Home | History | Annotate | Download | only in nv50
      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    struct nouveau_pushbuf_refn bo_refs[] = {
     51       { bsp_bo, NOUVEAU_BO_RD | NOUVEAU_BO_VRAM },
     52       { inter_bo, NOUVEAU_BO_WR | NOUVEAU_BO_VRAM },
     53 #if NOUVEAU_VP3_DEBUG_FENCE
     54       { dec->fence_bo, NOUVEAU_BO_WR | NOUVEAU_BO_GART },
     55 #endif
     56       { dec->bitplane_bo, NOUVEAU_BO_RDWR | NOUVEAU_BO_VRAM },
     57    };
     58    int num_refs = ARRAY_SIZE(bo_refs);
     59 
     60    if (!dec->bitplane_bo)
     61       num_refs--;
     62 
     63    bsp_size = NOUVEAU_VP3_BSP_RESERVED_SIZE;
     64    for (i = 0; i < num_buffers; i++)
     65       bsp_size += num_bytes[i];
     66    bsp_size += 256; /* the 4 end markers */
     67 
     68    if (!bsp_bo || bsp_size > bsp_bo->size) {
     69       struct nouveau_bo *tmp_bo = NULL;
     70 
     71       /* round up to the nearest mb */
     72       bsp_size += (1 << 20) - 1;
     73       bsp_size &= ~((1 << 20) - 1);
     74 
     75       ret = nouveau_bo_new(dec->client->device, NOUVEAU_BO_VRAM, 0, bsp_size, NULL, &tmp_bo);
     76       if (ret) {
     77          debug_printf("reallocating bsp %u -> %u failed with %i\n",
     78                       bsp_bo ? (unsigned)bsp_bo->size : 0, bsp_size, ret);
     79          return -1;
     80       }
     81       nouveau_bo_ref(NULL, &bsp_bo);
     82       bo_refs[0].bo = dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH] = bsp_bo = tmp_bo;
     83    }
     84 
     85    if (!inter_bo || bsp_bo->size * 4 > inter_bo->size) {
     86       struct nouveau_bo *tmp_bo = NULL;
     87 
     88       ret = nouveau_bo_new(dec->client->device, NOUVEAU_BO_VRAM, 0, bsp_bo->size * 4, NULL, &tmp_bo);
     89       if (ret) {
     90          debug_printf("reallocating inter %u -> %u failed with %i\n",
     91                       inter_bo ? (unsigned)inter_bo->size : 0, (unsigned)bsp_bo->size * 4, ret);
     92          return -1;
     93       }
     94       nouveau_bo_ref(NULL, &inter_bo);
     95       bo_refs[1].bo = dec->inter_bo[comm_seq & 1] = inter_bo = tmp_bo;
     96    }
     97 
     98    ret = nouveau_bo_map(bsp_bo, NOUVEAU_BO_WR, dec->client);
     99    if (ret) {
    100       debug_printf("map failed: %i %s\n", ret, strerror(-ret));
    101       return -1;
    102    }
    103 
    104    nouveau_vp3_bsp_begin(dec);
    105    nouveau_vp3_bsp_next(dec, num_buffers, data, num_bytes);
    106    caps = nouveau_vp3_bsp_end(dec, desc);
    107 
    108    nouveau_vp3_vp_caps(dec, desc, target, comm_seq, vp_caps, is_ref, refs);
    109 
    110    nouveau_pushbuf_space(push, 32, num_refs, 0);
    111    nouveau_pushbuf_refn(push, bo_refs, num_refs);
    112 
    113    bsp_addr = bsp_bo->offset >> 8;
    114    inter_addr = inter_bo->offset >> 8;
    115 
    116 #if NOUVEAU_VP3_DEBUG_FENCE
    117    memset(dec->comm, 0, 0x200);
    118    comm_addr = (dec->fence_bo->offset + COMM_OFFSET) >> 8;
    119 #else
    120    comm_addr = bsp_addr + (COMM_OFFSET>>8);
    121 #endif
    122 
    123    BEGIN_NV04(push, SUBC_BSP(0x700), 5);
    124    PUSH_DATA (push, caps); // 700 cmd
    125    PUSH_DATA (push, bsp_addr + 1); // 704 strparm_bsp
    126    PUSH_DATA (push, bsp_addr + 7); // 708 str addr
    127    PUSH_DATA (push, comm_addr); // 70c comm
    128    PUSH_DATA (push, comm_seq); // 710 seq
    129 
    130    if (codec != PIPE_VIDEO_FORMAT_MPEG4_AVC) {
    131       u32 bitplane_addr;
    132       int mpeg12 = (codec == PIPE_VIDEO_FORMAT_MPEG12);
    133 
    134       bitplane_addr = dec->bitplane_bo->offset >> 8;
    135 
    136       nouveau_vp3_inter_sizes(dec, 1, &slice_size, &bucket_size, &ring_size);
    137       BEGIN_NV04(push, SUBC_BSP(0x400), mpeg12 ? 5 : 7);
    138       PUSH_DATA (push, bsp_addr); // 400 picparm addr
    139       PUSH_DATA (push, inter_addr); // 404 interparm addr
    140       PUSH_DATA (push, inter_addr + slice_size + bucket_size); // 408 interdata addr
    141       PUSH_DATA (push, ring_size << 8); // 40c interdata_size
    142       if (!mpeg12) {
    143          PUSH_DATA (push, bitplane_addr); // 410 BITPLANE_DATA
    144          PUSH_DATA (push, 0x400); // 414 BITPLANE_DATA_SIZE
    145       }
    146       PUSH_DATA (push, 0); // dma idx
    147    } else {
    148       nouveau_vp3_inter_sizes(dec, desc.h264->slice_count, &slice_size, &bucket_size, &ring_size);
    149       BEGIN_NV04(push, SUBC_BSP(0x400), 8);
    150       PUSH_DATA (push, bsp_addr); // 400 picparm addr
    151       PUSH_DATA (push, inter_addr); // 404 interparm addr
    152       PUSH_DATA (push, slice_size << 8); // 408 interparm size?
    153       PUSH_DATA (push, inter_addr + slice_size + bucket_size); // 40c interdata addr
    154       PUSH_DATA (push, ring_size << 8); // 410 interdata size
    155       PUSH_DATA (push, inter_addr + slice_size); // 414 bucket?
    156       PUSH_DATA (push, bucket_size << 8); // 418 bucket size? unshifted..
    157       PUSH_DATA (push, 0); // 41c targets
    158       // TODO: Double check 414 / 418 with nvidia trace
    159    }
    160 
    161 #if NOUVEAU_VP3_DEBUG_FENCE
    162    BEGIN_NV04(push, SUBC_BSP(0x240), 3);
    163    PUSH_DATAh(push, dec->fence_bo->offset);
    164    PUSH_DATA (push, dec->fence_bo->offset);
    165    PUSH_DATA (push, dec->fence_seq);
    166 
    167    BEGIN_NV04(push, SUBC_BSP(0x300), 1);
    168    PUSH_DATA (push, 1);
    169    PUSH_KICK (push);
    170 
    171    {
    172       unsigned spin = 0;
    173       do {
    174          usleep(100);
    175          if ((spin++ & 0xff) == 0xff) {
    176             debug_printf("b%u: %u\n", dec->fence_seq, dec->fence_map[0]);
    177             dump_comm_bsp(dec->comm);
    178          }
    179       } while (dec->fence_seq > dec->fence_map[0]);
    180    }
    181 
    182    dump_comm_bsp(dec->comm);
    183    return dec->comm->status[comm_seq & 0xf];
    184 #else
    185    BEGIN_NV04(push, SUBC_BSP(0x300), 1);
    186    PUSH_DATA (push, 0);
    187    PUSH_KICK (push);
    188    return 2;
    189 #endif
    190 }
    191