max/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

avcodec/x86/vc1dsp_mc: Add size 8 horizontal SSSE3 mc functions

Browse Source 
pmaddubsw strikes again.

vc1dsp.avg_vc1_mspel_pixels_tab_mc10_8_c:              150.2 ( 1.00x)
vc1dsp.avg_vc1_mspel_pixels_tab_mc10_8_mmxext:          44.5 ( 3.38x)
vc1dsp.avg_vc1_mspel_pixels_tab_mc10_8_ssse3:           18.5 ( 8.12x)
vc1dsp.avg_vc1_mspel_pixels_tab_mc20_8_c:              288.2 ( 1.00x)
vc1dsp.avg_vc1_mspel_pixels_tab_mc20_8_mmxext:          37.7 ( 7.64x)
vc1dsp.avg_vc1_mspel_pixels_tab_mc20_8_ssse3:           18.1 (15.97x)
vc1dsp.avg_vc1_mspel_pixels_tab_mc30_8_c:              155.4 ( 1.00x)
vc1dsp.avg_vc1_mspel_pixels_tab_mc30_8_mmxext:          46.5 ( 3.34x)
vc1dsp.avg_vc1_mspel_pixels_tab_mc30_8_ssse3:           18.1 ( 8.60x)
vc1dsp.put_vc1_mspel_pixels_tab_mc10_8_c:              282.2 ( 1.00x)
vc1dsp.put_vc1_mspel_pixels_tab_mc10_8_mmx:             42.7 ( 6.61x)
vc1dsp.put_vc1_mspel_pixels_tab_mc10_8_ssse3:           16.4 (17.16x)
vc1dsp.put_vc1_mspel_pixels_tab_mc20_8_c:              223.4 ( 1.00x)
vc1dsp.put_vc1_mspel_pixels_tab_mc20_8_mmx:             36.3 ( 6.15x)
vc1dsp.put_vc1_mspel_pixels_tab_mc20_8_ssse3:           16.4 (13.59x)
vc1dsp.put_vc1_mspel_pixels_tab_mc30_8_c:              255.2 ( 1.00x)
vc1dsp.put_vc1_mspel_pixels_tab_mc30_8_mmx:             43.6 ( 5.85x)
vc1dsp.put_vc1_mspel_pixels_tab_mc30_8_ssse3:           16.4 (15.52x)

Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This commit is contained in:
Andreas Rheinhardt
2026-05-04 19:51:29 +02:00
parent f5b8a15b61
commit fd711adcb6
2 changed files with 77 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
libavcodec/x86/vc1dsp_init.c
+12
View File
@@ -89,6 +89,14 @@ void ff_vc1_inv_trans_8x4_dc_mmxext(uint8_t *dest, ptrdiff_t linesize,
void ff_vc1_inv_trans_8x8_dc_mmxext(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 \
(uint8_t *dst, const uint8_t *src, ptrdiff_t stride, int rnd); \
dsp->OP ## _vc1_mspel_pixels_tab[SIZE == 8][X + 4 * Y] = \
ff_vc1_ ## OP ## _mspel_mc ## X ## Y## _ ## SIZE ##_ ## XMM
#define MSPEL_FUNCS_SIZE(X, Y, SIZE, XMM) \
MSPEL_FUNC(put, X, Y, SIZE, XMM); \
MSPEL_FUNC(avg, X, Y, SIZE, XMM)
av_cold void ff_vc1dsp_init_x86(VC1DSPContext *dsp)
{
@@ -132,6 +140,10 @@ av_cold void ff_vc1dsp_init_x86(VC1DSPContext *dsp)
ASSIGN_LF816(ssse3);
dsp->put_no_rnd_vc1_chroma_pixels_tab[0] = ff_put_vc1_chroma_mc8_nornd_ssse3;
dsp->avg_no_rnd_vc1_chroma_pixels_tab[0] = ff_avg_vc1_chroma_mc8_nornd_ssse3;
MSPEL_FUNCS_SIZE(1, 0, 8, ssse3);
MSPEL_FUNCS_SIZE(2, 0, 8, ssse3);
MSPEL_FUNCS_SIZE(3, 0, 8, ssse3);
}
if (EXTERNAL_SSE4(cpu_flags)) {
dsp->vc1_h_loop_filter8 = ff_vc1_h_loop_filter8_sse4;
libavcodec/x86/vc1dsp_mc.asm
+65
View File
@@ -21,6 +21,12 @@
%include "libavutil/x86/x86util.asm"
SECTION_RODATA
pb_m4_36: times 8 db -4, 36
pb_m4_53: times 8 db -4, 53
pb_m3_18: times 8 db -3, 18
cextern pw_9
cextern pw_128
@@ -192,3 +198,62 @@ HOR_16B_SHIFT2 OP_PUT, put
INIT_MMX mmxext
HOR_16B_SHIFT2 OP_AVG, avg
%endif ; HAVE_MMX_INLINE
%define MOV8 movq
INIT_XMM ssse3
%macro HOR_8B 2
cglobal vc1_%1_mspel_mc10_%2, 4, 4, 6, dst, src, stride, rnd
mova m1, [pb_m4_53]
mova m2, [pb_m3_18]
sub rndd, 32
jmp vc1_%1_mspel_mc30_%2_after_prologue
cglobal vc1_%1_mspel_mc20_%2, 4, 4, 6, dst, src, stride, rnd
mova m1, [pb_m4_36]
lea rndd, [4*rndd-32]
mova m2, m1
jmp vc1_%1_mspel_mc30_%2_after_prologue
cglobal vc1_%1_mspel_mc30_%2, 4, 4, 6, dst, src, stride, rnd
mova m2, [pb_m4_53]
mova m1, [pb_m3_18]
sub rndd, 32
vc1_%1_mspel_mc30_%2_after_prologue:
movd m0, rndd
WIN64_SPILL_XMM 7
%define hd rndd
mov hd, %2
SPLATW m0, m0
.loop:
MOV%2 m3, [srcq-1]
MOV%2 m4, [srcq]
MOV%2 m5, [srcq+1]
MOV%2 m6, [srcq+2]
punpcklbw m3, m4
pmaddubsw m3, m1
%ifidn %1,avg
movq m4, [dstq]
%endif
punpcklbw m6, m5
pmaddubsw m6, m2
add srcq, strideq
psubw m3, m0
paddw m3, m6
psraw m3, 6
packuswb m3, m3
%ifidn %1,avg
pavgb m3, m4
%endif
movq [dstq], m3
add dstq, strideq
dec hd
jnz .loop
RET
%endmacro
HOR_8B put, 8
HOR_8B avg, 8