dilim/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2026-08-09 17:39:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

avcodec/x86/vc1dsp_inv_trans: Port to SSE2

Browse Source 
No change in performance here.

Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This commit is contained in:
Andreas Rheinhardt
2026-07-29 16:40:40 +02:00
parent 3db7a55280
commit 406c5a37aa
3 changed files with 35 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
libavcodec/x86/vc1dsp_init.c
View File

@@ -66,14 +66,14 @@ void ff_put_vc1_chroma_mc8_nornd_ssse3(uint8_t *dst, const uint8_t *src,
ptrdiff_t stride, int h, int x, int y);
void ff_avg_vc1_chroma_mc8_nornd_ssse3(uint8_t *dst, const uint8_t *src,
ptrdiff_t stride, int h, int x, int y);
void ff_vc1_inv_trans_4x4_dc_mmxext(uint8_t *dest, ptrdiff_t linesize,
int16_t *block);
void ff_vc1_inv_trans_4x8_dc_mmxext(uint8_t *dest, ptrdiff_t linesize,
int16_t *block);
void ff_vc1_inv_trans_8x4_dc_mmxext(uint8_t *dest, ptrdiff_t linesize,
int16_t *block);
void ff_vc1_inv_trans_8x8_dc_mmxext(uint8_t *dest, ptrdiff_t linesize,
int16_t *block);
void ff_vc1_inv_trans_4x4_dc_sse2(uint8_t *dest, ptrdiff_t linesize,
int16_t *block);
void ff_vc1_inv_trans_4x8_dc_sse2(uint8_t *dest, ptrdiff_t linesize,
int16_t *block);
void ff_vc1_inv_trans_8x4_dc_sse2(uint8_t *dest, ptrdiff_t linesize,
int16_t *block);
void ff_vc1_inv_trans_8x8_dc_sse2(uint8_t *dest, ptrdiff_t linesize,
int16_t *block);
#define MSPEL_FUNC(OP, X, Y, SIZE, XMM) \
void ff_vc1_ ## OP ## _mspel_mc ## X ## Y ## _ ## SIZE ##_ ## XMM \
@@ -110,13 +110,12 @@ av_cold void ff_vc1dsp_init_x86(VC1DSPContext *dsp)
dsp->vc1_v_loop_filter16 = vc1_v_loop_filter16_ ## EXT; \
dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_ ## EXT
if (EXTERNAL_MMXEXT(cpu_flags)) {
dsp->vc1_inv_trans_8x8_dc = ff_vc1_inv_trans_8x8_dc_mmxext;
dsp->vc1_inv_trans_4x8_dc = ff_vc1_inv_trans_4x8_dc_mmxext;
dsp->vc1_inv_trans_8x4_dc = ff_vc1_inv_trans_8x4_dc_mmxext;
dsp->vc1_inv_trans_4x4_dc = ff_vc1_inv_trans_4x4_dc_mmxext;
}
if (EXTERNAL_SSE2(cpu_flags)) {
dsp->vc1_inv_trans_8x8_dc = ff_vc1_inv_trans_8x8_dc_sse2;
dsp->vc1_inv_trans_4x8_dc = ff_vc1_inv_trans_4x8_dc_sse2;
dsp->vc1_inv_trans_8x4_dc = ff_vc1_inv_trans_8x4_dc_sse2;
dsp->vc1_inv_trans_4x4_dc = ff_vc1_inv_trans_4x4_dc_sse2;
ASSIGN_LF816(sse2);
MSPEL_FUNCS(0, 0, sse2);

41
libavcodec/x86/vc1dsp_inv_trans.asm
View File

@@ -23,9 +23,13 @@
SECTION .text
%macro INV_TRANS_INIT 0
%macro INV_TRANS_INIT 1 ; width
movd m0, blockd
%if %1 == 4
pshuflw m0, m0, 0
%else
SPLATW m0, m0
%endif
pxor m1, m1
psubw m1, m0
packuswb m0, m0
@@ -54,9 +58,9 @@ SECTION .text
mov%1 [linesize3q +destq], m5
%endmacro
; ff_vc1_inv_trans_?x?_dc_mmxext(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
INIT_MMX mmxext
cglobal vc1_inv_trans_4x4_dc, 3,4,0, dest, linesize, block
INIT_XMM sse2
; ff_vc1_inv_trans_?x?_dc_sse2(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
cglobal vc1_inv_trans_4x4_dc, 3,4,6, dest, linesize, block
movsx r3d, WORD [blockq]
mov blockd, r3d ; dc
shl blockd, 4 ; 16 * dc
@@ -67,13 +71,12 @@ cglobal vc1_inv_trans_4x4_dc, 3,4,0, dest, linesize, block
lea blockd, [blockq+r3+64] ; 17 * dc + 64
sar blockd, 7 ; >> 7
INV_TRANS_INIT
INV_TRANS_INIT 4
INV_TRANS_PROCESS h
INV_TRANS_PROCESS d
RET
INIT_MMX mmxext
cglobal vc1_inv_trans_4x8_dc, 3,4,0, dest, linesize, block
cglobal vc1_inv_trans_4x8_dc, 3,4,6, dest, linesize, block
movsx r3d, WORD [blockq]
mov blockd, r3d ; dc
shl blockd, 4 ; 16 * dc
@@ -83,15 +86,14 @@ cglobal vc1_inv_trans_4x8_dc, 3,4,0, dest, linesize, block
lea blockd, [blockq*3+64] ; 12 * dc + 64
sar blockd, 7 ; >> 7
INV_TRANS_INIT
INV_TRANS_INIT 4
INV_TRANS_PROCESS h
INV_TRANS_PROCESS d
lea destq, [destq+linesizeq*4]
INV_TRANS_PROCESS h
INV_TRANS_PROCESS d
RET
INIT_MMX mmxext
cglobal vc1_inv_trans_8x4_dc, 3,4,0, dest, linesize, block
cglobal vc1_inv_trans_8x4_dc, 3,4,6, dest, linesize, block
movsx blockd, WORD [blockq] ; dc
lea blockd, [blockq*3+1] ; 3 * dc + 1
sar blockd, 1 ; >> 1
@@ -100,22 +102,21 @@ cglobal vc1_inv_trans_8x4_dc, 3,4,0, dest, linesize, block
lea blockd, [blockq+r3+64] ; 17 * dc + 64
sar blockd, 7 ; >> 7
INV_TRANS_INIT
INV_TRANS_INIT 8
INV_TRANS_PROCESS a
INV_TRANS_PROCESS q
RET
INIT_MMX mmxext
cglobal vc1_inv_trans_8x8_dc, 3,3,0, dest, linesize, block
cglobal vc1_inv_trans_8x8_dc, 3,3,6, dest, linesize, block
movsx blockd, WORD [blockq] ; dc
lea blockd, [blockq*3+1] ; 3 * dc + 1
sar blockd, 1 ; >> 1
lea blockd, [blockq*3+16] ; 3 * dc + 16
sar blockd, 5 ; >> 5
INV_TRANS_INIT
INV_TRANS_INIT 8
INV_TRANS_PROCESS a
INV_TRANS_PROCESS q
lea destq, [destq+linesizeq*4]
INV_TRANS_PROCESS a
INV_TRANS_PROCESS q
RET

2
tests/checkasm/vc1dsp.c
View File

@@ -310,7 +310,7 @@ static void check_inv_trans_adding(void)
void (*func)(uint8_t *, ptrdiff_t, int16_t *) = *(void **)((intptr_t) &h + tests[k].offset);
if (check_func(func, "vc1dsp.%s", tests[k].name)) {
matrix *coeffs;
declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *, ptrdiff_t, int16_t *);
declare_func(void, uint8_t *, ptrdiff_t, int16_t *);
RANDOMIZE_BUFFER16(inv_trans_in, 8 * 8);
RANDOMIZE_BUFFER8(inv_trans_out, 10 * 24);
coeffs = generate_inverse_quantized_transform_coefficients(tests[k].width, tests[k].height);