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
  28 #include "vc4_screen.h"
  29 #include "vc4_context.h"
  30 #include "kernel/vc4_drv.h"
  31 #include "vc4_simulator_validate.h"
  32 #include "simpenrose/simpenrose.h"
  33
  34 #define OVERFLOW_SIZE (32 * 1024 * 1024)
  35
  36 static struct drm_gem_cma_object *
  37 vc4_wrap_bo_with_cma(struct drm_device *dev, struct vc4_bo *bo)
  38 {
  39         struct vc4_context *vc4 = dev->vc4;
  40         struct vc4_screen *screen = vc4->screen;
  41         struct drm_gem_cma_object *obj = CALLOC_STRUCT(drm_gem_cma_object);
  42         uint32_t size = align(bo->size, 4096);
  43
  44         obj->bo = bo;
  45         obj->base.size = size;
  46         obj->vaddr = screen->simulator_mem_base + dev->simulator_mem_next;
  47         obj->paddr = simpenrose_hw_addr(obj->vaddr);
  48
  49         dev->simulator_mem_next += size;
  50         dev->simulator_mem_next = align(dev->simulator_mem_next, 4096);
  51         assert(dev->simulator_mem_next <= screen->simulator_mem_size);
  52
  53         return obj;
  54 }
  55
  56 struct drm_gem_cma_object *
  57 drm_gem_cma_create(struct drm_device *dev, size_t size)
  58 {
  59         struct vc4_context *vc4 = dev->vc4;
  60         struct vc4_screen *screen = vc4->screen;
  61
  62         struct vc4_bo *bo = vc4_bo_alloc(screen, size, "simulator validate");
  63         return vc4_wrap_bo_with_cma(dev, bo);
  64 }
  65
  66 static int
  67 vc4_simulator_pin_bos(struct drm_device *dev, struct exec_info *exec)
  68 {
  69         struct drm_vc4_submit_cl *args = exec->args;
  70         struct vc4_context *vc4 = dev->vc4;
  71         struct vc4_bo **bos = vc4->bo_pointers.base;
  72
  73         exec->bo_count = args->bo_handle_count;
  74         exec->bo = calloc(exec->bo_count, sizeof(struct vc4_bo_exec_state));
  75         for (int i = 0; i < exec->bo_count; i++) {
  76                 struct vc4_bo *bo = bos[i];
  77                 struct drm_gem_cma_object *obj = vc4_wrap_bo_with_cma(dev, bo);
  78
  79 #if 0
  80                 fprintf(stderr, "bo hindex %d: %s\n", i, bo->name);
  81 #endif
  82
  83                 vc4_bo_map(bo);
  84                 memcpy(obj->vaddr, bo->map, bo->size);
  85
  86                 exec->bo[i].bo = obj;
  87         }
  88         return 0;
  89 }
  90
  91 static int
  92 vc4_simulator_unpin_bos(struct exec_info *exec)
  93 {
  94         for (int i = 0; i < exec->bo_count; i++) {
  95                 struct drm_gem_cma_object *obj = exec->bo[i].bo;
  96                 struct vc4_bo *bo = obj->bo;
  97
  98                 memcpy(bo->map, obj->vaddr, bo->size);
  99
 100                 free(obj);
 101         }
 102
 103         free(exec->bo);
 104
 105         return 0;
 106 }
 107
 108 int
 109 vc4_simulator_flush(struct vc4_context *vc4, struct drm_vc4_submit_cl *args)
 110 {
 111         struct vc4_screen *screen = vc4->screen;
 112         struct vc4_surface *csurf = vc4_surface(vc4->framebuffer.cbufs[0]);
 113         struct vc4_resource *ctex = csurf ? vc4_resource(csurf->base.texture) : NULL;
 114         uint32_t winsys_stride = ctex ? ctex->bo->simulator_winsys_stride : 0;
 115         uint32_t sim_stride = ctex ? ctex->slices[0].stride : 0;
 116         uint32_t row_len = MIN2(sim_stride, winsys_stride);
 117         struct exec_info exec;
 118         struct drm_device local_dev = {
 119                 .vc4 = vc4,
 120                 .simulator_mem_next = OVERFLOW_SIZE,
 121         };
 122         struct drm_device *dev = &local_dev;
 123         int ret;
 124
 125         memset(&exec, 0, sizeof(exec));
 126
 127         if (ctex && ctex->bo->simulator_winsys_map) {
 128 #if 0
 129                 fprintf(stderr, "%dx%d %d %d %d\n",
 130                         ctex->base.b.width0, ctex->base.b.height0,
 131                         winsys_stride,
 132                         sim_stride,
 133                         ctex->bo->size);
 134 #endif
 135
 136                 for (int y = 0; y < ctex->base.b.height0; y++) {
 137                         memcpy(ctex->bo->map + y * sim_stride,
 138                                ctex->bo->simulator_winsys_map + y * winsys_stride,
 139                                row_len);
 140                 }
 141         }
 142
 143         exec.args = args;
 144
 145         ret = vc4_simulator_pin_bos(dev, &exec);
 146         if (ret)
 147                 return ret;
 148
 149         ret = vc4_cl_validate(dev, &exec);
 150         if (ret)
 151                 return ret;
 152
 153         int bfc = simpenrose_do_binning(exec.ct0ca, exec.ct0ea);
 154         if (bfc != 1) {
 155                 fprintf(stderr, "Binning returned %d flushes, should be 1.\n",
 156                         bfc);
 157                 fprintf(stderr, "Relocated binning command list:\n");
 158                 vc4_dump_cl(screen->simulator_mem_base + exec.ct0ca,
 159                             exec.ct0ea - exec.ct0ca, false);
 160                 abort();
 161         }
 162         int rfc = simpenrose_do_rendering(exec.ct1ca, exec.ct1ea);
 163         if (rfc != 1) {
 164                 fprintf(stderr, "Rendering returned %d frames, should be 1.\n",
 165                         rfc);
 166                 fprintf(stderr, "Relocated render command list:\n");
 167                 vc4_dump_cl(screen->simulator_mem_base + exec.ct1ca,
 168                             exec.ct1ea - exec.ct1ca, true);
 169                 abort();
 170         }
 171
 172         ret = vc4_simulator_unpin_bos(&exec);
 173         if (ret)
 174                 return ret;
 175
 176         free(exec.exec_bo);
 177
 178         if (ctex && ctex->bo->simulator_winsys_map) {
 179                 for (int y = 0; y < ctex->base.b.height0; y++) {
 180                         memcpy(ctex->bo->simulator_winsys_map + y * winsys_stride,
 181                                ctex->bo->map + y * sim_stride,
 182                                row_len);
 183                 }
 184         }
 185
 186         return 0;
 187 }
 188
 189 void
 190 vc4_simulator_init(struct vc4_screen *screen)
 191 {
 192         screen->simulator_mem_size = 256 * 1024 * 1024;
 193         screen->simulator_mem_base = malloc(screen->simulator_mem_size);
 194
 195         /* We supply our own memory so that we can have more aperture
 196          * available (256MB instead of simpenrose's default 64MB).
 197          */
 198         simpenrose_init_hardware_supply_mem(screen->simulator_mem_base,
 199                                             screen->simulator_mem_size);
 200
 201         /* Carve out low memory for tile allocation overflow.  The kernel
 202          * should be automatically handling overflow memory setup on real
 203          * hardware, but for simulation we just get one shot to set up enough
 204          * overflow memory before execution.  This overflow mem will be used
 205          * up over the whole lifetime of simpenrose (not reused on each
 206          * flush), so it had better be big.
 207          */
 208         simpenrose_supply_overflow_mem(0, OVERFLOW_SIZE);
 209 }
 210
 211 #endif /* USE_VC4_SIMULATOR */