ref: 14011f037d0353d1bef1b21dd65a348d9792bfd3
parent: c02a4beed882521dd1ddcecbda88934868ea5844
author: Jingning Han <[email protected]>
date: Fri Jul 1 08:20:45 EDT 2016
Remove txfrm_block_to_raster_xy() from vp9 encoder The transform block row and column positions are always available outside the callees. There is no need to re-compute these values again. This approach has been used by the decoder. This commit removes txfrm_block_to_raster_xy() function. Change-Id: I5b90f91a0d8b7c35cfa7d171da9edf8202630108
--- a/vp9/common/vp9_blockd.c
+++ b/vp9/common/vp9_blockd.c
@@ -66,7 +66,7 @@
for (r = 0; r < max_blocks_high; r += (1 << tx_size)) {// Skip visiting the sub blocks that are wholly within the UMV.
for (c = 0; c < max_blocks_wide; c += (1 << tx_size)) {- visit(plane, i, plane_bsize, tx_size, arg);
+ visit(plane, i, r, c, plane_bsize, tx_size, arg);
i += step;
}
i += extra_step;
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -270,6 +270,7 @@
}
typedef void (*foreach_transformed_block_visitor)(int plane, int block,
+ int row, int col,
BLOCK_SIZE plane_bsize,
TX_SIZE tx_size,
void *arg);
@@ -282,17 +283,6 @@
void vp9_foreach_transformed_block(
const MACROBLOCKD* const xd, BLOCK_SIZE bsize,
foreach_transformed_block_visitor visit, void *arg);
-
-static INLINE void txfrm_block_to_raster_xy(BLOCK_SIZE plane_bsize,
- TX_SIZE tx_size, int block,
- int *x, int *y) {- const int bwl = b_width_log2_lookup[plane_bsize];
- const int tx_cols_log2 = bwl - tx_size;
- const int tx_cols = 1 << tx_cols_log2;
- const int raster_mb = block >> (tx_size << 1);
- *x = (raster_mb & (tx_cols - 1)) << tx_size;
- *y = (raster_mb >> tx_cols_log2) << tx_size;
-}
void vp9_set_contexts(const MACROBLOCKD *xd, struct macroblockd_plane *pd,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size, int has_eob,
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -335,7 +335,7 @@
}
#endif // CONFIG_VP9_HIGHBITDEPTH
-void vp9_xform_quant_fp(MACROBLOCK *x, int plane, int block,
+void vp9_xform_quant_fp(MACROBLOCK *x, int plane, int block, int row, int col,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size) {MACROBLOCKD *const xd = &x->e_mbd;
const struct macroblock_plane *const p = &x->plane[plane];
@@ -346,10 +346,8 @@
tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
uint16_t *const eob = &p->eobs[block];
const int diff_stride = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
- int i, j;
const int16_t *src_diff;
- txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
- src_diff = &p->src_diff[4 * (j * diff_stride + i)];
+ src_diff = &p->src_diff[4 * (row * diff_stride + col)];
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {@@ -425,7 +423,7 @@
}
}
-void vp9_xform_quant_dc(MACROBLOCK *x, int plane, int block,
+void vp9_xform_quant_dc(MACROBLOCK *x, int plane, int block, int row, int col,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size) {MACROBLOCKD *const xd = &x->e_mbd;
const struct macroblock_plane *const p = &x->plane[plane];
@@ -435,12 +433,8 @@
tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
uint16_t *const eob = &p->eobs[block];
const int diff_stride = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
- int i, j;
const int16_t *src_diff;
-
- txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
- src_diff = &p->src_diff[4 * (j * diff_stride + i)];
-
+ src_diff = &p->src_diff[4 * (row * diff_stride + col)];
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) { switch (tx_size) {@@ -506,7 +500,7 @@
}
}
-void vp9_xform_quant(MACROBLOCK *x, int plane, int block,
+void vp9_xform_quant(MACROBLOCK *x, int plane, int block, int row, int col,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size) {MACROBLOCKD *const xd = &x->e_mbd;
const struct macroblock_plane *const p = &x->plane[plane];
@@ -517,10 +511,8 @@
tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
uint16_t *const eob = &p->eobs[block];
const int diff_stride = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
- int i, j;
const int16_t *src_diff;
- txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
- src_diff = &p->src_diff[4 * (j * diff_stride + i)];
+ src_diff = &p->src_diff[4 * (row * diff_stride + col)];
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {@@ -595,7 +587,8 @@
}
}
-static void encode_block(int plane, int block, BLOCK_SIZE plane_bsize,
+static void encode_block(int plane, int block, int row, int col,
+ BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {struct encode_b_args *const args = arg;
MACROBLOCK *const x = args->x;
@@ -604,13 +597,11 @@
struct macroblock_plane *const p = &x->plane[plane];
struct macroblockd_plane *const pd = &xd->plane[plane];
tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
- int i, j;
uint8_t *dst;
ENTROPY_CONTEXT *a, *l;
- txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
- dst = &pd->dst.buf[4 * j * pd->dst.stride + 4 * i];
- a = &ctx->ta[plane][i];
- l = &ctx->tl[plane][j];
+ dst = &pd->dst.buf[4 * row * pd->dst.stride + 4 * col];
+ a = &ctx->ta[plane][col];
+ l = &ctx->tl[plane][row];
// TODO(jingning): per transformed block zero forcing only enabled for
// luma component. will integrate chroma components as well.
@@ -629,7 +620,7 @@
*a = *l = 0;
return;
} else {- vp9_xform_quant_fp(x, plane, block, plane_bsize, tx_size);
+ vp9_xform_quant_fp(x, plane, block, row, col, plane_bsize, tx_size);
}
} else { if (max_txsize_lookup[plane_bsize] == tx_size) {@@ -636,10 +627,10 @@
int txfm_blk_index = (plane << 2) + (block >> (tx_size << 1));
if (x->skip_txfm[txfm_blk_index] == SKIP_TXFM_NONE) {// full forward transform and quantization
- vp9_xform_quant(x, plane, block, plane_bsize, tx_size);
+ vp9_xform_quant(x, plane, block, row, col, plane_bsize, tx_size);
} else if (x->skip_txfm[txfm_blk_index] == SKIP_TXFM_AC_ONLY) {// fast path forward transform and quantization
- vp9_xform_quant_dc(x, plane, block, plane_bsize, tx_size);
+ vp9_xform_quant_dc(x, plane, block, row, col, plane_bsize, tx_size);
} else {// skip forward transform
p->eobs[block] = 0;
@@ -647,7 +638,7 @@
return;
}
} else {- vp9_xform_quant(x, plane, block, plane_bsize, tx_size);
+ vp9_xform_quant(x, plane, block, row, col, plane_bsize, tx_size);
}
}
}
@@ -715,7 +706,8 @@
}
}
-static void encode_block_pass1(int plane, int block, BLOCK_SIZE plane_bsize,
+static void encode_block_pass1(int plane, int block, int row, int col,
+ BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {MACROBLOCK *const x = (MACROBLOCK *)arg;
MACROBLOCKD *const xd = &x->e_mbd;
@@ -722,12 +714,10 @@
struct macroblock_plane *const p = &x->plane[plane];
struct macroblockd_plane *const pd = &xd->plane[plane];
tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
- int i, j;
uint8_t *dst;
- txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
- dst = &pd->dst.buf[4 * j * pd->dst.stride + 4 * i];
+ dst = &pd->dst.buf[4 * row * pd->dst.stride + 4 * col];
- vp9_xform_quant(x, plane, block, plane_bsize, tx_size);
+ vp9_xform_quant(x, plane, block, row, col, plane_bsize, tx_size);
if (p->eobs[block] > 0) {#if CONFIG_VP9_HIGHBITDEPTH
@@ -774,7 +764,8 @@
}
}
-void vp9_encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
+void vp9_encode_block_intra(int plane, int block, int row, int col,
+ BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {struct encode_b_args* const args = arg;
MACROBLOCK *const x = args->x;
@@ -795,18 +786,16 @@
uint16_t *eob = &p->eobs[block];
const int src_stride = p->src.stride;
const int dst_stride = pd->dst.stride;
- int i, j;
struct optimize_ctx *const ctx = args->ctx;
ENTROPY_CONTEXT *a = NULL;
ENTROPY_CONTEXT *l = NULL;
int entropy_ctx = 0;
- txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
- dst = &pd->dst.buf[4 * (j * dst_stride + i)];
- src = &p->src.buf[4 * (j * src_stride + i)];
- src_diff = &p->src_diff[4 * (j * diff_stride + i)];
+ dst = &pd->dst.buf[4 * (row * dst_stride + col)];
+ src = &p->src.buf[4 * (row * src_stride + col)];
+ src_diff = &p->src_diff[4 * (row * diff_stride + col)];
if (args->ctx != NULL) {- a = &ctx->ta[plane][i];
- l = &ctx->tl[plane][j];
+ a = &ctx->ta[plane][col];
+ l = &ctx->tl[plane][row];
entropy_ctx = combine_entropy_contexts(*a, *l);
}
@@ -826,7 +815,7 @@
vp9_predict_intra_block(xd, bwl, tx_size, mode, x->skip_encode ? src : dst,
x->skip_encode ? src_stride : dst_stride,
- dst, dst_stride, i, j, plane);
+ dst, dst_stride, col, row, plane);
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {--- a/vp9/encoder/vp9_encodemb.h
+++ b/vp9/encoder/vp9_encodemb.h
@@ -25,16 +25,17 @@
};
void vp9_encode_sb(MACROBLOCK *x, BLOCK_SIZE bsize);
void vp9_encode_sby_pass1(MACROBLOCK *x, BLOCK_SIZE bsize);
-void vp9_xform_quant_fp(MACROBLOCK *x, int plane, int block,
+void vp9_xform_quant_fp(MACROBLOCK *x, int plane, int block, int row, int col,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size);
-void vp9_xform_quant_dc(MACROBLOCK *x, int plane, int block,
+void vp9_xform_quant_dc(MACROBLOCK *x, int plane, int block, int row, int col,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size);
-void vp9_xform_quant(MACROBLOCK *x, int plane, int block,
+void vp9_xform_quant(MACROBLOCK *x, int plane, int block, int row, int col,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size);
void vp9_subtract_plane(MACROBLOCK *x, BLOCK_SIZE bsize, int plane);
-void vp9_encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
+void vp9_encode_block_intra(int plane, int block, int row, int col,
+ BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg);
void vp9_encode_intra_block_plane(MACROBLOCK *x, BLOCK_SIZE bsize, int plane,
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -932,7 +932,8 @@
RD_COST *rdc;
};
-static void estimate_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
+static void estimate_block_intra(int plane, int block, int row, int col,
+ BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {struct estimate_block_intra_args* const args = arg;
VP9_COMP *const cpi = args->cpi;
@@ -945,13 +946,12 @@
uint8_t *const dst_buf_base = pd->dst.buf;
const int src_stride = p->src.stride;
const int dst_stride = pd->dst.stride;
- int i, j;
RD_COST this_rdc;
- txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
+ (void)block;
- p->src.buf = &src_buf_base[4 * (j * src_stride + i)];
- pd->dst.buf = &dst_buf_base[4 * (j * dst_stride + i)];
+ p->src.buf = &src_buf_base[4 * (row * src_stride + col)];
+ pd->dst.buf = &dst_buf_base[4 * (row * dst_stride + col)];
// Use source buffer as an approximation for the fully reconstructed buffer.
vp9_predict_intra_block(xd, b_width_log2_lookup[plane_bsize],
tx_size, args->mode,
@@ -958,7 +958,7 @@
x->skip_encode ? p->src.buf : pd->dst.buf,
x->skip_encode ? src_stride : dst_stride,
pd->dst.buf, dst_stride,
- i, j, plane);
+ col, row, plane);
if (plane == 0) {int64_t this_sse = INT64_MAX;
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -498,18 +498,16 @@
}
}
-static int rate_block(int plane, int block, BLOCK_SIZE plane_bsize,
+static int rate_block(int plane, int block, int row, int col,
TX_SIZE tx_size, struct rdcost_block_args* args) {- int x_idx, y_idx;
- txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &x_idx, &y_idx);
-
- return cost_coeffs(args->x, plane, block, args->t_above + x_idx,
- args->t_left + y_idx, tx_size,
+ return cost_coeffs(args->x, plane, block, args->t_above + col,
+ args->t_left + row, tx_size,
args->so->scan, args->so->neighbors,
args->use_fast_coef_costing);
}
-static void block_rd_txfm(int plane, int block, BLOCK_SIZE plane_bsize,
+static void block_rd_txfm(int plane, int block, int row, int col,
+ BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {struct rdcost_block_args *args = arg;
MACROBLOCK *const x = args->x;
@@ -525,13 +523,13 @@
if (!is_inter_block(mi)) { struct encode_b_args arg = {x, NULL, &mi->skip};- vp9_encode_block_intra(plane, block, plane_bsize, tx_size, &arg);
+ vp9_encode_block_intra(plane, block, row, col, plane_bsize, tx_size, &arg);
dist_block(x, plane, block, tx_size, &dist, &sse);
} else if (max_txsize_lookup[plane_bsize] == tx_size) {if (x->skip_txfm[(plane << 2) + (block >> (tx_size << 1))] ==
SKIP_TXFM_NONE) {// full forward transform and quantization
- vp9_xform_quant(x, plane, block, plane_bsize, tx_size);
+ vp9_xform_quant(x, plane, block, row, col, plane_bsize, tx_size);
dist_block(x, plane, block, tx_size, &dist, &sse);
} else if (x->skip_txfm[(plane << 2) + (block >> (tx_size << 1))] ==
SKIP_TXFM_AC_ONLY) {@@ -538,7 +536,7 @@
// compute DC coefficient
tran_low_t *const coeff = BLOCK_OFFSET(x->plane[plane].coeff, block);
tran_low_t *const dqcoeff = BLOCK_OFFSET(xd->plane[plane].dqcoeff, block);
- vp9_xform_quant_dc(x, plane, block, plane_bsize, tx_size);
+ vp9_xform_quant_dc(x, plane, block, row, col, plane_bsize, tx_size);
sse = x->bsse[(plane << 2) + (block >> (tx_size << 1))] << 4;
dist = sse;
if (x->plane[plane].eobs[block]) {@@ -562,7 +560,7 @@
}
} else {// full forward transform and quantization
- vp9_xform_quant(x, plane, block, plane_bsize, tx_size);
+ vp9_xform_quant(x, plane, block, row, col, plane_bsize, tx_size);
dist_block(x, plane, block, tx_size, &dist, &sse);
}
@@ -572,7 +570,7 @@
return;
}
- rate = rate_block(plane, block, plane_bsize, tx_size, args);
+ rate = rate_block(plane, block, row, col, tx_size, args);
rd1 = RDCOST(x->rdmult, x->rddiv, rate, dist);
rd2 = RDCOST(x->rdmult, x->rddiv, 0, sse);
--- a/vp9/encoder/vp9_tokenize.c
+++ b/vp9/encoder/vp9_tokenize.c
@@ -319,7 +319,8 @@
TOKENEXTRA **tp;
};
-static void set_entropy_context_b(int plane, int block, BLOCK_SIZE plane_bsize,
+static void set_entropy_context_b(int plane, int block, int row, int col,
+ BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {struct tokenize_b_args* const args = arg;
ThreadData *const td = args->td;
@@ -327,10 +328,8 @@
MACROBLOCKD *const xd = &x->e_mbd;
struct macroblock_plane *p = &x->plane[plane];
struct macroblockd_plane *pd = &xd->plane[plane];
- int aoff, loff;
- txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &aoff, &loff);
vp9_set_contexts(xd, pd, plane_bsize, tx_size, p->eobs[block] > 0,
- aoff, loff);
+ col, row);
}
static INLINE void add_token(TOKENEXTRA **t, const vpx_prob *context_tree,
@@ -353,7 +352,8 @@
++counts[token];
}
-static void tokenize_b(int plane, int block, BLOCK_SIZE plane_bsize,
+static void tokenize_b(int plane, int block, int row, int col,
+ BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {struct tokenize_b_args* const args = arg;
VP9_COMP *cpi = args->cpi;
@@ -384,11 +384,8 @@
const int tx_eob = 16 << (tx_size << 1);
int16_t token;
EXTRABIT extra;
- int aoff, loff;
- txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &aoff, &loff);
-
- pt = get_entropy_context(tx_size, pd->above_context + aoff,
- pd->left_context + loff);
+ pt = get_entropy_context(tx_size, pd->above_context + col,
+ pd->left_context + row);
so = get_scan(xd, tx_size, type, block);
scan = so->scan;
nb = so->neighbors;
@@ -426,7 +423,7 @@
*tp = t;
- vp9_set_contexts(xd, pd, plane_bsize, tx_size, c > 0, aoff, loff);
+ vp9_set_contexts(xd, pd, plane_bsize, tx_size, c > 0, col, row);
}
struct is_skippable_args {@@ -433,7 +430,8 @@
uint16_t *eobs;
int *skippable;
};
-static void is_skippable(int plane, int block,
+
+static void is_skippable(int plane, int block, int row, int col,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
void *argv) {struct is_skippable_args *args = argv;
@@ -440,6 +438,8 @@
(void)plane;
(void)plane_bsize;
(void)tx_size;
+ (void)row;
+ (void)col;
args->skippable[0] &= (!args->eobs[block]);
}
@@ -453,7 +453,7 @@
return result;
}
-static void has_high_freq_coeff(int plane, int block,
+static void has_high_freq_coeff(int plane, int block, int row, int col,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
void *argv) {struct is_skippable_args *args = argv;
@@ -460,7 +460,8 @@
int eobs = (tx_size == TX_4X4) ? 3 : 10;
(void) plane;
(void) plane_bsize;
-
+ (void) row;
+ (void) col;
*(args->skippable) |= (args->eobs[block] > eobs);
}