vaapi_encode_mpeg2: Fix width/height columns/rows confusion
[ffmpeg.git] / libavcodec / nvdec_hevc.c
1 /*
2  * HEVC HW decode acceleration through NVDEC
3  *
4  * Copyright (c) 2017 Anton Khirnov
5  *
6  * This file is part of FFmpeg.
7  *
8  * FFmpeg is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU Lesser General Public
10  * License as published by the Free Software Foundation; either
11  * version 2.1 of the License, or (at your option) any later version.
12  *
13  * FFmpeg is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16  * Lesser General Public License for more details.
17  *
18  * You should have received a copy of the GNU Lesser General Public
19  * License along with FFmpeg; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21  */
22
23 #include <stdint.h>
24 #include <string.h>
25
26 #include "avcodec.h"
27 #include "nvdec.h"
28 #include "decode.h"
29 #include "internal.h"
30 #include "hevcdec.h"
31 #include "hevc_data.h"
32
33 static void dpb_add(CUVIDHEVCPICPARAMS *pp, int idx, const HEVCFrame *src)
34 {
35     FrameDecodeData *fdd = (FrameDecodeData*)src->frame->private_ref->data;
36     const NVDECFrame *cf = fdd->hwaccel_priv;
37
38     pp->RefPicIdx[idx]      = cf ? cf->idx : -1;
39     pp->PicOrderCntVal[idx] = src->poc;
40     pp->IsLongTerm[idx]     = !!(src->flags & HEVC_FRAME_FLAG_LONG_REF);
41 }
42
43 static void fill_scaling_lists(CUVIDHEVCPICPARAMS *ppc, const HEVCContext *s)
44 {
45     const ScalingList *sl = s->ps.pps->scaling_list_data_present_flag ?
46                             &s->ps.pps->scaling_list : &s->ps.sps->scaling_list;
47     int i, j, pos;
48
49     for (i = 0; i < 6; i++) {
50         for (j = 0; j < 16; j++) {
51             pos = 4 * ff_hevc_diag_scan4x4_y[j] + ff_hevc_diag_scan4x4_x[j];
52             ppc->ScalingList4x4[i][j] = sl->sl[0][i][pos];
53         }
54
55         for (j = 0; j < 64; j++) {
56             pos = 8 * ff_hevc_diag_scan8x8_y[j] + ff_hevc_diag_scan8x8_x[j];
57             ppc->ScalingList8x8[i][j]   = sl->sl[1][i][pos];
58             ppc->ScalingList16x16[i][j] = sl->sl[2][i][pos];
59
60             if (i < 2)
61                 ppc->ScalingList32x32[i][j] = sl->sl[3][i * 3][pos];
62         }
63
64         ppc->ScalingListDCCoeff16x16[i] = sl->sl_dc[0][i];
65         if (i < 2)
66             ppc->ScalingListDCCoeff32x32[i] = sl->sl_dc[1][i * 3];
67     }
68 }
69
70 static int nvdec_hevc_start_frame(AVCodecContext *avctx,
71                                   const uint8_t *buffer, uint32_t size)
72 {
73     const HEVCContext *s = avctx->priv_data;
74     const HEVCPPS *pps = s->ps.pps;
75     const HEVCSPS *sps = s->ps.sps;
76
77     NVDECContext       *ctx = avctx->internal->hwaccel_priv_data;
78     CUVIDPICPARAMS      *pp = &ctx->pic_params;
79     CUVIDHEVCPICPARAMS *ppc = &pp->CodecSpecific.hevc;
80     FrameDecodeData *fdd;
81     NVDECFrame *cf;
82
83     int i, j, dpb_size, ret;
84
85     ret = ff_nvdec_start_frame(avctx, s->ref->frame);
86     if (ret < 0)
87         return ret;
88
89     fdd = (FrameDecodeData*)s->ref->frame->private_ref->data;
90     cf  = (NVDECFrame*)fdd->hwaccel_priv;
91
92     *pp = (CUVIDPICPARAMS) {
93         .PicWidthInMbs     = sps->width  / 16,
94         .FrameHeightInMbs  = sps->height / 16,
95         .CurrPicIdx        = cf->idx,
96         .ref_pic_flag      = 1,
97         .intra_pic_flag    = IS_IRAP(s),
98
99         .CodecSpecific.hevc = {
100             .pic_width_in_luma_samples                    = sps->width,
101             .pic_height_in_luma_samples                   = sps->height,
102             .log2_min_luma_coding_block_size_minus3       = sps->log2_min_cb_size - 3,
103             .log2_diff_max_min_luma_coding_block_size     = sps->log2_diff_max_min_coding_block_size,
104             .log2_min_transform_block_size_minus2         = sps->log2_min_tb_size - 2,
105             .log2_diff_max_min_transform_block_size       = sps->log2_max_trafo_size - sps->log2_min_tb_size,
106             .pcm_enabled_flag                             = sps->pcm_enabled_flag,
107             .log2_min_pcm_luma_coding_block_size_minus3   = sps->pcm_enabled_flag ? sps->pcm.log2_min_pcm_cb_size - 3 : 0,
108             .log2_diff_max_min_pcm_luma_coding_block_size = sps->pcm.log2_max_pcm_cb_size - sps->pcm.log2_min_pcm_cb_size,
109             .pcm_sample_bit_depth_luma_minus1             = sps->pcm_enabled_flag ? sps->pcm.bit_depth - 1 : 0,
110             .pcm_sample_bit_depth_chroma_minus1           = sps->pcm_enabled_flag ? sps->pcm.bit_depth_chroma - 1 : 0,
111 #if NVDECAPI_CHECK_VERSION(8, 1)
112             .log2_max_transform_skip_block_size_minus2    = pps->log2_max_transform_skip_block_size - 2,
113             .log2_sao_offset_scale_luma                   = pps->log2_sao_offset_scale_luma,
114             .log2_sao_offset_scale_chroma                 = pps->log2_sao_offset_scale_chroma,
115             .high_precision_offsets_enabled_flag          = sps->high_precision_offsets_enabled_flag,
116 #endif
117             .pcm_loop_filter_disabled_flag                = sps->pcm.loop_filter_disable_flag,
118             .strong_intra_smoothing_enabled_flag          = sps->sps_strong_intra_smoothing_enable_flag,
119             .max_transform_hierarchy_depth_intra          = sps->max_transform_hierarchy_depth_intra,
120             .max_transform_hierarchy_depth_inter          = sps->max_transform_hierarchy_depth_inter,
121             .amp_enabled_flag                             = sps->amp_enabled_flag,
122             .separate_colour_plane_flag                   = sps->separate_colour_plane_flag,
123             .log2_max_pic_order_cnt_lsb_minus4            = sps->log2_max_poc_lsb - 4,
124             .num_short_term_ref_pic_sets                  = sps->nb_st_rps,
125             .long_term_ref_pics_present_flag              = sps->long_term_ref_pics_present_flag,
126             .num_long_term_ref_pics_sps                   = sps->num_long_term_ref_pics_sps,
127             .sps_temporal_mvp_enabled_flag                = sps->sps_temporal_mvp_enabled_flag,
128             .sample_adaptive_offset_enabled_flag          = sps->sao_enabled,
129             .scaling_list_enable_flag                     = sps->scaling_list_enable_flag,
130             .IrapPicFlag                                  = IS_IRAP(s),
131             .IdrPicFlag                                   = IS_IDR(s),
132             .bit_depth_luma_minus8                        = sps->bit_depth - 8,
133             .bit_depth_chroma_minus8                      = sps->bit_depth - 8,
134
135             .dependent_slice_segments_enabled_flag        = pps->dependent_slice_segments_enabled_flag,
136             .slice_segment_header_extension_present_flag  = pps->slice_header_extension_present_flag,
137             .sign_data_hiding_enabled_flag                = pps->sign_data_hiding_flag,
138             .cu_qp_delta_enabled_flag                     = pps->cu_qp_delta_enabled_flag,
139             .diff_cu_qp_delta_depth                       = pps->diff_cu_qp_delta_depth,
140             .init_qp_minus26                              = pps->pic_init_qp_minus26,
141             .pps_cb_qp_offset                             = pps->cb_qp_offset,
142             .pps_cr_qp_offset                             = pps->cr_qp_offset,
143             .constrained_intra_pred_flag                  = pps->constrained_intra_pred_flag,
144             .weighted_pred_flag                           = pps->weighted_pred_flag,
145             .weighted_bipred_flag                         = pps->weighted_bipred_flag,
146             .transform_skip_enabled_flag                  = pps->transform_skip_enabled_flag,
147             .transquant_bypass_enabled_flag               = pps->transquant_bypass_enable_flag,
148             .entropy_coding_sync_enabled_flag             = pps->entropy_coding_sync_enabled_flag,
149             .log2_parallel_merge_level_minus2             = pps->log2_parallel_merge_level - 2,
150             .num_extra_slice_header_bits                  = pps->num_extra_slice_header_bits,
151             .loop_filter_across_tiles_enabled_flag        = pps->loop_filter_across_tiles_enabled_flag,
152             .loop_filter_across_slices_enabled_flag       = pps->seq_loop_filter_across_slices_enabled_flag,
153             .output_flag_present_flag                     = pps->output_flag_present_flag,
154             .num_ref_idx_l0_default_active_minus1         = pps->num_ref_idx_l0_default_active - 1,
155             .num_ref_idx_l1_default_active_minus1         = pps->num_ref_idx_l1_default_active - 1,
156             .lists_modification_present_flag              = pps->lists_modification_present_flag,
157             .cabac_init_present_flag                      = pps->cabac_init_present_flag,
158             .pps_slice_chroma_qp_offsets_present_flag     = pps->pic_slice_level_chroma_qp_offsets_present_flag,
159             .deblocking_filter_override_enabled_flag      = pps->deblocking_filter_override_enabled_flag,
160             .pps_deblocking_filter_disabled_flag          = pps->disable_dbf,
161             .pps_beta_offset_div2                         = pps->beta_offset / 2,
162             .pps_tc_offset_div2                           = pps->tc_offset / 2,
163             .tiles_enabled_flag                           = pps->tiles_enabled_flag,
164             .uniform_spacing_flag                         = pps->uniform_spacing_flag,
165             .num_tile_columns_minus1                      = pps->num_tile_columns - 1,
166             .num_tile_rows_minus1                         = pps->num_tile_rows - 1,
167
168             .NumBitsForShortTermRPSInSlice                = s->sh.short_term_rps ? s->sh.short_term_ref_pic_set_size : 0,
169             .NumDeltaPocsOfRefRpsIdx                      = s->sh.short_term_rps ? s->sh.short_term_rps->rps_idx_num_delta_pocs : 0,
170             .NumPocTotalCurr                              = ff_hevc_frame_nb_refs(s),
171             .NumPocStCurrBefore                           = s->rps[ST_CURR_BEF].nb_refs,
172             .NumPocStCurrAfter                            = s->rps[ST_CURR_AFT].nb_refs,
173             .NumPocLtCurr                                 = s->rps[LT_CURR].nb_refs,
174             .CurrPicOrderCntVal                           = s->ref->poc,
175         },
176     };
177
178     if (pps->num_tile_columns > FF_ARRAY_ELEMS(ppc->column_width_minus1) ||
179         pps->num_tile_rows    > FF_ARRAY_ELEMS(ppc->row_height_minus1)) {
180         av_log(avctx, AV_LOG_ERROR, "Too many tiles\n");
181         return AVERROR(ENOSYS);
182     }
183     for (i = 0; i < pps->num_tile_columns; i++)
184         ppc->column_width_minus1[i] = pps->column_width[i] - 1;
185     for (i = 0; i < pps->num_tile_rows; i++)
186         ppc->row_height_minus1[i] = pps->row_height[i] - 1;
187
188     if (s->rps[LT_CURR].nb_refs     > FF_ARRAY_ELEMS(ppc->RefPicSetLtCurr)       ||
189         s->rps[ST_CURR_BEF].nb_refs > FF_ARRAY_ELEMS(ppc->RefPicSetStCurrBefore) ||
190         s->rps[ST_CURR_AFT].nb_refs > FF_ARRAY_ELEMS(ppc->RefPicSetStCurrAfter)) {
191         av_log(avctx, AV_LOG_ERROR, "Too many reference frames\n");
192         return AVERROR(ENOSYS);
193     }
194
195     dpb_size = 0;
196     for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
197         const HEVCFrame *ref = &s->DPB[i];
198         if (!(ref->flags & (HEVC_FRAME_FLAG_SHORT_REF | HEVC_FRAME_FLAG_LONG_REF)))
199             continue;
200         if (dpb_size >= FF_ARRAY_ELEMS(ppc->RefPicIdx)) {
201             av_log(avctx, AV_LOG_ERROR, "Too many reference frames\n");
202             return AVERROR_INVALIDDATA;
203         }
204         dpb_add(ppc, dpb_size++, ref);
205
206     }
207     for (i = dpb_size; i < FF_ARRAY_ELEMS(ppc->RefPicIdx); i++)
208         ppc->RefPicIdx[i] = -1;
209
210     for (i = 0; i < s->rps[ST_CURR_BEF].nb_refs; i++) {
211         for (j = 0; j < dpb_size; j++) {
212             if (ppc->PicOrderCntVal[j] == s->rps[ST_CURR_BEF].list[i]) {
213                 ppc->RefPicSetStCurrBefore[i] = j;
214                 break;
215             }
216         }
217     }
218     for (i = 0; i < s->rps[ST_CURR_AFT].nb_refs; i++) {
219         for (j = 0; j < dpb_size; j++) {
220             if (ppc->PicOrderCntVal[j] == s->rps[ST_CURR_AFT].list[i]) {
221                 ppc->RefPicSetStCurrAfter[i] = j;
222                 break;
223             }
224         }
225     }
226     for (i = 0; i < s->rps[LT_CURR].nb_refs; i++) {
227         for (j = 0; j < dpb_size; j++) {
228             if (ppc->PicOrderCntVal[j] == s->rps[LT_CURR].list[i]) {
229                 ppc->RefPicSetLtCurr[i] = j;
230                 break;
231             }
232         }
233     }
234
235     fill_scaling_lists(ppc, s);
236
237     return 0;
238 }
239
240 static int nvdec_hevc_decode_slice(AVCodecContext *avctx, const uint8_t *buffer,
241                                    uint32_t size)
242 {
243     NVDECContext *ctx = avctx->internal->hwaccel_priv_data;
244     void *tmp;
245
246     tmp = av_fast_realloc(ctx->bitstream, &ctx->bitstream_allocated,
247                           ctx->bitstream_len + size + 3);
248     if (!tmp)
249         return AVERROR(ENOMEM);
250     ctx->bitstream = tmp;
251
252     tmp = av_fast_realloc(ctx->slice_offsets, &ctx->slice_offsets_allocated,
253                           (ctx->nb_slices + 1) * sizeof(*ctx->slice_offsets));
254     if (!tmp)
255         return AVERROR(ENOMEM);
256     ctx->slice_offsets = tmp;
257
258     AV_WB24(ctx->bitstream + ctx->bitstream_len, 1);
259     memcpy(ctx->bitstream + ctx->bitstream_len + 3, buffer, size);
260     ctx->slice_offsets[ctx->nb_slices] = ctx->bitstream_len ;
261     ctx->bitstream_len += size + 3;
262     ctx->nb_slices++;
263
264     return 0;
265 }
266
267 static int nvdec_hevc_frame_params(AVCodecContext *avctx,
268                                    AVBufferRef *hw_frames_ctx)
269 {
270     const HEVCContext *s = avctx->priv_data;
271     const HEVCSPS *sps = s->ps.sps;
272     return ff_nvdec_frame_params(avctx, hw_frames_ctx, sps->temporal_layer[sps->max_sub_layers - 1].max_dec_pic_buffering + 1);
273 }
274
275 const AVHWAccel ff_hevc_nvdec_hwaccel = {
276     .name                 = "hevc_nvdec",
277     .type                 = AVMEDIA_TYPE_VIDEO,
278     .id                   = AV_CODEC_ID_HEVC,
279     .pix_fmt              = AV_PIX_FMT_CUDA,
280     .start_frame          = nvdec_hevc_start_frame,
281     .end_frame            = ff_nvdec_end_frame,
282     .decode_slice         = nvdec_hevc_decode_slice,
283     .frame_params         = nvdec_hevc_frame_params,
284     .init                 = ff_nvdec_decode_init,
285     .uninit               = ff_nvdec_decode_uninit,
286     .priv_data_size       = sizeof(NVDECContext),
287 };