src/gallium/drivers/nouveau/nvc0/nvc0_video_bsp.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 #if NOUVEAU_VP3_DEBUG_FENCE
  26 static void dump_comm_bsp(struct comm *comm)
  27 {
  28    unsigned idx = comm->bsp_cur_index & 0xf;
  29    debug_printf("Cur seq: %x, bsp byte ofs: %x\n", comm->bsp_cur_index, comm->byte_ofs);
  30    debug_printf("Status: %08x, pos: %08x\n", comm->status[idx], comm->pos[idx]);
  31 }
  32 #endif
  33
  34 unsigned
  35 nvc0_decoder_bsp(struct nouveau_vp3_decoder *dec, union pipe_desc desc,
  36                  struct nouveau_vp3_video_buffer *target,
  37                  unsigned comm_seq, unsigned num_buffers,
  38                  const void *const *data, const unsigned *num_bytes,
  39                  unsigned *vp_caps, unsigned *is_ref,
  40                  struct nouveau_vp3_video_buffer *refs[16])
  41 {
  42    struct nouveau_pushbuf *push = dec->pushbuf[0];
  43    enum pipe_video_format codec = u_reduce_video_profile(dec->base.profile);
  44    uint32_t bsp_addr, comm_addr, inter_addr;
  45    uint32_t slice_size, bucket_size, ring_size, bsp_size;
  46    uint32_t caps, i;
  47    int ret;
  48    struct nouveau_bo *bsp_bo = dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH];
  49    struct nouveau_bo *inter_bo = dec->inter_bo[comm_seq & 1];
  50    unsigned fence_extra = 0;
  51    struct nouveau_pushbuf_refn bo_refs[] = {
  52       { bsp_bo, NOUVEAU_BO_RD | NOUVEAU_BO_VRAM },
  53       { inter_bo, NOUVEAU_BO_WR | NOUVEAU_BO_VRAM },
  54 #if NOUVEAU_VP3_DEBUG_FENCE
  55       { dec->fence_bo, NOUVEAU_BO_WR | NOUVEAU_BO_GART },
  56 #endif
  57       { dec->bitplane_bo, NOUVEAU_BO_RDWR | NOUVEAU_BO_VRAM },
  58    };
  59    int num_refs = sizeof(bo_refs)/sizeof(*bo_refs);
  60
  61    if (!dec->bitplane_bo)
  62       num_refs--;
  63
  64 #if NOUVEAU_VP3_DEBUG_FENCE
  65    fence_extra = 4;
  66 #endif
  67
  68    bsp_size = NOUVEAU_VP3_BSP_RESERVED_SIZE;
  69    for (i = 0; i < num_buffers; i++)
  70       bsp_size += num_bytes[i];
  71    bsp_size += 256; /* the 4 end markers */
  72
  73    if (!bsp_bo || bsp_size > bsp_bo->size) {
  74       union nouveau_bo_config cfg;
  75       struct nouveau_bo *tmp_bo = NULL;
  76
  77       cfg.nvc0.tile_mode = 0x10;
  78       cfg.nvc0.memtype = 0xfe;
  79
  80       /* round up to the nearest mb */
  81       bsp_size += (1 << 20) - 1;
  82       bsp_size &= ~((1 << 20) - 1);
  83
  84       ret = nouveau_bo_new(dec->bitplane_bo->device, NOUVEAU_BO_VRAM, 0, bsp_size, &cfg, &tmp_bo);
  85       if (ret) {
  86          debug_printf("reallocating bsp %u -> %u failed with %i\n",
  87                       bsp_bo ? (unsigned)bsp_bo->size : 0, bsp_size, ret);
  88          return -1;
  89       }
  90       nouveau_bo_ref(NULL, &bsp_bo);
  91       bo_refs[0].bo = dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH] = bsp_bo = tmp_bo;
  92    }
  93
  94    if (!inter_bo || bsp_bo->size * 4 > inter_bo->size) {
  95       union nouveau_bo_config cfg;
  96       struct nouveau_bo *tmp_bo = NULL;
  97
  98       cfg.nvc0.tile_mode = 0x10;
  99       cfg.nvc0.memtype = 0xfe;
 100
 101       ret = nouveau_bo_new(dec->bitplane_bo->device, NOUVEAU_BO_VRAM, 0, bsp_bo->size * 4, &cfg, &tmp_bo);
 102       if (ret) {
 103          debug_printf("reallocating inter %u -> %u failed with %i\n",
 104                       inter_bo ? (unsigned)inter_bo->size : 0, (unsigned)bsp_bo->size * 4, ret);
 105          return -1;
 106       }
 107       nouveau_bo_ref(NULL, &inter_bo);
 108       bo_refs[1].bo = dec->inter_bo[comm_seq & 1] = inter_bo = tmp_bo;
 109    }
 110
 111    ret = nouveau_bo_map(bsp_bo, NOUVEAU_BO_WR, dec->client);
 112    if (ret) {
 113       debug_printf("map failed: %i %s\n", ret, strerror(-ret));
 114       return -1;
 115    }
 116
 117    caps = nouveau_vp3_bsp(dec, desc, target, comm_seq,
 118                           num_buffers, data, num_bytes);
 119
 120    nouveau_vp3_vp_caps(dec, desc, target, comm_seq, vp_caps, is_ref, refs);
 121
 122    nouveau_pushbuf_space(push, 6 + (codec == PIPE_VIDEO_FORMAT_MPEG4_AVC ? 9 : 7) + fence_extra + 2, num_refs, 0);
 123    nouveau_pushbuf_refn(push, bo_refs, num_refs);
 124
 125    bsp_addr = bsp_bo->offset >> 8;
 126    inter_addr = inter_bo->offset >> 8;
 127
 128 #if NOUVEAU_VP3_DEBUG_FENCE
 129    memset(dec->comm, 0, 0x200);
 130    comm_addr = (dec->fence_bo->offset + COMM_OFFSET) >> 8;
 131 #else
 132    comm_addr = bsp_addr + (COMM_OFFSET>>8);
 133 #endif
 134
 135    BEGIN_NVC0(push, SUBC_BSP(0x700), 5);
 136    PUSH_DATA (push, caps); // 700 cmd
 137    PUSH_DATA (push, bsp_addr + 1); // 704 strparm_bsp
 138    PUSH_DATA (push, bsp_addr + 7); // 708 str addr
 139    PUSH_DATA (push, comm_addr); // 70c comm
 140    PUSH_DATA (push, comm_seq); // 710 seq
 141
 142    if (codec != PIPE_VIDEO_FORMAT_MPEG4_AVC) {
 143       u32 bitplane_addr;
 144
 145       bitplane_addr = dec->bitplane_bo->offset >> 8;
 146
 147       nouveau_vp3_inter_sizes(dec, 1, &slice_size, &bucket_size, &ring_size);
 148       BEGIN_NVC0(push, SUBC_BSP(0x400), 6);
 149       PUSH_DATA (push, bsp_addr); // 400 picparm addr
 150       PUSH_DATA (push, inter_addr); // 404 interparm addr
 151       PUSH_DATA (push, inter_addr + slice_size + bucket_size); // 408 interdata addr
 152       PUSH_DATA (push, ring_size << 8); // 40c interdata_size
 153       PUSH_DATA (push, bitplane_addr); // 410 BITPLANE_DATA
 154       PUSH_DATA (push, 0x400); // 414 BITPLANE_DATA_SIZE
 155    } else {
 156       nouveau_vp3_inter_sizes(dec, desc.h264->slice_count, &slice_size, &bucket_size, &ring_size);
 157       BEGIN_NVC0(push, SUBC_BSP(0x400), 8);
 158       PUSH_DATA (push, bsp_addr); // 400 picparm addr
 159       PUSH_DATA (push, inter_addr); // 404 interparm addr
 160       PUSH_DATA (push, slice_size << 8); // 408 interparm size?
 161       PUSH_DATA (push, inter_addr + slice_size + bucket_size); // 40c interdata addr
 162       PUSH_DATA (push, ring_size << 8); // 410 interdata size
 163       PUSH_DATA (push, inter_addr + slice_size); // 414 bucket?
 164       PUSH_DATA (push, bucket_size << 8); // 418 bucket size? unshifted..
 165       PUSH_DATA (push, 0); // 41c targets
 166       // TODO: Double check 414 / 418 with nvidia trace
 167    }
 168
 169 #if NOUVEAU_VP3_DEBUG_FENCE
 170    BEGIN_NVC0(push, SUBC_BSP(0x240), 3);
 171    PUSH_DATAh(push, dec->fence_bo->offset);
 172    PUSH_DATA (push, dec->fence_bo->offset);
 173    PUSH_DATA (push, dec->fence_seq);
 174
 175    BEGIN_NVC0(push, SUBC_BSP(0x300), 1);
 176    PUSH_DATA (push, 1);
 177    PUSH_KICK (push);
 178
 179    {
 180       unsigned spin = 0;
 181       do {
 182          usleep(100);
 183          if ((spin++ & 0xff) == 0xff) {
 184             debug_printf("b%u: %u\n", dec->fence_seq, dec->fence_map[0]);
 185             dump_comm_bsp(dec->comm);
 186          }
 187       } while (dec->fence_seq > dec->fence_map[0]);
 188    }
 189
 190    dump_comm_bsp(dec->comm);
 191    return dec->comm->status[comm_seq & 0xf];
 192 #else
 193    BEGIN_NVC0(push, SUBC_BSP(0x300), 1);
 194    PUSH_DATA (push, 0);
 195    PUSH_KICK (push);
 196    return 2;
 197 #endif
 198 }