Merge "Change to prevent invalid memory access"
diff --git a/vp9/common/vp9_blockd.h b/vp9/common/vp9_blockd.h
index c8d677f..f7d5de1 100644
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -137,7 +137,7 @@
TX_SIZE tx_size;
int_mv mv[2]; // for each reference frame used
int_mv ref_mvs[MAX_REF_FRAMES][MAX_MV_REF_CANDIDATES];
- int_mv best_mv, best_second_mv;
+ int_mv best_mv[2];
uint8_t mode_context[MAX_REF_FRAMES];
diff --git a/vp9/encoder/vp9_bitstream.c b/vp9/encoder/vp9_bitstream.c
index 622f75f..20dd8e1 100644
--- a/vp9/encoder/vp9_bitstream.c
+++ b/vp9/encoder/vp9_bitstream.c
@@ -484,17 +484,13 @@
}
if (bsize < BLOCK_8X8) {
- int j;
- MB_PREDICTION_MODE blockmode;
- int_mv blockmv;
const int num_4x4_blocks_wide = num_4x4_blocks_wide_lookup[bsize];
const int num_4x4_blocks_high = num_4x4_blocks_high_lookup[bsize];
int idx, idy;
for (idy = 0; idy < 2; idy += num_4x4_blocks_high) {
for (idx = 0; idx < 2; idx += num_4x4_blocks_wide) {
- j = idy * 2 + idx;
- blockmode = x->partition_info->bmi[j].mode;
- blockmv = m->bmi[j].as_mv[0];
+ const int j = idy * 2 + idx;
+ const MB_PREDICTION_MODE blockmode = x->partition_info->bmi[j].mode;
write_sb_mv_ref(bc, blockmode, mv_ref_p);
++cm->counts.inter_mode[mi->mode_context[rf]]
[inter_mode_offset(blockmode)];
@@ -503,14 +499,12 @@
#ifdef ENTROPY_STATS
active_section = 11;
#endif
- vp9_encode_mv(cpi, bc, &blockmv.as_mv, &mi->best_mv.as_mv,
- nmvc, allow_hp);
+ vp9_encode_mv(cpi, bc, &m->bmi[j].as_mv[0].as_mv,
+ &mi->best_mv[0].as_mv, nmvc, allow_hp);
- if (mi->ref_frame[1] > INTRA_FRAME)
- vp9_encode_mv(cpi, bc,
- &m->bmi[j].as_mv[1].as_mv,
- &mi->best_second_mv.as_mv,
- nmvc, allow_hp);
+ if (has_second_ref(mi))
+ vp9_encode_mv(cpi, bc, &m->bmi[j].as_mv[1].as_mv,
+ &mi->best_mv[1].as_mv, nmvc, allow_hp);
}
}
}
@@ -518,12 +512,12 @@
#ifdef ENTROPY_STATS
active_section = 5;
#endif
- vp9_encode_mv(cpi, bc, &mi->mv[0].as_mv, &mi->best_mv.as_mv,
- nmvc, allow_hp);
+ vp9_encode_mv(cpi, bc, &mi->mv[0].as_mv,
+ &mi->best_mv[0].as_mv, nmvc, allow_hp);
- if (mi->ref_frame[1] > INTRA_FRAME)
- vp9_encode_mv(cpi, bc, &mi->mv[1].as_mv, &mi->best_second_mv.as_mv,
- nmvc, allow_hp);
+ if (has_second_ref(mi))
+ vp9_encode_mv(cpi, bc, &mi->mv[1].as_mv,
+ &mi->best_mv[1].as_mv, nmvc, allow_hp);
}
}
}
diff --git a/vp9/encoder/vp9_block.h b/vp9/encoder/vp9_block.h
index 013047e..5a0d746 100644
--- a/vp9/encoder/vp9_block.h
+++ b/vp9/encoder/vp9_block.h
@@ -34,6 +34,7 @@
typedef struct {
MODE_INFO mic;
PARTITION_INFO partition_info;
+ unsigned char zcoeff_blk[256];
int skip;
int_mv best_ref_mv;
int_mv second_best_ref_mv;
@@ -136,6 +137,7 @@
int mv_row_min;
int mv_row_max;
+ unsigned char zcoeff_blk[TX_SIZES][256];
int skip;
int encode_breakout;
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 0db5062..90a584e 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -390,6 +390,9 @@
}
x->skip = ctx->skip;
+ vpx_memcpy(x->zcoeff_blk[mbmi->tx_size], ctx->zcoeff_blk,
+ sizeof(ctx->zcoeff_blk));
+
if (!output_enabled)
return;
@@ -438,8 +441,8 @@
if (rf2 > 0)
best_second_mv.as_int = mbmi->ref_mvs[rf2][0].as_int;
}
- mbmi->best_mv.as_int = best_mv.as_int;
- mbmi->best_second_mv.as_int = best_second_mv.as_int;
+ mbmi->best_mv[0].as_int = best_mv.as_int;
+ mbmi->best_mv[1].as_int = best_second_mv.as_int;
vp9_update_nmv_count(cpi, x, &best_mv, &best_second_mv);
}
@@ -2745,7 +2748,6 @@
setup_pre_planes(xd, 1, second_ref_fb, mi_row, mi_col,
&xd->scale_factor[1]);
-
vp9_build_inter_predictors_sb(xd, mi_row, mi_col, MAX(bsize, BLOCK_8X8));
}
diff --git a/vp9/encoder/vp9_encodemb.c b/vp9/encoder/vp9_encodemb.c
index 2c12477..30d01d7 100644
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -482,6 +482,14 @@
int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
uint8_t *const dst = raster_block_offset_uint8(plane_bsize, raster_block,
pd->dst.buf, pd->dst.stride);
+
+ // TODO(jingning): per transformed block zero forcing only enabled for
+ // luma component. will integrate chroma components as well.
+ if (x->zcoeff_blk[tx_size][block] && plane == 0) {
+ pd->eobs[block] = 0;
+ return;
+ }
+
vp9_xform_quant(plane, block, plane_bsize, tx_size, arg);
if (x->optimize)
diff --git a/vp9/encoder/vp9_onyx_if.c b/vp9/encoder/vp9_onyx_if.c
index 1d0733e..d815dc9 100644
--- a/vp9/encoder/vp9_onyx_if.c
+++ b/vp9/encoder/vp9_onyx_if.c
@@ -215,7 +215,7 @@
afq_high_motion_minq[i] = calculate_minq_index(maxq,
0.0000021,
-0.00125,
- 0.57,
+ 0.55,
0.0);
}
}
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 3ef3eee..041316b 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -554,9 +554,13 @@
TX_SIZE tx_size;
int bw;
int bh;
- int rate;
- int64_t dist;
- int64_t sse;
+ int rate[256];
+ int64_t dist[256];
+ int64_t sse[256];
+ int this_rate;
+ int64_t this_dist;
+ int64_t this_sse;
+ int64_t this_rd;
int64_t best_rd;
int skip;
const int16_t *scan, *nb;
@@ -573,17 +577,17 @@
int shift = args->tx_size == TX_32X32 ? 0 : 2;
int16_t *const coeff = BLOCK_OFFSET(p->coeff, block);
int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
- args->dist += vp9_block_error(coeff, dqcoeff, 16 << ss_txfrm_size,
+ args->dist[block] = vp9_block_error(coeff, dqcoeff, 16 << ss_txfrm_size,
&this_sse) >> shift;
- args->sse += this_sse >> shift;
+ args->sse[block] = this_sse >> shift;
if (x->skip_encode &&
xd->this_mi->mbmi.ref_frame[0] == INTRA_FRAME) {
// TODO(jingning): tune the model to better capture the distortion.
int64_t p = (pd->dequant[1] * pd->dequant[1] *
(1 << ss_txfrm_size)) >> shift;
- args->dist += p;
- args->sse += p;
+ args->dist[block] = p;
+ args->sse[block] = p;
}
}
@@ -594,10 +598,10 @@
int x_idx, y_idx;
txfrm_block_to_raster_xy(plane_bsize, args->tx_size, block, &x_idx, &y_idx);
- args->rate += cost_coeffs(args->x, plane, block,
- args->t_above + x_idx,
- args->t_left + y_idx, args->tx_size,
- args->scan, args->nb);
+ args->rate[block] = cost_coeffs(args->x, plane, block,
+ args->t_above + x_idx,
+ args->t_left + y_idx, args->tx_size,
+ args->scan, args->nb);
}
static void block_yrd_txfm(int plane, int block, BLOCK_SIZE plane_bsize,
@@ -610,16 +614,6 @@
if (args->skip)
return;
- rd1 = RDCOST(x->rdmult, x->rddiv, args->rate, args->dist);
- rd2 = RDCOST(x->rdmult, x->rddiv, 0, args->sse);
- rd = MIN(rd1, rd2);
- if (rd > args->best_rd) {
- args->skip = 1;
- args->rate = INT_MAX;
- args->dist = INT64_MAX;
- args->sse = INT64_MAX;
- return;
- }
if (!is_inter_block(&xd->this_mi->mbmi))
vp9_encode_block_intra(plane, block, plane_bsize, tx_size, &encode_args);
@@ -628,6 +622,23 @@
dist_block(plane, block, tx_size, args);
rate_block(plane, block, plane_bsize, tx_size, args);
+ rd1 = RDCOST(x->rdmult, x->rddiv, args->rate[block], args->dist[block]);
+ rd2 = RDCOST(x->rdmult, x->rddiv, 0, args->sse[block]);
+
+ // TODO(jingning): temporarily enabled only for luma component
+ rd = MIN(rd1, rd2);
+ if (plane == 0)
+ x->zcoeff_blk[tx_size][block] = rd1 > rd2;
+
+ args->this_rate += args->rate[block];
+ args->this_dist += args->dist[block];
+ args->this_sse += args->sse[block];
+ args->this_rd += rd;
+
+ if (args->this_rd > args->best_rd) {
+ args->skip = 1;
+ return;
+ }
}
static void txfm_rd_in_plane(MACROBLOCK *x,
@@ -643,7 +654,8 @@
int i;
struct rdcost_block_args args = { x, { 0 }, { 0 }, tx_size,
num_4x4_blocks_wide, num_4x4_blocks_high,
- 0, 0, 0, ref_best_rd, 0 };
+ { 0 }, { 0 }, { 0 },
+ 0, 0, 0, 0, ref_best_rd, 0 };
if (plane == 0)
xd->this_mi->mbmi.tx_size = tx_size;
@@ -685,10 +697,17 @@
}
foreach_transformed_block_in_plane(xd, bsize, plane, block_yrd_txfm, &args);
- *distortion = args.dist;
- *rate = args.rate;
- *sse = args.sse;
- *skippable = vp9_is_skippable_in_plane(xd, bsize, plane) && (!args.skip);
+ if (args.skip) {
+ *rate = INT_MAX;
+ *distortion = INT64_MAX;
+ *sse = INT64_MAX;
+ *skippable = 0;
+ } else {
+ *distortion = args.this_dist;
+ *rate = args.this_rate;
+ *sse = args.this_sse;
+ *skippable = vp9_is_skippable_in_plane(xd, bsize, plane);
+ }
}
static void choose_largest_txfm_size(VP9_COMP *cpi, MACROBLOCK *x,
@@ -2220,6 +2239,9 @@
ctx->comp_pred_diff = (int)comp_pred_diff[COMP_PREDICTION_ONLY];
ctx->hybrid_pred_diff = (int)comp_pred_diff[HYBRID_PREDICTION];
+ vpx_memcpy(ctx->zcoeff_blk, x->zcoeff_blk[xd->this_mi->mbmi.tx_size],
+ sizeof(ctx->zcoeff_blk));
+
// FIXME(rbultje) does this memcpy the whole array? I believe sizeof()
// doesn't actually work this way
memcpy(ctx->tx_rd_diff, tx_size_diff, sizeof(ctx->tx_rd_diff));
@@ -3139,8 +3161,11 @@
const int bws = num_8x8_blocks_wide_lookup[bsize] / 2;
const int bhs = num_8x8_blocks_high_lookup[bsize] / 2;
int best_skip2 = 0;
+ unsigned char best_zcoeff_blk[256] = { 0 };
x->skip_encode = cpi->sf.skip_encode_frame && xd->q_index < QIDX_SKIP_THRESH;
+ vpx_memset(x->zcoeff_blk, 0, sizeof(x->zcoeff_blk));
+ vpx_memset(ctx->zcoeff_blk, 0, sizeof(ctx->zcoeff_blk));
for (i = 0; i < 4; i++) {
int j;
@@ -3812,6 +3837,8 @@
best_mbmode = *mbmi;
best_skip2 = this_skip2;
best_partition = *x->partition_info;
+ vpx_memcpy(best_zcoeff_blk, x->zcoeff_blk[mbmi->tx_size],
+ sizeof(best_zcoeff_blk));
if (this_mode == RD_I4X4_PRED || this_mode == RD_SPLITMV)
for (i = 0; i < 4; i++)
@@ -4007,6 +4034,9 @@
mbmi->mv[1].as_int = xd->this_mi->bmi[3].as_mv[1].as_int;
}
+ vpx_memcpy(x->zcoeff_blk[mbmi->tx_size], best_zcoeff_blk,
+ sizeof(best_zcoeff_blk));
+
for (i = 0; i < NB_PREDICTION_TYPES; ++i) {
if (best_pred_rd[i] == INT64_MAX)
best_pred_diff[i] = INT_MIN;
diff --git a/vp9/encoder/x86/vp9_dct_sse2.c b/vp9/encoder/x86/vp9_dct_sse2.c
index eb271fe..ad3d01d 100644
--- a/vp9/encoder/x86/vp9_dct_sse2.c
+++ b/vp9/encoder/x86/vp9_dct_sse2.c
@@ -171,22 +171,21 @@
void fdct4_1d_sse2(__m128i *in) {
const __m128i k__cospi_p16_p16 = _mm_set1_epi16(cospi_16_64);
const __m128i k__cospi_p16_m16 = pair_set_epi16(cospi_16_64, -cospi_16_64);
- const __m128i k__cospi_p24_p08 = pair_set_epi16(cospi_24_64, cospi_8_64);
- const __m128i k__cospi_m08_p24 = pair_set_epi16(-cospi_8_64, cospi_24_64);
+ const __m128i k__cospi_p08_p24 = pair_set_epi16(cospi_8_64, cospi_24_64);
+ const __m128i k__cospi_p24_m08 = pair_set_epi16(cospi_24_64, -cospi_8_64);
const __m128i k__DCT_CONST_ROUNDING = _mm_set1_epi32(DCT_CONST_ROUNDING);
__m128i u[4], v[4];
- u[0] = _mm_add_epi16(in[0], in[3]);
- u[1] = _mm_add_epi16(in[1], in[2]);
- u[2] = _mm_sub_epi16(in[1], in[2]);
- u[3] = _mm_sub_epi16(in[0], in[3]);
+ u[0]=_mm_unpacklo_epi16(in[0], in[1]);
+ u[1]=_mm_unpacklo_epi16(in[3], in[2]);
- v[0] = _mm_unpacklo_epi16(u[0], u[1]);
- v[1] = _mm_unpacklo_epi16(u[2], u[3]);
+ v[0] = _mm_add_epi16(u[0], u[1]);
+ v[1] = _mm_sub_epi16(u[0], u[1]);
+
u[0] = _mm_madd_epi16(v[0], k__cospi_p16_p16); // 0
u[1] = _mm_madd_epi16(v[0], k__cospi_p16_m16); // 2
- u[2] = _mm_madd_epi16(v[1], k__cospi_p24_p08); // 1
- u[3] = _mm_madd_epi16(v[1], k__cospi_m08_p24); // 3
+ u[2] = _mm_madd_epi16(v[1], k__cospi_p08_p24); // 1
+ u[3] = _mm_madd_epi16(v[1], k__cospi_p24_m08); // 3
v[0] = _mm_add_epi32(u[0], k__DCT_CONST_ROUNDING);
v[1] = _mm_add_epi32(u[1], k__DCT_CONST_ROUNDING);
diff --git a/vpxenc.c b/vpxenc.c
index c618ef8..71cf01f 100644
--- a/vpxenc.c
+++ b/vpxenc.c
@@ -725,14 +725,12 @@
{
unsigned int pixelWidth = cfg->g_w;
unsigned int pixelHeight = cfg->g_h;
- float frameRate = (float)fps->num / (float)fps->den;
EbmlLoc videoStart;
Ebml_StartSubElement(glob, &videoStart, Video);
Ebml_SerializeUnsigned(glob, PixelWidth, pixelWidth);
Ebml_SerializeUnsigned(glob, PixelHeight, pixelHeight);
Ebml_SerializeUnsigned(glob, StereoMode, stereo_fmt);
- Ebml_SerializeFloat(glob, FrameRate, frameRate);
Ebml_EndSubElement(glob, &videoStart);
}
Ebml_EndSubElement(glob, &start); /* Track Entry */
