| 28fff0d9 |
/*
* H.26L/H.264/AVC/JVT/14496-10/... decoder
* Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
* |
| 24823a76 |
* This file is part of FFmpeg. |
| 28fff0d9 |
* |
| 24823a76 |
* FFmpeg is free software; you can redistribute it and/or |
| 28fff0d9 |
* 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.
* |
| 24823a76 |
* FFmpeg is distributed in the hope that it will be useful, |
| 28fff0d9 |
* 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 |
| 24823a76 |
* License along with FFmpeg; if not, write to the Free Software |
| 28fff0d9 |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#undef FUNC
#undef PIXEL_SHIFT
#if SIMPLE
# define FUNC(n) AV_JOIN(n ## _simple_, BITS)
# define PIXEL_SHIFT (BITS >> 4)
#else
# define FUNC(n) n ## _complex
# define PIXEL_SHIFT h->pixel_shift
#endif
#undef CHROMA_IDC
#define CHROMA_IDC 1
#include "h264_mc_template.c"
#undef CHROMA_IDC
#define CHROMA_IDC 2
#include "h264_mc_template.c"
|
| 6490a0c0 |
static av_noinline void FUNC(hl_decode_mb)(const H264Context *h, H264SliceContext *sl) |
| 28fff0d9 |
{ |
| d4d9068c |
const int mb_x = sl->mb_x;
const int mb_y = sl->mb_y; |
| 0edbe6fa |
const int mb_xy = sl->mb_xy; |
| 759001c5 |
const int mb_type = h->cur_pic.mb_type[mb_xy]; |
| 28fff0d9 |
uint8_t *dest_y, *dest_cb, *dest_cr;
int linesize, uvlinesize /*dct_offset*/;
int i, j; |
| 51822879 |
const int *block_offset = &h->block_offset[0]; |
| 3176217c |
const int transform_bypass = !SIMPLE && (sl->qscale == 0 && h->ps.sps->transform_bypass); |
| 88bd7fdc |
void (*idct_add)(uint8_t *dst, int16_t *block, int stride); |
| 2c541554 |
const int block_h = 16 >> h->chroma_y_shift; |
| e962bd08 |
const int chroma422 = CHROMA422(h); |
| 28fff0d9 |
|
| a0f29460 |
dest_y = h->cur_pic.f->data[0] + ((mb_x << PIXEL_SHIFT) + mb_y * sl->linesize) * 16;
dest_cb = h->cur_pic.f->data[1] + (mb_x << PIXEL_SHIFT) * 8 + mb_y * sl->uvlinesize * block_h;
dest_cr = h->cur_pic.f->data[2] + (mb_x << PIXEL_SHIFT) * 8 + mb_y * sl->uvlinesize * block_h; |
| 28fff0d9 |
|
| c28ed1d7 |
h->vdsp.prefetch(dest_y + (sl->mb_x & 3) * 4 * sl->linesize + (64 << PIXEL_SHIFT), sl->linesize, 4);
h->vdsp.prefetch(dest_cb + (sl->mb_x & 7) * sl->uvlinesize + (64 << PIXEL_SHIFT), dest_cr - dest_cb, 2); |
| 28fff0d9 |
|
| 95eb35f3 |
h->list_counts[mb_xy] = sl->list_count; |
| 28fff0d9 |
|
| bc98e8c0 |
if (!SIMPLE && MB_FIELD(sl)) { |
| c28ed1d7 |
linesize = sl->mb_linesize = sl->linesize * 2;
uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2; |
| 28fff0d9 |
block_offset = &h->block_offset[48];
if (mb_y & 1) { // FIXME move out of this function? |
| c28ed1d7 |
dest_y -= sl->linesize * 15;
dest_cb -= sl->uvlinesize * (block_h - 1);
dest_cr -= sl->uvlinesize * (block_h - 1); |
| 28fff0d9 |
} |
| 7bece9b2 |
if (FRAME_MBAFF(h)) { |
| 28fff0d9 |
int list; |
| 95eb35f3 |
for (list = 0; list < sl->list_count; list++) { |
| 28fff0d9 |
if (!USES_LIST(mb_type, list))
continue;
if (IS_16X16(mb_type)) { |
| e6287f07 |
int8_t *ref = &sl->ref_cache[list][scan8[0]]; |
| d4d9068c |
fill_rectangle(ref, 4, 4, 8, (16 + *ref) ^ (sl->mb_y & 1), 1); |
| 28fff0d9 |
} else {
for (i = 0; i < 16; i += 4) { |
| e6287f07 |
int ref = sl->ref_cache[list][scan8[i]]; |
| 28fff0d9 |
if (ref >= 0) |
| e6287f07 |
fill_rectangle(&sl->ref_cache[list][scan8[i]], 2, 2, |
| d4d9068c |
8, (16 + ref) ^ (sl->mb_y & 1), 1); |
| 28fff0d9 |
}
}
}
}
} else { |
| c28ed1d7 |
linesize = sl->mb_linesize = sl->linesize;
uvlinesize = sl->mb_uvlinesize = sl->uvlinesize; |
| 28fff0d9 |
// dct_offset = s->linesize * 16;
}
if (!SIMPLE && IS_INTRA_PCM(mb_type)) { |
| 1534ef87 |
const int bit_depth = h->ps.sps->bit_depth_luma; |
| 28fff0d9 |
if (PIXEL_SHIFT) {
int j;
GetBitContext gb; |
| b063582e |
init_get_bits(&gb, sl->intra_pcm_ptr, |
| 3176217c |
ff_h264_mb_sizes[h->ps.sps->chroma_format_idc] * bit_depth); |
| 28fff0d9 |
for (i = 0; i < 16; i++) {
uint16_t *tmp_y = (uint16_t *)(dest_y + i * linesize);
for (j = 0; j < 16; j++)
tmp_y[j] = get_bits(&gb, bit_depth);
} |
| 7c6eb0a1 |
if (SIMPLE || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) { |
| 3176217c |
if (!h->ps.sps->chroma_format_idc) { |
| 28fff0d9 |
for (i = 0; i < block_h; i++) {
uint16_t *tmp_cb = (uint16_t *)(dest_cb + i * uvlinesize);
uint16_t *tmp_cr = (uint16_t *)(dest_cr + i * uvlinesize); |
| 24823a76 |
for (j = 0; j < 8; j++) {
tmp_cb[j] = tmp_cr[j] = 1 << (bit_depth - 1);
} |
| 28fff0d9 |
}
} else {
for (i = 0; i < block_h; i++) {
uint16_t *tmp_cb = (uint16_t *)(dest_cb + i * uvlinesize);
for (j = 0; j < 8; j++)
tmp_cb[j] = get_bits(&gb, bit_depth);
}
for (i = 0; i < block_h; i++) {
uint16_t *tmp_cr = (uint16_t *)(dest_cr + i * uvlinesize);
for (j = 0; j < 8; j++)
tmp_cr[j] = get_bits(&gb, bit_depth);
}
}
}
} else {
for (i = 0; i < 16; i++) |
| b063582e |
memcpy(dest_y + i * linesize, sl->intra_pcm_ptr + i * 16, 16); |
| 7c6eb0a1 |
if (SIMPLE || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) { |
| 3176217c |
if (!h->ps.sps->chroma_format_idc) { |
| 24823a76 |
for (i = 0; i < 8; i++) { |
| 4fb1221e |
memset(dest_cb + i * uvlinesize, 1 << (bit_depth - 1), 8);
memset(dest_cr + i * uvlinesize, 1 << (bit_depth - 1), 8); |
| 28fff0d9 |
}
} else { |
| b063582e |
const uint8_t *src_cb = sl->intra_pcm_ptr + 256;
const uint8_t *src_cr = sl->intra_pcm_ptr + 256 + block_h * 8; |
| 28fff0d9 |
for (i = 0; i < block_h; i++) {
memcpy(dest_cb + i * uvlinesize, src_cb + i * 8, 8);
memcpy(dest_cr + i * uvlinesize, src_cr + i * 8, 8);
}
}
}
}
} else {
if (IS_INTRA(mb_type)) { |
| e6c90ce9 |
if (sl->deblocking_filter) |
| 8b00f4df |
xchg_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize, |
| 28fff0d9 |
uvlinesize, 1, 0, SIMPLE, PIXEL_SHIFT);
|
| 7c6eb0a1 |
if (SIMPLE || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) { |
| 4bd5ac20 |
h->hpc.pred8x8[sl->chroma_pred_mode](dest_cb, uvlinesize);
h->hpc.pred8x8[sl->chroma_pred_mode](dest_cr, uvlinesize); |
| 28fff0d9 |
}
|
| 9b30f8dd |
hl_decode_mb_predict_luma(h, sl, mb_type, SIMPLE, |
| 28fff0d9 |
transform_bypass, PIXEL_SHIFT,
block_offset, linesize, dest_y, 0);
|
| e6c90ce9 |
if (sl->deblocking_filter) |
| 8b00f4df |
xchg_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize, |
| 28fff0d9 |
uvlinesize, 0, 0, SIMPLE, PIXEL_SHIFT); |
| 9b30f8dd |
} else { |
| 28fff0d9 |
if (chroma422) { |
| 92c6c2a6 |
FUNC(hl_motion_422)(h, sl, dest_y, dest_cb, dest_cr, |
| 56087ec0 |
h->h264qpel.put_h264_qpel_pixels_tab,
h->h264chroma.put_h264_chroma_pixels_tab,
h->h264qpel.avg_h264_qpel_pixels_tab,
h->h264chroma.avg_h264_chroma_pixels_tab, |
| 28fff0d9 |
h->h264dsp.weight_h264_pixels_tab,
h->h264dsp.biweight_h264_pixels_tab);
} else { |
| 92c6c2a6 |
FUNC(hl_motion_420)(h, sl, dest_y, dest_cb, dest_cr, |
| 56087ec0 |
h->h264qpel.put_h264_qpel_pixels_tab,
h->h264chroma.put_h264_chroma_pixels_tab,
h->h264qpel.avg_h264_qpel_pixels_tab,
h->h264chroma.avg_h264_chroma_pixels_tab, |
| 28fff0d9 |
h->h264dsp.weight_h264_pixels_tab,
h->h264dsp.biweight_h264_pixels_tab);
}
}
|
| 9b30f8dd |
hl_decode_mb_idct_luma(h, sl, mb_type, SIMPLE, transform_bypass, |
| 28fff0d9 |
PIXEL_SHIFT, block_offset, linesize, dest_y, 0);
|
| 7c6eb0a1 |
if ((SIMPLE || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) && |
| e7226984 |
(sl->cbp & 0x30)) { |
| 28fff0d9 |
uint8_t *dest[2] = { dest_cb, dest_cr };
if (transform_bypass) { |
| 3176217c |
if (IS_INTRA(mb_type) && h->ps.sps->profile_idc == 244 && |
| 4bd5ac20 |
(sl->chroma_pred_mode == VERT_PRED8x8 ||
sl->chroma_pred_mode == HOR_PRED8x8)) {
h->hpc.pred8x8_add[sl->chroma_pred_mode](dest[0], |
| 28fff0d9 |
block_offset + 16, |
| bf03a878 |
sl->mb + (16 * 16 * 1 << PIXEL_SHIFT), |
| 28fff0d9 |
uvlinesize); |
| 4bd5ac20 |
h->hpc.pred8x8_add[sl->chroma_pred_mode](dest[1], |
| 28fff0d9 |
block_offset + 32, |
| bf03a878 |
sl->mb + (16 * 16 * 2 << PIXEL_SHIFT), |
| 28fff0d9 |
uvlinesize);
} else { |
| 1acd7d59 |
idct_add = h->h264dsp.h264_add_pixels4_clear; |
| 28fff0d9 |
for (j = 1; j < 3; j++) {
for (i = j * 16; i < j * 16 + 4; i++) |
| f69574cf |
if (sl->non_zero_count_cache[scan8[i]] || |
| bf03a878 |
dctcoef_get(sl->mb, PIXEL_SHIFT, i * 16)) |
| 28fff0d9 |
idct_add(dest[j - 1] + block_offset[i], |
| bf03a878 |
sl->mb + (i * 16 << PIXEL_SHIFT), |
| 28fff0d9 |
uvlinesize);
if (chroma422) {
for (i = j * 16 + 4; i < j * 16 + 8; i++) |
| f69574cf |
if (sl->non_zero_count_cache[scan8[i + 4]] || |
| bf03a878 |
dctcoef_get(sl->mb, PIXEL_SHIFT, i * 16)) |
| 28fff0d9 |
idct_add(dest[j - 1] + block_offset[i + 4], |
| bf03a878 |
sl->mb + (i * 16 << PIXEL_SHIFT), |
| 28fff0d9 |
uvlinesize);
}
}
}
} else { |
| 9b30f8dd |
int qp[2];
if (chroma422) {
qp[0] = sl->chroma_qp[0] + 3;
qp[1] = sl->chroma_qp[1] + 3;
} else {
qp[0] = sl->chroma_qp[0];
qp[1] = sl->chroma_qp[1]; |
| 28fff0d9 |
} |
| 9b30f8dd |
if (sl->non_zero_count_cache[scan8[CHROMA_DC_BLOCK_INDEX + 0]])
h->h264dsp.h264_chroma_dc_dequant_idct(sl->mb + (16 * 16 * 1 << PIXEL_SHIFT), |
| 3176217c |
h->ps.pps->dequant4_coeff[IS_INTRA(mb_type) ? 1 : 4][qp[0]][0]); |
| 9b30f8dd |
if (sl->non_zero_count_cache[scan8[CHROMA_DC_BLOCK_INDEX + 1]])
h->h264dsp.h264_chroma_dc_dequant_idct(sl->mb + (16 * 16 * 2 << PIXEL_SHIFT), |
| 3176217c |
h->ps.pps->dequant4_coeff[IS_INTRA(mb_type) ? 2 : 5][qp[1]][0]); |
| 9b30f8dd |
h->h264dsp.h264_idct_add8(dest, block_offset,
sl->mb, uvlinesize,
sl->non_zero_count_cache); |
| 28fff0d9 |
}
}
}
}
#if !SIMPLE || BITS == 8
#undef CHROMA_IDC
#define CHROMA_IDC 3
#include "h264_mc_template.c"
|
| 6490a0c0 |
static av_noinline void FUNC(hl_decode_mb_444)(const H264Context *h, H264SliceContext *sl) |
| 28fff0d9 |
{ |
| d4d9068c |
const int mb_x = sl->mb_x;
const int mb_y = sl->mb_y; |
| 0edbe6fa |
const int mb_xy = sl->mb_xy; |
| 759001c5 |
const int mb_type = h->cur_pic.mb_type[mb_xy]; |
| 28fff0d9 |
uint8_t *dest[3];
int linesize;
int i, j, p; |
| 51822879 |
const int *block_offset = &h->block_offset[0]; |
| 3176217c |
const int transform_bypass = !SIMPLE && (sl->qscale == 0 && h->ps.sps->transform_bypass); |
| 7c6eb0a1 |
const int plane_count = (SIMPLE || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) ? 3 : 1; |
| 28fff0d9 |
for (p = 0; p < plane_count; p++) { |
| a0f29460 |
dest[p] = h->cur_pic.f->data[p] + |
| c28ed1d7 |
((mb_x << PIXEL_SHIFT) + mb_y * sl->linesize) * 16;
h->vdsp.prefetch(dest[p] + (sl->mb_x & 3) * 4 * sl->linesize + (64 << PIXEL_SHIFT),
sl->linesize, 4); |
| 28fff0d9 |
}
|
| 95eb35f3 |
h->list_counts[mb_xy] = sl->list_count; |
| 28fff0d9 |
|
| bc98e8c0 |
if (!SIMPLE && MB_FIELD(sl)) { |
| c28ed1d7 |
linesize = sl->mb_linesize = sl->mb_uvlinesize = sl->linesize * 2; |
| 28fff0d9 |
block_offset = &h->block_offset[48];
if (mb_y & 1) // FIXME move out of this function?
for (p = 0; p < 3; p++) |
| c28ed1d7 |
dest[p] -= sl->linesize * 15; |
| 7bece9b2 |
if (FRAME_MBAFF(h)) { |
| 28fff0d9 |
int list; |
| 95eb35f3 |
for (list = 0; list < sl->list_count; list++) { |
| 28fff0d9 |
if (!USES_LIST(mb_type, list))
continue;
if (IS_16X16(mb_type)) { |
| e6287f07 |
int8_t *ref = &sl->ref_cache[list][scan8[0]]; |
| d4d9068c |
fill_rectangle(ref, 4, 4, 8, (16 + *ref) ^ (sl->mb_y & 1), 1); |
| 28fff0d9 |
} else {
for (i = 0; i < 16; i += 4) { |
| e6287f07 |
int ref = sl->ref_cache[list][scan8[i]]; |
| 28fff0d9 |
if (ref >= 0) |
| e6287f07 |
fill_rectangle(&sl->ref_cache[list][scan8[i]], 2, 2, |
| d4d9068c |
8, (16 + ref) ^ (sl->mb_y & 1), 1); |
| 28fff0d9 |
}
}
}
}
} else { |
| c28ed1d7 |
linesize = sl->mb_linesize = sl->mb_uvlinesize = sl->linesize; |
| 28fff0d9 |
}
if (!SIMPLE && IS_INTRA_PCM(mb_type)) {
if (PIXEL_SHIFT) { |
| 3176217c |
const int bit_depth = h->ps.sps->bit_depth_luma; |
| 28fff0d9 |
GetBitContext gb; |
| b063582e |
init_get_bits(&gb, sl->intra_pcm_ptr, 768 * bit_depth); |
| 28fff0d9 |
for (p = 0; p < plane_count; p++)
for (i = 0; i < 16; i++) {
uint16_t *tmp = (uint16_t *)(dest[p] + i * linesize);
for (j = 0; j < 16; j++)
tmp[j] = get_bits(&gb, bit_depth);
}
} else {
for (p = 0; p < plane_count; p++)
for (i = 0; i < 16; i++)
memcpy(dest[p] + i * linesize, |
| b063582e |
sl->intra_pcm_ptr + p * 256 + i * 16, 16); |
| 28fff0d9 |
}
} else {
if (IS_INTRA(mb_type)) { |
| e6c90ce9 |
if (sl->deblocking_filter) |
| 8b00f4df |
xchg_mb_border(h, sl, dest[0], dest[1], dest[2], linesize, |
| 28fff0d9 |
linesize, 1, 1, SIMPLE, PIXEL_SHIFT);
for (p = 0; p < plane_count; p++) |
| 9b30f8dd |
hl_decode_mb_predict_luma(h, sl, mb_type, SIMPLE, |
| 28fff0d9 |
transform_bypass, PIXEL_SHIFT,
block_offset, linesize, dest[p], p);
|
| e6c90ce9 |
if (sl->deblocking_filter) |
| 8b00f4df |
xchg_mb_border(h, sl, dest[0], dest[1], dest[2], linesize, |
| 28fff0d9 |
linesize, 0, 1, SIMPLE, PIXEL_SHIFT);
} else { |
| 92c6c2a6 |
FUNC(hl_motion_444)(h, sl, dest[0], dest[1], dest[2], |
| 56087ec0 |
h->h264qpel.put_h264_qpel_pixels_tab,
h->h264chroma.put_h264_chroma_pixels_tab,
h->h264qpel.avg_h264_qpel_pixels_tab,
h->h264chroma.avg_h264_chroma_pixels_tab, |
| 28fff0d9 |
h->h264dsp.weight_h264_pixels_tab,
h->h264dsp.biweight_h264_pixels_tab);
}
for (p = 0; p < plane_count; p++) |
| 9b30f8dd |
hl_decode_mb_idct_luma(h, sl, mb_type, SIMPLE, transform_bypass, |
| 28fff0d9 |
PIXEL_SHIFT, block_offset, linesize,
dest[p], p);
}
}
#endif |