virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2025-08-15 14:13:16 +02:00
Code Issues Releases Activity

x86/synth_filter: add synth_filter_fma3

Browse Source 
Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Anton Khirnov <anton@khirnov.net>
This commit is contained in:
James Almer
2014-03-20 15:37:56 -03:00
committed by Anton Khirnov
parent 81e02fae6e
commit c74b86699c
2 changed files with 28 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
libavcodec/x86/dcadsp.asm
View File

@@ -230,6 +230,14 @@ DCA_LFE_FIR 1
SHUF m11, ptr2 + j + (15 - 3) * 4 - mmsize, m12 SHUF m11, ptr2 + j + (15 - 3) * 4 - mmsize, m12
mova m12, [ptr1 + j + mmsize] mova m12, [ptr1 + j + mmsize]
%endif %endif
%if cpuflag(fma3)
fmaddps m2, m6, [win + %1 + j + 16 * 4], m2
fnmaddps m1, m5, [win + %1 + j], m1
%if ARCH_X86_64
fmaddps m8, m12, [win + %1 + j + mmsize + 16 * 4], m8
fnmaddps m7, m11, [win + %1 + j + mmsize], m7
%endif
%else ; non-FMA
mulps m6, m6, [win + %1 + j + 16 * 4] mulps m6, m6, [win + %1 + j + 16 * 4]
mulps m5, m5, [win + %1 + j] mulps m5, m5, [win + %1 + j]
%if ARCH_X86_64 %if ARCH_X86_64
@@ -242,6 +250,7 @@ DCA_LFE_FIR 1
addps m8, m8, m12 addps m8, m8, m12
subps m7, m7, m11 subps m7, m7, m11
%endif %endif
%endif ; cpuflag(fma3)
;~ c += window[i + j + 32] * (synth_buf[16 + i + j]) ;~ c += window[i + j + 32] * (synth_buf[16 + i + j])
;~ d += window[i + j + 48] * (synth_buf[31 - i + j]) ;~ d += window[i + j + 48] * (synth_buf[31 - i + j])
SHUF m6, ptr2 + j + (31 - 3) * 4, m5 SHUF m6, ptr2 + j + (31 - 3) * 4, m5
@@ -250,6 +259,14 @@ DCA_LFE_FIR 1
SHUF m12, ptr2 + j + (31 - 3) * 4 - mmsize, m11 SHUF m12, ptr2 + j + (31 - 3) * 4 - mmsize, m11
mova m11, [ptr1 + j + mmsize + 16 * 4] mova m11, [ptr1 + j + mmsize + 16 * 4]
%endif %endif
%if cpuflag(fma3)
fmaddps m3, m5, [win + %1 + j + 32 * 4], m3
fmaddps m4, m6, [win + %1 + j + 48 * 4], m4
%if ARCH_X86_64
fmaddps m9, m11, [win + %1 + j + mmsize + 32 * 4], m9
fmaddps m10, m12, [win + %1 + j + mmsize + 48 * 4], m10
%endif
%else ; non-FMA
mulps m5, m5, [win + %1 + j + 32 * 4] mulps m5, m5, [win + %1 + j + 32 * 4]
mulps m6, m6, [win + %1 + j + 48 * 4] mulps m6, m6, [win + %1 + j + 48 * 4]
%if ARCH_X86_64 %if ARCH_X86_64
@@ -262,6 +279,7 @@ DCA_LFE_FIR 1
addps m9, m9, m11 addps m9, m9, m11
addps m10, m10, m12 addps m10, m10, m12
%endif %endif
%endif ; cpuflag(fma3)
sub j, 64 * 4 sub j, 64 * 4
%endmacro %endmacro
@@ -400,3 +418,7 @@ INIT_XMM sse2
SYNTH_FILTER SYNTH_FILTER
INIT_YMM avx INIT_YMM avx
SYNTH_FILTER SYNTH_FILTER
%if HAVE_FMA3_EXTERNAL
INIT_YMM fma3
SYNTH_FILTER
%endif

6
libavcodec/x86/dcadsp_init.c
View File

@@ -82,6 +82,9 @@ SYNTH_FILTER_FUNC(sse)
#endif #endif
SYNTH_FILTER_FUNC(sse2) SYNTH_FILTER_FUNC(sse2)
SYNTH_FILTER_FUNC(avx) SYNTH_FILTER_FUNC(avx)
#if HAVE_FMA3_EXTERNAL
SYNTH_FILTER_FUNC(fma3)
#endif
#endif /* HAVE_YASM */ #endif /* HAVE_YASM */
av_cold void ff_synth_filter_init_x86(SynthFilterContext *s) av_cold void ff_synth_filter_init_x86(SynthFilterContext *s)
@@ -100,5 +103,8 @@ av_cold void ff_synth_filter_init_x86(SynthFilterContext *s)
if (EXTERNAL_AVX(cpu_flags)) { if (EXTERNAL_AVX(cpu_flags)) {
s->synth_filter_float = synth_filter_avx; s->synth_filter_float = synth_filter_avx;
} }
if (EXTERNAL_FMA3(cpu_flags)) {
s->synth_filter_float = synth_filter_fma3;
}
#endif /* HAVE_YASM */ #endif /* HAVE_YASM */
} }