ffmpeg/FFmpeg
1
0
Fork 0
mirror of https://git.ffmpeg.org/ffmpeg.git synced 2024-09-20 13:26:39 +00:00
Code Issues Packages Projects Releases Wiki Activity

vorbis simd tweaks

Browse Source 
Originally committed as revision 5983 to svn://svn.ffmpeg.org/ffmpeg/trunk
This commit is contained in:
Loren Merritt 2006-08-11 18:19:37 +00:00
parent 49a0122c99
commit 069720565c
3 changed files with 69 additions and 53 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
libavcodec/dsputil.h
View File

@ -313,8 +313,8 @@ typedef struct DSPContext {
/* assume len is a multiple of 4, and arrays are 16-byte aligned */ /* assume len is a multiple of 4, and arrays are 16-byte aligned */
void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize); void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
void (*vector_fmul)(float *dst, const float *src, int len);
/* assume len is a multiple of 8, and arrays are 16-byte aligned */ /* assume len is a multiple of 8, and arrays are 16-byte aligned */
void (*vector_fmul)(float *dst, const float *src, int len);
void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len); void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
/* assume len is a multiple of 8, and src arrays are 16-byte aligned */ /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step); void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);

118
libavcodec/i386/dsputil_mmx.c
View File

@ -2773,33 +2773,39 @@ static void vorbis_inverse_coupling_sse(float *mag, float *ang, int blocksize)
} }
static void vector_fmul_3dnow(float *dst, const float *src, int len){ static void vector_fmul_3dnow(float *dst, const float *src, int len){
long i; long i = (len-4)*4;
len >>= 1; asm volatile(
for(i=0; i<len; i++) { "1: \n\t"
asm volatile( "movq (%1,%0), %%mm0 \n\t"
"movq %0, %%mm0 \n\t" "movq 8(%1,%0), %%mm1 \n\t"
"pfmul %1, %%mm0 \n\t" "pfmul (%2,%0), %%mm0 \n\t"
"movq %%mm0, %0 \n\t" "pfmul 8(%2,%0), %%mm1 \n\t"
:"+m"(dst[i*2]) "movq %%mm0, (%1,%0) \n\t"
:"m"(src[i*2]) "movq %%mm1, 8(%1,%0) \n\t"
:"memory" "sub $16, %0 \n\t"
); "jge 1b \n\t"
} "femms \n\t"
asm volatile("femms"); :"+r"(i)
:"r"(dst), "r"(src)
:"memory"
);
} }
static void vector_fmul_sse(float *dst, const float *src, int len){ static void vector_fmul_sse(float *dst, const float *src, int len){
long i; long i = (len-8)*4;
len >>= 2; asm volatile(
for(i=0; i<len; i++) { "1: \n\t"
asm volatile( "movaps (%1,%0), %%xmm0 \n\t"
"movaps %0, %%xmm0 \n\t" "movaps 16(%1,%0), %%xmm1 \n\t"
"mulps %1, %%xmm0 \n\t" "mulps (%2,%0), %%xmm0 \n\t"
"movaps %%xmm0, %0 \n\t" "mulps 16(%2,%0), %%xmm1 \n\t"
:"+m"(dst[i*4]) "movaps %%xmm0, (%1,%0) \n\t"
:"m"(src[i*4]) "movaps %%xmm1, 16(%1,%0) \n\t"
:"memory" "sub $32, %0 \n\t"
); "jge 1b \n\t"
} :"+r"(i)
:"r"(dst), "r"(src)
:"memory"
);
} }
static void vector_fmul_reverse_3dnow2(float *dst, const float *src0, const float *src1, int len){ static void vector_fmul_reverse_3dnow2(float *dst, const float *src0, const float *src1, int len){
@ -2842,9 +2848,8 @@ static void vector_fmul_reverse_sse(float *dst, const float *src0, const float *
static void vector_fmul_add_add_3dnow(float *dst, const float *src0, const float *src1, static void vector_fmul_add_add_3dnow(float *dst, const float *src0, const float *src1,
const float *src2, int src3, int len, int step){ const float *src2, int src3, int len, int step){
long i; long i = (len-4)*4;
if(step == 2 && src3 == 0){ if(step == 2 && src3 == 0){
i = (len-4)*4;
dst += (len-4)*2; dst += (len-4)*2;
asm volatile( asm volatile(
"1: \n\t" "1: \n\t"
@ -2869,26 +2874,31 @@ static void vector_fmul_add_add_3dnow(float *dst, const float *src0, const float
); );
} }
else if(step == 1 && src3 == 0){ else if(step == 1 && src3 == 0){
for(i=0; i<len; i+=2){ asm volatile(
asm volatile( "1: \n\t"
"movq %1, %%mm0 \n\t" "movq (%2,%0), %%mm0 \n\t"
"pfmul %2, %%mm0 \n\t" "movq 8(%2,%0), %%mm1 \n\t"
"pfadd %3, %%mm0 \n\t" "pfmul (%3,%0), %%mm0 \n\t"
"movq %%mm0, %0 \n\t" "pfmul 8(%3,%0), %%mm1 \n\t"
:"=m"(dst[i]) "pfadd (%4,%0), %%mm0 \n\t"
:"m"(src0[i]), "m"(src1[i]), "m"(src2[i]) "pfadd 8(%4,%0), %%mm1 \n\t"
); "movq %%mm0, (%1,%0) \n\t"
} "movq %%mm1, 8(%1,%0) \n\t"
"sub $16, %0 \n\t"
"jge 1b \n\t"
:"+r"(i)
:"r"(dst), "r"(src0), "r"(src1), "r"(src2)
:"memory"
);
} }
else else
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_sse(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, int src3, int len, int step){
long i; long i = (len-8)*4;
if(step == 2 && src3 == 0){ if(step == 2 && src3 == 0){
i = (len-8)*4;
dst += (len-8)*2; dst += (len-8)*2;
asm volatile( asm volatile(
"1: \n\t" "1: \n\t"
@ -2921,16 +2931,22 @@ static void vector_fmul_add_add_sse(float *dst, const float *src0, const float *
); );
} }
else if(step == 1 && src3 == 0){ else if(step == 1 && src3 == 0){
for(i=0; i<len; i+=4){ asm volatile(
asm volatile( "1: \n\t"
"movaps %1, %%xmm0 \n\t" "movaps (%2,%0), %%xmm0 \n\t"
"mulps %2, %%xmm0 \n\t" "movaps 16(%2,%0), %%xmm1 \n\t"
"addps %3, %%xmm0 \n\t" "mulps (%3,%0), %%xmm0 \n\t"
"movaps %%xmm0, %0 \n\t" "mulps 16(%3,%0), %%xmm1 \n\t"
:"=m"(dst[i]) "addps (%4,%0), %%xmm0 \n\t"
:"m"(src0[i]), "m"(src1[i]), "m"(src2[i]) "addps 16(%4,%0), %%xmm1 \n\t"
); "movaps %%xmm0, (%1,%0) \n\t"
} "movaps %%xmm1, 16(%1,%0) \n\t"
"sub $32, %0 \n\t"
"jge 1b \n\t"
:"+r"(i)
:"r"(dst), "r"(src0), "r"(src1), "r"(src2)
:"memory"
);
} }
else else
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);
@ -3409,7 +3425,7 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
c->vector_fmul_add_add = vector_fmul_add_add_sse; c->vector_fmul_add_add = vector_fmul_add_add_sse;
} }
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 sse
} }
#ifdef CONFIG_ENCODERS #ifdef CONFIG_ENCODERS

2
libavcodec/vorbis.c
View File

@ -762,7 +762,7 @@ static void create_map( vorbis_context * vc, uint_fast8_t floor_number )
for (blockflag=0;blockflag<2;++blockflag) for (blockflag=0;blockflag<2;++blockflag)
{ {
n=(blockflag ? vc->blocksize[1] : vc->blocksize[0]) / 2; n=vc->blocksize[blockflag]/2;
floors[floor_number].data.t0.map[blockflag]= floors[floor_number].data.t0.map[blockflag]=
av_malloc((n+1) * sizeof(int_fast32_t)); // n+sentinel av_malloc((n+1) * sizeof(int_fast32_t)); // n+sentinel