ffmpeg/libavcodec/vvc/inter_template.c

467 lines
19 KiB
C

/*
* VVC inter prediction DSP
*
* Copyright (C) 2022 Nuo Mi
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "libavcodec/h26x/h2656_inter_template.c"
static void FUNC(avg)(uint8_t *_dst, const ptrdiff_t _dst_stride,
const int16_t *src0, const int16_t *src1, const int width, const int height)
{
pixel *dst = (pixel*)_dst;
const ptrdiff_t dst_stride = _dst_stride / sizeof(pixel);
const int shift = FFMAX(3, 15 - BIT_DEPTH);
const int offset = 1 << (shift - 1);
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x++)
dst[x] = av_clip_pixel((src0[x] + src1[x] + offset) >> shift);
src0 += MAX_PB_SIZE;
src1 += MAX_PB_SIZE;
dst += dst_stride;
}
}
static void FUNC(w_avg)(uint8_t *_dst, const ptrdiff_t _dst_stride,
const int16_t *src0, const int16_t *src1, const int width, const int height,
const int denom, const int w0, const int w1, const int o0, const int o1)
{
pixel *dst = (pixel*)_dst;
const ptrdiff_t dst_stride = _dst_stride / sizeof(pixel);
const int shift = denom + FFMAX(3, 15 - BIT_DEPTH);
const int offset = ((o0 + o1) * (1 << (BIT_DEPTH - 8)) + 1) * (1 << (shift - 1));
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x++)
dst[x] = av_clip_pixel((src0[x] * w0 + src1[x] * w1 + offset) >> shift);
src0 += MAX_PB_SIZE;
src1 += MAX_PB_SIZE;
dst += dst_stride;
}
}
static void FUNC(put_ciip)(uint8_t *_dst, const ptrdiff_t _dst_stride,
const int width, const int height,
const uint8_t *_inter, const ptrdiff_t _inter_stride, const int intra_weight)
{
pixel *dst = (pixel *)_dst;
pixel *inter = (pixel *)_inter;
const size_t dst_stride = _dst_stride / sizeof(pixel);
const size_t inter_stride = _inter_stride / sizeof(pixel);
const int inter_weight = 4 - intra_weight;
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x++)
dst[x] = (dst[x] * intra_weight + inter[x] * inter_weight + 2) >> 2;
dst += dst_stride;
inter += inter_stride;
}
}
static void FUNC(put_gpm)(uint8_t *_dst, ptrdiff_t dst_stride,
const int width, const int height,
const int16_t *src0, const int16_t *src1,
const uint8_t *weights, const int step_x, const int step_y)
{
const int shift = FFMAX(5, 17 - BIT_DEPTH);
const int offset = 1 << (shift - 1);
pixel *dst = (pixel *)_dst;
dst_stride /= sizeof(pixel);
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x++) {
const uint8_t w = weights[x * step_x];
dst[x] = av_clip_pixel((src0[x] * w + src1[x] * (8 - w) + offset) >> shift);
}
dst += dst_stride;
src0 += MAX_PB_SIZE;
src1 += MAX_PB_SIZE;
weights += step_y;
}
}
//8.5.6.3.3 Luma integer sample fetching process, add one extra pad line
static void FUNC(bdof_fetch_samples)(int16_t *_dst, const uint8_t *_src, const ptrdiff_t _src_stride,
const int x_frac, const int y_frac, const int width, const int height)
{
const int x_off = (x_frac >> 3) - 1;
const int y_off = (y_frac >> 3) - 1;
const ptrdiff_t src_stride = _src_stride / sizeof(pixel);
const pixel *src = (pixel*)_src + (x_off) + y_off * src_stride;
int16_t *dst = _dst - 1 - MAX_PB_SIZE;
const int shift = 14 - BIT_DEPTH;
const int bdof_width = width + 2 * BDOF_BORDER_EXT;
// top
for (int i = 0; i < bdof_width; i++)
dst[i] = src[i] << shift;
dst += MAX_PB_SIZE;
src += src_stride;
for (int i = 0; i < height; i++) {
dst[0] = src[0] << shift;
dst[1 + width] = src[1 + width] << shift;
dst += MAX_PB_SIZE;
src += src_stride;
}
for (int i = 0; i < bdof_width; i++)
dst[i] = src[i] << shift;
}
//8.5.6.3.3 Luma integer sample fetching process
static void FUNC(fetch_samples)(int16_t *_dst, const uint8_t *_src, const ptrdiff_t _src_stride, const int x_frac, const int y_frac)
{
FUNC(bdof_fetch_samples)(_dst, _src, _src_stride, x_frac, y_frac, AFFINE_MIN_BLOCK_SIZE, AFFINE_MIN_BLOCK_SIZE);
}
static void FUNC(prof_grad_filter)(int16_t *_gradient_h, int16_t *_gradient_v, const ptrdiff_t gradient_stride,
const int16_t *_src, const ptrdiff_t src_stride, const int width, const int height, const int pad)
{
const int shift = 6;
const int16_t *src = _src;
int16_t *gradient_h = _gradient_h + pad * (1 + gradient_stride);
int16_t *gradient_v = _gradient_v + pad * (1 + gradient_stride);
for (int y = 0; y < height; y++) {
const int16_t *p = src;
for (int x = 0; x < width; x++) {
gradient_h[x] = (p[1] >> shift) - (p[-1] >> shift);
gradient_v[x] = (p[src_stride] >> shift) - (p[-src_stride] >> shift);
p++;
}
gradient_h += gradient_stride;
gradient_v += gradient_stride;
src += src_stride;
}
if (pad) {
pad_int16(_gradient_h + 1 + gradient_stride, gradient_stride, width, height);
pad_int16(_gradient_v + 1 + gradient_stride, gradient_stride, width, height);
}
}
static void FUNC(apply_prof)(int16_t *dst, const int16_t *src, const int16_t *diff_mv_x, const int16_t *diff_mv_y)
{
const int limit = (1 << FFMAX(13, BIT_DEPTH + 1)); ///< dILimit
int16_t gradient_h[AFFINE_MIN_BLOCK_SIZE * AFFINE_MIN_BLOCK_SIZE];
int16_t gradient_v[AFFINE_MIN_BLOCK_SIZE * AFFINE_MIN_BLOCK_SIZE];
FUNC(prof_grad_filter)(gradient_h, gradient_v, AFFINE_MIN_BLOCK_SIZE, src, MAX_PB_SIZE, AFFINE_MIN_BLOCK_SIZE, AFFINE_MIN_BLOCK_SIZE, 0);
for (int y = 0; y < AFFINE_MIN_BLOCK_SIZE; y++) {
for (int x = 0; x < AFFINE_MIN_BLOCK_SIZE; x++) {
const int o = y * AFFINE_MIN_BLOCK_SIZE + x;
const int di = gradient_h[o] * diff_mv_x[o] + gradient_v[o] * diff_mv_y[o];
const int val = src[x] + av_clip(di, -limit, limit - 1);
dst[x] = val;
}
src += MAX_PB_SIZE;
dst += MAX_PB_SIZE;
}
}
static void FUNC(apply_prof_uni)(uint8_t *_dst, const ptrdiff_t _dst_stride, const int16_t *src, const int16_t *diff_mv_x, const int16_t *diff_mv_y)
{
const int limit = (1 << FFMAX(13, BIT_DEPTH + 1)); ///< dILimit
pixel *dst = (pixel*)_dst;
const ptrdiff_t dst_stride = _dst_stride / sizeof(pixel);
const int shift = 14 - BIT_DEPTH;
#if BIT_DEPTH < 14
const int offset = 1 << (shift - 1);
#else
const int offset = 0;
#endif
int16_t gradient_h[AFFINE_MIN_BLOCK_SIZE * AFFINE_MIN_BLOCK_SIZE];
int16_t gradient_v[AFFINE_MIN_BLOCK_SIZE * AFFINE_MIN_BLOCK_SIZE];
FUNC(prof_grad_filter)(gradient_h, gradient_v, AFFINE_MIN_BLOCK_SIZE, src, MAX_PB_SIZE, AFFINE_MIN_BLOCK_SIZE, AFFINE_MIN_BLOCK_SIZE, 0);
for (int y = 0; y < AFFINE_MIN_BLOCK_SIZE; y++) {
for (int x = 0; x < AFFINE_MIN_BLOCK_SIZE; x++) {
const int o = y * AFFINE_MIN_BLOCK_SIZE + x;
const int di = gradient_h[o] * diff_mv_x[o] + gradient_v[o] * diff_mv_y[o];
const int val = src[x] + av_clip(di, -limit, limit - 1);
dst[x] = av_clip_pixel((val + offset) >> shift);
}
src += MAX_PB_SIZE;
dst += dst_stride;
}
}
static void FUNC(apply_prof_uni_w)(uint8_t *_dst, const ptrdiff_t _dst_stride,
const int16_t *src, const int16_t *diff_mv_x, const int16_t *diff_mv_y,
const int denom, const int wx, const int _ox)
{
const int limit = (1 << FFMAX(13, BIT_DEPTH + 1)); ///< dILimit
pixel *dst = (pixel*)_dst;
const ptrdiff_t dst_stride = _dst_stride / sizeof(pixel);
const int shift = denom + FFMAX(2, 14 - BIT_DEPTH);
const int offset = 1 << (shift - 1);
const int ox = _ox * (1 << (BIT_DEPTH - 8));
int16_t gradient_h[AFFINE_MIN_BLOCK_SIZE * AFFINE_MIN_BLOCK_SIZE];
int16_t gradient_v[AFFINE_MIN_BLOCK_SIZE * AFFINE_MIN_BLOCK_SIZE];
FUNC(prof_grad_filter)(gradient_h, gradient_v, AFFINE_MIN_BLOCK_SIZE, src, MAX_PB_SIZE, AFFINE_MIN_BLOCK_SIZE, AFFINE_MIN_BLOCK_SIZE, 0);
for (int y = 0; y < AFFINE_MIN_BLOCK_SIZE; y++) {
for (int x = 0; x < AFFINE_MIN_BLOCK_SIZE; x++) {
const int o = y * AFFINE_MIN_BLOCK_SIZE + x;
const int di = gradient_h[o] * diff_mv_x[o] + gradient_v[o] * diff_mv_y[o];
const int val = src[x] + av_clip(di, -limit, limit - 1);
dst[x] = av_clip_pixel(((val * wx + offset) >> shift) + ox);
}
src += MAX_PB_SIZE;
dst += dst_stride;
}
}
static void FUNC(derive_bdof_vx_vy)(const int16_t *_src0, const int16_t *_src1,
const int16_t **gradient_h, const int16_t **gradient_v, ptrdiff_t gradient_stride,
int* vx, int* vy)
{
const int shift2 = 4;
const int shift3 = 1;
const int thres = 1 << 4;
int sgx2 = 0, sgy2 = 0, sgxgy = 0, sgxdi = 0, sgydi = 0;
const int16_t *src0 = _src0 - 1 - MAX_PB_SIZE;
const int16_t *src1 = _src1 - 1 - MAX_PB_SIZE;
for (int y = 0; y < BDOF_GRADIENT_SIZE; y++) {
for (int x = 0; x < BDOF_GRADIENT_SIZE; x++) {
const int diff = (src0[x] >> shift2) - (src1[x] >> shift2);
const int idx = gradient_stride * y + x;
const int temph = (gradient_h[0][idx] + gradient_h[1][idx]) >> shift3;
const int tempv = (gradient_v[0][idx] + gradient_v[1][idx]) >> shift3;
sgx2 += FFABS(temph);
sgy2 += FFABS(tempv);
sgxgy += VVC_SIGN(tempv) * temph;
sgxdi += -VVC_SIGN(temph) * diff;
sgydi += -VVC_SIGN(tempv) * diff;
}
src0 += MAX_PB_SIZE;
src1 += MAX_PB_SIZE;
}
*vx = sgx2 > 0 ? av_clip((sgxdi * (1 << 2)) >> av_log2(sgx2) , -thres + 1, thres - 1) : 0;
*vy = sgy2 > 0 ? av_clip(((sgydi * (1 << 2)) - ((*vx * sgxgy) >> 1)) >> av_log2(sgy2), -thres + 1, thres - 1) : 0;
}
static void FUNC(apply_bdof_min_block)(pixel* dst, const ptrdiff_t dst_stride, const int16_t *src0, const int16_t *src1,
const int16_t **gradient_h, const int16_t **gradient_v, const int vx, const int vy)
{
const int shift4 = 15 - BIT_DEPTH;
const int offset4 = 1 << (shift4 - 1);
const int16_t* gh[] = { gradient_h[0] + 1 + BDOF_PADDED_SIZE, gradient_h[1] + 1 + BDOF_PADDED_SIZE };
const int16_t* gv[] = { gradient_v[0] + 1 + BDOF_PADDED_SIZE, gradient_v[1] + 1 + BDOF_PADDED_SIZE };
for (int y = 0; y < BDOF_BLOCK_SIZE; y++) {
for (int x = 0; x < BDOF_BLOCK_SIZE; x++) {
const int idx = y * BDOF_PADDED_SIZE + x;
const int bdof_offset = vx * (gh[0][idx] - gh[1][idx]) + vy * (gv[0][idx] - gv[1][idx]);
dst[x] = av_clip_pixel((src0[x] + offset4 + src1[x] + bdof_offset) >> shift4);
}
dst += dst_stride;
src0 += MAX_PB_SIZE;
src1 += MAX_PB_SIZE;
}
}
static void FUNC(apply_bdof)(uint8_t *_dst, const ptrdiff_t _dst_stride, int16_t *_src0, int16_t *_src1,
const int block_w, const int block_h)
{
int16_t gradient_h[2][BDOF_PADDED_SIZE * BDOF_PADDED_SIZE];
int16_t gradient_v[2][BDOF_PADDED_SIZE * BDOF_PADDED_SIZE];
int vx, vy;
const ptrdiff_t dst_stride = _dst_stride / sizeof(pixel);
pixel* dst = (pixel*)_dst;
FUNC(prof_grad_filter)(gradient_h[0], gradient_v[0], BDOF_PADDED_SIZE,
_src0, MAX_PB_SIZE, block_w, block_h, 1);
pad_int16(_src0, MAX_PB_SIZE, block_w, block_h);
FUNC(prof_grad_filter)(gradient_h[1], gradient_v[1], BDOF_PADDED_SIZE,
_src1, MAX_PB_SIZE, block_w, block_h, 1);
pad_int16(_src1, MAX_PB_SIZE, block_w, block_h);
for (int y = 0; y < block_h; y += BDOF_BLOCK_SIZE) {
for (int x = 0; x < block_w; x += BDOF_BLOCK_SIZE) {
const int16_t* src0 = _src0 + y * MAX_PB_SIZE + x;
const int16_t* src1 = _src1 + y * MAX_PB_SIZE + x;
pixel *d = dst + x;
const int idx = BDOF_PADDED_SIZE * y + x;
const int16_t* gh[] = { gradient_h[0] + idx, gradient_h[1] + idx };
const int16_t* gv[] = { gradient_v[0] + idx, gradient_v[1] + idx };
FUNC(derive_bdof_vx_vy)(src0, src1, gh, gv, BDOF_PADDED_SIZE, &vx, &vy);
FUNC(apply_bdof_min_block)(d, dst_stride, src0, src1, gh, gv, vx, vy);
}
dst += BDOF_BLOCK_SIZE * dst_stride;
}
}
#define DMVR_FILTER(src, stride) \
(filter[0] * src[x] + \
filter[1] * src[x + stride])
//8.5.3.2.2 Luma sample bilinear interpolation process
static void FUNC(dmvr)(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
const int height, const intptr_t mx, const intptr_t my, const int width)
{
const pixel *src = (const pixel *)_src;
const ptrdiff_t src_stride = _src_stride / sizeof(pixel);
#if BIT_DEPTH > 10
const int shift4 = BIT_DEPTH - 10;
const int offset4 = 1 << (shift4 - 1);
#define DMVR_SHIFT(s) (((s) + offset4) >> shift4)
#else
#define DMVR_SHIFT(s) ((s) << (10 - BIT_DEPTH))
#endif
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x++)
dst[x] = DMVR_SHIFT(src[x]);
src += src_stride;
dst += MAX_PB_SIZE;
}
#undef DMVR_SHIFT
}
//8.5.3.2.2 Luma sample bilinear interpolation process
static void FUNC(dmvr_h)(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
const int height, const intptr_t mx, const intptr_t my, const int width)
{
const pixel *src = (const pixel*)_src;
const ptrdiff_t src_stride = _src_stride / sizeof(pixel);
const int8_t *filter = ff_vvc_inter_luma_dmvr_filters[mx];
const int shift1 = BIT_DEPTH - 6;
const int offset1 = 1 << (shift1 - 1);
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x++)
dst[x] = (DMVR_FILTER(src, 1) + offset1) >> shift1;
src += src_stride;
dst += MAX_PB_SIZE;
}
}
//8.5.3.2.2 Luma sample bilinear interpolation process
static void FUNC(dmvr_v)(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
const int height, const intptr_t mx, const intptr_t my, const int width)
{
const pixel *src = (pixel*)_src;
const ptrdiff_t src_stride = _src_stride / sizeof(pixel);
const int8_t *filter = ff_vvc_inter_luma_dmvr_filters[my];
const int shift1 = BIT_DEPTH - 6;
const int offset1 = 1 << (shift1 - 1);
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x++)
dst[x] = (DMVR_FILTER(src, src_stride) + offset1) >> shift1;
src += src_stride;
dst += MAX_PB_SIZE;
}
}
//8.5.3.2.2 Luma sample bilinear interpolation process
static void FUNC(dmvr_hv)(int16_t *dst, const uint8_t *_src, const ptrdiff_t _src_stride,
const int height, const intptr_t mx, const intptr_t my, const int width)
{
int16_t tmp_array[(MAX_PB_SIZE + BILINEAR_EXTRA) * MAX_PB_SIZE];
int16_t *tmp = tmp_array;
const pixel *src = (const pixel*)_src;
const ptrdiff_t src_stride = _src_stride / sizeof(pixel);
const int8_t *filter = ff_vvc_inter_luma_dmvr_filters[mx];
const int shift1 = BIT_DEPTH - 6;
const int offset1 = 1 << (shift1 - 1);
const int shift2 = 4;
const int offset2 = 1 << (shift2 - 1);
src -= BILINEAR_EXTRA_BEFORE * src_stride;
for (int y = 0; y < height + BILINEAR_EXTRA; y++) {
for (int x = 0; x < width; x++)
tmp[x] = (DMVR_FILTER(src, 1) + offset1) >> shift1;
src += src_stride;
tmp += MAX_PB_SIZE;
}
tmp = tmp_array + BILINEAR_EXTRA_BEFORE * MAX_PB_SIZE;
filter = ff_vvc_inter_luma_dmvr_filters[my];
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x++)
dst[x] = (DMVR_FILTER(tmp, MAX_PB_SIZE) + offset2) >> shift2;
tmp += MAX_PB_SIZE;
dst += MAX_PB_SIZE;
}
}
#define PEL_FUNC(dst, C, idx1, idx2, a) \
do { \
for (int w = 0; w < 7; w++) \
inter->dst[C][w][idx1][idx2] = FUNC(a); \
} while (0) \
#define DIR_FUNCS(d, C, c) \
PEL_FUNC(put_##d, C, 0, 0, put_##d##_pixels); \
PEL_FUNC(put_##d, C, 0, 1, put_##d##_##c##_h); \
PEL_FUNC(put_##d, C, 1, 0, put_##d##_##c##_v); \
PEL_FUNC(put_##d, C, 1, 1, put_##d##_##c##_hv); \
PEL_FUNC(put_##d##_w, C, 0, 0, put_##d##_w_pixels); \
PEL_FUNC(put_##d##_w, C, 0, 1, put_##d##_##c##_w_h); \
PEL_FUNC(put_##d##_w, C, 1, 0, put_##d##_##c##_w_v); \
PEL_FUNC(put_##d##_w, C, 1, 1, put_##d##_##c##_w_hv);
#define FUNCS(C, c) \
PEL_FUNC(put, C, 0, 0, put_pixels); \
PEL_FUNC(put, C, 0, 1, put_##c##_h); \
PEL_FUNC(put, C, 1, 0, put_##c##_v); \
PEL_FUNC(put, C, 1, 1, put_##c##_hv); \
DIR_FUNCS(uni, C, c); \
static void FUNC(ff_vvc_inter_dsp_init)(VVCInterDSPContext *const inter)
{
FUNCS(LUMA, luma);
FUNCS(CHROMA, chroma);
inter->avg = FUNC(avg);
inter->w_avg = FUNC(w_avg);
inter->dmvr[0][0] = FUNC(dmvr);
inter->dmvr[0][1] = FUNC(dmvr_h);
inter->dmvr[1][0] = FUNC(dmvr_v);
inter->dmvr[1][1] = FUNC(dmvr_hv);
inter->put_ciip = FUNC(put_ciip);
inter->put_gpm = FUNC(put_gpm);
inter->fetch_samples = FUNC(fetch_samples);
inter->bdof_fetch_samples = FUNC(bdof_fetch_samples);
inter->apply_prof = FUNC(apply_prof);
inter->apply_prof_uni = FUNC(apply_prof_uni);
inter->apply_prof_uni_w = FUNC(apply_prof_uni_w);
inter->apply_bdof = FUNC(apply_bdof);
inter->prof_grad_filter = FUNC(prof_grad_filter);
inter->sad = vvc_sad;
}
#undef FUNCS
#undef PEL_FUNC
#undef DMVR_FUNCS