RepoMirrors
/
ffmpeg
mirror of https://git.ffmpeg.org/ffmpeg.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse 

please complain if they are slower on sse2 cpus ...

Originally committed as revision 5976 to svn://svn.ffmpeg.org/ffmpeg/trunk
This commit is contained in:
Michael Niedermayer 2006-08-10 20:24:58 +00:00
parent eb4825b5d4
commit 1f1aa1d955
1 changed files with 22 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

44
libavcodec/i386/dsputil_mmx.c
View File

@ -2820,12 +2820,14 @@ static void vector_fmul_reverse_3dnow2(float *dst, const float *src0, const floa
); );
asm volatile("femms"); asm volatile("femms");
} }
static void vector_fmul_reverse_sse2(float *dst, const float *src0, const float *src1, int len){ static void vector_fmul_reverse_sse(float *dst, const float *src0, const float *src1, int len){
long i = len*4-32; long i = len*4-32;
asm volatile( asm volatile(
"1: \n\t" "1: \n\t"
"pshufd $0x1b, 16(%1), %%xmm0 \n\t" "movaps 16(%1), %%xmm0 \n\t"
"pshufd $0x1b, (%1), %%xmm1 \n\t" "movaps (%1), %%xmm1 \n\t"
"shufps $0x1b, %%xmm0, %%xmm0 \n\t"
"shufps $0x1b, %%xmm1, %%xmm1 \n\t"
"mulps (%3,%0), %%xmm0 \n\t" "mulps (%3,%0), %%xmm0 \n\t"
"mulps 16(%3,%0), %%xmm1 \n\t" "mulps 16(%3,%0), %%xmm1 \n\t"
"movaps %%xmm0, (%2,%0) \n\t" "movaps %%xmm0, (%2,%0) \n\t"
@ -2882,7 +2884,7 @@ static void vector_fmul_add_add_3dnow(float *dst, const float *src0, const float
ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step); ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);
asm volatile("femms"); asm volatile("femms");
} }
static void vector_fmul_add_add_sse2(float *dst, const float *src0, const float *src1, static void vector_fmul_add_add_sse(float *dst, const float *src0, const float *src1,
const float *src2, float src3, int len, int step){ const float *src2, float src3, int len, int step){
long i; long i;
if(step == 2 && src3 == 0){ if(step == 2 && src3 == 0){
@ -2896,20 +2898,20 @@ static void vector_fmul_add_add_sse2(float *dst, const float *src0, const float
"mulps 16(%3,%0), %%xmm1 \n\t" "mulps 16(%3,%0), %%xmm1 \n\t"
"addps (%4,%0), %%xmm0 \n\t" "addps (%4,%0), %%xmm0 \n\t"
"addps 16(%4,%0), %%xmm1 \n\t" "addps 16(%4,%0), %%xmm1 \n\t"
"movd %%xmm0, (%1) \n\t" "movss %%xmm0, (%1) \n\t"
"movd %%xmm1, 32(%1) \n\t" "movss %%xmm1, 32(%1) \n\t"
"psrldq $4, %%xmm0 \n\t" "movhlps %%xmm0, %%xmm2 \n\t"
"psrldq $4, %%xmm1 \n\t" "movhlps %%xmm1, %%xmm3 \n\t"
"movd %%xmm0, 8(%1) \n\t" "movss %%xmm2, 16(%1) \n\t"
"movd %%xmm1, 40(%1) \n\t" "movss %%xmm3, 48(%1) \n\t"
"psrldq $4, %%xmm0 \n\t" "shufps $0xb1, %%xmm0, %%xmm0 \n\t"
"psrldq $4, %%xmm1 \n\t" "shufps $0xb1, %%xmm1, %%xmm1 \n\t"
"movd %%xmm0, 16(%1) \n\t" "movss %%xmm0, 8(%1) \n\t"
"movd %%xmm1, 48(%1) \n\t" "movss %%xmm1, 40(%1) \n\t"
"psrldq $4, %%xmm0 \n\t" "movhlps %%xmm0, %%xmm2 \n\t"
"psrldq $4, %%xmm1 \n\t" "movhlps %%xmm1, %%xmm3 \n\t"
"movd %%xmm0, 24(%1) \n\t" "movss %%xmm2, 24(%1) \n\t"
"movd %%xmm1, 56(%1) \n\t" "movss %%xmm3, 56(%1) \n\t"
"sub $64, %1 \n\t" "sub $64, %1 \n\t"
"sub $32, %0 \n\t" "sub $32, %0 \n\t"
"jge 1b \n\t" "jge 1b \n\t"
@ -3403,10 +3405,8 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse; c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse;
c->vector_fmul = vector_fmul_sse; c->vector_fmul = vector_fmul_sse;
c->float_to_int16 = float_to_int16_sse; c->float_to_int16 = float_to_int16_sse;
} c->vector_fmul_reverse = vector_fmul_reverse_sse;
if(mm_flags & MM_SSE2){ c->vector_fmul_add_add = vector_fmul_add_add_sse;
c->vector_fmul_reverse = vector_fmul_reverse_sse2;
c->vector_fmul_add_add = vector_fmul_add_add_sse2;
} }
if(mm_flags & MM_3DNOW) if(mm_flags & MM_3DNOW)
c->vector_fmul_add_add = vector_fmul_add_add_3dnow; // faster than sse2 c->vector_fmul_add_add = vector_fmul_add_add_3dnow; // faster than sse2