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 FFmpeg.
7 * FFmpeg 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 * FFmpeg 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 FFmpeg; 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)(H264Context *h)
43 MpegEncContext *const s = &h->s;
44 const int mb_x = s->mb_x;
45 const int mb_y = s->mb_y;
46 const int mb_xy = h->mb_xy;
47 const int mb_type = s->current_picture.f.mb_type[mb_xy];
48 uint8_t *dest_y, *dest_cb, *dest_cr;
49 int linesize, uvlinesize /*dct_offset*/;
51 int *block_offset = &h->block_offset[0];
52 const int transform_bypass = !SIMPLE && (s->qscale == 0 && h->sps.transform_bypass);
53 /* is_h264 should always be true if SVQ3 is disabled. */
54 const int is_h264 = !CONFIG_SVQ3_DECODER || SIMPLE || s->codec_id == AV_CODEC_ID_H264;
55 void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
56 const int block_h = 16 >> s->chroma_y_shift;
57 const int chroma422 = CHROMA422;
59 dest_y = s->current_picture.f.data[0] + ((mb_x << PIXEL_SHIFT) + mb_y * s->linesize) * 16;
60 dest_cb = s->current_picture.f.data[1] + (mb_x << PIXEL_SHIFT) * 8 + mb_y * s->uvlinesize * block_h;
61 dest_cr = s->current_picture.f.data[2] + (mb_x << PIXEL_SHIFT) * 8 + mb_y * s->uvlinesize * block_h;
63 s->vdsp.prefetch(dest_y + (s->mb_x & 3) * 4 * s->linesize + (64 << PIXEL_SHIFT), s->linesize, 4);
64 s->vdsp.prefetch(dest_cb + (s->mb_x & 7) * s->uvlinesize + (64 << PIXEL_SHIFT), dest_cr - dest_cb, 2);
66 h->list_counts[mb_xy] = h->list_count;
68 if (!SIMPLE && MB_FIELD) {
69 linesize = h->mb_linesize = s->linesize * 2;
70 uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
71 block_offset = &h->block_offset[48];
72 if (mb_y & 1) { // FIXME move out of this function?
73 dest_y -= s->linesize * 15;
74 dest_cb -= s->uvlinesize * (block_h - 1);
75 dest_cr -= s->uvlinesize * (block_h - 1);
79 for (list = 0; list < h->list_count; list++) {
80 if (!USES_LIST(mb_type, list))
82 if (IS_16X16(mb_type)) {
83 int8_t *ref = &h->ref_cache[list][scan8[0]];
84 fill_rectangle(ref, 4, 4, 8, (16 + *ref) ^ (s->mb_y & 1), 1);
86 for (i = 0; i < 16; i += 4) {
87 int ref = h->ref_cache[list][scan8[i]];
89 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2,
90 8, (16 + ref) ^ (s->mb_y & 1), 1);
96 linesize = h->mb_linesize = s->linesize;
97 uvlinesize = h->mb_uvlinesize = s->uvlinesize;
98 // dct_offset = s->linesize * 16;
101 if (!SIMPLE && IS_INTRA_PCM(mb_type)) {
102 const int bit_depth = h->sps.bit_depth_luma;
106 init_get_bits(&gb, (uint8_t *)h->mb,
107 ff_h264_mb_sizes[h->sps.chroma_format_idc] * bit_depth);
109 for (i = 0; i < 16; i++) {
110 uint16_t *tmp_y = (uint16_t *)(dest_y + i * linesize);
111 for (j = 0; j < 16; j++)
112 tmp_y[j] = get_bits(&gb, bit_depth);
114 if (SIMPLE || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
115 if (!h->sps.chroma_format_idc) {
116 for (i = 0; i < block_h; i++) {
117 uint16_t *tmp_cb = (uint16_t *)(dest_cb + i * uvlinesize);
118 uint16_t *tmp_cr = (uint16_t *)(dest_cr + i * uvlinesize);
119 for (j = 0; j < 8; j++) {
120 tmp_cb[j] = 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, (uint8_t *)h->mb + i * 16, 16);
139 if (SIMPLE || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
140 if (!h->sps.chroma_format_idc) {
141 for (i = 0; i < 8; i++) {
142 memset(dest_cb + i*uvlinesize, 1 << (bit_depth - 1), 8);
143 memset(dest_cr + i*uvlinesize, 1 << (bit_depth - 1), 8);
146 uint8_t *src_cb = (uint8_t *)h->mb + 256;
147 uint8_t *src_cr = (uint8_t *)h->mb + 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 (h->deblocking_filter)
158 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
159 uvlinesize, 1, 0, SIMPLE, PIXEL_SHIFT);
161 if (SIMPLE || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
162 h->hpc.pred8x8[h->chroma_pred_mode](dest_cb, uvlinesize);
163 h->hpc.pred8x8[h->chroma_pred_mode](dest_cr, uvlinesize);
166 hl_decode_mb_predict_luma(h, mb_type, is_h264, SIMPLE,
167 transform_bypass, PIXEL_SHIFT,
168 block_offset, linesize, dest_y, 0);
170 if (h->deblocking_filter)
171 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
172 uvlinesize, 0, 0, SIMPLE, PIXEL_SHIFT);
173 } else if (is_h264) {
175 FUNC(hl_motion_422)(h, dest_y, dest_cb, dest_cr,
176 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
177 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
178 h->h264dsp.weight_h264_pixels_tab,
179 h->h264dsp.biweight_h264_pixels_tab);
181 FUNC(hl_motion_420)(h, dest_y, dest_cb, dest_cr,
182 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
183 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
184 h->h264dsp.weight_h264_pixels_tab,
185 h->h264dsp.biweight_h264_pixels_tab);
189 hl_decode_mb_idct_luma(h, mb_type, is_h264, SIMPLE, transform_bypass,
190 PIXEL_SHIFT, block_offset, linesize, dest_y, 0);
192 if ((SIMPLE || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) &&
194 uint8_t *dest[2] = { dest_cb, dest_cr };
195 if (transform_bypass) {
196 if (IS_INTRA(mb_type) && h->sps.profile_idc == 244 &&
197 (h->chroma_pred_mode == VERT_PRED8x8 ||
198 h->chroma_pred_mode == HOR_PRED8x8)) {
199 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0],
201 h->mb + (16 * 16 * 1 << PIXEL_SHIFT),
203 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1],
205 h->mb + (16 * 16 * 2 << PIXEL_SHIFT),
208 idct_add = s->dsp.add_pixels4;
209 for (j = 1; j < 3; j++) {
210 for (i = j * 16; i < j * 16 + 4; i++)
211 if (h->non_zero_count_cache[scan8[i]] ||
212 dctcoef_get(h->mb, PIXEL_SHIFT, i * 16))
213 idct_add(dest[j - 1] + block_offset[i],
214 h->mb + (i * 16 << PIXEL_SHIFT),
217 for (i = j * 16 + 4; i < j * 16 + 8; i++)
218 if (h->non_zero_count_cache[scan8[i + 4]] ||
219 dctcoef_get(h->mb, PIXEL_SHIFT, i * 16))
220 idct_add(dest[j - 1] + block_offset[i + 4],
221 h->mb + (i * 16 << PIXEL_SHIFT),
230 qp[0] = h->chroma_qp[0] + 3;
231 qp[1] = h->chroma_qp[1] + 3;
233 qp[0] = h->chroma_qp[0];
234 qp[1] = h->chroma_qp[1];
236 if (h->non_zero_count_cache[scan8[CHROMA_DC_BLOCK_INDEX + 0]])
237 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16 * 16 * 1 << PIXEL_SHIFT),
238 h->dequant4_coeff[IS_INTRA(mb_type) ? 1 : 4][qp[0]][0]);
239 if (h->non_zero_count_cache[scan8[CHROMA_DC_BLOCK_INDEX + 1]])
240 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16 * 16 * 2 << PIXEL_SHIFT),
241 h->dequant4_coeff[IS_INTRA(mb_type) ? 2 : 5][qp[1]][0]);
242 h->h264dsp.h264_idct_add8(dest, block_offset,
244 h->non_zero_count_cache);
245 } else if (CONFIG_SVQ3_DECODER) {
246 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16 * 16 * 1,
247 h->dequant4_coeff[IS_INTRA(mb_type) ? 1 : 4][h->chroma_qp[0]][0]);
248 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16 * 16 * 2,
249 h->dequant4_coeff[IS_INTRA(mb_type) ? 2 : 5][h->chroma_qp[1]][0]);
250 for (j = 1; j < 3; j++) {
251 for (i = j * 16; i < j * 16 + 4; i++)
252 if (h->non_zero_count_cache[scan8[i]] || h->mb[i * 16]) {
253 uint8_t *const ptr = dest[j - 1] + block_offset[i];
254 ff_svq3_add_idct_c(ptr, h->mb + i * 16,
256 ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
263 if (h->cbp || IS_INTRA(mb_type)) {
264 s->dsp.clear_blocks(h->mb);
265 s->dsp.clear_blocks(h->mb + (24 * 16 << PIXEL_SHIFT));
269 #if !SIMPLE || BITS == 8
273 #include "h264_mc_template.c"
275 static av_noinline void FUNC(hl_decode_mb_444)(H264Context *h)
277 MpegEncContext *const s = &h->s;
278 const int mb_x = s->mb_x;
279 const int mb_y = s->mb_y;
280 const int mb_xy = h->mb_xy;
281 const int mb_type = s->current_picture.f.mb_type[mb_xy];
285 int *block_offset = &h->block_offset[0];
286 const int transform_bypass = !SIMPLE && (s->qscale == 0 && h->sps.transform_bypass);
287 const int plane_count = (SIMPLE || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) ? 3 : 1;
289 for (p = 0; p < plane_count; p++) {
290 dest[p] = s->current_picture.f.data[p] +
291 ((mb_x << PIXEL_SHIFT) + mb_y * s->linesize) * 16;
292 s->vdsp.prefetch(dest[p] + (s->mb_x & 3) * 4 * s->linesize + (64 << PIXEL_SHIFT),
296 h->list_counts[mb_xy] = h->list_count;
298 if (!SIMPLE && MB_FIELD) {
299 linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize * 2;
300 block_offset = &h->block_offset[48];
301 if (mb_y & 1) // FIXME move out of this function?
302 for (p = 0; p < 3; p++)
303 dest[p] -= s->linesize * 15;
306 for (list = 0; list < h->list_count; list++) {
307 if (!USES_LIST(mb_type, list))
309 if (IS_16X16(mb_type)) {
310 int8_t *ref = &h->ref_cache[list][scan8[0]];
311 fill_rectangle(ref, 4, 4, 8, (16 + *ref) ^ (s->mb_y & 1), 1);
313 for (i = 0; i < 16; i += 4) {
314 int ref = h->ref_cache[list][scan8[i]];
316 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2,
317 8, (16 + ref) ^ (s->mb_y & 1), 1);
323 linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize;
326 if (!SIMPLE && IS_INTRA_PCM(mb_type)) {
328 const int bit_depth = h->sps.bit_depth_luma;
330 init_get_bits(&gb, (uint8_t *)h->mb, 768 * bit_depth);
332 for (p = 0; p < plane_count; p++)
333 for (i = 0; i < 16; i++) {
334 uint16_t *tmp = (uint16_t *)(dest[p] + i * linesize);
335 for (j = 0; j < 16; j++)
336 tmp[j] = get_bits(&gb, bit_depth);
339 for (p = 0; p < plane_count; p++)
340 for (i = 0; i < 16; i++)
341 memcpy(dest[p] + i * linesize,
342 (uint8_t *)h->mb + p * 256 + i * 16, 16);
345 if (IS_INTRA(mb_type)) {
346 if (h->deblocking_filter)
347 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize,
348 linesize, 1, 1, SIMPLE, PIXEL_SHIFT);
350 for (p = 0; p < plane_count; p++)
351 hl_decode_mb_predict_luma(h, mb_type, 1, SIMPLE,
352 transform_bypass, PIXEL_SHIFT,
353 block_offset, linesize, dest[p], p);
355 if (h->deblocking_filter)
356 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize,
357 linesize, 0, 1, SIMPLE, PIXEL_SHIFT);
359 FUNC(hl_motion_444)(h, dest[0], dest[1], dest[2],
360 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
361 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
362 h->h264dsp.weight_h264_pixels_tab,
363 h->h264dsp.biweight_h264_pixels_tab);
366 for (p = 0; p < plane_count; p++)
367 hl_decode_mb_idct_luma(h, mb_type, 1, SIMPLE, transform_bypass,
368 PIXEL_SHIFT, block_offset, linesize,
371 if (h->cbp || IS_INTRA(mb_type)) {
372 s->dsp.clear_blocks(h->mb);
373 s->dsp.clear_blocks(h->mb + (24 * 16 << PIXEL_SHIFT));