Subversion Repositories Kolibri OS

Rev

Blame | Last modification | View Log | RSS feed

  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. #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. nv98_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.       struct nouveau_bo *tmp_bo = NULL;
  75.  
  76.       /* round up to the nearest mb */
  77.       bsp_size += (1 << 20) - 1;
  78.       bsp_size &= ~((1 << 20) - 1);
  79.  
  80.       ret = nouveau_bo_new(dec->bitplane_bo->device, NOUVEAU_BO_VRAM, 0, bsp_size, NULL, &tmp_bo);
  81.       if (ret) {
  82.          debug_printf("reallocating bsp %u -> %u failed with %i\n",
  83.                       bsp_bo ? (unsigned)bsp_bo->size : 0, bsp_size, ret);
  84.          return -1;
  85.       }
  86.       nouveau_bo_ref(NULL, &bsp_bo);
  87.       bo_refs[0].bo = dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH] = bsp_bo = tmp_bo;
  88.    }
  89.  
  90.    if (!inter_bo || bsp_bo->size * 4 > inter_bo->size) {
  91.       struct nouveau_bo *tmp_bo = NULL;
  92.  
  93.       ret = nouveau_bo_new(dec->bitplane_bo->device, NOUVEAU_BO_VRAM, 0, bsp_bo->size * 4, NULL, &tmp_bo);
  94.       if (ret) {
  95.          debug_printf("reallocating inter %u -> %u failed with %i\n",
  96.                       inter_bo ? (unsigned)inter_bo->size : 0, (unsigned)bsp_bo->size * 4, ret);
  97.          return -1;
  98.       }
  99.       nouveau_bo_ref(NULL, &inter_bo);
  100.       bo_refs[1].bo = dec->inter_bo[comm_seq & 1] = inter_bo = tmp_bo;
  101.    }
  102.  
  103.    ret = nouveau_bo_map(bsp_bo, NOUVEAU_BO_WR, dec->client);
  104.    if (ret) {
  105.       debug_printf("map failed: %i %s\n", ret, strerror(-ret));
  106.       return -1;
  107.    }
  108.  
  109.    caps = nouveau_vp3_bsp(dec, desc, target, comm_seq,
  110.                           num_buffers, data, num_bytes);
  111.  
  112.    nouveau_vp3_vp_caps(dec, desc, target, comm_seq, vp_caps, is_ref, refs);
  113.  
  114.    nouveau_pushbuf_space(push, 6 + (codec == PIPE_VIDEO_FORMAT_MPEG4_AVC ? 9 : 8) + fence_extra + 2, num_refs, 0);
  115.    nouveau_pushbuf_refn(push, bo_refs, num_refs);
  116.  
  117.    bsp_addr = bsp_bo->offset >> 8;
  118.    inter_addr = inter_bo->offset >> 8;
  119.  
  120. #if NOUVEAU_VP3_DEBUG_FENCE
  121.    memset(dec->comm, 0, 0x200);
  122.    comm_addr = (dec->fence_bo->offset + COMM_OFFSET) >> 8;
  123. #else
  124.    comm_addr = bsp_addr + (COMM_OFFSET>>8);
  125. #endif
  126.  
  127.    BEGIN_NV04(push, SUBC_BSP(0x700), 5);
  128.    PUSH_DATA (push, caps); // 700 cmd
  129.    PUSH_DATA (push, bsp_addr + 1); // 704 strparm_bsp
  130.    PUSH_DATA (push, bsp_addr + 7); // 708 str addr
  131.    PUSH_DATA (push, comm_addr); // 70c comm
  132.    PUSH_DATA (push, comm_seq); // 710 seq
  133.  
  134.    if (codec != PIPE_VIDEO_FORMAT_MPEG4_AVC) {
  135.       u32 bitplane_addr;
  136.       int mpeg12 = (codec == PIPE_VIDEO_FORMAT_MPEG12);
  137.  
  138.       bitplane_addr = dec->bitplane_bo->offset >> 8;
  139.  
  140.       nouveau_vp3_inter_sizes(dec, 1, &slice_size, &bucket_size, &ring_size);
  141.       BEGIN_NV04(push, SUBC_BSP(0x400), mpeg12 ? 5 : 7);
  142.       PUSH_DATA (push, bsp_addr); // 400 picparm addr
  143.       PUSH_DATA (push, inter_addr); // 404 interparm addr
  144.       PUSH_DATA (push, inter_addr + slice_size + bucket_size); // 408 interdata addr
  145.       PUSH_DATA (push, ring_size << 8); // 40c interdata_size
  146.       if (!mpeg12) {
  147.          PUSH_DATA (push, bitplane_addr); // 410 BITPLANE_DATA
  148.          PUSH_DATA (push, 0x400); // 414 BITPLANE_DATA_SIZE
  149.       }
  150.       PUSH_DATA (push, 0); // dma idx
  151.    } else {
  152.       nouveau_vp3_inter_sizes(dec, desc.h264->slice_count, &slice_size, &bucket_size, &ring_size);
  153.       BEGIN_NV04(push, SUBC_BSP(0x400), 8);
  154.       PUSH_DATA (push, bsp_addr); // 400 picparm addr
  155.       PUSH_DATA (push, inter_addr); // 404 interparm addr
  156.       PUSH_DATA (push, slice_size << 8); // 408 interparm size?
  157.       PUSH_DATA (push, inter_addr + slice_size + bucket_size); // 40c interdata addr
  158.       PUSH_DATA (push, ring_size << 8); // 410 interdata size
  159.       PUSH_DATA (push, inter_addr + slice_size); // 414 bucket?
  160.       PUSH_DATA (push, bucket_size << 8); // 418 bucket size? unshifted..
  161.       PUSH_DATA (push, 0); // 41c targets
  162.       // TODO: Double check 414 / 418 with nvidia trace
  163.    }
  164.  
  165. #if NOUVEAU_VP3_DEBUG_FENCE
  166.    BEGIN_NV04(push, SUBC_BSP(0x240), 3);
  167.    PUSH_DATAh(push, dec->fence_bo->offset);
  168.    PUSH_DATA (push, dec->fence_bo->offset);
  169.    PUSH_DATA (push, dec->fence_seq);
  170.  
  171.    BEGIN_NV04(push, SUBC_BSP(0x300), 1);
  172.    PUSH_DATA (push, 1);
  173.    PUSH_KICK (push);
  174.  
  175.    {
  176.       unsigned spin = 0;
  177.       do {
  178.          usleep(100);
  179.          if ((spin++ & 0xff) == 0xff) {
  180.             debug_printf("b%u: %u\n", dec->fence_seq, dec->fence_map[0]);
  181.             dump_comm_bsp(dec->comm);
  182.          }
  183.       } while (dec->fence_seq > dec->fence_map[0]);
  184.    }
  185.  
  186.    dump_comm_bsp(dec->comm);
  187.    return dec->comm->status[comm_seq & 0xf];
  188. #else
  189.    BEGIN_NV04(push, SUBC_BSP(0x300), 1);
  190.    PUSH_DATA (push, 0);
  191.    PUSH_KICK (push);
  192.    return 2;
  193. #endif
  194. }
  195.