Merge commit '89ef08c992c484a46711b1a68a988303679c288e'

* commit '89ef08c992c484a46711b1a68a988303679c288e':
  DNxHD: K&R formatting cosmetics

Conflicts:
	libavcodec/dnxhd_parser.c
	libavcodec/dnxhddec.c
	libavcodec/dnxhdenc.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>
pull/64/head
Michael Niedermayer 11 years ago
commit bad101ab8a
  1. 31
      libavcodec/dnxhddec.c
  2. 284
      libavcodec/dnxhdenc.c

@ -57,9 +57,12 @@ typedef struct DNXHDContext {
#define DNXHD_VLC_BITS 9
#define DNXHD_DC_VLC_BITS 7
static void dnxhd_decode_dct_block_8(DNXHDContext *ctx, int16_t *block, int n, int qscale);
static void dnxhd_decode_dct_block_10(DNXHDContext *ctx, int16_t *block, int n, int qscale);
static void dnxhd_decode_dct_block_10_444(DNXHDContext *ctx, int16_t *block, int n, int qscale);
static void dnxhd_decode_dct_block_8(DNXHDContext *ctx, int16_t *block,
int n, int qscale);
static void dnxhd_decode_dct_block_10(DNXHDContext *ctx, int16_t *block,
int n, int qscale);
static void dnxhd_decode_dct_block_10_444(DNXHDContext *ctx, int16_t *block,
int n, int qscale);
static av_cold int dnxhd_decode_init(AVCodecContext *avctx)
{
@ -99,14 +102,16 @@ static int dnxhd_init_vlc(DNXHDContext *ctx, uint32_t cid)
ctx->cid_table->run_bits, 1, 1,
ctx->cid_table->run_codes, 2, 2, 0);
ff_init_scantable(ctx->dsp.idct_permutation, &ctx->scantable, ff_zigzag_direct);
ff_init_scantable(ctx->dsp.idct_permutation, &ctx->scantable,
ff_zigzag_direct);
ctx->cid = cid;
}
return 0;
}
static int dnxhd_decode_header(DNXHDContext *ctx, AVFrame *frame,
const uint8_t *buf, int buf_size, int first_field)
const uint8_t *buf, int buf_size,
int first_field)
{
static const uint8_t header_prefix[] = { 0x00, 0x00, 0x02, 0x80, 0x01 };
static const uint8_t header_prefix444[] = { 0x00, 0x00, 0x02, 0x80, 0x02 };
@ -123,7 +128,8 @@ static int dnxhd_decode_header(DNXHDContext *ctx, AVFrame *frame,
ctx->cur_field = buf[5] & 1;
frame->interlaced_frame = 1;
frame->top_field_first = first_field ^ ctx->cur_field;
av_log(ctx->avctx, AV_LOG_DEBUG, "interlaced %d, cur field %d\n", buf[5] & 3, ctx->cur_field);
av_log(ctx->avctx, AV_LOG_DEBUG,
"interlaced %d, cur field %d\n", buf[5] & 3, ctx->cur_field);
}
ctx->height = AV_RB16(buf + 0x18);
@ -173,14 +179,16 @@ static int dnxhd_decode_header(DNXHDContext *ctx, AVFrame *frame,
ctx->mb_width = ctx->width >> 4;
ctx->mb_height = buf[0x16d];
av_dlog(ctx->avctx, "mb width %d, mb height %d\n", ctx->mb_width, ctx->mb_height);
av_dlog(ctx->avctx,
"mb width %d, mb height %d\n", ctx->mb_width, ctx->mb_height);
if ((ctx->height + 15) >> 4 == ctx->mb_height && frame->interlaced_frame)
ctx->height <<= 1;
if (ctx->mb_height > 68 ||
(ctx->mb_height << frame->interlaced_frame) > (ctx->height + 15) >> 4) {
av_log(ctx->avctx, AV_LOG_ERROR, "mb height too big: %d\n", ctx->mb_height);
av_log(ctx->avctx, AV_LOG_ERROR,
"mb height too big: %d\n", ctx->mb_height);
return AVERROR_INVALIDDATA;
}
@ -308,7 +316,8 @@ static void dnxhd_decode_dct_block_10_444(DNXHDContext *ctx, int16_t *block,
dnxhd_decode_dct_block(ctx, block, n, qscale, 6, 32, 6);
}
static int dnxhd_decode_macroblock(DNXHDContext *ctx, AVFrame *frame, int x, int y)
static int dnxhd_decode_macroblock(DNXHDContext *ctx, AVFrame *frame,
int x, int y)
{
int shift1 = ctx->bit_depth == 10;
int dct_linesize_luma = frame->linesize[0];
@ -409,8 +418,8 @@ static int dnxhd_decode_macroblocks(DNXHDContext *ctx, AVFrame *frame,
return 0;
}
static int dnxhd_decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
AVPacket *avpkt)
static int dnxhd_decode_frame(AVCodecContext *avctx, void *data,
int *got_frame, AVPacket *avpkt)
{
const uint8_t *buf = avpkt->data;
int buf_size = avpkt->size;

@ -23,23 +23,27 @@
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#define RC_VARIANCE 1 // use variance or ssd for fast rc
#include "libavutil/attributes.h"
#include "libavutil/internal.h"
#include "libavutil/opt.h"
#include "libavutil/timer.h"
#include "avcodec.h"
#include "dsputil.h"
#include "internal.h"
#include "mpegvideo.h"
#include "dnxhdenc.h"
#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
#define DNX10BIT_QMAT_SHIFT 18 // The largest value that will not lead to overflow for 10bit samples.
// The largest value that will not lead to overflow for 10bit samples.
#define DNX10BIT_QMAT_SHIFT 18
#define RC_VARIANCE 1 // use variance or ssd for fast rc
#define LAMBDA_FRAC_BITS 10
#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
static const AVOption options[] = {
{"nitris_compat", "encode with Avid Nitris compatibility", offsetof(DNXHDEncContext, nitris_compat), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, VE},
{ "nitris_compat", "encode with Avid Nitris compatibility",
offsetof(DNXHDEncContext, nitris_compat), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
{ NULL }
};
@ -50,16 +54,20 @@ static const AVClass dnxhd_class = {
.version = LIBAVUTIL_VERSION_INT,
};
#define LAMBDA_FRAC_BITS 10
static void dnxhd_8bit_get_pixels_8x4_sym(int16_t *av_restrict block, const uint8_t *pixels, ptrdiff_t line_size)
static void dnxhd_8bit_get_pixels_8x4_sym(int16_t *av_restrict block,
const uint8_t *pixels,
ptrdiff_t line_size)
{
int i;
for (i = 0; i < 4; i++) {
block[0] = pixels[0]; block[1] = pixels[1];
block[2] = pixels[2]; block[3] = pixels[3];
block[4] = pixels[4]; block[5] = pixels[5];
block[6] = pixels[6]; block[7] = pixels[7];
block[0] = pixels[0];
block[1] = pixels[1];
block[2] = pixels[2];
block[3] = pixels[3];
block[4] = pixels[4];
block[5] = pixels[5];
block[6] = pixels[6];
block[7] = pixels[7];
pixels += line_size;
block += 8;
}
@ -69,7 +77,10 @@ static void dnxhd_8bit_get_pixels_8x4_sym(int16_t *av_restrict block, const uint
memcpy(block + 24, block - 32, sizeof(*block) * 8);
}
static av_always_inline void dnxhd_10bit_get_pixels_8x4_sym(int16_t *av_restrict block, const uint8_t *pixels, ptrdiff_t line_size)
static av_always_inline
void dnxhd_10bit_get_pixels_8x4_sym(int16_t *av_restrict block,
const uint8_t *pixels,
ptrdiff_t line_size)
{
int i;
const uint16_t* pixels16 = (const uint16_t*)pixels;
@ -120,10 +131,14 @@ static av_cold int dnxhd_init_vlc(DNXHDEncContext *ctx)
int i, j, level, run;
int max_level = 1 << (ctx->cid_table->bit_depth + 2);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->vlc_codes, max_level*4*sizeof(*ctx->vlc_codes), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->vlc_bits, max_level*4*sizeof(*ctx->vlc_bits) , fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->run_codes, 63*2, fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->run_bits, 63, fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->vlc_codes,
max_level * 4 * sizeof(*ctx->vlc_codes), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->vlc_bits,
max_level * 4 * sizeof(*ctx->vlc_bits), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->run_codes,
63 * 2, fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->run_bits,
63, fail);
ctx->vlc_codes += max_level * 2;
ctx->vlc_bits += max_level * 2;
@ -143,7 +158,8 @@ static av_cold int dnxhd_init_vlc(DNXHDEncContext *ctx)
(!run || (ctx->cid_table->ac_flags[j] & 2) && run)) {
av_assert1(!ctx->vlc_codes[index]);
if (alevel) {
ctx->vlc_codes[index] = (ctx->cid_table->ac_codes[j]<<1)|(sign&1);
ctx->vlc_codes[index] =
(ctx->cid_table->ac_codes[j] << 1) | (sign & 1);
ctx->vlc_bits[index] = ctx->cid_table->ac_bits[j] + 1;
} else {
ctx->vlc_codes[index] = ctx->cid_table->ac_codes[j];
@ -154,7 +170,8 @@ static av_cold int dnxhd_init_vlc(DNXHDEncContext *ctx)
}
av_assert0(!alevel || j < 257);
if (offset) {
ctx->vlc_codes[index] = (ctx->vlc_codes[index]<<ctx->cid_table->index_bits)|offset;
ctx->vlc_codes[index] =
(ctx->vlc_codes[index] << ctx->cid_table->index_bits) | offset;
ctx->vlc_bits[index] += ctx->cid_table->index_bits;
}
}
@ -178,30 +195,41 @@ static av_cold int dnxhd_init_qmat(DNXHDEncContext *ctx, int lbias, int cbias)
const uint8_t *luma_weight_table = ctx->cid_table->luma_weight;
const uint8_t *chroma_weight_table = ctx->cid_table->chroma_weight;
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->qmatrix_l, (ctx->m.avctx->qmax+1) * 64 * sizeof(int), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->qmatrix_c, (ctx->m.avctx->qmax+1) * 64 * sizeof(int), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->qmatrix_l16, (ctx->m.avctx->qmax+1) * 64 * 2 * sizeof(uint16_t), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->qmatrix_c16, (ctx->m.avctx->qmax+1) * 64 * 2 * sizeof(uint16_t), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->qmatrix_l,
(ctx->m.avctx->qmax + 1) * 64 * sizeof(int), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->qmatrix_c,
(ctx->m.avctx->qmax + 1) * 64 * sizeof(int), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->qmatrix_l16,
(ctx->m.avctx->qmax + 1) * 64 * 2 * sizeof(uint16_t),
fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->qmatrix_c16,
(ctx->m.avctx->qmax + 1) * 64 * 2 * sizeof(uint16_t),
fail);
if (ctx->cid_table->bit_depth == 8) {
for (i = 1; i < 64; i++) {
int j = ctx->m.dsp.idct_permutation[ff_zigzag_direct[i]];
weight_matrix[j] = ctx->cid_table->luma_weight[i];
}
ff_convert_matrix(&ctx->m.dsp, ctx->qmatrix_l, ctx->qmatrix_l16, weight_matrix,
ctx->m.intra_quant_bias, 1, ctx->m.avctx->qmax, 1);
ff_convert_matrix(&ctx->m.dsp, ctx->qmatrix_l, ctx->qmatrix_l16,
weight_matrix, ctx->m.intra_quant_bias, 1,
ctx->m.avctx->qmax, 1);
for (i = 1; i < 64; i++) {
int j = ctx->m.dsp.idct_permutation[ff_zigzag_direct[i]];
weight_matrix[j] = ctx->cid_table->chroma_weight[i];
}
ff_convert_matrix(&ctx->m.dsp, ctx->qmatrix_c, ctx->qmatrix_c16, weight_matrix,
ctx->m.intra_quant_bias, 1, ctx->m.avctx->qmax, 1);
ff_convert_matrix(&ctx->m.dsp, ctx->qmatrix_c, ctx->qmatrix_c16,
weight_matrix, ctx->m.intra_quant_bias, 1,
ctx->m.avctx->qmax, 1);
for (qscale = 1; qscale <= ctx->m.avctx->qmax; qscale++) {
for (i = 0; i < 64; i++) {
ctx->qmatrix_l [qscale] [i] <<= 2; ctx->qmatrix_c [qscale] [i] <<= 2;
ctx->qmatrix_l16[qscale][0][i] <<= 2; ctx->qmatrix_l16[qscale][1][i] <<= 2;
ctx->qmatrix_c16[qscale][0][i] <<= 2; ctx->qmatrix_c16[qscale][1][i] <<= 2;
ctx->qmatrix_l[qscale][i] <<= 2;
ctx->qmatrix_c[qscale][i] <<= 2;
ctx->qmatrix_l16[qscale][0][i] <<= 2;
ctx->qmatrix_l16[qscale][1][i] <<= 2;
ctx->qmatrix_c16[qscale][0][i] <<= 2;
ctx->qmatrix_c16[qscale][1][i] <<= 2;
}
}
} else {
@ -210,16 +238,21 @@ static av_cold int dnxhd_init_qmat(DNXHDEncContext *ctx, int lbias, int cbias)
for (i = 1; i < 64; i++) {
int j = ctx->m.dsp.idct_permutation[ff_zigzag_direct[i]];
// The quantization formula from the VC-3 standard is:
// quantized = sign(block[i]) * floor(abs(block[i]/s) * p / (qscale * weight_table[i]))
// Where p is 32 for 8-bit samples and 8 for 10-bit ones.
// The s factor compensates scaling of DCT coefficients done by the DCT routines,
// and therefore is not present in standard. It's 8 for 8-bit samples and 4 for 10-bit ones.
// We want values of ctx->qtmatrix_l and ctx->qtmatrix_r to be:
// ((1 << DNX10BIT_QMAT_SHIFT) * (p / s)) / (qscale * weight_table[i])
// For 10-bit samples, p / s == 2
ctx->qmatrix_l[qscale][j] = (1 << (DNX10BIT_QMAT_SHIFT + 1)) / (qscale * luma_weight_table[i]);
ctx->qmatrix_c[qscale][j] = (1 << (DNX10BIT_QMAT_SHIFT + 1)) / (qscale * chroma_weight_table[i]);
/* The quantization formula from the VC-3 standard is:
* quantized = sign(block[i]) * floor(abs(block[i]/s) * p /
* (qscale * weight_table[i]))
* Where p is 32 for 8-bit samples and 8 for 10-bit ones.
* The s factor compensates scaling of DCT coefficients done by
* the DCT routines, and therefore is not present in standard.
* It's 8 for 8-bit samples and 4 for 10-bit ones.
* We want values of ctx->qtmatrix_l and ctx->qtmatrix_r to be:
* ((1 << DNX10BIT_QMAT_SHIFT) * (p / s)) /
* (qscale * weight_table[i])
* For 10-bit samples, p / s == 2 */
ctx->qmatrix_l[qscale][j] = (1 << (DNX10BIT_QMAT_SHIFT + 1)) /
(qscale * luma_weight_table[i]);
ctx->qmatrix_c[qscale][j] = (1 << (DNX10BIT_QMAT_SHIFT + 1)) /
(qscale * chroma_weight_table[i]);
}
}
}
@ -238,9 +271,11 @@ static av_cold int dnxhd_init_rc(DNXHDEncContext *ctx)
{
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->mb_rc, 8160 * (ctx->m.avctx->qmax + 1) * sizeof(RCEntry), fail);
if (ctx->m.avctx->mb_decision != FF_MB_DECISION_RD)
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->mb_cmp, ctx->m.mb_num*sizeof(RCCMPEntry), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->mb_cmp,
ctx->m.mb_num * sizeof(RCCMPEntry), fail);
ctx->frame_bits = (ctx->cid_table->coding_unit_size - 640 - 4 - ctx->min_padding) * 8;
ctx->frame_bits = (ctx->cid_table->coding_unit_size -
640 - 4 - ctx->min_padding) * 8;
ctx->qscale = 1;
ctx->lambda = 2 << LAMBDA_FRAC_BITS; // qscale 2
return 0;
@ -261,13 +296,15 @@ static av_cold int dnxhd_encode_init(AVCodecContext *avctx)
bit_depth = 10;
break;
default:
av_log(avctx, AV_LOG_ERROR, "pixel format is incompatible with DNxHD\n");
av_log(avctx, AV_LOG_ERROR,
"pixel format is incompatible with DNxHD\n");
return AVERROR(EINVAL);
}
ctx->cid = ff_dnxhd_find_cid(avctx, bit_depth);
if (!ctx->cid) {
av_log(avctx, AV_LOG_ERROR, "video parameters incompatible with DNxHD. Valid DNxHD profiles:\n");
av_log(avctx, AV_LOG_ERROR,
"video parameters incompatible with DNxHD. Valid DNxHD profiles:\n");
ff_dnxhd_print_profiles(avctx, AV_LOG_ERROR);
return AVERROR(EINVAL);
}
@ -313,10 +350,12 @@ static av_cold int dnxhd_encode_init(AVCodecContext *avctx)
if (avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
ctx->m.intra_quant_bias = avctx->intra_quant_bias;
if ((ret = dnxhd_init_qmat(ctx, ctx->m.intra_quant_bias, 0)) < 0) // XXX tune lbias/cbias
// XXX tune lbias/cbias
if ((ret = dnxhd_init_qmat(ctx, ctx->m.intra_quant_bias, 0)) < 0)
return ret;
// Avid Nitris hardware decoder requires a minimum amount of padding in the coding unit payload
/* Avid Nitris hardware decoder requires a minimum amount of padding
* in the coding unit payload */
if (ctx->nitris_compat)
ctx->min_padding = 1600;
@ -325,10 +364,14 @@ static av_cold int dnxhd_encode_init(AVCodecContext *avctx)
if ((ret = dnxhd_init_rc(ctx)) < 0)
return ret;
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->slice_size, ctx->m.mb_height*sizeof(uint32_t), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->slice_offs, ctx->m.mb_height*sizeof(uint32_t), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->mb_bits, ctx->m.mb_num *sizeof(uint16_t), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->mb_qscale, ctx->m.mb_num *sizeof(uint8_t), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->slice_size,
ctx->m.mb_height * sizeof(uint32_t), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->slice_offs,
ctx->m.mb_height * sizeof(uint32_t), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->mb_bits,
ctx->m.mb_num * sizeof(uint16_t), fail);
FF_ALLOCZ_OR_GOTO(ctx->m.avctx, ctx->mb_qscale,
ctx->m.mb_num * sizeof(uint8_t), fail);
avctx->coded_frame = av_frame_alloc();
if (!avctx->coded_frame)
@ -399,10 +442,13 @@ static av_always_inline void dnxhd_encode_dc(DNXHDEncContext *ctx, int diff)
nbits = av_log2_16bit(2 * diff);
}
put_bits(&ctx->m.pb, ctx->cid_table->dc_bits[nbits] + nbits,
(ctx->cid_table->dc_codes[nbits]<<nbits) + (diff & ((1 << nbits) - 1)));
(ctx->cid_table->dc_codes[nbits] << nbits) +
(diff & ((1 << nbits) - 1)));
}
static av_always_inline void dnxhd_encode_block(DNXHDEncContext *ctx, int16_t *block, int last_index, int n)
static av_always_inline
void dnxhd_encode_block(DNXHDEncContext *ctx, int16_t *block,
int last_index, int n)
{
int last_non_zero = 0;
int slevel, i, j;
@ -418,20 +464,24 @@ static av_always_inline void dnxhd_encode_block(DNXHDEncContext *ctx, int16_t *b
int rlevel = (slevel << 1) | !!run_level;
put_bits(&ctx->m.pb, ctx->vlc_bits[rlevel], ctx->vlc_codes[rlevel]);
if (run_level)
put_bits(&ctx->m.pb, ctx->run_bits[run_level], ctx->run_codes[run_level]);
put_bits(&ctx->m.pb, ctx->run_bits[run_level],
ctx->run_codes[run_level]);
last_non_zero = i;
}
}
put_bits(&ctx->m.pb, ctx->vlc_bits[0], ctx->vlc_codes[0]); // EOB
}
static av_always_inline void dnxhd_unquantize_c(DNXHDEncContext *ctx, int16_t *block, int n, int qscale, int last_index)
static av_always_inline
void dnxhd_unquantize_c(DNXHDEncContext *ctx, int16_t *block, int n,
int qscale, int last_index)
{
const uint8_t *weight_matrix;
int level;
int i;
weight_matrix = (n&2) ? ctx->cid_table->chroma_weight : ctx->cid_table->luma_weight;
weight_matrix = (n & 2) ? ctx->cid_table->chroma_weight
: ctx->cid_table->luma_weight;
for (i = 1; i <= last_index; i++) {
int j = ctx->m.intra_scantable.permutated[i];
@ -475,7 +525,8 @@ static av_always_inline int dnxhd_ssd_block(int16_t *qblock, int16_t *block)
return score;
}
static av_always_inline int dnxhd_calc_ac_bits(DNXHDEncContext *ctx, int16_t *block, int last_index)
static av_always_inline
int dnxhd_calc_ac_bits(DNXHDEncContext *ctx, int16_t *block, int last_index)
{
int last_non_zero = 0;
int bits = 0;
@ -485,20 +536,25 @@ static av_always_inline int dnxhd_calc_ac_bits(DNXHDEncContext *ctx, int16_t *bl
level = block[j];
if (level) {
int run_level = i - last_non_zero - 1;
bits += ctx->vlc_bits[(level<<1)|!!run_level]+ctx->run_bits[run_level];
bits += ctx->vlc_bits[(level << 1) |
!!run_level] + ctx->run_bits[run_level];
last_non_zero = i;
}
}
return bits;
}
static av_always_inline void dnxhd_get_blocks(DNXHDEncContext *ctx, int mb_x, int mb_y)
static av_always_inline
void dnxhd_get_blocks(DNXHDEncContext *ctx, int mb_x, int mb_y)
{
const int bs = ctx->block_width_l2;
const int bw = 1 << bs;
const uint8_t *ptr_y = ctx->thread[0]->src[0] + ((mb_y << 4) * ctx->m.linesize) + (mb_x << bs+1);
const uint8_t *ptr_u = ctx->thread[0]->src[1] + ((mb_y << 4) * ctx->m.uvlinesize) + (mb_x << bs);
const uint8_t *ptr_v = ctx->thread[0]->src[2] + ((mb_y << 4) * ctx->m.uvlinesize) + (mb_x << bs);
const uint8_t *ptr_y = ctx->thread[0]->src[0] +
((mb_y << 4) * ctx->m.linesize) + (mb_x << bs + 1);
const uint8_t *ptr_u = ctx->thread[0]->src[1] +
((mb_y << 4) * ctx->m.uvlinesize) + (mb_x << bs);
const uint8_t *ptr_v = ctx->thread[0]->src[2] +
((mb_y << 4) * ctx->m.uvlinesize) + (mb_x << bs);
DSPContext *dsp = &ctx->m.dsp;
dsp->get_pixels(ctx->blocks[0], ptr_y, ctx->m.linesize);
@ -508,10 +564,18 @@ static av_always_inline void dnxhd_get_blocks(DNXHDEncContext *ctx, int mb_x, in
if (mb_y + 1 == ctx->m.mb_height && ctx->m.avctx->height == 1080) {
if (ctx->interlaced) {
ctx->get_pixels_8x4_sym(ctx->blocks[4], ptr_y + ctx->dct_y_offset, ctx->m.linesize);
ctx->get_pixels_8x4_sym(ctx->blocks[5], ptr_y + ctx->dct_y_offset + bw, ctx->m.linesize);
ctx->get_pixels_8x4_sym(ctx->blocks[6], ptr_u + ctx->dct_uv_offset, ctx->m.uvlinesize);
ctx->get_pixels_8x4_sym(ctx->blocks[7], ptr_v + ctx->dct_uv_offset, ctx->m.uvlinesize);
ctx->get_pixels_8x4_sym(ctx->blocks[4],
ptr_y + ctx->dct_y_offset,
ctx->m.linesize);
ctx->get_pixels_8x4_sym(ctx->blocks[5],
ptr_y + ctx->dct_y_offset + bw,
ctx->m.linesize);
ctx->get_pixels_8x4_sym(ctx->blocks[6],
ptr_u + ctx->dct_uv_offset,
ctx->m.uvlinesize);
ctx->get_pixels_8x4_sym(ctx->blocks[7],
ptr_v + ctx->dct_uv_offset,
ctx->m.uvlinesize);
} else {
dsp->clear_block(ctx->blocks[4]);
dsp->clear_block(ctx->blocks[5]);
@ -519,20 +583,26 @@ static av_always_inline void dnxhd_get_blocks(DNXHDEncContext *ctx, int mb_x, in
dsp->clear_block(ctx->blocks[7]);
}
} else {
dsp->get_pixels(ctx->blocks[4], ptr_y + ctx->dct_y_offset, ctx->m.linesize);
dsp->get_pixels(ctx->blocks[5], ptr_y + ctx->dct_y_offset + bw, ctx->m.linesize);
dsp->get_pixels(ctx->blocks[6], ptr_u + ctx->dct_uv_offset, ctx->m.uvlinesize);
dsp->get_pixels(ctx->blocks[7], ptr_v + ctx->dct_uv_offset, ctx->m.uvlinesize);
dsp->get_pixels(ctx->blocks[4],
ptr_y + ctx->dct_y_offset, ctx->m.linesize);
dsp->get_pixels(ctx->blocks[5],
ptr_y + ctx->dct_y_offset + bw, ctx->m.linesize);
dsp->get_pixels(ctx->blocks[6],
ptr_u + ctx->dct_uv_offset, ctx->m.uvlinesize);
dsp->get_pixels(ctx->blocks[7],
ptr_v + ctx->dct_uv_offset, ctx->m.uvlinesize);
}
}
static av_always_inline int dnxhd_switch_matrix(DNXHDEncContext *ctx, int i)
static av_always_inline
int dnxhd_switch_matrix(DNXHDEncContext *ctx, int i)
{
const static uint8_t component[8]={0,0,1,2,0,0,1,2};
return component[i];
}
static int dnxhd_calc_bits_thread(AVCodecContext *avctx, void *arg, int jobnr, int threadnr)
static int dnxhd_calc_bits_thread(AVCodecContext *avctx, void *arg,
int jobnr, int threadnr)
{
DNXHDEncContext *ctx = avctx->priv_data;
int mb_y = jobnr, mb_x;
@ -559,12 +629,15 @@ static int dnxhd_calc_bits_thread(AVCodecContext *avctx, void *arg, int jobnr, i
int n = dnxhd_switch_matrix(ctx, i);
memcpy(block, src_block, 64 * sizeof(*block));
last_index = ctx->m.dct_quantize(&ctx->m, block, 4&(2*i), qscale, &overflow);
last_index = ctx->m.dct_quantize(&ctx->m, block, 4 & (2*i),
qscale, &overflow);
ac_bits += dnxhd_calc_ac_bits(ctx, block, last_index);
diff = block[0] - ctx->m.last_dc[n];
if (diff < 0) nbits = av_log2_16bit(-2*diff);
else nbits = av_log2_16bit( 2*diff);
if (diff < 0)
nbits = av_log2_16bit(-2 * diff);
else
nbits = av_log2_16bit(2 * diff);
av_assert1(nbits < ctx->cid_table->bit_depth + 4);
dc_bits += ctx->cid_table->dc_bits[nbits] + nbits;
@ -578,17 +651,20 @@ static int dnxhd_calc_bits_thread(AVCodecContext *avctx, void *arg, int jobnr, i
}
}
ctx->mb_rc[qscale][mb].ssd = ssd;
ctx->mb_rc[qscale][mb].bits = ac_bits+dc_bits+12+8*ctx->vlc_bits[0];
ctx->mb_rc[qscale][mb].bits = ac_bits + dc_bits + 12 +
8 * ctx->vlc_bits[0];
}
return 0;
}
static int dnxhd_encode_thread(AVCodecContext *avctx, void *arg, int jobnr, int threadnr)
static int dnxhd_encode_thread(AVCodecContext *avctx, void *arg,
int jobnr, int threadnr)
{
DNXHDEncContext *ctx = avctx->priv_data;
int mb_y = jobnr, mb_x;
ctx = ctx->thread[threadnr];
init_put_bits(&ctx->m.pb, (uint8_t *)arg + 640 + ctx->slice_offs[jobnr], ctx->slice_size[jobnr]);
init_put_bits(&ctx->m.pb, (uint8_t *)arg + 640 + ctx->slice_offs[jobnr],
ctx->slice_size[jobnr]);
ctx->m.last_dc[0] =
ctx->m.last_dc[1] =
@ -605,7 +681,8 @@ static int dnxhd_encode_thread(AVCodecContext *avctx, void *arg, int jobnr, int
for (i = 0; i < 8; i++) {
int16_t *block = ctx->blocks[i];
int overflow, n = dnxhd_switch_matrix(ctx, i);
int last_index = ctx->m.dct_quantize(&ctx->m, block, 4&(2*i), qscale, &overflow);
int last_index = ctx->m.dct_quantize(&ctx->m, block, 4 & (2*i),
qscale, &overflow);
// START_TIMER;
dnxhd_encode_block(ctx, block, last_index, n);
// STOP_TIMER("encode_block");
@ -636,7 +713,8 @@ static void dnxhd_setup_threads_slices(DNXHDEncContext *ctx)
}
}
static int dnxhd_mb_var_thread(AVCodecContext *avctx, void *arg, int jobnr, int threadnr)
static int dnxhd_mb_var_thread(AVCodecContext *avctx, void *arg,
int jobnr, int threadnr)
{
DNXHDEncContext *ctx = avctx->priv_data;
int mb_y = jobnr, mb_x, x, y;
@ -674,7 +752,8 @@ static int dnxhd_mb_var_thread(AVCodecContext *avctx, void *arg, int jobnr, int
} else { // 10-bit
int const linesize = ctx->m.linesize >> 1;
for (mb_x = 0; mb_x < ctx->m.mb_width; ++mb_x) {
uint16_t *pix = (uint16_t*)ctx->thread[0]->src[0] + ((mb_y << 4) * linesize) + (mb_x << 4);
uint16_t *pix = (uint16_t *)ctx->thread[0]->src[0] +
((mb_y << 4) * linesize) + (mb_x << 4);
unsigned mb = mb_y * ctx->m.mb_width + mb_x;
int sum = 0;
int sqsum = 0;
@ -708,7 +787,8 @@ static int dnxhd_encode_rdo(AVCodecContext *avctx, DNXHDEncContext *ctx)
for (q = 1; q < avctx->qmax; q++) {
ctx->qscale = q;
avctx->execute2(avctx, dnxhd_calc_bits_thread, NULL, NULL, ctx->m.mb_height);
avctx->execute2(avctx, dnxhd_calc_bits_thread,
NULL, NULL, ctx->m.mb_height);
}
up_step = down_step = 2 << LAMBDA_FRAC_BITS;
lambda = ctx->lambda;
@ -741,7 +821,8 @@ static int dnxhd_encode_rdo(AVCodecContext *avctx, DNXHDEncContext *ctx)
if (bits > ctx->frame_bits)
break;
}
//av_dlog(ctx->m.avctx, "lambda %d, up %u, down %u, bits %d, frame %d\n",
// av_dlog(ctx->m.avctx,
// "lambda %d, up %u, down %u, bits %d, frame %d\n",
// lambda, last_higher, last_lower, bits, ctx->frame_bits);
if (end) {
if (bits > ctx->frame_bits)
@ -791,7 +872,8 @@ static int dnxhd_find_qscale(DNXHDEncContext *ctx)
bits = 0;
ctx->qscale = qscale;
// XXX avoid recalculating bits
ctx->m.avctx->execute2(ctx->m.avctx, dnxhd_calc_bits_thread, NULL, NULL, ctx->m.mb_height);
ctx->m.avctx->execute2(ctx->m.avctx, dnxhd_calc_bits_thread,
NULL, NULL, ctx->m.mb_height);
for (y = 0; y < ctx->m.mb_height; y++) {
for (x = 0; x < ctx->m.mb_width; x++)
bits += ctx->mb_rc[qscale][y*ctx->m.mb_width+x].bits;
@ -799,8 +881,10 @@ static int dnxhd_find_qscale(DNXHDEncContext *ctx)
if (bits > ctx->frame_bits)
break;
}
//av_dlog(ctx->m.avctx, "%d, qscale %d, bits %d, frame %d, higher %d, lower %d\n",
// ctx->m.avctx->frame_number, qscale, bits, ctx->frame_bits, last_higher, last_lower);
// av_dlog(ctx->m.avctx,
// "%d, qscale %d, bits %d, frame %d, higher %d, lower %d\n",
// ctx->m.avctx->frame_number, qscale, bits, ctx->frame_bits,
// last_higher, last_lower);
if (bits < ctx->frame_bits) {
if (qscale == 1)
return 1;
@ -845,7 +929,8 @@ static inline int get_bucket(int value, int shift)
return NBUCKETS - 1 - value;
}
static void radix_count(const RCCMPEntry *data, int size, int buckets[RADIX_PASSES][NBUCKETS])
static void radix_count(const RCCMPEntry *data, int size,
int buckets[RADIX_PASSES][NBUCKETS])
{
int i, j;
memset(buckets, 0, sizeof(buckets[0][0]) * RADIX_PASSES * NBUCKETS);
@ -865,7 +950,8 @@ static void radix_count(const RCCMPEntry *data, int size, int buckets[RADIX_PASS
}
}
static void radix_sort_pass(RCCMPEntry *dst, const RCCMPEntry *data, int size, int buckets[NBUCKETS], int pass)
static void radix_sort_pass(RCCMPEntry *dst, const RCCMPEntry *data,
int size, int buckets[NBUCKETS], int pass)
{
int shift = pass * BUCKET_BITS;
int i;
@ -904,10 +990,13 @@ static int dnxhd_encode_fast(AVCodecContext *avctx, DNXHDEncContext *ctx)
ctx->mb_bits[mb] = ctx->mb_rc[ctx->qscale][mb].bits;
max_bits += ctx->mb_rc[ctx->qscale][mb].bits;
if (!RC_VARIANCE) {
delta_bits = ctx->mb_rc[ctx->qscale][mb].bits-ctx->mb_rc[ctx->qscale+1][mb].bits;
delta_bits = ctx->mb_rc[ctx->qscale][mb].bits -
ctx->mb_rc[ctx->qscale + 1][mb].bits;
ctx->mb_cmp[mb].mb = mb;
ctx->mb_cmp[mb].value = delta_bits ?
((ctx->mb_rc[ctx->qscale][mb].ssd-ctx->mb_rc[ctx->qscale+1][mb].ssd)*100)/delta_bits
ctx->mb_cmp[mb].value =
delta_bits ? ((ctx->mb_rc[ctx->qscale][mb].ssd -
ctx->mb_rc[ctx->qscale + 1][mb].ssd) * 100) /
delta_bits
: INT_MIN; // avoid increasing qscale
}
}
@ -915,11 +1004,13 @@ static int dnxhd_encode_fast(AVCodecContext *avctx, DNXHDEncContext *ctx)
}
if (!ret) {
if (RC_VARIANCE)
avctx->execute2(avctx, dnxhd_mb_var_thread, NULL, NULL, ctx->m.mb_height);
avctx->execute2(avctx, dnxhd_mb_var_thread,
NULL, NULL, ctx->m.mb_height);
radix_sort(ctx->mb_cmp, ctx->m.mb_num);
for (x = 0; x < ctx->m.mb_num && max_bits > ctx->frame_bits; x++) {
int mb = ctx->mb_cmp[x].mb;
max_bits -= ctx->mb_rc[ctx->qscale][mb].bits - ctx->mb_rc[ctx->qscale+1][mb].bits;
max_bits -= ctx->mb_rc[ctx->qscale][mb].bits -
ctx->mb_rc[ctx->qscale + 1][mb].bits;
ctx->mb_qscale[mb] = ctx->qscale + 1;
ctx->mb_bits[mb] = ctx->mb_rc[ctx->qscale + 1][mb].bits;
}
@ -987,7 +1078,8 @@ static int dnxhd_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
avctx->execute2(avctx, dnxhd_encode_thread, buf, NULL, ctx->m.mb_height);
av_assert1(640 + offset + 4 <= ctx->cid_table->coding_unit_size);
memset(buf + 640 + offset, 0, ctx->cid_table->coding_unit_size - 4 - offset - 640);
memset(buf + 640 + offset, 0,
ctx->cid_table->coding_unit_size - 4 - offset - 640);
AV_WB32(buf + ctx->cid_table->coding_unit_size - 4, 0x600DC0DE); // EOF
@ -1051,9 +1143,11 @@ AVCodec ff_dnxhd_encoder = {
.encode2 = dnxhd_encode_picture,
.close = dnxhd_encode_end,
.capabilities = CODEC_CAP_SLICE_THREADS,
.pix_fmts = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV422P,
.pix_fmts = (const enum AVPixelFormat[]) {
AV_PIX_FMT_YUV422P,
AV_PIX_FMT_YUV422P10,
AV_PIX_FMT_NONE },
AV_PIX_FMT_NONE
},
.priv_class = &dnxhd_class,
.defaults = dnxhd_defaults,
};

Loading…
Cancel
Save