virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2025-08-10 06:10:52 +02:00
Code Issues Releases Activity

hevc: simplify SAO computation, delay from one row its computation

Browse Source 
(cherry picked from commit f2c5f647cec786df26f442a85e6d685a131a50c9)

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This commit is contained in:
Mickaël Raulet
2014-07-15 10:23:20 +02:00
committed by Michael Niedermayer
parent d595361593
commit 1241eb8870
6 changed files with 191 additions and 450 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
libavcodec/hevc.c
View File

@@ -2092,7 +2092,7 @@ static int hls_decode_entry(AVCodecContext *avctxt, void *isFilterThread)
if (x_ctb + ctb_size >= s->sps->width && if (x_ctb + ctb_size >= s->sps->width &&
y_ctb + ctb_size >= s->sps->height) y_ctb + ctb_size >= s->sps->height)
ff_hevc_hls_filter(s, x_ctb, y_ctb); ff_hevc_hls_filter(s, x_ctb, y_ctb, ctb_size);
return ctb_addr_ts; return ctb_addr_ts;
} }
@@ -2167,7 +2167,7 @@ static int hls_decode_entry_wpp(AVCodecContext *avctxt, void *input_ctb_row, int
} }
if ((x_ctb+ctb_size) >= s->sps->width && (y_ctb+ctb_size) >= s->sps->height ) { if ((x_ctb+ctb_size) >= s->sps->width && (y_ctb+ctb_size) >= s->sps->height ) {
ff_hevc_hls_filter(s, x_ctb, y_ctb); ff_hevc_hls_filter(s, x_ctb, y_ctb, ctb_size);
ff_thread_report_progress2(s->avctx, ctb_row , thread, SHIFT_CTB_WPP); ff_thread_report_progress2(s->avctx, ctb_row , thread, SHIFT_CTB_WPP);
return ctb_addr_ts; return ctb_addr_ts;
} }

2
libavcodec/hevc.h
View File

@@ -1001,7 +1001,7 @@ void ff_hevc_deblocking_boundary_strengths(HEVCContext *s, int x0, int y0,
int log2_trafo_size); int log2_trafo_size);
int ff_hevc_cu_qp_delta_sign_flag(HEVCContext *s); int ff_hevc_cu_qp_delta_sign_flag(HEVCContext *s);
int ff_hevc_cu_qp_delta_abs(HEVCContext *s); int ff_hevc_cu_qp_delta_abs(HEVCContext *s);
void ff_hevc_hls_filter(HEVCContext *s, int x, int y); void ff_hevc_hls_filter(HEVCContext *s, int x, int y, int ctb_size);
void ff_hevc_hls_filters(HEVCContext *s, int x_ctb, int y_ctb, int ctb_size); void ff_hevc_hls_filters(HEVCContext *s, int x_ctb, int y_ctb, int ctb_size);
void ff_hevc_hls_residual_coding(HEVCContext *s, int x0, int y0, void ff_hevc_hls_residual_coding(HEVCContext *s, int x0, int y0,
int log2_trafo_size, enum ScanType scan_idx, int log2_trafo_size, enum ScanType scan_idx,

165
libavcodec/hevc_filter.c
View File

@@ -153,129 +153,97 @@ static void copy_CTB(uint8_t *dst, uint8_t *src,
static void sao_filter_CTB(HEVCContext *s, int x, int y) static void sao_filter_CTB(HEVCContext *s, int x, int y)
{ {
// TODO: This should be easily parallelizable int c_idx;
// TODO: skip CBs when (cu_transquant_bypass_flag || (pcm_loop_filter_disable_flag && pcm_flag))
int c_idx = 0;
int class = 1, class_index;
int edges[4]; // 0 left 1 top 2 right 3 bottom int edges[4]; // 0 left 1 top 2 right 3 bottom
SAOParams *sao[4];
int classes[4];
int x_shift = 0, y_shift = 0;
int x_ctb = x >> s->sps->log2_ctb_size; int x_ctb = x >> s->sps->log2_ctb_size;
int y_ctb = y >> s->sps->log2_ctb_size; int y_ctb = y >> s->sps->log2_ctb_size;
int ctb_addr_rs = y_ctb * s->sps->ctb_width + x_ctb; int ctb_addr_rs = y_ctb * s->sps->ctb_width + x_ctb;
int ctb_addr_ts = s->pps->ctb_addr_rs_to_ts[ctb_addr_rs]; int ctb_addr_ts = s->pps->ctb_addr_rs_to_ts[ctb_addr_rs];
SAOParams *sao = &CTB(s->sao, x_ctb, y_ctb);
// flags indicating unfilterable edges // flags indicating unfilterable edges
uint8_t vert_edge[] = { 0, 0, 0, 0 }; uint8_t vert_edge[] = { 0, 0 };
uint8_t horiz_edge[] = { 0, 0, 0, 0 }; uint8_t horiz_edge[] = { 0, 0 };
uint8_t diag_edge[] = { 0, 0, 0, 0 }; uint8_t diag_edge[] = { 0, 0, 0, 0 };
uint8_t lfase[3]; // current, above, left uint8_t lfase = CTB(s->filter_slice_edges, x_ctb, y_ctb);
uint8_t no_tile_filter = s->pps->tiles_enabled_flag && uint8_t no_tile_filter = s->pps->tiles_enabled_flag &&
!s->pps->loop_filter_across_tiles_enabled_flag; !s->pps->loop_filter_across_tiles_enabled_flag;
uint8_t restore = no_tile_filter || !lfase;
uint8_t left_tile_edge = 0; uint8_t left_tile_edge = 0;
uint8_t right_tile_edge = 0;
uint8_t up_tile_edge = 0; uint8_t up_tile_edge = 0;
uint8_t bottom_tile_edge = 0;
sao[0] = &CTB(s->sao, x_ctb, y_ctb);
edges[0] = x_ctb == 0; edges[0] = x_ctb == 0;
edges[1] = y_ctb == 0; edges[1] = y_ctb == 0;
edges[2] = x_ctb == s->sps->ctb_width - 1; edges[2] = x_ctb == s->sps->ctb_width - 1;
edges[3] = y_ctb == s->sps->ctb_height - 1; edges[3] = y_ctb == s->sps->ctb_height - 1;
lfase[0] = CTB(s->filter_slice_edges, x_ctb, y_ctb);
classes[0] = 0;
if (restore) {
if (!edges[0]) { if (!edges[0]) {
left_tile_edge = no_tile_filter && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs-1]]; left_tile_edge = no_tile_filter && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs-1]];
sao[class] = &CTB(s->sao, x_ctb - 1, y_ctb); vert_edge[0] = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb)) || left_tile_edge;
vert_edge[0] = (!lfase[0] && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb)) || left_tile_edge; }
vert_edge[2] = vert_edge[0]; if (!edges[2]) {
lfase[2] = CTB(s->filter_slice_edges, x_ctb - 1, y_ctb); right_tile_edge = no_tile_filter && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs+1]];
classes[class] = 2; vert_edge[1] = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb + 1, y_ctb)) || right_tile_edge;
class++;
x_shift = 8;
} }
if (!edges[1]) { if (!edges[1]) {
up_tile_edge = no_tile_filter && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs - s->sps->ctb_width]]; up_tile_edge = no_tile_filter && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs - s->sps->ctb_width]];
sao[class] = &CTB(s->sao, x_ctb, y_ctb - 1); horiz_edge[0] = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb, y_ctb - 1)) || up_tile_edge;
horiz_edge[0] = (!lfase[0] && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb, y_ctb - 1)) || up_tile_edge;
horiz_edge[1] = horiz_edge[0];
lfase[1] = CTB(s->filter_slice_edges, x_ctb, y_ctb - 1);
classes[class] = 1;
class++;
y_shift = 4;
if (!edges[0]) {
classes[class] = 3;
sao[class] = &CTB(s->sao, x_ctb - 1, y_ctb - 1);
class++;
// Tile check here is done current CTB row/col, not above/left like you'd expect,
//but that is because the tile boundary always extends through the whole pic
vert_edge[1] = (!lfase[1] && CTB(s->tab_slice_address, x_ctb, y_ctb - 1) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb - 1)) || left_tile_edge;
vert_edge[3] = vert_edge[1];
horiz_edge[2] = (!lfase[2] && CTB(s->tab_slice_address, x_ctb - 1, y_ctb) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb - 1)) || up_tile_edge;
horiz_edge[3] = horiz_edge[2];
diag_edge[0] = (!lfase[0] && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb - 1)) || left_tile_edge || up_tile_edge;
diag_edge[3] = diag_edge[0];
// Does left CTB comes after above CTB?
if (CTB(s->tab_slice_address, x_ctb - 1, y_ctb) >
CTB(s->tab_slice_address, x_ctb, y_ctb - 1)) {
diag_edge[2] = !lfase[2] || left_tile_edge || up_tile_edge;
diag_edge[1] = diag_edge[2];
} else if (CTB(s->tab_slice_address, x_ctb - 1, y_ctb) <
CTB(s->tab_slice_address, x_ctb, y_ctb - 1)) {
diag_edge[1] = !lfase[1] || left_tile_edge || up_tile_edge;
diag_edge[2] = diag_edge[1];
} else {
// Same slice, only consider tiles
diag_edge[2] = left_tile_edge || up_tile_edge;
diag_edge[1] = diag_edge[2];
} }
if (!edges[3]) {
bottom_tile_edge = no_tile_filter && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs + s->sps->ctb_width]];
horiz_edge[1] = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb, y_ctb + 1)) || bottom_tile_edge;
}
if (!edges[0] && !edges[1]) {
diag_edge[0] = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb - 1)) || left_tile_edge || up_tile_edge;
}
if (!edges[1] && !edges[2]) {
diag_edge[1] = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb + 1, y_ctb - 1)) || right_tile_edge || up_tile_edge;
}
if (!edges[2] && !edges[3]) {
diag_edge[2] = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb + 1, y_ctb + 1)) || right_tile_edge || bottom_tile_edge;
}
if (!edges[0] && !edges[3]) {
diag_edge[3] = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb + 1)) || left_tile_edge || bottom_tile_edge;
} }
} }
for (c_idx = 0; c_idx < 3; c_idx++) { for (c_idx = 0; c_idx < 3; c_idx++) {
int chroma = c_idx ? 1 : 0; int x0 = x >> s->sps->hshift[c_idx];
int x0 = x >> chroma; int y0 = y >> s->sps->vshift[c_idx];
int y0 = y >> chroma;
int stride = s->frame->linesize[c_idx]; int stride = s->frame->linesize[c_idx];
int ctb_size = (1 << (s->sps->log2_ctb_size)) >> s->sps->hshift[c_idx]; int ctb_size_h = (1 << (s->sps->log2_ctb_size)) >> s->sps->hshift[c_idx];
int width = FFMIN(ctb_size, int ctb_size_v = (1 << (s->sps->log2_ctb_size)) >> s->sps->vshift[c_idx];
int width = FFMIN(ctb_size_h,
(s->sps->width >> s->sps->hshift[c_idx]) - x0); (s->sps->width >> s->sps->hshift[c_idx]) - x0);
int height = FFMIN(ctb_size, int height = FFMIN(ctb_size_v,
(s->sps->height >> s->sps->vshift[c_idx]) - y0); (s->sps->height >> s->sps->vshift[c_idx]) - y0);
uint8_t *src = &s->frame->data[c_idx][y0 * stride + (x0 << s->sps->pixel_shift)]; uint8_t *src = &s->frame->data[c_idx][y0 * stride + (x0 << s->sps->pixel_shift)];
uint8_t *dst = &s->sao_frame->data[c_idx][y0 * stride + (x0 << s->sps->pixel_shift)]; uint8_t *dst = &s->sao_frame->data[c_idx][y0 * stride + (x0 << s->sps->pixel_shift)];
int offset = (y_shift >> chroma) * stride + ((x_shift >> chroma) << s->sps->pixel_shift);
copy_CTB(dst - offset, src - offset, switch (sao->type_idx[c_idx]) {
(edges[2] ? width + (x_shift >> chroma) : width) << s->sps->pixel_shift,
(edges[3] ? height + (y_shift >> chroma) : height), stride);
for (class_index = 0; class_index < class; class_index++) {
switch (sao[class_index]->type_idx[c_idx]) {
case SAO_BAND: case SAO_BAND:
s->hevcdsp.sao_band_filter[classes[class_index]](dst, src, s->hevcdsp.sao_band_filter(dst, src,
stride, stride,
sao[class_index], sao,
edges, width, edges, width,
height, c_idx); height, c_idx);
break; break;
case SAO_EDGE: case SAO_EDGE:
s->hevcdsp.sao_edge_filter[classes[class_index]](dst, src, s->hevcdsp.sao_edge_filter[restore](dst, src,
stride, stride,
sao[class_index], sao,
edges, width, edges, width,
height, c_idx, height, c_idx,
vert_edge[classes[class_index]], vert_edge,
horiz_edge[classes[class_index]], horiz_edge,
diag_edge[classes[class_index]]); diag_edge);
break;
default :
copy_CTB(dst, src, width << s->sps->pixel_shift, height, stride);
break; break;
}
} }
} }
} }
@@ -661,22 +629,41 @@ void ff_hevc_deblocking_boundary_strengths(HEVCContext *s, int x0, int y0,
#undef CB #undef CB
#undef CR #undef CR
void ff_hevc_hls_filter(HEVCContext *s, int x, int y) void ff_hevc_hls_filter(HEVCContext *s, int x, int y, int ctb_size)
{ {
deblocking_filter_CTB(s, x, y); deblocking_filter_CTB(s, x, y);
if (s->sps->sao_enabled) if (s->sps->sao_enabled) {
int x_end = x >= s->sps->width - ctb_size;
int y_end = y >= s->sps->height - ctb_size;
if (y && x)
sao_filter_CTB(s, x - ctb_size, y - ctb_size);
if (x && y_end)
sao_filter_CTB(s, x - ctb_size, y);
if (y && x_end) {
sao_filter_CTB(s, x, y - ctb_size);
if (s->threads_type & FF_THREAD_FRAME )
ff_thread_report_progress(&s->ref->tf, y - ctb_size, 0);
}
if (x_end && y_end) {
sao_filter_CTB(s, x , y); sao_filter_CTB(s, x , y);
if (s->threads_type & FF_THREAD_FRAME )
ff_thread_report_progress(&s->ref->tf, y, 0);
}
} else {
if (y && x >= s->sps->width - ctb_size)
if (s->threads_type & FF_THREAD_FRAME )
ff_thread_report_progress(&s->ref->tf, y, 0);
}
} }
void ff_hevc_hls_filters(HEVCContext *s, int x_ctb, int y_ctb, int ctb_size) void ff_hevc_hls_filters(HEVCContext *s, int x_ctb, int y_ctb, int ctb_size)
{ {
int x_end = x_ctb >= s->sps->width - ctb_size;
int y_end = y_ctb >= s->sps->height - ctb_size;
if (y_ctb && x_ctb) if (y_ctb && x_ctb)
ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb - ctb_size); ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb - ctb_size, ctb_size);
if (y_ctb && x_ctb >= s->sps->width - ctb_size) { if (y_ctb && x_end)
ff_hevc_hls_filter(s, x_ctb, y_ctb - ctb_size); ff_hevc_hls_filter(s, x_ctb, y_ctb - ctb_size, ctb_size);
if (s->threads_type == FF_THREAD_FRAME ) if (x_ctb && y_end)
ff_thread_report_progress(&s->ref->tf, y_ctb - ctb_size, 0); ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb, ctb_size);
}
if (x_ctb && y_ctb >= s->sps->height - ctb_size)
ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb);
} }

8
libavcodec/hevcdsp.c
View File

@@ -207,15 +207,9 @@ void ff_hevc_dsp_init(HEVCDSPContext *hevcdsp, int bit_depth)
hevcdsp->transform_dc_add[2] = FUNC(transform_16x16_dc_add, depth); \ hevcdsp->transform_dc_add[2] = FUNC(transform_16x16_dc_add, depth); \
hevcdsp->transform_dc_add[3] = FUNC(transform_32x32_dc_add, depth); \ hevcdsp->transform_dc_add[3] = FUNC(transform_32x32_dc_add, depth); \
\ \
hevcdsp->sao_band_filter[0] = FUNC(sao_band_filter_0, depth); \ hevcdsp->sao_band_filter = FUNC(sao_band_filter_0, depth); \
hevcdsp->sao_band_filter[1] = FUNC(sao_band_filter_1, depth); \
hevcdsp->sao_band_filter[2] = FUNC(sao_band_filter_2, depth); \
hevcdsp->sao_band_filter[3] = FUNC(sao_band_filter_3, depth); \
\
hevcdsp->sao_edge_filter[0] = FUNC(sao_edge_filter_0, depth); \ hevcdsp->sao_edge_filter[0] = FUNC(sao_edge_filter_0, depth); \
hevcdsp->sao_edge_filter[1] = FUNC(sao_edge_filter_1, depth); \ hevcdsp->sao_edge_filter[1] = FUNC(sao_edge_filter_1, depth); \
hevcdsp->sao_edge_filter[2] = FUNC(sao_edge_filter_2, depth); \
hevcdsp->sao_edge_filter[3] = FUNC(sao_edge_filter_3, depth); \
\ \
QPEL_FUNCS(depth); \ QPEL_FUNCS(depth); \
QPEL_UNI_FUNCS(depth); \ QPEL_UNI_FUNCS(depth); \

11
libavcodec/hevcdsp.h
View File

@@ -54,13 +54,14 @@ typedef struct HEVCDSPContext {
void (*transform_dc_add[4])(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride); void (*transform_dc_add[4])(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void (*sao_band_filter[4])(uint8_t *dst, uint8_t *src, ptrdiff_t stride, void (*sao_band_filter)(uint8_t *_dst, uint8_t *_src, ptrdiff_t _stride,
struct SAOParams *sao, int *borders, struct SAOParams *sao, int *borders,
int width, int height, int c_idx); int width, int height, int c_idx);
void (*sao_edge_filter[4])(uint8_t *dst, uint8_t *src, ptrdiff_t stride,
struct SAOParams *sao, int *borders, int width, void (*sao_edge_filter[2])(uint8_t *_dst, uint8_t *_src, ptrdiff_t _stride,
int height, int c_idx, uint8_t vert_edge, struct SAOParams *sao, int *borders, int _width,
uint8_t horiz_edge, uint8_t diag_edge); int _height, int c_idx, uint8_t *vert_edge,
uint8_t *horiz_edge, uint8_t *diag_edge);
void (*put_hevc_qpel[10][2][2])(int16_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride, void (*put_hevc_qpel[10][2][2])(int16_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride,
int height, intptr_t mx, intptr_t my, int width); int height, intptr_t mx, intptr_t my, int width);

385
libavcodec/hevcdsp_template.c
View File

@@ -294,133 +294,87 @@ TRANSFORM_DC_ADD( 8)
TRANSFORM_DC_ADD(16) TRANSFORM_DC_ADD(16)
TRANSFORM_DC_ADD(32) TRANSFORM_DC_ADD(32)
static void FUNC(sao_band_filter_0)(uint8_t *_dst, uint8_t *_src,
static void FUNC(sao_band_filter)(uint8_t *_dst, uint8_t *_src,
ptrdiff_t stride, SAOParams *sao, ptrdiff_t stride, SAOParams *sao,
int *borders, int width, int height, int *borders, int width, int height,
int c_idx, int class) int c_idx)
{ {
pixel *dst = (pixel *)_dst; pixel *dst = (pixel *)_dst;
pixel *src = (pixel *)_src; pixel *src = (pixel *)_src;
int offset_table[32] = { 0 }; int offset_table[32] = { 0 };
int k, y, x; int k, y, x;
int chroma = !!c_idx;
int shift = BIT_DEPTH - 5; int shift = BIT_DEPTH - 5;
int *sao_offset_val = sao->offset_val[c_idx]; int *sao_offset_val = sao->offset_val[c_idx];
int sao_left_class = sao->band_position[c_idx]; int sao_left_class = sao->band_position[c_idx];
int init_y = 0, init_x = 0;
stride /= sizeof(pixel); stride /= sizeof(pixel);
switch (class) {
case 0:
if (!borders[2])
width -= (8 >> chroma) + 2;
if (!borders[3])
height -= (4 >> chroma) + 2;
break;
case 1:
init_y = -(4 >> chroma) - 2;
if (!borders[2])
width -= (8 >> chroma) + 2;
height = (4 >> chroma) + 2;
break;
case 2:
init_x = -(8 >> chroma) - 2;
width = (8 >> chroma) + 2;
if (!borders[3])
height -= (4 >> chroma) + 2;
break;
case 3:
init_y = -(4 >> chroma) - 2;
init_x = -(8 >> chroma) - 2;
width = (8 >> chroma) + 2;
height = (4 >> chroma) + 2;
break;
}
dst = dst + (init_y * stride + init_x);
src = src + (init_y * stride + init_x);
for (k = 0; k < 4; k++) for (k = 0; k < 4; k++)
offset_table[(k + sao_left_class) & 31] = sao_offset_val[k + 1]; offset_table[(k + sao_left_class) & 31] = sao_offset_val[k + 1];
for (y = 0; y < height; y++) { for (y = 0; y < height; y++) {
for (x = 0; x < width; x++) for (x = 0; x < width; x++)
dst[x] = av_clip_pixel(src[x] + offset_table[src[x] >> shift]); dst[x] = av_clip_pixel(src[x] + offset_table[av_clip_pixel(src[x] >> shift)]);
dst += stride; dst += stride;
src += stride; src += stride;
} }
} }
static void FUNC(sao_band_filter_0)(uint8_t *dst, uint8_t *src, #define CMP(a, b) ((a) > (b) ? 1 : ((a) == (b) ? 0 : -1))
ptrdiff_t stride, SAOParams *sao,
int *borders, int width, int height,
int c_idx)
{
FUNC(sao_band_filter)(dst, src, stride, sao, borders,
width, height, c_idx, 0);
}
static void FUNC(sao_band_filter_1)(uint8_t *dst, uint8_t *src, static void FUNC(sao_edge_filter)(uint8_t *_dst, uint8_t *_src,
ptrdiff_t stride, SAOParams *sao, ptrdiff_t stride, SAOParams *sao,
int *borders, int width, int height, int width, int height,
int c_idx) int c_idx, int init_x, int init_y) {
{
FUNC(sao_band_filter)(dst, src, stride, sao, borders,
width, height, c_idx, 1);
}
static void FUNC(sao_band_filter_2)(uint8_t *dst, uint8_t *src, static const uint8_t edge_idx[] = { 1, 2, 0, 3, 4 };
ptrdiff_t stride, SAOParams *sao, static const int8_t pos[4][2][2] = {
int *borders, int width, int height, { { -1, 0 }, { 1, 0 } }, // horizontal
int c_idx) { { 0, -1 }, { 0, 1 } }, // vertical
{ { { -1, -1 }, { 1, 1 } }, // 45 degree
FUNC(sao_band_filter)(dst, src, stride, sao, borders, { { 1, -1 }, { -1, 1 } }, // 135 degree
width, height, c_idx, 2); };
} int *sao_offset_val = sao->offset_val[c_idx];
int sao_eo_class = sao->eo_class[c_idx];
pixel *dst = (pixel *)_dst;
pixel *src = (pixel *)_src;
static void FUNC(sao_band_filter_3)(uint8_t *_dst, uint8_t *_src, int y_stride = init_y * stride;
ptrdiff_t stride, SAOParams *sao, int pos_0_0 = pos[sao_eo_class][0][0];
int *borders, int width, int height, int pos_0_1 = pos[sao_eo_class][0][1];
int c_idx) int pos_1_0 = pos[sao_eo_class][1][0];
{ int pos_1_1 = pos[sao_eo_class][1][1];
FUNC(sao_band_filter)(_dst, _src, stride, sao, borders, int x, y;
width, height, c_idx, 3);
int y_stride_0_1 = (init_y + pos_0_1) * stride;
int y_stride_1_1 = (init_y + pos_1_1) * stride;
for (y = init_y; y < height; y++) {
for (x = init_x; x < width; x++) {
int diff0 = CMP(src[x + y_stride], src[x + pos_0_0 + y_stride_0_1]);
int diff1 = CMP(src[x + y_stride], src[x + pos_1_0 + y_stride_1_1]);
int offset_val = edge_idx[2 + diff0 + diff1];
dst[x + y_stride] = av_clip_pixel(src[x + y_stride] + sao_offset_val[offset_val]);
}
y_stride += stride;
y_stride_0_1 += stride;
y_stride_1_1 += stride;
}
} }
static void FUNC(sao_edge_filter_0)(uint8_t *_dst, uint8_t *_src, static void FUNC(sao_edge_filter_0)(uint8_t *_dst, uint8_t *_src,
ptrdiff_t stride, SAOParams *sao, ptrdiff_t stride, SAOParams *sao,
int *borders, int _width, int _height, int *borders, int _width, int _height,
int c_idx, uint8_t vert_edge, int c_idx, uint8_t *vert_edge,
uint8_t horiz_edge, uint8_t diag_edge) uint8_t *horiz_edge, uint8_t *diag_edge)
{ {
int x, y; int x, y;
pixel *dst = (pixel *)_dst; pixel *dst = (pixel *)_dst;
pixel *src = (pixel *)_src; pixel *src = (pixel *)_src;
int chroma = !!c_idx;
int *sao_offset_val = sao->offset_val[c_idx]; int *sao_offset_val = sao->offset_val[c_idx];
int sao_eo_class = sao->eo_class[c_idx]; int sao_eo_class = sao->eo_class[c_idx];
int init_x = 0, init_y = 0, width = _width, height = _height; int init_x = 0, init_y = 0, width = _width, height = _height;
static const int8_t pos[4][2][2] = {
{ { -1, 0 }, { 1, 0 } }, // horizontal
{ { 0, -1 }, { 0, 1 } }, // vertical
{ { -1, -1 }, { 1, 1 } }, // 45 degree
{ { 1, -1 }, { -1, 1 } }, // 135 degree
};
static const uint8_t edge_idx[] = { 1, 2, 0, 3, 4 };
#define CMP(a, b) ((a) > (b) ? 1 : ((a) == (b) ? 0 : -1))
stride /= sizeof(pixel); stride /= sizeof(pixel);
if (!borders[2])
width -= (8 >> chroma) + 2;
if (!borders[3])
height -= (4 >> chroma) + 2;
dst = dst + (init_y * stride + init_x);
src = src + (init_y * stride + init_x);
init_y = init_x = 0;
if (sao_eo_class != SAO_EO_VERT) { if (sao_eo_class != SAO_EO_VERT) {
if (borders[0]) { if (borders[0]) {
int offset_val = sao_offset_val[0]; int offset_val = sao_offset_val[0];
@@ -456,78 +410,25 @@ static void FUNC(sao_edge_filter_0)(uint8_t *_dst, uint8_t *_src,
height--; height--;
} }
} }
{
int y_stride = init_y * stride;
int pos_0_0 = pos[sao_eo_class][0][0];
int pos_0_1 = pos[sao_eo_class][0][1];
int pos_1_0 = pos[sao_eo_class][1][0];
int pos_1_1 = pos[sao_eo_class][1][1];
int y_stride_0_1 = (init_y + pos_0_1) * stride; FUNC(sao_edge_filter)((uint8_t *)dst, (uint8_t *)src, stride, sao, width, height, c_idx, init_x, init_y);
int y_stride_1_1 = (init_y + pos_1_1) * stride;
for (y = init_y; y < height; y++) {
for (x = init_x; x < width; x++) {
int diff0 = CMP(src[x + y_stride], src[x + pos_0_0 + y_stride_0_1]);
int diff1 = CMP(src[x + y_stride], src[x + pos_1_0 + y_stride_1_1]);
int offset_val = edge_idx[2 + diff0 + diff1];
dst[x + y_stride] = av_clip_pixel(src[x + y_stride] + sao_offset_val[offset_val]);
}
y_stride += stride;
y_stride_0_1 += stride;
y_stride_1_1 += stride;
}
}
{
// Restore pixels that can't be modified
int save_upper_left = !diag_edge && sao_eo_class == SAO_EO_135D && !borders[0] && !borders[1];
if (vert_edge && sao_eo_class != SAO_EO_VERT)
for (y = init_y+save_upper_left; y< height; y++)
dst[y*stride] = src[y*stride];
if(horiz_edge && sao_eo_class != SAO_EO_HORIZ)
for(x = init_x+save_upper_left; x<width; x++)
dst[x] = src[x];
if(diag_edge && sao_eo_class == SAO_EO_135D)
dst[0] = src[0];
}
#undef CMP
} }
static void FUNC(sao_edge_filter_1)(uint8_t *_dst, uint8_t *_src, static void FUNC(sao_edge_filter_1)(uint8_t *_dst, uint8_t *_src,
ptrdiff_t stride, SAOParams *sao, ptrdiff_t stride, SAOParams *sao,
int *borders, int _width, int _height, int *borders, int _width, int _height,
int c_idx, uint8_t vert_edge, int c_idx, uint8_t *vert_edge,
uint8_t horiz_edge, uint8_t diag_edge) uint8_t *horiz_edge, uint8_t *diag_edge)
{ {
int x, y; int x, y;
pixel *dst = (pixel *)_dst; pixel *dst = (pixel *)_dst;
pixel *src = (pixel *)_src; pixel *src = (pixel *)_src;
int chroma = !!c_idx;
int *sao_offset_val = sao->offset_val[c_idx]; int *sao_offset_val = sao->offset_val[c_idx];
int sao_eo_class = sao->eo_class[c_idx]; int sao_eo_class = sao->eo_class[c_idx];
int init_x = 0, init_y = 0, width = _width, height = _height; int init_x = 0, init_y = 0, width = _width, height = _height;
static const int8_t pos[4][2][2] = {
{ { -1, 0 }, { 1, 0 } }, // horizontal
{ { 0, -1 }, { 0, 1 } }, // vertical
{ { -1, -1 }, { 1, 1 } }, // 45 degree
{ { 1, -1 }, { -1, 1 } }, // 135 degree
};
static const uint8_t edge_idx[] = { 1, 2, 0, 3, 4 };
#define CMP(a, b) ((a) > (b) ? 1 : ((a) == (b) ? 0 : -1))
stride /= sizeof(pixel); stride /= sizeof(pixel);
init_y = -(4 >> chroma) - 2;
if (!borders[2])
width -= (8 >> chroma) + 2;
height = (4 >> chroma) + 2;
dst = dst + (init_y * stride + init_x);
src = src + (init_y * stride + init_x);
init_y = init_x = 0;
if (sao_eo_class != SAO_EO_VERT) { if (sao_eo_class != SAO_EO_VERT) {
if (borders[0]) { if (borders[0]) {
int offset_val = sao_offset_val[0]; int offset_val = sao_offset_val[0];
@@ -548,78 +449,6 @@ static void FUNC(sao_edge_filter_1)(uint8_t *_dst, uint8_t *_src,
width--; width--;
} }
} }
{
int y_stride = init_y * stride;
int pos_0_0 = pos[sao_eo_class][0][0];
int pos_0_1 = pos[sao_eo_class][0][1];
int pos_1_0 = pos[sao_eo_class][1][0];
int pos_1_1 = pos[sao_eo_class][1][1];
int y_stride_0_1 = (init_y + pos_0_1) * stride;
int y_stride_1_1 = (init_y + pos_1_1) * stride;
for (y = init_y; y < height; y++) {
for (x = init_x; x < width; x++) {
int diff0 = CMP(src[x + y_stride], src[x + pos_0_0 + y_stride_0_1]);
int diff1 = CMP(src[x + y_stride], src[x + pos_1_0 + y_stride_1_1]);
int offset_val = edge_idx[2 + diff0 + diff1];
dst[x + y_stride] = av_clip_pixel(src[x + y_stride] + sao_offset_val[offset_val]);
}
y_stride += stride;
y_stride_0_1 += stride;
y_stride_1_1 += stride;
}
}
{
// Restore pixels that can't be modified
int save_lower_left = !diag_edge && sao_eo_class == SAO_EO_45D && !borders[0];
if(vert_edge && sao_eo_class != SAO_EO_VERT)
for(y = init_y; y< height-save_lower_left; y++)
dst[y*stride] = src[y*stride];
if(horiz_edge && sao_eo_class != SAO_EO_HORIZ)
for(x = init_x+save_lower_left; x<width; x++)
dst[(height-1)*stride+x] = src[(height-1)*stride+x];
if(diag_edge && sao_eo_class == SAO_EO_45D)
dst[stride*(height-1)] = src[stride*(height-1)];
}
#undef CMP
}
static void FUNC(sao_edge_filter_2)(uint8_t *_dst, uint8_t *_src,
ptrdiff_t stride, SAOParams *sao,
int *borders, int _width, int _height,
int c_idx, uint8_t vert_edge,
uint8_t horiz_edge, uint8_t diag_edge)
{
int x, y;
pixel *dst = (pixel *)_dst;
pixel *src = (pixel *)_src;
int chroma = !!c_idx;
int *sao_offset_val = sao->offset_val[c_idx];
int sao_eo_class = sao->eo_class[c_idx];
int init_x = 0, init_y = 0, width = _width, height = _height;
static const int8_t pos[4][2][2] = {
{ { -1, 0 }, { 1, 0 } }, // horizontal
{ { 0, -1 }, { 0, 1 } }, // vertical
{ { -1, -1 }, { 1, 1 } }, // 45 degree
{ { 1, -1 }, { -1, 1 } }, // 135 degree
};
static const uint8_t edge_idx[] = { 1, 2, 0, 3, 4 };
#define CMP(a, b) ((a) > (b) ? 1 : ((a) == (b) ? 0 : -1))
stride /= sizeof(pixel);
init_x = -(8 >> chroma) - 2;
width = (8 >> chroma) + 2;
if (!borders[3])
height -= (4 >> chroma) + 2;
dst = dst + (init_y * stride + init_x);
src = src + (init_y * stride + init_x);
init_y = init_x = 0;
if (sao_eo_class != SAO_EO_HORIZ) { if (sao_eo_class != SAO_EO_HORIZ) {
if (borders[1]) { if (borders[1]) {
int offset_val = sao_offset_val[0]; int offset_val = sao_offset_val[0];
@@ -635,116 +464,46 @@ static void FUNC(sao_edge_filter_2)(uint8_t *_dst, uint8_t *_src,
height--; height--;
} }
} }
{
int y_stride = init_y * stride;
int pos_0_0 = pos[sao_eo_class][0][0];
int pos_0_1 = pos[sao_eo_class][0][1];
int pos_1_0 = pos[sao_eo_class][1][0];
int pos_1_1 = pos[sao_eo_class][1][1];
int y_stride_0_1 = (init_y + pos_0_1) * stride; FUNC(sao_edge_filter)((uint8_t *)dst, (uint8_t *)src, stride, sao, width, height, c_idx, init_x, init_y);
int y_stride_1_1 = (init_y + pos_1_1) * stride;
for (y = init_y; y < height; y++) {
for (x = init_x; x < width; x++) {
int diff0 = CMP(src[x + y_stride], src[x + pos_0_0 + y_stride_0_1]);
int diff1 = CMP(src[x + y_stride], src[x + pos_1_0 + y_stride_1_1]);
int offset_val = edge_idx[2 + diff0 + diff1];
dst[x + y_stride] = av_clip_pixel(src[x + y_stride] + sao_offset_val[offset_val]);
}
y_stride += stride;
y_stride_0_1 += stride;
y_stride_1_1 += stride;
}
}
{ {
int save_upper_left = !diag_edge[0] && sao_eo_class == SAO_EO_135D && !borders[0] && !borders[1];
int save_upper_right = !diag_edge[1] && sao_eo_class == SAO_EO_45D && !borders[1] && !borders[2];
int save_lower_right = !diag_edge[2] && sao_eo_class == SAO_EO_135D && !borders[2] && !borders[3];
int save_lower_left = !diag_edge[3] && sao_eo_class == SAO_EO_45D && !borders[0] && !borders[3];
// Restore pixels that can't be modified // Restore pixels that can't be modified
int save_upper_right = !diag_edge && sao_eo_class == SAO_EO_45D && !borders[1]; if(vert_edge[0] && sao_eo_class != SAO_EO_VERT) {
if(vert_edge && sao_eo_class != SAO_EO_VERT) for(y = init_y+save_upper_left; y< height-save_lower_left; y++)
for(y = init_y+save_upper_right; y< height; y++) dst[y*stride] = src[y*stride];
}
if(vert_edge[1] && sao_eo_class != SAO_EO_VERT) {
for(y = init_y+save_upper_right; y< height-save_lower_right; y++)
dst[y*stride+width-1] = src[y*stride+width-1]; dst[y*stride+width-1] = src[y*stride+width-1];
if(horiz_edge && sao_eo_class != SAO_EO_HORIZ) }
for(x = init_x; x<width-save_upper_right; x++)
if(horiz_edge[0] && sao_eo_class != SAO_EO_HORIZ) {
for(x = init_x+save_upper_left; x < width-save_upper_right; x++)
dst[x] = src[x]; dst[x] = src[x];
if(diag_edge && sao_eo_class == SAO_EO_45D)
dst[width-1] = src[width-1];
} }
#undef CMP if(horiz_edge[1] && sao_eo_class != SAO_EO_HORIZ) {
} for(x = init_x+save_lower_left; x < width-save_lower_right; x++)
static void FUNC(sao_edge_filter_3)(uint8_t *_dst, uint8_t *_src,
ptrdiff_t stride, SAOParams *sao,
int *borders, int _width, int _height,
int c_idx, uint8_t vert_edge,
uint8_t horiz_edge, uint8_t diag_edge)
{
int x, y;
pixel *dst = (pixel *)_dst;
pixel *src = (pixel *)_src;
int chroma = !!c_idx;
int *sao_offset_val = sao->offset_val[c_idx];
int sao_eo_class = sao->eo_class[c_idx];
int init_x = 0, init_y = 0, width = _width, height = _height;
static const int8_t pos[4][2][2] = {
{ { -1, 0 }, { 1, 0 } }, // horizontal
{ { 0, -1 }, { 0, 1 } }, // vertical
{ { -1, -1 }, { 1, 1 } }, // 45 degree
{ { 1, -1 }, { -1, 1 } }, // 135 degree
};
static const uint8_t edge_idx[] = { 1, 2, 0, 3, 4 };
#define CMP(a, b) ((a) > (b) ? 1 : ((a) == (b) ? 0 : -1))
stride /= sizeof(pixel);
init_y = -(4 >> chroma) - 2;
init_x = -(8 >> chroma) - 2;
width = (8 >> chroma) + 2;
height = (4 >> chroma) + 2;
dst = dst + (init_y * stride + init_x);
src = src + (init_y * stride + init_x);
init_y = init_x = 0;
{
int y_stride = init_y * stride;
int pos_0_0 = pos[sao_eo_class][0][0];
int pos_0_1 = pos[sao_eo_class][0][1];
int pos_1_0 = pos[sao_eo_class][1][0];
int pos_1_1 = pos[sao_eo_class][1][1];
int y_stride_0_1 = (init_y + pos_0_1) * stride;
int y_stride_1_1 = (init_y + pos_1_1) * stride;
for (y = init_y; y < height; y++) {
for (x = init_x; x < width; x++) {
int diff0 = CMP(src[x + y_stride], src[x + pos_0_0 + y_stride_0_1]);
int diff1 = CMP(src[x + y_stride], src[x + pos_1_0 + y_stride_1_1]);
int offset_val = edge_idx[2 + diff0 + diff1];
dst[x + y_stride] = av_clip_pixel(src[x + y_stride] + sao_offset_val[offset_val]);
}
y_stride += stride;
y_stride_0_1 += stride;
y_stride_1_1 += stride;
}
}
{
// Restore pixels that can't be modified
int save_lower_right = !diag_edge && sao_eo_class == SAO_EO_135D;
if(vert_edge && sao_eo_class != SAO_EO_VERT)
for(y = init_y; y< height-save_lower_right; y++)
dst[y*stride+width-1] = src[y*stride+width-1];
if(horiz_edge && sao_eo_class != SAO_EO_HORIZ)
for(x = init_x; x<width-save_lower_right; x++)
dst[(height-1)*stride+x] = src[(height-1)*stride+x]; dst[(height-1)*stride+x] = src[(height-1)*stride+x];
if(diag_edge && sao_eo_class == SAO_EO_135D) }
if(diag_edge[0] && sao_eo_class == SAO_EO_135D)
dst[0] = src[0];
if(diag_edge[1] && sao_eo_class == SAO_EO_45D)
dst[width-1] = src[width-1];
if(diag_edge[2] && sao_eo_class == SAO_EO_135D)
dst[stride*(height-1)+width-1] = src[stride*(height-1)+width-1]; dst[stride*(height-1)+width-1] = src[stride*(height-1)+width-1];
if(diag_edge[3] && sao_eo_class == SAO_EO_45D)
dst[stride*(height-1)] = src[stride*(height-1)];
} }
}
#undef CMP #undef CMP
}
#undef SET #undef SET
#undef SCALE #undef SCALE