ref: cfcd5c4f61a483c1913fde561619bf0eba31dd4b
parent: 8282c1a68dae80da79ff0e2bb936a6cfc0d9deb6
author: Deb Mukherjee <debargha@google.com>
date: Wed Nov 13 11:02:06 EST 2013
Simplifies band-getting with a static array Simplifies the code by implementing band mapping with static arrays. A lot of the code complexity introduced in a previous patch disappears. Change-Id: Ia3fac36e594fb5ad2d55ae141c58bba4c55c2d28
--- a/vp9/common/vp9_entropy.c
+++ b/vp9/common/vp9_entropy.c
@@ -37,15 +37,78 @@
};
DECLARE_ALIGNED(16, const uint8_t,
- vp9_coefband_trans_8x8plus[MAXBAND_INDEX + 1]) = {+ vp9_coefband_trans_8x8plus[1024]) = {0, 1, 1, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4,
- 4, 4, 4, 4, 4, 5
+ 4, 4, 4, 4, 4, 5,
+ // beyond MAXBAND_INDEX+1 all values are filled as 5
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+ 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
};
DECLARE_ALIGNED(16, const uint8_t,
- vp9_coefband_trans_4x4[MAXBAND_INDEX + 1]) = {+ vp9_coefband_trans_4x4[16]) = {0, 1, 1, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 5, 5, 5,
- 5, 5, 5, 5, 5, 5
};
DECLARE_ALIGNED(16, const uint8_t, vp9_pt_energy_class[MAX_ENTROPY_TOKENS]) = {--- a/vp9/common/vp9_entropy.h
+++ b/vp9/common/vp9_entropy.h
@@ -120,12 +120,16 @@
// This is the index in the scan order beyond which all coefficients for
// 8x8 transform and above are in the top band.
-// For 4x4 blocks the index is less but to keep things common the lookup
-// table for 4x4 is padded out to this index.
+// This macro is currently unused but may be used by certain implementations
#define MAXBAND_INDEX 21
-extern const uint8_t vp9_coefband_trans_8x8plus[MAXBAND_INDEX + 1];
-extern const uint8_t vp9_coefband_trans_4x4[MAXBAND_INDEX + 1];
+extern const uint8_t vp9_coefband_trans_8x8plus[1024];
+extern const uint8_t vp9_coefband_trans_4x4[16];
+
+static const uint8_t *get_band_translate(TX_SIZE tx_size) {+ return tx_size == TX_4X4 ? vp9_coefband_trans_4x4
+ : vp9_coefband_trans_8x8plus;
+}
// 128 lists of probabilities are stored for the following ONE node probs:
// 1, 3, 5, 7, ..., 253, 255
--- a/vp9/decoder/vp9_decodframe.c
+++ b/vp9/decoder/vp9_decodframe.c
@@ -45,7 +45,6 @@
DECLARE_ALIGNED(16, int16_t, qcoeff[MAX_MB_PLANE][64 * 64]);
DECLARE_ALIGNED(16, int16_t, dqcoeff[MAX_MB_PLANE][64 * 64]);
DECLARE_ALIGNED(16, uint16_t, eobs[MAX_MB_PLANE][256]);
- const uint8_t *band_translate[2];
} TileWorkerData;
static int read_be32(const uint8_t *p) {@@ -295,7 +294,6 @@
MACROBLOCKD *xd;
vp9_reader *r;
uint8_t *token_cache;
- const uint8_t *band_translate[2];
};
static void predict_and_reconstruct_intra_block(int plane, int block,
@@ -304,9 +302,6 @@
struct intra_args *const args = arg;
VP9_COMMON *const cm = args->cm;
MACROBLOCKD *const xd = args->xd;
- const uint8_t *band_translate[2] = {- args->band_translate[0], args->band_translate[1]
- };
struct macroblockd_plane *const pd = &xd->plane[plane];
MODE_INFO *const mi = xd->mi_8x8[0];
const MB_PREDICTION_MODE mode = (plane == 0)
@@ -327,7 +322,7 @@
if (!mi->mbmi.skip_coeff) {vp9_decode_block_tokens(cm, xd, plane, block, plane_bsize, x, y, tx_size,
- args->r, args->token_cache, band_translate);
+ args->r, args->token_cache);
inverse_transform_block(xd, plane, block, tx_size, x, y);
}
}
@@ -338,7 +333,6 @@
vp9_reader *r;
int *eobtotal;
uint8_t *token_cache;
- const uint8_t *band_translate[2];
};
static void reconstruct_inter_block(int plane, int block,
@@ -347,16 +341,12 @@
struct inter_args *args = arg;
VP9_COMMON *const cm = args->cm;
MACROBLOCKD *const xd = args->xd;
- const uint8_t *band_translate[2] = {- args->band_translate[0], args->band_translate[1]
- };
int x, y;
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,
- band_translate);
+ args->r, args->token_cache);
inverse_transform_block(xd, plane, block, tx_size, x, y);
}
@@ -408,8 +398,7 @@
const TileInfo *const tile,
int mi_row, int mi_col,
vp9_reader *r, BLOCK_SIZE bsize,
- uint8_t *token_cache,
- const uint8_t *band_translate[2]) {+ uint8_t *token_cache) {const int less8x8 = bsize < BLOCK_8X8;
MB_MODE_INFO *mbmi;
@@ -432,7 +421,7 @@
if (!is_inter_block(mbmi)) { struct intra_args arg = {- cm, xd, r, token_cache, {band_translate[0], band_translate[1]}+ cm, xd, r, token_cache
};
foreach_transformed_block(xd, bsize, predict_and_reconstruct_intra_block,
&arg);
@@ -452,8 +441,7 @@
if (!mbmi->skip_coeff) {int eobtotal = 0;
struct inter_args arg = {- cm, xd, r, &eobtotal, token_cache,
- {band_translate[0], band_translate[1]}+ cm, xd, r, &eobtotal, token_cache
};
foreach_transformed_block(xd, bsize, reconstruct_inter_block, &arg);
if (!less8x8 && eobtotal == 0)
@@ -494,8 +482,7 @@
const TileInfo *const tile,
int mi_row, int mi_col,
vp9_reader* r, BLOCK_SIZE bsize,
- uint8_t *token_cache,
- const uint8_t *band_translate[2]) {+ uint8_t *token_cache) {const int hbs = num_8x8_blocks_wide_lookup[bsize] / 2;
PARTITION_TYPE partition;
BLOCK_SIZE subsize;
@@ -506,37 +493,33 @@
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,
- band_translate);
+ decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, token_cache);
} else { switch (partition) {case PARTITION_NONE:
- decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, token_cache,
- band_translate);
+ decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, token_cache);
break;
case PARTITION_HORZ:
- decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, token_cache,
- band_translate);
+ decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, token_cache);
if (mi_row + hbs < cm->mi_rows)
decode_modes_b(cm, xd, tile, mi_row + hbs, mi_col, r, subsize,
- token_cache, band_translate);
+ token_cache);
break;
case PARTITION_VERT:
- decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, token_cache,
- band_translate);
+ decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, token_cache);
if (mi_col + hbs < cm->mi_cols)
decode_modes_b(cm, xd, tile, mi_row, mi_col + hbs, r, subsize,
- token_cache, band_translate);
+ token_cache);
break;
case PARTITION_SPLIT:
decode_modes_sb(cm, xd, tile, mi_row, mi_col, r, subsize,
- token_cache, band_translate);
+ token_cache);
decode_modes_sb(cm, xd, tile, mi_row, mi_col + hbs, r, subsize,
- token_cache, band_translate);
+ token_cache);
decode_modes_sb(cm, xd, tile, mi_row + hbs, mi_col, r, subsize,
- token_cache, band_translate);
+ token_cache);
decode_modes_sb(cm, xd, tile, mi_row + hbs, mi_col + hbs, r, subsize,
- token_cache, band_translate);
+ token_cache);
break;
default:
assert(!"Invalid partition type");
@@ -820,11 +803,8 @@
vp9_zero(xd->left_seg_context);
for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end;
mi_col += MI_BLOCK_SIZE) {- const uint8_t *band_translate[2] = {- vp9_coefband_trans_4x4, pbi->coefband_trans_8x8plus
- };
decode_modes_sb(cm, xd, tile, mi_row, mi_col, r, BLOCK_64X64,
- pbi->token_cache, band_translate);
+ pbi->token_cache);
}
if (pbi->do_loopfilter_inline) {@@ -985,8 +965,7 @@
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,
- tile_data->band_translate);
+ tile_data->token_cache);
}
}
return !tile_data->xd.corrupted;
@@ -1045,8 +1024,6 @@
tile_data->cm = cm;
tile_data->xd = pbi->mb;
tile_data->xd.corrupted = 0;
- tile_data->band_translate[0] = vp9_coefband_trans_4x4;
- tile_data->band_translate[1] = pbi->coefband_trans_8x8plus;
vp9_tile_init(tile, tile_data->cm, 0, tile_col);
setup_token_decoder(data, data_end, size, &cm->error,
--- a/vp9/decoder/vp9_detokenize.c
+++ b/vp9/decoder/vp9_detokenize.c
@@ -93,8 +93,7 @@
vp9_reader *r, int block_idx,
PLANE_TYPE type, int seg_eob, int16_t *dqcoeff_ptr,
TX_SIZE tx_size, const int16_t *dq, int pt,
- uint8_t *token_cache,
- const uint8_t *band_translate) {+ uint8_t *token_cache) {const FRAME_CONTEXT *const fc = &cm->fc;
FRAME_COUNTS *const counts = &cm->counts;
const int ref = is_inter_block(&xd->mi_8x8[0]->mbmi);
@@ -110,6 +109,7 @@
counts->eob_branch[tx_size][type][ref];
const int16_t *scan, *nb;
const uint8_t *cat6;
+ const uint8_t *band_translate = get_band_translate(tx_size);
get_scan(xd, tx_size, type, block_idx, &scan, &nb);
while (c < seg_eob) {@@ -219,8 +219,7 @@
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,
- const uint8_t *band_translate[2]) {+ uint8_t *token_cache) {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);
@@ -228,8 +227,7 @@
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,
- band_translate[tx_size != TX_4X4]);
+ pd->dequant, pt, token_cache);
set_contexts(xd, pd, plane_bsize, tx_size, eob > 0, x, y);
pd->eobs[block] = eob;
return eob;
--- a/vp9/decoder/vp9_detokenize.h
+++ b/vp9/decoder/vp9_detokenize.h
@@ -18,7 +18,6 @@
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,
- const uint8_t *band_translate[2]);
+ uint8_t *token_cache);
#endif // VP9_DECODER_VP9_DETOKENIZE_H_
--- a/vp9/encoder/vp9_block.h
+++ b/vp9/encoder/vp9_block.h
@@ -194,9 +194,6 @@
BLOCK_SIZE sb64_partitioning;
void (*fwd_txm4x4)(const int16_t *input, int16_t *output, int stride);
-
- // band cache
- DECLARE_ALIGNED(16, uint8_t, coefband_trans_8x8plus[1024]);
};
// TODO(jingning): the variables used here are little complicated. need further
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -137,9 +137,7 @@
const int mul = 1 + (tx_size == TX_32X32);
uint8_t token_cache[1024];
const int16_t *dequant_ptr = pd->dequant;
- const uint8_t *const band_translate = (tx_size == TX_4X4 ?
- vp9_coefband_trans_4x4 :
- mb->coefband_trans_8x8plus);
+ const uint8_t *const band_translate = get_band_translate(tx_size);
assert((!type && !plane) || (type && plane));
dqcoeff_ptr = BLOCK_OFFSET(pd->dqcoeff, block);
--- a/vp9/encoder/vp9_onyx_if.c
+++ b/vp9/encoder/vp9_onyx_if.c
@@ -1224,13 +1224,6 @@
cpi->fixed_divide[0] = 0;
for (i = 1; i < 512; i++)
cpi->fixed_divide[i] = 0x80000 / i;
-
- vpx_memset(cpi->mb.coefband_trans_8x8plus,
- (COEF_BANDS-1),
- sizeof(cpi->mb.coefband_trans_8x8plus));
- vpx_memcpy(cpi->mb.coefband_trans_8x8plus,
- vp9_coefband_trans_8x8plus,
- sizeof(vp9_coefband_trans_8x8plus));
}
--- a/vp9/encoder/vp9_tokenize.c
+++ b/vp9/encoder/vp9_tokenize.c
@@ -115,9 +115,7 @@
vp9_coeff_count *const counts = cpi->coef_counts[tx_size];
vp9_coeff_probs_model *const coef_probs = cpi->common.fc.coef_probs[tx_size];
const int ref = is_inter_block(mbmi);
- const uint8_t *const band_translate = (tx_size == TX_4X4 ?
- vp9_coefband_trans_4x4 :
- cpi->mb.coefband_trans_8x8plus);
+ const uint8_t *const band_translate = get_band_translate(tx_size);
const int seg_eob = get_tx_eob(&cpi->common.seg, segment_id, tx_size);
int aoff, loff;
txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &aoff, &loff);
--
⑨