mirror of
https://repo.dactyloidae.xyz/Dactyloidae/UXP.git
synced 2026-10-06 07:17:30 +09:00
Update aom to slightly newer commit ID
This commit is contained in:
trav90
Roy Tam
parent
1c8af26369
commit
76f1e6edca
311 changed files with 55292 additions and 33790 deletions
38
third_party/aom/aom_dsp/arm/avg_neon.c
vendored
38
third_party/aom/aom_dsp/arm/avg_neon.c
vendored
|
|
@ -25,44 +25,6 @@ static INLINE unsigned int horizontal_add_u16x8(const uint16x8_t v_16x8) {
|
|||
return vget_lane_u32(c, 0);
|
||||
}
|
||||
|
||||
unsigned int aom_avg_4x4_neon(const uint8_t *s, int p) {
|
||||
uint16x8_t v_sum;
|
||||
uint32x2_t v_s0 = vdup_n_u32(0);
|
||||
uint32x2_t v_s1 = vdup_n_u32(0);
|
||||
v_s0 = vld1_lane_u32((const uint32_t *)s, v_s0, 0);
|
||||
v_s0 = vld1_lane_u32((const uint32_t *)(s + p), v_s0, 1);
|
||||
v_s1 = vld1_lane_u32((const uint32_t *)(s + 2 * p), v_s1, 0);
|
||||
v_s1 = vld1_lane_u32((const uint32_t *)(s + 3 * p), v_s1, 1);
|
||||
v_sum = vaddl_u8(vreinterpret_u8_u32(v_s0), vreinterpret_u8_u32(v_s1));
|
||||
return (horizontal_add_u16x8(v_sum) + 8) >> 4;
|
||||
}
|
||||
|
||||
unsigned int aom_avg_8x8_neon(const uint8_t *s, int p) {
|
||||
uint8x8_t v_s0 = vld1_u8(s);
|
||||
const uint8x8_t v_s1 = vld1_u8(s + p);
|
||||
uint16x8_t v_sum = vaddl_u8(v_s0, v_s1);
|
||||
|
||||
v_s0 = vld1_u8(s + 2 * p);
|
||||
v_sum = vaddw_u8(v_sum, v_s0);
|
||||
|
||||
v_s0 = vld1_u8(s + 3 * p);
|
||||
v_sum = vaddw_u8(v_sum, v_s0);
|
||||
|
||||
v_s0 = vld1_u8(s + 4 * p);
|
||||
v_sum = vaddw_u8(v_sum, v_s0);
|
||||
|
||||
v_s0 = vld1_u8(s + 5 * p);
|
||||
v_sum = vaddw_u8(v_sum, v_s0);
|
||||
|
||||
v_s0 = vld1_u8(s + 6 * p);
|
||||
v_sum = vaddw_u8(v_sum, v_s0);
|
||||
|
||||
v_s0 = vld1_u8(s + 7 * p);
|
||||
v_sum = vaddw_u8(v_sum, v_s0);
|
||||
|
||||
return (horizontal_add_u16x8(v_sum) + 32) >> 6;
|
||||
}
|
||||
|
||||
// coeff: 16 bits, dynamic range [-32640, 32640].
|
||||
// length: value range {16, 64, 256, 1024}.
|
||||
int aom_satd_neon(const int16_t *coeff, int length) {
|
||||
|
|
|
|||
|
|
@ -1,240 +0,0 @@
|
|||
;
|
||||
; 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.
|
||||
;
|
||||
|
||||
;
|
||||
|
||||
|
||||
EXPORT |aom_filter_block2d_bil_first_pass_media|
|
||||
EXPORT |aom_filter_block2d_bil_second_pass_media|
|
||||
|
||||
AREA |.text|, CODE, READONLY ; name this block of code
|
||||
|
||||
;-------------------------------------
|
||||
; r0 unsigned char *src_ptr,
|
||||
; r1 unsigned short *dst_ptr,
|
||||
; r2 unsigned int src_pitch,
|
||||
; r3 unsigned int height,
|
||||
; stack unsigned int width,
|
||||
; stack const short *aom_filter
|
||||
;-------------------------------------
|
||||
; The output is transposed stroed in output array to make it easy for second pass filtering.
|
||||
|aom_filter_block2d_bil_first_pass_media| PROC
|
||||
stmdb sp!, {r4 - r11, lr}
|
||||
|
||||
ldr r11, [sp, #40] ; aom_filter address
|
||||
ldr r4, [sp, #36] ; width
|
||||
|
||||
mov r12, r3 ; outer-loop counter
|
||||
|
||||
add r7, r2, r4 ; preload next row
|
||||
pld [r0, r7]
|
||||
|
||||
sub r2, r2, r4 ; src increment for height loop
|
||||
|
||||
ldr r5, [r11] ; load up filter coefficients
|
||||
|
||||
mov r3, r3, lsl #1 ; height*2
|
||||
add r3, r3, #2 ; plus 2 to make output buffer 4-bit aligned since height is actually (height+1)
|
||||
|
||||
mov r11, r1 ; save dst_ptr for each row
|
||||
|
||||
cmp r5, #128 ; if filter coef = 128, then skip the filter
|
||||
beq bil_null_1st_filter
|
||||
|
||||
|bil_height_loop_1st_v6|
|
||||
ldrb r6, [r0] ; load source data
|
||||
ldrb r7, [r0, #1]
|
||||
ldrb r8, [r0, #2]
|
||||
mov lr, r4, lsr #2 ; 4-in-parellel loop counter
|
||||
|
||||
|bil_width_loop_1st_v6|
|
||||
ldrb r9, [r0, #3]
|
||||
ldrb r10, [r0, #4]
|
||||
|
||||
pkhbt r6, r6, r7, lsl #16 ; src[1] | src[0]
|
||||
pkhbt r7, r7, r8, lsl #16 ; src[2] | src[1]
|
||||
|
||||
smuad r6, r6, r5 ; apply the filter
|
||||
pkhbt r8, r8, r9, lsl #16 ; src[3] | src[2]
|
||||
smuad r7, r7, r5
|
||||
pkhbt r9, r9, r10, lsl #16 ; src[4] | src[3]
|
||||
|
||||
smuad r8, r8, r5
|
||||
smuad r9, r9, r5
|
||||
|
||||
add r0, r0, #4
|
||||
subs lr, lr, #1
|
||||
|
||||
add r6, r6, #0x40 ; round_shift_and_clamp
|
||||
add r7, r7, #0x40
|
||||
usat r6, #16, r6, asr #7
|
||||
usat r7, #16, r7, asr #7
|
||||
|
||||
strh r6, [r1], r3 ; result is transposed and stored
|
||||
|
||||
add r8, r8, #0x40 ; round_shift_and_clamp
|
||||
strh r7, [r1], r3
|
||||
add r9, r9, #0x40
|
||||
usat r8, #16, r8, asr #7
|
||||
usat r9, #16, r9, asr #7
|
||||
|
||||
strh r8, [r1], r3 ; result is transposed and stored
|
||||
|
||||
ldrneb r6, [r0] ; load source data
|
||||
strh r9, [r1], r3
|
||||
|
||||
ldrneb r7, [r0, #1]
|
||||
ldrneb r8, [r0, #2]
|
||||
|
||||
bne bil_width_loop_1st_v6
|
||||
|
||||
add r0, r0, r2 ; move to next input row
|
||||
subs r12, r12, #1
|
||||
|
||||
add r9, r2, r4, lsl #1 ; adding back block width
|
||||
pld [r0, r9] ; preload next row
|
||||
|
||||
add r11, r11, #2 ; move over to next column
|
||||
mov r1, r11
|
||||
|
||||
bne bil_height_loop_1st_v6
|
||||
|
||||
ldmia sp!, {r4 - r11, pc}
|
||||
|
||||
|bil_null_1st_filter|
|
||||
|bil_height_loop_null_1st|
|
||||
mov lr, r4, lsr #2 ; loop counter
|
||||
|
||||
|bil_width_loop_null_1st|
|
||||
ldrb r6, [r0] ; load data
|
||||
ldrb r7, [r0, #1]
|
||||
ldrb r8, [r0, #2]
|
||||
ldrb r9, [r0, #3]
|
||||
|
||||
strh r6, [r1], r3 ; store it to immediate buffer
|
||||
add r0, r0, #4
|
||||
strh r7, [r1], r3
|
||||
subs lr, lr, #1
|
||||
strh r8, [r1], r3
|
||||
strh r9, [r1], r3
|
||||
|
||||
bne bil_width_loop_null_1st
|
||||
|
||||
subs r12, r12, #1
|
||||
add r0, r0, r2 ; move to next input line
|
||||
add r11, r11, #2 ; move over to next column
|
||||
mov r1, r11
|
||||
|
||||
bne bil_height_loop_null_1st
|
||||
|
||||
ldmia sp!, {r4 - r11, pc}
|
||||
|
||||
ENDP ; |aom_filter_block2d_bil_first_pass_media|
|
||||
|
||||
|
||||
;---------------------------------
|
||||
; r0 unsigned short *src_ptr,
|
||||
; r1 unsigned char *dst_ptr,
|
||||
; r2 int dst_pitch,
|
||||
; r3 unsigned int height,
|
||||
; stack unsigned int width,
|
||||
; stack const short *aom_filter
|
||||
;---------------------------------
|
||||
|aom_filter_block2d_bil_second_pass_media| PROC
|
||||
stmdb sp!, {r4 - r11, lr}
|
||||
|
||||
ldr r11, [sp, #40] ; aom_filter address
|
||||
ldr r4, [sp, #36] ; width
|
||||
|
||||
ldr r5, [r11] ; load up filter coefficients
|
||||
mov r12, r4 ; outer-loop counter = width, since we work on transposed data matrix
|
||||
mov r11, r1
|
||||
|
||||
cmp r5, #128 ; if filter coef = 128, then skip the filter
|
||||
beq bil_null_2nd_filter
|
||||
|
||||
|bil_height_loop_2nd|
|
||||
ldr r6, [r0] ; load the data
|
||||
ldr r8, [r0, #4]
|
||||
ldrh r10, [r0, #8]
|
||||
mov lr, r3, lsr #2 ; loop counter
|
||||
|
||||
|bil_width_loop_2nd|
|
||||
pkhtb r7, r6, r8 ; src[1] | src[2]
|
||||
pkhtb r9, r8, r10 ; src[3] | src[4]
|
||||
|
||||
smuad r6, r6, r5 ; apply filter
|
||||
smuad r8, r8, r5 ; apply filter
|
||||
|
||||
subs lr, lr, #1
|
||||
|
||||
smuadx r7, r7, r5 ; apply filter
|
||||
smuadx r9, r9, r5 ; apply filter
|
||||
|
||||
add r0, r0, #8
|
||||
|
||||
add r6, r6, #0x40 ; round_shift_and_clamp
|
||||
add r7, r7, #0x40
|
||||
usat r6, #8, r6, asr #7
|
||||
usat r7, #8, r7, asr #7
|
||||
strb r6, [r1], r2 ; the result is transposed back and stored
|
||||
|
||||
add r8, r8, #0x40 ; round_shift_and_clamp
|
||||
strb r7, [r1], r2
|
||||
add r9, r9, #0x40
|
||||
usat r8, #8, r8, asr #7
|
||||
usat r9, #8, r9, asr #7
|
||||
strb r8, [r1], r2 ; the result is transposed back and stored
|
||||
|
||||
ldrne r6, [r0] ; load data
|
||||
strb r9, [r1], r2
|
||||
ldrne r8, [r0, #4]
|
||||
ldrneh r10, [r0, #8]
|
||||
|
||||
bne bil_width_loop_2nd
|
||||
|
||||
subs r12, r12, #1
|
||||
add r0, r0, #4 ; update src for next row
|
||||
add r11, r11, #1
|
||||
mov r1, r11
|
||||
|
||||
bne bil_height_loop_2nd
|
||||
ldmia sp!, {r4 - r11, pc}
|
||||
|
||||
|bil_null_2nd_filter|
|
||||
|bil_height_loop_null_2nd|
|
||||
mov lr, r3, lsr #2
|
||||
|
||||
|bil_width_loop_null_2nd|
|
||||
ldr r6, [r0], #4 ; load data
|
||||
subs lr, lr, #1
|
||||
ldr r8, [r0], #4
|
||||
|
||||
strb r6, [r1], r2 ; store data
|
||||
mov r7, r6, lsr #16
|
||||
strb r7, [r1], r2
|
||||
mov r9, r8, lsr #16
|
||||
strb r8, [r1], r2
|
||||
strb r9, [r1], r2
|
||||
|
||||
bne bil_width_loop_null_2nd
|
||||
|
||||
subs r12, r12, #1
|
||||
add r0, r0, #4
|
||||
add r11, r11, #1
|
||||
mov r1, r11
|
||||
|
||||
bne bil_height_loop_null_2nd
|
||||
|
||||
ldmia sp!, {r4 - r11, pc}
|
||||
ENDP ; |aom_filter_block2d_second_pass_media|
|
||||
|
||||
END
|
||||
98
third_party/aom/aom_dsp/arm/sad_media.asm
vendored
98
third_party/aom/aom_dsp/arm/sad_media.asm
vendored
|
|
@ -1,98 +0,0 @@
|
|||
;
|
||||
; 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.
|
||||
;
|
||||
|
||||
;
|
||||
|
||||
|
||||
EXPORT |aom_sad16x16_media|
|
||||
|
||||
ARM
|
||||
REQUIRE8
|
||||
PRESERVE8
|
||||
|
||||
AREA ||.text||, CODE, READONLY, ALIGN=2
|
||||
|
||||
; r0 const unsigned char *src_ptr
|
||||
; r1 int src_stride
|
||||
; r2 const unsigned char *ref_ptr
|
||||
; r3 int ref_stride
|
||||
|aom_sad16x16_media| PROC
|
||||
stmfd sp!, {r4-r12, lr}
|
||||
|
||||
pld [r0, r1, lsl #0]
|
||||
pld [r2, r3, lsl #0]
|
||||
pld [r0, r1, lsl #1]
|
||||
pld [r2, r3, lsl #1]
|
||||
|
||||
mov r4, #0 ; sad = 0;
|
||||
mov r5, #8 ; loop count
|
||||
|
||||
loop
|
||||
; 1st row
|
||||
ldr r6, [r0, #0x0] ; load 4 src pixels (1A)
|
||||
ldr r8, [r2, #0x0] ; load 4 ref pixels (1A)
|
||||
ldr r7, [r0, #0x4] ; load 4 src pixels (1A)
|
||||
ldr r9, [r2, #0x4] ; load 4 ref pixels (1A)
|
||||
ldr r10, [r0, #0x8] ; load 4 src pixels (1B)
|
||||
ldr r11, [r0, #0xC] ; load 4 src pixels (1B)
|
||||
|
||||
usada8 r4, r8, r6, r4 ; calculate sad for 4 pixels
|
||||
usad8 r8, r7, r9 ; calculate sad for 4 pixels
|
||||
|
||||
ldr r12, [r2, #0x8] ; load 4 ref pixels (1B)
|
||||
ldr lr, [r2, #0xC] ; load 4 ref pixels (1B)
|
||||
|
||||
add r0, r0, r1 ; set src pointer to next row
|
||||
add r2, r2, r3 ; set dst pointer to next row
|
||||
|
||||
pld [r0, r1, lsl #1]
|
||||
pld [r2, r3, lsl #1]
|
||||
|
||||
usada8 r4, r10, r12, r4 ; calculate sad for 4 pixels
|
||||
usada8 r8, r11, lr, r8 ; calculate sad for 4 pixels
|
||||
|
||||
ldr r6, [r0, #0x0] ; load 4 src pixels (2A)
|
||||
ldr r7, [r0, #0x4] ; load 4 src pixels (2A)
|
||||
add r4, r4, r8 ; add partial sad values
|
||||
|
||||
; 2nd row
|
||||
ldr r8, [r2, #0x0] ; load 4 ref pixels (2A)
|
||||
ldr r9, [r2, #0x4] ; load 4 ref pixels (2A)
|
||||
ldr r10, [r0, #0x8] ; load 4 src pixels (2B)
|
||||
ldr r11, [r0, #0xC] ; load 4 src pixels (2B)
|
||||
|
||||
usada8 r4, r6, r8, r4 ; calculate sad for 4 pixels
|
||||
usad8 r8, r7, r9 ; calculate sad for 4 pixels
|
||||
|
||||
ldr r12, [r2, #0x8] ; load 4 ref pixels (2B)
|
||||
ldr lr, [r2, #0xC] ; load 4 ref pixels (2B)
|
||||
|
||||
add r0, r0, r1 ; set src pointer to next row
|
||||
add r2, r2, r3 ; set dst pointer to next row
|
||||
|
||||
usada8 r4, r10, r12, r4 ; calculate sad for 4 pixels
|
||||
usada8 r8, r11, lr, r8 ; calculate sad for 4 pixels
|
||||
|
||||
pld [r0, r1, lsl #1]
|
||||
pld [r2, r3, lsl #1]
|
||||
|
||||
subs r5, r5, #1 ; decrement loop counter
|
||||
add r4, r4, r8 ; add partial sad values
|
||||
|
||||
bne loop
|
||||
|
||||
mov r0, r4 ; return sad
|
||||
ldmfd sp!, {r4-r12, pc}
|
||||
|
||||
ENDP
|
||||
|
||||
END
|
||||
|
||||
|
|
@ -1,81 +0,0 @@
|
|||
/*
|
||||
* 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_dsp_rtcd.h"
|
||||
#include "aom/aom_integer.h"
|
||||
#include "aom_ports/mem.h"
|
||||
|
||||
#if HAVE_MEDIA
|
||||
static const int16_t bilinear_filters_media[8][2] = { { 128, 0 }, { 112, 16 },
|
||||
{ 96, 32 }, { 80, 48 },
|
||||
{ 64, 64 }, { 48, 80 },
|
||||
{ 32, 96 }, { 16, 112 } };
|
||||
|
||||
extern void aom_filter_block2d_bil_first_pass_media(
|
||||
const uint8_t *src_ptr, uint16_t *dst_ptr, uint32_t src_pitch,
|
||||
uint32_t height, uint32_t width, const int16_t *filter);
|
||||
|
||||
extern void aom_filter_block2d_bil_second_pass_media(
|
||||
const uint16_t *src_ptr, uint8_t *dst_ptr, int32_t src_pitch,
|
||||
uint32_t height, uint32_t width, const int16_t *filter);
|
||||
|
||||
unsigned int aom_sub_pixel_variance8x8_media(
|
||||
const uint8_t *src_ptr, int src_pixels_per_line, int xoffset, int yoffset,
|
||||
const uint8_t *dst_ptr, int dst_pixels_per_line, unsigned int *sse) {
|
||||
uint16_t first_pass[10 * 8];
|
||||
uint8_t second_pass[8 * 8];
|
||||
const int16_t *HFilter, *VFilter;
|
||||
|
||||
HFilter = bilinear_filters_media[xoffset];
|
||||
VFilter = bilinear_filters_media[yoffset];
|
||||
|
||||
aom_filter_block2d_bil_first_pass_media(src_ptr, first_pass,
|
||||
src_pixels_per_line, 9, 8, HFilter);
|
||||
aom_filter_block2d_bil_second_pass_media(first_pass, second_pass, 8, 8, 8,
|
||||
VFilter);
|
||||
|
||||
return aom_variance8x8_media(second_pass, 8, dst_ptr, dst_pixels_per_line,
|
||||
sse);
|
||||
}
|
||||
|
||||
unsigned int aom_sub_pixel_variance16x16_media(
|
||||
const uint8_t *src_ptr, int src_pixels_per_line, int xoffset, int yoffset,
|
||||
const uint8_t *dst_ptr, int dst_pixels_per_line, unsigned int *sse) {
|
||||
uint16_t first_pass[36 * 16];
|
||||
uint8_t second_pass[20 * 16];
|
||||
const int16_t *HFilter, *VFilter;
|
||||
unsigned int var;
|
||||
|
||||
if (xoffset == 4 && yoffset == 0) {
|
||||
var = aom_variance_halfpixvar16x16_h_media(
|
||||
src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, sse);
|
||||
} else if (xoffset == 0 && yoffset == 4) {
|
||||
var = aom_variance_halfpixvar16x16_v_media(
|
||||
src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, sse);
|
||||
} else if (xoffset == 4 && yoffset == 4) {
|
||||
var = aom_variance_halfpixvar16x16_hv_media(
|
||||
src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, sse);
|
||||
} else {
|
||||
HFilter = bilinear_filters_media[xoffset];
|
||||
VFilter = bilinear_filters_media[yoffset];
|
||||
|
||||
aom_filter_block2d_bil_first_pass_media(
|
||||
src_ptr, first_pass, src_pixels_per_line, 17, 16, HFilter);
|
||||
aom_filter_block2d_bil_second_pass_media(first_pass, second_pass, 16, 16,
|
||||
16, VFilter);
|
||||
|
||||
var = aom_variance16x16_media(second_pass, 16, dst_ptr, dst_pixels_per_line,
|
||||
sse);
|
||||
}
|
||||
return var;
|
||||
}
|
||||
#endif // HAVE_MEDIA
|
||||
|
|
@ -1,185 +0,0 @@
|
|||
;
|
||||
; 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.
|
||||
;
|
||||
|
||||
;
|
||||
|
||||
|
||||
EXPORT |aom_variance_halfpixvar16x16_h_media|
|
||||
|
||||
ARM
|
||||
REQUIRE8
|
||||
PRESERVE8
|
||||
|
||||
AREA ||.text||, CODE, READONLY, ALIGN=2
|
||||
|
||||
; r0 unsigned char *src_ptr
|
||||
; r1 int source_stride
|
||||
; r2 unsigned char *ref_ptr
|
||||
; r3 int recon_stride
|
||||
; stack unsigned int *sse
|
||||
|aom_variance_halfpixvar16x16_h_media| PROC
|
||||
|
||||
stmfd sp!, {r4-r12, lr}
|
||||
|
||||
pld [r0, r1, lsl #0]
|
||||
pld [r2, r3, lsl #0]
|
||||
|
||||
mov r8, #0 ; initialize sum = 0
|
||||
ldr r10, c80808080
|
||||
mov r11, #0 ; initialize sse = 0
|
||||
mov r12, #16 ; set loop counter to 16 (=block height)
|
||||
mov lr, #0 ; constant zero
|
||||
loop
|
||||
; 1st 4 pixels
|
||||
ldr r4, [r0, #0] ; load 4 src pixels
|
||||
ldr r6, [r0, #1] ; load 4 src pixels with 1 byte offset
|
||||
ldr r5, [r2, #0] ; load 4 ref pixels
|
||||
|
||||
; bilinear interpolation
|
||||
mvn r6, r6
|
||||
uhsub8 r4, r4, r6
|
||||
eor r4, r4, r10
|
||||
|
||||
usub8 r6, r4, r5 ; calculate difference
|
||||
pld [r0, r1, lsl #1]
|
||||
sel r7, r6, lr ; select bytes with positive difference
|
||||
usub8 r6, r5, r4 ; calculate difference with reversed operands
|
||||
pld [r2, r3, lsl #1]
|
||||
sel r6, r6, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r4, r7, lr ; calculate sum of positive differences
|
||||
usad8 r5, r6, lr ; calculate sum of negative differences
|
||||
orr r6, r6, r7 ; differences of all 4 pixels
|
||||
; calculate total sum
|
||||
adds r8, r8, r4 ; add positive differences to sum
|
||||
subs r8, r8, r5 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r5, r6 ; byte (two pixels) to halfwords
|
||||
uxtb16 r7, r6, ror #8 ; another two pixels to halfwords
|
||||
smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1)
|
||||
|
||||
; 2nd 4 pixels
|
||||
ldr r4, [r0, #4] ; load 4 src pixels
|
||||
ldr r6, [r0, #5] ; load 4 src pixels with 1 byte offset
|
||||
ldr r5, [r2, #4] ; load 4 ref pixels
|
||||
|
||||
; bilinear interpolation
|
||||
mvn r6, r6
|
||||
uhsub8 r4, r4, r6
|
||||
eor r4, r4, r10
|
||||
|
||||
smlad r11, r7, r7, r11 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
usub8 r6, r4, r5 ; calculate difference
|
||||
sel r7, r6, lr ; select bytes with positive difference
|
||||
usub8 r6, r5, r4 ; calculate difference with reversed operands
|
||||
sel r6, r6, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r4, r7, lr ; calculate sum of positive differences
|
||||
usad8 r5, r6, lr ; calculate sum of negative differences
|
||||
orr r6, r6, r7 ; differences of all 4 pixels
|
||||
|
||||
; calculate total sum
|
||||
add r8, r8, r4 ; add positive differences to sum
|
||||
sub r8, r8, r5 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r5, r6 ; byte (two pixels) to halfwords
|
||||
uxtb16 r7, r6, ror #8 ; another two pixels to halfwords
|
||||
smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1)
|
||||
|
||||
; 3rd 4 pixels
|
||||
ldr r4, [r0, #8] ; load 4 src pixels
|
||||
ldr r6, [r0, #9] ; load 4 src pixels with 1 byte offset
|
||||
ldr r5, [r2, #8] ; load 4 ref pixels
|
||||
|
||||
; bilinear interpolation
|
||||
mvn r6, r6
|
||||
uhsub8 r4, r4, r6
|
||||
eor r4, r4, r10
|
||||
|
||||
smlad r11, r7, r7, r11 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
usub8 r6, r4, r5 ; calculate difference
|
||||
sel r7, r6, lr ; select bytes with positive difference
|
||||
usub8 r6, r5, r4 ; calculate difference with reversed operands
|
||||
sel r6, r6, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r4, r7, lr ; calculate sum of positive differences
|
||||
usad8 r5, r6, lr ; calculate sum of negative differences
|
||||
orr r6, r6, r7 ; differences of all 4 pixels
|
||||
|
||||
; calculate total sum
|
||||
add r8, r8, r4 ; add positive differences to sum
|
||||
sub r8, r8, r5 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r5, r6 ; byte (two pixels) to halfwords
|
||||
uxtb16 r7, r6, ror #8 ; another two pixels to halfwords
|
||||
smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1)
|
||||
|
||||
; 4th 4 pixels
|
||||
ldr r4, [r0, #12] ; load 4 src pixels
|
||||
ldr r6, [r0, #13] ; load 4 src pixels with 1 byte offset
|
||||
ldr r5, [r2, #12] ; load 4 ref pixels
|
||||
|
||||
; bilinear interpolation
|
||||
mvn r6, r6
|
||||
uhsub8 r4, r4, r6
|
||||
eor r4, r4, r10
|
||||
|
||||
smlad r11, r7, r7, r11 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
usub8 r6, r4, r5 ; calculate difference
|
||||
add r0, r0, r1 ; set src_ptr to next row
|
||||
sel r7, r6, lr ; select bytes with positive difference
|
||||
usub8 r6, r5, r4 ; calculate difference with reversed operands
|
||||
add r2, r2, r3 ; set dst_ptr to next row
|
||||
sel r6, r6, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r4, r7, lr ; calculate sum of positive differences
|
||||
usad8 r5, r6, lr ; calculate sum of negative differences
|
||||
orr r6, r6, r7 ; differences of all 4 pixels
|
||||
|
||||
; calculate total sum
|
||||
add r8, r8, r4 ; add positive differences to sum
|
||||
sub r8, r8, r5 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r5, r6 ; byte (two pixels) to halfwords
|
||||
uxtb16 r7, r6, ror #8 ; another two pixels to halfwords
|
||||
smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1)
|
||||
smlad r11, r7, r7, r11 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
subs r12, r12, #1
|
||||
|
||||
bne loop
|
||||
|
||||
; return stuff
|
||||
ldr r6, [sp, #40] ; get address of sse
|
||||
mul r0, r8, r8 ; sum * sum
|
||||
str r11, [r6] ; store sse
|
||||
sub r0, r11, r0, lsr #8 ; return (sse - ((sum * sum) >> 8))
|
||||
|
||||
ldmfd sp!, {r4-r12, pc}
|
||||
|
||||
ENDP
|
||||
|
||||
c80808080
|
||||
DCD 0x80808080
|
||||
|
||||
END
|
||||
|
||||
|
|
@ -1,225 +0,0 @@
|
|||
;
|
||||
; 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.
|
||||
;
|
||||
|
||||
;
|
||||
|
||||
|
||||
EXPORT |aom_variance_halfpixvar16x16_hv_media|
|
||||
|
||||
ARM
|
||||
REQUIRE8
|
||||
PRESERVE8
|
||||
|
||||
AREA ||.text||, CODE, READONLY, ALIGN=2
|
||||
|
||||
; r0 unsigned char *src_ptr
|
||||
; r1 int source_stride
|
||||
; r2 unsigned char *ref_ptr
|
||||
; r3 int recon_stride
|
||||
; stack unsigned int *sse
|
||||
|aom_variance_halfpixvar16x16_hv_media| PROC
|
||||
|
||||
stmfd sp!, {r4-r12, lr}
|
||||
|
||||
pld [r0, r1, lsl #0]
|
||||
pld [r2, r3, lsl #0]
|
||||
|
||||
mov r8, #0 ; initialize sum = 0
|
||||
ldr r10, c80808080
|
||||
mov r11, #0 ; initialize sse = 0
|
||||
mov r12, #16 ; set loop counter to 16 (=block height)
|
||||
mov lr, #0 ; constant zero
|
||||
loop
|
||||
add r9, r0, r1 ; pointer to pixels on the next row
|
||||
; 1st 4 pixels
|
||||
ldr r4, [r0, #0] ; load source pixels a, row N
|
||||
ldr r6, [r0, #1] ; load source pixels b, row N
|
||||
ldr r5, [r9, #0] ; load source pixels c, row N+1
|
||||
ldr r7, [r9, #1] ; load source pixels d, row N+1
|
||||
|
||||
; x = (a + b + 1) >> 1, interpolate pixels horizontally on row N
|
||||
mvn r6, r6
|
||||
uhsub8 r4, r4, r6
|
||||
eor r4, r4, r10
|
||||
; y = (c + d + 1) >> 1, interpolate pixels horizontally on row N+1
|
||||
mvn r7, r7
|
||||
uhsub8 r5, r5, r7
|
||||
eor r5, r5, r10
|
||||
; z = (x + y + 1) >> 1, interpolate half pixel values vertically
|
||||
mvn r5, r5
|
||||
uhsub8 r4, r4, r5
|
||||
ldr r5, [r2, #0] ; load 4 ref pixels
|
||||
eor r4, r4, r10
|
||||
|
||||
usub8 r6, r4, r5 ; calculate difference
|
||||
pld [r0, r1, lsl #1]
|
||||
sel r7, r6, lr ; select bytes with positive difference
|
||||
usub8 r6, r5, r4 ; calculate difference with reversed operands
|
||||
pld [r2, r3, lsl #1]
|
||||
sel r6, r6, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r4, r7, lr ; calculate sum of positive differences
|
||||
usad8 r5, r6, lr ; calculate sum of negative differences
|
||||
orr r6, r6, r7 ; differences of all 4 pixels
|
||||
; calculate total sum
|
||||
adds r8, r8, r4 ; add positive differences to sum
|
||||
subs r8, r8, r5 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r5, r6 ; byte (two pixels) to halfwords
|
||||
uxtb16 r7, r6, ror #8 ; another two pixels to halfwords
|
||||
smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1)
|
||||
|
||||
; 2nd 4 pixels
|
||||
ldr r4, [r0, #4] ; load source pixels a, row N
|
||||
ldr r6, [r0, #5] ; load source pixels b, row N
|
||||
ldr r5, [r9, #4] ; load source pixels c, row N+1
|
||||
|
||||
smlad r11, r7, r7, r11 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
ldr r7, [r9, #5] ; load source pixels d, row N+1
|
||||
|
||||
; x = (a + b + 1) >> 1, interpolate pixels horizontally on row N
|
||||
mvn r6, r6
|
||||
uhsub8 r4, r4, r6
|
||||
eor r4, r4, r10
|
||||
; y = (c + d + 1) >> 1, interpolate pixels horizontally on row N+1
|
||||
mvn r7, r7
|
||||
uhsub8 r5, r5, r7
|
||||
eor r5, r5, r10
|
||||
; z = (x + y + 1) >> 1, interpolate half pixel values vertically
|
||||
mvn r5, r5
|
||||
uhsub8 r4, r4, r5
|
||||
ldr r5, [r2, #4] ; load 4 ref pixels
|
||||
eor r4, r4, r10
|
||||
|
||||
usub8 r6, r4, r5 ; calculate difference
|
||||
sel r7, r6, lr ; select bytes with positive difference
|
||||
usub8 r6, r5, r4 ; calculate difference with reversed operands
|
||||
sel r6, r6, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r4, r7, lr ; calculate sum of positive differences
|
||||
usad8 r5, r6, lr ; calculate sum of negative differences
|
||||
orr r6, r6, r7 ; differences of all 4 pixels
|
||||
|
||||
; calculate total sum
|
||||
add r8, r8, r4 ; add positive differences to sum
|
||||
sub r8, r8, r5 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r5, r6 ; byte (two pixels) to halfwords
|
||||
uxtb16 r7, r6, ror #8 ; another two pixels to halfwords
|
||||
smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1)
|
||||
|
||||
; 3rd 4 pixels
|
||||
ldr r4, [r0, #8] ; load source pixels a, row N
|
||||
ldr r6, [r0, #9] ; load source pixels b, row N
|
||||
ldr r5, [r9, #8] ; load source pixels c, row N+1
|
||||
|
||||
smlad r11, r7, r7, r11 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
ldr r7, [r9, #9] ; load source pixels d, row N+1
|
||||
|
||||
; x = (a + b + 1) >> 1, interpolate pixels horizontally on row N
|
||||
mvn r6, r6
|
||||
uhsub8 r4, r4, r6
|
||||
eor r4, r4, r10
|
||||
; y = (c + d + 1) >> 1, interpolate pixels horizontally on row N+1
|
||||
mvn r7, r7
|
||||
uhsub8 r5, r5, r7
|
||||
eor r5, r5, r10
|
||||
; z = (x + y + 1) >> 1, interpolate half pixel values vertically
|
||||
mvn r5, r5
|
||||
uhsub8 r4, r4, r5
|
||||
ldr r5, [r2, #8] ; load 4 ref pixels
|
||||
eor r4, r4, r10
|
||||
|
||||
usub8 r6, r4, r5 ; calculate difference
|
||||
sel r7, r6, lr ; select bytes with positive difference
|
||||
usub8 r6, r5, r4 ; calculate difference with reversed operands
|
||||
sel r6, r6, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r4, r7, lr ; calculate sum of positive differences
|
||||
usad8 r5, r6, lr ; calculate sum of negative differences
|
||||
orr r6, r6, r7 ; differences of all 4 pixels
|
||||
|
||||
; calculate total sum
|
||||
add r8, r8, r4 ; add positive differences to sum
|
||||
sub r8, r8, r5 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r5, r6 ; byte (two pixels) to halfwords
|
||||
uxtb16 r7, r6, ror #8 ; another two pixels to halfwords
|
||||
smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1)
|
||||
|
||||
; 4th 4 pixels
|
||||
ldr r4, [r0, #12] ; load source pixels a, row N
|
||||
ldr r6, [r0, #13] ; load source pixels b, row N
|
||||
ldr r5, [r9, #12] ; load source pixels c, row N+1
|
||||
smlad r11, r7, r7, r11 ; dual signed multiply, add and accumulate (2)
|
||||
ldr r7, [r9, #13] ; load source pixels d, row N+1
|
||||
|
||||
; x = (a + b + 1) >> 1, interpolate pixels horizontally on row N
|
||||
mvn r6, r6
|
||||
uhsub8 r4, r4, r6
|
||||
eor r4, r4, r10
|
||||
; y = (c + d + 1) >> 1, interpolate pixels horizontally on row N+1
|
||||
mvn r7, r7
|
||||
uhsub8 r5, r5, r7
|
||||
eor r5, r5, r10
|
||||
; z = (x + y + 1) >> 1, interpolate half pixel values vertically
|
||||
mvn r5, r5
|
||||
uhsub8 r4, r4, r5
|
||||
ldr r5, [r2, #12] ; load 4 ref pixels
|
||||
eor r4, r4, r10
|
||||
|
||||
usub8 r6, r4, r5 ; calculate difference
|
||||
add r0, r0, r1 ; set src_ptr to next row
|
||||
sel r7, r6, lr ; select bytes with positive difference
|
||||
usub8 r6, r5, r4 ; calculate difference with reversed operands
|
||||
add r2, r2, r3 ; set dst_ptr to next row
|
||||
sel r6, r6, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r4, r7, lr ; calculate sum of positive differences
|
||||
usad8 r5, r6, lr ; calculate sum of negative differences
|
||||
orr r6, r6, r7 ; differences of all 4 pixels
|
||||
|
||||
; calculate total sum
|
||||
add r8, r8, r4 ; add positive differences to sum
|
||||
sub r8, r8, r5 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r5, r6 ; byte (two pixels) to halfwords
|
||||
uxtb16 r7, r6, ror #8 ; another two pixels to halfwords
|
||||
smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1)
|
||||
subs r12, r12, #1
|
||||
smlad r11, r7, r7, r11 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
bne loop
|
||||
|
||||
; return stuff
|
||||
ldr r6, [sp, #40] ; get address of sse
|
||||
mul r0, r8, r8 ; sum * sum
|
||||
str r11, [r6] ; store sse
|
||||
sub r0, r11, r0, lsr #8 ; return (sse - ((sum * sum) >> 8))
|
||||
|
||||
ldmfd sp!, {r4-r12, pc}
|
||||
|
||||
ENDP
|
||||
|
||||
c80808080
|
||||
DCD 0x80808080
|
||||
|
||||
END
|
||||
|
|
@ -1,187 +0,0 @@
|
|||
;
|
||||
; 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.
|
||||
;
|
||||
|
||||
;
|
||||
|
||||
|
||||
EXPORT |aom_variance_halfpixvar16x16_v_media|
|
||||
|
||||
ARM
|
||||
REQUIRE8
|
||||
PRESERVE8
|
||||
|
||||
AREA ||.text||, CODE, READONLY, ALIGN=2
|
||||
|
||||
; r0 unsigned char *src_ptr
|
||||
; r1 int source_stride
|
||||
; r2 unsigned char *ref_ptr
|
||||
; r3 int recon_stride
|
||||
; stack unsigned int *sse
|
||||
|aom_variance_halfpixvar16x16_v_media| PROC
|
||||
|
||||
stmfd sp!, {r4-r12, lr}
|
||||
|
||||
pld [r0, r1, lsl #0]
|
||||
pld [r2, r3, lsl #0]
|
||||
|
||||
mov r8, #0 ; initialize sum = 0
|
||||
ldr r10, c80808080
|
||||
mov r11, #0 ; initialize sse = 0
|
||||
mov r12, #16 ; set loop counter to 16 (=block height)
|
||||
mov lr, #0 ; constant zero
|
||||
loop
|
||||
add r9, r0, r1 ; set src pointer to next row
|
||||
; 1st 4 pixels
|
||||
ldr r4, [r0, #0] ; load 4 src pixels
|
||||
ldr r6, [r9, #0] ; load 4 src pixels from next row
|
||||
ldr r5, [r2, #0] ; load 4 ref pixels
|
||||
|
||||
; bilinear interpolation
|
||||
mvn r6, r6
|
||||
uhsub8 r4, r4, r6
|
||||
eor r4, r4, r10
|
||||
|
||||
usub8 r6, r4, r5 ; calculate difference
|
||||
pld [r0, r1, lsl #1]
|
||||
sel r7, r6, lr ; select bytes with positive difference
|
||||
usub8 r6, r5, r4 ; calculate difference with reversed operands
|
||||
pld [r2, r3, lsl #1]
|
||||
sel r6, r6, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r4, r7, lr ; calculate sum of positive differences
|
||||
usad8 r5, r6, lr ; calculate sum of negative differences
|
||||
orr r6, r6, r7 ; differences of all 4 pixels
|
||||
; calculate total sum
|
||||
adds r8, r8, r4 ; add positive differences to sum
|
||||
subs r8, r8, r5 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r5, r6 ; byte (two pixels) to halfwords
|
||||
uxtb16 r7, r6, ror #8 ; another two pixels to halfwords
|
||||
smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1)
|
||||
|
||||
; 2nd 4 pixels
|
||||
ldr r4, [r0, #4] ; load 4 src pixels
|
||||
ldr r6, [r9, #4] ; load 4 src pixels from next row
|
||||
ldr r5, [r2, #4] ; load 4 ref pixels
|
||||
|
||||
; bilinear interpolation
|
||||
mvn r6, r6
|
||||
uhsub8 r4, r4, r6
|
||||
eor r4, r4, r10
|
||||
|
||||
smlad r11, r7, r7, r11 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
usub8 r6, r4, r5 ; calculate difference
|
||||
sel r7, r6, lr ; select bytes with positive difference
|
||||
usub8 r6, r5, r4 ; calculate difference with reversed operands
|
||||
sel r6, r6, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r4, r7, lr ; calculate sum of positive differences
|
||||
usad8 r5, r6, lr ; calculate sum of negative differences
|
||||
orr r6, r6, r7 ; differences of all 4 pixels
|
||||
|
||||
; calculate total sum
|
||||
add r8, r8, r4 ; add positive differences to sum
|
||||
sub r8, r8, r5 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r5, r6 ; byte (two pixels) to halfwords
|
||||
uxtb16 r7, r6, ror #8 ; another two pixels to halfwords
|
||||
smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1)
|
||||
|
||||
; 3rd 4 pixels
|
||||
ldr r4, [r0, #8] ; load 4 src pixels
|
||||
ldr r6, [r9, #8] ; load 4 src pixels from next row
|
||||
ldr r5, [r2, #8] ; load 4 ref pixels
|
||||
|
||||
; bilinear interpolation
|
||||
mvn r6, r6
|
||||
uhsub8 r4, r4, r6
|
||||
eor r4, r4, r10
|
||||
|
||||
smlad r11, r7, r7, r11 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
usub8 r6, r4, r5 ; calculate difference
|
||||
sel r7, r6, lr ; select bytes with positive difference
|
||||
usub8 r6, r5, r4 ; calculate difference with reversed operands
|
||||
sel r6, r6, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r4, r7, lr ; calculate sum of positive differences
|
||||
usad8 r5, r6, lr ; calculate sum of negative differences
|
||||
orr r6, r6, r7 ; differences of all 4 pixels
|
||||
|
||||
; calculate total sum
|
||||
add r8, r8, r4 ; add positive differences to sum
|
||||
sub r8, r8, r5 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r5, r6 ; byte (two pixels) to halfwords
|
||||
uxtb16 r7, r6, ror #8 ; another two pixels to halfwords
|
||||
smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1)
|
||||
|
||||
; 4th 4 pixels
|
||||
ldr r4, [r0, #12] ; load 4 src pixels
|
||||
ldr r6, [r9, #12] ; load 4 src pixels from next row
|
||||
ldr r5, [r2, #12] ; load 4 ref pixels
|
||||
|
||||
; bilinear interpolation
|
||||
mvn r6, r6
|
||||
uhsub8 r4, r4, r6
|
||||
eor r4, r4, r10
|
||||
|
||||
smlad r11, r7, r7, r11 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
usub8 r6, r4, r5 ; calculate difference
|
||||
add r0, r0, r1 ; set src_ptr to next row
|
||||
sel r7, r6, lr ; select bytes with positive difference
|
||||
usub8 r6, r5, r4 ; calculate difference with reversed operands
|
||||
add r2, r2, r3 ; set dst_ptr to next row
|
||||
sel r6, r6, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r4, r7, lr ; calculate sum of positive differences
|
||||
usad8 r5, r6, lr ; calculate sum of negative differences
|
||||
orr r6, r6, r7 ; differences of all 4 pixels
|
||||
|
||||
; calculate total sum
|
||||
add r8, r8, r4 ; add positive differences to sum
|
||||
sub r8, r8, r5 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r5, r6 ; byte (two pixels) to halfwords
|
||||
uxtb16 r7, r6, ror #8 ; another two pixels to halfwords
|
||||
smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1)
|
||||
smlad r11, r7, r7, r11 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
|
||||
subs r12, r12, #1
|
||||
|
||||
bne loop
|
||||
|
||||
; return stuff
|
||||
ldr r6, [sp, #40] ; get address of sse
|
||||
mul r0, r8, r8 ; sum * sum
|
||||
str r11, [r6] ; store sse
|
||||
sub r0, r11, r0, lsr #8 ; return (sse - ((sum * sum) >> 8))
|
||||
|
||||
ldmfd sp!, {r4-r12, pc}
|
||||
|
||||
ENDP
|
||||
|
||||
c80808080
|
||||
DCD 0x80808080
|
||||
|
||||
END
|
||||
|
||||
361
third_party/aom/aom_dsp/arm/variance_media.asm
vendored
361
third_party/aom/aom_dsp/arm/variance_media.asm
vendored
|
|
@ -1,361 +0,0 @@
|
|||
;
|
||||
; 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.
|
||||
;
|
||||
|
||||
;
|
||||
|
||||
|
||||
EXPORT |aom_variance16x16_media|
|
||||
EXPORT |aom_variance8x8_media|
|
||||
EXPORT |aom_mse16x16_media|
|
||||
|
||||
ARM
|
||||
REQUIRE8
|
||||
PRESERVE8
|
||||
|
||||
AREA ||.text||, CODE, READONLY, ALIGN=2
|
||||
|
||||
; r0 unsigned char *src_ptr
|
||||
; r1 int source_stride
|
||||
; r2 unsigned char *ref_ptr
|
||||
; r3 int recon_stride
|
||||
; stack unsigned int *sse
|
||||
|aom_variance16x16_media| PROC
|
||||
|
||||
stmfd sp!, {r4-r12, lr}
|
||||
|
||||
pld [r0, r1, lsl #0]
|
||||
pld [r2, r3, lsl #0]
|
||||
|
||||
mov r8, #0 ; initialize sum = 0
|
||||
mov r11, #0 ; initialize sse = 0
|
||||
mov r12, #16 ; set loop counter to 16 (=block height)
|
||||
|
||||
loop16x16
|
||||
; 1st 4 pixels
|
||||
ldr r4, [r0, #0] ; load 4 src pixels
|
||||
ldr r5, [r2, #0] ; load 4 ref pixels
|
||||
|
||||
mov lr, #0 ; constant zero
|
||||
|
||||
usub8 r6, r4, r5 ; calculate difference
|
||||
pld [r0, r1, lsl #1]
|
||||
sel r7, r6, lr ; select bytes with positive difference
|
||||
usub8 r9, r5, r4 ; calculate difference with reversed operands
|
||||
pld [r2, r3, lsl #1]
|
||||
sel r6, r9, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r4, r7, lr ; calculate sum of positive differences
|
||||
usad8 r5, r6, lr ; calculate sum of negative differences
|
||||
orr r6, r6, r7 ; differences of all 4 pixels
|
||||
; calculate total sum
|
||||
adds r8, r8, r4 ; add positive differences to sum
|
||||
subs r8, r8, r5 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r5, r6 ; byte (two pixels) to halfwords
|
||||
uxtb16 r10, r6, ror #8 ; another two pixels to halfwords
|
||||
smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1)
|
||||
|
||||
; 2nd 4 pixels
|
||||
ldr r4, [r0, #4] ; load 4 src pixels
|
||||
ldr r5, [r2, #4] ; load 4 ref pixels
|
||||
smlad r11, r10, r10, r11 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
usub8 r6, r4, r5 ; calculate difference
|
||||
sel r7, r6, lr ; select bytes with positive difference
|
||||
usub8 r9, r5, r4 ; calculate difference with reversed operands
|
||||
sel r6, r9, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r4, r7, lr ; calculate sum of positive differences
|
||||
usad8 r5, r6, lr ; calculate sum of negative differences
|
||||
orr r6, r6, r7 ; differences of all 4 pixels
|
||||
|
||||
; calculate total sum
|
||||
add r8, r8, r4 ; add positive differences to sum
|
||||
sub r8, r8, r5 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r5, r6 ; byte (two pixels) to halfwords
|
||||
uxtb16 r10, r6, ror #8 ; another two pixels to halfwords
|
||||
smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1)
|
||||
|
||||
; 3rd 4 pixels
|
||||
ldr r4, [r0, #8] ; load 4 src pixels
|
||||
ldr r5, [r2, #8] ; load 4 ref pixels
|
||||
smlad r11, r10, r10, r11 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
usub8 r6, r4, r5 ; calculate difference
|
||||
sel r7, r6, lr ; select bytes with positive difference
|
||||
usub8 r9, r5, r4 ; calculate difference with reversed operands
|
||||
sel r6, r9, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r4, r7, lr ; calculate sum of positive differences
|
||||
usad8 r5, r6, lr ; calculate sum of negative differences
|
||||
orr r6, r6, r7 ; differences of all 4 pixels
|
||||
|
||||
; calculate total sum
|
||||
add r8, r8, r4 ; add positive differences to sum
|
||||
sub r8, r8, r5 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r5, r6 ; byte (two pixels) to halfwords
|
||||
uxtb16 r10, r6, ror #8 ; another two pixels to halfwords
|
||||
smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1)
|
||||
|
||||
; 4th 4 pixels
|
||||
ldr r4, [r0, #12] ; load 4 src pixels
|
||||
ldr r5, [r2, #12] ; load 4 ref pixels
|
||||
smlad r11, r10, r10, r11 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
usub8 r6, r4, r5 ; calculate difference
|
||||
add r0, r0, r1 ; set src_ptr to next row
|
||||
sel r7, r6, lr ; select bytes with positive difference
|
||||
usub8 r9, r5, r4 ; calculate difference with reversed operands
|
||||
add r2, r2, r3 ; set dst_ptr to next row
|
||||
sel r6, r9, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r4, r7, lr ; calculate sum of positive differences
|
||||
usad8 r5, r6, lr ; calculate sum of negative differences
|
||||
orr r6, r6, r7 ; differences of all 4 pixels
|
||||
|
||||
; calculate total sum
|
||||
add r8, r8, r4 ; add positive differences to sum
|
||||
sub r8, r8, r5 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r5, r6 ; byte (two pixels) to halfwords
|
||||
uxtb16 r10, r6, ror #8 ; another two pixels to halfwords
|
||||
smlad r11, r5, r5, r11 ; dual signed multiply, add and accumulate (1)
|
||||
smlad r11, r10, r10, r11 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
|
||||
subs r12, r12, #1
|
||||
|
||||
bne loop16x16
|
||||
|
||||
; return stuff
|
||||
ldr r6, [sp, #40] ; get address of sse
|
||||
mul r0, r8, r8 ; sum * sum
|
||||
str r11, [r6] ; store sse
|
||||
sub r0, r11, r0, lsr #8 ; return (sse - ((sum * sum) >> 8))
|
||||
|
||||
ldmfd sp!, {r4-r12, pc}
|
||||
|
||||
ENDP
|
||||
|
||||
; r0 unsigned char *src_ptr
|
||||
; r1 int source_stride
|
||||
; r2 unsigned char *ref_ptr
|
||||
; r3 int recon_stride
|
||||
; stack unsigned int *sse
|
||||
|aom_variance8x8_media| PROC
|
||||
|
||||
push {r4-r10, lr}
|
||||
|
||||
pld [r0, r1, lsl #0]
|
||||
pld [r2, r3, lsl #0]
|
||||
|
||||
mov r12, #8 ; set loop counter to 8 (=block height)
|
||||
mov r4, #0 ; initialize sum = 0
|
||||
mov r5, #0 ; initialize sse = 0
|
||||
|
||||
loop8x8
|
||||
; 1st 4 pixels
|
||||
ldr r6, [r0, #0x0] ; load 4 src pixels
|
||||
ldr r7, [r2, #0x0] ; load 4 ref pixels
|
||||
|
||||
mov lr, #0 ; constant zero
|
||||
|
||||
usub8 r8, r6, r7 ; calculate difference
|
||||
pld [r0, r1, lsl #1]
|
||||
sel r10, r8, lr ; select bytes with positive difference
|
||||
usub8 r9, r7, r6 ; calculate difference with reversed operands
|
||||
pld [r2, r3, lsl #1]
|
||||
sel r8, r9, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r6, r10, lr ; calculate sum of positive differences
|
||||
usad8 r7, r8, lr ; calculate sum of negative differences
|
||||
orr r8, r8, r10 ; differences of all 4 pixels
|
||||
; calculate total sum
|
||||
add r4, r4, r6 ; add positive differences to sum
|
||||
sub r4, r4, r7 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r7, r8 ; byte (two pixels) to halfwords
|
||||
uxtb16 r10, r8, ror #8 ; another two pixels to halfwords
|
||||
smlad r5, r7, r7, r5 ; dual signed multiply, add and accumulate (1)
|
||||
|
||||
; 2nd 4 pixels
|
||||
ldr r6, [r0, #0x4] ; load 4 src pixels
|
||||
ldr r7, [r2, #0x4] ; load 4 ref pixels
|
||||
smlad r5, r10, r10, r5 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
usub8 r8, r6, r7 ; calculate difference
|
||||
add r0, r0, r1 ; set src_ptr to next row
|
||||
sel r10, r8, lr ; select bytes with positive difference
|
||||
usub8 r9, r7, r6 ; calculate difference with reversed operands
|
||||
add r2, r2, r3 ; set dst_ptr to next row
|
||||
sel r8, r9, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r6, r10, lr ; calculate sum of positive differences
|
||||
usad8 r7, r8, lr ; calculate sum of negative differences
|
||||
orr r8, r8, r10 ; differences of all 4 pixels
|
||||
|
||||
; calculate total sum
|
||||
add r4, r4, r6 ; add positive differences to sum
|
||||
sub r4, r4, r7 ; subtract negative differences from sum
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r7, r8 ; byte (two pixels) to halfwords
|
||||
uxtb16 r10, r8, ror #8 ; another two pixels to halfwords
|
||||
smlad r5, r7, r7, r5 ; dual signed multiply, add and accumulate (1)
|
||||
subs r12, r12, #1 ; next row
|
||||
smlad r5, r10, r10, r5 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
bne loop8x8
|
||||
|
||||
; return stuff
|
||||
ldr r8, [sp, #32] ; get address of sse
|
||||
mul r1, r4, r4 ; sum * sum
|
||||
str r5, [r8] ; store sse
|
||||
sub r0, r5, r1, ASR #6 ; return (sse - ((sum * sum) >> 6))
|
||||
|
||||
pop {r4-r10, pc}
|
||||
|
||||
ENDP
|
||||
|
||||
; r0 unsigned char *src_ptr
|
||||
; r1 int source_stride
|
||||
; r2 unsigned char *ref_ptr
|
||||
; r3 int recon_stride
|
||||
; stack unsigned int *sse
|
||||
;
|
||||
;note: Based on aom_variance16x16_media. In this function, sum is never used.
|
||||
; So, we can remove this part of calculation.
|
||||
|
||||
|aom_mse16x16_media| PROC
|
||||
|
||||
push {r4-r9, lr}
|
||||
|
||||
pld [r0, r1, lsl #0]
|
||||
pld [r2, r3, lsl #0]
|
||||
|
||||
mov r12, #16 ; set loop counter to 16 (=block height)
|
||||
mov r4, #0 ; initialize sse = 0
|
||||
|
||||
loopmse
|
||||
; 1st 4 pixels
|
||||
ldr r5, [r0, #0x0] ; load 4 src pixels
|
||||
ldr r6, [r2, #0x0] ; load 4 ref pixels
|
||||
|
||||
mov lr, #0 ; constant zero
|
||||
|
||||
usub8 r8, r5, r6 ; calculate difference
|
||||
pld [r0, r1, lsl #1]
|
||||
sel r7, r8, lr ; select bytes with positive difference
|
||||
usub8 r9, r6, r5 ; calculate difference with reversed operands
|
||||
pld [r2, r3, lsl #1]
|
||||
sel r8, r9, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r5, r7, lr ; calculate sum of positive differences
|
||||
usad8 r6, r8, lr ; calculate sum of negative differences
|
||||
orr r8, r8, r7 ; differences of all 4 pixels
|
||||
|
||||
ldr r5, [r0, #0x4] ; load 4 src pixels
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r6, r8 ; byte (two pixels) to halfwords
|
||||
uxtb16 r7, r8, ror #8 ; another two pixels to halfwords
|
||||
smlad r4, r6, r6, r4 ; dual signed multiply, add and accumulate (1)
|
||||
|
||||
; 2nd 4 pixels
|
||||
ldr r6, [r2, #0x4] ; load 4 ref pixels
|
||||
smlad r4, r7, r7, r4 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
usub8 r8, r5, r6 ; calculate difference
|
||||
sel r7, r8, lr ; select bytes with positive difference
|
||||
usub8 r9, r6, r5 ; calculate difference with reversed operands
|
||||
sel r8, r9, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r5, r7, lr ; calculate sum of positive differences
|
||||
usad8 r6, r8, lr ; calculate sum of negative differences
|
||||
orr r8, r8, r7 ; differences of all 4 pixels
|
||||
ldr r5, [r0, #0x8] ; load 4 src pixels
|
||||
; calculate sse
|
||||
uxtb16 r6, r8 ; byte (two pixels) to halfwords
|
||||
uxtb16 r7, r8, ror #8 ; another two pixels to halfwords
|
||||
smlad r4, r6, r6, r4 ; dual signed multiply, add and accumulate (1)
|
||||
|
||||
; 3rd 4 pixels
|
||||
ldr r6, [r2, #0x8] ; load 4 ref pixels
|
||||
smlad r4, r7, r7, r4 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
usub8 r8, r5, r6 ; calculate difference
|
||||
sel r7, r8, lr ; select bytes with positive difference
|
||||
usub8 r9, r6, r5 ; calculate difference with reversed operands
|
||||
sel r8, r9, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r5, r7, lr ; calculate sum of positive differences
|
||||
usad8 r6, r8, lr ; calculate sum of negative differences
|
||||
orr r8, r8, r7 ; differences of all 4 pixels
|
||||
|
||||
ldr r5, [r0, #0xc] ; load 4 src pixels
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r6, r8 ; byte (two pixels) to halfwords
|
||||
uxtb16 r7, r8, ror #8 ; another two pixels to halfwords
|
||||
smlad r4, r6, r6, r4 ; dual signed multiply, add and accumulate (1)
|
||||
|
||||
; 4th 4 pixels
|
||||
ldr r6, [r2, #0xc] ; load 4 ref pixels
|
||||
smlad r4, r7, r7, r4 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
usub8 r8, r5, r6 ; calculate difference
|
||||
add r0, r0, r1 ; set src_ptr to next row
|
||||
sel r7, r8, lr ; select bytes with positive difference
|
||||
usub8 r9, r6, r5 ; calculate difference with reversed operands
|
||||
add r2, r2, r3 ; set dst_ptr to next row
|
||||
sel r8, r9, lr ; select bytes with negative difference
|
||||
|
||||
; calculate partial sums
|
||||
usad8 r5, r7, lr ; calculate sum of positive differences
|
||||
usad8 r6, r8, lr ; calculate sum of negative differences
|
||||
orr r8, r8, r7 ; differences of all 4 pixels
|
||||
|
||||
subs r12, r12, #1 ; next row
|
||||
|
||||
; calculate sse
|
||||
uxtb16 r6, r8 ; byte (two pixels) to halfwords
|
||||
uxtb16 r7, r8, ror #8 ; another two pixels to halfwords
|
||||
smlad r4, r6, r6, r4 ; dual signed multiply, add and accumulate (1)
|
||||
smlad r4, r7, r7, r4 ; dual signed multiply, add and accumulate (2)
|
||||
|
||||
bne loopmse
|
||||
|
||||
; return stuff
|
||||
ldr r1, [sp, #28] ; get address of sse
|
||||
mov r0, r4 ; return sse
|
||||
str r4, [r1] ; store sse
|
||||
|
||||
pop {r4-r9, pc}
|
||||
|
||||
ENDP
|
||||
|
||||
END
|
||||
Loading…
Add table
Add a link
Reference in a new issue