vc4: Simplify vc4_use_bo and make sure it's not a shader.
[mesa.git] / src / gallium / drivers / vc4 / vc4_simulator.c
1 /*
2 * Copyright © 2014 Broadcom
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 (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 #ifdef USE_VC4_SIMULATOR
25
26 #include "util/u_memory.h"
27 #include "util/ralloc.h"
28
29 #include "vc4_screen.h"
30 #include "vc4_context.h"
31 #include "kernel/vc4_drv.h"
32 #include "vc4_simulator_validate.h"
33 #include "simpenrose/simpenrose.h"
34
35 #define OVERFLOW_SIZE (32 * 1024 * 1024)
36
37 static struct drm_gem_cma_object *
38 vc4_wrap_bo_with_cma(struct drm_device *dev, struct vc4_bo *bo)
39 {
40 struct vc4_context *vc4 = dev->vc4;
41 struct vc4_screen *screen = vc4->screen;
42 struct drm_vc4_bo *drm_bo = CALLOC_STRUCT(drm_vc4_bo);
43 struct drm_gem_cma_object *obj = &drm_bo->base;
44 uint32_t size = align(bo->size, 4096);
45
46 drm_bo->bo = bo;
47 obj->base.size = size;
48 obj->base.dev = dev;
49 obj->vaddr = screen->simulator_mem_base + dev->simulator_mem_next;
50 obj->paddr = simpenrose_hw_addr(obj->vaddr);
51
52 dev->simulator_mem_next += size;
53 dev->simulator_mem_next = align(dev->simulator_mem_next, 4096);
54 assert(dev->simulator_mem_next <= screen->simulator_mem_size);
55
56 return obj;
57 }
58
59 struct drm_gem_cma_object *
60 drm_gem_cma_create(struct drm_device *dev, size_t size)
61 {
62 struct vc4_context *vc4 = dev->vc4;
63 struct vc4_screen *screen = vc4->screen;
64
65 struct vc4_bo *bo = vc4_bo_alloc(screen, size, "simulator validate");
66 return vc4_wrap_bo_with_cma(dev, bo);
67 }
68
69 static int
70 vc4_simulator_pin_bos(struct drm_device *dev, struct vc4_exec_info *exec)
71 {
72 struct drm_vc4_submit_cl *args = exec->args;
73 struct vc4_context *vc4 = dev->vc4;
74 struct vc4_bo **bos = vc4->bo_pointers.base;
75
76 exec->bo_count = args->bo_handle_count;
77 exec->bo = calloc(exec->bo_count, sizeof(void *));
78 for (int i = 0; i < exec->bo_count; i++) {
79 struct vc4_bo *bo = bos[i];
80 struct drm_gem_cma_object *obj = vc4_wrap_bo_with_cma(dev, bo);
81
82 struct drm_vc4_bo *drm_bo = to_vc4_bo(&obj->base);
83 #if 0
84 fprintf(stderr, "bo hindex %d: %s\n", i, bo->name);
85 #endif
86
87 vc4_bo_map(bo);
88 memcpy(obj->vaddr, bo->map, bo->size);
89
90 exec->bo[i] = obj;
91
92 /* The kernel does this validation at shader create ioctl
93 * time.
94 */
95 if (strcmp(bo->name, "code") == 0) {
96 drm_bo->validated_shader = vc4_validate_shader(obj);
97 if (!drm_bo->validated_shader)
98 abort();
99 }
100 }
101 return 0;
102 }
103
104 static int
105 vc4_simulator_unpin_bos(struct vc4_exec_info *exec)
106 {
107 for (int i = 0; i < exec->bo_count; i++) {
108 struct drm_gem_cma_object *obj = exec->bo[i];
109 struct vc4_bo *bo = to_vc4_bo(&obj->base)->bo;
110
111 memcpy(bo->map, obj->vaddr, bo->size);
112
113 free(obj);
114 }
115
116 free(exec->bo);
117
118 return 0;
119 }
120
121 int
122 vc4_simulator_flush(struct vc4_context *vc4, struct drm_vc4_submit_cl *args)
123 {
124 struct vc4_screen *screen = vc4->screen;
125 struct vc4_surface *csurf = vc4_surface(vc4->framebuffer.cbufs[0]);
126 struct vc4_resource *ctex = csurf ? vc4_resource(csurf->base.texture) : NULL;
127 uint32_t winsys_stride = ctex ? ctex->bo->simulator_winsys_stride : 0;
128 uint32_t sim_stride = ctex ? ctex->slices[0].stride : 0;
129 uint32_t row_len = MIN2(sim_stride, winsys_stride);
130 struct vc4_exec_info exec;
131 struct drm_device local_dev = {
132 .vc4 = vc4,
133 .simulator_mem_next = OVERFLOW_SIZE,
134 };
135 struct drm_device *dev = &local_dev;
136 int ret;
137
138 memset(&exec, 0, sizeof(exec));
139 list_inithead(&exec.unref_list);
140
141 if (ctex && ctex->bo->simulator_winsys_map) {
142 #if 0
143 fprintf(stderr, "%dx%d %d %d %d\n",
144 ctex->base.b.width0, ctex->base.b.height0,
145 winsys_stride,
146 sim_stride,
147 ctex->bo->size);
148 #endif
149
150 for (int y = 0; y < ctex->base.b.height0; y++) {
151 memcpy(ctex->bo->map + y * sim_stride,
152 ctex->bo->simulator_winsys_map + y * winsys_stride,
153 row_len);
154 }
155 }
156
157 exec.args = args;
158
159 ret = vc4_simulator_pin_bos(dev, &exec);
160 if (ret)
161 return ret;
162
163 ret = vc4_cl_validate(dev, &exec);
164 if (ret)
165 return ret;
166
167 if (exec.ct0ca != exec.ct0ea) {
168 int bfc = simpenrose_do_binning(exec.ct0ca, exec.ct0ea);
169 if (bfc != 1) {
170 fprintf(stderr, "Binning returned %d flushes, should be 1.\n",
171 bfc);
172 fprintf(stderr, "Relocated binning command list:\n");
173 vc4_dump_cl(screen->simulator_mem_base + exec.ct0ca,
174 exec.ct0ea - exec.ct0ca, false);
175 abort();
176 }
177 }
178 int rfc = simpenrose_do_rendering(exec.ct1ca, exec.ct1ea);
179 if (rfc != 1) {
180 fprintf(stderr, "Rendering returned %d frames, should be 1.\n",
181 rfc);
182 fprintf(stderr, "Relocated render command list:\n");
183 vc4_dump_cl(screen->simulator_mem_base + exec.ct1ca,
184 exec.ct1ea - exec.ct1ca, true);
185 abort();
186 }
187
188 ret = vc4_simulator_unpin_bos(&exec);
189 if (ret)
190 return ret;
191
192 list_for_each_entry_safe(struct drm_vc4_bo, bo, &exec.unref_list,
193 unref_head) {
194 list_del(&bo->unref_head);
195 vc4_bo_unreference(&bo->bo);
196 free(bo);
197 }
198
199 if (ctex && ctex->bo->simulator_winsys_map) {
200 for (int y = 0; y < ctex->base.b.height0; y++) {
201 memcpy(ctex->bo->simulator_winsys_map + y * winsys_stride,
202 ctex->bo->map + y * sim_stride,
203 row_len);
204 }
205 }
206
207 return 0;
208 }
209
210 void
211 vc4_simulator_init(struct vc4_screen *screen)
212 {
213 screen->simulator_mem_size = 256 * 1024 * 1024;
214 screen->simulator_mem_base = ralloc_size(screen,
215 screen->simulator_mem_size);
216
217 /* We supply our own memory so that we can have more aperture
218 * available (256MB instead of simpenrose's default 64MB).
219 */
220 simpenrose_init_hardware_supply_mem(screen->simulator_mem_base,
221 screen->simulator_mem_size);
222
223 /* Carve out low memory for tile allocation overflow. The kernel
224 * should be automatically handling overflow memory setup on real
225 * hardware, but for simulation we just get one shot to set up enough
226 * overflow memory before execution. This overflow mem will be used
227 * up over the whole lifetime of simpenrose (not reused on each
228 * flush), so it had better be big.
229 */
230 simpenrose_supply_overflow_mem(0, OVERFLOW_SIZE);
231 }
232
233 #endif /* USE_VC4_SIMULATOR */