Merge "Improve idct16x16: _256_add_sse2(x1.107)&_10_add_sse2(x1.012)"
diff --git a/test/test-data.sha1 b/test/test-data.sha1
index bb09b75..57c7eb3 100644
--- a/test/test-data.sha1
+++ b/test/test-data.sha1
@@ -61,6 +61,7 @@
aff51d865c2621b60510459244ea83e958e4baed vp80-05-sharpness-1439.ivf
da386e72b19b5485a6af199c5eb60ef25e510dd1 vp80-05-sharpness-1440.ivf
6759a095203d96ccd267ce09b1b050b8cc4c2f1f vp80-05-sharpness-1443.ivf
+b95d3cc1d0df991e63e150a801710a72f20d9ba0 vp80-06-smallsize.ivf
db55ec7fd02c864ba996ff060b25b1e08611330b vp80-00-comprehensive-001.ivf.md5
29db0ad011cba1e45f856d5623cd38dac3e3bf19 vp80-00-comprehensive-002.ivf.md5
e84f258f69e173e7d68f8f8c037a0a3766902182 vp80-00-comprehensive-003.ivf.md5
@@ -122,6 +123,7 @@
086c56378df81b6cee264d7540a7b8f2b405c7a4 vp80-05-sharpness-1439.ivf.md5
d32dc2c4165eb266ea4c23c14a45459b363def32 vp80-05-sharpness-1440.ivf.md5
8c69dc3d8e563f56ffab5ad1e400d9e689dd23df vp80-05-sharpness-1443.ivf.md5
+d6f246df012c241b5fa6c1345019a3703d85c419 vp80-06-smallsize.ivf.md5
ce881e567fe1d0fbcb2d3e9e6281a1a8d74d82e0 vp90-2-00-quantizer-00.webm
ac5eda33407d0521c7afca43a63fd305c0cd9d13 vp90-2-00-quantizer-00.webm.md5
2ca0463f2cfb93d25d7dded174db70b7cb87cb48 vp90-2-00-quantizer-01.webm
diff --git a/test/test.mk b/test/test.mk
index 4f877f4..8fd2d2f 100644
--- a/test/test.mk
+++ b/test/test.mk
@@ -175,6 +175,7 @@
LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-05-sharpness-1439.ivf
LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-05-sharpness-1440.ivf
LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-05-sharpness-1443.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-06-smallsize.ivf
LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-00-comprehensive-001.ivf.md5
LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-00-comprehensive-002.ivf.md5
LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-00-comprehensive-003.ivf.md5
@@ -236,6 +237,7 @@
LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-05-sharpness-1439.ivf.md5
LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-05-sharpness-1440.ivf.md5
LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-05-sharpness-1443.ivf.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP8_DECODER) += vp80-06-smallsize.ivf.md5
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-00-quantizer-00.webm
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-00-quantizer-00.webm.md5
LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-00-quantizer-01.webm
diff --git a/test/test_vector_test.cc b/test/test_vector_test.cc
index 4332420..b85c719 100644
--- a/test/test_vector_test.cc
+++ b/test/test_vector_test.cc
@@ -55,7 +55,8 @@
"vp80-05-sharpness-1430.ivf", "vp80-05-sharpness-1431.ivf",
"vp80-05-sharpness-1433.ivf", "vp80-05-sharpness-1434.ivf",
"vp80-05-sharpness-1438.ivf", "vp80-05-sharpness-1439.ivf",
- "vp80-05-sharpness-1440.ivf", "vp80-05-sharpness-1443.ivf"
+ "vp80-05-sharpness-1440.ivf", "vp80-05-sharpness-1443.ivf",
+ "vp80-06-smallsize.ivf"
};
#endif
#if CONFIG_VP9_DECODER
diff --git a/vp8/decoder/decodframe.c b/vp8/decoder/decodeframe.c
similarity index 99%
rename from vp8/decoder/decodframe.c
rename to vp8/decoder/decodeframe.c
index 16da78a..bfde599 100644
--- a/vp8/decoder/decodframe.c
+++ b/vp8/decoder/decodeframe.c
@@ -680,7 +680,6 @@
vp8_loop_filter_row_simple(pc, lf_mic, mb_row-1,
recon_y_stride, recon_uv_stride,
lf_dst[0], lf_dst[1], lf_dst[2]);
-
if(mb_row > 1)
{
yv12_extend_frame_left_right_c(yv12_fb_new,
@@ -691,10 +690,6 @@
eb_dst[0] += recon_y_stride * 16;
eb_dst[1] += recon_uv_stride * 8;
eb_dst[2] += recon_uv_stride * 8;
-
- if(mb_row == 2)
- yv12_extend_frame_top_c(yv12_fb_new);
-
}
lf_dst[0] += recon_y_stride * 16;
@@ -713,13 +708,9 @@
eb_dst[0],
eb_dst[1],
eb_dst[2]);
-
eb_dst[0] += recon_y_stride * 16;
eb_dst[1] += recon_uv_stride * 8;
eb_dst[2] += recon_uv_stride * 8;
-
- if(mb_row == 1)
- yv12_extend_frame_top_c(yv12_fb_new);
}
}
}
@@ -747,7 +738,7 @@
eb_dst[0],
eb_dst[1],
eb_dst[2]);
-
+ yv12_extend_frame_top_c(yv12_fb_new);
yv12_extend_frame_bottom_c(yv12_fb_new);
}
diff --git a/vp8/vp8dx.mk b/vp8/vp8dx.mk
index 4a8f467..892ed70 100644
--- a/vp8/vp8dx.mk
+++ b/vp8/vp8dx.mk
@@ -22,7 +22,7 @@
VP8_DX_SRCS-yes += decoder/dboolhuff.c
VP8_DX_SRCS-yes += decoder/decodemv.c
-VP8_DX_SRCS-yes += decoder/decodframe.c
+VP8_DX_SRCS-yes += decoder/decodeframe.c
VP8_DX_SRCS-yes += decoder/detokenize.c
VP8_DX_SRCS-$(CONFIG_ERROR_CONCEALMENT) += decoder/ec_types.h
VP8_DX_SRCS-$(CONFIG_ERROR_CONCEALMENT) += decoder/error_concealment.h
diff --git a/vp9/common/vp9_alloccommon.c b/vp9/common/vp9_alloccommon.c
index 28671c3..f495c29 100644
--- a/vp9/common/vp9_alloccommon.c
+++ b/vp9/common/vp9_alloccommon.c
@@ -200,9 +200,6 @@
void vp9_create_common(VP9_COMMON *cm) {
vp9_machine_specific_config(cm);
-
- cm->tx_mode = ONLY_4X4;
- cm->comp_pred_mode = REFERENCE_MODE_SELECT;
}
void vp9_remove_common(VP9_COMMON *cm) {
diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index bcd51f5..9b6740e 100644
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -41,7 +41,6 @@
VP9_COMMON *cm;
vp9_reader bit_reader;
DECLARE_ALIGNED(16, MACROBLOCKD, xd);
- DECLARE_ALIGNED(16, unsigned char, token_cache[1024]);
DECLARE_ALIGNED(16, int16_t, dqcoeff[MAX_MB_PLANE][64 * 64]);
DECLARE_ALIGNED(16, uint16_t, eobs[MAX_MB_PLANE][256]);
} TileWorkerData;
@@ -50,7 +49,7 @@
return (p[0] << 24) | (p[1] << 16) | (p[2] << 8) | p[3];
}
-static int is_compound_prediction_allowed(const VP9_COMMON *cm) {
+static int is_compound_reference_allowed(const VP9_COMMON *cm) {
int i;
for (i = 1; i < ALLOWED_REFS_PER_FRAME; ++i)
if (cm->ref_frame_sign_bias[i + 1] != cm->ref_frame_sign_bias[1])
@@ -59,7 +58,7 @@
return 0;
}
-static void setup_compound_prediction(VP9_COMMON *cm) {
+static void setup_compound_reference(VP9_COMMON *cm) {
if (cm->ref_frame_sign_bias[LAST_FRAME] ==
cm->ref_frame_sign_bias[GOLDEN_FRAME]) {
cm->comp_fixed_ref = ALTREF_FRAME;
@@ -94,7 +93,7 @@
return tx_mode;
}
-static void read_tx_probs(struct tx_probs *tx_probs, vp9_reader *r) {
+static void read_tx_mode_probs(struct tx_probs *tx_probs, vp9_reader *r) {
int i, j;
for (i = 0; i < TX_SIZE_CONTEXTS; ++i)
@@ -124,22 +123,20 @@
vp9_diff_update_prob(r, &fc->inter_mode_probs[i][j]);
}
-static INLINE REFERENCE_MODE read_comp_pred_mode(vp9_reader *r) {
- REFERENCE_MODE mode = vp9_read_bit(r);
- if (mode)
- mode += vp9_read_bit(r);
- return mode;
+static REFERENCE_MODE read_reference_mode(VP9_COMMON *cm, vp9_reader *r) {
+ if (is_compound_reference_allowed(cm)) {
+ REFERENCE_MODE mode = vp9_read_bit(r);
+ if (mode)
+ mode += vp9_read_bit(r);
+ setup_compound_reference(cm);
+ return mode;
+ } else {
+ return SINGLE_REFERENCE;
+ }
}
-static void read_comp_pred(VP9_COMMON *cm, vp9_reader *r) {
+static void read_reference_mode_probs(VP9_COMMON *cm, vp9_reader *r) {
int i;
-
- const int compound_allowed = is_compound_prediction_allowed(cm);
- cm->comp_pred_mode = compound_allowed ? read_comp_pred_mode(r)
- : SINGLE_REFERENCE;
- if (compound_allowed)
- setup_compound_prediction(cm);
-
if (cm->comp_pred_mode == REFERENCE_MODE_SELECT)
for (i = 0; i < COMP_INTER_CONTEXTS; i++)
vp9_diff_update_prob(r, &cm->fc.comp_inter_prob[i]);
@@ -241,8 +238,7 @@
}
static void inverse_transform_block(MACROBLOCKD* xd, int plane, int block,
- TX_SIZE tx_size, uint8_t *dst, int stride,
- uint8_t *token_cache) {
+ TX_SIZE tx_size, uint8_t *dst, int stride) {
struct macroblockd_plane *const pd = &xd->plane[plane];
const int eob = pd->eobs[block];
if (eob > 0) {
@@ -275,20 +271,13 @@
if (eob == 1) {
vpx_memset(dqcoeff, 0, 2 * sizeof(dqcoeff[0]));
- vpx_memset(token_cache, 0, 2 * sizeof(token_cache[0]));
} else {
- if (tx_type == DCT_DCT && tx_size <= TX_16X16 && eob <= 10) {
+ if (tx_type == DCT_DCT && tx_size <= TX_16X16 && eob <= 10)
vpx_memset(dqcoeff, 0, 4 * (4 << tx_size) * sizeof(dqcoeff[0]));
- vpx_memset(token_cache, 0,
- 4 * (4 << tx_size) * sizeof(token_cache[0]));
- } else if (tx_size == TX_32X32 && eob <= 34) {
+ else if (tx_size == TX_32X32 && eob <= 34)
vpx_memset(dqcoeff, 0, 256 * sizeof(dqcoeff[0]));
- vpx_memset(token_cache, 0, 256 * sizeof(token_cache[0]));
- } else {
+ else
vpx_memset(dqcoeff, 0, (16 << (tx_size << 1)) * sizeof(dqcoeff[0]));
- vpx_memset(token_cache, 0,
- (16 << (tx_size << 1)) * sizeof(token_cache[0]));
- }
}
}
}
@@ -297,7 +286,6 @@
VP9_COMMON *cm;
MACROBLOCKD *xd;
vp9_reader *r;
- uint8_t *token_cache;
};
static void predict_and_reconstruct_intra_block(int plane, int block,
@@ -326,9 +314,8 @@
if (!mi->mbmi.skip_coeff) {
vp9_decode_block_tokens(cm, xd, plane, block, plane_bsize, x, y, tx_size,
- args->r, args->token_cache);
- inverse_transform_block(xd, plane, block, tx_size, dst, pd->dst.stride,
- args->token_cache);
+ args->r);
+ inverse_transform_block(xd, plane, block, tx_size, dst, pd->dst.stride);
}
}
@@ -337,7 +324,6 @@
MACROBLOCKD *xd;
vp9_reader *r;
int *eobtotal;
- uint8_t *token_cache;
};
static void reconstruct_inter_block(int plane, int block,
@@ -351,10 +337,10 @@
txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &x, &y);
*args->eobtotal += vp9_decode_block_tokens(cm, xd, plane, block,
plane_bsize, x, y, tx_size,
- args->r, args->token_cache);
+ args->r);
inverse_transform_block(xd, plane, block, tx_size,
&pd->dst.buf[4 * y * pd->dst.stride + 4 * x],
- pd->dst.stride, args->token_cache);
+ pd->dst.stride);
}
static void set_offsets(VP9_COMMON *const cm, MACROBLOCKD *const xd,
@@ -404,8 +390,7 @@
static void decode_modes_b(VP9_COMMON *const cm, MACROBLOCKD *const xd,
const TileInfo *const tile,
int mi_row, int mi_col,
- vp9_reader *r, BLOCK_SIZE bsize,
- uint8_t *token_cache) {
+ vp9_reader *r, BLOCK_SIZE bsize) {
const int less8x8 = bsize < BLOCK_8X8;
MB_MODE_INFO *mbmi;
@@ -427,9 +412,7 @@
}
if (!is_inter_block(mbmi)) {
- struct intra_args arg = {
- cm, xd, r, token_cache
- };
+ struct intra_args arg = { cm, xd, r };
foreach_transformed_block(xd, bsize, predict_and_reconstruct_intra_block,
&arg);
} else {
@@ -447,9 +430,7 @@
// Reconstruction
if (!mbmi->skip_coeff) {
int eobtotal = 0;
- struct inter_args arg = {
- cm, xd, r, &eobtotal, token_cache
- };
+ struct inter_args arg = { cm, xd, r, &eobtotal };
foreach_transformed_block(xd, bsize, reconstruct_inter_block, &arg);
if (!less8x8 && eobtotal == 0)
mbmi->skip_coeff = 1; // skip loopfilter
@@ -488,8 +469,7 @@
static void decode_modes_sb(VP9_COMMON *const cm, MACROBLOCKD *const xd,
const TileInfo *const tile,
int mi_row, int mi_col,
- vp9_reader* r, BLOCK_SIZE bsize,
- uint8_t *token_cache) {
+ vp9_reader* r, BLOCK_SIZE bsize) {
const int hbs = num_8x8_blocks_wide_lookup[bsize] / 2;
PARTITION_TYPE partition;
BLOCK_SIZE subsize;
@@ -500,33 +480,27 @@
partition = read_partition(cm, xd, hbs, mi_row, mi_col, bsize, r);
subsize = get_subsize(bsize, partition);
if (subsize < BLOCK_8X8) {
- decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, token_cache);
+ decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize);
} else {
switch (partition) {
case PARTITION_NONE:
- decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, token_cache);
+ decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize);
break;
case PARTITION_HORZ:
- decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, token_cache);
+ decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize);
if (mi_row + hbs < cm->mi_rows)
- decode_modes_b(cm, xd, tile, mi_row + hbs, mi_col, r, subsize,
- token_cache);
+ decode_modes_b(cm, xd, tile, mi_row + hbs, mi_col, r, subsize);
break;
case PARTITION_VERT:
- decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, token_cache);
+ decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize);
if (mi_col + hbs < cm->mi_cols)
- decode_modes_b(cm, xd, tile, mi_row, mi_col + hbs, r, subsize,
- token_cache);
+ decode_modes_b(cm, xd, tile, mi_row, mi_col + hbs, r, subsize);
break;
case PARTITION_SPLIT:
- decode_modes_sb(cm, xd, tile, mi_row, mi_col, r, subsize,
- token_cache);
- decode_modes_sb(cm, xd, tile, mi_row, mi_col + hbs, r, subsize,
- token_cache);
- decode_modes_sb(cm, xd, tile, mi_row + hbs, mi_col, r, subsize,
- token_cache);
- decode_modes_sb(cm, xd, tile, mi_row + hbs, mi_col + hbs, r, subsize,
- token_cache);
+ decode_modes_sb(cm, xd, tile, mi_row, mi_col, r, subsize);
+ decode_modes_sb(cm, xd, tile, mi_row, mi_col + hbs, r, subsize);
+ decode_modes_sb(cm, xd, tile, mi_row + hbs, mi_col, r, subsize);
+ decode_modes_sb(cm, xd, tile, mi_row + hbs, mi_col + hbs, r, subsize);
break;
default:
assert(!"Invalid partition type");
@@ -809,8 +783,7 @@
vp9_zero(xd->left_seg_context);
for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end;
mi_col += MI_BLOCK_SIZE) {
- decode_modes_sb(cm, xd, tile, mi_row, mi_col, r, BLOCK_64X64,
- pbi->token_cache);
+ decode_modes_sb(cm, xd, tile, mi_row, mi_col, r, BLOCK_64X64);
}
if (pbi->do_loopfilter_inline) {
@@ -954,7 +927,6 @@
pd[i].dqcoeff = tile_data->dqcoeff[i];
pd[i].eobs = tile_data->eobs[i];
vpx_memset(xd->plane[i].dqcoeff, 0, 64 * 64 * sizeof(int16_t));
- vpx_memset(tile_data->token_cache, 0, sizeof(tile_data->token_cache));
}
}
@@ -970,8 +942,7 @@
for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end;
mi_col += MI_BLOCK_SIZE) {
decode_modes_sb(tile_data->cm, &tile_data->xd, tile,
- mi_row, mi_col, &tile_data->bit_reader, BLOCK_64X64,
- tile_data->token_cache);
+ mi_row, mi_col, &tile_data->bit_reader, BLOCK_64X64);
}
}
return !tile_data->xd.corrupted;
@@ -1215,7 +1186,7 @@
cm->tx_mode = xd->lossless ? ONLY_4X4 : read_tx_mode(&r);
if (cm->tx_mode == TX_MODE_SELECT)
- read_tx_probs(&fc->tx_probs, &r);
+ read_tx_mode_probs(&fc->tx_probs, &r);
read_coef_probs(fc, cm->tx_mode, &r);
for (k = 0; k < MBSKIP_CONTEXTS; ++k)
@@ -1233,7 +1204,8 @@
for (i = 0; i < INTRA_INTER_CONTEXTS; i++)
vp9_diff_update_prob(&r, &fc->intra_inter_prob[i]);
- read_comp_pred(cm, &r);
+ cm->comp_pred_mode = read_reference_mode(cm, &r);
+ read_reference_mode_probs(cm, &r);
for (j = 0; j < BLOCK_SIZE_GROUPS; j++)
for (i = 0; i < INTRA_MODES - 1; ++i)
diff --git a/vp9/decoder/vp9_detokenize.c b/vp9/decoder/vp9_detokenize.c
index 49da1a0..bdbe67d 100644
--- a/vp9/decoder/vp9_detokenize.c
+++ b/vp9/decoder/vp9_detokenize.c
@@ -60,16 +60,10 @@
254, 254, 254, 252, 249, 243, 230, 196, 177, 153, 140, 133, 130, 129, 0
};
-static const int token_to_counttoken[MAX_ENTROPY_TOKENS] = {
- ZERO_TOKEN, ONE_TOKEN, TWO_TOKEN, TWO_TOKEN,
- TWO_TOKEN, TWO_TOKEN, TWO_TOKEN, TWO_TOKEN,
- TWO_TOKEN, TWO_TOKEN, TWO_TOKEN, DCT_EOB_MODEL_TOKEN
-};
-
#define INCREMENT_COUNT(token) \
do { \
if (!cm->frame_parallel_decoding_mode) \
- ++coef_counts[band][pt][token_to_counttoken[token]]; \
+ ++coef_counts[band][pt][token]; \
} while (0)
@@ -77,7 +71,6 @@
{ \
v = (val * dqv) >> dq_shift; \
dqcoeff_ptr[scan[c]] = (vp9_read_bit(r) ? -v : v); \
- INCREMENT_COUNT(token); \
token_cache[scan[c]] = vp9_pt_energy_class[token]; \
++c; \
pt = get_coef_context(nb, token_cache, c); \
@@ -94,8 +87,7 @@
static int decode_coefs(VP9_COMMON *cm, const MACROBLOCKD *xd,
vp9_reader *r, int block_idx,
PLANE_TYPE type, int max_eob, int16_t *dqcoeff_ptr,
- TX_SIZE tx_size, const int16_t *dq, int pt,
- uint8_t *token_cache) {
+ TX_SIZE tx_size, const int16_t *dq, int pt) {
const FRAME_CONTEXT *const fc = &cm->fc;
FRAME_COUNTS *const counts = &cm->counts;
const int ref = is_inter_block(&xd->mi_8x8[0]->mbmi);
@@ -107,6 +99,7 @@
counts->coef[tx_size][type][ref];
unsigned int (*eob_branch_count)[PREV_COEF_CONTEXTS] =
counts->eob_branch[tx_size][type][ref];
+ uint8_t token_cache[32 * 32];
const uint8_t *cat6;
const uint8_t *band_translate = get_band_translate(tx_size);
const int dq_shift = (tx_size == TX_32X32);
@@ -123,14 +116,14 @@
if (!cm->frame_parallel_decoding_mode)
++eob_branch_count[band][pt];
if (!vp9_read(r, prob[EOB_CONTEXT_NODE])) {
- if (!cm->frame_parallel_decoding_mode)
- ++coef_counts[band][pt][DCT_EOB_MODEL_TOKEN];
+ INCREMENT_COUNT(DCT_EOB_MODEL_TOKEN);
break;
}
while (!vp9_read(r, prob[ZERO_CONTEXT_NODE])) {
INCREMENT_COUNT(ZERO_TOKEN);
dqv = dq[1];
+ token_cache[scan[c]] = 0;
++c;
if (c >= max_eob)
return c; // zero tokens at the end (no eob token)
@@ -141,12 +134,14 @@
// ONE_CONTEXT_NODE_0_
if (!vp9_read(r, prob[ONE_CONTEXT_NODE])) {
+ INCREMENT_COUNT(ONE_TOKEN);
WRITE_COEF_CONTINUE(1, ONE_TOKEN);
}
+ INCREMENT_COUNT(TWO_TOKEN);
+
prob = vp9_pareto8_full[prob[PIVOT_NODE] - 1];
- // LOW_VAL_CONTEXT_NODE_0_
if (!vp9_read(r, prob[LOW_VAL_CONTEXT_NODE])) {
if (!vp9_read(r, prob[TWO_CONTEXT_NODE])) {
WRITE_COEF_CONTINUE(2, TWO_TOKEN);
@@ -156,7 +151,7 @@
}
WRITE_COEF_CONTINUE(4, FOUR_TOKEN);
}
- // HIGH_LOW_CONTEXT_NODE_0_
+
if (!vp9_read(r, prob[HIGH_LOW_CONTEXT_NODE])) {
if (!vp9_read(r, prob[CAT_ONE_CONTEXT_NODE])) {
val = CAT1_MIN_VAL;
@@ -168,7 +163,7 @@
ADJUST_COEF(CAT2_PROB0, 0);
WRITE_COEF_CONTINUE(val, DCT_VAL_CATEGORY2);
}
- // CAT_THREEFOUR_CONTEXT_NODE_0_
+
if (!vp9_read(r, prob[CAT_THREEFOUR_CONTEXT_NODE])) {
if (!vp9_read(r, prob[CAT_THREE_CONTEXT_NODE])) {
val = CAT3_MIN_VAL;
@@ -184,7 +179,7 @@
ADJUST_COEF(CAT4_PROB0, 0);
WRITE_COEF_CONTINUE(val, DCT_VAL_CATEGORY4);
}
- // CAT_FIVE_CONTEXT_NODE_0_:
+
if (!vp9_read(r, prob[CAT_FIVE_CONTEXT_NODE])) {
val = CAT5_MIN_VAL;
ADJUST_COEF(CAT5_PROB4, 4);
@@ -208,8 +203,7 @@
int vp9_decode_block_tokens(VP9_COMMON *cm, MACROBLOCKD *xd,
int plane, int block, BLOCK_SIZE plane_bsize,
- int x, int y, TX_SIZE tx_size, vp9_reader *r,
- uint8_t *token_cache) {
+ int x, int y, TX_SIZE tx_size, vp9_reader *r) {
struct macroblockd_plane *const pd = &xd->plane[plane];
const int seg_eob = get_tx_eob(&cm->seg, xd->mi_8x8[0]->mbmi.segment_id,
tx_size);
@@ -217,7 +211,7 @@
pd->left_context + y);
const int eob = decode_coefs(cm, xd, r, block, pd->plane_type, seg_eob,
BLOCK_OFFSET(pd->dqcoeff, block), tx_size,
- pd->dequant, pt, token_cache);
+ pd->dequant, pt);
set_contexts(xd, pd, plane_bsize, tx_size, eob > 0, x, y);
pd->eobs[block] = eob;
return eob;
diff --git a/vp9/decoder/vp9_detokenize.h b/vp9/decoder/vp9_detokenize.h
index e858a19..2a88073 100644
--- a/vp9/decoder/vp9_detokenize.h
+++ b/vp9/decoder/vp9_detokenize.h
@@ -17,7 +17,6 @@
int vp9_decode_block_tokens(VP9_COMMON *cm, MACROBLOCKD *xd,
int plane, int block, BLOCK_SIZE plane_bsize,
- int x, int y, TX_SIZE tx_size, vp9_reader *r,
- uint8_t *token_cache);
+ int x, int y, TX_SIZE tx_size, vp9_reader *r);
#endif // VP9_DECODER_VP9_DETOKENIZE_H_
diff --git a/vp9/decoder/vp9_onyxd_int.h b/vp9/decoder/vp9_onyxd_int.h
index 3d1b97b..038cd96 100644
--- a/vp9/decoder/vp9_onyxd_int.h
+++ b/vp9/decoder/vp9_onyxd_int.h
@@ -52,8 +52,6 @@
ENTROPY_CONTEXT *above_context[MAX_MB_PLANE];
PARTITION_CONTEXT *above_seg_context;
-
- DECLARE_ALIGNED(16, uint8_t, token_cache[1024]);
} VP9D_COMP;
#endif // VP9_DECODER_VP9_ONYXD_INT_H_
diff --git a/vp9/encoder/vp9_encodemb.c b/vp9/encoder/vp9_encodemb.c
index 83f87b0..3691e7a 100644
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -559,8 +559,9 @@
mode = plane == 0 ? mbmi->mode : mbmi->uv_mode;
block >>= 6;
vp9_predict_intra_block(xd, block, bwl, TX_32X32, mode,
- dst, pd->dst.stride, dst, pd->dst.stride);
-
+ x->skip_encode ? src : dst,
+ x->skip_encode ? p->src.stride : pd->dst.stride,
+ dst, pd->dst.stride);
if (!x->skip_recode) {
vp9_subtract_block(32, 32, src_diff, diff_stride,
src, p->src.stride, dst, pd->dst.stride);
@@ -582,7 +583,9 @@
mode = plane == 0 ? mbmi->mode : mbmi->uv_mode;
block >>= 4;
vp9_predict_intra_block(xd, block, bwl, TX_16X16, mode,
- dst, pd->dst.stride, dst, pd->dst.stride);
+ x->skip_encode ? src : dst,
+ x->skip_encode ? p->src.stride : pd->dst.stride,
+ dst, pd->dst.stride);
if (!x->skip_recode) {
vp9_subtract_block(16, 16, src_diff, diff_stride,
src, p->src.stride, dst, pd->dst.stride);
@@ -600,7 +603,9 @@
mode = plane == 0 ? mbmi->mode : mbmi->uv_mode;
block >>= 2;
vp9_predict_intra_block(xd, block, bwl, TX_8X8, mode,
- dst, pd->dst.stride, dst, pd->dst.stride);
+ x->skip_encode ? src : dst,
+ x->skip_encode ? p->src.stride : pd->dst.stride,
+ dst, pd->dst.stride);
if (!x->skip_recode) {
vp9_subtract_block(8, 8, src_diff, diff_stride,
src, p->src.stride, dst, pd->dst.stride);
@@ -621,7 +626,9 @@
mode = plane == 0 ? mbmi->mode : mbmi->uv_mode;
vp9_predict_intra_block(xd, block, bwl, TX_4X4, mode,
- dst, pd->dst.stride, dst, pd->dst.stride);
+ x->skip_encode ? src : dst,
+ x->skip_encode ? p->src.stride : pd->dst.stride,
+ dst, pd->dst.stride);
if (!x->skip_recode) {
vp9_subtract_block(4, 4, src_diff, diff_stride,
diff --git a/vp9/encoder/vp9_onyx_if.c b/vp9/encoder/vp9_onyx_if.c
index 968cbfb..852cedf 100644
--- a/vp9/encoder/vp9_onyx_if.c
+++ b/vp9/encoder/vp9_onyx_if.c
@@ -159,6 +159,7 @@
if (!init_done) {
vp9_initialize_common();
+ vp9_coef_tree_initialize();
vp9_tokenize_initialize();
vp9_init_quant_tables();
vp9_init_me_luts();
@@ -166,7 +167,6 @@
// init_base_skip_probs();
vp9_entropy_mv_init();
vp9_entropy_mode_init();
- vp9_coef_tree_initialize();
init_done = 1;
}
}
diff --git a/vp9/encoder/vp9_quantize.c b/vp9/encoder/vp9_quantize.c
index 698130a..2591a57 100644
--- a/vp9/encoder/vp9_quantize.c
+++ b/vp9/encoder/vp9_quantize.c
@@ -137,45 +137,18 @@
*eob_ptr = eob + 1;
}
-struct plane_block_idx {
- int plane;
- int block;
-};
-
-// TODO(jkoleszar): returning a struct so it can be used in a const context,
-// expect to refactor this further later.
-static INLINE struct plane_block_idx plane_block_idx(int y_blocks,
- int b_idx) {
- const int v_offset = y_blocks * 5 / 4;
- struct plane_block_idx res;
-
- if (b_idx < y_blocks) {
- res.plane = 0;
- res.block = b_idx;
- } else if (b_idx < v_offset) {
- res.plane = 1;
- res.block = b_idx - y_blocks;
- } else {
- assert(b_idx < y_blocks * 3 / 2);
- res.plane = 2;
- res.block = b_idx - v_offset;
- }
- return res;
-}
-
-void vp9_regular_quantize_b_4x4(MACROBLOCK *x, int y_blocks, int b_idx,
+void vp9_regular_quantize_b_4x4(MACROBLOCK *x, int plane, int block,
const int16_t *scan, const int16_t *iscan) {
MACROBLOCKD *const xd = &x->e_mbd;
- const struct plane_block_idx pb_idx = plane_block_idx(y_blocks, b_idx);
- struct macroblock_plane* p = &x->plane[pb_idx.plane];
- struct macroblockd_plane* pd = &xd->plane[pb_idx.plane];
+ struct macroblock_plane* p = &x->plane[plane];
+ struct macroblockd_plane* pd = &xd->plane[plane];
- vp9_quantize_b(BLOCK_OFFSET(p->coeff, pb_idx.block),
+ vp9_quantize_b(BLOCK_OFFSET(p->coeff, block),
16, x->skip_block,
p->zbin, p->round, p->quant, p->quant_shift,
- BLOCK_OFFSET(p->qcoeff, pb_idx.block),
- BLOCK_OFFSET(pd->dqcoeff, pb_idx.block),
- pd->dequant, p->zbin_extra, &pd->eobs[pb_idx.block], scan, iscan);
+ BLOCK_OFFSET(p->qcoeff, block),
+ BLOCK_OFFSET(pd->dqcoeff, block),
+ pd->dequant, p->zbin_extra, &pd->eobs[block], scan, iscan);
}
static void invert_quant(int16_t *quant, int16_t *shift, int d) {
diff --git a/vp9/encoder/vp9_quantize.h b/vp9/encoder/vp9_quantize.h
index c078e1d..41cfa52 100644
--- a/vp9/encoder/vp9_quantize.h
+++ b/vp9/encoder/vp9_quantize.h
@@ -13,7 +13,7 @@
#include "vp9/encoder/vp9_block.h"
-void vp9_regular_quantize_b_4x4(MACROBLOCK *x, int y_blocks, int b_idx,
+void vp9_regular_quantize_b_4x4(MACROBLOCK *x, int plane, int block,
const int16_t *scan, const int16_t *iscan);
struct VP9_COMP;
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 84b7122..65cf5c7 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -283,7 +283,7 @@
cpi->mb.select_txfm_size = (cpi->sf.tx_size_search_method == USE_LARGESTALL &&
cm->frame_type != KEY_FRAME) ?
- 0 : 1;
+ 0 : 1;
set_block_thresholds(cpi);
@@ -1055,7 +1055,7 @@
else
x->fwd_txm4x4(src_diff, coeff, 8);
- vp9_regular_quantize_b_4x4(x, 4, block, so->scan, so->iscan);
+ vp9_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
ratey += cost_coeffs(x, 0, block, tempa + idx, templ + idy, TX_4X4,
so->scan, so->neighbors);
@@ -1299,11 +1299,7 @@
int this_rate_tokenonly, this_rate, s;
int64_t this_distortion, this_sse;
- // int mode_mask = (bsize <= BLOCK_8X8)
- // ? ALL_INTRA_MODES : cpi->sf.intra_uv_mode_mask;
-
- for (mode = DC_PRED; mode <= TM_PRED; mode ++) {
- // if (!(mode_mask & (1 << mode)))
+ for (mode = DC_PRED; mode <= TM_PRED; ++mode) {
if (!(cpi->sf.intra_uv_mode_mask[max_uv_txsize_lookup[bsize]]
& (1 << mode)))
continue;
@@ -1331,7 +1327,7 @@
struct macroblockd_plane *const pd = x->e_mbd.plane;
for (i = 1; i < MAX_MB_PLANE; ++i) {
p[i].coeff = ctx->coeff_pbuf[i][2];
- p[i].qcoeff = ctx->qcoeff_pbuf[i][2];
+ p[i].qcoeff = ctx->qcoeff_pbuf[i][2];
pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][2];
pd[i].eobs = ctx->eobs_pbuf[i][2];
@@ -1350,7 +1346,6 @@
}
x->e_mbd.mi_8x8[0]->mbmi.uv_mode = mode_selected;
-
return best_rd;
}
@@ -1546,7 +1541,7 @@
coeff = BLOCK_OFFSET(p->coeff, k);
x->fwd_txm4x4(raster_block_offset_int16(BLOCK_8X8, k, p->src_diff),
coeff, 8);
- vp9_regular_quantize_b_4x4(x, 4, k, so->scan, so->iscan);
+ vp9_regular_quantize_b_4x4(x, 0, k, so->scan, so->iscan);
thisdistortion += vp9_block_error(coeff, BLOCK_OFFSET(pd->dqcoeff, k),
16, &ssz);
thissse += ssz;
@@ -3675,7 +3670,9 @@
if (cpi->sf.use_uv_intra_rd_estimate) {
// Do Intra UV best rd mode selection if best mode choice above was intra.
if (vp9_mode_order[best_mode_index].ref_frame == INTRA_FRAME) {
- TX_SIZE uv_tx_size = get_uv_tx_size(mbmi);
+ TX_SIZE uv_tx_size;
+ *mbmi = best_mbmode;
+ uv_tx_size = get_uv_tx_size(mbmi);
rd_pick_intra_sbuv_mode(cpi, x, ctx, &rate_uv_intra[uv_tx_size],
&rate_uv_tokenonly[uv_tx_size],
&dist_uv[uv_tx_size],
@@ -4423,7 +4420,9 @@
if (cpi->sf.use_uv_intra_rd_estimate) {
// Do Intra UV best rd mode selection if best mode choice above was intra.
if (vp9_ref_order[best_mode_index].ref_frame == INTRA_FRAME) {
- TX_SIZE uv_tx_size = get_uv_tx_size(mbmi);
+ TX_SIZE uv_tx_size;
+ *mbmi = best_mbmode;
+ uv_tx_size = get_uv_tx_size(mbmi);
rd_pick_intra_sbuv_mode(cpi, x, ctx, &rate_uv_intra[uv_tx_size],
&rate_uv_tokenonly[uv_tx_size],
&dist_uv[uv_tx_size],
