#define FULLPEL_MODE 1
#define HALFPEL_MODE 2
#define THIRDPEL_MODE 3
+#define PREDICT_MODE 4
/* dual scan (from some older h264 draft)
o-->o-->o o
src[3+2*stride]=
src[2+3*stride]=
src[3+3*stride]=(l3 + t3)>>1;
-};
+}
static void pred16x16_plane_svq3_c(uint8_t *src, int stride){
pred16x16_plane_compat_c(src, stride, 1);
return 0;
}
-static void sixpel_mc_put (MpegEncContext *s,
- uint8_t *src, uint8_t *dst, int stride,
- int dxy, int width, int height) {
- int i, j;
-
- switch (dxy) {
- case 6*0+0:
- for (i=0; i < height; i++) {
- memcpy (dst, src, width);
- src += stride;
- dst += stride;
- }
- break;
- case 6*0+2:
- for (i=0; i < height; i++) {
- for (j=0; j < width; j++) {
- dst[j] = (683*(2*src[j] + src[j+1] + 1)) >> 11;
- }
- src += stride;
- dst += stride;
- }
- break;
- case 6*0+3:
- for (i=0; i < height; i++) {
- for (j=0; j < width; j++) {
- dst[j] = (src[j] + src[j+1] + 1) >> 1;
- }
- src += stride;
- dst += stride;
- }
- break;
- case 6*0+4:
- for (i=0; i < height; i++) {
- for (j=0; j < width; j++) {
- dst[j] = (683*(src[j] + 2*src[j+1] + 1)) >> 11;
- }
- src += stride;
- dst += stride;
- }
- break;
- case 6*2+0:
- for (i=0; i < height; i++) {
- for (j=0; j < width; j++) {
- dst[j] = (683*(2*src[j] + src[j+stride] + 1)) >> 11;
- }
- src += stride;
- dst += stride;
- }
- break;
- case 6*2+2:
- for (i=0; i < height; i++) {
- for (j=0; j < width; j++) {
- dst[j] = (2731*(4*src[j] + 3*src[j+1] + 3*src[j+stride] + 2*src[j+stride+1] + 6)) >> 15;
- }
- src += stride;
- dst += stride;
- }
- break;
- case 6*2+4:
- for (i=0; i < height; i++) {
- for (j=0; j < width; j++) {
- dst[j] = (2731*(3*src[j] + 4*src[j+1] + 2*src[j+stride] + 3*src[j+stride+1] + 6)) >> 15;
- }
- src += stride;
- dst += stride;
- }
- break;
- case 6*3+0:
- for (i=0; i < height; i++) {
- for (j=0; j < width; j++) {
- dst[j] = (src[j] + src[j+stride]+1) >> 1;
- }
- src += stride;
- dst += stride;
- }
- break;
- case 6*3+3:
- for (i=0; i < height; i++) {
- for (j=0; j < width; j++) {
- dst[j] = (src[j] + src[j+1] + src[j+stride] + src[j+stride+1] + 2) >> 2;
- }
- src += stride;
- dst += stride;
- }
- break;
- case 6*4+0:
- for (i=0; i < height; i++) {
- for (j=0; j < width; j++) {
- dst[j] = (683*(src[j] + 2*src[j+stride] + 1)) >> 11;
- }
- src += stride;
- dst += stride;
- }
- break;
- case 6*4+2:
- for (i=0; i < height; i++) {
- for (j=0; j < width; j++) {
- dst[j] = (2731*(3*src[j] + 2*src[j+1] + 4*src[j+stride] + 3*src[j+stride+1] + 6)) >> 15;
- }
- src += stride;
- dst += stride;
- }
- break;
- case 6*4+4:
- for (i=0; i < height; i++) {
- for (j=0; j < width; j++) {
- dst[j] = (2731*(2*src[j] + 3*src[j+1] + 3*src[j+stride] + 4*src[j+stride+1] + 6)) >> 15;
- }
- src += stride;
- dst += stride;
- }
- break;
- }
-}
+static inline void svq3_mc_dir_part (MpegEncContext *s,
+ int x, int y, int width, int height,
+ int mx, int my, int dxy,
+ int thirdpel, int dir, int avg) {
-static inline void svq3_mc_dir_part (MpegEncContext *s, int x, int y,
- int width, int height, int mx, int my, int dxy) {
+ const Picture *pic = (dir == 0) ? &s->last_picture : &s->next_picture;
uint8_t *src, *dest;
int i, emu = 0;
+ int blocksize= 2 - (width>>3); //16->0, 8->1, 4->2
mx += x;
my += y;
- if (mx < 0 || mx >= (s->width - width - 1) ||
- my < 0 || my >= (s->height - height - 1)) {
+ if (mx < 0 || mx >= (s->h_edge_pos - width - 1) ||
+ my < 0 || my >= (s->v_edge_pos - height - 1)) {
if ((s->flags & CODEC_FLAG_EMU_EDGE)) {
emu = 1;
}
- mx = clip (mx, -16, (s->width - width + 15));
- my = clip (my, -16, (s->height - height + 15));
+ mx = clip (mx, -16, (s->h_edge_pos - width + 15));
+ my = clip (my, -16, (s->v_edge_pos - height + 15));
}
/* form component predictions */
dest = s->current_picture.data[0] + x + y*s->linesize;
- src = s->last_picture.data[0] + mx + my*s->linesize;
+ src = pic->data[0] + mx + my*s->linesize;
if (emu) {
- ff_emulated_edge_mc (s, src, s->linesize, (width + 1), (height + 1),
- mx, my, s->width, s->height);
+ ff_emulated_edge_mc (s->edge_emu_buffer, src, s->linesize, (width + 1), (height + 1),
+ mx, my, s->h_edge_pos, s->v_edge_pos);
src = s->edge_emu_buffer;
}
- sixpel_mc_put (s, src, dest, s->linesize, dxy, width, height);
+ if(thirdpel)
+ (avg ? s->dsp.avg_tpel_pixels_tab : s->dsp.put_tpel_pixels_tab)[dxy](dest, src, s->linesize, width, height);
+ else
+ (avg ? s->dsp.avg_pixels_tab : s->dsp.put_pixels_tab)[blocksize][dxy](dest, src, s->linesize, height);
if (!(s->flags & CODEC_FLAG_GRAY)) {
mx = (mx + (mx < (int) x)) >> 1;
my = (my + (my < (int) y)) >> 1;
width = (width >> 1);
height = (height >> 1);
+ blocksize++;
for (i=1; i < 3; i++) {
dest = s->current_picture.data[i] + (x >> 1) + (y >> 1)*s->uvlinesize;
- src = s->last_picture.data[i] + mx + my*s->uvlinesize;
+ src = pic->data[i] + mx + my*s->uvlinesize;
if (emu) {
- ff_emulated_edge_mc (s, src, s->uvlinesize, (width + 1), (height + 1),
- mx, my, (s->width >> 1), (s->height >> 1));
- src = s->edge_emu_buffer;
+ ff_emulated_edge_mc (s->edge_emu_buffer, src, s->uvlinesize, (width + 1), (height + 1),
+ mx, my, (s->h_edge_pos >> 1), (s->v_edge_pos >> 1));
+ src = s->edge_emu_buffer;
}
- sixpel_mc_put (s, src, dest, s->uvlinesize, dxy, width, height);
+ if(thirdpel)
+ (avg ? s->dsp.avg_tpel_pixels_tab : s->dsp.put_tpel_pixels_tab)[dxy](dest, src, s->uvlinesize, width, height);
+ else
+ (avg ? s->dsp.avg_pixels_tab : s->dsp.put_pixels_tab)[blocksize][dxy](dest, src, s->uvlinesize, height);
}
}
}
+static inline int svq3_mc_dir (H264Context *h, int size, int mode, int dir, int avg) {
+
+ int i, j, k, mx, my, dx, dy, x, y;
+ MpegEncContext *const s = (MpegEncContext *) h;
+ const int part_width = ((size & 5) == 4) ? 4 : 16 >> (size & 1);
+ const int part_height = 16 >> ((unsigned) (size + 1) / 3);
+ const int extra_width = (mode == PREDICT_MODE) ? -16*6 : 0;
+ const int h_edge_pos = 6*(s->h_edge_pos - part_width ) - extra_width;
+ const int v_edge_pos = 6*(s->v_edge_pos - part_height) - extra_width;
+
+ for (i=0; i < 16; i+=part_height) {
+ for (j=0; j < 16; j+=part_width) {
+ const int b_xy = (4*s->mb_x+(j>>2)) + (4*s->mb_y+(i>>2))*h->b_stride;
+ int dxy;
+ x = 16*s->mb_x + j;
+ y = 16*s->mb_y + i;
+ k = ((j>>2)&1) + ((i>>1)&2) + ((j>>1)&4) + (i&8);
+
+ if (mode != PREDICT_MODE) {
+ pred_motion (h, k, (part_width >> 2), dir, 1, &mx, &my);
+ } else {
+ mx = s->next_picture.motion_val[0][b_xy][0]<<1;
+ my = s->next_picture.motion_val[0][b_xy][1]<<1;
+
+ if (dir == 0) {
+ mx = ((mx * h->frame_num_offset) / h->prev_frame_num_offset + 1)>>1;
+ my = ((my * h->frame_num_offset) / h->prev_frame_num_offset + 1)>>1;
+ } else {
+ mx = ((mx * (h->frame_num_offset - h->prev_frame_num_offset)) / h->prev_frame_num_offset + 1)>>1;
+ my = ((my * (h->frame_num_offset - h->prev_frame_num_offset)) / h->prev_frame_num_offset + 1)>>1;
+ }
+ }
+
+ /* clip motion vector prediction to frame border */
+ mx = clip (mx, extra_width - 6*x, h_edge_pos - 6*x);
+ my = clip (my, extra_width - 6*y, v_edge_pos - 6*y);
+
+ /* get (optional) motion vector differential */
+ if (mode == PREDICT_MODE) {
+ dx = dy = 0;
+ } else {
+ dy = svq3_get_se_golomb (&s->gb);
+ dx = svq3_get_se_golomb (&s->gb);
+
+ if (dx == INVALID_VLC || dy == INVALID_VLC) {
+ return -1;
+ }
+ }
+
+ /* compute motion vector */
+ if (mode == THIRDPEL_MODE) {
+ int fx, fy;
+ mx = ((mx + 1)>>1) + dx;
+ my = ((my + 1)>>1) + dy;
+ fx= ((unsigned)(mx + 0x3000))/3 - 0x1000;
+ fy= ((unsigned)(my + 0x3000))/3 - 0x1000;
+ dxy= (mx - 3*fx) + 4*(my - 3*fy);
+
+ svq3_mc_dir_part (s, x, y, part_width, part_height, fx, fy, dxy, 1, dir, avg);
+ mx += mx;
+ my += my;
+ } else if (mode == HALFPEL_MODE || mode == PREDICT_MODE) {
+ mx = ((unsigned)(mx + 1 + 0x3000))/3 + dx - 0x1000;
+ my = ((unsigned)(my + 1 + 0x3000))/3 + dy - 0x1000;
+ dxy= (mx&1) + 2*(my&1);
+
+ svq3_mc_dir_part (s, x, y, part_width, part_height, mx>>1, my>>1, dxy, 0, dir, avg);
+ mx *= 3;
+ my *= 3;
+ } else {
+ mx = ((unsigned)(mx + 3 + 0x6000))/6 + dx - 0x1000;
+ my = ((unsigned)(my + 3 + 0x6000))/6 + dy - 0x1000;
+
+ svq3_mc_dir_part (s, x, y, part_width, part_height, mx, my, 0, 0, dir, avg);
+ mx *= 6;
+ my *= 6;
+ }
+
+ /* update mv_cache */
+ if (mode != PREDICT_MODE) {
+ int32_t mv = pack16to32(mx,my);
+
+ if (part_height == 8 && i < 8) {
+ *(int32_t *) h->mv_cache[dir][scan8[k] + 1*8] = mv;
+
+ if (part_width == 8 && j < 8) {
+ *(int32_t *) h->mv_cache[dir][scan8[k] + 1 + 1*8] = mv;
+ }
+ }
+ if (part_width == 8 && j < 8) {
+ *(int32_t *) h->mv_cache[dir][scan8[k] + 1] = mv;
+ }
+ if (part_width == 4 || part_height == 4) {
+ *(int32_t *) h->mv_cache[dir][scan8[k]] = mv;
+ }
+ }
+
+ /* write back motion vectors */
+ fill_rectangle(s->current_picture.motion_val[dir][b_xy], part_width>>2, part_height>>2, h->b_stride, pack16to32(mx,my), 4);
+ }
+ }
+
+ return 0;
+}
+
static int svq3_decode_mb (H264Context *h, unsigned int mb_type) {
- int cbp, dir, mode, mx, my, dx, dy, x, y, part_width, part_height;
- int i, j, k, l, m;
+ int i, j, k, m, dir, mode;
+ int cbp = 0;
uint32_t vlc;
int8_t *top, *left;
MpegEncContext *const s = (MpegEncContext *) h;
h->topright_samples_available = 0xFFFF;
if (mb_type == 0) { /* SKIP */
- svq3_mc_dir_part (s, 16*s->mb_x, 16*s->mb_y, 16, 16, 0, 0, 0);
+ if (s->pict_type == P_TYPE || s->next_picture.mb_type[mb_xy] == -1) {
+ svq3_mc_dir_part (s, 16*s->mb_x, 16*s->mb_y, 16, 16, 0, 0, 0, 0, 0, 0);
+
+ if (s->pict_type == B_TYPE) {
+ svq3_mc_dir_part (s, 16*s->mb_x, 16*s->mb_y, 16, 16, 0, 0, 0, 0, 1, 1);
+ }
+
+ mb_type = MB_TYPE_SKIP;
+ } else {
+ svq3_mc_dir (h, s->next_picture.mb_type[mb_xy], PREDICT_MODE, 0, 0);
+ svq3_mc_dir (h, s->next_picture.mb_type[mb_xy], PREDICT_MODE, 1, 1);
- cbp = 0;
- mb_type = MB_TYPE_SKIP;
+ mb_type = MB_TYPE_16x16;
+ }
} else if (mb_type < 8) { /* INTER */
if (h->thirdpel_flag && h->halfpel_flag == !get_bits (&s->gb, 1)) {
mode = THIRDPEL_MODE;
}
/* fill caches */
- /* note ref_cache[0] should contain here:
+ /* note ref_cache should contain here:
????????
???11111
N??11111
N
*/
- if (s->mb_x > 0) {
- for (i=0; i < 4; i++) {
- *(uint32_t *) h->mv_cache[0][scan8[0] - 1 + i*8] = *(uint32_t *) s->current_picture.motion_val[0][b_xy - 1 + i*h->b_stride];
- }
- } else {
- for (i=0; i < 4; i++) {
- *(uint32_t *) h->mv_cache[0][scan8[0] - 1 + i*8] = 0;
+ for (m=0; m < 2; m++) {
+ if (s->mb_x > 0 && h->intra4x4_pred_mode[mb_xy - 1][0] != -1) {
+ for (i=0; i < 4; i++) {
+ *(uint32_t *) h->mv_cache[m][scan8[0] - 1 + i*8] = *(uint32_t *) s->current_picture.motion_val[m][b_xy - 1 + i*h->b_stride];
+ }
+ } else {
+ for (i=0; i < 4; i++) {
+ *(uint32_t *) h->mv_cache[m][scan8[0] - 1 + i*8] = 0;
+ }
}
- }
- if (s->mb_y > 0) {
- memcpy (h->mv_cache[0][scan8[0] - 1*8], s->current_picture.motion_val[0][b_xy - h->b_stride], 4*2*sizeof(int16_t));
- memset (&h->ref_cache[0][scan8[0] - 1*8], 1, 4);
-
- if (s->mb_x < (s->mb_width - 1)) {
- *(uint32_t *) h->mv_cache[0][scan8[0] + 4 - 1*8] = *(uint32_t *) s->current_picture.motion_val[0][b_xy - h->b_stride + 4];
- h->ref_cache[0][scan8[0] + 4 - 1*8] = 1;
+ if (s->mb_y > 0) {
+ memcpy (h->mv_cache[m][scan8[0] - 1*8], s->current_picture.motion_val[m][b_xy - h->b_stride], 4*2*sizeof(int16_t));
+ memset (&h->ref_cache[m][scan8[0] - 1*8], (h->intra4x4_pred_mode[mb_xy - s->mb_stride][4] == -1) ? PART_NOT_AVAILABLE : 1, 4);
+
+ if (s->mb_x < (s->mb_width - 1)) {
+ *(uint32_t *) h->mv_cache[m][scan8[0] + 4 - 1*8] = *(uint32_t *) s->current_picture.motion_val[m][b_xy - h->b_stride + 4];
+ h->ref_cache[m][scan8[0] + 4 - 1*8] =
+ (h->intra4x4_pred_mode[mb_xy - s->mb_stride + 1][0] == -1 ||
+ h->intra4x4_pred_mode[mb_xy - s->mb_stride][4] == -1) ? PART_NOT_AVAILABLE : 1;
+ }else
+ h->ref_cache[m][scan8[0] + 4 - 1*8] = PART_NOT_AVAILABLE;
+ if (s->mb_x > 0) {
+ *(uint32_t *) h->mv_cache[m][scan8[0] - 1 - 1*8] = *(uint32_t *) s->current_picture.motion_val[m][b_xy - h->b_stride - 1];
+ h->ref_cache[m][scan8[0] - 1 - 1*8] = (h->intra4x4_pred_mode[mb_xy - s->mb_stride - 1][3] == -1) ? PART_NOT_AVAILABLE : 1;
+ }else
+ h->ref_cache[m][scan8[0] - 1 - 1*8] = PART_NOT_AVAILABLE;
}else
- h->ref_cache[0][scan8[0] + 4 - 1*8] = PART_NOT_AVAILABLE;
- if (s->mb_x > 0) {
- *(uint32_t *) h->mv_cache[0][scan8[0] - 1 - 1*8] = *(uint32_t *) s->current_picture.motion_val[0][b_xy - h->b_stride - 1];
- h->ref_cache[0][scan8[0] - 1 - 1*8] = 1;
- }else
- h->ref_cache[0][scan8[0] - 1 - 1*8] = PART_NOT_AVAILABLE;
- }else
- memset (&h->ref_cache[0][scan8[0] - 1*8 - 1], PART_NOT_AVAILABLE, 8);
-
- /* decode motion vector(s) and form prediction(s) */
- part_width = ((mb_type & 5) == 5) ? 4 : 8 << (mb_type & 1);
- part_height = 16 >> ((unsigned) mb_type / 3);
+ memset (&h->ref_cache[m][scan8[0] - 1*8 - 1], PART_NOT_AVAILABLE, 8);
- for (i=0; i < 16; i+=part_height) {
- for (j=0; j < 16; j+=part_width) {
- int dxy;
- x = 16*s->mb_x + j;
- y = 16*s->mb_y + i;
- k = ((j>>2)&1) + ((i>>1)&2) + ((j>>1)&4) + (i&8);
-
- pred_motion (h, k, (part_width >> 2), 0, 1, &mx, &my);
-
- /* clip motion vector prediction to frame border */
- mx = clip (mx, -6*x, 6*(s->width - part_width - x));
- my = clip (my, -6*y, 6*(s->height - part_height - y));
-
- /* get motion vector differential */
- dy = svq3_get_se_golomb (&s->gb);
- dx = svq3_get_se_golomb (&s->gb);
+ if (s->pict_type != B_TYPE)
+ break;
+ }
- if (dx == INVALID_VLC || dy == INVALID_VLC) {
- return -1;
+ /* decode motion vector(s) and form prediction(s) */
+ if (s->pict_type == P_TYPE) {
+ svq3_mc_dir (h, (mb_type - 1), mode, 0, 0);
+ } else { /* B_TYPE */
+ if (mb_type != 2) {
+ svq3_mc_dir (h, 0, mode, 0, 0);
+ } else {
+ for (i=0; i < 4; i++) {
+ memset (s->current_picture.motion_val[0][b_xy + i*h->b_stride], 0, 4*2*sizeof(int16_t));
}
- /* compute motion vector */
- if (mode == THIRDPEL_MODE) {
- int fx, fy;
- mx = ((mx + 1)>>1) + dx;
- my = ((my + 1)>>1) + dy;
- fx= ((unsigned)(mx + 0x3000))/3 - 0x1000;
- fy= ((unsigned)(my + 0x3000))/3 - 0x1000;
- dxy= 2*(mx - 3*fx) + 2*6*(my - 3*fy);
-
- svq3_mc_dir_part (s, x, y, part_width, part_height, fx, fy, dxy);
- mx += mx;
- my += my;
- } else if (mode == HALFPEL_MODE) {
- mx = ((unsigned)(mx + 1 + 0x3000))/3 + dx - 0x1000;
- my = ((unsigned)(my + 1 + 0x3000))/3 + dy - 0x1000;
- dxy= 3*(mx&1) + 6*3*(my&1);
-
- svq3_mc_dir_part (s, x, y, part_width, part_height, mx>>1, my>>1, dxy);
- mx *= 3;
- my *= 3;
- } else {
- assert(mode == FULLPEL_MODE);
- mx = ((unsigned)(mx + 3 + 0x6000))/6 + dx - 0x1000;
- my = ((unsigned)(my + 3 + 0x6000))/6 + dy - 0x1000;
-
- svq3_mc_dir_part (s, x, y, part_width, part_height, mx, my, 0);
- mx *= 6;
- my *= 6;
+ }
+ if (mb_type != 1) {
+ svq3_mc_dir (h, 0, mode, 1, (mb_type == 3));
+ } else {
+ for (i=0; i < 4; i++) {
+ memset (s->current_picture.motion_val[1][b_xy + i*h->b_stride], 0, 4*2*sizeof(int16_t));
}
-
- /* update mv_cache */
- fill_rectangle(h->mv_cache[0][scan8[k]], part_width>>2, part_height>>2, 8, (mx&0xFFFF)+(my<<16), 4);
}
}
- for (i=0; i < 4; i++) {
- memcpy (s->current_picture.motion_val[0][b_xy + i*h->b_stride], h->mv_cache[0][scan8[0] + 8*i], 4*2*sizeof(int16_t));
- }
-
- if ((vlc = svq3_get_ue_golomb (&s->gb)) >= 48)
- return -1;
-
- cbp = golomb_to_inter_cbp[vlc];
mb_type = MB_TYPE_16x16;
- } else if (mb_type == 8) { /* INTRA4x4 */
+ } else if (mb_type == 8 || mb_type == 33) { /* INTRA4x4 */
memset (h->intra4x4_pred_mode_cache, -1, 8*5*sizeof(int8_t));
- if (s->mb_x > 0) {
- for (i=0; i < 4; i++) {
- h->intra4x4_pred_mode_cache[scan8[0] - 1 + i*8] = h->intra4x4_pred_mode[mb_xy - 1][i];
+ if (mb_type == 8) {
+ if (s->mb_x > 0) {
+ for (i=0; i < 4; i++) {
+ h->intra4x4_pred_mode_cache[scan8[0] - 1 + i*8] = h->intra4x4_pred_mode[mb_xy - 1][i];
+ }
+ if (h->intra4x4_pred_mode_cache[scan8[0] - 1] == -1) {
+ h->left_samples_available = 0x5F5F;
+ }
+ }
+ if (s->mb_y > 0) {
+ h->intra4x4_pred_mode_cache[4+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][4];
+ h->intra4x4_pred_mode_cache[5+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][5];
+ h->intra4x4_pred_mode_cache[6+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][6];
+ h->intra4x4_pred_mode_cache[7+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][3];
+
+ if (h->intra4x4_pred_mode_cache[4+8*0] == -1) {
+ h->top_samples_available = 0x33FF;
+ }
}
- }
- if (s->mb_y > 0) {
- h->intra4x4_pred_mode_cache[4+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][4];
- h->intra4x4_pred_mode_cache[5+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][5];
- h->intra4x4_pred_mode_cache[6+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][6];
- h->intra4x4_pred_mode_cache[7+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][3];
- }
- /* decode prediction codes for luma blocks */
- for (i=0; i < 16; i+=2) {
- vlc = svq3_get_ue_golomb (&s->gb);
+ /* decode prediction codes for luma blocks */
+ for (i=0; i < 16; i+=2) {
+ vlc = svq3_get_ue_golomb (&s->gb);
- if (vlc >= 25)
- return -1;
+ if (vlc >= 25)
+ return -1;
- left = &h->intra4x4_pred_mode_cache[scan8[i] - 1];
- top = &h->intra4x4_pred_mode_cache[scan8[i] - 8];
+ left = &h->intra4x4_pred_mode_cache[scan8[i] - 1];
+ top = &h->intra4x4_pred_mode_cache[scan8[i] - 8];
- left[1] = svq3_pred_1[top[0] + 1][left[0] + 1][svq3_pred_0[vlc][0]];
- left[2] = svq3_pred_1[top[1] + 1][left[1] + 1][svq3_pred_0[vlc][1]];
+ left[1] = svq3_pred_1[top[0] + 1][left[0] + 1][svq3_pred_0[vlc][0]];
+ left[2] = svq3_pred_1[top[1] + 1][left[1] + 1][svq3_pred_0[vlc][1]];
- if (left[1] == -1 || left[2] == -1)
- return -1;
+ if (left[1] == -1 || left[2] == -1)
+ return -1;
+ }
+ } else { /* mb_type == 33, DC_128_PRED block type */
+ for (i=0; i < 4; i++) {
+ memset (&h->intra4x4_pred_mode_cache[scan8[0] + 8*i], DC_PRED, 4);
+ }
}
write_back_intra_pred_mode (h);
- check_intra4x4_pred_mode (h);
- if ((vlc = svq3_get_ue_golomb (&s->gb)) >= 48)
- return -1;
+ if (mb_type == 8) {
+ check_intra4x4_pred_mode (h);
+
+ h->top_samples_available = (s->mb_y == 0) ? 0x33FF : 0xFFFF;
+ h->left_samples_available = (s->mb_x == 0) ? 0x5F5F : 0xFFFF;
+ } else {
+ for (i=0; i < 4; i++) {
+ memset (&h->intra4x4_pred_mode_cache[scan8[0] + 8*i], DC_128_PRED, 4);
+ }
+
+ h->top_samples_available = 0x33FF;
+ h->left_samples_available = 0x5F5F;
+ }
- cbp = golomb_to_intra4x4_cbp[vlc];
mb_type = MB_TYPE_INTRA4x4;
} else { /* INTRA16x16 */
dir = i_mb_type_info[mb_type - 8].pred_mode;
for (i=0; i < 4; i++) {
memset (s->current_picture.motion_val[0][b_xy + i*h->b_stride], 0, 4*2*sizeof(int16_t));
}
+ if (s->pict_type == B_TYPE) {
+ for (i=0; i < 4; i++) {
+ memset (s->current_picture.motion_val[1][b_xy + i*h->b_stride], 0, 4*2*sizeof(int16_t));
+ }
+ }
}
if (!IS_INTRA4x4(mb_type)) {
memset (h->intra4x4_pred_mode[mb_xy], DC_PRED, 8);
}
- if (!IS_SKIP(mb_type)) {
+ if (!IS_SKIP(mb_type) || s->pict_type == B_TYPE) {
memset (h->non_zero_count_cache + 8, 0, 4*9*sizeof(uint8_t));
s->dsp.clear_blocks(h->mb);
}
+ if (!IS_INTRA16x16(mb_type) && (!IS_SKIP(mb_type) || s->pict_type == B_TYPE)) {
+ if ((vlc = svq3_get_ue_golomb (&s->gb)) >= 48)
+ return -1;
+
+ cbp = IS_INTRA(mb_type) ? golomb_to_intra4x4_cbp[vlc] : golomb_to_inter_cbp[vlc];
+ }
if (IS_INTRA16x16(mb_type) || (s->pict_type != I_TYPE && s->adaptive_quant && cbp)) {
s->qscale += svq3_get_se_golomb (&s->gb);
return -1;
}
- if (!IS_SKIP(mb_type) && cbp) {
- l = IS_INTRA16x16(mb_type) ? 1 : 0;
- m = ((s->qscale < 24 && IS_INTRA4x4(mb_type)) ? 2 : 1);
+ if (cbp) {
+ const int index = IS_INTRA16x16(mb_type) ? 1 : 0;
+ const int type = ((s->qscale < 24 && IS_INTRA4x4(mb_type)) ? 2 : 1);
for (i=0; i < 4; i++) {
if ((cbp & (1 << i))) {
for (j=0; j < 4; j++) {
- k = l ? ((j&1) + 2*(i&1) + 2*(j&2) + 4*(i&2)) : (4*i + j);
+ k = index ? ((j&1) + 2*(i&1) + 2*(j&2) + 4*(i&2)) : (4*i + j);
h->non_zero_count_cache[ scan8[k] ] = 1;
- if (svq3_decode_block (&s->gb, &h->mb[16*k], l, m))
+ if (svq3_decode_block (&s->gb, &h->mb[16*k], index, type))
return -1;
}
}
return 0;
}
+static int svq3_decode_slice_header (H264Context *h) {
+ MpegEncContext *const s = (MpegEncContext *) h;
+ const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
+ int i, header;
+
+ header = get_bits (&s->gb, 8);
+
+ if (((header & 0x9F) != 1 && (header & 0x9F) != 2) || (header & 0x60) == 0) {
+ /* TODO: what? */
+ av_log(h->s.avctx, AV_LOG_ERROR, "unsupported slice header (%02X)\n", header);
+ return -1;
+ } else {
+ int length = (header >> 5) & 3;
+
+ h->next_slice_index = s->gb.index + 8*show_bits (&s->gb, 8*length) + 8*length;
+
+ if (h->next_slice_index > s->gb.size_in_bits){
+ av_log(h->s.avctx, AV_LOG_ERROR, "slice after bitstream end\n");
+ return -1;
+ }
+
+ s->gb.size_in_bits = h->next_slice_index - 8*(length - 1);
+ s->gb.index += 8;
+
+ if (length > 0) {
+ memcpy ((uint8_t *) &s->gb.buffer[s->gb.index >> 3],
+ &s->gb.buffer[s->gb.size_in_bits >> 3], (length - 1));
+ }
+ }
+
+ if ((i = svq3_get_ue_golomb (&s->gb)) == INVALID_VLC || i >= 3){
+ av_log(h->s.avctx, AV_LOG_ERROR, "illegal slice type %d \n", i);
+ return -1;
+ }
+
+ h->slice_type = golomb_to_pict_type[i];
+
+ if ((header & 0x9F) == 2) {
+ i = (s->mb_num < 64) ? 6 : (1 + av_log2 (s->mb_num - 1));
+ s->mb_skip_run = get_bits (&s->gb, i) - (s->mb_x + (s->mb_y * s->mb_width));
+ } else {
+ get_bits1 (&s->gb);
+ s->mb_skip_run = 0;
+ }
+
+ h->slice_num = get_bits (&s->gb, 8);
+ s->qscale = get_bits (&s->gb, 5);
+ s->adaptive_quant = get_bits1 (&s->gb);
+
+ /* unknown fields */
+ get_bits1 (&s->gb);
+
+ if (h->unknown_svq3_flag) {
+ get_bits1 (&s->gb);
+ }
+
+ get_bits1 (&s->gb);
+ get_bits (&s->gb, 2);
+
+ while (get_bits1 (&s->gb)) {
+ get_bits (&s->gb, 8);
+ }
+
+ /* reset intra predictors and invalidate motion vector references */
+ if (s->mb_x > 0) {
+ memset (h->intra4x4_pred_mode[mb_xy - 1], -1, 4*sizeof(int8_t));
+ memset (h->intra4x4_pred_mode[mb_xy - s->mb_x], -1, 8*sizeof(int8_t)*s->mb_x);
+ }
+ if (s->mb_y > 0) {
+ memset (h->intra4x4_pred_mode[mb_xy - s->mb_stride], -1, 8*sizeof(int8_t)*(s->mb_width - s->mb_x));
+
+ if (s->mb_x > 0) {
+ h->intra4x4_pred_mode[mb_xy - s->mb_stride - 1][3] = -1;
+ }
+ }
+
+ return 0;
+}
+
static int svq3_decode_frame (AVCodecContext *avctx,
void *data, int *data_size,
uint8_t *buf, int buf_size) {
MpegEncContext *const s = avctx->priv_data;
H264Context *const h = avctx->priv_data;
- int i;
+ int m, mb_type;
+
+ *data_size = 0;
s->flags = avctx->flags;
+ s->flags2 = avctx->flags2;
+ s->unrestricted_mv = 1;
if (!s->context_initialized) {
- s->width = (avctx->width + 15) & ~15;
- s->height = (avctx->height + 15) & ~15;
- h->b_stride = (s->width >> 2);
+ s->width = avctx->width;
+ s->height = avctx->height;
h->pred4x4[DIAG_DOWN_LEFT_PRED] = pred4x4_down_left_svq3_c;
h->pred16x16[PLANE_PRED8x8] = pred16x16_plane_svq3_c;
h->halfpel_flag = 1;
h->thirdpel_flag = 1;
+ h->unknown_svq3_flag = 0;
h->chroma_qp = 4;
if (MPV_common_init (s) < 0)
return -1;
+ h->b_stride = 4*s->mb_width;
+
alloc_tables (h);
- }
- if (avctx->extradata && avctx->extradata_size >= 0x63
- && !memcmp (avctx->extradata, "SVQ3", 4)) {
- uint8_t *stsd = (uint8_t *) avctx->extradata + 0x62;
+ if (avctx->extradata && avctx->extradata_size >= 0x64
+ && !memcmp (avctx->extradata, "SVQ3", 4)) {
+
+ GetBitContext gb;
- if ((*stsd >> 5) != 7 || avctx->extradata_size >= 0x66) {
+ init_get_bits (&gb, (uint8_t *) avctx->extradata + 0x62,
+ 8*(avctx->extradata_size - 0x62));
- if ((*stsd >> 5) == 7) {
- stsd += 3; /* skip width, height (12 bits each) */
+ /* 'frame size code' and optional 'width, height' */
+ if (get_bits (&gb, 3) == 7) {
+ get_bits (&gb, 12);
+ get_bits (&gb, 12);
}
- h->halfpel_flag = (*stsd >> 4) & 1;
- h->thirdpel_flag = (*stsd >> 3) & 1;
- }
- }
+ h->halfpel_flag = get_bits1 (&gb);
+ h->thirdpel_flag = get_bits1 (&gb);
- if ((buf[0] & 0x9F) != 1) {
- /* TODO: what? */
- fprintf (stderr, "unsupported header (%02X)\n", buf[0]);
- return -1;
- } else {
- int length = (buf[0] >> 5) & 3;
- int offset = 0;
+ /* unknown fields */
+ get_bits1 (&gb);
+ get_bits1 (&gb);
+ get_bits1 (&gb);
+ get_bits1 (&gb);
- for (i=0; i < length; i++) {
- offset = (offset << 8) | buf[i + 1];
- }
+ s->low_delay = get_bits1 (&gb);
- if (buf_size < (offset + length + 1) || length == 0)
- return -1;
+ /* unknown field */
+ get_bits1 (&gb);
- memcpy (&buf[2], &buf[offset + 2], (length - 1));
- }
+ while (get_bits1 (&gb)) {
+ get_bits (&gb, 8);
+ }
- init_get_bits (&s->gb, &buf[2], 8*(buf_size - 2));
+ h->unknown_svq3_flag = get_bits1 (&gb);
+ avctx->has_b_frames = !s->low_delay;
+ }
+ }
- if ((i = svq3_get_ue_golomb (&s->gb)) == INVALID_VLC || i >= 3)
- return -1;
+ /* special case for last picture */
+ if (buf_size == 0) {
+ if (s->next_picture_ptr && !s->low_delay) {
+ *(AVFrame *) data = *(AVFrame *) &s->next_picture;
+ *data_size = sizeof(AVFrame);
+ }
+ return 0;
+ }
- s->pict_type = golomb_to_pict_type[i];
+ init_get_bits (&s->gb, buf, 8*buf_size);
- /* unknown fields */
- get_bits (&s->gb, 1);
- get_bits (&s->gb, 8);
+ s->mb_x = s->mb_y = 0;
- s->qscale = get_bits (&s->gb, 5);
- s->adaptive_quant = get_bits (&s->gb, 1);
+ if (svq3_decode_slice_header (h))
+ return -1;
- /* unknown fields */
- get_bits (&s->gb, 1);
- get_bits (&s->gb, 1);
- get_bits (&s->gb, 2);
+ s->pict_type = h->slice_type;
+ s->picture_number = h->slice_num;
- while (get_bits (&s->gb, 1)) {
- get_bits (&s->gb, 8);
- }
-
if(avctx->debug&FF_DEBUG_PICT_INFO){
- printf("%c hpel:%d, tpel:%d aqp:%d qp:%d\n",
+ av_log(h->s.avctx, AV_LOG_DEBUG, "%c hpel:%d, tpel:%d aqp:%d qp:%d\n",
av_get_pict_type_char(s->pict_type), h->halfpel_flag, h->thirdpel_flag,
s->adaptive_quant, s->qscale
);
}
- /* B-frames are not supported */
- if (s->pict_type == B_TYPE/* && avctx->hurry_up*/)
- return buf_size;
+ /* for hurry_up==5 */
+ s->current_picture.pict_type = s->pict_type;
+ s->current_picture.key_frame = (s->pict_type == I_TYPE);
+
+ /* skip b frames if we dont have reference frames */
+ if (s->last_picture_ptr == NULL && s->pict_type == B_TYPE) return 0;
+ /* skip b frames if we are in a hurry */
+ if (avctx->hurry_up && s->pict_type == B_TYPE) return 0;
+ /* skip everything if we are in a hurry >= 5 */
+ if (avctx->hurry_up >= 5) return 0;
+
+ if (s->next_p_frame_damaged) {
+ if (s->pict_type == B_TYPE)
+ return 0;
+ else
+ s->next_p_frame_damaged = 0;
+ }
frame_start (h);
- for(i=0; i<4; i++){
- int j;
- for(j=-1; j<4; j++)
- h->ref_cache[0][scan8[0] + 8*i + j]= 1;
- h->ref_cache[0][scan8[0] + 8*i + j]= PART_NOT_AVAILABLE;
+ if (s->pict_type == B_TYPE) {
+ h->frame_num_offset = (h->slice_num - h->prev_frame_num);
+
+ if (h->frame_num_offset < 0) {
+ h->frame_num_offset += 256;
+ }
+ if (h->frame_num_offset == 0 || h->frame_num_offset >= h->prev_frame_num_offset) {
+ av_log(h->s.avctx, AV_LOG_ERROR, "error in B-frame picture id\n");
+ return -1;
+ }
+ } else {
+ h->prev_frame_num = h->frame_num;
+ h->frame_num = h->slice_num;
+ h->prev_frame_num_offset = (h->frame_num - h->prev_frame_num);
+
+ if (h->prev_frame_num_offset < 0) {
+ h->prev_frame_num_offset += 256;
+ }
+ }
+
+ for(m=0; m<2; m++){
+ int i;
+ for(i=0; i<4; i++){
+ int j;
+ for(j=-1; j<4; j++)
+ h->ref_cache[m][scan8[0] + 8*i + j]= 1;
+ h->ref_cache[m][scan8[0] + 8*i + j]= PART_NOT_AVAILABLE;
+ }
}
for (s->mb_y=0; s->mb_y < s->mb_height; s->mb_y++) {
for (s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
- int mb_type = svq3_get_ue_golomb (&s->gb);
+
+ if ( (s->gb.index + 7) >= s->gb.size_in_bits &&
+ ((s->gb.index & 7) == 0 || show_bits (&s->gb, (-s->gb.index & 7)) == 0)) {
+
+ s->gb.index = h->next_slice_index;
+ s->gb.size_in_bits = 8*buf_size;
+
+ if (svq3_decode_slice_header (h))
+ return -1;
+
+ /* TODO: support s->mb_skip_run */
+ }
+
+ mb_type = svq3_get_ue_golomb (&s->gb);
if (s->pict_type == I_TYPE) {
mb_type += 8;
+ } else if (s->pict_type == B_TYPE && mb_type >= 4) {
+ mb_type += 4;
}
- if (mb_type > 32 || svq3_decode_mb (h, mb_type)) {
- fprintf (stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
+ if (mb_type > 33 || svq3_decode_mb (h, mb_type)) {
+ av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
return -1;
}
if (mb_type != 0) {
hl_decode_mb (h);
}
+
+ if (s->pict_type != B_TYPE && !s->low_delay) {
+ s->current_picture.mb_type[s->mb_x + s->mb_y*s->mb_stride] =
+ (s->pict_type == P_TYPE && mb_type < 8) ? (mb_type - 1) : -1;
+ }
}
- }
- *(AVFrame *) data = *(AVFrame *) &s->current_picture;
- *data_size = sizeof(AVFrame);
+ ff_draw_horiz_band(s, 16*s->mb_y, 16);
+ }
MPV_frame_end(s);
-
+
+ if (s->pict_type == B_TYPE || s->low_delay) {
+ *(AVFrame *) data = *(AVFrame *) &s->current_picture;
+ } else {
+ *(AVFrame *) data = *(AVFrame *) &s->last_picture;
+ }
+
+ avctx->frame_number = s->picture_number - 1;
+
+ /* dont output the last pic after seeking */
+ if (s->last_picture_ptr || s->low_delay) {
+ *data_size = sizeof(AVFrame);
+ }
+
return buf_size;
}
NULL,
decode_end,
svq3_decode_frame,
- CODEC_CAP_DR1,
+ CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_DR1,
};
projects / ffmpeg / blobdiff