Rev 4358 | Go to most recent revision | Details | Compare with Previous | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
4358 | Serge | 1 | /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */ |
2 | |||
3 | /* |
||
4 | * Copyright (C) 2013 Rob Clark |
||
5 | * |
||
6 | * Permission is hereby granted, free of charge, to any person obtaining a |
||
7 | * copy of this software and associated documentation files (the "Software"), |
||
8 | * to deal in the Software without restriction, including without limitation |
||
9 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
||
10 | * and/or sell copies of the Software, and to permit persons to whom the |
||
11 | * Software is furnished to do so, subject to the following conditions: |
||
12 | * |
||
13 | * The above copyright notice and this permission notice (including the next |
||
14 | * paragraph) shall be included in all copies or substantial portions of the |
||
15 | * Software. |
||
16 | * |
||
17 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
||
18 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
||
19 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
||
20 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
||
21 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
||
22 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
||
23 | * SOFTWARE. |
||
24 | * |
||
25 | * Authors: |
||
26 | * Rob Clark |
||
27 | */ |
||
28 | |||
29 | #include "pipe/p_state.h" |
||
30 | #include "util/u_string.h" |
||
31 | #include "util/u_memory.h" |
||
32 | #include "util/u_helpers.h" |
||
33 | #include "util/u_format.h" |
||
34 | |||
35 | #include "freedreno_resource.h" |
||
36 | |||
37 | #include "fd3_emit.h" |
||
38 | #include "fd3_blend.h" |
||
39 | #include "fd3_context.h" |
||
40 | #include "fd3_program.h" |
||
41 | #include "fd3_rasterizer.h" |
||
42 | #include "fd3_texture.h" |
||
43 | #include "fd3_util.h" |
||
44 | #include "fd3_zsa.h" |
||
45 | |||
46 | /* regid: base const register |
||
47 | * prsc or dwords: buffer containing constant values |
||
48 | * sizedwords: size of const value buffer |
||
49 | */ |
||
50 | void |
||
51 | fd3_emit_constant(struct fd_ringbuffer *ring, |
||
52 | enum adreno_state_block sb, |
||
53 | uint32_t regid, uint32_t offset, uint32_t sizedwords, |
||
54 | const uint32_t *dwords, struct pipe_resource *prsc) |
||
55 | { |
||
56 | uint32_t i, sz; |
||
57 | enum adreno_state_src src; |
||
58 | |||
59 | if (prsc) { |
||
60 | sz = 0; |
||
61 | src = SS_INDIRECT; |
||
62 | } else { |
||
63 | sz = sizedwords; |
||
64 | src = SS_DIRECT; |
||
65 | } |
||
66 | |||
67 | /* we have this sometimes, not others.. perhaps we could be clever |
||
68 | * and figure out actually when we need to invalidate cache: |
||
69 | */ |
||
70 | OUT_PKT0(ring, REG_A3XX_UCHE_CACHE_INVALIDATE0_REG, 2); |
||
71 | OUT_RING(ring, A3XX_UCHE_CACHE_INVALIDATE0_REG_ADDR(0)); |
||
72 | OUT_RING(ring, A3XX_UCHE_CACHE_INVALIDATE1_REG_ADDR(0) | |
||
73 | A3XX_UCHE_CACHE_INVALIDATE1_REG_OPCODE(INVALIDATE) | |
||
74 | A3XX_UCHE_CACHE_INVALIDATE1_REG_ENTIRE_CACHE); |
||
75 | |||
76 | OUT_PKT3(ring, CP_LOAD_STATE, 2 + sz); |
||
77 | OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(regid/2) | |
||
78 | CP_LOAD_STATE_0_STATE_SRC(src) | |
||
79 | CP_LOAD_STATE_0_STATE_BLOCK(sb) | |
||
80 | CP_LOAD_STATE_0_NUM_UNIT(sizedwords/2)); |
||
81 | if (prsc) { |
||
82 | struct fd_bo *bo = fd_resource(prsc)->bo; |
||
83 | OUT_RELOC(ring, bo, offset, |
||
4401 | Serge | 84 | CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS), 0); |
4358 | Serge | 85 | } else { |
86 | OUT_RING(ring, CP_LOAD_STATE_1_EXT_SRC_ADDR(0) | |
||
87 | CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS)); |
||
88 | dwords = (uint32_t *)&((uint8_t *)dwords)[offset]; |
||
89 | } |
||
90 | for (i = 0; i < sz; i++) { |
||
91 | OUT_RING(ring, dwords[i]); |
||
92 | } |
||
93 | } |
||
94 | |||
95 | static void |
||
96 | emit_constants(struct fd_ringbuffer *ring, |
||
97 | enum adreno_state_block sb, |
||
98 | struct fd_constbuf_stateobj *constbuf, |
||
99 | struct fd3_shader_stateobj *shader) |
||
100 | { |
||
101 | uint32_t enabled_mask = constbuf->enabled_mask; |
||
102 | uint32_t base = 0; |
||
103 | unsigned i; |
||
104 | |||
105 | // XXX TODO only emit dirty consts.. but we need to keep track if |
||
106 | // they are clobbered by a clear, gmem2mem, or mem2gmem.. |
||
107 | constbuf->dirty_mask = enabled_mask; |
||
108 | |||
109 | /* emit user constants: */ |
||
110 | while (enabled_mask) { |
||
111 | unsigned index = ffs(enabled_mask) - 1; |
||
112 | struct pipe_constant_buffer *cb = &constbuf->cb[index]; |
||
113 | unsigned size = align(cb->buffer_size, 4) / 4; /* size in dwords */ |
||
114 | |||
115 | // I expect that size should be a multiple of vec4's: |
||
116 | assert(size == align(size, 4)); |
||
117 | |||
118 | /* gallium could have const-buffer still bound, even though the |
||
119 | * shader is not using it. Writing consts above constlen (or |
||
120 | * rather, HLSQ_{VS,FS}_CONTROL_REG.CONSTLENGTH) will cause a |
||
121 | * hang. |
||
122 | */ |
||
123 | if ((base / 4) >= shader->constlen) |
||
124 | break; |
||
125 | |||
126 | if (constbuf->dirty_mask & (1 << index)) { |
||
127 | fd3_emit_constant(ring, sb, base, |
||
128 | cb->buffer_offset, size, |
||
129 | cb->user_buffer, cb->buffer); |
||
130 | constbuf->dirty_mask &= ~(1 << index); |
||
131 | } |
||
132 | |||
133 | base += size; |
||
134 | enabled_mask &= ~(1 << index); |
||
135 | } |
||
136 | |||
137 | /* emit shader immediates: */ |
||
138 | if (shader) { |
||
139 | for (i = 0; i < shader->immediates_count; i++) { |
||
140 | fd3_emit_constant(ring, sb, |
||
141 | 4 * (shader->first_immediate + i), |
||
142 | 0, 4, shader->immediates[i].val, NULL); |
||
143 | } |
||
144 | } |
||
145 | } |
||
146 | |||
147 | #define VERT_TEX_OFF 0 |
||
148 | #define FRAG_TEX_OFF 16 |
||
149 | #define BASETABLE_SZ 14 |
||
150 | |||
151 | static void |
||
152 | emit_textures(struct fd_ringbuffer *ring, |
||
153 | enum adreno_state_block sb, |
||
154 | struct fd_texture_stateobj *tex) |
||
155 | { |
||
156 | static const unsigned tex_off[] = { |
||
157 | [SB_VERT_TEX] = VERT_TEX_OFF, |
||
158 | [SB_FRAG_TEX] = FRAG_TEX_OFF, |
||
159 | }; |
||
160 | static const enum adreno_state_block mipaddr[] = { |
||
161 | [SB_VERT_TEX] = SB_VERT_MIPADDR, |
||
162 | [SB_FRAG_TEX] = SB_FRAG_MIPADDR, |
||
163 | }; |
||
164 | unsigned i, j; |
||
165 | |||
166 | assert(tex->num_samplers == tex->num_textures); // TODO check.. |
||
167 | |||
168 | if (!tex->num_samplers) |
||
169 | return; |
||
170 | |||
171 | /* output sampler state: */ |
||
172 | OUT_PKT3(ring, CP_LOAD_STATE, 2 + (2 * tex->num_samplers)); |
||
173 | OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(tex_off[sb]) | |
||
174 | CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) | |
||
175 | CP_LOAD_STATE_0_STATE_BLOCK(sb) | |
||
176 | CP_LOAD_STATE_0_NUM_UNIT(tex->num_samplers)); |
||
177 | OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_SHADER) | |
||
178 | CP_LOAD_STATE_1_EXT_SRC_ADDR(0)); |
||
179 | for (i = 0; i < tex->num_samplers; i++) { |
||
180 | struct fd3_sampler_stateobj *sampler = |
||
181 | fd3_sampler_stateobj(tex->samplers[i]); |
||
182 | OUT_RING(ring, sampler->texsamp0); |
||
183 | OUT_RING(ring, sampler->texsamp1); |
||
184 | } |
||
185 | |||
186 | /* emit texture state: */ |
||
187 | OUT_PKT3(ring, CP_LOAD_STATE, 2 + (4 * tex->num_textures)); |
||
188 | OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(tex_off[sb]) | |
||
189 | CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) | |
||
190 | CP_LOAD_STATE_0_STATE_BLOCK(sb) | |
||
191 | CP_LOAD_STATE_0_NUM_UNIT(tex->num_textures)); |
||
192 | OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS) | |
||
193 | CP_LOAD_STATE_1_EXT_SRC_ADDR(0)); |
||
194 | for (i = 0; i < tex->num_textures; i++) { |
||
195 | struct fd3_pipe_sampler_view *view = |
||
196 | fd3_pipe_sampler_view(tex->textures[i]); |
||
197 | OUT_RING(ring, view->texconst0); |
||
198 | OUT_RING(ring, view->texconst1); |
||
199 | OUT_RING(ring, view->texconst2 | |
||
200 | A3XX_TEX_CONST_2_INDX(BASETABLE_SZ * i)); |
||
201 | OUT_RING(ring, view->texconst3); |
||
202 | } |
||
203 | |||
204 | /* emit mipaddrs: */ |
||
205 | OUT_PKT3(ring, CP_LOAD_STATE, 2 + (BASETABLE_SZ * tex->num_textures)); |
||
206 | OUT_RING(ring, CP_LOAD_STATE_0_DST_OFF(BASETABLE_SZ * tex_off[sb]) | |
||
207 | CP_LOAD_STATE_0_STATE_SRC(SS_DIRECT) | |
||
208 | CP_LOAD_STATE_0_STATE_BLOCK(mipaddr[sb]) | |
||
209 | CP_LOAD_STATE_0_NUM_UNIT(BASETABLE_SZ * tex->num_textures)); |
||
210 | OUT_RING(ring, CP_LOAD_STATE_1_STATE_TYPE(ST_CONSTANTS) | |
||
211 | CP_LOAD_STATE_1_EXT_SRC_ADDR(0)); |
||
212 | for (i = 0; i < tex->num_textures; i++) { |
||
213 | struct fd3_pipe_sampler_view *view = |
||
214 | fd3_pipe_sampler_view(tex->textures[i]); |
||
4401 | Serge | 215 | OUT_RELOC(ring, view->tex_resource->bo, 0, 0, 0); |
4358 | Serge | 216 | /* I think each entry is a ptr to mipmap level.. for now, just |
217 | * pad w/ null's until I get around to actually implementing |
||
218 | * mipmap support.. |
||
219 | */ |
||
220 | for (j = 1; j < BASETABLE_SZ; j++) { |
||
221 | OUT_RING(ring, 0x00000000); |
||
222 | } |
||
223 | } |
||
224 | } |
||
225 | |||
226 | static void |
||
227 | emit_cache_flush(struct fd_ringbuffer *ring) |
||
228 | { |
||
229 | OUT_PKT3(ring, CP_EVENT_WRITE, 1); |
||
230 | OUT_RING(ring, CACHE_FLUSH); |
||
231 | |||
232 | OUT_PKT3(ring, CP_DRAW_INDX, 3); |
||
233 | OUT_RING(ring, 0x00000000); |
||
234 | OUT_RING(ring, DRAW(DI_PT_POINTLIST, DI_SRC_SEL_AUTO_INDEX, |
||
235 | INDEX_SIZE_IGN, IGNORE_VISIBILITY)); |
||
236 | OUT_RING(ring, 0); /* NumIndices */ |
||
237 | |||
238 | OUT_PKT3(ring, CP_NOP, 4); |
||
239 | OUT_RING(ring, 0x00000000); |
||
240 | OUT_RING(ring, 0x00000000); |
||
241 | OUT_RING(ring, 0x00000000); |
||
242 | OUT_RING(ring, 0x00000000); |
||
243 | |||
244 | OUT_PKT3(ring, CP_WAIT_FOR_IDLE, 1); |
||
245 | OUT_RING(ring, 0x00000000); |
||
246 | } |
||
247 | |||
248 | /* emit texture state for mem->gmem restore operation.. eventually it would |
||
249 | * be good to get rid of this and use normal CSO/etc state for more of these |
||
250 | * special cases, but for now the compiler is not sufficient.. |
||
251 | */ |
||
252 | void |
||
253 | fd3_emit_gmem_restore_tex(struct fd_ringbuffer *ring, struct pipe_surface *psurf) |
||
254 | { |
||
255 | struct fd_resource *rsc = fd_resource(psurf->texture); |
||
256 | |||