112 #define QUALITY_THRESHOLD 100
113 #define THRESHOLD_MULTIPLIER 0.6
117 int threshold,
int lambda,
int intra)
119 int count, y, x, i, j,
split, best_mean, best_score, best_count;
121 int block_sum[7] = { 0, 0, 0, 0, 0, 0 };
122 int w = 2 << (level + 2 >> 1);
123 int h = 2 << (level + 1 >> 1);
125 int16_t
block[7][256];
126 const int8_t *codebook_sum, *codebook;
127 const uint16_t(*mean_vlc)[2];
128 const uint8_t(*multistage_vlc)[2];
137 for (y = 0; y < h; y++) {
138 for (x = 0; x < w; x++) {
139 int v = src[x + y *
stride];
140 block[0][x + w * y] = v;
150 for (y = 0; y < h; y++) {
151 for (x = 0; x < w; x++) {
153 block[0][x + w * y] = v;
161 best_score -= (int)((
unsigned)block_sum[0] * block_sum[0] >> (level + 3));
162 best_mean = block_sum[0] + (size >> 1) >> (level + 3);
165 for (count = 1; count < 7; count++) {
166 int best_vector_score = INT_MAX;
167 int best_vector_sum = -999, best_vector_mean = -999;
168 const int stage = count - 1;
169 const int8_t *vector;
171 for (i = 0; i < 16; i++) {
172 int sum = codebook_sum[stage * 16 + i];
173 int sqr, diff, score;
175 vector = codebook + stage * size * 16 + i *
size;
177 diff = block_sum[stage] - sum;
178 score = sqr - (diff * (int64_t)diff >> (level + 3));
179 if (score < best_vector_score) {
180 int mean = diff + (size >> 1) >> (level + 3);
181 assert(mean > -300 && mean < 300);
182 mean = av_clip(mean, intra ? 0 : -256, 255);
183 best_vector_score = score;
184 best_vector[stage] = i;
185 best_vector_sum = sum;
186 best_vector_mean = mean;
189 assert(best_vector_mean != -999);
190 vector = codebook + stage * size * 16 + best_vector[stage] *
size;
191 for (j = 0; j <
size; j++)
192 block[stage + 1][j] = block[stage][j] - vector[j];
193 block_sum[stage + 1] = block_sum[stage] - best_vector_sum;
194 best_vector_score += lambda *
196 multistage_vlc[1 + count][1]
197 + mean_vlc[best_vector_mean][1]);
199 if (best_vector_score < best_score) {
200 best_score = best_vector_score;
202 best_mean = best_vector_mean;
208 if (best_score > threshold && level) {
210 int offset = level & 1 ? stride * h / 2 : w / 2;
213 for (i = level - 1; i >= 0; i--)
215 score +=
encode_block(s, src, ref, decoded, stride, level - 1,
216 threshold >> 1, lambda, intra);
217 score +=
encode_block(s, src + offset, ref + offset, decoded + offset,
218 stride, level - 1, threshold >> 1, lambda, intra);
221 if (score < best_score) {
225 for (i = level - 1; i >= 0; i--)
233 assert(best_mean >= 0 && best_mean < 256 || !intra);
234 assert(best_mean >= -256 && best_mean < 256);
235 assert(best_count >= 0 && best_count < 7);
236 assert(level < 4 || best_count == 0);
240 multistage_vlc[1 + best_count][1],
241 multistage_vlc[1 + best_count][0]);
243 mean_vlc[best_mean][0]);
245 for (i = 0; i < best_count; i++) {
246 assert(best_vector[i] >= 0 && best_vector[i] < 16);
250 for (y = 0; y < h; y++)
251 for (x = 0; x < w; x++)
252 decoded[x + y * stride] = src[x + y * stride] -
253 block[best_count][x + w * y] +
261 unsigned char *src_plane,
262 unsigned char *ref_plane,
263 unsigned char *decoded_plane,
268 int block_width, block_height;
277 for (level = 4; level >= 0; level--)
280 block_width = (width + 15) / 16;
281 block_height = (height + 15) / 16;
315 block_height * 2 + 2) *
316 2 *
sizeof(int16_t));
318 (block_height + 2) + 1) *
319 2 *
sizeof(int16_t));
338 for (y = 0; y < block_height; y++) {
342 for (i = 0; i < 16 && i + 16 * y <
height; i++) {
343 memcpy(&src[i * stride], &src_plane[(i + 16 * y) * src_stride],
345 for (x = width; x < 16 * block_width; x++)
346 src[i * stride + x] = src[i * stride + x - 1];
348 for (; i < 16 && i + 16 * y < 16 * block_height; i++)
349 memcpy(&src[i * stride], &src[(i - 1) * stride],
352 for (x = 0; x < block_width; x++) {
368 for (y = 0; y < block_height; y++) {
369 for (i = 0; i < 16 && i + 16 * y <
height; i++) {
370 memcpy(&src[i * stride], &src_plane[(i + 16 * y) * src_stride],
372 for (x = width; x < 16 * block_width; x++)
373 src[i * stride + x] = src[i * stride + x - 1];
375 for (; i < 16 && i + 16 * y < 16 * block_height; i++)
376 memcpy(&src[i * stride], &src[(i - 1) * stride], 16 * block_width);
379 for (x = 0; x < block_width; x++) {
380 uint8_t reorder_buffer[3][6][7 * 32];
382 int offset = y * 16 * stride + x * 16;
383 uint8_t *decoded = decoded_plane + offset;
384 uint8_t *ref = ref_plane + offset;
385 int score[4] = { 0, 0, 0, 0 }, best;
401 for (i = 0; i < 6; i++)
407 score[0] = vlc[1] * lambda;
411 for (i = 0; i < 6; i++) {
422 int mx, my, pred_x, pred_y, dxy;
428 for (i = 0; i < 6; i++)
437 assert(mx >= -32 && mx <= 31);
438 assert(my >= -32 && my <= 31);
439 assert(pred_x >= -32 && pred_x <= 31);
440 assert(pred_y >= -32 && pred_y <= 31);
446 dxy = (mx & 1) + 2 * (my & 1);
454 decoded, stride, 5, 64, lambda, 0);
455 best = score[1] <= score[0];
458 score[2] = s->
dsp.
sse[0](
NULL, src + 16 * x, ref,
460 score[2] += vlc[1] * lambda;
461 if (score[2] < score[best] && mx == 0 && my == 0) {
464 for (i = 0; i < 6; i++)
471 for (i = 0; i < 6; i++) {
489 for (i = 5; i >= 0; i--)
534 const AVFrame *pict,
int *got_packet)
571 for (i = 0; i < 3; i++)
612 for (i = 0; i < 3; i++) {
uint8_t * scratchpad
data area for the ME algo, so that the ME does not need to malloc/free
void * av_malloc(size_t size)
Allocate a block of size bytes with alignment suitable for all memory accesses (including vectors if ...
void ff_init_block_index(MpegEncContext *s)
op_pixels_func put_pixels_tab[4][4]
Halfpel motion compensation with rounding (a+b+1)>>1.
av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)
This structure describes decoded (raw) audio or video data.
int16_t(* p_mv_table)[2]
MV table (1MV per MB) p-frame encoding.
void ff_estimate_p_frame_motion(MpegEncContext *s, int mb_x, int mb_y)
uint8_t * mb_mean
Table for MB luminance.
#define THRESHOLD_MULTIPLIER
void ff_h263_encode_init(MpegEncContext *s)
AVFrame * coded_frame
the picture in the bitstream
uint16_t * mb_var
Table for MB variances.
static int encode_block(SVQ1Context *s, uint8_t *src, uint8_t *ref, uint8_t *decoded, int stride, int level, int threshold, int lambda, int intra)
#define CANDIDATE_MB_TYPE_INTRA
#define FF_MIN_BUFFER_SIZE
void avpriv_copy_bits(PutBitContext *pb, const uint8_t *src, int length)
Copy the content of src to the bitstream.
enum AVPixelFormat pix_fmt
Pixel format, see AV_PIX_FMT_xxx.
uint32_t * score_map
map to store the scores
int16_t * ff_h263_pred_motion(MpegEncContext *s, int block, int dir, int *px, int *py)
static const int8_t svq1_intra_codebook_sum[4][16 *6]
void av_freep(void *arg)
Free a memory block which has been allocated with av_malloc(z)() or av_realloc() and set the pointer ...
static av_cold int svq1_encode_init(AVCodecContext *avctx)
static int svq1_encode_plane(SVQ1Context *s, int plane, unsigned char *src_plane, unsigned char *ref_plane, unsigned char *decoded_plane, int width, int height, int src_stride, int stride)
const int8_t *const ff_svq1_inter_codebooks[6]
const struct svq1_frame_size ff_svq1_frame_size_table[7]
Picture current_picture
copy of the current picture structure.
int mb_height
number of MBs horizontally & vertically
struct SVQ1Context SVQ1Context
static void ff_update_block_index(MpegEncContext *s)
static int init(AVCodecParserContext *s)
int av_new_packet(AVPacket *pkt, int size)
Allocate the payload of a packet and initialize its fields with default values.
static void svq1_write_header(SVQ1Context *s, int frame_type)
#define NULL_IF_CONFIG_SMALL(x)
Return NULL if CONFIG_SMALL is true, otherwise the argument without modification. ...
#define CANDIDATE_MB_TYPE_INTER
void av_log(void *avcl, int level, const char *fmt,...)
const char * name
Name of the codec implementation.
uint16_t * mb_type
Table for candidate MB types for encoding.
static void put_bits(PutBitContext *s, int n, unsigned int value)
Write up to 31 bits into a bitstream.
int flags
A combination of AV_PKT_FLAG values.
Sorenson Vector Quantizer #1 (SVQ1) video codec.
static int put_bits_count(PutBitContext *s)
static char * split(char *message, char delim)
void ff_h263_encode_motion(MpegEncContext *s, int val, int f_code)
enum AVPictureType pict_type
Picture type of the frame, see ?_TYPE below.
int me_method
ME algorithm.
Picture new_picture
copy of the source picture structure for encoding.
const uint8_t ff_svq1_block_type_vlc[4][2]
int width
picture width / height.
Picture * current_picture_ptr
pointer to the current picture
int ff_get_buffer(AVCodecContext *avctx, AVFrame *frame)
Get a buffer for a frame.
int quality
quality (between 1 (good) and FF_LAMBDA_MAX (bad))
const uint8_t ff_svq1_inter_multistage_vlc[6][8][2]
static av_cold int svq1_encode_end(AVCodecContext *avctx)
uint32_t * mb_type
macroblock type table mb_type_base + mb_width + 2
int first_slice_line
used in mpeg4 too to handle resync markers
uint16_t * mc_mb_var
Table for motion compensated MB variances.
unsigned int lambda2
(lambda*lambda) >> FF_LAMBDA_SHIFT
int linesize[AV_NUM_DATA_POINTERS]
Size, in bytes, of the data for each picture/channel plane.
main external API structure.
static void close(AVCodecParserContext *s)
int height
picture size. must be a multiple of 16
int16_t(*[2] motion_val)[2]
motion vector table
void ff_fix_long_p_mvs(MpegEncContext *s)
planar YUV 4:1:0, 9bpp, (1 Cr & Cb sample per 4x4 Y samples)
#define FF_ARRAY_ELEMS(a)
DSPContext dsp
pointers for accelerated dsp functions
int16_t(*[3] motion_val16)[2]
int f_code
forward MV resolution
int(* ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2, int size)
int pict_type
AV_PICTURE_TYPE_I, AV_PICTURE_TYPE_P, AV_PICTURE_TYPE_B, ...
int ff_init_me(MpegEncContext *s)
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
int16_t(*[3] motion_val8)[2]
int b8_stride
2*mb_width+1 used for some 8x8 block arrays to allow simple addressing
struct AVCodecContext * avctx
int gop_size
the number of pictures in a group of pictures, or 0 for intra_only
PutBitContext pb
bit output
const int8_t *const ff_svq1_intra_codebooks[6]
common internal api header.
int mb_stride
mb_width+1 used for some arrays to allow simple addressing of left & top MBs without sig11 ...
static void flush_put_bits(PutBitContext *s)
Pad the end of the output stream with zeros.
const uint8_t ff_svq1_intra_multistage_vlc[6][8][2]
Picture last_picture
copy of the previous picture structure.
Picture * last_picture_ptr
pointer to the previous picture.
static const int8_t svq1_inter_codebook_sum[4][16 *6]
const uint16_t ff_svq1_inter_mean_vlc[512][2]
static void init_put_bits(PutBitContext *s, uint8_t *buffer, int buffer_size)
Initialize the PutBitContext s.
uint32_t * map
map to avoid duplicate evaluations
int dia_size
ME diamond size & shape.
int key_frame
1 -> keyframe, 0-> not
int linesize
line size, in bytes, may be different from width
void ff_fix_long_mvs(MpegEncContext *s, uint8_t *field_select_table, int field_select, int16_t(*mv_table)[2], int f_code, int type, int truncate)
static int svq1_encode_frame(AVCodecContext *avctx, AVPacket *pkt, const AVFrame *pict, int *got_packet)
PutBitContext reorder_pb[6]
int frame_number
Frame counter, set by libavcodec.
int flags
AVCodecContext.flags (HQ, MV4, ...)
#define QUALITY_THRESHOLD
int me_method
Motion estimation algorithm used for video coding.
This structure stores compressed data.
const uint16_t ff_svq1_intra_mean_vlc[256][2]
void * av_mallocz(size_t size)
Allocate a block of size bytes with alignment suitable for all memory accesses (including vectors if ...
int ff_match_2uint16(const uint16_t(*tab)[2], int size, int a, int b)
Return the index into tab at which {a,b} match elements {[0],[1]} of tab.
unsigned int lambda
lagrange multipler used in rate distortion