From 3a3db14c5792e0892e4c0ca0e507a24048c6a1e2 Mon Sep 17 00:00:00 2001 From: Daniil Cherednik Date: Sat, 27 Jan 2018 23:19:55 +0000 Subject: [PATCH 1/2] avcodec/dcaenc: Use planar float frame format Signed-off-by: Daniil Cherednik --- libavcodec/dcaenc.c | 72 ++++++++++++++++++++++++++++----------------------- tests/fate/acodec.mak | 21 +++++---------- 2 files changed, 46 insertions(+), 47 deletions(-) diff --git a/libavcodec/dcaenc.c b/libavcodec/dcaenc.c index 186997c631..da1d7411de 100644 --- a/libavcodec/dcaenc.c +++ b/libavcodec/dcaenc.c @@ -29,6 +29,7 @@ #include "libavutil/common.h" #include "libavutil/ffmath.h" #include "libavutil/opt.h" +#include "libswresample/swresample.h" #include "avcodec.h" #include "dca.h" #include "dcaadpcm.h" @@ -64,6 +65,7 @@ typedef struct DCAEncContext { DCAADPCMEncContext adpcm_ctx; FFTContext mdct; CompressionOptions options; + SwrContext *swr; int frame_size; int frame_bits; int fullband_channels; @@ -242,6 +244,15 @@ static int encode_init(AVCodecContext *avctx) if ((ret = ff_mdct_init(&c->mdct, 9, 0, 1.0)) < 0) return ret; + c->swr = swr_alloc_set_opts(NULL, + AV_CH_LAYOUT_MONO, AV_SAMPLE_FMT_S32P, avctx->sample_rate, + AV_CH_LAYOUT_MONO, avctx->sample_fmt, avctx->sample_rate, + 0, NULL); + if ((ret = swr_init(c->swr)) < 0) { + av_log(avctx, AV_LOG_ERROR, "Error opening the resampler.\n"); + return ret; + } + /* Init all tables */ c->cos_table[0] = 0x7fffffff; c->cos_table[512] = 0; @@ -305,6 +316,7 @@ static int encode_init(AVCodecContext *avctx) static av_cold int encode_close(AVCodecContext *avctx) { DCAEncContext *c = avctx->priv_data; + swr_free(&c->swr); ff_mdct_end(&c->mdct); subband_bufer_free(c); ff_dcaadpcm_free(&c->adpcm_ctx); @@ -312,22 +324,21 @@ static av_cold int encode_close(AVCodecContext *avctx) return 0; } -static void subband_transform(DCAEncContext *c, const int32_t *input) +static void subband_transform(DCAEncContext *c, const float **input) { int ch, subs, i, k, j; for (ch = 0; ch < c->fullband_channels; ch++) { - /* History is copied because it is also needed for PSY */ - int32_t hist[512]; int hist_start = 0; const int chi = c->channel_order_tab[ch]; - - memcpy(hist, &c->history[ch][0], 512 * sizeof(int32_t)); + int32_t *hist = c->history[ch]; for (subs = 0; subs < SUBBAND_SAMPLES; subs++) { int32_t accum[64]; int32_t resp; int band; + int32_t * const to = hist + hist_start; + const float* from = &input[chi][subs * 32]; /* Calculate the convolutions at once */ memset(accum, 0, 64 * sizeof(int32_t)); @@ -354,26 +365,28 @@ static void subband_transform(DCAEncContext *c, const int32_t *input) } /* Copy in 32 new samples from input */ - for (i = 0; i < 32; i++) - hist[i + hist_start] = input[(subs * 32 + i) * c->channels + chi]; + swr_convert(c->swr, + (uint8_t**)&to, 32, + (const uint8_t**)&from, 32); hist_start = (hist_start + 32) & 511; } } } -static void lfe_downsample(DCAEncContext *c, const int32_t *input) +static void lfe_downsample(DCAEncContext *c, const float **input) { /* FIXME: make 128x LFE downsampling possible */ const int lfech = lfe_index[c->channel_config]; int i, j, lfes; - int32_t hist[512]; + int32_t *hist = c->history[c->channels - 1]; int32_t accum; int hist_start = 0; - memcpy(hist, &c->history[c->channels - 1][0], 512 * sizeof(int32_t)); - for (lfes = 0; lfes < DCA_LFE_SAMPLES; lfes++) { + int32_t * const to = hist + hist_start; + const float* from = &input[lfech][lfes * 64]; + /* Calculate the convolution */ accum = 0; @@ -385,8 +398,9 @@ static void lfe_downsample(DCAEncContext *c, const int32_t *input) c->downsampled_lfe[lfes] = accum; /* Copy in 64 new samples from input */ - for (i = 0; i < 64; i++) - hist[i + hist_start] = input[(lfes * 64 + i) * c->channels + lfech]; + swr_convert(c->swr, + (uint8_t**)&to, 64, + (const uint8_t**)&from, 64); hist_start = (hist_start + 64) & 511; } @@ -504,10 +518,11 @@ static void update_band_masking(DCAEncContext *c, int band1, int band2, c->band_masking_cb[band1] = value; } -static void calc_masking(DCAEncContext *c, const int32_t *input) +static void calc_masking(DCAEncContext *c, const float **input) { int i, k, band, ch, ssf; int32_t data[512]; + int32_t *p; for (i = 0; i < 256; i++) for (ssf = 0; ssf < SUBSUBFRAMES; ssf++) @@ -516,11 +531,15 @@ static void calc_masking(DCAEncContext *c, const int32_t *input) for (ssf = 0; ssf < SUBSUBFRAMES; ssf++) for (ch = 0; ch < c->fullband_channels; ch++) { const int chi = c->channel_order_tab[ch]; + const float *from = &input[chi][0]; for (i = 0, k = 128 + 256 * ssf; k < 512; i++, k++) data[i] = c->history[ch][k]; - for (k -= 512; i < 512; i++, k++) - data[i] = input[k * c->channels + chi]; + + p = &data[i]; + swr_convert(c->swr, + (uint8_t**)&p, 512 - i, + (const uint8_t**)&from, 512 - i); adjust_jnd(c, data, c->masking_curve_cb[ssf]); } for (i = 0; i < 256; i++) { @@ -881,18 +900,6 @@ out: c->worst_noise_ever = high; } -static void shift_history(DCAEncContext *c, const int32_t *input) -{ - int k, ch; - - for (k = 0; k < 512; k++) - for (ch = 0; ch < c->channels; ch++) { - const int chi = c->channel_order_tab[ch]; - - c->history[ch][k] = input[k * c->channels + chi]; - } -} - static void fill_in_adpcm_bufer(DCAEncContext *c) { int ch, band; @@ -1175,25 +1182,24 @@ static int encode_frame(AVCodecContext *avctx, AVPacket *avpkt, const AVFrame *frame, int *got_packet_ptr) { DCAEncContext *c = avctx->priv_data; - const int32_t *samples; + const float **samples; int ret, i; if ((ret = ff_alloc_packet2(avctx, avpkt, c->frame_size, 0)) < 0) return ret; - samples = (const int32_t *)frame->data[0]; + samples = (const float**)frame->extended_data; + calc_masking(c, samples); subband_transform(c, samples); if (c->lfe_channel) lfe_downsample(c, samples); - calc_masking(c, samples); if (c->options.adpcm_mode) adpcm_analysis(c); find_peaks(c); assign_bits(c); calc_lfe_scales(c); - shift_history(c, samples); init_put_bits(&c->pb, avpkt->data, avpkt->size); fill_in_adpcm_bufer(c); @@ -1245,7 +1251,7 @@ AVCodec ff_dca_encoder = { .encode2 = encode_frame, .capabilities = AV_CODEC_CAP_EXPERIMENTAL, .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP, - .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S32, + .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE }, .supported_samplerates = sample_rates, .channel_layouts = (const uint64_t[]) { AV_CH_LAYOUT_MONO, diff --git a/tests/fate/acodec.mak b/tests/fate/acodec.mak index 80d26de0f9..a072c203c2 100644 --- a/tests/fate/acodec.mak +++ b/tests/fate/acodec.mak @@ -99,20 +99,13 @@ FATE_ACODEC-$(call ENCDEC, ALAC, MOV) += fate-acodec-alac fate-acodec-alac: FMT = mov fate-acodec-alac: CODEC = alac -compression_level 1 -FATE_ACODEC-$(call ENCDEC, DCA, DTS) += fate-acodec-dca -fate-acodec-dca: tests/data/asynth-44100-2.wav -fate-acodec-dca: SRC = tests/data/asynth-44100-2.wav -fate-acodec-dca: CMD = md5 -i $(TARGET_PATH)/$(SRC) -c:a dca -strict -2 -f dts -flags +bitexact -fate-acodec-dca: CMP = oneline -fate-acodec-dca: REF = 2aa580ac67820fce4f581b96ebb34acc - -FATE_ACODEC-$(call ENCDEC, DCA, WAV) += fate-acodec-dca2 -fate-acodec-dca2: CMD = enc_dec_pcm dts wav s16le $(SRC) -c:a dca -strict -2 -flags +bitexact -fate-acodec-dca2: REF = $(SRC) -fate-acodec-dca2: CMP = stddev -fate-acodec-dca2: CMP_SHIFT = -2048 -fate-acodec-dca2: CMP_TARGET = 535 -fate-acodec-dca2: SIZE_TOLERANCE = 1632 +FATE_ACODEC-$(call ENCDEC, DCA, WAV) += fate-acodec-dca +fate-acodec-dca: CMD = enc_dec_pcm dts wav s16le $(SRC) -c:a dca -strict -2 +fate-acodec-dca: REF = $(SRC) +fate-acodec-dca: CMP = stddev +fate-acodec-dca: CMP_SHIFT = -2048 +fate-acodec-dca: CMP_TARGET = 535 +fate-acodec-dca: SIZE_TOLERANCE = 1632 FATE_ACODEC-$(call ENCDEC, FLAC, FLAC) += fate-acodec-flac fate-acodec-flac-exact-rice fate-acodec-flac: FMT = flac -- 2.13.5