Skip to content

F
ffmpeg.wasm-core
Commit f4e61cd1 authored by Michael Niedermayer's avatar Michael Niedermayer
Browse files
Options

more trivial optimizations 

Originally committed as revision 4326 to svn://svn.ffmpeg.org/ffmpeg/trunk
parent 568d4b81
Hide whitespace changes
Inline Side-by-side
Showing with 27 additions and 27 deletions
libavcodec/vorbis.c
View file @ f4e61cd1
...@@ -37,7 +37,8 @@ ...@@ -37,7 +37,8 @@
#define AV_DEBUG(...) #define AV_DEBUG(...)
#endif #endif
#undef NDEBUG
#include <assert.h>
/* Helper functions */ /* Helper functions */
...@@ -1099,10 +1100,11 @@ static int vorbis_residue_decode(vorbis_context *vc, vorbis_residue *vr, uint_fa ...@@ -1099,10 +1100,11 @@ static int vorbis_residue_decode(vorbis_context *vc, vorbis_residue *vr, uint_fa
AV_DEBUG("Classword: %d \n", temp); AV_DEBUG("Classword: %d \n", temp);
assert(vr->classifications > 1 && vr->classifications<256 && temp<=65536); //needed for inverse[]
for(i=0;i<c_p_c;++i) { for(i=0;i<c_p_c;++i) {
uint_fast32_t temp2; uint_fast32_t temp2;
temp2=temp/vr->classifications; temp2=(((uint_fast64_t)temp) * inverse[vr->classifications])>>32;
classifs[j_times_ptns_to_read+partition_count+c_p_c-1-i]=temp-temp2*vr->classifications; classifs[j_times_ptns_to_read+partition_count+c_p_c-1-i]=temp-temp2*vr->classifications;
temp=temp2; temp=temp2;
} }
...@@ -1120,41 +1122,40 @@ static int vorbis_residue_decode(vorbis_context *vc, vorbis_residue *vr, uint_fa ...@@ -1120,41 +1122,40 @@ static int vorbis_residue_decode(vorbis_context *vc, vorbis_residue *vr, uint_fa
if (vqbook>=0) { if (vqbook>=0) {
uint_fast16_t coffs; uint_fast16_t coffs;
uint_fast16_t step=vr->partition_size/vc->codebooks[vqbook].dimensions;
vorbis_codebook codebook= vc->codebooks[vqbook];
if (vr->type==0) { if (vr->type==0) {
uint_fast16_t step=vr->partition_size/vc->codebooks[vqbook].dimensions;
voffs=voffset+j*vlen; voffs=voffset+j*vlen;
for(k=0;k<step;++k) { for(k=0;k<step;++k) {
coffs=get_vlc2(gb, vc->codebooks[vqbook].vlc.table, coffs=get_vlc2(gb, codebook.vlc.table, V_NB_BITS, 3) * codebook.dimensions;
V_NB_BITS, 3) * vc->codebooks[vqbook].dimensions; for(l=0;l<codebook.dimensions;++l) {
for(l=0;l<vc->codebooks[vqbook].dimensions;++l) { vec[voffs+k+l*step]+=codebook.codevectors[coffs+l]; // FPMATH
vec[voffs+k+l*step]+=vc->codebooks[vqbook].codevectors[coffs+l]; // FPMATH
} }
} }
} }
else if (vr->type==1) { else if (vr->type==1) {
voffs=voffset+j*vlen; voffs=voffset+j*vlen;
for(k=0;k<vr->partition_size/vc->codebooks[vqbook].dimensions;++k) { for(k=0;k<step;++k) {
coffs=get_vlc2(gb, vc->codebooks[vqbook].vlc.table, coffs=get_vlc2(gb, codebook.vlc.table, V_NB_BITS, 3) * codebook.dimensions;
V_NB_BITS, 3) * vc->codebooks[vqbook].dimensions; for(l=0;l<codebook.dimensions;++l, ++voffs) {
for(l=0;l<vc->codebooks[vqbook].dimensions;++l, ++voffs) { vec[voffs]+=codebook.codevectors[coffs+l]; // FPMATH
vec[voffs]+=vc->codebooks[vqbook].codevectors[coffs+l]; // FPMATH
AV_DEBUG(" pass %d offs: %d curr: %f change: %f cv offs.: %d \n", pass, voffs, vec[voffs], vc->codebooks[vqbook].codevectors[coffs+l], coffs); AV_DEBUG(" pass %d offs: %d curr: %f change: %f cv offs.: %d \n", pass, voffs, vec[voffs], codebook.codevectors[coffs+l], coffs);
} }
} }
} }
else if (vr->type==2 && ch==2) { // most frequent case optimized else if (vr->type==2 && ch==2 && (voffset&1)==0 && (codebook.dimensions&1)==0) { // most frequent case optimized
voffs=voffset; voffs=voffset>>1;
for(k=0;k<vr->partition_size/vc->codebooks[vqbook].dimensions;++k) { for(k=0;k<step;++k) {
coffs=get_vlc2(gb, vc->codebooks[vqbook].vlc.table, coffs=get_vlc2(gb, codebook.vlc.table, V_NB_BITS, 3) * codebook.dimensions;
V_NB_BITS, 3) * vc->codebooks[vqbook].dimensions; for(l=0;l<codebook.dimensions;l+=2, voffs++) {
for(l=0;l<vc->codebooks[vqbook].dimensions;++l, ++voffs) { vec[voffs ]+=codebook.codevectors[coffs+l ]; // FPMATH
vec[(voffs>>1)+((voffs&1) ? vlen : 0)]+=vc->codebooks[vqbook].codevectors[coffs+l]; // FPMATH vec[voffs+vlen]+=codebook.codevectors[coffs+l+1]; // FPMATH
AV_DEBUG(" pass %d offs: %d curr: %f change: %f cv offs.: %d+%d \n", pass, voffset/ch+(voffs%ch)*vlen, vec[voffset/ch+(voffs%ch)*vlen], vc->codebooks[vqbook].codevectors[coffs+l], coffs, l); AV_DEBUG(" pass %d offs: %d curr: %f change: %f cv offs.: %d+%d \n", pass, voffset/ch+(voffs%ch)*vlen, vec[voffset/ch+(voffs%ch)*vlen], codebook.codevectors[coffs+l], coffs, l);
} }
} }
...@@ -1162,13 +1163,12 @@ static int vorbis_residue_decode(vorbis_context *vc, vorbis_residue *vr, uint_fa ...@@ -1162,13 +1163,12 @@ static int vorbis_residue_decode(vorbis_context *vc, vorbis_residue *vr, uint_fa
else if (vr->type==2) { else if (vr->type==2) {
voffs=voffset; voffs=voffset;
for(k=0;k<vr->partition_size/vc->codebooks[vqbook].dimensions;++k) { for(k=0;k<step;++k) {
coffs=get_vlc2(gb, vc->codebooks[vqbook].vlc.table, coffs=get_vlc2(gb, codebook.vlc.table, V_NB_BITS, 3) * codebook.dimensions;
V_NB_BITS, 3) * vc->codebooks[vqbook].dimensions; for(l=0;l<codebook.dimensions;++l, ++voffs) {
for(l=0;l<vc->codebooks[vqbook].dimensions;++l, ++voffs) { vec[voffs/ch+(voffs%ch)*vlen]+=codebook.codevectors[coffs+l]; // FPMATH FIXME use if and counter instead of / and %
vec[voffs/ch+(voffs%ch)*vlen]+=vc->codebooks[vqbook].codevectors[coffs+l]; // FPMATH FIXME use if and counter instead of / and %
AV_DEBUG(" pass %d offs: %d curr: %f change: %f cv offs.: %d+%d \n", pass, voffset/ch+(voffs%ch)*vlen, vec[voffset/ch+(voffs%ch)*vlen], vc->codebooks[vqbook].codevectors[coffs+l], coffs, l); AV_DEBUG(" pass %d offs: %d curr: %f change: %f cv offs.: %d+%d \n", pass, voffset/ch+(voffs%ch)*vlen, vec[voffset/ch+(voffs%ch)*vlen], codebook.codevectors[coffs+l], coffs, l);
} }
} }
} else { } else {
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment