[FFmpeg-devel] [PATCH 2/2] dirac: implement a native Dirac HQ encoder
Rostislav Pehlivanov
atomnuker at gmail.com
Fri Jan 22 19:15:43 CET 2016
Changes from the first vesion:
- Support 100% of the specifications.
- Enable 5 level transforms and custom quantization matrices.
- Raise the max slice size limit to 1024x1024 (works & supported)
- Change default settings to 5 level transforms with 128x64 slices
- Share tables between the encoder and decoders
- Require a strict level of -1 to enable features limited by level
- Fix a typo in the license headers - all headers are correct now
- Optimize coefficient encoding
- Optimize quantization index search
- Parallelize the DWT transform
The overall performance of the encoder is highly dependent of the number
of threads used.
The decoder can be nearly twice as slow as the encoder, so there's still
work to be done to optimize the dequantization process there.
Original commit message:
This commit adds a new encoder capable of creating BBC Dirac VC-2 HQ
profile files.
Dirac is a wavelet based codec created by the BBC a little more than 10
years ago. Since then, wavelets have mostly gone out of style as they
did not provide adequate encoding gains at lower bitrates. Dirac was a
fully featured video codec equipped with perceptual masking, support for
most popular pixel formats, interlacing, overlapped-block motion
compensation, and other features. It found new life after being stripped
of various features and standardized as the VC-2 codec by the SMPTE with
an extra profile, the HQ profile that this encoder supports, added.
The HQ profile was based off of the Low-Delay profile previously
existing in Dirac. The profile forbids DC prediction and arithmetic
coding to focus on high performance and low delay at higher bitrates.
The standard bitrates for this profile vary but generally 1:4
compression is expected (~525 Mbps vs the 2200 Mbps for uncompressed
1080p50). The codec only supports I-frames, hence the high bitrates.
The structure of this encoder is simple: do a DWT transform on the
entire image, split it into multiple slices (specified by the user) and
encode them in parallel. All of the slices are of the same size, making
rate control and threading very trivial. Although only in C, this
encoder
is capable of 30 frames per second on an 4 core 8 threads Ivy Bridge.
A lookup table is used to encode most of the coefficients.
No code was used from the GSoC encoder from 2007 except for the 2
transform functions in diracenc_transforms.c. All other code was written
from scratch.
This encoder outperforms any other encoders in quality, usability and in
features. Other existing implementations do not support 4 level
transforms or 64x64 blocks (slices), which greatly increase compression.
As previously said, the codec is meant for broadcasting, hence support
for non-broadcasting image widths, heights, bit depths, aspect ratios,
etc. are limited by the "level". Although this codec supports a few
chroma subsamplings (420, 422, 444), signalling those is generally
outside the specifications of the level used (3) and the reference
decoder will outright refuse to read any image with such a flag
signalled (it only supports 1920x1080 yuv422p10). However, most
implementations will happily read files with alternate dimensions,
framerates and formats signalled.
Therefore, in order to encode files other than 1080p50 yuv422p10le, you
need to provide an "-strict -2" argument to the command line. The FFmpeg
decoder will happily read any files made with non-standard parameters,
dimensions and subsamplings, and so will other implementations. IMO this
should be "-strict -1", but I'll leave that up for discussion.
There are still plenty of stuff to implement, for instance 5 more
wavelet transforms are still in the specs and supported by the decoder.
Signed-off-by: Rostislav Pehlivanov <atomnuker at gmail.com>
---
libavcodec/Makefile | 1 +
libavcodec/allcodecs.c | 2 +-
libavcodec/diracenc.c | 1149 ++++++++++++++++++++++++++++++++++++++
libavcodec/diracenc_transforms.c | 246 ++++++++
libavcodec/diracenc_transforms.h | 53 ++
5 files changed, 1450 insertions(+), 1 deletion(-)
create mode 100644 libavcodec/diracenc.c
create mode 100644 libavcodec/diracenc_transforms.c
create mode 100644 libavcodec/diracenc_transforms.h
diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index b9ffdb9..b8753de 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -228,6 +228,7 @@ OBJS-$(CONFIG_DCA_ENCODER) += dcaenc.o dca.o dcadata.o
OBJS-$(CONFIG_DDS_DECODER) += dds.o
OBJS-$(CONFIG_DIRAC_DECODER) += diracdec.o dirac.o diracdsp.o \
dirac_arith.o mpeg12data.o dirac_dwt.o
+OBJS-$(CONFIG_DIRAC_ENCODER) += diracenc.o diracenc_transforms.o
OBJS-$(CONFIG_DFA_DECODER) += dfa.o
OBJS-$(CONFIG_DNXHD_DECODER) += dnxhddec.o dnxhddata.o
OBJS-$(CONFIG_DNXHD_ENCODER) += dnxhdenc.o dnxhddata.o
diff --git a/libavcodec/allcodecs.c b/libavcodec/allcodecs.c
index 2128546..68fb24c 100644
--- a/libavcodec/allcodecs.c
+++ b/libavcodec/allcodecs.c
@@ -156,7 +156,7 @@ void avcodec_register_all(void)
REGISTER_DECODER(CYUV, cyuv);
REGISTER_DECODER(DDS, dds);
REGISTER_DECODER(DFA, dfa);
- REGISTER_DECODER(DIRAC, dirac);
+ REGISTER_ENCDEC (DIRAC, dirac);
REGISTER_ENCDEC (DNXHD, dnxhd);
REGISTER_ENCDEC (DPX, dpx);
REGISTER_DECODER(DSICINVIDEO, dsicinvideo);
diff --git a/libavcodec/diracenc.c b/libavcodec/diracenc.c
new file mode 100644
index 0000000..ffff64f
--- /dev/null
+++ b/libavcodec/diracenc.c
@@ -0,0 +1,1149 @@
+/*
+ * Copyright (C) 2016 Open Broadcast Systems Ltd.
+ * Author (C) 2016 Rostislav Pehlivanov <atomnuker at gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/ffversion.h"
+#include "libavutil/pixdesc.h"
+#include "libavutil/opt.h"
+#include "dirac.h"
+#include "put_bits.h"
+#include "internal.h"
+
+#include "diracenc_transforms.h"
+
+#define COEF_LUT_TAB 2048
+
+enum DiracQM {
+ DIRAC_QM_DEF = 0,
+ DIRAC_QM_COL,
+ DIRAC_QM_FLAT,
+
+ DIRAC_QM_NB
+};
+
+typedef struct SubBand {
+ dwtcoef *buf;
+ int stride;
+ int width;
+ int height;
+} SubBand;
+
+typedef struct Plane {
+ SubBand band[MAX_DWT_LEVELS][4];
+ dwtcoef *coef_buf;
+ int width;
+ int height;
+ int dwt_width;
+ int dwt_height;
+ int coef_stride;
+} Plane;
+
+typedef struct SliceArgs {
+ PutBitContext pb;
+ void *ctx;
+ int x;
+ int y;
+ int quant_idx;
+ int bits_ceil;
+ int bytes;
+} SliceArgs;
+
+typedef struct TransformArgs {
+ void *ctx;
+ Plane *plane;
+ void *idata;
+ int istride;
+ int field;
+ DiracTransforms t;
+} TransformArgs;
+
+typedef struct DiracEncContext {
+ AVClass *av_class;
+ PutBitContext pb;
+ Plane plane[3];
+ AVCodecContext *avctx;
+ DiracVersionInfo ver;
+
+ SliceArgs *slice_args;
+ TransformArgs transform_args[3];
+
+ /* For conversion from unsigned pixel values to signed */
+ int diff_offset;
+ int bpp;
+
+ /* Picture number */
+ uint32_t picture_number;
+
+ /* Base video format */
+ int base_vf;
+ int level;
+ int profile;
+
+ /* Quantization matrix */
+ uint8_t quant[MAX_DWT_LEVELS][4];
+
+ /* Coefficient LUT */
+ uint32_t *coef_lut_val;
+ uint8_t *coef_lut_len;
+
+ int num_x; /* #slices horizontally */
+ int num_y; /* #slices vertically */
+ int prefix_bytes;
+ int size_scaler;
+ int chroma_x_shift;
+ int chroma_y_shift;
+
+ /* Rate control stuff */
+ int slice_max_bytes;
+ int q_ceil;
+ int q_start;
+
+ /* Options */
+ double tolerance;
+ int wavelet_idx;
+ int wavelet_depth;
+ int strict_compliance;
+ int slice_height;
+ int slice_width;
+ int interlaced;
+ enum DiracQM quant_matrix;
+
+ /* Parse code state */
+ uint32_t next_parse_offset;
+ enum DiracParseCodes last_parse_code;
+} DiracEncContext;
+
+static av_always_inline void put_dirac_ue_uint(PutBitContext *pb, uint32_t val)
+{
+ int i;
+ int pbits = 0, bits = 0, topbit = 1, maxval = 1;
+
+ if (!val++) {
+ put_bits(pb, 1, 1);
+ return;
+ }
+
+ while (val > maxval) {
+ topbit <<= 1;
+ maxval <<= 1;
+ maxval |= 1;
+ }
+
+ bits = ff_log2(topbit);
+
+ for (i = 0; i < bits; i++) {
+ topbit >>= 1;
+ pbits <<= 2;
+ if (val & topbit)
+ pbits |= 0x1;
+ }
+
+ put_bits(pb, bits*2 + 1, (pbits << 1) | 1);
+}
+
+static av_always_inline int count_dirac_ue_uint(uint16_t val)
+{
+ int topbit = 1, maxval = 1;
+
+ if (!val++)
+ return 1;
+
+ while (val > maxval) {
+ topbit <<= 1;
+ maxval <<= 1;
+ maxval |= 1;
+ }
+
+ return ff_log2(topbit)*2 + 1;
+}
+
+static av_always_inline void get_dirac_ue_uint(uint16_t val, uint8_t *nbits,
+ uint32_t *eval)
+{
+ int i;
+ int pbits = 0, bits = 0, topbit = 1, maxval = 1;
+
+ if (!val++) {
+ *nbits = 1;
+ *eval = 1;
+ return;
+ }
+
+ while (val > maxval) {
+ topbit <<= 1;
+ maxval <<= 1;
+ maxval |= 1;
+ }
+
+ bits = ff_log2(topbit);
+
+ for (i = 0; i < bits; i++) {
+ topbit >>= 1;
+ pbits <<= 2;
+ if (val & topbit)
+ pbits |= 0x1;
+ }
+
+ *nbits = bits*2 + 1;
+ *eval = (pbits << 1) | 1;
+}
+
+/* VC-2 10.4 - parse_info() */
+static void encode_parse_info(DiracEncContext *s, enum DiracParseCodes pcode)
+{
+ uint32_t cur_pos, dist;
+
+ avpriv_align_put_bits(&s->pb);
+
+ cur_pos = put_bits_count(&s->pb) >> 3;
+
+ /* Magic string */
+ avpriv_put_string(&s->pb, "BBCD", 0);
+
+ /* Parse code */
+ put_bits(&s->pb, 8, pcode);
+
+ /* Next parse offset */
+ dist = cur_pos - s->next_parse_offset;
+ AV_WB32(s->pb.buf + s->next_parse_offset + 5, dist);
+ s->next_parse_offset = cur_pos;
+ put_bits32(&s->pb, pcode == DIRAC_PCODE_END_SEQ ? 13 : 0);
+
+ /* Last parse offset */
+ put_bits32(&s->pb, s->last_parse_code == DIRAC_PCODE_END_SEQ ? 13 : dist);
+
+ s->last_parse_code = pcode;
+}
+
+/* VC-2 11.1 - parse_parameters()
+ * The level dictates what the decoder should expect in terms of resolution
+ * and allows it to quickly reject whatever it can't support. Remember,
+ * this codec kinda targets cheapo FPGAs without much memory. Unfortunately
+ * it also limits us greatly in our choice of formats, hence the flag to disable
+ * strict_compliance */
+static void encode_parse_params(DiracEncContext *s)
+{
+ put_dirac_ue_uint(&s->pb, s->ver.major); /* VC-2 demands this to be 2 */
+ put_dirac_ue_uint(&s->pb, s->ver.minor); /* ^^ and this to be 0 */
+ put_dirac_ue_uint(&s->pb, s->profile); /* 3 to signal HQ profile */
+ put_dirac_ue_uint(&s->pb, s->level); /* 3 - 1080/720, 6 - 4K */
+}
+
+/* VC-2 11.3 - frame_size() */
+static void encode_frame_size(DiracEncContext *s)
+{
+ put_bits(&s->pb, 1, !s->strict_compliance);
+ if (!s->strict_compliance) {
+ AVCodecContext *avctx = s->avctx;
+ put_dirac_ue_uint(&s->pb, avctx->width);
+ put_dirac_ue_uint(&s->pb, avctx->height);
+ }
+}
+
+/* VC-2 11.3.3 - color_diff_sampling_format() */
+static void encode_sample_fmt(DiracEncContext *s)
+{
+ put_bits(&s->pb, 1, !s->strict_compliance);
+ if (!s->strict_compliance) {
+ int idx;
+ if (s->chroma_x_shift == 1 && s->chroma_y_shift == 0)
+ idx = 1; /* 422 */
+ else if (s->chroma_x_shift == 1 && s->chroma_y_shift == 1)
+ idx = 2; /* 420 */
+ else
+ idx = 0; /* 444 */
+ put_dirac_ue_uint(&s->pb, idx);
+ }
+}
+
+/* VC-2 11.3.4 - scan_format() */
+static void encode_scan_format(DiracEncContext *s)
+{
+ put_bits(&s->pb, 1, !s->strict_compliance);
+ if (!s->strict_compliance)
+ put_dirac_ue_uint(&s->pb, s->interlaced);
+}
+
+/* VC-2 11.3.5 - frame_rate() */
+static void encode_frame_rate(DiracEncContext *s)
+{
+ put_bits(&s->pb, 1, !s->strict_compliance);
+ if (!s->strict_compliance) {
+ AVCodecContext *avctx = s->avctx;
+ put_dirac_ue_uint(&s->pb, 0);
+ put_dirac_ue_uint(&s->pb, avctx->time_base.den);
+ put_dirac_ue_uint(&s->pb, avctx->time_base.num);
+ }
+}
+
+/* VC-2 11.3.6 - aspect_ratio() */
+static void encode_aspect_ratio(DiracEncContext *s)
+{
+ put_bits(&s->pb, 1, !s->strict_compliance);
+ if (!s->strict_compliance) {
+ AVCodecContext *avctx = s->avctx;
+ put_dirac_ue_uint(&s->pb, 0);
+ put_dirac_ue_uint(&s->pb, avctx->sample_aspect_ratio.num);
+ put_dirac_ue_uint(&s->pb, avctx->sample_aspect_ratio.den);
+ }
+}
+
+/* VC-2 11.3.7 - clean_area() */
+static void encode_clean_area(DiracEncContext *s)
+{
+ put_bits(&s->pb, 1, 0);
+}
+
+/* VC-2 11.3.8 - signal_range() */
+static void encode_signal_range(DiracEncContext *s)
+{
+ int idx;
+ AVCodecContext *avctx = s->avctx;
+ const AVPixFmtDescriptor *fmt = av_pix_fmt_desc_get(avctx->pix_fmt);
+ const int depth = fmt->comp[0].depth;
+ if (depth == 8 && avctx->color_range == AVCOL_RANGE_JPEG) {
+ idx = 1;
+ s->bpp = 1;
+ s->diff_offset = 128;
+ } else if (depth == 8 && (avctx->color_range == AVCOL_RANGE_MPEG ||
+ avctx->color_range == AVCOL_RANGE_UNSPECIFIED)) {
+ idx = 2;
+ s->bpp = 1;
+ s->diff_offset = 128;
+ } else if (depth == 10) {
+ idx = 3;
+ s->bpp = 2;
+ s->diff_offset = 512;
+ } else {
+ idx = 4;
+ s->bpp = 2;
+ s->diff_offset = 2048;
+ }
+ put_bits(&s->pb, 1, !s->strict_compliance);
+ if (!s->strict_compliance)
+ put_dirac_ue_uint(&s->pb, idx);
+}
+
+/* VC-2 11.3.9 - color_spec() */
+static void encode_color_spec(DiracEncContext *s)
+{
+ put_bits(&s->pb, 1, 0);
+}
+
+/* VC-2 11.3 - source_parameters() */
+static void encode_source_params(DiracEncContext *s)
+{
+ encode_frame_size(s);
+ encode_sample_fmt(s);
+ encode_scan_format(s);
+ encode_frame_rate(s);
+ encode_aspect_ratio(s);
+ encode_clean_area(s);
+ encode_signal_range(s);
+ encode_color_spec(s);
+}
+
+/* VC-2 11 - sequence_header() */
+static void encode_seq_header(DiracEncContext *s)
+{
+ avpriv_align_put_bits(&s->pb);
+ encode_parse_params(s);
+ put_dirac_ue_uint(&s->pb, s->base_vf);
+ encode_source_params(s);
+ put_dirac_ue_uint(&s->pb, s->interlaced); /* Frames or fields coding */
+}
+
+/* VC-2 12.1 - picture_header() */
+static void encode_picture_header(DiracEncContext *s)
+{
+ avpriv_align_put_bits(&s->pb);
+ put_bits32(&s->pb, s->picture_number++);
+}
+
+/* VC-2 12.3.4.1 - slice_parameters() */
+static void encode_slice_params(DiracEncContext *s)
+{
+ put_dirac_ue_uint(&s->pb, s->num_x);
+ put_dirac_ue_uint(&s->pb, s->num_y);
+ put_dirac_ue_uint(&s->pb, s->prefix_bytes);
+ put_dirac_ue_uint(&s->pb, s->size_scaler);
+}
+
+/* 1st idx = LL, second - vertical, third - horizontal, fourth - total */
+const uint8_t dirac_qm_col_tab[MAX_DWT_LEVELS][4] = {
+ {20, 9, 15, 4},
+ {0, 6, 6, 4},
+ {0, 3, 3, 5},
+ {0, 3, 5, 1},
+ {0, 11, 10, 11}
+};
+
+const uint8_t dirac_qm_flat_tab[MAX_DWT_LEVELS][4] = {
+ {0, 0, 0, 0},
+ {0, 0, 0, 0},
+ {0, 0, 0, 0},
+ {0, 0, 0, 0},
+ {0, 0, 0, 0}
+};
+
+static void init_custom_qm(DiracEncContext *s)
+{
+ int level, orientation;
+
+ if (s->quant_matrix == DIRAC_QM_DEF) {
+ for (level = 0; level < s->wavelet_depth; level++) {
+ for (orientation = 0; orientation < 4; orientation++) {
+ if (level < 5)
+ s->quant[level][orientation] = default_qmat[s->wavelet_idx][level][orientation];
+ else
+ s->quant[level][orientation] = dirac_qm_col_tab[level][orientation];
+ }
+ }
+ } else if (s->quant_matrix == DIRAC_QM_COL) {
+ for (level = 0; level < s->wavelet_depth; level++) {
+ for (orientation = 0; orientation < 4; orientation++) {
+ s->quant[level][orientation] = dirac_qm_col_tab[level][orientation];
+ }
+ }
+ } else {
+ for (level = 0; level < s->wavelet_depth; level++) {
+ for (orientation = 0; orientation < 4; orientation++) {
+ s->quant[level][orientation] = dirac_qm_flat_tab[level][orientation];
+ }
+ }
+ }
+}
+
+/* VC-2 12.3.4.2 - quant_matrix() */
+static void encode_quant_matrix(DiracEncContext *s)
+{
+ int level, custom_quant_matrix = 0;
+ if (s->wavelet_depth > 4 || s->quant_matrix != DIRAC_QM_DEF)
+ custom_quant_matrix = 1;
+ put_bits(&s->pb, 1, custom_quant_matrix);
+ if (custom_quant_matrix) {
+ init_custom_qm(s);
+ put_dirac_ue_uint(&s->pb, s->quant[0][0]);
+ for (level = 0; level < s->wavelet_depth; level++) {
+ put_dirac_ue_uint(&s->pb, s->quant[level][1]);
+ put_dirac_ue_uint(&s->pb, s->quant[level][2]);
+ put_dirac_ue_uint(&s->pb, s->quant[level][3]);
+ }
+ } else {
+ for (level = 0; level < s->wavelet_depth; level++) {
+ s->quant[level][0] = default_qmat[s->wavelet_idx][level][0];
+ s->quant[level][1] = default_qmat[s->wavelet_idx][level][1];
+ s->quant[level][2] = default_qmat[s->wavelet_idx][level][2];
+ s->quant[level][3] = default_qmat[s->wavelet_idx][level][3];
+ }
+ }
+}
+
+/* VC-2 12.3 - transform_parameters() */
+static void encode_transform_params(DiracEncContext *s)
+{
+ put_dirac_ue_uint(&s->pb, s->wavelet_idx);
+ put_dirac_ue_uint(&s->pb, s->wavelet_depth);
+
+ encode_slice_params(s);
+ encode_quant_matrix(s);
+}
+
+/* VC-2 12.2 - wavelet_transform() */
+static void encode_wavelet_transform(DiracEncContext *s)
+{
+ encode_transform_params(s);
+ avpriv_align_put_bits(&s->pb);
+ /* Continued after DWT in encode_transform_data() */
+}
+
+/* VC-2 12 - picture_parse() */
+static void encode_picture_start(DiracEncContext *s)
+{
+ avpriv_align_put_bits(&s->pb);
+ encode_picture_header(s);
+ avpriv_align_put_bits(&s->pb);
+ encode_wavelet_transform(s);
+}
+
+static av_always_inline void coeff_quantize_get(qcoef coeff, int qfactor, int qoffset,
+ uint8_t *len, uint32_t *eval)
+{
+ uint16_t acoef;
+ int sign = coeff < 0;
+ coeff -= sign;
+ coeff ^= -sign;
+ coeff <<= 2;
+ coeff = (coeff - qoffset)/qfactor;
+ acoef = abs(coeff);
+ get_dirac_ue_uint(acoef, len, eval);
+ if (acoef) {
+ *eval = (*eval << 1) | sign;
+ *len += 1;
+ }
+}
+
+static av_always_inline void coeff_quantize_encode(PutBitContext *pb, qcoef coeff,
+ int qfactor, int qoffset)
+{
+ uint16_t acoef;
+ int sign = coeff < 0;
+ coeff -= sign;
+ coeff ^= -sign;
+ coeff <<= 2;
+ coeff = (coeff - qoffset)/qfactor;
+ acoef = abs(coeff);
+ put_dirac_ue_uint(pb, acoef);
+ if (acoef)
+ put_bits(pb, 1, sign);
+}
+
+/* VC-2 13.5.5.2 - slice_band() */
+static void encode_subband(DiracEncContext *s, PutBitContext *pb, int sx, int sy,
+ SubBand *b, int quant)
+{
+ int x, y;
+
+ int left = b->width * (sx+0) / s->num_x;
+ int right = b->width * (sx+1) / s->num_x;
+ int top = b->height * (sy+0) / s->num_y;
+ int bottom = b->height * (sy+1) / s->num_y;
+
+ int qfactor = qscale_tab[quant];
+ int qoffset = qoffset_intra_tab[quant] + 2;
+ uint8_t *len_lut = &s->coef_lut_len[2*quant*COEF_LUT_TAB + COEF_LUT_TAB];
+ uint32_t *val_lut = &s->coef_lut_val[2*quant*COEF_LUT_TAB + COEF_LUT_TAB];
+
+ dwtcoef *coeff = b->buf + top * b->stride;
+
+ for (y = top; y < bottom; y++) {
+ for (x = left; x < right; x++) {
+ if (coeff[x] >= -COEF_LUT_TAB && coeff[x] < COEF_LUT_TAB)
+ put_bits(pb, len_lut[coeff[x]], val_lut[coeff[x]]);
+ else
+ coeff_quantize_encode(pb, coeff[x], qfactor, qoffset);
+ }
+ coeff += b->stride;
+ }
+}
+
+static int count_hq_slice(DiracEncContext *s, int slice_x,
+ int slice_y, int quant_idx)
+{
+ int x, y, left, right, top, bottom, qfactor, qoffset;
+ uint8_t quants[MAX_DWT_LEVELS][4];
+ int p, level, orientation;
+ int bits = 8*s->prefix_bytes + 40;
+
+ for (level = 0; level < s->wavelet_depth; level++)
+ for (orientation = !!level; orientation < 4; orientation++)
+ quants[level][orientation] = FFMAX(quant_idx - s->quant[level][orientation], 0);
+
+ for (p = 0; p < 3; p++) {
+ int bytes_start, bytes_end, bytes_len, bytes_pad;
+ bytes_start = bits >> 3;
+ for (level = 0; level < s->wavelet_depth; level++) {
+ for (orientation = !!level; orientation < 4; orientation++) {
+ SubBand *b = &s->plane[p].band[level][orientation];
+
+ quant_idx = quants[level][orientation];
+ qfactor = qscale_tab[quant_idx];
+ qoffset = qoffset_intra_tab[quant_idx] + 2;
+
+ left = b->width * slice_x / s->num_x;
+ right = b->width *(slice_x+1) / s->num_x;
+ top = b->height * slice_y / s->num_y;
+ bottom = b->height *(slice_y+1) / s->num_y;
+
+ dwtcoef *buf = b->buf + top * b->stride;
+
+ for (y = top; y < bottom; y++) {
+ for (x = left; x < right; x++) {
+ qcoef coeff = (qcoef)buf[x];
+ if (coeff >= -COEF_LUT_TAB && coeff < COEF_LUT_TAB) {
+ bits += s->coef_lut_len[2*quant_idx*COEF_LUT_TAB + coeff + COEF_LUT_TAB];
+ } else {
+ const int sign = coeff < 0;
+ coeff -= sign;
+ coeff ^= -sign;
+ coeff <<= 2;
+ coeff = (coeff - qoffset)/qfactor;
+ bits += count_dirac_ue_uint(abs(coeff));
+ bits += !!coeff;
+ }
+ }
+ buf += b->stride;
+ }
+ }
+ }
+ bytes_end = bits >> 3;
+ bytes_len = FFALIGN((bytes_end - bytes_start), 64)/s->size_scaler;
+ bytes_pad = (bytes_len*s->size_scaler) - (bytes_end - bytes_start);
+ bits += bytes_pad*8;
+ }
+
+ return bits;
+}
+
+/* Approaches the best possible quantizer asymptotically, its kinda exaustive
+ * but we have a LUT to get the coefficient size in bits. Guaranteed to never
+ * overshoot, which is apparently very important when streaming */
+static int rate_control(AVCodecContext *avctx, void *arg)
+{
+ SliceArgs *slice_dat = arg;
+ DiracEncContext *s = slice_dat->ctx;
+ const int sx = slice_dat->x;
+ const int sy = slice_dat->y;
+ int quant_buf[2], bits_buf[2], quant = s->q_start, range = s->q_start/3;
+ const int64_t top = slice_dat->bits_ceil;
+ const double percent = s->tolerance;
+ const double bottom = top - top*(percent/100.0f);
+ int bits = count_hq_slice(s, sx, sy, quant);
+ range -= range & 1; /* Make it an even number */
+ while ((bits > top) || (bits < bottom)) {
+ range *= bits > top ? +1 : -1;
+ quant = av_clip(quant + range, 0, s->q_ceil);
+ bits = count_hq_slice(s, sx, sy, quant);
+ range = av_clip(range/2, 1, s->q_ceil);
+ if (quant_buf[1] == quant) {
+ quant = bits_buf[0] < bits ? quant_buf[0] : quant;
+ bits = bits_buf[0] < bits ? bits_buf[0] : bits;
+ break;
+ }
+ quant_buf[1] = quant_buf[0];
+ quant_buf[0] = quant;
+ bits_buf[1] = bits_buf[0];
+ bits_buf[0] = bits;
+ }
+ slice_dat->quant_idx = av_clip(++quant, 0, s->q_ceil);
+ slice_dat->bytes = FFALIGN((bits >> 3), s->size_scaler) + 4;
+
+ return 0;
+}
+
+/* VC-2 13.5.3 - hq_slice */
+static int encode_hq_slice(AVCodecContext *avctx, void *arg)
+{
+ SliceArgs *slice_dat = arg;
+ DiracEncContext *s = slice_dat->ctx;
+ PutBitContext *pb = &slice_dat->pb;
+ const int slice_x = slice_dat->x;
+ const int slice_y = slice_dat->y;
+ const int quant_idx = slice_dat->quant_idx;
+ const int slice_bytes_max = slice_dat->bytes;
+ uint8_t quants[MAX_DWT_LEVELS][4];
+ int p, j, level, orientation;
+ int slice_bytes_start = put_bits_count(pb) >> 3;
+
+ avpriv_align_put_bits(pb);
+ put_bits(pb, 8*s->prefix_bytes, 0);
+ put_bits(pb, 8, quant_idx);
+
+ /* Slice quantization (slice_quantizers() in the specs) */
+ for (level = 0; level < s->wavelet_depth; level++)
+ for (orientation = !!level; orientation < 4; orientation++)
+ quants[level][orientation] = FFMAX(quant_idx - s->quant[level][orientation], 0);
+
+ /* Luma + 2 Chroma planes */
+ for (p = 0; p < 3; p++) {
+ int bytes_start, bytes_len, pad_s, pad_c;
+ avpriv_align_put_bits(pb);
+ bytes_start = put_bits_count(pb) >> 3;
+ put_bits(pb, 8, 0);
+ /*
+ * Orientation 0 (LL) only exists for the very first level, hence why
+ * orientation starts at !!level.
+ */
+ for (level = 0; level < s->wavelet_depth; level++) {
+ for (orientation = !!level; orientation < 4; orientation++) {
+ encode_subband(s, pb, slice_x, slice_y,
+ &s->plane[p].band[level][orientation],
+ quants[level][orientation]);
+ }
+ }
+ avpriv_align_put_bits(pb);
+ bytes_len = (put_bits_count(pb) >> 3) - bytes_start - 1;
+ if (p == 2) {
+ int slice_bytes_end = (put_bits_count(pb) >> 3);
+ int slice_bytes_len = (slice_bytes_end - slice_bytes_start);
+ int len_diff = slice_bytes_max - slice_bytes_len;
+ pad_s = FFALIGN((bytes_len + len_diff), s->size_scaler)/s->size_scaler;
+ pad_c = (pad_s*s->size_scaler) - bytes_len;
+ } else {
+ pad_s = FFALIGN(bytes_len, s->size_scaler)/s->size_scaler;
+ pad_c = (pad_s*s->size_scaler) - bytes_len;
+ }
+ pb->buf[bytes_start] = pad_s;
+ for (j = 0; j < pad_c; j++)
+ put_bits(pb, 8, 0);
+ }
+
+ return 0;
+}
+
+/* VC-2 13.5.1 - low_delay_transform_data() */
+static int encode_slices(DiracEncContext *s)
+{
+ uint8_t *buf;
+ int slice_x, slice_y, skip = 0;
+ SliceArgs *enc_args = s->slice_args;
+
+ avpriv_align_put_bits(&s->pb);
+
+ flush_put_bits(&s->pb);
+
+ buf = put_bits_ptr(&s->pb);
+
+ for (slice_y = 0; slice_y < s->num_y; slice_y++) {
+ for (slice_x = 0; slice_x < s->num_x; slice_x++) {
+ SliceArgs *args = &enc_args[s->num_x*slice_y + slice_x];
+ args->ctx = s;
+ args->x = slice_x;
+ args->y = slice_y;
+ args->bits_ceil = s->slice_max_bytes << 3;
+ }
+ }
+
+ /* Determine quantization indices and bytes per slice */
+ s->avctx->execute(s->avctx, rate_control, enc_args, NULL, s->num_x*s->num_y,
+ sizeof(SliceArgs));
+
+ for (slice_y = 0; slice_y < s->num_y; slice_y++) {
+ for (slice_x = 0; slice_x < s->num_x; slice_x++) {
+ SliceArgs *args = &enc_args[s->num_x*slice_y + slice_x];
+ init_put_bits(&args->pb, buf + skip, 2*args->bytes);
+ s->q_start = (s->q_start + args->quant_idx)/2;
+ skip += args->bytes;
+ }
+ }
+
+ s->avctx->execute(s->avctx, encode_hq_slice, enc_args, NULL, s->num_x*s->num_y,
+ sizeof(SliceArgs));
+
+ skip_put_bytes(&s->pb, skip);
+
+ return 0;
+}
+
+/*
+ * Transform basics for a 3 level transform
+ * |---------------------------------------------------------------------|
+ * | LL-0 | HL-0 | | |
+ * |--------|-------| HL-1 | |
+ * | LH-0 | HH-0 | | |
+ * |----------------|-----------------| HL-2 |
+ * | | | |
+ * | LH-1 | HH-1 | |
+ * | | | |
+ * |----------------------------------|----------------------------------|
+ * | | |
+ * | | |
+ * | | |
+ * | LH-2 | HH-2 |
+ * | | |
+ * | | |
+ * | | |
+ * |---------------------------------------------------------------------|
+ *
+ * DWT transforms are generally applied by splitting the image in two vertically
+ * and applying a low pass transform on the left part and a corresponding high
+ * pass transform on the right hand side. This is known as the horizontal filter
+ * stage.
+ * After that, the same operation is performed except the image is divided
+ * horizontally, with the high pass on the lower and the low pass on the higher
+ * side.
+ * Therefore, you're left with 4 subdivisions - known as low-low, low-high,
+ * high-low and high-high. They're referred to as orientations in the decoder
+ * and encoder.
+ *
+ * The LL (low-low) area contains the original image downsampled by the amount
+ * of levels. The rest of the areas can be thought as the details needed
+ * to restore the image perfectly to its original size.
+ */
+
+
+static int dwt_plane(AVCodecContext *avctx, void *arg)
+{
+ TransformArgs *transform_dat = arg;
+ DiracEncContext *s = transform_dat->ctx;
+ const void *frame_data = transform_dat->idata;
+ const int linesize = transform_dat->istride;
+ const int field = transform_dat->field;
+ const Plane *p = transform_dat->plane;
+ DiracTransforms *t = &transform_dat->t;
+ dwtcoef *buf = p->coef_buf;
+ const int idx = s->wavelet_idx;
+ const int skip = 1 + s->interlaced;
+
+ int x, y, level, offset, pix_stride = linesize >> (s->bpp - 1);
+
+ if (field == 1) {
+ offset = 0;
+ pix_stride <<= 1;
+ } else if (field == 2) {
+ offset = pix_stride;
+ pix_stride <<= 1;
+ } else {
+ offset = 0;
+ }
+
+ if (s->bpp == 1) {
+ const uint8_t *pix = (const uint8_t *)frame_data + offset;
+ for (y = 0; y < p->height*skip; y+=skip) {
+ for (x = 0; x < p->width; x++) {
+ buf[x] = pix[x] - s->diff_offset;
+ }
+ buf += p->coef_stride;
+ pix += pix_stride;
+ }
+ } else {
+ const uint16_t *pix = (const uint16_t *)frame_data + offset;
+ for (y = 0; y < p->height*skip; y+=skip) {
+ for (x = 0; x < p->width; x++) {
+ buf[x] = pix[x] - s->diff_offset;
+ }
+ buf += p->coef_stride;
+ pix += pix_stride;
+ }
+ }
+
+ memset(buf, 0, (p->coef_stride*p->dwt_height - p->height*p->width)*sizeof(dwtcoef));
+
+ for (level = s->wavelet_depth-1; level >= 0; level--) {
+ const SubBand *b = &p->band[level][0];
+ t->dirac_subband_dwt[idx](t, p->coef_buf, p->coef_stride,
+ b->width, b->height);
+ }
+
+ return 0;
+}
+
+static void encode_frame(DiracEncContext *s, const AVFrame *frame,
+ const char *aux_data, int field)
+{
+ int i;
+
+ /* Sequence header */
+ encode_parse_info(s, DIRAC_PCODE_SEQ_HEADER);
+ encode_seq_header(s);
+
+ /* Encoder version */
+ if (aux_data) {
+ encode_parse_info(s, DIRAC_PCODE_AUX);
+ avpriv_put_string(&s->pb, aux_data, 1);
+ }
+
+ /* Picture header */
+ encode_parse_info(s, DIRAC_PCODE_PICTURE_HQ);
+ encode_picture_start(s);
+
+ for (i = 0; i < 3; i++) {
+ s->transform_args[i].ctx = s;
+ s->transform_args[i].field = field;
+ s->transform_args[i].plane = &s->plane[i];
+ s->transform_args[i].idata = frame->data[i];
+ s->transform_args[i].istride = frame->linesize[i];
+ }
+
+ /* Do a DWT transform */
+ s->avctx->execute(s->avctx, dwt_plane, s->transform_args, NULL, 3,
+ sizeof(TransformArgs));
+
+ /* Init planes and encode slices */
+ encode_slices(s);
+
+ /* End sequence */
+ encode_parse_info(s, DIRAC_PCODE_END_SEQ);
+}
+
+static av_cold int dirac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
+ const AVFrame *frame, int *got_packet_ptr)
+{
+ int ret;
+ int max_frame_bytes, sig_size = 256;
+ DiracEncContext *s = avctx->priv_data;
+ const char aux_data[] = "FFmpeg version "FFMPEG_VERSION;
+ const int aux_data_size = sizeof(aux_data);
+ const int header_size = 100 + aux_data_size;
+ int64_t r_bitrate = avctx->bit_rate >> (s->interlaced);
+
+ s->avctx = avctx;
+ s->size_scaler = 1;
+ s->prefix_bytes = 0;
+ s->last_parse_code = 0;
+ s->next_parse_offset = 0;
+
+ /* Rate control */
+ max_frame_bytes = (av_rescale(r_bitrate, s->avctx->time_base.num,
+ s->avctx->time_base.den) >> 3) - header_size;
+
+ /* Find an appropriate size scaler */
+ while (sig_size > 255) {
+ s->slice_max_bytes = FFALIGN(av_rescale(max_frame_bytes, 1, s->num_x*s->num_y), s->size_scaler) + 4;
+ sig_size = s->slice_max_bytes/s->size_scaler; /* Signalled slize size */
+ s->size_scaler <<= 1;
+ }
+
+ ret = ff_alloc_packet2(avctx, avpkt, max_frame_bytes*2, 0);
+ if (ret < 0) {
+ av_log(avctx, AV_LOG_ERROR, "Error getting output packet.\n");
+ return ret;
+ }
+
+ /* Init PB */
+ init_put_bits(&s->pb, avpkt->data, avpkt->size);
+
+ encode_frame(s, frame, aux_data, s->interlaced);
+ if (s->interlaced)
+ encode_frame(s, frame, NULL, 2);
+
+ flush_put_bits(&s->pb);
+ avpkt->size = put_bits_count(&s->pb) >> 3;
+
+ *got_packet_ptr = 1;
+
+ return 0;
+}
+
+static av_cold int dirac_encode_end(AVCodecContext *avctx)
+{
+ int i;
+ DiracEncContext *s = avctx->priv_data;
+
+ for (i = 0; i < 3; i++) {
+ ff_diracenc_deinit_transforms(&s->transform_args[i].t);
+ av_freep(&s->plane[i].coef_buf);
+ }
+
+ av_freep(&s->slice_args);
+ av_freep(&s->coef_lut_len);
+ av_freep(&s->coef_lut_val);
+
+ return 0;
+}
+
+
+static av_cold int dirac_encode_init(AVCodecContext *avctx)
+{
+ Plane *p;
+ SubBand *b;
+ int i, j, level, o, shift;
+ DiracEncContext *s = avctx->priv_data;
+
+ s->picture_number = 0;
+
+ /* Total allowed quantization range */
+ s->q_ceil = 49;
+
+ s->ver.major = 2;
+ s->ver.minor = 0;
+ s->profile = 3;
+ s->level = 3;
+
+ s->base_vf = -1;
+ s->strict_compliance = 1;
+
+ /* Mark unknown as progressive */
+ s->interlaced = !((avctx->field_order == AV_FIELD_UNKNOWN) ||
+ (avctx->field_order == AV_FIELD_PROGRESSIVE));
+
+ if (avctx->pix_fmt == AV_PIX_FMT_YUV422P10) {
+ if (avctx->width == 1280 && avctx->height == 720) {
+ s->level = 3;
+ if (avctx->time_base.num == 1001 && avctx->time_base.den == 60000)
+ s->base_vf = 9;
+ if (avctx->time_base.num == 1 && avctx->time_base.den == 50)
+ s->base_vf = 10;
+ } else if (avctx->width == 1920 && avctx->height == 1080) {
+ s->level = 3;
+ if (s->interlaced) {
+ if (avctx->time_base.num == 1001 && avctx->time_base.den == 30000)
+ s->base_vf = 11;
+ if (avctx->time_base.num == 1 && avctx->time_base.den == 50)
+ s->base_vf = 12;
+ } else {
+ if (avctx->time_base.num == 1001 && avctx->time_base.den == 60000)
+ s->base_vf = 13;
+ if (avctx->time_base.num == 1 && avctx->time_base.den == 50)
+ s->base_vf = 14;
+ if (avctx->time_base.num == 1001 && avctx->time_base.den == 24000)
+ s->base_vf = 21;
+ }
+ } else if (avctx->width == 3840 && avctx->height == 2160) {
+ s->level = 6;
+ if (avctx->time_base.num == 1001 && avctx->time_base.den == 60000)
+ s->base_vf = 17;
+ if (avctx->time_base.num == 1 && avctx->time_base.den == 50)
+ s->base_vf = 18;
+ }
+ }
+
+ if (s->interlaced && s->base_vf <= 0) {
+ av_log(avctx, AV_LOG_ERROR, "Interlacing not supported with non standard formats!\n");
+ return AVERROR_UNKNOWN;
+ }
+
+ if (s->interlaced)
+ av_log(avctx, AV_LOG_WARNING, "Interlacing enabled!\n");
+
+ if ((s->slice_width & (s->slice_width - 1)) ||
+ (s->slice_height & (s->slice_height - 1))) {
+ av_log(avctx, AV_LOG_ERROR, "Slice size is not a power of two!\n");
+ return AVERROR_UNKNOWN;
+ }
+
+ if ((s->slice_width > avctx->width) ||
+ (s->slice_height > avctx->height)) {
+ av_log(avctx, AV_LOG_ERROR, "Slice size is bigger than the image!\n");
+ return AVERROR_UNKNOWN;
+ }
+
+ if (s->base_vf <= 0) {
+ if (avctx->strict_std_compliance <= FF_COMPLIANCE_UNOFFICIAL) {
+ s->strict_compliance = s->base_vf = 0;
+ av_log(avctx, AV_LOG_WARNING, "Disabling strict compliance\n");
+ } else {
+ av_log(avctx, AV_LOG_WARNING, "Given format does not strictly comply with "
+ "the specifications, please add a -strict -1 flag to use it\n");
+ return AVERROR_UNKNOWN;
+ }
+ } else {
+ av_log(avctx, AV_LOG_INFO, "Selected base video format = %i\n", s->base_vf);
+ }
+
+ avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
+
+ /* Planes initialization */
+ for (i = 0; i < 3; i++) {
+ int w, h;
+ p = &s->plane[i];
+ p->width = avctx->width >> (i ? s->chroma_x_shift : 0);
+ p->height = avctx->height >> (i ? s->chroma_y_shift : 0);
+ if (s->interlaced)
+ p->height >>= 1;
+ p->dwt_width = w = FFALIGN(p->width, (1 << s->wavelet_depth));
+ p->dwt_height = h = FFALIGN(p->height, (1 << s->wavelet_depth));
+ p->coef_stride = FFALIGN(p->dwt_width, 32);
+ p->coef_buf = av_malloc(p->coef_stride*p->dwt_height*sizeof(dwtcoef));
+ if (!p->coef_buf)
+ goto alloc_fail;
+ for (level = s->wavelet_depth-1; level >= 0; level--) {
+ w = w >> 1;
+ h = h >> 1;
+ for (o = 0; o < 4; o++) {
+ b = &p->band[level][o];
+ b->width = w;
+ b->height = h;
+ b->stride = p->coef_stride;
+ shift = (o > 1)*b->height*b->stride + (o & 1)*b->width;
+ b->buf = p->coef_buf + shift;
+ }
+ }
+
+ /* DWT init */
+ if (ff_diracenc_init_transforms(&s->transform_args[i].t,
+ s->plane[0].coef_stride,
+ s->plane[0].dwt_height))
+ goto alloc_fail;
+ }
+
+ /* Slices */
+ s->num_x = s->plane[0].dwt_width/s->slice_width;
+ s->num_y = s->plane[0].dwt_height/s->slice_height;
+
+ s->slice_args = av_malloc(s->num_x*s->num_y*sizeof(SliceArgs));
+ if (!s->slice_args)
+ goto alloc_fail;
+
+ /* Lookup tables */
+ s->coef_lut_len = av_malloc(2*COEF_LUT_TAB*s->q_ceil*sizeof(*s->coef_lut_len));
+ if (!s->coef_lut_len)
+ goto alloc_fail;
+
+ s->coef_lut_val = av_malloc(2*COEF_LUT_TAB*s->q_ceil*sizeof(*s->coef_lut_val));
+ if (!s->coef_lut_val)
+ goto alloc_fail;
+
+ for (i = 0; i < s->q_ceil; i++) {
+ for (j = -COEF_LUT_TAB; j < COEF_LUT_TAB; j++) {
+ uint8_t *len_lut = &s->coef_lut_len[2*i*COEF_LUT_TAB + COEF_LUT_TAB];
+ uint32_t *val_lut = &s->coef_lut_val[2*i*COEF_LUT_TAB + COEF_LUT_TAB];
+ coeff_quantize_get(j, qscale_tab[i], qoffset_intra_tab[i] + 2, &len_lut[j], &val_lut[j]);
+ }
+ }
+
+ return 0;
+
+alloc_fail:
+ dirac_encode_end(avctx);
+ av_log(avctx, AV_LOG_ERROR, "Unable to allocate memory!\n");
+ return AVERROR(ENOMEM);
+}
+
+#define DIRACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_VIDEO_PARAM
+static const AVOption diracenc_options[] = {
+ {"tolerance", "Max undershoot in percent", offsetof(DiracEncContext, tolerance), AV_OPT_TYPE_DOUBLE, {.dbl = 10.0f}, 0.0f, 45.0f, DIRACENC_FLAGS, "tolerance"},
+ {"slice_width", "Slice width", offsetof(DiracEncContext, slice_width), AV_OPT_TYPE_INT, {.i64 = 128}, 32, 1024, DIRACENC_FLAGS, "slice_width"},
+ {"slice_height", "Slice height", offsetof(DiracEncContext, slice_height), AV_OPT_TYPE_INT, {.i64 = 64}, 8, 1024, DIRACENC_FLAGS, "slice_height"},
+ {"wavelet_depth", "Transform depth", offsetof(DiracEncContext, wavelet_depth), AV_OPT_TYPE_INT, {.i64 = 5}, 1, 5, DIRACENC_FLAGS, "wavelet_depth"},
+ {"wavelet_type", "Transform type", offsetof(DiracEncContext, wavelet_idx), AV_OPT_TYPE_INT, {.i64 = DIRAC_TRANSFORM_9_7}, 0, DIRAC_TRANSFORMS_NB, DIRACENC_FLAGS, "wavelet_idx"},
+ {"9_7", "Deslauriers-Dubuc (9,7)", 0, AV_OPT_TYPE_CONST, {.i64 = DIRAC_TRANSFORM_9_7}, INT_MIN, INT_MAX, DIRACENC_FLAGS, "wavelet_idx"},
+ {"5_3", "LeGall (5,3)", 0, AV_OPT_TYPE_CONST, {.i64 = DIRAC_TRANSFORM_5_3}, INT_MIN, INT_MAX, DIRACENC_FLAGS, "wavelet_idx"},
+ {"quant_matrix", "Custom quantization matrix", offsetof(DiracEncContext, quant_matrix), AV_OPT_TYPE_INT, {.i64 = DIRAC_QM_DEF}, 0, DIRAC_QM_NB, DIRACENC_FLAGS, "quant_matrix"},
+ {"default", "Default from the specifications", 0, AV_OPT_TYPE_CONST, {.i64 = DIRAC_QM_DEF}, INT_MIN, INT_MAX, DIRACENC_FLAGS, "quant_matrix"},
+ {"color", "Prevents low bitrate discoloration", 0, AV_OPT_TYPE_CONST, {.i64 = DIRAC_QM_COL}, INT_MIN, INT_MAX, DIRACENC_FLAGS, "quant_matrix"},
+ {"flat", "Optimize for PSNR", 0, AV_OPT_TYPE_CONST, {.i64 = DIRAC_QM_FLAT}, INT_MIN, INT_MAX, DIRACENC_FLAGS, "quant_matrix"},
+ {NULL}
+};
+
+static const AVClass diracenc_class = {
+ .class_name = "BBC Dirac VC-2 encoder",
+ .category = AV_CLASS_CATEGORY_ENCODER,
+ .option = diracenc_options,
+ .item_name = av_default_item_name,
+ .version = LIBAVUTIL_VERSION_INT
+};
+
+static const AVCodecDefault diracenc_defaults[] = {
+ { "b", "600000000" },
+ { NULL },
+};
+
+static const enum AVPixelFormat allowed_pix_fmts[] = {
+ AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P,
+ AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
+ AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12,
+ AV_PIX_FMT_NONE
+};
+
+AVCodec ff_dirac_encoder = {
+ .name = "dirac",
+ .long_name = NULL_IF_CONFIG_SMALL("BBC Dirac VC-2"),
+ .type = AVMEDIA_TYPE_VIDEO,
+ .id = AV_CODEC_ID_DIRAC,
+ .priv_data_size = sizeof(DiracEncContext),
+ .init = dirac_encode_init,
+ .close = dirac_encode_end,
+ .capabilities = AV_CODEC_CAP_SLICE_THREADS,
+ .encode2 = dirac_encode_frame,
+ .priv_class = &diracenc_class,
+ .defaults = diracenc_defaults,
+ .pix_fmts = allowed_pix_fmts
+};
diff --git a/libavcodec/diracenc_transforms.c b/libavcodec/diracenc_transforms.c
new file mode 100644
index 0000000..1646c30
--- /dev/null
+++ b/libavcodec/diracenc_transforms.c
@@ -0,0 +1,246 @@
+/*
+ * Copyright (C) 2007 Marco Gerards <marco at gnu.org>
+ * Copyright (C) 2016 Open Broadcast Systems Ltd.
+ * Author (C) 2016 Rostislav Pehlivanov <atomnuker at gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "diracenc_transforms.h"
+#include "libavutil/avassert.h"
+
+/* Since the transforms spit out interleaved coefficients, this function
+ * rearranges the coefficients into the more traditional subdivision,
+ * making it easier to encode and perform another level. */
+static av_always_inline void deinterleave(dwtcoef *linell, int stride,
+ int width, int height, dwtcoef *synthl)
+{
+ int x, y;
+ int synthw = width << 1;
+ dwtcoef *linehl = linell + width;
+ dwtcoef *linelh = linell + height*stride;
+ dwtcoef *linehh = linelh + width;
+
+ /* Deinterleave the coefficients. */
+ for (y = 0; y < height; y++) {
+ for (x = 0; x < width; x++) {
+ linell[x] = synthl[(x << 1)];
+ linehl[x] = synthl[(x << 1) + 1];
+ linelh[x] = synthl[(x << 1) + synthw];
+ linehh[x] = synthl[(x << 1) + synthw + 1];
+ }
+ synthl += synthw << 1;
+ linell += stride;
+ linelh += stride;
+ linehl += stride;
+ linehh += stride;
+ }
+}
+
+static void dirac_subband_dwt_97(DiracTransforms *t,
+ dwtcoef *data, int stride,
+ int width, int height)
+{
+ int x, y;
+ dwtcoef *datal = data, *synth = t->buffer, *synthl = synth;
+ const int synth_width = width << 1;
+ const int synth_height = height << 1;
+
+ /*
+ * Shift in one bit that is used for additional precision and copy
+ * the data to the buffer.
+ */
+ for (y = 0; y < synth_height; y++) {
+ for (x = 0; x < synth_width; x++)
+ synthl[x] = datal[x] << 1;
+ synthl += synth_width;
+ datal += stride;
+ }
+
+ /* Horizontal synthesis. */
+ synthl = synth;
+ for (y = 0; y < synth_height; y++) {
+ /* Lifting stage 2. */
+ synthl[1] -= (8*synthl[0] + 9*synthl[2] - synthl[4] + 8) >> 4;
+ for (x = 1; x < width - 2; x++)
+ synthl[2*x + 1] -= (9*synthl[2*x] + 9*synthl[2*x + 2] - synthl[2*x + 4] -
+ synthl[2 * x - 2] + 8) >> 4;
+ synthl[synth_width - 1] -= (17*synthl[synth_width - 2] -
+ synthl[synth_width - 4] + 8) >> 4;
+ synthl[synth_width - 3] -= (8*synthl[synth_width - 2] +
+ 9*synthl[synth_width - 4] -
+ synthl[synth_width - 6] + 8) >> 4;
+ /* Lifting stage 1. */
+ synthl[0] += (synthl[1] + synthl[1] + 2) >> 2;
+ for (x = 1; x < width - 1; x++)
+ synthl[2*x] += (synthl[2*x - 1] + synthl[2*x + 1] + 2) >> 2;
+
+ synthl[synth_width - 2] += (synthl[synth_width - 3] +
+ synthl[synth_width - 1] + 2) >> 2;
+ synthl += synth_width;
+ }
+
+ /* Vertical synthesis: Lifting stage 2. */
+ synthl = synth + synth_width;
+ for (x = 0; x < synth_width; x++)
+ synthl[x] -= (8*synthl[x - synth_width] + 9*synthl[x + synth_width] -
+ synthl[x + 3 * synth_width] + 8) >> 4;
+
+ synthl = synth + (synth_width << 1);
+ for (y = 1; y < height - 2; y++) {
+ for (x = 0; x < synth_width; x++)
+ synthl[x + synth_width] -= (9*synthl[x] +
+ 9*synthl[x + 2 * synth_width] -
+ synthl[x - 2 * synth_width] -
+ synthl[x + 4 * synth_width] + 8) >> 4;
+ synthl += synth_width << 1;
+ }
+
+ synthl = synth + (synth_height - 1) * synth_width;
+ for (x = 0; x < synth_width; x++) {
+ synthl[x] -= (17*synthl[x - synth_width] -
+ synthl[x - 3*synth_width] + 8) >> 4;
+ synthl[x - 2*synth_width] -= (9*synthl[x - 3*synth_width] +
+ 8*synthl[x - 1*synth_width] -
+ synthl[x - 5*synth_width] + 8) >> 4;
+ }
+
+ /* Vertical synthesis: Lifting stage 1. */
+ synthl = synth;
+ for (x = 0; x < synth_width; x++) {
+ synthl[x] += (synthl[x + synth_width] + synthl[x + synth_width] + 2) >> 2;
+ }
+
+ synthl = synth + (synth_width << 1);
+ for (y = 1; y < height - 1; y++) {
+ for (x = 0; x < synth_width; x++)
+ synthl[x] += (synthl[x - synth_width] + synthl[x + synth_width] + 2) >> 2;
+ synthl += synth_width << 1;
+ }
+
+ synthl = synth + (synth_height - 2) * synth_width;
+ for (x = 0; x < synth_width; x++)
+ synthl[x] += (synthl[x - synth_width] + synthl[x + synth_width] + 2) >> 2;
+
+ deinterleave(data, stride, width, height, synth);
+}
+
+static void dirac_subband_dwt_53(DiracTransforms *t,
+ dwtcoef *data, int stride,
+ int width, int height)
+{
+ int x, y;
+ dwtcoef *synth = t->buffer, *synthl = synth, *datal = data;
+ const int synth_width = width << 1;
+ const int synth_height = height << 1;
+
+ /*
+ * Shift in one bit that is used for additional precision and copy
+ * the data to the buffer.
+ */
+ for (y = 0; y < synth_height; y++) {
+ for (x = 0; x < synth_width; x++)
+ synthl[x] = datal[x] << 1;
+ synthl += synth_width;
+ datal += stride;
+ }
+
+ /* Horizontal synthesis. */
+ synthl = synth;
+ for (y = 0; y < synth_height; y++) {
+ /* Lifting stage 2. */
+ for (x = 0; x < width - 1; x++)
+ synthl[2 * x + 1] -= (synthl[2 * x] + synthl[2 * x + 2] + 1) >> 1;
+
+ synthl[synth_width - 1] -= (2*synthl[synth_width - 2] + 1) >> 1;
+
+ /* Lifting stage 1. */
+ synthl[0] += (2*synthl[1] + 2) >> 2;
+ for (x = 1; x < width - 1; x++)
+ synthl[2 * x] += (synthl[2 * x - 1] + synthl[2 * x + 1] + 2) >> 2;
+
+ synthl[synth_width - 2] += (synthl[synth_width - 3] + synthl[synth_width - 1] + 2) >> 2;
+
+ synthl += synth_width;
+ }
+
+ /* Vertical synthesis: Lifting stage 2. */
+ synthl = synth + synth_width;
+ for (x = 0; x < synth_width; x++)
+ synthl[x] -= (synthl[x - synth_width] + synthl[x + synth_width] + 1) >> 1;
+
+ synthl = synth + (synth_width << 1);
+ for (y = 1; y < height - 1; y++) {
+ for (x = 0; x < synth_width; x++) {
+ synthl[x + synth_width] -= (synthl[x] + synthl[x + synth_width * 2] + 1) >> 1;
+ }
+ synthl += (synth_width << 1);
+ }
+
+ synthl = synth + (synth_height - 1) * synth_width;
+ for (x = 0; x < synth_width; x++)
+ synthl[x] -= (2*synthl[x - synth_width] + 1) >> 1;
+
+ /* Vertical synthesis: Lifting stage 1. */
+ synthl = synth;
+ for (x = 0; x < synth_width; x++) {
+ synthl[x] += (2*synthl[synth_width + x] + 2) >> 2;
+ }
+
+ synthl = synth + (synth_width << 1);
+ for (y = 1; y < height - 1; y++) {
+ for (x = 0; x < synth_width; x++) {
+ synthl[x] += (synthl[x + synth_width] + synthl[x - synth_width] + 2) >> 2;
+ }
+ synthl += (synth_width << 1);
+ }
+
+ synthl = synth + (synth_height - 2)*synth_width;
+ for (x = 0; x < synth_width; x++)
+ synthl[x] += (synthl[x - synth_width] + synthl[x + synth_width] + 2) >> 2;
+
+
+ deinterleave(data, stride, width, height, synth);
+}
+
+static void dirac_subband_noop(DiracTransforms *t, dwtcoef *data, int stride,
+ int width, int height)
+{
+ av_assert0(0);
+}
+
+int ff_diracenc_init_transforms(DiracTransforms *s, int p_width, int p_height)
+{
+ s->dirac_subband_dwt[DIRAC_TRANSFORM_9_7] = dirac_subband_dwt_97;
+ s->dirac_subband_dwt[DIRAC_TRANSFORM_5_3] = dirac_subband_dwt_53;
+ s->dirac_subband_dwt[DIRAC_TRANSFORM_13_7] = dirac_subband_noop;
+ s->dirac_subband_dwt[DIRAC_TRANSFORM_HAAR] = dirac_subband_noop;
+ s->dirac_subband_dwt[DIRAC_TRANSFORM_HAAR_S] = dirac_subband_noop;
+ s->dirac_subband_dwt[DIRAC_TRANSFORM_FIDEL] = dirac_subband_noop;
+ s->dirac_subband_dwt[DIRAC_TRANSFORM_9_7_I] = dirac_subband_noop;
+
+ s->buffer = av_malloc(2*p_width*p_height*sizeof(dwtcoef));
+ if (!s->buffer)
+ return 1;
+
+ return 0;
+}
+
+void ff_diracenc_deinit_transforms(DiracTransforms *s)
+{
+ av_freep(&s->buffer);
+}
diff --git a/libavcodec/diracenc_transforms.h b/libavcodec/diracenc_transforms.h
new file mode 100644
index 0000000..3b723b2
--- /dev/null
+++ b/libavcodec/diracenc_transforms.h
@@ -0,0 +1,53 @@
+/*
+ * Copyright (C) 2016 Open Broadcast Systems Ltd.
+ * Author (C) 2016 Rostislav Pehlivanov <atomnuker at gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_DIRAC_WAVELET_H
+#define AVCODEC_DIRAC_WAVELET_H
+
+#include <stdint.h>
+
+typedef int16_t dwtcoef;
+typedef int32_t qcoef; /* Quantization needs more precision */
+
+/* Only Deslauriers-Dubuc (9,7) and LeGall (5,3) supported! */
+
+enum DiracTransformType {
+ DIRAC_TRANSFORM_9_7 = 0, /* Deslauriers-Dubuc (9,7) */
+ DIRAC_TRANSFORM_5_3 = 1, /* LeGall (5,3) */
+ DIRAC_TRANSFORM_13_7 = 2, /* Deslauriers-Dubuc (13,7) */
+ DIRAC_TRANSFORM_HAAR = 3, /* Haar without shift */
+ DIRAC_TRANSFORM_HAAR_S = 4, /* Haar with 1 shift/lvl */
+ DIRAC_TRANSFORM_FIDEL = 5, /* Fidelity filter */
+ DIRAC_TRANSFORM_9_7_I = 6, /* Daubechies (9,7) */
+ DIRAC_TRANSFORMS_NB
+};
+
+typedef struct DiracTransforms {
+ dwtcoef *buffer;
+ void (*dirac_subband_dwt[DIRAC_TRANSFORMS_NB])(struct DiracTransforms *t,
+ dwtcoef *data, int stride,
+ int width, int height);
+} DiracTransforms;
+
+int ff_diracenc_init_transforms(DiracTransforms *t, int p_width, int p_height);
+void ff_diracenc_deinit_transforms(DiracTransforms *t);
+
+#endif /* AVCODEC_DIRACWAVELET_H */
--
2.7.0
More information about the ffmpeg-devel
mailing list