2 * H.26L/H.264/AVC/JVT/14496-10/... decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of Libav.
7 * Libav is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * Libav is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 # define FUNC(n) AV_JOIN(n ## _simple_, BITS)
27 # define PIXEL_SHIFT (BITS >> 4)
29 # define FUNC(n) n ## _complex
30 # define PIXEL_SHIFT h->pixel_shift
35 #include "h264_mc_template.c"
39 #include "h264_mc_template.c"
41 static av_noinline void FUNC(hl_decode_mb)(const H264Context *h, H264SliceContext *sl)
43 const int mb_x = sl->mb_x;
44 const int mb_y = sl->mb_y;
45 const int mb_xy = sl->mb_xy;
46 const int mb_type = h->cur_pic.mb_type[mb_xy];
47 uint8_t *dest_y, *dest_cb, *dest_cr;
48 int linesize, uvlinesize /*dct_offset*/;
50 const int *block_offset = &h->block_offset[0];
51 const int transform_bypass = !SIMPLE && (sl->qscale == 0 && h->ps.sps->transform_bypass);
52 void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
53 const int block_h = 16 >> h->chroma_y_shift;
54 const int chroma422 = CHROMA422(h);
56 dest_y = h->cur_pic.f->data[0] + ((mb_x << PIXEL_SHIFT) + mb_y * sl->linesize) * 16;
57 dest_cb = h->cur_pic.f->data[1] + (mb_x << PIXEL_SHIFT) * 8 + mb_y * sl->uvlinesize * block_h;
58 dest_cr = h->cur_pic.f->data[2] + (mb_x << PIXEL_SHIFT) * 8 + mb_y * sl->uvlinesize * block_h;
60 h->vdsp.prefetch(dest_y + (sl->mb_x & 3) * 4 * sl->linesize + (64 << PIXEL_SHIFT), sl->linesize, 4);
61 h->vdsp.prefetch(dest_cb + (sl->mb_x & 7) * sl->uvlinesize + (64 << PIXEL_SHIFT), dest_cr - dest_cb, 2);
63 h->list_counts[mb_xy] = sl->list_count;
65 if (!SIMPLE && MB_FIELD(sl)) {
66 linesize = sl->mb_linesize = sl->linesize * 2;
67 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
68 block_offset = &h->block_offset[48];
69 if (mb_y & 1) { // FIXME move out of this function?
70 dest_y -= sl->linesize * 15;
71 dest_cb -= sl->uvlinesize * (block_h - 1);
72 dest_cr -= sl->uvlinesize * (block_h - 1);
76 for (list = 0; list < sl->list_count; list++) {
77 if (!USES_LIST(mb_type, list))
79 if (IS_16X16(mb_type)) {
80 int8_t *ref = &sl->ref_cache[list][scan8[0]];
81 fill_rectangle(ref, 4, 4, 8, (16 + *ref) ^ (sl->mb_y & 1), 1);
83 for (i = 0; i < 16; i += 4) {
84 int ref = sl->ref_cache[list][scan8[i]];
86 fill_rectangle(&sl->ref_cache[list][scan8[i]], 2, 2,
87 8, (16 + ref) ^ (sl->mb_y & 1), 1);
93 linesize = sl->mb_linesize = sl->linesize;
94 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
95 // dct_offset = s->linesize * 16;
98 if (!SIMPLE && IS_INTRA_PCM(mb_type)) {
100 const int bit_depth = h->ps.sps->bit_depth_luma;
103 init_get_bits(&gb, sl->intra_pcm_ptr,
104 ff_h264_mb_sizes[h->ps.sps->chroma_format_idc] * bit_depth);
106 for (i = 0; i < 16; i++) {
107 uint16_t *tmp_y = (uint16_t *)(dest_y + i * linesize);
108 for (j = 0; j < 16; j++)
109 tmp_y[j] = get_bits(&gb, bit_depth);
111 if (SIMPLE || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
112 if (!h->ps.sps->chroma_format_idc) {
113 for (i = 0; i < block_h; i++) {
114 uint16_t *tmp_cb = (uint16_t *)(dest_cb + i * uvlinesize);
115 for (j = 0; j < 8; j++)
116 tmp_cb[j] = 1 << (bit_depth - 1);
118 for (i = 0; i < block_h; i++) {
119 uint16_t *tmp_cr = (uint16_t *)(dest_cr + i * uvlinesize);
120 for (j = 0; j < 8; j++)
121 tmp_cr[j] = 1 << (bit_depth - 1);
124 for (i = 0; i < block_h; i++) {
125 uint16_t *tmp_cb = (uint16_t *)(dest_cb + i * uvlinesize);
126 for (j = 0; j < 8; j++)
127 tmp_cb[j] = get_bits(&gb, bit_depth);
129 for (i = 0; i < block_h; i++) {
130 uint16_t *tmp_cr = (uint16_t *)(dest_cr + i * uvlinesize);
131 for (j = 0; j < 8; j++)
132 tmp_cr[j] = get_bits(&gb, bit_depth);
137 for (i = 0; i < 16; i++)
138 memcpy(dest_y + i * linesize, sl->intra_pcm_ptr + i * 16, 16);
139 if (SIMPLE || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
140 if (!h->ps.sps->chroma_format_idc) {
141 for (i = 0; i < block_h; i++) {
142 memset(dest_cb + i * uvlinesize, 128, 8);
143 memset(dest_cr + i * uvlinesize, 128, 8);
146 const uint8_t *src_cb = sl->intra_pcm_ptr + 256;
147 const uint8_t *src_cr = sl->intra_pcm_ptr + 256 + block_h * 8;
148 for (i = 0; i < block_h; i++) {
149 memcpy(dest_cb + i * uvlinesize, src_cb + i * 8, 8);
150 memcpy(dest_cr + i * uvlinesize, src_cr + i * 8, 8);
156 if (IS_INTRA(mb_type)) {
157 if (sl->deblocking_filter)
158 xchg_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
159 uvlinesize, 1, 0, SIMPLE, PIXEL_SHIFT);
161 if (SIMPLE || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
162 h->hpc.pred8x8[sl->chroma_pred_mode](dest_cb, uvlinesize);
163 h->hpc.pred8x8[sl->chroma_pred_mode](dest_cr, uvlinesize);
166 hl_decode_mb_predict_luma(h, sl, mb_type, SIMPLE,
167 transform_bypass, PIXEL_SHIFT,
168 block_offset, linesize, dest_y, 0);
170 if (sl->deblocking_filter)
171 xchg_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
172 uvlinesize, 0, 0, SIMPLE, PIXEL_SHIFT);
175 FUNC(hl_motion_422)(h, sl, dest_y, dest_cb, dest_cr,
176 h->h264qpel.put_h264_qpel_pixels_tab,
177 h->h264chroma.put_h264_chroma_pixels_tab,
178 h->h264qpel.avg_h264_qpel_pixels_tab,
179 h->h264chroma.avg_h264_chroma_pixels_tab,
180 h->h264dsp.weight_h264_pixels_tab,
181 h->h264dsp.biweight_h264_pixels_tab);
183 FUNC(hl_motion_420)(h, sl, dest_y, dest_cb, dest_cr,
184 h->h264qpel.put_h264_qpel_pixels_tab,
185 h->h264chroma.put_h264_chroma_pixels_tab,
186 h->h264qpel.avg_h264_qpel_pixels_tab,
187 h->h264chroma.avg_h264_chroma_pixels_tab,
188 h->h264dsp.weight_h264_pixels_tab,
189 h->h264dsp.biweight_h264_pixels_tab);
193 hl_decode_mb_idct_luma(h, sl, mb_type, SIMPLE, transform_bypass,
194 PIXEL_SHIFT, block_offset, linesize, dest_y, 0);
196 if ((SIMPLE || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) &&
198 uint8_t *dest[2] = { dest_cb, dest_cr };
199 if (transform_bypass) {
200 if (IS_INTRA(mb_type) && h->ps.sps->profile_idc == 244 &&
201 (sl->chroma_pred_mode == VERT_PRED8x8 ||
202 sl->chroma_pred_mode == HOR_PRED8x8)) {
203 h->hpc.pred8x8_add[sl->chroma_pred_mode](dest[0],
205 sl->mb + (16 * 16 * 1 << PIXEL_SHIFT),
207 h->hpc.pred8x8_add[sl->chroma_pred_mode](dest[1],
209 sl->mb + (16 * 16 * 2 << PIXEL_SHIFT),
212 idct_add = h->h264dsp.h264_add_pixels4_clear;
213 for (j = 1; j < 3; j++) {
214 for (i = j * 16; i < j * 16 + 4; i++)
215 if (sl->non_zero_count_cache[scan8[i]] ||
216 dctcoef_get(sl->mb, PIXEL_SHIFT, i * 16))
217 idct_add(dest[j - 1] + block_offset[i],
218 sl->mb + (i * 16 << PIXEL_SHIFT),
221 for (i = j * 16 + 4; i < j * 16 + 8; i++)
222 if (sl->non_zero_count_cache[scan8[i + 4]] ||
223 dctcoef_get(sl->mb, PIXEL_SHIFT, i * 16))
224 idct_add(dest[j - 1] + block_offset[i + 4],
225 sl->mb + (i * 16 << PIXEL_SHIFT),
233 qp[0] = sl->chroma_qp[0] + 3;
234 qp[1] = sl->chroma_qp[1] + 3;
236 qp[0] = sl->chroma_qp[0];
237 qp[1] = sl->chroma_qp[1];
239 if (sl->non_zero_count_cache[scan8[CHROMA_DC_BLOCK_INDEX + 0]])
240 h->h264dsp.h264_chroma_dc_dequant_idct(sl->mb + (16 * 16 * 1 << PIXEL_SHIFT),
241 h->ps.pps->dequant4_coeff[IS_INTRA(mb_type) ? 1 : 4][qp[0]][0]);
242 if (sl->non_zero_count_cache[scan8[CHROMA_DC_BLOCK_INDEX + 1]])
243 h->h264dsp.h264_chroma_dc_dequant_idct(sl->mb + (16 * 16 * 2 << PIXEL_SHIFT),
244 h->ps.pps->dequant4_coeff[IS_INTRA(mb_type) ? 2 : 5][qp[1]][0]);
245 h->h264dsp.h264_idct_add8(dest, block_offset,
247 sl->non_zero_count_cache);
253 #if !SIMPLE || BITS == 8
257 #include "h264_mc_template.c"
259 static av_noinline void FUNC(hl_decode_mb_444)(const H264Context *h, H264SliceContext *sl)
261 const int mb_x = sl->mb_x;
262 const int mb_y = sl->mb_y;
263 const int mb_xy = sl->mb_xy;
264 const int mb_type = h->cur_pic.mb_type[mb_xy];
268 const int *block_offset = &h->block_offset[0];
269 const int transform_bypass = !SIMPLE && (sl->qscale == 0 && h->ps.sps->transform_bypass);
270 const int plane_count = (SIMPLE || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) ? 3 : 1;
272 for (p = 0; p < plane_count; p++) {
273 dest[p] = h->cur_pic.f->data[p] +
274 ((mb_x << PIXEL_SHIFT) + mb_y * sl->linesize) * 16;
275 h->vdsp.prefetch(dest[p] + (sl->mb_x & 3) * 4 * sl->linesize + (64 << PIXEL_SHIFT),
279 h->list_counts[mb_xy] = sl->list_count;
281 if (!SIMPLE && MB_FIELD(sl)) {
282 linesize = sl->mb_linesize = sl->mb_uvlinesize = sl->linesize * 2;
283 block_offset = &h->block_offset[48];
284 if (mb_y & 1) // FIXME move out of this function?
285 for (p = 0; p < 3; p++)
286 dest[p] -= sl->linesize * 15;
287 if (FRAME_MBAFF(h)) {
289 for (list = 0; list < sl->list_count; list++) {
290 if (!USES_LIST(mb_type, list))
292 if (IS_16X16(mb_type)) {
293 int8_t *ref = &sl->ref_cache[list][scan8[0]];
294 fill_rectangle(ref, 4, 4, 8, (16 + *ref) ^ (sl->mb_y & 1), 1);
296 for (i = 0; i < 16; i += 4) {
297 int ref = sl->ref_cache[list][scan8[i]];
299 fill_rectangle(&sl->ref_cache[list][scan8[i]], 2, 2,
300 8, (16 + ref) ^ (sl->mb_y & 1), 1);
306 linesize = sl->mb_linesize = sl->mb_uvlinesize = sl->linesize;
309 if (!SIMPLE && IS_INTRA_PCM(mb_type)) {
311 const int bit_depth = h->ps.sps->bit_depth_luma;
313 init_get_bits(&gb, sl->intra_pcm_ptr, 768 * bit_depth);
315 for (p = 0; p < plane_count; p++)
316 for (i = 0; i < 16; i++) {
317 uint16_t *tmp = (uint16_t *)(dest[p] + i * linesize);
318 for (j = 0; j < 16; j++)
319 tmp[j] = get_bits(&gb, bit_depth);
322 for (p = 0; p < plane_count; p++)
323 for (i = 0; i < 16; i++)
324 memcpy(dest[p] + i * linesize,
325 sl->intra_pcm_ptr + p * 256 + i * 16, 16);
328 if (IS_INTRA(mb_type)) {
329 if (sl->deblocking_filter)
330 xchg_mb_border(h, sl, dest[0], dest[1], dest[2], linesize,
331 linesize, 1, 1, SIMPLE, PIXEL_SHIFT);
333 for (p = 0; p < plane_count; p++)
334 hl_decode_mb_predict_luma(h, sl, mb_type, SIMPLE,
335 transform_bypass, PIXEL_SHIFT,
336 block_offset, linesize, dest[p], p);
338 if (sl->deblocking_filter)
339 xchg_mb_border(h, sl, dest[0], dest[1], dest[2], linesize,
340 linesize, 0, 1, SIMPLE, PIXEL_SHIFT);
342 FUNC(hl_motion_444)(h, sl, dest[0], dest[1], dest[2],
343 h->h264qpel.put_h264_qpel_pixels_tab,
344 h->h264chroma.put_h264_chroma_pixels_tab,
345 h->h264qpel.avg_h264_qpel_pixels_tab,
346 h->h264chroma.avg_h264_chroma_pixels_tab,
347 h->h264dsp.weight_h264_pixels_tab,
348 h->h264dsp.biweight_h264_pixels_tab);
351 for (p = 0; p < plane_count; p++)
352 hl_decode_mb_idct_luma(h, sl, mb_type, SIMPLE, transform_bypass,
353 PIXEL_SHIFT, block_offset, linesize,