vpx/vp9/common/vp9_reconintra.c

769 lines
27 KiB
C
Raw Normal View History

2010-05-18 17:58:33 +02:00
/*
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
2010-05-18 17:58:33 +02:00
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
2010-05-18 17:58:33 +02:00
*/
#include <stdio.h>
#include "./vpx_config.h"
#include "vp9_rtcd.h"
#include "vp9/common/vp9_reconintra.h"
2010-05-18 17:58:33 +02:00
#include "vpx_mem/vpx_mem.h"
/* For skip_recon_mb(), add vp9_build_intra_predictors_mby_s(MACROBLOCKD *xd)
* and vp9_build_intra_predictors_mbuv_s(MACROBLOCKD *xd).
*/
static void d27_predictor(uint8_t *ypred_ptr, int y_stride, int n,
uint8_t *yabove_row, uint8_t *yleft_col) {
int r, c, h, w, v;
int a, b;
r = 0;
for (c = 0; c < n - 2; c++) {
if (c & 1)
a = yleft_col[r + 1];
else
a = (yleft_col[r] + yleft_col[r + 1] + 1) >> 1;
b = yabove_row[c + 2];
ypred_ptr[c] = (2 * a + (c + 1) * b + (c + 3) / 2) / (c + 3);
}
for (r = 1; r < n / 2 - 1; r++) {
for (c = 0; c < n - 2 - 2 * r; c++) {
if (c & 1)
a = yleft_col[r + 1];
else
a = (yleft_col[r] + yleft_col[r + 1] + 1) >> 1;
b = ypred_ptr[(r - 1) * y_stride + c + 2];
ypred_ptr[r * y_stride + c] = (2 * a + (c + 1) * b + (c + 3) / 2) / (c + 3);
}
}
for (; r < n - 1; ++r) {
for (c = 0; c < n; c++) {
v = (c & 1 ? yleft_col[r + 1] : (yleft_col[r] + yleft_col[r + 1] + 1) >> 1);
h = r - c / 2;
ypred_ptr[h * y_stride + c] = v;
}
}
c = 0;
r = n - 1;
ypred_ptr[r * y_stride] = (ypred_ptr[(r - 1) * y_stride] +
yleft_col[r] + 1) >> 1;
for (r = n - 2; r >= n / 2; --r) {
w = c + (n - 1 - r) * 2;
ypred_ptr[r * y_stride + w] = (ypred_ptr[(r - 1) * y_stride + w] +
ypred_ptr[r * y_stride + w - 1] + 1) >> 1;
}
for (c = 1; c < n; c++) {
for (r = n - 1; r >= n / 2 + c / 2; --r) {
w = c + (n - 1 - r) * 2;
ypred_ptr[r * y_stride + w] = (ypred_ptr[(r - 1) * y_stride + w] +
ypred_ptr[r * y_stride + w - 1] + 1) >> 1;
}
}
}
static void d63_predictor(uint8_t *ypred_ptr, int y_stride, int n,
uint8_t *yabove_row, uint8_t *yleft_col) {
int r, c, h, w, v;
int a, b;
c = 0;
for (r = 0; r < n - 2; r++) {
if (r & 1)
a = yabove_row[c + 1];
else
a = (yabove_row[c] + yabove_row[c + 1] + 1) >> 1;
b = yleft_col[r + 2];
ypred_ptr[r * y_stride] = (2 * a + (r + 1) * b + (r + 3) / 2) / (r + 3);
}
for (c = 1; c < n / 2 - 1; c++) {
for (r = 0; r < n - 2 - 2 * c; r++) {
if (r & 1)
a = yabove_row[c + 1];
else
a = (yabove_row[c] + yabove_row[c + 1] + 1) >> 1;
b = ypred_ptr[(r + 2) * y_stride + c - 1];
ypred_ptr[r * y_stride + c] = (2 * a + (c + 1) * b + (c + 3) / 2) / (c + 3);
}
}
for (; c < n - 1; ++c) {
for (r = 0; r < n; r++) {
v = (r & 1 ? yabove_row[c + 1] : (yabove_row[c] + yabove_row[c + 1] + 1) >> 1);
w = c - r / 2;
ypred_ptr[r * y_stride + w] = v;
}
}
r = 0;
c = n - 1;
ypred_ptr[c] = (ypred_ptr[(c - 1)] + yabove_row[c] + 1) >> 1;
for (c = n - 2; c >= n / 2; --c) {
h = r + (n - 1 - c) * 2;
ypred_ptr[h * y_stride + c] = (ypred_ptr[h * y_stride + c - 1] +
ypred_ptr[(h - 1) * y_stride + c] + 1) >> 1;
}
for (r = 1; r < n; r++) {
for (c = n - 1; c >= n / 2 + r / 2; --c) {
h = r + (n - 1 - c) * 2;
ypred_ptr[h * y_stride + c] = (ypred_ptr[h * y_stride + c - 1] +
ypred_ptr[(h - 1) * y_stride + c] + 1) >> 1;
}
}
}
static void d45_predictor(uint8_t *ypred_ptr, int y_stride, int n,
uint8_t *yabove_row, uint8_t *yleft_col) {
int r, c;
for (r = 0; r < n - 1; ++r) {
for (c = 0; c <= r; ++c) {
ypred_ptr[(r - c) * y_stride + c] =
(yabove_row[r + 1] * (c + 1) +
yleft_col[r + 1] * (r - c + 1) + r / 2 + 1) / (r + 2);
}
}
for (c = 0; c <= r; ++c) {
int yabove_ext = yabove_row[r]; // clip_pixel(2 * yabove_row[r] -
// yabove_row[r - 1]);
int yleft_ext = yleft_col[r]; // clip_pixel(2 * yleft_col[r] -
// yleft_col[r-1]);
ypred_ptr[(r - c) * y_stride + c] =
(yabove_ext * (c + 1) +
yleft_ext * (r - c + 1) + r / 2 + 1) / (r + 2);
}
for (r = 1; r < n; ++r) {
for (c = n - r; c < n; ++c) {
const int yabove_ext = ypred_ptr[(r - 1) * y_stride + c];
const int yleft_ext = ypred_ptr[r * y_stride + c - 1];
ypred_ptr[r * y_stride + c] = (yabove_ext + yleft_ext + 1) >> 1;
}
}
}
static void d117_predictor(uint8_t *ypred_ptr, int y_stride, int n,
uint8_t *yabove_row, uint8_t *yleft_col) {
int r, c;
for (c = 0; c < n; c++)
ypred_ptr[c] = (yabove_row[c - 1] + yabove_row[c] + 1) >> 1;
ypred_ptr += y_stride;
for (c = 0; c < n; c++)
ypred_ptr[c] = yabove_row[c - 1];
ypred_ptr += y_stride;
for (r = 2; r < n; ++r) {
ypred_ptr[0] = yleft_col[r - 2];
for (c = 1; c < n; c++)
ypred_ptr[c] = ypred_ptr[-2 * y_stride + c - 1];
ypred_ptr += y_stride;
}
}
static void d135_predictor(uint8_t *ypred_ptr, int y_stride, int n,
uint8_t *yabove_row, uint8_t *yleft_col) {
int r, c;
ypred_ptr[0] = yabove_row[-1];
for (c = 1; c < n; c++)
ypred_ptr[c] = yabove_row[c - 1];
for (r = 1; r < n; ++r)
ypred_ptr[r * y_stride] = yleft_col[r - 1];
ypred_ptr += y_stride;
for (r = 1; r < n; ++r) {
for (c = 1; c < n; c++) {
ypred_ptr[c] = ypred_ptr[-y_stride + c - 1];
}
ypred_ptr += y_stride;
}
}
static void d153_predictor(uint8_t *ypred_ptr, int y_stride, int n,
uint8_t *yabove_row, uint8_t *yleft_col) {
int r, c;
ypred_ptr[0] = (yabove_row[-1] + yleft_col[0] + 1) >> 1;
for (r = 1; r < n; r++)
ypred_ptr[r * y_stride] = (yleft_col[r - 1] + yleft_col[r] + 1) >> 1;
ypred_ptr++;
ypred_ptr[0] = yabove_row[-1];
for (r = 1; r < n; r++)
ypred_ptr[r * y_stride] = yleft_col[r - 1];
ypred_ptr++;
for (c = 0; c < n - 2; c++)
ypred_ptr[c] = yabove_row[c];
ypred_ptr += y_stride;
for (r = 1; r < n; ++r) {
for (c = 0; c < n - 2; c++)
ypred_ptr[c] = ypred_ptr[-y_stride + c - 2];
ypred_ptr += y_stride;
}
}
static void corner_predictor(uint8_t *ypred_ptr, int y_stride, int n,
uint8_t *yabove_row,
uint8_t *yleft_col) {
int mh, mv, maxgradh, maxgradv, x, y, nx, ny;
int i, j;
int top_left = yabove_row[-1];
mh = mv = 0;
maxgradh = yabove_row[1] - top_left;
maxgradv = yleft_col[1] - top_left;
for (i = 2; i < n; ++i) {
int gh = yabove_row[i] - yabove_row[i - 2];
int gv = yleft_col[i] - yleft_col[i - 2];
if (gh > maxgradh) {
maxgradh = gh;
mh = i - 1;
}
if (gv > maxgradv) {
maxgradv = gv;
mv = i - 1;
}
}
nx = mh + mv + 3;
ny = 2 * n + 1 - nx;
x = top_left;
for (i = 0; i <= mh; ++i) x += yabove_row[i];
for (i = 0; i <= mv; ++i) x += yleft_col[i];
x += (nx >> 1);
x /= nx;
y = 0;
for (i = mh + 1; i < n; ++i) y += yabove_row[i];
for (i = mv + 1; i < n; ++i) y += yleft_col[i];
y += (ny >> 1);
y /= ny;
for (i = 0; i < n; ++i) {
for (j = 0; j < n; ++j)
ypred_ptr[j] = (i <= mh && j <= mv ? x : y);
ypred_ptr += y_stride;
}
}
void vp9_recon_intra_mbuv(MACROBLOCKD *xd) {
int i;
for (i = 16; i < 24; i += 2) {
BLOCKD *b = &xd->block[i];
vp9_recon2b(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
}
2010-05-18 17:58:33 +02:00
}
void vp9_build_intra_predictors_internal(uint8_t *src, int src_stride,
uint8_t *ypred_ptr,
int y_stride, int mode, int bsize,
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
int up_available, int left_available,
int right_available) {
int r, c, i;
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
uint8_t yleft_col[64], yabove_data[65], ytop_left;
uint8_t *yabove_row = yabove_data + 1;
/*
* 127 127 127 .. 127 127 127 127 127 127
* 129 A B .. Y Z
* 129 C D .. W X
* 129 E F .. U V
* 129 G H .. S T T T T T
* ..
*/
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
if (left_available) {
for (i = 0; i < bsize; i++)
yleft_col[i] = src[i * src_stride - 1];
} else {
vpx_memset(yleft_col, 129, bsize);
}
if (up_available) {
uint8_t *yabove_ptr = src - src_stride;
vpx_memcpy(yabove_row, yabove_ptr, bsize);
if (left_available) {
ytop_left = yabove_ptr[-1];
} else {
ytop_left = 127;
}
} else {
vpx_memset(yabove_row, 127, bsize);
ytop_left = 127;
}
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
yabove_row[-1] = ytop_left;
/* for Y */
switch (mode) {
case DC_PRED: {
int expected_dc;
int i;
int shift;
int average = 0;
int log2_bsize_minus_1;
assert(bsize == 4 || bsize == 8 || bsize == 16 || bsize == 32 ||
bsize == 64);
if (bsize == 4) {
log2_bsize_minus_1 = 1;
} else if (bsize == 8) {
log2_bsize_minus_1 = 2;
} else if (bsize == 16) {
log2_bsize_minus_1 = 3;
} else if (bsize == 32) {
log2_bsize_minus_1 = 4;
} else {
assert(bsize == 64);
log2_bsize_minus_1 = 5;
}
if (up_available || left_available) {
if (up_available) {
for (i = 0; i < bsize; i++) {
average += yabove_row[i];
}
2010-05-18 17:58:33 +02:00
}
if (left_available) {
for (i = 0; i < bsize; i++) {
average += yleft_col[i];
}
2010-05-18 17:58:33 +02:00
}
shift = log2_bsize_minus_1 + up_available + left_available;
expected_dc = (average + (1 << (shift - 1))) >> shift;
} else {
expected_dc = 128;
}
for (r = 0; r < bsize; r++) {
vpx_memset(ypred_ptr, expected_dc, bsize);
ypred_ptr += y_stride;
}
2010-05-18 17:58:33 +02:00
}
break;
case V_PRED: {
for (r = 0; r < bsize; r++) {
memcpy(ypred_ptr, yabove_row, bsize);
ypred_ptr += y_stride;
}
2010-05-18 17:58:33 +02:00
}
break;
case H_PRED: {
for (r = 0; r < bsize; r++) {
vpx_memset(ypred_ptr, yleft_col[r], bsize);
ypred_ptr += y_stride;
}
2010-05-18 17:58:33 +02:00
}
break;
case TM_PRED: {
for (r = 0; r < bsize; r++) {
for (c = 0; c < bsize; c++) {
ypred_ptr[c] = clip_pixel(yleft_col[r] + yabove_row[c] - ytop_left);
2010-05-18 17:58:33 +02:00
}
ypred_ptr += y_stride;
}
2010-05-18 17:58:33 +02:00
}
break;
case D45_PRED: {
d45_predictor(ypred_ptr, y_stride, bsize, yabove_row, yleft_col);
}
break;
case D135_PRED: {
d135_predictor(ypred_ptr, y_stride, bsize, yabove_row, yleft_col);
}
break;
case D117_PRED: {
d117_predictor(ypred_ptr, y_stride, bsize, yabove_row, yleft_col);
}
break;
case D153_PRED: {
d153_predictor(ypred_ptr, y_stride, bsize, yabove_row, yleft_col);
}
break;
case D27_PRED: {
d27_predictor(ypred_ptr, y_stride, bsize, yabove_row, yleft_col);
}
break;
case D63_PRED: {
d63_predictor(ypred_ptr, y_stride, bsize, yabove_row, yleft_col);
}
break;
case I8X8_PRED:
2010-05-18 17:58:33 +02:00
case B_PRED:
case NEARESTMV:
case NEARMV:
case ZEROMV:
case NEWMV:
case SPLITMV:
case MB_MODE_COUNT:
break;
}
2010-05-18 17:58:33 +02:00
}
#if CONFIG_COMP_INTERINTRA_PRED
static void combine_interintra(MB_PREDICTION_MODE mode,
uint8_t *interpred,
int interstride,
uint8_t *intrapred,
int intrastride,
int size) {
// TODO(debargha): Explore different ways of combining predictors
// or designing the tables below
static const int scale_bits = 8;
static const int scale_max = 256; // 1 << scale_bits;
static const int scale_round = 127; // (1 << (scale_bits - 1));
// This table is a function A + B*exp(-kx), where x is hor. index
static const int weights1d[64] = {
128, 125, 122, 119, 116, 114, 111, 109,
107, 105, 103, 101, 99, 97, 96, 94,
93, 91, 90, 89, 88, 86, 85, 84,
83, 82, 81, 81, 80, 79, 78, 78,
77, 76, 76, 75, 75, 74, 74, 73,
73, 72, 72, 71, 71, 71, 70, 70,
70, 70, 69, 69, 69, 69, 68, 68,
68, 68, 68, 67, 67, 67, 67, 67,
};
int size_scale = (size >= 64 ? 1:
size == 32 ? 2 :
size == 16 ? 4 :
size == 8 ? 8 : 16);
int i, j;
switch (mode) {
case V_PRED:
for (i = 0; i < size; ++i) {
for (j = 0; j < size; ++j) {
int k = i * interstride + j;
int scale = weights1d[i * size_scale];
interpred[k] =
((scale_max - scale) * interpred[k] +
scale * intrapred[i * intrastride + j] + scale_round)
>> scale_bits;
}
}
break;
case H_PRED:
for (i = 0; i < size; ++i) {
for (j = 0; j < size; ++j) {
int k = i * interstride + j;
int scale = weights1d[j * size_scale];
interpred[k] =
((scale_max - scale) * interpred[k] +
scale * intrapred[i * intrastride + j] + scale_round)
>> scale_bits;
}
}
break;
case D63_PRED:
case D117_PRED:
for (i = 0; i < size; ++i) {
for (j = 0; j < size; ++j) {
int k = i * interstride + j;
int scale = (weights1d[i * size_scale] * 3 +
weights1d[j * size_scale]) >> 2;
interpred[k] =
((scale_max - scale) * interpred[k] +
scale * intrapred[i * intrastride + j] + scale_round)
>> scale_bits;
}
}
break;
case D27_PRED:
case D153_PRED:
for (i = 0; i < size; ++i) {
for (j = 0; j < size; ++j) {
int k = i * interstride + j;
int scale = (weights1d[j * size_scale] * 3 +
weights1d[i * size_scale]) >> 2;
interpred[k] =
((scale_max - scale) * interpred[k] +
scale * intrapred[i * intrastride + j] + scale_round)
>> scale_bits;
}
}
break;
case D135_PRED:
for (i = 0; i < size; ++i) {
for (j = 0; j < size; ++j) {
int k = i * interstride + j;
int scale = weights1d[(i < j ? i : j) * size_scale];
interpred[k] =
((scale_max - scale) * interpred[k] +
scale * intrapred[i * intrastride + j] + scale_round)
>> scale_bits;
}
}
break;
case D45_PRED:
for (i = 0; i < size; ++i) {
for (j = 0; j < size; ++j) {
int k = i * interstride + j;
int scale = (weights1d[i * size_scale] +
weights1d[j * size_scale]) >> 1;
interpred[k] =
((scale_max - scale) * interpred[k] +
scale * intrapred[i * intrastride + j] + scale_round)
>> scale_bits;
}
}
break;
case TM_PRED:
case DC_PRED:
default:
// simple average
for (i = 0; i < size; ++i) {
for (j = 0; j < size; ++j) {
int k = i * interstride + j;
interpred[k] = (interpred[k] + intrapred[i * intrastride + j]) >> 1;
}
}
break;
}
}
void vp9_build_interintra_16x16_predictors_mb(MACROBLOCKD *xd,
uint8_t *ypred,
uint8_t *upred,
uint8_t *vpred,
int ystride, int uvstride) {
vp9_build_interintra_16x16_predictors_mby(xd, ypred, ystride);
vp9_build_interintra_16x16_predictors_mbuv(xd, upred, vpred, uvstride);
}
void vp9_build_interintra_16x16_predictors_mby(MACROBLOCKD *xd,
uint8_t *ypred,
int ystride) {
uint8_t intrapredictor[256];
vp9_build_intra_predictors_internal(
xd->dst.y_buffer, xd->dst.y_stride,
intrapredictor, 16,
xd->mode_info_context->mbmi.interintra_mode, 16,
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
xd->up_available, xd->left_available, xd->right_available);
combine_interintra(xd->mode_info_context->mbmi.interintra_mode,
ypred, ystride, intrapredictor, 16, 16);
}
void vp9_build_interintra_16x16_predictors_mbuv(MACROBLOCKD *xd,
uint8_t *upred,
uint8_t *vpred,
int uvstride) {
uint8_t uintrapredictor[64];
uint8_t vintrapredictor[64];
vp9_build_intra_predictors_internal(
xd->dst.u_buffer, xd->dst.uv_stride,
uintrapredictor, 8,
xd->mode_info_context->mbmi.interintra_uv_mode, 8,
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
xd->up_available, xd->left_available, xd->right_available);
vp9_build_intra_predictors_internal(
xd->dst.v_buffer, xd->dst.uv_stride,
vintrapredictor, 8,
xd->mode_info_context->mbmi.interintra_uv_mode, 8,
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
xd->up_available, xd->left_available, xd->right_available);
combine_interintra(xd->mode_info_context->mbmi.interintra_uv_mode,
upred, uvstride, uintrapredictor, 8, 8);
combine_interintra(xd->mode_info_context->mbmi.interintra_uv_mode,
vpred, uvstride, vintrapredictor, 8, 8);
}
void vp9_build_interintra_32x32_predictors_sby(MACROBLOCKD *xd,
uint8_t *ypred,
int ystride) {
uint8_t intrapredictor[1024];
vp9_build_intra_predictors_internal(
xd->dst.y_buffer, xd->dst.y_stride,
intrapredictor, 32,
xd->mode_info_context->mbmi.interintra_mode, 32,
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
xd->up_available, xd->left_available, xd->right_available);
combine_interintra(xd->mode_info_context->mbmi.interintra_mode,
ypred, ystride, intrapredictor, 32, 32);
}
void vp9_build_interintra_32x32_predictors_sbuv(MACROBLOCKD *xd,
uint8_t *upred,
uint8_t *vpred,
int uvstride) {
uint8_t uintrapredictor[256];
uint8_t vintrapredictor[256];
vp9_build_intra_predictors_internal(
xd->dst.u_buffer, xd->dst.uv_stride,
uintrapredictor, 16,
xd->mode_info_context->mbmi.interintra_uv_mode, 16,
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
xd->up_available, xd->left_available, xd->right_available);
vp9_build_intra_predictors_internal(
xd->dst.v_buffer, xd->dst.uv_stride,
vintrapredictor, 16,
xd->mode_info_context->mbmi.interintra_uv_mode, 16,
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
xd->up_available, xd->left_available, xd->right_available);
combine_interintra(xd->mode_info_context->mbmi.interintra_uv_mode,
upred, uvstride, uintrapredictor, 16, 16);
combine_interintra(xd->mode_info_context->mbmi.interintra_uv_mode,
vpred, uvstride, vintrapredictor, 16, 16);
}
void vp9_build_interintra_32x32_predictors_sb(MACROBLOCKD *xd,
uint8_t *ypred,
uint8_t *upred,
uint8_t *vpred,
int ystride,
int uvstride) {
vp9_build_interintra_32x32_predictors_sby(xd, ypred, ystride);
vp9_build_interintra_32x32_predictors_sbuv(xd, upred, vpred, uvstride);
}
void vp9_build_interintra_64x64_predictors_sby(MACROBLOCKD *xd,
uint8_t *ypred,
int ystride) {
uint8_t intrapredictor[4096];
const int mode = xd->mode_info_context->mbmi.interintra_mode;
vp9_build_intra_predictors_internal(xd->dst.y_buffer, xd->dst.y_stride,
intrapredictor, 64, mode, 64,
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
xd->up_available, xd->left_available,
xd->right_available);
combine_interintra(xd->mode_info_context->mbmi.interintra_mode,
ypred, ystride, intrapredictor, 64, 64);
}
void vp9_build_interintra_64x64_predictors_sbuv(MACROBLOCKD *xd,
uint8_t *upred,
uint8_t *vpred,
int uvstride) {
uint8_t uintrapredictor[1024];
uint8_t vintrapredictor[1024];
const int mode = xd->mode_info_context->mbmi.interintra_uv_mode;
vp9_build_intra_predictors_internal(xd->dst.u_buffer, xd->dst.uv_stride,
uintrapredictor, 32, mode, 32,
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
xd->up_available, xd->left_available,
xd->right_available);
vp9_build_intra_predictors_internal(xd->dst.v_buffer, xd->dst.uv_stride,
vintrapredictor, 32, mode, 32,
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
xd->up_available, xd->left_available,
xd->right_available);
combine_interintra(xd->mode_info_context->mbmi.interintra_uv_mode,
upred, uvstride, uintrapredictor, 32, 32);
combine_interintra(xd->mode_info_context->mbmi.interintra_uv_mode,
vpred, uvstride, vintrapredictor, 32, 32);
}
void vp9_build_interintra_64x64_predictors_sb(MACROBLOCKD *xd,
uint8_t *ypred,
uint8_t *upred,
uint8_t *vpred,
int ystride,
int uvstride) {
vp9_build_interintra_64x64_predictors_sby(xd, ypred, ystride);
vp9_build_interintra_64x64_predictors_sbuv(xd, upred, vpred, uvstride);
}
#endif // CONFIG_COMP_INTERINTRA_PRED
void vp9_build_intra_predictors_mby(MACROBLOCKD *xd) {
vp9_build_intra_predictors_internal(xd->dst.y_buffer, xd->dst.y_stride,
xd->predictor, 16,
xd->mode_info_context->mbmi.mode, 16,
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
xd->up_available, xd->left_available,
xd->right_available);
}
2010-05-18 17:58:33 +02:00
void vp9_build_intra_predictors_mby_s(MACROBLOCKD *xd) {
vp9_build_intra_predictors_internal(xd->dst.y_buffer, xd->dst.y_stride,
xd->dst.y_buffer, xd->dst.y_stride,
xd->mode_info_context->mbmi.mode, 16,
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
xd->up_available, xd->left_available,
xd->right_available);
}
void vp9_build_intra_predictors_sby_s(MACROBLOCKD *xd) {
vp9_build_intra_predictors_internal(xd->dst.y_buffer, xd->dst.y_stride,
xd->dst.y_buffer, xd->dst.y_stride,
xd->mode_info_context->mbmi.mode, 32,
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
xd->up_available, xd->left_available,
xd->right_available);
}
void vp9_build_intra_predictors_sb64y_s(MACROBLOCKD *xd) {
vp9_build_intra_predictors_internal(xd->dst.y_buffer, xd->dst.y_stride,
xd->dst.y_buffer, xd->dst.y_stride,
xd->mode_info_context->mbmi.mode, 64,
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
xd->up_available, xd->left_available,
xd->right_available);
}
2010-05-18 17:58:33 +02:00
void vp9_build_intra_predictors_mbuv_internal(MACROBLOCKD *xd,
uint8_t *upred_ptr,
uint8_t *vpred_ptr,
int uv_stride,
int mode, int bsize) {
vp9_build_intra_predictors_internal(xd->dst.u_buffer, xd->dst.uv_stride,
upred_ptr, uv_stride, mode, bsize,
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
xd->up_available, xd->left_available,
xd->right_available);
vp9_build_intra_predictors_internal(xd->dst.v_buffer, xd->dst.uv_stride,
vpred_ptr, uv_stride, mode, bsize,
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
xd->up_available, xd->left_available,
xd->right_available);
2010-05-18 17:58:33 +02:00
}
void vp9_build_intra_predictors_mbuv(MACROBLOCKD *xd) {
vp9_build_intra_predictors_mbuv_internal(xd, &xd->predictor[256],
&xd->predictor[320], 8,
xd->mode_info_context->mbmi.uv_mode,
8);
}
void vp9_build_intra_predictors_mbuv_s(MACROBLOCKD *xd) {
vp9_build_intra_predictors_mbuv_internal(xd, xd->dst.u_buffer,
xd->dst.v_buffer,
xd->dst.uv_stride,
xd->mode_info_context->mbmi.uv_mode,
8);
}
void vp9_build_intra_predictors_sbuv_s(MACROBLOCKD *xd) {
vp9_build_intra_predictors_mbuv_internal(xd, xd->dst.u_buffer,
xd->dst.v_buffer, xd->dst.uv_stride,
xd->mode_info_context->mbmi.uv_mode,
16);
}
void vp9_build_intra_predictors_sb64uv_s(MACROBLOCKD *xd) {
vp9_build_intra_predictors_mbuv_internal(xd, xd->dst.u_buffer,
xd->dst.v_buffer, xd->dst.uv_stride,
xd->mode_info_context->mbmi.uv_mode,
32);
}
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
void vp9_intra8x8_predict(MACROBLOCKD *xd,
BLOCKD *b,
int mode,
uint8_t *predictor) {
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
const int block4x4_idx = (b - xd->block);
const int block_idx = (block4x4_idx >> 2) | !!(block4x4_idx & 2);
const int have_top = (block_idx >> 1) || xd->up_available;
const int have_left = (block_idx & 1) || xd->left_available;
const int have_right = !(block_idx & 1) || xd->right_available;
vp9_build_intra_predictors_internal(*(b->base_dst) + b->dst,
b->dst_stride, predictor, 16,
mode, 8, have_top, have_left,
have_right);
}
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
void vp9_intra_uv4x4_predict(MACROBLOCKD *xd,
BLOCKD *b,
int mode,
uint8_t *predictor) {
[WIP] Add column-based tiling. This patch adds column-based tiling. The idea is to make each tile independently decodable (after reading the common frame header) and also independendly encodable (minus within-frame cost adjustments in the RD loop) to speed-up hardware & software en/decoders if they used multi-threading. Column-based tiling has the added advantage (over other tiling methods) that it minimizes realtime use-case latency, since all threads can start encoding data as soon as the first SB-row worth of data is available to the encoder. There is some test code that does random tile ordering in the decoder, to confirm that each tile is indeed independently decodable from other tiles in the same frame. At tile edges, all contexts assume default values (i.e. 0, 0 motion vector, no coefficients, DC intra4x4 mode), and motion vector search and ordering do not cross tiles in the same frame. t log Tile independence is not maintained between frames ATM, i.e. tile 0 of frame 1 is free to use motion vectors that point into any tile of frame 0. We support 1 (i.e. no tiling), 2 or 4 column-tiles. The loopfilter crosses tile boundaries. I discussed this briefly with Aki and he says that's OK. An in-loop loopfilter would need to do some sync between tile threads, but that shouldn't be a big issue. Resuls: with tiling disabled, we go up slightly because of improved edge use in the intra4x4 prediction. With 2 tiles, we lose about ~1% on derf, ~0.35% on HD and ~0.55% on STD/HD. With 4 tiles, we lose another ~1.5% on derf ~0.77% on HD and ~0.85% on STD/HD. Most of this loss is concentrated in the low-bitrate end of clips, and most of it is because of the loss of edges at tile boundaries and the resulting loss of intra predictors. TODO: - more tiles (perhaps allow row-based tiling also, and max. 8 tiles)? - maybe optionally (for EC purposes), motion vectors themselves should not cross tile edges, or we should emulate such borders as if they were off-frame, to limit error propagation to within one tile only. This doesn't have to be the default behaviour but could be an optional bitstream flag. Change-Id: I5951c3a0742a767b20bc9fb5af685d9892c2c96f
2013-02-01 18:35:28 +01:00
const int block_idx = (b - xd->block) & 3;
const int have_top = (block_idx >> 1) || xd->up_available;
const int have_left = (block_idx & 1) || xd->left_available;
const int have_right = !(block_idx & 1) || xd->right_available;
vp9_build_intra_predictors_internal(*(b->base_dst) + b->dst,
b->dst_stride, predictor, 8,
mode, 4, have_top, have_left,
have_right);
}
/* TODO: try different ways of use Y-UV mode correlation
Current code assumes that a uv 4x4 block use same mode
as corresponding Y 8x8 area
*/