26 #define RC_VARIANCE 1 // use variance or ssd for fast rc
28 #include "libavutil/attributes.h"
30 #include "libavutil/opt.h"
37 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
38 #define DNX10BIT_QMAT_SHIFT 18 // The largest value that will not lead to overflow for 10bit samples.
52 #define LAMBDA_FRAC_BITS 10
57 for (i = 0; i < 4; i++) {
58 block[0] = pixels[0]; block[1] = pixels[1];
59 block[2] = pixels[2]; block[3] = pixels[3];
60 block[4] = pixels[4]; block[5] = pixels[5];
61 block[6] = pixels[6]; block[7] = pixels[7];
65 memcpy(block, block - 8,
sizeof(*block) * 8);
66 memcpy(block + 8, block - 16,
sizeof(*block) * 8);
67 memcpy(block + 16, block - 24,
sizeof(*block) * 8);
68 memcpy(block + 24, block - 32,
sizeof(*block) * 8);
74 const uint16_t* pixels16 = (
const uint16_t*)pixels;
77 for (i = 0; i < 4; i++) {
78 block[0] = pixels16[0]; block[1] = pixels16[1];
79 block[2] = pixels16[2]; block[3] = pixels16[3];
80 block[4] = pixels16[4]; block[5] = pixels16[5];
81 block[6] = pixels16[6]; block[7] = pixels16[7];
82 pixels16 += line_size;
85 memcpy(block, block - 8,
sizeof(*block) * 8);
86 memcpy(block + 8, block - 16,
sizeof(*block) * 8);
87 memcpy(block + 16, block - 24,
sizeof(*block) * 8);
88 memcpy(block + 24, block - 32,
sizeof(*block) * 8);
92 int n,
int qscale,
int *overflow)
96 int last_non_zero = 0;
102 block[0] = (block[0] + 2) >> 2;
104 for (i = 1; i < 64; ++i) {
105 int j = scantable[i];
106 int sign = block[j] >> 31;
107 int level = (block[j] ^ sign) - sign;
109 block[j] = (level ^ sign) - sign;
114 return last_non_zero;
129 for (level = -max_level; level < max_level; level++) {
130 for (run = 0; run < 2; run++) {
131 int index = (level<<1)|run;
136 offset = (alevel-1)>>6;
139 for (j = 0; j < 257; j++) {
161 for (i = 0; i < 62; i++) {
175 uint16_t weight_matrix[64] = {1,};
186 for (i = 1; i < 64; i++) {
192 for (i = 1; i < 64; i++) {
199 for (qscale = 1; qscale <= ctx->
m.
avctx->
qmax; qscale++) {
200 for (i = 0; i < 64; i++) {
208 for (qscale = 1; qscale <= ctx->
m.
avctx->
qmax; qscale++) {
209 for (i = 1; i < 64; i++) {
269 av_log(avctx,
AV_LOG_ERROR,
"video parameters incompatible with DNxHD. Valid DNxHD profiles:\n");
355 static const uint8_t header_prefix[5] = { 0x00,0x00,0x02,0x80,0x01 };
359 memcpy(buf, header_prefix, 5);
379 ctx->
msip = buf + 0x170;
398 int last_non_zero = 0;
404 for (i = 1; i <= last_index; i++) {
408 int run_level = i - last_non_zero - 1;
409 int rlevel = (slevel<<1)|!!run_level;
427 for (i = 1; i <= last_index; i++) {
432 level = (1-2*
level) * qscale * weight_matrix[i];
434 if (weight_matrix[i] != 8)
438 if (weight_matrix[i] != 32)
444 level = (2*level+1) * qscale * weight_matrix[i];
446 if (weight_matrix[i] != 8)
450 if (weight_matrix[i] != 32)
464 for (i = 0; i < 64; i++)
465 score += (block[i] - qblock[i]) * (block[i] - qblock[i]);
471 int last_non_zero = 0;
474 for (i = 1; i <= last_index; i++) {
478 int run_level = i - last_non_zero - 1;
489 const int bw = 1 << bs;
522 const static uint8_t component[8]={0,0,1,2,0,0,1,2};
529 int mb_y = jobnr, mb_x;
532 ctx = ctx->
thread[threadnr];
538 for (mb_x = 0; mb_x < ctx->
m.
mb_width; mb_x++) {
547 for (i = 0; i < 8; i++) {
548 int16_t *src_block = ctx->
blocks[i];
549 int overflow, nbits, diff, last_index;
580 int mb_y = jobnr, mb_x;
581 ctx = ctx->
thread[threadnr];
587 for (mb_x = 0; mb_x < ctx->
m.
mb_width; mb_x++) {
596 for (i = 0; i < 8; i++) {
599 int last_index = ctx->
m.
dct_quantize(&ctx->
m, block, 4&(2*i), qscale, &overflow);
615 for (mb_y = 0; mb_y < ctx->
m.
mb_height; mb_y++) {
619 for (mb_x = 0; mb_x < ctx->
m.
mb_width; mb_x++) {
626 offset += thread_size;
633 int mb_y = jobnr, mb_x, x,
y;
634 int partial_last_row = (mb_y == ctx->
m.
mb_height - 1) &&
637 ctx = ctx->
thread[threadnr];
640 for (mb_x = 0; mb_x < ctx->
m.
mb_width; ++mb_x, pix += 16) {
645 if (!partial_last_row && mb_x * 16 <= avctx->
width - 16) {
652 for (y = 0; y < bh; y++) {
653 for (x = 0; x < bw; x++) {
660 varc = (varc - (((unsigned)sum * sum) >> 8) + 128) >> 8;
666 int const linesize = ctx->
m.
linesize >> 1;
667 for (mb_x = 0; mb_x < ctx->
m.
mb_width; ++mb_x) {
668 uint16_t *pix = (uint16_t*)ctx->
thread[0]->
src[0] + ((mb_y << 4) * linesize) + (mb_x << 4);
675 for (i = 0; i < 16; ++i) {
676 for (j = 0; j < 16; ++j) {
678 int const sample = (unsigned)pix[j] >> 6;
696 int lambda, up_step, down_step;
697 int last_lower = INT_MAX, last_higher = 0;
700 for (q = 1; q < avctx->
qmax; q++) {
710 if (lambda == last_higher) {
716 unsigned min = UINT_MAX;
719 for (q = 1; q < avctx->
qmax; q++) {
731 bits = (bits+31)&~31;
742 if (bits < ctx->frame_bits) {
743 last_lower =
FFMIN(lambda, last_lower);
744 if (last_higher != 0)
745 lambda = (lambda+last_higher)>>1;
748 down_step =
FFMIN((int64_t)down_step*5, INT_MAX);
750 lambda =
FFMAX(1, lambda);
751 if (lambda == last_lower)
754 last_higher =
FFMAX(lambda, last_higher);
755 if (last_lower != INT_MAX)
756 lambda = (lambda+last_lower)>>1;
757 else if ((int64_t)lambda + up_step > INT_MAX)
761 up_step =
FFMIN((int64_t)up_step*5, INT_MAX);
776 int last_lower = INT_MAX;
789 bits = (bits+31)&~31;
795 if (bits < ctx->frame_bits) {
798 if (last_higher == qscale - 1) {
799 qscale = last_higher;
802 last_lower =
FFMIN(qscale, last_lower);
803 if (last_higher != 0)
804 qscale = (qscale+last_higher)>>1;
806 qscale -= down_step++;
811 if (last_lower == qscale + 1)
813 last_higher =
FFMAX(qscale, last_higher);
814 if (last_lower != INT_MAX)
815 qscale = (qscale+last_lower)>>1;
828 #define BUCKET_BITS 8
829 #define RADIX_PASSES 4
830 #define NBUCKETS (1 << BUCKET_BITS)
842 memset(buckets, 0,
sizeof(buckets[0][0]) *
RADIX_PASSES * NBUCKETS);
843 for (i = 0; i <
size; i++) {
853 for (i = NBUCKETS - 1; i >= 0; i--)
854 buckets[j][i] = offset -= buckets[j][i];
863 for (i = 0; i <
size; i++) {
865 int pos = buckets[
v]++;
925 for (i = 0; i < 3; i++) {
956 for (i = 0; i < 3; i++) {
970 "picture could not fit ratecontrol constraints, increase qmax\n");
994 goto encode_coding_unit;
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])
int ff_alloc_packet2(AVCodecContext *avctx, AVPacket *avpkt, int64_t size)
Check AVPacket size and/or allocate data.
const char const char void * val
static av_always_inline void dnxhd_get_blocks(DNXHDEncContext *ctx, int mb_x, int mb_y)
static int shift(int a, int b)
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)
static void put_bits(Jpeg2000EncoderContext *s, int val, int n)
put n times val bit
const uint8_t * luma_weight
#define LIBAVUTIL_VERSION_INT
static av_cold int init(AVCodecContext *avctx)
uint16_t(* q_chroma_intra_matrix16)[2][64]
planar YUV 4:2:2, 16bpp, (1 Cr & Cb sample per 2x1 Y samples)
const CIDEntry ff_dnxhd_cid_table[]
const uint16_t * run_codes
static const AVClass dnxhd_class
void av_log(void *avcl, int level, const char *fmt,...) av_printf_format(3
Send the specified message to the log if the level is less than or equal to the current av_log_level...
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.
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)
void av_freep(void *ptr)
Free a memory block which has been allocated with av_malloc(z)() or av_realloc() and set the pointer ...
#define FF_MB_DECISION_RD
rate distortion
void(* get_pixels_8x4_sym)(int16_t *, const uint8_t *, int)
const char * av_default_item_name(void *ctx)
Return the context name.
const char * class_name
The name of the class; usually it is the same name as the context structure type to which the AVClass...
static int get_bucket(int value, int shift)
static int first_field(const struct video_data *s, int fd)
static const uint8_t offset[511][2]
static av_cold int end(AVCodecContext *avctx)
static int dnxhd_encode_fast(AVCodecContext *avctx, DNXHDEncContext *ctx)
const char * name
Name of the codec implementation.
int(* q_chroma_intra_matrix)[64]
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
static av_cold int dnxhd_init_qmat(DNXHDEncContext *ctx, int lbias, int cbias)
unsigned int coding_unit_size
const OptionDef options[]
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)
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)
Libavcodec external API header.
AVPixelFormat
Pixel format.
void(* idct)(int16_t *block)
int flags
A combination of AV_PKT_FLAG values.
static int put_bits_count(PutBitContext *s)
const uint8_t * chroma_weight
common internal API header
enum AVPictureType pict_type
Picture type of the frame.
static const AVCodecDefault dnxhd_defaults[]
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.
void * av_malloc(size_t size) av_malloc_attrib 1(1)
Allocate a block of size bytes with alignment suitable for all memory accesses (including vectors if ...
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))
#define CODEC_FLAG_INTERLACED_DCT
Use interlaced DCT.
uint8_t * msip
Macroblock Scan Indexes Payload.
int mb_decision
macroblock decision mode
void ff_dnxhd_print_profiles(AVCodecContext *avctx, int loglevel)
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)
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
static int dnxhd_encode_picture(AVCodecContext *avctx, AVPacket *pkt, const AVFrame *frame, int *got_packet)
main external API structure.
static void close(AVCodecParserContext *s)
ScanTable intra_scantable
int ff_dct_encode_init(MpegEncContext *s)
static av_cold int dnxhd_init_rc(DNXHDEncContext *ctx)
void ff_dnxhdenc_init_x86(DNXHDEncContext *ctx)
#define FF_DEFAULT_QUANT_BIAS
Describe the class of an AVClass context structure.
AVFrame * coded_frame
the picture in the bitstream
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)
int linesize[AV_NUM_DATA_POINTERS]
For video, size in bytes of each picture line.
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 *av_restrict block, const uint8_t *pixels, int line_size)
#define av_assert1(cond)
assert() equivalent, that does not lie in speed critical code.
static void dnxhd_load_picture(DNXHDEncContext *ctx, const AVFrame *frame)
static void dnxhd_8bit_get_pixels_8x4_sym(int16_t *av_restrict block, const uint8_t *pixels, int line_size)
#define DNX10BIT_QMAT_SHIFT
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.
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.
int key_frame
1 -> keyframe, 0-> not
#define FF_QP2LAMBDA
factor to convert from H.263 QP to lambda
const CIDEntry * cid_table
#define LOCAL_ALIGNED_16(t, v,...)
#define AV_PIX_FMT_YUV422P10
#define av_assert0(cond)
assert() equivalent, that is always enabled.
int(* dct_quantize)(struct MpegEncContext *s, int16_t *block, int n, int qscale, int *overflow)
This structure stores compressed data.
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
#define FF_ALLOCZ_OR_GOTO(ctx, p, size, label)