May 2015 - ffmpeg-cvslog - lists.ffmpeg.org

mov: reject zero bytes_per_frame with non-zero samples_per_frame
by git＠videolan.org 31 May '15

31 May '15

ffmpeg | branch: master | Andreas Cadhalpun <Andreas.Cadhalpun(a)googlemail.com> | Tue May 26 14:24:37 2015 +0100| [cffb9ea81bd16d2ed8bce187b6bf0279c07b3b9b] | committer: Luca Barbato mov: reject zero bytes_per_frame with non-zero samples_per_frame In this case the mov demuxer can return a large number of empty packets. Signed-off-by: Andreas Cadhalpun <Andreas.Cadhalpun(a)googlemail.com> Signed-off-by: Luca Barbato <lu_zero(a)gentoo.org> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=cffb9ea81bd16d2ed… --- libavformat/mov.c | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/libavformat/mov.c b/libavformat/mov.c index fbbfef0..2cf6e8e 100644 --- a/libavformat/mov.c +++ b/libavformat/mov.c @@ -2263,6 +2263,13 @@ static void mov_build_index(MOVContext *mov, AVStream *st) AVIndexEntry *e; unsigned size, samples; + if (sc->samples_per_frame > 1 && !sc->bytes_per_frame) { + avpriv_request_sample(mov->fc, + "Zero bytes per frame, but %d samples per frame", + sc->samples_per_frame); + return; + } + if (sc->samples_per_frame >= 160) { // gsm samples = sc->samples_per_frame; size = sc->bytes_per_frame;

1 0

mov: abort on EOF in ff_mov_read_chan
by git＠videolan.org 31 May '15

31 May '15

ffmpeg | branch: master | Andreas Cadhalpun <Andreas.Cadhalpun(a)googlemail.com> | Tue May 26 14:24:35 2015 +0100| [8e9c39e81fe5ba34010a7ba05cbe4ae31f177d89] | committer: Luca Barbato mov: abort on EOF in ff_mov_read_chan Otherwise the loop can take a lot of time if num_descr is very large. Signed-off-by: Andreas Cadhalpun <Andreas.Cadhalpun(a)googlemail.com> Signed-off-by: Luca Barbato <lu_zero(a)gentoo.org> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=8e9c39e81fe5ba340… --- libavformat/mov_chan.c | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/libavformat/mov_chan.c b/libavformat/mov_chan.c index 2c54920..94b93cf 100644 --- a/libavformat/mov_chan.c +++ b/libavformat/mov_chan.c @@ -565,6 +565,11 @@ int ff_mov_read_chan(AVFormatContext *s, AVIOContext *pb, AVStream *st, label_mask = 0; for (i = 0; i < num_descr; i++) { uint32_t label; + if (pb->eof_reached) { + av_log(s, AV_LOG_ERROR, + "reached EOF while reading channel layout\n"); + return AVERROR_INVALIDDATA; + } label = avio_rb32(pb); // mChannelLabel avio_rb32(pb); // mChannelFlags avio_rl32(pb); // mCoordinates[0]

1 0

mov: Correctly allocate ctts_data
by git＠videolan.org 31 May '15

31 May '15

ffmpeg | branch: master | Luca Barbato <lu_zero(a)gentoo.org> | Sun May 31 12:28:33 2015 +0200| [a8a90906fb0ad0f8d3c68a97f182528ee3b5198c] | committer: Luca Barbato mov: Correctly allocate ctts_data It can be reallocated. CC: libav-stable(a)libav.org > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=a8a90906fb0ad0f8d… --- libavformat/mov.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/libavformat/mov.c b/libavformat/mov.c index 33e1054..fbbfef0 100644 --- a/libavformat/mov.c +++ b/libavformat/mov.c @@ -2035,7 +2035,7 @@ static int mov_read_ctts(MOVContext *c, AVIOContext *pb, MOVAtom atom) return 0; if (entries >= UINT_MAX / sizeof(*sc->ctts_data)) return AVERROR_INVALIDDATA; - sc->ctts_data = av_malloc(entries * sizeof(*sc->ctts_data)); + sc->ctts_data = av_realloc(NULL, entries * sizeof(*sc->ctts_data)); if (!sc->ctts_data) return AVERROR(ENOMEM);

1 0

Merge commit 'd68c05380cebf563915412182643a8be04ef890b'
by git＠videolan.org 31 May '15

31 May '15

ffmpeg | branch: master | Michael Niedermayer <michaelni(a)gmx.at> | Mon Jun 1 00:22:23 2015 +0200| [b66b1a103fff88682fab7960e698080f6b847098] | committer: Michael Niedermayer Merge commit 'd68c05380cebf563915412182643a8be04ef890b' * commit 'd68c05380cebf563915412182643a8be04ef890b': x86: check for AV_CPU_FLAG_AVXSLOW where useful See: c16e99e3b3c02edcf33245468731d414eab97dac Merged-by: Michael Niedermayer <michaelni(a)gmx.at> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=b66b1a103fff88682… ---

1 0

Merge commit 'f261a55d72753bcec33369954809732abae8f8f7'
by git＠videolan.org 31 May '15

31 May '15

ffmpeg | branch: master | Michael Niedermayer <michaelni(a)gmx.at> | Mon Jun 1 00:25:30 2015 +0200| [c2876c0181d170ed9d4d962599f5b65ba4e85178] | committer: Michael Niedermayer Merge commit 'f261a55d72753bcec33369954809732abae8f8f7' * commit 'f261a55d72753bcec33369954809732abae8f8f7': mov: Fix two memleaks See: 8e77f8ab4abc199a5cc9d6c2c206b73fc2921af0 Merged-by: Michael Niedermayer <michaelni(a)gmx.at> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=c2876c0181d170ed9… ---

1 0

Merge commit '3f38d4b816b2aeca15c0b90000ea7ed40377eb0c'
by git＠videolan.org 31 May '15

31 May '15

ffmpeg | branch: master | Michael Niedermayer <michaelni(a)gmx.at> | Mon Jun 1 00:24:45 2015 +0200| [61e1287551cd912b9744b7ffa813686afa1f97e2] | committer: Michael Niedermayer Merge commit '3f38d4b816b2aeca15c0b90000ea7ed40377eb0c' * commit '3f38d4b816b2aeca15c0b90000ea7ed40377eb0c': vp9: Parse subsampling and report missing feature Conflicts: libavcodec/vp9.c libavcodec/vp9.h No change as FFmpeg already supports subsampling Merged-by: Michael Niedermayer <michaelni(a)gmx.at> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=61e1287551cd912b9… ---

1 0

mov: Fix two memleaks
by git＠videolan.org 31 May '15

31 May '15

ffmpeg | branch: master | Andreas Cadhalpun <Andreas.Cadhalpun(a)googlemail.com> | Tue May 26 14:24:36 2015 +0100| [f261a55d72753bcec33369954809732abae8f8f7] | committer: Luca Barbato mov: Fix two memleaks Signed-off-by: Andreas Cadhalpun <Andreas.Cadhalpun(a)googlemail.com> Signed-off-by: Luca Barbato <lu_zero(a)gentoo.org> CC: libav-stable(a)libav.org > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=f261a55d72753bcec… --- libavformat/mov.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/libavformat/mov.c b/libavformat/mov.c index 70fa1e0..33e1054 100644 --- a/libavformat/mov.c +++ b/libavformat/mov.c @@ -1942,10 +1942,11 @@ static int mov_read_stsz(MOVContext *c, AVIOContext *pb, MOVAtom atom) sc->sample_count = i; + av_free(buf); + if (pb->eof_reached) return AVERROR_EOF; - av_free(buf); return 0; } @@ -2028,6 +2029,8 @@ static int mov_read_ctts(MOVContext *c, AVIOContext *pb, MOVAtom atom) av_log(c->fc, AV_LOG_TRACE, "track[%i].ctts.entries = %i\n", c->fc->nb_streams-1, entries); + av_freep(&sc->ctts_data); + if (!entries) return 0; if (entries >= UINT_MAX / sizeof(*sc->ctts_data))

1 0

vp9: Parse subsampling and report missing feature
by git＠videolan.org 31 May '15

31 May '15

ffmpeg | branch: master | Vittorio Giovara <vittorio.giovara(a)gmail.com> | Wed May 27 15:06:57 2015 +0100| [3f38d4b816b2aeca15c0b90000ea7ed40377eb0c] | committer: Vittorio Giovara vp9: Parse subsampling and report missing feature > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=3f38d4b816b2aeca1… --- libavcodec/vp9.c | 24 +++++++++++++++++++++++- libavcodec/vp9.h | 2 ++ 2 files changed, 25 insertions(+), 1 deletion(-) diff --git a/libavcodec/vp9.c b/libavcodec/vp9.c index 50b84ae..b9397f5 100644 --- a/libavcodec/vp9.c +++ b/libavcodec/vp9.c @@ -219,7 +219,29 @@ static int decode_frame_header(AVCodecContext *avctx, return AVERROR_INVALIDDATA; } s->fullrange = get_bits1(&s->gb); - // for profile 1, here follows the subsampling bits + + // subsampling bits + if (s->profile == 1 || s->profile == 3) { + s->sub_x = get_bits1(&s->gb); + s->sub_y = get_bits1(&s->gb); + if (s->sub_x && s->sub_y) { + av_log(avctx, AV_LOG_ERROR, + "4:2:0 color not supported in profile 1 or 3\n"); + return AVERROR_INVALIDDATA; + } + if (get_bits1(&s->gb)) { // reserved bit + av_log(avctx, AV_LOG_ERROR, "Reserved bit should be zero\n"); + return AVERROR_INVALIDDATA; + } + } else { + s->sub_x = s->sub_y = 1; + } + if (!s->sub_x || !s->sub_y) { + avpriv_report_missing_feature(avctx, "Subsampling %d:%d", + s->sub_x, s->sub_y); + return AVERROR_PATCHWELCOME; + } + s->refreshrefmask = 0xff; w = get_bits(&s->gb, 16) + 1; h = get_bits(&s->gb, 16) + 1; diff --git a/libavcodec/vp9.h b/libavcodec/vp9.h index 724288d..c6f395e 100644 --- a/libavcodec/vp9.h +++ b/libavcodec/vp9.h @@ -277,6 +277,8 @@ typedef struct VP9Context { uint8_t use_last_frame_mvs; uint8_t errorres; uint8_t colorspace; + uint8_t sub_x; + uint8_t sub_y; uint8_t fullrange; uint8_t intraonly; uint8_t resetctx;

1 0

x86: check for AV_CPU_FLAG_AVXSLOW where useful
by git＠videolan.org 31 May '15

31 May '15

ffmpeg | branch: master | James Almer <jamrial(a)gmail.com> | Tue May 26 14:29:08 2015 -0300| [d68c05380cebf563915412182643a8be04ef890b] | committer: Luca Barbato x86: check for AV_CPU_FLAG_AVXSLOW where useful Signed-off-by: James Almer <jamrial(a)gmail.com> Signed-off-by: Luca Barbato <lu_zero(a)gentoo.org> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=d68c05380cebf5639… --- libavcodec/x86/dcadsp_init.c | 4 ++-- libavcodec/x86/dct_init.c | 2 +- libavcodec/x86/fft_init.c | 2 +- libavfilter/x86/af_volume_init.c | 2 +- libavresample/x86/audio_convert_init.c | 4 +++- libavresample/x86/audio_mix_init.c | 4 +++- libavresample/x86/dither_init.c | 4 ++-- libavutil/x86/float_dsp_init.c | 2 +- libavutil/x86/lls_init.c | 2 +- 9 files changed, 15 insertions(+), 11 deletions(-) diff --git a/libavcodec/x86/dcadsp_init.c b/libavcodec/x86/dcadsp_init.c index 9acb818..7c2bec1 100644 --- a/libavcodec/x86/dcadsp_init.c +++ b/libavcodec/x86/dcadsp_init.c @@ -98,10 +98,10 @@ av_cold void ff_synth_filter_init_x86(SynthFilterContext *s) if (EXTERNAL_SSE2(cpu_flags)) { s->synth_filter_float = synth_filter_sse2; } - if (EXTERNAL_AVX(cpu_flags)) { + if (EXTERNAL_AVX_FAST(cpu_flags)) { s->synth_filter_float = synth_filter_avx; } - if (EXTERNAL_FMA3(cpu_flags)) { + if (EXTERNAL_FMA3(cpu_flags) && !(cpu_flags & AV_CPU_FLAG_AVXSLOW)) { s->synth_filter_float = synth_filter_fma3; } #endif /* HAVE_YASM */ diff --git a/libavcodec/x86/dct_init.c b/libavcodec/x86/dct_init.c index 7bda5e8..ca9fbc7 100644 --- a/libavcodec/x86/dct_init.c +++ b/libavcodec/x86/dct_init.c @@ -34,6 +34,6 @@ av_cold void ff_dct_init_x86(DCTContext *s) s->dct32 = ff_dct32_float_sse; if (EXTERNAL_SSE2(cpu_flags)) s->dct32 = ff_dct32_float_sse2; - if (EXTERNAL_AVX(cpu_flags)) + if (EXTERNAL_AVX_FAST(cpu_flags)) s->dct32 = ff_dct32_float_avx; } diff --git a/libavcodec/x86/fft_init.c b/libavcodec/x86/fft_init.c index 7ca72c5..5c0273d 100644 --- a/libavcodec/x86/fft_init.c +++ b/libavcodec/x86/fft_init.c @@ -48,7 +48,7 @@ av_cold void ff_fft_init_x86(FFTContext *s) s->fft_calc = ff_fft_calc_sse; s->fft_permutation = FF_FFT_PERM_SWAP_LSBS; } - if (EXTERNAL_AVX(cpu_flags) && s->nbits >= 5) { + if (EXTERNAL_AVX_FAST(cpu_flags) && s->nbits >= 5) { /* AVX for SB */ s->imdct_half = ff_imdct_half_avx; s->fft_calc = ff_fft_calc_avx; diff --git a/libavfilter/x86/af_volume_init.c b/libavfilter/x86/af_volume_init.c index c59e0ed..26605fb 100644 --- a/libavfilter/x86/af_volume_init.c +++ b/libavfilter/x86/af_volume_init.c @@ -52,7 +52,7 @@ av_cold void ff_volume_init_x86(VolumeContext *vol) vol->scale_samples = ff_scale_samples_s32_ssse3_atom; vol->samples_align = 4; } - if (EXTERNAL_AVX(cpu_flags)) { + if (EXTERNAL_AVX_FAST(cpu_flags)) { vol->scale_samples = ff_scale_samples_s32_avx; vol->samples_align = 8; } diff --git a/libavresample/x86/audio_convert_init.c b/libavresample/x86/audio_convert_init.c index d85ca84..ae6c319 100644 --- a/libavresample/x86/audio_convert_init.c +++ b/libavresample/x86/audio_convert_init.c @@ -226,11 +226,13 @@ av_cold void ff_audio_convert_init_x86(AudioConvert *ac) ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP, 6, 16, 4, "SSE4", ff_conv_fltp_to_flt_6ch_sse4); } - if (EXTERNAL_AVX(cpu_flags)) { + if (EXTERNAL_AVX_FAST(cpu_flags)) { ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S32, 0, 32, 16, "AVX", ff_conv_s32_to_flt_avx); ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_FLT, 0, 32, 32, "AVX", ff_conv_flt_to_s32_avx); + } + if (EXTERNAL_AVX(cpu_flags)) { ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16P, 2, 16, 16, "AVX", ff_conv_s16p_to_s16_2ch_avx); ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16P, diff --git a/libavresample/x86/audio_mix_init.c b/libavresample/x86/audio_mix_init.c index 7fc530e..e14a540 100644 --- a/libavresample/x86/audio_mix_init.c +++ b/libavresample/x86/audio_mix_init.c @@ -195,11 +195,13 @@ av_cold void ff_audio_mix_init_x86(AudioMix *am) ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT, 1, 2, 16, 8, "SSE4", ff_mix_1_to_2_s16p_flt_sse4); } - if (EXTERNAL_AVX(cpu_flags)) { + if (EXTERNAL_AVX_FAST(cpu_flags)) { ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT, 2, 1, 32, 16, "AVX", ff_mix_2_to_1_fltp_flt_avx); ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT, 1, 2, 32, 8, "AVX", ff_mix_1_to_2_fltp_flt_avx); + } + if (EXTERNAL_AVX(cpu_flags)) { ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT, 1, 2, 16, 8, "AVX", ff_mix_1_to_2_s16p_flt_avx); } diff --git a/libavresample/x86/dither_init.c b/libavresample/x86/dither_init.c index 8349d5e..2a351c5 100644 --- a/libavresample/x86/dither_init.c +++ b/libavresample/x86/dither_init.c @@ -46,14 +46,14 @@ av_cold void ff_dither_init_x86(DitherDSPContext *ddsp, if (EXTERNAL_SSE2(cpu_flags)) { ddsp->dither_int_to_float = ff_dither_int_to_float_rectangular_sse2; } - if (EXTERNAL_AVX(cpu_flags)) { + if (EXTERNAL_AVX_FAST(cpu_flags)) { ddsp->dither_int_to_float = ff_dither_int_to_float_rectangular_avx; } } else { if (EXTERNAL_SSE2(cpu_flags)) { ddsp->dither_int_to_float = ff_dither_int_to_float_triangular_sse2; } - if (EXTERNAL_AVX(cpu_flags)) { + if (EXTERNAL_AVX_FAST(cpu_flags)) { ddsp->dither_int_to_float = ff_dither_int_to_float_triangular_avx; } } diff --git a/libavutil/x86/float_dsp_init.c b/libavutil/x86/float_dsp_init.c index a04d91c..b704330 100644 --- a/libavutil/x86/float_dsp_init.c +++ b/libavutil/x86/float_dsp_init.c @@ -146,7 +146,7 @@ av_cold void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp) if (EXTERNAL_SSE2(cpu_flags)) { fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_sse2; } - if (EXTERNAL_AVX(cpu_flags)) { + if (EXTERNAL_AVX_FAST(cpu_flags)) { fdsp->vector_fmul = ff_vector_fmul_avx; fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_avx; fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_avx; diff --git a/libavutil/x86/lls_init.c b/libavutil/x86/lls_init.c index 888bc54..80cda29 100644 --- a/libavutil/x86/lls_init.c +++ b/libavutil/x86/lls_init.c @@ -35,7 +35,7 @@ av_cold void ff_init_lls_x86(LLSModel *m) if (m->indep_count >= 4) m->evaluate_lls = ff_evaluate_lls_sse2; } - if (EXTERNAL_AVX(cpu_flags)) { + if (EXTERNAL_AVX_FAST(cpu_flags)) { m->update_lls = ff_update_lls_avx; } }

1 0

x86: check for AV_CPU_FLAG_AVXSLOW where useful
by git＠videolan.org 31 May '15

31 May '15

ffmpeg | branch: master | James Almer <jamrial(a)gmail.com> | Sun May 31 14:20:29 2015 -0300| [c16e99e3b3c02edcf33245468731d414eab97dac] | committer: Michael Niedermayer x86: check for AV_CPU_FLAG_AVXSLOW where useful Signed-off-by: James Almer <jamrial(a)gmail.com> Signed-off-by: Michael Niedermayer <michaelni(a)gmx.at> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=c16e99e3b3c02edcf… --- libavcodec/x86/dcadsp_init.c | 4 ++-- libavcodec/x86/dct_init.c | 2 +- libavcodec/x86/fft_init.c | 2 +- libavcodec/x86/vp9dsp_init.c | 6 ++++-- libavfilter/x86/af_volume_init.c | 2 +- libavutil/x86/float_dsp_init.c | 4 ++-- libavutil/x86/lls_init.c | 2 +- libswresample/x86/audio_convert_init.c | 4 +++- libswresample/x86/rematrix_init.c | 2 +- libswresample/x86/resample_init.c | 4 ++-- 10 files changed, 18 insertions(+), 14 deletions(-) diff --git a/libavcodec/x86/dcadsp_init.c b/libavcodec/x86/dcadsp_init.c index bb86c26..1a19f6b 100644 --- a/libavcodec/x86/dcadsp_init.c +++ b/libavcodec/x86/dcadsp_init.c @@ -103,10 +103,10 @@ av_cold void ff_synth_filter_init_x86(SynthFilterContext *s) if (EXTERNAL_SSE2(cpu_flags)) { s->synth_filter_float = synth_filter_sse2; } - if (EXTERNAL_AVX(cpu_flags)) { + if (EXTERNAL_AVX_FAST(cpu_flags)) { s->synth_filter_float = synth_filter_avx; } - if (EXTERNAL_FMA3(cpu_flags)) { + if (EXTERNAL_FMA3(cpu_flags) && !(cpu_flags & AV_CPU_FLAG_AVXSLOW)) { s->synth_filter_float = synth_filter_fma3; } #endif /* HAVE_YASM */ diff --git a/libavcodec/x86/dct_init.c b/libavcodec/x86/dct_init.c index 30c8f12..daf2bb4 100644 --- a/libavcodec/x86/dct_init.c +++ b/libavcodec/x86/dct_init.c @@ -34,6 +34,6 @@ av_cold void ff_dct_init_x86(DCTContext *s) s->dct32 = ff_dct32_float_sse; if (EXTERNAL_SSE2(cpu_flags)) s->dct32 = ff_dct32_float_sse2; - if (EXTERNAL_AVX(cpu_flags)) + if (EXTERNAL_AVX_FAST(cpu_flags)) s->dct32 = ff_dct32_float_avx; } diff --git a/libavcodec/x86/fft_init.c b/libavcodec/x86/fft_init.c index 5682230..5085f11 100644 --- a/libavcodec/x86/fft_init.c +++ b/libavcodec/x86/fft_init.c @@ -48,7 +48,7 @@ av_cold void ff_fft_init_x86(FFTContext *s) s->fft_calc = ff_fft_calc_sse; s->fft_permutation = FF_FFT_PERM_SWAP_LSBS; } - if (EXTERNAL_AVX(cpu_flags) && s->nbits >= 5) { + if (EXTERNAL_AVX_FAST(cpu_flags) && s->nbits >= 5) { /* AVX for SB */ s->imdct_half = ff_imdct_half_avx; s->fft_calc = ff_fft_calc_avx; diff --git a/libavcodec/x86/vp9dsp_init.c b/libavcodec/x86/vp9dsp_init.c index 979bd93..00e7125 100644 --- a/libavcodec/x86/vp9dsp_init.c +++ b/libavcodec/x86/vp9dsp_init.c @@ -483,12 +483,14 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp) dsp->itxfm_add[TX_32X32][ADST_DCT] = dsp->itxfm_add[TX_32X32][DCT_ADST] = dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_avx; - init_fpel(1, 0, 32, put, avx); - init_fpel(0, 0, 64, put, avx); init_lpf(avx); init_dir_tm_h_ipred(8, avx); init_dir_tm_h_ipred(16, avx); init_dir_tm_h_ipred(32, avx); + } + if (EXTERNAL_AVX_FAST(cpu_flags)) { + init_fpel(1, 0, 32, put, avx); + init_fpel(0, 0, 64, put, avx); init_ipred(32, avx, v, VERT); } diff --git a/libavfilter/x86/af_volume_init.c b/libavfilter/x86/af_volume_init.c index 57c7eab..88f5a96 100644 --- a/libavfilter/x86/af_volume_init.c +++ b/libavfilter/x86/af_volume_init.c @@ -52,7 +52,7 @@ av_cold void ff_volume_init_x86(VolumeContext *vol) vol->scale_samples = ff_scale_samples_s32_ssse3_atom; vol->samples_align = 4; } - if (EXTERNAL_AVX(cpu_flags)) { + if (EXTERNAL_AVX_FAST(cpu_flags)) { vol->scale_samples = ff_scale_samples_s32_avx; vol->samples_align = 8; } diff --git a/libavutil/x86/float_dsp_init.c b/libavutil/x86/float_dsp_init.c index 64b3a4d..f211f23 100644 --- a/libavutil/x86/float_dsp_init.c +++ b/libavutil/x86/float_dsp_init.c @@ -85,14 +85,14 @@ av_cold void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp) if (EXTERNAL_SSE2(cpu_flags)) { fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_sse2; } - if (EXTERNAL_AVX(cpu_flags)) { + if (EXTERNAL_AVX_FAST(cpu_flags)) { fdsp->vector_fmul = ff_vector_fmul_avx; fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_avx; fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_avx; fdsp->vector_fmul_add = ff_vector_fmul_add_avx; fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_avx; } - if (EXTERNAL_FMA3(cpu_flags)) { + if (EXTERNAL_FMA3(cpu_flags) && !(cpu_flags & AV_CPU_FLAG_AVXSLOW)) { fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_fma3; fdsp->vector_fmul_add = ff_vector_fmul_add_fma3; } diff --git a/libavutil/x86/lls_init.c b/libavutil/x86/lls_init.c index f531904..81f141c 100644 --- a/libavutil/x86/lls_init.c +++ b/libavutil/x86/lls_init.c @@ -35,7 +35,7 @@ av_cold void ff_init_lls_x86(LLSModel *m) if (m->indep_count >= 4) m->evaluate_lls = ff_evaluate_lls_sse2; } - if (EXTERNAL_AVX(cpu_flags)) { + if (EXTERNAL_AVX_FAST(cpu_flags)) { m->update_lls = ff_update_lls_avx; } } diff --git a/libswresample/x86/audio_convert_init.c b/libswresample/x86/audio_convert_init.c index 7f25d98..5e5e91d 100644 --- a/libswresample/x86/audio_convert_init.c +++ b/libswresample/x86/audio_convert_init.c @@ -145,9 +145,11 @@ MULTI_CAPS_FUNC(SSE2, sse2) ac->simd_f = ff_unpack_2ch_int16_to_float_a_ssse3; } } - if(EXTERNAL_AVX(mm_flags)) { + if(EXTERNAL_AVX_FAST(mm_flags)) { if( out_fmt == AV_SAMPLE_FMT_FLT && in_fmt == AV_SAMPLE_FMT_S32 || out_fmt == AV_SAMPLE_FMT_FLTP && in_fmt == AV_SAMPLE_FMT_S32P) ac->simd_f = ff_int32_to_float_a_avx; + } + if(EXTERNAL_AVX(mm_flags)) { if(channels == 6) { if( out_fmt == AV_SAMPLE_FMT_FLT && in_fmt == AV_SAMPLE_FMT_FLTP || out_fmt == AV_SAMPLE_FMT_S32 && in_fmt == AV_SAMPLE_FMT_S32P) ac->simd_f = ff_pack_6ch_float_to_float_a_avx; diff --git a/libswresample/x86/rematrix_init.c b/libswresample/x86/rematrix_init.c index 918479a..5f2c5fe 100644 --- a/libswresample/x86/rematrix_init.c +++ b/libswresample/x86/rematrix_init.c @@ -73,7 +73,7 @@ av_cold int swri_rematrix_init_x86(struct SwrContext *s){ s->mix_1_1_simd = ff_mix_1_1_a_float_sse; s->mix_2_1_simd = ff_mix_2_1_a_float_sse; } - if(EXTERNAL_AVX(mm_flags)) { + if(EXTERNAL_AVX_FAST(mm_flags)) { s->mix_1_1_simd = ff_mix_1_1_a_float_avx; s->mix_2_1_simd = ff_mix_2_1_a_float_avx; } diff --git a/libswresample/x86/resample_init.c b/libswresample/x86/resample_init.c index 93001d6..bc444cf 100644 --- a/libswresample/x86/resample_init.c +++ b/libswresample/x86/resample_init.c @@ -67,11 +67,11 @@ av_cold void swri_resample_dsp_x86_init(ResampleContext *c) c->dsp.resample = c->linear ? ff_resample_linear_float_sse : ff_resample_common_float_sse; } - if (EXTERNAL_AVX(mm_flags)) { + if (EXTERNAL_AVX_FAST(mm_flags)) { c->dsp.resample = c->linear ? ff_resample_linear_float_avx : ff_resample_common_float_avx; } - if (EXTERNAL_FMA3(mm_flags)) { + if (EXTERNAL_FMA3(mm_flags) && !(mm_flags & AV_CPU_FLAG_AVXSLOW)) { c->dsp.resample = c->linear ? ff_resample_linear_float_fma3 : ff_resample_common_float_fma3; }

1 0