wuggy/Dactyloidae
wuggy/Dactyloidae
1
0
Fork 0
mirror of https://repo.dactyloidae.xyz/Dactyloidae/UXP.git synced 2026-09-25 18:07:31 +09:00
Code Issues Projects Releases Packages Wiki Activity

[ffvpx] Update ffvp9/ffvp8 to 3.4.2-release

Browse source 
Structure of code was slightly modified so that it should be no longer necessary to re-generate the config_*.h files, greatly simplifying the resync process in the future.
This commit is contained in:
trav90 2018-04-26 16:49:15 -05:00 • committed by Roy Tam
commit d1361d3c21
183 changed files with 17773 additions and 28489 deletions
Download patch file Download diff file Expand all files Collapse all files

86
media/ffvpx/libavutil/x86/float_dsp.asm
View file

@ -22,6 +22,9 @@
%include "x86util.asm"
SECTION_RODATA 32
pd_reverse: dd 7, 6, 5, 4, 3, 2, 1, 0
SECTION .text
;-----------------------------------------------------------------------------
@ -148,6 +151,69 @@ cglobal vector_fmul_scalar, 4,4,3, dst, src, mul, len
INIT_XMM sse
VECTOR_FMUL_SCALAR
;------------------------------------------------------------------------------
; void ff_vector_dmac_scalar(double *dst, const double *src, double mul,
; int len)
;------------------------------------------------------------------------------
%macro VECTOR_DMAC_SCALAR 0
%if ARCH_X86_32
cglobal vector_dmac_scalar, 2,4,5, dst, src, mul, len, lenaddr
mov lenq, lenaddrm
VBROADCASTSD m0, mulm
%else
%if UNIX64
cglobal vector_dmac_scalar, 3,3,5, dst, src, len
%else
cglobal vector_dmac_scalar, 4,4,5, dst, src, mul, len
SWAP 0, 2
%endif
movlhps xm0, xm0
%if cpuflag(avx)
vinsertf128 m0, m0, xm0, 1
%endif
%endif
lea lenq, [lend*8-mmsize*4]
.loop:
%if cpuflag(fma3)
movaps m1, [dstq+lenq]
movaps m2, [dstq+lenq+1*mmsize]
movaps m3, [dstq+lenq+2*mmsize]
movaps m4, [dstq+lenq+3*mmsize]
fmaddpd m1, m0, [srcq+lenq], m1
fmaddpd m2, m0, [srcq+lenq+1*mmsize], m2
fmaddpd m3, m0, [srcq+lenq+2*mmsize], m3
fmaddpd m4, m0, [srcq+lenq+3*mmsize], m4
%else ; cpuflag
mulpd m1, m0, [srcq+lenq]
mulpd m2, m0, [srcq+lenq+1*mmsize]
mulpd m3, m0, [srcq+lenq+2*mmsize]
mulpd m4, m0, [srcq+lenq+3*mmsize]
addpd m1, m1, [dstq+lenq]
addpd m2, m2, [dstq+lenq+1*mmsize]
addpd m3, m3, [dstq+lenq+2*mmsize]
addpd m4, m4, [dstq+lenq+3*mmsize]
%endif ; cpuflag
movaps [dstq+lenq], m1
movaps [dstq+lenq+1*mmsize], m2
movaps [dstq+lenq+2*mmsize], m3
movaps [dstq+lenq+3*mmsize], m4
sub lenq, mmsize*4
jge .loop
REP_RET
%endmacro
INIT_XMM sse2
VECTOR_DMAC_SCALAR
%if HAVE_AVX_EXTERNAL
INIT_YMM avx
VECTOR_DMAC_SCALAR
%endif
%if HAVE_FMA3_EXTERNAL
INIT_YMM fma3
VECTOR_DMAC_SCALAR
%endif
;------------------------------------------------------------------------------
; void ff_vector_dmul_scalar(double *dst, const double *src, double mul,
; int len)
@ -177,8 +243,8 @@ cglobal vector_dmul_scalar, 4,4,3, dst, src, mul, len
.loop:
mulpd m1, m0, [srcq+lenq ]
mulpd m2, m0, [srcq+lenq+mmsize]
mova [dstq+lenq ], m1
mova [dstq+lenq+mmsize], m2
movaps [dstq+lenq ], m1
movaps [dstq+lenq+mmsize], m2
sub lenq, 2*mmsize
jge .loop
REP_RET
@ -296,10 +362,16 @@ VECTOR_FMUL_ADD
;-----------------------------------------------------------------------------
%macro VECTOR_FMUL_REVERSE 0
cglobal vector_fmul_reverse, 4,4,2, dst, src0, src1, len
%if cpuflag(avx2)
movaps m2, [pd_reverse]
%endif
lea lenq, [lend*4 - 2*mmsize]
ALIGN 16
.loop:
%if cpuflag(avx)
%if cpuflag(avx2)
vpermps m0, m2, [src1q]
vpermps m1, m2, [src1q+mmsize]
%elif cpuflag(avx)
vmovaps xmm0, [src1q + 16]
vinsertf128 m0, m0, [src1q], 1
vshufps m0, m0, m0, q0123
@ -314,8 +386,8 @@ ALIGN 16
%endif
mulps m0, m0, [src0q + lenq + mmsize]
mulps m1, m1, [src0q + lenq]
mova [dstq + lenq + mmsize], m0
mova [dstq + lenq], m1
movaps [dstq + lenq + mmsize], m0
movaps [dstq + lenq], m1
add src1q, 2*mmsize
sub lenq, 2*mmsize
jge .loop
@ -328,6 +400,10 @@ VECTOR_FMUL_REVERSE
INIT_YMM avx
VECTOR_FMUL_REVERSE
%endif
%if HAVE_AVX2_EXTERNAL
INIT_YMM avx2
VECTOR_FMUL_REVERSE
%endif
; float scalarproduct_float_sse(const float *v1, const float *v2, int len)
INIT_XMM sse