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
dump_comm_bsp(struct comm * comm)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
nv98_decoder_bsp(struct nouveau_vp3_decoder * dec,union pipe_desc desc,struct nouveau_vp3_video_buffer * target,unsigned comm_seq,unsigned num_buffers,const void * const * data,const unsigned * num_bytes,unsigned * vp_caps,unsigned * is_ref,struct nouveau_vp3_video_buffer * refs[16])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_screen *screen = nouveau_screen(dec->base.context->screen);
43 struct nouveau_pushbuf *push = dec->pushbuf[0];
44 enum pipe_video_format codec = u_reduce_video_profile(dec->base.profile);
45 uint32_t bsp_addr, comm_addr, inter_addr;
46 uint32_t slice_size, bucket_size, ring_size, bsp_size;
47 uint32_t caps, i;
48 int ret;
49 struct nouveau_bo *bsp_bo = dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH];
50 struct nouveau_bo *inter_bo = dec->inter_bo[comm_seq & 1];
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 = ARRAY_SIZE(bo_refs);
60
61 if (!dec->bitplane_bo)
62 num_refs--;
63
64 bsp_size = NOUVEAU_VP3_BSP_RESERVED_SIZE;
65 for (i = 0; i < num_buffers; i++)
66 bsp_size += num_bytes[i];
67 bsp_size += 256; /* the 4 end markers */
68
69 if (!bsp_bo || bsp_size > bsp_bo->size) {
70 struct nouveau_bo *tmp_bo = NULL;
71
72 /* round up to the nearest mb */
73 bsp_size += (1 << 20) - 1;
74 bsp_size &= ~((1 << 20) - 1);
75
76 ret = nouveau_bo_new(dec->client->device, NOUVEAU_BO_VRAM, 0, bsp_size, NULL, &tmp_bo);
77 if (ret) {
78 debug_printf("reallocating bsp %u -> %u failed with %i\n",
79 bsp_bo ? (unsigned)bsp_bo->size : 0, bsp_size, ret);
80 return -1;
81 }
82 nouveau_bo_ref(NULL, &bsp_bo);
83 bo_refs[0].bo = dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH] = bsp_bo = tmp_bo;
84 }
85
86 if (!inter_bo || bsp_bo->size * 4 > inter_bo->size) {
87 struct nouveau_bo *tmp_bo = NULL;
88
89 ret = nouveau_bo_new(dec->client->device, NOUVEAU_BO_VRAM, 0, bsp_bo->size * 4, NULL, &tmp_bo);
90 if (ret) {
91 debug_printf("reallocating inter %u -> %u failed with %i\n",
92 inter_bo ? (unsigned)inter_bo->size : 0, (unsigned)bsp_bo->size * 4, ret);
93 return -1;
94 }
95 nouveau_bo_ref(NULL, &inter_bo);
96 bo_refs[1].bo = dec->inter_bo[comm_seq & 1] = inter_bo = tmp_bo;
97 }
98
99 ret = BO_MAP(screen, bsp_bo, NOUVEAU_BO_WR, dec->client);
100 if (ret) {
101 debug_printf("map failed: %i %s\n", ret, strerror(-ret));
102 return -1;
103 }
104
105 nouveau_vp3_bsp_begin(dec);
106 nouveau_vp3_bsp_next(dec, num_buffers, data, num_bytes);
107 caps = nouveau_vp3_bsp_end(dec, desc);
108
109 nouveau_vp3_vp_caps(dec, desc, target, comm_seq, vp_caps, is_ref, refs);
110
111 PUSH_SPACE_EX(push, 32, num_refs, 0);
112 PUSH_REFN(push, bo_refs, num_refs);
113
114 bsp_addr = bsp_bo->offset >> 8;
115 inter_addr = inter_bo->offset >> 8;
116
117 #if NOUVEAU_VP3_DEBUG_FENCE
118 memset(dec->comm, 0, 0x200);
119 comm_addr = (dec->fence_bo->offset + COMM_OFFSET) >> 8;
120 #else
121 comm_addr = bsp_addr + (COMM_OFFSET>>8);
122 #endif
123
124 BEGIN_NV04(push, SUBC_BSP(0x700), 5);
125 PUSH_DATA (push, caps); // 700 cmd
126 PUSH_DATA (push, bsp_addr + 1); // 704 strparm_bsp
127 PUSH_DATA (push, bsp_addr + 7); // 708 str addr
128 PUSH_DATA (push, comm_addr); // 70c comm
129 PUSH_DATA (push, comm_seq); // 710 seq
130
131 if (codec != PIPE_VIDEO_FORMAT_MPEG4_AVC) {
132 u32 bitplane_addr;
133 int mpeg12 = (codec == PIPE_VIDEO_FORMAT_MPEG12);
134
135 bitplane_addr = dec->bitplane_bo->offset >> 8;
136
137 nouveau_vp3_inter_sizes(dec, 1, &slice_size, &bucket_size, &ring_size);
138 BEGIN_NV04(push, SUBC_BSP(0x400), mpeg12 ? 5 : 7);
139 PUSH_DATA (push, bsp_addr); // 400 picparm addr
140 PUSH_DATA (push, inter_addr); // 404 interparm addr
141 PUSH_DATA (push, inter_addr + slice_size + bucket_size); // 408 interdata addr
142 PUSH_DATA (push, ring_size << 8); // 40c interdata_size
143 if (!mpeg12) {
144 PUSH_DATA (push, bitplane_addr); // 410 BITPLANE_DATA
145 PUSH_DATA (push, 0x400); // 414 BITPLANE_DATA_SIZE
146 }
147 PUSH_DATA (push, 0); // dma idx
148 } else {
149 nouveau_vp3_inter_sizes(dec, desc.h264->slice_count, &slice_size, &bucket_size, &ring_size);
150 BEGIN_NV04(push, SUBC_BSP(0x400), 8);
151 PUSH_DATA (push, bsp_addr); // 400 picparm addr
152 PUSH_DATA (push, inter_addr); // 404 interparm addr
153 PUSH_DATA (push, slice_size << 8); // 408 interparm size?
154 PUSH_DATA (push, inter_addr + slice_size + bucket_size); // 40c interdata addr
155 PUSH_DATA (push, ring_size << 8); // 410 interdata size
156 PUSH_DATA (push, inter_addr + slice_size); // 414 bucket?
157 PUSH_DATA (push, bucket_size << 8); // 418 bucket size? unshifted..
158 PUSH_DATA (push, 0); // 41c targets
159 // TODO: Double check 414 / 418 with nvidia trace
160 }
161
162 #if NOUVEAU_VP3_DEBUG_FENCE
163 BEGIN_NV04(push, SUBC_BSP(0x240), 3);
164 PUSH_DATAh(push, dec->fence_bo->offset);
165 PUSH_DATA (push, dec->fence_bo->offset);
166 PUSH_DATA (push, dec->fence_seq);
167
168 BEGIN_NV04(push, SUBC_BSP(0x300), 1);
169 PUSH_DATA (push, 1);
170 PUSH_KICK (push);
171
172 {
173 unsigned spin = 0;
174 do {
175 usleep(100);
176 if ((spin++ & 0xff) == 0xff) {
177 debug_printf("b%u: %u\n", dec->fence_seq, dec->fence_map[0]);
178 dump_comm_bsp(dec->comm);
179 }
180 } while (dec->fence_seq > dec->fence_map[0]);
181 }
182
183 dump_comm_bsp(dec->comm);
184 return dec->comm->status[comm_seq & 0xf];
185 #else
186 BEGIN_NV04(push, SUBC_BSP(0x300), 1);
187 PUSH_DATA (push, 0);
188 PUSH_KICK (push);
189 return 2;
190 #endif
191 }
192