mozilla
/
aom
зеркало из
1
0
Форкнуть 0
Код Задачи Пакеты Проекты Релизы Вики Активность

Reformatting the deringing code 

Manally removed the "clang-format off" lines. The rest is done by clang

Change-Id: I88a2028b55a541729b4e8896cdf66b544e9898bb
This commit is contained in:
Jean-Marc Valin 2016-11-02 02:33:46 -04:00 коммит произвёл Yaowu Xu
Родитель e254241ce7
Коммит 39d92a071d
5 изменённых файлов: 101 добавлений и 112 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

88
av1/common/dering.c
Просмотреть файл

@ -9,8 +9,6 @@
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
// clang-format off
#include <string.h>
#include <math.h>
@ -72,39 +70,37 @@ int sb_compute_dering_list(const AV1_COMMON *const cm, int mi_row, int mi_col,
return count;
}
static INLINE void copy_8x8_16bit_to_8bit(uint8_t *dst, int dstride, int16_t *src, int sstride) {
static INLINE void copy_8x8_16bit_to_8bit(uint8_t *dst, int dstride,
int16_t *src, int sstride) {
int i, j;
for (i = 0; i < 8; i++)
for (j = 0; j < 8; j++)
dst[i * dstride + j] = src[i * sstride + j];
for (j = 0; j < 8; j++) dst[i * dstride + j] = src[i * sstride + j];
}
static INLINE void copy_4x4_16bit_to_8bit(uint8_t *dst, int dstride, int16_t *src, int sstride) {
static INLINE void copy_4x4_16bit_to_8bit(uint8_t *dst, int dstride,
int16_t *src, int sstride) {
int i, j;
for (i = 0; i < 4; i++)
for (j = 0; j < 4; j++)
dst[i * dstride + j] = src[i * sstride + j];
for (j = 0; j < 4; j++) dst[i * dstride + j] = src[i * sstride + j];
}
/* TODO: Optimize this function for SSE. */
void copy_dering_16bit_to_8bit(uint8_t *dst, int dstride, int16_t *src,
dering_list *dlist, int dering_count, int bsize)
{
dering_list *dlist, int dering_count,
int bsize) {
int bi, bx, by;
if (bsize == 3) {
for (bi = 0; bi < dering_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
copy_8x8_16bit_to_8bit(&dst[(by << 3) * dstride + (bx << 3)],
dstride,
copy_8x8_16bit_to_8bit(&dst[(by << 3) * dstride + (bx << 3)], dstride,
&src[bi << 2 * bsize], 1 << bsize);
}
} else {
for (bi = 0; bi < dering_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
copy_4x4_16bit_to_8bit(&dst[(by << 2) * dstride + (bx << 2)],
dstride,
copy_4x4_16bit_to_8bit(&dst[(by << 2) * dstride + (bx << 2)], dstride,
&src[bi << 2 * bsize], 1 << bsize);
}
}
@ -112,15 +108,14 @@ void copy_dering_16bit_to_8bit(uint8_t *dst, int dstride, int16_t *src,
/* TODO: Optimize this function for SSE. */
static void copy_sb8_16(AV1_COMMON *cm, int16_t *dst, int dstride,
const uint8_t *src, int src_voffset, int src_hoffset, int sstride,
int vsize, int hsize)
{
const uint8_t *src, int src_voffset, int src_hoffset,
int sstride, int vsize, int hsize) {
int r, c;
(void)cm;
#if CONFIG_AOM_HIGHBITDEPTH
if (cm->use_highbitdepth) {
const uint16_t *base = &CONVERT_TO_SHORTPTR(src)[src_voffset * sstride
+ src_hoffset];
const uint16_t *base =
&CONVERT_TO_SHORTPTR(src)[src_voffset * sstride + src_hoffset];
for (r = 0; r < vsize; r++) {
for (c = 0; c < hsize; c++) {
dst[r * dstride + c] = base[r * sstride + c];
@ -190,8 +185,7 @@ void av1_dering_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
int level;
int nhb, nvb;
int cstart = 0;
if (!dering_left)
cstart = -OD_FILT_HBORDER;
if (!dering_left) cstart = -OD_FILT_HBORDER;
nhb = AOMMIN(MAX_MIB_SIZE, cm->mi_cols - MAX_MIB_SIZE * sbc);
nvb = AOMMIN(MAX_MIB_SIZE, cm->mi_rows - MAX_MIB_SIZE * sbr);
level = compute_level_from_index(
@ -200,8 +194,8 @@ void av1_dering_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
->mbmi.dering_gain);
curr_row_dering[sbc] = 0;
if (level == 0 ||
(dering_count = sb_compute_dering_list(cm, sbr * MAX_MIB_SIZE,
sbc * MAX_MIB_SIZE, dlist)) == 0) {
(dering_count = sb_compute_dering_list(
cm, sbr * MAX_MIB_SIZE, sbc * MAX_MIB_SIZE, dlist)) == 0) {
dering_left = 0;
continue;
}
@ -225,8 +219,8 @@ void av1_dering_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
OD_DERING_VERY_LARGE to avoid filtering with the outside. */
for (r = 0; r < rend + OD_FILT_VBORDER; r++) {
for (c = cend; c < (nhb << bsize[pli]) + OD_FILT_HBORDER; ++c) {
src[r * OD_FILT_BSTRIDE + c + OD_FILT_HBORDER]
= OD_DERING_VERY_LARGE;
src[r * OD_FILT_BSTRIDE + c + OD_FILT_HBORDER] =
OD_DERING_VERY_LARGE;
}
}
}
@ -242,15 +236,18 @@ void av1_dering_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
}
/* Copy in the pixels we need from the current superblock for
deringing.*/
copy_sb8_16(cm, &src[OD_FILT_VBORDER*OD_FILT_BSTRIDE + OD_FILT_HBORDER
+ cstart], OD_FILT_BSTRIDE, xd->plane[pli].dst.buf,
copy_sb8_16(
cm,
&src[OD_FILT_VBORDER * OD_FILT_BSTRIDE + OD_FILT_HBORDER + cstart],
OD_FILT_BSTRIDE, xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << bsize[pli]) * sbr, coffset + cstart,
xd->plane[pli].dst.stride, rend, cend - cstart);
if (!prev_row_dering[sbc]) {
copy_sb8_16(cm, &src[OD_FILT_HBORDER], OD_FILT_BSTRIDE,
xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << bsize[pli]) * sbr - OD_FILT_VBORDER, coffset,
xd->plane[pli].dst.stride, OD_FILT_VBORDER, nhb << bsize[pli]);
(MAX_MIB_SIZE << bsize[pli]) * sbr - OD_FILT_VBORDER,
coffset, xd->plane[pli].dst.stride, OD_FILT_VBORDER,
nhb << bsize[pli]);
} else if (sbr > 0) {
for (r = 0; r < OD_FILT_VBORDER; r++) {
for (c = 0; c < nhb << bsize[pli]; c++) {
@ -267,11 +264,10 @@ void av1_dering_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
}
}
if (!prev_row_dering[sbc - 1]) {
copy_sb8_16(cm, src, OD_FILT_BSTRIDE,
xd->plane[pli].dst.buf,
copy_sb8_16(cm, src, OD_FILT_BSTRIDE, xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << bsize[pli]) * sbr - OD_FILT_VBORDER,
coffset - OD_FILT_HBORDER,
xd->plane[pli].dst.stride, OD_FILT_VBORDER, OD_FILT_HBORDER);
coffset - OD_FILT_HBORDER, xd->plane[pli].dst.stride,
OD_FILT_VBORDER, OD_FILT_HBORDER);
} else if (sbr > 0 && sbc > 0) {
for (r = 0; r < OD_FILT_VBORDER; r++) {
for (c = -OD_FILT_HBORDER; c < 0; c++) {
@ -291,8 +287,8 @@ void av1_dering_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
copy_sb8_16(cm, &src[OD_FILT_HBORDER + (nhb << bsize[pli])],
OD_FILT_BSTRIDE, xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << bsize[pli]) * sbr - OD_FILT_VBORDER,
coffset + (nhb << bsize[pli]),
xd->plane[pli].dst.stride, OD_FILT_VBORDER, OD_FILT_HBORDER);
coffset + (nhb << bsize[pli]), xd->plane[pli].dst.stride,
OD_FILT_VBORDER, OD_FILT_HBORDER);
} else if (sbr > 0 && sbc < nhsb - 1) {
for (r = 0; r < OD_FILT_VBORDER; r++) {
for (c = nhb << bsize[pli];
@ -323,13 +319,13 @@ void av1_dering_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
for (c = 0; c < OD_FILT_HBORDER; c++) {
/* Saving pixels in case we need to dering the superblock on the
right. */
colbuf[pli][r][c] = src[r * OD_FILT_BSTRIDE + c
+ (nhb << bsize[pli])];
colbuf[pli][r][c] =
src[r * OD_FILT_BSTRIDE + c + (nhb << bsize[pli])];
}
}
copy_sb8_16(cm, &linebuf[pli][coffset], stride, xd->plane[pli].dst.buf,
(MAX_MIB_SIZE << bsize[pli]) * (sbr + 1) - OD_FILT_VBORDER, coffset,
xd->plane[pli].dst.stride, OD_FILT_VBORDER,
(MAX_MIB_SIZE << bsize[pli]) * (sbr + 1) - OD_FILT_VBORDER,
coffset, xd->plane[pli].dst.stride, OD_FILT_VBORDER,
(nhb << bsize[pli]));
/* FIXME: This is a temporary hack that uses more conservative
@ -339,26 +335,26 @@ void av1_dering_frame(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm,
else
threshold = level << coeff_shift;
if (threshold == 0) continue;
od_dering(dst, &src[OD_FILT_VBORDER * OD_FILT_BSTRIDE
+ OD_FILT_HBORDER],
od_dering(
dst, &src[OD_FILT_VBORDER * OD_FILT_BSTRIDE + OD_FILT_HBORDER],
dec[pli], dir, pli, dlist, dering_count, threshold, coeff_shift);
#if CONFIG_AOM_HIGHBITDEPTH
if (cm->use_highbitdepth) {
copy_dering_16bit_to_16bit(
(int16_t *)&CONVERT_TO_SHORTPTR(
xd->plane[pli].dst.buf)[xd->plane[pli].dst.stride *
xd->plane[pli]
.dst.buf)[xd->plane[pli].dst.stride *
(MAX_MIB_SIZE * sbr << bsize[pli]) +
(sbc * MAX_MIB_SIZE << bsize[pli])],
xd->plane[pli].dst.stride, dst, dlist,
dering_count, 3 - dec[pli]);
xd->plane[pli].dst.stride, dst, dlist, dering_count,
3 - dec[pli]);
} else {
#endif
copy_dering_16bit_to_8bit(
&xd->plane[pli].dst.buf[xd->plane[pli].dst.stride *
(MAX_MIB_SIZE * sbr << bsize[pli]) +
(sbc * MAX_MIB_SIZE << bsize[pli])],
xd->plane[pli].dst.stride, dst, dlist,
dering_count, bsize[pli]);
xd->plane[pli].dst.stride, dst, dlist, dering_count, bsize[pli]);
#if CONFIG_AOM_HIGHBITDEPTH
}
#endif

2
av1/common/dering.h
Просмотреть файл

@ -11,8 +11,6 @@
#ifndef AV1_COMMON_DERING_H_
#define AV1_COMMON_DERING_H_
// clang-format off
#include "av1/common/od_dering.h"
#include "av1/common/onyxc_int.h"
#include "aom/aom_integer.h"

32
av1/common/od_dering.c
Просмотреть файл

@ -12,8 +12,6 @@
#include "config.h"
#endif
// clang-format off
#include <stdlib.h>
#include <math.h>
#include "dering.h"
@ -258,39 +256,37 @@ static INLINE int od_adjust_thresh(int threshold, int32_t var) {
return (threshold * OD_THRESH_TABLE_Q8[OD_ILOG(v1)] + 128) >> 8;
}
static INLINE void copy_8x8_16bit_to_16bit(int16_t *dst, int dstride, int16_t *src, int sstride) {
static INLINE void copy_8x8_16bit_to_16bit(int16_t *dst, int dstride,
int16_t *src, int sstride) {
int i, j;
for (i = 0; i < 8; i++)
for (j = 0; j < 8; j++)
dst[i * dstride + j] = src[i * sstride + j];
for (j = 0; j < 8; j++) dst[i * dstride + j] = src[i * sstride + j];
}
static INLINE void copy_4x4_16bit_to_16bit(int16_t *dst, int dstride, int16_t *src, int sstride) {
static INLINE void copy_4x4_16bit_to_16bit(int16_t *dst, int dstride,
int16_t *src, int sstride) {
int i, j;
for (i = 0; i < 4; i++)
for (j = 0; j < 4; j++)
dst[i * dstride + j] = src[i * sstride + j];
for (j = 0; j < 4; j++) dst[i * dstride + j] = src[i * sstride + j];
}
/* TODO: Optimize this function for SSE. */
void copy_dering_16bit_to_16bit(int16_t *dst, int dstride, int16_t *src,
dering_list *dlist, int dering_count, int bsize)
{
dering_list *dlist, int dering_count,
int bsize) {
int bi, bx, by;
if (bsize == 3) {
for (bi = 0; bi < dering_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
copy_8x8_16bit_to_16bit(&dst[(by << 3) * dstride + (bx << 3)],
dstride,
copy_8x8_16bit_to_16bit(&dst[(by << 3) * dstride + (bx << 3)], dstride,
&src[bi << 2 * bsize], 1 << bsize);
}
} else {
for (bi = 0; bi < dering_count; bi++) {
by = dlist[bi].by;
bx = dlist[bi].bx;
copy_4x4_16bit_to_16bit(&dst[(by << 2) * dstride + (bx << 2)],
dstride,
copy_4x4_16bit_to_16bit(&dst[(by << 2) * dstride + (bx << 2)], dstride,
&src[bi << 2 * bsize], 1 << bsize);
}
}
@ -317,8 +313,8 @@ void od_dering(int16_t *y, int16_t *in, int xdec,
int32_t var;
by = dlist[bi].by;
bx = dlist[bi].bx;
dir[by][bx] = od_dir_find8(&in[8 * by * OD_FILT_BSTRIDE + 8 * bx], OD_FILT_BSTRIDE,
&var, coeff_shift);
dir[by][bx] = od_dir_find8(&in[8 * by * OD_FILT_BSTRIDE + 8 * bx],
OD_FILT_BSTRIDE, &var, coeff_shift);
/* Deringing orthogonal to the direction uses a tighter threshold
because we want to be conservative. We've presumably already
achieved some deringing, so the amount of change is expected
@ -350,7 +346,7 @@ void od_dering(int16_t *y, int16_t *in, int xdec,
if (filter2_thresh[by][bx] == 0) continue;
(filter_dering_orthogonal[bsize - OD_LOG_BSIZE0])(
&y[bi << 2 * bsize], 1 << bsize,
&in[(by * OD_FILT_BSTRIDE << bsize) + (bx << bsize)], filter2_thresh[by][bx],
dir[by][bx]);
&in[(by * OD_FILT_BSTRIDE << bsize) + (bx << bsize)],
filter2_thresh[by][bx], dir[by][bx]);
}
}

5
av1/common/od_dering.h
Просмотреть файл

@ -12,8 +12,6 @@
#if !defined(_dering_H)
#define _dering_H (1)
// clang-format off
#include "odintrin.h"
#define OD_DERINGSIZES (2)
@ -47,7 +45,8 @@ typedef void (*od_filter_dering_orthogonal_func)(int16_t *y, int ystride,
const int16_t *in,
int threshold, int dir);
void copy_dering_16bit_to_16bit(int16_t *dst, int dstride, int16_t *src,
dering_list *dlist, int dering_count, int bsize);
dering_list *dlist, int dering_count,
int bsize);
void od_dering(int16_t *y, int16_t *in, int xdec,
int dir[OD_DERING_NBLOCKS][OD_DERING_NBLOCKS], int pli,

22
av1/encoder/pickdering.c
Просмотреть файл

@ -9,8 +9,6 @@
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
// clang-format off
#include <string.h>
#include <math.h>
@ -101,10 +99,9 @@ int av1_dering_search(YV12_BUFFER_CONFIG *frame, const YV12_BUFFER_CONFIG *ref,
int16_t tmp_dst[MAX_MIB_SIZE * MAX_MIB_SIZE * 8 * 8];
nhb = AOMMIN(MAX_MIB_SIZE, cm->mi_cols - MAX_MIB_SIZE * sbc);
nvb = AOMMIN(MAX_MIB_SIZE, cm->mi_rows - MAX_MIB_SIZE * sbr);
dering_count = sb_compute_dering_list(cm, sbr * MAX_MIB_SIZE, sbc * MAX_MIB_SIZE,
dlist);
if (dering_count == 0)
continue;
dering_count = sb_compute_dering_list(cm, sbr * MAX_MIB_SIZE,
sbc * MAX_MIB_SIZE, dlist);
if (dering_count == 0) continue;
best_gi = 0;
for (gi = 0; gi < DERING_REFINEMENT_LEVELS; gi++) {
int cur_mse;
@ -123,13 +120,16 @@ int av1_dering_search(YV12_BUFFER_CONFIG *frame, const YV12_BUFFER_CONFIG *ref,
}
in = inbuf + OD_FILT_VBORDER * OD_FILT_BSTRIDE + OD_FILT_HBORDER;
/* We avoid filtering the pixels for which some of the pixels to average
are outside the frame. We could change the filter instead, but it would
are outside the frame. We could change the filter instead, but it
would
add special cases for any future vectorization. */
for (i = 0; i < OD_DERING_INBUF_SIZE; i++) inbuf[i] = OD_DERING_VERY_LARGE;
for (i = 0; i < OD_DERING_INBUF_SIZE; i++)
inbuf[i] = OD_DERING_VERY_LARGE;
for (i = -OD_FILT_VBORDER * (sbr != 0);
i < (nvb << bsize[0]) + OD_FILT_VBORDER * (sbr != nvsb - 1); i++) {
for (j = -OD_FILT_HBORDER * (sbc != 0);
j < (nhb << bsize[0]) + OD_FILT_HBORDER * (sbc != nhsb - 1); j++) {
j < (nhb << bsize[0]) + OD_FILT_HBORDER * (sbc != nhsb - 1);
j++) {
int16_t *x;
x = &src[(sbr * stride * MAX_MIB_SIZE << bsize[0]) +
(sbc * MAX_MIB_SIZE << bsize[0])];
@ -138,8 +138,8 @@ int av1_dering_search(YV12_BUFFER_CONFIG *frame, const YV12_BUFFER_CONFIG *ref,
}
od_dering(tmp_dst, in, 0, dir, 0, dlist, dering_count, threshold,
coeff_shift);
copy_dering_16bit_to_16bit(dst, MAX_MIB_SIZE << bsize[0], tmp_dst, dlist,
dering_count, bsize[0]);
copy_dering_16bit_to_16bit(dst, MAX_MIB_SIZE << bsize[0], tmp_dst,
dlist, dering_count, bsize[0]);
cur_mse = (int)compute_dist(
dst, MAX_MIB_SIZE << bsize[0],
&ref_coeff[(sbr * stride * MAX_MIB_SIZE << bsize[0]) +