diff options
| author | Loren Merritt <lorenm@u.washington.edu> | 2008-07-13 15:03:58 +0000 |
|---|---|---|
| committer | Loren Merritt <lorenm@u.washington.edu> | 2008-07-13 15:03:58 +0000 |
| commit | b9fa32082c71013e90eab9e9997967d2939cf4a6 (patch) | |
| tree | 83edd135988c73a75b017fbd12396e156de5e0a4 /libavcodec/vorbis_dec.c | |
| parent | eb2cd99c73df74cba8ce0173f9ee2b70313adaa6 (diff) | |
| download | ffmpeg-b9fa32082c71013e90eab9e9997967d2939cf4a6.tar.gz | |
exploit mdct symmetry
2% faster vorbis on conroe, k8. 7% on celeron.
Originally committed as revision 14207 to svn://svn.ffmpeg.org/ffmpeg/trunk
Diffstat (limited to 'libavcodec/vorbis_dec.c')
| -rw-r--r-- | libavcodec/vorbis_dec.c | 20 |
1 files changed, 10 insertions, 10 deletions
diff --git a/libavcodec/vorbis_dec.c b/libavcodec/vorbis_dec.c index ed13229cd4..d4a5402d86 100644 --- a/libavcodec/vorbis_dec.c +++ b/libavcodec/vorbis_dec.c @@ -899,10 +899,10 @@ static int vorbis_parse_id_hdr(vorbis_context *vc){ vc->channel_residues= av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float)); vc->channel_floors = av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float)); - vc->saved = av_mallocz((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float)); + vc->saved = av_mallocz((vc->blocksize[1]/4)*vc->audio_channels * sizeof(float)); vc->ret = av_malloc((vc->blocksize[1]/2)*vc->audio_channels * sizeof(float)); - vc->buf = av_malloc( vc->blocksize[1] * sizeof(float)); - vc->buf_tmp = av_malloc( vc->blocksize[1] * sizeof(float)); + vc->buf = av_malloc( vc->blocksize[1]/2 * sizeof(float)); + vc->buf_tmp = av_malloc( vc->blocksize[1]/2 * sizeof(float)); vc->previous_window=0; ff_mdct_init(&vc->mdct[0], bl0, 1); @@ -1520,23 +1520,23 @@ static int vorbis_parse_audio_packet(vorbis_context *vc) { for(j=0;j<vc->audio_channels;++j) { uint_fast16_t bs0=vc->blocksize[0]; uint_fast16_t bs1=vc->blocksize[1]; - float *saved=vc->saved+j*bs1/2; + float *saved=vc->saved+j*bs1/4; float *ret=vc->ret+j*retlen; float *buf=vc->buf; const float *win=vc->win[blockflag&previous_window]; - vc->mdct[0].fft.imdct_calc(&vc->mdct[blockflag], buf, vc->channel_floors+j*blocksize/2, vc->buf_tmp); + vc->mdct[0].fft.imdct_half(&vc->mdct[blockflag], buf, vc->channel_floors+j*blocksize/2, vc->buf_tmp); if(blockflag == previous_window) { - vc->dsp.vector_fmul_window(ret, saved, buf, win, fadd_bias, blocksize/2); + vc->dsp.vector_fmul_window(ret, saved, buf, win, fadd_bias, blocksize/4); } else if(blockflag > previous_window) { - vc->dsp.vector_fmul_window(ret, saved, buf+(bs1-bs0)/4, win, fadd_bias, bs0/2); - copy_normalize(ret+bs0/2, buf+(bs1+bs0)/4, (bs1-bs0)/4, vc->exp_bias, fadd_bias); + vc->dsp.vector_fmul_window(ret, saved, buf, win, fadd_bias, bs0/4); + copy_normalize(ret+bs0/2, buf+bs0/4, (bs1-bs0)/4, vc->exp_bias, fadd_bias); } else { copy_normalize(ret, saved, (bs1-bs0)/4, vc->exp_bias, fadd_bias); - vc->dsp.vector_fmul_window(ret+(bs1-bs0)/4, saved+(bs1-bs0)/4, buf, win, fadd_bias, bs0/2); + vc->dsp.vector_fmul_window(ret+(bs1-bs0)/4, saved+(bs1-bs0)/4, buf, win, fadd_bias, bs0/4); } - memcpy(saved, buf+blocksize/2, blocksize/2*sizeof(float)); + memcpy(saved, buf+blocksize/4, blocksize/4*sizeof(float)); } vc->previous_window = blockflag; |