*/
#include "libavutil/opt.h"
+#include "libavutil/pixdesc.h"
#include "avcodec.h"
-#include "dsputil.h"
+#include "fdctdsp.h"
#include "put_bits.h"
#include "bytestream.h"
#include "internal.h"
-#include "proresdsp.h"
#include "proresdata.h"
#define CFACTOR_Y422 2
#define MAX_MBS_PER_SLICE 8
-#define MAX_PLANES 3 // should be increased to 4 when there's AV_PIX_FMT_YUV444AP10
+#define MAX_PLANES 4
enum {
PRORES_PROFILE_PROXY = 0,
PRORES_PROFILE_LT,
PRORES_PROFILE_STANDARD,
PRORES_PROFILE_HQ,
+ PRORES_PROFILE_4444,
};
enum {
int max_quant;
int br_tab[NUM_MB_LIMITS];
int quant;
-} prores_profile_info[4] = {
+} prores_profile_info[5] = {
{
.full_name = "proxy",
.tag = MKTAG('a', 'p', 'c', 'o'),
.max_quant = 6,
.br_tab = { 1566, 1216, 1070, 950 },
.quant = QUANT_MAT_HQ,
+ },
+ {
+ .full_name = "4444",
+ .tag = MKTAG('a', 'p', '4', 'h'),
+ .min_quant = 1,
+ .max_quant = 6,
+ .br_tab = { 2350, 1828, 1600, 1425 },
+ .quant = QUANT_MAT_HQ,
}
-// for 4444 profile bitrate numbers are { 2350, 1828, 1600, 1425 }
};
#define TRELLIS_WIDTH 16
int16_t quants[MAX_STORED_Q][64];
int16_t custom_q[64];
const uint8_t *quant_mat;
+ const uint8_t *scantable;
- ProresDSPContext dsp;
- ScanTable scantable;
+ void (*fdct)(FDCTDSPContext *fdsp, const uint16_t *src,
+ int linesize, int16_t *block);
+ FDCTDSPContext fdsp;
int mb_width, mb_height;
int mbs_per_slice;
int num_planes;
int bits_per_mb;
int force_quant;
+ int alpha_bits;
+ int warn;
char *vendor;
int quant_sel;
mb_width * sizeof(*emu_buf));
}
if (!is_chroma) {
- ctx->dsp.fdct(esrc, elinesize, blocks);
+ ctx->fdct(&ctx->fdsp, esrc, elinesize, blocks);
blocks += 64;
if (blocks_per_mb > 2) {
- ctx->dsp.fdct(esrc + 8, elinesize, blocks);
+ ctx->fdct(&ctx->fdsp, esrc + 8, elinesize, blocks);
blocks += 64;
}
- ctx->dsp.fdct(esrc + elinesize * 4, elinesize, blocks);
+ ctx->fdct(&ctx->fdsp, esrc + elinesize * 4, elinesize, blocks);
blocks += 64;
if (blocks_per_mb > 2) {
- ctx->dsp.fdct(esrc + elinesize * 4 + 8, elinesize, blocks);
+ ctx->fdct(&ctx->fdsp, esrc + elinesize * 4 + 8, elinesize, blocks);
blocks += 64;
}
} else {
- ctx->dsp.fdct(esrc, elinesize, blocks);
+ ctx->fdct(&ctx->fdsp, esrc, elinesize, blocks);
blocks += 64;
- ctx->dsp.fdct(esrc + elinesize * 4, elinesize, blocks);
+ ctx->fdct(&ctx->fdsp, esrc + elinesize * 4, elinesize, blocks);
blocks += 64;
if (blocks_per_mb > 2) {
- ctx->dsp.fdct(esrc + 8, elinesize, blocks);
+ ctx->fdct(&ctx->fdsp, esrc + 8, elinesize, blocks);
blocks += 64;
- ctx->dsp.fdct(esrc + elinesize * 4 + 8, elinesize, blocks);
+ ctx->fdct(&ctx->fdsp, esrc + elinesize * 4 + 8, elinesize, blocks);
blocks += 64;
}
}
}
}
+static void get_alpha_data(ProresContext *ctx, const uint16_t *src,
+ int linesize, int x, int y, int w, int h,
+ int16_t *blocks, int mbs_per_slice, int abits)
+{
+ const int slice_width = 16 * mbs_per_slice;
+ int i, j, copy_w, copy_h;
+
+ copy_w = FFMIN(w - x, slice_width);
+ copy_h = FFMIN(h - y, 16);
+ for (i = 0; i < copy_h; i++) {
+ memcpy(blocks, src, copy_w * sizeof(*src));
+ if (abits == 8)
+ for (j = 0; j < copy_w; j++)
+ blocks[j] >>= 2;
+ else
+ for (j = 0; j < copy_w; j++)
+ blocks[j] = (blocks[j] << 6) | (blocks[j] >> 4);
+ for (j = copy_w; j < slice_width; j++)
+ blocks[j] = blocks[copy_w - 1];
+ blocks += slice_width;
+ src += linesize >> 1;
+ }
+ for (; i < 16; i++) {
+ memcpy(blocks, blocks - slice_width, slice_width * sizeof(*blocks));
+ blocks += slice_width;
+ }
+}
+
/**
* Write an unsigned rice/exp golomb codeword.
*/
encode_dcs(pb, blocks, blocks_per_slice, qmat[0]);
encode_acs(pb, blocks, blocks_per_slice, plane_size_factor,
- ctx->scantable.permutated, qmat);
+ ctx->scantable, qmat);
flush_put_bits(pb);
return (put_bits_count(pb) - saved_pos) >> 3;
}
+static void put_alpha_diff(PutBitContext *pb, int cur, int prev, int abits)
+{
+ const int mask = (1 << abits) - 1;
+ const int dbits = (abits == 8) ? 4 : 7;
+ const int dsize = 1 << dbits - 1;
+ int diff = cur - prev;
+
+ diff &= mask;
+ if (diff >= (1 << abits) - dsize)
+ diff -= 1 << abits;
+ if (diff < -dsize || diff > dsize || !diff) {
+ put_bits(pb, 1, 1);
+ put_bits(pb, abits, diff);
+ } else {
+ put_bits(pb, 1, 0);
+ put_bits(pb, dbits - 1, FFABS(diff) - 1);
+ put_bits(pb, 1, diff < 0);
+ }
+}
+
+static void put_alpha_run(PutBitContext *pb, int run)
+{
+ if (run) {
+ put_bits(pb, 1, 0);
+ if (run < 0x10)
+ put_bits(pb, 4, run);
+ else
+ put_bits(pb, 15, run);
+ } else {
+ put_bits(pb, 1, 1);
+ }
+}
+
+// todo alpha quantisation for high quants
+static int encode_alpha_plane(ProresContext *ctx, PutBitContext *pb,
+ int mbs_per_slice, uint16_t *blocks,
+ int quant)
+{
+ const int abits = ctx->alpha_bits;
+ const int mask = (1 << abits) - 1;
+ const int num_coeffs = mbs_per_slice * 256;
+ int saved_pos = put_bits_count(pb);
+ int prev = mask, cur;
+ int idx = 0;
+ int run = 0;
+
+ cur = blocks[idx++];
+ put_alpha_diff(pb, cur, prev, abits);
+ prev = cur;
+ do {
+ cur = blocks[idx++];
+ if (cur != prev) {
+ put_alpha_run (pb, run);
+ put_alpha_diff(pb, cur, prev, abits);
+ prev = cur;
+ run = 0;
+ } else {
+ run++;
+ }
+ } while (idx < num_coeffs);
+ if (run)
+ put_alpha_run(pb, run);
+ flush_put_bits(pb);
+ return (put_bits_count(pb) - saved_pos) >> 3;
+}
+
static int encode_slice(AVCodecContext *avctx, const AVFrame *pic,
PutBitContext *pb,
int sizes[4], int x, int y, int quant,
src = (const uint16_t*)(pic->data[i] + yp * linesize +
line_add * pic->linesize[i]) + xp;
- get_slice_data(ctx, src, linesize, xp, yp,
- pwidth, avctx->height / ctx->pictures_per_frame,
- ctx->blocks[0], ctx->emu_buf,
- mbs_per_slice, num_cblocks, is_chroma);
- sizes[i] = encode_slice_plane(ctx, pb, src, linesize,
- mbs_per_slice, ctx->blocks[0],
- num_cblocks, plane_factor,
- qmat);
+ if (i < 3) {
+ get_slice_data(ctx, src, linesize, xp, yp,
+ pwidth, avctx->height / ctx->pictures_per_frame,
+ ctx->blocks[0], ctx->emu_buf,
+ mbs_per_slice, num_cblocks, is_chroma);
+ sizes[i] = encode_slice_plane(ctx, pb, src, linesize,
+ mbs_per_slice, ctx->blocks[0],
+ num_cblocks, plane_factor,
+ qmat);
+ } else {
+ get_alpha_data(ctx, src, linesize, xp, yp,
+ pwidth, avctx->height / ctx->pictures_per_frame,
+ ctx->blocks[0], mbs_per_slice, ctx->alpha_bits);
+ sizes[i] = encode_alpha_plane(ctx, pb, mbs_per_slice,
+ ctx->blocks[0], quant);
+ }
total_size += sizes[i];
+ if (put_bits_left(pb) < 0) {
+ av_log(avctx, AV_LOG_ERROR,
+ "Underestimated required buffer size.\n");
+ return AVERROR_BUG;
+ }
}
return total_size;
}
bits = estimate_dcs(error, td->blocks[plane], blocks_per_slice, qmat[0]);
bits += estimate_acs(error, td->blocks[plane], blocks_per_slice,
- plane_size_factor, ctx->scantable.permutated, qmat);
+ plane_size_factor, ctx->scantable, qmat);
return FFALIGN(bits, 8);
}
+static int est_alpha_diff(int cur, int prev, int abits)
+{
+ const int mask = (1 << abits) - 1;
+ const int dbits = (abits == 8) ? 4 : 7;
+ const int dsize = 1 << dbits - 1;
+ int diff = cur - prev;
+
+ diff &= mask;
+ if (diff >= (1 << abits) - dsize)
+ diff -= 1 << abits;
+ if (diff < -dsize || diff > dsize || !diff)
+ return abits + 1;
+ else
+ return dbits + 1;
+}
+
+static int estimate_alpha_plane(ProresContext *ctx, int *error,
+ const uint16_t *src, int linesize,
+ int mbs_per_slice, int quant,
+ int16_t *blocks)
+{
+ const int abits = ctx->alpha_bits;
+ const int mask = (1 << abits) - 1;
+ const int num_coeffs = mbs_per_slice * 256;
+ int prev = mask, cur;
+ int idx = 0;
+ int run = 0;
+ int bits;
+
+ *error = 0;
+ cur = blocks[idx++];
+ bits = est_alpha_diff(cur, prev, abits);
+ prev = cur;
+ do {
+ cur = blocks[idx++];
+ if (cur != prev) {
+ if (!run)
+ bits++;
+ else if (run < 0x10)
+ bits += 4;
+ else
+ bits += 15;
+ bits += est_alpha_diff(cur, prev, abits);
+ prev = cur;
+ run = 0;
+ } else {
+ run++;
+ }
+ } while (idx < num_coeffs);
+
+ if (run) {
+ if (run < 0x10)
+ bits += 4;
+ else
+ bits += 15;
+ }
+
+ return bits;
+}
+
static int find_slice_quant(AVCodecContext *avctx, const AVFrame *pic,
int trellis_node, int x, int y, int mbs_per_slice,
ProresThreadData *td)
src = (const uint16_t*)(pic->data[i] + yp * linesize[i] +
line_add * pic->linesize[i]) + xp;
- get_slice_data(ctx, src, linesize[i], xp, yp,
- pwidth, avctx->height / ctx->pictures_per_frame,
- td->blocks[i], td->emu_buf,
- mbs_per_slice, num_cblocks[i], is_chroma[i]);
+ if (i < 3) {
+ get_slice_data(ctx, src, linesize[i], xp, yp,
+ pwidth, avctx->height / ctx->pictures_per_frame,
+ td->blocks[i], td->emu_buf,
+ mbs_per_slice, num_cblocks[i], is_chroma[i]);
+ } else {
+ get_alpha_data(ctx, src, linesize[i], xp, yp,
+ pwidth, avctx->height / ctx->pictures_per_frame,
+ td->blocks[i], mbs_per_slice, ctx->alpha_bits);
+ }
}
for (q = min_quant; q < max_quant + 2; q++) {
for (q = min_quant; q <= max_quant; q++) {
bits = 0;
error = 0;
- for (i = 0; i < ctx->num_planes; i++) {
+ for (i = 0; i < ctx->num_planes - !!ctx->alpha_bits; i++) {
bits += estimate_slice_plane(ctx, &error, i,
src, linesize[i],
mbs_per_slice,
num_cblocks[i], plane_factor[i],
ctx->quants[q], td);
}
- if (bits > 65000 * 8) {
+ if (ctx->alpha_bits)
+ bits += estimate_alpha_plane(ctx, &error, src, linesize[3],
+ mbs_per_slice, q, td->blocks[3]);
+ if (bits > 65000 * 8)
error = SCORE_LIMIT;
- break;
- }
+
slice_bits[q] = bits;
slice_score[q] = error;
}
for (i = 0; i < 64; i++)
qmat[i] = ctx->quant_mat[i] * q;
}
- for (i = 0; i < ctx->num_planes; i++) {
+ for (i = 0; i < ctx->num_planes - !!ctx->alpha_bits; i++) {
bits += estimate_slice_plane(ctx, &error, i,
src, linesize[i],
mbs_per_slice,
num_cblocks[i], plane_factor[i],
qmat, td);
}
+ if (ctx->alpha_bits)
+ bits += estimate_alpha_plane(ctx, &error, src, linesize[3],
+ mbs_per_slice, q, td->blocks[3]);
if (bits <= ctx->bits_per_mb * mbs_per_slice)
break;
}
int sizes[4] = { 0 };
int slice_hdr_size = 2 + 2 * (ctx->num_planes - 1);
int frame_size, picture_size, slice_size;
- int pkt_size, ret;
+ int pkt_size, ret, max_slice_size = 0;
uint8_t frame_flags;
*avctx->coded_frame = *pic;
avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I;
avctx->coded_frame->key_frame = 1;
- pkt_size = ctx->frame_size_upper_bound + FF_MIN_BUFFER_SIZE;
+ pkt_size = ctx->frame_size_upper_bound;
- if ((ret = ff_alloc_packet(pkt, pkt_size)) < 0) {
+ if ((ret = ff_alloc_packet(pkt, pkt_size + FF_MIN_BUFFER_SIZE)) < 0) {
av_log(avctx, AV_LOG_ERROR, "Error getting output packet.\n");
return ret;
}
bytestream_put_byte (&buf, avctx->color_primaries);
bytestream_put_byte (&buf, avctx->color_trc);
bytestream_put_byte (&buf, avctx->colorspace);
- bytestream_put_byte (&buf, 0x40); // source format and alpha information
+ bytestream_put_byte (&buf, 0x40 | (ctx->alpha_bits >> 3));
bytestream_put_byte (&buf, 0); // reserved
if (ctx->quant_sel != QUANT_MAT_DEFAULT) {
bytestream_put_byte (&buf, 0x03); // matrix flags - both matrices are present
bytestream_put_byte(&buf, slice_hdr_size << 3);
slice_hdr = buf;
buf += slice_hdr_size - 1;
+ if (pkt_size <= buf - orig_buf + 2 * max_slice_size) {
+ uint8_t *start = pkt->data;
+ // Recompute new size according to max_slice_size
+ // and deduce delta
+ int delta = 200 + ctx->pictures_per_frame *
+ ctx->slices_per_picture * max_slice_size -
+ pkt_size;
+
+ delta = FFMAX(delta, 2 * max_slice_size);
+ ctx->frame_size_upper_bound += delta;
+
+ if (!ctx->warn) {
+ avpriv_request_sample(avctx,
+ "Packet too small: is %i,"
+ " needs %i (slice: %i). "
+ "Correct allocation",
+ pkt_size, delta, max_slice_size);
+ ctx->warn = 1;
+ }
+
+ ret = av_grow_packet(pkt, delta);
+ if (ret < 0)
+ return ret;
+
+ pkt_size += delta;
+ // restore pointers
+ orig_buf = pkt->data + (orig_buf - start);
+ buf = pkt->data + (buf - start);
+ picture_size_pos = pkt->data + (picture_size_pos - start);
+ slice_sizes = pkt->data + (slice_sizes - start);
+ slice_hdr = pkt->data + (slice_hdr - start);
+ tmp = pkt->data + (tmp - start);
+ }
init_put_bits(&pb, buf, (pkt_size - (buf - orig_buf)) * 8);
- encode_slice(avctx, pic, &pb, sizes, x, y, q, mbs_per_slice);
+ ret = encode_slice(avctx, pic, &pb, sizes, x, y, q,
+ mbs_per_slice);
+ if (ret < 0)
+ return ret;
bytestream_put_byte(&slice_hdr, q);
slice_size = slice_hdr_size + sizes[ctx->num_planes - 1];
}
bytestream_put_be16(&slice_sizes, slice_size);
buf += slice_size - slice_hdr_size;
+ if (max_slice_size < slice_size)
+ max_slice_size = slice_size;
}
}
return 0;
}
+static void prores_fdct(FDCTDSPContext *fdsp, const uint16_t *src,
+ int linesize, int16_t *block)
+{
+ int x, y;
+ const uint16_t *tsrc = src;
+
+ for (y = 0; y < 8; y++) {
+ for (x = 0; x < 8; x++)
+ block[y * 8 + x] = tsrc[x];
+ tsrc += linesize >> 1;
+ }
+ fdsp->fdct(block);
+}
+
static av_cold int encode_init(AVCodecContext *avctx)
{
ProresContext *ctx = avctx->priv_data;
int interlaced = !!(avctx->flags & CODEC_FLAG_INTERLACED_DCT);
avctx->bits_per_raw_sample = 10;
- avctx->coded_frame = avcodec_alloc_frame();
+ avctx->coded_frame = av_frame_alloc();
if (!avctx->coded_frame)
return AVERROR(ENOMEM);
- ff_proresdsp_init(&ctx->dsp);
- ff_init_scantable(ctx->dsp.dct_permutation, &ctx->scantable,
- interlaced ? ff_prores_interlaced_scan
- : ff_prores_progressive_scan);
+ ctx->fdct = prores_fdct;
+ ctx->scantable = interlaced ? ff_prores_interlaced_scan
+ : ff_prores_progressive_scan;
+ ff_fdctdsp_init(&ctx->fdsp, avctx);
mps = ctx->mbs_per_slice;
if (mps & (mps - 1)) {
"there should be an integer power of two MBs per slice\n");
return AVERROR(EINVAL);
}
+ if (av_pix_fmt_desc_get(avctx->pix_fmt)->flags & AV_PIX_FMT_FLAG_ALPHA) {
+ if (ctx->alpha_bits & 7) {
+ av_log(avctx, AV_LOG_ERROR, "alpha bits should be 0, 8 or 16\n");
+ return AVERROR(EINVAL);
+ }
+ } else {
+ ctx->alpha_bits = 0;
+ }
ctx->chroma_factor = avctx->pix_fmt == AV_PIX_FMT_YUV422P10
? CFACTOR_Y422
: CFACTOR_Y444;
ctx->profile_info = prores_profile_info + ctx->profile;
- ctx->num_planes = 3;
+ ctx->num_planes = 3 + !!ctx->alpha_bits;
ctx->mb_width = FFALIGN(avctx->width, 16) >> 4;
ctx->bits_per_mb = ls * 8;
if (ctx->chroma_factor == CFACTOR_Y444)
ctx->bits_per_mb += ls * 4;
- if (ctx->num_planes == 4)
- ctx->bits_per_mb += ls * 4;
}
ctx->frame_size_upper_bound = ctx->pictures_per_frame *
(mps * ctx->bits_per_mb) / 8)
+ 200;
+ if (ctx->alpha_bits) {
+ // The alpha plane is run-coded and might exceed the bit budget.
+ ctx->frame_size_upper_bound += ctx->pictures_per_frame *
+ ctx->slices_per_picture *
+ /* num pixels per slice */ (ctx->mbs_per_slice * 256 *
+ /* bits per pixel */ (1 + ctx->alpha_bits + 1) + 7 >> 3);
+ }
+
avctx->codec_tag = ctx->profile_info->tag;
av_log(avctx, AV_LOG_DEBUG,
AV_OPT_TYPE_INT, { .i64 = 8 }, 1, MAX_MBS_PER_SLICE, VE },
{ "profile", NULL, OFFSET(profile), AV_OPT_TYPE_INT,
{ .i64 = PRORES_PROFILE_STANDARD },
- PRORES_PROFILE_PROXY, PRORES_PROFILE_HQ, VE, "profile" },
+ PRORES_PROFILE_PROXY, PRORES_PROFILE_4444, VE, "profile" },
{ "proxy", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRORES_PROFILE_PROXY },
0, 0, VE, "profile" },
{ "lt", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRORES_PROFILE_LT },
0, 0, VE, "profile" },
{ "hq", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRORES_PROFILE_HQ },
0, 0, VE, "profile" },
+ { "4444", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = PRORES_PROFILE_4444 },
+ 0, 0, VE, "profile" },
{ "vendor", "vendor ID", OFFSET(vendor),
AV_OPT_TYPE_STRING, { .str = "Lavc" }, CHAR_MIN, CHAR_MAX, VE },
{ "bits_per_mb", "desired bits per macroblock", OFFSET(bits_per_mb),
0, 0, VE, "quant_mat" },
{ "default", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = QUANT_MAT_DEFAULT },
0, 0, VE, "quant_mat" },
+ { "alpha_bits", "bits for alpha plane", OFFSET(alpha_bits), AV_OPT_TYPE_INT,
+ { .i64 = 16 }, 0, 16, VE },
{ NULL }
};
AVCodec ff_prores_encoder = {
.name = "prores",
+ .long_name = NULL_IF_CONFIG_SMALL("Apple ProRes (iCodec Pro)"),
.type = AVMEDIA_TYPE_VIDEO,
.id = AV_CODEC_ID_PRORES,
.priv_data_size = sizeof(ProresContext),
.close = encode_close,
.encode2 = encode_frame,
.capabilities = CODEC_CAP_SLICE_THREADS,
- .long_name = NULL_IF_CONFIG_SMALL("Apple ProRes (iCodec Pro)"),
.pix_fmts = (const enum AVPixelFormat[]) {
- AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10, AV_PIX_FMT_NONE
+ AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
+ AV_PIX_FMT_YUVA444P10, AV_PIX_FMT_NONE
},
.priv_class = &proresenc_class,
};
projects / ffmpeg / blobdiff