* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
+#include "libavcore/imgutils.h"
#include "avcodec.h"
#include "vp56.h"
#include "vp8data.h"
VP8Macroblock *macroblocks;
VP8Macroblock *macroblocks_base;
VP8FilterStrength *filter_strength;
- int mb_stride;
uint8_t *intra4x4_pred_mode_top;
uint8_t intra4x4_pred_mode_left[4];
uint8_t *segmentation_map;
- int b4_stride;
/**
* Cache of the top row needed for intra prediction
static int update_dimensions(VP8Context *s, int width, int height)
{
- if (avcodec_check_dimensions(s->avctx, width, height))
+ if (av_image_check_size(width, height, 0, s->avctx))
return AVERROR_INVALIDDATA;
vp8_decode_flush(s->avctx);
s->mb_width = (s->avctx->coded_width +15) / 16;
s->mb_height = (s->avctx->coded_height+15) / 16;
- // we allocate a border around the top/left of intra4x4 modes
- // this is 4 blocks for intra4x4 to keep 4-byte alignment for fill_rectangle
- s->mb_stride = s->mb_width+1;
- s->b4_stride = 4*s->mb_stride;
-
- s->macroblocks_base = av_mallocz((s->mb_stride+s->mb_height*2+2)*sizeof(*s->macroblocks));
- s->filter_strength = av_mallocz(s->mb_stride*sizeof(*s->filter_strength));
- s->intra4x4_pred_mode_top = av_mallocz(s->b4_stride*4);
+ s->macroblocks_base = av_mallocz((s->mb_width+s->mb_height*2+1)*sizeof(*s->macroblocks));
+ s->filter_strength = av_mallocz(s->mb_width*sizeof(*s->filter_strength));
+ s->intra4x4_pred_mode_top = av_mallocz(s->mb_width*4);
s->top_nnz = av_mallocz(s->mb_width*sizeof(*s->top_nnz));
s->top_border = av_mallocz((s->mb_width+1)*sizeof(*s->top_border));
- s->segmentation_map = av_mallocz(s->mb_stride*s->mb_height);
+ s->segmentation_map = av_mallocz(s->mb_width*s->mb_height);
if (!s->macroblocks_base || !s->filter_strength || !s->intra4x4_pred_mode_top ||
!s->top_nnz || !s->top_border || !s->segmentation_map)
if (buf_size - size < 0)
return -1;
- vp56_init_range_decoder(&s->coeff_partition[i], buf, size);
+ ff_vp56_init_range_decoder(&s->coeff_partition[i], buf, size);
buf += size;
buf_size -= size;
}
- vp56_init_range_decoder(&s->coeff_partition[i], buf, buf_size);
+ ff_vp56_init_range_decoder(&s->coeff_partition[i], buf, buf_size);
return 0;
}
return ret;
}
- vp56_init_range_decoder(c, buf, header_size);
+ ff_vp56_init_range_decoder(c, buf, header_size);
buf += header_size;
buf_size -= header_size;
}
static av_always_inline
-void find_near_mvs(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y,
+void find_near_mvs(VP8Context *s, VP8Macroblock *mb,
VP56mv near[2], VP56mv *best, uint8_t cnt[4])
{
VP8Macroblock *mb_edge[3] = { mb + 2 /* top */,
s->ref_count[mb->ref_frame-1]++;
// motion vectors, 16.3
- find_near_mvs(s, mb, mb_x, mb_y, near, &best, cnt);
+ find_near_mvs(s, mb, near, &best, cnt);
if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[0]][0])) {
if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[1]][1])) {
if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[2]][2])) {
} else {
mb->mode = VP8_MVMODE_NEW;
clamp_mv(s, &mb->mv, &mb->mv, mb_x, mb_y);
- mb->mv.y += + read_mv_component(c, s->prob->mvc[0]);
- mb->mv.x += + read_mv_component(c, s->prob->mvc[1]);
+ mb->mv.y += read_mv_component(c, s->prob->mvc[0]);
+ mb->mv.x += read_mv_component(c, s->prob->mvc[1]);
}
} else {
mb->mode = VP8_MVMODE_NEAR;
* @return 0 if no coeffs were decoded
* otherwise, the index of the last coeff decoded plus one
*/
-static int decode_block_coeffs(VP56RangeCoder *c, DCTELEM block[16],
- uint8_t probs[8][3][NUM_DCT_TOKENS-1],
- int i, int zero_nhood, int16_t qmul[2])
+static int decode_block_coeffs_internal(VP56RangeCoder *c, DCTELEM block[16],
+ uint8_t probs[8][3][NUM_DCT_TOKENS-1],
+ int i, uint8_t *token_prob, int16_t qmul[2])
{
- uint8_t *token_prob = probs[i][zero_nhood];
- int nonzero = 0;
- int coeff;
-
+ goto skip_eob;
do {
+ int coeff;
if (!vp56_rac_get_prob_branchy(c, token_prob[0])) // DCT_EOB
- return nonzero;
+ return i;
skip_eob:
if (!vp56_rac_get_prob_branchy(c, token_prob[1])) { // DCT_0
if (++i == 16)
- return nonzero; // invalid input; blocks should end with EOB
+ return i; // invalid input; blocks should end with EOB
token_prob = probs[i][0];
goto skip_eob;
}
}
token_prob = probs[i+1][2];
}
-
- // todo: full [16] qmat? load into register?
block[zigzag_scan[i]] = (vp8_rac_get(c) ? -coeff : coeff) * qmul[!!i];
- nonzero = ++i;
- } while (i < 16);
+ } while (++i < 16);
- return nonzero;
+ return i;
+}
+
+static av_always_inline
+int decode_block_coeffs(VP56RangeCoder *c, DCTELEM block[16],
+ uint8_t probs[8][3][NUM_DCT_TOKENS-1],
+ int i, int zero_nhood, int16_t qmul[2])
+{
+ uint8_t *token_prob = probs[i][zero_nhood];
+ if (!vp56_rac_get_prob_branchy(c, token_prob[0])) // DCT_EOB
+ return 0;
+ return decode_block_coeffs_internal(c, block, probs, i, token_prob, qmul);
}
static av_always_inline
}
static av_always_inline
-int check_intra_pred_mode(int mode, int mb_x, int mb_y)
+int check_dc_pred8x8_mode(int mode, int mb_x, int mb_y)
+{
+ if (!mb_x) {
+ return mb_y ? TOP_DC_PRED8x8 : DC_128_PRED8x8;
+ } else {
+ return mb_y ? mode : LEFT_DC_PRED8x8;
+ }
+}
+
+static av_always_inline
+int check_tm_pred8x8_mode(int mode, int mb_x, int mb_y)
+{
+ if (!mb_x) {
+ return mb_y ? VERT_PRED8x8 : DC_129_PRED8x8;
+ } else {
+ return mb_y ? mode : HOR_PRED8x8;
+ }
+}
+
+static av_always_inline
+int check_intra_pred8x8_mode(int mode, int mb_x, int mb_y)
{
if (mode == DC_PRED8x8) {
- if (!mb_x) {
- mode = mb_y ? TOP_DC_PRED8x8 : DC_128_PRED8x8;
- } else if (!mb_y) {
- mode = LEFT_DC_PRED8x8;
+ return check_dc_pred8x8_mode(mode, mb_x, mb_y);
+ } else {
+ return mode;
+ }
+}
+
+static av_always_inline
+int check_intra_pred8x8_mode_emuedge(int mode, int mb_x, int mb_y)
+{
+ switch (mode) {
+ case DC_PRED8x8:
+ return check_dc_pred8x8_mode(mode, mb_x, mb_y);
+ case VERT_PRED8x8:
+ return !mb_y ? DC_127_PRED8x8 : mode;
+ case HOR_PRED8x8:
+ return !mb_x ? DC_129_PRED8x8 : mode;
+ case PLANE_PRED8x8 /*TM*/:
+ return check_tm_pred8x8_mode(mode, mb_x, mb_y);
+ }
+ return mode;
+}
+
+static av_always_inline
+int check_tm_pred4x4_mode(int mode, int mb_x, int mb_y)
+{
+ if (!mb_x) {
+ return mb_y ? VERT_VP8_PRED : DC_129_PRED;
+ } else {
+ return mb_y ? mode : HOR_VP8_PRED;
+ }
+}
+
+static av_always_inline
+int check_intra_pred4x4_mode_emuedge(int mode, int mb_x, int mb_y, int *copy_buf)
+{
+ switch (mode) {
+ case VERT_PRED:
+ if (!mb_x && mb_y) {
+ *copy_buf = 1;
+ return mode;
}
+ /* fall-through */
+ case DIAG_DOWN_LEFT_PRED:
+ case VERT_LEFT_PRED:
+ return !mb_y ? DC_127_PRED : mode;
+ case HOR_PRED:
+ if (!mb_y) {
+ *copy_buf = 1;
+ return mode;
+ }
+ /* fall-through */
+ case HOR_UP_PRED:
+ return !mb_x ? DC_129_PRED : mode;
+ case TM_VP8_PRED:
+ return check_tm_pred4x4_mode(mode, mb_x, mb_y);
+ case DC_PRED: // 4x4 DC doesn't use the same "H.264-style" exceptions as 16x16/8x8 DC
+ case DIAG_DOWN_RIGHT_PRED:
+ case VERT_RIGHT_PRED:
+ case HOR_DOWN_PRED:
+ if (!mb_y || !mb_x)
+ *copy_buf = 1;
+ return mode;
}
return mode;
}
void intra_predict(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb,
int mb_x, int mb_y)
{
+ AVCodecContext *avctx = s->avctx;
int x, y, mode, nnz, tr;
// for the first row, we need to run xchg_mb_border to init the top edge to 127
// otherwise, skip it if we aren't going to deblock
- if (s->deblock_filter || !mb_y)
+ if (!(avctx->flags & CODEC_FLAG_EMU_EDGE && !mb_y) && (s->deblock_filter || !mb_y))
xchg_mb_border(s->top_border[mb_x+1], dst[0], dst[1], dst[2],
s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width,
s->filter.simple, 1);
if (mb->mode < MODE_I4x4) {
- mode = check_intra_pred_mode(mb->mode, mb_x, mb_y);
+ if (avctx->flags & CODEC_FLAG_EMU_EDGE) { // tested
+ mode = check_intra_pred8x8_mode_emuedge(mb->mode, mb_x, mb_y);
+ } else {
+ mode = check_intra_pred8x8_mode(mb->mode, mb_x, mb_y);
+ }
s->hpc.pred16x16[mode](dst[0], s->linesize);
} else {
uint8_t *ptr = dst[0];
uint8_t *intra4x4 = s->intra4x4_pred_mode_mb;
+ uint8_t tr_top[4] = { 127, 127, 127, 127 };
// all blocks on the right edge of the macroblock use bottom edge
// the top macroblock for their topright edge
// if we're on the right edge of the frame, said edge is extended
// from the top macroblock
- if (mb_x == s->mb_width-1) {
+ if (!(!mb_y && avctx->flags & CODEC_FLAG_EMU_EDGE) &&
+ mb_x == s->mb_width-1) {
tr = tr_right[-1]*0x01010101;
tr_right = (uint8_t *)&tr;
}
for (y = 0; y < 4; y++) {
uint8_t *topright = ptr + 4 - s->linesize;
for (x = 0; x < 4; x++) {
- if (x == 3)
+ int copy = 0, linesize = s->linesize;
+ uint8_t *dst = ptr+4*x;
+ DECLARE_ALIGNED(4, uint8_t, copy_dst)[5*8];
+
+ if ((y == 0 || x == 3) && mb_y == 0 && avctx->flags & CODEC_FLAG_EMU_EDGE) {
+ topright = tr_top;
+ } else if (x == 3)
topright = tr_right;
- s->hpc.pred4x4[intra4x4[x]](ptr+4*x, topright, s->linesize);
+ if (avctx->flags & CODEC_FLAG_EMU_EDGE) { // mb_x+x or mb_y+y is a hack but works
+ mode = check_intra_pred4x4_mode_emuedge(intra4x4[x], mb_x + x, mb_y + y, ©);
+ if (copy) {
+ dst = copy_dst + 12;
+ linesize = 8;
+ if (!(mb_y + y)) {
+ copy_dst[3] = 127U;
+ * (uint32_t *) (copy_dst + 4) = 127U * 0x01010101U;
+ } else {
+ * (uint32_t *) (copy_dst + 4) = * (uint32_t *) (ptr+4*x-s->linesize);
+ if (!(mb_x + x)) {
+ copy_dst[3] = 129U;
+ } else {
+ copy_dst[3] = ptr[4*x-s->linesize-1];
+ }
+ }
+ if (!(mb_x + x)) {
+ copy_dst[11] =
+ copy_dst[19] =
+ copy_dst[27] =
+ copy_dst[35] = 129U;
+ } else {
+ copy_dst[11] = ptr[4*x -1];
+ copy_dst[19] = ptr[4*x+s->linesize -1];
+ copy_dst[27] = ptr[4*x+s->linesize*2-1];
+ copy_dst[35] = ptr[4*x+s->linesize*3-1];
+ }
+ }
+ } else {
+ mode = intra4x4[x];
+ }
+ s->hpc.pred4x4[mode](dst, topright, linesize);
+ if (copy) {
+ * (uint32_t *) (ptr+4*x) = * (uint32_t *) (copy_dst + 12);
+ * (uint32_t *) (ptr+4*x+s->linesize) = * (uint32_t *) (copy_dst + 20);
+ * (uint32_t *) (ptr+4*x+s->linesize*2) = * (uint32_t *) (copy_dst + 28);
+ * (uint32_t *) (ptr+4*x+s->linesize*3) = * (uint32_t *) (copy_dst + 36);
+ }
nnz = s->non_zero_count_cache[y][x];
if (nnz) {
}
}
- mode = check_intra_pred_mode(s->chroma_pred_mode, mb_x, mb_y);
+ if (avctx->flags & CODEC_FLAG_EMU_EDGE) {
+ mode = check_intra_pred8x8_mode_emuedge(s->chroma_pred_mode, mb_x, mb_y);
+ } else {
+ mode = check_intra_pred8x8_mode(s->chroma_pred_mode, mb_x, mb_y);
+ }
s->hpc.pred8x8[mode](dst[1], s->uvlinesize);
s->hpc.pred8x8[mode](dst[2], s->uvlinesize);
- if (s->deblock_filter || !mb_y)
+ if (!(avctx->flags & CODEC_FLAG_EMU_EDGE && !mb_y) && (s->deblock_filter || !mb_y))
xchg_mb_border(s->top_border[mb_x+1], dst[0], dst[1], dst[2],
s->linesize, s->uvlinesize, mb_x, mb_y, s->mb_width,
s->filter.simple, 0);
/* Don't prefetch refs that haven't been used very often this frame. */
if (s->ref_count[ref-1] > (mb_xy >> 5)) {
int x_off = mb_x << 4, y_off = mb_y << 4;
- int mx = mb->mv.x + x_off + 8;
- int my = mb->mv.y + y_off;
+ int mx = (mb->mv.x>>2) + x_off + 8;
+ int my = (mb->mv.y>>2) + y_off;
uint8_t **src= s->framep[ref]->data;
int off= mx + (my + (mb_x&3)*4)*s->linesize + 64;
s->dsp.prefetch(src[0]+off, s->linesize, 4);
memset(s->top_nnz, 0, s->mb_width*sizeof(*s->top_nnz));
- /* Zero macroblock structures for top/left prediction from outside the frame. */
- memset(s->macroblocks, 0, (s->mb_width + s->mb_height*2)*sizeof(*s->macroblocks));
+ /* Zero macroblock structures for top/top-left prediction from outside the frame. */
+ memset(s->macroblocks + s->mb_height*2 - 1, 0, (s->mb_width+1)*sizeof(*s->macroblocks));
// top edge of 127 for intra prediction
- memset(s->top_border, 127, (s->mb_width+1)*sizeof(*s->top_border));
+ if (!(avctx->flags & CODEC_FLAG_EMU_EDGE)) {
+ s->top_border[0][15] = s->top_border[0][23] = 127;
+ memset(s->top_border[1]-1, 127, s->mb_width*sizeof(*s->top_border)+1);
+ }
memset(s->ref_count, 0, sizeof(s->ref_count));
if (s->keyframe)
- memset(s->intra4x4_pred_mode_top, DC_PRED, s->b4_stride*4);
+ memset(s->intra4x4_pred_mode_top, DC_PRED, s->mb_width*4);
for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
VP56RangeCoder *c = &s->coeff_partition[mb_y & (s->num_coeff_partitions-1)];
VP8Macroblock *mb = s->macroblocks + (s->mb_height - mb_y - 1)*2;
- uint8_t *segment_map = s->segmentation_map + mb_y*s->mb_stride;
- int mb_xy = mb_y * s->mb_stride;
+ int mb_xy = mb_y*s->mb_width;
uint8_t *dst[3] = {
curframe->data[0] + 16*mb_y*s->linesize,
curframe->data[1] + 8*mb_y*s->uvlinesize,
curframe->data[2] + 8*mb_y*s->uvlinesize
};
+ memset(mb - 1, 0, sizeof(*mb)); // zero left macroblock
memset(s->left_nnz, 0, sizeof(s->left_nnz));
AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED*0x01010101);
// left edge of 129 for intra prediction
- if (!(avctx->flags & CODEC_FLAG_EMU_EDGE))
+ if (!(avctx->flags & CODEC_FLAG_EMU_EDGE)) {
for (i = 0; i < 3; i++)
for (y = 0; y < 16>>!!i; y++)
dst[i][y*curframe->linesize[i]-1] = 129;
- if (mb_y)
- memset(s->top_border, 129, sizeof(*s->top_border));
+ if (mb_y == 1) // top left edge is also 129
+ s->top_border[0][15] = s->top_border[0][23] = s->top_border[0][31] = 129;
+ }
for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) {
- uint8_t *segment_mb = segment_map+mb_x;
-
/* Prefetch the current frame, 4 MBs ahead */
s->dsp.prefetch(dst[0] + (mb_x&3)*4*s->linesize + 64, s->linesize, 4);
s->dsp.prefetch(dst[1] + (mb_x&7)*s->uvlinesize + 64, dst[2] - dst[1], 2);
- decode_mb_mode(s, mb, mb_x, mb_y, segment_mb);
+ decode_mb_mode(s, mb, mb_x, mb_y, s->segmentation_map + mb_xy);
prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_PREVIOUS);
ff_h264_pred_init(&s->hpc, CODEC_ID_VP8);
ff_vp8dsp_init(&s->vp8dsp);
- // intra pred needs edge emulation among other things
- if (avctx->flags&CODEC_FLAG_EMU_EDGE) {
- av_log(avctx, AV_LOG_ERROR, "Edge emulation not supported\n");
- return AVERROR_PATCHWELCOME;
- }
-
return 0;
}
projects / ffmpeg / blobdiff