dxva2_vc1.c 16.9 KB
Newer Older
1 2 3 4 5
/*
 * DXVA2 WMV3/VC-1 HW acceleration.
 *
 * copyright (c) 2010 Laurent Aimar
 *
6
 * This file is part of Libav.
7
 *
8
 * Libav is free software; you can redistribute it and/or
9 10 11 12
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
13
 * Libav is distributed in the hope that it will be useful,
14 15 16 17 18
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
19
 * License along with Libav; if not, write to the Free Software
20 21 22
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

23
#include "mpegutils.h"
24 25 26
#include "vc1.h"
#include "vc1data.h"

27 28 29 30 31
// The headers above may include w32threads.h, which uses the original
// _WIN32_WINNT define, while dxva2_internal.h redefines it to target a
// potentially newer version.
#include "dxva2_internal.h"

32 33 34 35 36 37 38 39 40
struct dxva2_picture_context {
    DXVA_PictureParameters pp;
    DXVA_SliceInfo         si;

    const uint8_t          *bitstream;
    unsigned               bitstream_size;
};

static void fill_picture_parameters(AVCodecContext *avctx,
41
                                    AVDXVAContext *ctx, const VC1Context *v,
42 43 44 45 46 47 48
                                    DXVA_PictureParameters *pp)
{
    const MpegEncContext *s = &v->s;
    const Picture *current_picture = s->current_picture_ptr;

    memset(pp, 0, sizeof(*pp));
    pp->wDecodedPictureIndex    =
49
    pp->wDeblockedPictureIndex  = ff_dxva2_get_surface_index(avctx, ctx, current_picture->f);
50
    if (s->pict_type != AV_PICTURE_TYPE_I && !v->bi_type)
51
        pp->wForwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->last_picture.f);
52 53
    else
        pp->wForwardRefPictureIndex = 0xffff;
54
    if (s->pict_type == AV_PICTURE_TYPE_B && !v->bi_type)
55
        pp->wBackwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->next_picture.f);
56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75
    else
        pp->wBackwardRefPictureIndex = 0xffff;
    if (v->profile == PROFILE_ADVANCED) {
        /* It is the cropped width/height -1 of the frame */
        pp->wPicWidthInMBminus1 = avctx->width  - 1;
        pp->wPicHeightInMBminus1= avctx->height - 1;
    } else {
        /* It is the coded width/height in macroblock -1 of the frame */
        pp->wPicWidthInMBminus1 = s->mb_width  - 1;
        pp->wPicHeightInMBminus1= s->mb_height - 1;
    }
    pp->bMacroblockWidthMinus1  = 15;
    pp->bMacroblockHeightMinus1 = 15;
    pp->bBlockWidthMinus1       = 7;
    pp->bBlockHeightMinus1      = 7;
    pp->bBPPminus1              = 7;
    if (s->picture_structure & PICT_TOP_FIELD)
        pp->bPicStructure      |= 0x01;
    if (s->picture_structure & PICT_BOTTOM_FIELD)
        pp->bPicStructure      |= 0x02;
76
    pp->bSecondField            = v->interlace && v->fcm != ILACE_FIELD && !s->first_field;
77 78
    pp->bPicIntra               = s->pict_type == AV_PICTURE_TYPE_I || v->bi_type;
    pp->bPicBackwardPrediction  = s->pict_type == AV_PICTURE_TYPE_B && !v->bi_type;
79
    pp->bBidirectionalAveragingMode = (1                                           << 7) |
80 81
                                      ((DXVA_CONTEXT_CFG_INTRARESID(avctx, ctx) != 0) << 6) |
                                      ((DXVA_CONTEXT_CFG_RESIDACCEL(avctx, ctx) != 0) << 5) |
82 83 84 85 86 87 88
                                      ((v->lumscale != 32 || v->lumshift != 0)     << 4) |
                                      ((v->profile == PROFILE_ADVANCED)            << 3);
    pp->bMVprecisionAndChromaRelation = ((v->mv_mode == MV_PMODE_1MV_HPEL_BILIN) << 3) |
                                        (1                                       << 2) |
                                        (0                                       << 1) |
                                        (!s->quarter_sample                          );
    pp->bChromaFormat           = v->chromaformat;
89 90 91 92 93
    DXVA_CONTEXT_REPORT_ID(avctx, ctx)++;
    if (DXVA_CONTEXT_REPORT_ID(avctx, ctx) >= (1 << 16))
        DXVA_CONTEXT_REPORT_ID(avctx, ctx) = 1;
    pp->bPicScanFixed           = DXVA_CONTEXT_REPORT_ID(avctx, ctx) >> 8;
    pp->bPicScanMethod          = DXVA_CONTEXT_REPORT_ID(avctx, ctx) & 0xff;
94 95 96 97 98 99 100 101 102 103 104
    pp->bPicReadbackRequests    = 0;
    pp->bRcontrol               = v->rnd;
    pp->bPicSpatialResid8       = (v->panscanflag  << 7) |
                                  (v->refdist_flag << 6) |
                                  (s->loop_filter  << 5) |
                                  (v->fastuvmc     << 4) |
                                  (v->extended_mv  << 3) |
                                  (v->dquant       << 1) |
                                  (v->vstransform      );
    pp->bPicOverflowBlocks      = (v->quantizer_mode << 6) |
                                  (v->multires       << 5) |
105
                                  (v->resync_marker  << 4) |
106 107
                                  (v->rangered       << 3) |
                                  (s->max_b_frames       );
108
    pp->bPicExtrapolation       = (!v->interlace || v->fcm == PROGRESSIVE) ? 1 : 2;
109 110
    pp->bPicDeblocked           = ((!pp->bPicBackwardPrediction && v->overlap)        << 6) |
                                  ((v->profile != PROFILE_ADVANCED && v->rangeredfrm) << 5) |
111 112 113 114 115 116
                                  (s->loop_filter                                     << 1);
    pp->bPicDeblockConfined     = (v->postprocflag             << 7) |
                                  (v->broadcast                << 6) |
                                  (v->interlace                << 5) |
                                  (v->tfcntrflag               << 4) |
                                  (v->finterpflag              << 3) |
117
                                  ((s->pict_type != AV_PICTURE_TYPE_B) << 2) |
118 119
                                  (v->psf                      << 1) |
                                  (v->extended_dmv                 );
120
    if (s->pict_type != AV_PICTURE_TYPE_I)
121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163
        pp->bPic4MVallowed      = v->mv_mode == MV_PMODE_MIXED_MV ||
                                  (v->mv_mode == MV_PMODE_INTENSITY_COMP &&
                                   v->mv_mode2 == MV_PMODE_MIXED_MV);
    if (v->profile == PROFILE_ADVANCED)
        pp->bPicOBMC            = (v->range_mapy_flag  << 7) |
                                  (v->range_mapy       << 4) |
                                  (v->range_mapuv_flag << 3) |
                                  (v->range_mapuv          );
    pp->bPicBinPB               = 0;
    pp->bMV_RPS                 = 0;
    pp->bReservedBits           = 0;
    if (s->picture_structure == PICT_FRAME) {
        pp->wBitstreamFcodes        = v->lumscale;
        pp->wBitstreamPCEelements   = v->lumshift;
    } else {
        /* Syntax: (top_field_param << 8) | bottom_field_param */
        pp->wBitstreamFcodes        = (v->lumscale << 8) | v->lumscale;
        pp->wBitstreamPCEelements   = (v->lumshift << 8) | v->lumshift;
    }
    pp->bBitstreamConcealmentNeed   = 0;
    pp->bBitstreamConcealmentMethod = 0;
}

static void fill_slice(AVCodecContext *avctx, DXVA_SliceInfo *slice,
                       unsigned position, unsigned size)
{
    const VC1Context *v = avctx->priv_data;
    const MpegEncContext *s = &v->s;

    memset(slice, 0, sizeof(*slice));
    slice->wHorizontalPosition = 0;
    slice->wVerticalPosition   = s->mb_y;
    slice->dwSliceBitsInBuffer = 8 * size;
    slice->dwSliceDataLocation = position;
    slice->bStartCodeBitOffset = 0;
    slice->bReservedBits       = 0;
    slice->wMBbitOffset        = get_bits_count(&s->gb);
    slice->wNumberMBsInSlice   = s->mb_width * s->mb_height; /* XXX We assume 1 slice */
    slice->wQuantizerScaleCode = v->pq;
    slice->wBadSliceChopping   = 0;
}

static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx,
164 165
                                             DECODER_BUFFER_DESC *bs,
                                             DECODER_BUFFER_DESC *sc)
166 167
{
    const VC1Context *v = avctx->priv_data;
168
    AVDXVAContext *ctx = DXVA_CONTEXT(avctx);
169
    const MpegEncContext *s = &v->s;
170
    struct dxva2_picture_context *ctx_pic = s->current_picture_ptr->hwaccel_picture_private;
171 172 173 174

    DXVA_SliceInfo *slice = &ctx_pic->si;

    static const uint8_t start_code[] = { 0, 0, 1, 0x0d };
175
    const unsigned start_code_size = avctx->codec_id == AV_CODEC_ID_VC1 ? sizeof(start_code) : 0;
176 177 178 179
    const unsigned slice_size = slice->dwSliceBitsInBuffer / 8;
    const unsigned padding = 128 - ((start_code_size + slice_size) & 127);
    const unsigned data_size = start_code_size + slice_size + padding;

180
    void     *dxva_data_ptr;
181 182 183
    uint8_t  *dxva_data;
    unsigned dxva_size;
    int result;
184
    unsigned type;
185

186
#if CONFIG_D3D11VA
187
    if (ff_dxva2_is_d3d11(avctx)) {
188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204
        type = D3D11_VIDEO_DECODER_BUFFER_BITSTREAM;
        if (FAILED(ID3D11VideoContext_GetDecoderBuffer(D3D11VA_CONTEXT(ctx)->video_context,
                                                       D3D11VA_CONTEXT(ctx)->decoder,
                                                       type,
                                                       &dxva_size, &dxva_data_ptr)))
            return -1;
    }
#endif
#if CONFIG_DXVA2
    if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) {
        type = DXVA2_BitStreamDateBufferType;
        if (FAILED(IDirectXVideoDecoder_GetBuffer(DXVA2_CONTEXT(ctx)->decoder,
                                                  type,
                                                  &dxva_data_ptr, &dxva_size)))
            return -1;
    }
#endif
205

206
    dxva_data = dxva_data_ptr;
207 208 209 210 211 212 213 214 215 216
    result = data_size <= dxva_size ? 0 : -1;
    if (!result) {
        if (start_code_size > 0)
            memcpy(dxva_data, start_code, start_code_size);
        memcpy(dxva_data + start_code_size,
               ctx_pic->bitstream + slice->dwSliceDataLocation, slice_size);
        if (padding > 0)
            memset(dxva_data + start_code_size + slice_size, 0, padding);
        slice->dwSliceBitsInBuffer = 8 * data_size;
    }
217
#if CONFIG_D3D11VA
218
    if (ff_dxva2_is_d3d11(avctx))
219 220 221 222 223 224 225 226
        if (FAILED(ID3D11VideoContext_ReleaseDecoderBuffer(D3D11VA_CONTEXT(ctx)->video_context, D3D11VA_CONTEXT(ctx)->decoder, type)))
            return -1;
#endif
#if CONFIG_DXVA2
    if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD)
        if (FAILED(IDirectXVideoDecoder_ReleaseBuffer(DXVA2_CONTEXT(ctx)->decoder, type)))
            return -1;
#endif
227 228 229
    if (result)
        return result;

230
#if CONFIG_D3D11VA
231
    if (ff_dxva2_is_d3d11(avctx)) {
232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252
        D3D11_VIDEO_DECODER_BUFFER_DESC *dsc11 = bs;
        memset(dsc11, 0, sizeof(*dsc11));
        dsc11->BufferType           = type;
        dsc11->DataSize             = data_size;
        dsc11->NumMBsInBuffer       = s->mb_width * s->mb_height;

        type = D3D11_VIDEO_DECODER_BUFFER_SLICE_CONTROL;
    }
#endif
#if CONFIG_DXVA2
    if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) {
        DXVA2_DecodeBufferDesc *dsc2 = bs;
        memset(dsc2, 0, sizeof(*dsc2));
        dsc2->CompressedBufferType = type;
        dsc2->DataSize             = data_size;
        dsc2->NumMBsInBuffer       = s->mb_width * s->mb_height;

        type = DXVA2_SliceControlBufferType;
    }
#endif
    assert((data_size & 127) == 0);
253 254

    return ff_dxva2_commit_buffer(avctx, ctx, sc,
255 256
                                  type,
                                  slice, sizeof(*slice), s->mb_width * s->mb_height);
257 258
}

259 260 261
static int dxva2_vc1_start_frame(AVCodecContext *avctx,
                                 av_unused const uint8_t *buffer,
                                 av_unused uint32_t size)
262 263
{
    const VC1Context *v = avctx->priv_data;
264
    AVDXVAContext *ctx = DXVA_CONTEXT(avctx);
265
    struct dxva2_picture_context *ctx_pic = v->s.current_picture_ptr->hwaccel_picture_private;
266

267
    if (!DXVA_CONTEXT_VALID(avctx, ctx))
268 269 270 271 272 273 274 275 276 277
        return -1;
    assert(ctx_pic);

    fill_picture_parameters(avctx, ctx, v, &ctx_pic->pp);

    ctx_pic->bitstream_size = 0;
    ctx_pic->bitstream      = NULL;
    return 0;
}

278 279 280
static int dxva2_vc1_decode_slice(AVCodecContext *avctx,
                                  const uint8_t *buffer,
                                  uint32_t size)
281 282 283
{
    const VC1Context *v = avctx->priv_data;
    const Picture *current_picture = v->s.current_picture_ptr;
284
    struct dxva2_picture_context *ctx_pic = current_picture->hwaccel_picture_private;
285 286 287 288

    if (ctx_pic->bitstream_size > 0)
        return -1;

289
    if (avctx->codec_id == AV_CODEC_ID_VC1 &&
290 291 292 293 294 295 296 297 298 299 300 301
        size >= 4 && IS_MARKER(AV_RB32(buffer))) {
        buffer += 4;
        size   -= 4;
    }

    ctx_pic->bitstream_size = size;
    ctx_pic->bitstream      = buffer;

    fill_slice(avctx, &ctx_pic->si, 0, size);
    return 0;
}

302
static int dxva2_vc1_end_frame(AVCodecContext *avctx)
303 304
{
    VC1Context *v = avctx->priv_data;
305
    struct dxva2_picture_context *ctx_pic = v->s.current_picture_ptr->hwaccel_picture_private;
306
    int ret;
307 308 309 310

    if (ctx_pic->bitstream_size <= 0)
        return -1;

311
    ret = ff_dxva2_common_end_frame(avctx, v->s.current_picture_ptr->f,
312 313 314 315 316 317
                                    &ctx_pic->pp, sizeof(ctx_pic->pp),
                                    NULL, 0,
                                    commit_bitstream_and_slice_buffer);
    if (!ret)
        ff_mpeg_draw_horiz_band(&v->s, 0, avctx->height);
    return ret;
318 319 320
}

#if CONFIG_WMV3_DXVA2_HWACCEL
321
const AVHWAccel ff_wmv3_dxva2_hwaccel = {
322
    .name           = "wmv3_dxva2",
323
    .type           = AVMEDIA_TYPE_VIDEO,
324
    .id             = AV_CODEC_ID_WMV3,
325
    .pix_fmt        = AV_PIX_FMT_DXVA2_VLD,
326 327
    .init           = ff_dxva2_decode_init,
    .uninit         = ff_dxva2_decode_uninit,
328 329 330
    .start_frame    = dxva2_vc1_start_frame,
    .decode_slice   = dxva2_vc1_decode_slice,
    .end_frame      = dxva2_vc1_end_frame,
331
    .frame_params   = ff_dxva2_common_frame_params,
332
    .frame_priv_data_size = sizeof(struct dxva2_picture_context),
333
    .priv_data_size = sizeof(FFDXVASharedContext),
334 335 336
};
#endif

337
#if CONFIG_VC1_DXVA2_HWACCEL
338
const AVHWAccel ff_vc1_dxva2_hwaccel = {
339
    .name           = "vc1_dxva2",
340
    .type           = AVMEDIA_TYPE_VIDEO,
341
    .id             = AV_CODEC_ID_VC1,
342
    .pix_fmt        = AV_PIX_FMT_DXVA2_VLD,
343 344
    .init           = ff_dxva2_decode_init,
    .uninit         = ff_dxva2_decode_uninit,
345 346 347
    .start_frame    = dxva2_vc1_start_frame,
    .decode_slice   = dxva2_vc1_decode_slice,
    .end_frame      = dxva2_vc1_end_frame,
348
    .frame_params   = ff_dxva2_common_frame_params,
349
    .frame_priv_data_size = sizeof(struct dxva2_picture_context),
350
    .priv_data_size = sizeof(FFDXVASharedContext),
351
};
352 353 354
#endif

#if CONFIG_WMV3_D3D11VA_HWACCEL
355
const AVHWAccel ff_wmv3_d3d11va_hwaccel = {
356 357 358 359
    .name           = "wmv3_d3d11va",
    .type           = AVMEDIA_TYPE_VIDEO,
    .id             = AV_CODEC_ID_WMV3,
    .pix_fmt        = AV_PIX_FMT_D3D11VA_VLD,
360 361
    .init           = ff_dxva2_decode_init,
    .uninit         = ff_dxva2_decode_uninit,
362 363 364
    .start_frame    = dxva2_vc1_start_frame,
    .decode_slice   = dxva2_vc1_decode_slice,
    .end_frame      = dxva2_vc1_end_frame,
365
    .frame_params   = ff_dxva2_common_frame_params,
366
    .frame_priv_data_size = sizeof(struct dxva2_picture_context),
367 368 369 370 371
    .priv_data_size = sizeof(FFDXVASharedContext),
};
#endif

#if CONFIG_WMV3_D3D11VA2_HWACCEL
372
const AVHWAccel ff_wmv3_d3d11va2_hwaccel = {
373 374 375 376 377 378 379 380 381
    .name           = "wmv3_d3d11va2",
    .type           = AVMEDIA_TYPE_VIDEO,
    .id             = AV_CODEC_ID_WMV3,
    .pix_fmt        = AV_PIX_FMT_D3D11,
    .init           = ff_dxva2_decode_init,
    .uninit         = ff_dxva2_decode_uninit,
    .start_frame    = dxva2_vc1_start_frame,
    .decode_slice   = dxva2_vc1_decode_slice,
    .end_frame      = dxva2_vc1_end_frame,
382
    .frame_params   = ff_dxva2_common_frame_params,
383 384
    .frame_priv_data_size = sizeof(struct dxva2_picture_context),
    .priv_data_size = sizeof(FFDXVASharedContext),
385 386 387 388
};
#endif

#if CONFIG_VC1_D3D11VA_HWACCEL
389
const AVHWAccel ff_vc1_d3d11va_hwaccel = {
390 391 392 393
    .name           = "vc1_d3d11va",
    .type           = AVMEDIA_TYPE_VIDEO,
    .id             = AV_CODEC_ID_VC1,
    .pix_fmt        = AV_PIX_FMT_D3D11VA_VLD,
394 395 396 397 398
    .init           = ff_dxva2_decode_init,
    .uninit         = ff_dxva2_decode_uninit,
    .start_frame    = dxva2_vc1_start_frame,
    .decode_slice   = dxva2_vc1_decode_slice,
    .end_frame      = dxva2_vc1_end_frame,
399
    .frame_params   = ff_dxva2_common_frame_params,
400 401 402 403 404 405
    .frame_priv_data_size = sizeof(struct dxva2_picture_context),
    .priv_data_size = sizeof(FFDXVASharedContext),
};
#endif

#if CONFIG_VC1_D3D11VA2_HWACCEL
406
const AVHWAccel ff_vc1_d3d11va2_hwaccel = {
407 408 409 410 411 412
    .name           = "vc1_d3d11va2",
    .type           = AVMEDIA_TYPE_VIDEO,
    .id             = AV_CODEC_ID_VC1,
    .pix_fmt        = AV_PIX_FMT_D3D11,
    .init           = ff_dxva2_decode_init,
    .uninit         = ff_dxva2_decode_uninit,
413 414 415 416
    .start_frame    = dxva2_vc1_start_frame,
    .decode_slice   = dxva2_vc1_decode_slice,
    .end_frame      = dxva2_vc1_end_frame,
    .frame_priv_data_size = sizeof(struct dxva2_picture_context),
417
    .priv_data_size = sizeof(FFDXVASharedContext),
418 419
};
#endif