76 #define QUALITY_THRESHOLD 100
77 #define THRESHOLD_MULTIPLIER 0.6
84 for (i = 0; i <
size; i++)
85 score += (pix1[i] - pix2[i]) * (pix1[i] - pix2[i]);
91 int threshold,
int lambda,
int intra)
93 int count,
y, x, i, j,
split, best_mean, best_score, best_count;
95 int block_sum[7] = { 0, 0, 0, 0, 0, 0 };
96 int w = 2 << (level + 2 >> 1);
97 int h = 2 << (level + 1 >> 1);
100 const int8_t *codebook_sum, *codebook;
101 const uint16_t(*mean_vlc)[2];
102 const uint8_t(*multistage_vlc)[2];
111 for (y = 0; y <
h; y++) {
112 for (x = 0; x < w; x++) {
124 for (y = 0; y <
h; y++) {
125 for (x = 0; x < w; x++) {
135 best_score -= (int)((
unsigned)block_sum[0] * block_sum[0] >> (level + 3));
136 best_mean = block_sum[0] + (size >> 1) >> (level + 3);
139 for (count = 1; count < 7; count++) {
140 int best_vector_score = INT_MAX;
141 int best_vector_sum = -999, best_vector_mean = -999;
142 const int stage = count - 1;
143 const int8_t *vector;
145 for (i = 0; i < 16; i++) {
146 int sum = codebook_sum[stage * 16 + i];
147 int sqr,
diff, score;
149 vector = codebook + stage * size * 16 + i *
size;
151 diff = block_sum[stage] - sum;
152 score = sqr - (diff * (int64_t)diff >> (level + 3));
153 if (score < best_vector_score) {
154 int mean = diff + (size >> 1) >> (level + 3);
156 mean = av_clip(mean, intra ? 0 : -256, 255);
157 best_vector_score = score;
158 best_vector[stage] = i;
159 best_vector_sum = sum;
160 best_vector_mean = mean;
164 vector = codebook + stage * size * 16 + best_vector[stage] *
size;
165 for (j = 0; j <
size; j++)
166 block[stage + 1][j] =
block[stage][j] - vector[j];
167 block_sum[stage + 1] = block_sum[stage] - best_vector_sum;
168 best_vector_score += lambda *
170 multistage_vlc[1 +
count][1]
171 + mean_vlc[best_vector_mean][1]);
173 if (best_vector_score < best_score) {
174 best_score = best_vector_score;
176 best_mean = best_vector_mean;
182 if (best_score > threshold && level) {
184 int offset = level & 1 ? stride * h / 2 : w / 2;
187 for (i = level - 1; i >= 0; i--)
189 score +=
encode_block(s, src, ref, decoded, stride, level - 1,
190 threshold >> 1, lambda, intra);
191 score +=
encode_block(s, src + offset, ref + offset, decoded + offset,
192 stride, level - 1, threshold >> 1, lambda, intra);
195 if (score < best_score) {
199 for (i = level - 1; i >= 0; i--)
207 av_assert1(best_mean >= 0 && best_mean < 256 || !intra);
208 av_assert1(best_mean >= -256 && best_mean < 256);
209 av_assert1(best_count >= 0 && best_count < 7);
214 multistage_vlc[1 + best_count][1],
215 multistage_vlc[1 + best_count][0]);
217 mean_vlc[best_mean][0]);
219 for (i = 0; i < best_count; i++) {
220 av_assert2(best_vector[i] >= 0 && best_vector[i] < 16);
224 for (y = 0; y <
h; y++)
225 for (x = 0; x < w; x++)
226 decoded[x + y * stride] = src[x + y * stride] -
227 block[best_count][x + w * y] +
244 unsigned char *src_plane,
245 unsigned char *ref_plane,
246 unsigned char *decoded_plane,
252 int block_width, block_height;
261 for (level = 4; level >= 0; level--)
264 block_width = (width + 15) / 16;
265 block_height = (height + 15) / 16;
298 block_height * 2 + 2) *
299 2 *
sizeof(int16_t));
301 (block_height + 2) + 1) *
302 2 *
sizeof(int16_t));
323 for (y = 0; y < block_height; y++) {
327 for (i = 0; i < 16 && i + 16 * y <
height; i++) {
328 memcpy(&src[i * stride], &src_plane[(i + 16 * y) * src_stride],
330 for (x = width; x < 16 * block_width; x++)
331 src[i * stride + x] = src[i * stride + x - 1];
333 for (; i < 16 && i + 16 * y < 16 * block_height; i++)
334 memcpy(&src[i * stride], &src[(i - 1) * stride],
337 for (x = 0; x < block_width; x++) {
352 for (y = 0; y < block_height; y++) {
353 for (i = 0; i < 16 && i + 16 * y <
height; i++) {
354 memcpy(&src[i * stride], &src_plane[(i + 16 * y) * src_stride],
356 for (x = width; x < 16 * block_width; x++)
357 src[i * stride + x] = src[i * stride + x - 1];
359 for (; i < 16 && i + 16 * y < 16 * block_height; i++)
360 memcpy(&src[i * stride], &src[(i - 1) * stride], 16 * block_width);
363 for (x = 0; x < block_width; x++) {
364 uint8_t reorder_buffer[2][6][7 * 32];
366 int offset = y * 16 * stride + x * 16;
369 int score[4] = { 0, 0, 0, 0 }, best;
384 for (i = 0; i < 6; i++)
390 score[0] = vlc[1] * lambda;
394 for (i = 0; i < 6; i++) {
405 int mx, my, pred_x, pred_y, dxy;
411 for (i = 0; i < 6; i++)
429 dxy = (mx & 1) + 2 * (my & 1);
436 score[1] +=
encode_block(s, src + 16 * x, temp + 16*stride,
437 decoded, stride, 5, 64, lambda, 0);
438 best = score[1] <= score[0];
443 score[2] += vlc[1] * lambda;
444 if (score[2] < score[best] && mx == 0 && my == 0) {
452 for (i = 0; i < 6; i++) {
471 for (i = 5; i >= 0; i--)
501 for (i = 0; i < 3; i++) {
576 const AVFrame *pict,
int *got_packet)
617 for (i = 0; i < 3; i++)
627 for (j = 0; j < i; j++) {
uint8_t * scratchpad
data area for the ME algo, so that the ME does not need to malloc/free.
av_cold void ff_me_cmp_init(MECmpContext *c, AVCodecContext *avctx)
int ff_alloc_packet2(AVCodecContext *avctx, AVPacket *avpkt, int64_t size)
Check AVPacket size and/or allocate data.
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
static void put_bits(Jpeg2000EncoderContext *s, int val, int n)
put n times val bit
void ff_h263_encode_init(MpegEncContext *s)
AVFrame * coded_frame
the picture in the bitstream
static av_cold int init(AVCodecContext *avctx)
uint16_t * mb_var
Table for MB variances.
AVFrame * current_picture
void avpriv_copy_bits(PutBitContext *pb, const uint8_t *src, int length)
Copy the content of src to the bitstream.
static int svq1_encode_plane(SVQ1EncContext *s, int plane, unsigned char *src_plane, unsigned char *ref_plane, unsigned char *decoded_plane, int width, int height, int src_stride, int stride)
enum AVPixelFormat pix_fmt
Pixel format, see AV_PIX_FMT_xxx.
uint32_t * score_map
map to store the scores
#define FF_ARRAY_ELEMS(a)
int16_t * ff_h263_pred_motion(MpegEncContext *s, int block, int dir, int *px, int *py)
void ff_h263_encode_motion(PutBitContext *pb, int val, int f_code)
static const int8_t svq1_intra_codebook_sum[4][16 *6]
static void init_block_index(MpegEncContext *s)
const uint16_t ff_svq1_frame_size_table[7][2]
#define CANDIDATE_MB_TYPE_INTER
#define av_assert0(cond)
assert() equivalent, that is always enabled.
static av_cold int svq1_encode_init(AVCodecContext *avctx)
av_cold void ff_mpegvideoencdsp_init(MpegvideoEncDSPContext *c, AVCodecContext *avctx)
AVFrame * av_frame_alloc(void)
Allocate an AVFrame and set its fields to default values.
#define av_assert2(cond)
assert() equivalent, that does lie in speed critical code.
const int8_t *const ff_svq1_inter_codebooks[6]
int16_t encoded_block_levels[6][7][256]
Picture current_picture
copy of the current picture structure.
int mb_height
number of MBs horizontally & vertically
#define AV_PKT_FLAG_KEY
The packet contains a keyframe.
static int ssd_int8_vs_int16_c(const int8_t *pix1, const int16_t *pix2, intptr_t size)
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
void av_frame_free(AVFrame **frame)
Free the frame and any dynamically allocated objects in it, e.g.
#define NULL_IF_CONFIG_SMALL(x)
Return NULL if CONFIG_SMALL is true, otherwise the argument without modification. ...
#define AV_LOG_DEBUG
Stuff which is only useful for libav* developers.
simple assert() macros that are a bit more flexible than ISO C assert().
MpegvideoEncDSPContext mpvencdsp
const char * name
Name of the codec implementation.
int16_t(*[3] motion_val16)[2]
uint16_t * mb_type
Table for candidate MB types for encoding (defines in mpegutils.h)
static const uint8_t offset[127][2]
void ff_mpv_common_end(MpegEncContext *s)
Libavcodec external API header.
static int encode_block(SVQ1EncContext *s, uint8_t *src, uint8_t *ref, uint8_t *decoded, int stride, int level, int threshold, int lambda, int intra)
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)
av_cold void ff_hpeldsp_init(HpelDSPContext *c, int flags)
PutBitContext reorder_pb[6]
enum AVPictureType pict_type
Picture type of the frame.
#define av_assert1(cond)
assert() equivalent, that does not lie in speed critical code.
#define FF_MIN_BUFFER_SIZE
minimum encoding buffer size Used to avoid some checks during header writing.
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.
int16_t(*[2] motion_val)[2]
Picture * current_picture_ptr
pointer to the current picture
int(* ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2, intptr_t size)
int quality
quality (between 1 (good) and FF_LAMBDA_MAX (bad))
const uint8_t ff_svq1_inter_multistage_vlc[6][8][2]
int block_index[6]
index to current MB in block based arrays with edges
static av_cold int svq1_encode_end(AVCodecContext *avctx)
int first_slice_line
used in mpeg4 too to handle resync markers
uint16_t * mc_mb_var
Table for motion compensated MB variances.
int16_t(*[3] motion_val8)[2]
unsigned int lambda2
(lambda*lambda) >> FF_LAMBDA_SHIFT
ptrdiff_t linesize
line size, in bytes, may be different from width
static void svq1_write_header(SVQ1EncContext *s, int frame_type)
int linesize[AV_NUM_DATA_POINTERS]
For video, size in bytes of each picture line.
av_cold void ff_svq1enc_init_ppc(SVQ1EncContext *c)
main external API structure.
int height
picture size. must be a multiple of 16
int ff_get_buffer(AVCodecContext *avctx, AVFrame *frame, int flags)
Get a buffer for a frame.
op_pixels_func put_pixels_tab[4][4]
Halfpel motion compensation with rounding (a+b+1)>>1.
void ff_fix_long_p_mvs(MpegEncContext *s)
BYTE int const BYTE int int int height
planar YUV 4:1:0, 9bpp, (1 Cr & Cb sample per 4x4 Y samples)
int f_code
forward MV resolution
int pict_type
AV_PICTURE_TYPE_I, AV_PICTURE_TYPE_P, AV_PICTURE_TYPE_B, ...
void ff_svq1enc_init_x86(SVQ1EncContext *c)
int ff_init_me(MpegEncContext *s)
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
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
GLint GLenum GLboolean GLsizei stride
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
#define CANDIDATE_MB_TYPE_INTRA
av_cold int ff_mpv_common_init(MpegEncContext *s)
init common structure for both encoder and decoder.
static av_always_inline int diff(const uint32_t a, const uint32_t b)
int dia_size
ME diamond size & shape.
int key_frame
1 -> keyframe, 0-> not
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)
int frame_number
Frame counter, set by libavcodec.
uint32_t * mb_type
types and macros are defined in mpegutils.h
#define av_malloc_array(a, b)
#define FFSWAP(type, a, b)
#define QUALITY_THRESHOLD
int me_method
Motion estimation algorithm used for video coding.
AVPixelFormat
Pixel format.
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