src/gallium/drivers/nouveau/nvc0/nvc0_video.c

   1 /*
   2  * Copyright 2011-2013 Maarten Lankhorst
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice shall be included in
  12  * all copies or substantial portions of the Software.
  13  *
  14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
  18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  20  * OTHER DEALINGS IN THE SOFTWARE.
  21  */
  22
  23 #include "nvc0/nvc0_video.h"
  24
  25 #include "util/u_sampler.h"
  26 #include "util/u_format.h"
  27
  28 static void
  29 nvc0_decoder_begin_frame(struct pipe_video_codec *decoder,
  30                          struct pipe_video_buffer *target,
  31                          struct pipe_picture_desc *picture)
  32 {
  33    struct nouveau_vp3_decoder *dec = (struct nouveau_vp3_decoder *)decoder;
  34    uint32_t comm_seq = ++dec->fence_seq;
  35    unsigned ret = 0;
  36
  37    assert(dec);
  38    assert(target);
  39    assert(target->buffer_format == PIPE_FORMAT_NV12);
  40
  41    ret = nvc0_decoder_bsp_begin(dec, comm_seq);
  42
  43    assert(ret == 2);
  44 }
  45
  46 static void
  47 nvc0_decoder_decode_bitstream(struct pipe_video_codec *decoder,
  48                               struct pipe_video_buffer *video_target,
  49                               struct pipe_picture_desc *picture,
  50                               unsigned num_buffers,
  51                               const void *const *data,
  52                               const unsigned *num_bytes)
  53 {
  54    struct nouveau_vp3_decoder *dec = (struct nouveau_vp3_decoder *)decoder;
  55    uint32_t comm_seq = dec->fence_seq;
  56    unsigned ret = 0;
  57
  58    assert(decoder);
  59
  60    ret = nvc0_decoder_bsp_next(dec, comm_seq, num_buffers, data, num_bytes);
  61
  62    assert(ret == 2);
  63 }
  64
  65 static void
  66 nvc0_decoder_end_frame(struct pipe_video_codec *decoder,
  67                        struct pipe_video_buffer *video_target,
  68                        struct pipe_picture_desc *picture)
  69 {
  70    struct nouveau_vp3_decoder *dec = (struct nouveau_vp3_decoder *)decoder;
  71    struct nouveau_vp3_video_buffer *target = (struct nouveau_vp3_video_buffer *)video_target;
  72    uint32_t comm_seq = dec->fence_seq;
  73    union pipe_desc desc;
  74
  75    unsigned vp_caps, is_ref, ret;
  76    struct nouveau_vp3_video_buffer *refs[16] = {};
  77
  78    desc.base = picture;
  79
  80    ret = nvc0_decoder_bsp_end(dec, desc, target, comm_seq, &vp_caps, &is_ref, refs);
  81
  82    /* did we decode bitstream correctly? */
  83    assert(ret == 2);
  84
  85    nvc0_decoder_vp(dec, desc, target, comm_seq, vp_caps, is_ref, refs);
  86    nvc0_decoder_ppp(dec, desc, target, comm_seq);
  87 }
  88
  89 struct pipe_video_codec *
  90 nvc0_create_decoder(struct pipe_context *context,
  91                     const struct pipe_video_codec *templ)
  92 {
  93    struct nouveau_screen *screen = &((struct nvc0_context *)context)->screen->base;
  94    struct nouveau_vp3_decoder *dec;
  95    struct nouveau_pushbuf **push;
  96    union nouveau_bo_config cfg;
  97    bool kepler = screen->device->chipset >= 0xe0;
  98
  99    cfg.nvc0.tile_mode = 0x10;
 100    cfg.nvc0.memtype = 0xfe;
 101
 102    int ret, i;
 103    uint32_t codec = 1, ppp_codec = 3;
 104    uint32_t timeout;
 105    u32 tmp_size = 0;
 106
 107    if (getenv("XVMC_VL"))
 108        return vl_create_decoder(context, templ);
 109
 110    if (templ->entrypoint != PIPE_VIDEO_ENTRYPOINT_BITSTREAM) {
 111       debug_printf("%x\n", templ->entrypoint);
 112       return NULL;
 113    }
 114
 115    dec = CALLOC_STRUCT(nouveau_vp3_decoder);
 116    if (!dec)
 117       return NULL;
 118    dec->client = screen->client;
 119    dec->base = *templ;
 120    nouveau_vp3_decoder_init_common(&dec->base);
 121
 122    if (!kepler) {
 123       dec->bsp_idx = 5;
 124       dec->vp_idx = 6;
 125       dec->ppp_idx = 7;
 126    } else {
 127       dec->bsp_idx = 2;
 128       dec->vp_idx = 2;
 129       dec->ppp_idx = 2;
 130    }
 131
 132    for (i = 0; i < 3; ++i)
 133       if (i && !kepler) {
 134          dec->channel[i] = dec->channel[0];
 135          dec->pushbuf[i] = dec->pushbuf[0];
 136       } else {
 137          void *data;
 138          u32 size;
 139          struct nvc0_fifo nvc0_args = {};
 140          struct nve0_fifo nve0_args = {};
 141
 142          if (!kepler) {
 143             size = sizeof(nvc0_args);
 144             data = &nvc0_args;
 145          } else {
 146             unsigned engine[] = {
 147                NVE0_FIFO_ENGINE_BSP,
 148                NVE0_FIFO_ENGINE_VP,
 149                NVE0_FIFO_ENGINE_PPP
 150             };
 151
 152             nve0_args.engine = engine[i];
 153             size = sizeof(nve0_args);
 154             data = &nve0_args;
 155          }
 156
 157          ret = nouveau_object_new(&screen->device->object, 0,
 158                                   NOUVEAU_FIFO_CHANNEL_CLASS,
 159                                   data, size, &dec->channel[i]);
 160
 161          if (!ret)
 162             ret = nouveau_pushbuf_new(screen->client, dec->channel[i], 4,
 163                                    32 * 1024, true, &dec->pushbuf[i]);
 164          if (ret)
 165             break;
 166       }
 167    push = dec->pushbuf;
 168
 169    if (!kepler) {
 170       if (!ret)
 171          ret = nouveau_object_new(dec->channel[0], 0x390b1, 0x90b1, NULL, 0, &dec->bsp);
 172       if (!ret)
 173          ret = nouveau_object_new(dec->channel[1], 0x190b2, 0x90b2, NULL, 0, &dec->vp);
 174       if (!ret)
 175          ret = nouveau_object_new(dec->channel[2], 0x290b3, 0x90b3, NULL, 0, &dec->ppp);
 176    } else {
 177       if (!ret)
 178          ret = nouveau_object_new(dec->channel[0], 0x95b1, 0x95b1, NULL, 0, &dec->bsp);
 179       if (!ret)
 180          ret = nouveau_object_new(dec->channel[1], 0x95b2, 0x95b2, NULL, 0, &dec->vp);
 181       if (!ret)
 182          ret = nouveau_object_new(dec->channel[2], 0x90b3, 0x90b3, NULL, 0, &dec->ppp);
 183    }
 184    if (ret)
 185       goto fail;
 186
 187    BEGIN_NVC0(push[0], SUBC_BSP(NV01_SUBCHAN_OBJECT), 1);
 188    PUSH_DATA (push[0], dec->bsp->handle);
 189
 190    BEGIN_NVC0(push[1], SUBC_VP(NV01_SUBCHAN_OBJECT), 1);
 191    PUSH_DATA (push[1], dec->vp->handle);
 192
 193    BEGIN_NVC0(push[2], SUBC_PPP(NV01_SUBCHAN_OBJECT), 1);
 194    PUSH_DATA (push[2], dec->ppp->handle);
 195
 196    dec->base.context = context;
 197    dec->base.begin_frame = nvc0_decoder_begin_frame;
 198    dec->base.decode_bitstream = nvc0_decoder_decode_bitstream;
 199    dec->base.end_frame = nvc0_decoder_end_frame;
 200
 201    for (i = 0; i < NOUVEAU_VP3_VIDEO_QDEPTH && !ret; ++i)
 202       ret = nouveau_bo_new(screen->device, NOUVEAU_BO_VRAM,
 203                            0, 1 << 20, &cfg, &dec->bsp_bo[i]);
 204    if (!ret) {
 205       /* total fudge factor... just has to be bigger for higher bitrates? */
 206       unsigned inter_size = align(templ->width * templ->height * 2, 4 << 20);
 207       ret = nouveau_bo_new(screen->device, NOUVEAU_BO_VRAM,
 208                            0x100, inter_size, &cfg, &dec->inter_bo[0]);
 209    }
 210    if (!ret) {
 211       ret = nouveau_bo_new(screen->device, NOUVEAU_BO_VRAM,
 212                            0x100, dec->inter_bo[0]->size, &cfg,
 213                            &dec->inter_bo[1]);
 214    }
 215    if (ret)
 216       goto fail;
 217    switch (u_reduce_video_profile(templ->profile)) {
 218    case PIPE_VIDEO_FORMAT_MPEG12: {
 219       codec = 1;
 220       assert(templ->max_references <= 2);
 221       break;
 222    }
 223    case PIPE_VIDEO_FORMAT_MPEG4: {
 224       codec = 4;
 225       tmp_size = mb(templ->height)*16 * mb(templ->width)*16;
 226       assert(templ->max_references <= 2);
 227       break;
 228    }
 229    case PIPE_VIDEO_FORMAT_VC1: {
 230       ppp_codec = codec = 2;
 231       tmp_size = mb(templ->height)*16 * mb(templ->width)*16;
 232       assert(templ->max_references <= 2);
 233       break;
 234    }
 235    case PIPE_VIDEO_FORMAT_MPEG4_AVC: {
 236       codec = 3;
 237       dec->tmp_stride = 16 * mb_half(templ->width) * nouveau_vp3_video_align(templ->height) * 3 / 2;
 238       tmp_size = dec->tmp_stride * (templ->max_references + 1);
 239       assert(templ->max_references <= 16);
 240       break;
 241    }
 242    default:
 243       fprintf(stderr, "invalid codec\n");
 244       goto fail;
 245    }
 246
 247    if (screen->device->chipset < 0xd0) {
 248       ret = nouveau_bo_new(screen->device, NOUVEAU_BO_VRAM, 0,
 249                            0x4000, &cfg, &dec->fw_bo);
 250       if (ret)
 251          goto fail;
 252
 253       ret = nouveau_vp3_load_firmware(dec, templ->profile, screen->device->chipset);
 254       if (ret)
 255          goto fw_fail;
 256    }
 257
 258    if (codec != 3) {
 259       ret = nouveau_bo_new(screen->device, NOUVEAU_BO_VRAM, 0,
 260                            0x400, &cfg, &dec->bitplane_bo);
 261       if (ret)
 262          goto fail;
 263    }
 264
 265    dec->ref_stride = mb(templ->width)*16 * (mb_half(templ->height)*32 + nouveau_vp3_video_align(templ->height)/2);
 266    ret = nouveau_bo_new(screen->device, NOUVEAU_BO_VRAM, 0,
 267                         dec->ref_stride * (templ->max_references+2) + tmp_size,
 268                         &cfg, &dec->ref_bo);
 269    if (ret)
 270       goto fail;
 271
 272    timeout = 0;
 273
 274    BEGIN_NVC0(push[0], SUBC_BSP(0x200), 2);
 275    PUSH_DATA (push[0], codec);
 276    PUSH_DATA (push[0], timeout);
 277
 278    BEGIN_NVC0(push[1], SUBC_VP(0x200), 2);
 279    PUSH_DATA (push[1], codec);
 280    PUSH_DATA (push[1], timeout);
 281
 282    BEGIN_NVC0(push[2], SUBC_PPP(0x200), 2);
 283    PUSH_DATA (push[2], ppp_codec);
 284    PUSH_DATA (push[2], timeout);
 285
 286    ++dec->fence_seq;
 287
 288 #if NOUVEAU_VP3_DEBUG_FENCE
 289    ret = nouveau_bo_new(screen->device, NOUVEAU_BO_GART|NOUVEAU_BO_MAP,
 290                         0, 0x1000, NULL, &dec->fence_bo);
 291    if (ret)
 292       goto fail;
 293
 294    nouveau_bo_map(dec->fence_bo, NOUVEAU_BO_RDWR, screen->client);
 295    dec->fence_map = dec->fence_bo->map;
 296    dec->fence_map[0] = dec->fence_map[4] = dec->fence_map[8] = 0;
 297    dec->comm = (struct comm *)(dec->fence_map + (COMM_OFFSET/sizeof(*dec->fence_map)));
 298
 299    /* So lets test if the fence is working? */
 300    nouveau_pushbuf_space(push[0], 6, 1, 0);
 301    PUSH_REFN (push[0], dec->fence_bo, NOUVEAU_BO_GART|NOUVEAU_BO_RDWR);
 302    BEGIN_NVC0(push[0], SUBC_BSP(0x240), 3);
 303    PUSH_DATAh(push[0], dec->fence_bo->offset);
 304    PUSH_DATA (push[0], dec->fence_bo->offset);
 305    PUSH_DATA (push[0], dec->fence_seq);
 306
 307    BEGIN_NVC0(push[0], SUBC_BSP(0x304), 1);
 308    PUSH_DATA (push[0], 0);
 309    PUSH_KICK (push[0]);
 310
 311    nouveau_pushbuf_space(push[1], 6, 1, 0);
 312    PUSH_REFN (push[1], dec->fence_bo, NOUVEAU_BO_GART|NOUVEAU_BO_RDWR);
 313    BEGIN_NVC0(push[1], SUBC_VP(0x240), 3);
 314    PUSH_DATAh(push[1], (dec->fence_bo->offset + 0x10));
 315    PUSH_DATA (push[1], (dec->fence_bo->offset + 0x10));
 316    PUSH_DATA (push[1], dec->fence_seq);
 317
 318    BEGIN_NVC0(push[1], SUBC_VP(0x304), 1);
 319    PUSH_DATA (push[1], 0);
 320    PUSH_KICK (push[1]);
 321
 322    nouveau_pushbuf_space(push[2], 6, 1, 0);
 323    PUSH_REFN (push[2], dec->fence_bo, NOUVEAU_BO_GART|NOUVEAU_BO_RDWR);
 324    BEGIN_NVC0(push[2], SUBC_PPP(0x240), 3);
 325    PUSH_DATAh(push[2], (dec->fence_bo->offset + 0x20));
 326    PUSH_DATA (push[2], (dec->fence_bo->offset + 0x20));
 327    PUSH_DATA (push[2], dec->fence_seq);
 328
 329    BEGIN_NVC0(push[2], SUBC_PPP(0x304), 1);
 330    PUSH_DATA (push[2], 0);
 331    PUSH_KICK (push[2]);
 332
 333    usleep(100);
 334    while (dec->fence_seq > dec->fence_map[0] ||
 335           dec->fence_seq > dec->fence_map[4] ||
 336           dec->fence_seq > dec->fence_map[8]) {
 337       debug_printf("%u: %u %u %u\n", dec->fence_seq, dec->fence_map[0], dec->fence_map[4], dec->fence_map[8]);
 338       usleep(100);
 339    }
 340    debug_printf("%u: %u %u %u\n", dec->fence_seq, dec->fence_map[0], dec->fence_map[4], dec->fence_map[8]);
 341 #endif
 342
 343    return &dec->base;
 344
 345 fw_fail:
 346    debug_printf("Cannot create decoder without firmware..\n");
 347    dec->base.destroy(&dec->base);
 348    return NULL;
 349
 350 fail:
 351    debug_printf("Creation failed: %s (%i)\n", strerror(-ret), ret);
 352    dec->base.destroy(&dec->base);
 353    return NULL;
 354 }
 355
 356 struct pipe_video_buffer *
 357 nvc0_video_buffer_create(struct pipe_context *pipe,
 358                          const struct pipe_video_buffer *templat)
 359 {
 360    return nouveau_vp3_video_buffer_create(
 361          pipe, templat, NVC0_RESOURCE_FLAG_VIDEO);
 362 }