2010-05-18 17:58:33 +02:00
|
|
|
/*
|
2010-09-09 14:16:39 +02:00
|
|
|
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
|
2010-05-18 17:58:33 +02:00
|
|
|
*
|
2010-06-18 18:39:21 +02:00
|
|
|
* Use of this source code is governed by a BSD-style license
|
2010-06-04 22:19:40 +02:00
|
|
|
* that can be found in the LICENSE file in the root of the source
|
|
|
|
* tree. An additional intellectual property rights grant can be found
|
2010-06-18 18:39:21 +02:00
|
|
|
* in the file PATENTS. All contributing project authors may
|
2010-06-04 22:19:40 +02:00
|
|
|
* be found in the AUTHORS file in the root of the source tree.
|
2010-05-18 17:58:33 +02:00
|
|
|
*/
|
|
|
|
|
2012-11-30 01:36:10 +01:00
|
|
|
#ifndef VP9_COMMON_VP9_RECONINTER_H_
|
|
|
|
#define VP9_COMMON_VP9_RECONINTER_H_
|
2010-05-18 17:58:33 +02:00
|
|
|
|
2012-12-19 00:31:19 +01:00
|
|
|
#include "vpx/vpx_integer.h"
|
2012-11-28 19:41:40 +01:00
|
|
|
#include "vp9/common/vp9_onyxc_int.h"
|
2012-07-18 22:43:01 +02:00
|
|
|
|
Convert subpixel filters to use convolve framework
Update the code to call the new convolution functions to do subpixel
prediction rather than the existing functions. Remove the old C and
assembly code, since it is unused. This causes a 50% performance
reduction on the decoder, but that will be resolved when the asm for
the new functions is available.
There is no consensus for whether 6-tap or 2-tap predictors will be
supported in the final codec, so these filters are implemented in
terms of the 8-tap code, so that quality testing of these modes
can continue. Implementing the lower complexity algorithms is a
simple exercise, should it be necessary.
This code produces slightly better results in the EIGHTTAP_SMOOTH
case, since the filter is now applied in only one direction when
the subpel motion is only in one direction. Like the previous code,
the filtering is skipped entirely on full-pel MVs. This combination
seems to give the best quality gains, but this may be indicative of a
bug in the encoder's filter selection, since the encoder could
achieve the result of skipping the filtering on full-pel by selecting
one of the other filters. This should be revisited.
Quality gains on derf positive on almost all clips. The only clip
that seemed to be hurt at all datarates was football
(-0.115% PSNR average, -0.587% min). Overall averages 0.375% PSNR,
0.347% SSIM.
Change-Id: I7d469716091b1d89b4b08adde5863999319d69ff
2013-01-29 01:59:03 +01:00
|
|
|
struct subpix_fn_table;
|
|
|
|
|
2013-04-17 22:41:18 +02:00
|
|
|
void vp9_build_inter_predictors_sby(MACROBLOCKD *xd,
|
2013-04-15 18:31:27 +02:00
|
|
|
int mb_row,
|
|
|
|
int mb_col,
|
|
|
|
BLOCK_SIZE_TYPE bsize);
|
|
|
|
|
2013-04-17 22:41:18 +02:00
|
|
|
void vp9_build_inter_predictors_sbuv(MACROBLOCKD *xd,
|
2013-04-15 18:31:27 +02:00
|
|
|
int mb_row,
|
|
|
|
int mb_col,
|
|
|
|
BLOCK_SIZE_TYPE bsize);
|
2013-04-19 19:45:50 +02:00
|
|
|
|
2013-04-12 20:25:53 +02:00
|
|
|
void vp9_build_inter_predictors_sb(MACROBLOCKD *mb,
|
|
|
|
int mb_row, int mb_col,
|
|
|
|
BLOCK_SIZE_TYPE bsize);
|
2012-08-21 02:45:36 +02:00
|
|
|
|
2013-02-27 00:52:05 +01:00
|
|
|
void vp9_setup_interp_filters(MACROBLOCKD *xd,
|
|
|
|
INTERPOLATIONFILTERTYPE filter,
|
|
|
|
VP9_COMMON *cm);
|
2010-05-18 17:58:33 +02:00
|
|
|
|
Spatial resamping of ZEROMV predictors
This patch allows coding frames using references of different
resolution, in ZEROMV mode. For compound prediction, either
reference may be scaled.
To test, I use the resize_test and enable WRITE_RECON_BUFFER
in vp9_onyxd_if.c. It's also useful to apply this patch to
test/i420_video_source.h:
--- a/test/i420_video_source.h
+++ b/test/i420_video_source.h
@@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource {
virtual void FillFrame() {
// Read a frame from input_file.
+ if (frame_ != 3)
if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) {
limit_ = frame_;
}
This forces the frame that the resolution changes on to be coded
with no motion, only scaling, and improves the quality of the
result.
Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
|
|
|
void vp9_setup_scale_factors_for_frame(struct scale_factors *scale,
|
2013-05-08 23:11:47 +02:00
|
|
|
int other_w, int other_h,
|
Spatial resamping of ZEROMV predictors
This patch allows coding frames using references of different
resolution, in ZEROMV mode. For compound prediction, either
reference may be scaled.
To test, I use the resize_test and enable WRITE_RECON_BUFFER
in vp9_onyxd_if.c. It's also useful to apply this patch to
test/i420_video_source.h:
--- a/test/i420_video_source.h
+++ b/test/i420_video_source.h
@@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource {
virtual void FillFrame() {
// Read a frame from input_file.
+ if (frame_ != 3)
if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) {
limit_ = frame_;
}
This forces the frame that the resolution changes on to be coded
with no motion, only scaling, and improves the quality of the
result.
Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
|
|
|
int this_w, int this_h);
|
|
|
|
|
2013-02-09 02:49:44 +01:00
|
|
|
void vp9_build_inter_predictor(const uint8_t *src, int src_stride,
|
|
|
|
uint8_t *dst, int dst_stride,
|
|
|
|
const int_mv *mv_q3,
|
|
|
|
const struct scale_factors *scale,
|
|
|
|
int w, int h, int do_avg,
|
|
|
|
const struct subpix_fn_table *subpix);
|
|
|
|
|
|
|
|
void vp9_build_inter_predictor_q4(const uint8_t *src, int src_stride,
|
|
|
|
uint8_t *dst, int dst_stride,
|
2013-04-15 22:18:24 +02:00
|
|
|
const int_mv *mv_q4,
|
2013-02-09 02:49:44 +01:00
|
|
|
const struct scale_factors *scale,
|
|
|
|
int w, int h, int do_avg,
|
|
|
|
const struct subpix_fn_table *subpix);
|
2013-02-27 00:52:05 +01:00
|
|
|
|
2013-04-04 18:56:02 +02:00
|
|
|
static int scale_value_x_with_scaling(int val,
|
|
|
|
const struct scale_factors *scale) {
|
Spatial resamping of ZEROMV predictors
This patch allows coding frames using references of different
resolution, in ZEROMV mode. For compound prediction, either
reference may be scaled.
To test, I use the resize_test and enable WRITE_RECON_BUFFER
in vp9_onyxd_if.c. It's also useful to apply this patch to
test/i420_video_source.h:
--- a/test/i420_video_source.h
+++ b/test/i420_video_source.h
@@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource {
virtual void FillFrame() {
// Read a frame from input_file.
+ if (frame_ != 3)
if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) {
limit_ = frame_;
}
This forces the frame that the resolution changes on to be coded
with no motion, only scaling, and improves the quality of the
result.
Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
|
|
|
return val * scale->x_num / scale->x_den;
|
|
|
|
}
|
|
|
|
|
2013-04-04 18:56:02 +02:00
|
|
|
static int scale_value_y_with_scaling(int val,
|
|
|
|
const struct scale_factors *scale) {
|
Spatial resamping of ZEROMV predictors
This patch allows coding frames using references of different
resolution, in ZEROMV mode. For compound prediction, either
reference may be scaled.
To test, I use the resize_test and enable WRITE_RECON_BUFFER
in vp9_onyxd_if.c. It's also useful to apply this patch to
test/i420_video_source.h:
--- a/test/i420_video_source.h
+++ b/test/i420_video_source.h
@@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource {
virtual void FillFrame() {
// Read a frame from input_file.
+ if (frame_ != 3)
if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) {
limit_ = frame_;
}
This forces the frame that the resolution changes on to be coded
with no motion, only scaling, and improves the quality of the
result.
Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
|
|
|
return val * scale->y_num / scale->y_den;
|
|
|
|
}
|
|
|
|
|
2013-04-04 18:56:02 +02:00
|
|
|
static int unscaled_value(int val, const struct scale_factors *scale) {
|
|
|
|
(void) scale;
|
|
|
|
return val;
|
|
|
|
}
|
|
|
|
|
2013-05-08 23:11:47 +02:00
|
|
|
static int scaled_buffer_offset(int x_offset, int y_offset, int stride,
|
Spatial resamping of ZEROMV predictors
This patch allows coding frames using references of different
resolution, in ZEROMV mode. For compound prediction, either
reference may be scaled.
To test, I use the resize_test and enable WRITE_RECON_BUFFER
in vp9_onyxd_if.c. It's also useful to apply this patch to
test/i420_video_source.h:
--- a/test/i420_video_source.h
+++ b/test/i420_video_source.h
@@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource {
virtual void FillFrame() {
// Read a frame from input_file.
+ if (frame_ != 3)
if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) {
limit_ = frame_;
}
This forces the frame that the resolution changes on to be coded
with no motion, only scaling, and improves the quality of the
result.
Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
|
|
|
const struct scale_factors *scale) {
|
2013-05-08 23:11:47 +02:00
|
|
|
const int x = scale ? scale->scale_value_x(x_offset, scale) : x_offset;
|
|
|
|
const int y = scale ? scale->scale_value_y(y_offset, scale) : y_offset;
|
|
|
|
return y * stride + x;
|
2013-04-20 00:52:17 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
static void setup_pred_plane(struct buf_2d *dst,
|
|
|
|
uint8_t *src, int stride,
|
2013-04-26 20:57:17 +02:00
|
|
|
int mi_row, int mi_col,
|
2013-04-20 00:52:17 +02:00
|
|
|
const struct scale_factors *scale,
|
|
|
|
int subsampling_x, int subsampling_y) {
|
2013-04-26 20:57:17 +02:00
|
|
|
const int x = (MI_SIZE * mi_col) >> subsampling_x;
|
|
|
|
const int y = (MI_SIZE * mi_row) >> subsampling_y;
|
2013-04-20 00:52:17 +02:00
|
|
|
dst->buf = src + scaled_buffer_offset(x, y, stride, scale);
|
|
|
|
dst->stride = stride;
|
|
|
|
}
|
|
|
|
|
|
|
|
// TODO(jkoleszar): audit all uses of this that don't set mb_row, mb_col
|
|
|
|
static void setup_dst_planes(MACROBLOCKD *xd,
|
|
|
|
const YV12_BUFFER_CONFIG *src,
|
2013-04-26 20:57:17 +02:00
|
|
|
int mi_row, int mi_col) {
|
2013-05-08 22:39:16 +02:00
|
|
|
uint8_t *buffers[3] = {src->y_buffer, src->u_buffer, src->v_buffer};
|
|
|
|
int strides[3] = {src->y_stride, src->uv_stride, src->uv_stride};
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < MAX_MB_PLANE; ++i) {
|
|
|
|
struct macroblockd_plane *pd = &xd->plane[i];
|
|
|
|
setup_pred_plane(&pd->dst, buffers[i], strides[i], mi_row, mi_col, NULL,
|
|
|
|
pd->subsampling_x, pd->subsampling_y);
|
|
|
|
}
|
Spatial resamping of ZEROMV predictors
This patch allows coding frames using references of different
resolution, in ZEROMV mode. For compound prediction, either
reference may be scaled.
To test, I use the resize_test and enable WRITE_RECON_BUFFER
in vp9_onyxd_if.c. It's also useful to apply this patch to
test/i420_video_source.h:
--- a/test/i420_video_source.h
+++ b/test/i420_video_source.h
@@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource {
virtual void FillFrame() {
// Read a frame from input_file.
+ if (frame_ != 3)
if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) {
limit_ = frame_;
}
This forces the frame that the resolution changes on to be coded
with no motion, only scaling, and improves the quality of the
result.
Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
|
|
|
}
|
|
|
|
|
2013-04-20 04:16:14 +02:00
|
|
|
static void setup_pre_planes(MACROBLOCKD *xd,
|
|
|
|
const YV12_BUFFER_CONFIG *src0,
|
|
|
|
const YV12_BUFFER_CONFIG *src1,
|
2013-04-26 20:57:17 +02:00
|
|
|
int mi_row, int mi_col,
|
2013-04-20 04:16:14 +02:00
|
|
|
const struct scale_factors *scale,
|
|
|
|
const struct scale_factors *scale_uv) {
|
2013-05-08 22:39:16 +02:00
|
|
|
const YV12_BUFFER_CONFIG *srcs[2] = {src0, src1};
|
|
|
|
int i, j;
|
|
|
|
|
|
|
|
for (i = 0; i < 2; ++i) {
|
|
|
|
const YV12_BUFFER_CONFIG *src = srcs[i];
|
|
|
|
if (src) {
|
|
|
|
uint8_t* buffers[3] = {src->y_buffer, src->u_buffer, src->v_buffer};
|
|
|
|
int strides[3] = {src->y_stride, src->uv_stride, src->uv_stride};
|
|
|
|
|
|
|
|
for (j = 0; j < MAX_MB_PLANE; ++j) {
|
|
|
|
struct macroblockd_plane *pd = &xd->plane[j];
|
|
|
|
const struct scale_factors *sf = j ? scale_uv : scale;
|
|
|
|
setup_pred_plane(&pd->pre[i],
|
|
|
|
buffers[j], strides[j],
|
|
|
|
mi_row, mi_col, sf ? &sf[i] : NULL,
|
|
|
|
pd->subsampling_x, pd->subsampling_y);
|
|
|
|
}
|
|
|
|
}
|
2013-04-20 04:16:14 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
Spatial resamping of ZEROMV predictors
This patch allows coding frames using references of different
resolution, in ZEROMV mode. For compound prediction, either
reference may be scaled.
To test, I use the resize_test and enable WRITE_RECON_BUFFER
in vp9_onyxd_if.c. It's also useful to apply this patch to
test/i420_video_source.h:
--- a/test/i420_video_source.h
+++ b/test/i420_video_source.h
@@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource {
virtual void FillFrame() {
// Read a frame from input_file.
+ if (frame_ != 3)
if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) {
limit_ = frame_;
}
This forces the frame that the resolution changes on to be coded
with no motion, only scaling, and improves the quality of the
result.
Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496
2013-02-25 05:55:14 +01:00
|
|
|
static void set_scale_factors(MACROBLOCKD *xd,
|
|
|
|
int ref0, int ref1,
|
|
|
|
struct scale_factors scale_factor[MAX_REF_FRAMES]) {
|
|
|
|
|
|
|
|
xd->scale_factor[0] = scale_factor[ref0 >= 0 ? ref0 : 0];
|
|
|
|
xd->scale_factor[1] = scale_factor[ref1 >= 0 ? ref1 : 0];
|
|
|
|
xd->scale_factor_uv[0] = xd->scale_factor[0];
|
|
|
|
xd->scale_factor_uv[1] = xd->scale_factor[1];
|
|
|
|
}
|
|
|
|
|
2013-04-04 18:56:02 +02:00
|
|
|
static void set_offsets_with_scaling(struct scale_factors *scale,
|
|
|
|
int row, int col) {
|
|
|
|
const int x_q4 = 16 * col;
|
|
|
|
const int y_q4 = 16 * row;
|
|
|
|
|
|
|
|
scale->x_offset_q4 = (x_q4 * scale->x_num / scale->x_den) & 0xf;
|
|
|
|
scale->y_offset_q4 = (y_q4 * scale->y_num / scale->y_den) & 0xf;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void set_offsets_without_scaling(struct scale_factors *scale,
|
|
|
|
int row, int col) {
|
|
|
|
scale->x_offset_q4 = 0;
|
|
|
|
scale->y_offset_q4 = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int_mv32 motion_vector_q3_to_q4_with_scaling(
|
|
|
|
const int_mv *src_mv,
|
|
|
|
const struct scale_factors *scale) {
|
|
|
|
// returns mv * scale + offset
|
|
|
|
int_mv32 result;
|
|
|
|
const int32_t mv_row_q4 = src_mv->as_mv.row << 1;
|
|
|
|
const int32_t mv_col_q4 = src_mv->as_mv.col << 1;
|
|
|
|
|
|
|
|
/* TODO(jkoleszar): make fixed point, or as a second multiply? */
|
|
|
|
result.as_mv.row = mv_row_q4 * scale->y_num / scale->y_den
|
|
|
|
+ scale->y_offset_q4;
|
|
|
|
result.as_mv.col = mv_col_q4 * scale->x_num / scale->x_den
|
|
|
|
+ scale->x_offset_q4;
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int_mv32 motion_vector_q3_to_q4_without_scaling(
|
|
|
|
const int_mv *src_mv,
|
|
|
|
const struct scale_factors *scale) {
|
|
|
|
// returns mv * scale + offset
|
|
|
|
int_mv32 result;
|
|
|
|
|
|
|
|
result.as_mv.row = src_mv->as_mv.row << 1;
|
|
|
|
result.as_mv.col = src_mv->as_mv.col << 1;
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int32_t motion_vector_component_q4_with_scaling(int mv_q4,
|
|
|
|
int num,
|
|
|
|
int den,
|
|
|
|
int offset_q4) {
|
|
|
|
// returns the scaled and offset value of the mv component.
|
|
|
|
|
|
|
|
/* TODO(jkoleszar): make fixed point, or as a second multiply? */
|
|
|
|
return mv_q4 * num / den + offset_q4;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int32_t motion_vector_component_q4_without_scaling(int mv_q4,
|
|
|
|
int num,
|
|
|
|
int den,
|
|
|
|
int offset_q4) {
|
|
|
|
// returns the scaled and offset value of the mv component.
|
|
|
|
(void)num;
|
|
|
|
(void)den;
|
|
|
|
(void)offset_q4;
|
|
|
|
return mv_q4;
|
|
|
|
}
|
2012-12-19 00:31:19 +01:00
|
|
|
#endif // VP9_COMMON_VP9_RECONINTER_H_
|