Skip to content

experiment: separate out chroma deblock and start unrolling #285

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
stone-d-chen wants to merge 1 commit into
base: deblock_asm_20250223
Choose a base branch
from
Draft

experiment: separate out chroma deblock and start unrolling #285

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
128 changes: 115 additions & 13 deletions libavcodec/vvc/filter.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -780,6 +780,7 @@ static int get_qp(const VVCFrameContext *fc, const uint8_t *src, const int x, co
return get_qp_c(fc, x, y, c_idx, vertical);
}


static void vvc_deblock(const VVCLocalContext *lc, int x0, int y0, const int rs, const int vertical)
{
VVCFrameContext *fc = lc->fc;
Expand All @@ -790,6 +791,7 @@ static void vvc_deblock(const VVCLocalContext *lc, int x0, int y0, const int rs,
int x_end = FFMIN(x0 + ctb_size, fc->ps.pps->width);
int y_end = FFMIN(y0 + ctb_size, fc->ps.pps->height);


//not use this yet, may needed by plt.
const uint8_t no_p[4] = { 0 };
const uint8_t no_q[4] = { 0 } ;
Expand All @@ -801,21 +803,26 @@ static void vvc_deblock(const VVCLocalContext *lc, int x0, int y0, const int rs,
FFSWAP(int, x0, y0);
}

for (int c_idx = 0; c_idx < c_end; c_idx++) {
for (int c_idx = 0; c_idx < 1; c_idx++)
{
// c_idx != 0 is chroma, probably from the two chroma
const int hs = (vertical ? sps->hshift : sps->vshift)[c_idx];
const int vs = (vertical ? sps->vshift : sps->hshift)[c_idx];
const int grid = c_idx ? (CHROMA_GRID << hs) : LUMA_GRID;
const int tc_offset = params->tc_offset[c_idx];
const int beta_offset = params->beta_offset[c_idx];
const int src_stride = fc->frame->linesize[c_idx];

for (int y = y0; y < y_end; y += (DEBLOCK_STEP << vs)) {
for (int x = x0 ? x0 : grid; x < x_end; x += grid) {
for (int y = y0; y < y_end; y += (DEBLOCK_STEP << vs))
{
for (int x = x0 ? x0 : grid; x < x_end; x += grid)
{
const uint8_t horizontal_ctu_edge = !vertical && !(x % ctb_size);
int32_t bs[4], beta[4], tc[4] = { 0 }, all_zero_bs = 1;
uint8_t max_len_p[4], max_len_q[4];
int32_t bs[8], beta[8], tc[8] = { 0 }, all_zero_bs = 1;
uint8_t max_len_p[8], max_len_q[8];

for (int i = 0; i < DEBLOCK_STEP >> (2 - vs); i++) {
for (int i = 0; i < DEBLOCK_STEP >> (2 - vs); i++)
{
int tx = x;
int ty = y + (i << 2);
const int end = ty >= y_end;
Expand All @@ -824,7 +831,8 @@ static void vvc_deblock(const VVCLocalContext *lc, int x0, int y0, const int rs,
FFSWAP(int, tx, ty);

bs[i] = end ? 0 : TAB_BS(fc->tab.bs[vertical][c_idx], tx, ty);
if (bs[i]) {
if (bs[i])
{
const int qp = get_qp(fc, POS(c_idx, tx, ty), tx, ty, c_idx, vertical);
beta[i] = betatable[av_clip(qp + beta_offset, 0, MAX_QP)];
tc[i] = TC_CALC(qp, bs[i]) ;
Expand All @@ -833,26 +841,120 @@ static void vvc_deblock(const VVCLocalContext *lc, int x0, int y0, const int rs,
}
}

if (!all_zero_bs) {
if (!all_zero_bs)
{
uint8_t *src = vertical ? POS(c_idx, x, y) : POS(c_idx, y, x);
if (!c_idx) {
if (!c_idx)
{
int use_c = 0;
for (int i = 0; i < 2; i++) {
for (int i = 0; i < 2; i++)
{
if (max_len_p[i] > 3 || max_len_q[i])
use_c = 1;
}
if (use_c) {
if (use_c)
{
fc->vvcdsp.lf.filter_luma_c[vertical](src, src_stride, beta, tc, no_p, no_q, max_len_p, max_len_q, horizontal_ctu_edge);
} else {
}
else
{
fc->vvcdsp.lf.filter_luma[vertical](src, src_stride, beta, tc, no_p, no_q, max_len_p, max_len_q, horizontal_ctu_edge);
}
} else {
}
else
{
// printf("y0 %d y_end %d y_size = %d\n", y0, y_end, y_end - y0);
printf("x0 %d x_end %d x_size = %d\n", x0, x_end, x_end - x0);
printf("grid %d\n", grid);
assert(x_end - x0 > 16);

// printf("grid %d\n", grid);
fc->vvcdsp.lf.filter_chroma[vertical](src, src_stride, beta, tc, no_p, no_q, max_len_p, max_len_q, vs);
}
}
}
}
}

for (int c_idx = 1; c_idx < c_end; c_idx++)
{
// c_idx != 0 is chroma, probably from the two chroma
const int hs = (vertical ? sps->hshift : sps->vshift)[c_idx];
const int vs = (vertical ? sps->vshift : sps->hshift)[c_idx];
const int grid = (CHROMA_GRID << hs);
const int tc_offset = params->tc_offset[c_idx];
const int beta_offset = params->beta_offset[c_idx];
const int src_stride = fc->frame->linesize[c_idx];


for (int y = y0; y < y_end; y += (DEBLOCK_STEP << vs))
{
for (int x = x0 ? x0 : grid; x < x_end; x += 2 * grid)
{
const uint8_t horizontal_ctu_edge = !vertical && !(x % ctb_size);
int32_t bs[4], beta[4], tc[4] = {0}, all_zero_bs = 1;
uint8_t max_len_p[4], max_len_q[4];

const uint8_t horizontal_ctu_edge2 = !vertical && !((x + grid) % ctb_size);
int32_t bs2[4], beta2[4], tc2[4] = {0}, all_zero_bs2 = 1;
uint8_t max_len_p2[4], max_len_q2[4];

for (int i = 0; i < DEBLOCK_STEP >> (2 - vs); i++)
{
int tx = x;
int ty = y + (i << 2);
const int end = ty >= y_end;

if (!vertical)
FFSWAP(int, tx, ty);

bs[i] = end ? 0 : TAB_BS(fc->tab.bs[vertical][c_idx], tx, ty);
if (bs[i])
{
const int qp = get_qp(fc, POS(c_idx, tx, ty), tx, ty, c_idx, vertical);
beta[i] = betatable[av_clip(qp + beta_offset, 0, MAX_QP)];
tc[i] = TC_CALC(qp, bs[i]);
max_filter_length(fc, tx, ty, c_idx, vertical, horizontal_ctu_edge, bs[i], &max_len_p[i], &max_len_q[i]);
all_zero_bs = 0;
}
}

for (int i = 0; i < DEBLOCK_STEP >> (2 - vs); i++)
{
int tx = x + grid;
int ty = y + (i << 2);
const int end = ty >= y_end;

if (!vertical)
FFSWAP(int, tx, ty);

bs2[i] = end ? 0 : TAB_BS(fc->tab.bs[vertical][c_idx], tx, ty);
if (bs2[i])
{
const int qp = get_qp(fc, POS(c_idx, tx, ty), tx, ty, c_idx, vertical);
beta2[i] = betatable[av_clip(qp + beta_offset, 0, MAX_QP)];
tc2[i] = TC_CALC(qp, bs2[i]);
max_filter_length(fc, tx, ty, c_idx, vertical, horizontal_ctu_edge2, bs2[i], &max_len_p2[i], &max_len_q2[i]);
all_zero_bs2 = 0;
}
}

if (!all_zero_bs)
{
uint8_t *src = vertical ? POS(c_idx, x, y) : POS(c_idx, y, x);

fc->vvcdsp.lf.filter_chroma[vertical](src, src_stride, beta, tc, no_p, no_q, max_len_p, max_len_q, vs);
}

if (!all_zero_bs2)
{
uint8_t *src = vertical ? POS(c_idx, x + grid, y) : POS(c_idx, y, x + grid);

fc->vvcdsp.lf.filter_chroma[vertical](src, src_stride, beta2, tc2, no_p, no_q, max_len_p2, max_len_q2, vs);
}
}
}
}
}

void ff_vvc_deblock_vertical(const VVCLocalContext *lc, const int x0, const int y0, const int rs)
Expand Down