April 2017 - ffmpeg-cvslog - lists.ffmpeg.org

tools/target_dec_fuzzer: Fix return code on open failure
by Michael Niedermayer 27 Apr '17

27 Apr '17

ffmpeg | branch: master | Michael Niedermayer <michael(a)niedermayer.cc> | Thu Apr 27 02:18:03 2017 +0200| [966cbfbc83db34f50616fe582d1a53f81d292282] | committer: Michael Niedermayer tools/target_dec_fuzzer: Fix return code on open failure Fixes: 1271/clusterfuzz-testcase-minimized-6095220498235392 Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/targets/ffmpeg Signed-off-by: Michael Niedermayer <michael(a)niedermayer.cc> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=966cbfbc83db34f50… --- tools/target_dec_fuzzer.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tools/target_dec_fuzzer.c b/tools/target_dec_fuzzer.c index dc72fd8b5d..e6a942cb30 100644 --- a/tools/target_dec_fuzzer.c +++ b/tools/target_dec_fuzzer.c @@ -170,7 +170,7 @@ int LLVMFuzzerTestOneInput(const uint8_t *data, size_t size) { int res = avcodec_open2(ctx, c, NULL); if (res < 0) - return res; + return 0; // Failure of avcodec_open2() does not imply that a issue was found FDBCreate(&buffer); int got_frame;

1 0

avcodec/x86/vc1dsp_init: Fix build failure with --disable-optimizations and clang
by Michael Niedermayer 27 Apr '17

27 Apr '17

ffmpeg | branch: master | Michael Niedermayer <michael(a)niedermayer.cc> | Thu Apr 27 02:08:54 2017 +0200| [fa8fd0808f1086fc85abba5cf123faf41da49305] | committer: Michael Niedermayer avcodec/x86/vc1dsp_init: Fix build failure with --disable-optimizations and clang compilers doing DCE at -O0 do not necessarily understand "complex" boolean expressions Build succeeds with this change, this was the only failure Signed-off-by: Michael Niedermayer <michael(a)niedermayer.cc> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=fa8fd0808f1086fc8… --- libavcodec/x86/vc1dsp_init.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/libavcodec/x86/vc1dsp_init.c b/libavcodec/x86/vc1dsp_init.c index 79d22a294f..0b00f8ac11 100644 --- a/libavcodec/x86/vc1dsp_init.c +++ b/libavcodec/x86/vc1dsp_init.c @@ -106,10 +106,12 @@ av_cold void ff_vc1dsp_init_x86(VC1DSPContext *dsp) { int cpu_flags = av_get_cpu_flags(); - if (HAVE_6REGS && INLINE_MMX(cpu_flags) && EXTERNAL_MMX(cpu_flags)) + if (HAVE_6REGS && INLINE_MMX(cpu_flags)) + if (EXTERNAL_MMX(cpu_flags)) ff_vc1dsp_init_mmx(dsp); - if (HAVE_6REGS && INLINE_MMXEXT(cpu_flags) && EXTERNAL_MMXEXT(cpu_flags)) + if (HAVE_6REGS && INLINE_MMXEXT(cpu_flags)) + if (EXTERNAL_MMXEXT(cpu_flags)) ff_vc1dsp_init_mmxext(dsp); #define ASSIGN_LF(EXT) \

1 0

libavcodec/exr : fix float to uint16 conversion for negative float value
by Martin Vignali 27 Apr '17

27 Apr '17

ffmpeg | branch: master | Martin Vignali <martin.vignali(a)gmail.com> | Tue Apr 25 22:52:50 2017 +0200| [e46d63745215c04637e7797228bad36bce49d881] | committer: Michael Niedermayer libavcodec/exr : fix float to uint16 conversion for negative float value Signed-off-by: Michael Niedermayer <michael(a)niedermayer.cc> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=e46d63745215c0463… --- libavcodec/exr.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/libavcodec/exr.c b/libavcodec/exr.c index 71946400c5..c32eea1b8f 100644 --- a/libavcodec/exr.c +++ b/libavcodec/exr.c @@ -220,9 +220,9 @@ static union av_intfloat32 exr_half2float(uint16_t hf) * * @return normalized 16-bit unsigned int */ -static inline uint16_t exr_flt2uint(uint32_t v) +static inline uint16_t exr_flt2uint(int32_t v) { - unsigned int exp = v >> 23; + int32_t exp = v >> 23; // "HACK": negative values result in exp< 0, so clipping them to 0 // is also handled by this condition, avoids explicit check for sign bit. if (exp <= 127 + 7 - 24) // we would shift out all bits anyway

1 0

opus_pvq: merge band encoding and decoding into one function
by Rostislav Pehlivanov 27 Apr '17

27 Apr '17

ffmpeg | branch: master | Rostislav Pehlivanov <atomnuker(a)gmail.com> | Thu Apr 27 02:15:36 2017 +0100| [f16180f44859587e0211170bd82b1bfd26f24b5a] | committer: Rostislav Pehlivanov opus_pvq: merge band encoding and decoding into one function Most code between the 2 functions was duplicated which made keeping both in sync difficult. This also fixes some discovered issues with encoding (incorrect TF switching buffers) and reduces stack usage (reuse the already allocated CeltFrame->scratch buffer for the quantized coefficients). Signed-off-by: Rostislav Pehlivanov <atomnuker(a)gmail.com> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=f16180f44859587e0… --- libavcodec/opus_pvq.c | 523 +++++++++++--------------------------------------- 1 file changed, 112 insertions(+), 411 deletions(-) diff --git a/libavcodec/opus_pvq.c b/libavcodec/opus_pvq.c index d1091677ad..3109e70f17 100644 --- a/libavcodec/opus_pvq.c +++ b/libavcodec/opus_pvq.c @@ -412,9 +412,10 @@ static int celt_pvq_search(float *X, int *y, int K, int N) } static uint32_t celt_alg_quant(OpusRangeCoder *rc, float *X, uint32_t N, uint32_t K, - enum CeltSpread spread, uint32_t blocks, float gain) + enum CeltSpread spread, uint32_t blocks, float gain, + void *scratch) { - int y[176]; + int *y = scratch; celt_exp_rotation(X, N, blocks, K, spread, 1); gain /= sqrtf(celt_pvq_search(X, y, K, N)); @@ -427,9 +428,10 @@ static uint32_t celt_alg_quant(OpusRangeCoder *rc, float *X, uint32_t N, uint32_ /** Decode pulse vector and combine the result with the pitch vector to produce the final normalised signal in the current band. */ static uint32_t celt_alg_unquant(OpusRangeCoder *rc, float *X, uint32_t N, uint32_t K, - enum CeltSpread spread, uint32_t blocks, float gain) + enum CeltSpread spread, uint32_t blocks, float gain, + void *scratch) { - int y[176]; + int *y = scratch; gain /= sqrtf(celt_decode_pulses(rc, y, N, K)); celt_normalize_residual(y, X, N, gain); @@ -437,354 +439,6 @@ static uint32_t celt_alg_unquant(OpusRangeCoder *rc, float *X, uint32_t N, uint3 return celt_extract_collapse_mask(y, N, blocks); } -uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, - float *X, float *Y, int N, int b, uint32_t blocks, - float *lowband, int duration, float *lowband_out, int level, - float gain, float *lowband_scratch, int fill) -{ - int i; - const uint8_t *cache; - int stereo = !!Y, split = !!Y; - int imid = 0, iside = 0; - uint32_t N0 = N; - int N_B = N / blocks; - int N_B0 = N_B; - int B0 = blocks; - int time_divide = 0; - int recombine = 0; - int inv = 0; - float mid = 0, side = 0; - int longblocks = (B0 == 1); - uint32_t cm = 0; - - if (N == 1) { - /* special case for one sample */ - float *x = X; - for (i = 0; i <= stereo; i++) { - int sign = 0; - if (f->remaining2 >= 1<<3) { - sign = ff_opus_rc_get_raw(rc, 1); - f->remaining2 -= 1 << 3; - b -= 1 << 3; - } - x[0] = sign ? -1.0f : 1.0f; - x = Y; - } - if (lowband_out) - lowband_out[0] = X[0]; - return 1; - } - - if (!stereo && level == 0) { - int tf_change = f->tf_change[band]; - int k; - if (tf_change > 0) - recombine = tf_change; - /* Band recombining to increase frequency resolution */ - - if (lowband && - (recombine || ((N_B & 1) == 0 && tf_change < 0) || B0 > 1)) { - for (i = 0; i < N; i++) - lowband_scratch[i] = lowband[i]; - lowband = lowband_scratch; - } - - for (k = 0; k < recombine; k++) { - if (lowband) - celt_haar1(lowband, N >> k, 1 << k); - fill = ff_celt_bit_interleave[fill & 0xF] | ff_celt_bit_interleave[fill >> 4] << 2; - } - blocks >>= recombine; - N_B <<= recombine; - - /* Increasing the time resolution */ - while ((N_B & 1) == 0 && tf_change < 0) { - if (lowband) - celt_haar1(lowband, N_B, blocks); - fill |= fill << blocks; - blocks <<= 1; - N_B >>= 1; - time_divide++; - tf_change++; - } - B0 = blocks; - N_B0 = N_B; - - /* Reorganize the samples in time order instead of frequency order */ - if (B0 > 1 && lowband) - celt_deinterleave_hadamard(f->scratch, lowband, N_B >> recombine, - B0 << recombine, longblocks); - } - - /* If we need 1.5 more bit than we can produce, split the band in two. */ - cache = ff_celt_cache_bits + - ff_celt_cache_index[(duration + 1) * CELT_MAX_BANDS + band]; - if (!stereo && duration >= 0 && b > cache[cache[0]] + 12 && N > 2) { - N >>= 1; - Y = X + N; - split = 1; - duration -= 1; - if (blocks == 1) - fill = (fill & 1) | (fill << 1); - blocks = (blocks + 1) >> 1; - } - - if (split) { - int qn; - int itheta = 0; - int mbits, sbits, delta; - int qalloc; - int pulse_cap; - int offset; - int orig_fill; - int tell; - - /* Decide on the resolution to give to the split parameter theta */ - pulse_cap = ff_celt_log_freq_range[band] + duration * 8; - offset = (pulse_cap >> 1) - (stereo && N == 2 ? CELT_QTHETA_OFFSET_TWOPHASE : - CELT_QTHETA_OFFSET); - qn = (stereo && band >= f->intensity_stereo) ? 1 : - celt_compute_qn(N, b, offset, pulse_cap, stereo); - tell = opus_rc_tell_frac(rc); - if (qn != 1) { - /* Entropy coding of the angle. We use a uniform pdf for the - time split, a step for stereo, and a triangular one for the rest. */ - if (stereo && N > 2) - itheta = ff_opus_rc_dec_uint_step(rc, qn/2); - else if (stereo || B0 > 1) - itheta = ff_opus_rc_dec_uint(rc, qn+1); - else - itheta = ff_opus_rc_dec_uint_tri(rc, qn); - itheta = itheta * 16384 / qn; - /* NOTE: Renormalising X and Y *may* help fixed-point a bit at very high rate. - Let's do that at higher complexity */ - } else if (stereo) { - inv = (b > 2 << 3 && f->remaining2 > 2 << 3) ? ff_opus_rc_dec_log(rc, 2) : 0; - itheta = 0; - } - qalloc = opus_rc_tell_frac(rc) - tell; - b -= qalloc; - - orig_fill = fill; - if (itheta == 0) { - imid = 32767; - iside = 0; - fill = av_mod_uintp2(fill, blocks); - delta = -16384; - } else if (itheta == 16384) { - imid = 0; - iside = 32767; - fill &= ((1 << blocks) - 1) << blocks; - delta = 16384; - } else { - imid = celt_cos(itheta); - iside = celt_cos(16384-itheta); - /* This is the mid vs side allocation that minimizes squared error - in that band. */ - delta = ROUND_MUL16((N - 1) << 7, celt_log2tan(iside, imid)); - } - - mid = imid / 32768.0f; - side = iside / 32768.0f; - - /* This is a special case for N=2 that only works for stereo and takes - advantage of the fact that mid and side are orthogonal to encode - the side with just one bit. */ - if (N == 2 && stereo) { - int c; - int sign = 0; - float tmp; - float *x2, *y2; - mbits = b; - /* Only need one bit for the side */ - sbits = (itheta != 0 && itheta != 16384) ? 1 << 3 : 0; - mbits -= sbits; - c = (itheta > 8192); - f->remaining2 -= qalloc+sbits; - - x2 = c ? Y : X; - y2 = c ? X : Y; - if (sbits) - sign = ff_opus_rc_get_raw(rc, 1); - sign = 1 - 2 * sign; - /* We use orig_fill here because we want to fold the side, but if - itheta==16384, we'll have cleared the low bits of fill. */ - cm = ff_celt_decode_band(f, rc, band, x2, NULL, N, mbits, blocks, - lowband, duration, lowband_out, level, gain, - lowband_scratch, orig_fill); - /* We don't split N=2 bands, so cm is either 1 or 0 (for a fold-collapse), - and there's no need to worry about mixing with the other channel. */ - y2[0] = -sign * x2[1]; - y2[1] = sign * x2[0]; - X[0] *= mid; - X[1] *= mid; - Y[0] *= side; - Y[1] *= side; - tmp = X[0]; - X[0] = tmp - Y[0]; - Y[0] = tmp + Y[0]; - tmp = X[1]; - X[1] = tmp - Y[1]; - Y[1] = tmp + Y[1]; - } else { - /* "Normal" split code */ - float *next_lowband2 = NULL; - float *next_lowband_out1 = NULL; - int next_level = 0; - int rebalance; - - /* Give more bits to low-energy MDCTs than they would - * otherwise deserve */ - if (B0 > 1 && !stereo && (itheta & 0x3fff)) { - if (itheta > 8192) - /* Rough approximation for pre-echo masking */ - delta -= delta >> (4 - duration); - else - /* Corresponds to a forward-masking slope of - * 1.5 dB per 10 ms */ - delta = FFMIN(0, delta + (N << 3 >> (5 - duration))); - } - mbits = av_clip((b - delta) / 2, 0, b); - sbits = b - mbits; - f->remaining2 -= qalloc; - - if (lowband && !stereo) - next_lowband2 = lowband + N; /* >32-bit split case */ - - /* Only stereo needs to pass on lowband_out. - * Otherwise, it's handled at the end */ - if (stereo) - next_lowband_out1 = lowband_out; - else - next_level = level + 1; - - rebalance = f->remaining2; - if (mbits >= sbits) { - /* In stereo mode, we do not apply a scaling to the mid - * because we need the normalized mid for folding later */ - cm = ff_celt_decode_band(f, rc, band, X, NULL, N, mbits, blocks, - lowband, duration, next_lowband_out1, - next_level, stereo ? 1.0f : (gain * mid), - lowband_scratch, fill); - - rebalance = mbits - (rebalance - f->remaining2); - if (rebalance > 3 << 3 && itheta != 0) - sbits += rebalance - (3 << 3); - - /* For a stereo split, the high bits of fill are always zero, - * so no folding will be done to the side. */ - cm |= ff_celt_decode_band(f, rc, band, Y, NULL, N, sbits, blocks, - next_lowband2, duration, NULL, - next_level, gain * side, NULL, - fill >> blocks) << ((B0 >> 1) & (stereo - 1)); - } else { - /* For a stereo split, the high bits of fill are always zero, - * so no folding will be done to the side. */ - cm = ff_celt_decode_band(f, rc, band, Y, NULL, N, sbits, blocks, - next_lowband2, duration, NULL, - next_level, gain * side, NULL, - fill >> blocks) << ((B0 >> 1) & (stereo - 1)); - - rebalance = sbits - (rebalance - f->remaining2); - if (rebalance > 3 << 3 && itheta != 16384) - mbits += rebalance - (3 << 3); - - /* In stereo mode, we do not apply a scaling to the mid because - * we need the normalized mid for folding later */ - cm |= ff_celt_decode_band(f, rc, band, X, NULL, N, mbits, blocks, - lowband, duration, next_lowband_out1, - next_level, stereo ? 1.0f : (gain * mid), - lowband_scratch, fill); - } - } - } else { - /* This is the basic no-split case */ - uint32_t q = celt_bits2pulses(cache, b); - uint32_t curr_bits = celt_pulses2bits(cache, q); - f->remaining2 -= curr_bits; - - /* Ensures we can never bust the budget */ - while (f->remaining2 < 0 && q > 0) { - f->remaining2 += curr_bits; - curr_bits = celt_pulses2bits(cache, --q); - f->remaining2 -= curr_bits; - } - - if (q != 0) { - /* Finally do the actual quantization */ - cm = celt_alg_unquant(rc, X, N, (q < 8) ? q : (8 + (q & 7)) << ((q >> 3) - 1), - f->spread, blocks, gain); - } else { - /* If there's no pulse, fill the band anyway */ - uint32_t cm_mask = (1 << blocks) - 1; - fill &= cm_mask; - if (fill) { - if (!lowband) { - /* Noise */ - for (i = 0; i < N; i++) - X[i] = (((int32_t)celt_rng(f)) >> 20); - cm = cm_mask; - } else { - /* Folded spectrum */ - for (i = 0; i < N; i++) { - /* About 48 dB below the "normal" folding level */ - X[i] = lowband[i] + (((celt_rng(f)) & 0x8000) ? 1.0f / 256 : -1.0f / 256); - } - cm = fill; - } - celt_renormalize_vector(X, N, gain); - } else { - memset(X, 0, N*sizeof(float)); - } - } - } - - /* This code is used by the decoder and by the resynthesis-enabled encoder */ - if (stereo) { - if (N > 2) - celt_stereo_merge(X, Y, mid, N); - if (inv) { - for (i = 0; i < N; i++) - Y[i] *= -1; - } - } else if (level == 0) { - int k; - - /* Undo the sample reorganization going from time order to frequency order */ - if (B0 > 1) - celt_interleave_hadamard(f->scratch, X, N_B >> recombine, - B0 << recombine, longblocks); - - /* Undo time-freq changes that we did earlier */ - N_B = N_B0; - blocks = B0; - for (k = 0; k < time_divide; k++) { - blocks >>= 1; - N_B <<= 1; - cm |= cm >> blocks; - celt_haar1(X, N_B, blocks); - } - - for (k = 0; k < recombine; k++) { - cm = ff_celt_bit_deinterleave[cm]; - celt_haar1(X, N0>>k, 1<<k); - } - blocks <<= recombine; - - /* Scale output for later folding */ - if (lowband_out) { - float n = sqrtf(N0); - for (i = 0; i < N0; i++) - lowband_out[i] = n * X[i]; - } - cm = av_mod_uintp2(cm, blocks); - } - - return cm; -} - -/* This has to be, AND MUST BE done by the psychoacoustic system, this has a very - * big impact on the entire quantization and especially huge on transients */ static int celt_calc_theta(const float *X, const float *Y, int coupling, int N) { int i; @@ -823,14 +477,15 @@ static void celt_stereo_ms_decouple(float *X, float *Y, int N) } } -uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, - float *X, float *Y, int N, int b, uint32_t blocks, - float *lowband, int duration, float *lowband_out, int level, - float gain, float *lowband_scratch, int fill) +static av_always_inline uint32_t quant_band_template(CeltFrame *f, OpusRangeCoder *rc, const int band, + float *X, float *Y, int N, int b, uint32_t blocks, + float *lowband, int duration, float *lowband_out, + int level, float gain, float *lowband_scratch, + int fill, int quant) { int i; const uint8_t *cache; - int stereo = !!Y, split = !!Y; + int stereo = !!Y, split = stereo; int imid = 0, iside = 0; uint32_t N0 = N; int N_B = N / blocks; @@ -844,15 +499,19 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, uint32_t cm = 0; if (N == 1) { - /* special case for one sample - the decoder's output will be +- 1.0f!!! */ float *x = X; for (i = 0; i <= stereo; i++) { - if (f->remaining2 >= 1<<3) { - ff_opus_rc_put_raw(rc, x[0] < 0, 1); + int sign = 0; + if (f->remaining2 >= 1 << 3) { + if (quant) { + sign = x[0] < 0; + ff_opus_rc_put_raw(rc, sign, 1); + } else { + sign = ff_opus_rc_get_raw(rc, 1); + } f->remaining2 -= 1 << 3; - b -= 1 << 3; } - x[0] = 1.0f - 2.0f*(x[0] < 0); + x[0] = 1.0f - 2.0f*sign; x = Y; } if (lowband_out) @@ -875,7 +534,8 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, } for (k = 0; k < recombine; k++) { - celt_haar1(X, N >> k, 1 << k); + if (quant || lowband) + celt_haar1(quant ? X : lowband, N >> k, 1 << k); fill = ff_celt_bit_interleave[fill & 0xF] | ff_celt_bit_interleave[fill >> 4] << 2; } blocks >>= recombine; @@ -883,7 +543,8 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, /* Increasing the time resolution */ while ((N_B & 1) == 0 && tf_change < 0) { - celt_haar1(X, N_B, blocks); + if (quant || lowband) + celt_haar1(quant ? X : lowband, N_B, blocks); fill |= fill << blocks; blocks <<= 1; N_B >>= 1; @@ -894,9 +555,10 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, N_B0 = N_B; /* Reorganize the samples in time order instead of frequency order */ - if (B0 > 1) - celt_deinterleave_hadamard(f->scratch, X, N_B >> recombine, - B0 << recombine, longblocks); + if (B0 > 1 && (quant || lowband)) + celt_deinterleave_hadamard(f->scratch, quant ? X : lowband, + N_B >> recombine, B0 << recombine, + longblocks); } /* If we need 1.5 more bit than we can produce, split the band in two. */ @@ -914,7 +576,7 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, if (split) { int qn; - int itheta = celt_calc_theta(X, Y, stereo, N); + int itheta = quant ? celt_calc_theta(X, Y, stereo, N) : 0; int mbits, sbits, delta; int qalloc; int pulse_cap; @@ -929,43 +591,53 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, qn = (stereo && band >= f->intensity_stereo) ? 1 : celt_compute_qn(N, b, offset, pulse_cap, stereo); tell = opus_rc_tell_frac(rc); - if (qn != 1) { - - itheta = (itheta*qn + 8192) >> 14; - + if (quant) + itheta = (itheta*qn + 8192) >> 14; /* Entropy coding of the angle. We use a uniform pdf for the * time split, a step for stereo, and a triangular one for the rest. */ - if (stereo && N > 2) - ff_opus_rc_enc_uint_step(rc, itheta, qn / 2); - else if (stereo || B0 > 1) - ff_opus_rc_enc_uint(rc, itheta, qn + 1); - else - ff_opus_rc_enc_uint_tri(rc, itheta, qn); - itheta = itheta * 16384 / qn; - - if (stereo) { - if (itheta == 0) - celt_stereo_is_decouple(X, Y, f->block[0].lin_energy[band], - f->block[1].lin_energy[band], N); + if (quant) { + if (stereo && N > 2) + ff_opus_rc_enc_uint_step(rc, itheta, qn / 2); + else if (stereo || B0 > 1) + ff_opus_rc_enc_uint(rc, itheta, qn + 1); + else + ff_opus_rc_enc_uint_tri(rc, itheta, qn); + itheta = itheta * 16384 / qn; + if (stereo) { + if (itheta == 0) + celt_stereo_is_decouple(X, Y, f->block[0].lin_energy[band], + f->block[1].lin_energy[band], N); + else + celt_stereo_ms_decouple(X, Y, N); + } + } else { + if (stereo && N > 2) + itheta = ff_opus_rc_dec_uint_step(rc, qn / 2); + else if (stereo || B0 > 1) + itheta = ff_opus_rc_dec_uint(rc, qn+1); else - celt_stereo_ms_decouple(X, Y, N); + itheta = ff_opus_rc_dec_uint_tri(rc, qn); + itheta = itheta * 16384 / qn; } } else if (stereo) { - inv = itheta > 8192; - if (inv) { - for (i = 0; i < N; i++) - Y[i] *= -1; - } - celt_stereo_is_decouple(X, Y, f->block[0].lin_energy[band], - f->block[1].lin_energy[band], N); - - if (b > 2 << 3 && f->remaining2 > 2 << 3) { - ff_opus_rc_enc_log(rc, inv, 2); + if (quant) { + inv = itheta > 8192; + if (inv) { + for (i = 0; i < N; i++) + Y[i] *= -1; + } + celt_stereo_is_decouple(X, Y, f->block[0].lin_energy[band], + f->block[1].lin_energy[band], N); + + if (b > 2 << 3 && f->remaining2 > 2 << 3) { + ff_opus_rc_enc_log(rc, inv, 2); + } else { + inv = 0; + } } else { - inv = 0; + inv = (b > 2 << 3 && f->remaining2 > 2 << 3) ? ff_opus_rc_dec_log(rc, 2) : 0; } - itheta = 0; } qalloc = opus_rc_tell_frac(rc) - tell; @@ -1011,13 +683,17 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, x2 = c ? Y : X; y2 = c ? X : Y; if (sbits) { - sign = x2[0]*y2[1] - x2[1]*y2[0] < 0; - ff_opus_rc_put_raw(rc, sign, 1); + if (quant) { + sign = x2[0]*y2[1] - x2[1]*y2[0] < 0; + ff_opus_rc_put_raw(rc, sign, 1); + } else { + sign = ff_opus_rc_get_raw(rc, 1); + } } sign = 1 - 2 * sign; /* We use orig_fill here because we want to fold the side, but if itheta==16384, we'll have cleared the low bits of fill. */ - cm = ff_celt_encode_band(f, rc, band, x2, NULL, N, mbits, blocks, + cm = ff_celt_decode_band(f, rc, band, x2, NULL, N, mbits, blocks, lowband, duration, lowband_out, level, gain, lowband_scratch, orig_fill); /* We don't split N=2 bands, so cm is either 1 or 0 (for a fold-collapse), @@ -1070,10 +746,10 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, if (mbits >= sbits) { /* In stereo mode, we do not apply a scaling to the mid * because we need the normalized mid for folding later */ - cm = ff_celt_encode_band(f, rc, band, X, NULL, N, mbits, blocks, + cm = quant_band_template(f, rc, band, X, NULL, N, mbits, blocks, lowband, duration, next_lowband_out1, next_level, stereo ? 1.0f : (gain * mid), - lowband_scratch, fill); + lowband_scratch, fill, quant); rebalance = mbits - (rebalance - f->remaining2); if (rebalance > 3 << 3 && itheta != 0) @@ -1081,17 +757,17 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, /* For a stereo split, the high bits of fill are always zero, * so no folding will be done to the side. */ - cm |= ff_celt_encode_band(f, rc, band, Y, NULL, N, sbits, blocks, + cm |= quant_band_template(f, rc, band, Y, NULL, N, sbits, blocks, next_lowband2, duration, NULL, next_level, gain * side, NULL, - fill >> blocks) << ((B0 >> 1) & (stereo - 1)); + fill >> blocks, quant) << ((B0 >> 1) & (stereo - 1)); } else { /* For a stereo split, the high bits of fill are always zero, * so no folding will be done to the side. */ - cm = ff_celt_encode_band(f, rc, band, Y, NULL, N, sbits, blocks, + cm = quant_band_template(f, rc, band, Y, NULL, N, sbits, blocks, next_lowband2, duration, NULL, next_level, gain * side, NULL, - fill >> blocks) << ((B0 >> 1) & (stereo - 1)); + fill >> blocks, quant) << ((B0 >> 1) & (stereo - 1)); rebalance = sbits - (rebalance - f->remaining2); if (rebalance > 3 << 3 && itheta != 16384) @@ -1099,10 +775,10 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, /* In stereo mode, we do not apply a scaling to the mid because * we need the normalized mid for folding later */ - cm |= ff_celt_encode_band(f, rc, band, X, NULL, N, mbits, blocks, + cm |= quant_band_template(f, rc, band, X, NULL, N, mbits, blocks, lowband, duration, next_lowband_out1, next_level, stereo ? 1.0f : (gain * mid), - lowband_scratch, fill); + lowband_scratch, fill, quant); } } } else { @@ -1119,9 +795,14 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, } if (q != 0) { - /* Finally do the actual quantization */ - cm = celt_alg_quant(rc, X, N, (q < 8) ? q : (8 + (q & 7)) << ((q >> 3) - 1), - f->spread, blocks, gain); + /* Finally do the actual (de)quantization */ + if (quant) { + cm = celt_alg_quant(rc, X, N, (q < 8) ? q : (8 + (q & 7)) << ((q >> 3) - 1), + f->spread, blocks, gain, f->scratch); + } else { + cm = celt_alg_unquant(rc, X, N, (q < 8) ? q : (8 + (q & 7)) << ((q >> 3) - 1), + f->spread, blocks, gain, f->scratch); + } } else { /* If there's no pulse, fill the band anyway */ uint32_t cm_mask = (1 << blocks) - 1; @@ -1191,6 +872,26 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, return cm; } +uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, + float *X, float *Y, int N, int b, uint32_t blocks, + float *lowband, int duration, float *lowband_out, + int level, float gain, float *lowband_scratch, + int fill) +{ + return quant_band_template(f, rc, band, X, Y, N, b, blocks, lowband, duration, + lowband_out, level, gain, lowband_scratch, fill, 0); +} + +uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, + float *X, float *Y, int N, int b, uint32_t blocks, + float *lowband, int duration, float *lowband_out, + int level, float gain, float *lowband_scratch, + int fill) +{ + return quant_band_template(f, rc, band, X, Y, N, b, blocks, lowband, duration, + lowband_out, level, gain, lowband_scratch, fill, 1); +} + float ff_celt_quant_band_cost(CeltFrame *f, OpusRangeCoder *rc, int band, float *bits, float lambda) {

1 0

opus_pvq: remove outdated/incorrect comments and redundant variables
by Rostislav Pehlivanov 27 Apr '17

27 Apr '17

ffmpeg | branch: master | Rostislav Pehlivanov <atomnuker(a)gmail.com> | Wed Apr 26 23:28:16 2017 +0100| [18a0d9d7ad0292ce37f5262bda92ebfdd33bf31d] | committer: Rostislav Pehlivanov opus_pvq: remove outdated/incorrect comments and redundant variables Removes the last style issues with opus_pvq.c Signed-off-by: Rostislav Pehlivanov <atomnuker(a)gmail.com> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=18a0d9d7ad0292ce3… --- libavcodec/opus_pvq.c | 25 +++++++------------------ 1 file changed, 7 insertions(+), 18 deletions(-) diff --git a/libavcodec/opus_pvq.c b/libavcodec/opus_pvq.c index e37eccae16..d1091677ad 100644 --- a/libavcodec/opus_pvq.c +++ b/libavcodec/opus_pvq.c @@ -89,18 +89,16 @@ static void celt_exp_rotation_impl(float *X, uint32_t len, uint32_t stride, Xptr = X; for (i = 0; i < len - stride; i++) { - float x1, x2; - x1 = Xptr[0]; - x2 = Xptr[stride]; + float x1 = Xptr[0]; + float x2 = Xptr[stride]; Xptr[stride] = c * x2 + s * x1; *Xptr++ = c * x1 - s * x2; } Xptr = &X[len - 2 * stride - 1]; for (i = len - 2 * stride - 1; i >= 0; i--) { - float x1, x2; - x1 = Xptr[0]; - x2 = Xptr[stride]; + float x1 = Xptr[0]; + float x2 = Xptr[stride]; Xptr[stride] = c * x2 + s * x1; *Xptr-- = c * x1 - s * x2; } @@ -132,8 +130,6 @@ static inline void celt_exp_rotation(float *X, uint32_t len, stride2++; } - /*NOTE: As a minor optimization, we could be passing around log2(B), not B, for both this and for - extract_collapse_mask().*/ len /= stride; for (i = 0; i < stride; i++) { if (encode) { @@ -150,17 +146,12 @@ static inline void celt_exp_rotation(float *X, uint32_t len, static inline uint32_t celt_extract_collapse_mask(const int *iy, uint32_t N, uint32_t B) { - uint32_t collapse_mask; - int N0; - int i, j; + int i, j, N0 = N / B; + uint32_t collapse_mask = 0; if (B <= 1) return 1; - /*NOTE: As a minor optimization, we could be passing around log2(B), not B, for both this and for - exp_rotation().*/ - N0 = N/B; - collapse_mask = 0; for (i = 0; i < B; i++) for (j = 0; j < N0; j++) collapse_mask |= (!!iy[i*N0+j]) << i; @@ -277,13 +268,11 @@ static inline uint32_t celt_icwrsi(uint32_t N, uint32_t K, const int *y) static inline uint64_t celt_cwrsi(uint32_t N, uint32_t K, uint32_t i, int *y) { uint64_t norm = 0; - uint32_t p; + uint32_t q, p; int s, val; int k0; while (N > 2) { - uint32_t q; - /*Lots of pulses case:*/ if (K >= N) { const uint32_t *row = ff_celt_pvq_u_row[N];

1 0

avformat/concatdec: fix the h264 annexb extradata check
by James Almer 27 Apr '17

27 Apr '17

ffmpeg | branch: release/3.3 | James Almer <jamrial(a)gmail.com> | Tue Apr 25 20:23:12 2017 -0300| [cfca0b91399bfc72218bd47412bc8e57c19742b9] | committer: James Almer avformat/concatdec: fix the h264 annexb extradata check The start code can be either in the first three or four bytes. (cherry picked from commit b4330a0e02fcbef61d630a369abe5f4421ced659) > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=cfca0b91399bfc722… --- libavformat/concatdec.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) diff --git a/libavformat/concatdec.c b/libavformat/concatdec.c index 8649916ff2..dd52e4d366 100644 --- a/libavformat/concatdec.c +++ b/libavformat/concatdec.c @@ -199,8 +199,11 @@ static int detect_stream_specific(AVFormatContext *avf, int idx) AVBitStreamFilterContext *bsf; int ret; - if (cat->auto_convert && st->codecpar->codec_id == AV_CODEC_ID_H264 && - (st->codecpar->extradata_size < 4 || AV_RB32(st->codecpar->extradata) != 1)) { + if (cat->auto_convert && st->codecpar->codec_id == AV_CODEC_ID_H264) { + if (!st->codecpar->extradata_size || + (st->codecpar->extradata_size >= 3 && AV_RB24(st->codecpar->extradata) == 1) || + (st->codecpar->extradata_size >= 4 && AV_RB32(st->codecpar->extradata) == 1)) + return 0; av_log(cat->avf, AV_LOG_INFO, "Auto-inserting h264_mp4toannexb bitstream filter\n"); if (!(bsf = av_bitstream_filter_init("h264_mp4toannexb"))) {

1 0

opus_pvq: minor cleanups
by Rostislav Pehlivanov 27 Apr '17

27 Apr '17

ffmpeg | branch: master | Rostislav Pehlivanov <atomnuker(a)gmail.com> | Thu Apr 20 20:00:25 2017 +0100| [4d59de39915099c5646af773ff98909dbea31799] | committer: Rostislav Pehlivanov opus_pvq: minor cleanups Removes unneeded variables, renames confusing and innacurate variables and rewrites and slightly optimizes hadamard interleave/deinterleave functions. Signed-off-by: Rostislav Pehlivanov <atomnuker(a)gmail.com> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=4d59de39915099c56… --- libavcodec/opus_pvq.c | 249 ++++++++++++++++++++++---------------------------- libavcodec/opustab.c | 5 +- libavcodec/opustab.h | 2 +- 3 files changed, 112 insertions(+), 144 deletions(-) diff --git a/libavcodec/opus_pvq.c b/libavcodec/opus_pvq.c index e424cb2e03..e37eccae16 100644 --- a/libavcodec/opus_pvq.c +++ b/libavcodec/opus_pvq.c @@ -33,7 +33,7 @@ static inline int16_t celt_cos(int16_t x) { x = (MUL16(x, x) + 4096) >> 13; x = (32767-x) + ROUND_MUL16(x, (-7651 + ROUND_MUL16(x, (8277 + ROUND_MUL16(-626, x))))); - return 1+x; + return x + 1; } static inline int celt_log2tan(int isin, int icos) @@ -163,7 +163,7 @@ static inline uint32_t celt_extract_collapse_mask(const int *iy, uint32_t N, uin collapse_mask = 0; for (i = 0; i < B; i++) for (j = 0; j < N0; j++) - collapse_mask |= (iy[i*N0+j]!=0)<<i; + collapse_mask |= (!!iy[i*N0+j]) << i; return collapse_mask; } @@ -173,7 +173,7 @@ static inline void celt_stereo_merge(float *X, float *Y, float mid, int N) float xp = 0, side = 0; float E[2]; float mid2; - float t, gain[2]; + float gain[2]; /* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */ for (i = 0; i < N; i++) { @@ -192,10 +192,8 @@ static inline void celt_stereo_merge(float *X, float *Y, float mid, int N) return; } - t = E[0]; - gain[0] = 1.0f / sqrtf(t); - t = E[1]; - gain[1] = 1.0f / sqrtf(t); + gain[0] = 1.0f / sqrtf(E[0]); + gain[1] = 1.0f / sqrtf(E[1]); for (i = 0; i < N; i++) { float value[2]; @@ -210,43 +208,27 @@ static inline void celt_stereo_merge(float *X, float *Y, float mid, int N) static void celt_interleave_hadamard(float *tmp, float *X, int N0, int stride, int hadamard) { - int i, j; - int N = N0*stride; + int i, j, N = N0*stride; + const uint8_t *order = &ff_celt_hadamard_order[hadamard ? stride - 2 : 30]; - if (hadamard) { - const uint8_t *ordery = ff_celt_hadamard_ordery + stride - 2; - for (i = 0; i < stride; i++) - for (j = 0; j < N0; j++) - tmp[j*stride+i] = X[ordery[i]*N0+j]; - } else { - for (i = 0; i < stride; i++) - for (j = 0; j < N0; j++) - tmp[j*stride+i] = X[i*N0+j]; - } + for (i = 0; i < stride; i++) + for (j = 0; j < N0; j++) + tmp[j*stride+i] = X[order[i]*N0+j]; - for (i = 0; i < N; i++) - X[i] = tmp[i]; + memcpy(X, tmp, N*sizeof(float)); } static void celt_deinterleave_hadamard(float *tmp, float *X, int N0, int stride, int hadamard) { - int i, j; - int N = N0*stride; + int i, j, N = N0*stride; + const uint8_t *order = &ff_celt_hadamard_order[hadamard ? stride - 2 : 30]; - if (hadamard) { - const uint8_t *ordery = ff_celt_hadamard_ordery + stride - 2; - for (i = 0; i < stride; i++) - for (j = 0; j < N0; j++) - tmp[ordery[i]*N0+j] = X[j*stride+i]; - } else { - for (i = 0; i < stride; i++) - for (j = 0; j < N0; j++) - tmp[i*N0+j] = X[j*stride+i]; - } + for (i = 0; i < stride; i++) + for (j = 0; j < N0; j++) + tmp[order[i]*N0+j] = X[j*stride+i]; - for (i = 0; i < N; i++) - X[i] = tmp[i]; + memcpy(X, tmp, N*sizeof(float)); } static void celt_haar1(float *X, int N0, int stride) @@ -264,11 +246,11 @@ static void celt_haar1(float *X, int N0, int stride) } static inline int celt_compute_qn(int N, int b, int offset, int pulse_cap, - int dualstereo) + int stereo) { int qn, qb; int N2 = 2 * N - 1; - if (dualstereo && N == 2) + if (stereo && N == 2) N2--; /* The upper limit ensures that in a stereo split with itheta==16384, we'll @@ -471,12 +453,13 @@ uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, float *lowband, int duration, float *lowband_out, int level, float gain, float *lowband_scratch, int fill) { + int i; const uint8_t *cache; - int dualstereo, split; + int stereo = !!Y, split = !!Y; int imid = 0, iside = 0; uint32_t N0 = N; - int N_B; - int N_B0; + int N_B = N / blocks; + int N_B0 = N_B; int B0 = blocks; int time_divide = 0; int recombine = 0; @@ -485,14 +468,10 @@ uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, int longblocks = (B0 == 1); uint32_t cm = 0; - N_B0 = N_B = N / blocks; - split = dualstereo = (Y != NULL); - if (N == 1) { /* special case for one sample */ - int i; float *x = X; - for (i = 0; i <= dualstereo; i++) { + for (i = 0; i <= stereo; i++) { int sign = 0; if (f->remaining2 >= 1<<3) { sign = ff_opus_rc_get_raw(rc, 1); @@ -507,7 +486,7 @@ uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, return 1; } - if (!dualstereo && level == 0) { + if (!stereo && level == 0) { int tf_change = f->tf_change[band]; int k; if (tf_change > 0) @@ -516,9 +495,8 @@ uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, if (lowband && (recombine || ((N_B & 1) == 0 && tf_change < 0) || B0 > 1)) { - int j; - for (j = 0; j < N; j++) - lowband_scratch[j] = lowband[j]; + for (i = 0; i < N; i++) + lowband_scratch[i] = lowband[i]; lowband = lowband_scratch; } @@ -552,7 +530,7 @@ uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, /* If we need 1.5 more bit than we can produce, split the band in two. */ cache = ff_celt_cache_bits + ff_celt_cache_index[(duration + 1) * CELT_MAX_BANDS + band]; - if (!dualstereo && duration >= 0 && b > cache[cache[0]] + 12 && N > 2) { + if (!stereo && duration >= 0 && b > cache[cache[0]] + 12 && N > 2) { N >>= 1; Y = X + N; split = 1; @@ -574,24 +552,24 @@ uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, /* Decide on the resolution to give to the split parameter theta */ pulse_cap = ff_celt_log_freq_range[band] + duration * 8; - offset = (pulse_cap >> 1) - (dualstereo && N == 2 ? CELT_QTHETA_OFFSET_TWOPHASE : + offset = (pulse_cap >> 1) - (stereo && N == 2 ? CELT_QTHETA_OFFSET_TWOPHASE : CELT_QTHETA_OFFSET); - qn = (dualstereo && band >= f->intensity_stereo) ? 1 : - celt_compute_qn(N, b, offset, pulse_cap, dualstereo); + qn = (stereo && band >= f->intensity_stereo) ? 1 : + celt_compute_qn(N, b, offset, pulse_cap, stereo); tell = opus_rc_tell_frac(rc); if (qn != 1) { /* Entropy coding of the angle. We use a uniform pdf for the time split, a step for stereo, and a triangular one for the rest. */ - if (dualstereo && N > 2) + if (stereo && N > 2) itheta = ff_opus_rc_dec_uint_step(rc, qn/2); - else if (dualstereo || B0 > 1) + else if (stereo || B0 > 1) itheta = ff_opus_rc_dec_uint(rc, qn+1); else itheta = ff_opus_rc_dec_uint_tri(rc, qn); itheta = itheta * 16384 / qn; /* NOTE: Renormalising X and Y *may* help fixed-point a bit at very high rate. Let's do that at higher complexity */ - } else if (dualstereo) { + } else if (stereo) { inv = (b > 2 << 3 && f->remaining2 > 2 << 3) ? ff_opus_rc_dec_log(rc, 2) : 0; itheta = 0; } @@ -623,7 +601,7 @@ uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, /* This is a special case for N=2 that only works for stereo and takes advantage of the fact that mid and side are orthogonal to encode the side with just one bit. */ - if (N == 2 && dualstereo) { + if (N == 2 && stereo) { int c; int sign = 0; float tmp; @@ -668,7 +646,7 @@ uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, /* Give more bits to low-energy MDCTs than they would * otherwise deserve */ - if (B0 > 1 && !dualstereo && (itheta & 0x3fff)) { + if (B0 > 1 && !stereo && (itheta & 0x3fff)) { if (itheta > 8192) /* Rough approximation for pre-echo masking */ delta -= delta >> (4 - duration); @@ -681,12 +659,12 @@ uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, sbits = b - mbits; f->remaining2 -= qalloc; - if (lowband && !dualstereo) + if (lowband && !stereo) next_lowband2 = lowband + N; /* >32-bit split case */ /* Only stereo needs to pass on lowband_out. * Otherwise, it's handled at the end */ - if (dualstereo) + if (stereo) next_lowband_out1 = lowband_out; else next_level = level + 1; @@ -697,7 +675,7 @@ uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, * because we need the normalized mid for folding later */ cm = ff_celt_decode_band(f, rc, band, X, NULL, N, mbits, blocks, lowband, duration, next_lowband_out1, - next_level, dualstereo ? 1.0f : (gain * mid), + next_level, stereo ? 1.0f : (gain * mid), lowband_scratch, fill); rebalance = mbits - (rebalance - f->remaining2); @@ -709,14 +687,14 @@ uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, cm |= ff_celt_decode_band(f, rc, band, Y, NULL, N, sbits, blocks, next_lowband2, duration, NULL, next_level, gain * side, NULL, - fill >> blocks) << ((B0 >> 1) & (dualstereo - 1)); + fill >> blocks) << ((B0 >> 1) & (stereo - 1)); } else { /* For a stereo split, the high bits of fill are always zero, * so no folding will be done to the side. */ cm = ff_celt_decode_band(f, rc, band, Y, NULL, N, sbits, blocks, next_lowband2, duration, NULL, next_level, gain * side, NULL, - fill >> blocks) << ((B0 >> 1) & (dualstereo - 1)); + fill >> blocks) << ((B0 >> 1) & (stereo - 1)); rebalance = sbits - (rebalance - f->remaining2); if (rebalance > 3 << 3 && itheta != 16384) @@ -726,7 +704,7 @@ uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, * we need the normalized mid for folding later */ cm |= ff_celt_decode_band(f, rc, band, X, NULL, N, mbits, blocks, lowband, duration, next_lowband_out1, - next_level, dualstereo ? 1.0f : (gain * mid), + next_level, stereo ? 1.0f : (gain * mid), lowband_scratch, fill); } } @@ -749,47 +727,44 @@ uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, f->spread, blocks, gain); } else { /* If there's no pulse, fill the band anyway */ - int j; uint32_t cm_mask = (1 << blocks) - 1; fill &= cm_mask; - if (!fill) { - for (j = 0; j < N; j++) - X[j] = 0.0f; - } else { + if (fill) { if (!lowband) { /* Noise */ - for (j = 0; j < N; j++) - X[j] = (((int32_t)celt_rng(f)) >> 20); + for (i = 0; i < N; i++) + X[i] = (((int32_t)celt_rng(f)) >> 20); cm = cm_mask; } else { /* Folded spectrum */ - for (j = 0; j < N; j++) { + for (i = 0; i < N; i++) { /* About 48 dB below the "normal" folding level */ - X[j] = lowband[j] + (((celt_rng(f)) & 0x8000) ? 1.0f / 256 : -1.0f / 256); + X[i] = lowband[i] + (((celt_rng(f)) & 0x8000) ? 1.0f / 256 : -1.0f / 256); } cm = fill; } celt_renormalize_vector(X, N, gain); + } else { + memset(X, 0, N*sizeof(float)); } } } /* This code is used by the decoder and by the resynthesis-enabled encoder */ - if (dualstereo) { - int j; - if (N != 2) + if (stereo) { + if (N > 2) celt_stereo_merge(X, Y, mid, N); if (inv) { - for (j = 0; j < N; j++) - Y[j] *= -1; + for (i = 0; i < N; i++) + Y[i] *= -1; } } else if (level == 0) { int k; /* Undo the sample reorganization going from time order to frequency order */ if (B0 > 1) - celt_interleave_hadamard(f->scratch, X, N_B>>recombine, - B0<<recombine, longblocks); + celt_interleave_hadamard(f->scratch, X, N_B >> recombine, + B0 << recombine, longblocks); /* Undo time-freq changes that we did earlier */ N_B = N_B0; @@ -809,10 +784,9 @@ uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, /* Scale output for later folding */ if (lowband_out) { - int j; float n = sqrtf(N0); - for (j = 0; j < N0; j++) - lowband_out[j] = n * X[j]; + for (i = 0; i < N0; i++) + lowband_out[i] = n * X[i]; } cm = av_mod_uintp2(cm, blocks); } @@ -824,15 +798,17 @@ uint32_t ff_celt_decode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, * big impact on the entire quantization and especially huge on transients */ static int celt_calc_theta(const float *X, const float *Y, int coupling, int N) { - int j; + int i; float e[2] = { 0.0f, 0.0f }; - for (j = 0; j < N; j++) { - if (coupling) { /* Coupling case */ - e[0] += (X[j] + Y[j])*(X[j] + Y[j]); - e[1] += (X[j] - Y[j])*(X[j] - Y[j]); - } else { - e[0] += X[j]*X[j]; - e[1] += Y[j]*Y[j]; + if (coupling) { /* Coupling case */ + for (i = 0; i < N; i++) { + e[0] += (X[i] + Y[i])*(X[i] + Y[i]); + e[1] += (X[i] - Y[i])*(X[i] - Y[i]); + } + } else { + for (i = 0; i < N; i++) { + e[0] += X[i]*X[i]; + e[1] += Y[i]*Y[i]; } } return lrintf(32768.0f*atan2f(sqrtf(e[1]), sqrtf(e[0]))/M_PI); @@ -851,11 +827,10 @@ static void celt_stereo_is_decouple(float *X, float *Y, float e_l, float e_r, in static void celt_stereo_ms_decouple(float *X, float *Y, int N) { int i; - const float decouple_norm = 1.0f/sqrtf(1.0f + 1.0f); for (i = 0; i < N; i++) { const float Xret = X[i]; - X[i] = (X[i] + Y[i])*decouple_norm; - Y[i] = (Y[i] - Xret)*decouple_norm; + X[i] = (X[i] + Y[i])*M_SQRT1_2; + Y[i] = (Y[i] - Xret)*M_SQRT1_2; } } @@ -864,8 +839,9 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, float *lowband, int duration, float *lowband_out, int level, float gain, float *lowband_scratch, int fill) { + int i; const uint8_t *cache; - int dualstereo, split; + int stereo = !!Y, split = !!Y; int imid = 0, iside = 0; uint32_t N0 = N; int N_B = N / blocks; @@ -878,13 +854,10 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, int longblocks = (B0 == 1); uint32_t cm = 0; - split = dualstereo = (Y != NULL); - if (N == 1) { /* special case for one sample - the decoder's output will be +- 1.0f!!! */ - int i; float *x = X; - for (i = 0; i <= dualstereo; i++) { + for (i = 0; i <= stereo; i++) { if (f->remaining2 >= 1<<3) { ff_opus_rc_put_raw(rc, x[0] < 0, 1); f->remaining2 -= 1 << 3; @@ -898,7 +871,7 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, return 1; } - if (!dualstereo && level == 0) { + if (!stereo && level == 0) { int tf_change = f->tf_change[band]; int k; if (tf_change > 0) @@ -907,9 +880,8 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, if (lowband && (recombine || ((N_B & 1) == 0 && tf_change < 0) || B0 > 1)) { - int j; - for (j = 0; j < N; j++) - lowband_scratch[j] = lowband[j]; + for (i = 0; i < N; i++) + lowband_scratch[i] = lowband[i]; lowband = lowband_scratch; } @@ -941,7 +913,7 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, /* If we need 1.5 more bit than we can produce, split the band in two. */ cache = ff_celt_cache_bits + ff_celt_cache_index[(duration + 1) * CELT_MAX_BANDS + band]; - if (!dualstereo && duration >= 0 && b > cache[cache[0]] + 12 && N > 2) { + if (!stereo && duration >= 0 && b > cache[cache[0]] + 12 && N > 2) { N >>= 1; Y = X + N; split = 1; @@ -953,7 +925,7 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, if (split) { int qn; - int itheta = celt_calc_theta(X, Y, dualstereo, N); + int itheta = celt_calc_theta(X, Y, stereo, N); int mbits, sbits, delta; int qalloc; int pulse_cap; @@ -963,10 +935,10 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, /* Decide on the resolution to give to the split parameter theta */ pulse_cap = ff_celt_log_freq_range[band] + duration * 8; - offset = (pulse_cap >> 1) - (dualstereo && N == 2 ? CELT_QTHETA_OFFSET_TWOPHASE : + offset = (pulse_cap >> 1) - (stereo && N == 2 ? CELT_QTHETA_OFFSET_TWOPHASE : CELT_QTHETA_OFFSET); - qn = (dualstereo && band >= f->intensity_stereo) ? 1 : - celt_compute_qn(N, b, offset, pulse_cap, dualstereo); + qn = (stereo && band >= f->intensity_stereo) ? 1 : + celt_compute_qn(N, b, offset, pulse_cap, stereo); tell = opus_rc_tell_frac(rc); if (qn != 1) { @@ -975,27 +947,26 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, /* Entropy coding of the angle. We use a uniform pdf for the * time split, a step for stereo, and a triangular one for the rest. */ - if (dualstereo && N > 2) + if (stereo && N > 2) ff_opus_rc_enc_uint_step(rc, itheta, qn / 2); - else if (dualstereo || B0 > 1) + else if (stereo || B0 > 1) ff_opus_rc_enc_uint(rc, itheta, qn + 1); else ff_opus_rc_enc_uint_tri(rc, itheta, qn); itheta = itheta * 16384 / qn; - if (dualstereo) { + if (stereo) { if (itheta == 0) celt_stereo_is_decouple(X, Y, f->block[0].lin_energy[band], f->block[1].lin_energy[band], N); else celt_stereo_ms_decouple(X, Y, N); } - } else if (dualstereo) { + } else if (stereo) { inv = itheta > 8192; if (inv) { - int j; - for (j = 0; j < N; j++) - Y[j] = -Y[j]; + for (i = 0; i < N; i++) + Y[i] *= -1; } celt_stereo_is_decouple(X, Y, f->block[0].lin_energy[band], f->block[1].lin_energy[band], N); @@ -1036,7 +1007,7 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, /* This is a special case for N=2 that only works for stereo and takes advantage of the fact that mid and side are orthogonal to encode the side with just one bit. */ - if (N == 2 && dualstereo) { + if (N == 2 && stereo) { int c; int sign = 0; float tmp; @@ -1083,7 +1054,7 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, /* Give more bits to low-energy MDCTs than they would * otherwise deserve */ - if (B0 > 1 && !dualstereo && (itheta & 0x3fff)) { + if (B0 > 1 && !stereo && (itheta & 0x3fff)) { if (itheta > 8192) /* Rough approximation for pre-echo masking */ delta -= delta >> (4 - duration); @@ -1096,12 +1067,12 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, sbits = b - mbits; f->remaining2 -= qalloc; - if (lowband && !dualstereo) + if (lowband && !stereo) next_lowband2 = lowband + N; /* >32-bit split case */ /* Only stereo needs to pass on lowband_out. * Otherwise, it's handled at the end */ - if (dualstereo) + if (stereo) next_lowband_out1 = lowband_out; else next_level = level + 1; @@ -1112,7 +1083,7 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, * because we need the normalized mid for folding later */ cm = ff_celt_encode_band(f, rc, band, X, NULL, N, mbits, blocks, lowband, duration, next_lowband_out1, - next_level, dualstereo ? 1.0f : (gain * mid), + next_level, stereo ? 1.0f : (gain * mid), lowband_scratch, fill); rebalance = mbits - (rebalance - f->remaining2); @@ -1124,14 +1095,14 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, cm |= ff_celt_encode_band(f, rc, band, Y, NULL, N, sbits, blocks, next_lowband2, duration, NULL, next_level, gain * side, NULL, - fill >> blocks) << ((B0 >> 1) & (dualstereo - 1)); + fill >> blocks) << ((B0 >> 1) & (stereo - 1)); } else { /* For a stereo split, the high bits of fill are always zero, * so no folding will be done to the side. */ cm = ff_celt_encode_band(f, rc, band, Y, NULL, N, sbits, blocks, next_lowband2, duration, NULL, next_level, gain * side, NULL, - fill >> blocks) << ((B0 >> 1) & (dualstereo - 1)); + fill >> blocks) << ((B0 >> 1) & (stereo - 1)); rebalance = sbits - (rebalance - f->remaining2); if (rebalance > 3 << 3 && itheta != 16384) @@ -1141,7 +1112,7 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, * we need the normalized mid for folding later */ cm |= ff_celt_encode_band(f, rc, band, X, NULL, N, mbits, blocks, lowband, duration, next_lowband_out1, - next_level, dualstereo ? 1.0f : (gain * mid), + next_level, stereo ? 1.0f : (gain * mid), lowband_scratch, fill); } } @@ -1164,39 +1135,36 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, f->spread, blocks, gain); } else { /* If there's no pulse, fill the band anyway */ - int j; uint32_t cm_mask = (1 << blocks) - 1; fill &= cm_mask; - if (!fill) { - for (j = 0; j < N; j++) - X[j] = 0.0f; - } else { + if (fill) { if (!lowband) { /* Noise */ - for (j = 0; j < N; j++) - X[j] = (((int32_t)celt_rng(f)) >> 20); + for (i = 0; i < N; i++) + X[i] = (((int32_t)celt_rng(f)) >> 20); cm = cm_mask; } else { /* Folded spectrum */ - for (j = 0; j < N; j++) { + for (i = 0; i < N; i++) { /* About 48 dB below the "normal" folding level */ - X[j] = lowband[j] + (((celt_rng(f)) & 0x8000) ? 1.0f / 256 : -1.0f / 256); + X[i] = lowband[i] + (((celt_rng(f)) & 0x8000) ? 1.0f / 256 : -1.0f / 256); } cm = fill; } celt_renormalize_vector(X, N, gain); + } else { + memset(X, 0, N*sizeof(float)); } } } /* This code is used by the decoder and by the resynthesis-enabled encoder */ - if (dualstereo) { - int j; - if (N != 2) + if (stereo) { + if (N > 2) celt_stereo_merge(X, Y, mid, N); if (inv) { - for (j = 0; j < N; j++) - Y[j] *= -1; + for (i = 0; i < N; i++) + Y[i] *= -1; } } else if (level == 0) { int k; @@ -1204,7 +1172,7 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, /* Undo the sample reorganization going from time order to frequency order */ if (B0 > 1) celt_interleave_hadamard(f->scratch, X, N_B >> recombine, - B0<<recombine, longblocks); + B0 << recombine, longblocks); /* Undo time-freq changes that we did earlier */ N_B = N_B0; @@ -1224,10 +1192,9 @@ uint32_t ff_celt_encode_band(CeltFrame *f, OpusRangeCoder *rc, const int band, /* Scale output for later folding */ if (lowband_out) { - int j; float n = sqrtf(N0); - for (j = 0; j < N0; j++) - lowband_out[j] = n * X[j]; + for (i = 0; i < N0; i++) + lowband_out[i] = n * X[i]; } cm = av_mod_uintp2(cm, blocks); } diff --git a/libavcodec/opustab.c b/libavcodec/opustab.c index 02d548e71a..d0b4bf4bd5 100644 --- a/libavcodec/opustab.c +++ b/libavcodec/opustab.c @@ -930,11 +930,12 @@ const uint8_t ff_celt_bit_deinterleave[] = { 0xC0, 0xC3, 0xCC, 0xCF, 0xF0, 0xF3, 0xFC, 0xFF }; -const uint8_t ff_celt_hadamard_ordery[] = { +const uint8_t ff_celt_hadamard_order[] = { 1, 0, 3, 0, 2, 1, 7, 0, 4, 3, 6, 1, 5, 2, - 15, 0, 8, 7, 12, 3, 11, 4, 14, 1, 9, 6, 13, 2, 10, 5 + 15, 0, 8, 7, 12, 3, 11, 4, 14, 1, 9, 6, 13, 2, 10, 5, + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 }; const uint16_t ff_celt_qn_exp2[] = { diff --git a/libavcodec/opustab.h b/libavcodec/opustab.h index b6be073659..2701a392fd 100644 --- a/libavcodec/opustab.h +++ b/libavcodec/opustab.h @@ -146,7 +146,7 @@ extern const uint8_t ff_celt_log2_frac[]; extern const uint8_t ff_celt_bit_interleave[]; extern const uint8_t ff_celt_bit_deinterleave[]; -extern const uint8_t ff_celt_hadamard_ordery[]; +extern const uint8_t ff_celt_hadamard_order[]; extern const uint16_t ff_celt_qn_exp2[]; extern const uint32_t ff_celt_pvq_u[1272];

1 0

avcodec/nvenc: surface allocation reduction
by Ben Chang 26 Apr '17

26 Apr '17

ffmpeg | branch: master | Ben Chang <benc(a)nvidia.com> | Tue Apr 25 14:57:56 2017 -0700| [8de3458a07376b0a96772e586b6dba5e93432f52] | committer: Timo Rothenpieler avcodec/nvenc: surface allocation reduction This patch aims to reduce the number of input/output surfaces NVENC allocates per session. Previous default sets allocated surfaces to 32 (unless there is user specified param or lookahead involved). Having large number of surfaces consumes extra video memory (esp for higher resolution encoding). The patch changes the surfaces calculation for default, B-frames, lookahead scenario respectively. The other change involves surface selection. Previously, if a session allocates x surfaces, only x-1 surfaces are used (due to combination of output delay and lock toggle logic). To prevent unused surfaces, changing surface rotation to using predefined fifo. Signed-off-by: Timo Rothenpieler <timo(a)rothenpieler.org> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=8de3458a07376b0a9… --- libavcodec/nvenc.c | 55 ++++++++++++++++++++++++++++++++++--------------- libavcodec/nvenc.h | 2 +- libavcodec/nvenc_h264.c | 4 ++-- libavcodec/nvenc_hevc.c | 4 ++-- 4 files changed, 43 insertions(+), 22 deletions(-) diff --git a/libavcodec/nvenc.c b/libavcodec/nvenc.c index cf054550c1..00766c25d0 100644 --- a/libavcodec/nvenc.c +++ b/libavcodec/nvenc.c @@ -644,16 +644,34 @@ static void nvenc_override_rate_control(AVCodecContext *avctx) static av_cold int nvenc_recalc_surfaces(AVCodecContext *avctx) { NvencContext *ctx = avctx->priv_data; - int nb_surfaces = 0; + // default minimum of 4 surfaces + // multiply by 2 for number of NVENCs on gpu (hardcode to 2) + // another multiply by 2 to avoid blocking next PBB group + int nb_surfaces = FFMAX(4, ctx->encode_config.frameIntervalP * 2 * 2); + // lookahead enabled if (ctx->rc_lookahead > 0) { - nb_surfaces = ctx->rc_lookahead + ((ctx->encode_config.frameIntervalP > 0) ? ctx->encode_config.frameIntervalP : 0) + 1 + 4; - if (ctx->nb_surfaces < nb_surfaces) { + // +1 is to account for lkd_bound calculation later + // +4 is to allow sufficient pipelining with lookahead + nb_surfaces = FFMAX(1, FFMAX(nb_surfaces, ctx->rc_lookahead + ctx->encode_config.frameIntervalP + 1 + 4)); + if (nb_surfaces > ctx->nb_surfaces && ctx->nb_surfaces > 0) + { av_log(avctx, AV_LOG_WARNING, "Defined rc_lookahead requires more surfaces, " "increasing used surfaces %d -> %d\n", ctx->nb_surfaces, nb_surfaces); - ctx->nb_surfaces = nb_surfaces; } + ctx->nb_surfaces = FFMAX(nb_surfaces, ctx->nb_surfaces); + } else { + if (ctx->encode_config.frameIntervalP > 1 && ctx->nb_surfaces < nb_surfaces && ctx->nb_surfaces > 0) + { + av_log(avctx, AV_LOG_WARNING, + "Defined b-frame requires more surfaces, " + "increasing used surfaces %d -> %d\n", ctx->nb_surfaces, nb_surfaces); + ctx->nb_surfaces = FFMAX(ctx->nb_surfaces, nb_surfaces); + } + else if (ctx->nb_surfaces <= 0) + ctx->nb_surfaces = nb_surfaces; + // otherwise use user specified value } ctx->nb_surfaces = FFMAX(1, FFMIN(MAX_REGISTERED_FRAMES, ctx->nb_surfaces)); @@ -1086,6 +1104,7 @@ static av_cold int nvenc_alloc_surface(AVCodecContext *avctx, int idx) NvencContext *ctx = avctx->priv_data; NvencDynLoadFunctions *dl_fn = &ctx->nvenc_dload_funcs; NV_ENCODE_API_FUNCTION_LIST *p_nvenc = &dl_fn->nvenc_funcs; + NvencSurface* tmp_surface = &ctx->surfaces[idx]; NVENCSTATUS nv_status; NV_ENC_CREATE_BITSTREAM_BUFFER allocOut = { 0 }; @@ -1121,8 +1140,6 @@ static av_cold int nvenc_alloc_surface(AVCodecContext *avctx, int idx) ctx->surfaces[idx].height = allocSurf.height; } - ctx->surfaces[idx].lockCount = 0; - /* 1MB is large enough to hold most output frames. * NVENC increases this automaticaly if it is not enough. */ allocOut.size = 1024 * 1024; @@ -1141,6 +1158,8 @@ static av_cold int nvenc_alloc_surface(AVCodecContext *avctx, int idx) ctx->surfaces[idx].output_surface = allocOut.bitstreamBuffer; ctx->surfaces[idx].size = allocOut.size; + av_fifo_generic_write(ctx->unused_surface_queue, &tmp_surface, sizeof(tmp_surface), NULL); + return 0; } @@ -1156,6 +1175,11 @@ static av_cold int nvenc_setup_surfaces(AVCodecContext *avctx) ctx->timestamp_list = av_fifo_alloc(ctx->nb_surfaces * sizeof(int64_t)); if (!ctx->timestamp_list) return AVERROR(ENOMEM); + + ctx->unused_surface_queue = av_fifo_alloc(ctx->nb_surfaces * sizeof(NvencSurface*)); + if (!ctx->unused_surface_queue) + return AVERROR(ENOMEM); + ctx->output_surface_queue = av_fifo_alloc(ctx->nb_surfaces * sizeof(NvencSurface*)); if (!ctx->output_surface_queue) return AVERROR(ENOMEM); @@ -1222,6 +1246,7 @@ av_cold int ff_nvenc_encode_close(AVCodecContext *avctx) av_fifo_freep(&ctx->timestamp_list); av_fifo_freep(&ctx->output_surface_ready_queue); av_fifo_freep(&ctx->output_surface_queue); + av_fifo_freep(&ctx->unused_surface_queue); if (ctx->surfaces && avctx->pix_fmt == AV_PIX_FMT_CUDA) { for (i = 0; i < ctx->nb_surfaces; ++i) { @@ -1305,16 +1330,14 @@ av_cold int ff_nvenc_encode_init(AVCodecContext *avctx) static NvencSurface *get_free_frame(NvencContext *ctx) { - int i; + NvencSurface *tmp_surf; - for (i = 0; i < ctx->nb_surfaces; i++) { - if (!ctx->surfaces[i].lockCount) { - ctx->surfaces[i].lockCount = 1; - return &ctx->surfaces[i]; - } - } + if (!(av_fifo_size(ctx->unused_surface_queue) > 0)) + // queue empty + return NULL; - return NULL; + av_fifo_generic_read(ctx->unused_surface_queue, &tmp_surf, sizeof(tmp_surf), NULL); + return tmp_surf; } static int nvenc_copy_frame(AVCodecContext *avctx, NvencSurface *nv_surface, @@ -1712,7 +1735,6 @@ int ff_nvenc_encode_frame(AVCodecContext *avctx, AVPacket *pkt, } if (res) { - inSurf->lockCount = 0; return res; } @@ -1790,8 +1812,7 @@ int ff_nvenc_encode_frame(AVCodecContext *avctx, AVPacket *pkt, if (res) return res; - av_assert0(tmpoutsurf->lockCount); - tmpoutsurf->lockCount--; + av_fifo_generic_write(ctx->unused_surface_queue, &tmpoutsurf, sizeof(tmpoutsurf), NULL); *got_packet = 1; } else { diff --git a/libavcodec/nvenc.h b/libavcodec/nvenc.h index 7dec5cc685..763647b283 100644 --- a/libavcodec/nvenc.h +++ b/libavcodec/nvenc.h @@ -44,7 +44,6 @@ typedef struct NvencSurface NV_ENC_OUTPUT_PTR output_surface; NV_ENC_BUFFER_FORMAT format; int size; - int lockCount; } NvencSurface; typedef struct NvencDynLoadFunctions @@ -110,6 +109,7 @@ typedef struct NvencContext int nb_surfaces; NvencSurface *surfaces; + AVFifoBuffer *unused_surface_queue; AVFifoBuffer *output_surface_queue; AVFifoBuffer *output_surface_ready_queue; AVFifoBuffer *timestamp_list; diff --git a/libavcodec/nvenc_h264.c b/libavcodec/nvenc_h264.c index 2c55b60789..8d44b1f350 100644 --- a/libavcodec/nvenc_h264.c +++ b/libavcodec/nvenc_h264.c @@ -79,8 +79,8 @@ static const AVOption options[] = { 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_PARAMS_RC_2_PASS_FRAMESIZE_CAP }, 0, 0, VE, "rc" }, { "vbr_2pass", "Multi-pass variable bitrate mode", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_PARAMS_RC_2_PASS_VBR }, 0, 0, VE, "rc" }, { "rc-lookahead", "Number of frames to look ahead for rate-control", - OFFSET(rc_lookahead), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, INT_MAX, VE }, - { "surfaces", "Number of concurrent surfaces", OFFSET(nb_surfaces), AV_OPT_TYPE_INT, { .i64 = 32 }, 0, MAX_REGISTERED_FRAMES, VE }, + OFFSET(rc_lookahead), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, INT_MAX, VE }, + { "surfaces", "Number of concurrent surfaces", OFFSET(nb_surfaces), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, MAX_REGISTERED_FRAMES, VE }, { "cbr", "Use cbr encoding mode", OFFSET(cbr), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE }, { "2pass", "Use 2pass encoding mode", OFFSET(twopass), AV_OPT_TYPE_BOOL, { .i64 = -1 }, -1, 1, VE }, { "gpu", "Selects which NVENC capable GPU to use. First GPU is 0, second is 1, and so on.", diff --git a/libavcodec/nvenc_hevc.c b/libavcodec/nvenc_hevc.c index c32ba4220b..6d6750a3d4 100644 --- a/libavcodec/nvenc_hevc.c +++ b/libavcodec/nvenc_hevc.c @@ -78,8 +78,8 @@ static const AVOption options[] = { 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_PARAMS_RC_2_PASS_FRAMESIZE_CAP }, 0, 0, VE, "rc" }, { "vbr_2pass", "Multi-pass variable bitrate mode", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_PARAMS_RC_2_PASS_VBR }, 0, 0, VE, "rc" }, { "rc-lookahead", "Number of frames to look ahead for rate-control", - OFFSET(rc_lookahead), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, INT_MAX, VE }, - { "surfaces", "Number of concurrent surfaces", OFFSET(nb_surfaces), AV_OPT_TYPE_INT, { .i64 = 32 }, 0, MAX_REGISTERED_FRAMES, VE }, + OFFSET(rc_lookahead), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, INT_MAX, VE }, + { "surfaces", "Number of concurrent surfaces", OFFSET(nb_surfaces), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, MAX_REGISTERED_FRAMES, VE }, { "cbr", "Use cbr encoding mode", OFFSET(cbr), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE }, { "2pass", "Use 2pass encoding mode", OFFSET(twopass), AV_OPT_TYPE_BOOL, { .i64 = -1 }, -1, 1, VE }, { "gpu", "Selects which NVENC capable GPU to use. First GPU is 0, second is 1, and so on.",

1 0

lavc/hevcdec: fix invalid use of ff_get_format()
by Clément Bœsch 26 Apr '17

26 Apr '17

ffmpeg | branch: master | Clément Bœsch <u(a)pkh.me> | Wed Apr 26 17:35:28 2017 +0200| [78a5fc4579deb63e1e6b93cd4d6e2ec2dceff931] | committer: Clément Bœsch lavc/hevcdec: fix invalid use of ff_get_format() Regression since 76cc100afba25308eaa909acd8804cc0b42057f6. Spotted-by: James Almer <jamrial(a)gmail.com> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=78a5fc4579deb63e1… --- libavcodec/hevcdec.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/libavcodec/hevcdec.c b/libavcodec/hevcdec.c index b698c0bbc7..2fb08d81d2 100644 --- a/libavcodec/hevcdec.c +++ b/libavcodec/hevcdec.c @@ -367,7 +367,7 @@ static enum AVPixelFormat get_format(HEVCContext *s, const HEVCSPS *sps) *fmt++ = sps->pix_fmt; *fmt = AV_PIX_FMT_NONE; - return ff_get_format(s->avctx, pix_fmts); + return ff_thread_get_format(s->avctx, pix_fmts); } static int set_sps(HEVCContext *s, const HEVCSPS *sps,

1 0

Merge commit 'd4a91e65343be5d79a4afa61c791191e1b57499a'
by Clément Bœsch 26 Apr '17

26 Apr '17

ffmpeg | branch: master | Clément Bœsch <u(a)pkh.me> | Wed Apr 26 17:32:40 2017 +0200| [540b8760e84cda1ecab807c808e61a4cfaa1783c] | committer: Clément Bœsch Merge commit 'd4a91e65343be5d79a4afa61c791191e1b57499a' * commit 'd4a91e65343be5d79a4afa61c791191e1b57499a': pthread_frame: do not run hwaccel decoding asynchronously unless it's safe This commit is a noop, see e0cd598bc4684654d63942e9ff4872c0b48a7dc2 Merged-by: Clément Bœsch <u(a)pkh.me> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=540b8760e84cda1ec… ---

1 0