src/gallium/drivers/nouveau/nv50/nv98_video_vp.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 #include <sys/mman.h>
  25
  26 #if NOUVEAU_VP3_DEBUG_FENCE
  27 static void dump_comm_vp(struct nouveau_vp3_decoder *dec, struct comm *comm, u32 comm_seq,
  28                          struct nouveau_bo *inter_bo, unsigned slice_size)
  29 {
  30    unsigned i, idx = comm->pvp_cur_index & 0xf;
  31    debug_printf("Status: %08x, stage: %08x\n", comm->status_vp[idx], comm->pvp_stage);
  32 #if 0
  33    debug_printf("Acked byte ofs: %x, bsp byte ofs: %x\n", comm->acked_byte_ofs, comm->byte_ofs);
  34    debug_printf("Irq/parse indexes: %i %i\n", comm->irq_index, comm->parse_endpos_index);
  35
  36    for (i = 0; i != comm->irq_index; ++i)
  37       debug_printf("irq[%i] = { @ %08x -> %04x }\n", i, comm->irq_pos[i], comm->irq_470[i]);
  38    for (i = 0; i != comm->parse_endpos_index; ++i)
  39       debug_printf("parse_endpos[%i] = { @ %08x}\n", i, comm->parse_endpos[i]);
  40 #endif
  41    debug_printf("mb_y = %u\n", comm->mb_y[idx]);
  42    if (comm->status_vp[idx] <= 1)
  43       return;
  44
  45    if ((comm->pvp_stage & 0xff) != 0xff) {
  46       unsigned *map;
  47       int ret = nouveau_bo_map(inter_bo, NOUVEAU_BO_RD|NOUVEAU_BO_NOBLOCK, dec->client);
  48       assert(ret >= 0);
  49       map = inter_bo->map;
  50       for (i = 0; i < comm->byte_ofs + slice_size; i += 0x10) {
  51          debug_printf("%05x: %08x %08x %08x %08x\n", i, map[i/4], map[i/4+1], map[i/4+2], map[i/4+3]);
  52       }
  53       munmap(inter_bo->map, inter_bo->size);
  54       inter_bo->map = NULL;
  55    }
  56    assert((comm->pvp_stage & 0xff) == 0xff);
  57 }
  58 #endif
  59
  60 static void
  61 nv98_decoder_kick_ref(struct nouveau_vp3_decoder *dec, struct nouveau_vp3_video_buffer *target)
  62 {
  63    dec->refs[target->valid_ref].last_used = 0;
  64 //   debug_printf("Unreffed %p\n", target);
  65 }
  66
  67 void
  68 nv98_decoder_vp(struct nouveau_vp3_decoder *dec, union pipe_desc desc,
  69                 struct nouveau_vp3_video_buffer *target, unsigned comm_seq,
  70                 unsigned caps, unsigned is_ref,
  71                 struct nouveau_vp3_video_buffer *refs[16])
  72 {
  73    struct nouveau_pushbuf *push = dec->pushbuf[1];
  74    uint32_t bsp_addr, comm_addr, inter_addr, ucode_addr, pic_addr[17], last_addr, null_addr;
  75    uint32_t slice_size, bucket_size, ring_size, i;
  76    enum pipe_video_format codec = u_reduce_video_profile(dec->base.profile);
  77    struct nouveau_bo *bsp_bo = dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH];
  78    struct nouveau_bo *inter_bo = dec->inter_bo[comm_seq & 1];
  79    u32 fence_extra = 0, codec_extra = 0;
  80    struct nouveau_pushbuf_refn bo_refs[] = {
  81       { inter_bo, NOUVEAU_BO_WR | NOUVEAU_BO_VRAM },
  82       { dec->ref_bo, NOUVEAU_BO_WR | NOUVEAU_BO_VRAM },
  83       { bsp_bo, NOUVEAU_BO_RD | NOUVEAU_BO_VRAM },
  84 #if NOUVEAU_VP3_DEBUG_FENCE
  85       { dec->fence_bo, NOUVEAU_BO_WR | NOUVEAU_BO_GART },
  86 #endif
  87       { dec->fw_bo, NOUVEAU_BO_RD | NOUVEAU_BO_VRAM },
  88    };
  89    int num_refs = ARRAY_SIZE(bo_refs) - !dec->fw_bo;
  90
  91 #if NOUVEAU_VP3_DEBUG_FENCE
  92    fence_extra = 4;
  93 #endif
  94
  95    if (codec == PIPE_VIDEO_FORMAT_MPEG4_AVC) {
  96       nouveau_vp3_inter_sizes(dec, desc.h264->slice_count, &slice_size, &bucket_size, &ring_size);
  97       codec_extra += 2;
  98    } else
  99       nouveau_vp3_inter_sizes(dec, 1, &slice_size, &bucket_size, &ring_size);
 100
 101    if (dec->base.max_references > 2)
 102       codec_extra += 1 + (dec->base.max_references - 2);
 103
 104    pic_addr[16] = nouveau_vp3_video_addr(dec, target) >> 8;
 105    last_addr = null_addr = nouveau_vp3_video_addr(dec, NULL) >> 8;
 106
 107    for (i = 0; i < dec->base.max_references; ++i) {
 108       if (!refs[i])
 109          pic_addr[i] = last_addr;
 110       else if (dec->refs[refs[i]->valid_ref].vidbuf == refs[i])
 111          last_addr = pic_addr[i] = nouveau_vp3_video_addr(dec, refs[i]) >> 8;
 112       else
 113          pic_addr[i] = null_addr;
 114    }
 115    if (!is_ref && (dec->refs[target->valid_ref].decoded_top && dec->refs[target->valid_ref].decoded_bottom))
 116       nv98_decoder_kick_ref(dec, target);
 117
 118    nouveau_pushbuf_space(push, 8 + 3 * (codec != PIPE_VIDEO_FORMAT_MPEG12) +
 119               6 + codec_extra + fence_extra + 2, num_refs, 0);
 120
 121    nouveau_pushbuf_refn(push, bo_refs, num_refs);
 122
 123    bsp_addr = bsp_bo->offset >> 8;
 124 #if NOUVEAU_VP3_DEBUG_FENCE
 125    comm_addr = (dec->fence_bo->offset + COMM_OFFSET)>>8;
 126 #else
 127    comm_addr = bsp_addr + (COMM_OFFSET>>8);
 128 #endif
 129    inter_addr = inter_bo->offset >> 8;
 130    if (dec->fw_bo)
 131       ucode_addr = dec->fw_bo->offset >> 8;
 132    else
 133       ucode_addr = 0;
 134
 135    BEGIN_NV04(push, SUBC_VP(0x700), 7);
 136    PUSH_DATA (push, caps); // 700
 137    PUSH_DATA (push, comm_seq); // 704
 138    PUSH_DATA (push, 0); // 708 fuc targets, ignored for nv98
 139    PUSH_DATA (push, dec->fw_sizes); // 70c
 140    PUSH_DATA (push, bsp_addr+(VP_OFFSET>>8)); // 710 picparm_addr
 141    PUSH_DATA (push, inter_addr); // 714 inter_parm
 142    PUSH_DATA (push, inter_addr + slice_size + bucket_size); // 718 inter_data_ofs
 143
 144    if (bucket_size) {
 145       uint64_t tmpimg_addr = dec->ref_bo->offset + dec->ref_stride * (dec->base.max_references+2);
 146
 147       BEGIN_NV04(push, SUBC_VP(0x71c), 2);
 148       PUSH_DATA (push, tmpimg_addr >> 8); // 71c
 149       PUSH_DATA (push, inter_addr + slice_size); // 720 bucket_ofs
 150    }
 151
 152    BEGIN_NV04(push, SUBC_VP(0x724), 5);
 153    PUSH_DATA (push, comm_addr); // 724
 154    PUSH_DATA (push, ucode_addr); // 728
 155    PUSH_DATA (push, pic_addr[16]); // 734
 156    PUSH_DATA (push, pic_addr[0]); // 72c
 157    PUSH_DATA (push, pic_addr[1]); // 730
 158
 159    if (dec->base.max_references > 2) {
 160       int i;
 161
 162       BEGIN_NV04(push, SUBC_VP(0x400), dec->base.max_references - 2);
 163       for (i = 2; i < dec->base.max_references; ++i) {
 164          assert(0x400 + (i - 2) * 4 < 0x438);
 165          PUSH_DATA (push, pic_addr[i]);
 166       }
 167    }
 168
 169    if (codec == PIPE_VIDEO_FORMAT_MPEG4_AVC) {
 170       BEGIN_NV04(push, SUBC_VP(0x438), 1);
 171       PUSH_DATA (push, desc.h264->slice_count);
 172    }
 173
 174    //debug_printf("Decoding %08lx with %08lx and %08lx\n", pic_addr[16], pic_addr[0], pic_addr[1]);
 175
 176 #if NOUVEAU_VP3_DEBUG_FENCE
 177    BEGIN_NV04(push, SUBC_VP(0x240), 3);
 178    PUSH_DATAh(push, (dec->fence_bo->offset + 0x10));
 179    PUSH_DATA (push, (dec->fence_bo->offset + 0x10));
 180    PUSH_DATA (push, dec->fence_seq);
 181
 182    BEGIN_NV04(push, SUBC_VP(0x300), 1);
 183    PUSH_DATA (push, 1);
 184    PUSH_KICK(push);
 185
 186    {
 187       unsigned spin = 0;
 188       do {
 189          usleep(100);
 190          if ((spin++ & 0xff) == 0xff) {
 191             debug_printf("v%u: %u\n", dec->fence_seq, dec->fence_map[4]);
 192             dump_comm_vp(dec, dec->comm, comm_seq, inter_bo, slice_size << 8);
 193          }
 194       } while (dec->fence_seq > dec->fence_map[4]);
 195    }
 196    dump_comm_vp(dec, dec->comm, comm_seq, inter_bo, slice_size << 8);
 197 #else
 198    BEGIN_NV04(push, SUBC_VP(0x300), 1);
 199    PUSH_DATA (push, 0);
 200    PUSH_KICK (push);
 201 #endif
 202 }