*
* Copyright (C) 2010 David Conrad
* Copyright (C) 2010 Ronald S. Bultje
+ * Copyright (C) 2010 Jason Garrett-Glaser
*
* This file is part of FFmpeg.
*
* 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"
#include "rectangle.h"
typedef struct {
- uint8_t segment;
+ uint8_t filter_level;
+ uint8_t inner_limit;
+ uint8_t inner_filter;
+} VP8FilterStrength;
+
+typedef struct {
uint8_t skip;
// todo: make it possible to check for at least (i4x4 or split_mv)
// in one op. are others needed?
DSPContext dsp;
VP8DSPContext vp8dsp;
H264PredContext hpc;
+ vp8_mc_func put_pixels_tab[3][3][3];
AVFrame frames[4];
AVFrame *framep[4];
uint8_t *edge_emu_buffer;
int update_last; ///< update VP56_FRAME_PREVIOUS with the current one
int update_golden; ///< VP56_FRAME_NONE if not updated, or which frame to copy if so
int update_altref;
+ int deblock_filter;
/**
* If this flag is not set, all the probability updates
VP8Macroblock *macroblocks;
VP8Macroblock *macroblocks_base;
- int mb_stride;
+ VP8FilterStrength *filter_strength;
+
+ uint8_t *intra4x4_pred_mode_top;
+ uint8_t intra4x4_pred_mode_left[4];
+ uint8_t *segmentation_map;
- uint8_t *intra4x4_pred_mode;
- uint8_t *intra4x4_pred_mode_base;
- int b4_stride;
+ /**
+ * Cache of the top row needed for intra prediction
+ * 16 for luma, 8 for each chroma plane
+ */
+ uint8_t (*top_border)[16+8+8];
/**
* For coeff decode, we need to know whether the above block had non-zero
*/
DECLARE_ALIGNED(16, uint8_t, non_zero_count_cache)[6][4];
DECLARE_ALIGNED(16, DCTELEM, block)[6][4][16];
+ DECLARE_ALIGNED(16, DCTELEM, block_dc)[16];
+ uint8_t intra4x4_pred_mode_mb[16];
int chroma_pred_mode; ///< 8x8c pred mode of the current macroblock
+ int segment; ///< segment of the current macroblock
int mbskip_enabled;
int sign_bias[4]; ///< one state [0, 1] per ref frame type
+ int ref_count[3];
/**
* Base parameters for segmentation, i.e. per-macroblock parameters.
uint8_t golden;
uint8_t pred16x16[4];
uint8_t pred8x8c[3];
- uint8_t token[4][8][3][NUM_DCT_TOKENS-1];
+ /* Padded to allow overreads */
+ uint8_t token[4][17][3][NUM_DCT_TOKENS-1];
uint8_t mvc[2][19];
} prob[2];
} VP8Context;
-#define RL24(p) (AV_RL16(p) + ((p)[2] << 16))
-
static void vp8_decode_flush(AVCodecContext *avctx)
{
VP8Context *s = avctx->priv_data;
memset(s->framep, 0, sizeof(s->framep));
av_freep(&s->macroblocks_base);
- av_freep(&s->intra4x4_pred_mode_base);
+ av_freep(&s->filter_strength);
+ av_freep(&s->intra4x4_pred_mode_top);
av_freep(&s->top_nnz);
av_freep(&s->edge_emu_buffer);
+ av_freep(&s->top_border);
+ av_freep(&s->segmentation_map);
s->macroblocks = NULL;
- s->intra4x4_pred_mode = NULL;
}
static int update_dimensions(VP8Context *s, int width, int height)
{
- int i;
-
- 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+1)*sizeof(*s->macroblocks));
- s->intra4x4_pred_mode_base = av_mallocz(s->b4_stride*(4*s->mb_height+1));
+ 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_width*s->mb_height);
- s->macroblocks = s->macroblocks_base + 1 + s->mb_stride;
- s->intra4x4_pred_mode = s->intra4x4_pred_mode_base + 4 + s->b4_stride;
+ if (!s->macroblocks_base || !s->filter_strength || !s->intra4x4_pred_mode_top ||
+ !s->top_nnz || !s->top_border || !s->segmentation_map)
+ return AVERROR(ENOMEM);
- memset(s->intra4x4_pred_mode_base, DC_PRED, s->b4_stride);
- for (i = 0; i < 4*s->mb_height; i++)
- s->intra4x4_pred_mode[i*s->b4_stride-1] = DC_PRED;
+ s->macroblocks = s->macroblocks_base + 1;
return 0;
}
return -1;
for (i = 0; i < s->num_coeff_partitions-1; i++) {
- int size = RL24(sizes + 3*i);
+ int size = AV_RL24(sizes + 3*i);
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;
}
static int decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_size)
{
VP56RangeCoder *c = &s->c;
- int header_size, hscale, vscale, i, j, k, l, ret;
+ int header_size, hscale, vscale, i, j, k, l, m, ret;
int width = s->avctx->width;
int height = s->avctx->height;
s->keyframe = !(buf[0] & 1);
s->profile = (buf[0]>>1) & 7;
s->invisible = !(buf[0] & 0x10);
- header_size = RL24(buf) >> 5;
+ header_size = AV_RL24(buf) >> 5;
buf += 3;
buf_size -= 3;
- if (s->profile)
- av_log(s->avctx, AV_LOG_WARNING, "Profile %d not fully handled\n", s->profile);
+ if (s->profile > 3)
+ av_log(s->avctx, AV_LOG_WARNING, "Unknown profile %d\n", s->profile);
+
+ if (!s->profile)
+ memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_epel_pixels_tab, sizeof(s->put_pixels_tab));
+ else // profile 1-3 use bilinear, 4+ aren't defined so whatever
+ memcpy(s->put_pixels_tab, s->vp8dsp.put_vp8_bilinear_pixels_tab, sizeof(s->put_pixels_tab));
if (header_size > buf_size - 7*s->keyframe) {
av_log(s->avctx, AV_LOG_ERROR, "Header size larger than data provided\n");
}
if (s->keyframe) {
- if (RL24(buf) != 0x2a019d) {
- av_log(s->avctx, AV_LOG_ERROR, "Invalid start code 0x%x\n", RL24(buf));
+ if (AV_RL24(buf) != 0x2a019d) {
+ av_log(s->avctx, AV_LOG_ERROR, "Invalid start code 0x%x\n", AV_RL24(buf));
return AVERROR_INVALIDDATA;
}
width = AV_RL16(buf+3) & 0x3fff;
buf += 7;
buf_size -= 7;
+ if (hscale || vscale)
+ av_log_missing_feature(s->avctx, "Upscaling", 1);
+
s->update_golden = s->update_altref = VP56_FRAME_CURRENT;
- memcpy(s->prob->token , vp8_token_default_probs , sizeof(s->prob->token));
+ for (i = 0; i < 4; i++)
+ for (j = 0; j < 16; j++)
+ memcpy(s->prob->token[i][j], vp8_token_default_probs[i][vp8_coeff_band[j]],
+ sizeof(s->prob->token[i][j]));
memcpy(s->prob->pred16x16, vp8_pred16x16_prob_inter, sizeof(s->prob->pred16x16));
memcpy(s->prob->pred8x8c , vp8_pred8x8c_prob_inter , sizeof(s->prob->pred8x8c));
memcpy(s->prob->mvc , vp8_mv_default_prob , sizeof(s->prob->mvc));
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;
for (j = 0; j < 8; j++)
for (k = 0; k < 3; k++)
for (l = 0; l < NUM_DCT_TOKENS-1; l++)
- if (vp56_rac_get_prob(c, vp8_token_update_probs[i][j][k][l]))
- s->prob->token[i][j][k][l] = vp8_rac_get_uint(c, 8);
+ if (vp56_rac_get_prob_branchy(c, vp8_token_update_probs[i][j][k][l])) {
+ int prob = vp8_rac_get_uint(c, 8);
+ for (m = 0; vp8_coeff_band_indexes[j][m] >= 0; m++)
+ s->prob->token[i][vp8_coeff_band_indexes[j][m]][k][l] = prob;
+ }
if ((s->mbskip_enabled = vp8_rac_get(c)))
s->prob->mbskip = vp8_rac_get_uint(c, 8);
// 17.2 MV probability update
for (i = 0; i < 2; i++)
for (j = 0; j < 19; j++)
- if (vp56_rac_get_prob(c, vp8_mv_update_prob[i][j]))
+ if (vp56_rac_get_prob_branchy(c, vp8_mv_update_prob[i][j]))
s->prob->mvc[i][j] = vp8_rac_get_nn(c);
}
return 0;
}
-static inline void clamp_mv(VP8Context *s, VP56mv *dst, const VP56mv *src,
- int mb_x, int mb_y)
+static av_always_inline
+void clamp_mv(VP8Context *s, VP56mv *dst, const VP56mv *src, int mb_x, int mb_y)
{
#define MARGIN (16 << 2)
dst->x = av_clip(src->x, -((mb_x << 6) + MARGIN),
((s->mb_height - 1 - mb_y) << 6) + MARGIN);
}
-static void find_near_mvs(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y,
- VP56mv near[2], VP56mv *best, int cnt[4])
+static av_always_inline
+void find_near_mvs(VP8Context *s, VP8Macroblock *mb,
+ VP56mv near[2], VP56mv *best, uint8_t cnt[4])
{
- VP8Macroblock *mb_edge[3] = { mb - s->mb_stride /* top */,
- mb - 1 /* left */,
- mb - s->mb_stride - 1 /* top-left */ };
+ VP8Macroblock *mb_edge[3] = { mb + 2 /* top */,
+ mb - 1 /* left */,
+ mb + 1 /* top-left */ };
enum { EDGE_TOP, EDGE_LEFT, EDGE_TOPLEFT };
VP56mv near_mv[4] = {{ 0 }};
enum { CNT_ZERO, CNT_NEAREST, CNT_NEAR, CNT_SPLITMV };
- int idx = CNT_ZERO, n;
+ int idx = CNT_ZERO;
int best_idx = CNT_ZERO;
+ int cur_sign_bias = s->sign_bias[mb->ref_frame];
+ int *sign_bias = s->sign_bias;
/* Process MB on top, left and top-left */
- for (n = 0; n < 3; n++) {
- VP8Macroblock *edge = mb_edge[n];
- if (edge->ref_frame != VP56_FRAME_CURRENT) {
- if (edge->mv.x | edge->mv.y) {
- VP56mv tmp = edge->mv;
- if (s->sign_bias[mb->ref_frame] != s->sign_bias[edge->ref_frame]) {
- tmp.x *= -1;
- tmp.y *= -1;
- }
- if ((tmp.x ^ near_mv[idx].x) | (tmp.y ^ near_mv[idx].y))
- near_mv[++idx] = tmp;
- cnt[idx] += 1 + (n != 2);
- } else
- cnt[CNT_ZERO] += 1 + (n != 2);
- }
+ #define MV_EDGE_CHECK(n)\
+ {\
+ VP8Macroblock *edge = mb_edge[n];\
+ int edge_ref = edge->ref_frame;\
+ if (edge_ref != VP56_FRAME_CURRENT) {\
+ uint32_t mv = AV_RN32A(&edge->mv);\
+ if (mv) {\
+ if (cur_sign_bias != sign_bias[edge_ref]) {\
+ /* SWAR negate of the values in mv. */\
+ mv = ~mv;\
+ mv = ((mv&0x7fff7fff) + 0x00010001) ^ (mv&0x80008000);\
+ }\
+ if (!n || mv != AV_RN32A(&near_mv[idx]))\
+ AV_WN32A(&near_mv[++idx], mv);\
+ cnt[idx] += 1 + (n != 2);\
+ } else\
+ cnt[CNT_ZERO] += 1 + (n != 2);\
+ }\
}
+ MV_EDGE_CHECK(0)
+ MV_EDGE_CHECK(1)
+ MV_EDGE_CHECK(2)
- /* If we have three distinct MV's, merge first and last if they're the same */
- if (cnt[CNT_SPLITMV] &&
- !((near_mv[1+EDGE_TOP].x ^ near_mv[1+EDGE_TOPLEFT].x) |
- (near_mv[1+EDGE_TOP].y ^ near_mv[1+EDGE_TOPLEFT].y)))
+ /* If we have three distinct MVs, merge first and last if they're the same */
+ if (cnt[CNT_SPLITMV] && AV_RN32A(&near_mv[1+EDGE_TOP]) == AV_RN32A(&near_mv[1+EDGE_TOPLEFT]))
cnt[CNT_NEAREST] += 1;
cnt[CNT_SPLITMV] = ((mb_edge[EDGE_LEFT]->mode == VP8_MVMODE_SPLIT) +
/* Swap near and nearest if necessary */
if (cnt[CNT_NEAR] > cnt[CNT_NEAREST]) {
- FFSWAP(int, cnt[CNT_NEAREST], cnt[CNT_NEAR]);
- FFSWAP(VP56mv, near_mv[CNT_NEAREST], near_mv[CNT_NEAR]);
+ FFSWAP(uint8_t, cnt[CNT_NEAREST], cnt[CNT_NEAR]);
+ FFSWAP( VP56mv, near_mv[CNT_NEAREST], near_mv[CNT_NEAR]);
}
/* Choose the best mv out of 0,0 and the nearest mv */
if (cnt[CNT_NEAREST] >= cnt[CNT_ZERO])
best_idx = CNT_NEAREST;
- clamp_mv(s, best, &near_mv[best_idx], mb_x, mb_y);
+ mb->mv = near_mv[best_idx];
near[0] = near_mv[CNT_NEAREST];
near[1] = near_mv[CNT_NEAR];
}
*/
static int read_mv_component(VP56RangeCoder *c, const uint8_t *p)
{
- int x = 0;
+ int bit, x = 0;
- if (vp56_rac_get_prob(c, p[0])) {
+ if (vp56_rac_get_prob_branchy(c, p[0])) {
int i;
for (i = 0; i < 3; i++)
x += vp56_rac_get_prob(c, p[9 + i]) << i;
if (!(x & 0xFFF0) || vp56_rac_get_prob(c, p[12]))
x += 8;
- } else
- x = vp8_rac_get_tree(c, vp8_small_mvtree, &p[2]);
+ } else {
+ // small_mvtree
+ const uint8_t *ps = p+2;
+ bit = vp56_rac_get_prob(c, *ps);
+ ps += 1 + 3*bit;
+ x += 4*bit;
+ bit = vp56_rac_get_prob(c, *ps);
+ ps += 1 + bit;
+ x += 2*bit;
+ x += vp56_rac_get_prob(c, *ps);
+ }
return (x && vp56_rac_get_prob(c, p[1])) ? -x : x;
}
-static const uint8_t *get_submv_prob(const VP56mv *left, const VP56mv *top)
+static av_always_inline
+const uint8_t *get_submv_prob(uint32_t left, uint32_t top)
{
- int l_is_zero = !(left->x | left->y);
- int t_is_zero = !(top->x | top->y);
- int equal = !((left->x ^ top->x) | (left->y ^ top->y));
-
- if (equal)
- return l_is_zero ? vp8_submv_prob[4] : vp8_submv_prob[3];
- if (t_is_zero)
+ if (left == top)
+ return vp8_submv_prob[4-!!left];
+ if (!top)
return vp8_submv_prob[2];
- return l_is_zero ? vp8_submv_prob[1] : vp8_submv_prob[0];
+ return vp8_submv_prob[1-!!left];
}
/**
* Split motion vector prediction, 16.4.
+ * @returns the number of motion vectors parsed (2, 4 or 16)
*/
-static void decode_splitmvs(VP8Context *s, VP56RangeCoder *c,
- VP8Macroblock *mb, VP56mv *base_mv)
+static av_always_inline
+int decode_splitmvs(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb)
{
- int part_idx = mb->partitioning =
- vp8_rac_get_tree(c, vp8_mbsplit_tree, vp8_mbsplit_prob);
- int n, num = vp8_mbsplit_count[part_idx];
- VP56mv part_mv[16];
+ int part_idx;
+ int n, num;
+ VP8Macroblock *top_mb = &mb[2];
+ VP8Macroblock *left_mb = &mb[-1];
+ const uint8_t *mbsplits_left = vp8_mbsplits[left_mb->partitioning],
+ *mbsplits_top = vp8_mbsplits[top_mb->partitioning],
+ *mbsplits_cur, *firstidx;
+ VP56mv *top_mv = top_mb->bmv;
+ VP56mv *left_mv = left_mb->bmv;
+ VP56mv *cur_mv = mb->bmv;
+
+ if (vp56_rac_get_prob_branchy(c, vp8_mbsplit_prob[0])) {
+ if (vp56_rac_get_prob_branchy(c, vp8_mbsplit_prob[1])) {
+ part_idx = VP8_SPLITMVMODE_16x8 + vp56_rac_get_prob(c, vp8_mbsplit_prob[2]);
+ } else {
+ part_idx = VP8_SPLITMVMODE_8x8;
+ }
+ } else {
+ part_idx = VP8_SPLITMVMODE_4x4;
+ }
+
+ num = vp8_mbsplit_count[part_idx];
+ mbsplits_cur = vp8_mbsplits[part_idx],
+ firstidx = vp8_mbfirstidx[part_idx];
+ mb->partitioning = part_idx;
for (n = 0; n < num; n++) {
- int k = vp8_mbfirstidx[part_idx][n];
- const VP56mv *left = (k & 3) ? &mb->bmv[k - 1] : &mb[-1].bmv[k + 3],
- *above = (k > 3) ? &mb->bmv[k - 4] : &mb[-s->mb_stride].bmv[k + 12];
- const uint8_t *submv_prob = get_submv_prob(left, above);
-
- switch (vp8_rac_get_tree(c, vp8_submv_ref_tree, submv_prob)) {
- case VP8_SUBMVMODE_NEW4X4:
- part_mv[n].y = base_mv->y + read_mv_component(c, s->prob->mvc[0]);
- part_mv[n].x = base_mv->x + read_mv_component(c, s->prob->mvc[1]);
- break;
- case VP8_SUBMVMODE_ZERO4X4:
- part_mv[n].x = 0;
- part_mv[n].y = 0;
- break;
- case VP8_SUBMVMODE_LEFT4X4:
- part_mv[n] = *left;
- break;
- case VP8_SUBMVMODE_TOP4X4:
- part_mv[n] = *above;
- break;
- }
+ int k = firstidx[n];
+ uint32_t left, above;
+ const uint8_t *submv_prob;
- /* fill out over the 4x4 blocks in MB */
- for (k = 0; k < 16; k++)
- if (vp8_mbsplits[part_idx][k] == n) {
- mb->bmv[k] = part_mv[n];
+ if (!(k & 3))
+ left = AV_RN32A(&left_mv[mbsplits_left[k + 3]]);
+ else
+ left = AV_RN32A(&cur_mv[mbsplits_cur[k - 1]]);
+ if (k <= 3)
+ above = AV_RN32A(&top_mv[mbsplits_top[k + 12]]);
+ else
+ above = AV_RN32A(&cur_mv[mbsplits_cur[k - 4]]);
+
+ submv_prob = get_submv_prob(left, above);
+
+ if (vp56_rac_get_prob_branchy(c, submv_prob[0])) {
+ if (vp56_rac_get_prob_branchy(c, submv_prob[1])) {
+ if (vp56_rac_get_prob_branchy(c, submv_prob[2])) {
+ mb->bmv[n].y = mb->mv.y + read_mv_component(c, s->prob->mvc[0]);
+ mb->bmv[n].x = mb->mv.x + read_mv_component(c, s->prob->mvc[1]);
+ } else {
+ AV_ZERO32(&mb->bmv[n]);
+ }
+ } else {
+ AV_WN32A(&mb->bmv[n], above);
}
+ } else {
+ AV_WN32A(&mb->bmv[n], left);
+ }
}
+
+ return num;
}
-static inline void decode_intra4x4_modes(VP56RangeCoder *c, uint8_t *intra4x4,
- int stride, int keyframe)
+static av_always_inline
+void decode_intra4x4_modes(VP8Context *s, VP56RangeCoder *c,
+ int mb_x, int keyframe)
{
- int x, y, t, l;
- const uint8_t *ctx = vp8_pred4x4_prob_inter;
-
- for (y = 0; y < 4; y++) {
- for (x = 0; x < 4; x++) {
- if (keyframe) {
- t = intra4x4[x - stride];
- l = intra4x4[x - 1];
- ctx = vp8_pred4x4_prob_intra[t][l];
+ uint8_t *intra4x4 = s->intra4x4_pred_mode_mb;
+ if (keyframe) {
+ int x, y;
+ uint8_t* const top = s->intra4x4_pred_mode_top + 4 * mb_x;
+ uint8_t* const left = s->intra4x4_pred_mode_left;
+ for (y = 0; y < 4; y++) {
+ for (x = 0; x < 4; x++) {
+ const uint8_t *ctx;
+ ctx = vp8_pred4x4_prob_intra[top[x]][left[y]];
+ *intra4x4 = vp8_rac_get_tree(c, vp8_pred4x4_tree, ctx);
+ left[y] = top[x] = *intra4x4;
+ intra4x4++;
}
- intra4x4[x] = vp8_rac_get_tree(c, vp8_pred4x4_tree, ctx);
}
- intra4x4 += stride;
+ } else {
+ int i;
+ for (i = 0; i < 16; i++)
+ intra4x4[i] = vp8_rac_get_tree(c, vp8_pred4x4_tree, vp8_pred4x4_prob_inter);
}
}
-static void decode_mb_mode(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y,
- uint8_t *intra4x4)
+static av_always_inline
+void decode_mb_mode(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y, uint8_t *segment)
{
VP56RangeCoder *c = &s->c;
- int n;
if (s->segmentation.update_map)
- mb->segment = vp8_rac_get_tree(c, vp8_segmentid_tree, s->prob->segmentid);
+ *segment = vp8_rac_get_tree(c, vp8_segmentid_tree, s->prob->segmentid);
+ s->segment = *segment;
mb->skip = s->mbskip_enabled ? vp56_rac_get_prob(c, s->prob->mbskip) : 0;
mb->mode = vp8_rac_get_tree(c, vp8_pred16x16_tree_intra, vp8_pred16x16_prob_intra);
if (mb->mode == MODE_I4x4) {
- decode_intra4x4_modes(c, intra4x4, s->b4_stride, 1);
- } else
- fill_rectangle(intra4x4, 4, 4, s->b4_stride, vp8_pred4x4_mode[mb->mode], 1);
+ decode_intra4x4_modes(s, c, mb_x, 1);
+ } else {
+ const uint32_t modes = vp8_pred4x4_mode[mb->mode] * 0x01010101u;
+ AV_WN32A(s->intra4x4_pred_mode_top + 4 * mb_x, modes);
+ AV_WN32A(s->intra4x4_pred_mode_left, modes);
+ }
s->chroma_pred_mode = vp8_rac_get_tree(c, vp8_pred8x8c_tree, vp8_pred8x8c_prob_intra);
mb->ref_frame = VP56_FRAME_CURRENT;
- } else if (vp56_rac_get_prob(c, s->prob->intra)) {
+ } else if (vp56_rac_get_prob_branchy(c, s->prob->intra)) {
VP56mv near[2], best;
- int cnt[4] = { 0 };
- uint8_t p[4];
+ uint8_t cnt[4] = { 0 };
// inter MB, 16.2
- if (vp56_rac_get_prob(c, s->prob->last))
+ if (vp56_rac_get_prob_branchy(c, s->prob->last))
mb->ref_frame = vp56_rac_get_prob(c, s->prob->golden) ?
VP56_FRAME_GOLDEN2 /* altref */ : VP56_FRAME_GOLDEN;
else
mb->ref_frame = VP56_FRAME_PREVIOUS;
+ s->ref_count[mb->ref_frame-1]++;
// motion vectors, 16.3
- find_near_mvs(s, mb, mb_x, mb_y, near, &best, cnt);
- for (n = 0; n < 4; n++)
- p[n] = vp8_mode_contexts[cnt[n]][n];
- mb->mode = vp8_rac_get_tree(c, vp8_pred16x16_tree_mvinter, p);
- switch (mb->mode) {
- case VP8_MVMODE_SPLIT:
- decode_splitmvs(s, c, mb, &best);
- mb->mv = mb->bmv[15];
- break;
- case VP8_MVMODE_ZERO:
- mb->mv.x = 0;
- mb->mv.y = 0;
- break;
- case VP8_MVMODE_NEAREST:
- clamp_mv(s, &mb->mv, &near[0], mb_x, mb_y);
- break;
- case VP8_MVMODE_NEAR:
- clamp_mv(s, &mb->mv, &near[1], mb_x, mb_y);
- break;
- case VP8_MVMODE_NEW:
- mb->mv.y = best.y + read_mv_component(c, s->prob->mvc[0]);
- mb->mv.x = best.x + read_mv_component(c, s->prob->mvc[1]);
- break;
+ 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])) {
+ if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[3]][3])) {
+ mb->mode = VP8_MVMODE_SPLIT;
+ clamp_mv(s, &mb->mv, &mb->mv, mb_x, mb_y);
+ mb->mv = mb->bmv[decode_splitmvs(s, c, mb) - 1];
+ } 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]);
+ }
+ } else {
+ mb->mode = VP8_MVMODE_NEAR;
+ clamp_mv(s, &mb->mv, &near[1], mb_x, mb_y);
+ }
+ } else {
+ mb->mode = VP8_MVMODE_NEAREST;
+ clamp_mv(s, &mb->mv, &near[0], mb_x, mb_y);
+ }
+ } else {
+ mb->mode = VP8_MVMODE_ZERO;
+ AV_ZERO32(&mb->mv);
}
if (mb->mode != VP8_MVMODE_SPLIT) {
- for (n = 0; n < 16; n++)
- mb->bmv[n] = mb->mv;
+ mb->partitioning = VP8_SPLITMVMODE_NONE;
+ mb->bmv[0] = mb->mv;
}
} else {
// intra MB, 16.1
mb->mode = vp8_rac_get_tree(c, vp8_pred16x16_tree_inter, s->prob->pred16x16);
- if (mb->mode == MODE_I4x4) {
- decode_intra4x4_modes(c, intra4x4, s->b4_stride, 0);
- } else
- fill_rectangle(intra4x4, 4, 4, s->b4_stride, vp8_pred4x4_mode[mb->mode], 1);
+ if (mb->mode == MODE_I4x4)
+ decode_intra4x4_modes(s, c, mb_x, 0);
s->chroma_pred_mode = vp8_rac_get_tree(c, vp8_pred8x8c_tree, s->prob->pred8x8c);
mb->ref_frame = VP56_FRAME_CURRENT;
+ mb->partitioning = VP8_SPLITMVMODE_NONE;
+ AV_ZERO32(&mb->bmv[0]);
}
}
/**
+ * @param c arithmetic bitstream reader context
+ * @param block destination for block coefficients
+ * @param probs probabilities to use when reading trees from the bitstream
* @param i initial coeff index, 0 unless a separate DC block is coded
* @param zero_nhood the initial prediction context for number of surrounding
* all-zero blocks (only left/top, so 0-2)
- * @param qmul[0] dc dequant factor
- * @param qmul[1] ac dequant factor
+ * @param qmul array holding the dc/ac dequant factor at position 0/1
* @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])
{
- int token, nonzero = 0;
- int offset = 0;
-
- for (; i < 16; i++) {
- token = vp8_rac_get_tree_with_offset(c, vp8_coeff_tree, probs[vp8_coeff_band[i]][zero_nhood], offset);
+ goto skip_eob;
+ do {
+ int coeff;
+ if (!vp56_rac_get_prob_branchy(c, token_prob[0])) // DCT_EOB
+ return i;
+
+skip_eob:
+ if (!vp56_rac_get_prob_branchy(c, token_prob[1])) { // DCT_0
+ if (++i == 16)
+ return i; // invalid input; blocks should end with EOB
+ token_prob = probs[i][0];
+ goto skip_eob;
+ }
- if (token == DCT_EOB)
- break;
- else if (token >= DCT_CAT1) {
- int cat = token-DCT_CAT1;
- token = vp8_rac_get_coeff(c, vp8_dct_cat_prob[cat]);
- token += vp8_dct_cat_offset[cat];
+ if (!vp56_rac_get_prob_branchy(c, token_prob[2])) { // DCT_1
+ coeff = 1;
+ token_prob = probs[i+1][1];
+ } else {
+ if (!vp56_rac_get_prob_branchy(c, token_prob[3])) { // DCT 2,3,4
+ coeff = vp56_rac_get_prob_branchy(c, token_prob[4]);
+ if (coeff)
+ coeff += vp56_rac_get_prob(c, token_prob[5]);
+ coeff += 2;
+ } else {
+ // DCT_CAT*
+ if (!vp56_rac_get_prob_branchy(c, token_prob[6])) {
+ if (!vp56_rac_get_prob_branchy(c, token_prob[7])) { // DCT_CAT1
+ coeff = 5 + vp56_rac_get_prob(c, vp8_dct_cat1_prob[0]);
+ } else { // DCT_CAT2
+ coeff = 7;
+ coeff += vp56_rac_get_prob(c, vp8_dct_cat2_prob[0]) << 1;
+ coeff += vp56_rac_get_prob(c, vp8_dct_cat2_prob[1]);
+ }
+ } else { // DCT_CAT3 and up
+ int a = vp56_rac_get_prob(c, token_prob[8]);
+ int b = vp56_rac_get_prob(c, token_prob[9+a]);
+ int cat = (a<<1) + b;
+ coeff = 3 + (8<<cat);
+ coeff += vp8_rac_get_coeff(c, vp8_dct_cat_prob[cat]);
+ }
+ }
+ token_prob = probs[i+1][2];
}
+ block[zigzag_scan[i]] = (vp8_rac_get(c) ? -coeff : coeff) * qmul[!!i];
+ } while (++i < 16);
- // after the first token, the non-zero prediction context becomes
- // based on the last decoded coeff
- if (!token) {
- zero_nhood = 0;
- offset = 1;
- continue;
- } else if (token == 1)
- zero_nhood = 1;
- else
- zero_nhood = 2;
+ return i;
+}
- // todo: full [16] qmat? load into register?
- block[zigzag_scan[i]] = (vp8_rac_get(c) ? -token : token) * qmul[!!i];
- nonzero = i+1;
- offset = 0;
- }
- return nonzero;
+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 void decode_mb_coeffs(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb,
- uint8_t t_nnz[9], uint8_t l_nnz[9])
+static av_always_inline
+void decode_mb_coeffs(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb,
+ uint8_t t_nnz[9], uint8_t l_nnz[9])
{
- LOCAL_ALIGNED_16(DCTELEM, dc,[16]);
int i, x, y, luma_start = 0, luma_ctx = 3;
int nnz_pred, nnz, nnz_total = 0;
- int segment = s->segmentation.enabled ? mb->segment : 0;
-
- s->dsp.clear_blocks((DCTELEM *)s->block);
+ int segment = s->segment;
+ int block_dc = 0;
if (mb->mode != MODE_I4x4 && mb->mode != VP8_MVMODE_SPLIT) {
- AV_ZERO128(dc);
- AV_ZERO128(dc+8);
nnz_pred = t_nnz[8] + l_nnz[8];
// decode DC values and do hadamard
- nnz = decode_block_coeffs(c, dc, s->prob->token[1], 0, nnz_pred,
+ nnz = decode_block_coeffs(c, s->block_dc, s->prob->token[1], 0, nnz_pred,
s->qmat[segment].luma_dc_qmul);
l_nnz[8] = t_nnz[8] = !!nnz;
- nnz_total += nnz;
- s->vp8dsp.vp8_luma_dc_wht(s->block, dc);
+ if (nnz) {
+ nnz_total += nnz;
+ block_dc = 1;
+ if (nnz == 1)
+ s->vp8dsp.vp8_luma_dc_wht_dc(s->block, s->block_dc);
+ else
+ s->vp8dsp.vp8_luma_dc_wht(s->block, s->block_dc);
+ }
luma_start = 1;
luma_ctx = 0;
}
nnz_pred = l_nnz[y] + t_nnz[x];
nnz = decode_block_coeffs(c, s->block[y][x], s->prob->token[luma_ctx], luma_start,
nnz_pred, s->qmat[segment].luma_qmul);
- // nnz+luma_start may be one more than the actual last index, but we don't care
- s->non_zero_count_cache[y][x] = nnz + luma_start;
+ // nnz+block_dc may be one more than the actual last index, but we don't care
+ s->non_zero_count_cache[y][x] = nnz + block_dc;
t_nnz[x] = l_nnz[y] = !!nnz;
nnz_total += nnz;
}
mb->skip = 1;
}
-static int check_intra_pred_mode(int mode, int mb_x, int mb_y)
+static av_always_inline
+void backup_mb_border(uint8_t *top_border, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr,
+ int linesize, int uvlinesize, int simple)
+{
+ AV_COPY128(top_border, src_y + 15*linesize);
+ if (!simple) {
+ AV_COPY64(top_border+16, src_cb + 7*uvlinesize);
+ AV_COPY64(top_border+24, src_cr + 7*uvlinesize);
+ }
+}
+
+static av_always_inline
+void xchg_mb_border(uint8_t *top_border, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr,
+ int linesize, int uvlinesize, int mb_x, int mb_y, int mb_width,
+ int simple, int xchg)
+{
+ uint8_t *top_border_m1 = top_border-32; // for TL prediction
+ src_y -= linesize;
+ src_cb -= uvlinesize;
+ src_cr -= uvlinesize;
+
+#define XCHG(a,b,xchg) do { \
+ if (xchg) AV_SWAP64(b,a); \
+ else AV_COPY64(b,a); \
+ } while (0)
+
+ XCHG(top_border_m1+8, src_y-8, xchg);
+ XCHG(top_border, src_y, xchg);
+ XCHG(top_border+8, src_y+8, 1);
+ if (mb_x < mb_width-1)
+ XCHG(top_border+32, src_y+16, 1);
+
+ // only copy chroma for normal loop filter
+ // or to initialize the top row to 127
+ if (!simple || !mb_y) {
+ XCHG(top_border_m1+16, src_cb-8, xchg);
+ XCHG(top_border_m1+24, src_cr-8, xchg);
+ XCHG(top_border+16, src_cb, 1);
+ XCHG(top_border+24, src_cr, 1);
+ }
+}
+
+static av_always_inline
+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|mb_y))
- mode = DC_128_PRED8x8;
- else if (!mb_y)
- mode = LEFT_DC_PRED8x8;
- else if (!mb_x)
- mode = TOP_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 void intra_predict(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb,
- uint8_t *bmode, int mb_x, int mb_y)
+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;
+}
+
+static av_always_inline
+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 (!(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;
}
+ if (mb->skip)
+ AV_ZERO128(s->non_zero_count_cache);
+
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[bmode[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) {
}
ptr += 4*s->linesize;
- bmode += s->b4_stride;
+ intra4x4 += 4;
}
}
- 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 (!(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);
}
/**
* @param width width of src/dst plane data
* @param height height of src/dst plane data
* @param linesize size of a single line of plane data, including padding
+ * @param mc_func motion compensation function pointers (bilinear or sixtap MC)
*/
-static inline void vp8_mc(VP8Context *s, int luma,
- uint8_t *dst, uint8_t *src, const VP56mv *mv,
- int x_off, int y_off, int block_w, int block_h,
- int width, int height, int linesize,
- h264_chroma_mc_func mc_func[3][3])
+static av_always_inline
+void vp8_mc(VP8Context *s, int luma,
+ uint8_t *dst, uint8_t *src, const VP56mv *mv,
+ int x_off, int y_off, int block_w, int block_h,
+ int width, int height, int linesize,
+ vp8_mc_func mc_func[3][3])
{
- static const uint8_t idx[8] = { 0, 1, 2, 1, 2, 1, 2, 1 };
- int mx = (mv->x << luma)&7, mx_idx = idx[mx];
- int my = (mv->y << luma)&7, my_idx = idx[my];
-
- x_off += mv->x >> (3 - luma);
- y_off += mv->y >> (3 - luma);
-
- // edge emulation
- src += y_off * linesize + x_off;
- if (x_off < 2 || x_off >= width - block_w - 3 ||
- y_off < 2 || y_off >= height - block_h - 3) {
- ff_emulated_edge_mc(s->edge_emu_buffer, src - 2 * linesize - 2, linesize,
- block_w + 5, block_h + 5,
- x_off - 2, y_off - 2, width, height);
- src = s->edge_emu_buffer + 2 + linesize * 2;
+ if (AV_RN32A(mv)) {
+ static const uint8_t idx[8] = { 0, 1, 2, 1, 2, 1, 2, 1 };
+ int mx = (mv->x << luma)&7, mx_idx = idx[mx];
+ int my = (mv->y << luma)&7, my_idx = idx[my];
+
+ x_off += mv->x >> (3 - luma);
+ y_off += mv->y >> (3 - luma);
+
+ // edge emulation
+ src += y_off * linesize + x_off;
+ if (x_off < 2 || x_off >= width - block_w - 3 ||
+ y_off < 2 || y_off >= height - block_h - 3) {
+ ff_emulated_edge_mc(s->edge_emu_buffer, src - 2 * linesize - 2, linesize,
+ block_w + 5, block_h + 5,
+ x_off - 2, y_off - 2, width, height);
+ src = s->edge_emu_buffer + 2 + linesize * 2;
+ }
+ mc_func[my_idx][mx_idx](dst, linesize, src, linesize, block_h, mx, my);
+ } else
+ mc_func[0][0](dst, linesize, src + y_off * linesize + x_off, linesize, block_h, 0, 0);
+}
+
+static av_always_inline
+void vp8_mc_part(VP8Context *s, uint8_t *dst[3],
+ AVFrame *ref_frame, int x_off, int y_off,
+ int bx_off, int by_off,
+ int block_w, int block_h,
+ int width, int height, VP56mv *mv)
+{
+ VP56mv uvmv = *mv;
+
+ /* Y */
+ vp8_mc(s, 1, dst[0] + by_off * s->linesize + bx_off,
+ ref_frame->data[0], mv, x_off + bx_off, y_off + by_off,
+ block_w, block_h, width, height, s->linesize,
+ s->put_pixels_tab[block_w == 8]);
+
+ /* U/V */
+ if (s->profile == 3) {
+ uvmv.x &= ~7;
+ uvmv.y &= ~7;
}
+ x_off >>= 1; y_off >>= 1;
+ bx_off >>= 1; by_off >>= 1;
+ width >>= 1; height >>= 1;
+ block_w >>= 1; block_h >>= 1;
+ vp8_mc(s, 0, dst[1] + by_off * s->uvlinesize + bx_off,
+ ref_frame->data[1], &uvmv, x_off + bx_off, y_off + by_off,
+ block_w, block_h, width, height, s->uvlinesize,
+ s->put_pixels_tab[1 + (block_w == 4)]);
+ vp8_mc(s, 0, dst[2] + by_off * s->uvlinesize + bx_off,
+ ref_frame->data[2], &uvmv, x_off + bx_off, y_off + by_off,
+ block_w, block_h, width, height, s->uvlinesize,
+ s->put_pixels_tab[1 + (block_w == 4)]);
+}
- mc_func[my_idx][mx_idx](dst, src, linesize, block_h, mx, my);
+/* Fetch pixels for estimated mv 4 macroblocks ahead.
+ * Optimized for 64-byte cache lines. Inspired by ffh264 prefetch_motion. */
+static av_always_inline void prefetch_motion(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y, int mb_xy, int ref)
+{
+ /* 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>>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);
+ off= (mx>>1) + ((my>>1) + (mb_x&7))*s->uvlinesize + 64;
+ s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
+ }
}
/**
* Apply motion vectors to prediction buffer, chapter 18.
*/
-static void inter_predict(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb,
- int mb_x, int mb_y)
+static av_always_inline
+void inter_predict(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb,
+ int mb_x, int mb_y)
{
int x_off = mb_x << 4, y_off = mb_y << 4;
int width = 16*s->mb_width, height = 16*s->mb_height;
- VP56mv uvmv;
+ AVFrame *ref = s->framep[mb->ref_frame];
+ VP56mv *bmv = mb->bmv;
if (mb->mode < VP8_MVMODE_SPLIT) {
- /* Y */
- vp8_mc(s, 1, dst[0], s->framep[mb->ref_frame]->data[0], &mb->mv,
- x_off, y_off, 16, 16, width, height, s->linesize,
- s->vp8dsp.put_vp8_epel_pixels_tab[0]);
-
- /* U/V */
- uvmv = mb->mv;
- if (s->profile == 3) {
- uvmv.x &= ~7;
- uvmv.y &= ~7;
- }
- x_off >>= 1; y_off >>= 1; width >>= 1; height >>= 1;
- vp8_mc(s, 0, dst[1], s->framep[mb->ref_frame]->data[1], &uvmv,
- x_off, y_off, 8, 8, width, height, s->uvlinesize,
- s->vp8dsp.put_vp8_epel_pixels_tab[1]);
- vp8_mc(s, 0, dst[2], s->framep[mb->ref_frame]->data[2], &uvmv,
- x_off, y_off, 8, 8, width, height, s->uvlinesize,
- s->vp8dsp.put_vp8_epel_pixels_tab[1]);
- } else {
+ vp8_mc_part(s, dst, ref, x_off, y_off,
+ 0, 0, 16, 16, width, height, &mb->mv);
+ } else switch (mb->partitioning) {
+ case VP8_SPLITMVMODE_4x4: {
int x, y;
+ VP56mv uvmv;
/* Y */
for (y = 0; y < 4; y++) {
for (x = 0; x < 4; x++) {
vp8_mc(s, 1, dst[0] + 4*y*s->linesize + x*4,
- s->framep[mb->ref_frame]->data[0], &mb->bmv[4*y + x],
+ ref->data[0], &bmv[4*y + x],
4*x + x_off, 4*y + y_off, 4, 4,
width, height, s->linesize,
- s->vp8dsp.put_vp8_epel_pixels_tab[2]);
+ s->put_pixels_tab[2]);
}
}
mb->bmv[ 2*y * 4 + 2*x+1].y +
mb->bmv[(2*y+1) * 4 + 2*x ].y +
mb->bmv[(2*y+1) * 4 + 2*x+1].y;
- uvmv.x = (uvmv.x + (uvmv.x < 0 ? -2 : 2)) / 4;
- uvmv.y = (uvmv.y + (uvmv.y < 0 ? -2 : 2)) / 4;
+ uvmv.x = (uvmv.x + 2 + (uvmv.x >> (INT_BIT-1))) >> 2;
+ uvmv.y = (uvmv.y + 2 + (uvmv.y >> (INT_BIT-1))) >> 2;
if (s->profile == 3) {
uvmv.x &= ~7;
uvmv.y &= ~7;
}
vp8_mc(s, 0, dst[1] + 4*y*s->uvlinesize + x*4,
- s->framep[mb->ref_frame]->data[1], &uvmv,
+ ref->data[1], &uvmv,
4*x + x_off, 4*y + y_off, 4, 4,
width, height, s->uvlinesize,
- s->vp8dsp.put_vp8_epel_pixels_tab[2]);
+ s->put_pixels_tab[2]);
vp8_mc(s, 0, dst[2] + 4*y*s->uvlinesize + x*4,
- s->framep[mb->ref_frame]->data[2], &uvmv,
+ ref->data[2], &uvmv,
4*x + x_off, 4*y + y_off, 4, 4,
width, height, s->uvlinesize,
- s->vp8dsp.put_vp8_epel_pixels_tab[2]);
+ s->put_pixels_tab[2]);
}
}
+ break;
+ }
+ case VP8_SPLITMVMODE_16x8:
+ vp8_mc_part(s, dst, ref, x_off, y_off,
+ 0, 0, 16, 8, width, height, &bmv[0]);
+ vp8_mc_part(s, dst, ref, x_off, y_off,
+ 0, 8, 16, 8, width, height, &bmv[1]);
+ break;
+ case VP8_SPLITMVMODE_8x16:
+ vp8_mc_part(s, dst, ref, x_off, y_off,
+ 0, 0, 8, 16, width, height, &bmv[0]);
+ vp8_mc_part(s, dst, ref, x_off, y_off,
+ 8, 0, 8, 16, width, height, &bmv[1]);
+ break;
+ case VP8_SPLITMVMODE_8x8:
+ vp8_mc_part(s, dst, ref, x_off, y_off,
+ 0, 0, 8, 8, width, height, &bmv[0]);
+ vp8_mc_part(s, dst, ref, x_off, y_off,
+ 8, 0, 8, 8, width, height, &bmv[1]);
+ vp8_mc_part(s, dst, ref, x_off, y_off,
+ 0, 8, 8, 8, width, height, &bmv[2]);
+ vp8_mc_part(s, dst, ref, x_off, y_off,
+ 8, 8, 8, 8, width, height, &bmv[3]);
+ break;
}
}
-static void idct_mb(VP8Context *s, uint8_t *y_dst, uint8_t *u_dst, uint8_t *v_dst,
- VP8Macroblock *mb)
+static av_always_inline void idct_mb(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb)
{
- int x, y, nnz;
+ int x, y, ch;
- if (mb->mode != MODE_I4x4)
+ if (mb->mode != MODE_I4x4) {
+ uint8_t *y_dst = dst[0];
for (y = 0; y < 4; y++) {
- for (x = 0; x < 4; x++) {
- nnz = s->non_zero_count_cache[y][x];
- if (nnz) {
- if (nnz == 1)
- s->vp8dsp.vp8_idct_dc_add(y_dst+4*x, s->block[y][x], s->linesize);
- else
- s->vp8dsp.vp8_idct_add(y_dst+4*x, s->block[y][x], s->linesize);
+ uint32_t nnz4 = AV_RN32A(s->non_zero_count_cache[y]);
+ if (nnz4) {
+ if (nnz4&~0x01010101) {
+ for (x = 0; x < 4; x++) {
+ int nnz = s->non_zero_count_cache[y][x];
+ if (nnz) {
+ if (nnz == 1)
+ s->vp8dsp.vp8_idct_dc_add(y_dst+4*x, s->block[y][x], s->linesize);
+ else
+ s->vp8dsp.vp8_idct_add(y_dst+4*x, s->block[y][x], s->linesize);
+ }
+ }
+ } else {
+ s->vp8dsp.vp8_idct_dc_add4y(y_dst, s->block[y], s->linesize);
}
}
y_dst += 4*s->linesize;
}
+ }
- for (y = 0; y < 2; y++) {
- for (x = 0; x < 2; x++) {
- nnz = s->non_zero_count_cache[4][(y<<1)+x];
- if (nnz) {
- if (nnz == 1)
- s->vp8dsp.vp8_idct_dc_add(u_dst+4*x, s->block[4][(y<<1)+x], s->uvlinesize);
- else
- s->vp8dsp.vp8_idct_add(u_dst+4*x, s->block[4][(y<<1)+x], s->uvlinesize);
- }
-
- nnz = s->non_zero_count_cache[5][(y<<1)+x];
- if (nnz) {
- if (nnz == 1)
- s->vp8dsp.vp8_idct_dc_add(v_dst+4*x, s->block[5][(y<<1)+x], s->uvlinesize);
- else
- s->vp8dsp.vp8_idct_add(v_dst+4*x, s->block[5][(y<<1)+x], s->uvlinesize);
+ for (ch = 0; ch < 2; ch++) {
+ uint32_t nnz4 = AV_RN32A(s->non_zero_count_cache[4+ch]);
+ if (nnz4) {
+ uint8_t *ch_dst = dst[1+ch];
+ if (nnz4&~0x01010101) {
+ for (y = 0; y < 2; y++) {
+ for (x = 0; x < 2; x++) {
+ int nnz = s->non_zero_count_cache[4+ch][(y<<1)+x];
+ if (nnz) {
+ if (nnz == 1)
+ s->vp8dsp.vp8_idct_dc_add(ch_dst+4*x, s->block[4+ch][(y<<1)+x], s->uvlinesize);
+ else
+ s->vp8dsp.vp8_idct_add(ch_dst+4*x, s->block[4+ch][(y<<1)+x], s->uvlinesize);
+ }
+ }
+ ch_dst += 4*s->uvlinesize;
+ }
+ } else {
+ s->vp8dsp.vp8_idct_dc_add4uv(ch_dst, s->block[4+ch], s->uvlinesize);
}
}
- u_dst += 4*s->uvlinesize;
- v_dst += 4*s->uvlinesize;
}
}
-static void filter_level_for_mb(VP8Context *s, VP8Macroblock *mb, int *level, int *inner, int *hev_thresh)
+static av_always_inline void filter_level_for_mb(VP8Context *s, VP8Macroblock *mb, VP8FilterStrength *f )
{
int interior_limit, filter_level;
if (s->segmentation.enabled) {
- filter_level = s->segmentation.filter_level[mb->segment];
+ filter_level = s->segmentation.filter_level[s->segment];
if (!s->segmentation.absolute_vals)
filter_level += s->filter.level;
} else
}
interior_limit = FFMAX(interior_limit, 1);
- *level = filter_level;
- *inner = interior_limit;
-
- if (hev_thresh) {
- *hev_thresh = filter_level >= 15;
-
- if (s->keyframe) {
- if (filter_level >= 40)
- *hev_thresh = 2;
- } else {
- if (filter_level >= 40)
- *hev_thresh = 3;
- else if (filter_level >= 20)
- *hev_thresh = 2;
- }
- }
+ f->filter_level = filter_level;
+ f->inner_limit = interior_limit;
+ f->inner_filter = !mb->skip || mb->mode == MODE_I4x4 || mb->mode == VP8_MVMODE_SPLIT;
}
-// TODO: look at backup_mb_border / xchg_mb_border in h264.c
-static void filter_mb(VP8Context *s, uint8_t *dst[3], VP8Macroblock *mb, int mb_x, int mb_y)
+static av_always_inline void filter_mb(VP8Context *s, uint8_t *dst[3], VP8FilterStrength *f, int mb_x, int mb_y)
{
- int filter_level, inner_limit, hev_thresh;
+ int mbedge_lim, bedge_lim, hev_thresh;
+ int filter_level = f->filter_level;
+ int inner_limit = f->inner_limit;
+ int inner_filter = f->inner_filter;
+ int linesize = s->linesize;
+ int uvlinesize = s->uvlinesize;
- filter_level_for_mb(s, mb, &filter_level, &inner_limit, &hev_thresh);
if (!filter_level)
return;
+ mbedge_lim = 2*(filter_level+2) + inner_limit;
+ bedge_lim = 2* filter_level + inner_limit;
+ hev_thresh = filter_level >= 15;
+
+ if (s->keyframe) {
+ if (filter_level >= 40)
+ hev_thresh = 2;
+ } else {
+ if (filter_level >= 40)
+ hev_thresh = 3;
+ else if (filter_level >= 20)
+ hev_thresh = 2;
+ }
+
if (mb_x) {
- s->vp8dsp.vp8_h_loop_filter16(dst[0], s->linesize, filter_level+2, inner_limit, hev_thresh);
- s->vp8dsp.vp8_h_loop_filter8 (dst[1], s->uvlinesize, filter_level+2, inner_limit, hev_thresh);
- s->vp8dsp.vp8_h_loop_filter8 (dst[2], s->uvlinesize, filter_level+2, inner_limit, hev_thresh);
+ s->vp8dsp.vp8_h_loop_filter16y(dst[0], linesize,
+ mbedge_lim, inner_limit, hev_thresh);
+ s->vp8dsp.vp8_h_loop_filter8uv(dst[1], dst[2], uvlinesize,
+ mbedge_lim, inner_limit, hev_thresh);
}
- if (!mb->skip || mb->mode == MODE_I4x4 || mb->mode == VP8_MVMODE_SPLIT) {
- s->vp8dsp.vp8_h_loop_filter16_inner(dst[0]+ 4, s->linesize, filter_level, inner_limit, hev_thresh);
- s->vp8dsp.vp8_h_loop_filter16_inner(dst[0]+ 8, s->linesize, filter_level, inner_limit, hev_thresh);
- s->vp8dsp.vp8_h_loop_filter16_inner(dst[0]+12, s->linesize, filter_level, inner_limit, hev_thresh);
- s->vp8dsp.vp8_h_loop_filter8_inner (dst[1]+ 4, s->uvlinesize, filter_level, inner_limit, hev_thresh);
- s->vp8dsp.vp8_h_loop_filter8_inner (dst[2]+ 4, s->uvlinesize, filter_level, inner_limit, hev_thresh);
+ if (inner_filter) {
+ s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0]+ 4, linesize, bedge_lim,
+ inner_limit, hev_thresh);
+ s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0]+ 8, linesize, bedge_lim,
+ inner_limit, hev_thresh);
+ s->vp8dsp.vp8_h_loop_filter16y_inner(dst[0]+12, linesize, bedge_lim,
+ inner_limit, hev_thresh);
+ s->vp8dsp.vp8_h_loop_filter8uv_inner(dst[1] + 4, dst[2] + 4,
+ uvlinesize, bedge_lim,
+ inner_limit, hev_thresh);
}
if (mb_y) {
- s->vp8dsp.vp8_v_loop_filter16(dst[0], s->linesize, filter_level+2, inner_limit, hev_thresh);
- s->vp8dsp.vp8_v_loop_filter8 (dst[1], s->uvlinesize, filter_level+2, inner_limit, hev_thresh);
- s->vp8dsp.vp8_v_loop_filter8 (dst[2], s->uvlinesize, filter_level+2, inner_limit, hev_thresh);
+ s->vp8dsp.vp8_v_loop_filter16y(dst[0], linesize,
+ mbedge_lim, inner_limit, hev_thresh);
+ s->vp8dsp.vp8_v_loop_filter8uv(dst[1], dst[2], uvlinesize,
+ mbedge_lim, inner_limit, hev_thresh);
}
- if (!mb->skip || mb->mode == MODE_I4x4 || mb->mode == VP8_MVMODE_SPLIT) {
- s->vp8dsp.vp8_v_loop_filter16_inner(dst[0]+ 4*s->linesize, s->linesize, filter_level, inner_limit, hev_thresh);
- s->vp8dsp.vp8_v_loop_filter16_inner(dst[0]+ 8*s->linesize, s->linesize, filter_level, inner_limit, hev_thresh);
- s->vp8dsp.vp8_v_loop_filter16_inner(dst[0]+12*s->linesize, s->linesize, filter_level, inner_limit, hev_thresh);
- s->vp8dsp.vp8_v_loop_filter8_inner (dst[1]+ 4*s->uvlinesize, s->uvlinesize, filter_level, inner_limit, hev_thresh);
- s->vp8dsp.vp8_v_loop_filter8_inner (dst[2]+ 4*s->uvlinesize, s->uvlinesize, filter_level, inner_limit, hev_thresh);
+ if (inner_filter) {
+ s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0]+ 4*linesize,
+ linesize, bedge_lim,
+ inner_limit, hev_thresh);
+ s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0]+ 8*linesize,
+ linesize, bedge_lim,
+ inner_limit, hev_thresh);
+ s->vp8dsp.vp8_v_loop_filter16y_inner(dst[0]+12*linesize,
+ linesize, bedge_lim,
+ inner_limit, hev_thresh);
+ s->vp8dsp.vp8_v_loop_filter8uv_inner(dst[1] + 4 * uvlinesize,
+ dst[2] + 4 * uvlinesize,
+ uvlinesize, bedge_lim,
+ inner_limit, hev_thresh);
}
}
-static void filter_mb_simple(VP8Context *s, uint8_t *dst, VP8Macroblock *mb, int mb_x, int mb_y)
+static av_always_inline void filter_mb_simple(VP8Context *s, uint8_t *dst, VP8FilterStrength *f, int mb_x, int mb_y)
{
- int filter_level, inner_limit, mbedge_lim, bedge_lim;
+ int mbedge_lim, bedge_lim;
+ int filter_level = f->filter_level;
+ int inner_limit = f->inner_limit;
+ int inner_filter = f->inner_filter;
+ int linesize = s->linesize;
- filter_level_for_mb(s, mb, &filter_level, &inner_limit, NULL);
if (!filter_level)
return;
bedge_lim = 2* filter_level + inner_limit;
if (mb_x)
- s->vp8dsp.vp8_h_loop_filter_simple(dst, s->linesize, mbedge_lim);
- if (!mb->skip || mb->mode == MODE_I4x4 || mb->mode == VP8_MVMODE_SPLIT) {
- s->vp8dsp.vp8_h_loop_filter_simple(dst+ 4, s->linesize, bedge_lim);
- s->vp8dsp.vp8_h_loop_filter_simple(dst+ 8, s->linesize, bedge_lim);
- s->vp8dsp.vp8_h_loop_filter_simple(dst+12, s->linesize, bedge_lim);
+ s->vp8dsp.vp8_h_loop_filter_simple(dst, linesize, mbedge_lim);
+ if (inner_filter) {
+ s->vp8dsp.vp8_h_loop_filter_simple(dst+ 4, linesize, bedge_lim);
+ s->vp8dsp.vp8_h_loop_filter_simple(dst+ 8, linesize, bedge_lim);
+ s->vp8dsp.vp8_h_loop_filter_simple(dst+12, linesize, bedge_lim);
}
if (mb_y)
- s->vp8dsp.vp8_v_loop_filter_simple(dst, s->linesize, mbedge_lim);
- if (!mb->skip || mb->mode == MODE_I4x4 || mb->mode == VP8_MVMODE_SPLIT) {
- s->vp8dsp.vp8_v_loop_filter_simple(dst+ 4*s->linesize, s->linesize, bedge_lim);
- s->vp8dsp.vp8_v_loop_filter_simple(dst+ 8*s->linesize, s->linesize, bedge_lim);
- s->vp8dsp.vp8_v_loop_filter_simple(dst+12*s->linesize, s->linesize, bedge_lim);
+ s->vp8dsp.vp8_v_loop_filter_simple(dst, linesize, mbedge_lim);
+ if (inner_filter) {
+ s->vp8dsp.vp8_v_loop_filter_simple(dst+ 4*linesize, linesize, bedge_lim);
+ s->vp8dsp.vp8_v_loop_filter_simple(dst+ 8*linesize, linesize, bedge_lim);
+ s->vp8dsp.vp8_v_loop_filter_simple(dst+12*linesize, linesize, bedge_lim);
}
}
static void filter_mb_row(VP8Context *s, int mb_y)
{
- VP8Macroblock *mb = s->macroblocks + mb_y*s->mb_stride;
+ VP8FilterStrength *f = s->filter_strength;
uint8_t *dst[3] = {
s->framep[VP56_FRAME_CURRENT]->data[0] + 16*mb_y*s->linesize,
s->framep[VP56_FRAME_CURRENT]->data[1] + 8*mb_y*s->uvlinesize,
int mb_x;
for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
- filter_mb(s, dst, mb++, mb_x, mb_y);
+ backup_mb_border(s->top_border[mb_x+1], dst[0], dst[1], dst[2], s->linesize, s->uvlinesize, 0);
+ filter_mb(s, dst, f++, mb_x, mb_y);
dst[0] += 16;
dst[1] += 8;
dst[2] += 8;
static void filter_mb_row_simple(VP8Context *s, int mb_y)
{
+ VP8FilterStrength *f = s->filter_strength;
uint8_t *dst = s->framep[VP56_FRAME_CURRENT]->data[0] + 16*mb_y*s->linesize;
- VP8Macroblock *mb = s->macroblocks + mb_y*s->mb_stride;
int mb_x;
for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
- filter_mb_simple(s, dst, mb++, mb_x, mb_y);
+ backup_mb_border(s->top_border[mb_x+1], dst, NULL, NULL, s->linesize, 0, 1);
+ filter_mb_simple(s, dst, f++, mb_x, mb_y);
dst += 16;
}
}
VP8Context *s = avctx->priv_data;
int ret, mb_x, mb_y, i, y, referenced;
enum AVDiscard skip_thresh;
- AVFrame *curframe;
+ AVFrame *av_uninit(curframe);
if ((ret = decode_frame_header(s, avpkt->data, avpkt->size)) < 0)
return ret;
s->invisible = 1;
goto skip_decode;
}
+ s->deblock_filter = s->filter.level && avctx->skip_loop_filter < skip_thresh;
for (i = 0; i < 4; i++)
if (&s->frames[i] != s->framep[VP56_FRAME_PREVIOUS] &&
memset(s->top_nnz, 0, s->mb_width*sizeof(*s->top_nnz));
+ /* 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
if (!(avctx->flags & CODEC_FLAG_EMU_EDGE)) {
- memset(curframe->data[0] - s->linesize -1, 127, s->linesize +1);
- memset(curframe->data[1] - s->uvlinesize-1, 127, s->uvlinesize+1);
- memset(curframe->data[2] - s->uvlinesize-1, 127, s->uvlinesize+1);
+ 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->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 + mb_y*s->mb_stride;
- uint8_t *intra4x4 = s->intra4x4_pred_mode + 4*mb_y*s->b4_stride;
+ VP8Macroblock *mb = s->macroblocks + (s->mb_height - mb_y - 1)*2;
+ 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 == 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++) {
+ /* 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);
- for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
- decode_mb_mode(s, mb, mb_x, mb_y, intra4x4 + 4*mb_x);
+ 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);
if (!mb->skip)
decode_mb_coeffs(s, c, mb, s->top_nnz[mb_x], s->left_nnz);
- else {
- AV_ZERO128(s->non_zero_count_cache); // luma
- AV_ZERO64(s->non_zero_count_cache[4]); // chroma
- }
- if (mb->mode <= MODE_I4x4) {
- intra_predict(s, dst, mb, intra4x4 + 4*mb_x, mb_x, mb_y);
- memset(mb->bmv, 0, sizeof(mb->bmv));
- } else {
+ if (mb->mode <= MODE_I4x4)
+ intra_predict(s, dst, mb, mb_x, mb_y);
+ else
inter_predict(s, dst, mb, mb_x, mb_y);
- }
+
+ prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_GOLDEN);
if (!mb->skip) {
- idct_mb(s, dst[0], dst[1], dst[2], mb);
+ idct_mb(s, dst, mb);
} else {
AV_ZERO64(s->left_nnz);
AV_WN64(s->top_nnz[mb_x], 0); // array of 9, so unaligned
}
}
+ if (s->deblock_filter)
+ filter_level_for_mb(s, mb, &s->filter_strength[mb_x]);
+
+ prefetch_motion(s, mb, mb_x, mb_y, mb_xy, VP56_FRAME_GOLDEN2);
+
dst[0] += 16;
dst[1] += 8;
dst[2] += 8;
- mb++;
}
- if (mb_y && s->filter.level && avctx->skip_loop_filter < skip_thresh) {
+ if (s->deblock_filter) {
if (s->filter.simple)
- filter_mb_row_simple(s, mb_y-1);
+ filter_mb_row_simple(s, mb_y);
else
- filter_mb_row(s, mb_y-1);
+ filter_mb_row(s, mb_y);
}
}
- if (s->filter.level && avctx->skip_loop_filter < skip_thresh) {
- if (s->filter.simple)
- filter_mb_row_simple(s, mb_y-1);
- else
- filter_mb_row(s, mb_y-1);
- }
skip_decode:
// if future frames don't use the updated probabilities,
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 supproted\n");
- return AVERROR_PATCHWELCOME;
- }
-
return 0;
}