virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2025-10-06 05:47:18 +02:00
Code Issues Releases Activity

avcodec/x86/h264_qpel_10bit: Remove SSE2 "cache64" duplicates

Browse Source 
The horizontal 10bit MC SSE2 functions are currently duplicated:
They exist both in ordinary form as well as with a "sse2_cache64"
suffix. A comment in ff_h264qpel_init_x86() indicates that this
is due to older processors not liking accesses that cross cache
lines, yet these functions are identical to the non-cache64
functions (apart from the unavoidable changes in the rip-offset).

The only difference between these functions and the ordinary ones
are that the cache64 ones are created via a special form of the
INIT_XMM macro: "INIT_XMM sse2, cache64". This affects the name
and apparently defines cpuflags_cache64, yet nothing checks for
this, so both versions are identical. So remove the cache64 ones
and treat the remaining ones like ordinary SSE2 functions.

Reviewed-by: James Almer <jamrial@gmail.com>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This commit is contained in:
Andreas Rheinhardt
2025-09-29 23:08:11 +02:00
parent 697da64c8e
commit 495c3d03ae
2 changed files with 3 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
libavcodec/x86/h264_qpel.c
View File

@@ -384,13 +384,10 @@ LUMA_MC_4(10, mc33, mmxext)
LUMA_MC_816(10, mc00, sse2) LUMA_MC_816(10, mc00, sse2)
LUMA_MC_816(10, mc10, sse2) LUMA_MC_816(10, mc10, sse2)
LUMA_MC_816(10, mc10, sse2_cache64)
LUMA_MC_816(10, mc10, ssse3_cache64) LUMA_MC_816(10, mc10, ssse3_cache64)
LUMA_MC_816(10, mc20, sse2) LUMA_MC_816(10, mc20, sse2)
LUMA_MC_816(10, mc20, sse2_cache64)
LUMA_MC_816(10, mc20, ssse3_cache64) LUMA_MC_816(10, mc20, ssse3_cache64)
LUMA_MC_816(10, mc30, sse2) LUMA_MC_816(10, mc30, sse2)
LUMA_MC_816(10, mc30, sse2_cache64)
LUMA_MC_816(10, mc30, ssse3_cache64) LUMA_MC_816(10, mc30, ssse3_cache64)
LUMA_MC_816(10, mc01, sse2) LUMA_MC_816(10, mc01, sse2)
LUMA_MC_816(10, mc11, sse2) LUMA_MC_816(10, mc11, sse2)
@@ -488,9 +485,9 @@ av_cold void ff_h264qpel_init_x86(H264QpelContext *c, int bit_depth)
SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 10_sse2, ff_); SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 10_sse2, ff_);
SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_sse2, ff_); SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_sse2, ff_);
SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 10_sse2, ff_); SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 10_sse2, ff_);
H264_QPEL_FUNCS_10(1, 0, sse2_cache64); H264_QPEL_FUNCS_10(1, 0, sse2);
H264_QPEL_FUNCS_10(2, 0, sse2_cache64); H264_QPEL_FUNCS_10(2, 0, sse2);
H264_QPEL_FUNCS_10(3, 0, sse2_cache64); H264_QPEL_FUNCS_10(3, 0, sse2);
} }
} }
@@ -516,17 +513,5 @@ av_cold void ff_h264qpel_init_x86(H264QpelContext *c, int bit_depth)
H264_QPEL_FUNCS_10(3, 0, ssse3_cache64); H264_QPEL_FUNCS_10(3, 0, ssse3_cache64);
} }
} }
if (EXTERNAL_AVX(cpu_flags)) {
/* AVX implies 64 byte cache lines without the need to avoid unaligned
* memory accesses that cross the boundary between two cache lines.
* TODO: Port X264_CPU_CACHELINE_32/64 detection from x264 to avoid
* having to treat SSE2 functions with such properties as AVX. */
if (bit_depth == 10) {
H264_QPEL_FUNCS_10(1, 0, sse2);
H264_QPEL_FUNCS_10(2, 0, sse2);
H264_QPEL_FUNCS_10(3, 0, sse2);
}
}
#endif #endif
} }

4
libavcodec/x86/h264_qpel_10bit.asm
View File

@@ -227,8 +227,6 @@ MC00 avg
%define OP_MOV mova %define OP_MOV mova
INIT_MMX mmxext INIT_MMX mmxext
%1 put, 4 %1 put, 4
INIT_XMM sse2, cache64
%1 put, 8
INIT_XMM ssse3, cache64 INIT_XMM ssse3, cache64
%1 put, 8 %1 put, 8
INIT_XMM sse2 INIT_XMM sse2
@@ -237,8 +235,6 @@ INIT_XMM sse2
%define OP_MOV AVG_MOV %define OP_MOV AVG_MOV
INIT_MMX mmxext INIT_MMX mmxext
%1 avg, 4 %1 avg, 4
INIT_XMM sse2, cache64
%1 avg, 8
INIT_XMM ssse3, cache64 INIT_XMM ssse3, cache64
%1 avg, 8 %1 avg, 8
INIT_XMM sse2 INIT_XMM sse2