[Ffmpeg-devel] [PATCH] another vorbis optimization
Loren Merritt
lorenm
Thu Aug 10 21:38:39 CEST 2006
On Thu, 10 Aug 2006, Michael Niedermayer wrote:
> On Wed, Aug 09, 2006 at 10:21:16PM -0700, Loren Merritt wrote:
> [...]
>> + if(vc->exp_bias){
>> + int i;
>> + float *win;
>> + win = av_malloc(vc->blocksize_0/2 * sizeof(float));
>> + for(i=0; i<vc->blocksize_0/2; i++)
>> + win[i] = vc->swin[i] * (1<<15);
>> + vc->swin = win;
>> + win = av_malloc(vc->blocksize_1/2 * sizeof(float));
>> + for(i=0; i<vc->blocksize_1/2; i++)
>> + win[i] = vc->lwin[i] * (1<<15);
>> + vc->lwin = win;
>> + }
>
> offtopic, but maybe we should change vc->lwin / vc->swin to vc->win[2]?
like so?
>> + int float_to_int16_expbias;
>> + int float_to_int16_addbias;
>
> hmm, what do these do in dsputil.h? typo?
yes, typo.
> exept that iam fine with the patch (maybe some of the sse functions
> contained sse2 instructions, iam not sure but its no problem ill fix
> it if it breaks unless someone else is quicker)
fixed.
--Loren Merritt
-------------- next part --------------
Index: vorbis.c
===================================================================
--- vorbis.c (revision 5975)
+++ vorbis.c (working copy)
@@ -161,8 +161,8 @@
av_freep(&vc->residues);
av_freep(&vc->modes);
- ff_mdct_end(&vc->mdct0);
- ff_mdct_end(&vc->mdct1);
+ ff_mdct_end(&vc->mdct[0]);
+ ff_mdct_end(&vc->mdct[1]);
for(i=0;i<vc->codebook_count;++i) {
av_free(vc->codebooks[i].codevectors);
@@ -194,8 +194,8 @@
av_freep(&vc->mappings);
if(vc->exp_bias){
- av_freep(&vc->swin);
- av_freep(&vc->lwin);
+ av_freep(&vc->win[0]);
+ av_freep(&vc->win[1]);
}
}
@@ -762,7 +762,7 @@
for (blockflag=0;blockflag<2;++blockflag)
{
- n=(blockflag ? vc->blocksize_1 : vc->blocksize_0) / 2;
+ n=(blockflag ? vc->blocksize[1] : vc->blocksize[0]) / 2;
floors[floor_number].data.t0.map[blockflag]=
av_malloc((n+1) * sizeof(int_fast32_t)); // n+sentinel
@@ -877,33 +877,30 @@
vc->bitrate_minimum=get_bits_long_le(gb, 32);
bl0=get_bits(gb, 4);
bl1=get_bits(gb, 4);
- vc->blocksize_0=(1<<bl0);
- vc->blocksize_1=(1<<bl1);
+ vc->blocksize[0]=(1<<bl0);
+ vc->blocksize[1]=(1<<bl1);
if (bl0>13 || bl0<6 || bl1>13 || bl1<6 || bl1<bl0) {
av_log(vc->avccontext, AV_LOG_ERROR, " Vorbis id header packet corrupt (illegal blocksize). \n");
return 3;
}
// output format int16
- if (vc->blocksize_1/2 * vc->audio_channels * 2 >
+ if (vc->blocksize[1]/2 * vc->audio_channels * 2 >
AVCODEC_MAX_AUDIO_FRAME_SIZE) {
av_log(vc->avccontext, AV_LOG_ERROR, "Vorbis channel count makes "
"output packets too large.\n");
return 4;
}
- vc->swin=vwin[bl0-6];
- vc->lwin=vwin[bl1-6];
+ vc->win[0]=vwin[bl0-6];
+ vc->win[1]=vwin[bl1-6];
if(vc->exp_bias){
- int i;
- float *win;
- win = av_malloc(vc->blocksize_0/2 * sizeof(float));
- for(i=0; i<vc->blocksize_0/2; i++)
- win[i] = vc->swin[i] * (1<<15);
- vc->swin = win;
- win = av_malloc(vc->blocksize_1/2 * sizeof(float));
- for(i=0; i<vc->blocksize_1/2; i++)
- win[i] = vc->lwin[i] * (1<<15);
- vc->lwin = win;
+ int i, j;
+ for(j=0; j<2; j++){
+ float *win = av_malloc(vc->blocksize[j]/2 * sizeof(float));
+ for(i=0; i<vc->blocksize[j]/2; i++)
+ win[i] = vc->win[j][i] * (1<<15);
+ vc->win[j] = win;
+ }
}
if ((get_bits1(gb)) == 0) {
@@ -911,24 +908,24 @@
return 2;
}
- vc->channel_residues=(float *)av_malloc((vc->blocksize_1/2)*vc->audio_channels * sizeof(float));
- vc->channel_floors=(float *)av_malloc((vc->blocksize_1/2)*vc->audio_channels * sizeof(float));
- vc->saved=(float *)av_malloc((vc->blocksize_1/2)*vc->audio_channels * sizeof(float));
- vc->ret=(float *)av_malloc((vc->blocksize_1/2)*vc->audio_channels * sizeof(float));
- vc->buf=(float *)av_malloc(vc->blocksize_1 * sizeof(float));
- vc->buf_tmp=(float *)av_malloc(vc->blocksize_1 * sizeof(float));
+ vc->channel_residues=(float *)av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float));
+ vc->channel_floors=(float *)av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float));
+ vc->saved=(float *)av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float));
+ vc->ret=(float *)av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float));
+ vc->buf=(float *)av_malloc(vc->blocksize[1] * sizeof(float));
+ vc->buf_tmp=(float *)av_malloc(vc->blocksize[1] * sizeof(float));
vc->saved_start=0;
- ff_mdct_init(&vc->mdct0, bl0, 1);
- ff_mdct_init(&vc->mdct1, bl1, 1);
+ ff_mdct_init(&vc->mdct[0], bl0, 1);
+ ff_mdct_init(&vc->mdct[1], bl1, 1);
AV_DEBUG(" vorbis version %d \n audio_channels %d \n audio_samplerate %d \n bitrate_max %d \n bitrate_nom %d \n bitrate_min %d \n blk_0 %d blk_1 %d \n ",
- vc->version, vc->audio_channels, vc->audio_samplerate, vc->bitrate_maximum, vc->bitrate_nominal, vc->bitrate_minimum, vc->blocksize_0, vc->blocksize_1);
+ vc->version, vc->audio_channels, vc->audio_samplerate, vc->bitrate_maximum, vc->bitrate_nominal, vc->bitrate_minimum, vc->blocksize[0], vc->blocksize[1]);
/*
- BLK=vc->blocksize_0;
+ BLK=vc->blocksize[0];
for(i=0;i<BLK/2;++i) {
- vc->swin[i]=sin(0.5*3.14159265358*(sin(((float)i+0.5)/(float)BLK*3.14159265358))*(sin(((float)i+0.5)/(float)BLK*3.14159265358)));
+ vc->win[0][i]=sin(0.5*3.14159265358*(sin(((float)i+0.5)/(float)BLK*3.14159265358))*(sin(((float)i+0.5)/(float)BLK*3.14159265358)));
}
*/
@@ -1537,7 +1534,7 @@
next_window=get_bits1(gb);
}
- blocksize=vc->modes[mode_number].blockflag ? vc->blocksize_1 : vc->blocksize_0;
+ blocksize=vc->blocksize[vc->modes[mode_number].blockflag];
memset(ch_res_ptr, 0, sizeof(float)*vc->audio_channels*blocksize/2); //FIXME can this be removed ?
memset(ch_floor_ptr, 0, sizeof(float)*vc->audio_channels*blocksize/2); //FIXME can this be removed ?
@@ -1610,10 +1607,10 @@
for(j=0;j<vc->audio_channels;++j) {
uint_fast8_t step=vc->audio_channels;
uint_fast16_t k;
- float *saved=vc->saved+j*vc->blocksize_1/2;
+ float *saved=vc->saved+j*vc->blocksize[1]/2;
float *ret=vc->ret;
- const float *lwin=vc->lwin;
- const float *swin=vc->swin;
+ const float *lwin=vc->win[1];
+ const float *swin=vc->win[0];
float *buf=vc->buf;
float *buf_tmp=vc->buf_tmp;
@@ -1621,20 +1618,20 @@
saved_start=vc->saved_start;
- vc->mdct0.fft.imdct_calc(vc->modes[mode_number].blockflag ? &vc->mdct1 : &vc->mdct0, buf, ch_floor_ptr, buf_tmp);
+ vc->mdct[0].fft.imdct_calc(&vc->mdct[vc->modes[mode_number].blockflag], buf, ch_floor_ptr, buf_tmp);
//FIXME process channels together, to allow faster simd vector_fmul_add_add?
if (vc->modes[mode_number].blockflag) {
// -- overlap/add
if (previous_window) {
- vc->dsp.vector_fmul_add_add(ret+j, buf, lwin, saved, vc->add_bias, vc->blocksize_1/2, step);
- retlen=vc->blocksize_1/2;
+ vc->dsp.vector_fmul_add_add(ret+j, buf, lwin, saved, vc->add_bias, vc->blocksize[1]/2, step);
+ retlen=vc->blocksize[1]/2;
} else {
- int len = (vc->blocksize_1-vc->blocksize_0)/4;
+ int len = (vc->blocksize[1]-vc->blocksize[0])/4;
buf += len;
- vc->dsp.vector_fmul_add_add(ret+j, buf, swin, saved, vc->add_bias, vc->blocksize_0/2, step);
- k = vc->blocksize_0/2*step + j;
- buf += vc->blocksize_0/2;
+ vc->dsp.vector_fmul_add_add(ret+j, buf, swin, saved, vc->add_bias, vc->blocksize[0]/2, step);
+ k = vc->blocksize[0]/2*step + j;
+ buf += vc->blocksize[0]/2;
if(vc->exp_bias){
for(i=0; i<len; i++, k+=step)
((uint32_t*)ret)[k] = ((uint32_t*)buf)[i] + vc->exp_bias; // ret[k]=buf[i]*(1<<bias)
@@ -1643,19 +1640,19 @@
ret[k] = buf[i] + fadd_bias;
}
buf=vc->buf;
- retlen=vc->blocksize_0/2+len;
+ retlen=vc->blocksize[0]/2+len;
}
// -- save
if (next_window) {
- buf += vc->blocksize_1/2;
- vc->dsp.vector_fmul_reverse(saved, buf, lwin, vc->blocksize_1/2);
+ buf += vc->blocksize[1]/2;
+ vc->dsp.vector_fmul_reverse(saved, buf, lwin, vc->blocksize[1]/2);
saved_start=0;
} else {
- saved_start=(vc->blocksize_1-vc->blocksize_0)/4;
- buf += vc->blocksize_1/2;
+ saved_start=(vc->blocksize[1]-vc->blocksize[0])/4;
+ buf += vc->blocksize[1]/2;
for(i=0; i<saved_start; i++)
((uint32_t*)saved)[i] = ((uint32_t*)buf)[i] + vc->exp_bias;
- vc->dsp.vector_fmul_reverse(saved+saved_start, buf+saved_start, swin, vc->blocksize_0/2);
+ vc->dsp.vector_fmul_reverse(saved+saved_start, buf+saved_start, swin, vc->blocksize[0]/2);
}
} else {
// --overlap/add
@@ -1666,11 +1663,11 @@
for(k=j, i=0;i<saved_start;++i, k+=step)
ret[k] = saved[i];
}
- vc->dsp.vector_fmul_add_add(ret+k, buf, swin, saved+saved_start, vc->add_bias, vc->blocksize_0/2, step);
- retlen=saved_start+vc->blocksize_0/2;
+ vc->dsp.vector_fmul_add_add(ret+k, buf, swin, saved+saved_start, vc->add_bias, vc->blocksize[0]/2, step);
+ retlen=saved_start+vc->blocksize[0]/2;
// -- save
- buf += vc->blocksize_0/2;
- vc->dsp.vector_fmul_reverse(saved, buf, swin, vc->blocksize_0/2);
+ buf += vc->blocksize[0]/2;
+ vc->dsp.vector_fmul_reverse(saved, buf, swin, vc->blocksize[0]/2);
saved_start=0;
}
}
Index: vorbis.h
===================================================================
--- vorbis.h (revision 5975)
+++ vorbis.h (working copy)
@@ -89,8 +89,7 @@
GetBitContext gb;
DSPContext dsp;
- MDCTContext mdct0;
- MDCTContext mdct1;
+ MDCTContext mdct[2];
uint_fast8_t first_frame;
uint_fast32_t version;
uint_fast8_t audio_channels;
@@ -98,10 +97,8 @@
uint_fast32_t bitrate_maximum;
uint_fast32_t bitrate_nominal;
uint_fast32_t bitrate_minimum;
- uint_fast32_t blocksize_0;
- uint_fast32_t blocksize_1;
- const float * swin;
- const float * lwin;
+ uint_fast32_t blocksize[2];
+ const float * win[2];
uint_fast16_t codebook_count;
vorbis_codebook *codebooks;
uint_fast8_t floor_count;
More information about the ffmpeg-devel
mailing list