAX_TX_SIZE_LOG2)
#define MIN_TX_SIZE_LOG2 2
#define MIN_TX_SIZE (1 << MIN_TX_SIZE_LOG2)
#define MAX_TX_SQUARE (MAX_TX_SIZE * MAX_TX_SIZE)
// Number of maxium size transform blocks in the maximum size superblock
#define MAX_TX_BLOCKS_IN_MAX_SB_LOG2 ((MAX_SB_SIZE_LOG2 - MAX_TX_SIZE_LOG2) * 2)
#define MAX_TX_BLOCKS_IN_MAX_SB (1 << MAX_TX_BLOCKS_IN_MAX_SB_LOG2)
#define MAX_NUM_TXB (1 << (MAX_SB_SIZE_LOG2 - MIN_TX_SIZE_LOG2))
// frame transform mode
typedef enum {
ONLY_4X4 = 0, // only 4x4 transform used
ALLOW_8X8 = 1, // allow block transform size up to 8x8
ALLOW_16X16 = 2, // allow block transform size up to 16x16
ALLOW_32X32 = 3, // allow block transform size up to 32x32
#if CONFIG_TX64X64
ALLOW_64X64 = 4, // allow block transform size up to 64x64
#endif
TX_MODE_SELECT, // transform specified for each block
TX_MODES,
} TX_MODE;
// 1D tx types
typedef enum {
DCT_1D = 0,
ADST_1D = 1,
FLIPADST_1D = 2,
IDTX_1D = 3,
TX_TYPES_1D = 4,
} TX_TYPE_1D;
typedef enum {
DCT_DCT = 0, // DCT in both horizontal and vertical
ADST_DCT = 1, // ADST in vertical, DCT in horizontal
DCT_ADST = 2, // DCT in vertical, ADST in horizontal
ADST_ADST = 3, // ADST in both directions
#if CONFIG_EXT_TX
FLIPADST_DCT = 4,
DCT_FLIPADST = 5,
FLIPADST_FLIPADST = 6,
ADST_FLIPADST = 7,
FLIPADST_ADST = 8,
IDTX = 9,
V_DCT = 10,
H_DCT = 11,
V_ADST = 12,
H_ADST = 13,
V_FLIPADST = 14,
H_FLIPADST = 15,
#endif // CONFIG_EXT_TX
TX_TYPES,
} TX_TYPE;
typedef enum {
TILE_LEFT_BOUNDARY = 1,
TILE_RIGHT_BOUNDARY = 2,
TILE_ABOVE_BOUNDARY = 4,
TILE_BOTTOM_BOUNDARY = 8,
FRAME_LEFT_BOUNDARY = 16,
FRAME_RIGHT_BOUNDARY = 32,
FRAME_ABOVE_BOUNDARY = 64,
FRAME_BOTTOM_BOUNDARY = 128,
} BOUNDARY_TYPE;
#if CONFIG_EXT_TX
#if CONFIG_CB4X4
#define EXT_TX_SIZES 5 // number of sizes that use extended transforms
#else
#define EXT_TX_SIZES 4 // number of sizes that use extended transforms
#endif // CONFIG_CB4X4
#define EXT_TX_SETS_INTER 4 // Sets of transform selections for INTER
#define EXT_TX_SETS_INTRA 3 // Sets of transform selections for INTRA
#else
#if CONFIG_CB4X4
#define EXT_TX_SIZES 4 // number of sizes that use extended transforms
#else
#define EXT_TX_SIZES 3 // number of sizes that use extended transforms
#endif
#endif // CONFIG_EXT_TX
typedef enum {
AOM_LAST_FLAG = 1 << 0,
#if CONFIG_EXT_REFS
AOM_LAST2_FLAG = 1 << 1,
AOM_LAST3_FLAG = 1 << 2,
AOM_GOLD_FLAG = 1 << 3,
AOM_BWD_FLAG = 1 << 4,
AOM_ALT_FLAG = 1 << 5,
AOM_REFFRAME_ALL = (1 << 6) - 1
#else
AOM_GOLD_FLAG = 1 << 1,
AOM_ALT_FLAG = 1 << 2,
AOM_REFFRAME_ALL = (1 << 3) - 1
#endif // CONFIG_EXT_REFS
} AOM_REFFRAME;
typedef enum { PLANE_TYPE_Y = 0, PLANE_TYPE_UV = 1, PLANE_TYPES } PLANE_TYPE;
#if CONFIG_CFL
typedef enum { CFL_PRED_U = 0, CFL_PRED_V = 1, CFL_PRED_PLANES } CFL_PRED_TYPE;
#endif
#if CONFIG_PALETTE
typedef enum {
TWO_COLORS,
THREE_COLORS,
FOUR_COLORS,
FIVE_COLORS,
SIX_COLORS,
SEVEN_COLORS,
EIGHT_COLORS,
PALETTE_SIZES
} PALETTE_SIZE;
typedef enum {
PALETTE_COLOR_ONE,
PALETTE_COLOR_TWO,
PALETTE_COLOR_THREE,
PALETTE_COLOR_FOUR,
PALETTE_COLOR_FIVE,
PALETTE_COLOR_SIX,
PALETTE_COLOR_SEVEN,
PALETTE_COLOR_EIGHT,
PALETTE_COLORS
} PALETTE_COLOR;
#endif // CONFIG_PALETTE
typedef enum ATTRIBUTE_PACKED {
DC_PRED, // Average of above and left pixels
V_PRED, // Vertical
H_PRED, // Horizontal
D45_PRED, // Directional 45 deg = round(arctan(1/1) * 180/pi)
D135_PRED, // Directional 135 deg = 180 - 45
D117_PRED, // Directional 117 deg = 180 - 63
D153_PRED, // Directional 153 deg = 180 - 27
D207_PRED, // Directional 207 deg = 180 + 27
D63_PRED, // Directional 63 deg = round(arctan(2/1) * 180/pi)
#if CONFIG_ALT_INTRA
SMOOTH_PRED, // Combination of horizontal and vertical interpolation
#endif // CONFIG_ALT_INTRA
TM_PRED, // True-motion
NEARESTMV,
NEARMV,
ZEROMV,
NEWMV,
#if CONFIG_EXT_INTER
#if CONFIG_COMPOUND_SINGLEREF
// Single ref compound modes
SR_NEAREST_NEARMV,
SR_NEAREST_NEWMV,
SR_NEAR_NEWMV,
SR_ZERO_NEWMV,
SR_NEW_NEWMV,
#endif // CONFIG_COMPOUND_SINGLEREF
// Compound ref compound modes
NEAREST_NEARESTMV,
NEAREST_NEARMV,
NEAR_NEARESTMV,
NEAR_NEARMV,
NEAREST_NEWMV,
NEW_NEARESTMV,
NEAR_NEWMV,
NEW_NEARMV,
ZERO_ZEROMV,
NEW_NEWMV,
#endif // CONFIG_EXT_INTER
MB_MODE_COUNT,
INTRA_MODES = TM_PRED + 1,
INTRA_INVALID = MB_MODE_COUNT // For uv_mode in inter blocks
} PREDICTION_MODE;
typedef enum {
SIMPLE_TRANSLATION = 0,
#if CONFIG_MOTION_VAR
OBMC_CAUSAL, // 2-sided OBMC
#endif // CONFIG_MOTION_VAR
#if CONFIG_WARPED_MOTION
WARPED_CAUSAL, // 2-sided WARPED
#endif // CONFIG_WARPED_MOTION
MOTION_MODES
} MOTION_MODE;
// TODO(urvang): Consider adding II_SMOOTH_PRED if it's helpful.
#if CONFIG_EXT_INTER
typedef enum {
II_DC_PRED = 0,
II_V_PRED,
II_H_PRED,
II_D45_PRED,
II_D135_PRED,
II_D117_PRED,
II_D153_PRED,
II_D207_PRED,
II_D63_PRED,
II_TM_PRED,
INTERINTRA_MODES
} INTERINTRA_MODE;
typedef enum {
COMPOUND_AVERAGE = 0,
#if CONFIG_WEDGE
COMPOUND_WEDGE,
#endif // CONFIG_WEDGE
#if CONFIG_COMPOUND_SEGMENT
COMPOUND_SEG,
#endif // CONFIG_COMPOUND_SEGMENT
COMPOUND_TYPES,
} COMPOUND_TYPE;
#endif // CONFIG_EXT_INTER
// TODO(huisu): Consider adding FILTER_SMOOTH_PRED to "FILTER_INTRA_MODE".
#if CONFIG_FILTER_INTRA
typedef enum {
FILTER_DC_PRED,
FILTER_V_PRED,
FILTER_H_PRED,
FILTER_D45_PRED,
FILTER_D135_PRED,
FILTER_D117_PRED,
FILTER_D153_PRED,
FILTER_D207_PRED,
FILTER_D63_PRED,
FILTER_TM_PRED,
FILTER_INTRA_MODES,
} FILTER_INTRA_MODE;
#endif // CONFIG_FILTER_INTRA
#if CONFIG_EXT_INTRA
#define DIRECTIONAL_MODES (INTRA_MODES - 2)
#endif // CONFIG_EXT_INTRA
#define INTER_MODES (1 + NEWMV - NEARESTMV)
#if CONFIG_EXT_INTER
#if CONFIG_COMPOUND_SINGLEREF
#define INTER_SINGLEREF_COMP_MODES (1 + SR_NEW_NEWMV - SR_NEAREST_NEARMV)
#endif // CONFIG_COMPOUND_SINGLEREF
#define INTER_COMPOUND_MODES (1 + NEW_NEWMV - NEAREST_NEARESTMV)
#endif // CONFIG_EXT_INTER
#define SKIP_CONTEXTS 3
#if CONFIG_REF_MV
#define NMV_CONTEXTS 3
#define NEWMV_MODE_CONTEXTS 7
#define ZEROMV_MODE_CONTEXTS 2
#define REFMV_MODE_CONTEXTS 9
#define DRL_MODE_CONTEXTS 5
#define ZEROMV_OFFSET 3
#define REFMV_OFFSET 4
#define NEWMV_CTX_MASK ((1 << ZEROMV_OFFSET) - 1)
#define ZEROMV_CTX_MASK ((1 << (REFMV_OFFSET - ZEROMV_OFFSET)) - 1)
#define REFMV_CTX_MASK ((1 << (8 - REFMV_OFFSET)) - 1)
#define ALL_ZERO_FLAG_OFFSET 8
#define SKIP_NEARESTMV_OFFSET 9
#define SKIP_NEARMV_OFFSET 10
#define SKIP_NEARESTMV_SUB8X8_OFFSET 11
#endif
#define INTER_MODE_CONTEXTS 7
#if CONFIG_DELTA_Q
#define DELTA_Q_SMALL 3
#define DELTA_Q_PROBS (DELTA_Q_SMALL)
#define DEFAULT_DELTA_Q_RES 4
#if CONFIG_EXT_DELTA_Q
#define DELTA_LF_SMALL 3
#define DELTA_LF_PROBS (DELTA_LF_SMALL)
#define DEFAULT_DELTA_LF_RES 2
#endif
#endif
/* Segment Feature Masks */
#define MAX_MV_REF_CANDIDATES 2
#if CONFIG_REF_MV
#define MAX_REF_MV_STACK_SIZE 16
#if CONFIG_EXT_PARTITION
#define REF_CAT_LEVEL 640
#else
#define REF_CAT_LEVEL 255
#endif // CONFIG_EXT_PARTITION
#endif // CONFIG_REF_MV
#define INTRA_INTER_CONTEXTS 4
#define COMP_INTER_CONTEXTS 5
#define REF_CONTEXTS 5
#if CONFIG_EXT_INTER && CONFIG_COMPOUND_SINGLEREF
#define COMP_INTER_MODE_CONTEXTS 4
#endif // CONFIG_EXT_INTER && CONFIG_COMPOUND_SINGLEREF
#if CONFIG_VAR_TX
#define TXFM_PARTITION_CONTEXTS ((TX_SIZES - TX_8X8) * 6 - 2)
typedef uint8_t TXFM_CONTEXT;
#endif
#define NONE_FRAME -1
#define INTRA_FRAME 0
#define LAST_FRAME 1
#if CONFIG_EXT_REFS
#define LAST2_FRAME 2
#define LAST3_FRAME 3
#define GOLDEN_FRAME 4
#define BWDREF_FRAME 5
#define ALTREF_FRAME 6
#define LAST_REF_FRAMES (LAST3_FRAME - LAST_FRAME + 1)
#else
#define GOLDEN_FRAME 2
#define ALTREF_FRAME 3
#endif // CONFIG_EXT_REFS
#define INTER_REFS_PER_FRAME (ALTREF_FRAME - LAST_FRAME + 1)
#define TOTAL_REFS_PER_FRAME (ALTREF_FRAME - INTRA_FRAME + 1)
#define FWD_REFS (GOLDEN_FRAME - LAST_FRAME + 1)
#define FWD_RF_OFFSET(ref) (ref - LAST_FRAME)
#if CONFIG_EXT_REFS
#define BWD_REFS (ALTREF_FRAME - BWDREF_FRAME + 1)
#define BWD_RF_OFFSET(ref) (ref - BWDREF_FRAME)
#else
#define BWD_REFS 1
#define BWD_RF_OFFSET(ref) (ref - ALTREF_FRAME)
#endif // CONFIG_EXT_REFS
#define SINGLE_REFS (FWD_REFS + BWD_REFS)
#define COMP_REFS (FWD_REFS * BWD_REFS)
#if CONFIG_REF_MV
#define MODE_CTX_REF_FRAMES (TOTAL_REFS_PER_FRAME + COMP_REFS)
#else
#define MODE_CTX_REF_FRAMES TOTAL_REFS_PER_FRAME
#endif
#if CONFIG_SUPERTX
#define PARTITION_SUPERTX_CONTEXTS 2
#define MAX_SUPERTX_BLOCK_SIZE BLOCK_32X32
#endif // CONFIG_SUPERTX
#if CONFIG_LOOP_RESTORATION
typedef enum {
RESTORE_NONE = 0,
RESTORE_WIENER = 1,
RESTORE_SGRPROJ = 2,
RESTORE_SWITCHABLE,
RESTORE_SWITCHABLE_TYPES = RESTORE_SWITCHABLE,
RESTORE_TYPES,
} RestorationType;
#endif // CONFIG_LOOP_RESTORATION
#if CONFIG_FRAME_SUPERRES
#define SUPERRES_SCALE_DENOMINATOR 16
#define SUPERRES_SCALE_BITS 3
#define SUPERRES_SCALE_NUMERATOR_MIN 8
#endif // CONFIG_FRAME_SUPERRES
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_ENUMS_H_

360
third_party/aom/av1/common/filter.c vendored Normal file
View file

@ -0,0 +1,360 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <assert.h>
#include "av1/common/filter.h"
DECLARE_ALIGNED(256, static const InterpKernel,
bilinear_filters[SUBPEL_SHIFTS]) = {
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, 0, 0, 120, 8, 0, 0, 0 },
{ 0, 0, 0, 112, 16, 0, 0, 0 }, { 0, 0, 0, 104, 24, 0, 0, 0 },
{ 0, 0, 0, 96, 32, 0, 0, 0 }, { 0, 0, 0, 88, 40, 0, 0, 0 },
{ 0, 0, 0, 80, 48, 0, 0, 0 }, { 0, 0, 0, 72, 56, 0, 0, 0 },
{ 0, 0, 0, 64, 64, 0, 0, 0 }, { 0, 0, 0, 56, 72, 0, 0, 0 },
{ 0, 0, 0, 48, 80, 0, 0, 0 }, { 0, 0, 0, 40, 88, 0, 0, 0 },
{ 0, 0, 0, 32, 96, 0, 0, 0 }, { 0, 0, 0, 24, 104, 0, 0, 0 },
{ 0, 0, 0, 16, 112, 0, 0, 0 }, { 0, 0, 0, 8, 120, 0, 0, 0 }
};
#if USE_TEMPORALFILTER_12TAP
DECLARE_ALIGNED(16, static const int16_t,
sub_pel_filters_temporalfilter_12[SUBPEL_SHIFTS][12]) = {
// intfilt 0.8
{ 0, 0, 0, 0, 0, 128, 0, 0, 0, 0, 0, 0 },
{ 0, 1, -1, 3, -7, 127, 8, -4, 2, -1, 0, 0 },
{ 0, 1, -3, 5, -12, 124, 18, -8, 4, -2, 1, 0 },
{ -1, 2, -4, 8, -17, 120, 28, -11, 6, -3, 1, -1 },
{ -1, 2, -4, 10, -21, 114, 38, -15, 8, -4, 2, -1 },
{ -1, 3, -5, 11, -23, 107, 49, -18, 9, -5, 2, -1 },
{ -1, 3, -6, 12, -25, 99, 60, -21, 11, -6, 3, -1 },
{ -1, 3, -6, 12, -25, 90, 70, -23, 12, -6, 3, -1 },
{ -1, 3, -6, 12, -24, 80, 80, -24, 12, -6, 3, -1 },
{ -1, 3, -6, 12, -23, 70, 90, -25, 12, -6, 3, -1 },
{ -1, 3, -6, 11, -21, 60, 99, -25, 12, -6, 3, -1 },
{ -1, 2, -5, 9, -18, 49, 107, -23, 11, -5, 3, -1 },
{ -1, 2, -4, 8, -15, 38, 114, -21, 10, -4, 2, -1 },
{ -1, 1, -3, 6, -11, 28, 120, -17, 8, -4, 2, -1 },
{ 0, 1, -2, 4, -8, 18, 124, -12, 5, -3, 1, 0 },
{ 0, 0, -1, 2, -4, 8, 127, -7, 3, -1, 1, 0 },
};
#endif // USE_TEMPORALFILTER_12TAP
#if CONFIG_DUAL_FILTER
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_8[SUBPEL_SHIFTS]) = {
// intfilt 0.575
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, 1, -5, 126, 8, -3, 1, 0 },
{ -1, 3, -10, 123, 18, -6, 2, -1 }, { -1, 4, -14, 118, 27, -9, 3, 0 },
{ -1, 5, -16, 112, 37, -12, 4, -1 }, { -1, 5, -18, 105, 48, -14, 4, -1 },
{ -1, 6, -19, 97, 58, -17, 5, -1 }, { -1, 6, -20, 88, 68, -18, 6, -1 },
{ -1, 6, -19, 78, 78, -19, 6, -1 }, { -1, 6, -18, 68, 88, -20, 6, -1 },
{ -1, 5, -17, 58, 97, -19, 6, -1 }, { -1, 4, -14, 48, 105, -18, 5, -1 },
{ -1, 4, -12, 37, 112, -16, 5, -1 }, { 0, 3, -9, 27, 118, -14, 4, -1 },
{ -1, 2, -6, 18, 123, -10, 3, -1 }, { 0, 1, -3, 8, 126, -5, 1, 0 },
};
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_regular_uv[SUBPEL_SHIFTS]) = {
// intfilt 0.575
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, 1, -5, 126, 8, -3, 1, 0 },
{ -1, 3, -10, 123, 18, -6, 2, -1 }, { -1, 4, -14, 118, 27, -9, 3, 0 },
{ -1, 5, -16, 112, 37, -12, 4, -1 }, { -1, 5, -18, 105, 48, -14, 4, -1 },
{ -1, 6, -19, 97, 58, -17, 5, -1 }, { -1, 6, -20, 88, 68, -18, 6, -1 },
{ -1, 6, -19, 78, 78, -19, 6, -1 }, { -1, 6, -18, 68, 88, -20, 6, -1 },
{ -1, 5, -17, 58, 97, -19, 6, -1 }, { -1, 4, -14, 48, 105, -18, 5, -1 },
{ -1, 4, -12, 37, 112, -16, 5, -1 }, { 0, 3, -9, 27, 118, -14, 4, -1 },
{ -1, 2, -6, 18, 123, -10, 3, -1 }, { 0, 1, -3, 8, 126, -5, 1, 0 },
};
#if USE_12TAP_FILTER
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_8sharp[SUBPEL_SHIFTS]) = {
// intfilt 0.8
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { -1, 2, -6, 127, 9, -4, 2, -1 },
{ -2, 5, -12, 124, 18, -7, 4, -2 }, { -2, 7, -16, 119, 28, -11, 5, -2 },
{ -3, 8, -19, 114, 38, -14, 7, -3 }, { -3, 9, -22, 107, 49, -17, 8, -3 },
{ -4, 10, -23, 99, 60, -20, 10, -4 }, { -4, 11, -23, 90, 70, -22, 10, -4 },
{ -4, 11, -23, 80, 80, -23, 11, -4 }, { -4, 10, -22, 70, 90, -23, 11, -4 },
{ -4, 10, -20, 60, 99, -23, 10, -4 }, { -3, 8, -17, 49, 107, -22, 9, -3 },
{ -3, 7, -14, 38, 114, -19, 8, -3 }, { -2, 5, -11, 28, 119, -16, 7, -2 },
{ -2, 4, -7, 18, 124, -12, 5, -2 }, { -1, 2, -4, 9, 127, -6, 2, -1 },
};
DECLARE_ALIGNED(256, static const int16_t,
sub_pel_filters_10sharp[SUBPEL_SHIFTS][12]) = {
// intfilt 0.85
{ 0, 0, 0, 0, 0, 128, 0, 0, 0, 0, 0, 0 },
{ 0, 1, -2, 3, -7, 127, 8, -4, 2, -1, 1, 0 },
{ 0, 1, -3, 6, -13, 124, 18, -8, 4, -2, 1, 0 },
{ 0, 2, -4, 8, -18, 120, 28, -12, 6, -4, 2, 0 },
{ 0, 2, -5, 10, -21, 114, 38, -15, 8, -5, 2, 0 },
{ 0, 3, -6, 11, -24, 107, 49, -19, 10, -6, 3, 0 },
{ 0, 3, -7, 12, -25, 99, 59, -21, 11, -6, 3, 0 },
{ 0, 3, -7, 12, -25, 90, 70, -23, 12, -7, 3, 0 },
{ 0, 3, -7, 12, -25, 81, 81, -25, 12, -7, 3, 0 },
{ 0, 3, -7, 12, -23, 70, 90, -25, 12, -7, 3, 0 },
{ 0, 3, -6, 11, -21, 59, 99, -25, 12, -7, 3, 0 },
{ 0, 3, -6, 10, -19, 49, 107, -24, 11, -6, 3, 0 },
{ 0, 2, -5, 8, -15, 38, 114, -21, 10, -5, 2, 0 },
{ 0, 2, -4, 6, -12, 28, 120, -18, 8, -4, 2, 0 },
{ 0, 1, -2, 4, -8, 18, 124, -13, 6, -3, 1, 0 },
{ 0, 1, -1, 2, -4, 8, 127, -7, 3, -2, 1, 0 },
};
#else
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_8sharp[SUBPEL_SHIFTS]) = {
#if CONFIG_FILTER_7BIT
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { -2, 2, -6, 126, 8, -2, 2, 0 },
{ -2, 6, -12, 124, 16, -6, 4, -2 }, { -2, 8, -18, 120, 26, -10, 6, -2 },
{ -4, 10, -22, 116, 38, -14, 6, -2 }, { -4, 10, -22, 108, 48, -18, 8, -2 },
{ -4, 10, -24, 100, 60, -20, 8, -2 }, { -4, 10, -24, 90, 70, -22, 10, -2 },
{ -4, 12, -24, 80, 80, -24, 12, -4 }, { -2, 10, -22, 70, 90, -24, 10, -4 },
{ -2, 8, -20, 60, 100, -24, 10, -4 }, { -2, 8, -18, 48, 108, -22, 10, -4 },
{ -2, 6, -14, 38, 116, -22, 10, -4 }, { -2, 6, -10, 26, 120, -18, 8, -2 },
{ -2, 4, -6, 16, 124, -12, 6, -2 }, { 0, 2, -2, 8, 126, -6, 2, -2 }
#else
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { -1, 3, -7, 127, 8, -3, 1, 0 },
{ -2, 5, -13, 125, 17, -6, 3, -1 }, { -3, 7, -17, 121, 27, -10, 5, -2 },
{ -4, 9, -20, 115, 37, -13, 6, -2 }, { -4, 10, -23, 108, 48, -16, 8, -3 },
{ -4, 10, -24, 100, 59, -19, 9, -3 }, { -4, 11, -24, 90, 70, -21, 10, -4 },
{ -4, 11, -23, 80, 80, -23, 11, -4 }, { -4, 10, -21, 70, 90, -24, 11, -4 },
{ -3, 9, -19, 59, 100, -24, 10, -4 }, { -3, 8, -16, 48, 108, -23, 10, -4 },
{ -2, 6, -13, 37, 115, -20, 9, -4 }, { -2, 5, -10, 27, 121, -17, 7, -3 },
{ -1, 3, -6, 17, 125, -13, 5, -2 }, { 0, 1, -3, 8, 127, -7, 3, -1 }
#endif
};
#endif
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_8smooth2[SUBPEL_SHIFTS]) = {
// freqmultiplier = 0.2
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, 9, 30, 44, 32, 11, 2, 0 },
{ 0, 8, 28, 44, 34, 12, 2, 0 }, { 0, 7, 27, 44, 35, 13, 2, 0 },
{ 0, 6, 26, 43, 37, 14, 2, 0 }, { 0, 5, 24, 43, 38, 16, 2, 0 },
{ 0, 5, 23, 42, 38, 17, 3, 0 }, { 0, 4, 21, 41, 40, 19, 3, 0 },
{ 0, 4, 20, 40, 40, 20, 4, 0 }, { 0, 3, 19, 40, 41, 21, 4, 0 },
{ 0, 3, 17, 38, 42, 23, 5, 0 }, { 0, 2, 16, 38, 43, 24, 5, 0 },
{ 0, 2, 14, 37, 43, 26, 6, 0 }, { 0, 2, 13, 35, 44, 27, 7, 0 },
{ 0, 2, 12, 34, 44, 28, 8, 0 }, { 0, 2, 11, 32, 44, 30, 9, 0 },
};
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_smooth2_uv[SUBPEL_SHIFTS]) = {
// freqmultiplier = 0.2
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, 9, 30, 44, 32, 11, 2, 0 },
{ 0, 8, 28, 44, 34, 12, 2, 0 }, { 0, 7, 27, 44, 35, 13, 2, 0 },
{ 0, 6, 26, 43, 37, 14, 2, 0 }, { 0, 5, 24, 43, 38, 16, 2, 0 },
{ 0, 5, 23, 42, 38, 17, 3, 0 }, { 0, 4, 21, 41, 40, 19, 3, 0 },
{ 0, 4, 20, 40, 40, 20, 4, 0 }, { 0, 3, 19, 40, 41, 21, 4, 0 },
{ 0, 3, 17, 38, 42, 23, 5, 0 }, { 0, 2, 16, 38, 43, 24, 5, 0 },
{ 0, 2, 14, 37, 43, 26, 6, 0 }, { 0, 2, 13, 35, 44, 27, 7, 0 },
{ 0, 2, 12, 34, 44, 28, 8, 0 }, { 0, 2, 11, 32, 44, 30, 9, 0 },
};
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_8smooth[SUBPEL_SHIFTS]) = {
// freqmultiplier = 0.8
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, -5, 13, 102, 24, -7, 1, 0 },
{ 0, -4, 8, 100, 31, -8, 1, 0 }, { 0, -3, 4, 97, 37, -8, 1, 0 },
{ 0, -2, 0, 94, 44, -9, 1, 0 }, { 0, -2, -3, 90, 51, -9, 1, 0 },
{ 0, -1, -5, 84, 59, -9, 0, 0 }, { 0, 0, -7, 79, 65, -9, 0, 0 },
{ 0, 0, -8, 72, 72, -8, 0, 0 }, { 0, 0, -9, 65, 79, -7, 0, 0 },
{ 0, 0, -9, 59, 84, -5, -1, 0 }, { 0, 1, -9, 51, 90, -3, -2, 0 },
{ 0, 1, -9, 44, 94, 0, -2, 0 }, { 0, 1, -8, 37, 97, 4, -3, 0 },
{ 0, 1, -8, 31, 100, 8, -4, 0 }, { 0, 1, -7, 24, 102, 13, -5, 0 },
};
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_smooth_uv[SUBPEL_SHIFTS]) = {
// freqmultiplier = 0.8
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, -5, 13, 102, 24, -7, 1, 0 },
{ 0, -4, 8, 100, 31, -8, 1, 0 }, { 0, -3, 4, 97, 37, -8, 1, 0 },
{ 0, -2, 0, 94, 44, -9, 1, 0 }, { 0, -2, -3, 90, 51, -9, 1, 0 },
{ 0, -1, -5, 84, 59, -9, 0, 0 }, { 0, 0, -7, 79, 65, -9, 0, 0 },
{ 0, 0, -8, 72, 72, -8, 0, 0 }, { 0, 0, -9, 65, 79, -7, 0, 0 },
{ 0, 0, -9, 59, 84, -5, -1, 0 }, { 0, 1, -9, 51, 90, -3, -2, 0 },
{ 0, 1, -9, 44, 94, 0, -2, 0 }, { 0, 1, -8, 37, 97, 4, -3, 0 },
{ 0, 1, -8, 31, 100, 8, -4, 0 }, { 0, 1, -7, 24, 102, 13, -5, 0 },
};
#else // CONFIG_DUAL_FILTER
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_8[SUBPEL_SHIFTS]) = {
#if CONFIG_FILTER_7BIT
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, 2, -6, 126, 8, -2, 0, 0 },
{ 0, 2, -10, 122, 18, -4, 0, 0 }, { 0, 2, -12, 116, 28, -8, 2, 0 },
{ 0, 2, -14, 110, 38, -10, 2, 0 }, { 0, 2, -14, 102, 48, -12, 2, 0 },
{ 0, 2, -16, 94, 58, -12, 2, 0 }, { 0, 2, -14, 84, 66, -12, 2, 0 },
{ 0, 2, -14, 76, 76, -14, 2, 0 }, { 0, 2, -12, 66, 84, -14, 2, 0 },
{ 0, 2, -12, 58, 94, -16, 2, 0 }, { 0, 2, -12, 48, 102, -14, 2, 0 },
{ 0, 2, -10, 38, 110, -14, 2, 0 }, { 0, 2, -8, 28, 116, -12, 2, 0 },
{ 0, 0, -4, 18, 122, -10, 2, 0 }, { 0, 0, -2, 8, 126, -6, 2, 0 }
#else
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, 1, -5, 126, 8, -3, 1, 0 },
{ -1, 3, -10, 122, 18, -6, 2, 0 }, { -1, 4, -13, 118, 27, -9, 3, -1 },
{ -1, 4, -16, 112, 37, -11, 4, -1 }, { -1, 5, -18, 105, 48, -14, 4, -1 },
{ -1, 5, -19, 97, 58, -16, 5, -1 }, { -1, 6, -19, 88, 68, -18, 5, -1 },
{ -1, 6, -19, 78, 78, -19, 6, -1 }, { -1, 5, -18, 68, 88, -19, 6, -1 },
{ -1, 5, -16, 58, 97, -19, 5, -1 }, { -1, 4, -14, 48, 105, -18, 5, -1 },
{ -1, 4, -11, 37, 112, -16, 4, -1 }, { -1, 3, -9, 27, 118, -13, 4, -1 },
{ 0, 2, -6, 18, 122, -10, 3, -1 }, { 0, 1, -3, 8, 126, -5, 1, 0 }
#endif
};
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_8sharp[SUBPEL_SHIFTS]) = {
#if CONFIG_FILTER_7BIT
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { -2, 2, -6, 126, 8, -2, 2, 0 },
{ -2, 6, -12, 124, 16, -6, 4, -2 }, { -2, 8, -18, 120, 26, -10, 6, -2 },
{ -4, 10, -22, 116, 38, -14, 6, -2 }, { -4, 10, -22, 108, 48, -18, 8, -2 },
{ -4, 10, -24, 100, 60, -20, 8, -2 }, { -4, 10, -24, 90, 70, -22, 10, -2 },
{ -4, 12, -24, 80, 80, -24, 12, -4 }, { -2, 10, -22, 70, 90, -24, 10, -4 },
{ -2, 8, -20, 60, 100, -24, 10, -4 }, { -2, 8, -18, 48, 108, -22, 10, -4 },
{ -2, 6, -14, 38, 116, -22, 10, -4 }, { -2, 6, -10, 26, 120, -18, 8, -2 },
{ -2, 4, -6, 16, 124, -12, 6, -2 }, { 0, 2, -2, 8, 126, -6, 2, -2 }
#else
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { -1, 3, -7, 127, 8, -3, 1, 0 },
{ -2, 5, -13, 125, 17, -6, 3, -1 }, { -3, 7, -17, 121, 27, -10, 5, -2 },
{ -4, 9, -20, 115, 37, -13, 6, -2 }, { -4, 10, -23, 108, 48, -16, 8, -3 },
{ -4, 10, -24, 100, 59, -19, 9, -3 }, { -4, 11, -24, 90, 70, -21, 10, -4 },
{ -4, 11, -23, 80, 80, -23, 11, -4 }, { -4, 10, -21, 70, 90, -24, 11, -4 },
{ -3, 9, -19, 59, 100, -24, 10, -4 }, { -3, 8, -16, 48, 108, -23, 10, -4 },
{ -2, 6, -13, 37, 115, -20, 9, -4 }, { -2, 5, -10, 27, 121, -17, 7, -3 },
{ -1, 3, -6, 17, 125, -13, 5, -2 }, { 0, 1, -3, 8, 127, -7, 3, -1 }
#endif
};
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_8smooth[SUBPEL_SHIFTS]) = {
#if CONFIG_FILTER_7BIT
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, 2, 28, 62, 34, 2, 0, 0 },
{ 0, 0, 26, 62, 36, 4, 0, 0 }, { 0, 0, 22, 62, 40, 4, 0, 0 },
{ 0, 0, 20, 60, 42, 6, 0, 0 }, { 0, 0, 18, 58, 44, 8, 0, 0 },
{ 0, 0, 16, 56, 46, 10, 0, 0 }, { 0, -2, 16, 54, 48, 12, 0, 0 },
{ 0, -2, 14, 52, 52, 14, -2, 0 }, { 0, 0, 12, 48, 54, 16, -2, 0 },
{ 0, 0, 10, 46, 56, 16, 0, 0 }, { 0, 0, 8, 44, 58, 18, 0, 0 },
{ 0, 0, 6, 42, 60, 20, 0, 0 }, { 0, 0, 4, 40, 62, 22, 0, 0 },
{ 0, 0, 4, 36, 62, 26, 0, 0 }, { 0, 0, 2, 34, 62, 28, 2, 0 }
#else
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { -3, -1, 32, 64, 38, 1, -3, 0 },
{ -2, -2, 29, 63, 41, 2, -3, 0 }, { -2, -2, 26, 63, 43, 4, -4, 0 },
{ -2, -3, 24, 62, 46, 5, -4, 0 }, { -2, -3, 21, 60, 49, 7, -4, 0 },
{ -1, -4, 18, 59, 51, 9, -4, 0 }, { -1, -4, 16, 57, 53, 12, -4, -1 },
{ -1, -4, 14, 55, 55, 14, -4, -1 }, { -1, -4, 12, 53, 57, 16, -4, -1 },
{ 0, -4, 9, 51, 59, 18, -4, -1 }, { 0, -4, 7, 49, 60, 21, -3, -2 },
{ 0, -4, 5, 46, 62, 24, -3, -2 }, { 0, -4, 4, 43, 63, 26, -2, -2 },
{ 0, -3, 2, 41, 63, 29, -2, -2 }, { 0, -3, 1, 38, 64, 32, -1, -3 }
#endif
};
#endif // CONFIG_DUAL_FILTER
#if CONFIG_EXT_INTRA
#if CONFIG_INTRA_INTERP
const InterpKernel *av1_intra_filter_kernels[INTRA_FILTERS] = {
bilinear_filters, // INTRA_FILTER_LINEAR
sub_pel_filters_8, // INTRA_FILTER_8TAP
sub_pel_filters_8sharp, // INTRA_FILTER_8TAP_SHARP
sub_pel_filters_8smooth, // INTRA_FILTER_8TAP_SMOOTH
};
#endif // CONFIG_INTRA_INTERP
#endif // CONFIG_EXT_INTRA
#if CONFIG_DUAL_FILTER
static const InterpFilterParams
av1_interp_filter_params_list[SWITCHABLE_FILTERS + EXTRA_FILTERS] = {
{ (const int16_t *)sub_pel_filters_8, SUBPEL_TAPS, SUBPEL_SHIFTS,
EIGHTTAP_REGULAR },
{ (const int16_t *)sub_pel_filters_8smooth, SUBPEL_TAPS, SUBPEL_SHIFTS,
EIGHTTAP_SMOOTH },
#if USE_12TAP_FILTER
{ (const int16_t *)sub_pel_filters_10sharp, 12, SUBPEL_SHIFTS,
MULTITAP_SHARP },
#else
{ (const int16_t *)sub_pel_filters_8sharp, SUBPEL_TAPS, SUBPEL_SHIFTS,
EIGHTTAP_SHARP },
#endif
{ (const int16_t *)sub_pel_filters_8smooth2, SUBPEL_TAPS, SUBPEL_SHIFTS,
EIGHTTAP_SMOOTH2 },
{ (const int16_t *)bilinear_filters, SUBPEL_TAPS, SUBPEL_SHIFTS,
BILINEAR },
{ (const int16_t *)sub_pel_filters_8sharp, SUBPEL_TAPS, SUBPEL_SHIFTS,
EIGHTTAP_SHARP },
{ (const int16_t *)sub_pel_filters_regular_uv, SUBPEL_TAPS, SUBPEL_SHIFTS,
FILTER_REGULAR_UV },
{ (const int16_t *)sub_pel_filters_smooth_uv, SUBPEL_TAPS, SUBPEL_SHIFTS,
FILTER_SMOOTH_UV },
{ (const int16_t *)sub_pel_filters_8sharp, SUBPEL_TAPS, SUBPEL_SHIFTS,
FILTER_SHARP_UV },
{ (const int16_t *)sub_pel_filters_smooth2_uv, SUBPEL_TAPS, SUBPEL_SHIFTS,
FILTER_SMOOTH2_UV },
};
#else
static const InterpFilterParams
av1_interp_filter_params_list[SWITCHABLE_FILTERS + 1] = {
{ (const int16_t *)sub_pel_filters_8, SUBPEL_TAPS, SUBPEL_SHIFTS,
EIGHTTAP_REGULAR },
{ (const int16_t *)sub_pel_filters_8smooth, SUBPEL_TAPS, SUBPEL_SHIFTS,
EIGHTTAP_SMOOTH },
{ (const int16_t *)sub_pel_filters_8sharp, SUBPEL_TAPS, SUBPEL_SHIFTS,
MULTITAP_SHARP },
{ (const int16_t *)bilinear_filters, SUBPEL_TAPS, SUBPEL_SHIFTS,
BILINEAR }
};
#endif // CONFIG_DUAL_FILTER
#if USE_TEMPORALFILTER_12TAP
static const InterpFilterParams av1_interp_temporalfilter_12tap = {
(const int16_t *)sub_pel_filters_temporalfilter_12, 12, SUBPEL_SHIFTS,
TEMPORALFILTER_12TAP
};
#endif // USE_TEMPORALFILTER_12TAP
InterpFilterParams av1_get_interp_filter_params(
const InterpFilter interp_filter) {
#if USE_TEMPORALFILTER_12TAP
if (interp_filter == TEMPORALFILTER_12TAP)
return av1_interp_temporalfilter_12tap;
#endif // USE_TEMPORALFILTER_12TAP
return av1_interp_filter_params_list[interp_filter];
}
const int16_t *av1_get_interp_filter_kernel(const InterpFilter interp_filter) {
#if USE_TEMPORALFILTER_12TAP
if (interp_filter == TEMPORALFILTER_12TAP)
return av1_interp_temporalfilter_12tap.filter_ptr;
#endif // USE_TEMPORALFILTER_12TAP
return (const int16_t *)av1_interp_filter_params_list[interp_filter]
.filter_ptr;
}
#if CONFIG_DUAL_FILTER
InterpFilter av1_get_plane_interp_filter(InterpFilter interp_filter,
int plane) {
#if USE_TEMPORALFILTER_12TAP
assert(interp_filter <= EIGHTTAP_SHARP ||
interp_filter == TEMPORALFILTER_12TAP);
#else
assert(interp_filter <= EIGHTTAP_SHARP);
#endif
if (plane == 0) {
return interp_filter;
} else {
switch (interp_filter) {
case EIGHTTAP_REGULAR: return FILTER_REGULAR_UV;
case EIGHTTAP_SMOOTH: return FILTER_SMOOTH_UV;
case MULTITAP_SHARP: return FILTER_SHARP_UV;
case EIGHTTAP_SMOOTH2: return FILTER_SMOOTH2_UV;
default: return interp_filter;
}
}
}
#endif

111
third_party/aom/av1/common/filter.h vendored Normal file
View file

@ -0,0 +1,111 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_FILTER_H_
#define AV1_COMMON_FILTER_H_
#include "./aom_config.h"
#include "aom/aom_integer.h"
#include "aom_dsp/aom_filter.h"
#include "aom_ports/mem.h"
#ifdef __cplusplus
extern "C" {
#endif
#define USE_TEMPORALFILTER_12TAP 1
#define MAX_FILTER_TAP 12
#define USE_12TAP_FILTER 0
typedef enum {
EIGHTTAP_REGULAR,
EIGHTTAP_SMOOTH,
MULTITAP_SHARP,
#if CONFIG_DUAL_FILTER
EIGHTTAP_SMOOTH2,
#endif // CONFIG_DUAL_FILTER
BILINEAR,
#if CONFIG_DUAL_FILTER
EIGHTTAP_SHARP,
FILTER_REGULAR_UV,
FILTER_SMOOTH_UV,
FILTER_SHARP_UV,
FILTER_SMOOTH2_UV,
#endif // CONFIG_DUAL_FILTER
INTERP_FILTERS_ALL,
SWITCHABLE_FILTERS = BILINEAR,
SWITCHABLE = SWITCHABLE_FILTERS + 1, /* the last switchable one */
EXTRA_FILTERS = INTERP_FILTERS_ALL - SWITCHABLE_FILTERS,
#if USE_TEMPORALFILTER_12TAP
TEMPORALFILTER_12TAP = SWITCHABLE_FILTERS + EXTRA_FILTERS,
#endif
} InterpFilter;
#if CONFIG_DUAL_FILTER
#define MAX_SUBPEL_TAPS 12
#define LOG_SWITCHABLE_FILTERS \
3 /* (1 << LOG_SWITCHABLE_FILTERS) > SWITCHABLE_FILTERS */
#define SWITCHABLE_FILTER_CONTEXTS ((SWITCHABLE_FILTERS + 1) * 4)
#define INTER_FILTER_COMP_OFFSET (SWITCHABLE_FILTERS + 1)
#define INTER_FILTER_DIR_OFFSET ((SWITCHABLE_FILTERS + 1) * 2)
#else // CONFIG_DUAL_FILTER
#define LOG_SWITCHABLE_FILTERS \
2 /* (1 << LOG_SWITCHABLE_FILTERS) > SWITCHABLE_FILTERS */
#define SWITCHABLE_FILTER_CONTEXTS (SWITCHABLE_FILTERS + 1)
#endif // CONFIG_DUAL_FILTER
#if CONFIG_EXT_INTRA
#if CONFIG_INTRA_INTERP
typedef enum {
INTRA_FILTER_LINEAR,
INTRA_FILTER_8TAP,
INTRA_FILTER_8TAP_SHARP,
INTRA_FILTER_8TAP_SMOOTH,
INTRA_FILTERS,
} INTRA_FILTER;
extern const InterpKernel *av1_intra_filter_kernels[INTRA_FILTERS];
#endif // CONFIG_INTRA_INTERP
#endif // CONFIG_EXT_INTRA
typedef struct InterpFilterParams {
const int16_t *filter_ptr;
uint16_t taps;
uint16_t subpel_shifts;
InterpFilter interp_filter;
} InterpFilterParams;
InterpFilterParams av1_get_interp_filter_params(
const InterpFilter interp_filter);
const int16_t *av1_get_interp_filter_kernel(const InterpFilter interp_filter);
static INLINE const int16_t *av1_get_interp_filter_subpel_kernel(
const InterpFilterParams filter_params, const int subpel) {
return filter_params.filter_ptr + filter_params.taps * subpel;
}
static INLINE int av1_is_interpolating_filter(
const InterpFilter interp_filter) {
const InterpFilterParams ip = av1_get_interp_filter_params(interp_filter);
return (ip.filter_ptr[ip.taps / 2 - 1] == 128);
}
#if CONFIG_DUAL_FILTER
InterpFilter av1_get_plane_interp_filter(InterpFilter interp_filter, int plane);
#endif
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_FILTER_H_

79
third_party/aom/av1/common/frame_buffers.c vendored Normal file
View file

@ -0,0 +1,79 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <assert.h>
#include "av1/common/frame_buffers.h"
#include "aom_mem/aom_mem.h"
int av1_alloc_internal_frame_buffers(InternalFrameBufferList *list) {
assert(list != NULL);
av1_free_internal_frame_buffers(list);
list->num_internal_frame_buffers =
AOM_MAXIMUM_REF_BUFFERS + AOM_MAXIMUM_WORK_BUFFERS;
list->int_fb = (InternalFrameBuffer *)aom_calloc(
list->num_internal_frame_buffers, sizeof(*list->int_fb));
return (list->int_fb == NULL);
}
void av1_free_internal_frame_buffers(InternalFrameBufferList *list) {
int i;
assert(list != NULL);
for (i = 0; i < list->num_internal_frame_buffers; ++i) {
aom_free(list->int_fb[i].data);
list->int_fb[i].data = NULL;
}
aom_free(list->int_fb);
list->int_fb = NULL;
}
int av1_get_frame_buffer(void *cb_priv, size_t min_size,
aom_codec_frame_buffer_t *fb) {
int i;
InternalFrameBufferList *const int_fb_list =
(InternalFrameBufferList *)cb_priv;
if (int_fb_list == NULL) return -1;
// Find a free frame buffer.
for (i = 0; i < int_fb_list->num_internal_frame_buffers; ++i) {
if (!int_fb_list->int_fb[i].in_use) break;
}
if (i == int_fb_list->num_internal_frame_buffers) return -1;
if (int_fb_list->int_fb[i].size < min_size) {
aom_free(int_fb_list->int_fb[i].data);
// The data must be zeroed to fix a valgrind error from the C loop filter
// due to access uninitialized memory in frame border. It could be
// skipped if border were totally removed.
int_fb_list->int_fb[i].data = (uint8_t *)aom_calloc(1, min_size);
if (!int_fb_list->int_fb[i].data) return -1;
int_fb_list->int_fb[i].size = min_size;
}
fb->data = int_fb_list->int_fb[i].data;
fb->size = int_fb_list->int_fb[i].size;
int_fb_list->int_fb[i].in_use = 1;
// Set the frame buffer's private data to point at the internal frame buffer.
fb->priv = &int_fb_list->int_fb[i];
return 0;
}
int av1_release_frame_buffer(void *cb_priv, aom_codec_frame_buffer_t *fb) {
InternalFrameBuffer *const int_fb = (InternalFrameBuffer *)fb->priv;
(void)cb_priv;
if (int_fb) int_fb->in_use = 0;
return 0;
}

54
third_party/aom/av1/common/frame_buffers.h vendored Normal file
View file

@ -0,0 +1,54 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_FRAME_BUFFERS_H_
#define AV1_COMMON_FRAME_BUFFERS_H_
#include "aom/aom_frame_buffer.h"
#include "aom/aom_integer.h"
#ifdef __cplusplus
extern "C" {
#endif
typedef struct InternalFrameBuffer {
uint8_t *data;
size_t size;
int in_use;
} InternalFrameBuffer;
typedef struct InternalFrameBufferList {
int num_internal_frame_buffers;
InternalFrameBuffer *int_fb;
} InternalFrameBufferList;
// Initializes |list|. Returns 0 on success.
int av1_alloc_internal_frame_buffers(InternalFrameBufferList *list);
// Free any data allocated to the frame buffers.
void av1_free_internal_frame_buffers(InternalFrameBufferList *list);
// Callback used by libaom to request an external frame buffer. |cb_priv|
// Callback private data, which points to an InternalFrameBufferList.
// |min_size| is the minimum size in bytes needed to decode the next frame.
// |fb| pointer to the frame buffer.
int av1_get_frame_buffer(void *cb_priv, size_t min_size,
aom_codec_frame_buffer_t *fb);
// Callback used by libaom when there are no references to the frame buffer.
// |cb_priv| is not used. |fb| pointer to the frame buffer.
int av1_release_frame_buffer(void *cb_priv, aom_codec_frame_buffer_t *fb);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_FRAME_BUFFERS_H_

114
third_party/aom/av1/common/generic_code.c vendored Normal file
View file

@ -0,0 +1,114 @@
/*
* Copyright (c) 2001-2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
/* clang-format off */
#ifdef HAVE_CONFIG_H
# include "config.h"
#endif
#include "generic_code.h"
void aom_cdf_init_q15_1D(uint16_t *cdf, int nsyms, int cdf_size) {
int i;
for (i = 0; i < nsyms; i++)
cdf[i] = AOM_ICDF((i + 1)*CDF_PROB_TOP/nsyms);
#if CONFIG_EC_ADAPT
cdf[cdf_size - 1] = 0;
#endif
}
/** Adapts a Q15 cdf after encoding/decoding a symbol. */
void aom_cdf_adapt_q15(int val, uint16_t *cdf, int n, int *count, int rate) {
int i;
*count = OD_MINI(*count + 1, 1 << rate);
OD_ASSERT(AOM_ICDF(cdf[n - 1]) == 32768);
if (*count >= 1 << rate) {
/* Steady-state adaptation based on a simple IIR with dyadic rate. */
for (i = 0; i < n; i++) {
int tmp;
/* When (i < val), we want the adjustment ((cdf[i] - tmp) >> rate) to be
positive so long as (cdf[i] > i + 1), and 0 when (cdf[i] == i + 1),
to ensure we don't drive any probabilities to 0. Replacing cdf[i] with
(i + 2) and solving ((i + 2 - tmp) >> rate == 1) for tmp produces
tmp == i + 2 - (1 << rate). Using this value of tmp with
cdf[i] == i + 1 instead gives an adjustment of 0 as desired.
When (i >= val), we want ((cdf[i] - tmp) >> rate) to be negative so
long as cdf[i] < 32768 - (n - 1 - i), and 0 when
cdf[i] == 32768 - (n - 1 - i), again to ensure we don't drive any
probabilities to 0. Since right-shifting any negative value is still
negative, we can solve (32768 - (n - 1 - i) - tmp == 0) for tmp,
producing tmp = 32769 - n + i. Using this value of tmp with smaller
values of cdf[i] instead gives negative adjustments, as desired.
Combining the two cases gives the expression below. These could be
stored in a lookup table indexed by n and rate to avoid the
arithmetic. */
tmp = 2 - (1<<rate) + i + (32767 + (1<<rate) - n)*(i >= val);
cdf[i] = AOM_ICDF(AOM_ICDF(cdf[i]) - ((AOM_ICDF(cdf[i]) - tmp) >> rate));
}
}
else {
int alpha;
/* Initial adaptation for the first symbols. The adaptation rate is
computed to be equivalent to what od_{en,de}code_cdf_adapt() does
when the initial cdf is set to increment/4. */
alpha = 4*32768/(n + 4**count);
for (i = 0; i < n; i++) {
int tmp;
tmp = (32768 - n)*(i >= val) + i + 1;
cdf[i] = AOM_ICDF(AOM_ICDF(cdf[i])
- (((AOM_ICDF(cdf[i]) - tmp)*alpha) >> 15));
}
}
OD_ASSERT(AOM_ICDF(cdf[n - 1]) == 32768);
}
/** Takes the base-2 log of E(x) in Q1.
*
* @param [in] ExQ16 expectation of x in Q16
*
* @retval 2*log2(ExQ16/2^16)
*/
int log_ex(int ex_q16) {
int lg;
int lg_q1;
int odd;
lg = OD_ILOG(ex_q16);
if (lg < 15) {
odd = ex_q16*ex_q16 > 2 << 2*lg;
}
else {
int tmp;
tmp = ex_q16 >> (lg - 8);
odd = tmp*tmp > (1 << 15);
}
lg_q1 = OD_MAXI(0, 2*lg - 33 + odd);
return lg_q1;
}
/** Updates the probability model based on the encoded/decoded value
*
* @param [in,out] model generic prob model
* @param [in,out] ExQ16 expectation of x
* @param [in] x variable encoded/decoded (used for ExQ16)
* @param [in] xs variable x after shift (used for the model)
* @param [in] id id of the icdf to adapt
* @param [in] integration integration period of ExQ16 (leaky average over
* 1<<integration samples)
*/
void generic_model_update(int *ex_q16, int x, int integration) {
/* We could have saturated ExQ16 directly, but this is safe and simpler */
x = OD_MINI(x, 32767);
OD_IIR_DIADIC(*ex_q16, x << 16, integration);
}

81
third_party/aom/av1/common/generic_code.h vendored Normal file
View file

@ -0,0 +1,81 @@
/*
* Copyright (c) 2001-2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
/* clang-format off */
#if !defined(_generic_code_H)
# define _generic_code_H
# include "aom_dsp/bitreader.h"
# include "aom_dsp/bitwriter.h"
# define GENERIC_TABLES 12
#define generic_decode(r, model, ex_q16, integration, ACCT_STR_NAME) \
generic_decode_(r, model, ex_q16, integration ACCT_STR_ARG(ACCT_STR_NAME))
#define aom_decode_cdf_adapt_q15(r, cdf, n, count, rate, ACCT_STR_NAME) \
aom_decode_cdf_adapt_q15_(r, cdf, n, count, rate ACCT_STR_ARG(ACCT_STR_NAME))
#define aom_decode_cdf_adapt(r, cdf, n, increment, ACCT_STR_NAME) \
aom_decode_cdf_adapt_(r, cdf, n, increment ACCT_STR_ARG(ACCT_STR_NAME))
typedef struct {
/** cdf for multiple expectations of x */
uint16_t cdf[GENERIC_TABLES][CDF_SIZE(16)];
} generic_encoder;
#define OD_IIR_DIADIC(y, x, shift) ((y) += ((x) - (y)) >> (shift))
void generic_model_init(generic_encoder *model);
/* Initialize a CDF for use by aom_write_symbol_pvq()/aom_read_symbol_pvq().
This is used for CDFs whose size might not match the declared array size.
The only real requirement is that the first value of every CDF be zero.
Then aom_cdf_init_q15_1D() will be called with the real size the first time
the CDF is used. */
#define OD_CDFS_INIT_DYNAMIC(cdf) (memset(cdf, 0, sizeof(cdf)))
// WARNING: DO NOT USE this init function,
// if the size of cdf is different from what is declared by code.
#define OD_CDFS_INIT_Q15(cdfs) \
{ int n_cdfs = sizeof(cdfs)/sizeof(cdfs[0]); \
int cdf_size = sizeof(cdfs[0])/sizeof(cdfs[0][0]); \
int nsyms = cdf_size - CONFIG_EC_ADAPT; \
int i_; \
for (i_ = 0; i_ < n_cdfs; i_++) \
aom_cdf_init_q15_1D(cdfs[i_], nsyms, cdf_size); \
}
void aom_cdf_init(uint16_t *cdf, int ncdfs, int nsyms, int val, int first);
void aom_cdf_init_q15_1D(uint16_t *cdf, int nsyms, int cdf_size);
void aom_cdf_adapt_q15(int val, uint16_t *cdf, int n, int *count, int rate);
void aom_encode_cdf_adapt_q15(aom_writer *w, int val, uint16_t *cdf, int n,
int *count, int rate);
void generic_encode(aom_writer *w, generic_encoder *model, int x,
int *ex_q16, int integration);
double generic_encode_cost(generic_encoder *model, int x, int *ex_q16);
double od_encode_cdf_cost(int val, uint16_t *cdf, int n);
int aom_decode_cdf_adapt_q15_(aom_reader *r, uint16_t *cdf, int n,
int *count, int rate ACCT_STR_PARAM);
int generic_decode_(aom_reader *r, generic_encoder *model,
int *ex_q16, int integration ACCT_STR_PARAM);
int log_ex(int ex_q16);
void generic_model_update(int *ex_q16, int x, int integration);
#endif

3067
third_party/aom/av1/common/idct.c vendored Normal file
View file

File diff suppressed because it is too large Load diff

99
third_party/aom/av1/common/idct.h vendored Normal file
View file

@ -0,0 +1,99 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_IDCT_H_
#define AV1_COMMON_IDCT_H_
#include <assert.h>
#include "./aom_config.h"
#include "av1/common/blockd.h"
#include "av1/common/common.h"
#include "av1/common/enums.h"
#include "aom_dsp/inv_txfm.h"
#include "aom_dsp/txfm_common.h"
#include "aom_ports/mem.h"
#ifdef __cplusplus
extern "C" {
#endif
typedef struct INV_TXFM_PARAM {
#if CONFIG_ADAPT_SCAN
const int16_t *eob_threshold;
#endif
TX_TYPE tx_type;
TX_SIZE tx_size;
int eob;
int lossless;
#if CONFIG_HIGHBITDEPTH
int bd;
#endif
} INV_TXFM_PARAM;
typedef void (*transform_1d)(const tran_low_t *, tran_low_t *);
typedef struct {
transform_1d cols, rows; // vertical and horizontal
} transform_2d;
#if CONFIG_HIGHBITDEPTH
typedef void (*highbd_transform_1d)(const tran_low_t *, tran_low_t *,
const int8_t *cos_bit,
const int8_t *stage_range, int bd);
typedef struct {
highbd_transform_1d cols, rows; // vertical and horizontal
} highbd_transform_2d;
#endif // CONFIG_HIGHBITDEPTH
#define MAX_TX_SCALE 1
int av1_get_tx_scale(const TX_SIZE tx_size);
void av1_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob);
void av1_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob);
void av1_inv_txfm_add_4x4(const tran_low_t *input, uint8_t *dest, int stride,
int eob, TX_TYPE tx_type, int lossless);
void av1_inv_txfm_add_8x4(const tran_low_t *input, uint8_t *dest, int stride,
int eob, TX_TYPE tx_type);
void av1_inv_txfm_add_4x8(const tran_low_t *input, uint8_t *dest, int stride,
int eob, TX_TYPE tx_type);
void av1_inv_txfm_add(const tran_low_t *input, uint8_t *dest, int stride,
INV_TXFM_PARAM *inv_txfm_param);
void av1_inverse_transform_block(const MACROBLOCKD *xd,
const tran_low_t *dqcoeff, TX_TYPE tx_type,
TX_SIZE tx_size, uint8_t *dst, int stride,
int eob);
void av1_inverse_transform_block_facade(MACROBLOCKD *xd, int plane, int block,
int blk_row, int blk_col, int eob);
#if CONFIG_HIGHBITDEPTH
void av1_highbd_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob, int bd);
void av1_highbd_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob, int bd);
void av1_highbd_inv_txfm_add_4x4(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd, TX_TYPE tx_type,
int lossless);
void av1_highbd_inv_txfm_add_4x8(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd, TX_TYPE tx_type);
void av1_highbd_inv_txfm_add_8x4(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd, TX_TYPE tx_type);
void av1_highbd_inv_txfm_add(const tran_low_t *input, uint8_t *dest, int stride,
INV_TXFM_PARAM *inv_txfm_param);
#endif // CONFIG_HIGHBITDEPTH
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_IDCT_H_

657
third_party/aom/av1/common/laplace_tables.c vendored Normal file
View file

@ -0,0 +1,657 @@
/* This file is auto-generated using "gen_laplace_tables 128 7" */
/* clang-format off */
#ifdef HAVE_CONFIG_H
# include "config.h"
#endif
#include "aom_dsp/prob.h"
#include "pvq.h"
const uint16_t EXP_CDF_TABLE[128][16] = {
{AOM_ICDF(32753), AOM_ICDF(32754), AOM_ICDF(32755), AOM_ICDF(32756),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(32499), AOM_ICDF(32753), AOM_ICDF(32755), AOM_ICDF(32756),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(32243), AOM_ICDF(32747), AOM_ICDF(32755), AOM_ICDF(32756),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(31987), AOM_ICDF(32737), AOM_ICDF(32755), AOM_ICDF(32756),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(31732), AOM_ICDF(32724), AOM_ICDF(32755), AOM_ICDF(32756),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(31476), AOM_ICDF(32706), AOM_ICDF(32754), AOM_ICDF(32756),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(31220), AOM_ICDF(32684), AOM_ICDF(32753), AOM_ICDF(32756),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(30964), AOM_ICDF(32658), AOM_ICDF(32751), AOM_ICDF(32756),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(30708), AOM_ICDF(32628), AOM_ICDF(32748), AOM_ICDF(32756),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(30452), AOM_ICDF(32594), AOM_ICDF(32745), AOM_ICDF(32756),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(30198), AOM_ICDF(32558), AOM_ICDF(32742), AOM_ICDF(32756),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(29941), AOM_ICDF(32515), AOM_ICDF(32736), AOM_ICDF(32755),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(29686), AOM_ICDF(32470), AOM_ICDF(32731), AOM_ICDF(32755),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(29429), AOM_ICDF(32419), AOM_ICDF(32723), AOM_ICDF(32754),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(29174), AOM_ICDF(32366), AOM_ICDF(32715), AOM_ICDF(32753),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(28918), AOM_ICDF(32308), AOM_ICDF(32705), AOM_ICDF(32752),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(28662), AOM_ICDF(32246), AOM_ICDF(32694), AOM_ICDF(32750),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(28406), AOM_ICDF(32180), AOM_ICDF(32681), AOM_ICDF(32748),
AOM_ICDF(32757), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(28150), AOM_ICDF(32110), AOM_ICDF(32667), AOM_ICDF(32745),
AOM_ICDF(32756), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(27894), AOM_ICDF(32036), AOM_ICDF(32651), AOM_ICDF(32742),
AOM_ICDF(32756), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(27639), AOM_ICDF(31959), AOM_ICDF(32634), AOM_ICDF(32739),
AOM_ICDF(32755), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(27383), AOM_ICDF(31877), AOM_ICDF(32614), AOM_ICDF(32735),
AOM_ICDF(32755), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(27126), AOM_ICDF(31790), AOM_ICDF(32592), AOM_ICDF(32730),
AOM_ICDF(32754), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(26871), AOM_ICDF(31701), AOM_ICDF(32569), AOM_ICDF(32725),
AOM_ICDF(32753), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(26615), AOM_ICDF(31607), AOM_ICDF(32543), AOM_ICDF(32719),
AOM_ICDF(32752), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(26361), AOM_ICDF(31511), AOM_ICDF(32517), AOM_ICDF(32713),
AOM_ICDF(32751), AOM_ICDF(32758), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(26104), AOM_ICDF(31408), AOM_ICDF(32485), AOM_ICDF(32704),
AOM_ICDF(32748), AOM_ICDF(32757), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(25848), AOM_ICDF(31302), AOM_ICDF(32452), AOM_ICDF(32695),
AOM_ICDF(32746), AOM_ICDF(32757), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(25591), AOM_ICDF(31191), AOM_ICDF(32416), AOM_ICDF(32684),
AOM_ICDF(32743), AOM_ICDF(32756), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(25336), AOM_ICDF(31078), AOM_ICDF(32379), AOM_ICDF(32674),
AOM_ICDF(32741), AOM_ICDF(32756), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(25080), AOM_ICDF(30960), AOM_ICDF(32338), AOM_ICDF(32661),
AOM_ICDF(32737), AOM_ICDF(32755), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(24824), AOM_ICDF(30838), AOM_ICDF(32295), AOM_ICDF(32648),
AOM_ICDF(32733), AOM_ICDF(32754), AOM_ICDF(32759), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(24568), AOM_ICDF(30712), AOM_ICDF(32248), AOM_ICDF(32632),
AOM_ICDF(32728), AOM_ICDF(32752), AOM_ICDF(32758), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(24313), AOM_ICDF(30583), AOM_ICDF(32199), AOM_ICDF(32616),
AOM_ICDF(32723), AOM_ICDF(32751), AOM_ICDF(32758), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(24057), AOM_ICDF(30449), AOM_ICDF(32147), AOM_ICDF(32598),
AOM_ICDF(32718), AOM_ICDF(32750), AOM_ICDF(32758), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(23801), AOM_ICDF(30311), AOM_ICDF(32091), AOM_ICDF(32578),
AOM_ICDF(32711), AOM_ICDF(32747), AOM_ICDF(32757), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(23546), AOM_ICDF(30170), AOM_ICDF(32033), AOM_ICDF(32557),
AOM_ICDF(32704), AOM_ICDF(32745), AOM_ICDF(32757), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(23288), AOM_ICDF(30022), AOM_ICDF(31969), AOM_ICDF(32532),
AOM_ICDF(32695), AOM_ICDF(32742), AOM_ICDF(32756), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(23033), AOM_ICDF(29873), AOM_ICDF(31904), AOM_ICDF(32507),
AOM_ICDF(32686), AOM_ICDF(32739), AOM_ICDF(32755), AOM_ICDF(32760),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(22778), AOM_ICDF(29720), AOM_ICDF(31835), AOM_ICDF(32479),
AOM_ICDF(32675), AOM_ICDF(32735), AOM_ICDF(32753), AOM_ICDF(32759),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(22521), AOM_ICDF(29561), AOM_ICDF(31761), AOM_ICDF(32449),
AOM_ICDF(32664), AOM_ICDF(32731), AOM_ICDF(32752), AOM_ICDF(32759),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(22267), AOM_ICDF(29401), AOM_ICDF(31686), AOM_ICDF(32418),
AOM_ICDF(32652), AOM_ICDF(32727), AOM_ICDF(32751), AOM_ICDF(32759),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(22011), AOM_ICDF(29235), AOM_ICDF(31605), AOM_ICDF(32383),
AOM_ICDF(32638), AOM_ICDF(32722), AOM_ICDF(32749), AOM_ICDF(32758),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(21754), AOM_ICDF(29064), AOM_ICDF(31520), AOM_ICDF(32345),
AOM_ICDF(32622), AOM_ICDF(32715), AOM_ICDF(32746), AOM_ICDF(32757),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(21501), AOM_ICDF(28893), AOM_ICDF(31434), AOM_ICDF(32307),
AOM_ICDF(32607), AOM_ICDF(32710), AOM_ICDF(32745), AOM_ICDF(32757),
AOM_ICDF(32761), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(21243), AOM_ICDF(28713), AOM_ICDF(31339), AOM_ICDF(32262),
AOM_ICDF(32587), AOM_ICDF(32701), AOM_ICDF(32741), AOM_ICDF(32755),
AOM_ICDF(32760), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(20988), AOM_ICDF(28532), AOM_ICDF(31243), AOM_ICDF(32217),
AOM_ICDF(32567), AOM_ICDF(32693), AOM_ICDF(32738), AOM_ICDF(32754),
AOM_ICDF(32760), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(20730), AOM_ICDF(28344), AOM_ICDF(31140), AOM_ICDF(32167),
AOM_ICDF(32544), AOM_ICDF(32682), AOM_ICDF(32733), AOM_ICDF(32752),
AOM_ICDF(32759), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(20476), AOM_ICDF(28156), AOM_ICDF(31036), AOM_ICDF(32116),
AOM_ICDF(32521), AOM_ICDF(32673), AOM_ICDF(32730), AOM_ICDF(32751),
AOM_ICDF(32759), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(20220), AOM_ICDF(27962), AOM_ICDF(30926), AOM_ICDF(32061),
AOM_ICDF(32495), AOM_ICDF(32661), AOM_ICDF(32725), AOM_ICDF(32749),
AOM_ICDF(32758), AOM_ICDF(32762), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(19963), AOM_ICDF(27763), AOM_ICDF(30810), AOM_ICDF(32000),
AOM_ICDF(32465), AOM_ICDF(32647), AOM_ICDF(32718), AOM_ICDF(32746),
AOM_ICDF(32757), AOM_ICDF(32761), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(19708), AOM_ICDF(27562), AOM_ICDF(30691), AOM_ICDF(31938),
AOM_ICDF(32435), AOM_ICDF(32633), AOM_ICDF(32712), AOM_ICDF(32743),
AOM_ICDF(32756), AOM_ICDF(32761), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(19454), AOM_ICDF(27358), AOM_ICDF(30569), AOM_ICDF(31873),
AOM_ICDF(32403), AOM_ICDF(32618), AOM_ICDF(32705), AOM_ICDF(32741),
AOM_ICDF(32755), AOM_ICDF(32761), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(19196), AOM_ICDF(27146), AOM_ICDF(30438), AOM_ICDF(31801),
AOM_ICDF(32365), AOM_ICDF(32599), AOM_ICDF(32696), AOM_ICDF(32736),
AOM_ICDF(32753), AOM_ICDF(32760), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(18942), AOM_ICDF(26934), AOM_ICDF(30306), AOM_ICDF(31728),
AOM_ICDF(32328), AOM_ICDF(32581), AOM_ICDF(32688), AOM_ICDF(32733),
AOM_ICDF(32752), AOM_ICDF(32760), AOM_ICDF(32763), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(18684), AOM_ICDF(26714), AOM_ICDF(30164), AOM_ICDF(31647),
AOM_ICDF(32284), AOM_ICDF(32558), AOM_ICDF(32676), AOM_ICDF(32727),
AOM_ICDF(32749), AOM_ICDF(32758), AOM_ICDF(32762), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(18429), AOM_ICDF(26493), AOM_ICDF(30021), AOM_ICDF(31565),
AOM_ICDF(32240), AOM_ICDF(32535), AOM_ICDF(32664), AOM_ICDF(32721),
AOM_ICDF(32746), AOM_ICDF(32757), AOM_ICDF(32762), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(18174), AOM_ICDF(26268), AOM_ICDF(29872), AOM_ICDF(31477),
AOM_ICDF(32192), AOM_ICDF(32510), AOM_ICDF(32652), AOM_ICDF(32715),
AOM_ICDF(32743), AOM_ICDF(32756), AOM_ICDF(32762), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(17920), AOM_ICDF(26040), AOM_ICDF(29719), AOM_ICDF(31386),
AOM_ICDF(32141), AOM_ICDF(32483), AOM_ICDF(32638), AOM_ICDF(32708),
AOM_ICDF(32740), AOM_ICDF(32754), AOM_ICDF(32761), AOM_ICDF(32764),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(17661), AOM_ICDF(25803), AOM_ICDF(29556), AOM_ICDF(31286),
AOM_ICDF(32083), AOM_ICDF(32451), AOM_ICDF(32620), AOM_ICDF(32698),
AOM_ICDF(32734), AOM_ICDF(32751), AOM_ICDF(32759), AOM_ICDF(32763),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(17406), AOM_ICDF(25566), AOM_ICDF(29391), AOM_ICDF(31184),
AOM_ICDF(32024), AOM_ICDF(32418), AOM_ICDF(32603), AOM_ICDF(32690),
AOM_ICDF(32731), AOM_ICDF(32750), AOM_ICDF(32759), AOM_ICDF(32763),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(17151), AOM_ICDF(25325), AOM_ICDF(29220), AOM_ICDF(31076),
AOM_ICDF(31961), AOM_ICDF(32383), AOM_ICDF(32584), AOM_ICDF(32680),
AOM_ICDF(32726), AOM_ICDF(32748), AOM_ICDF(32758), AOM_ICDF(32763),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(16896), AOM_ICDF(25080), AOM_ICDF(29044), AOM_ICDF(30964),
AOM_ICDF(31894), AOM_ICDF(32344), AOM_ICDF(32562), AOM_ICDF(32668),
AOM_ICDF(32719), AOM_ICDF(32744), AOM_ICDF(32756), AOM_ICDF(32762),
AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(16639), AOM_ICDF(24829), AOM_ICDF(28860), AOM_ICDF(30844),
AOM_ICDF(31821), AOM_ICDF(32302), AOM_ICDF(32539), AOM_ICDF(32655),
AOM_ICDF(32712), AOM_ICDF(32740), AOM_ICDF(32754), AOM_ICDF(32761),
AOM_ICDF(32764), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(16384), AOM_ICDF(24576), AOM_ICDF(28672), AOM_ICDF(30720),
AOM_ICDF(31744), AOM_ICDF(32256), AOM_ICDF(32512), AOM_ICDF(32640),
AOM_ICDF(32704), AOM_ICDF(32736), AOM_ICDF(32752), AOM_ICDF(32760),
AOM_ICDF(32764), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(16130), AOM_ICDF(24320), AOM_ICDF(28479), AOM_ICDF(30591),
AOM_ICDF(31663), AOM_ICDF(32208), AOM_ICDF(32485), AOM_ICDF(32625),
AOM_ICDF(32696), AOM_ICDF(32732), AOM_ICDF(32750), AOM_ICDF(32759),
AOM_ICDF(32764), AOM_ICDF(32766), AOM_ICDF(32767), AOM_ICDF(32768)},
{AOM_ICDF(15872), AOM_ICDF(24056), AOM_ICDF(28276), AOM_ICDF(30452),
AOM_ICDF(31574), AOM_ICDF(32152), AOM_ICDF(32450), AOM_ICDF(32604),
AOM_ICDF(32683), AOM_ICDF(32724), AOM_ICDF(32745), AOM_ICDF(32756),
AOM_ICDF(32762), AOM_ICDF(32765), AOM_ICDF(32766), AOM_ICDF(32768)},
{AOM_ICDF(15615), AOM_ICDF(23789), AOM_ICDF(28068), AOM_ICDF(30308),
AOM_ICDF(31480), AOM_ICDF(32094), AOM_ICDF(32415), AOM_ICDF(32583),
AOM_ICDF(32671), AOM_ICDF(32717), AOM_ICDF(32741), AOM_ICDF(32754),
AOM_ICDF(32761), AOM_ICDF(32764), AOM_ICDF(32766), AOM_ICDF(32768)},
{AOM_ICDF(15361), AOM_ICDF(23521), AOM_ICDF(27856), AOM_ICDF(30159),
AOM_ICDF(31382), AOM_ICDF(32032), AOM_ICDF(32377), AOM_ICDF(32560),
AOM_ICDF(32657), AOM_ICDF(32709), AOM_ICDF(32737), AOM_ICDF(32752),
AOM_ICDF(32760), AOM_ICDF(32764), AOM_ICDF(32766), AOM_ICDF(32768)},
{AOM_ICDF(15103), AOM_ICDF(23245), AOM_ICDF(27634), AOM_ICDF(30000),
AOM_ICDF(31275), AOM_ICDF(31963), AOM_ICDF(32334), AOM_ICDF(32534),
AOM_ICDF(32642), AOM_ICDF(32700), AOM_ICDF(32731), AOM_ICDF(32748),
AOM_ICDF(32757), AOM_ICDF(32762), AOM_ICDF(32765), AOM_ICDF(32768)},
{AOM_ICDF(14848), AOM_ICDF(22968), AOM_ICDF(27409), AOM_ICDF(29837),
AOM_ICDF(31165), AOM_ICDF(31891), AOM_ICDF(32288), AOM_ICDF(32505),
AOM_ICDF(32624), AOM_ICDF(32689), AOM_ICDF(32725), AOM_ICDF(32744),
AOM_ICDF(32755), AOM_ICDF(32761), AOM_ICDF(32764), AOM_ICDF(32768)},
{AOM_ICDF(14592), AOM_ICDF(22686), AOM_ICDF(27176), AOM_ICDF(29666),
AOM_ICDF(31047), AOM_ICDF(31813), AOM_ICDF(32238), AOM_ICDF(32474),
AOM_ICDF(32605), AOM_ICDF(32678), AOM_ICDF(32718), AOM_ICDF(32740),
AOM_ICDF(32752), AOM_ICDF(32759), AOM_ICDF(32763), AOM_ICDF(32768)},
{AOM_ICDF(14336), AOM_ICDF(22400), AOM_ICDF(26936), AOM_ICDF(29488),
AOM_ICDF(30923), AOM_ICDF(31730), AOM_ICDF(32184), AOM_ICDF(32439),
AOM_ICDF(32583), AOM_ICDF(32664), AOM_ICDF(32709), AOM_ICDF(32735),
AOM_ICDF(32749), AOM_ICDF(32757), AOM_ICDF(32762), AOM_ICDF(32768)},
{AOM_ICDF(14079), AOM_ICDF(22109), AOM_ICDF(26689), AOM_ICDF(29301),
AOM_ICDF(30791), AOM_ICDF(31641), AOM_ICDF(32125), AOM_ICDF(32401),
AOM_ICDF(32559), AOM_ICDF(32649), AOM_ICDF(32700), AOM_ICDF(32729),
AOM_ICDF(32746), AOM_ICDF(32756), AOM_ICDF(32761), AOM_ICDF(32768)},
{AOM_ICDF(13825), AOM_ICDF(21817), AOM_ICDF(26437), AOM_ICDF(29108),
AOM_ICDF(30652), AOM_ICDF(31545), AOM_ICDF(32061), AOM_ICDF(32359),
AOM_ICDF(32532), AOM_ICDF(32632), AOM_ICDF(32690), AOM_ICDF(32723),
AOM_ICDF(32742), AOM_ICDF(32753), AOM_ICDF(32759), AOM_ICDF(32768)},
{AOM_ICDF(13568), AOM_ICDF(21518), AOM_ICDF(26176), AOM_ICDF(28905),
AOM_ICDF(30504), AOM_ICDF(31441), AOM_ICDF(31990), AOM_ICDF(32312),
AOM_ICDF(32501), AOM_ICDF(32611), AOM_ICDF(32676), AOM_ICDF(32714),
AOM_ICDF(32736), AOM_ICDF(32749), AOM_ICDF(32757), AOM_ICDF(32768)},
{AOM_ICDF(13314), AOM_ICDF(21218), AOM_ICDF(25911), AOM_ICDF(28697),
AOM_ICDF(30351), AOM_ICDF(31333), AOM_ICDF(31916), AOM_ICDF(32262),
AOM_ICDF(32468), AOM_ICDF(32590), AOM_ICDF(32662), AOM_ICDF(32705),
AOM_ICDF(32731), AOM_ICDF(32746), AOM_ICDF(32755), AOM_ICDF(32768)},
{AOM_ICDF(13054), AOM_ICDF(20908), AOM_ICDF(25633), AOM_ICDF(28475),
AOM_ICDF(30185), AOM_ICDF(31214), AOM_ICDF(31833), AOM_ICDF(32205),
AOM_ICDF(32429), AOM_ICDF(32564), AOM_ICDF(32645), AOM_ICDF(32694),
AOM_ICDF(32723), AOM_ICDF(32741), AOM_ICDF(32752), AOM_ICDF(32768)},
{AOM_ICDF(12803), AOM_ICDF(20603), AOM_ICDF(25356), AOM_ICDF(28252),
AOM_ICDF(30017), AOM_ICDF(31093), AOM_ICDF(31748), AOM_ICDF(32147),
AOM_ICDF(32390), AOM_ICDF(32538), AOM_ICDF(32628), AOM_ICDF(32683),
AOM_ICDF(32717), AOM_ICDF(32737), AOM_ICDF(32749), AOM_ICDF(32768)},
{AOM_ICDF(12544), AOM_ICDF(20286), AOM_ICDF(25064), AOM_ICDF(28013),
AOM_ICDF(29833), AOM_ICDF(30956), AOM_ICDF(31649), AOM_ICDF(32077),
AOM_ICDF(32341), AOM_ICDF(32504), AOM_ICDF(32605), AOM_ICDF(32667),
AOM_ICDF(32705), AOM_ICDF(32729), AOM_ICDF(32744), AOM_ICDF(32768)},
{AOM_ICDF(12288), AOM_ICDF(19968), AOM_ICDF(24768), AOM_ICDF(27768),
AOM_ICDF(29643), AOM_ICDF(30815), AOM_ICDF(31547), AOM_ICDF(32005),
AOM_ICDF(32291), AOM_ICDF(32470), AOM_ICDF(32582), AOM_ICDF(32652),
AOM_ICDF(32696), AOM_ICDF(32723), AOM_ICDF(32740), AOM_ICDF(32768)},
{AOM_ICDF(12033), AOM_ICDF(19647), AOM_ICDF(24465), AOM_ICDF(27514),
AOM_ICDF(29443), AOM_ICDF(30664), AOM_ICDF(31437), AOM_ICDF(31926),
AOM_ICDF(32235), AOM_ICDF(32431), AOM_ICDF(32555), AOM_ICDF(32633),
AOM_ICDF(32683), AOM_ICDF(32714), AOM_ICDF(32734), AOM_ICDF(32768)},
{AOM_ICDF(11777), AOM_ICDF(19321), AOM_ICDF(24154), AOM_ICDF(27250),
AOM_ICDF(29233), AOM_ICDF(30504), AOM_ICDF(31318), AOM_ICDF(31839),
AOM_ICDF(32173), AOM_ICDF(32387), AOM_ICDF(32524), AOM_ICDF(32612),
AOM_ICDF(32668), AOM_ICDF(32704), AOM_ICDF(32727), AOM_ICDF(32768)},
{AOM_ICDF(11521), AOM_ICDF(18991), AOM_ICDF(23835), AOM_ICDF(26976),
AOM_ICDF(29013), AOM_ICDF(30334), AOM_ICDF(31190), AOM_ICDF(31745),
AOM_ICDF(32105), AOM_ICDF(32338), AOM_ICDF(32489), AOM_ICDF(32587),
AOM_ICDF(32651), AOM_ICDF(32692), AOM_ICDF(32719), AOM_ICDF(32768)},
{AOM_ICDF(11265), AOM_ICDF(18657), AOM_ICDF(23508), AOM_ICDF(26691),
AOM_ICDF(28780), AOM_ICDF(30151), AOM_ICDF(31051), AOM_ICDF(31641),
AOM_ICDF(32028), AOM_ICDF(32282), AOM_ICDF(32449), AOM_ICDF(32559),
AOM_ICDF(32631), AOM_ICDF(32678), AOM_ICDF(32709), AOM_ICDF(32768)},
{AOM_ICDF(11006), AOM_ICDF(18316), AOM_ICDF(23170), AOM_ICDF(26394),
AOM_ICDF(28535), AOM_ICDF(29957), AOM_ICDF(30901), AOM_ICDF(31528),
AOM_ICDF(31944), AOM_ICDF(32220), AOM_ICDF(32404), AOM_ICDF(32526),
AOM_ICDF(32607), AOM_ICDF(32661), AOM_ICDF(32697), AOM_ICDF(32768)},
{AOM_ICDF(10752), AOM_ICDF(17976), AOM_ICDF(22830), AOM_ICDF(26091),
AOM_ICDF(28282), AOM_ICDF(29754), AOM_ICDF(30743), AOM_ICDF(31408),
AOM_ICDF(31854), AOM_ICDF(32154), AOM_ICDF(32356), AOM_ICDF(32491),
AOM_ICDF(32582), AOM_ICDF(32643), AOM_ICDF(32684), AOM_ICDF(32768)},
{AOM_ICDF(10496), AOM_ICDF(17630), AOM_ICDF(22479), AOM_ICDF(25775),
AOM_ICDF(28015), AOM_ICDF(29538), AOM_ICDF(30573), AOM_ICDF(31276),
AOM_ICDF(31754), AOM_ICDF(32079), AOM_ICDF(32300), AOM_ICDF(32450),
AOM_ICDF(32552), AOM_ICDF(32621), AOM_ICDF(32668), AOM_ICDF(32768)},
{AOM_ICDF(10240), AOM_ICDF(17280), AOM_ICDF(22120), AOM_ICDF(25448),
AOM_ICDF(27736), AOM_ICDF(29309), AOM_ICDF(30390), AOM_ICDF(31133),
AOM_ICDF(31644), AOM_ICDF(31995), AOM_ICDF(32237), AOM_ICDF(32403),
AOM_ICDF(32517), AOM_ICDF(32595), AOM_ICDF(32649), AOM_ICDF(32768)},
{ AOM_ICDF(9984), AOM_ICDF(16926), AOM_ICDF(21753), AOM_ICDF(25109),
AOM_ICDF(27443), AOM_ICDF(29066), AOM_ICDF(30194), AOM_ICDF(30978),
AOM_ICDF(31523), AOM_ICDF(31902), AOM_ICDF(32166), AOM_ICDF(32349),
AOM_ICDF(32476), AOM_ICDF(32565), AOM_ICDF(32627), AOM_ICDF(32768)},
{ AOM_ICDF(9728), AOM_ICDF(16568), AOM_ICDF(21377), AOM_ICDF(24759),
AOM_ICDF(27137), AOM_ICDF(28809), AOM_ICDF(29984), AOM_ICDF(30811),
AOM_ICDF(31392), AOM_ICDF(31801), AOM_ICDF(32088), AOM_ICDF(32290),
AOM_ICDF(32432), AOM_ICDF(32532), AOM_ICDF(32602), AOM_ICDF(32768)},
{ AOM_ICDF(9474), AOM_ICDF(16208), AOM_ICDF(20995), AOM_ICDF(24399),
AOM_ICDF(26819), AOM_ICDF(28539), AOM_ICDF(29762), AOM_ICDF(30631),
AOM_ICDF(31249), AOM_ICDF(31688), AOM_ICDF(32000), AOM_ICDF(32222),
AOM_ICDF(32380), AOM_ICDF(32492), AOM_ICDF(32572), AOM_ICDF(32768)},
{ AOM_ICDF(9216), AOM_ICDF(15840), AOM_ICDF(20601), AOM_ICDF(24023),
AOM_ICDF(26483), AOM_ICDF(28251), AOM_ICDF(29522), AOM_ICDF(30435),
AOM_ICDF(31091), AOM_ICDF(31563), AOM_ICDF(31902), AOM_ICDF(32146),
AOM_ICDF(32321), AOM_ICDF(32447), AOM_ICDF(32537), AOM_ICDF(32768)},
{ AOM_ICDF(8959), AOM_ICDF(15469), AOM_ICDF(20199), AOM_ICDF(23636),
AOM_ICDF(26133), AOM_ICDF(27947), AOM_ICDF(29265), AOM_ICDF(30223),
AOM_ICDF(30919), AOM_ICDF(31425), AOM_ICDF(31792), AOM_ICDF(32059),
AOM_ICDF(32253), AOM_ICDF(32394), AOM_ICDF(32496), AOM_ICDF(32768)},
{ AOM_ICDF(8705), AOM_ICDF(15097), AOM_ICDF(19791), AOM_ICDF(23238),
AOM_ICDF(25770), AOM_ICDF(27629), AOM_ICDF(28994), AOM_ICDF(29997),
AOM_ICDF(30733), AOM_ICDF(31274), AOM_ICDF(31671), AOM_ICDF(31963),
AOM_ICDF(32177), AOM_ICDF(32334), AOM_ICDF(32449), AOM_ICDF(32768)},
{ AOM_ICDF(8449), AOM_ICDF(14719), AOM_ICDF(19373), AOM_ICDF(22827),
AOM_ICDF(25390), AOM_ICDF(27292), AOM_ICDF(28704), AOM_ICDF(29752),
AOM_ICDF(30530), AOM_ICDF(31107), AOM_ICDF(31535), AOM_ICDF(31853),
AOM_ICDF(32089), AOM_ICDF(32264), AOM_ICDF(32394), AOM_ICDF(32768)},
{ AOM_ICDF(8192), AOM_ICDF(14336), AOM_ICDF(18944), AOM_ICDF(22400),
AOM_ICDF(24992), AOM_ICDF(26936), AOM_ICDF(28394), AOM_ICDF(29488),
AOM_ICDF(30308), AOM_ICDF(30923), AOM_ICDF(31384), AOM_ICDF(31730),
AOM_ICDF(31989), AOM_ICDF(32184), AOM_ICDF(32330), AOM_ICDF(32768)},
{ AOM_ICDF(7936), AOM_ICDF(13950), AOM_ICDF(18507), AOM_ICDF(21961),
AOM_ICDF(24578), AOM_ICDF(26561), AOM_ICDF(28064), AOM_ICDF(29203),
AOM_ICDF(30066), AOM_ICDF(30720), AOM_ICDF(31216), AOM_ICDF(31592),
AOM_ICDF(31877), AOM_ICDF(32093), AOM_ICDF(32256), AOM_ICDF(32768)},
{ AOM_ICDF(7678), AOM_ICDF(13558), AOM_ICDF(18060), AOM_ICDF(21507),
AOM_ICDF(24146), AOM_ICDF(26166), AOM_ICDF(27713), AOM_ICDF(28897),
AOM_ICDF(29804), AOM_ICDF(30498), AOM_ICDF(31030), AOM_ICDF(31437),
AOM_ICDF(31749), AOM_ICDF(31988), AOM_ICDF(32171), AOM_ICDF(32768)},
{ AOM_ICDF(7423), AOM_ICDF(13165), AOM_ICDF(17606), AOM_ICDF(21041),
AOM_ICDF(23698), AOM_ICDF(25753), AOM_ICDF(27342), AOM_ICDF(28571),
AOM_ICDF(29522), AOM_ICDF(30257), AOM_ICDF(30826), AOM_ICDF(31266),
AOM_ICDF(31606), AOM_ICDF(31869), AOM_ICDF(32073), AOM_ICDF(32768)},
{ AOM_ICDF(7168), AOM_ICDF(12768), AOM_ICDF(17143), AOM_ICDF(20561),
AOM_ICDF(23231), AOM_ICDF(25317), AOM_ICDF(26947), AOM_ICDF(28220),
AOM_ICDF(29215), AOM_ICDF(29992), AOM_ICDF(30599), AOM_ICDF(31073),
AOM_ICDF(31444), AOM_ICDF(31734), AOM_ICDF(31960), AOM_ICDF(32768)},
{ AOM_ICDF(6911), AOM_ICDF(12365), AOM_ICDF(16669), AOM_ICDF(20065),
AOM_ICDF(22744), AOM_ICDF(24858), AOM_ICDF(26526), AOM_ICDF(27842),
AOM_ICDF(28881), AOM_ICDF(29701), AOM_ICDF(30348), AOM_ICDF(30858),
AOM_ICDF(31261), AOM_ICDF(31579), AOM_ICDF(31830), AOM_ICDF(32768)},
{ AOM_ICDF(6657), AOM_ICDF(11961), AOM_ICDF(16188), AOM_ICDF(19556),
AOM_ICDF(22240), AOM_ICDF(24379), AOM_ICDF(26083), AOM_ICDF(27441),
AOM_ICDF(28523), AOM_ICDF(29385), AOM_ICDF(30072), AOM_ICDF(30620),
AOM_ICDF(31056), AOM_ICDF(31404), AOM_ICDF(31681), AOM_ICDF(32768)},
{ AOM_ICDF(6400), AOM_ICDF(11550), AOM_ICDF(15694), AOM_ICDF(19029),
AOM_ICDF(21712), AOM_ICDF(23871), AOM_ICDF(25609), AOM_ICDF(27007),
AOM_ICDF(28132), AOM_ICDF(29037), AOM_ICDF(29766), AOM_ICDF(30352),
AOM_ICDF(30824), AOM_ICDF(31204), AOM_ICDF(31509), AOM_ICDF(32768)},
{ AOM_ICDF(6142), AOM_ICDF(11134), AOM_ICDF(15190), AOM_ICDF(18486),
AOM_ICDF(21164), AOM_ICDF(23340), AOM_ICDF(25108), AOM_ICDF(26544),
AOM_ICDF(27711), AOM_ICDF(28659), AOM_ICDF(29429), AOM_ICDF(30055),
AOM_ICDF(30564), AOM_ICDF(30977), AOM_ICDF(31313), AOM_ICDF(32768)},
{ AOM_ICDF(5890), AOM_ICDF(10720), AOM_ICDF(14682), AOM_ICDF(17932),
AOM_ICDF(20598), AOM_ICDF(22785), AOM_ICDF(24579), AOM_ICDF(26051),
AOM_ICDF(27258), AOM_ICDF(28248), AOM_ICDF(29060), AOM_ICDF(29726),
AOM_ICDF(30273), AOM_ICDF(30721), AOM_ICDF(31089), AOM_ICDF(32768)},
{ AOM_ICDF(5631), AOM_ICDF(10295), AOM_ICDF(14157), AOM_ICDF(17356),
AOM_ICDF(20005), AOM_ICDF(22199), AOM_ICDF(24016), AOM_ICDF(25520),
AOM_ICDF(26766), AOM_ICDF(27798), AOM_ICDF(28652), AOM_ICDF(29359),
AOM_ICDF(29945), AOM_ICDF(30430), AOM_ICDF(30832), AOM_ICDF(32768)},
{ AOM_ICDF(5377), AOM_ICDF(9871), AOM_ICDF(13628), AOM_ICDF(16768),
AOM_ICDF(19393), AOM_ICDF(21587), AOM_ICDF(23421), AOM_ICDF(24954),
AOM_ICDF(26236), AOM_ICDF(27308), AOM_ICDF(28204), AOM_ICDF(28953),
AOM_ICDF(29579), AOM_ICDF(30102), AOM_ICDF(30539), AOM_ICDF(32768)},
{ AOM_ICDF(5121), AOM_ICDF(9441), AOM_ICDF(13086), AOM_ICDF(16161),
AOM_ICDF(18756), AOM_ICDF(20945), AOM_ICDF(22792), AOM_ICDF(24351),
AOM_ICDF(25666), AOM_ICDF(26776), AOM_ICDF(27712), AOM_ICDF(28502),
AOM_ICDF(29169), AOM_ICDF(29731), AOM_ICDF(30206), AOM_ICDF(32768)},
{ AOM_ICDF(4865), AOM_ICDF(9007), AOM_ICDF(12534), AOM_ICDF(15538),
AOM_ICDF(18096), AOM_ICDF(20274), AOM_ICDF(22129), AOM_ICDF(23708),
AOM_ICDF(25053), AOM_ICDF(26198), AOM_ICDF(27173), AOM_ICDF(28004),
AOM_ICDF(28711), AOM_ICDF(29313), AOM_ICDF(29826), AOM_ICDF(32768)},
{ AOM_ICDF(4608), AOM_ICDF(8568), AOM_ICDF(11971), AOM_ICDF(14896),
AOM_ICDF(17409), AOM_ICDF(19569), AOM_ICDF(21425), AOM_ICDF(23020),
AOM_ICDF(24391), AOM_ICDF(25569), AOM_ICDF(26581), AOM_ICDF(27451),
AOM_ICDF(28199), AOM_ICDF(28842), AOM_ICDF(29394), AOM_ICDF(32768)},
{ AOM_ICDF(4351), AOM_ICDF(8125), AOM_ICDF(11398), AOM_ICDF(14236),
AOM_ICDF(16697), AOM_ICDF(18831), AOM_ICDF(20682), AOM_ICDF(22287),
AOM_ICDF(23679), AOM_ICDF(24886), AOM_ICDF(25933), AOM_ICDF(26841),
AOM_ICDF(27628), AOM_ICDF(28311), AOM_ICDF(28903), AOM_ICDF(32768)},
{ AOM_ICDF(4096), AOM_ICDF(7680), AOM_ICDF(10816), AOM_ICDF(13560),
AOM_ICDF(15961), AOM_ICDF(18062), AOM_ICDF(19900), AOM_ICDF(21508),
AOM_ICDF(22915), AOM_ICDF(24146), AOM_ICDF(25224), AOM_ICDF(26167),
AOM_ICDF(26992), AOM_ICDF(27714), AOM_ICDF(28346), AOM_ICDF(32768)},
{ AOM_ICDF(3840), AOM_ICDF(7230), AOM_ICDF(10223), AOM_ICDF(12865),
AOM_ICDF(15197), AOM_ICDF(17256), AOM_ICDF(19074), AOM_ICDF(20679),
AOM_ICDF(22096), AOM_ICDF(23347), AOM_ICDF(24451), AOM_ICDF(25426),
AOM_ICDF(26287), AOM_ICDF(27047), AOM_ICDF(27718), AOM_ICDF(32768)},
{ AOM_ICDF(3584), AOM_ICDF(6776), AOM_ICDF(9619), AOM_ICDF(12151),
AOM_ICDF(14406), AOM_ICDF(16414), AOM_ICDF(18203), AOM_ICDF(19796),
AOM_ICDF(21215), AOM_ICDF(22479), AOM_ICDF(23604), AOM_ICDF(24606),
AOM_ICDF(25499), AOM_ICDF(26294), AOM_ICDF(27002), AOM_ICDF(32768)},
{ AOM_ICDF(3328), AOM_ICDF(6318), AOM_ICDF(9004), AOM_ICDF(11417),
AOM_ICDF(13585), AOM_ICDF(15533), AOM_ICDF(17283), AOM_ICDF(18856),
AOM_ICDF(20269), AOM_ICDF(21538), AOM_ICDF(22678), AOM_ICDF(23703),
AOM_ICDF(24624), AOM_ICDF(25451), AOM_ICDF(26194), AOM_ICDF(32768)},
{ AOM_ICDF(3072), AOM_ICDF(5856), AOM_ICDF(8379), AOM_ICDF(10665),
AOM_ICDF(12737), AOM_ICDF(14615), AOM_ICDF(16317), AOM_ICDF(17859),
AOM_ICDF(19257), AOM_ICDF(20524), AOM_ICDF(21672), AOM_ICDF(22712),
AOM_ICDF(23655), AOM_ICDF(24509), AOM_ICDF(25283), AOM_ICDF(32768)},
{ AOM_ICDF(2816), AOM_ICDF(5390), AOM_ICDF(7743), AOM_ICDF(9894),
AOM_ICDF(11860), AOM_ICDF(13657), AOM_ICDF(15299), AOM_ICDF(16800),
AOM_ICDF(18172), AOM_ICDF(19426), AOM_ICDF(20573), AOM_ICDF(21621),
AOM_ICDF(22579), AOM_ICDF(23455), AOM_ICDF(24255), AOM_ICDF(32768)},
{ AOM_ICDF(2560), AOM_ICDF(4920), AOM_ICDF(7096), AOM_ICDF(9102),
AOM_ICDF(10951), AOM_ICDF(12656), AOM_ICDF(14227), AOM_ICDF(15676),
AOM_ICDF(17011), AOM_ICDF(18242), AOM_ICDF(19377), AOM_ICDF(20423),
AOM_ICDF(21388), AOM_ICDF(22277), AOM_ICDF(23097), AOM_ICDF(32768)},
{ AOM_ICDF(2304), AOM_ICDF(4446), AOM_ICDF(6437), AOM_ICDF(8288),
AOM_ICDF(10009), AOM_ICDF(11609), AOM_ICDF(13097), AOM_ICDF(14480),
AOM_ICDF(15766), AOM_ICDF(16961), AOM_ICDF(18072), AOM_ICDF(19105),
AOM_ICDF(20066), AOM_ICDF(20959), AOM_ICDF(21789), AOM_ICDF(32768)},
{ AOM_ICDF(2048), AOM_ICDF(3968), AOM_ICDF(5768), AOM_ICDF(7456),
AOM_ICDF(9038), AOM_ICDF(10521), AOM_ICDF(11911), AOM_ICDF(13215),
AOM_ICDF(14437), AOM_ICDF(15583), AOM_ICDF(16657), AOM_ICDF(17664),
AOM_ICDF(18608), AOM_ICDF(19493), AOM_ICDF(20323), AOM_ICDF(32768)},
{ AOM_ICDF(1792), AOM_ICDF(3486), AOM_ICDF(5087), AOM_ICDF(6601),
AOM_ICDF(8032), AOM_ICDF(9385), AOM_ICDF(10664), AOM_ICDF(11873),
AOM_ICDF(13016), AOM_ICDF(14096), AOM_ICDF(15117), AOM_ICDF(16082),
AOM_ICDF(16995), AOM_ICDF(17858), AOM_ICDF(18673), AOM_ICDF(32768)},
{ AOM_ICDF(1536), AOM_ICDF(3000), AOM_ICDF(4395), AOM_ICDF(5725),
AOM_ICDF(6993), AOM_ICDF(8201), AOM_ICDF(9353), AOM_ICDF(10451),
AOM_ICDF(11497), AOM_ICDF(12494), AOM_ICDF(13444), AOM_ICDF(14350),
AOM_ICDF(15213), AOM_ICDF(16036), AOM_ICDF(16820), AOM_ICDF(32768)},
{ AOM_ICDF(1280), AOM_ICDF(2510), AOM_ICDF(3692), AOM_ICDF(4828),
AOM_ICDF(5919), AOM_ICDF(6968), AOM_ICDF(7976), AOM_ICDF(8944),
AOM_ICDF(9875), AOM_ICDF(10769), AOM_ICDF(11628), AOM_ICDF(12454),
AOM_ICDF(13248), AOM_ICDF(14011), AOM_ICDF(14744), AOM_ICDF(32768)},
{ AOM_ICDF(1024), AOM_ICDF(2016), AOM_ICDF(2977), AOM_ICDF(3908),
AOM_ICDF(4810), AOM_ICDF(5684), AOM_ICDF(6530), AOM_ICDF(7350),
AOM_ICDF(8144), AOM_ICDF(8913), AOM_ICDF(9658), AOM_ICDF(10380),
AOM_ICDF(11080), AOM_ICDF(11758), AOM_ICDF(12415), AOM_ICDF(32768)},
{ AOM_ICDF(768), AOM_ICDF(1518), AOM_ICDF(2250), AOM_ICDF(2965),
AOM_ICDF(3663), AOM_ICDF(4345), AOM_ICDF(5011), AOM_ICDF(5662),
AOM_ICDF(6297), AOM_ICDF(6917), AOM_ICDF(7523), AOM_ICDF(8115),
AOM_ICDF(8693), AOM_ICDF(9257), AOM_ICDF(9808), AOM_ICDF(32768)},
{ AOM_ICDF(512), AOM_ICDF(1016), AOM_ICDF(1512), AOM_ICDF(2000),
AOM_ICDF(2481), AOM_ICDF(2954), AOM_ICDF(3420), AOM_ICDF(3879),
AOM_ICDF(4330), AOM_ICDF(4774), AOM_ICDF(5211), AOM_ICDF(5642),
AOM_ICDF(6066), AOM_ICDF(6483), AOM_ICDF(6894), AOM_ICDF(32768)},
{ AOM_ICDF(256), AOM_ICDF(510), AOM_ICDF(762), AOM_ICDF(1012),
AOM_ICDF(1260), AOM_ICDF(1506), AOM_ICDF(1750), AOM_ICDF(1992),
AOM_ICDF(2232), AOM_ICDF(2471), AOM_ICDF(2708), AOM_ICDF(2943),
AOM_ICDF(3176), AOM_ICDF(3407), AOM_ICDF(3636), AOM_ICDF(32768)},
};
const uint16_t LAPLACE_OFFSET[128] = {
0,
29871,
28672,
27751,
26975,
26291,
25673,
25105,
24576,
24079,
23609,
23162,
22734,
22325,
21931,
21550,
21182,
20826,
20480,
20143,
19815,
19495,
19183,
18877,
18579,
18286,
17999,
17718,
17442,
17170,
16904,
16642,
16384,
16129,
15879,
15633,
15390,
15150,
14913,
14680,
14450,
14222,
13997,
13775,
13556,
13338,
13124,
12911,
12701,
12493,
12288,
12084,
11882,
11682,
11484,
11288,
11094,
10901,
10710,
10521,
10333,
10147,
9962,
9779,
9597,
9417,
9238,
9060,
8884,
8709,
8535,
8363,
8192,
8021,
7853,
7685,
7518,
7352,
7188,
7025,
6862,
6701,
6540,
6381,
6222,
6065,
5908,
5753,
5598,
5444,
5291,
5138,
4987,
4837,
4687,
4538,
4390,
4242,
4096,
3950,
3804,
3660,
3516,
3373,
3231,
3089,
2948,
2808,
2668,
2529,
2391,
2253,
2116,
1979,
1843,
1708,
1573,
1439,
1306,
1172,
1040,
908,
777,
646,
516,
386,
257,
128,
};

97
third_party/aom/av1/common/mips/dspr2/av1_itrans16_dspr2.c vendored Normal file
View file

@ -0,0 +1,97 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <assert.h>
#include <stdio.h>
#include "./aom_config.h"
#include "./av1_rtcd.h"
#include "av1/common/common.h"
#include "av1/common/blockd.h"
#include "av1/common/idct.h"
#include "aom_dsp/mips/inv_txfm_dspr2.h"
#include "aom_dsp/txfm_common.h"
#include "aom_ports/mem.h"
#if HAVE_DSPR2
void av1_iht16x16_256_add_dspr2(const int16_t *input, uint8_t *dest, int pitch,
int tx_type) {
int i, j;
DECLARE_ALIGNED(32, int16_t, out[16 * 16]);
int16_t *outptr = out;
int16_t temp_out[16];
uint32_t pos = 45;
/* bit positon for extract from acc */
__asm__ __volatile__("wrdsp %[pos], 1 \n\t" : : [pos] "r"(pos));
switch (tx_type) {
case DCT_DCT: // DCT in both horizontal and vertical
idct16_rows_dspr2(input, outptr, 16);
idct16_cols_add_blk_dspr2(out, dest, pitch);
break;
case ADST_DCT: // ADST in vertical, DCT in horizontal
idct16_rows_dspr2(input, outptr, 16);
outptr = out;
for (i = 0; i < 16; ++i) {
iadst16_dspr2(outptr, temp_out);
for (j = 0; j < 16; ++j)
dest[j * pitch + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6) +
dest[j * pitch + i]);
outptr += 16;
}
break;
case DCT_ADST: // DCT in vertical, ADST in horizontal
{
int16_t temp_in[16 * 16];
for (i = 0; i < 16; ++i) {
/* prefetch row */
prefetch_load((const uint8_t *)(input + 16));
iadst16_dspr2(input, outptr);
input += 16;
outptr += 16;
}
for (i = 0; i < 16; ++i)
for (j = 0; j < 16; ++j) temp_in[j * 16 + i] = out[i * 16 + j];
idct16_cols_add_blk_dspr2(temp_in, dest, pitch);
} break;
case ADST_ADST: // ADST in both directions
{
int16_t temp_in[16];
for (i = 0; i < 16; ++i) {
/* prefetch row */
prefetch_load((const uint8_t *)(input + 16));
iadst16_dspr2(input, outptr);
input += 16;
outptr += 16;
}
for (i = 0; i < 16; ++i) {
for (j = 0; j < 16; ++j) temp_in[j] = out[j * 16 + i];
iadst16_dspr2(temp_in, temp_out);
for (j = 0; j < 16; ++j)
dest[j * pitch + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6) +
dest[j * pitch + i]);
}
} break;
default: printf("av1_short_iht16x16_add_dspr2 : Invalid tx_type\n"); break;
}
}
#endif // #if HAVE_DSPR2

91
third_party/aom/av1/common/mips/dspr2/av1_itrans4_dspr2.c vendored Normal file
View file

@ -0,0 +1,91 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <assert.h>
#include <stdio.h>
#include "./aom_config.h"
#include "./av1_rtcd.h"
#include "av1/common/common.h"
#include "av1/common/blockd.h"
#include "av1/common/idct.h"
#include "aom_dsp/mips/inv_txfm_dspr2.h"
#include "aom_dsp/txfm_common.h"
#include "aom_ports/mem.h"
#if HAVE_DSPR2
void av1_iht4x4_16_add_dspr2(const int16_t *input, uint8_t *dest,
int dest_stride, int tx_type) {
int i, j;
DECLARE_ALIGNED(32, int16_t, out[4 * 4]);
int16_t *outptr = out;
int16_t temp_in[4 * 4], temp_out[4];
uint32_t pos = 45;
/* bit positon for extract from acc */
__asm__ __volatile__("wrdsp %[pos], 1 \n\t"
:
: [pos] "r"(pos));
switch (tx_type) {
case DCT_DCT: // DCT in both horizontal and vertical
aom_idct4_rows_dspr2(input, outptr);
aom_idct4_columns_add_blk_dspr2(&out[0], dest, dest_stride);
break;
case ADST_DCT: // ADST in vertical, DCT in horizontal
aom_idct4_rows_dspr2(input, outptr);
outptr = out;
for (i = 0; i < 4; ++i) {
iadst4_dspr2(outptr, temp_out);
for (j = 0; j < 4; ++j)
dest[j * dest_stride + i] = clip_pixel(
ROUND_POWER_OF_TWO(temp_out[j], 4) + dest[j * dest_stride + i]);
outptr += 4;
}
break;
case DCT_ADST: // DCT in vertical, ADST in horizontal
for (i = 0; i < 4; ++i) {
iadst4_dspr2(input, outptr);
input += 4;
outptr += 4;
}
for (i = 0; i < 4; ++i) {
for (j = 0; j < 4; ++j) {
temp_in[i * 4 + j] = out[j * 4 + i];
}
}
aom_idct4_columns_add_blk_dspr2(&temp_in[0], dest, dest_stride);
break;
case ADST_ADST: // ADST in both directions
for (i = 0; i < 4; ++i) {
iadst4_dspr2(input, outptr);
input += 4;
outptr += 4;
}
for (i = 0; i < 4; ++i) {
for (j = 0; j < 4; ++j) temp_in[j] = out[j * 4 + i];
iadst4_dspr2(temp_in, temp_out);
for (j = 0; j < 4; ++j)
dest[j * dest_stride + i] = clip_pixel(
ROUND_POWER_OF_TWO(temp_out[j], 4) + dest[j * dest_stride + i]);
}
break;
default: printf("av1_short_iht4x4_add_dspr2 : Invalid tx_type\n"); break;
}
}
#endif // #if HAVE_DSPR2

85
third_party/aom/av1/common/mips/dspr2/av1_itrans8_dspr2.c vendored Normal file
View file

@ -0,0 +1,85 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <assert.h>
#include <stdio.h>
#include "./aom_config.h"
#include "./av1_rtcd.h"
#include "av1/common/common.h"
#include "av1/common/blockd.h"
#include "aom_dsp/mips/inv_txfm_dspr2.h"
#include "aom_dsp/txfm_common.h"
#include "aom_ports/mem.h"
#if HAVE_DSPR2
void av1_iht8x8_64_add_dspr2(const int16_t *input, uint8_t *dest,
int dest_stride, int tx_type) {
int i, j;
DECLARE_ALIGNED(32, int16_t, out[8 * 8]);
int16_t *outptr = out;
int16_t temp_in[8 * 8], temp_out[8];
uint32_t pos = 45;
/* bit positon for extract from acc */
__asm__ __volatile__("wrdsp %[pos], 1 \n\t" : : [pos] "r"(pos));
switch (tx_type) {
case DCT_DCT: // DCT in both horizontal and vertical
idct8_rows_dspr2(input, outptr, 8);
idct8_columns_add_blk_dspr2(&out[0], dest, dest_stride);
break;
case ADST_DCT: // ADST in vertical, DCT in horizontal
idct8_rows_dspr2(input, outptr, 8);
for (i = 0; i < 8; ++i) {
iadst8_dspr2(&out[i * 8], temp_out);
for (j = 0; j < 8; ++j)
dest[j * dest_stride + i] = clip_pixel(
ROUND_POWER_OF_TWO(temp_out[j], 5) + dest[j * dest_stride + i]);
}
break;
case DCT_ADST: // DCT in vertical, ADST in horizontal
for (i = 0; i < 8; ++i) {
iadst8_dspr2(input, outptr);
input += 8;
outptr += 8;
}
for (i = 0; i < 8; ++i) {
for (j = 0; j < 8; ++j) {
temp_in[i * 8 + j] = out[j * 8 + i];
}
}
idct8_columns_add_blk_dspr2(&temp_in[0], dest, dest_stride);
break;
case ADST_ADST: // ADST in both directions
for (i = 0; i < 8; ++i) {
iadst8_dspr2(input, outptr);
input += 8;
outptr += 8;
}
for (i = 0; i < 8; ++i) {
for (j = 0; j < 8; ++j) temp_in[j] = out[j * 8 + i];
iadst8_dspr2(temp_in, temp_out);
for (j = 0; j < 8; ++j)
dest[j * dest_stride + i] = clip_pixel(
ROUND_POWER_OF_TWO(temp_out[j], 5) + dest[j * dest_stride + i]);
}
break;
default: printf("av1_short_iht8x8_add_dspr2 : Invalid tx_type\n"); break;
}
}
#endif // #if HAVE_DSPR2

80
third_party/aom/av1/common/mips/msa/av1_idct16x16_msa.c vendored Normal file
View file

@ -0,0 +1,80 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <assert.h>
#include "av1/common/enums.h"
#include "aom_dsp/mips/inv_txfm_msa.h"
void av1_iht16x16_256_add_msa(const int16_t *input, uint8_t *dst,
int32_t dst_stride, int32_t tx_type) {
int32_t i;
DECLARE_ALIGNED(32, int16_t, out[16 * 16]);
int16_t *out_ptr = &out[0];
switch (tx_type) {
case DCT_DCT:
/* transform rows */
for (i = 0; i < 2; ++i) {
/* process 16 * 8 block */
aom_idct16_1d_rows_msa((input + (i << 7)), (out_ptr + (i << 7)));
}
/* transform columns */
for (i = 0; i < 2; ++i) {
/* process 8 * 16 block */
aom_idct16_1d_columns_addblk_msa((out_ptr + (i << 3)), (dst + (i << 3)),
dst_stride);
}
break;
case ADST_DCT:
/* transform rows */
for (i = 0; i < 2; ++i) {
/* process 16 * 8 block */
aom_idct16_1d_rows_msa((input + (i << 7)), (out_ptr + (i << 7)));
}
/* transform columns */
for (i = 0; i < 2; ++i) {
aom_iadst16_1d_columns_addblk_msa((out_ptr + (i << 3)),
(dst + (i << 3)), dst_stride);
}
break;
case DCT_ADST:
/* transform rows */
for (i = 0; i < 2; ++i) {
/* process 16 * 8 block */
aom_iadst16_1d_rows_msa((input + (i << 7)), (out_ptr + (i << 7)));
}
/* transform columns */
for (i = 0; i < 2; ++i) {
/* process 8 * 16 block */
aom_idct16_1d_columns_addblk_msa((out_ptr + (i << 3)), (dst + (i << 3)),
dst_stride);
}
break;
case ADST_ADST:
/* transform rows */
for (i = 0; i < 2; ++i) {
/* process 16 * 8 block */
aom_iadst16_1d_rows_msa((input + (i << 7)), (out_ptr + (i << 7)));
}
/* transform columns */
for (i = 0; i < 2; ++i) {
aom_iadst16_1d_columns_addblk_msa((out_ptr + (i << 3)),
(dst + (i << 3)), dst_stride);
}
break;
default: assert(0); break;
}
}

61
third_party/aom/av1/common/mips/msa/av1_idct4x4_msa.c vendored Normal file
View file

@ -0,0 +1,61 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <assert.h>
#include "av1/common/enums.h"
#include "aom_dsp/mips/inv_txfm_msa.h"
void av1_iht4x4_16_add_msa(const int16_t *input, uint8_t *dst,
int32_t dst_stride, int32_t tx_type) {
v8i16 in0, in1, in2, in3;
/* load vector elements of 4x4 block */
LD4x4_SH(input, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
switch (tx_type) {
case DCT_DCT:
/* DCT in horizontal */
AOM_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3);
/* DCT in vertical */
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
AOM_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case ADST_DCT:
/* DCT in horizontal */
AOM_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3);
/* ADST in vertical */
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
AOM_IADST4x4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case DCT_ADST:
/* ADST in horizontal */
AOM_IADST4x4(in0, in1, in2, in3, in0, in1, in2, in3);
/* DCT in vertical */
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
AOM_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case ADST_ADST:
/* ADST in horizontal */
AOM_IADST4x4(in0, in1, in2, in3, in0, in1, in2, in3);
/* ADST in vertical */
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
AOM_IADST4x4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
default: assert(0); break;
}
/* final rounding (add 2^3, divide by 2^4) and shift */
SRARI_H4_SH(in0, in1, in2, in3, 4);
/* add block and store 4x4 */
ADDBLK_ST4x4_UB(in0, in1, in2, in3, dst, dst_stride);
}

79
third_party/aom/av1/common/mips/msa/av1_idct8x8_msa.c vendored Normal file
View file

@ -0,0 +1,79 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <assert.h>
#include "av1/common/enums.h"
#include "aom_dsp/mips/inv_txfm_msa.h"
void av1_iht8x8_64_add_msa(const int16_t *input, uint8_t *dst,
int32_t dst_stride, int32_t tx_type) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
/* load vector elements of 8x8 block */
LD_SH8(input, 8, in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
in4, in5, in6, in7);
switch (tx_type) {
case DCT_DCT:
/* DCT in horizontal */
AOM_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
in4, in5, in6, in7);
/* DCT in vertical */
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2,
in3, in4, in5, in6, in7);
AOM_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
in4, in5, in6, in7);
break;
case ADST_DCT:
/* DCT in horizontal */
AOM_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
in4, in5, in6, in7);
/* ADST in vertical */
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2,
in3, in4, in5, in6, in7);
AOM_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
in5, in6, in7);
break;
case DCT_ADST:
/* ADST in horizontal */
AOM_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
in5, in6, in7);
/* DCT in vertical */
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2,
in3, in4, in5, in6, in7);
AOM_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
in4, in5, in6, in7);
break;
case ADST_ADST:
/* ADST in horizontal */
AOM_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
in5, in6, in7);
/* ADST in vertical */
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2,
in3, in4, in5, in6, in7);
AOM_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
in5, in6, in7);
break;
default: assert(0); break;
}
/* final rounding (add 2^4, divide by 2^5) and shift */
SRARI_H4_SH(in0, in1, in2, in3, 5);
SRARI_H4_SH(in4, in5, in6, in7, 5);
/* add block and store 8x8 */
AOM_ADDBLK_ST8x4_UB(dst, dst_stride, in0, in1, in2, in3);
dst += (4 * dst_stride);
AOM_ADDBLK_ST8x4_UB(dst, dst_stride, in4, in5, in6, in7);
}

302
third_party/aom/av1/common/mv.h vendored Normal file
View file

@ -0,0 +1,302 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_MV_H_
#define AV1_COMMON_MV_H_
#include "av1/common/common.h"
#include "av1/common/common_data.h"
#include "aom_dsp/aom_filter.h"
#ifdef __cplusplus
extern "C" {
#endif
typedef struct mv {
int16_t row;
int16_t col;
} MV;
typedef union int_mv {
uint32_t as_int;
MV as_mv;
} int_mv; /* facilitates faster equality tests and copies */
typedef struct mv32 {
int32_t row;
int32_t col;
} MV32;
#if (CONFIG_WARPED_MOTION || CONFIG_MOTION_VAR) && CONFIG_GLOBAL_MOTION
#define SEPARATE_GLOBAL_MOTION 1
#endif // (CONFIG_WARPED_MOTION || CONFIG_MOTION_VAR) && CONFIG_GLOBAL_MOTION
#if CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
// Bits of precision used for the model
#define WARPEDMODEL_PREC_BITS 16
#define WARPEDMODEL_ROW3HOMO_PREC_BITS 16
#define WARPEDMODEL_TRANS_CLAMP (128 << WARPEDMODEL_PREC_BITS)
#define WARPEDMODEL_DIAGAFFINE_CLAMP (1 << (WARPEDMODEL_PREC_BITS + 1))
#define WARPEDMODEL_NONDIAGAFFINE_CLAMP (1 << (WARPEDMODEL_PREC_BITS - 1))
#define WARPEDMODEL_ROW3HOMO_CLAMP (1 << (WARPEDMODEL_PREC_BITS - 1))
// Bits of subpel precision for warped interpolation
#define WARPEDPIXEL_PREC_BITS 6
#define WARPEDPIXEL_PREC_SHIFTS (1 << WARPEDPIXEL_PREC_BITS)
// Taps for ntap filter
#define WARPEDPIXEL_FILTER_TAPS 6
// Precision of filter taps
#define WARPEDPIXEL_FILTER_BITS 7
// Precision bits reduction after horizontal shear
#define HORSHEAR_REDUCE_PREC_BITS 5
#define VERSHEAR_REDUCE_PREC_BITS \
(2 * WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS)
#define WARPEDDIFF_PREC_BITS (WARPEDMODEL_PREC_BITS - WARPEDPIXEL_PREC_BITS)
/* clang-format off */
typedef enum {
IDENTITY = 0, // identity transformation, 0-parameter
TRANSLATION = 1, // translational motion 2-parameter
ROTZOOM = 2, // simplified affine with rotation + zoom only, 4-parameter
AFFINE = 3, // affine, 6-parameter
HORTRAPEZOID = 4, // constrained homography, hor trapezoid, 6-parameter
VERTRAPEZOID = 5, // constrained homography, ver trapezoid, 6-parameter
HOMOGRAPHY = 6, // homography, 8-parameter
TRANS_TYPES = 7,
} TransformationType;
/* clang-format on */
// Number of types used for global motion (must be >= 3 and <= TRANS_TYPES)
// The following can be useful:
// GLOBAL_TRANS_TYPES 3 - up to rotation-zoom
// GLOBAL_TRANS_TYPES 4 - up to affine
// GLOBAL_TRANS_TYPES 6 - up to hor/ver trapezoids
// GLOBAL_TRANS_TYPES 7 - up to full homography
#define GLOBAL_TRANS_TYPES 4
typedef struct {
#if CONFIG_GLOBAL_MOTION
int global_warp_allowed;
#endif // CONFIG_GLOBAL_MOTION
#if CONFIG_WARPED_MOTION
int local_warp_allowed;
#endif // CONFIG_WARPED_MOTION
} WarpTypesAllowed;
// number of parameters used by each transformation in TransformationTypes
static const int trans_model_params[TRANS_TYPES] = { 0, 2, 4, 6, 6, 6, 8 };
// The order of values in the wmmat matrix below is best described
// by the homography:
// [x' (m2 m3 m0 [x
// z . y' = m4 m5 m1 * y
// 1] m6 m7 1) 1]
typedef struct {
TransformationType wmtype;
int32_t wmmat[8];
int16_t alpha, beta, gamma, delta;
} WarpedMotionParams;
static INLINE void set_default_warp_params(WarpedMotionParams *wm) {
static const int32_t default_wm_mat[8] = {
0, 0, (1 << WARPEDMODEL_PREC_BITS), 0, 0, (1 << WARPEDMODEL_PREC_BITS), 0, 0
};
memset(wm, 0, sizeof(*wm));
memcpy(wm->wmmat, default_wm_mat, sizeof(wm->wmmat));
wm->wmtype = IDENTITY;
}
#endif // CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
#if CONFIG_GLOBAL_MOTION
// The following constants describe the various precisions
// of different parameters in the global motion experiment.
//
// Given the general homography:
// [x' (a b c [x
// z . y' = d e f * y
// 1] g h i) 1]
//
// Constants using the name ALPHA here are related to parameters
// a, b, d, e. Constants using the name TRANS are related
// to parameters c and f.
//
// Anything ending in PREC_BITS is the number of bits of precision
// to maintain when converting from double to integer.
//
// The ABS parameters are used to create an upper and lower bound
// for each parameter. In other words, after a parameter is integerized
// it is clamped between -(1 << ABS_XXX_BITS) and (1 << ABS_XXX_BITS).
//
// XXX_PREC_DIFF and XXX_DECODE_FACTOR
// are computed once here to prevent repetitive
// computation on the decoder side. These are
// to allow the global motion parameters to be encoded in a lower
// precision than the warped model precision. This means that they
// need to be changed to warped precision when they are decoded.
//
// XX_MIN, XX_MAX are also computed to avoid repeated computation
#define SUBEXPFIN_K 3
#define GM_TRANS_PREC_BITS 6
#define GM_ABS_TRANS_BITS 12
#define GM_ABS_TRANS_ONLY_BITS (GM_ABS_TRANS_BITS - GM_TRANS_PREC_BITS + 3)
#define GM_TRANS_PREC_DIFF (WARPEDMODEL_PREC_BITS - GM_TRANS_PREC_BITS)
#define GM_TRANS_ONLY_PREC_DIFF (WARPEDMODEL_PREC_BITS - 3)
#define GM_TRANS_DECODE_FACTOR (1 << GM_TRANS_PREC_DIFF)
#define GM_TRANS_ONLY_DECODE_FACTOR (1 << GM_TRANS_ONLY_PREC_DIFF)
#define GM_ALPHA_PREC_BITS 15
#define GM_ABS_ALPHA_BITS 12
#define GM_ALPHA_PREC_DIFF (WARPEDMODEL_PREC_BITS - GM_ALPHA_PREC_BITS)
#define GM_ALPHA_DECODE_FACTOR (1 << GM_ALPHA_PREC_DIFF)
#define GM_ROW3HOMO_PREC_BITS 16
#define GM_ABS_ROW3HOMO_BITS 11
#define GM_ROW3HOMO_PREC_DIFF \
(WARPEDMODEL_ROW3HOMO_PREC_BITS - GM_ROW3HOMO_PREC_BITS)
#define GM_ROW3HOMO_DECODE_FACTOR (1 << GM_ROW3HOMO_PREC_DIFF)
#define GM_TRANS_MAX (1 << GM_ABS_TRANS_BITS)
#define GM_ALPHA_MAX (1 << GM_ABS_ALPHA_BITS)
#define GM_ROW3HOMO_MAX (1 << GM_ABS_ROW3HOMO_BITS)
#define GM_TRANS_MIN -GM_TRANS_MAX
#define GM_ALPHA_MIN -GM_ALPHA_MAX
#define GM_ROW3HOMO_MIN -GM_ROW3HOMO_MAX
// Use global motion parameters for sub8x8 blocks
#define GLOBAL_SUB8X8_USED 0
static INLINE int block_center_x(int mi_col, BLOCK_SIZE bs) {
const int bw = block_size_wide[bs];
return mi_col * MI_SIZE + bw / 2 - 1;
}
static INLINE int block_center_y(int mi_row, BLOCK_SIZE bs) {
const int bh = block_size_high[bs];
return mi_row * MI_SIZE + bh / 2 - 1;
}
static INLINE int convert_to_trans_prec(int allow_hp, int coor) {
if (allow_hp)
return ROUND_POWER_OF_TWO_SIGNED(coor, WARPEDMODEL_PREC_BITS - 3);
else
return ROUND_POWER_OF_TWO_SIGNED(coor, WARPEDMODEL_PREC_BITS - 2) * 2;
}
// Convert a global motion translation vector (which may have more bits than a
// regular motion vector) into a motion vector
static INLINE int_mv gm_get_motion_vector(const WarpedMotionParams *gm,
int allow_hp, BLOCK_SIZE bsize,
int mi_col, int mi_row,
int block_idx) {
const int unify_bsize = CONFIG_CB4X4;
int_mv res;
const int32_t *mat = gm->wmmat;
int x, y, tx, ty;
if (gm->wmtype == TRANSLATION) {
res.as_mv.row = gm->wmmat[0] >> GM_TRANS_ONLY_PREC_DIFF;
res.as_mv.col = gm->wmmat[1] >> GM_TRANS_ONLY_PREC_DIFF;
return res;
}
if (bsize >= BLOCK_8X8 || unify_bsize) {
x = block_center_x(mi_col, bsize);
y = block_center_y(mi_row, bsize);
} else {
x = block_center_x(mi_col, bsize);
y = block_center_y(mi_row, bsize);
x += (block_idx & 1) * MI_SIZE / 2;
y += (block_idx & 2) * MI_SIZE / 4;
}
if (gm->wmtype == ROTZOOM) {
assert(gm->wmmat[5] == gm->wmmat[2]);
assert(gm->wmmat[4] == -gm->wmmat[3]);
}
if (gm->wmtype > AFFINE) {
int xc = (int)((int64_t)mat[2] * x + (int64_t)mat[3] * y + mat[0]);
int yc = (int)((int64_t)mat[4] * x + (int64_t)mat[5] * y + mat[1]);
const int Z = (int)((int64_t)mat[6] * x + (int64_t)mat[7] * y +
(1 << WARPEDMODEL_ROW3HOMO_PREC_BITS));
xc *= 1 << (WARPEDMODEL_ROW3HOMO_PREC_BITS - WARPEDMODEL_PREC_BITS);
yc *= 1 << (WARPEDMODEL_ROW3HOMO_PREC_BITS - WARPEDMODEL_PREC_BITS);
xc = (int)(xc > 0 ? ((int64_t)xc + Z / 2) / Z : ((int64_t)xc - Z / 2) / Z);
yc = (int)(yc > 0 ? ((int64_t)yc + Z / 2) / Z : ((int64_t)yc - Z / 2) / Z);
tx = convert_to_trans_prec(allow_hp, xc) - (x << 3);
ty = convert_to_trans_prec(allow_hp, yc) - (y << 3);
} else {
const int xc =
(mat[2] - (1 << WARPEDMODEL_PREC_BITS)) * x + mat[3] * y + mat[0];
const int yc =
mat[4] * x + (mat[5] - (1 << WARPEDMODEL_PREC_BITS)) * y + mat[1];
tx = convert_to_trans_prec(allow_hp, xc);
ty = convert_to_trans_prec(allow_hp, yc);
}
res.as_mv.row = ty;
res.as_mv.col = tx;
return res;
}
static INLINE TransformationType get_gmtype(const WarpedMotionParams *gm) {
if (gm->wmmat[6] != 0 || gm->wmmat[7] != 0) {
if (!gm->wmmat[6] && !gm->wmmat[4]) return HORTRAPEZOID;
if (!gm->wmmat[7] && !gm->wmmat[3]) return VERTRAPEZOID;
return HOMOGRAPHY;
}
if (gm->wmmat[5] == (1 << WARPEDMODEL_PREC_BITS) && !gm->wmmat[4] &&
gm->wmmat[2] == (1 << WARPEDMODEL_PREC_BITS) && !gm->wmmat[3]) {
return ((!gm->wmmat[1] && !gm->wmmat[0]) ? IDENTITY : TRANSLATION);
}
if (gm->wmmat[2] == gm->wmmat[5] && gm->wmmat[3] == -gm->wmmat[4])
return ROTZOOM;
else
return AFFINE;
}
#endif // CONFIG_GLOBAL_MOTION
#if CONFIG_REF_MV
typedef struct candidate_mv {
int_mv this_mv;
int_mv comp_mv;
uint8_t pred_diff[2];
int weight;
} CANDIDATE_MV;
#endif
static INLINE int is_zero_mv(const MV *mv) {
return *((const uint32_t *)mv) == 0;
}
static INLINE int is_equal_mv(const MV *a, const MV *b) {
return *((const uint32_t *)a) == *((const uint32_t *)b);
}
static INLINE void clamp_mv(MV *mv, int min_col, int max_col, int min_row,
int max_row) {
mv->col = clamp(mv->col, min_col, max_col);
mv->row = clamp(mv->row, min_row, max_row);
}
static INLINE int mv_has_subpel(const MV *mv) {
return (mv->row & SUBPEL_MASK) || (mv->col & SUBPEL_MASK);
}
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_MV_H_

1164
third_party/aom/av1/common/mvref_common.c vendored Normal file
View file

File diff suppressed because it is too large Load diff

580
third_party/aom/av1/common/mvref_common.h vendored Normal file
View file

@ -0,0 +1,580 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_MVREF_COMMON_H_
#define AV1_COMMON_MVREF_COMMON_H_
#include "av1/common/onyxc_int.h"
#include "av1/common/blockd.h"
#ifdef __cplusplus
extern "C" {
#endif
#if CONFIG_REF_MV
#define MVREF_NEIGHBOURS 9
#else
#define MVREF_NEIGHBOURS 8
#endif
typedef struct position {
int row;
int col;
} POSITION;
typedef enum {
BOTH_ZERO = 0,
ZERO_PLUS_PREDICTED = 1,
BOTH_PREDICTED = 2,
NEW_PLUS_NON_INTRA = 3,
BOTH_NEW = 4,
INTRA_PLUS_NON_INTRA = 5,
BOTH_INTRA = 6,
INVALID_CASE = 9
} motion_vector_context;
// This is used to figure out a context for the ref blocks. The code flattens
// an array that would have 3 possible counts (0, 1 & 2) for 3 choices by
// adding 9 for each intra block, 3 for each zero mv and 1 for each new
// motion vector. This single number is then converted into a context
// with a single lookup ( counter_to_context ).
static const int mode_2_counter[MB_MODE_COUNT] = {
9, // DC_PRED
9, // V_PRED
9, // H_PRED
9, // D45_PRED
9, // D135_PRED
9, // D117_PRED
9, // D153_PRED
9, // D207_PRED
9, // D63_PRED
#if CONFIG_ALT_INTRA
9, // SMOOTH_PRED
#endif // CONFIG_ALT_INTRA
9, // TM_PRED
0, // NEARESTMV
0, // NEARMV
3, // ZEROMV
1, // NEWMV
#if CONFIG_EXT_INTER
#if CONFIG_COMPOUND_SINGLEREF
0, // SR_NEAREST_NEARMV
1, // SR_NEAREST_NEWMV
1, // SR_NEAR_NEWMV
3, // SR_ZERO_NEWMV
1, // SR_NEW_NEWMV
#endif // CONFIG_COMPOUND_SINGLEREF
0, // NEAREST_NEARESTMV
0, // NEAREST_NEARMV
0, // NEAR_NEARESTMV
0, // NEAR_NEARMV
1, // NEAREST_NEWMV
1, // NEW_NEARESTMV
1, // NEAR_NEWMV
1, // NEW_NEARMV
3, // ZERO_ZEROMV
1, // NEW_NEWMV
#endif // CONFIG_EXT_INTER
};
// There are 3^3 different combinations of 3 counts that can be either 0,1 or
// 2. However the actual count can never be greater than 2 so the highest
// counter we need is 18. 9 is an invalid counter that's never used.
static const int counter_to_context[19] = {
BOTH_PREDICTED, // 0
NEW_PLUS_NON_INTRA, // 1
BOTH_NEW, // 2
ZERO_PLUS_PREDICTED, // 3
NEW_PLUS_NON_INTRA, // 4
INVALID_CASE, // 5
BOTH_ZERO, // 6
INVALID_CASE, // 7
INVALID_CASE, // 8
INTRA_PLUS_NON_INTRA, // 9
INTRA_PLUS_NON_INTRA, // 10
INVALID_CASE, // 11
INTRA_PLUS_NON_INTRA, // 12
INVALID_CASE, // 13
INVALID_CASE, // 14
INVALID_CASE, // 15
INVALID_CASE, // 16
INVALID_CASE, // 17
BOTH_INTRA // 18
};
#if !CONFIG_REF_MV
static const POSITION mv_ref_blocks[BLOCK_SIZES][MVREF_NEIGHBOURS] = {
// 4X4
{ { -1, 0 },
{ 0, -1 },
{ -1, -1 },
{ -2, 0 },
{ 0, -2 },
{ -2, -1 },
{ -1, -2 },
{ -2, -2 } },
// 4X8
{ { -1, 0 },
{ 0, -1 },
{ -1, -1 },
{ -2, 0 },
{ 0, -2 },
{ -2, -1 },
{ -1, -2 },
{ -2, -2 } },
// 8X4
{ { -1, 0 },
{ 0, -1 },
{ -1, -1 },
{ -2, 0 },
{ 0, -2 },
{ -2, -1 },
{ -1, -2 },
{ -2, -2 } },
// 8X8
{ { -1, 0 },
{ 0, -1 },
{ -1, -1 },
{ -2, 0 },
{ 0, -2 },
{ -2, -1 },
{ -1, -2 },
{ -2, -2 } },
// 8X16
{ { 0, -1 },
{ -1, 0 },
{ 1, -1 },
{ -1, -1 },
{ 0, -2 },
{ -2, 0 },
{ -2, -1 },
{ -1, -2 } },
// 16X8
{ { -1, 0 },
{ 0, -1 },
{ -1, 1 },
{ -1, -1 },
{ -2, 0 },
{ 0, -2 },
{ -1, -2 },
{ -2, -1 } },
// 16X16
{ { -1, 0 },
{ 0, -1 },
{ -1, 1 },
{ 1, -1 },
{ -1, -1 },
{ -3, 0 },
{ 0, -3 },
{ -3, -3 } },
// 16X32
{ { 0, -1 },
{ -1, 0 },
{ 2, -1 },
{ -1, -1 },
{ -1, 1 },
{ 0, -3 },
{ -3, 0 },
{ -3, -3 } },
// 32X16
{ { -1, 0 },
{ 0, -1 },
{ -1, 2 },
{ -1, -1 },
{ 1, -1 },
{ -3, 0 },
{ 0, -3 },
{ -3, -3 } },
// 32X32
{ { -1, 1 },
{ 1, -1 },
{ -1, 2 },
{ 2, -1 },
{ -1, -1 },
{ -3, 0 },
{ 0, -3 },
{ -3, -3 } },
// 32X64
{ { 0, -1 },
{ -1, 0 },
{ 4, -1 },
{ -1, 2 },
{ -1, -1 },
{ 0, -3 },
{ -3, 0 },
{ 2, -1 } },
// 64X32
{ { -1, 0 },
{ 0, -1 },
{ -1, 4 },
{ 2, -1 },
{ -1, -1 },
{ -3, 0 },
{ 0, -3 },
{ -1, 2 } },
// 64X64
{ { -1, 3 },
{ 3, -1 },
{ -1, 4 },
{ 4, -1 },
{ -1, -1 },
{ -1, 0 },
{ 0, -1 },
{ -1, 6 } },
#if CONFIG_EXT_PARTITION
// TODO(debargha/jingning) Making them twice the 32x64, .. ones above
// 64x128
{ { 0, -2 },
{ -2, 0 },
{ 8, -2 },
{ -2, 4 },
{ -2, -2 },
{ 0, -6 },
{ -6, 0 },
{ 4, -2 } },
// 128x64
{ { -2, 0 },
{ 0, -2 },
{ -2, 8 },
{ 4, -2 },
{ -2, -2 },
{ -6, 0 },
{ 0, -6 },
{ -2, 4 } },
// 128x128
{ { -2, 6 },
{ 6, -2 },
{ -2, 8 },
{ 8, -2 },
{ -2, -2 },
{ -2, 0 },
{ 0, -2 },
{ -2, 12 } },
#endif // CONFIG_EXT_PARTITION
};
#endif
static const int idx_n_column_to_subblock[4][2] = {
{ 1, 2 }, { 1, 3 }, { 3, 2 }, { 3, 3 }
};
// clamp_mv_ref
#if CONFIG_EXT_PARTITION
#define MV_BORDER (16 << 3) // Allow 16 pels in 1/8th pel units
#else
#define MV_BORDER (8 << 3) // Allow 8 pels in 1/8th pel units
#endif // CONFIG_EXT_PARTITION
static INLINE void clamp_mv_ref(MV *mv, int bw, int bh, const MACROBLOCKD *xd) {
clamp_mv(mv, xd->mb_to_left_edge - bw * 8 - MV_BORDER,
xd->mb_to_right_edge + bw * 8 + MV_BORDER,
xd->mb_to_top_edge - bh * 8 - MV_BORDER,
xd->mb_to_bottom_edge + bh * 8 + MV_BORDER);
}
// This function returns either the appropriate sub block or block's mv
// on whether the block_size < 8x8 and we have check_sub_blocks set.
static INLINE int_mv get_sub_block_mv(const MODE_INFO *candidate, int which_mv,
int search_col, int block_idx) {
#if CONFIG_REF_MV
(void)search_col;
(void)block_idx;
return candidate->mbmi.mv[which_mv];
#else
return block_idx >= 0 && candidate->mbmi.sb_type < BLOCK_8X8
? candidate
->bmi[idx_n_column_to_subblock[block_idx][search_col == 0]]
.as_mv[which_mv]
: candidate->mbmi.mv[which_mv];
#endif
}
#if CONFIG_REF_MV
static INLINE int_mv get_sub_block_pred_mv(const MODE_INFO *candidate,
int which_mv, int search_col,
int block_idx) {
(void)search_col;
(void)block_idx;
return candidate->mbmi.mv[which_mv];
}
#endif
// Performs mv sign inversion if indicated by the reference frame combination.
static INLINE int_mv scale_mv(const MB_MODE_INFO *mbmi, int ref,
const MV_REFERENCE_FRAME this_ref_frame,
const int *ref_sign_bias) {
int_mv mv = mbmi->mv[ref];
if (ref_sign_bias[mbmi->ref_frame[ref]] != ref_sign_bias[this_ref_frame]) {
mv.as_mv.row *= -1;
mv.as_mv.col *= -1;
}
return mv;
}
#define CLIP_IN_ADD(mv, bw, bh, xd) clamp_mv_ref(mv, bw, bh, xd)
// This macro is used to add a motion vector mv_ref list if it isn't
// already in the list. If it's the second motion vector it will also
// skip all additional processing and jump to done!
#define ADD_MV_REF_LIST(mv, refmv_count, mv_ref_list, bw, bh, xd, Done) \
do { \
(mv_ref_list)[(refmv_count)] = (mv); \
CLIP_IN_ADD(&(mv_ref_list)[(refmv_count)].as_mv, (bw), (bh), (xd)); \
if (refmv_count && (mv_ref_list)[1].as_int != (mv_ref_list)[0].as_int) { \
(refmv_count) = 2; \
goto Done; \
} \
(refmv_count) = 1; \
} while (0)
// If either reference frame is different, not INTRA, and they
// are different from each other scale and add the mv to our list.
#define IF_DIFF_REF_FRAME_ADD_MV(mbmi, ref_frame, ref_sign_bias, refmv_count, \
mv_ref_list, bw, bh, xd, Done) \
do { \
if (is_inter_block(mbmi)) { \
if ((mbmi)->ref_frame[0] != ref_frame) \
ADD_MV_REF_LIST(scale_mv((mbmi), 0, ref_frame, ref_sign_bias), \
refmv_count, mv_ref_list, bw, bh, xd, Done); \
if (has_second_ref(mbmi) && (mbmi)->ref_frame[1] != ref_frame) \
ADD_MV_REF_LIST(scale_mv((mbmi), 1, ref_frame, ref_sign_bias), \
refmv_count, mv_ref_list, bw, bh, xd, Done); \
} \
} while (0)
// Checks that the given mi_row, mi_col and search point
// are inside the borders of the tile.
static INLINE int is_inside(const TileInfo *const tile, int mi_col, int mi_row,
int mi_rows, const AV1_COMMON *cm,
const POSITION *mi_pos) {
#if CONFIG_DEPENDENT_HORZTILES
const int dependent_horz_tile_flag = cm->dependent_horz_tiles;
#else
const int dependent_horz_tile_flag = 0;
(void)cm;
#endif
#if CONFIG_TILE_GROUPS
if (dependent_horz_tile_flag && !tile->tg_horz_boundary) {
#else
if (dependent_horz_tile_flag) {
#endif
return !(mi_row + mi_pos->row < 0 ||
mi_col + mi_pos->col < tile->mi_col_start ||
mi_row + mi_pos->row >= mi_rows ||
mi_col + mi_pos->col >= tile->mi_col_end);
} else {
return !(mi_row + mi_pos->row < tile->mi_row_start ||
mi_col + mi_pos->col < tile->mi_col_start ||
mi_row + mi_pos->row >= tile->mi_row_end ||
mi_col + mi_pos->col >= tile->mi_col_end);
}
}
static INLINE void lower_mv_precision(MV *mv, int allow_hp) {
if (!allow_hp) {
if (mv->row & 1) mv->row += (mv->row > 0 ? -1 : 1);
if (mv->col & 1) mv->col += (mv->col > 0 ? -1 : 1);
}
}
#if CONFIG_REF_MV
static INLINE uint8_t av1_get_pred_diff_ctx(const int_mv pred_mv,
const int_mv this_mv) {
if (abs(this_mv.as_mv.row - pred_mv.as_mv.row) <= 4 &&
abs(this_mv.as_mv.col - pred_mv.as_mv.col) <= 4)
return 2;
else
return 1;
}
static INLINE int av1_nmv_ctx(const uint8_t ref_mv_count,
const CANDIDATE_MV *ref_mv_stack, int ref,
int ref_mv_idx) {
if (ref_mv_stack[ref_mv_idx].weight >= REF_CAT_LEVEL && ref_mv_count > 0)
return ref_mv_stack[ref_mv_idx].pred_diff[ref];
return 0;
}
static INLINE int8_t av1_ref_frame_type(const MV_REFERENCE_FRAME *const rf) {
if (rf[1] > INTRA_FRAME) {
return TOTAL_REFS_PER_FRAME + FWD_RF_OFFSET(rf[0]) +
BWD_RF_OFFSET(rf[1]) * FWD_REFS;
}
return rf[0];
}
// clang-format off
static MV_REFERENCE_FRAME ref_frame_map[COMP_REFS][2] = {
#if CONFIG_EXT_REFS
{ LAST_FRAME, BWDREF_FRAME }, { LAST2_FRAME, BWDREF_FRAME },
{ LAST3_FRAME, BWDREF_FRAME }, { GOLDEN_FRAME, BWDREF_FRAME },
{ LAST_FRAME, ALTREF_FRAME }, { LAST2_FRAME, ALTREF_FRAME },
{ LAST3_FRAME, ALTREF_FRAME }, { GOLDEN_FRAME, ALTREF_FRAME }
#else
{ LAST_FRAME, ALTREF_FRAME }, { GOLDEN_FRAME, ALTREF_FRAME }
#endif
};
// clang-format on
static INLINE void av1_set_ref_frame(MV_REFERENCE_FRAME *rf,
int8_t ref_frame_type) {
if (ref_frame_type >= TOTAL_REFS_PER_FRAME) {
rf[0] = ref_frame_map[ref_frame_type - TOTAL_REFS_PER_FRAME][0];
rf[1] = ref_frame_map[ref_frame_type - TOTAL_REFS_PER_FRAME][1];
} else {
rf[0] = ref_frame_type;
rf[1] = NONE_FRAME;
assert(ref_frame_type > INTRA_FRAME &&
ref_frame_type < TOTAL_REFS_PER_FRAME);
}
}
static INLINE int16_t av1_mode_context_analyzer(
const int16_t *const mode_context, const MV_REFERENCE_FRAME *const rf,
BLOCK_SIZE bsize, int block) {
int16_t mode_ctx = 0;
int8_t ref_frame_type = av1_ref_frame_type(rf);
if (block >= 0) {
mode_ctx = mode_context[rf[0]] & 0x00ff;
#if !CONFIG_CB4X4
if (block > 0 && bsize < BLOCK_8X8 && bsize > BLOCK_4X4)
mode_ctx |= (1 << SKIP_NEARESTMV_SUB8X8_OFFSET);
#else
(void)block;
(void)bsize;
#endif
return mode_ctx;
}
return mode_context[ref_frame_type];
}
static INLINE uint8_t av1_drl_ctx(const CANDIDATE_MV *ref_mv_stack,
int ref_idx) {
if (ref_mv_stack[ref_idx].weight >= REF_CAT_LEVEL &&
ref_mv_stack[ref_idx + 1].weight >= REF_CAT_LEVEL)
return 0;
if (ref_mv_stack[ref_idx].weight >= REF_CAT_LEVEL &&
ref_mv_stack[ref_idx + 1].weight < REF_CAT_LEVEL)
return 2;
if (ref_mv_stack[ref_idx].weight < REF_CAT_LEVEL &&
ref_mv_stack[ref_idx + 1].weight < REF_CAT_LEVEL)
return 3;
return 0;
}
#endif
typedef void (*find_mv_refs_sync)(void *const data, int mi_row);
void av1_find_mv_refs(const AV1_COMMON *cm, const MACROBLOCKD *xd,
MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
#if CONFIG_REF_MV
uint8_t *ref_mv_count, CANDIDATE_MV *ref_mv_stack,
#if CONFIG_EXT_INTER
int16_t *compound_mode_context,
#endif // CONFIG_EXT_INTER
#endif
int_mv *mv_ref_list, int mi_row, int mi_col,
find_mv_refs_sync sync, void *const data,
int16_t *mode_context);
// check a list of motion vectors by sad score using a number rows of pixels
// above and a number cols of pixels in the left to select the one with best
// score to use as ref motion vector
void av1_find_best_ref_mvs(int allow_hp, int_mv *mvlist, int_mv *nearest_mv,
int_mv *near_mv);
void av1_append_sub8x8_mvs_for_idx(const AV1_COMMON *cm, MACROBLOCKD *xd,
int block, int ref, int mi_row, int mi_col,
#if CONFIG_REF_MV
CANDIDATE_MV *ref_mv_stack,
uint8_t *ref_mv_count,
#endif
#if CONFIG_EXT_INTER
int_mv *mv_list,
#endif // CONFIG_EXT_INTER
int_mv *nearest_mv, int_mv *near_mv);
#if CONFIG_EXT_INTER
// This function keeps a mode count for a given MB/SB
void av1_update_mv_context(const AV1_COMMON *cm, const MACROBLOCKD *xd,
MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
int_mv *mv_ref_list, int block, int mi_row,
int mi_col, int16_t *mode_context);
#endif // CONFIG_EXT_INTER
#if CONFIG_WARPED_MOTION
int findSamples(const AV1_COMMON *cm, MACROBLOCKD *xd, int mi_row, int mi_col,
int *pts, int *pts_inref);
#endif // CONFIG_WARPED_MOTION
#if CONFIG_INTRABC
static INLINE void av1_find_ref_dv(int_mv *ref_dv, int mi_row, int mi_col) {
// TODO(aconverse@google.com): Handle tiles and such
(void)mi_col;
if (mi_row < MAX_MIB_SIZE) {
ref_dv->as_mv.row = 0;
ref_dv->as_mv.col = -MI_SIZE * MAX_MIB_SIZE;
} else {
ref_dv->as_mv.row = -MI_SIZE * MAX_MIB_SIZE;
ref_dv->as_mv.col = 0;
}
}
static INLINE int is_dv_valid(const MV dv, const TileInfo *const tile,
int mi_row, int mi_col, BLOCK_SIZE bsize) {
const int bw = block_size_wide[bsize];
const int bh = block_size_high[bsize];
const int SCALE_PX_TO_MV = 8;
// Disallow subpixel for now
// SUBPEL_MASK is not the correct scale
if ((dv.row & (SCALE_PX_TO_MV - 1) || dv.col & (SCALE_PX_TO_MV - 1)))
return 0;
// Is the source top-left inside the current tile?
const int src_top_edge = mi_row * MI_SIZE * SCALE_PX_TO_MV + dv.row;
const int tile_top_edge = tile->mi_row_start * MI_SIZE * SCALE_PX_TO_MV;
if (src_top_edge < tile_top_edge) return 0;
const int src_left_edge = mi_col * MI_SIZE * SCALE_PX_TO_MV + dv.col;
const int tile_left_edge = tile->mi_col_start * MI_SIZE * SCALE_PX_TO_MV;
if (src_left_edge < tile_left_edge) return 0;
// Is the bottom right inside the current tile?
const int src_bottom_edge = (mi_row * MI_SIZE + bh) * SCALE_PX_TO_MV + dv.row;
const int tile_bottom_edge = tile->mi_row_end * MI_SIZE * SCALE_PX_TO_MV;
if (src_bottom_edge > tile_bottom_edge) return 0;
const int src_right_edge = (mi_col * MI_SIZE + bw) * SCALE_PX_TO_MV + dv.col;
const int tile_right_edge = tile->mi_col_end * MI_SIZE * SCALE_PX_TO_MV;
if (src_right_edge > tile_right_edge) return 0;
// Is the bottom right within an already coded SB?
const int active_sb_top_edge =
(mi_row & ~MAX_MIB_MASK) * MI_SIZE * SCALE_PX_TO_MV;
const int active_sb_bottom_edge =
((mi_row & ~MAX_MIB_MASK) + MAX_MIB_SIZE) * MI_SIZE * SCALE_PX_TO_MV;
const int active_sb_left_edge =
(mi_col & ~MAX_MIB_MASK) * MI_SIZE * SCALE_PX_TO_MV;
if (src_bottom_edge > active_sb_bottom_edge) return 0;
if (src_bottom_edge > active_sb_top_edge &&
src_right_edge > active_sb_left_edge)
return 0;
return 1;
}
#endif // CONFIG_INTRABC
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_MVREF_COMMON_H_

416
third_party/aom/av1/common/od_dering.c vendored Normal file
View file

@ -0,0 +1,416 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <math.h>
#include <stdlib.h>
#ifdef HAVE_CONFIG_H
#include "./config.h"
#endif
#include "./aom_dsp_rtcd.h"
#include "./av1_rtcd.h"
#include "./cdef.h"
/* Generated from gen_filter_tables.c. */
const int OD_DIRECTION_OFFSETS_TABLE[8][3] = {
{ -1 * OD_FILT_BSTRIDE + 1, -2 * OD_FILT_BSTRIDE + 2,
-3 * OD_FILT_BSTRIDE + 3 },
{ 0 * OD_FILT_BSTRIDE + 1, -1 * OD_FILT_BSTRIDE + 2,
-1 * OD_FILT_BSTRIDE + 3 },
{ 0 * OD_FILT_BSTRIDE + 1, 0 * OD_FILT_BSTRIDE + 2, 0 * OD_FILT_BSTRIDE + 3 },
{ 0 * OD_FILT_BSTRIDE + 1, 1 * OD_FILT_BSTRIDE + 2, 1 * OD_FILT_BSTRIDE + 3 },
{ 1 * OD_FILT_BSTRIDE + 1, 2 * OD_FILT_BSTRIDE + 2, 3 * OD_FILT_BSTRIDE + 3 },
{ 1 * OD_FILT_BSTRIDE + 0, 2 * OD_FILT_BSTRIDE + 1, 3 * OD_FILT_BSTRIDE + 1 },
{ 1 * OD_FILT_BSTRIDE + 0, 2 * OD_FILT_BSTRIDE + 0, 3 * OD_FILT_BSTRIDE + 0 },
{ 1 * OD_FILT_BSTRIDE + 0, 2 * OD_FILT_BSTRIDE - 1, 3 * OD_FILT_BSTRIDE - 1 },
};
/* Detect direction. 0 means 45-degree up-right, 2 is horizontal, and so on.
The search minimizes the weighted variance along all the lines in a
particular direction, i.e. the squared error between the input and a
"predicted" block where each pixel is replaced by the average along a line
in a particular direction. Since each direction have the same sum(x^2) term,
that term is never computed. See Section 2, step 2, of:
http://jmvalin.ca/notes/intra_paint.pdf */
int od_dir_find8_c(const uint16_t *img, int stride, int32_t *var,
int coeff_shift) {
int i;
int32_t cost[8] = { 0 };
int partial[8][15] = { { 0 } };
int32_t best_cost = 0;
int best_dir = 0;
/* Instead of dividing by n between 2 and 8, we multiply by 3*5*7*8/n.
The output is then 840 times larger, but we don't care for finding
the max. */
static const int div_table[] = { 0, 840, 420, 280, 210, 168, 140, 120, 105 };
for (i = 0; i < 8; i++) {
int j;
for (j = 0; j < 8; j++) {
int x;
/* We subtract 128 here to reduce the maximum range of the squared
partial sums. */
x = (img[i * stride + j] >> coeff_shift) - 128;
partial[0][i + j] += x;
partial[1][i + j / 2] += x;
partial[2][i] += x;
partial[3][3 + i - j / 2] += x;
partial[4][7 + i - j] += x;
partial[5][3 - i / 2 + j] += x;
partial[6][j] += x;
partial[7][i / 2 + j] += x;
}
}
for (i = 0; i < 8; i++) {
cost[2] += partial[2][i] * partial[2][i];
cost[6] += partial[6][i] * partial[6][i];
}
cost[2] *= div_table[8];
cost[6] *= div_table[8];
for (i = 0; i < 7; i++) {
cost[0] += (partial[0][i] * partial[0][i] +
partial[0][14 - i] * partial[0][14 - i]) *
div_table[i + 1];
cost[4] += (partial[4][i] * partial[4][i] +
partial[4][14 - i] * partial[4][14 - i]) *
div_table[i + 1];
}
cost[0] += partial[0][7] * partial[0][7] * div_table[8];
cost[4] += partial[4][7] * partial[4][7] * div_table[8];
for (i = 1; i < 8; i += 2) {
int j;
for (j = 0; j < 4 + 1; j++) {
cost[i] += partial[i][3 + j] * partial[i][3 + j];
}
cost[i] *= div_table[8];
for (j = 0; j < 4 - 1; j++) {
cost[i] += (partial[i][j] * partial[i][j] +
partial[i][10 - j] * partial[i][10 - j]) *
div_table[2 * j + 2];
}
}
for (i = 0; i < 8; i++) {
if (cost[i] > best_cost) {
best_cost = cost[i];
best_dir = i;
}
}
/* Difference between the optimal variance and the variance along the
orthogonal direction. Again, the sum(x^2) terms cancel out. */
*var = best_cost - cost[(best_dir + 4) & 7];
/* We'd normally divide by 840, but dividing by 1024 is close enough
for what we're going to do with this. */
*var >>= 10;
return best_dir;
}
/* Smooth in the direction detected. */
void od_filter_dering_direction_8x8_c(uint16_t *y, int ystride,
const uint16_t *in, int threshold,
int dir, int damping) {
int i;
int j;
int k;
static const int taps[3] = { 3, 2, 1 };
for (i = 0; i < 8; i++) {
for (j = 0; j < 8; j++) {
int16_t sum;
int16_t xx;
int16_t yy;
xx = in[i * OD_FILT_BSTRIDE + j];
sum = 0;
for (k = 0; k < 3; k++) {
int16_t p0;
int16_t p1;
p0 = in[i * OD_FILT_BSTRIDE + j + OD_DIRECTION_OFFSETS_TABLE[dir][k]] -
xx;
p1 = in[i * OD_FILT_BSTRIDE + j - OD_DIRECTION_OFFSETS_TABLE[dir][k]] -
xx;
sum += taps[k] * constrain(p0, threshold, damping);
sum += taps[k] * constrain(p1, threshold, damping);
}
sum = (sum + 8) >> 4;
yy = xx + sum;
y[i * ystride + j] = yy;
}
}
}
/* Smooth in the direction detected. */
void od_filter_dering_direction_4x4_c(uint16_t *y, int ystride,
const uint16_t *in, int threshold,
int dir, int damping) {
int i;
int j;
int k;
static const int taps[2] = { 4, 1 };
for (i = 0; i < 4; i++) {
for (j = 0; j < 4; j++) {
int16_t sum;
int16_t xx;
int16_t yy;
xx = in[i * OD_FILT_BSTRIDE + j];
sum = 0;
for (k = 0; k < 2; k++) {
int16_t p0;
int16_t p1;
p0 = in[i * OD_FILT_BSTRIDE + j + OD_DIRECTION_OFFSETS_TABLE[dir][k]] -
xx;
p1 = in[i * OD_FILT_BSTRIDE + j - OD_DIRECTION_OFFSETS_TABLE[dir][k]] -
xx;
sum += taps[k] * constrain(p0, threshold, damping);
sum += taps[k] * constrain(p1, threshold, damping);
}
sum = (sum + 8) >> 4;
yy = xx + sum;
y[i * ystride + j] = yy;
}
}
}
/* Compute deringing filter threshold for an 8x8 block based on the
directional variance difference. A high variance difference means that we
have a highly directional pattern (e.g. a high contrast edge), so we can
apply more deringing. A low variance means that we either have a low
contrast edge, or a non-directional texture, so we want to be careful not
to blur. */
static INLINE int od_adjust_thresh(int threshold, int32_t var) {
const int i = var >> 6 ? AOMMIN(get_msb(var >> 6), 12) : 0;
/* We use the variance of 8x8 blocks to adjust the threshold. */
return var ? (threshold * (4 + i) + 8) >> 4 : 0;
}
void copy_8x8_16bit_to_16bit_c(uint16_t *dst, int dstride, const uint16_t *src,
int sstride) {
int i, j;
for (i = 0; i < 8; i++)
for (j = 0; j < 8; j++) dst[i * dstride + j] = src[i * sstride + j];
}
void copy_4x4_16bit_to_16bit_c(uint16_t *dst, int dstride, const uint16_t *src,
int sstride) {
int i, j;
for (i = 0; i < 4; i++)
for (j = 0; j < 4; j++) dst[i * dstride + j] = src[i * sstride + j];
}
void copy_dering_16bit_to_16bit(uint16_t *dst, int dstride, uint16_t *src,
dering_list *dlist, int dering_count,
int bsize) {
int bi, bx, by;
if (bsize == BLOCK_8X8) {
for (bi = 0; bi < dering_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
copy_8x8_16bit_to_16bit(&dst[(by << 3) * dstride + (bx << 3)], dstride,
&src[bi << (3 + 3)], 8);
}
} else if (bsize == BLOCK_4X8) {
for (bi = 0; bi < dering_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
copy_4x4_16bit_to_16bit(&dst[(by << 3) * dstride + (bx << 2)], dstride,
&src[bi << (3 + 2)], 4);
copy_4x4_16bit_to_16bit(&dst[((by << 3) + 4) * dstride + (bx << 2)],
dstride, &src[(bi << (3 + 2)) + 4 * 4], 4);
}
} else if (bsize == BLOCK_8X4) {
for (bi = 0; bi < dering_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
copy_4x4_16bit_to_16bit(&dst[(by << 2) * dstride + (bx << 3)], dstride,
&src[bi << (2 + 3)], 8);
copy_4x4_16bit_to_16bit(&dst[(by << 2) * dstride + (bx << 3) + 4],
dstride, &src[(bi << (2 + 3)) + 4], 8);
}
} else {
assert(bsize == BLOCK_4X4);
for (bi = 0; bi < dering_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
copy_4x4_16bit_to_16bit(&dst[(by << 2) * dstride + (bx << 2)], dstride,
&src[bi << (2 + 2)], 4);
}
}
}
void copy_8x8_16bit_to_8bit_c(uint8_t *dst, int dstride, const uint16_t *src,
int sstride) {
int i, j;
for (i = 0; i < 8; i++)
for (j = 0; j < 8; j++)
dst[i * dstride + j] = (uint8_t)src[i * sstride + j];
}
void copy_4x4_16bit_to_8bit_c(uint8_t *dst, int dstride, const uint16_t *src,
int sstride) {
int i, j;
for (i = 0; i < 4; i++)
for (j = 0; j < 4; j++)
dst[i * dstride + j] = (uint8_t)src[i * sstride + j];
}
static void copy_dering_16bit_to_8bit(uint8_t *dst, int dstride,
const uint16_t *src, dering_list *dlist,
int dering_count, int bsize) {
int bi, bx, by;
if (bsize == BLOCK_8X8) {
for (bi = 0; bi < dering_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
copy_8x8_16bit_to_8bit(&dst[(by << 3) * dstride + (bx << 3)], dstride,
&src[bi << (3 + 3)], 8);
}
} else if (bsize == BLOCK_4X8) {
for (bi = 0; bi < dering_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
copy_4x4_16bit_to_8bit(&dst[(by << 3) * dstride + (bx << 2)], dstride,
&src[bi << (3 + 2)], 4);
copy_4x4_16bit_to_8bit(&dst[((by << 3) + 4) * dstride + (bx << 2)],
dstride, &src[(bi << (3 + 2)) + 4 * 4], 4);
}
} else if (bsize == BLOCK_8X4) {
for (bi = 0; bi < dering_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
copy_4x4_16bit_to_8bit(&dst[(by << 2) * dstride + (bx << 3)], dstride,
&src[bi << (2 + 3)], 8);
copy_4x4_16bit_to_8bit(&dst[(by << 2) * dstride + (bx << 3) + 4], dstride,
&src[(bi << (2 + 3)) + 4], 8);
}
} else {
assert(bsize == BLOCK_4X4);
for (bi = 0; bi < dering_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
copy_4x4_16bit_to_8bit(&dst[(by << 2) * dstride + (bx << 2)], dstride,
&src[bi << (2 * 2)], 4);
}
}
}
int get_filter_skip(int level) {
int filter_skip = level & 1;
if (level == 1) filter_skip = 0;
return filter_skip;
}
void od_dering(uint8_t *dst, int dstride, uint16_t *y, uint16_t *in, int xdec,
int ydec, int dir[OD_DERING_NBLOCKS][OD_DERING_NBLOCKS],
int *dirinit, int var[OD_DERING_NBLOCKS][OD_DERING_NBLOCKS],
int pli, dering_list *dlist, int dering_count, int level,
int clpf_strength, int clpf_damping, int dering_damping,
int coeff_shift, int skip_dering, int hbd) {
int bi;
int bx;
int by;
int bsize, bsizex, bsizey;
int threshold = (level >> 1) << coeff_shift;
int filter_skip = get_filter_skip(level);
if (level == 1) threshold = 31 << coeff_shift;
od_filter_dering_direction_func filter_dering_direction[] = {
od_filter_dering_direction_4x4, od_filter_dering_direction_8x8
};
clpf_damping += coeff_shift - (pli != AOM_PLANE_Y);
dering_damping += coeff_shift - (pli != AOM_PLANE_Y);
bsize =
ydec ? (xdec ? BLOCK_4X4 : BLOCK_8X4) : (xdec ? BLOCK_4X8 : BLOCK_8X8);
bsizex = 3 - xdec;
bsizey = 3 - ydec;
if (!skip_dering) {
if (pli == 0) {
if (!dirinit || !*dirinit) {
for (bi = 0; bi < dering_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
dir[by][bx] =
od_dir_find8(&in[8 * by * OD_FILT_BSTRIDE + 8 * bx],
OD_FILT_BSTRIDE, &var[by][bx], coeff_shift);
}
if (dirinit) *dirinit = 1;
}
}
// Only run dering for non-zero threshold (which is always the case for
// 4:2:2 or 4:4:0). If we don't dering, we still need to eventually write
// something out in y[] later.
if (threshold != 0) {
assert(bsize == BLOCK_8X8 || bsize == BLOCK_4X4);
for (bi = 0; bi < dering_count; bi++) {
int t = !filter_skip && dlist[bi].skip ? 0 : threshold;
by = dlist[bi].by;
bx = dlist[bi].bx;
(filter_dering_direction[bsize == BLOCK_8X8])(
&y[bi << (bsizex + bsizey)], 1 << bsizex,
&in[(by * OD_FILT_BSTRIDE << bsizey) + (bx << bsizex)],
pli ? t : od_adjust_thresh(t, var[by][bx]), dir[by][bx],
dering_damping);
}
}
}
if (clpf_strength) {
if (threshold && !skip_dering)
copy_dering_16bit_to_16bit(in, OD_FILT_BSTRIDE, y, dlist, dering_count,
bsize);
for (bi = 0; bi < dering_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
int py = by << bsizey;
int px = bx << bsizex;
if (!filter_skip && dlist[bi].skip) continue;
if (!dst || hbd) {
// 16 bit destination if high bitdepth or 8 bit destination not given
(!threshold || (dir[by][bx] < 4 && dir[by][bx]) ? aom_clpf_block_hbd
: aom_clpf_hblock_hbd)(
dst ? (uint16_t *)dst + py * dstride + px
: &y[bi << (bsizex + bsizey)],
in + py * OD_FILT_BSTRIDE + px, dst && hbd ? dstride : 1 << bsizex,
OD_FILT_BSTRIDE, 1 << bsizex, 1 << bsizey,
clpf_strength << coeff_shift, clpf_damping);
} else {
// Do clpf and write the result to an 8 bit destination
(!threshold || (dir[by][bx] < 4 && dir[by][bx]) ? aom_clpf_block
: aom_clpf_hblock)(
dst + py * dstride + px, in + py * OD_FILT_BSTRIDE + px, dstride,
OD_FILT_BSTRIDE, 1 << bsizex, 1 << bsizey,
clpf_strength << coeff_shift, clpf_damping);
}
}
} else if (threshold != 0) {
// No clpf, so copy instead
if (hbd) {
copy_dering_16bit_to_16bit((uint16_t *)dst, dstride, y, dlist,
dering_count, bsize);
} else {
copy_dering_16bit_to_8bit(dst, dstride, y, dlist, dering_count, bsize);
}
} else if (dirinit) {
// If we're here, both dering and clpf are off, and we still haven't written
// anything to y[] yet, so we just copy the input to y[]. This is necessary
// only for av1_cdef_search() and only av1_cdef_search() sets dirinit.
for (bi = 0; bi < dering_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
int iy, ix;
// TODO(stemidts/jmvalin): SIMD optimisations
for (iy = 0; iy < 1 << bsizey; iy++)
for (ix = 0; ix < 1 << bsizex; ix++)
y[(bi << (bsizex + bsizey)) + (iy << bsizex) + ix] =
in[((by << bsizey) + iy) * OD_FILT_BSTRIDE + (bx << bsizex) + ix];
}
}
}

54
third_party/aom/av1/common/od_dering.h vendored Normal file
View file

@ -0,0 +1,54 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#if !defined(_dering_H)
#define _dering_H (1)
#include "odintrin.h"
#define OD_DERING_NBLOCKS (MAX_SB_SIZE / 8)
/* We need to buffer three vertical lines. */
#define OD_FILT_VBORDER (3)
/* We only need to buffer three horizontal pixels too, but let's align to
16 bytes (8 x 16 bits) to make vectorization easier. */
#define OD_FILT_HBORDER (8)
#define OD_FILT_BSTRIDE ALIGN_POWER_OF_TWO(MAX_SB_SIZE + 2 * OD_FILT_HBORDER, 3)
#define OD_DERING_VERY_LARGE (30000)
#define OD_DERING_INBUF_SIZE \
(OD_FILT_BSTRIDE * (MAX_SB_SIZE + 2 * OD_FILT_VBORDER))
extern const int OD_DIRECTION_OFFSETS_TABLE[8][3];
typedef struct {
uint8_t by;
uint8_t bx;
uint8_t skip;
} dering_list;
typedef void (*od_filter_dering_direction_func)(uint16_t *y, int ystride,
const uint16_t *in,
int threshold, int dir,
int damping);
void copy_dering_16bit_to_16bit(uint16_t *dst, int dstride, uint16_t *src,
dering_list *dlist, int dering_count,
int bsize);
int get_filter_skip(int level);
void od_dering(uint8_t *dst, int dstride, uint16_t *y, uint16_t *in, int xdec,
int ydec, int dir[OD_DERING_NBLOCKS][OD_DERING_NBLOCKS],
int *dirinit, int var[OD_DERING_NBLOCKS][OD_DERING_NBLOCKS],
int pli, dering_list *dlist, int dering_count, int level,
int clpf_strength, int clpf_damping, int dering_damping,
int coeff_shift, int skip_dering, int hbd);
#endif

14
third_party/aom/av1/common/od_dering_neon.c vendored Normal file
View file

@ -0,0 +1,14 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "aom_dsp/aom_simd.h"
#define SIMD_FUNC(name) name##_neon
#include "./od_dering_simd.h"

390
third_party/aom/av1/common/od_dering_simd.h vendored Normal file
View file

@ -0,0 +1,390 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "./av1_rtcd.h"
#include "./cdef_simd.h"
#include "./od_dering.h"
/* partial A is a 16-bit vector of the form:
[x8 x7 x6 x5 x4 x3 x2 x1] and partial B has the form:
[0 y1 y2 y3 y4 y5 y6 y7].
This function computes (x1^2+y1^2)*C1 + (x2^2+y2^2)*C2 + ...
(x7^2+y2^7)*C7 + (x8^2+0^2)*C8 where the C1..C8 constants are in const1
and const2. */
static INLINE v128 fold_mul_and_sum(v128 partiala, v128 partialb, v128 const1,
v128 const2) {
v128 tmp;
/* Reverse partial B. */
partialb = v128_shuffle_8(
partialb, v128_from_32(0x0f0e0100, 0x03020504, 0x07060908, 0x0b0a0d0c));
/* Interleave the x and y values of identical indices and pair x8 with 0. */
tmp = partiala;
partiala = v128_ziplo_16(partialb, partiala);
partialb = v128_ziphi_16(partialb, tmp);
/* Square and add the corresponding x and y values. */
partiala = v128_madd_s16(partiala, partiala);
partialb = v128_madd_s16(partialb, partialb);
/* Multiply by constant. */
partiala = v128_mullo_s32(partiala, const1);
partialb = v128_mullo_s32(partialb, const2);
/* Sum all results. */
partiala = v128_add_32(partiala, partialb);
return partiala;
}
static INLINE v128 hsum4(v128 x0, v128 x1, v128 x2, v128 x3) {
v128 t0, t1, t2, t3;
t0 = v128_ziplo_32(x1, x0);
t1 = v128_ziplo_32(x3, x2);
t2 = v128_ziphi_32(x1, x0);
t3 = v128_ziphi_32(x3, x2);
x0 = v128_ziplo_64(t1, t0);
x1 = v128_ziphi_64(t1, t0);
x2 = v128_ziplo_64(t3, t2);
x3 = v128_ziphi_64(t3, t2);
return v128_add_32(v128_add_32(x0, x1), v128_add_32(x2, x3));
}
/* Computes cost for directions 0, 5, 6 and 7. We can call this function again
to compute the remaining directions. */
static INLINE v128 compute_directions(v128 lines[8], int32_t tmp_cost1[4]) {
v128 partial4a, partial4b, partial5a, partial5b, partial7a, partial7b;
v128 partial6;
v128 tmp;
/* Partial sums for lines 0 and 1. */
partial4a = v128_shl_n_byte(lines[0], 14);
partial4b = v128_shr_n_byte(lines[0], 2);
partial4a = v128_add_16(partial4a, v128_shl_n_byte(lines[1], 12));
partial4b = v128_add_16(partial4b, v128_shr_n_byte(lines[1], 4));
tmp = v128_add_16(lines[0], lines[1]);
partial5a = v128_shl_n_byte(tmp, 10);
partial5b = v128_shr_n_byte(tmp, 6);
partial7a = v128_shl_n_byte(tmp, 4);
partial7b = v128_shr_n_byte(tmp, 12);
partial6 = tmp;
/* Partial sums for lines 2 and 3. */
partial4a = v128_add_16(partial4a, v128_shl_n_byte(lines[2], 10));
partial4b = v128_add_16(partial4b, v128_shr_n_byte(lines[2], 6));
partial4a = v128_add_16(partial4a, v128_shl_n_byte(lines[3], 8));
partial4b = v128_add_16(partial4b, v128_shr_n_byte(lines[3], 8));
tmp = v128_add_16(lines[2], lines[3]);
partial5a = v128_add_16(partial5a, v128_shl_n_byte(tmp, 8));
partial5b = v128_add_16(partial5b, v128_shr_n_byte(tmp, 8));
partial7a = v128_add_16(partial7a, v128_shl_n_byte(tmp, 6));
partial7b = v128_add_16(partial7b, v128_shr_n_byte(tmp, 10));
partial6 = v128_add_16(partial6, tmp);
/* Partial sums for lines 4 and 5. */
partial4a = v128_add_16(partial4a, v128_shl_n_byte(lines[4], 6));
partial4b = v128_add_16(partial4b, v128_shr_n_byte(lines[4], 10));
partial4a = v128_add_16(partial4a, v128_shl_n_byte(lines[5], 4));
partial4b = v128_add_16(partial4b, v128_shr_n_byte(lines[5], 12));
tmp = v128_add_16(lines[4], lines[5]);
partial5a = v128_add_16(partial5a, v128_shl_n_byte(tmp, 6));
partial5b = v128_add_16(partial5b, v128_shr_n_byte(tmp, 10));
partial7a = v128_add_16(partial7a, v128_shl_n_byte(tmp, 8));
partial7b = v128_add_16(partial7b, v128_shr_n_byte(tmp, 8));
partial6 = v128_add_16(partial6, tmp);
/* Partial sums for lines 6 and 7. */
partial4a = v128_add_16(partial4a, v128_shl_n_byte(lines[6], 2));
partial4b = v128_add_16(partial4b, v128_shr_n_byte(lines[6], 14));
partial4a = v128_add_16(partial4a, lines[7]);
tmp = v128_add_16(lines[6], lines[7]);
partial5a = v128_add_16(partial5a, v128_shl_n_byte(tmp, 4));
partial5b = v128_add_16(partial5b, v128_shr_n_byte(tmp, 12));
partial7a = v128_add_16(partial7a, v128_shl_n_byte(tmp, 10));
partial7b = v128_add_16(partial7b, v128_shr_n_byte(tmp, 6));
partial6 = v128_add_16(partial6, tmp);
/* Compute costs in terms of partial sums. */
partial4a =
fold_mul_and_sum(partial4a, partial4b, v128_from_32(210, 280, 420, 840),
v128_from_32(105, 120, 140, 168));
partial7a =
fold_mul_and_sum(partial7a, partial7b, v128_from_32(210, 420, 0, 0),
v128_from_32(105, 105, 105, 140));
partial5a =
fold_mul_and_sum(partial5a, partial5b, v128_from_32(210, 420, 0, 0),
v128_from_32(105, 105, 105, 140));
partial6 = v128_madd_s16(partial6, partial6);
partial6 = v128_mullo_s32(partial6, v128_dup_32(105));
partial4a = hsum4(partial4a, partial5a, partial6, partial7a);
v128_store_unaligned(tmp_cost1, partial4a);
return partial4a;
}
/* transpose and reverse the order of the lines -- equivalent to a 90-degree
counter-clockwise rotation of the pixels. */
static INLINE void array_reverse_transpose_8x8(v128 *in, v128 *res) {
const v128 tr0_0 = v128_ziplo_16(in[1], in[0]);
const v128 tr0_1 = v128_ziplo_16(in[3], in[2]);
const v128 tr0_2 = v128_ziphi_16(in[1], in[0]);
const v128 tr0_3 = v128_ziphi_16(in[3], in[2]);
const v128 tr0_4 = v128_ziplo_16(in[5], in[4]);
const v128 tr0_5 = v128_ziplo_16(in[7], in[6]);
const v128 tr0_6 = v128_ziphi_16(in[5], in[4]);
const v128 tr0_7 = v128_ziphi_16(in[7], in[6]);
const v128 tr1_0 = v128_ziplo_32(tr0_1, tr0_0);
const v128 tr1_1 = v128_ziplo_32(tr0_5, tr0_4);
const v128 tr1_2 = v128_ziphi_32(tr0_1, tr0_0);
const v128 tr1_3 = v128_ziphi_32(tr0_5, tr0_4);
const v128 tr1_4 = v128_ziplo_32(tr0_3, tr0_2);
const v128 tr1_5 = v128_ziplo_32(tr0_7, tr0_6);
const v128 tr1_6 = v128_ziphi_32(tr0_3, tr0_2);
const v128 tr1_7 = v128_ziphi_32(tr0_7, tr0_6);
res[7] = v128_ziplo_64(tr1_1, tr1_0);
res[6] = v128_ziphi_64(tr1_1, tr1_0);
res[5] = v128_ziplo_64(tr1_3, tr1_2);
res[4] = v128_ziphi_64(tr1_3, tr1_2);
res[3] = v128_ziplo_64(tr1_5, tr1_4);
res[2] = v128_ziphi_64(tr1_5, tr1_4);
res[1] = v128_ziplo_64(tr1_7, tr1_6);
res[0] = v128_ziphi_64(tr1_7, tr1_6);
}
int SIMD_FUNC(od_dir_find8)(const od_dering_in *img, int stride, int32_t *var,
int coeff_shift) {
int i;
int32_t cost[8];
int32_t best_cost = 0;
int best_dir = 0;
v128 lines[8];
for (i = 0; i < 8; i++) {
lines[i] = v128_load_unaligned(&img[i * stride]);
lines[i] =
v128_sub_16(v128_shr_s16(lines[i], coeff_shift), v128_dup_16(128));
}
#if defined(__SSE4_1__)
/* Compute "mostly vertical" directions. */
__m128i dir47 = compute_directions(lines, cost + 4);
array_reverse_transpose_8x8(lines, lines);
/* Compute "mostly horizontal" directions. */
__m128i dir03 = compute_directions(lines, cost);
__m128i max = _mm_max_epi32(dir03, dir47);
max = _mm_max_epi32(max, _mm_shuffle_epi32(max, _MM_SHUFFLE(1, 0, 3, 2)));
max = _mm_max_epi32(max, _mm_shuffle_epi32(max, _MM_SHUFFLE(2, 3, 0, 1)));
best_cost = _mm_cvtsi128_si32(max);
__m128i t =
_mm_packs_epi32(_mm_cmpeq_epi32(max, dir03), _mm_cmpeq_epi32(max, dir47));
best_dir = _mm_movemask_epi8(_mm_packs_epi16(t, t));
best_dir = get_msb(best_dir ^ (best_dir - 1)); // Count trailing zeros
#else
/* Compute "mostly vertical" directions. */
compute_directions(lines, cost + 4);
array_reverse_transpose_8x8(lines, lines);
/* Compute "mostly horizontal" directions. */
compute_directions(lines, cost);
for (i = 0; i < 8; i++) {
if (cost[i] > best_cost) {
best_cost = cost[i];
best_dir = i;
}
}
#endif
/* Difference between the optimal variance and the variance along the
orthogonal direction. Again, the sum(x^2) terms cancel out. */
*var = best_cost - cost[(best_dir + 4) & 7];
/* We'd normally divide by 840, but dividing by 1024 is close enough
for what we're going to do with this. */
*var >>= 10;
return best_dir;
}
void SIMD_FUNC(od_filter_dering_direction_4x4)(uint16_t *y, int ystride,
const uint16_t *in,
int threshold, int dir,
int damping) {
int i;
v128 p0, p1, sum, row, res;
int o1 = OD_DIRECTION_OFFSETS_TABLE[dir][0];
int o2 = OD_DIRECTION_OFFSETS_TABLE[dir][1];
if (threshold) damping -= get_msb(threshold);
for (i = 0; i < 4; i += 2) {
sum = v128_zero();
row = v128_from_v64(v64_load_aligned(&in[i * OD_FILT_BSTRIDE]),
v64_load_aligned(&in[(i + 1) * OD_FILT_BSTRIDE]));
// p0 = constrain16(in[i*OD_FILT_BSTRIDE + offset], row, threshold, damping)
p0 = v128_from_v64(v64_load_unaligned(&in[i * OD_FILT_BSTRIDE + o1]),
v64_load_unaligned(&in[(i + 1) * OD_FILT_BSTRIDE + o1]));
p0 = constrain16(p0, row, threshold, damping);
// p1 = constrain16(in[i*OD_FILT_BSTRIDE - offset], row, threshold, damping)
p1 = v128_from_v64(v64_load_unaligned(&in[i * OD_FILT_BSTRIDE - o1]),
v64_load_unaligned(&in[(i + 1) * OD_FILT_BSTRIDE - o1]));
p1 = constrain16(p1, row, threshold, damping);
// sum += 4 * (p0 + p1)
sum = v128_add_16(sum, v128_shl_n_16(v128_add_16(p0, p1), 2));
// p0 = constrain16(in[i*OD_FILT_BSTRIDE + offset], row, threshold, damping)
p0 = v128_from_v64(v64_load_unaligned(&in[i * OD_FILT_BSTRIDE + o2]),
v64_load_unaligned(&in[(i + 1) * OD_FILT_BSTRIDE + o2]));
p0 = constrain16(p0, row, threshold, damping);
// p1 = constrain16(in[i*OD_FILT_BSTRIDE - offset], row, threshold, damping)
p1 = v128_from_v64(v64_load_unaligned(&in[i * OD_FILT_BSTRIDE - o2]),
v64_load_unaligned(&in[(i + 1) * OD_FILT_BSTRIDE - o2]));
p1 = constrain16(p1, row, threshold, damping);
// sum += 1 * (p0 + p1)
sum = v128_add_16(sum, v128_add_16(p0, p1));
// res = row + ((sum + 8) >> 4)
res = v128_add_16(sum, v128_dup_16(8));
res = v128_shr_n_s16(res, 4);
res = v128_add_16(row, res);
v64_store_aligned(&y[i * ystride], v128_high_v64(res));
v64_store_aligned(&y[(i + 1) * ystride], v128_low_v64(res));
}
}
void SIMD_FUNC(od_filter_dering_direction_8x8)(uint16_t *y, int ystride,
const uint16_t *in,
int threshold, int dir,
int damping) {
int i;
v128 sum, p0, p1, row, res;
int o1 = OD_DIRECTION_OFFSETS_TABLE[dir][0];
int o2 = OD_DIRECTION_OFFSETS_TABLE[dir][1];
int o3 = OD_DIRECTION_OFFSETS_TABLE[dir][2];
if (threshold) damping -= get_msb(threshold);
for (i = 0; i < 8; i++) {
sum = v128_zero();
row = v128_load_aligned(&in[i * OD_FILT_BSTRIDE]);
// p0 = constrain16(in[i*OD_FILT_BSTRIDE + offset], row, threshold, damping)
p0 = v128_load_unaligned(&in[i * OD_FILT_BSTRIDE + o1]);
p0 = constrain16(p0, row, threshold, damping);
// p1 = constrain16(in[i*OD_FILT_BSTRIDE - offset], row, threshold, damping)
p1 = v128_load_unaligned(&in[i * OD_FILT_BSTRIDE - o1]);
p1 = constrain16(p1, row, threshold, damping);
// sum += 3 * (p0 + p1)
p0 = v128_add_16(p0, p1);
p0 = v128_add_16(p0, v128_shl_n_16(p0, 1));
sum = v128_add_16(sum, p0);
// p0 = constrain16(in[i*OD_FILT_BSTRIDE + offset], row, threshold, damping)
p0 = v128_load_unaligned(&in[i * OD_FILT_BSTRIDE + o2]);
p0 = constrain16(p0, row, threshold, damping);
// p1 = constrain16(in[i*OD_FILT_BSTRIDE - offset], row, threshold, damping)
p1 = v128_load_unaligned(&in[i * OD_FILT_BSTRIDE - o2]);
p1 = constrain16(p1, row, threshold, damping);
// sum += 2 * (p0 + p1)
p0 = v128_shl_n_16(v128_add_16(p0, p1), 1);
sum = v128_add_16(sum, p0);
// p0 = constrain16(in[i*OD_FILT_BSTRIDE + offset], row, threshold, damping)
p0 = v128_load_unaligned(&in[i * OD_FILT_BSTRIDE + o3]);
p0 = constrain16(p0, row, threshold, damping);
// p1 = constrain16(in[i*OD_FILT_BSTRIDE - offset], row, threshold, damping)
p1 = v128_load_unaligned(&in[i * OD_FILT_BSTRIDE - o3]);
p1 = constrain16(p1, row, threshold, damping);
// sum += (p0 + p1)
p0 = v128_add_16(p0, p1);
sum = v128_add_16(sum, p0);
// res = row + ((sum + 8) >> 4)
res = v128_add_16(sum, v128_dup_16(8));
res = v128_shr_n_s16(res, 4);
res = v128_add_16(row, res);
v128_store_unaligned(&y[i * ystride], res);
}
}
void SIMD_FUNC(copy_8x8_16bit_to_8bit)(uint8_t *dst, int dstride,
const uint16_t *src, int sstride) {
int i;
for (i = 0; i < 8; i++) {
v128 row = v128_load_unaligned(&src[i * sstride]);
row = v128_pack_s16_u8(row, row);
v64_store_unaligned(&dst[i * dstride], v128_low_v64(row));
}
}
void SIMD_FUNC(copy_4x4_16bit_to_8bit)(uint8_t *dst, int dstride,
const uint16_t *src, int sstride) {
int i;
for (i = 0; i < 4; i++) {
v128 row = v128_load_unaligned(&src[i * sstride]);
row = v128_pack_s16_u8(row, row);
u32_store_unaligned(&dst[i * dstride], v128_low_u32(row));
}
}
void SIMD_FUNC(copy_8x8_16bit_to_16bit)(uint16_t *dst, int dstride,
const uint16_t *src, int sstride) {
int i;
for (i = 0; i < 8; i++) {
v128 row = v128_load_unaligned(&src[i * sstride]);
v128_store_unaligned(&dst[i * dstride], row);
}
}
void SIMD_FUNC(copy_4x4_16bit_to_16bit)(uint16_t *dst, int dstride,
const uint16_t *src, int sstride) {
int i;
for (i = 0; i < 4; i++) {
v64 row = v64_load_unaligned(&src[i * sstride]);
v64_store_unaligned(&dst[i * dstride], row);
}
}
void SIMD_FUNC(copy_rect8_8bit_to_16bit)(uint16_t *dst, int dstride,
const uint8_t *src, int sstride, int v,
int h) {
int i, j;
for (i = 0; i < v; i++) {
for (j = 0; j < (h & ~0x7); j += 8) {
v64 row = v64_load_unaligned(&src[i * sstride + j]);
v128_store_unaligned(&dst[i * dstride + j], v128_unpack_u8_s16(row));
}
for (; j < h; j++) {
dst[i * dstride + j] = src[i * sstride + j];
}
}
}
void SIMD_FUNC(copy_rect8_16bit_to_16bit)(uint16_t *dst, int dstride,
const uint16_t *src, int sstride,
int v, int h) {
int i, j;
for (i = 0; i < v; i++) {
for (j = 0; j < (h & ~0x7); j += 8) {
v128 row = v128_load_unaligned(&src[i * sstride + j]);
v128_store_unaligned(&dst[i * dstride + j], row);
}
for (; j < h; j++) {
dst[i * dstride + j] = src[i * sstride + j];
}
}
}

14
third_party/aom/av1/common/od_dering_sse2.c vendored Normal file
View file

@ -0,0 +1,14 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "aom_dsp/aom_simd.h"
#define SIMD_FUNC(name) name##_sse2
#include "./od_dering_simd.h"

14
third_party/aom/av1/common/od_dering_sse4.c vendored Normal file
View file

@ -0,0 +1,14 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "aom_dsp/aom_simd.h"
#define SIMD_FUNC(name) name##_sse4_1
#include "./od_dering_simd.h"

14
third_party/aom/av1/common/od_dering_ssse3.c vendored Normal file
View file

@ -0,0 +1,14 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "aom_dsp/aom_simd.h"
#define SIMD_FUNC(name) name##_ssse3
#include "./od_dering_simd.h"

551
third_party/aom/av1/common/odintrin.c vendored Normal file
View file

@ -0,0 +1,551 @@
/*
* Copyright (c) 2001-2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
/* clang-format off */
#include "av1/common/odintrin.h"
#if defined(OD_ENABLE_ASSERTIONS)
# include <stdio.h>
void od_fatal_impl(const char *_str, const char *_file, int _line) {
fprintf(stderr, "Fatal (internal) error in %s, line %d: %s\n",
_file, _line, _str);
abort();
}
#endif
/*Constants for use with OD_DIVU_SMALL().
See \cite{Rob05} for details on computing these constants.
@INPROCEEDINGS{Rob05,
author="Arch D. Robison",
title="{N}-bit Unsigned Division via {N}-bit Multiply-Add",
booktitle="Proc. of the 17th IEEE Symposium on Computer Arithmetic
(ARITH'05)",
pages="131--139",
address="Cape Cod, MA",
month=Jun,
year=2005
}*/
uint32_t OD_DIVU_SMALL_CONSTS[OD_DIVU_DMAX][2] = {
{ 0xFFFFFFFF, 0xFFFFFFFF }, { 0xFFFFFFFF, 0xFFFFFFFF },
{ 0xAAAAAAAB, 0 }, { 0xFFFFFFFF, 0xFFFFFFFF },
{ 0xCCCCCCCD, 0 }, { 0xAAAAAAAB, 0 },
{ 0x92492492, 0x92492492 }, { 0xFFFFFFFF, 0xFFFFFFFF },
{ 0xE38E38E4, 0 }, { 0xCCCCCCCD, 0 },
{ 0xBA2E8BA3, 0 }, { 0xAAAAAAAB, 0 },
{ 0x9D89D89E, 0 }, { 0x92492492, 0x92492492 },
{ 0x88888889, 0 }, { 0xFFFFFFFF, 0xFFFFFFFF },
{ 0xF0F0F0F1, 0 }, { 0xE38E38E4, 0 },
{ 0xD79435E5, 0xD79435E5 }, { 0xCCCCCCCD, 0 },
{ 0xC30C30C3, 0xC30C30C3 }, { 0xBA2E8BA3, 0 },
{ 0xB21642C9, 0 }, { 0xAAAAAAAB, 0 },
{ 0xA3D70A3E, 0 }, { 0x9D89D89E, 0 },
{ 0x97B425ED, 0x97B425ED }, { 0x92492492, 0x92492492 },
{ 0x8D3DCB09, 0 }, { 0x88888889, 0 },
{ 0x84210842, 0x84210842 }, { 0xFFFFFFFF, 0xFFFFFFFF },
{ 0xF83E0F84, 0 }, { 0xF0F0F0F1, 0 },
{ 0xEA0EA0EA, 0xEA0EA0EA }, { 0xE38E38E4, 0 },
{ 0xDD67C8A6, 0xDD67C8A6 }, { 0xD79435E5, 0xD79435E5 },
{ 0xD20D20D2, 0xD20D20D2 }, { 0xCCCCCCCD, 0 },
{ 0xC7CE0C7D, 0 }, { 0xC30C30C3, 0xC30C30C3 },
{ 0xBE82FA0C, 0 }, { 0xBA2E8BA3, 0 },
{ 0xB60B60B6, 0xB60B60B6 }, { 0xB21642C9, 0 },
{ 0xAE4C415D, 0 }, { 0xAAAAAAAB, 0 },
{ 0xA72F053A, 0 }, { 0xA3D70A3E, 0 },
{ 0xA0A0A0A1, 0 }, { 0x9D89D89E, 0 },
{ 0x9A90E7D9, 0x9A90E7D9 }, { 0x97B425ED, 0x97B425ED },
{ 0x94F2094F, 0x94F2094F }, { 0x92492492, 0x92492492 },
{ 0x8FB823EE, 0x8FB823EE }, { 0x8D3DCB09, 0 },
{ 0x8AD8F2FC, 0 }, { 0x88888889, 0 },
{ 0x864B8A7E, 0 }, { 0x84210842, 0x84210842 },
{ 0x82082082, 0x82082082 }, { 0xFFFFFFFF, 0xFFFFFFFF },
{ 0xFC0FC0FD, 0 }, { 0xF83E0F84, 0 },
{ 0xF4898D60, 0 }, { 0xF0F0F0F1, 0 },
{ 0xED7303B6, 0 }, { 0xEA0EA0EA, 0xEA0EA0EA },
{ 0xE6C2B449, 0 }, { 0xE38E38E4, 0 },
{ 0xE070381C, 0xE070381C }, { 0xDD67C8A6, 0xDD67C8A6 },
{ 0xDA740DA8, 0 }, { 0xD79435E5, 0xD79435E5 },
{ 0xD4C77B04, 0 }, { 0xD20D20D2, 0xD20D20D2 },
{ 0xCF6474A9, 0 }, { 0xCCCCCCCD, 0 },
{ 0xCA4587E7, 0 }, { 0xC7CE0C7D, 0 },
{ 0xC565C87C, 0 }, { 0xC30C30C3, 0xC30C30C3 },
{ 0xC0C0C0C1, 0 }, { 0xBE82FA0C, 0 },
{ 0xBC52640C, 0 }, { 0xBA2E8BA3, 0 },
{ 0xB81702E1, 0 }, { 0xB60B60B6, 0xB60B60B6 },
{ 0xB40B40B4, 0xB40B40B4 }, { 0xB21642C9, 0 },
{ 0xB02C0B03, 0 }, { 0xAE4C415D, 0 },
{ 0xAC769184, 0xAC769184 }, { 0xAAAAAAAB, 0 },
{ 0xA8E83F57, 0xA8E83F57 }, { 0xA72F053A, 0 },
{ 0xA57EB503, 0 }, { 0xA3D70A3E, 0 },
{ 0xA237C32B, 0xA237C32B }, { 0xA0A0A0A1, 0 },
{ 0x9F1165E7, 0x9F1165E7 }, { 0x9D89D89E, 0 },
{ 0x9C09C09C, 0x9C09C09C }, { 0x9A90E7D9, 0x9A90E7D9 },
{ 0x991F1A51, 0x991F1A51 }, { 0x97B425ED, 0x97B425ED },
{ 0x964FDA6C, 0x964FDA6C }, { 0x94F2094F, 0x94F2094F },
{ 0x939A85C4, 0x939A85C4 }, { 0x92492492, 0x92492492 },
{ 0x90FDBC09, 0x90FDBC09 }, { 0x8FB823EE, 0x8FB823EE },
{ 0x8E78356D, 0x8E78356D }, { 0x8D3DCB09, 0 },
{ 0x8C08C08C, 0x8C08C08C }, { 0x8AD8F2FC, 0 },
{ 0x89AE408A, 0 }, { 0x88888889, 0 },
{ 0x8767AB5F, 0x8767AB5F }, { 0x864B8A7E, 0 },
{ 0x85340853, 0x85340853 }, { 0x84210842, 0x84210842 },
{ 0x83126E98, 0 }, { 0x82082082, 0x82082082 },
{ 0x81020408, 0x81020408 }, { 0xFFFFFFFF, 0xFFFFFFFF },
{ 0xFE03F810, 0 }, { 0xFC0FC0FD, 0 },
{ 0xFA232CF3, 0 }, { 0xF83E0F84, 0 },
{ 0xF6603D99, 0 }, { 0xF4898D60, 0 },
{ 0xF2B9D649, 0 }, { 0xF0F0F0F1, 0 },
{ 0xEF2EB720, 0 }, { 0xED7303B6, 0 },
{ 0xEBBDB2A6, 0 }, { 0xEA0EA0EA, 0xEA0EA0EA },
{ 0xE865AC7C, 0 }, { 0xE6C2B449, 0 },
{ 0xE525982B, 0 }, { 0xE38E38E4, 0 },
{ 0xE1FC780F, 0 }, { 0xE070381C, 0xE070381C },
{ 0xDEE95C4D, 0 }, { 0xDD67C8A6, 0xDD67C8A6 },
{ 0xDBEB61EF, 0 }, { 0xDA740DA8, 0 },
{ 0xD901B204, 0 }, { 0xD79435E5, 0xD79435E5 },
{ 0xD62B80D7, 0 }, { 0xD4C77B04, 0 },
{ 0xD3680D37, 0 }, { 0xD20D20D2, 0xD20D20D2 },
{ 0xD0B69FCC, 0 }, { 0xCF6474A9, 0 },
{ 0xCE168A77, 0xCE168A77 }, { 0xCCCCCCCD, 0 },
{ 0xCB8727C1, 0 }, { 0xCA4587E7, 0 },
{ 0xC907DA4F, 0 }, { 0xC7CE0C7D, 0 },
{ 0xC6980C6A, 0 }, { 0xC565C87C, 0 },
{ 0xC4372F86, 0 }, { 0xC30C30C3, 0xC30C30C3 },
{ 0xC1E4BBD6, 0 }, { 0xC0C0C0C1, 0 },
{ 0xBFA02FE8, 0xBFA02FE8 }, { 0xBE82FA0C, 0 },
{ 0xBD691047, 0xBD691047 }, { 0xBC52640C, 0 },
{ 0xBB3EE722, 0 }, { 0xBA2E8BA3, 0 },
{ 0xB92143FA, 0xB92143FA }, { 0xB81702E1, 0 },
{ 0xB70FBB5A, 0xB70FBB5A }, { 0xB60B60B6, 0xB60B60B6 },
{ 0xB509E68B, 0 }, { 0xB40B40B4, 0xB40B40B4 },
{ 0xB30F6353, 0 }, { 0xB21642C9, 0 },
{ 0xB11FD3B8, 0xB11FD3B8 }, { 0xB02C0B03, 0 },
{ 0xAF3ADDC7, 0 }, { 0xAE4C415D, 0 },
{ 0xAD602B58, 0xAD602B58 }, { 0xAC769184, 0xAC769184 },
{ 0xAB8F69E3, 0 }, { 0xAAAAAAAB, 0 },
{ 0xA9C84A48, 0 }, { 0xA8E83F57, 0xA8E83F57 },
{ 0xA80A80A8, 0xA80A80A8 }, { 0xA72F053A, 0 },
{ 0xA655C439, 0xA655C439 }, { 0xA57EB503, 0 },
{ 0xA4A9CF1E, 0 }, { 0xA3D70A3E, 0 },
{ 0xA3065E40, 0 }, { 0xA237C32B, 0xA237C32B },
{ 0xA16B312F, 0 }, { 0xA0A0A0A1, 0 },
{ 0x9FD809FE, 0 }, { 0x9F1165E7, 0x9F1165E7 },
{ 0x9E4CAD24, 0 }, { 0x9D89D89E, 0 },
{ 0x9CC8E161, 0 }, { 0x9C09C09C, 0x9C09C09C },
{ 0x9B4C6F9F, 0 }, { 0x9A90E7D9, 0x9A90E7D9 },
{ 0x99D722DB, 0 }, { 0x991F1A51, 0x991F1A51 },
{ 0x9868C80A, 0 }, { 0x97B425ED, 0x97B425ED },
{ 0x97012E02, 0x97012E02 }, { 0x964FDA6C, 0x964FDA6C },
{ 0x95A02568, 0x95A02568 }, { 0x94F2094F, 0x94F2094F },
{ 0x94458094, 0x94458094 }, { 0x939A85C4, 0x939A85C4 },
{ 0x92F11384, 0x92F11384 }, { 0x92492492, 0x92492492 },
{ 0x91A2B3C5, 0 }, { 0x90FDBC09, 0x90FDBC09 },
{ 0x905A3863, 0x905A3863 }, { 0x8FB823EE, 0x8FB823EE },
{ 0x8F1779DA, 0 }, { 0x8E78356D, 0x8E78356D },
{ 0x8DDA5202, 0x8DDA5202 }, { 0x8D3DCB09, 0 },
{ 0x8CA29C04, 0x8CA29C04 }, { 0x8C08C08C, 0x8C08C08C },
{ 0x8B70344A, 0x8B70344A }, { 0x8AD8F2FC, 0 },
{ 0x8A42F870, 0x8A42F870 }, { 0x89AE408A, 0 },
{ 0x891AC73B, 0 }, { 0x88888889, 0 },
{ 0x87F78088, 0 }, { 0x8767AB5F, 0x8767AB5F },
{ 0x86D90545, 0 }, { 0x864B8A7E, 0 },
{ 0x85BF3761, 0x85BF3761 }, { 0x85340853, 0x85340853 },
{ 0x84A9F9C8, 0x84A9F9C8 }, { 0x84210842, 0x84210842 },
{ 0x83993052, 0x83993052 }, { 0x83126E98, 0 },
{ 0x828CBFBF, 0 }, { 0x82082082, 0x82082082 },
{ 0x81848DA9, 0 }, { 0x81020408, 0x81020408 },
{ 0x80808081, 0 }, { 0xFFFFFFFF, 0xFFFFFFFF },
{ 0xFF00FF01, 0 }, { 0xFE03F810, 0 },
{ 0xFD08E551, 0 }, { 0xFC0FC0FD, 0 },
{ 0xFB188566, 0 }, { 0xFA232CF3, 0 },
{ 0xF92FB222, 0 }, { 0xF83E0F84, 0 },
{ 0xF74E3FC3, 0 }, { 0xF6603D99, 0 },
{ 0xF57403D6, 0 }, { 0xF4898D60, 0 },
{ 0xF3A0D52D, 0 }, { 0xF2B9D649, 0 },
{ 0xF1D48BCF, 0 }, { 0xF0F0F0F1, 0 },
{ 0xF00F00F0, 0xF00F00F0 }, { 0xEF2EB720, 0 },
{ 0xEE500EE5, 0xEE500EE5 }, { 0xED7303B6, 0 },
{ 0xEC979119, 0 }, { 0xEBBDB2A6, 0 },
{ 0xEAE56404, 0 }, { 0xEA0EA0EA, 0xEA0EA0EA },
{ 0xE9396520, 0 }, { 0xE865AC7C, 0 },
{ 0xE79372E3, 0 }, { 0xE6C2B449, 0 },
{ 0xE5F36CB0, 0xE5F36CB0 }, { 0xE525982B, 0 },
{ 0xE45932D8, 0 }, { 0xE38E38E4, 0 },
{ 0xE2C4A689, 0 }, { 0xE1FC780F, 0 },
{ 0xE135A9CA, 0 }, { 0xE070381C, 0xE070381C },
{ 0xDFAC1F75, 0 }, { 0xDEE95C4D, 0 },
{ 0xDE27EB2D, 0 }, { 0xDD67C8A6, 0xDD67C8A6 },
{ 0xDCA8F159, 0 }, { 0xDBEB61EF, 0 },
{ 0xDB2F171E, 0 }, { 0xDA740DA8, 0 },
{ 0xD9BA4257, 0 }, { 0xD901B204, 0 },
{ 0xD84A598F, 0 }, { 0xD79435E5, 0xD79435E5 },
{ 0xD6DF43FD, 0 }, { 0xD62B80D7, 0 },
{ 0xD578E97D, 0 }, { 0xD4C77B04, 0 },
{ 0xD417328A, 0 }, { 0xD3680D37, 0 },
{ 0xD2BA083C, 0 }, { 0xD20D20D2, 0xD20D20D2 },
{ 0xD161543E, 0xD161543E }, { 0xD0B69FCC, 0 },
{ 0xD00D00D0, 0xD00D00D0 }, { 0xCF6474A9, 0 },
{ 0xCEBCF8BC, 0 }, { 0xCE168A77, 0xCE168A77 },
{ 0xCD712753, 0 }, { 0xCCCCCCCD, 0 },
{ 0xCC29786D, 0 }, { 0xCB8727C1, 0 },
{ 0xCAE5D85F, 0xCAE5D85F }, { 0xCA4587E7, 0 },
{ 0xC9A633FD, 0 }, { 0xC907DA4F, 0 },
{ 0xC86A7890, 0xC86A7890 }, { 0xC7CE0C7D, 0 },
{ 0xC73293D8, 0 }, { 0xC6980C6A, 0 },
{ 0xC5FE7403, 0xC5FE7403 }, { 0xC565C87C, 0 },
{ 0xC4CE07B0, 0xC4CE07B0 }, { 0xC4372F86, 0 },
{ 0xC3A13DE6, 0xC3A13DE6 }, { 0xC30C30C3, 0xC30C30C3 },
{ 0xC2780614, 0 }, { 0xC1E4BBD6, 0 },
{ 0xC152500C, 0xC152500C }, { 0xC0C0C0C1, 0 },
{ 0xC0300C03, 0xC0300C03 }, { 0xBFA02FE8, 0xBFA02FE8 },
{ 0xBF112A8B, 0 }, { 0xBE82FA0C, 0 },
{ 0xBDF59C92, 0 }, { 0xBD691047, 0xBD691047 },
{ 0xBCDD535E, 0 }, { 0xBC52640C, 0 },
{ 0xBBC8408D, 0 }, { 0xBB3EE722, 0 },
{ 0xBAB65610, 0xBAB65610 }, { 0xBA2E8BA3, 0 },
{ 0xB9A7862A, 0xB9A7862A }, { 0xB92143FA, 0xB92143FA },
{ 0xB89BC36D, 0 }, { 0xB81702E1, 0 },
{ 0xB79300B8, 0 }, { 0xB70FBB5A, 0xB70FBB5A },
{ 0xB68D3134, 0xB68D3134 }, { 0xB60B60B6, 0xB60B60B6 },
{ 0xB58A4855, 0xB58A4855 }, { 0xB509E68B, 0 },
{ 0xB48A39D4, 0xB48A39D4 }, { 0xB40B40B4, 0xB40B40B4 },
{ 0xB38CF9B0, 0xB38CF9B0 }, { 0xB30F6353, 0 },
{ 0xB2927C2A, 0 }, { 0xB21642C9, 0 },
{ 0xB19AB5C5, 0 }, { 0xB11FD3B8, 0xB11FD3B8 },
{ 0xB0A59B42, 0 }, { 0xB02C0B03, 0 },
{ 0xAFB321A1, 0xAFB321A1 }, { 0xAF3ADDC7, 0 },
{ 0xAEC33E20, 0 }, { 0xAE4C415D, 0 },
{ 0xADD5E632, 0xADD5E632 }, { 0xAD602B58, 0xAD602B58 },
{ 0xACEB0F89, 0xACEB0F89 }, { 0xAC769184, 0xAC769184 },
{ 0xAC02B00B, 0 }, { 0xAB8F69E3, 0 },
{ 0xAB1CBDD4, 0 }, { 0xAAAAAAAB, 0 },
{ 0xAA392F36, 0 }, { 0xA9C84A48, 0 },
{ 0xA957FAB5, 0xA957FAB5 }, { 0xA8E83F57, 0xA8E83F57 },
{ 0xA8791709, 0 }, { 0xA80A80A8, 0xA80A80A8 },
{ 0xA79C7B17, 0 }, { 0xA72F053A, 0 },
{ 0xA6C21DF7, 0 }, { 0xA655C439, 0xA655C439 },
{ 0xA5E9F6ED, 0xA5E9F6ED }, { 0xA57EB503, 0 },
{ 0xA513FD6C, 0 }, { 0xA4A9CF1E, 0 },
{ 0xA4402910, 0xA4402910 }, { 0xA3D70A3E, 0 },
{ 0xA36E71A3, 0 }, { 0xA3065E40, 0 },
{ 0xA29ECF16, 0xA29ECF16 }, { 0xA237C32B, 0xA237C32B },
{ 0xA1D13986, 0 }, { 0xA16B312F, 0 },
{ 0xA105A933, 0 }, { 0xA0A0A0A1, 0 },
{ 0xA03C1689, 0 }, { 0x9FD809FE, 0 },
{ 0x9F747A15, 0x9F747A15 }, { 0x9F1165E7, 0x9F1165E7 },
{ 0x9EAECC8D, 0x9EAECC8D }, { 0x9E4CAD24, 0 },
{ 0x9DEB06C9, 0x9DEB06C9 }, { 0x9D89D89E, 0 },
{ 0x9D2921C4, 0 }, { 0x9CC8E161, 0 },
{ 0x9C69169B, 0x9C69169B }, { 0x9C09C09C, 0x9C09C09C },
{ 0x9BAADE8E, 0x9BAADE8E }, { 0x9B4C6F9F, 0 },
{ 0x9AEE72FD, 0 }, { 0x9A90E7D9, 0x9A90E7D9 },
{ 0x9A33CD67, 0x9A33CD67 }, { 0x99D722DB, 0 },
{ 0x997AE76B, 0x997AE76B }, { 0x991F1A51, 0x991F1A51 },
{ 0x98C3BAC7, 0x98C3BAC7 }, { 0x9868C80A, 0 },
{ 0x980E4156, 0x980E4156 }, { 0x97B425ED, 0x97B425ED },
{ 0x975A7510, 0 }, { 0x97012E02, 0x97012E02 },
{ 0x96A8500A, 0 }, { 0x964FDA6C, 0x964FDA6C },
{ 0x95F7CC73, 0 }, { 0x95A02568, 0x95A02568 },
{ 0x9548E498, 0 }, { 0x94F2094F, 0x94F2094F },
{ 0x949B92DE, 0 }, { 0x94458094, 0x94458094 },
{ 0x93EFD1C5, 0x93EFD1C5 }, { 0x939A85C4, 0x939A85C4 },
{ 0x93459BE7, 0 }, { 0x92F11384, 0x92F11384 },
{ 0x929CEBF5, 0 }, { 0x92492492, 0x92492492 },
{ 0x91F5BCB9, 0 }, { 0x91A2B3C5, 0 },
{ 0x91500915, 0x91500915 }, { 0x90FDBC09, 0x90FDBC09 },
{ 0x90ABCC02, 0x90ABCC02 }, { 0x905A3863, 0x905A3863 },
{ 0x90090090, 0x90090090 }, { 0x8FB823EE, 0x8FB823EE },
{ 0x8F67A1E4, 0 }, { 0x8F1779DA, 0 },
{ 0x8EC7AB3A, 0 }, { 0x8E78356D, 0x8E78356D },
{ 0x8E2917E1, 0 }, { 0x8DDA5202, 0x8DDA5202 },
{ 0x8D8BE340, 0 }, { 0x8D3DCB09, 0 },
{ 0x8CF008CF, 0x8CF008CF }, { 0x8CA29C04, 0x8CA29C04 },
{ 0x8C55841D, 0 }, { 0x8C08C08C, 0x8C08C08C },
{ 0x8BBC50C9, 0 }, { 0x8B70344A, 0x8B70344A },
{ 0x8B246A88, 0 }, { 0x8AD8F2FC, 0 },
{ 0x8A8DCD20, 0 }, { 0x8A42F870, 0x8A42F870 },
{ 0x89F8746A, 0 }, { 0x89AE408A, 0 },
{ 0x89645C4F, 0x89645C4F }, { 0x891AC73B, 0 },
{ 0x88D180CD, 0x88D180CD }, { 0x88888889, 0 },
{ 0x883FDDF0, 0x883FDDF0 }, { 0x87F78088, 0 },
{ 0x87AF6FD6, 0 }, { 0x8767AB5F, 0x8767AB5F },
{ 0x872032AC, 0x872032AC }, { 0x86D90545, 0 },
{ 0x869222B2, 0 }, { 0x864B8A7E, 0 },
{ 0x86053C34, 0x86053C34 }, { 0x85BF3761, 0x85BF3761 },
{ 0x85797B91, 0x85797B91 }, { 0x85340853, 0x85340853 },
{ 0x84EEDD36, 0 }, { 0x84A9F9C8, 0x84A9F9C8 },
{ 0x84655D9C, 0 }, { 0x84210842, 0x84210842 },
{ 0x83DCF94E, 0 }, { 0x83993052, 0x83993052 },
{ 0x8355ACE4, 0 }, { 0x83126E98, 0 },
{ 0x82CF7504, 0 }, { 0x828CBFBF, 0 },
{ 0x824A4E61, 0 }, { 0x82082082, 0x82082082 },
{ 0x81C635BC, 0x81C635BC }, { 0x81848DA9, 0 },
{ 0x814327E4, 0 }, { 0x81020408, 0x81020408 },
{ 0x80C121B3, 0 }, { 0x80808081, 0 },
{ 0x80402010, 0x80402010 }, { 0xFFFFFFFF, 0xFFFFFFFF },
{ 0xFF803FE1, 0 }, { 0xFF00FF01, 0 },
{ 0xFE823CA6, 0 }, { 0xFE03F810, 0 },
{ 0xFD863087, 0 }, { 0xFD08E551, 0 },
{ 0xFC8C15B5, 0 }, { 0xFC0FC0FD, 0 },
{ 0xFB93E673, 0 }, { 0xFB188566, 0 },
{ 0xFA9D9D20, 0 }, { 0xFA232CF3, 0 },
{ 0xF9A9342D, 0 }, { 0xF92FB222, 0 },
{ 0xF8B6A622, 0xF8B6A622 }, { 0xF83E0F84, 0 },
{ 0xF7C5ED9D, 0 }, { 0xF74E3FC3, 0 },
{ 0xF6D7054E, 0 }, { 0xF6603D99, 0 },
{ 0xF5E9E7FD, 0 }, { 0xF57403D6, 0 },
{ 0xF4FE9083, 0 }, { 0xF4898D60, 0 },
{ 0xF414F9CE, 0 }, { 0xF3A0D52D, 0 },
{ 0xF32D1EE0, 0 }, { 0xF2B9D649, 0 },
{ 0xF246FACC, 0 }, { 0xF1D48BCF, 0 },
{ 0xF16288B9, 0 }, { 0xF0F0F0F1, 0 },
{ 0xF07FC3E0, 0xF07FC3E0 }, { 0xF00F00F0, 0xF00F00F0 },
{ 0xEF9EA78C, 0 }, { 0xEF2EB720, 0 },
{ 0xEEBF2F19, 0 }, { 0xEE500EE5, 0xEE500EE5 },
{ 0xEDE155F4, 0 }, { 0xED7303B6, 0 },
{ 0xED05179C, 0xED05179C }, { 0xEC979119, 0 },
{ 0xEC2A6FA0, 0xEC2A6FA0 }, { 0xEBBDB2A6, 0 },
{ 0xEB5159A0, 0 }, { 0xEAE56404, 0 },
{ 0xEA79D14A, 0 }, { 0xEA0EA0EA, 0xEA0EA0EA },
{ 0xE9A3D25E, 0xE9A3D25E }, { 0xE9396520, 0 },
{ 0xE8CF58AB, 0 }, { 0xE865AC7C, 0 },
{ 0xE7FC600F, 0 }, { 0xE79372E3, 0 },
{ 0xE72AE476, 0 }, { 0xE6C2B449, 0 },
{ 0xE65AE1DC, 0 }, { 0xE5F36CB0, 0xE5F36CB0 },
{ 0xE58C544A, 0 }, { 0xE525982B, 0 },
{ 0xE4BF37D9, 0 }, { 0xE45932D8, 0 },
{ 0xE3F388AF, 0 }, { 0xE38E38E4, 0 },
{ 0xE32942FF, 0 }, { 0xE2C4A689, 0 },
{ 0xE260630B, 0 }, { 0xE1FC780F, 0 },
{ 0xE198E520, 0 }, { 0xE135A9CA, 0 },
{ 0xE0D2C59A, 0 }, { 0xE070381C, 0xE070381C },
{ 0xE00E00E0, 0xE00E00E0 }, { 0xDFAC1F75, 0 },
{ 0xDF4A9369, 0 }, { 0xDEE95C4D, 0 },
{ 0xDE8879B3, 0 }, { 0xDE27EB2D, 0 },
{ 0xDDC7B04D, 0 }, { 0xDD67C8A6, 0xDD67C8A6 },
{ 0xDD0833CE, 0 }, { 0xDCA8F159, 0 },
{ 0xDC4A00DD, 0 }, { 0xDBEB61EF, 0 },
{ 0xDB8D1428, 0 }, { 0xDB2F171E, 0 },
{ 0xDAD16A6B, 0 }, { 0xDA740DA8, 0 },
{ 0xDA17006D, 0xDA17006D }, { 0xD9BA4257, 0 },
{ 0xD95DD300, 0 }, { 0xD901B204, 0 },
{ 0xD8A5DEFF, 0 }, { 0xD84A598F, 0 },
{ 0xD7EF2152, 0 }, { 0xD79435E5, 0xD79435E5 },
{ 0xD73996E9, 0 }, { 0xD6DF43FD, 0 },
{ 0xD6853CC1, 0 }, { 0xD62B80D7, 0 },
{ 0xD5D20FDF, 0 }, { 0xD578E97D, 0 },
{ 0xD5200D52, 0xD5200D52 }, { 0xD4C77B04, 0 },
{ 0xD46F3235, 0 }, { 0xD417328A, 0 },
{ 0xD3BF7BA9, 0 }, { 0xD3680D37, 0 },
{ 0xD310E6DB, 0 }, { 0xD2BA083C, 0 },
{ 0xD2637101, 0 }, { 0xD20D20D2, 0xD20D20D2 },
{ 0xD1B71759, 0 }, { 0xD161543E, 0xD161543E },
{ 0xD10BD72C, 0 }, { 0xD0B69FCC, 0 },
{ 0xD061ADCA, 0 }, { 0xD00D00D0, 0xD00D00D0 },
{ 0xCFB8988C, 0 }, { 0xCF6474A9, 0 },
{ 0xCF1094D4, 0 }, { 0xCEBCF8BC, 0 },
{ 0xCE69A00D, 0 }, { 0xCE168A77, 0xCE168A77 },
{ 0xCDC3B7A9, 0xCDC3B7A9 }, { 0xCD712753, 0 },
{ 0xCD1ED924, 0 }, { 0xCCCCCCCD, 0 },
{ 0xCC7B0200, 0 }, { 0xCC29786D, 0 },
{ 0xCBD82FC7, 0 }, { 0xCB8727C1, 0 },
{ 0xCB36600D, 0 }, { 0xCAE5D85F, 0xCAE5D85F },
{ 0xCA95906C, 0 }, { 0xCA4587E7, 0 },
{ 0xC9F5BE86, 0 }, { 0xC9A633FD, 0 },
{ 0xC956E803, 0xC956E803 }, { 0xC907DA4F, 0 },
{ 0xC8B90A96, 0 }, { 0xC86A7890, 0xC86A7890 },
{ 0xC81C23F5, 0xC81C23F5 }, { 0xC7CE0C7D, 0 },
{ 0xC78031E0, 0xC78031E0 }, { 0xC73293D8, 0 },
{ 0xC6E5321D, 0 }, { 0xC6980C6A, 0 },
{ 0xC64B2278, 0xC64B2278 }, { 0xC5FE7403, 0xC5FE7403 },
{ 0xC5B200C6, 0 }, { 0xC565C87C, 0 },
{ 0xC519CAE0, 0xC519CAE0 }, { 0xC4CE07B0, 0xC4CE07B0 },
{ 0xC4827EA8, 0xC4827EA8 }, { 0xC4372F86, 0 },
{ 0xC3EC1A06, 0 }, { 0xC3A13DE6, 0xC3A13DE6 },
{ 0xC3569AE6, 0 }, { 0xC30C30C3, 0xC30C30C3 },
{ 0xC2C1FF3E, 0 }, { 0xC2780614, 0 },
{ 0xC22E4507, 0 }, { 0xC1E4BBD6, 0 },
{ 0xC19B6A42, 0 }, { 0xC152500C, 0xC152500C },
{ 0xC1096CF6, 0 }, { 0xC0C0C0C1, 0 },
{ 0xC0784B2F, 0 }, { 0xC0300C03, 0xC0300C03 },
{ 0xBFE80300, 0 }, { 0xBFA02FE8, 0xBFA02FE8 },
{ 0xBF589280, 0 }, { 0xBF112A8B, 0 },
{ 0xBEC9F7CE, 0 }, { 0xBE82FA0C, 0 },
{ 0xBE3C310C, 0 }, { 0xBDF59C92, 0 },
{ 0xBDAF3C64, 0 }, { 0xBD691047, 0xBD691047 },
{ 0xBD231803, 0 }, { 0xBCDD535E, 0 },
{ 0xBC97C21E, 0xBC97C21E }, { 0xBC52640C, 0 },
{ 0xBC0D38EE, 0xBC0D38EE }, { 0xBBC8408D, 0 },
{ 0xBB837AB1, 0 }, { 0xBB3EE722, 0 },
{ 0xBAFA85A9, 0xBAFA85A9 }, { 0xBAB65610, 0xBAB65610 },
{ 0xBA725820, 0xBA725820 }, { 0xBA2E8BA3, 0 },
{ 0xB9EAF063, 0 }, { 0xB9A7862A, 0xB9A7862A },
{ 0xB9644CC4, 0 }, { 0xB92143FA, 0xB92143FA },
{ 0xB8DE6B9A, 0 }, { 0xB89BC36D, 0 },
{ 0xB8594B41, 0 }, { 0xB81702E1, 0 },
{ 0xB7D4EA19, 0xB7D4EA19 }, { 0xB79300B8, 0 },
{ 0xB7514689, 0 }, { 0xB70FBB5A, 0xB70FBB5A },
{ 0xB6CE5EF9, 0xB6CE5EF9 }, { 0xB68D3134, 0xB68D3134 },
{ 0xB64C31D9, 0 }, { 0xB60B60B6, 0xB60B60B6 },
{ 0xB5CABD9B, 0 }, { 0xB58A4855, 0xB58A4855 },
{ 0xB54A00B5, 0xB54A00B5 }, { 0xB509E68B, 0 },
{ 0xB4C9F9A5, 0 }, { 0xB48A39D4, 0xB48A39D4 },
{ 0xB44AA6E9, 0xB44AA6E9 }, { 0xB40B40B4, 0xB40B40B4 },
{ 0xB3CC0706, 0 }, { 0xB38CF9B0, 0xB38CF9B0 },
{ 0xB34E1884, 0 }, { 0xB30F6353, 0 },
{ 0xB2D0D9EF, 0 }, { 0xB2927C2A, 0 },
{ 0xB25449D7, 0 }, { 0xB21642C9, 0 },
{ 0xB1D866D1, 0xB1D866D1 }, { 0xB19AB5C5, 0 },
{ 0xB15D2F76, 0 }, { 0xB11FD3B8, 0xB11FD3B8 },
{ 0xB0E2A260, 0xB0E2A260 }, { 0xB0A59B42, 0 },
{ 0xB068BE31, 0 }, { 0xB02C0B03, 0 },
{ 0xAFEF818C, 0 }, { 0xAFB321A1, 0xAFB321A1 },
{ 0xAF76EB19, 0 }, { 0xAF3ADDC7, 0 },
{ 0xAEFEF982, 0 }, { 0xAEC33E20, 0 },
{ 0xAE87AB76, 0xAE87AB76 }, { 0xAE4C415D, 0 },
{ 0xAE10FFA9, 0 }, { 0xADD5E632, 0xADD5E632 },
{ 0xAD9AF4D0, 0 }, { 0xAD602B58, 0xAD602B58 },
{ 0xAD2589A4, 0 }, { 0xACEB0F89, 0xACEB0F89 },
{ 0xACB0BCE1, 0xACB0BCE1 }, { 0xAC769184, 0xAC769184 },
{ 0xAC3C8D4A, 0 }, { 0xAC02B00B, 0 },
{ 0xABC8F9A0, 0xABC8F9A0 }, { 0xAB8F69E3, 0 },
{ 0xAB5600AC, 0 }, { 0xAB1CBDD4, 0 },
{ 0xAAE3A136, 0 }, { 0xAAAAAAAB, 0 },
{ 0xAA71DA0D, 0 }, { 0xAA392F36, 0 },
{ 0xAA00AA01, 0 }, { 0xA9C84A48, 0 },
{ 0xA9900FE6, 0 }, { 0xA957FAB5, 0xA957FAB5 },
{ 0xA9200A92, 0xA9200A92 }, { 0xA8E83F57, 0xA8E83F57 },
{ 0xA8B098E0, 0xA8B098E0 }, { 0xA8791709, 0 },
{ 0xA841B9AD, 0 }, { 0xA80A80A8, 0xA80A80A8 },
{ 0xA7D36BD8, 0 }, { 0xA79C7B17, 0 },
{ 0xA765AE44, 0 }, { 0xA72F053A, 0 },
{ 0xA6F87FD6, 0xA6F87FD6 }, { 0xA6C21DF7, 0 },
{ 0xA68BDF79, 0 }, { 0xA655C439, 0xA655C439 },
{ 0xA61FCC16, 0xA61FCC16 }, { 0xA5E9F6ED, 0xA5E9F6ED },
{ 0xA5B4449D, 0 }, { 0xA57EB503, 0 },
{ 0xA54947FE, 0 }, { 0xA513FD6C, 0 },
{ 0xA4DED52C, 0xA4DED52C }, { 0xA4A9CF1E, 0 },
{ 0xA474EB1F, 0xA474EB1F }, { 0xA4402910, 0xA4402910 },
{ 0xA40B88D0, 0 }, { 0xA3D70A3E, 0 },
{ 0xA3A2AD39, 0xA3A2AD39 }, { 0xA36E71A3, 0 },
{ 0xA33A575A, 0xA33A575A }, { 0xA3065E40, 0 },
{ 0xA2D28634, 0 }, { 0xA29ECF16, 0xA29ECF16 },
{ 0xA26B38C9, 0 }, { 0xA237C32B, 0xA237C32B },
{ 0xA2046E1F, 0xA2046E1F }, { 0xA1D13986, 0 },
{ 0xA19E2540, 0 }, { 0xA16B312F, 0 },
{ 0xA1385D35, 0 }, { 0xA105A933, 0 },
{ 0xA0D3150C, 0 }, { 0xA0A0A0A1, 0 },
{ 0xA06E4BD4, 0xA06E4BD4 }, { 0xA03C1689, 0 },
{ 0xA00A00A0, 0xA00A00A0 }, { 0x9FD809FE, 0 },
{ 0x9FA63284, 0 }, { 0x9F747A15, 0x9F747A15 },
{ 0x9F42E095, 0x9F42E095 }, { 0x9F1165E7, 0x9F1165E7 },
{ 0x9EE009EE, 0x9EE009EE }, { 0x9EAECC8D, 0x9EAECC8D },
{ 0x9E7DADA9, 0 }, { 0x9E4CAD24, 0 },
{ 0x9E1BCAE3, 0 }, { 0x9DEB06C9, 0x9DEB06C9 },
{ 0x9DBA60BB, 0x9DBA60BB }, { 0x9D89D89E, 0 },
{ 0x9D596E54, 0x9D596E54 }, { 0x9D2921C4, 0 },
{ 0x9CF8F2D1, 0x9CF8F2D1 }, { 0x9CC8E161, 0 },
{ 0x9C98ED58, 0 }, { 0x9C69169B, 0x9C69169B },
{ 0x9C395D10, 0x9C395D10 }, { 0x9C09C09C, 0x9C09C09C },
{ 0x9BDA4124, 0x9BDA4124 }, { 0x9BAADE8E, 0x9BAADE8E },
{ 0x9B7B98C0, 0 }, { 0x9B4C6F9F, 0 },
{ 0x9B1D6311, 0x9B1D6311 }, { 0x9AEE72FD, 0 },
{ 0x9ABF9F48, 0x9ABF9F48 }, { 0x9A90E7D9, 0x9A90E7D9 },
{ 0x9A624C97, 0 }, { 0x9A33CD67, 0x9A33CD67 },
{ 0x9A056A31, 0 }, { 0x99D722DB, 0 },
{ 0x99A8F74C, 0 }, { 0x997AE76B, 0x997AE76B },
{ 0x994CF320, 0x994CF320 }, { 0x991F1A51, 0x991F1A51 },
{ 0x98F15CE7, 0 }, { 0x98C3BAC7, 0x98C3BAC7 },
{ 0x989633DB, 0x989633DB }, { 0x9868C80A, 0 },
{ 0x983B773B, 0 }, { 0x980E4156, 0x980E4156 },
{ 0x97E12644, 0x97E12644 }, { 0x97B425ED, 0x97B425ED },
{ 0x97874039, 0 }, { 0x975A7510, 0 },
{ 0x972DC45B, 0 }, { 0x97012E02, 0x97012E02 },
{ 0x96D4B1EF, 0 }, { 0x96A8500A, 0 },
{ 0x967C083B, 0 }, { 0x964FDA6C, 0x964FDA6C },
{ 0x9623C686, 0x9623C686 }, { 0x95F7CC73, 0 },
{ 0x95CBEC1B, 0 }, { 0x95A02568, 0x95A02568 },
{ 0x95747844, 0 }, { 0x9548E498, 0 },
{ 0x951D6A4E, 0 }, { 0x94F2094F, 0x94F2094F },
{ 0x94C6C187, 0 }, { 0x949B92DE, 0 },
{ 0x94707D3F, 0 }, { 0x94458094, 0x94458094 },
{ 0x941A9CC8, 0x941A9CC8 }, { 0x93EFD1C5, 0x93EFD1C5 },
{ 0x93C51F76, 0 }, { 0x939A85C4, 0x939A85C4 },
{ 0x9370049C, 0 }, { 0x93459BE7, 0 },
{ 0x931B4B91, 0 }, { 0x92F11384, 0x92F11384 },
{ 0x92C6F3AC, 0x92C6F3AC }, { 0x929CEBF5, 0 },
{ 0x9272FC48, 0x9272FC48 }, { 0x92492492, 0x92492492 },
{ 0x921F64BF, 0 }, { 0x91F5BCB9, 0 },
{ 0x91CC2C6C, 0x91CC2C6C }, { 0x91A2B3C5, 0 },
{ 0x917952AF, 0 }, { 0x91500915, 0x91500915 },
{ 0x9126D6E5, 0 }, { 0x90FDBC09, 0x90FDBC09 },
{ 0x90D4B86F, 0 }, { 0x90ABCC02, 0x90ABCC02 },
{ 0x9082F6B0, 0 }, { 0x905A3863, 0x905A3863 },
{ 0x9031910A, 0 }, { 0x90090090, 0x90090090 },
{ 0x8FE086E3, 0 }, { 0x8FB823EE, 0x8FB823EE },
{ 0x8F8FD7A0, 0 }, { 0x8F67A1E4, 0 },
{ 0x8F3F82A8, 0x8F3F82A8 }, { 0x8F1779DA, 0 },
{ 0x8EEF8766, 0 }, { 0x8EC7AB3A, 0 },
{ 0x8E9FE542, 0x8E9FE542 }, { 0x8E78356D, 0x8E78356D },
{ 0x8E509BA8, 0x8E509BA8 }, { 0x8E2917E1, 0 },
{ 0x8E01AA05, 0 }, { 0x8DDA5202, 0x8DDA5202 },
{ 0x8DB30FC6, 0x8DB30FC6 }, { 0x8D8BE340, 0 },
{ 0x8D64CC5C, 0 }, { 0x8D3DCB09, 0 },
{ 0x8D16DF35, 0x8D16DF35 }, { 0x8CF008CF, 0x8CF008CF },
{ 0x8CC947C5, 0 }, { 0x8CA29C04, 0x8CA29C04 },
{ 0x8C7C057D, 0 }, { 0x8C55841D, 0 },
{ 0x8C2F17D2, 0x8C2F17D2 }, { 0x8C08C08C, 0x8C08C08C },
{ 0x8BE27E39, 0x8BE27E39 }, { 0x8BBC50C9, 0 },
{ 0x8B963829, 0x8B963829 }, { 0x8B70344A, 0x8B70344A },
{ 0x8B4A451A, 0 }, { 0x8B246A88, 0 },
{ 0x8AFEA483, 0x8AFEA483 }, { 0x8AD8F2FC, 0 },
{ 0x8AB355E0, 0x8AB355E0 }, { 0x8A8DCD20, 0 },
{ 0x8A6858AB, 0 }, { 0x8A42F870, 0x8A42F870 },
{ 0x8A1DAC60, 0x8A1DAC60 }, { 0x89F8746A, 0 },
{ 0x89D3507D, 0 }, { 0x89AE408A, 0 },
{ 0x89894480, 0 }, { 0x89645C4F, 0x89645C4F },
{ 0x893F87E8, 0x893F87E8 }, { 0x891AC73B, 0 },
{ 0x88F61A37, 0x88F61A37 }, { 0x88D180CD, 0x88D180CD },
{ 0x88ACFAEE, 0 }, { 0x88888889, 0 },
{ 0x8864298F, 0 }, { 0x883FDDF0, 0x883FDDF0 },
{ 0x881BA59E, 0 }, { 0x87F78088, 0 },
{ 0x87D36EA0, 0 }, { 0x87AF6FD6, 0 },
{ 0x878B841B, 0 }, { 0x8767AB5F, 0x8767AB5F },
{ 0x8743E595, 0 }, { 0x872032AC, 0x872032AC },
{ 0x86FC9296, 0x86FC9296 }, { 0x86D90545, 0 },
{ 0x86B58AA8, 0 }, { 0x869222B2, 0 },
{ 0x866ECD53, 0x866ECD53 }, { 0x864B8A7E, 0 },
{ 0x86285A23, 0x86285A23 }, { 0x86053C34, 0x86053C34 },
{ 0x85E230A3, 0x85E230A3 }, { 0x85BF3761, 0x85BF3761 },
{ 0x859C5060, 0x859C5060 }, { 0x85797B91, 0x85797B91 },
{ 0x8556B8E7, 0x8556B8E7 }, { 0x85340853, 0x85340853 },
{ 0x851169C7, 0x851169C7 }, { 0x84EEDD36, 0 },
{ 0x84CC6290, 0 }, { 0x84A9F9C8, 0x84A9F9C8 },
{ 0x8487A2D1, 0 }, { 0x84655D9C, 0 },
{ 0x84432A1B, 0x84432A1B }, { 0x84210842, 0x84210842 },
{ 0x83FEF802, 0x83FEF802 }, { 0x83DCF94E, 0 },
{ 0x83BB0C18, 0 }, { 0x83993052, 0x83993052 },
{ 0x837765F0, 0x837765F0 }, { 0x8355ACE4, 0 },
{ 0x83340520, 0x83340520 }, { 0x83126E98, 0 },
{ 0x82F0E93D, 0x82F0E93D }, { 0x82CF7504, 0 },
{ 0x82AE11DE, 0 }, { 0x828CBFBF, 0 },
{ 0x826B7E99, 0x826B7E99 }, { 0x824A4E61, 0 },
{ 0x82292F08, 0 }, { 0x82082082, 0x82082082 },
{ 0x81E722C2, 0x81E722C2 }, { 0x81C635BC, 0x81C635BC },
{ 0x81A55963, 0 }, { 0x81848DA9, 0 },
{ 0x8163D283, 0 }, { 0x814327E4, 0 },
{ 0x81228DBF, 0 }, { 0x81020408, 0x81020408 },
{ 0x80E18AB3, 0 }, { 0x80C121B3, 0 },
{ 0x80A0C8FB, 0x80A0C8FB }, { 0x80808081, 0 },
{ 0x80604836, 0x80604836 }, { 0x80402010, 0x80402010 },
{ 0x80200802, 0x80200802 }, { 0xFFFFFFFF, 0xFFFFFFFF }
};

267
third_party/aom/av1/common/odintrin.h vendored Normal file
View file

@ -0,0 +1,267 @@
/*
* Copyright (c) 2001-2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
/* clang-format off */
#ifndef AV1_COMMON_ODINTRIN_H_
#define AV1_COMMON_ODINTRIN_H_
#include <math.h>
#include <stdlib.h>
#include <string.h>
#include "aom/aom_integer.h"
#include "aom_dsp/aom_dsp_common.h"
#include "aom_ports/bitops.h"
#include "av1/common/enums.h"
#ifdef __cplusplus
extern "C" {
#endif
# if !defined(M_PI)
# define M_PI (3.1415926535897932384626433832795)
# endif
# if !defined(M_SQRT2)
# define M_SQRT2 (1.41421356237309504880168872420970)
# endif
# if !defined(M_SQRT1_2)
# define M_SQRT1_2 (0.70710678118654752440084436210485)
# endif
# if !defined(M_LOG2E)
# define M_LOG2E (1.4426950408889634073599246810019)
# endif
# if !defined(M_LN2)
# define M_LN2 (0.69314718055994530941723212145818)
# endif
/*Smallest blocks are 4x4*/
#define OD_LOG_BSIZE0 (2)
/*There are 5 block sizes total (4x4, 8x8, 16x16, 32x32 and 64x64).*/
#define OD_NBSIZES (5)
/*There are 4 transform sizes total in AV1 (4x4, 8x8, 16x16 and 32x32).*/
#define OD_TXSIZES TX_SIZES
/*The log of the maximum length of the side of a transform.*/
#define OD_LOG_TXSIZE_MAX (OD_LOG_BSIZE0 + OD_TXSIZES - 1)
/*The maximum length of the side of a transform.*/
#define OD_TXSIZE_MAX (1 << OD_LOG_TXSIZE_MAX)
/**The maximum number of color planes allowed in a single frame.*/
# define OD_NPLANES_MAX (3)
# define OD_COEFF_SHIFT (4)
# define OD_DISABLE_CFL (1)
# define OD_DISABLE_FILTER (1)
#if !defined(NDEBUG)
# define OD_ENABLE_ASSERTIONS (1)
#endif
# define OD_LOG(a)
# define OD_LOG_PARTIAL(a)
/*Possible block sizes, note that OD_BLOCK_NXN = log2(N) - 2.*/
#define OD_BLOCK_4X4 (0)
#define OD_BLOCK_8X8 (1)
#define OD_BLOCK_16X16 (2)
#define OD_BLOCK_32X32 (3)
#define OD_BLOCK_SIZES (OD_BLOCK_32X32 + 1)
# define OD_LIMIT_BSIZE_MIN (OD_BLOCK_4X4)
# define OD_LIMIT_BSIZE_MAX (OD_BLOCK_32X32)
typedef int od_coeff;
#define OD_DIVU_DMAX (1024)
extern uint32_t OD_DIVU_SMALL_CONSTS[OD_DIVU_DMAX][2];
/*Translate unsigned division by small divisors into multiplications.*/
#define OD_DIVU_SMALL(_x, _d) \
((uint32_t)((OD_DIVU_SMALL_CONSTS[(_d)-1][0] * (uint64_t)(_x) + \
OD_DIVU_SMALL_CONSTS[(_d)-1][1]) >> \
32) >> \
(OD_ILOG_NZ(_d) - 1))
#define OD_DIVU(_x, _d) \
(((_d) < OD_DIVU_DMAX) ? (OD_DIVU_SMALL((_x), (_d))) : ((_x) / (_d)))
#define OD_MINI AOMMIN
#define OD_MAXI AOMMAX
#define OD_CLAMPI(min, val, max) (OD_MAXI(min, OD_MINI(val, max)))
#define OD_CLZ0 (1)
#define OD_CLZ(x) (-get_msb(x))
#define OD_ILOG_NZ(x) (OD_CLZ0 - OD_CLZ(x))
/*Note that __builtin_clz is not defined when x == 0, according to the gcc
documentation (and that of the x86 BSR instruction that implements it), so
we have to special-case it.
We define a special version of the macro to use when x can be zero.*/
#define OD_ILOG(x) ((x) ? OD_ILOG_NZ(x) : 0)
#define OD_LOG2(x) (M_LOG2E*log(x))
#define OD_EXP2(x) (exp(M_LN2*(x)))
/*Enable special features for gcc and compatible compilers.*/
#if defined(__GNUC__) && defined(__GNUC_MINOR__) && defined(__GNUC_PATCHLEVEL__)
#define OD_GNUC_PREREQ(maj, min, pat) \
((__GNUC__ << 16) + (__GNUC_MINOR__ << 8) + __GNUC_PATCHLEVEL__ >= \
((maj) << 16) + ((min) << 8) + pat) // NOLINT
#else
#define OD_GNUC_PREREQ(maj, min, pat) (0)
#endif
#if OD_GNUC_PREREQ(3, 4, 0)
#define OD_WARN_UNUSED_RESULT __attribute__((__warn_unused_result__))
#else
#define OD_WARN_UNUSED_RESULT
#endif
#if OD_GNUC_PREREQ(3, 4, 0)
#define OD_ARG_NONNULL(x) __attribute__((__nonnull__(x)))
#else
#define OD_ARG_NONNULL(x)
#endif
#if defined(OD_ENABLE_ASSERTIONS)
#if OD_GNUC_PREREQ(2, 5, 0)
__attribute__((noreturn))
#endif
void od_fatal_impl(const char *_str, const char *_file, int _line);
#define OD_FATAL(_str) (od_fatal_impl(_str, __FILE__, __LINE__))
#define OD_ASSERT(_cond) \
do { \
if (!(_cond)) { \
OD_FATAL("assertion failed: " #_cond); \
} \
} while (0)
#define OD_ASSERT2(_cond, _message) \
do { \
if (!(_cond)) { \
OD_FATAL("assertion failed: " #_cond "\n" _message); \
} \
} while (0)
#define OD_ALWAYS_TRUE(_cond) OD_ASSERT(_cond)
#else
#define OD_ASSERT(_cond)
#define OD_ASSERT2(_cond, _message)
#define OD_ALWAYS_TRUE(_cond) ((void)(_cond))
#endif
/** Copy n elements of memory from src to dst. The 0* term provides
compile-time type checking */
#if !defined(OVERRIDE_OD_COPY)
#define OD_COPY(dst, src, n) \
(memcpy((dst), (src), sizeof(*(dst)) * (n) + 0 * ((dst) - (src))))
#endif
/** Copy n elements of memory from src to dst, allowing overlapping regions.
The 0* term provides compile-time type checking */
#if !defined(OVERRIDE_OD_MOVE)
# define OD_MOVE(dst, src, n) \
(memmove((dst), (src), sizeof(*(dst))*(n) + 0*((dst) - (src)) ))
#endif
/** Linkage will break without this if using a C++ compiler, and will issue
* warnings without this for a C compiler*/
#if defined(__cplusplus)
# define OD_EXTERN extern
#else
# define OD_EXTERN
#endif
/** Set n elements of dst to zero */
#if !defined(OVERRIDE_OD_CLEAR)
# define OD_CLEAR(dst, n) (memset((dst), 0, sizeof(*(dst))*(n)))
#endif
/** Silence unused parameter/variable warnings */
# define OD_UNUSED(expr) (void)(expr)
#if defined(OD_FLOAT_PVQ)
typedef double od_val16;
typedef double od_val32;
# define OD_QCONST32(x, bits) (x)
# define OD_ROUND16(x) (x)
# define OD_ROUND32(x) (x)
# define OD_SHL(x, shift) (x)
# define OD_SHR(x, shift) (x)
# define OD_SHR_ROUND(x, shift) (x)
# define OD_ABS(x) (fabs(x))
# define OD_MULT16_16(a, b) ((a)*(b))
# define OD_MULT16_32_Q16(a, b) ((a)*(b))
#else
typedef int16_t od_val16;
typedef int32_t od_val32;
/** Compile-time conversion of float constant to 32-bit value */
# define OD_QCONST32(x, bits) ((od_val32)(.5 + (x)*(((od_val32)1) << (bits))))
# define OD_ROUND16(x) (int16_t)(floor(.5 + (x)))
# define OD_ROUND32(x) (int32_t)(floor(.5 + (x)))
/*Shift x left by shift*/
# define OD_SHL(a, shift) ((int32_t)((uint32_t)(a) << (shift)))
/*Shift x right by shift (without rounding)*/
# define OD_SHR(x, shift) \
((int32_t)((x) >> (shift)))
/*Shift x right by shift (with rounding)*/
# define OD_SHR_ROUND(x, shift) \
((int32_t)(((x) + (1 << (shift) >> 1)) >> (shift)))
/*Shift x right by shift (without rounding) or left by -shift if shift
is negative.*/
# define OD_VSHR(x, shift) \
(((shift) > 0) ? OD_SHR(x, shift) : OD_SHL(x, -(shift)))
/*Shift x right by shift (with rounding) or left by -shift if shift
is negative.*/
# define OD_VSHR_ROUND(x, shift) \
(((shift) > 0) ? OD_SHR_ROUND(x, shift) : OD_SHL(x, -(shift)))
# define OD_ABS(x) (abs(x))
/* (od_val32)(od_val16) gives TI compiler a hint that it's 16x16->32 multiply */
/** 16x16 multiplication where the result fits in 32 bits */
# define OD_MULT16_16(a, b) \
(((od_val32)(od_val16)(a))*((od_val32)(od_val16)(b)))
/* Multiplies 16-bit a by 32-bit b and keeps bits [16:47]. */
# define OD_MULT16_32_Q16(a, b) ((int16_t)(a)*(int64_t)(int32_t)(b) >> 16)
/*16x16 multiplication where the result fits in 16 bits, without rounding.*/
# define OD_MULT16_16_Q15(a, b) \
(((int16_t)(a)*((int32_t)(int16_t)(b))) >> 15)
/*16x16 multiplication where the result fits in 16 bits, without rounding.*/
# define OD_MULT16_16_Q16(a, b) \
((((int16_t)(a))*((int32_t)(int16_t)(b))) >> 16)
#endif
/*All of these macros should expect floats as arguments.*/
/*These two should compile as a single SSE instruction.*/
# define OD_MINF(a, b) ((a) < (b) ? (a) : (b))
# define OD_MAXF(a, b) ((a) > (b) ? (a) : (b))
# define OD_DIV_R0(x, y) (((x) + OD_FLIPSIGNI((((y) + 1) >> 1) - 1, (x)))/(y))
# define OD_SIGNMASK(a) (-((a) < 0))
# define OD_FLIPSIGNI(a, b) (((a) + OD_SIGNMASK(b)) ^ OD_SIGNMASK(b))
# define OD_MULT16_16_Q15(a, b) \
(((int16_t)(a)*((int32_t)(int16_t)(b))) >> 15)
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_ODINTRIN_H_

1027
third_party/aom/av1/common/onyxc_int.h vendored Normal file
View file

File diff suppressed because it is too large Load diff

256
third_party/aom/av1/common/partition.c vendored Normal file
View file

@ -0,0 +1,256 @@
/*
* Copyright (c) 2001-2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
/* clang-format off */
#ifdef HAVE_CONFIG_H
# include "config.h"
#endif
#include "enums.h"
#include "odintrin.h"
#include "partition.h"
#include "zigzag.h"
OD_EXTERN const index_pair *OD_ZIGZAG4[4] = {
OD_ZIGZAG4_DCT_DCT,
OD_ZIGZAG4_ADST_DCT,
OD_ZIGZAG4_DCT_ADST,
OD_ZIGZAG4_ADST_ADST
};
OD_EXTERN const index_pair *OD_ZIGZAG8[4] = {
OD_ZIGZAG8_DCT_DCT,
OD_ZIGZAG8_ADST_DCT,
OD_ZIGZAG8_DCT_ADST,
OD_ZIGZAG8_ADST_ADST
};
OD_EXTERN const index_pair *OD_ZIGZAG16[4] = {
OD_ZIGZAG16_DCT_DCT,
OD_ZIGZAG16_ADST_DCT,
OD_ZIGZAG16_DCT_ADST,
OD_ZIGZAG16_ADST_ADST
};
OD_EXTERN const index_pair *OD_ZIGZAG32[4] = {
OD_ZIGZAG32_DCT_DCT,
OD_ZIGZAG32_DCT_DCT,
OD_ZIGZAG32_DCT_DCT,
OD_ZIGZAG32_DCT_DCT
};
/* The tables below specify how coefficient blocks are translated to
and from PVQ partition coding scan order for 4x4, 8x8 and 16x16 */
static const int OD_LAYOUT32_OFFSETS[4] = { 0, 128, 256, 768 };
const band_layout OD_LAYOUT32 = {
OD_ZIGZAG32,
32,
3,
OD_LAYOUT32_OFFSETS
};
static const int OD_LAYOUT16_OFFSETS[4] = { 0, 32, 64, 192 };
const band_layout OD_LAYOUT16 = {
OD_ZIGZAG16,
16,
3,
OD_LAYOUT16_OFFSETS
};
const int OD_LAYOUT8_OFFSETS[4] = { 0, 8, 16, 48 };
const band_layout OD_LAYOUT8 = {
OD_ZIGZAG8,
8,
3,
OD_LAYOUT8_OFFSETS
};
static const int OD_LAYOUT4_OFFSETS[2] = { 0, 15 };
const band_layout OD_LAYOUT4 = {
OD_ZIGZAG4,
4,
1,
OD_LAYOUT4_OFFSETS
};
/* First element is the number of bands, followed by the list all the band
boundaries. */
static const int OD_BAND_OFFSETS4[] = {1, 1, 16};
static const int OD_BAND_OFFSETS8[] = {4, 1, 16, 24, 32, 64};
static const int OD_BAND_OFFSETS16[] = {7, 1, 16, 24, 32, 64, 96, 128, 256};
static const int OD_BAND_OFFSETS32[] = {10, 1, 16, 24, 32, 64, 96, 128, 256,
384, 512, 1024};
static const int OD_BAND_OFFSETS64[] = {13, 1, 16, 24, 32, 64, 96, 128, 256,
384, 512, 1024, 1536, 2048, 4096};
const int *const OD_BAND_OFFSETS[OD_TXSIZES + 1] = {
OD_BAND_OFFSETS4,
OD_BAND_OFFSETS8,
OD_BAND_OFFSETS16,
OD_BAND_OFFSETS32,
OD_BAND_OFFSETS64
};
/** Perform a single stage of conversion from a coefficient block in
* raster order into coding scan order
*
* @param [in] layout scan order specification
* @param [out] dst destination vector
* @param [in] src source coefficient block
* @param [int] int source vector row stride
*/
static void od_band_from_raster(const band_layout *layout, tran_low_t *dst,
const tran_low_t *src, int stride, TX_TYPE tx_type) {
int i;
int len;
len = layout->band_offsets[layout->nb_bands];
for (i = 0; i < len; i++) {
dst[i] = src[layout->dst_table[tx_type][i][1]*stride + layout->dst_table[tx_type][i][0]];
}
}
/** Perform a single stage of conversion from a vector in coding scan
order back into a coefficient block in raster order
*
* @param [in] layout scan order specification
* @param [out] dst destination coefficient block
* @param [in] src source vector
* @param [int] stride destination vector row stride
*/
static void od_raster_from_band(const band_layout *layout, tran_low_t *dst,
int stride, TX_TYPE tx_type, const tran_low_t *src) {
int i;
int len;
len = layout->band_offsets[layout->nb_bands];
for (i = 0; i < len; i++) {
dst[layout->dst_table[tx_type][i][1]*stride + layout->dst_table[tx_type][i][0]] = src[i];
}
}
static const band_layout *const OD_LAYOUTS[] = {&OD_LAYOUT4, &OD_LAYOUT8,
&OD_LAYOUT16, &OD_LAYOUT32};
/** Converts a coefficient block in raster order into a vector in
* coding scan order with the PVQ partitions laid out one after
* another. This works in stages; the 4x4 conversion is applied to
* the coefficients nearest DC, then the 8x8 applied to the 8x8 block
* nearest DC that was not already coded by 4x4, then 16x16 following
* the same pattern.
*
* @param [out] dst destination vector
* @param [in] n block size (along one side)
* @param [in] ty_type transfrom type
* @param [in] src source coefficient block
* @param [in] stride source vector row stride
*/
void od_raster_to_coding_order(tran_low_t *dst, int n, TX_TYPE ty_type,
const tran_low_t *src, int stride) {
int bs;
/* dst + 1 because DC is not included for 4x4 blocks. */
od_band_from_raster(OD_LAYOUTS[0], dst + 1, src, stride, ty_type);
for (bs = 1; bs < OD_TXSIZES; bs++) {
int size;
int offset;
/* Length of block size > 4. */
size = 1 << (OD_LOG_BSIZE0 + bs);
/* Offset is the size of the previous block squared. */
offset = 1 << 2*(OD_LOG_BSIZE0 - 1 + bs);
if (n >= size) {
/* 3 16x16 bands come after 3 8x8 bands, which come after 2 4x4 bands. */
od_band_from_raster(OD_LAYOUTS[bs], dst + offset, src, stride, ty_type);
}
}
dst[0] = src[0];
}
/** Converts a vector in coding scan order witht he PVQ partitions
* laid out one after another into a coefficient block in raster
* order. This works in stages in the reverse order of raster->scan
* order; the 16x16 conversion is applied to the coefficients that
* don't appear in an 8x8 block, then the 8x8 applied to the 8x8 block
* sans the 4x4 block it contains, then 4x4 is converted sans DC.
*
* @param [out] dst destination coefficient block
* @param [in] stride destination vector row stride
* @param [in] src source vector
* @param [in] n block size (along one side)
*/
void od_coding_order_to_raster(tran_low_t *dst, int stride, TX_TYPE ty_type,
const tran_low_t *src, int n) {
int bs;
/* src + 1 because DC is not included for 4x4 blocks. */
od_raster_from_band(OD_LAYOUTS[0], dst, stride, ty_type, src + 1);
for (bs = 1; bs < OD_TXSIZES; bs++) {
int size;
int offset;
/* Length of block size > 4 */
size = 1 << (OD_LOG_BSIZE0 + bs);
/* Offset is the size of the previous block squared. */
offset = 1 << 2*(OD_LOG_BSIZE0 - 1 + bs);
if (n >= size) {
/* 3 16x16 bands come after 3 8x8 bands, which come after 2 4x4 bands. */
od_raster_from_band(OD_LAYOUTS[bs], dst, stride, ty_type, src + offset);
}
}
dst[0] = src[0];
}
/** Perform a single stage of conversion from a coefficient block in
* raster order into coding scan order
*
* @param [in] layout scan order specification
* @param [out] dst destination vector
* @param [in] src source coefficient block
* @param [int] int source vector row stride
*/
static void od_band_from_raster_16(const band_layout *layout, int16_t *dst,
const int16_t *src, int stride) {
int i;
int len;
len = layout->band_offsets[layout->nb_bands];
for (i = 0; i < len; i++) {
dst[i] = src[layout->dst_table[DCT_DCT][i][1]*stride + layout->dst_table[DCT_DCT][i][0]];
}
}
/** Converts a coefficient block in raster order into a vector in
* coding scan order with the PVQ partitions laid out one after
* another. This works in stages; the 4x4 conversion is applied to
* the coefficients nearest DC, then the 8x8 applied to the 8x8 block
* nearest DC that was not already coded by 4x4, then 16x16 following
* the same pattern.
*
* @param [out] dst destination vector
* @param [in] n block size (along one side)
* @param [in] src source coefficient block
* @param [in] stride source vector row stride
*/
void od_raster_to_coding_order_16(int16_t *dst, int n, const int16_t *src,
int stride) {
int bs;
/* dst + 1 because DC is not included for 4x4 blocks. */
od_band_from_raster_16(OD_LAYOUTS[0], dst + 1, src, stride);
for (bs = 1; bs < OD_TXSIZES; bs++) {
int size;
int offset;
/* Length of block size > 4. */
size = 1 << (OD_LOG_BSIZE0 + bs);
/* Offset is the size of the previous block squared. */
offset = 1 << 2*(OD_LOG_BSIZE0 - 1 + bs);
if (n >= size) {
/* 3 16x16 bands come after 3 8x8 bands, which come after 2 4x4 bands. */
od_band_from_raster_16(OD_LAYOUTS[bs], dst + offset, src, stride);
}
}
dst[0] = src[0];
}

40
third_party/aom/av1/common/partition.h vendored Normal file
View file

@ -0,0 +1,40 @@
/*
* Copyright (c) 2001-2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
/* clang-format off */
#if !defined(_partition_H)
# define _partition_H
#include "av1/common/enums.h"
#include "odintrin.h"
typedef unsigned char index_pair[2];
typedef struct {
const index_pair **const dst_table;
int size;
int nb_bands;
const int *const band_offsets;
} band_layout;
extern const int *const OD_BAND_OFFSETS[OD_TXSIZES + 1];
void od_raster_to_coding_order(tran_low_t *dst, int n, TX_TYPE ty_type,
const tran_low_t *src, int stride);
void od_coding_order_to_raster(tran_low_t *dst, int stride, TX_TYPE ty_type,
const tran_low_t *src, int n);
void od_raster_to_coding_order_16(int16_t *dst, int n, const int16_t *src,
int stride);
#endif

1408
third_party/aom/av1/common/pred_common.c vendored Normal file
View file

File diff suppressed because it is too large Load diff

266
third_party/aom/av1/common/pred_common.h vendored Normal file
View file

@ -0,0 +1,266 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_PRED_COMMON_H_
#define AV1_COMMON_PRED_COMMON_H_
#include "av1/common/blockd.h"
#include "av1/common/onyxc_int.h"
#include "aom_dsp/aom_dsp_common.h"
#ifdef __cplusplus
extern "C" {
#endif
static INLINE int get_segment_id(const AV1_COMMON *const cm,
const uint8_t *segment_ids, BLOCK_SIZE bsize,
int mi_row, int mi_col) {
const int mi_offset = mi_row * cm->mi_cols + mi_col;
const int bw = mi_size_wide[bsize];
const int bh = mi_size_high[bsize];
const int xmis = AOMMIN(cm->mi_cols - mi_col, bw);
const int ymis = AOMMIN(cm->mi_rows - mi_row, bh);
int x, y, segment_id = MAX_SEGMENTS;
for (y = 0; y < ymis; ++y)
for (x = 0; x < xmis; ++x)
segment_id =
AOMMIN(segment_id, segment_ids[mi_offset + y * cm->mi_cols + x]);
assert(segment_id >= 0 && segment_id < MAX_SEGMENTS);
return segment_id;
}
static INLINE int av1_get_pred_context_seg_id(const MACROBLOCKD *xd) {
const MODE_INFO *const above_mi = xd->above_mi;
const MODE_INFO *const left_mi = xd->left_mi;
const int above_sip =
(above_mi != NULL) ? above_mi->mbmi.seg_id_predicted : 0;
const int left_sip = (left_mi != NULL) ? left_mi->mbmi.seg_id_predicted : 0;
return above_sip + left_sip;
}
static INLINE aom_prob av1_get_pred_prob_seg_id(
const struct segmentation_probs *segp, const MACROBLOCKD *xd) {
return segp->pred_probs[av1_get_pred_context_seg_id(xd)];
}
static INLINE int av1_get_skip_context(const MACROBLOCKD *xd) {
const MODE_INFO *const above_mi = xd->above_mi;
const MODE_INFO *const left_mi = xd->left_mi;
const int above_skip = (above_mi != NULL) ? above_mi->mbmi.skip : 0;
const int left_skip = (left_mi != NULL) ? left_mi->mbmi.skip : 0;
return above_skip + left_skip;
}
static INLINE aom_prob av1_get_skip_prob(const AV1_COMMON *cm,
const MACROBLOCKD *xd) {
return cm->fc->skip_probs[av1_get_skip_context(xd)];
}
#if CONFIG_DUAL_FILTER
int av1_get_pred_context_switchable_interp(const MACROBLOCKD *xd, int dir);
#else
int av1_get_pred_context_switchable_interp(const MACROBLOCKD *xd);
#endif
#if CONFIG_EXT_INTRA
#if CONFIG_INTRA_INTERP
int av1_get_pred_context_intra_interp(const MACROBLOCKD *xd);
#endif // CONFIG_INTRA_INTERP
#endif // CONFIG_EXT_INTRA
int av1_get_intra_inter_context(const MACROBLOCKD *xd);
static INLINE aom_prob av1_get_intra_inter_prob(const AV1_COMMON *cm,
const MACROBLOCKD *xd) {
return cm->fc->intra_inter_prob[av1_get_intra_inter_context(xd)];
}
int av1_get_reference_mode_context(const AV1_COMMON *cm, const MACROBLOCKD *xd);
static INLINE aom_prob av1_get_reference_mode_prob(const AV1_COMMON *cm,
const MACROBLOCKD *xd) {
return cm->fc->comp_inter_prob[av1_get_reference_mode_context(cm, xd)];
}
int av1_get_pred_context_comp_ref_p(const AV1_COMMON *cm,
const MACROBLOCKD *xd);
static INLINE aom_prob av1_get_pred_prob_comp_ref_p(const AV1_COMMON *cm,
const MACROBLOCKD *xd) {
const int pred_context = av1_get_pred_context_comp_ref_p(cm, xd);
return cm->fc->comp_ref_prob[pred_context][0];
}
#if CONFIG_EXT_REFS
int av1_get_pred_context_comp_ref_p1(const AV1_COMMON *cm,
const MACROBLOCKD *xd);
static INLINE aom_prob av1_get_pred_prob_comp_ref_p1(const AV1_COMMON *cm,
const MACROBLOCKD *xd) {
const int pred_context = av1_get_pred_context_comp_ref_p1(cm, xd);
return cm->fc->comp_ref_prob[pred_context][1];
}
int av1_get_pred_context_comp_ref_p2(const AV1_COMMON *cm,
const MACROBLOCKD *xd);
static INLINE aom_prob av1_get_pred_prob_comp_ref_p2(const AV1_COMMON *cm,
const MACROBLOCKD *xd) {
const int pred_context = av1_get_pred_context_comp_ref_p2(cm, xd);
return cm->fc->comp_ref_prob[pred_context][2];
}
int av1_get_pred_context_comp_bwdref_p(const AV1_COMMON *cm,
const MACROBLOCKD *xd);
static INLINE aom_prob av1_get_pred_prob_comp_bwdref_p(const AV1_COMMON *cm,
const MACROBLOCKD *xd) {
const int pred_context = av1_get_pred_context_comp_bwdref_p(cm, xd);
return cm->fc->comp_bwdref_prob[pred_context][0];
}
#endif // CONFIG_EXT_REFS
int av1_get_pred_context_single_ref_p1(const MACROBLOCKD *xd);
static INLINE aom_prob av1_get_pred_prob_single_ref_p1(const AV1_COMMON *cm,
const MACROBLOCKD *xd) {
return cm->fc->single_ref_prob[av1_get_pred_context_single_ref_p1(xd)][0];
}
int av1_get_pred_context_single_ref_p2(const MACROBLOCKD *xd);
static INLINE aom_prob av1_get_pred_prob_single_ref_p2(const AV1_COMMON *cm,
const MACROBLOCKD *xd) {
return cm->fc->single_ref_prob[av1_get_pred_context_single_ref_p2(xd)][1];
}
#if CONFIG_EXT_REFS
int av1_get_pred_context_single_ref_p3(const MACROBLOCKD *xd);
static INLINE aom_prob av1_get_pred_prob_single_ref_p3(const AV1_COMMON *cm,
const MACROBLOCKD *xd) {
return cm->fc->single_ref_prob[av1_get_pred_context_single_ref_p3(xd)][2];
}
int av1_get_pred_context_single_ref_p4(const MACROBLOCKD *xd);
static INLINE aom_prob av1_get_pred_prob_single_ref_p4(const AV1_COMMON *cm,
const MACROBLOCKD *xd) {
return cm->fc->single_ref_prob[av1_get_pred_context_single_ref_p4(xd)][3];
}
int av1_get_pred_context_single_ref_p5(const MACROBLOCKD *xd);
static INLINE aom_prob av1_get_pred_prob_single_ref_p5(const AV1_COMMON *cm,
const MACROBLOCKD *xd) {
return cm->fc->single_ref_prob[av1_get_pred_context_single_ref_p5(xd)][4];
}
#endif // CONFIG_EXT_REFS
#if CONFIG_EXT_INTER && CONFIG_COMPOUND_SINGLEREF
int av1_get_inter_mode_context(const MACROBLOCKD *xd);
static INLINE aom_prob av1_get_inter_mode_prob(const AV1_COMMON *cm,
const MACROBLOCKD *xd) {
return cm->fc->comp_inter_mode_prob[av1_get_inter_mode_context(xd)];
}
#endif // CONFIG_EXT_INTER && CONFIG_COMPOUND_SINGLEREF
// Returns a context number for the given MB prediction signal
// The mode info data structure has a one element border above and to the
// left of the entries corresponding to real blocks.
// The prediction flags in these dummy entries are initialized to 0.
static INLINE int get_tx_size_context(const MACROBLOCKD *xd) {
const int max_tx_size = max_txsize_lookup[xd->mi[0]->mbmi.sb_type];
const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
const int has_above = xd->up_available;
const int has_left = xd->left_available;
int above_ctx = (has_above && !above_mbmi->skip)
? (int)txsize_sqr_map[above_mbmi->tx_size]
: max_tx_size;
int left_ctx = (has_left && !left_mbmi->skip)
? (int)txsize_sqr_map[left_mbmi->tx_size]
: max_tx_size;
if (!has_left) left_ctx = above_ctx;
if (!has_above) above_ctx = left_ctx;
#if CONFIG_CB4X4
// TODO(jingning): Temporary setup. Will rework this after the cb4x4
// framework is up running.
return (above_ctx + left_ctx) > max_tx_size + 1;
#else
return (above_ctx + left_ctx) > max_tx_size;
#endif
}
#if CONFIG_VAR_TX
static void update_tx_counts(AV1_COMMON *cm, MACROBLOCKD *xd,
MB_MODE_INFO *mbmi, BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, int blk_row, int blk_col,
TX_SIZE max_tx_size, int ctx) {
const struct macroblockd_plane *const pd = &xd->plane[0];
const BLOCK_SIZE bsize = txsize_to_bsize[tx_size];
const int tx_row = blk_row >> (1 - pd->subsampling_y);
const int tx_col = blk_col >> (1 - pd->subsampling_x);
const TX_SIZE plane_tx_size = mbmi->inter_tx_size[tx_row][tx_col];
const int max_blocks_high = max_block_high(xd, plane_bsize, 0);
const int max_blocks_wide = max_block_wide(xd, plane_bsize, 0);
if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
if (tx_size == plane_tx_size) {
++xd->counts->tx_size[max_tx_size - TX_8X8][ctx][tx_size];
mbmi->tx_size = tx_size;
} else {
int bsl = b_width_log2_lookup[bsize];
int i;
assert(bsl > 0);
--bsl;
for (i = 0; i < 4; ++i) {
const int offsetr = blk_row + ((i >> 1) << bsl);
const int offsetc = blk_col + ((i & 0x01) << bsl);
if (offsetr >= max_blocks_high || offsetc >= max_blocks_wide) continue;
update_tx_counts(cm, xd, mbmi, plane_bsize, (TX_SIZE)(tx_size - 1),
offsetr, offsetc, max_tx_size, ctx);
}
}
}
static INLINE void inter_block_tx_count_update(AV1_COMMON *cm, MACROBLOCKD *xd,
MB_MODE_INFO *mbmi,
BLOCK_SIZE plane_bsize,
int ctx) {
const int mi_width = block_size_wide[plane_bsize] >> tx_size_wide_log2[0];
const int mi_height = block_size_high[plane_bsize] >> tx_size_wide_log2[0];
TX_SIZE max_tx_size = max_txsize_lookup[plane_bsize];
int bh = tx_size_wide_unit[max_tx_size];
int idx, idy;
for (idy = 0; idy < mi_height; idy += bh)
for (idx = 0; idx < mi_width; idx += bh)
update_tx_counts(cm, xd, mbmi, plane_bsize, max_tx_size, idy, idx,
max_tx_size, ctx);
}
#endif
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_PRED_COMMON_H_

1007
third_party/aom/av1/common/pvq.c vendored Normal file
View file

File diff suppressed because it is too large Load diff

183
third_party/aom/av1/common/pvq.h vendored Normal file
View file

@ -0,0 +1,183 @@
/*
* Copyright (c) 2001-2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
/* clang-format off */
#if !defined(_pvq_H)
# define _pvq_H (1)
# include "generic_code.h"
# include "odintrin.h"
extern const uint16_t EXP_CDF_TABLE[][16];
extern const uint16_t LAPLACE_OFFSET[];
#if CONFIG_DAALA_DIST
#define AV1_PVQ_ENABLE_ACTIVITY_MASKING (1)
#else
#define AV1_PVQ_ENABLE_ACTIVITY_MASKING (0)
#endif
# define PVQ_MAX_PARTITIONS (1 + 3*(OD_TXSIZES-1))
# define OD_NOREF_ADAPT_SPEED (4)
/* Normalized lambda for PVQ quantizer. Since we normalize the gain by q, the
distortion is normalized by q^2 and lambda does not need the q^2 factor.
At high rate, this would be log(2)/6, but we're using a slightly more
aggressive value, closer to:
Li, Xiang, et al. "Laplace distribution based Lagrangian rate distortion
optimization for hybrid video coding." Circuits and Systems for Video
Technology, IEEE Transactions on 19.2 (2009): 193-205.
*/
# define OD_PVQ_LAMBDA (.1146)
#define OD_PVQ_SKIP_ZERO 1
#define OD_PVQ_SKIP_COPY 2
/* Maximum size for coding a PVQ band. */
#define OD_MAX_PVQ_SIZE (1024)
#if defined(OD_FLOAT_PVQ)
#define OD_QM_SHIFT (15)
#else
#define OD_QM_SHIFT (11)
#endif
#define OD_QM_SCALE (1 << OD_QM_SHIFT)
#if defined(OD_FLOAT_PVQ)
#define OD_QM_SCALE_1 (1./OD_QM_SCALE)
#endif
#define OD_QM_SCALE_MAX 32767
#define OD_QM_INV_SHIFT (12)
#define OD_QM_INV_SCALE (1 << OD_QM_INV_SHIFT)
#if defined(OD_FLOAT_PVQ)
#define OD_QM_INV_SCALE_1 (1./OD_QM_INV_SCALE)
#endif
#define OD_QM_OFFSET(bs) ((((1 << 2*bs) - 1) << 2*OD_LOG_BSIZE0)/3)
#define OD_QM_STRIDE (OD_QM_OFFSET(OD_TXSIZES))
#define OD_QM_BUFFER_SIZE (2*OD_QM_STRIDE)
#if !defined(OD_FLOAT_PVQ)
#define OD_THETA_SHIFT (15)
#define OD_THETA_SCALE ((1 << OD_THETA_SHIFT)*2./M_PI)
#define OD_MAX_THETA_SCALE (1 << OD_THETA_SHIFT)
#define OD_TRIG_SCALE (32768)
#define OD_BETA_SHIFT (12)
#define OD_BETA_SCALE_1 (1./(1 << OD_BETA_SHIFT))
/*Multiplies 16-bit a by 32-bit b and keeps bits [16:64-OD_BETA_SHIFT-1].*/
#define OD_MULT16_32_QBETA(a, b) \
((int16_t)(a)*(int64_t)(int32_t)(b) >> OD_BETA_SHIFT)
# define OD_MULT16_16_QBETA(a, b) \
((((int16_t)(a))*((int32_t)(int16_t)(b))) >> OD_BETA_SHIFT)
#define OD_CGAIN_SHIFT (8)
#define OD_CGAIN_SCALE (1 << OD_CGAIN_SHIFT)
#else
#define OD_BETA_SCALE_1 (1.)
#define OD_THETA_SCALE (1)
#define OD_TRIG_SCALE (1)
#define OD_CGAIN_SCALE (1)
#endif
#define OD_THETA_SCALE_1 (1./OD_THETA_SCALE)
#define OD_TRIG_SCALE_1 (1./OD_TRIG_SCALE)
#define OD_CGAIN_SCALE_1 (1./OD_CGAIN_SCALE)
#define OD_CGAIN_SCALE_2 (OD_CGAIN_SCALE_1*OD_CGAIN_SCALE_1)
/* Largest PVQ partition is half the coefficients of largest block size. */
#define MAXN (OD_TXSIZE_MAX*OD_TXSIZE_MAX/2)
#define OD_COMPAND_SHIFT (8 + OD_COEFF_SHIFT)
#define OD_COMPAND_SCALE (1 << OD_COMPAND_SHIFT)
#define OD_COMPAND_SCALE_1 (1./OD_COMPAND_SCALE)
#define OD_QM_SIZE (OD_TXSIZES*(OD_TXSIZES + 1))
#define OD_FLAT_QM 0
#define OD_HVS_QM 1
# define OD_NSB_ADAPT_CTXS (4)
# define OD_ADAPT_K_Q8 0
# define OD_ADAPT_SUM_EX_Q8 1
# define OD_ADAPT_COUNT_Q8 2
# define OD_ADAPT_COUNT_EX_Q8 3
# define OD_ADAPT_NO_VALUE (-2147483647-1)
typedef enum {
PVQ_SKIP = 0x0,
DC_CODED = 0x1,
AC_CODED = 0x2,
AC_DC_CODED = 0x3,
} PVQ_SKIP_TYPE;
typedef struct od_pvq_adapt_ctx od_pvq_adapt_ctx;
typedef struct od_pvq_codeword_ctx od_pvq_codeword_ctx;
struct od_pvq_codeword_ctx {
int pvq_adapt[2*OD_TXSIZES*OD_NSB_ADAPT_CTXS];
/* CDFs are size 16 despite the fact that we're using less than that. */
uint16_t pvq_k1_cdf[12][CDF_SIZE(16)];
uint16_t pvq_split_cdf[22*7][CDF_SIZE(8)];
};
struct od_pvq_adapt_ctx {
od_pvq_codeword_ctx pvq_codeword_ctx;
generic_encoder pvq_param_model[3];
int pvq_ext[OD_TXSIZES*PVQ_MAX_PARTITIONS];
int pvq_exg[OD_NPLANES_MAX][OD_TXSIZES][PVQ_MAX_PARTITIONS];
uint16_t pvq_gaintheta_cdf[2*OD_TXSIZES*PVQ_MAX_PARTITIONS][CDF_SIZE(16)];
uint16_t pvq_skip_dir_cdf[2*(OD_TXSIZES-1)][CDF_SIZE(7)];
};
typedef struct od_qm_entry {
int interp_q;
int scale_q8;
const unsigned char *qm_q4;
} od_qm_entry;
extern const od_qm_entry OD_DEFAULT_QMS[2][2][OD_NPLANES_MAX];
void od_adapt_pvq_ctx_reset(od_pvq_adapt_ctx *state, int is_keyframe);
int od_pvq_size_ctx(int n);
int od_pvq_k1_ctx(int n, int orig_size);
od_val16 od_pvq_sin(od_val32 x);
od_val16 od_pvq_cos(od_val32 x);
#if !defined(OD_FLOAT_PVQ)
int od_vector_log_mag(const od_coeff *x, int n);
#endif
void od_interp_qm(unsigned char *out, int q, const od_qm_entry *entry1,
const od_qm_entry *entry2);
int od_qm_get_index(int bs, int band);
extern const od_val16 *const OD_PVQ_BETA[2][OD_NPLANES_MAX][OD_TXSIZES + 1];
void od_init_qm(int16_t *x, int16_t *x_inv, const int *qm);
int od_compute_householder(od_val16 *r, int n, od_val32 gr, int *sign,
int shift);
void od_apply_householder(od_val16 *out, const od_val16 *x, const od_val16 *r,
int n);
void od_pvq_synthesis_partial(od_coeff *xcoeff, const od_coeff *ypulse,
const od_val16 *r, int n,
int noref, od_val32 g,
od_val32 theta, int m, int s,
const int16_t *qm_inv);
od_val32 od_gain_expand(od_val32 cg, int q0, od_val16 beta);
od_val32 od_pvq_compute_gain(const od_val16 *x, int n, int q0, od_val32 *g,
od_val16 beta, int bshift);
int od_pvq_compute_max_theta(od_val32 qcg, od_val16 beta);
od_val32 od_pvq_compute_theta(int t, int max_theta);
int od_pvq_compute_k(od_val32 qcg, int itheta, int noref, int n, od_val16 beta);
int od_vector_is_null(const od_coeff *x, int len);
int od_qm_offset(int bs, int xydec);
#endif

50
third_party/aom/av1/common/pvq_state.c vendored Normal file
View file

@ -0,0 +1,50 @@
/*
* Copyright (c) 2001-2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "av1/common/pvq_state.h"
#include "av1/common/odintrin.h"
void od_adapt_ctx_reset(od_adapt_ctx *adapt, int is_keyframe) {
int pli;
od_adapt_pvq_ctx_reset(&adapt->pvq, is_keyframe);
OD_CDFS_INIT_Q15(adapt->skip_cdf);
for (pli = 0; pli < OD_NPLANES_MAX; pli++) {
int i;
OD_CDFS_INIT_DYNAMIC(adapt->model_dc[pli].cdf);
for (i = 0; i < OD_TXSIZES; i++) {
int j;
adapt->ex_g[pli][i] = 8;
for (j = 0; j < 3; j++) {
adapt->ex_dc[pli][i][j] = pli > 0 ? 8 : 32768;
}
}
}
}
void od_init_skipped_coeffs(int16_t *d, int16_t *pred, int is_keyframe, int bo,
int n, int w) {
int i;
int j;
if (is_keyframe) {
for (i = 0; i < n; i++) {
for (j = 0; j < n; j++) {
/* skip DC */
if (i || j) d[bo + i * w + j] = 0;
}
}
} else {
for (i = 0; i < n; i++) {
for (j = 0; j < n; j++) {
d[bo + i * w + j] = pred[i * n + j];
}
}
}
}

52
third_party/aom/av1/common/pvq_state.h vendored Normal file
View file

@ -0,0 +1,52 @@
/*
* Copyright (c) 2001-2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
/* clang-format off */
#if !defined(_state_H)
# define _state_H (1)
typedef struct od_state od_state;
typedef struct od_adapt_ctx od_adapt_ctx;
# include "generic_code.h"
# include "odintrin.h"
# include "pvq.h"
/*Adaptation speed of scalar Laplace encoding.*/
# define OD_SCALAR_ADAPT_SPEED (4)
struct od_adapt_ctx {
/* Support for PVQ encode/decode */
od_pvq_adapt_ctx pvq;
generic_encoder model_dc[OD_NPLANES_MAX];
int ex_dc[OD_NPLANES_MAX][OD_TXSIZES][3];
int ex_g[OD_NPLANES_MAX][OD_TXSIZES];
/* Joint skip flag for DC and AC */
uint16_t skip_cdf[OD_TXSIZES*2][CDF_SIZE(4)];
};
struct od_state {
od_adapt_ctx *adapt;
unsigned char pvq_qm_q4[OD_NPLANES_MAX][OD_QM_SIZE];
/* Quantization matrices and their inverses. */
int16_t qm[OD_QM_BUFFER_SIZE];
int16_t qm_inv[OD_QM_BUFFER_SIZE];
};
void od_adapt_ctx_reset(od_adapt_ctx *state, int is_keyframe);
void od_init_skipped_coeffs(int16_t *d, int16_t *pred, int is_keyframe,
int bo, int n, int w);
#endif

11369
third_party/aom/av1/common/quant_common.c vendored Normal file
View file

File diff suppressed because it is too large Load diff

111
third_party/aom/av1/common/quant_common.h vendored Normal file
View file

@ -0,0 +1,111 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_QUANT_COMMON_H_
#define AV1_COMMON_QUANT_COMMON_H_
#include "aom/aom_codec.h"
#include "av1/common/seg_common.h"
#include "av1/common/enums.h"
#include "av1/common/entropy.h"
#ifdef __cplusplus
extern "C" {
#endif
#define MINQ 0
#define MAXQ 255
#define QINDEX_RANGE (MAXQ - MINQ + 1)
#define QINDEX_BITS 8
#if CONFIG_AOM_QM
// Total number of QM sets stored
#define QM_LEVEL_BITS 4
#define NUM_QM_LEVELS (1 << QM_LEVEL_BITS)
/* Offset into the list of QMs. Actual number of levels used is
(NUM_QM_LEVELS-AOM_QM_OFFSET)
Lower value of AOM_QM_OFFSET implies more heavily weighted matrices.*/
#define DEFAULT_QM_FIRST (NUM_QM_LEVELS / 2)
#define DEFAULT_QM_LAST (NUM_QM_LEVELS - 1)
#endif
struct AV1Common;
int16_t av1_dc_quant(int qindex, int delta, aom_bit_depth_t bit_depth);
int16_t av1_ac_quant(int qindex, int delta, aom_bit_depth_t bit_depth);
int16_t av1_qindex_from_ac(int ac, aom_bit_depth_t bit_depth);
int av1_get_qindex(const struct segmentation *seg, int segment_id,
int base_qindex);
#if CONFIG_AOM_QM
// Reduce the large number of quantizers to a smaller number of levels for which
// different matrices may be defined
static INLINE int aom_get_qmlevel(int qindex, int first, int last) {
int qmlevel = (qindex * (last + 1 - first) + QINDEX_RANGE / 2) / QINDEX_RANGE;
qmlevel = AOMMIN(qmlevel + first, NUM_QM_LEVELS - 1);
return qmlevel;
}
void aom_qm_init(struct AV1Common *cm);
qm_val_t *aom_iqmatrix(struct AV1Common *cm, int qindex, int comp,
int log2sizem2, int is_intra);
qm_val_t *aom_qmatrix(struct AV1Common *cm, int qindex, int comp,
int log2sizem2, int is_intra);
#endif
#if CONFIG_NEW_QUANT
#define QUANT_PROFILES 4
#define QUANT_RANGES 2
#define NUQ_KNOTS 3
typedef tran_low_t dequant_val_type_nuq[NUQ_KNOTS + 1];
typedef tran_low_t cuml_bins_type_nuq[NUQ_KNOTS];
void av1_get_dequant_val_nuq(int q, int band, tran_low_t *dq,
tran_low_t *cuml_bins, int dq_off_index);
tran_low_t av1_dequant_abscoeff_nuq(int v, int q, const tran_low_t *dq);
tran_low_t av1_dequant_coeff_nuq(int v, int q, const tran_low_t *dq);
static INLINE int qindex_to_qrange(int qindex) {
return (qindex < 140 ? 1 : 0);
}
static INLINE int get_dq_profile_from_ctx(int qindex, int q_ctx, int is_inter,
PLANE_TYPE plane_type) {
// intra/inter, Y/UV, ctx, qrange
static const int
def_dq_profile_lookup[REF_TYPES][PLANE_TYPES][COEFF_CONTEXTS0]
[QUANT_RANGES] = {
{
// intra
{ { 2, 1 }, { 2, 1 }, { 2, 1 } }, // Y
{ { 3, 1 }, { 3, 1 }, { 3, 1 } }, // UV
},
{
// inter
{ { 3, 1 }, { 2, 1 }, { 2, 1 } }, // Y
{ { 3, 1 }, { 3, 1 }, { 3, 1 } }, // UV
},
};
if (!qindex) return 0; // lossless
return def_dq_profile_lookup[is_inter][plane_type][q_ctx]
[qindex_to_qrange(qindex)];
}
#endif // CONFIG_NEW_QUANT
#if CONFIG_PVQ || CONFIG_DAALA_DIST
extern const int OD_QM8_Q4_FLAT[];
extern const int OD_QM8_Q4_HVS[];
#endif
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_QUANT_COMMON_H_

3083
third_party/aom/av1/common/reconinter.c vendored Normal file
View file

File diff suppressed because it is too large Load diff

828
third_party/aom/av1/common/reconinter.h vendored Normal file
View file

@ -0,0 +1,828 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_RECONINTER_H_
#define AV1_COMMON_RECONINTER_H_
#include "av1/common/filter.h"
#include "av1/common/onyxc_int.h"
#include "av1/common/convolve.h"
#if CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
#include "av1/common/warped_motion.h"
#endif // CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
#include "aom/aom_integer.h"
#if CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
#define WARP_WM_NEIGHBORS_WITH_OBMC 0
#endif // CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
#if CONFIG_MOTION_VAR && CONFIG_GLOBAL_MOTION
#define WARP_GM_NEIGHBORS_WITH_OBMC 0
#endif // CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
#ifdef __cplusplus
extern "C" {
#endif
static INLINE void inter_predictor(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride,
const int subpel_x, const int subpel_y,
const struct scale_factors *sf, int w, int h,
ConvolveParams *conv_params,
#if CONFIG_DUAL_FILTER
const InterpFilter *interp_filter,
#else
const InterpFilter interp_filter,
#endif
int xs, int ys) {
#if CONFIG_DUAL_FILTER
InterpFilter filter_x = av1_get_plane_interp_filter(
interp_filter[1 + 2 * conv_params->ref], conv_params->plane);
InterpFilter filter_y = av1_get_plane_interp_filter(
interp_filter[0 + 2 * conv_params->ref], conv_params->plane);
InterpFilterParams interp_filter_params_x =
av1_get_interp_filter_params(filter_x);
InterpFilterParams interp_filter_params_y =
av1_get_interp_filter_params(filter_y);
#else
InterpFilterParams interp_filter_params =
av1_get_interp_filter_params(interp_filter);
#endif
assert(sf);
#if CONFIG_DUAL_FILTER
if (interp_filter_params_x.taps == SUBPEL_TAPS &&
interp_filter_params_y.taps == SUBPEL_TAPS && w > 2 && h > 2 &&
conv_params->round == CONVOLVE_OPT_ROUND && xs == 16 && ys == 16) {
const int16_t *kernel_x =
av1_get_interp_filter_subpel_kernel(interp_filter_params_x, subpel_x);
const int16_t *kernel_y =
av1_get_interp_filter_subpel_kernel(interp_filter_params_y, subpel_y);
#else
if (interp_filter_params.taps == SUBPEL_TAPS && w > 2 && h > 2 &&
conv_params->round == CONVOLVE_OPT_ROUND && xs == 16 && ys == 16) {
const int16_t *kernel_x =
av1_get_interp_filter_subpel_kernel(interp_filter_params, subpel_x);
const int16_t *kernel_y =
av1_get_interp_filter_subpel_kernel(interp_filter_params, subpel_y);
#endif
sf->predict[subpel_x != 0][subpel_y != 0][conv_params->ref](
src, src_stride, dst, dst_stride, kernel_x, xs, kernel_y, ys, w, h);
} else {
// ref_idx > 0 means this is the second reference frame
// first reference frame's prediction result is already in dst
// therefore we need to average the first and second results
#if CONFIG_CONVOLVE_ROUND
if (conv_params->round == CONVOLVE_OPT_NO_ROUND && xs == 16 && ys == 16)
av1_convolve_2d_facade(src, src_stride, dst, dst_stride, w, h,
#if CONFIG_DUAL_FILTER
interp_filter,
#else
&interp_filter,
#endif
subpel_x, xs, subpel_y, ys, conv_params);
else
#endif
{
if (xs == 16 && ys == 16) {
av1_convolve(src, src_stride, dst, dst_stride, w, h, interp_filter,
subpel_x, xs, subpel_y, ys, conv_params);
} else {
// If xs == 16 || ys == 16 scaling is happening and the SSE2
// instructions don't support scaling; use the C versions to be safe.
av1_convolve_c(src, src_stride, dst, dst_stride, w, h, interp_filter,
subpel_x, xs, subpel_y, ys, conv_params);
}
}
}
}
#if CONFIG_HIGHBITDEPTH
static INLINE void highbd_inter_predictor(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride,
const int subpel_x,
const int subpel_y,
const struct scale_factors *sf, int w,
int h, int ref,
#if CONFIG_DUAL_FILTER
const InterpFilter *interp_filter,
#else
const InterpFilter interp_filter,
#endif
int xs, int ys, int bd) {
#if CONFIG_DUAL_FILTER
InterpFilterParams interp_filter_params_x =
av1_get_interp_filter_params(interp_filter[1 + 2 * ref]);
InterpFilterParams interp_filter_params_y =
av1_get_interp_filter_params(interp_filter[0 + 2 * ref]);
#else
InterpFilterParams interp_filter_params =
av1_get_interp_filter_params(interp_filter);
#endif
#if CONFIG_DUAL_FILTER
if (interp_filter_params_x.taps == SUBPEL_TAPS &&
interp_filter_params_y.taps == SUBPEL_TAPS && w > 2 && h > 2) {
const int16_t *kernel_x =
av1_get_interp_filter_subpel_kernel(interp_filter_params_x, subpel_x);
const int16_t *kernel_y =
av1_get_interp_filter_subpel_kernel(interp_filter_params_y, subpel_y);
#else
if (interp_filter_params.taps == SUBPEL_TAPS && w > 2 && h > 2) {
const int16_t *kernel_x =
av1_get_interp_filter_subpel_kernel(interp_filter_params, subpel_x);
const int16_t *kernel_y =
av1_get_interp_filter_subpel_kernel(interp_filter_params, subpel_y);
#endif // CONFIG_DUAL_FILTER
sf->highbd_predict[subpel_x != 0][subpel_y != 0][ref](
src, src_stride, dst, dst_stride, kernel_x, xs, kernel_y, ys, w, h, bd);
} else {
// ref > 0 means this is the second reference frame
// first reference frame's prediction result is already in dst
// therefore we need to average the first and second results
int avg = ref > 0;
av1_highbd_convolve(src, src_stride, dst, dst_stride, w, h, interp_filter,
subpel_x, xs, subpel_y, ys, avg, bd);
}
}
#endif // CONFIG_HIGHBITDEPTH
#if CONFIG_EXT_INTER
// Set to (1 << 5) if the 32-ary codebooks are used for any bock size
#define MAX_WEDGE_TYPES (1 << 4)
#define MAX_WEDGE_SIZE_LOG2 5 // 32x32
#define MAX_WEDGE_SIZE (1 << MAX_WEDGE_SIZE_LOG2)
#define MAX_WEDGE_SQUARE (MAX_WEDGE_SIZE * MAX_WEDGE_SIZE)
#define WEDGE_WEIGHT_BITS 6
#define WEDGE_NONE -1
// Angles are with respect to horizontal anti-clockwise
typedef enum {
WEDGE_HORIZONTAL = 0,
WEDGE_VERTICAL = 1,
WEDGE_OBLIQUE27 = 2,
WEDGE_OBLIQUE63 = 3,
WEDGE_OBLIQUE117 = 4,
WEDGE_OBLIQUE153 = 5,
WEDGE_DIRECTIONS
} WedgeDirectionType;
// 3-tuple: {direction, x_offset, y_offset}
typedef struct {
WedgeDirectionType direction;
int x_offset;
int y_offset;
} wedge_code_type;
typedef uint8_t *wedge_masks_type[MAX_WEDGE_TYPES];
typedef struct {
int bits;
const wedge_code_type *codebook;
uint8_t *signflip;
int smoother;
wedge_masks_type *masks;
} wedge_params_type;
extern const wedge_params_type wedge_params_lookup[BLOCK_SIZES];
static INLINE int is_interinter_compound_used(COMPOUND_TYPE type,
BLOCK_SIZE sb_type) {
(void)sb_type;
switch (type) {
case COMPOUND_AVERAGE: return 1;
#if CONFIG_WEDGE
case COMPOUND_WEDGE: return wedge_params_lookup[sb_type].bits > 0;
#endif // CONFIG_WEDGE
#if CONFIG_COMPOUND_SEGMENT
case COMPOUND_SEG: return sb_type >= BLOCK_8X8;
#endif // CONFIG_COMPOUND_SEGMENT
default: assert(0); return 0;
}
}
static INLINE int is_any_masked_compound_used(BLOCK_SIZE sb_type) {
COMPOUND_TYPE comp_type;
for (comp_type = 0; comp_type < COMPOUND_TYPES; comp_type++) {
if (is_masked_compound_type(comp_type) &&
is_interinter_compound_used(comp_type, sb_type))
return 1;
}
return 0;
}
static INLINE int get_wedge_bits_lookup(BLOCK_SIZE sb_type) {
return wedge_params_lookup[sb_type].bits;
}
static INLINE int get_interinter_wedge_bits(BLOCK_SIZE sb_type) {
const int wbits = wedge_params_lookup[sb_type].bits;
return (wbits > 0) ? wbits + 1 : 0;
}
static INLINE int is_interintra_wedge_used(BLOCK_SIZE sb_type) {
(void)sb_type;
return wedge_params_lookup[sb_type].bits > 0;
}
static INLINE int get_interintra_wedge_bits(BLOCK_SIZE sb_type) {
return wedge_params_lookup[sb_type].bits;
}
#if CONFIG_COMPOUND_SEGMENT
void build_compound_seg_mask(uint8_t *mask, SEG_MASK_TYPE mask_type,
const uint8_t *src0, int src0_stride,
const uint8_t *src1, int src1_stride,
BLOCK_SIZE sb_type, int h, int w);
#if CONFIG_HIGHBITDEPTH
void build_compound_seg_mask_highbd(uint8_t *mask, SEG_MASK_TYPE mask_type,
const uint8_t *src0, int src0_stride,
const uint8_t *src1, int src1_stride,
BLOCK_SIZE sb_type, int h, int w, int bd);
#endif // CONFIG_HIGHBITDEPTH
#endif // CONFIG_COMPOUND_SEGMENT
#endif // CONFIG_EXT_INTER
void build_inter_predictors(MACROBLOCKD *xd, int plane,
#if CONFIG_MOTION_VAR
int mi_col_offset, int mi_row_offset,
#endif // CONFIG_MOTION_VAR
int block, int bw, int bh, int x, int y, int w,
int h,
#if CONFIG_SUPERTX && CONFIG_EXT_INTER
int wedge_offset_x, int wedge_offset_y,
#endif // CONFIG_SUPERTX && CONFIG_EXT_INTER
int mi_x, int mi_y);
#if CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
// This function will determine whether or not to create a warped
// prediction and return the appropriate motion model depending
// on the configuration. Behavior will change with different
// combinations of GLOBAL_MOTION, WARPED_MOTION and MOTION_VAR.
static INLINE int allow_warp(const MODE_INFO *const mi,
const WarpTypesAllowed *const warp_types,
#if CONFIG_GLOBAL_MOTION
const WarpedMotionParams *const gm_params,
#endif // CONFIG_GLOBAL_MOTION
#if CONFIG_MOTION_VAR
int mi_col_offset, int mi_row_offset,
#endif // CONFIG_MOTION_VAR
WarpedMotionParams *final_warp_params) {
const MB_MODE_INFO *const mbmi = &mi->mbmi;
set_default_warp_params(final_warp_params);
// Only global motion configured
#if CONFIG_GLOBAL_MOTION && !CONFIG_WARPED_MOTION && !CONFIG_MOTION_VAR
(void)mbmi;
if (warp_types->global_warp_allowed) {
memcpy(final_warp_params, gm_params, sizeof(*final_warp_params));
return 1;
}
#endif // CONFIG_GLOBAL_MOTION && !CONFIG_WARPED_MOTION && !CONFIG_MOTION_VAR
// Only warped motion configured
#if CONFIG_WARPED_MOTION && !CONFIG_GLOBAL_MOTION && !CONFIG_MOTION_VAR
if (warp_types->local_warp_allowed) {
memcpy(final_warp_params, &mbmi->wm_params[0], sizeof(*final_warp_params));
return 1;
}
#endif // CONFIG_WARPED_MOTION && !CONFIG_GLOBAL_MOTION && !CONFIG_MOTION_VAR
// Warped and global motion configured
#if CONFIG_GLOBAL_MOTION && CONFIG_WARPED_MOTION && !CONFIG_MOTION_VAR
// When both are enabled, warped will take priority. The global parameters
// will only be used to compute projection samples to find the warped model.
// Note that, if SEPARATE_GLOBAL_MOTION is enabled and a block chooses
// global, it will not be possible to select WARPED_CAUSAL.
if (warp_types->local_warp_allowed) {
memcpy(final_warp_params, &mbmi->wm_params[0], sizeof(*final_warp_params));
return 1;
} else if (warp_types->global_warp_allowed) {
memcpy(final_warp_params, gm_params, sizeof(*final_warp_params));
return 1;
}
#endif // CONFIG_GLOBAL_MOTION && CONFIG_WARPED_MOTION && !CONFIG_MOTION_VAR
// Motion var and global motion configured
#if CONFIG_GLOBAL_MOTION && CONFIG_MOTION_VAR && !CONFIG_WARPED_MOTION
// We warp if either case is true:
// 1.) We are predicting a block which uses global motion
// 2.) We are predicting a neighboring block of a block using OBMC,
// the neighboring block uses global motion, and we have enabled
// WARP_GM_NEIGHBORS_WITH_OBMC
const int build_for_obmc = !(mi_col_offset == 0 && mi_row_offset == 0);
(void)mbmi;
if (warp_types->global_warp_allowed &&
(WARP_GM_NEIGHBORS_WITH_OBMC || !build_for_obmc)) {
memcpy(final_warp_params, gm_params, sizeof(*final_warp_params));
return 1;
}
#endif // CONFIG_GLOBAL_MOTION && CONFIG_MOTION_VAR && !CONFIG_WARPED_MOTION
// Motion var and warped motion configured
#if CONFIG_WARPED_MOTION && CONFIG_MOTION_VAR && !CONFIG_GLOBAL_MOTION
// We warp if either case is true:
// 1.) We are predicting a block with motion mode WARPED_CAUSAL
// 2.) We are predicting a neighboring block of a block using OBMC,
// the neighboring block has mode WARPED_CAUSAL, and we have enabled
// WARP_WM_NEIGHBORS_WITH_OBMC
const int build_for_obmc = !(mi_col_offset == 0 && mi_row_offset == 0);
if (warp_types->local_warp_allowed) {
if ((build_for_obmc && WARP_WM_NEIGHBORS_WITH_OBMC) || (!build_for_obmc)) {
memcpy(final_warp_params, &mbmi->wm_params[0],
sizeof(*final_warp_params));
return 1;
}
}
#endif // CONFIG_WARPED_MOTION && CONFIG_MOTION_VAR && !CONFIG_GLOBAL_MOTION
// Motion var, warped motion and global motion all configured
#if CONFIG_WARPED_MOTION && CONFIG_MOTION_VAR && CONFIG_GLOBAL_MOTION
const int build_for_obmc = !(mi_col_offset == 0 && mi_row_offset == 0);
if (warp_types->local_warp_allowed) {
if ((build_for_obmc && WARP_WM_NEIGHBORS_WITH_OBMC) || (!build_for_obmc)) {
memcpy(final_warp_params, &mbmi->wm_params[0],
sizeof(*final_warp_params));
return 1;
}
} else if (warp_types->global_warp_allowed &&
(WARP_GM_NEIGHBORS_WITH_OBMC || !build_for_obmc)) {
memcpy(final_warp_params, gm_params, sizeof(*final_warp_params));
return 1;
}
#endif // CONFIG_WARPED_MOTION && CONFIG_MOTION_VAR && CONFIG_GLOBAL_MOTION
return 0;
}
#endif // CONFIG_GLOBAL_MOTION ||CONFIG_WARPED_MOTION
static INLINE void av1_make_inter_predictor(
const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride,
const int subpel_x, const int subpel_y, const struct scale_factors *sf,
int w, int h, ConvolveParams *conv_params,
#if CONFIG_DUAL_FILTER
const InterpFilter *interp_filter,
#else
const InterpFilter interp_filter,
#endif
#if CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
const WarpTypesAllowed *warp_types, int p_col, int p_row, int plane,
int ref,
#endif // CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
#if CONFIG_MOTION_VAR
int mi_col_offset, int mi_row_offset,
#endif
int xs, int ys, const MACROBLOCKD *xd) {
(void)xd;
#if CONFIG_MOTION_VAR
const MODE_INFO *mi = xd->mi[mi_col_offset + xd->mi_stride * mi_row_offset];
#else
const MODE_INFO *mi = xd->mi[0];
(void)mi;
#endif // CONFIG_MOTION_VAR
// Make sure the selected motion mode is valid for this configuration
#if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
assert_motion_mode_valid(mi->mbmi.motion_mode,
#if CONFIG_GLOBAL_MOTION && SEPARATE_GLOBAL_MOTION
0, xd->global_motion,
#endif // CONFIG_GLOBAL_MOTION && SEPARATE_GLOBAL_MOTION
mi);
#endif // CONFIG MOTION_VAR || CONFIG_WARPED_MOTION
#if CONFIG_WARPED_MOTION || CONFIG_GLOBAL_MOTION
WarpedMotionParams final_warp_params;
const int do_warp = allow_warp(mi, warp_types,
#if CONFIG_GLOBAL_MOTION
&xd->global_motion[mi->mbmi.ref_frame[ref]],
#endif // CONFIG_GLOBAL_MOTION
#if CONFIG_MOTION_VAR
mi_col_offset, mi_row_offset,
#endif // CONFIG_MOTION_VAR
&final_warp_params);
if (do_warp) {
const struct macroblockd_plane *const pd = &xd->plane[plane];
const struct buf_2d *const pre_buf = &pd->pre[ref];
av1_warp_plane(&final_warp_params,
#if CONFIG_HIGHBITDEPTH
xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH, xd->bd,
#endif // CONFIG_HIGHBITDEPTH
pre_buf->buf0, pre_buf->width, pre_buf->height,
pre_buf->stride, dst, p_col, p_row, w, h, dst_stride,
pd->subsampling_x, pd->subsampling_y, xs, ys, ref);
return;
}
#endif // CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
#if CONFIG_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
highbd_inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y,
sf, w, h, conv_params->ref, interp_filter, xs, ys,
xd->bd);
return;
}
#endif // CONFIG_HIGHBITDEPTH
inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y, sf, w,
h, conv_params, interp_filter, xs, ys);
}
#if CONFIG_EXT_INTER
void av1_make_masked_inter_predictor(const uint8_t *pre, int pre_stride,
uint8_t *dst, int dst_stride,
const int subpel_x, const int subpel_y,
const struct scale_factors *sf, int w,
int h,
#if CONFIG_DUAL_FILTER
const InterpFilter *interp_filter,
#else
const InterpFilter interp_filter,
#endif
int xs, int ys,
#if CONFIG_SUPERTX
int wedge_offset_x, int wedge_offset_y,
#endif // CONFIG_SUPERTX
int plane,
#if CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
const WarpTypesAllowed *warp_types,
int p_col, int p_row, int ref,
#endif // CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
MACROBLOCKD *xd);
#endif // CONFIG_EXT_INTER
static INLINE int round_mv_comp_q4(int value) {
return (value < 0 ? value - 2 : value + 2) / 4;
}
static MV mi_mv_pred_q4(const MODE_INFO *mi, int idx) {
MV res = {
round_mv_comp_q4(
mi->bmi[0].as_mv[idx].as_mv.row + mi->bmi[1].as_mv[idx].as_mv.row +
mi->bmi[2].as_mv[idx].as_mv.row + mi->bmi[3].as_mv[idx].as_mv.row),
round_mv_comp_q4(
mi->bmi[0].as_mv[idx].as_mv.col + mi->bmi[1].as_mv[idx].as_mv.col +
mi->bmi[2].as_mv[idx].as_mv.col + mi->bmi[3].as_mv[idx].as_mv.col)
};
return res;
}
static INLINE int round_mv_comp_q2(int value) {
return (value < 0 ? value - 1 : value + 1) / 2;
}
static MV mi_mv_pred_q2(const MODE_INFO *mi, int idx, int block0, int block1) {
MV res = { round_mv_comp_q2(mi->bmi[block0].as_mv[idx].as_mv.row +
mi->bmi[block1].as_mv[idx].as_mv.row),
round_mv_comp_q2(mi->bmi[block0].as_mv[idx].as_mv.col +
mi->bmi[block1].as_mv[idx].as_mv.col) };
return res;
}
// TODO(jkoleszar): yet another mv clamping function :-(
static INLINE MV clamp_mv_to_umv_border_sb(const MACROBLOCKD *xd,
const MV *src_mv, int bw, int bh,
int ss_x, int ss_y) {
// If the MV points so far into the UMV border that no visible pixels
// are used for reconstruction, the subpel part of the MV can be
// discarded and the MV limited to 16 pixels with equivalent results.
const int spel_left = (AOM_INTERP_EXTEND + bw) << SUBPEL_BITS;
const int spel_right = spel_left - SUBPEL_SHIFTS;
const int spel_top = (AOM_INTERP_EXTEND + bh) << SUBPEL_BITS;
const int spel_bottom = spel_top - SUBPEL_SHIFTS;
MV clamped_mv = { src_mv->row * (1 << (1 - ss_y)),
src_mv->col * (1 << (1 - ss_x)) };
assert(ss_x <= 1);
assert(ss_y <= 1);
clamp_mv(&clamped_mv, xd->mb_to_left_edge * (1 << (1 - ss_x)) - spel_left,
xd->mb_to_right_edge * (1 << (1 - ss_x)) + spel_right,
xd->mb_to_top_edge * (1 << (1 - ss_y)) - spel_top,
xd->mb_to_bottom_edge * (1 << (1 - ss_y)) + spel_bottom);
return clamped_mv;
}
static INLINE MV average_split_mvs(const struct macroblockd_plane *pd,
const MODE_INFO *mi, int ref, int block) {
const int ss_idx = ((pd->subsampling_x > 0) << 1) | (pd->subsampling_y > 0);
MV res = { 0, 0 };
switch (ss_idx) {
case 0: res = mi->bmi[block].as_mv[ref].as_mv; break;
case 1: res = mi_mv_pred_q2(mi, ref, block, block + 2); break;
case 2: res = mi_mv_pred_q2(mi, ref, block, block + 1); break;
case 3: res = mi_mv_pred_q4(mi, ref); break;
default: assert(ss_idx <= 3 && ss_idx >= 0);
}
return res;
}
void av1_build_inter_predictor_sub8x8(MACROBLOCKD *xd, int plane, int i, int ir,
int ic, int mi_row, int mi_col);
void av1_build_inter_predictors_sby(MACROBLOCKD *xd, int mi_row, int mi_col,
BUFFER_SET *ctx, BLOCK_SIZE bsize);
void av1_build_inter_predictors_sbuv(MACROBLOCKD *xd, int mi_row, int mi_col,
BUFFER_SET *ctx, BLOCK_SIZE bsize);
void av1_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col,
BUFFER_SET *ctx, BLOCK_SIZE bsize);
#if CONFIG_SUPERTX
void av1_build_inter_predictors_sb_sub8x8_extend(MACROBLOCKD *xd,
#if CONFIG_EXT_INTER
int mi_row_ori, int mi_col_ori,
#endif // CONFIG_EXT_INTER
int mi_row, int mi_col,
BLOCK_SIZE bsize, int block);
void av1_build_inter_predictors_sb_extend(MACROBLOCKD *xd,
#if CONFIG_EXT_INTER
int mi_row_ori, int mi_col_ori,
#endif // CONFIG_EXT_INTER
int mi_row, int mi_col,
BLOCK_SIZE bsize);
struct macroblockd_plane;
void av1_build_masked_inter_predictor_complex(
MACROBLOCKD *xd, uint8_t *dst, int dst_stride, const uint8_t *pre,
int pre_stride, int mi_row, int mi_col, int mi_row_ori, int mi_col_ori,
BLOCK_SIZE bsize, BLOCK_SIZE top_bsize, PARTITION_TYPE partition,
int plane);
#endif // CONFIG_SUPERTX
void av1_build_inter_predictor(const uint8_t *src, int src_stride, uint8_t *dst,
int dst_stride, const MV *src_mv,
const struct scale_factors *sf, int w, int h,
ConvolveParams *conv_params,
#if CONFIG_DUAL_FILTER
const InterpFilter *interp_filter,
#else
const InterpFilter interp_filter,
#endif
#if CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
const WarpTypesAllowed *warp_types, int p_col,
int p_row, int plane, int ref,
#endif // CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
enum mv_precision precision, int x, int y,
const MACROBLOCKD *xd);
#if CONFIG_HIGHBITDEPTH
void av1_highbd_build_inter_predictor(
const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride,
const MV *mv_q3, const struct scale_factors *sf, int w, int h, int do_avg,
#if CONFIG_DUAL_FILTER
const InterpFilter *interp_filter,
#else
const InterpFilter interp_filter,
#endif
#if CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
const WarpTypesAllowed *warp_types, int p_col, int p_row,
#endif // CONFIG_GLOBAL_MOTION || CONFIG_WARPED_MOTION
int plane, enum mv_precision precision, int x, int y,
const MACROBLOCKD *xd);
#endif
static INLINE int scaled_buffer_offset(int x_offset, int y_offset, int stride,
const struct scale_factors *sf) {
const int x = sf ? sf->scale_value_x(x_offset, sf) : x_offset;
const int y = sf ? sf->scale_value_y(y_offset, sf) : y_offset;
return y * stride + x;
}
static INLINE void setup_pred_plane(struct buf_2d *dst, BLOCK_SIZE bsize,
uint8_t *src, int width, int height,
int stride, int mi_row, int mi_col,
const struct scale_factors *scale,
int subsampling_x, int subsampling_y) {
#if CONFIG_CHROMA_SUB8X8
if (bsize < BLOCK_8X8) {
// Offset the buffer pointer
if (subsampling_y && (mi_row & 0x01)) mi_row -= 1;
if (subsampling_x && (mi_col & 0x01)) mi_col -= 1;
}
#else
(void)bsize;
#endif
const int x = (MI_SIZE * mi_col) >> subsampling_x;
const int y = (MI_SIZE * mi_row) >> subsampling_y;
dst->buf = src + scaled_buffer_offset(x, y, stride, scale);
dst->buf0 = src;
dst->width = width;
dst->height = height;
dst->stride = stride;
}
void av1_setup_dst_planes(struct macroblockd_plane planes[MAX_MB_PLANE],
BLOCK_SIZE bsize, const YV12_BUFFER_CONFIG *src,
int mi_row, int mi_col);
void av1_setup_pre_planes(MACROBLOCKD *xd, int idx,
const YV12_BUFFER_CONFIG *src, int mi_row, int mi_col,
const struct scale_factors *sf);
// Detect if the block have sub-pixel level motion vectors
// per component.
#define CHECK_SUBPEL 0
static INLINE int has_subpel_mv_component(const MODE_INFO *const mi,
const MACROBLOCKD *const xd,
int dir) {
#if CHECK_SUBPEL
const MB_MODE_INFO *const mbmi = &mi->mbmi;
const BLOCK_SIZE bsize = mbmi->sb_type;
int plane;
int ref = (dir >> 1);
#if CONFIG_CB4X4
const int unify_bsize = 1;
#else
const int unify_bsize = 0;
#endif
if (bsize >= BLOCK_8X8 || unify_bsize) {
if (dir & 0x01) {
if (mbmi->mv[ref].as_mv.col & SUBPEL_MASK) return 1;
} else {
if (mbmi->mv[ref].as_mv.row & SUBPEL_MASK) return 1;
}
} else {
for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
const PARTITION_TYPE bp = BLOCK_8X8 - bsize;
const struct macroblockd_plane *const pd = &xd->plane[plane];
const int have_vsplit = bp != PARTITION_HORZ;
const int have_hsplit = bp != PARTITION_VERT;
const int num_4x4_w = 2 >> ((!have_vsplit) | pd->subsampling_x);
const int num_4x4_h = 2 >> ((!have_hsplit) | pd->subsampling_y);
int x, y;
for (y = 0; y < num_4x4_h; ++y) {
for (x = 0; x < num_4x4_w; ++x) {
const MV mv = average_split_mvs(pd, mi, ref, y * 2 + x);
if (dir & 0x01) {
if (mv.col & SUBPEL_MASK) return 1;
} else {
if (mv.row & SUBPEL_MASK) return 1;
}
}
}
}
}
return 0;
#else
(void)mi;
(void)xd;
(void)dir;
return 1;
#endif
}
static INLINE void set_default_interp_filters(
MB_MODE_INFO *const mbmi, InterpFilter frame_interp_filter) {
#if CONFIG_DUAL_FILTER
int dir;
for (dir = 0; dir < 4; ++dir)
mbmi->interp_filter[dir] = frame_interp_filter == SWITCHABLE
? EIGHTTAP_REGULAR
: frame_interp_filter;
#else
mbmi->interp_filter = frame_interp_filter == SWITCHABLE ? EIGHTTAP_REGULAR
: frame_interp_filter;
#endif // CONFIG_DUAL_FILTER
}
static INLINE int av1_is_interp_needed(const MACROBLOCKD *const xd) {
(void)xd;
#if CONFIG_WARPED_MOTION
const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
if (mbmi->motion_mode == WARPED_CAUSAL) return 0;
#endif // CONFIG_WARPED_MOTION
#if CONFIG_GLOBAL_MOTION
if (is_nontrans_global_motion(xd)) return 0;
#endif // CONFIG_GLOBAL_MOTION
return 1;
}
static INLINE int av1_is_interp_search_needed(const MACROBLOCKD *const xd) {
MODE_INFO *const mi = xd->mi[0];
const int is_compound = has_second_ref(&mi->mbmi);
int ref;
for (ref = 0; ref < 1 + is_compound; ++ref) {
int row_col;
for (row_col = 0; row_col < 2; ++row_col) {
const int dir = (ref << 1) + row_col;
if (has_subpel_mv_component(mi, xd, dir)) {
return 1;
}
}
}
return 0;
}
#if CONFIG_MOTION_VAR
const uint8_t *av1_get_obmc_mask(int length);
void av1_count_overlappable_neighbors(const AV1_COMMON *cm, MACROBLOCKD *xd,
int mi_row, int mi_col);
void av1_build_obmc_inter_prediction(const AV1_COMMON *cm, MACROBLOCKD *xd,
int mi_row, int mi_col,
uint8_t *above[MAX_MB_PLANE],
int above_stride[MAX_MB_PLANE],
uint8_t *left[MAX_MB_PLANE],
int left_stride[MAX_MB_PLANE]);
void av1_build_prediction_by_above_preds(const AV1_COMMON *cm, MACROBLOCKD *xd,
int mi_row, int mi_col,
uint8_t *tmp_buf[MAX_MB_PLANE],
int tmp_width[MAX_MB_PLANE],
int tmp_height[MAX_MB_PLANE],
int tmp_stride[MAX_MB_PLANE]);
void av1_build_prediction_by_left_preds(const AV1_COMMON *cm, MACROBLOCKD *xd,
int mi_row, int mi_col,
uint8_t *tmp_buf[MAX_MB_PLANE],
int tmp_width[MAX_MB_PLANE],
int tmp_height[MAX_MB_PLANE],
int tmp_stride[MAX_MB_PLANE]);
void av1_build_obmc_inter_predictors_sb(const AV1_COMMON *cm, MACROBLOCKD *xd,
int mi_row, int mi_col);
#if CONFIG_NCOBMC
void av1_build_ncobmc_inter_predictors_sb(const AV1_COMMON *cm, MACROBLOCKD *xd,
int mi_row, int mi_col);
#endif
#endif // CONFIG_MOTION_VAR
#if CONFIG_EXT_INTER
#define MASK_MASTER_SIZE ((MAX_WEDGE_SIZE) << 1)
#define MASK_MASTER_STRIDE (MASK_MASTER_SIZE)
void av1_init_wedge_masks();
static INLINE const uint8_t *av1_get_contiguous_soft_mask(int wedge_index,
int wedge_sign,
BLOCK_SIZE sb_type) {
return wedge_params_lookup[sb_type].masks[wedge_sign][wedge_index];
}
const uint8_t *av1_get_soft_mask(int wedge_index, int wedge_sign,
BLOCK_SIZE sb_type, int wedge_offset_x,
int wedge_offset_y);
const uint8_t *av1_get_compound_type_mask_inverse(
const INTERINTER_COMPOUND_DATA *const comp_data,
#if CONFIG_COMPOUND_SEGMENT
uint8_t *mask_buffer, int h, int w, int stride,
#endif
BLOCK_SIZE sb_type);
const uint8_t *av1_get_compound_type_mask(
const INTERINTER_COMPOUND_DATA *const comp_data, BLOCK_SIZE sb_type);
void av1_build_interintra_predictors(MACROBLOCKD *xd, uint8_t *ypred,
uint8_t *upred, uint8_t *vpred,
int ystride, int ustride, int vstride,
BUFFER_SET *ctx, BLOCK_SIZE bsize);
void av1_build_interintra_predictors_sby(MACROBLOCKD *xd, uint8_t *ypred,
int ystride, BUFFER_SET *ctx,
BLOCK_SIZE bsize);
void av1_build_interintra_predictors_sbc(MACROBLOCKD *xd, uint8_t *upred,
int ustride, BUFFER_SET *ctx,
int plane, BLOCK_SIZE bsize);
void av1_build_interintra_predictors_sbuv(MACROBLOCKD *xd, uint8_t *upred,
uint8_t *vpred, int ustride,
int vstride, BUFFER_SET *ctx,
BLOCK_SIZE bsize);
void av1_build_intra_predictors_for_interintra(MACROBLOCKD *xd,
BLOCK_SIZE bsize, int plane,
BUFFER_SET *ctx,
uint8_t *intra_pred,
int intra_stride);
void av1_combine_interintra(MACROBLOCKD *xd, BLOCK_SIZE bsize, int plane,
const uint8_t *inter_pred, int inter_stride,
const uint8_t *intra_pred, int intra_stride);
// Encoder only
void av1_build_inter_predictors_for_planes_single_buf(
MACROBLOCKD *xd, BLOCK_SIZE bsize, int plane_from, int plane_to, int mi_row,
int mi_col, int ref, uint8_t *ext_dst[3], int ext_dst_stride[3]);
void av1_build_wedge_inter_predictor_from_buf(
MACROBLOCKD *xd, BLOCK_SIZE bsize, int plane_from, int plane_to,
#if CONFIG_SUPERTX
int wedge_offset_x, int wedge_offset_y,
#endif // CONFIG_SUPERTX
uint8_t *ext_dst0[3], int ext_dst_stride0[3], uint8_t *ext_dst1[3],
int ext_dst_stride1[3]);
#endif // CONFIG_EXT_INTER
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_RECONINTER_H_

2467
third_party/aom/av1/common/reconintra.c vendored Normal file
View file

File diff suppressed because it is too large Load diff

67
third_party/aom/av1/common/reconintra.h vendored Normal file
View file

@ -0,0 +1,67 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_RECONINTRA_H_
#define AV1_COMMON_RECONINTRA_H_
#include "aom/aom_integer.h"
#include "av1/common/blockd.h"
#ifdef __cplusplus
extern "C" {
#endif
void av1_init_intra_predictors(void);
void av1_predict_intra_block_facade(MACROBLOCKD *xd, int plane, int block_idx,
int blk_col, int blk_row, TX_SIZE tx_size);
void av1_predict_intra_block(const MACROBLOCKD *xd, int bw, int bh,
BLOCK_SIZE bsize, PREDICTION_MODE mode,
const uint8_t *ref, int ref_stride, uint8_t *dst,
int dst_stride, int aoff, int loff, int plane);
#if CONFIG_EXT_INTER
// Mapping of interintra to intra mode for use in the intra component
static const PREDICTION_MODE interintra_to_intra_mode[INTERINTRA_MODES] = {
DC_PRED, V_PRED, H_PRED, D45_PRED, D135_PRED,
D117_PRED, D153_PRED, D207_PRED, D63_PRED, TM_PRED
};
// Mapping of intra mode to the interintra mode
static const INTERINTRA_MODE intra_to_interintra_mode[INTRA_MODES] = {
II_DC_PRED, II_V_PRED, II_H_PRED, II_D45_PRED, II_D135_PRED,
II_D117_PRED, II_D153_PRED, II_D207_PRED, II_D63_PRED,
#if CONFIG_ALT_INTRA
II_DC_PRED, // Note: Filler value, as there's no II_SMOOTH_PRED.
#endif // CONFIG_ALT_INTRA
II_TM_PRED
};
#endif // CONFIG_EXT_INTER
#ifdef __cplusplus
} // extern "C"
#endif
#if CONFIG_FILTER_INTRA
#define FILTER_INTRA_PREC_BITS 10
extern int av1_filter_intra_taps_4[TX_SIZES][INTRA_MODES][4];
#endif // CONFIG_FILTER_INTRA
#if CONFIG_EXT_INTRA
static INLINE int av1_is_directional_mode(PREDICTION_MODE mode,
BLOCK_SIZE bsize) {
return mode != DC_PRED && mode != TM_PRED &&
#if CONFIG_ALT_INTRA
mode != SMOOTH_PRED &&
#endif // CONFIG_ALT_INTRA
bsize >= BLOCK_8X8;
}
#endif // CONFIG_EXT_INTRA
#endif // AV1_COMMON_RECONINTRA_H_

821
third_party/aom/av1/common/resize.c vendored Normal file
View file

@ -0,0 +1,821 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <assert.h>
#include <limits.h>
#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "./aom_config.h"
#if CONFIG_HIGHBITDEPTH
#include "aom_dsp/aom_dsp_common.h"
#endif // CONFIG_HIGHBITDEPTH
#include "aom_ports/mem.h"
#include "av1/common/common.h"
#include "av1/common/resize.h"
#define FILTER_BITS 7
#define INTERP_TAPS 8
#define SUBPEL_BITS 5
#define SUBPEL_MASK ((1 << SUBPEL_BITS) - 1)
#define INTERP_PRECISION_BITS 32
typedef int16_t interp_kernel[INTERP_TAPS];
// Filters for interpolation (0.5-band) - note this also filters integer pels.
static const interp_kernel filteredinterp_filters500[(1 << SUBPEL_BITS)] = {
{ -3, 0, 35, 64, 35, 0, -3, 0 }, { -3, -1, 34, 64, 36, 1, -3, 0 },
{ -3, -1, 32, 64, 38, 1, -3, 0 }, { -2, -2, 31, 63, 39, 2, -3, 0 },
{ -2, -2, 29, 63, 41, 2, -3, 0 }, { -2, -2, 28, 63, 42, 3, -4, 0 },
{ -2, -3, 27, 63, 43, 4, -4, 0 }, { -2, -3, 25, 62, 45, 5, -4, 0 },
{ -2, -3, 24, 62, 46, 5, -4, 0 }, { -2, -3, 23, 61, 47, 6, -4, 0 },
{ -2, -3, 21, 60, 49, 7, -4, 0 }, { -1, -4, 20, 60, 50, 8, -4, -1 },
{ -1, -4, 19, 59, 51, 9, -4, -1 }, { -1, -4, 17, 58, 52, 10, -4, 0 },
{ -1, -4, 16, 57, 53, 12, -4, -1 }, { -1, -4, 15, 56, 54, 13, -4, -1 },
{ -1, -4, 14, 55, 55, 14, -4, -1 }, { -1, -4, 13, 54, 56, 15, -4, -1 },
{ -1, -4, 12, 53, 57, 16, -4, -1 }, { 0, -4, 10, 52, 58, 17, -4, -1 },
{ -1, -4, 9, 51, 59, 19, -4, -1 }, { -1, -4, 8, 50, 60, 20, -4, -1 },
{ 0, -4, 7, 49, 60, 21, -3, -2 }, { 0, -4, 6, 47, 61, 23, -3, -2 },
{ 0, -4, 5, 46, 62, 24, -3, -2 }, { 0, -4, 5, 45, 62, 25, -3, -2 },
{ 0, -4, 4, 43, 63, 27, -3, -2 }, { 0, -4, 3, 42, 63, 28, -2, -2 },
{ 0, -3, 2, 41, 63, 29, -2, -2 }, { 0, -3, 2, 39, 63, 31, -2, -2 },
{ 0, -3, 1, 38, 64, 32, -1, -3 }, { 0, -3, 1, 36, 64, 34, -1, -3 }
};
// Filters for interpolation (0.625-band) - note this also filters integer pels.
static const interp_kernel filteredinterp_filters625[(1 << SUBPEL_BITS)] = {
{ -1, -8, 33, 80, 33, -8, -1, 0 }, { -1, -8, 30, 80, 35, -8, -1, 1 },
{ -1, -8, 28, 80, 37, -7, -2, 1 }, { 0, -8, 26, 79, 39, -7, -2, 1 },
{ 0, -8, 24, 79, 41, -7, -2, 1 }, { 0, -8, 22, 78, 43, -6, -2, 1 },
{ 0, -8, 20, 78, 45, -5, -3, 1 }, { 0, -8, 18, 77, 48, -5, -3, 1 },
{ 0, -8, 16, 76, 50, -4, -3, 1 }, { 0, -8, 15, 75, 52, -3, -4, 1 },
{ 0, -7, 13, 74, 54, -3, -4, 1 }, { 0, -7, 11, 73, 56, -2, -4, 1 },
{ 0, -7, 10, 71, 58, -1, -4, 1 }, { 1, -7, 8, 70, 60, 0, -5, 1 },
{ 1, -6, 6, 68, 62, 1, -5, 1 }, { 1, -6, 5, 67, 63, 2, -5, 1 },
{ 1, -6, 4, 65, 65, 4, -6, 1 }, { 1, -5, 2, 63, 67, 5, -6, 1 },
{ 1, -5, 1, 62, 68, 6, -6, 1 }, { 1, -5, 0, 60, 70, 8, -7, 1 },
{ 1, -4, -1, 58, 71, 10, -7, 0 }, { 1, -4, -2, 56, 73, 11, -7, 0 },
{ 1, -4, -3, 54, 74, 13, -7, 0 }, { 1, -4, -3, 52, 75, 15, -8, 0 },
{ 1, -3, -4, 50, 76, 16, -8, 0 }, { 1, -3, -5, 48, 77, 18, -8, 0 },
{ 1, -3, -5, 45, 78, 20, -8, 0 }, { 1, -2, -6, 43, 78, 22, -8, 0 },
{ 1, -2, -7, 41, 79, 24, -8, 0 }, { 1, -2, -7, 39, 79, 26, -8, 0 },
{ 1, -2, -7, 37, 80, 28, -8, -1 }, { 1, -1, -8, 35, 80, 30, -8, -1 },
};
// Filters for interpolation (0.75-band) - note this also filters integer pels.
static const interp_kernel filteredinterp_filters750[(1 << SUBPEL_BITS)] = {
{ 2, -11, 25, 96, 25, -11, 2, 0 }, { 2, -11, 22, 96, 28, -11, 2, 0 },
{ 2, -10, 19, 95, 31, -11, 2, 0 }, { 2, -10, 17, 95, 34, -12, 2, 0 },
{ 2, -9, 14, 94, 37, -12, 2, 0 }, { 2, -8, 12, 93, 40, -12, 1, 0 },
{ 2, -8, 9, 92, 43, -12, 1, 1 }, { 2, -7, 7, 91, 46, -12, 1, 0 },
{ 2, -7, 5, 90, 49, -12, 1, 0 }, { 2, -6, 3, 88, 52, -12, 0, 1 },
{ 2, -5, 1, 86, 55, -12, 0, 1 }, { 2, -5, -1, 84, 58, -11, 0, 1 },
{ 2, -4, -2, 82, 61, -11, -1, 1 }, { 2, -4, -4, 80, 64, -10, -1, 1 },
{ 1, -3, -5, 77, 67, -9, -1, 1 }, { 1, -3, -6, 75, 70, -8, -2, 1 },
{ 1, -2, -7, 72, 72, -7, -2, 1 }, { 1, -2, -8, 70, 75, -6, -3, 1 },
{ 1, -1, -9, 67, 77, -5, -3, 1 }, { 1, -1, -10, 64, 80, -4, -4, 2 },
{ 1, -1, -11, 61, 82, -2, -4, 2 }, { 1, 0, -11, 58, 84, -1, -5, 2 },
{ 1, 0, -12, 55, 86, 1, -5, 2 }, { 1, 0, -12, 52, 88, 3, -6, 2 },
{ 0, 1, -12, 49, 90, 5, -7, 2 }, { 0, 1, -12, 46, 91, 7, -7, 2 },
{ 1, 1, -12, 43, 92, 9, -8, 2 }, { 0, 1, -12, 40, 93, 12, -8, 2 },
{ 0, 2, -12, 37, 94, 14, -9, 2 }, { 0, 2, -12, 34, 95, 17, -10, 2 },
{ 0, 2, -11, 31, 95, 19, -10, 2 }, { 0, 2, -11, 28, 96, 22, -11, 2 }
};
// Filters for interpolation (0.875-band) - note this also filters integer pels.
static const interp_kernel filteredinterp_filters875[(1 << SUBPEL_BITS)] = {
{ 3, -8, 13, 112, 13, -8, 3, 0 }, { 3, -7, 10, 112, 17, -9, 3, -1 },
{ 2, -6, 7, 111, 21, -9, 3, -1 }, { 2, -5, 4, 111, 24, -10, 3, -1 },
{ 2, -4, 1, 110, 28, -11, 3, -1 }, { 1, -3, -1, 108, 32, -12, 4, -1 },
{ 1, -2, -3, 106, 36, -13, 4, -1 }, { 1, -1, -6, 105, 40, -14, 4, -1 },
{ 1, -1, -7, 102, 44, -14, 4, -1 }, { 1, 0, -9, 100, 48, -15, 4, -1 },
{ 1, 1, -11, 97, 53, -16, 4, -1 }, { 0, 1, -12, 95, 57, -16, 4, -1 },
{ 0, 2, -13, 91, 61, -16, 4, -1 }, { 0, 2, -14, 88, 65, -16, 4, -1 },
{ 0, 3, -15, 84, 69, -17, 4, 0 }, { 0, 3, -16, 81, 73, -16, 3, 0 },
{ 0, 3, -16, 77, 77, -16, 3, 0 }, { 0, 3, -16, 73, 81, -16, 3, 0 },
{ 0, 4, -17, 69, 84, -15, 3, 0 }, { -1, 4, -16, 65, 88, -14, 2, 0 },
{ -1, 4, -16, 61, 91, -13, 2, 0 }, { -1, 4, -16, 57, 95, -12, 1, 0 },
{ -1, 4, -16, 53, 97, -11, 1, 1 }, { -1, 4, -15, 48, 100, -9, 0, 1 },
{ -1, 4, -14, 44, 102, -7, -1, 1 }, { -1, 4, -14, 40, 105, -6, -1, 1 },
{ -1, 4, -13, 36, 106, -3, -2, 1 }, { -1, 4, -12, 32, 108, -1, -3, 1 },
{ -1, 3, -11, 28, 110, 1, -4, 2 }, { -1, 3, -10, 24, 111, 4, -5, 2 },
{ -1, 3, -9, 21, 111, 7, -6, 2 }, { -1, 3, -9, 17, 112, 10, -7, 3 }
};
// Filters for interpolation (full-band) - no filtering for integer pixels
static const interp_kernel filteredinterp_filters1000[(1 << SUBPEL_BITS)] = {
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, 1, -3, 128, 3, -1, 0, 0 },
{ -1, 2, -6, 127, 7, -2, 1, 0 }, { -1, 3, -9, 126, 12, -4, 1, 0 },
{ -1, 4, -12, 125, 16, -5, 1, 0 }, { -1, 4, -14, 123, 20, -6, 2, 0 },
{ -1, 5, -15, 120, 25, -8, 2, 0 }, { -1, 5, -17, 118, 30, -9, 3, -1 },
{ -1, 6, -18, 114, 35, -10, 3, -1 }, { -1, 6, -19, 111, 41, -12, 3, -1 },
{ -1, 6, -20, 107, 46, -13, 4, -1 }, { -1, 6, -21, 103, 52, -14, 4, -1 },
{ -1, 6, -21, 99, 57, -16, 5, -1 }, { -1, 6, -21, 94, 63, -17, 5, -1 },
{ -1, 6, -20, 89, 68, -18, 5, -1 }, { -1, 6, -20, 84, 73, -19, 6, -1 },
{ -1, 6, -20, 79, 79, -20, 6, -1 }, { -1, 6, -19, 73, 84, -20, 6, -1 },
{ -1, 5, -18, 68, 89, -20, 6, -1 }, { -1, 5, -17, 63, 94, -21, 6, -1 },
{ -1, 5, -16, 57, 99, -21, 6, -1 }, { -1, 4, -14, 52, 103, -21, 6, -1 },
{ -1, 4, -13, 46, 107, -20, 6, -1 }, { -1, 3, -12, 41, 111, -19, 6, -1 },
{ -1, 3, -10, 35, 114, -18, 6, -1 }, { -1, 3, -9, 30, 118, -17, 5, -1 },
{ 0, 2, -8, 25, 120, -15, 5, -1 }, { 0, 2, -6, 20, 123, -14, 4, -1 },
{ 0, 1, -5, 16, 125, -12, 4, -1 }, { 0, 1, -4, 12, 126, -9, 3, -1 },
{ 0, 1, -2, 7, 127, -6, 2, -1 }, { 0, 0, -1, 3, 128, -3, 1, 0 }
};
// Filters for factor of 2 downsampling.
static const int16_t av1_down2_symeven_half_filter[] = { 56, 12, -3, -1 };
static const int16_t av1_down2_symodd_half_filter[] = { 64, 35, 0, -3 };
static const interp_kernel *choose_interp_filter(int inlength, int outlength) {
int outlength16 = outlength * 16;
if (outlength16 >= inlength * 16)
return filteredinterp_filters1000;
else if (outlength16 >= inlength * 13)
return filteredinterp_filters875;
else if (outlength16 >= inlength * 11)
return filteredinterp_filters750;
else if (outlength16 >= inlength * 9)
return filteredinterp_filters625;
else
return filteredinterp_filters500;
}
static void interpolate(const uint8_t *const input, int inlength,
uint8_t *output, int outlength) {
const int64_t delta =
(((uint64_t)inlength << 32) + outlength / 2) / outlength;
const int64_t offset =
inlength > outlength
? (((int64_t)(inlength - outlength) << 31) + outlength / 2) /
outlength
: -(((int64_t)(outlength - inlength) << 31) + outlength / 2) /
outlength;
uint8_t *optr = output;
int x, x1, x2, sum, k, int_pel, sub_pel;
int64_t y;
const interp_kernel *interp_filters =
choose_interp_filter(inlength, outlength);
x = 0;
y = offset;
while ((y >> INTERP_PRECISION_BITS) < (INTERP_TAPS / 2 - 1)) {
x++;
y += delta;
}
x1 = x;
x = outlength - 1;
y = delta * x + offset;
while ((y >> INTERP_PRECISION_BITS) + (int64_t)(INTERP_TAPS / 2) >=
inlength) {
x--;
y -= delta;
}
x2 = x;
if (x1 > x2) {
for (x = 0, y = offset; x < outlength; ++x, y += delta) {
const int16_t *filter;
int_pel = y >> INTERP_PRECISION_BITS;
sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k) {
const int pk = int_pel - INTERP_TAPS / 2 + 1 + k;
sum += filter[k] *
input[(pk < 0 ? 0 : (pk >= inlength ? inlength - 1 : pk))];
}
*optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
}
} else {
// Initial part.
for (x = 0, y = offset; x < x1; ++x, y += delta) {
const int16_t *filter;
int_pel = y >> INTERP_PRECISION_BITS;
sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k)
sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k < 0
? 0
: int_pel - INTERP_TAPS / 2 + 1 + k)];
*optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
}
// Middle part.
for (; x <= x2; ++x, y += delta) {
const int16_t *filter;
int_pel = y >> INTERP_PRECISION_BITS;
sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k)
sum += filter[k] * input[int_pel - INTERP_TAPS / 2 + 1 + k];
*optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
}
// End part.
for (; x < outlength; ++x, y += delta) {
const int16_t *filter;
int_pel = y >> INTERP_PRECISION_BITS;
sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k)
sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k >= inlength
? inlength - 1
: int_pel - INTERP_TAPS / 2 + 1 + k)];
*optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
}
}
}
static void down2_symeven(const uint8_t *const input, int length,
uint8_t *output) {
// Actual filter len = 2 * filter_len_half.
const int16_t *filter = av1_down2_symeven_half_filter;
const int filter_len_half = sizeof(av1_down2_symeven_half_filter) / 2;
int i, j;
uint8_t *optr = output;
int l1 = filter_len_half;
int l2 = (length - filter_len_half);
l1 += (l1 & 1);
l2 += (l2 & 1);
if (l1 > l2) {
// Short input length.
for (i = 0; i < length; i += 2) {
int sum = (1 << (FILTER_BITS - 1));
for (j = 0; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] +
input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel(sum);
}
} else {
// Initial part.
for (i = 0; i < l1; i += 2) {
int sum = (1 << (FILTER_BITS - 1));
for (j = 0; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] + input[i + 1 + j]) * filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel(sum);
}
// Middle part.
for (; i < l2; i += 2) {
int sum = (1 << (FILTER_BITS - 1));
for (j = 0; j < filter_len_half; ++j) {
sum += (input[i - j] + input[i + 1 + j]) * filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel(sum);
}
// End part.
for (; i < length; i += 2) {
int sum = (1 << (FILTER_BITS - 1));
for (j = 0; j < filter_len_half; ++j) {
sum += (input[i - j] +
input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel(sum);
}
}
}
static void down2_symodd(const uint8_t *const input, int length,
uint8_t *output) {
// Actual filter len = 2 * filter_len_half - 1.
const int16_t *filter = av1_down2_symodd_half_filter;
const int filter_len_half = sizeof(av1_down2_symodd_half_filter) / 2;
int i, j;
uint8_t *optr = output;
int l1 = filter_len_half - 1;
int l2 = (length - filter_len_half + 1);
l1 += (l1 & 1);
l2 += (l2 & 1);
if (l1 > l2) {
// Short input length.
for (i = 0; i < length; i += 2) {
int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
for (j = 1; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] +
input[(i + j >= length ? length - 1 : i + j)]) *
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel(sum);
}
} else {
// Initial part.
for (i = 0; i < l1; i += 2) {
int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
for (j = 1; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] + input[i + j]) * filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel(sum);
}
// Middle part.
for (; i < l2; i += 2) {
int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
for (j = 1; j < filter_len_half; ++j) {
sum += (input[i - j] + input[i + j]) * filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel(sum);
}
// End part.
for (; i < length; i += 2) {
int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
for (j = 1; j < filter_len_half; ++j) {
sum += (input[i - j] + input[(i + j >= length ? length - 1 : i + j)]) *
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel(sum);
}
}
}
static int get_down2_length(int length, int steps) {
int s;
for (s = 0; s < steps; ++s) length = (length + 1) >> 1;
return length;
}
static int get_down2_steps(int in_length, int out_length) {
int steps = 0;
int proj_in_length;
while ((proj_in_length = get_down2_length(in_length, 1)) >= out_length) {
++steps;
in_length = proj_in_length;
}
return steps;
}
static void resize_multistep(const uint8_t *const input, int length,
uint8_t *output, int olength, uint8_t *otmp) {
int steps;
if (length == olength) {
memcpy(output, input, sizeof(output[0]) * length);
return;
}
steps = get_down2_steps(length, olength);
if (steps > 0) {
int s;
uint8_t *out = NULL;
uint8_t *otmp2;
int filteredlength = length;
assert(otmp != NULL);
otmp2 = otmp + get_down2_length(length, 1);
for (s = 0; s < steps; ++s) {
const int proj_filteredlength = get_down2_length(filteredlength, 1);
const uint8_t *const in = (s == 0 ? input : out);
if (s == steps - 1 && proj_filteredlength == olength)
out = output;
else
out = (s & 1 ? otmp2 : otmp);
if (filteredlength & 1)
down2_symodd(in, filteredlength, out);
else
down2_symeven(in, filteredlength, out);
filteredlength = proj_filteredlength;
}
if (filteredlength != olength) {
interpolate(out, filteredlength, output, olength);
}
} else {
interpolate(input, length, output, olength);
}
}
static void fill_col_to_arr(uint8_t *img, int stride, int len, uint8_t *arr) {
int i;
uint8_t *iptr = img;
uint8_t *aptr = arr;
for (i = 0; i < len; ++i, iptr += stride) {
*aptr++ = *iptr;
}
}
static void fill_arr_to_col(uint8_t *img, int stride, int len, uint8_t *arr) {
int i;
uint8_t *iptr = img;
uint8_t *aptr = arr;
for (i = 0; i < len; ++i, iptr += stride) {
*iptr = *aptr++;
}
}
void av1_resize_plane(const uint8_t *const input, int height, int width,
int in_stride, uint8_t *output, int height2, int width2,
int out_stride) {
int i;
uint8_t *intbuf = (uint8_t *)malloc(sizeof(uint8_t) * width2 * height);
uint8_t *tmpbuf =
(uint8_t *)malloc(sizeof(uint8_t) * (width < height ? height : width));
uint8_t *arrbuf = (uint8_t *)malloc(sizeof(uint8_t) * height);
uint8_t *arrbuf2 = (uint8_t *)malloc(sizeof(uint8_t) * height2);
if (intbuf == NULL || tmpbuf == NULL || arrbuf == NULL || arrbuf2 == NULL)
goto Error;
assert(width > 0);
assert(height > 0);
assert(width2 > 0);
assert(height2 > 0);
for (i = 0; i < height; ++i)
resize_multistep(input + in_stride * i, width, intbuf + width2 * i, width2,
tmpbuf);
for (i = 0; i < width2; ++i) {
fill_col_to_arr(intbuf + i, width2, height, arrbuf);
resize_multistep(arrbuf, height, arrbuf2, height2, tmpbuf);
fill_arr_to_col(output + i, out_stride, height2, arrbuf2);
}
Error:
free(intbuf);
free(tmpbuf);
free(arrbuf);
free(arrbuf2);
}
#if CONFIG_HIGHBITDEPTH
static void highbd_interpolate(const uint16_t *const input, int inlength,
uint16_t *output, int outlength, int bd) {
const int64_t delta =
(((uint64_t)inlength << 32) + outlength / 2) / outlength;
const int64_t offset =
inlength > outlength
? (((int64_t)(inlength - outlength) << 31) + outlength / 2) /
outlength
: -(((int64_t)(outlength - inlength) << 31) + outlength / 2) /
outlength;
uint16_t *optr = output;
int x, x1, x2, sum, k, int_pel, sub_pel;
int64_t y;
const interp_kernel *interp_filters =
choose_interp_filter(inlength, outlength);
x = 0;
y = offset;
while ((y >> INTERP_PRECISION_BITS) < (INTERP_TAPS / 2 - 1)) {
x++;
y += delta;
}
x1 = x;
x = outlength - 1;
y = delta * x + offset;
while ((y >> INTERP_PRECISION_BITS) + (int64_t)(INTERP_TAPS / 2) >=
inlength) {
x--;
y -= delta;
}
x2 = x;
if (x1 > x2) {
for (x = 0, y = offset; x < outlength; ++x, y += delta) {
const int16_t *filter;
int_pel = y >> INTERP_PRECISION_BITS;
sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k) {
const int pk = int_pel - INTERP_TAPS / 2 + 1 + k;
sum += filter[k] *
input[(pk < 0 ? 0 : (pk >= inlength ? inlength - 1 : pk))];
}
*optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
}
} else {
// Initial part.
for (x = 0, y = offset; x < x1; ++x, y += delta) {
const int16_t *filter;
int_pel = y >> INTERP_PRECISION_BITS;
sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k)
sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k < 0
? 0
: int_pel - INTERP_TAPS / 2 + 1 + k)];
*optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
}
// Middle part.
for (; x <= x2; ++x, y += delta) {
const int16_t *filter;
int_pel = y >> INTERP_PRECISION_BITS;
sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k)
sum += filter[k] * input[int_pel - INTERP_TAPS / 2 + 1 + k];
*optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
}
// End part.
for (; x < outlength; ++x, y += delta) {
const int16_t *filter;
int_pel = y >> INTERP_PRECISION_BITS;
sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k)
sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k >= inlength
? inlength - 1
: int_pel - INTERP_TAPS / 2 + 1 + k)];
*optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
}
}
}
static void highbd_down2_symeven(const uint16_t *const input, int length,
uint16_t *output, int bd) {
// Actual filter len = 2 * filter_len_half.
static const int16_t *filter = av1_down2_symeven_half_filter;
const int filter_len_half = sizeof(av1_down2_symeven_half_filter) / 2;
int i, j;
uint16_t *optr = output;
int l1 = filter_len_half;
int l2 = (length - filter_len_half);
l1 += (l1 & 1);
l2 += (l2 & 1);
if (l1 > l2) {
// Short input length.
for (i = 0; i < length; i += 2) {
int sum = (1 << (FILTER_BITS - 1));
for (j = 0; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] +
input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel_highbd(sum, bd);
}
} else {
// Initial part.
for (i = 0; i < l1; i += 2) {
int sum = (1 << (FILTER_BITS - 1));
for (j = 0; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] + input[i + 1 + j]) * filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel_highbd(sum, bd);
}
// Middle part.
for (; i < l2; i += 2) {
int sum = (1 << (FILTER_BITS - 1));
for (j = 0; j < filter_len_half; ++j) {
sum += (input[i - j] + input[i + 1 + j]) * filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel_highbd(sum, bd);
}
// End part.
for (; i < length; i += 2) {
int sum = (1 << (FILTER_BITS - 1));
for (j = 0; j < filter_len_half; ++j) {
sum += (input[i - j] +
input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel_highbd(sum, bd);
}
}
}
static void highbd_down2_symodd(const uint16_t *const input, int length,
uint16_t *output, int bd) {
// Actual filter len = 2 * filter_len_half - 1.
static const int16_t *filter = av1_down2_symodd_half_filter;
const int filter_len_half = sizeof(av1_down2_symodd_half_filter) / 2;
int i, j;
uint16_t *optr = output;
int l1 = filter_len_half - 1;
int l2 = (length - filter_len_half + 1);
l1 += (l1 & 1);
l2 += (l2 & 1);
if (l1 > l2) {
// Short input length.
for (i = 0; i < length; i += 2) {
int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
for (j = 1; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] +
input[(i + j >= length ? length - 1 : i + j)]) *
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel_highbd(sum, bd);
}
} else {
// Initial part.
for (i = 0; i < l1; i += 2) {
int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
for (j = 1; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] + input[i + j]) * filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel_highbd(sum, bd);
}
// Middle part.
for (; i < l2; i += 2) {
int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
for (j = 1; j < filter_len_half; ++j) {
sum += (input[i - j] + input[i + j]) * filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel_highbd(sum, bd);
}
// End part.
for (; i < length; i += 2) {
int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
for (j = 1; j < filter_len_half; ++j) {
sum += (input[i - j] + input[(i + j >= length ? length - 1 : i + j)]) *
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel_highbd(sum, bd);
}
}
}
static void highbd_resize_multistep(const uint16_t *const input, int length,
uint16_t *output, int olength,
uint16_t *otmp, int bd) {
int steps;
if (length == olength) {
memcpy(output, input, sizeof(output[0]) * length);
return;
}
steps = get_down2_steps(length, olength);
if (steps > 0) {
int s;
uint16_t *out = NULL;
uint16_t *otmp2;
int filteredlength = length;
assert(otmp != NULL);
otmp2 = otmp + get_down2_length(length, 1);
for (s = 0; s < steps; ++s) {
const int proj_filteredlength = get_down2_length(filteredlength, 1);
const uint16_t *const in = (s == 0 ? input : out);
if (s == steps - 1 && proj_filteredlength == olength)
out = output;
else
out = (s & 1 ? otmp2 : otmp);
if (filteredlength & 1)
highbd_down2_symodd(in, filteredlength, out, bd);
else
highbd_down2_symeven(in, filteredlength, out, bd);
filteredlength = proj_filteredlength;
}
if (filteredlength != olength) {
highbd_interpolate(out, filteredlength, output, olength, bd);
}
} else {
highbd_interpolate(input, length, output, olength, bd);
}
}
static void highbd_fill_col_to_arr(uint16_t *img, int stride, int len,
uint16_t *arr) {
int i;
uint16_t *iptr = img;
uint16_t *aptr = arr;
for (i = 0; i < len; ++i, iptr += stride) {
*aptr++ = *iptr;
}
}
static void highbd_fill_arr_to_col(uint16_t *img, int stride, int len,
uint16_t *arr) {
int i;
uint16_t *iptr = img;
uint16_t *aptr = arr;
for (i = 0; i < len; ++i, iptr += stride) {
*iptr = *aptr++;
}
}
void av1_highbd_resize_plane(const uint8_t *const input, int height, int width,
int in_stride, uint8_t *output, int height2,
int width2, int out_stride, int bd) {
int i;
uint16_t *intbuf = (uint16_t *)malloc(sizeof(uint16_t) * width2 * height);
uint16_t *tmpbuf =
(uint16_t *)malloc(sizeof(uint16_t) * (width < height ? height : width));
uint16_t *arrbuf = (uint16_t *)malloc(sizeof(uint16_t) * height);
uint16_t *arrbuf2 = (uint16_t *)malloc(sizeof(uint16_t) * height2);
if (intbuf == NULL || tmpbuf == NULL || arrbuf == NULL || arrbuf2 == NULL)
goto Error;
for (i = 0; i < height; ++i) {
highbd_resize_multistep(CONVERT_TO_SHORTPTR(input + in_stride * i), width,
intbuf + width2 * i, width2, tmpbuf, bd);
}
for (i = 0; i < width2; ++i) {
highbd_fill_col_to_arr(intbuf + i, width2, height, arrbuf);
highbd_resize_multistep(arrbuf, height, arrbuf2, height2, tmpbuf, bd);
highbd_fill_arr_to_col(CONVERT_TO_SHORTPTR(output + i), out_stride, height2,
arrbuf2);
}
Error:
free(intbuf);
free(tmpbuf);
free(arrbuf);
free(arrbuf2);
}
#endif // CONFIG_HIGHBITDEPTH
void av1_resize_frame420(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width, uint8_t *oy,
int oy_stride, uint8_t *ou, uint8_t *ov,
int ouv_stride, int oheight, int owidth) {
av1_resize_plane(y, height, width, y_stride, oy, oheight, owidth, oy_stride);
av1_resize_plane(u, height / 2, width / 2, uv_stride, ou, oheight / 2,
owidth / 2, ouv_stride);
av1_resize_plane(v, height / 2, width / 2, uv_stride, ov, oheight / 2,
owidth / 2, ouv_stride);
}
void av1_resize_frame422(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width, uint8_t *oy,
int oy_stride, uint8_t *ou, uint8_t *ov,
int ouv_stride, int oheight, int owidth) {
av1_resize_plane(y, height, width, y_stride, oy, oheight, owidth, oy_stride);
av1_resize_plane(u, height, width / 2, uv_stride, ou, oheight, owidth / 2,
ouv_stride);
av1_resize_plane(v, height, width / 2, uv_stride, ov, oheight, owidth / 2,
ouv_stride);
}
void av1_resize_frame444(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width, uint8_t *oy,
int oy_stride, uint8_t *ou, uint8_t *ov,
int ouv_stride, int oheight, int owidth) {
av1_resize_plane(y, height, width, y_stride, oy, oheight, owidth, oy_stride);
av1_resize_plane(u, height, width, uv_stride, ou, oheight, owidth,
ouv_stride);
av1_resize_plane(v, height, width, uv_stride, ov, oheight, owidth,
ouv_stride);
}
#if CONFIG_HIGHBITDEPTH
void av1_highbd_resize_frame420(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width,
uint8_t *oy, int oy_stride, uint8_t *ou,
uint8_t *ov, int ouv_stride, int oheight,
int owidth, int bd) {
av1_highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth,
oy_stride, bd);
av1_highbd_resize_plane(u, height / 2, width / 2, uv_stride, ou, oheight / 2,
owidth / 2, ouv_stride, bd);
av1_highbd_resize_plane(v, height / 2, width / 2, uv_stride, ov, oheight / 2,
owidth / 2, ouv_stride, bd);
}
void av1_highbd_resize_frame422(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width,
uint8_t *oy, int oy_stride, uint8_t *ou,
uint8_t *ov, int ouv_stride, int oheight,
int owidth, int bd) {
av1_highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth,
oy_stride, bd);
av1_highbd_resize_plane(u, height, width / 2, uv_stride, ou, oheight,
owidth / 2, ouv_stride, bd);
av1_highbd_resize_plane(v, height, width / 2, uv_stride, ov, oheight,
owidth / 2, ouv_stride, bd);
}
void av1_highbd_resize_frame444(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width,
uint8_t *oy, int oy_stride, uint8_t *ou,
uint8_t *ov, int ouv_stride, int oheight,
int owidth, int bd) {
av1_highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth,
oy_stride, bd);
av1_highbd_resize_plane(u, height, width, uv_stride, ou, oheight, owidth,
ouv_stride, bd);
av1_highbd_resize_plane(v, height, width, uv_stride, ov, oheight, owidth,
ouv_stride, bd);
}
#endif // CONFIG_HIGHBITDEPTH

69
third_party/aom/av1/common/resize.h vendored Normal file
View file

@ -0,0 +1,69 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_ENCODER_RESIZE_H_
#define AV1_ENCODER_RESIZE_H_
#include <stdio.h>
#include "aom/aom_integer.h"
#ifdef __cplusplus
extern "C" {
#endif
void av1_resize_plane(const uint8_t *const input, int height, int width,
int in_stride, uint8_t *output, int height2, int width2,
int out_stride);
void av1_resize_frame420(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width, uint8_t *oy,
int oy_stride, uint8_t *ou, uint8_t *ov,
int ouv_stride, int oheight, int owidth);
void av1_resize_frame422(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width, uint8_t *oy,
int oy_stride, uint8_t *ou, uint8_t *ov,
int ouv_stride, int oheight, int owidth);
void av1_resize_frame444(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width, uint8_t *oy,
int oy_stride, uint8_t *ou, uint8_t *ov,
int ouv_stride, int oheight, int owidth);
#if CONFIG_HIGHBITDEPTH
void av1_highbd_resize_plane(const uint8_t *const input, int height, int width,
int in_stride, uint8_t *output, int height2,
int width2, int out_stride, int bd);
void av1_highbd_resize_frame420(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width,
uint8_t *oy, int oy_stride, uint8_t *ou,
uint8_t *ov, int ouv_stride, int oheight,
int owidth, int bd);
void av1_highbd_resize_frame422(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width,
uint8_t *oy, int oy_stride, uint8_t *ou,
uint8_t *ov, int ouv_stride, int oheight,
int owidth, int bd);
void av1_highbd_resize_frame444(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width,
uint8_t *oy, int oy_stride, uint8_t *ou,
uint8_t *ov, int ouv_stride, int oheight,
int owidth, int bd);
#endif // CONFIG_HIGHBITDEPTH
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_ENCODER_RESIZE_H_

1401
third_party/aom/av1/common/restoration.c vendored Normal file
View file

File diff suppressed because it is too large Load diff

257
third_party/aom/av1/common/restoration.h vendored Normal file
View file

@ -0,0 +1,257 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_RESTORATION_H_
#define AV1_COMMON_RESTORATION_H_
#include "aom_ports/mem.h"
#include "./aom_config.h"
#include "av1/common/blockd.h"
#ifdef __cplusplus
extern "C" {
#endif
#define CLIP(x, lo, hi) ((x) < (lo) ? (lo) : (x) > (hi) ? (hi) : (x))
#define RINT(x) ((x) < 0 ? (int)((x)-0.5) : (int)((x) + 0.5))
#define RESTORATION_TILESIZE_MAX 256
#define RESTORATION_TILEPELS_MAX \
(RESTORATION_TILESIZE_MAX * RESTORATION_TILESIZE_MAX * 9 / 4)
// 4 32-bit buffers needed for the filter:
// 2 for the restored versions of the frame and
// 2 for each restoration operation
#define SGRPROJ_OUTBUF_SIZE \
((RESTORATION_TILESIZE_MAX * 3 / 2) * (RESTORATION_TILESIZE_MAX * 3 / 2 + 16))
#define SGRPROJ_TMPBUF_SIZE \
(RESTORATION_TILEPELS_MAX * 2 * sizeof(int32_t) + \
SGRPROJ_OUTBUF_SIZE * 2 * sizeof(int32_t))
#define SGRPROJ_EXTBUF_SIZE (0)
#define SGRPROJ_PARAMS_BITS 4
#define SGRPROJ_PARAMS (1 << SGRPROJ_PARAMS_BITS)
#define USE_HIGHPASS_IN_SGRPROJ 0
// Precision bits for projection
#define SGRPROJ_PRJ_BITS 7
// Restoration precision bits generated higher than source before projection
#define SGRPROJ_RST_BITS 4
// Internal precision bits for core selfguided_restoration
#define SGRPROJ_SGR_BITS 8
#define SGRPROJ_SGR (1 << SGRPROJ_SGR_BITS)
#if USE_HIGHPASS_IN_SGRPROJ
#define SGRPROJ_PRJ_MIN0 (-(1 << SGRPROJ_PRJ_BITS) / 8)
#define SGRPROJ_PRJ_MAX0 (SGRPROJ_PRJ_MIN0 + (1 << SGRPROJ_PRJ_BITS) - 1)
#define SGRPROJ_PRJ_MIN1 (-(1 << SGRPROJ_PRJ_BITS) / 2)
#define SGRPROJ_PRJ_MAX1 (SGRPROJ_PRJ_MIN1 + (1 << SGRPROJ_PRJ_BITS) - 1)
#else
#define SGRPROJ_PRJ_MIN0 (-(1 << SGRPROJ_PRJ_BITS) * 3 / 4)
#define SGRPROJ_PRJ_MAX0 (SGRPROJ_PRJ_MIN0 + (1 << SGRPROJ_PRJ_BITS) - 1)
#define SGRPROJ_PRJ_MIN1 (-(1 << SGRPROJ_PRJ_BITS) / 4)
#define SGRPROJ_PRJ_MAX1 (SGRPROJ_PRJ_MIN1 + (1 << SGRPROJ_PRJ_BITS) - 1)
#endif // USE_HIGHPASS_IN_SGRPROJ
#define SGRPROJ_PRJ_SUBEXP_K 4
#define SGRPROJ_BITS (SGRPROJ_PRJ_BITS * 2 + SGRPROJ_PARAMS_BITS)
#define MAX_RADIUS 3 // Only 1, 2, 3 allowed
#define MAX_EPS 80 // Max value of eps
#define MAX_NELEM ((2 * MAX_RADIUS + 1) * (2 * MAX_RADIUS + 1))
#define SGRPROJ_MTABLE_BITS 20
#define SGRPROJ_RECIP_BITS 12
#define WIENER_HALFWIN 3
#define WIENER_HALFWIN1 (WIENER_HALFWIN + 1)
#define WIENER_WIN (2 * WIENER_HALFWIN + 1)
#define WIENER_WIN2 ((WIENER_WIN) * (WIENER_WIN))
#define WIENER_TMPBUF_SIZE (0)
#define WIENER_EXTBUF_SIZE (0)
#define WIENER_FILT_PREC_BITS 7
#define WIENER_FILT_STEP (1 << WIENER_FILT_PREC_BITS)
// Central values for the taps
#define WIENER_FILT_TAP0_MIDV (3)
#define WIENER_FILT_TAP1_MIDV (-7)
#define WIENER_FILT_TAP2_MIDV (15)
#define WIENER_FILT_TAP0_BITS 4
#define WIENER_FILT_TAP1_BITS 5
#define WIENER_FILT_TAP2_BITS 6
#define WIENER_FILT_BITS \
((WIENER_FILT_TAP0_BITS + WIENER_FILT_TAP1_BITS + WIENER_FILT_TAP2_BITS) * 2)
#define WIENER_FILT_TAP0_MINV \
(WIENER_FILT_TAP0_MIDV - (1 << WIENER_FILT_TAP0_BITS) / 2)
#define WIENER_FILT_TAP1_MINV \
(WIENER_FILT_TAP1_MIDV - (1 << WIENER_FILT_TAP1_BITS) / 2)
#define WIENER_FILT_TAP2_MINV \
(WIENER_FILT_TAP2_MIDV - (1 << WIENER_FILT_TAP2_BITS) / 2)
#define WIENER_FILT_TAP0_MAXV \
(WIENER_FILT_TAP0_MIDV - 1 + (1 << WIENER_FILT_TAP0_BITS) / 2)
#define WIENER_FILT_TAP1_MAXV \
(WIENER_FILT_TAP1_MIDV - 1 + (1 << WIENER_FILT_TAP1_BITS) / 2)
#define WIENER_FILT_TAP2_MAXV \
(WIENER_FILT_TAP2_MIDV - 1 + (1 << WIENER_FILT_TAP2_BITS) / 2)
#define WIENER_FILT_TAP0_SUBEXP_K 1
#define WIENER_FILT_TAP1_SUBEXP_K 2
#define WIENER_FILT_TAP2_SUBEXP_K 3
// Max of SGRPROJ_TMPBUF_SIZE, DOMAINTXFMRF_TMPBUF_SIZE, WIENER_TMPBUF_SIZE
#define RESTORATION_TMPBUF_SIZE (SGRPROJ_TMPBUF_SIZE)
// Max of SGRPROJ_EXTBUF_SIZE, WIENER_EXTBUF_SIZE
#define RESTORATION_EXTBUF_SIZE (WIENER_EXTBUF_SIZE)
// Check the assumptions of the existing code
#if SUBPEL_TAPS != WIENER_WIN + 1
#error "Wiener filter currently only works if SUBPEL_TAPS == WIENER_WIN + 1"
#endif
#if WIENER_FILT_PREC_BITS != 7
#error "Wiener filter currently only works if WIENER_FILT_PREC_BITS == 7"
#endif
typedef struct {
DECLARE_ALIGNED(16, InterpKernel, vfilter);
DECLARE_ALIGNED(16, InterpKernel, hfilter);
} WienerInfo;
typedef struct {
#if USE_HIGHPASS_IN_SGRPROJ
int corner;
int edge;
#else
int r1;
int e1;
#endif // USE_HIGHPASS_IN_SGRPROJ
int r2;
int e2;
} sgr_params_type;
typedef struct {
int ep;
int xqd[2];
} SgrprojInfo;
typedef struct {
int restoration_tilesize;
RestorationType frame_restoration_type;
RestorationType *restoration_type;
// Wiener filter
WienerInfo *wiener_info;
// Selfguided proj filter
SgrprojInfo *sgrproj_info;
} RestorationInfo;
typedef struct {
RestorationInfo *rsi;
int keyframe;
int ntiles;
int tile_width, tile_height;
int nhtiles, nvtiles;
int32_t *tmpbuf;
} RestorationInternal;
static INLINE void set_default_sgrproj(SgrprojInfo *sgrproj_info) {
sgrproj_info->xqd[0] = (SGRPROJ_PRJ_MIN0 + SGRPROJ_PRJ_MAX0) / 2;
sgrproj_info->xqd[1] = (SGRPROJ_PRJ_MIN1 + SGRPROJ_PRJ_MAX1) / 2;
}
static INLINE void set_default_wiener(WienerInfo *wiener_info) {
wiener_info->vfilter[0] = wiener_info->hfilter[0] = WIENER_FILT_TAP0_MIDV;
wiener_info->vfilter[1] = wiener_info->hfilter[1] = WIENER_FILT_TAP1_MIDV;
wiener_info->vfilter[2] = wiener_info->hfilter[2] = WIENER_FILT_TAP2_MIDV;
wiener_info->vfilter[WIENER_HALFWIN] = wiener_info->hfilter[WIENER_HALFWIN] =
-2 *
(WIENER_FILT_TAP2_MIDV + WIENER_FILT_TAP1_MIDV + WIENER_FILT_TAP0_MIDV);
wiener_info->vfilter[4] = wiener_info->hfilter[4] = WIENER_FILT_TAP2_MIDV;
wiener_info->vfilter[5] = wiener_info->hfilter[5] = WIENER_FILT_TAP1_MIDV;
wiener_info->vfilter[6] = wiener_info->hfilter[6] = WIENER_FILT_TAP0_MIDV;
}
static INLINE int av1_get_rest_ntiles(int width, int height, int tilesize,
int *tile_width, int *tile_height,
int *nhtiles, int *nvtiles) {
int nhtiles_, nvtiles_;
int tile_width_, tile_height_;
tile_width_ = (tilesize < 0) ? width : AOMMIN(tilesize, width);
tile_height_ = (tilesize < 0) ? height : AOMMIN(tilesize, height);
nhtiles_ = (width + (tile_width_ >> 1)) / tile_width_;
nvtiles_ = (height + (tile_height_ >> 1)) / tile_height_;
if (tile_width) *tile_width = tile_width_;
if (tile_height) *tile_height = tile_height_;
if (nhtiles) *nhtiles = nhtiles_;
if (nvtiles) *nvtiles = nvtiles_;
return (nhtiles_ * nvtiles_);
}
static INLINE void av1_get_rest_tile_limits(
int tile_idx, int subtile_idx, int subtile_bits, int nhtiles, int nvtiles,
int tile_width, int tile_height, int im_width, int im_height, int clamp_h,
int clamp_v, int *h_start, int *h_end, int *v_start, int *v_end) {
const int htile_idx = tile_idx % nhtiles;
const int vtile_idx = tile_idx / nhtiles;
*h_start = htile_idx * tile_width;
*v_start = vtile_idx * tile_height;
*h_end = (htile_idx < nhtiles - 1) ? *h_start + tile_width : im_width;
*v_end = (vtile_idx < nvtiles - 1) ? *v_start + tile_height : im_height;
if (subtile_bits) {
const int num_subtiles_1d = (1 << subtile_bits);
const int subtile_width = (*h_end - *h_start) >> subtile_bits;
const int subtile_height = (*v_end - *v_start) >> subtile_bits;
const int subtile_idx_h = subtile_idx & (num_subtiles_1d - 1);
const int subtile_idx_v = subtile_idx >> subtile_bits;
*h_start += subtile_idx_h * subtile_width;
*v_start += subtile_idx_v * subtile_height;
*h_end = subtile_idx_h == num_subtiles_1d - 1 ? *h_end
: *h_start + subtile_width;
*v_end = subtile_idx_v == num_subtiles_1d - 1 ? *v_end
: *v_start + subtile_height;
}
if (clamp_h) {
*h_start = AOMMAX(*h_start, clamp_h);
*h_end = AOMMIN(*h_end, im_width - clamp_h);
}
if (clamp_v) {
*v_start = AOMMAX(*v_start, clamp_v);
*v_end = AOMMIN(*v_end, im_height - clamp_v);
}
}
extern const sgr_params_type sgr_params[SGRPROJ_PARAMS];
extern int sgrproj_mtable[MAX_EPS][MAX_NELEM];
extern const int32_t x_by_xplus1[256];
extern const int32_t one_by_x[MAX_NELEM];
int av1_alloc_restoration_struct(struct AV1Common *cm,
RestorationInfo *rst_info, int width,
int height);
void av1_free_restoration_struct(RestorationInfo *rst_info);
void extend_frame(uint8_t *data, int width, int height, int stride);
#if CONFIG_HIGHBITDEPTH
void extend_frame_highbd(uint16_t *data, int width, int height, int stride);
#endif // CONFIG_HIGHBITDEPTH
void decode_xq(int *xqd, int *xq);
void av1_loop_restoration_frame(YV12_BUFFER_CONFIG *frame, struct AV1Common *cm,
RestorationInfo *rsi, int components_pattern,
int partial_frame, YV12_BUFFER_CONFIG *dst);
void av1_loop_restoration_precal();
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_RESTORATION_H_

164
third_party/aom/av1/common/scale.c vendored Normal file
View file

@ -0,0 +1,164 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include "./aom_dsp_rtcd.h"
#include "av1/common/filter.h"
#include "av1/common/scale.h"
#include "aom_dsp/aom_filter.h"
static INLINE int scaled_x(int val, const struct scale_factors *sf) {
return (int)((int64_t)val * sf->x_scale_fp >> REF_SCALE_SHIFT);
}
static INLINE int scaled_y(int val, const struct scale_factors *sf) {
return (int)((int64_t)val * sf->y_scale_fp >> REF_SCALE_SHIFT);
}
static int unscaled_value(int val, const struct scale_factors *sf) {
(void)sf;
return val;
}
static int get_fixed_point_scale_factor(int other_size, int this_size) {
// Calculate scaling factor once for each reference frame
// and use fixed point scaling factors in decoding and encoding routines.
// Hardware implementations can calculate scale factor in device driver
// and use multiplication and shifting on hardware instead of division.
return (other_size << REF_SCALE_SHIFT) / this_size;
}
MV32 av1_scale_mv(const MV *mv, int x, int y, const struct scale_factors *sf) {
const int x_off_q4 = scaled_x(x << SUBPEL_BITS, sf) & SUBPEL_MASK;
const int y_off_q4 = scaled_y(y << SUBPEL_BITS, sf) & SUBPEL_MASK;
const MV32 res = { scaled_y(mv->row, sf) + y_off_q4,
scaled_x(mv->col, sf) + x_off_q4 };
return res;
}
#if CONFIG_HIGHBITDEPTH
void av1_setup_scale_factors_for_frame(struct scale_factors *sf, int other_w,
int other_h, int this_w, int this_h,
int use_highbd) {
#else
void av1_setup_scale_factors_for_frame(struct scale_factors *sf, int other_w,
int other_h, int this_w, int this_h) {
#endif
if (!valid_ref_frame_size(other_w, other_h, this_w, this_h)) {
sf->x_scale_fp = REF_INVALID_SCALE;
sf->y_scale_fp = REF_INVALID_SCALE;
return;
}
sf->x_scale_fp = get_fixed_point_scale_factor(other_w, this_w);
sf->y_scale_fp = get_fixed_point_scale_factor(other_h, this_h);
sf->x_step_q4 = scaled_x(16, sf);
sf->y_step_q4 = scaled_y(16, sf);
if (av1_is_scaled(sf)) {
sf->scale_value_x = scaled_x;
sf->scale_value_y = scaled_y;
} else {
sf->scale_value_x = unscaled_value;
sf->scale_value_y = unscaled_value;
}
// TODO(agrange): Investigate the best choice of functions to use here
// for EIGHTTAP_SMOOTH. Since it is not interpolating, need to choose what
// to do at full-pel offsets. The current selection, where the filter is
// applied in one direction only, and not at all for 0,0, seems to give the
// best quality, but it may be worth trying an additional mode that does
// do the filtering on full-pel.
if (sf->x_step_q4 == 16) {
if (sf->y_step_q4 == 16) {
// No scaling in either direction.
sf->predict[0][0][0] = aom_convolve_copy;
sf->predict[0][0][1] = aom_convolve_avg;
sf->predict[0][1][0] = aom_convolve8_vert;
sf->predict[0][1][1] = aom_convolve8_avg_vert;
sf->predict[1][0][0] = aom_convolve8_horiz;
sf->predict[1][0][1] = aom_convolve8_avg_horiz;
} else {
// No scaling in x direction. Must always scale in the y direction.
sf->predict[0][0][0] = aom_convolve8_vert;
sf->predict[0][0][1] = aom_convolve8_avg_vert;
sf->predict[0][1][0] = aom_convolve8_vert;
sf->predict[0][1][1] = aom_convolve8_avg_vert;
sf->predict[1][0][0] = aom_convolve8;
sf->predict[1][0][1] = aom_convolve8_avg;
}
} else {
if (sf->y_step_q4 == 16) {
// No scaling in the y direction. Must always scale in the x direction.
sf->predict[0][0][0] = aom_convolve8_horiz;
sf->predict[0][0][1] = aom_convolve8_avg_horiz;
sf->predict[0][1][0] = aom_convolve8;
sf->predict[0][1][1] = aom_convolve8_avg;
sf->predict[1][0][0] = aom_convolve8_horiz;
sf->predict[1][0][1] = aom_convolve8_avg_horiz;
} else {
// Must always scale in both directions.
sf->predict[0][0][0] = aom_convolve8;
sf->predict[0][0][1] = aom_convolve8_avg;
sf->predict[0][1][0] = aom_convolve8;
sf->predict[0][1][1] = aom_convolve8_avg;
sf->predict[1][0][0] = aom_convolve8;
sf->predict[1][0][1] = aom_convolve8_avg;
}
}
// 2D subpel motion always gets filtered in both directions
sf->predict[1][1][0] = aom_convolve8;
sf->predict[1][1][1] = aom_convolve8_avg;
#if CONFIG_HIGHBITDEPTH
if (use_highbd) {
if (sf->x_step_q4 == 16) {
if (sf->y_step_q4 == 16) {
// No scaling in either direction.
sf->highbd_predict[0][0][0] = aom_highbd_convolve_copy;
sf->highbd_predict[0][0][1] = aom_highbd_convolve_avg;
sf->highbd_predict[0][1][0] = aom_highbd_convolve8_vert;
sf->highbd_predict[0][1][1] = aom_highbd_convolve8_avg_vert;
sf->highbd_predict[1][0][0] = aom_highbd_convolve8_horiz;
sf->highbd_predict[1][0][1] = aom_highbd_convolve8_avg_horiz;
} else {
// No scaling in x direction. Must always scale in the y direction.
sf->highbd_predict[0][0][0] = aom_highbd_convolve8_vert;
sf->highbd_predict[0][0][1] = aom_highbd_convolve8_avg_vert;
sf->highbd_predict[0][1][0] = aom_highbd_convolve8_vert;
sf->highbd_predict[0][1][1] = aom_highbd_convolve8_avg_vert;
sf->highbd_predict[1][0][0] = aom_highbd_convolve8;
sf->highbd_predict[1][0][1] = aom_highbd_convolve8_avg;
}
} else {
if (sf->y_step_q4 == 16) {
// No scaling in the y direction. Must always scale in the x direction.
sf->highbd_predict[0][0][0] = aom_highbd_convolve8_horiz;
sf->highbd_predict[0][0][1] = aom_highbd_convolve8_avg_horiz;
sf->highbd_predict[0][1][0] = aom_highbd_convolve8;
sf->highbd_predict[0][1][1] = aom_highbd_convolve8_avg;
sf->highbd_predict[1][0][0] = aom_highbd_convolve8_horiz;
sf->highbd_predict[1][0][1] = aom_highbd_convolve8_avg_horiz;
} else {
// Must always scale in both directions.
sf->highbd_predict[0][0][0] = aom_highbd_convolve8;
sf->highbd_predict[0][0][1] = aom_highbd_convolve8_avg;
sf->highbd_predict[0][1][0] = aom_highbd_convolve8;
sf->highbd_predict[0][1][1] = aom_highbd_convolve8_avg;
sf->highbd_predict[1][0][0] = aom_highbd_convolve8;
sf->highbd_predict[1][0][1] = aom_highbd_convolve8_avg;
}
}
// 2D subpel motion always gets filtered in both directions.
sf->highbd_predict[1][1][0] = aom_highbd_convolve8;
sf->highbd_predict[1][1][1] = aom_highbd_convolve8_avg;
}
#endif // CONFIG_HIGHBITDEPTH
}

72
third_party/aom/av1/common/scale.h vendored Normal file
View file

@ -0,0 +1,72 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_SCALE_H_
#define AV1_COMMON_SCALE_H_
#include "av1/common/mv.h"
#include "aom_dsp/aom_convolve.h"
#ifdef __cplusplus
extern "C" {
#endif
#define REF_SCALE_SHIFT 14
#define REF_NO_SCALE (1 << REF_SCALE_SHIFT)
#define REF_INVALID_SCALE -1
struct scale_factors {
int x_scale_fp; // horizontal fixed point scale factor
int y_scale_fp; // vertical fixed point scale factor
int x_step_q4;
int y_step_q4;
int (*scale_value_x)(int val, const struct scale_factors *sf);
int (*scale_value_y)(int val, const struct scale_factors *sf);
convolve_fn_t predict[2][2][2]; // horiz, vert, avg
#if CONFIG_HIGHBITDEPTH
highbd_convolve_fn_t highbd_predict[2][2][2]; // horiz, vert, avg
#endif // CONFIG_HIGHBITDEPTH
};
MV32 av1_scale_mv(const MV *mv, int x, int y, const struct scale_factors *sf);
#if CONFIG_HIGHBITDEPTH
void av1_setup_scale_factors_for_frame(struct scale_factors *sf, int other_w,
int other_h, int this_w, int this_h,
int use_high);
#else
void av1_setup_scale_factors_for_frame(struct scale_factors *sf, int other_w,
int other_h, int this_w, int this_h);
#endif // CONFIG_HIGHBITDEPTH
static INLINE int av1_is_valid_scale(const struct scale_factors *sf) {
return sf->x_scale_fp != REF_INVALID_SCALE &&
sf->y_scale_fp != REF_INVALID_SCALE;
}
static INLINE int av1_is_scaled(const struct scale_factors *sf) {
return av1_is_valid_scale(sf) &&
(sf->x_scale_fp != REF_NO_SCALE || sf->y_scale_fp != REF_NO_SCALE);
}
static INLINE int valid_ref_frame_size(int ref_width, int ref_height,
int this_width, int this_height) {
return 2 * this_width >= ref_width && 2 * this_height >= ref_height &&
this_width <= 16 * ref_width && this_height <= 16 * ref_height;
}
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_SCALE_H_

6853
third_party/aom/av1/common/scan.c vendored Normal file
View file

File diff suppressed because it is too large Load diff

95
third_party/aom/av1/common/scan.h vendored Normal file
View file

@ -0,0 +1,95 @@
/*
* Copyright (c) 2016, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#ifndef AV1_COMMON_SCAN_H_
#define AV1_COMMON_SCAN_H_
#include "aom/aom_integer.h"
#include "aom_ports/mem.h"
#include "av1/common/enums.h"
#include "av1/common/onyxc_int.h"
#include "av1/common/blockd.h"
#ifdef __cplusplus
extern "C" {
#endif
#define MAX_NEIGHBORS 2
extern const SCAN_ORDER av1_default_scan_orders[TX_SIZES];
extern const SCAN_ORDER av1_intra_scan_orders[TX_SIZES_ALL][TX_TYPES];
extern const SCAN_ORDER av1_inter_scan_orders[TX_SIZES_ALL][TX_TYPES];
#if CONFIG_ADAPT_SCAN
void av1_update_scan_count_facade(AV1_COMMON *cm, FRAME_COUNTS *counts,
TX_SIZE tx_size, TX_TYPE tx_type,
const tran_low_t *dqcoeffs, int max_scan);
// embed r + c and coeff_idx info with nonzero probabilities. When sorting the
// nonzero probabilities, if there is a tie, the coefficient with smaller r + c
// will be scanned first
void av1_augment_prob(TX_SIZE tx_size, TX_TYPE tx_type, uint32_t *prob);
// apply quick sort on nonzero probabilities to obtain a sort order
void av1_update_sort_order(TX_SIZE tx_size, TX_TYPE tx_type,
const uint32_t *non_zero_prob, int16_t *sort_order);
// apply topological sort on the nonzero probabilities sorting order to
// guarantee each to-be-scanned coefficient's upper and left coefficient will be
// scanned before the to-be-scanned coefficient.
void av1_update_scan_order(TX_SIZE tx_size, int16_t *sort_order, int16_t *scan,
int16_t *iscan);
// For each coeff_idx in scan[], update its above and left neighbors in
// neighbors[] accordingly.
void av1_update_neighbors(int tx_size, const int16_t *scan,
const int16_t *iscan, int16_t *neighbors);
void av1_init_scan_order(AV1_COMMON *cm);
void av1_adapt_scan_order(AV1_COMMON *cm);
#endif
void av1_deliver_eob_threshold(const AV1_COMMON *cm, MACROBLOCKD *xd);
static INLINE int get_coef_context(const int16_t *neighbors,
const uint8_t *token_cache, int c) {
return (1 + token_cache[neighbors[MAX_NEIGHBORS * c + 0]] +
token_cache[neighbors[MAX_NEIGHBORS * c + 1]]) >>
1;
}
static INLINE const SCAN_ORDER *get_default_scan(TX_SIZE tx_size,
TX_TYPE tx_type,
int is_inter) {
#if CONFIG_EXT_TX || CONFIG_VAR_TX
return is_inter ? &av1_inter_scan_orders[tx_size][tx_type]
: &av1_intra_scan_orders[tx_size][tx_type];
#else
(void)is_inter;
return &av1_intra_scan_orders[tx_size][tx_type];
#endif // CONFIG_EXT_TX
}
static INLINE const SCAN_ORDER *get_scan(const AV1_COMMON *cm, TX_SIZE tx_size,
TX_TYPE tx_type, int is_inter) {
#if CONFIG_ADAPT_SCAN
(void)is_inter;
return &cm->fc->sc[tx_size][tx_type];
#else // CONFIG_ADAPT_SCAN
(void)cm;
return get_default_scan(tx_size, tx_type, is_inter);
#endif // CONFIG_ADAPT_SCAN
}
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_COMMON_SCAN_H_

Some files were not shown because too many files have changed in this diff Show more