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)(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)) {
103 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 for (j = 0; j < 8; j++)
119 tmp_cb[j] = 1 << (bit_depth - 1);
121 for (i = 0; i < block_h; i++) {
122 uint16_t *tmp_cr = (uint16_t *)(dest_cr + i * uvlinesize);
123 for (j = 0; j < 8; j++)
124 tmp_cr[j] = 1 << (bit_depth - 1);
127 for (i = 0; i < block_h; i++) {
128 uint16_t *tmp_cb = (uint16_t *)(dest_cb + i * uvlinesize);
129 for (j = 0; j < 8; j++)
130 tmp_cb[j] = get_bits(&gb, bit_depth);
132 for (i = 0; i < block_h; i++) {
133 uint16_t *tmp_cr = (uint16_t *)(dest_cr + i * uvlinesize);
134 for (j = 0; j < 8; j++)
135 tmp_cr[j] = get_bits(&gb, bit_depth);
140 for (i = 0; i < 16; i++)
141 memcpy(dest_y + i * linesize, (uint8_t *)h->mb + i * 16, 16);
142 if (SIMPLE || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
143 if (!h->sps.chroma_format_idc) {
144 for (i = 0; i < block_h; i++) {
145 memset(dest_cb + i * uvlinesize, 128, 8);
146 memset(dest_cr + i * uvlinesize, 128, 8);
149 uint8_t *src_cb = (uint8_t *)h->mb + 256;
150 uint8_t *src_cr = (uint8_t *)h->mb + 256 + block_h * 8;
151 for (i = 0; i < block_h; i++) {
152 memcpy(dest_cb + i * uvlinesize, src_cb + i * 8, 8);
153 memcpy(dest_cr + i * uvlinesize, src_cr + i * 8, 8);
159 if (IS_INTRA(mb_type)) {
160 if (h->deblocking_filter)
161 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
162 uvlinesize, 1, 0, SIMPLE, PIXEL_SHIFT);
164 if (SIMPLE || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
165 h->hpc.pred8x8[h->chroma_pred_mode](dest_cb, uvlinesize);
166 h->hpc.pred8x8[h->chroma_pred_mode](dest_cr, uvlinesize);
169 hl_decode_mb_predict_luma(h, mb_type, is_h264, SIMPLE,
170 transform_bypass, PIXEL_SHIFT,
171 block_offset, linesize, dest_y, 0);
173 if (h->deblocking_filter)
174 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
175 uvlinesize, 0, 0, SIMPLE, PIXEL_SHIFT);
176 } else if (is_h264) {
178 FUNC(hl_motion_422)(h, dest_y, dest_cb, dest_cr,
179 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
180 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
181 h->h264dsp.weight_h264_pixels_tab,
182 h->h264dsp.biweight_h264_pixels_tab);
184 FUNC(hl_motion_420)(h, dest_y, dest_cb, dest_cr,
185 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
186 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
187 h->h264dsp.weight_h264_pixels_tab,
188 h->h264dsp.biweight_h264_pixels_tab);
192 hl_decode_mb_idct_luma(h, mb_type, is_h264, SIMPLE, transform_bypass,
193 PIXEL_SHIFT, block_offset, linesize, dest_y, 0);
195 if ((SIMPLE || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) &&
197 uint8_t *dest[2] = { dest_cb, dest_cr };
198 if (transform_bypass) {
199 if (IS_INTRA(mb_type) && h->sps.profile_idc == 244 &&
200 (h->chroma_pred_mode == VERT_PRED8x8 ||
201 h->chroma_pred_mode == HOR_PRED8x8)) {
202 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0],
204 h->mb + (16 * 16 * 1 << PIXEL_SHIFT),
206 h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1],
208 h->mb + (16 * 16 * 2 << PIXEL_SHIFT),
211 idct_add = s->dsp.add_pixels4;
212 for (j = 1; j < 3; j++) {
213 for (i = j * 16; i < j * 16 + 4; i++)
214 if (h->non_zero_count_cache[scan8[i]] ||
215 dctcoef_get(h->mb, PIXEL_SHIFT, i * 16))
216 idct_add(dest[j - 1] + block_offset[i],
217 h->mb + (i * 16 << PIXEL_SHIFT),
220 for (i = j * 16 + 4; i < j * 16 + 8; i++)
221 if (h->non_zero_count_cache[scan8[i + 4]] ||
222 dctcoef_get(h->mb, PIXEL_SHIFT, i * 16))
223 idct_add(dest[j - 1] + block_offset[i + 4],
224 h->mb + (i * 16 << PIXEL_SHIFT),
233 qp[0] = h->chroma_qp[0] + 3;
234 qp[1] = h->chroma_qp[1] + 3;
236 qp[0] = h->chroma_qp[0];
237 qp[1] = h->chroma_qp[1];
239 if (h->non_zero_count_cache[scan8[CHROMA_DC_BLOCK_INDEX + 0]])
240 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16 * 16 * 1 << PIXEL_SHIFT),
241 h->dequant4_coeff[IS_INTRA(mb_type) ? 1 : 4][qp[0]][0]);
242 if (h->non_zero_count_cache[scan8[CHROMA_DC_BLOCK_INDEX + 1]])
243 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16 * 16 * 2 << PIXEL_SHIFT),
244 h->dequant4_coeff[IS_INTRA(mb_type) ? 2 : 5][qp[1]][0]);
245 h->h264dsp.h264_idct_add8(dest, block_offset,
247 h->non_zero_count_cache);
248 } else if (CONFIG_SVQ3_DECODER) {
249 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16 * 16 * 1,
250 h->dequant4_coeff[IS_INTRA(mb_type) ? 1 : 4][h->chroma_qp[0]][0]);
251 h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16 * 16 * 2,
252 h->dequant4_coeff[IS_INTRA(mb_type) ? 2 : 5][h->chroma_qp[1]][0]);
253 for (j = 1; j < 3; j++) {
254 for (i = j * 16; i < j * 16 + 4; i++)
255 if (h->non_zero_count_cache[scan8[i]] || h->mb[i * 16]) {
256 uint8_t *const ptr = dest[j - 1] + block_offset[i];
257 ff_svq3_add_idct_c(ptr, h->mb + i * 16,
259 ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
266 if (h->cbp || IS_INTRA(mb_type)) {
267 s->dsp.clear_blocks(h->mb);
268 s->dsp.clear_blocks(h->mb + (24 * 16 << PIXEL_SHIFT));
272 #if !SIMPLE || BITS == 8
276 #include "h264_mc_template.c"
278 static av_noinline void FUNC(hl_decode_mb_444)(H264Context *h)
280 MpegEncContext *const s = &h->s;
281 const int mb_x = s->mb_x;
282 const int mb_y = s->mb_y;
283 const int mb_xy = h->mb_xy;
284 const int mb_type = s->current_picture.f.mb_type[mb_xy];
288 int *block_offset = &h->block_offset[0];
289 const int transform_bypass = !SIMPLE && (s->qscale == 0 && h->sps.transform_bypass);
290 const int plane_count = (SIMPLE || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) ? 3 : 1;
292 for (p = 0; p < plane_count; p++) {
293 dest[p] = s->current_picture.f.data[p] +
294 ((mb_x << PIXEL_SHIFT) + mb_y * s->linesize) * 16;
295 s->vdsp.prefetch(dest[p] + (s->mb_x & 3) * 4 * s->linesize + (64 << PIXEL_SHIFT),
299 h->list_counts[mb_xy] = h->list_count;
301 if (!SIMPLE && MB_FIELD) {
302 linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize * 2;
303 block_offset = &h->block_offset[48];
304 if (mb_y & 1) // FIXME move out of this function?
305 for (p = 0; p < 3; p++)
306 dest[p] -= s->linesize * 15;
309 for (list = 0; list < h->list_count; list++) {
310 if (!USES_LIST(mb_type, list))
312 if (IS_16X16(mb_type)) {
313 int8_t *ref = &h->ref_cache[list][scan8[0]];
314 fill_rectangle(ref, 4, 4, 8, (16 + *ref) ^ (s->mb_y & 1), 1);
316 for (i = 0; i < 16; i += 4) {
317 int ref = h->ref_cache[list][scan8[i]];
319 fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2,
320 8, (16 + ref) ^ (s->mb_y & 1), 1);
326 linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize;
329 if (!SIMPLE && IS_INTRA_PCM(mb_type)) {
331 const int bit_depth = h->sps.bit_depth_luma;
333 init_get_bits(&gb, (uint8_t *)h->mb, 768 * bit_depth);
335 for (p = 0; p < plane_count; p++)
336 for (i = 0; i < 16; i++) {
337 uint16_t *tmp = (uint16_t *)(dest[p] + i * linesize);
338 for (j = 0; j < 16; j++)
339 tmp[j] = get_bits(&gb, bit_depth);
342 for (p = 0; p < plane_count; p++)
343 for (i = 0; i < 16; i++)
344 memcpy(dest[p] + i * linesize,
345 (uint8_t *)h->mb + p * 256 + i * 16, 16);
348 if (IS_INTRA(mb_type)) {
349 if (h->deblocking_filter)
350 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize,
351 linesize, 1, 1, SIMPLE, PIXEL_SHIFT);
353 for (p = 0; p < plane_count; p++)
354 hl_decode_mb_predict_luma(h, mb_type, 1, SIMPLE,
355 transform_bypass, PIXEL_SHIFT,
356 block_offset, linesize, dest[p], p);
358 if (h->deblocking_filter)
359 xchg_mb_border(h, dest[0], dest[1], dest[2], linesize,
360 linesize, 0, 1, SIMPLE, PIXEL_SHIFT);
362 FUNC(hl_motion_444)(h, dest[0], dest[1], dest[2],
363 s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
364 s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
365 h->h264dsp.weight_h264_pixels_tab,
366 h->h264dsp.biweight_h264_pixels_tab);
369 for (p = 0; p < plane_count; p++)
370 hl_decode_mb_idct_luma(h, mb_type, 1, SIMPLE, transform_bypass,
371 PIXEL_SHIFT, block_offset, linesize,
374 if (h->cbp || IS_INTRA(mb_type)) {
375 s->dsp.clear_blocks(h->mb);
376 s->dsp.clear_blocks(h->mb + (24 * 16 << PIXEL_SHIFT));