943f69a6 |
/*
* H.26L/H.264/AVC/JVT/14496-10/... direct mb/block decoding
* Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/** |
ba87f080 |
* @file |
943f69a6 |
* H.264 / AVC / MPEG4 part10 direct mb/block decoding.
* @author Michael Niedermayer <michaelni@gmx.at>
*/
#include "internal.h"
#include "avcodec.h"
#include "h264.h" |
e0c16e4e |
#include "mpegutils.h" |
943f69a6 |
#include "rectangle.h" |
d375c104 |
#include "thread.h" |
943f69a6 |
#include <assert.h>
|
d27b8248 |
static int get_scale_factor(H264SliceContext *sl, |
95eb35f3 |
int poc, int poc1, int i) |
2db953f8 |
{ |
95eb35f3 |
int poc0 = sl->ref_list[0][i].poc; |
9abc80f1 |
int td = av_clip_int8(poc1 - poc0); |
a12d3188 |
if (td == 0 || sl->ref_list[0][i].parent->long_ref) { |
943f69a6 |
return 256; |
2db953f8 |
} else { |
9abc80f1 |
int tb = av_clip_int8(poc - poc0); |
943f69a6 |
int tx = (16384 + (FFABS(td) >> 1)) / td; |
9abc80f1 |
return av_clip_intp2((tb * tx + 32) >> 6, 10); |
943f69a6 |
}
}
|
d27b8248 |
void ff_h264_direct_dist_scale_factor(const H264Context *const h, |
ee0d774d |
H264SliceContext *sl) |
2db953f8 |
{ |
c39059be |
const int poc = FIELD_PICTURE(h) ? h->cur_pic_ptr->field_poc[h->picture_structure == PICT_BOTTOM_FIELD]
: h->cur_pic_ptr->poc; |
95eb35f3 |
const int poc1 = sl->ref_list[1][0].poc; |
943f69a6 |
int i, field;
|
7bece9b2 |
if (FRAME_MBAFF(h)) |
2db953f8 |
for (field = 0; field < 2; field++) { |
2c541554 |
const int poc = h->cur_pic_ptr->field_poc[field]; |
a12d3188 |
const int poc1 = sl->ref_list[1][0].parent->field_poc[field]; |
95eb35f3 |
for (i = 0; i < 2 * sl->ref_count[0]; i++) |
ee0d774d |
sl->dist_scale_factor_field[field][i ^ field] = |
d27b8248 |
get_scale_factor(sl, poc, poc1, i + 16); |
73ad2c2f |
}
|
95eb35f3 |
for (i = 0; i < sl->ref_count[0]; i++) |
d27b8248 |
sl->dist_scale_factor[i] = get_scale_factor(sl, poc, poc1, i); |
943f69a6 |
}
|
d27b8248 |
static void fill_colmap(const H264Context *h, H264SliceContext *sl, |
95eb35f3 |
int map[2][16 + 32], int list, |
2db953f8 |
int field, int colfield, int mbafi)
{ |
a12d3188 |
H264Picture *const ref1 = sl->ref_list[1][0].parent; |
943f69a6 |
int j, old_ref, rfield; |
2db953f8 |
int start = mbafi ? 16 : 0; |
95eb35f3 |
int end = mbafi ? 16 + 2 * sl->ref_count[0] : sl->ref_count[0]; |
2db953f8 |
int interl = mbafi || h->picture_structure != PICT_FRAME; |
943f69a6 |
/* bogus; fills in for missing frames */
memset(map[list], 0, sizeof(map[list]));
|
2db953f8 |
for (rfield = 0; rfield < 2; rfield++) {
for (old_ref = 0; old_ref < ref1->ref_count[colfield][list]; old_ref++) { |
943f69a6 |
int poc = ref1->ref_poc[colfield][list][old_ref];
|
2db953f8 |
if (!interl) |
943f69a6 |
poc |= 3; |
2db953f8 |
// FIXME: store all MBAFF references so this is not needed
else if (interl && (poc & 3) == 3)
poc = (poc & ~3) + rfield + 1;
for (j = start; j < end; j++) { |
a12d3188 |
if (4 * sl->ref_list[0][j].parent->frame_num + |
95eb35f3 |
(sl->ref_list[0][j].reference & 3) == poc) { |
2db953f8 |
int cur_ref = mbafi ? (j - 16) ^ field : j; |
c9aab8a1 |
if (ref1->mbaff) |
2db953f8 |
map[list][2 * old_ref + (rfield ^ field) + 16] = cur_ref;
if (rfield == field || !interl) |
943f69a6 |
map[list][old_ref] = cur_ref;
break;
}
}
}
}
}
|
d27b8248 |
void ff_h264_direct_ref_list_init(const H264Context *const h, H264SliceContext *sl) |
2db953f8 |
{ |
a12d3188 |
H264Ref *const ref1 = &sl->ref_list[1][0]; |
2db953f8 |
H264Picture *const cur = h->cur_pic_ptr; |
943f69a6 |
int list, j, field; |
2db953f8 |
int sidx = (h->picture_structure & 1) ^ 1;
int ref1sidx = (ref1->reference & 1) ^ 1; |
943f69a6 |
|
a12d3188 |
for (list = 0; list < sl->list_count; list++) { |
95eb35f3 |
cur->ref_count[sidx][list] = sl->ref_count[list];
for (j = 0; j < sl->ref_count[list]; j++) |
a12d3188 |
cur->ref_poc[sidx][list][j] = 4 * sl->ref_list[list][j].parent->frame_num + |
95eb35f3 |
(sl->ref_list[list][j].reference & 3); |
943f69a6 |
}
|
2db953f8 |
if (h->picture_structure == PICT_FRAME) { |
943f69a6 |
memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0])); |
2db953f8 |
memcpy(cur->ref_poc[1], cur->ref_poc[0], sizeof(cur->ref_poc[0])); |
943f69a6 |
}
|
7bece9b2 |
cur->mbaff = FRAME_MBAFF(h); |
943f69a6 |
|
be69f0a8 |
sl->col_fieldoff = 0; |
a12d3188 |
if (sl->list_count != 2 || !sl->ref_count[1])
return;
|
2db953f8 |
if (h->picture_structure == PICT_FRAME) {
int cur_poc = h->cur_pic_ptr->poc; |
a12d3188 |
int *col_poc = sl->ref_list[1][0].parent->field_poc; |
77a644e6 |
if (col_poc[0] == INT_MAX && col_poc[1] == INT_MAX) {
av_log(h->avctx, AV_LOG_ERROR, "co located POCs unavailable\n");
sl->col_parity = 1;
} else |
be69f0a8 |
sl->col_parity = (FFABS(col_poc[0] - cur_poc) >=
FFABS(col_poc[1] - cur_poc)); |
2db953f8 |
ref1sidx = |
be69f0a8 |
sidx = sl->col_parity; |
2db953f8 |
// FL -> FL & differ parity |
95eb35f3 |
} else if (!(h->picture_structure & sl->ref_list[1][0].reference) && |
a12d3188 |
!sl->ref_list[1][0].parent->mbaff) { |
95eb35f3 |
sl->col_fieldoff = 2 * sl->ref_list[1][0].reference - 3; |
d88ea354 |
}
|
066aafce |
if (sl->slice_type_nos != AV_PICTURE_TYPE_B || sl->direct_spatial_mv_pred) |
0a8dca87 |
return;
|
2db953f8 |
for (list = 0; list < 2; list++) { |
95eb35f3 |
fill_colmap(h, sl, sl->map_col_to_list0, list, sidx, ref1sidx, 0); |
7bece9b2 |
if (FRAME_MBAFF(h)) |
2db953f8 |
for (field = 0; field < 2; field++) |
95eb35f3 |
fill_colmap(h, sl, sl->map_col_to_list0_field[field], list, field, |
2db953f8 |
field, 1); |
943f69a6 |
}
}
|
d8151a7e |
static void await_reference_mb_row(const H264Context *const h, H264Ref *ref, |
2db953f8 |
int mb_y) |
d375c104 |
{ |
2db953f8 |
int ref_field = ref->reference - 1; |
d8151a7e |
int ref_field_picture = ref->parent->field_picture; |
2db953f8 |
int ref_height = 16 * h->mb_height >> ref_field_picture; |
d375c104 |
|
2db953f8 |
if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_FRAME)) |
d375c104 |
return;
|
2db953f8 |
/* FIXME: It can be safe to access mb stuff
* even if pixels aren't deblocked yet. */ |
d375c104 |
|
d8151a7e |
ff_thread_await_progress(&ref->parent->tf, |
2db953f8 |
FFMIN(16 * mb_y >> ref_field_picture,
ref_height - 1), |
d375c104 |
ref_field_picture && ref_field);
}
|
d27b8248 |
static void pred_spatial_direct_motion(const H264Context *const h, H264SliceContext *sl, |
e6287f07 |
int *mb_type) |
2db953f8 |
{ |
358b5b1a |
int b8_stride = 2; |
943f69a6 |
int b4_stride = h->b_stride; |
d4d9068c |
int mb_xy = sl->mb_xy, mb_y = sl->mb_y; |
943f69a6 |
int mb_type_col[2];
const int16_t (*l1mv0)[2], (*l1mv1)[2];
const int8_t *l1ref0, *l1ref1;
const int is_b8x8 = IS_8X8(*mb_type); |
2db953f8 |
unsigned int sub_mb_type = MB_TYPE_L0L1; |
943f69a6 |
int i8, i4; |
af0d6581 |
int ref[2];
int mv[2];
int list; |
943f69a6 |
|
95eb35f3 |
assert(sl->ref_list[1][0].reference & 3); |
943f69a6 |
|
95eb35f3 |
await_reference_mb_row(h, &sl->ref_list[1][0], |
d4d9068c |
sl->mb_y + !!IS_INTERLACED(*mb_type)); |
943f69a6 |
|
2db953f8 |
#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16 | MB_TYPE_INTRA4x4 | \
MB_TYPE_INTRA16x16 | MB_TYPE_INTRA_PCM) |
af0d6581 |
/* ref = min(neighbors) */ |
2db953f8 |
for (list = 0; list < 2; list++) { |
e6287f07 |
int left_ref = sl->ref_cache[list][scan8[0] - 1];
int top_ref = sl->ref_cache[list][scan8[0] - 8];
int refc = sl->ref_cache[list][scan8[0] - 8 + 4];
const int16_t *C = sl->mv_cache[list][scan8[0] - 8 + 4]; |
2db953f8 |
if (refc == PART_NOT_AVAILABLE) { |
e6287f07 |
refc = sl->ref_cache[list][scan8[0] - 8 - 1];
C = sl->mv_cache[list][scan8[0] - 8 - 1]; |
af0d6581 |
} |
2db953f8 |
ref[list] = FFMIN3((unsigned)left_ref,
(unsigned)top_ref,
(unsigned)refc);
if (ref[list] >= 0) {
/* This is just pred_motion() but with the cases removed that
* cannot happen for direct blocks. */ |
e6287f07 |
const int16_t *const A = sl->mv_cache[list][scan8[0] - 1];
const int16_t *const B = sl->mv_cache[list][scan8[0] - 8]; |
2db953f8 |
int match_count = (left_ref == ref[list]) +
(top_ref == ref[list]) +
(refc == ref[list]);
if (match_count > 1) { // most common
mv[list] = pack16to32(mid_pred(A[0], B[0], C[0]),
mid_pred(A[1], B[1], C[1]));
} else {
assert(match_count == 1);
if (left_ref == ref[list])
mv[list] = AV_RN32A(A);
else if (top_ref == ref[list])
mv[list] = AV_RN32A(B);
else
mv[list] = AV_RN32A(C); |
af0d6581 |
} |
cc2cfdc4 |
av_assert2(ref[list] < (sl->ref_count[list] << !!FRAME_MBAFF(h))); |
2db953f8 |
} else {
int mask = ~(MB_TYPE_L0 << (2 * list));
mv[list] = 0; |
af0d6581 |
ref[list] = -1; |
2db953f8 |
if (!is_b8x8) |
af0d6581 |
*mb_type &= mask;
sub_mb_type &= mask;
}
} |
2db953f8 |
if (ref[0] < 0 && ref[1] < 0) { |
af0d6581 |
ref[0] = ref[1] = 0; |
2db953f8 |
if (!is_b8x8) |
af0d6581 |
*mb_type |= MB_TYPE_L0L1;
sub_mb_type |= MB_TYPE_L0L1;
}
|
2db953f8 |
if (!(is_b8x8 | mv[0] | mv[1])) { |
e6287f07 |
fill_rectangle(&sl->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
fill_rectangle(&sl->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
fill_rectangle(&sl->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
fill_rectangle(&sl->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4); |
2db953f8 |
*mb_type = (*mb_type & ~(MB_TYPE_8x8 | MB_TYPE_16x8 | MB_TYPE_8x16 |
MB_TYPE_P1L0 | MB_TYPE_P1L1)) |
MB_TYPE_16x16 | MB_TYPE_DIRECT2; |
af0d6581 |
return;
}
|
a12d3188 |
if (IS_INTERLACED(sl->ref_list[1][0].parent->mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL |
2db953f8 |
if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL |
d4d9068c |
mb_y = (sl->mb_y & ~1) + sl->col_parity;
mb_xy = sl->mb_x +
((sl->mb_y & ~1) + sl->col_parity) * h->mb_stride; |
943f69a6 |
b8_stride = 0; |
2db953f8 |
} else { |
be69f0a8 |
mb_y += sl->col_fieldoff;
mb_xy += h->mb_stride * sl->col_fieldoff; // non-zero for FL -> FL & differ parity |
943f69a6 |
}
goto single_col; |
2db953f8 |
} else { // AFL/AFR/FR/FL -> AFR/FR
if (IS_INTERLACED(*mb_type)) { // AFL /FL -> AFR/FR |
d4d9068c |
mb_y = sl->mb_y & ~1;
mb_xy = (sl->mb_y & ~1) * h->mb_stride + sl->mb_x; |
a12d3188 |
mb_type_col[0] = sl->ref_list[1][0].parent->mb_type[mb_xy];
mb_type_col[1] = sl->ref_list[1][0].parent->mb_type[mb_xy + h->mb_stride]; |
2db953f8 |
b8_stride = 2 + 4 * h->mb_stride;
b4_stride *= 6;
if (IS_INTERLACED(mb_type_col[0]) !=
IS_INTERLACED(mb_type_col[1])) { |
833a1959 |
mb_type_col[0] &= ~MB_TYPE_INTERLACED;
mb_type_col[1] &= ~MB_TYPE_INTERLACED;
} |
668fd142 |
|
2db953f8 |
sub_mb_type |= MB_TYPE_16x16 | MB_TYPE_DIRECT2; /* B_SUB_8x8 */
if ((mb_type_col[0] & MB_TYPE_16x16_OR_INTRA) &&
(mb_type_col[1] & MB_TYPE_16x16_OR_INTRA) &&
!is_b8x8) {
*mb_type |= MB_TYPE_16x8 | MB_TYPE_DIRECT2; /* B_16x8 */
} else {
*mb_type |= MB_TYPE_8x8; |
943f69a6 |
} |
2db953f8 |
} else { // AFR/FR -> AFR/FR |
943f69a6 |
single_col:
mb_type_col[0] = |
a12d3188 |
mb_type_col[1] = sl->ref_list[1][0].parent->mb_type[mb_xy]; |
daaf6e3e |
|
2db953f8 |
sub_mb_type |= MB_TYPE_16x16 | MB_TYPE_DIRECT2; /* B_SUB_8x8 */
if (!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)) {
*mb_type |= MB_TYPE_16x16 | MB_TYPE_DIRECT2; /* B_16x16 */
} else if (!is_b8x8 &&
(mb_type_col[0] & (MB_TYPE_16x8 | MB_TYPE_8x16))) {
*mb_type |= MB_TYPE_DIRECT2 |
(mb_type_col[0] & (MB_TYPE_16x8 | MB_TYPE_8x16));
} else {
if (!h->sps.direct_8x8_inference_flag) {
/* FIXME: Save sub mb types from previous frames (or derive
* from MVs) so we know exactly what block size to use. */
sub_mb_type += (MB_TYPE_8x8 - MB_TYPE_16x16); /* B_SUB_4x4 */ |
daaf6e3e |
} |
2db953f8 |
*mb_type |= MB_TYPE_8x8; |
943f69a6 |
}
}
}
|
95eb35f3 |
await_reference_mb_row(h, &sl->ref_list[1][0], mb_y); |
d375c104 |
|
d8151a7e |
l1mv0 = (void*)&sl->ref_list[1][0].parent->motion_val[0][h->mb2b_xy[mb_xy]];
l1mv1 = (void*)&sl->ref_list[1][0].parent->motion_val[1][h->mb2b_xy[mb_xy]]; |
a12d3188 |
l1ref0 = &sl->ref_list[1][0].parent->ref_index[0][4 * mb_xy];
l1ref1 = &sl->ref_list[1][0].parent->ref_index[1][4 * mb_xy]; |
2db953f8 |
if (!b8_stride) { |
d4d9068c |
if (sl->mb_y & 1) { |
358b5b1a |
l1ref0 += 2;
l1ref1 += 2; |
2db953f8 |
l1mv0 += 2 * b4_stride;
l1mv1 += 2 * b4_stride; |
943f69a6 |
}
}
|
2db953f8 |
if (IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])) {
int n = 0;
for (i8 = 0; i8 < 4; i8++) {
int x8 = i8 & 1;
int y8 = i8 >> 1;
int xy8 = x8 + y8 * b8_stride;
int xy4 = x8 * 3 + y8 * b4_stride;
int a, b;
|
ed451a08 |
if (is_b8x8 && !IS_DIRECT(sl->sub_mb_type[i8])) |
2db953f8 |
continue; |
ed451a08 |
sl->sub_mb_type[i8] = sub_mb_type; |
2db953f8 |
|
e6287f07 |
fill_rectangle(&sl->ref_cache[0][scan8[i8 * 4]], 2, 2, 8, |
2db953f8 |
(uint8_t)ref[0], 1); |
e6287f07 |
fill_rectangle(&sl->ref_cache[1][scan8[i8 * 4]], 2, 2, 8, |
2db953f8 |
(uint8_t)ref[1], 1); |
a12d3188 |
if (!IS_INTRA(mb_type_col[y8]) && !sl->ref_list[1][0].parent->long_ref && |
2db953f8 |
((l1ref0[xy8] == 0 &&
FFABS(l1mv0[xy4][0]) <= 1 &&
FFABS(l1mv0[xy4][1]) <= 1) ||
(l1ref0[xy8] < 0 &&
l1ref1[xy8] == 0 &&
FFABS(l1mv1[xy4][0]) <= 1 &&
FFABS(l1mv1[xy4][1]) <= 1))) {
a =
b = 0;
if (ref[0] > 0)
a = mv[0];
if (ref[1] > 0)
b = mv[1];
n++;
} else {
a = mv[0];
b = mv[1]; |
943f69a6 |
} |
e6287f07 |
fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2, 8, a, 4);
fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2, 8, b, 4); |
2db953f8 |
}
if (!is_b8x8 && !(n & 3))
*mb_type = (*mb_type & ~(MB_TYPE_8x8 | MB_TYPE_16x8 | MB_TYPE_8x16 |
MB_TYPE_P1L0 | MB_TYPE_P1L1)) |
MB_TYPE_16x16 | MB_TYPE_DIRECT2;
} else if (IS_16X16(*mb_type)) {
int a, b; |
943f69a6 |
|
e6287f07 |
fill_rectangle(&sl->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
fill_rectangle(&sl->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1); |
a12d3188 |
if (!IS_INTRA(mb_type_col[0]) && !sl->ref_list[1][0].parent->long_ref && |
2db953f8 |
((l1ref0[0] == 0 &&
FFABS(l1mv0[0][0]) <= 1 &&
FFABS(l1mv0[0][1]) <= 1) ||
(l1ref0[0] < 0 && !l1ref1[0] &&
FFABS(l1mv1[0][0]) <= 1 &&
FFABS(l1mv1[0][1]) <= 1 &&
h->x264_build > 33U))) {
a = b = 0;
if (ref[0] > 0)
a = mv[0];
if (ref[1] > 0)
b = mv[1];
} else {
a = mv[0];
b = mv[1];
} |
e6287f07 |
fill_rectangle(&sl->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
fill_rectangle(&sl->mv_cache[1][scan8[0]], 4, 4, 8, b, 4); |
2db953f8 |
} else {
int n = 0;
for (i8 = 0; i8 < 4; i8++) {
const int x8 = i8 & 1;
const int y8 = i8 >> 1;
|
ed451a08 |
if (is_b8x8 && !IS_DIRECT(sl->sub_mb_type[i8])) |
2db953f8 |
continue; |
ed451a08 |
sl->sub_mb_type[i8] = sub_mb_type; |
2db953f8 |
|
e6287f07 |
fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2, 8, mv[0], 4);
fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2, 8, mv[1], 4);
fill_rectangle(&sl->ref_cache[0][scan8[i8 * 4]], 2, 2, 8, |
2db953f8 |
(uint8_t)ref[0], 1); |
e6287f07 |
fill_rectangle(&sl->ref_cache[1][scan8[i8 * 4]], 2, 2, 8, |
2db953f8 |
(uint8_t)ref[1], 1);
assert(b8_stride == 2);
/* col_zero_flag */ |
a12d3188 |
if (!IS_INTRA(mb_type_col[0]) && !sl->ref_list[1][0].parent->long_ref && |
2db953f8 |
(l1ref0[i8] == 0 ||
(l1ref0[i8] < 0 &&
l1ref1[i8] == 0 &&
h->x264_build > 33U))) {
const int16_t (*l1mv)[2] = l1ref0[i8] == 0 ? l1mv0 : l1mv1;
if (IS_SUB_8X8(sub_mb_type)) {
const int16_t *mv_col = l1mv[x8 * 3 + y8 * 3 * b4_stride];
if (FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1) {
if (ref[0] == 0) |
e6287f07 |
fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2, |
2db953f8 |
8, 0, 4);
if (ref[1] == 0) |
e6287f07 |
fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2, |
2db953f8 |
8, 0, 4);
n += 4;
}
} else {
int m = 0;
for (i4 = 0; i4 < 4; i4++) {
const int16_t *mv_col = l1mv[x8 * 2 + (i4 & 1) +
(y8 * 2 + (i4 >> 1)) * b4_stride];
if (FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1) {
if (ref[0] == 0) |
e6287f07 |
AV_ZERO32(sl->mv_cache[0][scan8[i8 * 4 + i4]]); |
2db953f8 |
if (ref[1] == 0) |
e6287f07 |
AV_ZERO32(sl->mv_cache[1][scan8[i8 * 4 + i4]]); |
8b9eb2df |
m++; |
943f69a6 |
}
} |
2db953f8 |
if (!(m & 3)) |
ed451a08 |
sl->sub_mb_type[i8] += MB_TYPE_16x16 - MB_TYPE_8x8; |
2db953f8 |
n += m; |
943f69a6 |
}
}
} |
2db953f8 |
if (!is_b8x8 && !(n & 15))
*mb_type = (*mb_type & ~(MB_TYPE_8x8 | MB_TYPE_16x8 | MB_TYPE_8x16 |
MB_TYPE_P1L0 | MB_TYPE_P1L1)) |
MB_TYPE_16x16 | MB_TYPE_DIRECT2;
} |
edb4df43 |
}
|
d27b8248 |
static void pred_temp_direct_motion(const H264Context *const h, H264SliceContext *sl, |
e6287f07 |
int *mb_type) |
2db953f8 |
{ |
358b5b1a |
int b8_stride = 2; |
edb4df43 |
int b4_stride = h->b_stride; |
d4d9068c |
int mb_xy = sl->mb_xy, mb_y = sl->mb_y; |
edb4df43 |
int mb_type_col[2];
const int16_t (*l1mv0)[2], (*l1mv1)[2];
const int8_t *l1ref0, *l1ref1;
const int is_b8x8 = IS_8X8(*mb_type);
unsigned int sub_mb_type;
int i8, i4;
|
95eb35f3 |
assert(sl->ref_list[1][0].reference & 3); |
edb4df43 |
|
95eb35f3 |
await_reference_mb_row(h, &sl->ref_list[1][0], |
d4d9068c |
sl->mb_y + !!IS_INTERLACED(*mb_type)); |
d375c104 |
|
a12d3188 |
if (IS_INTERLACED(sl->ref_list[1][0].parent->mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL |
2db953f8 |
if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL |
d4d9068c |
mb_y = (sl->mb_y & ~1) + sl->col_parity;
mb_xy = sl->mb_x +
((sl->mb_y & ~1) + sl->col_parity) * h->mb_stride; |
edb4df43 |
b8_stride = 0; |
2db953f8 |
} else { |
be69f0a8 |
mb_y += sl->col_fieldoff;
mb_xy += h->mb_stride * sl->col_fieldoff; // non-zero for FL -> FL & differ parity |
edb4df43 |
}
goto single_col; |
2db953f8 |
} else { // AFL/AFR/FR/FL -> AFR/FR
if (IS_INTERLACED(*mb_type)) { // AFL /FL -> AFR/FR |
d4d9068c |
mb_y = sl->mb_y & ~1;
mb_xy = sl->mb_x + (sl->mb_y & ~1) * h->mb_stride; |
a12d3188 |
mb_type_col[0] = sl->ref_list[1][0].parent->mb_type[mb_xy];
mb_type_col[1] = sl->ref_list[1][0].parent->mb_type[mb_xy + h->mb_stride]; |
2db953f8 |
b8_stride = 2 + 4 * h->mb_stride;
b4_stride *= 6;
if (IS_INTERLACED(mb_type_col[0]) !=
IS_INTERLACED(mb_type_col[1])) { |
758ec111 |
mb_type_col[0] &= ~MB_TYPE_INTERLACED;
mb_type_col[1] &= ~MB_TYPE_INTERLACED;
} |
edb4df43 |
|
2db953f8 |
sub_mb_type = MB_TYPE_16x16 | MB_TYPE_P0L0 | MB_TYPE_P0L1 |
MB_TYPE_DIRECT2; /* B_SUB_8x8 */ |
edb4df43 |
|
2db953f8 |
if ((mb_type_col[0] & MB_TYPE_16x16_OR_INTRA) &&
(mb_type_col[1] & MB_TYPE_16x16_OR_INTRA) &&
!is_b8x8) {
*mb_type |= MB_TYPE_16x8 | MB_TYPE_L0L1 |
MB_TYPE_DIRECT2; /* B_16x8 */
} else {
*mb_type |= MB_TYPE_8x8 | MB_TYPE_L0L1; |
edb4df43 |
} |
2db953f8 |
} else { // AFR/FR -> AFR/FR |
edb4df43 |
single_col: |
2db953f8 |
mb_type_col[0] = |
a12d3188 |
mb_type_col[1] = sl->ref_list[1][0].parent->mb_type[mb_xy]; |
2db953f8 |
sub_mb_type = MB_TYPE_16x16 | MB_TYPE_P0L0 | MB_TYPE_P0L1 |
MB_TYPE_DIRECT2; /* B_SUB_8x8 */
if (!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)) {
*mb_type |= MB_TYPE_16x16 | MB_TYPE_P0L0 | MB_TYPE_P0L1 |
MB_TYPE_DIRECT2; /* B_16x16 */
} else if (!is_b8x8 &&
(mb_type_col[0] & (MB_TYPE_16x8 | MB_TYPE_8x16))) {
*mb_type |= MB_TYPE_L0L1 | MB_TYPE_DIRECT2 |
(mb_type_col[0] & (MB_TYPE_16x8 | MB_TYPE_8x16));
} else {
if (!h->sps.direct_8x8_inference_flag) {
/* FIXME: save sub mb types from previous frames (or derive
* from MVs) so we know exactly what block size to use */
sub_mb_type = MB_TYPE_8x8 | MB_TYPE_P0L0 | MB_TYPE_P0L1 |
MB_TYPE_DIRECT2; /* B_SUB_4x4 */ |
edb4df43 |
} |
2db953f8 |
*mb_type |= MB_TYPE_8x8 | MB_TYPE_L0L1; |
edb4df43 |
}
}
}
|
95eb35f3 |
await_reference_mb_row(h, &sl->ref_list[1][0], mb_y); |
d375c104 |
|
d8151a7e |
l1mv0 = (void*)&sl->ref_list[1][0].parent->motion_val[0][h->mb2b_xy[mb_xy]];
l1mv1 = (void*)&sl->ref_list[1][0].parent->motion_val[1][h->mb2b_xy[mb_xy]]; |
a12d3188 |
l1ref0 = &sl->ref_list[1][0].parent->ref_index[0][4 * mb_xy];
l1ref1 = &sl->ref_list[1][0].parent->ref_index[1][4 * mb_xy]; |
2db953f8 |
if (!b8_stride) { |
d4d9068c |
if (sl->mb_y & 1) { |
358b5b1a |
l1ref0 += 2;
l1ref1 += 2; |
2db953f8 |
l1mv0 += 2 * b4_stride;
l1mv1 += 2 * b4_stride; |
edb4df43 |
}
}
{ |
77477266 |
const int *map_col_to_list0[2] = { sl->map_col_to_list0[0],
sl->map_col_to_list0[1] }; |
ee0d774d |
const int *dist_scale_factor = sl->dist_scale_factor; |
542d4a75 |
int ref_offset; |
943f69a6 |
|
7bece9b2 |
if (FRAME_MBAFF(h) && IS_INTERLACED(*mb_type)) { |
d4d9068c |
map_col_to_list0[0] = sl->map_col_to_list0_field[sl->mb_y & 1][0];
map_col_to_list0[1] = sl->map_col_to_list0_field[sl->mb_y & 1][1];
dist_scale_factor = sl->dist_scale_factor_field[sl->mb_y & 1]; |
943f69a6 |
} |
a12d3188 |
ref_offset = (sl->ref_list[1][0].parent->mbaff << 4) & (mb_type_col[0] >> 3); |
943f69a6 |
|
2db953f8 |
if (IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])) {
int y_shift = 2 * !IS_INTERLACED(*mb_type); |
5aa1d759 |
assert(h->sps.direct_8x8_inference_flag); |
943f69a6 |
|
2db953f8 |
for (i8 = 0; i8 < 4; i8++) {
const int x8 = i8 & 1;
const int y8 = i8 >> 1; |
943f69a6 |
int ref0, scale; |
2db953f8 |
const int16_t (*l1mv)[2] = l1mv0; |
943f69a6 |
|
ed451a08 |
if (is_b8x8 && !IS_DIRECT(sl->sub_mb_type[i8])) |
943f69a6 |
continue; |
ed451a08 |
sl->sub_mb_type[i8] = sub_mb_type; |
943f69a6 |
|
e6287f07 |
fill_rectangle(&sl->ref_cache[1][scan8[i8 * 4]], 2, 2, 8, 0, 1); |
2db953f8 |
if (IS_INTRA(mb_type_col[y8])) { |
e6287f07 |
fill_rectangle(&sl->ref_cache[0][scan8[i8 * 4]], 2, 2, 8, 0, 1);
fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2, 8, 0, 4);
fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2, 8, 0, 4); |
943f69a6 |
continue;
}
|
2db953f8 |
ref0 = l1ref0[x8 + y8 * b8_stride];
if (ref0 >= 0) |
943f69a6 |
ref0 = map_col_to_list0[0][ref0 + ref_offset]; |
2db953f8 |
else {
ref0 = map_col_to_list0[1][l1ref1[x8 + y8 * b8_stride] +
ref_offset];
l1mv = l1mv1; |
943f69a6 |
}
scale = dist_scale_factor[ref0]; |
e6287f07 |
fill_rectangle(&sl->ref_cache[0][scan8[i8 * 4]], 2, 2, 8, |
2db953f8 |
ref0, 1); |
943f69a6 |
{ |
2db953f8 |
const int16_t *mv_col = l1mv[x8 * 3 + y8 * b4_stride];
int my_col = (mv_col[1] << y_shift) / 2;
int mx = (scale * mv_col[0] + 128) >> 8;
int my = (scale * my_col + 128) >> 8; |
e6287f07 |
fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2, 8, |
2db953f8 |
pack16to32(mx, my), 4); |
e6287f07 |
fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2, 8, |
2db953f8 |
pack16to32(mx - mv_col[0], my - my_col), 4); |
943f69a6 |
}
}
return;
}
/* one-to-one mv scaling */
|
2db953f8 |
if (IS_16X16(*mb_type)) { |
943f69a6 |
int ref, mv0, mv1;
|
e6287f07 |
fill_rectangle(&sl->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1); |
2db953f8 |
if (IS_INTRA(mb_type_col[0])) {
ref = mv0 = mv1 = 0;
} else { |
943f69a6 |
const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
: map_col_to_list0[1][l1ref1[0] + ref_offset];
const int scale = dist_scale_factor[ref0];
const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
int mv_l0[2];
mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
mv_l0[1] = (scale * mv_col[1] + 128) >> 8; |
2db953f8 |
ref = ref0;
mv0 = pack16to32(mv_l0[0], mv_l0[1]);
mv1 = pack16to32(mv_l0[0] - mv_col[0], mv_l0[1] - mv_col[1]); |
943f69a6 |
} |
e6287f07 |
fill_rectangle(&sl->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
fill_rectangle(&sl->mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
fill_rectangle(&sl->mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4); |
2db953f8 |
} else {
for (i8 = 0; i8 < 4; i8++) {
const int x8 = i8 & 1;
const int y8 = i8 >> 1; |
943f69a6 |
int ref0, scale; |
2db953f8 |
const int16_t (*l1mv)[2] = l1mv0; |
943f69a6 |
|
ed451a08 |
if (is_b8x8 && !IS_DIRECT(sl->sub_mb_type[i8])) |
943f69a6 |
continue; |
ed451a08 |
sl->sub_mb_type[i8] = sub_mb_type; |
e6287f07 |
fill_rectangle(&sl->ref_cache[1][scan8[i8 * 4]], 2, 2, 8, 0, 1); |
2db953f8 |
if (IS_INTRA(mb_type_col[0])) { |
e6287f07 |
fill_rectangle(&sl->ref_cache[0][scan8[i8 * 4]], 2, 2, 8, 0, 1);
fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2, 8, 0, 4);
fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2, 8, 0, 4); |
943f69a6 |
continue;
}
|
358b5b1a |
assert(b8_stride == 2);
ref0 = l1ref0[i8]; |
2db953f8 |
if (ref0 >= 0) |
50f8e985 |
ref0 = map_col_to_list0[0][ref0 + ref_offset]; |
2db953f8 |
else { |
358b5b1a |
ref0 = map_col_to_list0[1][l1ref1[i8] + ref_offset]; |
2db953f8 |
l1mv = l1mv1; |
943f69a6 |
}
scale = dist_scale_factor[ref0];
|
e6287f07 |
fill_rectangle(&sl->ref_cache[0][scan8[i8 * 4]], 2, 2, 8, |
2db953f8 |
ref0, 1);
if (IS_SUB_8X8(sub_mb_type)) {
const int16_t *mv_col = l1mv[x8 * 3 + y8 * 3 * b4_stride];
int mx = (scale * mv_col[0] + 128) >> 8;
int my = (scale * mv_col[1] + 128) >> 8; |
e6287f07 |
fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2, 8, |
2db953f8 |
pack16to32(mx, my), 4); |
e6287f07 |
fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2, 8, |
2db953f8 |
pack16to32(mx - mv_col[0], my - mv_col[1]), 4);
} else {
for (i4 = 0; i4 < 4; i4++) {
const int16_t *mv_col = l1mv[x8 * 2 + (i4 & 1) +
(y8 * 2 + (i4 >> 1)) * b4_stride]; |
e6287f07 |
int16_t *mv_l0 = sl->mv_cache[0][scan8[i8 * 4 + i4]]; |
2db953f8 |
mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
mv_l0[1] = (scale * mv_col[1] + 128) >> 8; |
e6287f07 |
AV_WN32A(sl->mv_cache[1][scan8[i8 * 4 + i4]], |
2db953f8 |
pack16to32(mv_l0[0] - mv_col[0],
mv_l0[1] - mv_col[1]));
} |
943f69a6 |
}
}
}
}
} |
edb4df43 |
|
d27b8248 |
void ff_h264_pred_direct_motion(const H264Context *const h, H264SliceContext *sl, |
e6287f07 |
int *mb_type) |
2db953f8 |
{ |
066aafce |
if (sl->direct_spatial_mv_pred) |
e6287f07 |
pred_spatial_direct_motion(h, sl, mb_type); |
2db953f8 |
else |
e6287f07 |
pred_temp_direct_motion(h, sl, mb_type); |
edb4df43 |
} |