har0ke/FFmpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

avcodec/me_cmp, motion_est: Move me_(pre_)?_cmp etc. to MotionEstContext

Browse Source 
MECmpContext has several arrays of function pointers that
are not set by ff_me_cmp_init(), but that are set by users
to one of the other arrays via ff_set_cmp().

One of these other users is the motion estimation API.
It uses MECmpContext.(me_pre|me|me_sub|mb)_cmp. It is
basically the only user of these arrays.

This commit therefore moves these arrays to MotionEstContext;
this has the additional advantage of making motion_est.c
more independent from MpegEncContext.

Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This commit is contained in:
Andreas Rheinhardt 2024-05-12 18:47:25 +02:00
parent 1367ef198a
commit 182e647a64
6 changed files with 60 additions and 63 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
libavcodec/me_cmp.h
View File

@ -70,10 +70,6 @@ typedef struct MECmpContext {
me_cmp_func dct_max[6]; me_cmp_func dct_max[6];
me_cmp_func dct264_sad[6]; me_cmp_func dct264_sad[6];
me_cmp_func me_pre_cmp[6];
me_cmp_func me_cmp[6];
me_cmp_func me_sub_cmp[6];
me_cmp_func mb_cmp[6];
me_cmp_func ildct_cmp[6]; // only width 16 used me_cmp_func ildct_cmp[6]; // only width 16 used
me_cmp_func frame_skip_cmp[6]; // only width 8 used me_cmp_func frame_skip_cmp[6]; // only width 8 used

36
libavcodec/motion_est.c
View File

@ -305,7 +305,7 @@ static int zero_cmp(MpegEncContext *s, const uint8_t *a, const uint8_t *b,
static void zero_hpel(uint8_t *a, const uint8_t *b, ptrdiff_t stride, int h){ static void zero_hpel(uint8_t *a, const uint8_t *b, ptrdiff_t stride, int h){
} }
av_cold int ff_me_init(MotionEstContext *c, AVCodecContext *avctx, MECmpContext *mecc) av_cold int ff_me_init(MotionEstContext *c, AVCodecContext *avctx, const MECmpContext *mecc)
{ {
int cache_size = FFMIN(ME_MAP_SIZE>>ME_MAP_SHIFT, 1<<ME_MAP_SHIFT); int cache_size = FFMIN(ME_MAP_SIZE>>ME_MAP_SHIFT, 1<<ME_MAP_SHIFT);
int dia_size = FFMAX(FFABS(avctx->dia_size) & 255, FFABS(avctx->pre_dia_size) & 255); int dia_size = FFMAX(FFABS(avctx->dia_size) & 255, FFABS(avctx->pre_dia_size) & 255);
@ -324,10 +324,10 @@ av_cold int ff_me_init(MotionEstContext *c, AVCodecContext *avctx, MECmpContext
if (cache_size < 2 * dia_size) if (cache_size < 2 * dia_size)
av_log(avctx, AV_LOG_INFO, "ME_MAP size may be a little small for the selected diamond size\n"); av_log(avctx, AV_LOG_INFO, "ME_MAP size may be a little small for the selected diamond size\n");
ret = ff_set_cmp(mecc, mecc->me_pre_cmp, avctx->me_pre_cmp); ret = ff_set_cmp(mecc, c->me_pre_cmp, avctx->me_pre_cmp);
ret |= ff_set_cmp(mecc, mecc->me_cmp, avctx->me_cmp); ret |= ff_set_cmp(mecc, c->me_cmp, avctx->me_cmp);
ret |= ff_set_cmp(mecc, mecc->me_sub_cmp, avctx->me_sub_cmp); ret |= ff_set_cmp(mecc, c->me_sub_cmp, avctx->me_sub_cmp);
ret |= ff_set_cmp(mecc, mecc->mb_cmp, avctx->mb_cmp); ret |= ff_set_cmp(mecc, c->mb_cmp, avctx->mb_cmp);
if (ret < 0) if (ret < 0)
return ret; return ret;
@ -354,10 +354,10 @@ av_cold int ff_me_init(MotionEstContext *c, AVCodecContext *avctx, MECmpContext
* not have yet, and even if we had, the motion estimation code * not have yet, and even if we had, the motion estimation code
* does not expect it. */ * does not expect it. */
if (avctx->codec_id != AV_CODEC_ID_SNOW) { if (avctx->codec_id != AV_CODEC_ID_SNOW) {
if ((avctx->me_cmp & FF_CMP_CHROMA) /* && !s->mecc.me_cmp[2] */) if ((avctx->me_cmp & FF_CMP_CHROMA) /* && !c->me_cmp[2] */)
mecc->me_cmp[2] = zero_cmp; c->me_cmp[2] = zero_cmp;
if ((avctx->me_sub_cmp & FF_CMP_CHROMA) && !mecc->me_sub_cmp[2]) if ((avctx->me_sub_cmp & FF_CMP_CHROMA) && !c->me_sub_cmp[2])
mecc->me_sub_cmp[2] = zero_cmp; c->me_sub_cmp[2] = zero_cmp;
} }
return 0; return 0;
@ -649,7 +649,7 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
dmin4= c->sub_motion_search(s, &mx4, &my4, dmin4, block, block, size, h); dmin4= c->sub_motion_search(s, &mx4, &my4, dmin4, block, block, size, h);
if (s->mecc.me_sub_cmp[0] != s->mecc.mb_cmp[0]) { if (c->me_sub_cmp[0] != c->mb_cmp[0]) {
int dxy; int dxy;
const int offset= ((block&1) + (block>>1)*stride)*8; const int offset= ((block&1) + (block>>1)*stride)*8;
uint8_t *dest_y = c->scratchpad + offset; uint8_t *dest_y = c->scratchpad + offset;
@ -691,8 +691,8 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
if(same) if(same)
return INT_MAX; return INT_MAX;
if (s->mecc.me_sub_cmp[0] != s->mecc.mb_cmp[0]) { if (c->me_sub_cmp[0] != c->mb_cmp[0]) {
dmin_sum += s->mecc.mb_cmp[0](s, dmin_sum += c->mb_cmp[0](s,
s->new_pic->data[0] + s->new_pic->data[0] +
s->mb_x * 16 + s->mb_y * 16 * stride, s->mb_x * 16 + s->mb_y * 16 * stride,
c->scratchpad, stride, 16); c->scratchpad, stride, 16);
@ -717,8 +717,8 @@ static inline int h263_mv4_search(MpegEncContext *s, int mx, int my, int shift)
s->hdsp.put_pixels_tab [1][dxy](c->scratchpad + 8, s->last_pic.data[2] + offset, s->uvlinesize, 8); s->hdsp.put_pixels_tab [1][dxy](c->scratchpad + 8, s->last_pic.data[2] + offset, s->uvlinesize, 8);
} }
dmin_sum += s->mecc.mb_cmp[1](s, s->new_pic->data[1] + s->mb_x * 8 + s->mb_y * 8 * s->uvlinesize, c->scratchpad, s->uvlinesize, 8); dmin_sum += c->mb_cmp[1](s, s->new_pic->data[1] + s->mb_x * 8 + s->mb_y * 8 * s->uvlinesize, c->scratchpad, s->uvlinesize, 8);
dmin_sum += s->mecc.mb_cmp[1](s, s->new_pic->data[2] + s->mb_x * 8 + s->mb_y * 8 * s->uvlinesize, c->scratchpad + 8, s->uvlinesize, 8); dmin_sum += c->mb_cmp[1](s, s->new_pic->data[2] + s->mb_x * 8 + s->mb_y * 8 * s->uvlinesize, c->scratchpad + 8, s->uvlinesize, 8);
} }
c->pred_x= mx; c->pred_x= mx;
@ -814,7 +814,7 @@ static int interlaced_search(MpegEncContext *s, int ref_index,
mv_table[xy][0]= mx_i; mv_table[xy][0]= mx_i;
mv_table[xy][1]= my_i; mv_table[xy][1]= my_i;
if (s->mecc.me_sub_cmp[0] != s->mecc.mb_cmp[0]) { if (c->me_sub_cmp[0] != c->mb_cmp[0]) {
int dxy; int dxy;
//FIXME chroma ME //FIXME chroma ME
@ -826,7 +826,7 @@ static int interlaced_search(MpegEncContext *s, int ref_index,
}else{ }else{
s->hdsp.put_pixels_tab [size][dxy](c->scratchpad, ref , stride, h); s->hdsp.put_pixels_tab [size][dxy](c->scratchpad, ref , stride, h);
} }
dmin = s->mecc.mb_cmp[size](s, c->src[block][0], c->scratchpad, stride, h); dmin = c->mb_cmp[size](s, c->src[block][0], c->scratchpad, stride, h);
dmin+= (mv_penalty[mx_i-c->pred_x] + mv_penalty[my_i-c->pred_y] + 1)*c->mb_penalty_factor; dmin+= (mv_penalty[mx_i-c->pred_x] + mv_penalty[my_i-c->pred_y] + 1)*c->mb_penalty_factor;
}else }else
dmin+= c->mb_penalty_factor; //field_select bits dmin+= c->mb_penalty_factor; //field_select bits
@ -1049,7 +1049,7 @@ void ff_estimate_p_frame_motion(MpegEncContext * s,
*(uint32_t*)(&c->scratchpad[i*s->linesize+12]) = mean; *(uint32_t*)(&c->scratchpad[i*s->linesize+12]) = mean;
} }
intra_score= s->mecc.mb_cmp[0](s, c->scratchpad, pix, s->linesize, 16); intra_score= c->mb_cmp[0](s, c->scratchpad, pix, s->linesize, 16);
} }
intra_score += c->mb_penalty_factor*16 + s->intra_penalty; intra_score += c->mb_penalty_factor*16 + s->intra_penalty;
@ -1236,7 +1236,7 @@ static inline int check_bidir_mv(MpegEncContext * s,
fbmin = (mv_penalty_f[motion_fx-pred_fx] + mv_penalty_f[motion_fy-pred_fy])*c->mb_penalty_factor fbmin = (mv_penalty_f[motion_fx-pred_fx] + mv_penalty_f[motion_fy-pred_fy])*c->mb_penalty_factor
+(mv_penalty_b[motion_bx-pred_bx] + mv_penalty_b[motion_by-pred_by])*c->mb_penalty_factor +(mv_penalty_b[motion_bx-pred_bx] + mv_penalty_b[motion_by-pred_by])*c->mb_penalty_factor
+ s->mecc.mb_cmp[size](s, src_data[0], dest_y, stride, h); // FIXME new_pic + c->mb_cmp[size](s, src_data[0], dest_y, stride, h); // FIXME new_pic
if(c->avctx->mb_cmp&FF_CMP_CHROMA){ if(c->avctx->mb_cmp&FF_CMP_CHROMA){
} }

7
libavcodec/motion_est.h
View File

@ -84,6 +84,11 @@ typedef struct MotionEstContext {
int64_t mb_var_sum_temp; int64_t mb_var_sum_temp;
int scene_change_score; int scene_change_score;
me_cmp_func me_pre_cmp[6];
me_cmp_func me_cmp[6];
me_cmp_func me_sub_cmp[6];
me_cmp_func mb_cmp[6];
op_pixels_func(*hpel_put)[4]; op_pixels_func(*hpel_put)[4];
op_pixels_func(*hpel_avg)[4]; op_pixels_func(*hpel_avg)[4];
qpel_mc_func(*qpel_put)[16]; qpel_mc_func(*qpel_put)[16];
@ -110,7 +115,7 @@ static inline int ff_h263_round_chroma(int x)
* Performs one-time initialization of the MotionEstContext. * Performs one-time initialization of the MotionEstContext.
*/ */
int ff_me_init(MotionEstContext *c, struct AVCodecContext *avctx, int ff_me_init(MotionEstContext *c, struct AVCodecContext *avctx,
struct MECmpContext *mecc); const struct MECmpContext *mecc);
void ff_me_init_pic(struct MpegEncContext *s); void ff_me_init_pic(struct MpegEncContext *s);

60
libavcodec/motion_est_template.c
View File

@ -64,8 +64,8 @@ static int hpel_motion_search(MpegEncContext * s,
//FIXME factorize //FIXME factorize
cmp_sub = s->mecc.me_sub_cmp[size]; cmp_sub = c->me_sub_cmp[size];
chroma_cmp_sub = s->mecc.me_sub_cmp[size + 1]; chroma_cmp_sub = c->me_sub_cmp[size + 1];
if(c->skip){ //FIXME move out of hpel? if(c->skip){ //FIXME move out of hpel?
*mx_ptr = 0; *mx_ptr = 0;
@ -178,8 +178,8 @@ static inline int get_mb_score(MpegEncContext *s, int mx, int my,
//FIXME factorize //FIXME factorize
cmp_sub = s->mecc.mb_cmp[size]; cmp_sub = c->mb_cmp[size];
chroma_cmp_sub = s->mecc.mb_cmp[size + 1]; chroma_cmp_sub = c->mb_cmp[size + 1];
d= cmp(s, mx>>(qpel+1), my>>(qpel+1), mx&mask, my&mask, size, h, ref_index, src_index, cmp_sub, chroma_cmp_sub, flags); d= cmp(s, mx>>(qpel+1), my>>(qpel+1), mx&mask, my&mask, size, h, ref_index, src_index, cmp_sub, chroma_cmp_sub, flags);
//FIXME check cbp before adding penalty for (0,0) vector //FIXME check cbp before adding penalty for (0,0) vector
@ -222,12 +222,12 @@ static int qpel_motion_search(MpegEncContext * s,
LOAD_COMMON LOAD_COMMON
int flags= c->sub_flags; int flags= c->sub_flags;
cmpf = s->mecc.me_cmp[size]; cmpf = c->me_cmp[size];
chroma_cmpf = s->mecc.me_cmp[size + 1]; // FIXME: factorize chroma_cmpf = c->me_cmp[size + 1]; // FIXME: factorize
//FIXME factorize //FIXME factorize
cmp_sub = s->mecc.me_sub_cmp[size]; cmp_sub = c->me_sub_cmp[size];
chroma_cmp_sub = s->mecc.me_sub_cmp[size + 1]; chroma_cmp_sub = c->me_sub_cmp[size + 1];
if(c->skip){ //FIXME somehow move up (benchmark) if(c->skip){ //FIXME somehow move up (benchmark)
*mx_ptr = 0; *mx_ptr = 0;
@ -424,8 +424,8 @@ static av_always_inline int small_diamond_search(MpegEncContext * s, int *best,
LOAD_COMMON2 LOAD_COMMON2
unsigned map_generation = c->map_generation; unsigned map_generation = c->map_generation;
cmpf = s->mecc.me_cmp[size]; cmpf = c->me_cmp[size];
chroma_cmpf = s->mecc.me_cmp[size + 1]; chroma_cmpf = c->me_cmp[size + 1];
{ /* ensure that the best point is in the MAP as h/qpel refinement needs it */ { /* ensure that the best point is in the MAP as h/qpel refinement needs it */
const unsigned key = ((unsigned)best[1]<<ME_MAP_MV_BITS) + best[0] + map_generation; const unsigned key = ((unsigned)best[1]<<ME_MAP_MV_BITS) + best[0] + map_generation;
@ -465,8 +465,8 @@ static int funny_diamond_search(MpegEncContext * s, int *best, int dmin,
LOAD_COMMON2 LOAD_COMMON2
unsigned map_generation = c->map_generation; unsigned map_generation = c->map_generation;
cmpf = s->mecc.me_cmp[size]; cmpf = c->me_cmp[size];
chroma_cmpf = s->mecc.me_cmp[size + 1]; chroma_cmpf = c->me_cmp[size + 1];
for(dia_size=1; dia_size<=4; dia_size++){ for(dia_size=1; dia_size<=4; dia_size++){
int dir; int dir;
@ -508,8 +508,8 @@ static int hex_search(MpegEncContext * s, int *best, int dmin,
int x,y,d; int x,y,d;
const int dec= dia_size & (dia_size-1); const int dec= dia_size & (dia_size-1);
cmpf = s->mecc.me_cmp[size]; cmpf = c->me_cmp[size];
chroma_cmpf = s->mecc.me_cmp[size + 1]; chroma_cmpf = c->me_cmp[size + 1];
for(;dia_size; dia_size= dec ? dia_size-1 : dia_size>>1){ for(;dia_size; dia_size= dec ? dia_size-1 : dia_size>>1){
do{ do{
@ -545,8 +545,8 @@ static int l2s_dia_search(MpegEncContext * s, int *best, int dmin,
static const int hex[8][2]={{-2, 0}, {-1,-1}, { 0,-2}, { 1,-1}, static const int hex[8][2]={{-2, 0}, {-1,-1}, { 0,-2}, { 1,-1},
{ 2, 0}, { 1, 1}, { 0, 2}, {-1, 1}}; { 2, 0}, { 1, 1}, { 0, 2}, {-1, 1}};
cmpf = s->mecc.me_cmp[size]; cmpf = c->me_cmp[size];
chroma_cmpf = s->mecc.me_cmp[size + 1]; chroma_cmpf = c->me_cmp[size + 1];
for(; dia_size; dia_size= dec ? dia_size-1 : dia_size>>1){ for(; dia_size; dia_size= dec ? dia_size-1 : dia_size>>1){
do{ do{
@ -584,8 +584,8 @@ static int umh_search(MpegEncContext * s, int *best, int dmin,
{-2, 3}, { 0, 4}, { 2, 3}, {-2, 3}, { 0, 4}, { 2, 3},
{-2,-3}, { 0,-4}, { 2,-3},}; {-2,-3}, { 0,-4}, { 2,-3},};
cmpf = s->mecc.me_cmp[size]; cmpf = c->me_cmp[size];
chroma_cmpf = s->mecc.me_cmp[size + 1]; chroma_cmpf = c->me_cmp[size + 1];
x= best[0]; x= best[0];
y= best[1]; y= best[1];
@ -627,8 +627,8 @@ static int full_search(MpegEncContext * s, int *best, int dmin,
int x,y, d; int x,y, d;
const int dia_size= c->dia_size&0xFF; const int dia_size= c->dia_size&0xFF;
cmpf = s->mecc.me_cmp[size]; cmpf = c->me_cmp[size];
chroma_cmpf = s->mecc.me_cmp[size + 1]; chroma_cmpf = c->me_cmp[size + 1];
for(y=FFMAX(-dia_size, ymin); y<=FFMIN(dia_size,ymax); y++){ for(y=FFMAX(-dia_size, ymin); y<=FFMIN(dia_size,ymax); y++){
for(x=FFMAX(-dia_size, xmin); x<=FFMIN(dia_size,xmax); x++){ for(x=FFMAX(-dia_size, xmin); x<=FFMIN(dia_size,xmax); x++){
@ -693,8 +693,8 @@ static int sab_diamond_search(MpegEncContext * s, int *best, int dmin,
av_assert1(minima_count <= MAX_SAB_SIZE); av_assert1(minima_count <= MAX_SAB_SIZE);
cmpf = s->mecc.me_cmp[size]; cmpf = c->me_cmp[size];
chroma_cmpf = s->mecc.me_cmp[size + 1]; chroma_cmpf = c->me_cmp[size + 1];
/*Note j<MAX_SAB_SIZE is needed if MAX_SAB_SIZE < ME_MAP_SIZE as j can /*Note j<MAX_SAB_SIZE is needed if MAX_SAB_SIZE < ME_MAP_SIZE as j can
become larger due to MVs overflowing their ME_MAP_MV_BITS bits space in map become larger due to MVs overflowing their ME_MAP_MV_BITS bits space in map
@ -779,8 +779,8 @@ static int var_diamond_search(MpegEncContext * s, int *best, int dmin,
LOAD_COMMON2 LOAD_COMMON2
unsigned map_generation = c->map_generation; unsigned map_generation = c->map_generation;
cmpf = s->mecc.me_cmp[size]; cmpf = c->me_cmp[size];
chroma_cmpf = s->mecc.me_cmp[size + 1]; chroma_cmpf = c->me_cmp[size + 1];
for(dia_size=1; dia_size<=c->dia_size; dia_size++){ for(dia_size=1; dia_size<=c->dia_size; dia_size++){
int dir, start, end; int dir, start, end;
@ -880,12 +880,12 @@ static av_always_inline int epzs_motion_search_internal(MpegEncContext * s, int
if(c->pre_pass){ if(c->pre_pass){
penalty_factor= c->pre_penalty_factor; penalty_factor= c->pre_penalty_factor;
cmpf = s->mecc.me_pre_cmp[size]; cmpf = c->me_pre_cmp[size];
chroma_cmpf = s->mecc.me_pre_cmp[size + 1]; chroma_cmpf = c->me_pre_cmp[size + 1];
}else{ }else{
penalty_factor= c->penalty_factor; penalty_factor= c->penalty_factor;
cmpf = s->mecc.me_cmp[size]; cmpf = c->me_cmp[size];
chroma_cmpf = s->mecc.me_cmp[size + 1]; chroma_cmpf = c->me_cmp[size + 1];
} }
map_generation= update_map_generation(c); map_generation= update_map_generation(c);
@ -1008,8 +1008,8 @@ static int epzs_motion_search2(MpegEncContext * s,
int flags= c->flags; int flags= c->flags;
LOAD_COMMON2 LOAD_COMMON2
cmpf = s->mecc.me_cmp[size]; cmpf = c->me_cmp[size];
chroma_cmpf = s->mecc.me_cmp[size + 1]; chroma_cmpf = c->me_cmp[size + 1];
map_generation= update_map_generation(c); map_generation= update_map_generation(c);

6
libavcodec/snowenc.c
View File

@ -832,12 +832,12 @@ static int get_block_rd(SnowEncContext *enc, int mb_x, int mb_y,
distortion = 0; distortion = 0;
for(i=0; i<4; i++){ for(i=0; i<4; i++){
int off = sx+16*(i&1) + (sy+16*(i>>1))*ref_stride; int off = sx+16*(i&1) + (sy+16*(i>>1))*ref_stride;
distortion += enc->mecc.me_cmp[0](&enc->m, src + off, dst + off, ref_stride, 16); distortion += enc->m.me.me_cmp[0](&enc->m, src + off, dst + off, ref_stride, 16);
} }
} }
}else{ }else{
av_assert2(block_w==8); av_assert2(block_w==8);
distortion = enc->mecc.me_cmp[0](&enc->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, block_w*2); distortion = enc->m.me.me_cmp[0](&enc->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, block_w*2);
} }
if(plane_index==0){ if(plane_index==0){
@ -903,7 +903,7 @@ static int get_4block_rd(SnowEncContext *enc, int mb_x, int mb_y, int plane_inde
} }
av_assert1(block_w== 8 || block_w==16); av_assert1(block_w== 8 || block_w==16);
distortion += enc->mecc.me_cmp[block_w==8](&enc->m, src + x + y*ref_stride, dst + x + y*ref_stride, ref_stride, block_h); distortion += enc->m.me.me_cmp[block_w==8](&enc->m, src + x + y*ref_stride, dst + x + y*ref_stride, ref_stride, block_h);
} }
if(plane_index==0){ if(plane_index==0){

4
tests/checkasm/motion.c
View File

@ -94,10 +94,6 @@ static void test_motion(const char *name, me_cmp_func test_func)
XX(vsad) \ XX(vsad) \
XX(vsse) \ XX(vsse) \
XX(nsse) \ XX(nsse) \
XX(me_pre_cmp) \
XX(me_cmp) \
XX(me_sub_cmp) \
XX(mb_cmp) \
XX(ildct_cmp) \ XX(ildct_cmp) \
XX(frame_skip_cmp) \ XX(frame_skip_cmp) \
XX(median_sad) XX(median_sad)