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