Skip to content

F
ffmpeg
Commit d8edf1b5 authored by Mans Rullgard's avatar Mans Rullgard
Browse files
Options

rv40: rearrange loop filter functions 

This splits the loop filter functions into smaller, more SIMD-friendly
functions.
Signed-off-by: 's avatarMans Rullgard <mans@mansr.com>
parent 3e0b1c9a
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 64 additions and 26 deletions
libavcodec/rv34dsp.h
View file @ d8edf1b5
...@@ -36,10 +36,18 @@ typedef void (*rv40_weight_func)(uint8_t *dst/*align width (8 or 16)*/, ...@@ -36,10 +36,18 @@ typedef void (*rv40_weight_func)(uint8_t *dst/*align width (8 or 16)*/,
typedef void (*rv34_inv_transform_func)(DCTELEM *block); typedef void (*rv34_inv_transform_func)(DCTELEM *block);
typedef void (*rv40_loop_filter_func)(uint8_t *src, int stride, int dmode, typedef void (*rv40_weak_loop_filter_func)(uint8_t *src, int stride,
int lim_q1, int lim_p1, int alpha, int filter_p1, int filter_q1,
int beta, int beta2, int chroma, int alpha, int beta,
int edge); int lims, int lim_q1, int lim_p1);
typedef void (*rv40_strong_loop_filter_func)(uint8_t *src, int stride,
int alpha, int lims,
int dmode, int chroma);
typedef int (*rv40_loop_filter_strength_func)(uint8_t *src, int stride,
int beta, int beta2, int edge,
int *p1, int *q1);
typedef struct RV34DSPContext { typedef struct RV34DSPContext {
qpel_mc_func put_pixels_tab[4][16]; qpel_mc_func put_pixels_tab[4][16];
...@@ -49,8 +57,9 @@ typedef struct RV34DSPContext { ...@@ -49,8 +57,9 @@ typedef struct RV34DSPContext {
rv40_weight_func rv40_weight_pixels_tab[2]; rv40_weight_func rv40_weight_pixels_tab[2];
rv34_inv_transform_func rv34_inv_transform_tab[2]; rv34_inv_transform_func rv34_inv_transform_tab[2];
void (*rv34_dequant4x4)(DCTELEM *block, int Qdc, int Q); void (*rv34_dequant4x4)(DCTELEM *block, int Qdc, int Q);
rv40_loop_filter_func rv40_h_loop_filter; rv40_weak_loop_filter_func rv40_weak_loop_filter[2];
rv40_loop_filter_func rv40_v_loop_filter; rv40_strong_loop_filter_func rv40_strong_loop_filter[2];
rv40_loop_filter_strength_func rv40_loop_filter_strength[2];
} RV34DSPContext; } RV34DSPContext;
void ff_rv30dsp_init(RV34DSPContext *c, DSPContext* dsp); void ff_rv30dsp_init(RV34DSPContext *c, DSPContext* dsp);
......
libavcodec/rv40.c
View file @ d8edf1b5
...@@ -294,6 +294,34 @@ enum RV40BlockPos{ ...@@ -294,6 +294,34 @@ enum RV40BlockPos{
static const int neighbour_offs_x[4] = { 0, 0, -1, 0 }; static const int neighbour_offs_x[4] = { 0, 0, -1, 0 };
static const int neighbour_offs_y[4] = { 0, -1, 0, 1 }; static const int neighbour_offs_y[4] = { 0, -1, 0, 1 };
static void rv40_adaptive_loop_filter(RV34DSPContext *rdsp,
uint8_t *src, int stride, int dmode,
int lim_q1, int lim_p1,
int alpha, int beta, int beta2,
int chroma, int edge, int dir)
{
int filter_p1, filter_q1;
int strong;
int lims;
strong = rdsp->rv40_loop_filter_strength[dir](src, stride, beta, beta2,
edge, &filter_p1, &filter_q1);
lims = filter_p1 + filter_q1 + ((lim_q1 + lim_p1) >> 1) + 1;
if (strong) {
rdsp->rv40_strong_loop_filter[dir](src, stride, alpha,
lims, dmode, chroma);
} else if (filter_p1 & filter_q1) {
rdsp->rv40_weak_loop_filter[dir](src, stride, 1, 1, alpha, beta,
lims, lim_q1, lim_p1);
} else if (filter_p1 | filter_q1) {
rdsp->rv40_weak_loop_filter[dir](src, stride, filter_p1, filter_q1,
alpha, beta, lims >> 1, lim_q1 >> 1,
lim_p1 >> 1);
}
}
/** /**
* RV40 loop filtering function * RV40 loop filtering function
*/ */
...@@ -430,10 +458,11 @@ static void rv40_loop_filter(RV34DecContext *r, int row) ...@@ -430,10 +458,11 @@ static void rv40_loop_filter(RV34DecContext *r, int row)
// if bottom block is coded then we can filter its top edge // if bottom block is coded then we can filter its top edge
// (or bottom edge of this block, which is the same) // (or bottom edge of this block, which is the same)
if(y_h_deblock & (MASK_BOTTOM << ij)){ if(y_h_deblock & (MASK_BOTTOM << ij)){
r->rdsp.rv40_h_loop_filter(Y+4*s->linesize, s->linesize, dither, rv40_adaptive_loop_filter(&r->rdsp, Y+4*s->linesize,
y_to_deblock & (MASK_BOTTOM << ij) ? clip[POS_CUR] : 0, s->linesize, dither,
clip_cur, y_to_deblock & (MASK_BOTTOM << ij) ? clip[POS_CUR] : 0,
alpha, beta, betaY, 0, 0); clip_cur, alpha, beta, betaY,
0, 0, 0);
} }
// filter left block edge in ordinary mode (with low filtering strength) // filter left block edge in ordinary mode (with low filtering strength)
if(y_v_deblock & (MASK_CUR << ij) && (i || !(mb_strong[POS_CUR] || mb_strong[POS_LEFT]))){ if(y_v_deblock & (MASK_CUR << ij) && (i || !(mb_strong[POS_CUR] || mb_strong[POS_LEFT]))){
...@@ -441,25 +470,25 @@ static void rv40_loop_filter(RV34DecContext *r, int row) ...@@ -441,25 +470,25 @@ static void rv40_loop_filter(RV34DecContext *r, int row)
clip_left = mvmasks[POS_LEFT] & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0; clip_left = mvmasks[POS_LEFT] & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0;
else else
clip_left = y_to_deblock & (MASK_CUR << (ij-1)) ? clip[POS_CUR] : 0; clip_left = y_to_deblock & (MASK_CUR << (ij-1)) ? clip[POS_CUR] : 0;
r->rdsp.rv40_v_loop_filter(Y, s->linesize, dither, rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
clip_cur, clip_cur,
clip_left, clip_left,
alpha, beta, betaY, 0, 0); alpha, beta, betaY, 0, 0, 1);
} }
// filter top edge of the current macroblock when filtering strength is high // filter top edge of the current macroblock when filtering strength is high
if(!j && y_h_deblock & (MASK_CUR << i) && (mb_strong[POS_CUR] || mb_strong[POS_TOP])){ if(!j && y_h_deblock & (MASK_CUR << i) && (mb_strong[POS_CUR] || mb_strong[POS_TOP])){
r->rdsp.rv40_h_loop_filter(Y, s->linesize, dither, rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
clip_cur, clip_cur,
mvmasks[POS_TOP] & (MASK_TOP << i) ? clip[POS_TOP] : 0, mvmasks[POS_TOP] & (MASK_TOP << i) ? clip[POS_TOP] : 0,
alpha, beta, betaY, 0, 1); alpha, beta, betaY, 0, 1, 0);
} }
// filter left block edge in edge mode (with high filtering strength) // filter left block edge in edge mode (with high filtering strength)
if(y_v_deblock & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] || mb_strong[POS_LEFT])){ if(y_v_deblock & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] || mb_strong[POS_LEFT])){
clip_left = mvmasks[POS_LEFT] & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0; clip_left = mvmasks[POS_LEFT] & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0;
r->rdsp.rv40_v_loop_filter(Y, s->linesize, dither, rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
clip_cur, clip_cur,
clip_left, clip_left,
alpha, beta, betaY, 0, 1); alpha, beta, betaY, 0, 1, 1);
} }
} }
} }
...@@ -471,34 +500,34 @@ static void rv40_loop_filter(RV34DecContext *r, int row) ...@@ -471,34 +500,34 @@ static void rv40_loop_filter(RV34DecContext *r, int row)
int clip_cur = c_to_deblock[k] & (MASK_CUR << ij) ? clip[POS_CUR] : 0; int clip_cur = c_to_deblock[k] & (MASK_CUR << ij) ? clip[POS_CUR] : 0;
if(c_h_deblock[k] & (MASK_CUR << (ij+2))){ if(c_h_deblock[k] & (MASK_CUR << (ij+2))){
int clip_bot = c_to_deblock[k] & (MASK_CUR << (ij+2)) ? clip[POS_CUR] : 0; int clip_bot = c_to_deblock[k] & (MASK_CUR << (ij+2)) ? clip[POS_CUR] : 0;
r->rdsp.rv40_h_loop_filter(C+4*s->uvlinesize, s->uvlinesize, i*8, rv40_adaptive_loop_filter(&r->rdsp, C+4*s->uvlinesize, s->uvlinesize, i*8,
clip_bot, clip_bot,
clip_cur, clip_cur,
alpha, beta, betaC, 1, 0); alpha, beta, betaC, 1, 0, 0);
} }
if((c_v_deblock[k] & (MASK_CUR << ij)) && (i || !(mb_strong[POS_CUR] || mb_strong[POS_LEFT]))){ if((c_v_deblock[k] & (MASK_CUR << ij)) && (i || !(mb_strong[POS_CUR] || mb_strong[POS_LEFT]))){
if(!i) if(!i)
clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0; clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
else else
clip_left = c_to_deblock[k] & (MASK_CUR << (ij-1)) ? clip[POS_CUR] : 0; clip_left = c_to_deblock[k] & (MASK_CUR << (ij-1)) ? clip[POS_CUR] : 0;
r->rdsp.rv40_v_loop_filter(C, s->uvlinesize, j*8, rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, j*8,
clip_cur, clip_cur,
clip_left, clip_left,
alpha, beta, betaC, 1, 0); alpha, beta, betaC, 1, 0, 1);
} }
if(!j && c_h_deblock[k] & (MASK_CUR << ij) && (mb_strong[POS_CUR] || mb_strong[POS_TOP])){ if(!j && c_h_deblock[k] & (MASK_CUR << ij) && (mb_strong[POS_CUR] || mb_strong[POS_TOP])){
int clip_top = uvcbp[POS_TOP][k] & (MASK_CUR << (ij+2)) ? clip[POS_TOP] : 0; int clip_top = uvcbp[POS_TOP][k] & (MASK_CUR << (ij+2)) ? clip[POS_TOP] : 0;
r->rdsp.rv40_h_loop_filter(C, s->uvlinesize, i*8, rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, i*8,
clip_cur, clip_cur,
clip_top, clip_top,
alpha, beta, betaC, 1, 1); alpha, beta, betaC, 1, 1, 0);
} }
if(c_v_deblock[k] & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] || mb_strong[POS_LEFT])){ if(c_v_deblock[k] & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] || mb_strong[POS_LEFT])){
clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0; clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
r->rdsp.rv40_v_loop_filter(C, s->uvlinesize, j*8, rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, j*8,
clip_cur, clip_cur,
clip_left, clip_left,
alpha, beta, betaC, 1, 1); alpha, beta, betaC, 1, 1, 1);
} }
} }
} }
......
libavcodec/rv40dsp.c
View file @ d8edf1b5
This diff is collapsed.
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment