| 1 | /* |
| 2 | * H.264 HW decode acceleration through VA API |
| 3 | * |
| 4 | * Copyright (C) 2008-2009 Splitted-Desktop Systems |
| 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 "vaapi_internal.h" |
| 24 | #include "h264.h" |
| 25 | #include "mpegutils.h" |
| 26 | |
| 27 | /** |
| 28 | * @file |
| 29 | * This file implements the glue code between FFmpeg's and VA API's |
| 30 | * structures for H.264 decoding. |
| 31 | */ |
| 32 | |
| 33 | /** |
| 34 | * Initialize an empty VA API picture. |
| 35 | * |
| 36 | * VA API requires a fixed-size reference picture array. |
| 37 | */ |
| 38 | static void init_vaapi_pic(VAPictureH264 *va_pic) |
| 39 | { |
| 40 | va_pic->picture_id = VA_INVALID_ID; |
| 41 | va_pic->flags = VA_PICTURE_H264_INVALID; |
| 42 | va_pic->TopFieldOrderCnt = 0; |
| 43 | va_pic->BottomFieldOrderCnt = 0; |
| 44 | } |
| 45 | |
| 46 | /** |
| 47 | * Translate an FFmpeg Picture into its VA API form. |
| 48 | * |
| 49 | * @param[out] va_pic A pointer to VA API's own picture struct |
| 50 | * @param[in] pic A pointer to the FFmpeg picture struct to convert |
| 51 | * @param[in] pic_structure The picture field type (as defined in mpegvideo.h), |
| 52 | * supersedes pic's field type if nonzero. |
| 53 | */ |
| 54 | static void fill_vaapi_pic(VAPictureH264 *va_pic, |
| 55 | H264Picture *pic, |
| 56 | int pic_structure) |
| 57 | { |
| 58 | if (pic_structure == 0) |
| 59 | pic_structure = pic->reference; |
| 60 | pic_structure &= PICT_FRAME; /* PICT_TOP_FIELD|PICT_BOTTOM_FIELD */ |
| 61 | |
| 62 | va_pic->picture_id = ff_vaapi_get_surface_id(&pic->f); |
| 63 | va_pic->frame_idx = pic->long_ref ? pic->pic_id : pic->frame_num; |
| 64 | |
| 65 | va_pic->flags = 0; |
| 66 | if (pic_structure != PICT_FRAME) |
| 67 | va_pic->flags |= (pic_structure & PICT_TOP_FIELD) ? VA_PICTURE_H264_TOP_FIELD : VA_PICTURE_H264_BOTTOM_FIELD; |
| 68 | if (pic->reference) |
| 69 | va_pic->flags |= pic->long_ref ? VA_PICTURE_H264_LONG_TERM_REFERENCE : VA_PICTURE_H264_SHORT_TERM_REFERENCE; |
| 70 | |
| 71 | va_pic->TopFieldOrderCnt = 0; |
| 72 | if (pic->field_poc[0] != INT_MAX) |
| 73 | va_pic->TopFieldOrderCnt = pic->field_poc[0]; |
| 74 | |
| 75 | va_pic->BottomFieldOrderCnt = 0; |
| 76 | if (pic->field_poc[1] != INT_MAX) |
| 77 | va_pic->BottomFieldOrderCnt = pic->field_poc[1]; |
| 78 | } |
| 79 | |
| 80 | /** Decoded Picture Buffer (DPB). */ |
| 81 | typedef struct DPB { |
| 82 | int size; ///< Current number of reference frames in the DPB |
| 83 | int max_size; ///< Max number of reference frames. This is FF_ARRAY_ELEMS(VAPictureParameterBufferH264.ReferenceFrames) |
| 84 | VAPictureH264 *va_pics; ///< Pointer to VAPictureParameterBufferH264.ReferenceFrames array |
| 85 | } DPB; |
| 86 | |
| 87 | /** |
| 88 | * Append picture to the decoded picture buffer, in a VA API form that |
| 89 | * merges the second field picture attributes with the first, if |
| 90 | * available. The decoded picture buffer's size must be large enough |
| 91 | * to receive the new VA API picture object. |
| 92 | */ |
| 93 | static int dpb_add(DPB *dpb, H264Picture *pic) |
| 94 | { |
| 95 | int i; |
| 96 | |
| 97 | if (dpb->size >= dpb->max_size) |
| 98 | return -1; |
| 99 | |
| 100 | for (i = 0; i < dpb->size; i++) { |
| 101 | VAPictureH264 * const va_pic = &dpb->va_pics[i]; |
| 102 | if (va_pic->picture_id == ff_vaapi_get_surface_id(&pic->f)) { |
| 103 | VAPictureH264 temp_va_pic; |
| 104 | fill_vaapi_pic(&temp_va_pic, pic, 0); |
| 105 | |
| 106 | if ((temp_va_pic.flags ^ va_pic->flags) & (VA_PICTURE_H264_TOP_FIELD | VA_PICTURE_H264_BOTTOM_FIELD)) { |
| 107 | va_pic->flags |= temp_va_pic.flags & (VA_PICTURE_H264_TOP_FIELD | VA_PICTURE_H264_BOTTOM_FIELD); |
| 108 | /* Merge second field */ |
| 109 | if (temp_va_pic.flags & VA_PICTURE_H264_TOP_FIELD) { |
| 110 | va_pic->TopFieldOrderCnt = temp_va_pic.TopFieldOrderCnt; |
| 111 | } else { |
| 112 | va_pic->BottomFieldOrderCnt = temp_va_pic.BottomFieldOrderCnt; |
| 113 | } |
| 114 | } |
| 115 | return 0; |
| 116 | } |
| 117 | } |
| 118 | |
| 119 | fill_vaapi_pic(&dpb->va_pics[dpb->size++], pic, 0); |
| 120 | return 0; |
| 121 | } |
| 122 | |
| 123 | /** Fill in VA API reference frames array. */ |
| 124 | static int fill_vaapi_ReferenceFrames(VAPictureParameterBufferH264 *pic_param, |
| 125 | H264Context *h) |
| 126 | { |
| 127 | DPB dpb; |
| 128 | int i; |
| 129 | |
| 130 | dpb.size = 0; |
| 131 | dpb.max_size = FF_ARRAY_ELEMS(pic_param->ReferenceFrames); |
| 132 | dpb.va_pics = pic_param->ReferenceFrames; |
| 133 | for (i = 0; i < dpb.max_size; i++) |
| 134 | init_vaapi_pic(&dpb.va_pics[i]); |
| 135 | |
| 136 | for (i = 0; i < h->short_ref_count; i++) { |
| 137 | H264Picture * const pic = h->short_ref[i]; |
| 138 | if (pic && pic->reference && dpb_add(&dpb, pic) < 0) |
| 139 | return -1; |
| 140 | } |
| 141 | |
| 142 | for (i = 0; i < 16; i++) { |
| 143 | H264Picture * const pic = h->long_ref[i]; |
| 144 | if (pic && pic->reference && dpb_add(&dpb, pic) < 0) |
| 145 | return -1; |
| 146 | } |
| 147 | return 0; |
| 148 | } |
| 149 | |
| 150 | /** |
| 151 | * Fill in VA API reference picture lists from the FFmpeg reference |
| 152 | * picture list. |
| 153 | * |
| 154 | * @param[out] RefPicList VA API internal reference picture list |
| 155 | * @param[in] ref_list A pointer to the FFmpeg reference list |
| 156 | * @param[in] ref_count The number of reference pictures in ref_list |
| 157 | */ |
| 158 | static void fill_vaapi_RefPicList(VAPictureH264 RefPicList[32], |
| 159 | H264Picture *ref_list, |
| 160 | unsigned int ref_count) |
| 161 | { |
| 162 | unsigned int i, n = 0; |
| 163 | for (i = 0; i < ref_count; i++) |
| 164 | if (ref_list[i].reference) |
| 165 | fill_vaapi_pic(&RefPicList[n++], &ref_list[i], 0); |
| 166 | |
| 167 | for (; n < 32; n++) |
| 168 | init_vaapi_pic(&RefPicList[n]); |
| 169 | } |
| 170 | |
| 171 | /** |
| 172 | * Fill in prediction weight table. |
| 173 | * |
| 174 | * VA API requires a plain prediction weight table as it does not infer |
| 175 | * any value. |
| 176 | * |
| 177 | * @param[in] h A pointer to the current H.264 context |
| 178 | * @param[in] list The reference frame list index to use |
| 179 | * @param[out] luma_weight_flag VA API plain luma weight flag |
| 180 | * @param[out] luma_weight VA API plain luma weight table |
| 181 | * @param[out] luma_offset VA API plain luma offset table |
| 182 | * @param[out] chroma_weight_flag VA API plain chroma weight flag |
| 183 | * @param[out] chroma_weight VA API plain chroma weight table |
| 184 | * @param[out] chroma_offset VA API plain chroma offset table |
| 185 | */ |
| 186 | static void fill_vaapi_plain_pred_weight_table(H264Context *h, |
| 187 | int list, |
| 188 | unsigned char *luma_weight_flag, |
| 189 | short luma_weight[32], |
| 190 | short luma_offset[32], |
| 191 | unsigned char *chroma_weight_flag, |
| 192 | short chroma_weight[32][2], |
| 193 | short chroma_offset[32][2]) |
| 194 | { |
| 195 | unsigned int i, j; |
| 196 | |
| 197 | *luma_weight_flag = h->luma_weight_flag[list]; |
| 198 | *chroma_weight_flag = h->chroma_weight_flag[list]; |
| 199 | |
| 200 | for (i = 0; i < h->ref_count[list]; i++) { |
| 201 | /* VA API also wants the inferred (default) values, not |
| 202 | only what is available in the bitstream (7.4.3.2). */ |
| 203 | if (h->luma_weight_flag[list]) { |
| 204 | luma_weight[i] = h->luma_weight[i][list][0]; |
| 205 | luma_offset[i] = h->luma_weight[i][list][1]; |
| 206 | } else { |
| 207 | luma_weight[i] = 1 << h->luma_log2_weight_denom; |
| 208 | luma_offset[i] = 0; |
| 209 | } |
| 210 | for (j = 0; j < 2; j++) { |
| 211 | if (h->chroma_weight_flag[list]) { |
| 212 | chroma_weight[i][j] = h->chroma_weight[i][list][j][0]; |
| 213 | chroma_offset[i][j] = h->chroma_weight[i][list][j][1]; |
| 214 | } else { |
| 215 | chroma_weight[i][j] = 1 << h->chroma_log2_weight_denom; |
| 216 | chroma_offset[i][j] = 0; |
| 217 | } |
| 218 | } |
| 219 | } |
| 220 | } |
| 221 | |
| 222 | /** Initialize and start decoding a frame with VA API. */ |
| 223 | static int vaapi_h264_start_frame(AVCodecContext *avctx, |
| 224 | av_unused const uint8_t *buffer, |
| 225 | av_unused uint32_t size) |
| 226 | { |
| 227 | H264Context * const h = avctx->priv_data; |
| 228 | struct vaapi_context * const vactx = avctx->hwaccel_context; |
| 229 | VAPictureParameterBufferH264 *pic_param; |
| 230 | VAIQMatrixBufferH264 *iq_matrix; |
| 231 | |
| 232 | av_dlog(avctx, "vaapi_h264_start_frame()\n"); |
| 233 | |
| 234 | vactx->slice_param_size = sizeof(VASliceParameterBufferH264); |
| 235 | |
| 236 | /* Fill in VAPictureParameterBufferH264. */ |
| 237 | pic_param = ff_vaapi_alloc_pic_param(vactx, sizeof(VAPictureParameterBufferH264)); |
| 238 | if (!pic_param) |
| 239 | return -1; |
| 240 | fill_vaapi_pic(&pic_param->CurrPic, h->cur_pic_ptr, h->picture_structure); |
| 241 | if (fill_vaapi_ReferenceFrames(pic_param, h) < 0) |
| 242 | return -1; |
| 243 | pic_param->picture_width_in_mbs_minus1 = h->mb_width - 1; |
| 244 | pic_param->picture_height_in_mbs_minus1 = h->mb_height - 1; |
| 245 | pic_param->bit_depth_luma_minus8 = h->sps.bit_depth_luma - 8; |
| 246 | pic_param->bit_depth_chroma_minus8 = h->sps.bit_depth_chroma - 8; |
| 247 | pic_param->num_ref_frames = h->sps.ref_frame_count; |
| 248 | pic_param->seq_fields.value = 0; /* reset all bits */ |
| 249 | pic_param->seq_fields.bits.chroma_format_idc = h->sps.chroma_format_idc; |
| 250 | pic_param->seq_fields.bits.residual_colour_transform_flag = h->sps.residual_color_transform_flag; /* XXX: only for 4:4:4 high profile? */ |
| 251 | pic_param->seq_fields.bits.gaps_in_frame_num_value_allowed_flag = h->sps.gaps_in_frame_num_allowed_flag; |
| 252 | pic_param->seq_fields.bits.frame_mbs_only_flag = h->sps.frame_mbs_only_flag; |
| 253 | pic_param->seq_fields.bits.mb_adaptive_frame_field_flag = h->sps.mb_aff; |
| 254 | pic_param->seq_fields.bits.direct_8x8_inference_flag = h->sps.direct_8x8_inference_flag; |
| 255 | pic_param->seq_fields.bits.MinLumaBiPredSize8x8 = h->sps.level_idc >= 31; /* A.3.3.2 */ |
| 256 | pic_param->seq_fields.bits.log2_max_frame_num_minus4 = h->sps.log2_max_frame_num - 4; |
| 257 | pic_param->seq_fields.bits.pic_order_cnt_type = h->sps.poc_type; |
| 258 | pic_param->seq_fields.bits.log2_max_pic_order_cnt_lsb_minus4 = h->sps.log2_max_poc_lsb - 4; |
| 259 | pic_param->seq_fields.bits.delta_pic_order_always_zero_flag = h->sps.delta_pic_order_always_zero_flag; |
| 260 | pic_param->num_slice_groups_minus1 = h->pps.slice_group_count - 1; |
| 261 | pic_param->slice_group_map_type = h->pps.mb_slice_group_map_type; |
| 262 | pic_param->slice_group_change_rate_minus1 = 0; /* XXX: unimplemented in FFmpeg */ |
| 263 | pic_param->pic_init_qp_minus26 = h->pps.init_qp - 26; |
| 264 | pic_param->pic_init_qs_minus26 = h->pps.init_qs - 26; |
| 265 | pic_param->chroma_qp_index_offset = h->pps.chroma_qp_index_offset[0]; |
| 266 | pic_param->second_chroma_qp_index_offset = h->pps.chroma_qp_index_offset[1]; |
| 267 | pic_param->pic_fields.value = 0; /* reset all bits */ |
| 268 | pic_param->pic_fields.bits.entropy_coding_mode_flag = h->pps.cabac; |
| 269 | pic_param->pic_fields.bits.weighted_pred_flag = h->pps.weighted_pred; |
| 270 | pic_param->pic_fields.bits.weighted_bipred_idc = h->pps.weighted_bipred_idc; |
| 271 | pic_param->pic_fields.bits.transform_8x8_mode_flag = h->pps.transform_8x8_mode; |
| 272 | pic_param->pic_fields.bits.field_pic_flag = h->picture_structure != PICT_FRAME; |
| 273 | pic_param->pic_fields.bits.constrained_intra_pred_flag = h->pps.constrained_intra_pred; |
| 274 | pic_param->pic_fields.bits.pic_order_present_flag = h->pps.pic_order_present; |
| 275 | pic_param->pic_fields.bits.deblocking_filter_control_present_flag = h->pps.deblocking_filter_parameters_present; |
| 276 | pic_param->pic_fields.bits.redundant_pic_cnt_present_flag = h->pps.redundant_pic_cnt_present; |
| 277 | pic_param->pic_fields.bits.reference_pic_flag = h->nal_ref_idc != 0; |
| 278 | pic_param->frame_num = h->frame_num; |
| 279 | |
| 280 | /* Fill in VAIQMatrixBufferH264. */ |
| 281 | iq_matrix = ff_vaapi_alloc_iq_matrix(vactx, sizeof(VAIQMatrixBufferH264)); |
| 282 | if (!iq_matrix) |
| 283 | return -1; |
| 284 | memcpy(iq_matrix->ScalingList4x4, h->pps.scaling_matrix4, sizeof(iq_matrix->ScalingList4x4)); |
| 285 | memcpy(iq_matrix->ScalingList8x8[0], h->pps.scaling_matrix8[0], sizeof(iq_matrix->ScalingList8x8[0])); |
| 286 | memcpy(iq_matrix->ScalingList8x8[1], h->pps.scaling_matrix8[3], sizeof(iq_matrix->ScalingList8x8[0])); |
| 287 | return 0; |
| 288 | } |
| 289 | |
| 290 | /** End a hardware decoding based frame. */ |
| 291 | static int vaapi_h264_end_frame(AVCodecContext *avctx) |
| 292 | { |
| 293 | struct vaapi_context * const vactx = avctx->hwaccel_context; |
| 294 | H264Context * const h = avctx->priv_data; |
| 295 | int ret; |
| 296 | |
| 297 | av_dlog(avctx, "vaapi_h264_end_frame()\n"); |
| 298 | ret = ff_vaapi_commit_slices(vactx); |
| 299 | if (ret < 0) |
| 300 | goto finish; |
| 301 | |
| 302 | ret = ff_vaapi_render_picture(vactx, ff_vaapi_get_surface_id(&h->cur_pic_ptr->f)); |
| 303 | if (ret < 0) |
| 304 | goto finish; |
| 305 | |
| 306 | ff_h264_draw_horiz_band(h, 0, h->avctx->height); |
| 307 | |
| 308 | finish: |
| 309 | ff_vaapi_common_end_frame(avctx); |
| 310 | return ret; |
| 311 | } |
| 312 | |
| 313 | /** Decode the given H.264 slice with VA API. */ |
| 314 | static int vaapi_h264_decode_slice(AVCodecContext *avctx, |
| 315 | const uint8_t *buffer, |
| 316 | uint32_t size) |
| 317 | { |
| 318 | H264Context * const h = avctx->priv_data; |
| 319 | VASliceParameterBufferH264 *slice_param; |
| 320 | |
| 321 | av_dlog(avctx, "vaapi_h264_decode_slice(): buffer %p, size %d\n", |
| 322 | buffer, size); |
| 323 | |
| 324 | /* Fill in VASliceParameterBufferH264. */ |
| 325 | slice_param = (VASliceParameterBufferH264 *)ff_vaapi_alloc_slice(avctx->hwaccel_context, buffer, size); |
| 326 | if (!slice_param) |
| 327 | return -1; |
| 328 | slice_param->slice_data_bit_offset = get_bits_count(&h->gb) + 8; /* bit buffer started beyond nal_unit_type */ |
| 329 | slice_param->first_mb_in_slice = (h->mb_y >> FIELD_OR_MBAFF_PICTURE(h)) * h->mb_width + h->mb_x; |
| 330 | slice_param->slice_type = ff_h264_get_slice_type(h); |
| 331 | slice_param->direct_spatial_mv_pred_flag = h->slice_type == AV_PICTURE_TYPE_B ? h->direct_spatial_mv_pred : 0; |
| 332 | slice_param->num_ref_idx_l0_active_minus1 = h->list_count > 0 ? h->ref_count[0] - 1 : 0; |
| 333 | slice_param->num_ref_idx_l1_active_minus1 = h->list_count > 1 ? h->ref_count[1] - 1 : 0; |
| 334 | slice_param->cabac_init_idc = h->cabac_init_idc; |
| 335 | slice_param->slice_qp_delta = h->qscale - h->pps.init_qp; |
| 336 | slice_param->disable_deblocking_filter_idc = h->deblocking_filter < 2 ? !h->deblocking_filter : h->deblocking_filter; |
| 337 | slice_param->slice_alpha_c0_offset_div2 = h->slice_alpha_c0_offset / 2; |
| 338 | slice_param->slice_beta_offset_div2 = h->slice_beta_offset / 2; |
| 339 | slice_param->luma_log2_weight_denom = h->luma_log2_weight_denom; |
| 340 | slice_param->chroma_log2_weight_denom = h->chroma_log2_weight_denom; |
| 341 | |
| 342 | fill_vaapi_RefPicList(slice_param->RefPicList0, h->ref_list[0], h->list_count > 0 ? h->ref_count[0] : 0); |
| 343 | fill_vaapi_RefPicList(slice_param->RefPicList1, h->ref_list[1], h->list_count > 1 ? h->ref_count[1] : 0); |
| 344 | |
| 345 | fill_vaapi_plain_pred_weight_table(h, 0, |
| 346 | &slice_param->luma_weight_l0_flag, slice_param->luma_weight_l0, slice_param->luma_offset_l0, |
| 347 | &slice_param->chroma_weight_l0_flag, slice_param->chroma_weight_l0, slice_param->chroma_offset_l0); |
| 348 | fill_vaapi_plain_pred_weight_table(h, 1, |
| 349 | &slice_param->luma_weight_l1_flag, slice_param->luma_weight_l1, slice_param->luma_offset_l1, |
| 350 | &slice_param->chroma_weight_l1_flag, slice_param->chroma_weight_l1, slice_param->chroma_offset_l1); |
| 351 | return 0; |
| 352 | } |
| 353 | |
| 354 | AVHWAccel ff_h264_vaapi_hwaccel = { |
| 355 | .name = "h264_vaapi", |
| 356 | .type = AVMEDIA_TYPE_VIDEO, |
| 357 | .id = AV_CODEC_ID_H264, |
| 358 | .pix_fmt = AV_PIX_FMT_VAAPI_VLD, |
| 359 | .start_frame = vaapi_h264_start_frame, |
| 360 | .end_frame = vaapi_h264_end_frame, |
| 361 | .decode_slice = vaapi_h264_decode_slice, |
| 362 | }; |