avcodec/aacdec_fixed: use 64bit to avoid overflow in rounding in apply_dependent_coup...
[ffmpeg.git] / libavcodec / vdpau_h264.c
1 /*
2  * MPEG-4 Part 10 / AVC / H.264 HW decode acceleration through VDPAU
3  *
4  * Copyright (c) 2008 NVIDIA
5  * Copyright (c) 2013 RĂ©mi Denis-Courmont
6  *
7  * This file is part of FFmpeg.
8  *
9  * FFmpeg is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU Lesser General Public
11  * License as published by the Free Software Foundation; either
12  * version 2.1 of the License, or (at your option) any later version.
13  *
14  * FFmpeg is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17  * Lesser General Public License for more details.
18  *
19  * You should have received a copy of the GNU Lesser General Public
20  * License along with FFmpeg; if not, write to the Free Software Foundation,
21  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22  */
23
24 #include <vdpau/vdpau.h>
25
26 #include "avcodec.h"
27 #include "internal.h"
28 #include "h264dec.h"
29 #include "h264_ps.h"
30 #include "hwaccel.h"
31 #include "mpegutils.h"
32 #include "vdpau.h"
33 #include "vdpau_internal.h"
34
35 static int32_t h264_foc(int foc)
36 {
37     if (foc == INT_MAX)
38         foc = 0;
39     return foc;
40 }
41
42 static void vdpau_h264_clear_rf(VdpReferenceFrameH264 *rf)
43 {
44     rf->surface             = VDP_INVALID_HANDLE;
45     rf->is_long_term        = VDP_FALSE;
46     rf->top_is_reference    = VDP_FALSE;
47     rf->bottom_is_reference = VDP_FALSE;
48     rf->field_order_cnt[0]  = 0;
49     rf->field_order_cnt[1]  = 0;
50     rf->frame_idx           = 0;
51 }
52
53 static void vdpau_h264_set_rf(VdpReferenceFrameH264 *rf, H264Picture *pic,
54                               int pic_structure)
55 {
56     VdpVideoSurface surface = ff_vdpau_get_surface_id(pic->f);
57
58     if (pic_structure == 0)
59         pic_structure = pic->reference;
60
61     rf->surface             = surface;
62     rf->is_long_term        = pic->reference && pic->long_ref;
63     rf->top_is_reference    = (pic_structure & PICT_TOP_FIELD)    != 0;
64     rf->bottom_is_reference = (pic_structure & PICT_BOTTOM_FIELD) != 0;
65     rf->field_order_cnt[0]  = h264_foc(pic->field_poc[0]);
66     rf->field_order_cnt[1]  = h264_foc(pic->field_poc[1]);
67     rf->frame_idx           = pic->long_ref ? pic->pic_id : pic->frame_num;
68 }
69
70 static void vdpau_h264_set_reference_frames(AVCodecContext *avctx)
71 {
72     H264Context * const h = avctx->priv_data;
73     struct vdpau_picture_context *pic_ctx = h->cur_pic_ptr->hwaccel_picture_private;
74     VdpPictureInfoH264 *info = &pic_ctx->info.h264;
75     int list;
76
77     VdpReferenceFrameH264 *rf = &info->referenceFrames[0];
78 #define H264_RF_COUNT FF_ARRAY_ELEMS(info->referenceFrames)
79
80     for (list = 0; list < 2; ++list) {
81         H264Picture **lp = list ? h->long_ref : h->short_ref;
82         int i, ls    = list ? 16          : h->short_ref_count;
83
84         for (i = 0; i < ls; ++i) {
85             H264Picture *pic = lp[i];
86             VdpReferenceFrameH264 *rf2;
87             VdpVideoSurface surface_ref;
88             int pic_frame_idx;
89
90             if (!pic || !pic->reference)
91                 continue;
92             pic_frame_idx = pic->long_ref ? pic->pic_id : pic->frame_num;
93             surface_ref = ff_vdpau_get_surface_id(pic->f);
94
95             rf2 = &info->referenceFrames[0];
96             while (rf2 != rf) {
97                 if ((rf2->surface      == surface_ref)   &&
98                     (rf2->is_long_term == pic->long_ref) &&
99                     (rf2->frame_idx    == pic_frame_idx))
100                     break;
101                 ++rf2;
102             }
103             if (rf2 != rf) {
104                 rf2->top_is_reference    |= (pic->reference & PICT_TOP_FIELD)    ? VDP_TRUE : VDP_FALSE;
105                 rf2->bottom_is_reference |= (pic->reference & PICT_BOTTOM_FIELD) ? VDP_TRUE : VDP_FALSE;
106                 continue;
107             }
108
109             if (rf >= &info->referenceFrames[H264_RF_COUNT])
110                 continue;
111
112             vdpau_h264_set_rf(rf, pic, pic->reference);
113             ++rf;
114         }
115     }
116
117     for (; rf < &info->referenceFrames[H264_RF_COUNT]; ++rf)
118         vdpau_h264_clear_rf(rf);
119 }
120
121 static int vdpau_h264_start_frame(AVCodecContext *avctx,
122                                   const uint8_t *buffer, uint32_t size)
123 {
124     H264Context * const h = avctx->priv_data;
125     const PPS *pps = h->ps.pps;
126     const SPS *sps = h->ps.sps;
127     H264Picture *pic = h->cur_pic_ptr;
128     struct vdpau_picture_context *pic_ctx = pic->hwaccel_picture_private;
129     VdpPictureInfoH264 *info = &pic_ctx->info.h264;
130 #ifdef VDP_DECODER_PROFILE_H264_HIGH_444_PREDICTIVE
131     VdpPictureInfoH264Predictive *info2 = &pic_ctx->info.h264_predictive;
132 #endif
133
134     /* init VdpPictureInfoH264 */
135     info->slice_count                            = 0;
136     info->field_order_cnt[0]                     = h264_foc(pic->field_poc[0]);
137     info->field_order_cnt[1]                     = h264_foc(pic->field_poc[1]);
138     info->is_reference                           = h->nal_ref_idc != 0;
139     info->frame_num                              = h->poc.frame_num;
140     info->field_pic_flag                         = h->picture_structure != PICT_FRAME;
141     info->bottom_field_flag                      = h->picture_structure == PICT_BOTTOM_FIELD;
142     info->num_ref_frames                         = sps->ref_frame_count;
143     info->mb_adaptive_frame_field_flag           = sps->mb_aff && !info->field_pic_flag;
144     info->constrained_intra_pred_flag            = pps->constrained_intra_pred;
145     info->weighted_pred_flag                     = pps->weighted_pred;
146     info->weighted_bipred_idc                    = pps->weighted_bipred_idc;
147     info->frame_mbs_only_flag                    = sps->frame_mbs_only_flag;
148     info->transform_8x8_mode_flag                = pps->transform_8x8_mode;
149     info->chroma_qp_index_offset                 = pps->chroma_qp_index_offset[0];
150     info->second_chroma_qp_index_offset          = pps->chroma_qp_index_offset[1];
151     info->pic_init_qp_minus26                    = pps->init_qp - 26;
152     info->num_ref_idx_l0_active_minus1           = pps->ref_count[0] - 1;
153     info->num_ref_idx_l1_active_minus1           = pps->ref_count[1] - 1;
154     info->log2_max_frame_num_minus4              = sps->log2_max_frame_num - 4;
155     info->pic_order_cnt_type                     = sps->poc_type;
156     info->log2_max_pic_order_cnt_lsb_minus4      = sps->poc_type ? 0 : sps->log2_max_poc_lsb - 4;
157     info->delta_pic_order_always_zero_flag       = sps->delta_pic_order_always_zero_flag;
158     info->direct_8x8_inference_flag              = sps->direct_8x8_inference_flag;
159 #ifdef VDP_DECODER_PROFILE_H264_HIGH_444_PREDICTIVE
160     info2->qpprime_y_zero_transform_bypass_flag  = sps->transform_bypass;
161     info2->separate_colour_plane_flag            = sps->residual_color_transform_flag;
162 #endif
163     info->entropy_coding_mode_flag               = pps->cabac;
164     info->pic_order_present_flag                 = pps->pic_order_present;
165     info->deblocking_filter_control_present_flag = pps->deblocking_filter_parameters_present;
166     info->redundant_pic_cnt_present_flag         = pps->redundant_pic_cnt_present;
167
168     memcpy(info->scaling_lists_4x4, pps->scaling_matrix4,
169            sizeof(info->scaling_lists_4x4));
170     memcpy(info->scaling_lists_8x8[0], pps->scaling_matrix8[0],
171            sizeof(info->scaling_lists_8x8[0]));
172     memcpy(info->scaling_lists_8x8[1], pps->scaling_matrix8[3],
173            sizeof(info->scaling_lists_8x8[1]));
174
175     vdpau_h264_set_reference_frames(avctx);
176
177     return ff_vdpau_common_start_frame(pic_ctx, buffer, size);
178 }
179
180 static const uint8_t start_code_prefix[3] = { 0x00, 0x00, 0x01 };
181
182 static int vdpau_h264_decode_slice(AVCodecContext *avctx,
183                                    const uint8_t *buffer, uint32_t size)
184 {
185     H264Context *h = avctx->priv_data;
186     H264Picture *pic = h->cur_pic_ptr;
187     struct vdpau_picture_context *pic_ctx = pic->hwaccel_picture_private;
188     int val;
189
190     val = ff_vdpau_add_buffer(pic_ctx, start_code_prefix, 3);
191     if (val)
192         return val;
193
194     val = ff_vdpau_add_buffer(pic_ctx, buffer, size);
195     if (val)
196         return val;
197
198     pic_ctx->info.h264.slice_count++;
199     return 0;
200 }
201
202 static int vdpau_h264_end_frame(AVCodecContext *avctx)
203 {
204     H264Context *h = avctx->priv_data;
205     H264SliceContext *sl = &h->slice_ctx[0];
206     H264Picture *pic = h->cur_pic_ptr;
207     struct vdpau_picture_context *pic_ctx = pic->hwaccel_picture_private;
208     int val;
209
210     val = ff_vdpau_common_end_frame(avctx, pic->f, pic_ctx);
211     if (val < 0)
212         return val;
213
214     ff_h264_draw_horiz_band(h, sl, 0, h->avctx->height);
215     return 0;
216 }
217
218 static int vdpau_h264_init(AVCodecContext *avctx)
219 {
220     VdpDecoderProfile profile;
221     uint32_t level = avctx->level;
222
223     switch (avctx->profile & ~FF_PROFILE_H264_INTRA) {
224     case FF_PROFILE_H264_BASELINE:
225         profile = VDP_DECODER_PROFILE_H264_BASELINE;
226         break;
227     case FF_PROFILE_H264_CONSTRAINED_BASELINE:
228 #ifdef VDP_DECODER_PROFILE_H264_CONSTRAINED_BASELINE
229         profile = VDP_DECODER_PROFILE_H264_CONSTRAINED_BASELINE;
230         break;
231 #endif
232     case FF_PROFILE_H264_MAIN:
233         profile = VDP_DECODER_PROFILE_H264_MAIN;
234         break;
235     case FF_PROFILE_H264_HIGH:
236         profile = VDP_DECODER_PROFILE_H264_HIGH;
237         break;
238 #ifdef VDP_DECODER_PROFILE_H264_EXTENDED
239     case FF_PROFILE_H264_EXTENDED:
240         profile = VDP_DECODER_PROFILE_H264_EXTENDED;
241         break;
242 #endif
243     case FF_PROFILE_H264_HIGH_10:
244         /* XXX: High 10 can be treated as High so long as only 8 bits per
245          * format are supported. */
246         profile = VDP_DECODER_PROFILE_H264_HIGH;
247         break;
248 #ifdef VDP_DECODER_PROFILE_H264_HIGH_444_PREDICTIVE
249     case FF_PROFILE_H264_HIGH_422:
250     case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
251     case FF_PROFILE_H264_CAVLC_444:
252         profile = VDP_DECODER_PROFILE_H264_HIGH_444_PREDICTIVE;
253         break;
254 #endif
255     default:
256         return AVERROR(ENOTSUP);
257     }
258
259     if ((avctx->profile & FF_PROFILE_H264_INTRA) && avctx->level == 11)
260         level = VDP_DECODER_LEVEL_H264_1b;
261
262     return ff_vdpau_common_init(avctx, profile, level);
263 }
264
265 const AVHWAccel ff_h264_vdpau_hwaccel = {
266     .name           = "h264_vdpau",
267     .type           = AVMEDIA_TYPE_VIDEO,
268     .id             = AV_CODEC_ID_H264,
269     .pix_fmt        = AV_PIX_FMT_VDPAU,
270     .start_frame    = vdpau_h264_start_frame,
271     .end_frame      = vdpau_h264_end_frame,
272     .decode_slice   = vdpau_h264_decode_slice,
273     .frame_priv_data_size = sizeof(struct vdpau_picture_context),
274     .init           = vdpau_h264_init,
275     .uninit         = ff_vdpau_common_uninit,
276     .frame_params   = ff_vdpau_common_frame_params,
277     .priv_data_size = sizeof(VDPAUContext),
278     .caps_internal  = HWACCEL_CAP_ASYNC_SAFE,
279 };