ref: 7b101d4fb3d0942d4b180b8e5a7b1e35af06d0bc
dir: /src/looprestoration.c/
/*
* Copyright © 2018, VideoLAN and dav1d authors
* Copyright © 2018, Two Orioles, LLC
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice, this
* list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
* ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include "config.h"
#include <stdlib.h>
#include "common/intops.h"
#include "src/looprestoration.h"
#include "src/tables.h"
// 256 * 1.5 + 3 + 3 = 390
#define REST_UNIT_STRIDE (390)
// TODO Reuse p when no padding is needed (add and remove lpf pixels in p)
// TODO Chroma only requires 2 rows of padding.
static void padding(pixel *dst, const pixel *p, const ptrdiff_t p_stride,
const pixel (*left)[4],
const pixel *lpf, const ptrdiff_t lpf_stride,
int unit_w, const int stripe_h, const enum LrEdgeFlags edges)
{
const int have_left = !!(edges & LR_HAVE_LEFT);
const int have_right = !!(edges & LR_HAVE_RIGHT);
// Copy more pixels if we don't have to pad them
unit_w += 3 * have_left + 3 * have_right;
pixel *dst_l = dst + 3 * !have_left;
p -= 3 * have_left;
lpf -= 3 * have_left;
if (edges & LR_HAVE_TOP) {
// Copy previous loop filtered rows
const pixel *const above_1 = lpf;
const pixel *const above_2 = above_1 + PXSTRIDE(lpf_stride);
pixel_copy(dst_l, above_1, unit_w);
pixel_copy(dst_l + REST_UNIT_STRIDE, above_1, unit_w);
pixel_copy(dst_l + 2 * REST_UNIT_STRIDE, above_2, unit_w);
} else {
// Pad with first row
pixel_copy(dst_l, p, unit_w);
pixel_copy(dst_l + REST_UNIT_STRIDE, p, unit_w);
pixel_copy(dst_l + 2 * REST_UNIT_STRIDE, p, unit_w);
if (have_left) {
pixel_copy(dst_l, &left[0][1], 3);
pixel_copy(dst_l + REST_UNIT_STRIDE, &left[0][1], 3);
pixel_copy(dst_l + 2 * REST_UNIT_STRIDE, &left[0][1], 3);
}
}
pixel *dst_tl = dst_l + 3 * REST_UNIT_STRIDE;
if (edges & LR_HAVE_BOTTOM) {
// Copy next loop filtered rows
const pixel *const below_1 = lpf + 6 * PXSTRIDE(lpf_stride);
const pixel *const below_2 = below_1 + PXSTRIDE(lpf_stride);
pixel_copy(dst_tl + stripe_h * REST_UNIT_STRIDE, below_1, unit_w);
pixel_copy(dst_tl + (stripe_h + 1) * REST_UNIT_STRIDE, below_2, unit_w);
pixel_copy(dst_tl + (stripe_h + 2) * REST_UNIT_STRIDE, below_2, unit_w);
} else {
// Pad with last row
const pixel *const src = p + (stripe_h - 1) * PXSTRIDE(p_stride);
pixel_copy(dst_tl + stripe_h * REST_UNIT_STRIDE, src, unit_w);
pixel_copy(dst_tl + (stripe_h + 1) * REST_UNIT_STRIDE, src, unit_w);
pixel_copy(dst_tl + (stripe_h + 2) * REST_UNIT_STRIDE, src, unit_w);
if (have_left) {
pixel_copy(dst_tl + stripe_h * REST_UNIT_STRIDE, &left[stripe_h - 1][1], 3);
pixel_copy(dst_tl + (stripe_h + 1) * REST_UNIT_STRIDE, &left[stripe_h - 1][1], 3);
pixel_copy(dst_tl + (stripe_h + 2) * REST_UNIT_STRIDE, &left[stripe_h - 1][1], 3);
}
}
// Inner UNIT_WxSTRIPE_H
for (int j = 0; j < stripe_h; j++) {
pixel_copy(dst_tl + 3 * have_left, p + 3 * have_left, unit_w - 3 * have_left);
dst_tl += REST_UNIT_STRIDE;
p += PXSTRIDE(p_stride);
}
if (!have_right) {
pixel *pad = dst_l + unit_w;
pixel *row_last = &dst_l[unit_w - 1];
// Pad 3x(STRIPE_H+6) with last column
for (int j = 0; j < stripe_h + 6; j++) {
pixel_set(pad, *row_last, 3);
pad += REST_UNIT_STRIDE;
row_last += REST_UNIT_STRIDE;
}
}
if (!have_left) {
// Pad 3x(STRIPE_H+6) with first column
for (int j = 0; j < stripe_h + 6; j++) {
pixel_set(dst, *dst_l, 3);
dst += REST_UNIT_STRIDE;
dst_l += REST_UNIT_STRIDE;
}
} else {
dst += 3 * REST_UNIT_STRIDE;
for (int j = 0; j < stripe_h; j++) {
pixel_copy(dst, &left[j][1], 3);
dst += REST_UNIT_STRIDE;
}
}
}
// FIXME Could split into luma and chroma specific functions,
// (since first and last tops are always 0 for chroma)
// FIXME Could implement a version that requires less temporary memory
// (should be possible to implement with only 6 rows of temp storage)
static void wiener_c(pixel *p, const ptrdiff_t p_stride,
const pixel (*const left)[4],
const pixel *lpf, const ptrdiff_t lpf_stride,
const int w, const int h,
const int16_t filterh[7], const int16_t filterv[7],
const enum LrEdgeFlags edges)
{
// Wiener filtering is applied to a maximum stripe height of 64 + 3 pixels
// of padding above and below
pixel tmp[70 /*(64 + 3 + 3)*/ * REST_UNIT_STRIDE];
pixel *tmp_ptr = tmp;
padding(tmp, p, p_stride, left, lpf, lpf_stride, w, h, edges);
// Values stored between horizontal and vertical filtering don't
// fit in a uint8_t.
uint16_t hor[70 /*(64 + 3 + 3)*/ * REST_UNIT_STRIDE];
uint16_t *hor_ptr = hor;
const int round_bits_h = 3 + (BITDEPTH == 12) * 2;
const int rounding_off_h = 1 << (round_bits_h - 1);
const int clip_limit = 1 << ((BITDEPTH) + 1 + 7 - round_bits_h);
for (int j = 0; j < h + 6; j++) {
for (int i = 0; i < w; i++) {
int sum = (tmp_ptr[i + 3] << 7) + (1 << (BITDEPTH + 6));
for (int k = 0; k < 7; k++) {
sum += tmp_ptr[i + k] * filterh[k];
}
hor_ptr[i] =
iclip((sum + rounding_off_h) >> round_bits_h, 0, clip_limit - 1);
}
tmp_ptr += REST_UNIT_STRIDE;
hor_ptr += REST_UNIT_STRIDE;
}
const int round_bits_v = 11 - (BITDEPTH == 12) * 2;
const int rounding_off_v = 1 << (round_bits_v - 1);
const int round_offset = 1 << (BITDEPTH + (round_bits_v - 1));
for (int i = 0; i < w; i++) {
for (int j = 0; j < h; j++) {
int sum = (hor[(j + 3) * REST_UNIT_STRIDE + i] << 7) - round_offset;
for (int k = 0; k < 7; k++) {
sum += hor[(j + k) * REST_UNIT_STRIDE + i] * filterv[k];
}
p[j * PXSTRIDE(p_stride) + i] =
iclip_pixel((sum + rounding_off_v) >> round_bits_v);
}
}
}
// Sum over a 3x3 area
// The dst and src pointers are positioned 3 pixels above and 3 pixels to the
// left of the top left corner. However, the self guided filter only needs 1
// pixel above and one pixel to the left. As for the pixels below and to the
// right they must be computed in the sums, but don't need to be stored.
//
// Example for a 4x4 block:
// x x x x x x x x x x
// x c c c c c c c c x
// x i s s s s s s i x
// x i s s s s s s i x
// x i s s s s s s i x
// x i s s s s s s i x
// x i s s s s s s i x
// x i s s s s s s i x
// x c c c c c c c c x
// x x x x x x x x x x
//
// s: Pixel summed and stored
// i: Pixel summed and stored (between loops)
// c: Pixel summed not stored
// x: Pixel not summed not stored
static void boxsum3(coef *dst, const pixel *src, const int w, const int h) {
// We skip the first row, as it is never used
src += REST_UNIT_STRIDE;
dst += REST_UNIT_STRIDE;
// We skip the first and last columns, as they are never used
for (int x = 1; x < w - 1; x++) {
coef *ds = dst + x;
const pixel *s = src + x;
int a = s[0], b = s[REST_UNIT_STRIDE];
// We skip the first 2 rows, as they are skipped in the next loop and
// we don't need the last 2 row as it is skipped in the next loop
for (int y = 2; y < h - 2; y++) {
s += REST_UNIT_STRIDE;
const int c = s[REST_UNIT_STRIDE];
ds += REST_UNIT_STRIDE;
*ds = a + b + c;
a = b;
b = c;
}
}
// We skip the first 2 rows as they are never read
dst += REST_UNIT_STRIDE;
// We skip the last 2 rows as it is never read
for (int y = 2; y < h - 2; y++) {
int a = dst[1], b = dst[2];
// We don't store the first column as it is never read and
// we don't store the last 2 columns as they are never read
for (int x = 2; x < w - 2; x++) {
const int c = dst[x + 1];
dst[x] = a + b + c;
a = b;
b = c;
}
dst += REST_UNIT_STRIDE;
}
}
// Sum over a 5x5 area
// The dst and src pointers are positioned 3 pixels above and 3 pixels to the
// left of the top left corner. However, the self guided filter only needs 1
// pixel above and one pixel to the left. As for the pixels below and to the
// right they must be computed in the sums, but don't need to be stored.
//
// Example for a 4x4 block:
// c c c c c c c c c c
// c c c c c c c c c c
// i i s s s s s s i i
// i i s s s s s s i i
// i i s s s s s s i i
// i i s s s s s s i i
// i i s s s s s s i i
// i i s s s s s s i i
// c c c c c c c c c c
// c c c c c c c c c c
//
// s: Pixel summed and stored
// i: Pixel summed and stored (between loops)
// c: Pixel summed not stored
// x: Pixel not summed not stored
static void boxsum5(coef *dst, const pixel *const src, const int w, const int h) {
// We skip the first row, as it is never used
dst += REST_UNIT_STRIDE;
for (int x = 0; x < w; x++) {
coef *ds = dst + x;
const pixel *s = src + 3 * REST_UNIT_STRIDE + x;
int a = s[-3 * REST_UNIT_STRIDE];
int b = s[-2 * REST_UNIT_STRIDE];
int c = s[-1 * REST_UNIT_STRIDE];
int d = s[0];
// We skip the first 2 rows, as they are skipped in the next loop and
// we don't need the last 2 row as it is skipped in the next loop
for (int y = 2; y < h - 2; y++) {
s += REST_UNIT_STRIDE;
const int e = *s;
ds += REST_UNIT_STRIDE;
*ds = a + b + c + d + e;
a = b;
b = c;
c = d;
d = e;
}
}
// We skip the first 2 rows as they are never read
dst += REST_UNIT_STRIDE;
for (int y = 2; y < h - 2; y++) {
int a = dst[0];
int b = dst[1];
int c = dst[2];
int d = dst[3];
for (int x = 2; x < w - 2; x++) {
const int e = dst[x + 2];
dst[x] = a + b + c + d + e;
a = b;
b = c;
c = d;
d = e;
}
dst += REST_UNIT_STRIDE;
}
}
// See boxsum3 function comments for details on row and column skipping
static void boxsum3sqr(int32_t *dst, const pixel *src, const int w, const int h) {
// We skip the first row, as it is never used
src += REST_UNIT_STRIDE;
dst += REST_UNIT_STRIDE;
// We skip the first and last columns, as they are never used
for (int x = 1; x < w - 1; x++) {
int *ds = dst + x;
const pixel *s = src + x;
int a = s[0] * s[0];
int b = s[REST_UNIT_STRIDE] * s[REST_UNIT_STRIDE];
// We skip the first row, as it is skipped in the next loop and
// we don't need the last row as it is skipped in the next loop
for (int y = 2; y < h - 2; y++) {
s += REST_UNIT_STRIDE;
const int c = s[REST_UNIT_STRIDE] * s[REST_UNIT_STRIDE];
ds += REST_UNIT_STRIDE;
*ds = a + b + c;
a = b;
b = c;
}
}
// We skip the first row as it is never read
dst += REST_UNIT_STRIDE;
// We skip the last row as it is never read
for (int y = 2; y < h - 2; y++) {
int a = dst[1], b = dst[2];
// We don't store the first column as it is never read and
// we don't store the last 2 columns as they are never read
for (int x = 2; x < w - 2; x++) {
const int c = dst[x + 1];
dst[x] = a + b + c;
a = b;
b = c;
}
dst += REST_UNIT_STRIDE;
}
}
// See boxsum5 function comments for details on row and column skipping
static void boxsum5sqr(int32_t *dst, const pixel *const src, const int w,
const int h)
{
// We skip the first row, as it is never used
dst += REST_UNIT_STRIDE;
for (int x = 0; x < w; x++) {
int *ds = dst + x;
const pixel *s = src + 3 * REST_UNIT_STRIDE + x;
int a = s[-3 * REST_UNIT_STRIDE] * s[-3 * REST_UNIT_STRIDE];
int b = s[-2 * REST_UNIT_STRIDE] * s[-2 * REST_UNIT_STRIDE];
int c = s[-1 * REST_UNIT_STRIDE] * s[-1 * REST_UNIT_STRIDE];
int d = s[0] * s[0];
// We skip the first 2 rows, as they are skipped in the next loop and
// we don't need the last 2 row as it is skipped in the next loop
for (int y = 2; y < h - 2; y++) {
s += REST_UNIT_STRIDE;
const int e = s[0] * s[0];
ds += REST_UNIT_STRIDE;
*ds = a + b + c + d + e;
a = b;
b = c;
c = d;
d = e;
}
}
// We skip the first 2 rows as they are never read
dst += REST_UNIT_STRIDE;
for (int y = 2; y < h - 2; y++) {
int a = dst[0];
int b = dst[1];
int c = dst[2];
int d = dst[3];
for (int x = 2; x < w - 2; x++) {
const int e = dst[x + 2];
dst[x] = a + b + c + d + e;
a = b;
b = c;
c = d;
d = e;
}
dst += REST_UNIT_STRIDE;
}
}
static void selfguided_filter(int32_t *dst, const pixel *src,
const ptrdiff_t src_stride, const int w,
const int h, const int n, const int s)
{
// Selfguided filter is applied to a maximum stripe height of 64 + 3 pixels
// of padding above and below
int32_t A_[70 /*(64 + 3 + 3)*/ * REST_UNIT_STRIDE];
int32_t *A = A_ + 3 * REST_UNIT_STRIDE + 3;
// By inverting A and B after the boxsums, B can be of size coef instead
// of int32_t
coef B_[70 /*(64 + 3 + 3)*/ * REST_UNIT_STRIDE];
coef *B = B_ + 3 * REST_UNIT_STRIDE + 3;
const int step = (n == 25) + 1;
if (n == 25) {
boxsum5(B_, src, w + 6, h + 6);
boxsum5sqr(A_, src, w + 6, h + 6);
} else {
boxsum3(B_, src, w + 6, h + 6);
boxsum3sqr(A_, src, w + 6, h + 6);
}
int32_t *AA = A - REST_UNIT_STRIDE;
coef *BB = B - REST_UNIT_STRIDE;
for (int j = -1; j < h + 1; j+= step) {
for (int i = -1; i < w + 1; i++) {
const int a =
(AA[i] + (1 << (2 * (BITDEPTH - 8)) >> 1)) >> (2 * (BITDEPTH - 8));
const int b =
(BB[i] + (1 << (BITDEPTH - 8) >> 1)) >> (BITDEPTH - 8);
const uint32_t p = (a * n >= b * b) * (a * n - b * b);
const uint32_t z = (p * s + (1 << 19)) >> 20;
const int x = dav1d_sgr_x_by_xplus1[imin(z, 255)];
// This is where we invert A and B, so that B is of size coef.
AA[i] = (((1 << 8) - x) * BB[i] * dav1d_sgr_one_by_x[n - 1] + (1 << 11)) >> 12;
BB[i] = x;
}
AA += step * REST_UNIT_STRIDE;
BB += step * REST_UNIT_STRIDE;
}
src += 3 * REST_UNIT_STRIDE + 3;
if (n == 25) {
int j = 0;
#define SIX_NEIGHBORS(P, i)\
((P[i - REST_UNIT_STRIDE] + P[i + REST_UNIT_STRIDE]) * 6 + \
(P[i - 1 - REST_UNIT_STRIDE] + P[i - 1 + REST_UNIT_STRIDE] + \
P[i + 1 - REST_UNIT_STRIDE] + P[i + 1 + REST_UNIT_STRIDE]) * 5)
for (; j < h - 1; j+=2) {
for (int i = 0; i < w; i++) {
const int32_t a = SIX_NEIGHBORS(B, i);
const int32_t b = SIX_NEIGHBORS(A, i);
dst[i] = (a * src[i] + b + (1 << 8)) >> 9;
}
dst += 384 /* Maximum restoration width is 384 (256 * 1.5) */;
src += REST_UNIT_STRIDE;
B += REST_UNIT_STRIDE;
A += REST_UNIT_STRIDE;
for (int i = 0; i < w; i++) {
const int32_t a = B[i] * 6 + (B[i - 1] + B[i + 1]) * 5;
const int32_t b = A[i] * 6 + (A[i - 1] + A[i + 1]) * 5;
dst[i] = (a * src[i] + b + (1 << 7)) >> 8;
}
dst += 384 /* Maximum restoration width is 384 (256 * 1.5) */;
src += REST_UNIT_STRIDE;
B += REST_UNIT_STRIDE;
A += REST_UNIT_STRIDE;
}
if (j + 1 == h) { // Last row, when number of rows is odd
for (int i = 0; i < w; i++) {
const int32_t a = SIX_NEIGHBORS(B, i);
const int32_t b = SIX_NEIGHBORS(A, i);
dst[i] = (a * src[i] + b + (1 << 8)) >> 9;
}
}
#undef SIX_NEIGHBORS
} else {
#define EIGHT_NEIGHBORS(P, i)\
((P[i] + P[i - 1] + P[i + 1] + P[i - REST_UNIT_STRIDE] + P[i + REST_UNIT_STRIDE]) * 4 + \
(P[i - 1 - REST_UNIT_STRIDE] + P[i - 1 + REST_UNIT_STRIDE] + \
P[i + 1 - REST_UNIT_STRIDE] + P[i + 1 + REST_UNIT_STRIDE]) * 3)
for (int j = 0; j < h; j++) {
for (int i = 0; i < w; i++) {
const int32_t a = EIGHT_NEIGHBORS(B, i);
const int32_t b = EIGHT_NEIGHBORS(A, i);
dst[i] = (a * src[i] + b + (1 << 8)) >> 9;
}
dst += 384;
src += REST_UNIT_STRIDE;
B += REST_UNIT_STRIDE;
A += REST_UNIT_STRIDE;
}
}
#undef NINE_NEIGHBORS
}
static void selfguided_c(pixel *p, const ptrdiff_t p_stride,
const pixel (*const left)[4],
const pixel *lpf, const ptrdiff_t lpf_stride,
const int w, const int h, const int sgr_idx,
const int16_t sgr_w[2], const enum LrEdgeFlags edges)
{
// Selfguided filter is applied to a maximum stripe height of 64 + 3 pixels
// of padding above and below
pixel tmp[70 /*(64 + 3 + 3)*/ * REST_UNIT_STRIDE];
padding(tmp, p, p_stride, left, lpf, lpf_stride, w, h, edges);
// Selfguided filter outputs to a maximum stripe height of 64 and a
// maximum restoration width of 384 (256 * 1.5)
int32_t dst[64 * 384];
// both r1 and r0 can't be zero
if (!dav1d_sgr_params[sgr_idx][0]) {
const int s1 = dav1d_sgr_params[sgr_idx][3];
selfguided_filter(dst, tmp, REST_UNIT_STRIDE, w, h, 9, s1);
const int w1 = (1 << 7) - sgr_w[1];
for (int j = 0; j < h; j++) {
for (int i = 0; i < w; i++) {
const int32_t u = (p[i] << 4);
const int32_t v = (u << 7) + w1 * (dst[j * 384 + i] - u);
p[i] = iclip_pixel((v + (1 << 10)) >> 11);
}
p += PXSTRIDE(p_stride);
}
} else if (!dav1d_sgr_params[sgr_idx][1]) {
const int s0 = dav1d_sgr_params[sgr_idx][2];
selfguided_filter(dst, tmp, REST_UNIT_STRIDE, w, h, 25, s0);
const int w0 = sgr_w[0];
for (int j = 0; j < h; j++) {
for (int i = 0; i < w; i++) {
const int32_t u = (p[i] << 4);
const int32_t v = (u << 7) + w0 * (dst[j * 384 + i] - u);
p[i] = iclip_pixel((v + (1 << 10)) >> 11);
}
p += PXSTRIDE(p_stride);
}
} else {
int32_t dst1[64 * 384];
const int s0 = dav1d_sgr_params[sgr_idx][2];
const int s1 = dav1d_sgr_params[sgr_idx][3];
const int w0 = sgr_w[0];
const int w1 = (1 << 7) - w0 - sgr_w[1];
selfguided_filter(dst, tmp, REST_UNIT_STRIDE, w, h, 25, s0);
selfguided_filter(dst1, tmp, REST_UNIT_STRIDE, w, h, 9, s1);
for (int j = 0; j < h; j++) {
for (int i = 0; i < w; i++) {
const int32_t u = (p[i] << 4);
const int32_t v = (u << 7) + w0 * (dst[j * 384 + i] - u) +
w1 * (dst1[j * 384 + i] - u);
p[i] = iclip_pixel((v + (1 << 10)) >> 11);
}
p += PXSTRIDE(p_stride);
}
}
}
void bitfn(dav1d_loop_restoration_dsp_init)(Dav1dLoopRestorationDSPContext *const c) {
c->wiener = wiener_c;
c->selfguided = selfguided_c;
#if HAVE_ASM && ARCH_X86 && BITDEPTH == 8
bitfn(dav1d_loop_restoration_dsp_init_x86)(c);
#endif
}