26 #define RC_VARIANCE 1 // use variance or ssd for fast rc
38 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
39 #define DNX10BIT_QMAT_SHIFT 18 // The largest value that will not lead to overflow for 10bit samples.
47 #define LAMBDA_FRAC_BITS 10
52 for (i = 0; i < 4; i++) {
53 block[0] = pixels[0]; block[1] = pixels[1];
54 block[2] = pixels[2]; block[3] = pixels[3];
55 block[4] = pixels[4]; block[5] = pixels[5];
56 block[6] = pixels[6]; block[7] = pixels[7];
60 memcpy(block, block - 8,
sizeof(*block) * 8);
61 memcpy(block + 8, block - 16,
sizeof(*block) * 8);
62 memcpy(block + 16, block - 24,
sizeof(*block) * 8);
63 memcpy(block + 24, block - 32,
sizeof(*block) * 8);
72 for (i = 0; i < 4; i++) {
73 memcpy(block + i * 8, pixels + i * line_size, 8 *
sizeof(*block));
74 memcpy(block - (i+1) * 8, pixels + i * line_size, 8 *
sizeof(*block));
79 int n,
int qscale,
int *overflow)
83 int last_non_zero = 0;
89 block[0] = (block[0] + 2) >> 2;
91 for (i = 1; i < 64; ++i) {
93 int sign = block[j] >> 31;
94 int level = (block[j] ^ sign) - sign;
96 block[j] = (level ^ sign) - sign;
101 return last_non_zero;
116 for (level = -max_level; level < max_level; level++) {
117 for (run = 0; run < 2; run++) {
118 int index = (level<<1)|run;
119 int sign, offset = 0, alevel =
level;
123 offset = (alevel-1)>>6;
126 for (j = 0; j < 257; j++) {
141 assert(!alevel || j < 257);
148 for (i = 0; i < 62; i++) {
162 uint16_t weight_matrix[64] = {1,};
173 for (i = 1; i < 64; i++) {
179 for (i = 1; i < 64; i++) {
186 for (qscale = 1; qscale <= ctx->
m.
avctx->
qmax; qscale++) {
187 for (i = 0; i < 64; i++) {
195 for (qscale = 1; qscale <= ctx->
m.
avctx->
qmax; qscale++) {
196 for (i = 1; i < 64; i++) {
235 int i,
index, bit_depth;
337 const uint8_t header_prefix[5] = { 0x00,0x00,0x02,0x80,0x01 };
341 memcpy(buf, header_prefix, 5);
361 ctx->
msip = buf + 0x170;
380 int last_non_zero = 0;
386 for (i = 1; i <= last_index; i++) {
390 int run_level = i - last_non_zero - 1;
391 int rlevel = (slevel<<1)|!!run_level;
409 for (i = 1; i <= last_index; i++) {
414 level = (1-2*
level) * qscale * weight_matrix[i];
416 if (weight_matrix[i] != 8)
420 if (weight_matrix[i] != 32)
426 level = (2*level+1) * qscale * weight_matrix[i];
428 if (weight_matrix[i] != 8)
432 if (weight_matrix[i] != 32)
446 for (i = 0; i < 64; i++)
447 score += (block[i] - qblock[i]) * (block[i] - qblock[i]);
453 int last_non_zero = 0;
456 for (i = 1; i <= last_index; i++) {
460 int run_level = i - last_non_zero - 1;
471 const int bw = 1 << bs;
518 int mb_y = jobnr, mb_x;
521 ctx = ctx->
thread[threadnr];
527 for (mb_x = 0; mb_x < ctx->
m.
mb_width; mb_x++) {
528 unsigned mb = mb_y * ctx->
m.
mb_width + mb_x;
536 for (i = 0; i < 8; i++) {
537 int16_t *src_block = ctx->
blocks[i];
538 int overflow, nbits, diff, last_index;
549 assert(nbits < ctx->cid_table->bit_depth + 4);
569 int mb_y = jobnr, mb_x;
570 ctx = ctx->
thread[threadnr];
576 for (mb_x = 0; mb_x < ctx->
m.
mb_width; mb_x++) {
577 unsigned mb = mb_y * ctx->
m.
mb_width + mb_x;
585 for (i = 0; i < 8; i++) {
605 for (mb_y = 0; mb_y < ctx->
m.
mb_height; mb_y++) {
609 for (mb_x = 0; mb_x < ctx->
m.
mb_width; mb_x++) {
610 unsigned mb = mb_y * ctx->
m.
mb_width + mb_x;
616 offset += thread_size;
623 int mb_y = jobnr, mb_x, x, y;
624 int partial_last_row = (mb_y == ctx->
m.
mb_height - 1) &&
627 ctx = ctx->
thread[threadnr];
630 for (mb_x = 0; mb_x < ctx->
m.
mb_width; ++mb_x, pix += 16) {
631 unsigned mb = mb_y * ctx->
m.
mb_width + mb_x;
635 if (!partial_last_row && mb_x * 16 <= avctx->
width - 16) {
642 for (y = 0; y < bh; y++) {
643 for (x = 0; x < bw; x++) {
650 varc = (varc - (((unsigned)sum * sum) >> 8) + 128) >> 8;
656 int const linesize = ctx->
m.
linesize >> 1;
657 for (mb_x = 0; mb_x < ctx->
m.
mb_width; ++mb_x) {
658 uint16_t *pix = (uint16_t*)ctx->
thread[0]->
src[0] + ((mb_y << 4) * linesize) + (mb_x << 4);
659 unsigned mb = mb_y * ctx->
m.
mb_width + mb_x;
665 for (i = 0; i < 16; ++i) {
666 for (j = 0; j < 16; ++j) {
668 int const sample = (unsigned)pix[j] >> 6;
686 int lambda, up_step, down_step;
687 int last_lower = INT_MAX, last_higher = 0;
690 for (q = 1; q < avctx->
qmax; q++) {
700 if (lambda == last_higher) {
706 unsigned min = UINT_MAX;
709 for (q = 1; q < avctx->
qmax; q++) {
710 unsigned score = ctx->
mb_rc[q][mb].
bits*lambda+
721 bits = (bits+31)&~31;
732 if (bits < ctx->frame_bits) {
733 last_lower =
FFMIN(lambda, last_lower);
734 if (last_higher != 0)
735 lambda = (lambda+last_higher)>>1;
738 down_step =
FFMIN((int64_t)down_step*5, INT_MAX);
740 lambda =
FFMAX(1, lambda);
741 if (lambda == last_lower)
744 last_higher =
FFMAX(lambda, last_higher);
745 if (last_lower != INT_MAX)
746 lambda = (lambda+last_lower)>>1;
747 else if ((int64_t)lambda + up_step > INT_MAX)
751 up_step =
FFMIN((int64_t)up_step*5, INT_MAX);
766 int last_lower = INT_MAX;
779 bits = (bits+31)&~31;
785 if (bits < ctx->frame_bits) {
788 if (last_higher == qscale - 1) {
789 qscale = last_higher;
792 last_lower =
FFMIN(qscale, last_lower);
793 if (last_higher != 0)
794 qscale = (qscale+last_higher)>>1;
796 qscale -= down_step++;
801 if (last_lower == qscale + 1)
803 last_higher =
FFMAX(qscale, last_higher);
804 if (last_lower != INT_MAX)
805 qscale = (qscale+last_lower)>>1;
818 #define BUCKET_BITS 8
819 #define RADIX_PASSES 4
820 #define NBUCKETS (1 << BUCKET_BITS)
832 memset(buckets, 0,
sizeof(buckets[0][0]) *
RADIX_PASSES * NBUCKETS);
833 for (i = 0; i <
size; i++) {
834 int v = data[i].
value;
843 for (i = NBUCKETS - 1; i >= 0; i--)
844 buckets[j][i] = offset -= buckets[j][i];
845 assert(!buckets[j][0]);
853 for (i = 0; i <
size; i++) {
855 int pos = buckets[v]++;
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;
1035 .priv_class = &
class,
static av_always_inline int dnxhd_ssd_block(int16_t *qblock, int16_t *block)
#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)
static av_always_inline void dnxhd_get_blocks(DNXHDEncContext *ctx, int mb_x, int mb_y)
This structure describes decoded (raw) audio or video data.
av_cold int ff_dct_common_init(MpegEncContext *s)
void(* get_pixels)(int16_t *block, const uint8_t *pixels, int line_size)
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)
static av_cold int dnxhd_init_vlc(DNXHDEncContext *ctx)
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]
static av_always_inline int dnxhd_calc_ac_bits(DNXHDEncContext *ctx, int16_t *block, int last_index)
int h263_aic
Advanded INTRA Coding (AIC)
int(* pix_sum)(uint8_t *pix, int line_size)
Macro definitions for various function/variable attributes.
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(* get_pixels_8x4_sym)(int16_t *, const uint8_t *, int)
AVFrame * av_frame_alloc(void)
Allocate an AVFrame and set its fields to default values.
static int get_bucket(int value, int shift)
static int dnxhd_encode_fast(AVCodecContext *avctx, DNXHDEncContext *ctx)
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.
av_cold void ff_dnxhdenc_init_x86(DNXHDEncContext *ctx)
int mb_height
number of MBs horizontally & vertically
static av_cold int dnxhd_init_qmat(DNXHDEncContext *ctx, int lbias, int cbias)
unsigned int coding_unit_size
high precision timer, useful to profile code
const uint8_t * ac_index_flag
static av_cold int dnxhd_encode_end(AVCodecContext *avctx)
#define AV_PKT_FLAG_KEY
The packet contains a keyframe.
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)
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
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)
void av_frame_free(AVFrame **frame)
Free the frame and any dynamically allocated objects in it, e.g.
int qmax
maximum quantizer
#define NULL_IF_CONFIG_SMALL(x)
Return NULL if CONFIG_SMALL is true, otherwise the argument without modification. ...
static av_always_inline void dnxhd_unquantize_c(DNXHDEncContext *ctx, int16_t *block, int n, int qscale, int last_index)
#define AV_LOG_DEBUG
Stuff which is only useful for libav* developers.
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 void put_bits(PutBitContext *s, int n, unsigned int value)
Write up to 31 bits into a bitstream.
void(* idct)(int16_t *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
common internal API header
enum AVPictureType pict_type
Picture type of the frame.
MpegEncContext m
Used for quantization dsp functions.
static av_always_inline void dnxhd_encode_block(DNXHDEncContext *ctx, int16_t *block, int last_index, int n)
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.
#define CODEC_FLAG_INTERLACED_DCT
Use interlaced DCT.
uint8_t * msip
Macroblock Scan Indexes Payload.
int mb_decision
macroblock decision mode
int thread_count
thread count is used to decide how many independent tasks should be passed to execute() ...
int ff_dct_quantize_c(MpegEncContext *s, int16_t *block, int n, int qscale, int *overflow)
static void radix_sort(RCCMPEntry *data, int size)
static int dnxhd_encode_rdo(AVCodecContext *avctx, DNXHDEncContext *ctx)
Libavcodec external API header.
static int dnxhd_10bit_dct_quantize(MpegEncContext *ctx, int16_t *block, int n, int qscale, int *overflow)
ptrdiff_t linesize
line size, in bytes, may be different from width
int linesize[AV_NUM_DATA_POINTERS]
For video, size in bytes of each picture line.
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)
ScanTable intra_scantable
#define FF_DEFAULT_QUANT_BIAS
static av_cold int dnxhd_init_rc(DNXHDEncContext *ctx)
Describe the class of an AVClass context structure.
#define FF_MB_DECISION_RD
rate distortion
ptrdiff_t uvlinesize
line size, for chroma in bytes, may be different from width
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]
static av_always_inline void dnxhd_10bit_get_pixels_8x4_sym(int16_t *restrict block, const uint8_t *pixels, int line_size)
static void dnxhd_load_picture(DNXHDEncContext *ctx, const AVFrame *frame)
#define AV_PIX_FMT_YUV422P10
#define DNX10BIT_QMAT_SHIFT
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
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)
static av_cold int dnxhd_encode_init(AVCodecContext *avctx)
#define CODEC_CAP_SLICE_THREADS
Codec supports slice-based (or partition-based) multithreading.
common internal api header.
const uint8_t ff_zigzag_direct[64]
static void flush_put_bits(PutBitContext *s)
Pad the end of the output stream with zeros.
void(* clear_block)(int16_t *block)
static void init_put_bits(PutBitContext *s, uint8_t *buffer, int buffer_size)
Initialize the PutBitContext s.
static av_cold int init(AVCodecParserContext *s)
int top_field_first
If the content is interlaced, is top field displayed first.
void(* fdct)(int16_t *block)
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 void dnxhd_8bit_get_pixels_8x4_sym(int16_t *restrict block, const uint8_t *pixels, int line_size)
int key_frame
1 -> keyframe, 0-> not
#define FF_QP2LAMBDA
factor to convert from H.263 QP to lambda
#define LOCAL_ALIGNED_16(t, v,...)
static int first_field(int fd)
int(* dct_quantize)(struct MpegEncContext *s, int16_t *block, int n, int qscale, int *overflow)
AVPixelFormat
Pixel format.
This structure stores compressed data.
#define FF_ALLOCZ_OR_GOTO(ctx, p, size, label)