diff options
| author | Michael Niedermayer <michaelni@gmx.at> | 2011-11-08 02:03:14 +0100 |
|---|---|---|
| committer | Michael Niedermayer <michaelni@gmx.at> | 2011-11-08 02:03:14 +0100 |
| commit | 0bd42ae72c0e88826b823d8106286d5abe2028e7 (patch) | |
| tree | da7087d1f67ea54580026e30dcedb4f58db6aba6 /libavcodec/x86/dsputil_yasm.asm | |
| parent | 5955c63c36e9b822f1186cd45a8dec5056b9c575 (diff) | |
| parent | 1c8c41ff079953d332f61a9d26656e4e36af606c (diff) | |
Merge remote-tracking branch 'qatar/master'
* qatar/master:
avformat: Avoid a warning about mixed declarations and code
BMV demuxer and decoder
matroskaenc: Make sure the seekhead struct is freed even on seek failure
mpeg12enc: Remove write-only variables.
mpeg12enc: Don't set up run-level info for level 0.
msmpeg4: Don't set up run-level info for level 0.
avformat: Warn about using network functions without calling avformat_network_init
avformat: Revise wording
rdt: Set AVFMT_NOFILE on ff_rdt_demuxer
rdt: Check the return value of avformat_open
rtsp: Discard the dynamic handler, if it has an alloc function which failed
dsputil: use cpuflags in x86 versions of vector_clip_int32()
Conflicts:
libavcodec/avcodec.h
libavcodec/version.h
libavformat/Makefile
libavformat/allformats.c
libavformat/version.h
Merged-by: Michael Niedermayer <michaelni@gmx.at>
Diffstat (limited to 'libavcodec/x86/dsputil_yasm.asm')
| -rw-r--r-- | libavcodec/x86/dsputil_yasm.asm | 40 |
1 files changed, 23 insertions, 17 deletions
diff --git a/libavcodec/x86/dsputil_yasm.asm b/libavcodec/x86/dsputil_yasm.asm index 6627d21bd8..82678f1880 100644 --- a/libavcodec/x86/dsputil_yasm.asm +++ b/libavcodec/x86/dsputil_yasm.asm @@ -1055,9 +1055,14 @@ emu_edge mmx ; int32_t max, unsigned int len) ;----------------------------------------------------------------------------- -%macro VECTOR_CLIP_INT32 4 -cglobal vector_clip_int32_%1, 5,5,%2, dst, src, min, max, len -%ifidn %1, sse2 +; %1 = number of xmm registers used +; %2 = number of inline load/process/store loops per asm loop +; %3 = process 4*mmsize (%3=0) or 8*mmsize (%3=1) bytes per loop +; %4 = CLIPD function takes min/max as float instead of int (CLIPD_SSE2) +; %5 = suffix +%macro VECTOR_CLIP_INT32 4-5 +cglobal vector_clip_int32%5, 5,5,%2, dst, src, min, max, len +%if %4 cvtsi2ss m4, minm cvtsi2ss m5, maxm %else @@ -1068,12 +1073,12 @@ cglobal vector_clip_int32_%1, 5,5,%2, dst, src, min, max, len SPLATD m5 .loop: %assign %%i 1 -%rep %3 +%rep %2 mova m0, [srcq+mmsize*0*%%i] mova m1, [srcq+mmsize*1*%%i] mova m2, [srcq+mmsize*2*%%i] mova m3, [srcq+mmsize*3*%%i] -%if %4 +%if %3 mova m7, [srcq+mmsize*4*%%i] mova m8, [srcq+mmsize*5*%%i] mova m9, [srcq+mmsize*6*%%i] @@ -1083,7 +1088,7 @@ cglobal vector_clip_int32_%1, 5,5,%2, dst, src, min, max, len CLIPD m1, m4, m5, m6 CLIPD m2, m4, m5, m6 CLIPD m3, m4, m5, m6 -%if %4 +%if %3 CLIPD m7, m4, m5, m6 CLIPD m8, m4, m5, m6 CLIPD m9, m4, m5, m6 @@ -1093,7 +1098,7 @@ cglobal vector_clip_int32_%1, 5,5,%2, dst, src, min, max, len mova [dstq+mmsize*1*%%i], m1 mova [dstq+mmsize*2*%%i], m2 mova [dstq+mmsize*3*%%i], m3 -%if %4 +%if %3 mova [dstq+mmsize*4*%%i], m7 mova [dstq+mmsize*5*%%i], m8 mova [dstq+mmsize*6*%%i], m9 @@ -1101,25 +1106,26 @@ cglobal vector_clip_int32_%1, 5,5,%2, dst, src, min, max, len %endif %assign %%i %%i+1 %endrep - add srcq, mmsize*4*(%3+%4) - add dstq, mmsize*4*(%3+%4) - sub lend, mmsize*(%3+%4) + add srcq, mmsize*4*(%2+%3) + add dstq, mmsize*4*(%2+%3) + sub lend, mmsize*(%2+%3) jg .loop REP_RET %endmacro -INIT_MMX +INIT_MMX mmx %define SPLATD SPLATD_MMX %define CLIPD CLIPD_MMX -VECTOR_CLIP_INT32 mmx, 0, 1, 0 -INIT_XMM +VECTOR_CLIP_INT32 0, 1, 0, 0 +INIT_XMM sse2 %define SPLATD SPLATD_SSE2 -VECTOR_CLIP_INT32 sse2_int, 6, 1, 0 +VECTOR_CLIP_INT32 6, 1, 0, 0, _int %define CLIPD CLIPD_SSE2 -VECTOR_CLIP_INT32 sse2, 6, 2, 0 +VECTOR_CLIP_INT32 6, 2, 0, 1 +INIT_XMM sse4 %define CLIPD CLIPD_SSE41 %ifdef m8 -VECTOR_CLIP_INT32 sse41, 11, 1, 1 +VECTOR_CLIP_INT32 11, 1, 1, 0 %else -VECTOR_CLIP_INT32 sse41, 6, 1, 0 +VECTOR_CLIP_INT32 6, 1, 0, 0 %endif |