aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorJason Garrett-Glaser <jason@x264.com>2011-06-29 13:27:36 -0700
committerJason Garrett-Glaser <jason@x264.com>2011-07-03 15:05:55 -0700
commit3b7ebeb4d52a25c7e1038ae90c6c19b0d6f11877 (patch)
tree2582b35da48add54c1087c62014a36478fe114a9
parentcb5469462d427ea38625e255306f07b37d75280f (diff)
downloadffmpeg-3b7ebeb4d52a25c7e1038ae90c6c19b0d6f11877.tar.gz
H.264: faster write_back_*
Avoid aliasing, unroll loops, and inline more functions.
-rw-r--r--libavcodec/h264.c9
-rw-r--r--libavcodec/h264.h123
-rw-r--r--libavcodec/h264_cabac.c21
-rw-r--r--libavcodec/h264_cavlc.c2
-rw-r--r--libavcodec/svq3.c2
5 files changed, 81 insertions, 76 deletions
diff --git a/libavcodec/h264.c b/libavcodec/h264.c
index b5a9bc0dfc..db3d93d5d7 100644
--- a/libavcodec/h264.c
+++ b/libavcodec/h264.c
@@ -60,15 +60,6 @@ static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
PIX_FMT_NONE
};
-void ff_h264_write_back_intra_pred_mode(H264Context *h){
- int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
-
- AV_COPY32(mode, h->intra4x4_pred_mode_cache + 4 + 8*4);
- mode[4]= h->intra4x4_pred_mode_cache[7+8*3];
- mode[5]= h->intra4x4_pred_mode_cache[7+8*2];
- mode[6]= h->intra4x4_pred_mode_cache[7+8*1];
-}
-
/**
* checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
*/
diff --git a/libavcodec/h264.h b/libavcodec/h264.h
index e3cc815565..8e04db4f22 100644
--- a/libavcodec/h264.h
+++ b/libavcodec/h264.h
@@ -658,7 +658,6 @@ int ff_h264_check_intra4x4_pred_mode(H264Context *h);
*/
int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
-void ff_h264_write_back_intra_pred_mode(H264Context *h);
void ff_h264_hl_decode_mb(H264Context *h);
int ff_h264_frame_start(H264Context *h);
int ff_h264_decode_extradata(H264Context *h);
@@ -1185,7 +1184,7 @@ static void fill_decode_caches(H264Context *h, int mb_type){
/**
* gets the predicted intra4x4 prediction mode.
*/
-static inline int pred_intra_mode(H264Context *h, int n){
+static av_always_inline int pred_intra_mode(H264Context *h, int n){
const int index8= scan8[n];
const int left= h->intra4x4_pred_mode_cache[index8 - 1];
const int top = h->intra4x4_pred_mode_cache[index8 - 8];
@@ -1197,69 +1196,83 @@ static inline int pred_intra_mode(H264Context *h, int n){
else return min;
}
-static inline void write_back_non_zero_count(H264Context *h){
- const int mb_xy= h->mb_xy;
+static av_always_inline void write_back_intra_pred_mode(H264Context *h){
+ int8_t *i4x4= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
+ int8_t *i4x4_cache= h->intra4x4_pred_mode_cache;
+
+ AV_COPY32(i4x4, i4x4_cache + 4 + 8*4);
+ i4x4[4]= i4x4_cache[7+8*3];
+ i4x4[5]= i4x4_cache[7+8*2];
+ i4x4[6]= i4x4_cache[7+8*1];
+}
- AV_COPY32(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[4+8* 1]);
- AV_COPY32(&h->non_zero_count[mb_xy][ 4], &h->non_zero_count_cache[4+8* 2]);
- AV_COPY32(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[4+8* 3]);
- AV_COPY32(&h->non_zero_count[mb_xy][12], &h->non_zero_count_cache[4+8* 4]);
- AV_COPY32(&h->non_zero_count[mb_xy][16], &h->non_zero_count_cache[4+8* 6]);
- AV_COPY32(&h->non_zero_count[mb_xy][20], &h->non_zero_count_cache[4+8* 7]);
- AV_COPY32(&h->non_zero_count[mb_xy][32], &h->non_zero_count_cache[4+8*11]);
- AV_COPY32(&h->non_zero_count[mb_xy][36], &h->non_zero_count_cache[4+8*12]);
+static av_always_inline void write_back_non_zero_count(H264Context *h){
+ const int mb_xy= h->mb_xy;
+ uint8_t *nnz = h->non_zero_count[mb_xy];
+ uint8_t *nnz_cache = h->non_zero_count_cache;
+
+ AV_COPY32(&nnz[ 0], &nnz_cache[4+8* 1]);
+ AV_COPY32(&nnz[ 4], &nnz_cache[4+8* 2]);
+ AV_COPY32(&nnz[ 8], &nnz_cache[4+8* 3]);
+ AV_COPY32(&nnz[12], &nnz_cache[4+8* 4]);
+ AV_COPY32(&nnz[16], &nnz_cache[4+8* 6]);
+ AV_COPY32(&nnz[20], &nnz_cache[4+8* 7]);
+ AV_COPY32(&nnz[32], &nnz_cache[4+8*11]);
+ AV_COPY32(&nnz[36], &nnz_cache[4+8*12]);
if(CHROMA444){
- AV_COPY32(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[4+8* 8]);
- AV_COPY32(&h->non_zero_count[mb_xy][28], &h->non_zero_count_cache[4+8* 9]);
- AV_COPY32(&h->non_zero_count[mb_xy][40], &h->non_zero_count_cache[4+8*13]);
- AV_COPY32(&h->non_zero_count[mb_xy][44], &h->non_zero_count_cache[4+8*14]);
+ AV_COPY32(&nnz[24], &nnz_cache[4+8* 8]);
+ AV_COPY32(&nnz[28], &nnz_cache[4+8* 9]);
+ AV_COPY32(&nnz[40], &nnz_cache[4+8*13]);
+ AV_COPY32(&nnz[44], &nnz_cache[4+8*14]);
+ }
+}
+
+static av_always_inline void write_back_motion_list(H264Context *h, MpegEncContext * const s, int b_stride,
+ int b_xy, int b8_xy, int mb_type, int list )
+{
+ int16_t (*mv_dst)[2] = &s->current_picture.motion_val[list][b_xy];
+ int16_t (*mv_src)[2] = &h->mv_cache[list][scan8[0]];
+ AV_COPY128(mv_dst + 0*b_stride, mv_src + 8*0);
+ AV_COPY128(mv_dst + 1*b_stride, mv_src + 8*1);
+ AV_COPY128(mv_dst + 2*b_stride, mv_src + 8*2);
+ AV_COPY128(mv_dst + 3*b_stride, mv_src + 8*3);
+ if( CABAC ) {
+ uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
+ uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
+ if(IS_SKIP(mb_type))
+ AV_ZERO128(mvd_dst);
+ else{
+ AV_COPY64(mvd_dst, mvd_src + 8*3);
+ AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
+ AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
+ AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
+ }
+ }
+
+ {
+ int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
+ int8_t *ref_cache = h->ref_cache[list];
+ ref_index[0+0*2]= ref_cache[scan8[0]];
+ ref_index[1+0*2]= ref_cache[scan8[4]];
+ ref_index[0+1*2]= ref_cache[scan8[8]];
+ ref_index[1+1*2]= ref_cache[scan8[12]];
}
}
-static inline void write_back_motion(H264Context *h, int mb_type){
+static av_always_inline void write_back_motion(H264Context *h, int mb_type){
MpegEncContext * const s = &h->s;
+ const int b_stride = h->b_stride;
const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; //try mb2b(8)_xy
const int b8_xy= 4*h->mb_xy;
- int list;
- if(!USES_LIST(mb_type, 0))
+ if(USES_LIST(mb_type, 0)){
+ write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 0);
+ }else{
fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
-
- for(list=0; list<h->list_count; list++){
- int y, b_stride;
- int16_t (*mv_dst)[2];
- int16_t (*mv_src)[2];
-
- if(!USES_LIST(mb_type, list))
- continue;
-
- b_stride = h->b_stride;
- mv_dst = &s->current_picture.motion_val[list][b_xy];
- mv_src = &h->mv_cache[list][scan8[0]];
- for(y=0; y<4; y++){
- AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
- }
- if( CABAC ) {
- uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
- uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
- if(IS_SKIP(mb_type))
- AV_ZERO128(mvd_dst);
- else{
- AV_COPY64(mvd_dst, mvd_src + 8*3);
- AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
- AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
- AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
- }
- }
-
- {
- int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
- ref_index[0+0*2]= h->ref_cache[list][scan8[0]];
- ref_index[1+0*2]= h->ref_cache[list][scan8[4]];
- ref_index[0+1*2]= h->ref_cache[list][scan8[8]];
- ref_index[1+1*2]= h->ref_cache[list][scan8[12]];
- }
+ }
+ if(USES_LIST(mb_type, 1)){
+ write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 1);
}
if(h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC){
@@ -1272,7 +1285,7 @@ static inline void write_back_motion(H264Context *h, int mb_type){
}
}
-static inline int get_dct8x8_allowed(H264Context *h){
+static av_always_inline int get_dct8x8_allowed(H264Context *h){
if(h->sps.direct_8x8_inference_flag)
return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8 )*0x0001000100010001ULL));
else
diff --git a/libavcodec/h264_cabac.c b/libavcodec/h264_cabac.c
index f30f4e1c9c..a643297f8a 100644
--- a/libavcodec/h264_cabac.c
+++ b/libavcodec/h264_cabac.c
@@ -1999,7 +1999,7 @@ decode_intra_mb:
//av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );
}
}
- ff_h264_write_back_intra_pred_mode(h);
+ write_back_intra_pred_mode(h);
if( ff_h264_check_intra4x4_pred_mode(h) < 0 ) return -1;
} else {
h->intra16x16_pred_mode= ff_h264_check_intra_pred_mode( h, h->intra16x16_pred_mode );
@@ -2248,21 +2248,22 @@ decode_intra_mb:
* the transform mode of the current macroblock there. */
if (CHROMA444 && IS_8x8DCT(mb_type)){
int i;
+ uint8_t *nnz_cache = h->non_zero_count_cache;
for (i = 0; i < 2; i++){
if (h->left_type[i] && !IS_8x8DCT(h->left_type[i])){
- h->non_zero_count_cache[3+8* 1 + 2*8*i]=
- h->non_zero_count_cache[3+8* 2 + 2*8*i]=
- h->non_zero_count_cache[3+8* 6 + 2*8*i]=
- h->non_zero_count_cache[3+8* 7 + 2*8*i]=
- h->non_zero_count_cache[3+8*11 + 2*8*i]=
- h->non_zero_count_cache[3+8*12 + 2*8*i]= IS_INTRA(mb_type) ? 64 : 0;
+ nnz_cache[3+8* 1 + 2*8*i]=
+ nnz_cache[3+8* 2 + 2*8*i]=
+ nnz_cache[3+8* 6 + 2*8*i]=
+ nnz_cache[3+8* 7 + 2*8*i]=
+ nnz_cache[3+8*11 + 2*8*i]=
+ nnz_cache[3+8*12 + 2*8*i]= IS_INTRA(mb_type) ? 64 : 0;
}
}
if (h->top_type && !IS_8x8DCT(h->top_type)){
uint32_t top_empty = CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040;
- AV_WN32A(&h->non_zero_count_cache[4+8* 0], top_empty);
- AV_WN32A(&h->non_zero_count_cache[4+8* 5], top_empty);
- AV_WN32A(&h->non_zero_count_cache[4+8*10], top_empty);
+ AV_WN32A(&nnz_cache[4+8* 0], top_empty);
+ AV_WN32A(&nnz_cache[4+8* 5], top_empty);
+ AV_WN32A(&nnz_cache[4+8*10], top_empty);
}
}
s->current_picture.mb_type[mb_xy]= mb_type;
diff --git a/libavcodec/h264_cavlc.c b/libavcodec/h264_cavlc.c
index 497166b423..90c411002e 100644
--- a/libavcodec/h264_cavlc.c
+++ b/libavcodec/h264_cavlc.c
@@ -731,7 +731,7 @@ decode_intra_mb:
else
h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
}
- ff_h264_write_back_intra_pred_mode(h);
+ write_back_intra_pred_mode(h);
if( ff_h264_check_intra4x4_pred_mode(h) < 0)
return -1;
}else{
diff --git a/libavcodec/svq3.c b/libavcodec/svq3.c
index 23ab209312..a88b069daf 100644
--- a/libavcodec/svq3.c
+++ b/libavcodec/svq3.c
@@ -589,7 +589,7 @@ static int svq3_decode_mb(SVQ3Context *svq3, unsigned int mb_type)
}
}
- ff_h264_write_back_intra_pred_mode(h);
+ write_back_intra_pred_mode(h);
if (mb_type == 8) {
ff_h264_check_intra4x4_pred_mode(h);