diff options
| author | Michael Niedermayer <michaelni@gmx.at> | 2014-11-22 20:38:30 +0100 |
|---|---|---|
| committer | Michael Niedermayer <michaelni@gmx.at> | 2014-11-22 20:40:36 +0100 |
| commit | 4327088da355509b9f97e4bdb658456ee56de72b (patch) | |
| tree | 427b036667704cf182955f5e8b35f4bc80449654 | |
| parent | 2f6550bb9a8d71f56fefbdbf849f17de22b57e66 (diff) | |
| download | ffmpeg-4327088da355509b9f97e4bdb658456ee56de72b.tar.gz | |
avcodec/x86/lossless_audiodsp: support len %16 == 8 in scalarproduct_and_madd_int16()
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
| -rw-r--r-- | libavcodec/lossless_audiodsp.h | 2 | ||||
| -rw-r--r-- | libavcodec/x86/lossless_audiodsp.asm | 6 |
2 files changed, 7 insertions, 1 deletions
diff --git a/libavcodec/lossless_audiodsp.h b/libavcodec/lossless_audiodsp.h index 4c27502841..c3ee2bec54 100644 --- a/libavcodec/lossless_audiodsp.h +++ b/libavcodec/lossless_audiodsp.h @@ -29,7 +29,7 @@ typedef struct LLAudDSPContext { /** * Calculate scalar product of v1 and v2, * and v1[i] += v3[i] * mul - * @param len length of vectors, should be multiple of 16 + * @param len length of vectors, should be multiple of 8 */ int32_t (*scalarproduct_and_madd_int16)(int16_t *v1 /* align 16 */, const int16_t *v2, diff --git a/libavcodec/x86/lossless_audiodsp.asm b/libavcodec/x86/lossless_audiodsp.asm index 64b769f7d4..cc5e5e5fac 100644 --- a/libavcodec/x86/lossless_audiodsp.asm +++ b/libavcodec/x86/lossless_audiodsp.asm @@ -26,6 +26,10 @@ SECTION_TEXT ; int ff_scalarproduct_and_madd_int16(int16_t *v1, int16_t *v2, int16_t *v3, ; int order, int mul) cglobal scalarproduct_and_madd_int16, 4,4,8, v1, v2, v3, order, mul +%if mmsize == 16 + test orderq, 8 + jnz scalarproduct_and_madd_int16_mmxext; +%endif shl orderq, 1 movd m7, mulm %if mmsize == 16 @@ -117,6 +121,8 @@ align 16 ; int order, int mul) INIT_XMM ssse3 cglobal scalarproduct_and_madd_int16, 4,5,10, v1, v2, v3, order, mul + test orderq, 8 + jnz scalarproduct_and_madd_int16_mmxext; shl orderq, 1 movd m7, mulm pshuflw m7, m7, 0 |