27 #define RC_VARIANCE 1 // use variance or ssd for fast rc
36 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
37 #define DNX10BIT_QMAT_SHIFT 18 // The largest value that will not lead to overflow for 10bit samples.
45 #define LAMBDA_FRAC_BITS 10
50 for (i = 0; i < 4; i++) {
51 block[0] = pixels[0]; block[1] = pixels[1];
52 block[2] = pixels[2]; block[3] = pixels[3];
53 block[4] = pixels[4]; block[5] = pixels[5];
54 block[6] = pixels[6]; block[7] = pixels[7];
58 memcpy(block, block - 8,
sizeof(*block) * 8);
59 memcpy(block + 8, block - 16,
sizeof(*block) * 8);
60 memcpy(block + 16, block - 24,
sizeof(*block) * 8);
61 memcpy(block + 24, block - 32,
sizeof(*block) * 8);
70 for (i = 0; i < 4; i++) {
71 memcpy(block + i * 8, pixels + i * line_size, 8 *
sizeof(*block));
72 memcpy(block - (i+1) * 8, pixels + i * line_size, 8 *
sizeof(*block));
77 int n,
int qscale,
int *overflow)
81 int last_non_zero = 0;
87 block[0] = (block[0] + 2) >> 2;
89 for (i = 1; i < 64; ++i) {
91 int sign = block[j] >> 31;
92 int level = (block[j] ^ sign) - sign;
94 block[j] = (level ^ sign) - sign;
114 for (level = -max_level; level < max_level; level++) {
115 for (run = 0; run < 2; run++) {
116 int index = (level<<1)|run;
117 int sign, offset = 0, alevel =
level;
121 offset = (alevel-1)>>6;
124 for (j = 0; j < 257; j++) {
139 assert(!alevel || j < 257);
146 for (i = 0; i < 62; i++) {
160 uint16_t weight_matrix[64] = {1,};
171 for (i = 1; i < 64; i++) {
177 for (i = 1; i < 64; i++) {
184 for (qscale = 1; qscale <= ctx->
m.
avctx->
qmax; qscale++) {
185 for (i = 0; i < 64; i++) {
193 for (qscale = 1; qscale <= ctx->
m.
avctx->
qmax; qscale++) {
194 for (i = 1; i < 64; i++) {
233 int i,
index, bit_depth;
332 const uint8_t header_prefix[5] = { 0x00,0x00,0x02,0x80,0x01 };
336 memcpy(buf, header_prefix, 5);
356 ctx->
msip = buf + 0x170;
375 int last_non_zero = 0;
381 for (i = 1; i <= last_index; i++) {
385 int run_level = i - last_non_zero - 1;
386 int rlevel = (slevel<<1)|!!run_level;
404 for (i = 1; i <= last_index; i++) {
409 level = (1-2*
level) * qscale * weight_matrix[i];
411 if (weight_matrix[i] != 8)
415 if (weight_matrix[i] != 32)
421 level = (2*level+1) * qscale * weight_matrix[i];
423 if (weight_matrix[i] != 8)
427 if (weight_matrix[i] != 32)
441 for (i = 0; i < 64; i++)
442 score += (block[i] - qblock[i]) * (block[i] - qblock[i]);
448 int last_non_zero = 0;
451 for (i = 1; i <= last_index; i++) {
455 int run_level = i - last_non_zero - 1;
466 const int bw = 1 << bs;
513 int mb_y = jobnr, mb_x;
516 ctx = ctx->
thread[threadnr];
522 for (mb_x = 0; mb_x < ctx->
m.
mb_width; mb_x++) {
523 unsigned mb = mb_y * ctx->
m.
mb_width + mb_x;
531 for (i = 0; i < 8; i++) {
533 int overflow, nbits, diff, last_index;
544 assert(nbits < ctx->cid_table->bit_depth + 4);
564 int mb_y = jobnr, mb_x;
565 ctx = ctx->
thread[threadnr];
571 for (mb_x = 0; mb_x < ctx->
m.
mb_width; mb_x++) {
572 unsigned mb = mb_y * ctx->
m.
mb_width + mb_x;
580 for (i = 0; i < 8; i++) {
600 for (mb_y = 0; mb_y < ctx->
m.
mb_height; mb_y++) {
604 for (mb_x = 0; mb_x < ctx->
m.
mb_width; mb_x++) {
605 unsigned mb = mb_y * ctx->
m.
mb_width + mb_x;
611 offset += thread_size;
618 int mb_y = jobnr, mb_x, x, y;
619 int partial_last_row = (mb_y == ctx->
m.
mb_height - 1) &&
622 ctx = ctx->
thread[threadnr];
625 for (mb_x = 0; mb_x < ctx->
m.
mb_width; ++mb_x, pix += 16) {
626 unsigned mb = mb_y * ctx->
m.
mb_width + mb_x;
630 if (!partial_last_row && mb_x * 16 <= avctx->
width - 16) {
637 for (y = 0; y < bh; y++) {
638 for (x = 0; x < bw; x++) {
645 varc = (varc - (((unsigned)sum * sum) >> 8) + 128) >> 8;
651 int const linesize = ctx->
m.
linesize >> 1;
652 for (mb_x = 0; mb_x < ctx->
m.
mb_width; ++mb_x) {
653 uint16_t *pix = (uint16_t*)ctx->
thread[0]->
src[0] + ((mb_y << 4) * linesize) + (mb_x << 4);
654 unsigned mb = mb_y * ctx->
m.
mb_width + mb_x;
660 for (i = 0; i < 16; ++i) {
661 for (j = 0; j < 16; ++j) {
663 int const sample = (unsigned)pix[j] >> 6;
681 int lambda, up_step, down_step;
682 int last_lower = INT_MAX, last_higher = 0;
685 for (q = 1; q < avctx->
qmax; q++) {
695 if (lambda == last_higher) {
701 unsigned min = UINT_MAX;
704 for (q = 1; q < avctx->
qmax; q++) {
705 unsigned score = ctx->
mb_rc[q][mb].
bits*lambda+
716 bits = (bits+31)&~31;
727 if (bits < ctx->frame_bits) {
728 last_lower =
FFMIN(lambda, last_lower);
729 if (last_higher != 0)
730 lambda = (lambda+last_higher)>>1;
733 down_step =
FFMIN((int64_t)down_step*5, INT_MAX);
735 lambda =
FFMAX(1, lambda);
736 if (lambda == last_lower)
739 last_higher =
FFMAX(lambda, last_higher);
740 if (last_lower != INT_MAX)
741 lambda = (lambda+last_lower)>>1;
742 else if ((int64_t)lambda + up_step > INT_MAX)
746 up_step =
FFMIN((int64_t)up_step*5, INT_MAX);
761 int last_lower = INT_MAX;
774 bits = (bits+31)&~31;
780 if (bits < ctx->frame_bits) {
783 if (last_higher == qscale - 1) {
784 qscale = last_higher;
787 last_lower =
FFMIN(qscale, last_lower);
788 if (last_higher != 0)
789 qscale = (qscale+last_higher)>>1;
791 qscale -= down_step++;
796 if (last_lower == qscale + 1)
798 last_higher =
FFMAX(qscale, last_higher);
799 if (last_lower != INT_MAX)
800 qscale = (qscale+last_lower)>>1;
813 #define BUCKET_BITS 8
814 #define RADIX_PASSES 4
815 #define NBUCKETS (1 << BUCKET_BITS)
827 memset(buckets, 0,
sizeof(buckets[0][0]) *
RADIX_PASSES * NBUCKETS);
828 for (i = 0; i <
size; i++) {
829 int v = data[i].
value;
838 for (i = NBUCKETS - 1; i >= 0; i--)
839 buckets[j][i] = offset -= buckets[j][i];
840 assert(!buckets[j][0]);
848 for (i = 0; i <
size; i++) {
850 int pos = buckets[v]++;
910 for (i = 0; i < 3; i++) {
927 const AVFrame *frame,
int *got_packet)
943 for (i = 0; i < 3; i++) {
957 "picture could not fit ratecontrol constraints, increase qmax\n");
981 goto encode_coding_unit;
1033 .priv_class = &
class,
#define CODEC_FLAG_INTERLACED_DCT
static int dnxhd_encode_end(AVCodecContext *avctx)
static av_always_inline int dnxhd_calc_ac_bits(DNXHDEncContext *ctx, DCTELEM *block, int last_index)
#define MASK_ABS(mask, level)
static void radix_count(const RCCMPEntry *data, int size, int buckets[RADIX_PASSES][NBUCKETS])
void * av_malloc(size_t size)
Allocate a block of size bytes with alignment suitable for all memory accesses (including vectors if ...
av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)
int(* dct_quantize)(struct MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow)
static av_always_inline void dnxhd_get_blocks(DNXHDEncContext *ctx, int mb_x, int mb_y)
static int dnxhd_encode_init(AVCodecContext *avctx)
const uint8_t ff_zigzag_direct[64]
This structure describes decoded (raw) audio or video data.
av_cold int ff_dct_common_init(MpegEncContext *s)
const uint8_t * luma_weight
AVFrame * coded_frame
the picture in the bitstream
const CIDEntry ff_dnxhd_cid_table[]
const uint16_t * run_codes
enum AVPixelFormat pix_fmt
Pixel format, see AV_PIX_FMT_xxx.
void ff_convert_matrix(DSPContext *dsp, int(*qmat)[64], uint16_t(*qmat16)[2][64], const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
static av_always_inline void dnxhd_encode_dc(DNXHDEncContext *ctx, int diff)
int ff_dnxhd_find_cid(AVCodecContext *avctx, int bit_depth)
int intra_quant_bias
intra quantizer bias
int bits_per_raw_sample
Bits per sample/pixel of internal libavcodec pixel/sample format.
static const AVOption options[]
int mb_num
number of MBs of a picture
struct DNXHDEncContext * thread[MAX_THREADS]
int h263_aic
Advanded INTRA Coding (AIC)
int(* pix_sum)(uint8_t *pix, int line_size)
void av_freep(void *arg)
Free a memory block which has been allocated with av_malloc(z)() or av_realloc() and set the pointer ...
void ff_dnxhdenc_init_x86(DNXHDEncContext *ctx)
int ff_dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow)
static int get_bucket(int value, int shift)
static int dnxhd_encode_fast(AVCodecContext *avctx, DNXHDEncContext *ctx)
void(* get_pixels)(DCTELEM *block, const uint8_t *pixels, int line_size)
static void dnxhd_8bit_get_pixels_8x4_sym(DCTELEM *restrict block, const uint8_t *pixels, int line_size)
static void radix_sort_pass(RCCMPEntry *dst, const RCCMPEntry *data, int size, int buckets[NBUCKETS], int pass)
uint8_t idct_permutation[64]
idct input permutation.
const uint8_t * scantable
int interlaced_frame
The content of the picture is interlaced.
int mb_height
number of MBs horizontally & vertically
unsigned int coding_unit_size
#define LOCAL_ALIGNED_16(t, v,...)
const uint8_t * ac_index_flag
static int init(AVCodecParserContext *s)
int(* q_intra_matrix)[64]
precomputed matrix (combine qscale and DCT renorm)
static av_always_inline int dnxhd_switch_matrix(DNXHDEncContext *ctx, int i)
const uint16_t * ac_codes
static int dnxhd_calc_bits_thread(AVCodecContext *avctx, void *arg, int jobnr, int threadnr)
void av_free(void *ptr)
Free a memory block which has been allocated with av_malloc(z)() or av_realloc(). ...
int ff_dnxhd_get_cid_table(int cid)
int last_dc[3]
last DC values for MPEG1
static int dnxhd_find_qscale(DNXHDEncContext *ctx)
static int dnxhd_init_vlc(DNXHDEncContext *ctx)
static int dnxhd_init_rc(DNXHDEncContext *ctx)
int qmax
maximum quantizer
#define NULL_IF_CONFIG_SMALL(x)
Return NULL if CONFIG_SMALL is true, otherwise the argument without modification. ...
static int dnxhd_write_header(AVCodecContext *avctx, uint8_t *buf)
void av_log(void *avcl, int level, const char *fmt,...)
const char * name
Name of the codec implementation.
static av_always_inline void dnxhd_10bit_get_pixels_8x4_sym(DCTELEM *restrict block, const uint8_t *pixels, int line_size)
static void put_bits(PutBitContext *s, int n, unsigned int value)
Write up to 31 bits into a bitstream.
void(* clear_block)(DCTELEM *block)
void(* idct)(DCTELEM *block)
int flags
A combination of AV_PKT_FLAG values.
static int put_bits_count(PutBitContext *s)
planar YUV 4:2:2, 16bpp, (1 Cr & Cb sample per 2x1 Y samples)
const uint8_t * chroma_weight
enum AVPictureType pict_type
Picture type of the frame, see ?_TYPE below.
MpegEncContext m
Used for quantization dsp functions.
int intra_quant_bias
bias for the quantizer
int width
picture width / height.
uint16_t(* q_intra_matrix16)[2][64]
identical to the above but for MMX & these are not permutated, second 64 entries are bias ...
uint16_t(* qmatrix_l16)[2][64]
int quality
quality (between 1 (good) and FF_LAMBDA_MAX (bad))
int ff_alloc_packet(AVPacket *avpkt, int size)
Check AVPacket size and/or allocate data.
uint8_t * msip
Macroblock Scan Indexes Payload.
int mb_decision
macroblock decision mode
void(* get_pixels_8x4_sym)(DCTELEM *, const uint8_t *, int)
static av_always_inline void dnxhd_unquantize_c(DNXHDEncContext *ctx, DCTELEM *block, int n, int qscale, int last_index)
int thread_count
thread count is used to decide how many independent tasks should be passed to execute() ...
static void radix_sort(RCCMPEntry *data, int size)
static int dnxhd_encode_rdo(AVCodecContext *avctx, DNXHDEncContext *ctx)
int linesize[AV_NUM_DATA_POINTERS]
Size, in bytes, of the data for each picture/channel plane.
static int dnxhd_encode_picture(AVCodecContext *avctx, AVPacket *pkt, const AVFrame *frame, int *got_packet)
const uint8_t * ac_run_flag
main external API structure.
static void close(AVCodecParserContext *s)
struct DNXHDEncContext DNXHDEncContext
ScanTable intra_scantable
Describe the class of an AVClass context structure.
DSPContext dsp
pointers for accelerated dsp functions
int(* pix_norm1)(uint8_t *pix, int line_size)
static void dnxhd_setup_threads_slices(DNXHDEncContext *ctx)
const CIDEntry * cid_table
uint16_t(* qmatrix_c16)[2][64]
#define CODEC_CAP_SLICE_THREADS
static int dnxhd_10bit_dct_quantize(MpegEncContext *ctx, DCTELEM *block, int n, int qscale, int *overflow)
static void dnxhd_load_picture(DNXHDEncContext *ctx, const AVFrame *frame)
#define DNX10BIT_QMAT_SHIFT
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
#define AV_PIX_FMT_YUV422P10
struct AVCodecContext * avctx
PutBitContext pb
bit output
static int dnxhd_mb_var_thread(AVCodecContext *avctx, void *arg, int jobnr, int threadnr)
static int dnxhd_encode_thread(AVCodecContext *avctx, void *arg, int jobnr, int threadnr)
common internal api header.
static void flush_put_bits(PutBitContext *s)
Pad the end of the output stream with zeros.
void(* fdct)(DCTELEM *block)
static int dnxhd_init_qmat(DNXHDEncContext *ctx, int lbias, int cbias)
static void init_put_bits(PutBitContext *s, uint8_t *buffer, int buffer_size)
Initialize the PutBitContext s.
int top_field_first
If the content is interlaced, is top field displayed first.
int(* execute2)(struct AVCodecContext *c, int(*func)(struct AVCodecContext *c2, void *arg, int jobnr, int threadnr), void *arg2, int *ret, int count)
The codec may call this to execute several independent things.
static av_always_inline void dnxhd_encode_block(DNXHDEncContext *ctx, DCTELEM *block, int last_index, int n)
int key_frame
1 -> keyframe, 0-> not
int linesize
line size, in bytes, may be different from width
#define FF_DEFAULT_QUANT_BIAS
static int first_field(int fd)
int uvlinesize
line size, for chroma in bytes, may be different from width
AVPixelFormat
Pixel format.
This structure stores compressed data.
#define FF_MB_DECISION_RD
#define FF_ALLOCZ_OR_GOTO(ctx, p, size, label)
static av_always_inline int dnxhd_ssd_block(DCTELEM *qblock, DCTELEM *block)