Store intra4x4_pred_mode per row only.

about 5 cpu cycles slower in the local code but should be overall faster
due to reduced cache use. (my sample though has too few intra4x4 blocks
for this to be meassureable easily either way)

Originally committed as revision 22052 to svn://svn.ffmpeg.org/ffmpeg/trunk
This commit is contained in:
Michael Niedermayer 2010-02-25 14:02:39 +00:00
parent c2186cbddc
commit 5b0fb5244d
2 changed files with 17 additions and 15 deletions

View File

@ -52,15 +52,15 @@ static const uint8_t div6[52]={
};
void ff_h264_write_back_intra_pred_mode(H264Context *h){
const int mb_xy= h->mb_xy;
int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
mode[0]= h->intra4x4_pred_mode_cache[7+8*1];
mode[1]= h->intra4x4_pred_mode_cache[7+8*2];
mode[2]= h->intra4x4_pred_mode_cache[7+8*3];
mode[3]= h->intra4x4_pred_mode_cache[7+8*4];
mode[4]= h->intra4x4_pred_mode_cache[4+8*4];
mode[5]= h->intra4x4_pred_mode_cache[5+8*4];
mode[6]= h->intra4x4_pred_mode_cache[6+8*4];
}
/**

View File

@ -298,7 +298,7 @@ typedef struct H264Context{
int topleft_partition;
int8_t intra4x4_pred_mode_cache[5*8];
int8_t (*intra4x4_pred_mode)[8];
int8_t (*intra4x4_pred_mode);
H264PredContext hpc;
unsigned int topleft_samples_available;
unsigned int top_samples_available;
@ -886,10 +886,11 @@ static void fill_decode_caches(H264Context *h, int mb_type){
if(IS_INTRA4x4(mb_type)){
if(IS_INTRA4x4(top_type)){
h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[top_xy];
h->intra4x4_pred_mode_cache[4+8*0]= mode[4];
h->intra4x4_pred_mode_cache[5+8*0]= mode[5];
h->intra4x4_pred_mode_cache[6+8*0]= mode[6];
h->intra4x4_pred_mode_cache[7+8*0]= mode[3];
}else{
int pred;
if(!(top_type & type_mask))
@ -904,8 +905,9 @@ static void fill_decode_caches(H264Context *h, int mb_type){
}
for(i=0; i<2; i++){
if(IS_INTRA4x4(left_type[i])){
h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[i]];
h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[left_block[0+2*i]];
h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[left_block[1+2*i]];
}else{
int pred;
if(!(left_type[i] & type_mask))