src/gallium/drivers/nouveau/nv50/nv98_video.c

   1 /*
   2  * Copyright 2011-2013 Maarten Lankhorst, Ilia Mirkin
   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 "nv50/nv98_video.h"
  24
  25 #include "util/u_sampler.h"
  26 #include "util/u_format.h"
  27
  28 static void
  29 nv98_decoder_decode_bitstream(struct pipe_video_codec *decoder,
  30                               struct pipe_video_buffer *video_target,
  31                               struct pipe_picture_desc *picture,
  32                               unsigned num_buffers,
  33                               const void *const *data,
  34                               const unsigned *num_bytes)
  35 {
  36    struct nouveau_vp3_decoder *dec = (struct nouveau_vp3_decoder *)decoder;
  37    struct nouveau_vp3_video_buffer *target = (struct nouveau_vp3_video_buffer *)video_target;
  38    uint32_t comm_seq = ++dec->fence_seq;
  39    union pipe_desc desc;
  40
  41    unsigned vp_caps, is_ref, ret;
  42    struct nouveau_vp3_video_buffer *refs[16] = {};
  43
  44    desc.base = picture;
  45
  46    assert(target->base.buffer_format == PIPE_FORMAT_NV12);
  47
  48    ret = nv98_decoder_bsp(dec, desc, target, comm_seq,
  49                           num_buffers, data, num_bytes,
  50                           &vp_caps, &is_ref, refs);
  51
  52    /* did we decode bitstream correctly? */
  53    assert(ret == 2);
  54
  55    nv98_decoder_vp(dec, desc, target, comm_seq, vp_caps, is_ref, refs);
  56    nv98_decoder_ppp(dec, desc, target, comm_seq);
  57 }
  58
  59 struct pipe_video_codec *
  60 nv98_create_decoder(struct pipe_context *context,
  61                     const struct pipe_video_codec *templ)
  62 {
  63    struct nouveau_screen *screen = &((struct nv50_context *)context)->screen->base;
  64    struct nouveau_vp3_decoder *dec;
  65    struct nouveau_pushbuf **push;
  66    struct nv04_fifo nv04_data = {.vram = 0xbeef0201, .gart = 0xbeef0202};
  67    union nouveau_bo_config cfg;
  68
  69    cfg.nv50.tile_mode = 0x20;
  70    cfg.nv50.memtype = 0x70;
  71
  72    int ret, i;
  73    uint32_t codec = 1, ppp_codec = 3;
  74    uint32_t timeout;
  75    u32 tmp_size = 0;
  76
  77    if (getenv("XVMC_VL"))
  78        return vl_create_decoder(context, templ);
  79
  80    if (templ->entrypoint != PIPE_VIDEO_ENTRYPOINT_BITSTREAM) {
  81       debug_printf("%x\n", templ->entrypoint);
  82       return NULL;
  83    }
  84
  85    dec = CALLOC_STRUCT(nouveau_vp3_decoder);
  86    if (!dec)
  87       return NULL;
  88    dec->client = screen->client;
  89    dec->base = *templ;
  90    nouveau_vp3_decoder_init_common(&dec->base);
  91
  92    dec->bsp_idx = 5;
  93    dec->vp_idx = 6;
  94    dec->ppp_idx = 7;
  95
  96    ret = nouveau_object_new(&screen->device->object, 0,
  97                             NOUVEAU_FIFO_CHANNEL_CLASS,
  98                             &nv04_data, sizeof(nv04_data), &dec->channel[0]);
  99
 100    if (!ret)
 101       ret = nouveau_pushbuf_new(screen->client, dec->channel[0], 4,
 102                                 32 * 1024, true, &dec->pushbuf[0]);
 103
 104    for (i = 1; i < 3; ++i) {
 105       dec->channel[i] = dec->channel[0];
 106       dec->pushbuf[i] = dec->pushbuf[0];
 107    }
 108    push = dec->pushbuf;
 109
 110    if (!ret)
 111       ret = nouveau_object_new(dec->channel[0], 0x390b1, 0x85b1, NULL, 0, &dec->bsp);
 112    if (!ret)
 113       ret = nouveau_object_new(dec->channel[1], 0x190b2, 0x85b2, NULL, 0, &dec->vp);
 114    if (!ret)
 115       ret = nouveau_object_new(dec->channel[2], 0x290b3, 0x85b3, NULL, 0, &dec->ppp);
 116    if (ret)
 117       goto fail;
 118
 119    BEGIN_NV04(push[0], SUBC_BSP(NV01_SUBCHAN_OBJECT), 1);
 120    PUSH_DATA (push[0], dec->bsp->handle);
 121
 122    BEGIN_NV04(push[0], SUBC_BSP(0x180), 5);
 123    for (i = 0; i < 5; i++)
 124       PUSH_DATA (push[0], nv04_data.vram);
 125
 126    BEGIN_NV04(push[1], SUBC_VP(NV01_SUBCHAN_OBJECT), 1);
 127    PUSH_DATA (push[1], dec->vp->handle);
 128
 129    BEGIN_NV04(push[1], SUBC_VP(0x180), 6);
 130    for (i = 0; i < 6; i++)
 131       PUSH_DATA (push[1], nv04_data.vram);
 132
 133    BEGIN_NV04(push[2], SUBC_PPP(NV01_SUBCHAN_OBJECT), 1);
 134    PUSH_DATA (push[2], dec->ppp->handle);
 135
 136    BEGIN_NV04(push[2], SUBC_PPP(0x180), 5);
 137    for (i = 0; i < 5; i++)
 138       PUSH_DATA (push[2], nv04_data.vram);
 139
 140    dec->base.context = context;
 141    dec->base.decode_bitstream = nv98_decoder_decode_bitstream;
 142
 143    for (i = 0; i < NOUVEAU_VP3_VIDEO_QDEPTH && !ret; ++i)
 144       ret = nouveau_bo_new(screen->device, NOUVEAU_BO_VRAM,
 145                            0, 1 << 20, NULL, &dec->bsp_bo[i]);
 146    if (!ret)
 147       ret = nouveau_bo_new(screen->device, NOUVEAU_BO_VRAM,
 148                            0x100, 4 << 20, NULL, &dec->inter_bo[0]);
 149    if (!ret)
 150       nouveau_bo_ref(dec->inter_bo[0], &dec->inter_bo[1]);
 151    if (ret)
 152       goto fail;
 153
 154    switch (u_reduce_video_profile(templ->profile)) {
 155    case PIPE_VIDEO_FORMAT_MPEG12: {
 156       codec = 1;
 157       assert(templ->max_references <= 2);
 158       break;
 159    }
 160    case PIPE_VIDEO_FORMAT_MPEG4: {
 161       codec = 4;
 162       tmp_size = mb(templ->height)*16 * mb(templ->width)*16;
 163       assert(templ->max_references <= 2);
 164       break;
 165    }
 166    case PIPE_VIDEO_FORMAT_VC1: {
 167       ppp_codec = codec = 2;
 168       tmp_size = mb(templ->height)*16 * mb(templ->width)*16;
 169       assert(templ->max_references <= 2);
 170       break;
 171    }
 172    case PIPE_VIDEO_FORMAT_MPEG4_AVC: {
 173       codec = 3;
 174       dec->tmp_stride = 16 * mb_half(templ->width) * nouveau_vp3_video_align(templ->height) * 3 / 2;
 175       tmp_size = dec->tmp_stride * (templ->max_references + 1);
 176       assert(templ->max_references <= 16);
 177       break;
 178    }
 179    default:
 180       fprintf(stderr, "invalid codec\n");
 181       goto fail;
 182    }
 183
 184    ret = nouveau_bo_new(screen->device, NOUVEAU_BO_VRAM, 0,
 185                            0x4000, NULL, &dec->fw_bo);
 186    if (ret)
 187       goto fail;
 188
 189    ret = nouveau_vp3_load_firmware(dec, templ->profile, screen->device->chipset);
 190    if (ret)
 191       goto fw_fail;
 192
 193    if (codec != 3) {
 194       ret = nouveau_bo_new(screen->device, NOUVEAU_BO_VRAM, 0,
 195                            0x400, NULL, &dec->bitplane_bo);
 196       if (ret)
 197          goto fail;
 198    }
 199
 200    dec->ref_stride = mb(templ->width)*16 * (mb_half(templ->height)*32 + nouveau_vp3_video_align(templ->height)/2);
 201    ret = nouveau_bo_new(screen->device, NOUVEAU_BO_VRAM, 0,
 202                         dec->ref_stride * (templ->max_references+2) + tmp_size,
 203                         &cfg, &dec->ref_bo);
 204    if (ret)
 205       goto fail;
 206
 207    timeout = 0;
 208
 209    BEGIN_NV04(push[0], SUBC_BSP(0x200), 2);
 210    PUSH_DATA (push[0], codec);
 211    PUSH_DATA (push[0], timeout);
 212
 213    BEGIN_NV04(push[1], SUBC_VP(0x200), 2);
 214    PUSH_DATA (push[1], codec);
 215    PUSH_DATA (push[1], timeout);
 216
 217    BEGIN_NV04(push[2], SUBC_PPP(0x200), 2);
 218    PUSH_DATA (push[2], ppp_codec);
 219    PUSH_DATA (push[2], timeout);
 220
 221    ++dec->fence_seq;
 222
 223 #if NOUVEAU_VP3_DEBUG_FENCE
 224    ret = nouveau_bo_new(screen->device, NOUVEAU_BO_GART|NOUVEAU_BO_MAP,
 225                         0, 0x1000, NULL, &dec->fence_bo);
 226    if (ret)
 227       goto fail;
 228
 229    nouveau_bo_map(dec->fence_bo, NOUVEAU_BO_RDWR, screen->client);
 230    dec->fence_map = dec->fence_bo->map;
 231    dec->fence_map[0] = dec->fence_map[4] = dec->fence_map[8] = 0;
 232    dec->comm = (struct comm *)(dec->fence_map + (COMM_OFFSET/sizeof(*dec->fence_map)));
 233
 234    /* So lets test if the fence is working? */
 235    nouveau_pushbuf_space(push[0], 6, 1, 0);
 236    PUSH_REFN (push[0], dec->fence_bo, NOUVEAU_BO_GART|NOUVEAU_BO_RDWR);
 237    BEGIN_NV04(push[0], SUBC_BSP(0x240), 3);
 238    PUSH_DATAh(push[0], dec->fence_bo->offset);
 239    PUSH_DATA (push[0], dec->fence_bo->offset);
 240    PUSH_DATA (push[0], dec->fence_seq);
 241
 242    BEGIN_NV04(push[0], SUBC_BSP(0x304), 1);
 243    PUSH_DATA (push[0], 0);
 244    PUSH_KICK (push[0]);
 245
 246    nouveau_pushbuf_space(push[1], 6, 1, 0);
 247    PUSH_REFN (push[1], dec->fence_bo, NOUVEAU_BO_GART|NOUVEAU_BO_RDWR);
 248    BEGIN_NV04(push[1], SUBC_VP(0x240), 3);
 249    PUSH_DATAh(push[1], (dec->fence_bo->offset + 0x10));
 250    PUSH_DATA (push[1], (dec->fence_bo->offset + 0x10));
 251    PUSH_DATA (push[1], dec->fence_seq);
 252
 253    BEGIN_NV04(push[1], SUBC_VP(0x304), 1);
 254    PUSH_DATA (push[1], 0);
 255    PUSH_KICK (push[1]);
 256
 257    nouveau_pushbuf_space(push[2], 6, 1, 0);
 258    PUSH_REFN (push[2], dec->fence_bo, NOUVEAU_BO_GART|NOUVEAU_BO_RDWR);
 259    BEGIN_NV04(push[2], SUBC_PPP(0x240), 3);
 260    PUSH_DATAh(push[2], (dec->fence_bo->offset + 0x20));
 261    PUSH_DATA (push[2], (dec->fence_bo->offset + 0x20));
 262    PUSH_DATA (push[2], dec->fence_seq);
 263
 264    BEGIN_NV04(push[2], SUBC_PPP(0x304), 1);
 265    PUSH_DATA (push[2], 0);
 266    PUSH_KICK (push[2]);
 267
 268    usleep(100);
 269    while (dec->fence_seq > dec->fence_map[0] ||
 270           dec->fence_seq > dec->fence_map[4] ||
 271           dec->fence_seq > dec->fence_map[8]) {
 272       debug_printf("%u: %u %u %u\n", dec->fence_seq, dec->fence_map[0], dec->fence_map[4], dec->fence_map[8]);
 273       usleep(100);
 274    }
 275    debug_printf("%u: %u %u %u\n", dec->fence_seq, dec->fence_map[0], dec->fence_map[4], dec->fence_map[8]);
 276 #endif
 277
 278    return &dec->base;
 279
 280 fw_fail:
 281    debug_printf("Cannot create decoder without firmware..\n");
 282    dec->base.destroy(&dec->base);
 283    return NULL;
 284
 285 fail:
 286    debug_printf("Creation failed: %s (%i)\n", strerror(-ret), ret);
 287    dec->base.destroy(&dec->base);
 288    return NULL;
 289 }
 290
 291 struct pipe_video_buffer *
 292 nv98_video_buffer_create(struct pipe_context *pipe,
 293                          const struct pipe_video_buffer *templat)
 294 {
 295    return nouveau_vp3_video_buffer_create(
 296          pipe, templat, NV50_RESOURCE_FLAG_VIDEO);
 297 }