src/freedreno/vulkan/tu_shader.c

   1 /*
   2  * Copyright © 2019 Google LLC
   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
  21  * DEALINGS IN THE SOFTWARE.
  22  */
  23
  24 #include "tu_private.h"
  25
  26 #include "spirv/nir_spirv.h"
  27 #include "util/mesa-sha1.h"
  28
  29 #include "ir3/ir3_nir.h"
  30
  31 static nir_function *
  32 tu_spirv_to_nir(struct ir3_compiler *compiler,
  33                 const uint32_t *words,
  34                 size_t word_count,
  35                 gl_shader_stage stage,
  36                 const char *entry_point_name,
  37                 const VkSpecializationInfo *spec_info)
  38 {
  39    /* TODO these are made-up */
  40    const struct spirv_to_nir_options spirv_options = {
  41       .lower_workgroup_access_to_offsets = true,
  42       .lower_ubo_ssbo_access_to_offsets = true,
  43       .caps = { false },
  44    };
  45    const nir_shader_compiler_options *nir_options =
  46       ir3_get_compiler_options(compiler);
  47
  48    /* convert VkSpecializationInfo */
  49    struct nir_spirv_specialization *spec = NULL;
  50    uint32_t num_spec = 0;
  51    if (spec_info && spec_info->mapEntryCount) {
  52       spec = malloc(sizeof(*spec) * spec_info->mapEntryCount);
  53       if (!spec)
  54          return NULL;
  55
  56       for (uint32_t i = 0; i < spec_info->mapEntryCount; i++) {
  57          const VkSpecializationMapEntry *entry = &spec_info->pMapEntries[i];
  58          const void *data = spec_info->pData + entry->offset;
  59          assert(data + entry->size <= spec_info->pData + spec_info->dataSize);
  60          spec[i].id = entry->constantID;
  61          if (entry->size == 8)
  62             spec[i].data64 = *(const uint64_t *) data;
  63          else
  64             spec[i].data32 = *(const uint32_t *) data;
  65          spec[i].defined_on_module = false;
  66       }
  67
  68       num_spec = spec_info->mapEntryCount;
  69    }
  70
  71    nir_function *entry_point =
  72       spirv_to_nir(words, word_count, spec, num_spec, stage, entry_point_name,
  73                    &spirv_options, nir_options);
  74
  75    free(spec);
  76
  77    assert(entry_point->shader->info.stage == stage);
  78    nir_validate_shader(entry_point->shader, "after spirv_to_nir");
  79
  80    return entry_point;
  81 }
  82
  83 static void
  84 tu_sort_variables_by_location(struct exec_list *variables)
  85 {
  86    struct exec_list sorted;
  87    exec_list_make_empty(&sorted);
  88
  89    nir_foreach_variable_safe(var, variables)
  90    {
  91       exec_node_remove(&var->node);
  92
  93       /* insert the variable into the sorted list */
  94       nir_variable *next = NULL;
  95       nir_foreach_variable(tmp, &sorted)
  96       {
  97          if (var->data.location < tmp->data.location) {
  98             next = tmp;
  99             break;
 100          }
 101       }
 102       if (next)
 103          exec_node_insert_node_before(&next->node, &var->node);
 104       else
 105          exec_list_push_tail(&sorted, &var->node);
 106    }
 107
 108    exec_list_move_nodes_to(&sorted, variables);
 109 }
 110
 111 struct tu_shader *
 112 tu_shader_create(struct tu_device *dev,
 113                  gl_shader_stage stage,
 114                  const VkPipelineShaderStageCreateInfo *stage_info,
 115                  const VkAllocationCallbacks *alloc)
 116 {
 117    const struct tu_shader_module *module =
 118       tu_shader_module_from_handle(stage_info->module);
 119    struct tu_shader *shader;
 120
 121    const uint32_t max_variant_count = (stage == MESA_SHADER_VERTEX) ? 2 : 1;
 122    shader = vk_zalloc2(
 123       &dev->alloc, alloc,
 124       sizeof(*shader) + sizeof(struct ir3_shader_variant) * max_variant_count,
 125       8, VK_SYSTEM_ALLOCATION_SCOPE_COMMAND);
 126    if (!shader)
 127       return NULL;
 128
 129    /* translate SPIR-V to NIR */
 130    assert(module->code_size % 4 == 0);
 131    nir_function *entry_point = tu_spirv_to_nir(
 132       dev->compiler, (const uint32_t *) module->code, module->code_size / 4,
 133       stage, stage_info->pName, stage_info->pSpecializationInfo);
 134    if (!entry_point) {
 135       vk_free2(&dev->alloc, alloc, shader);
 136       return NULL;
 137    }
 138
 139    nir_shader *nir = entry_point->shader;
 140
 141    if (unlikely(dev->physical_device->instance->debug_flags & TU_DEBUG_NIR)) {
 142       fprintf(stderr, "translated nir:\n");
 143       nir_print_shader(nir, stderr);
 144    }
 145
 146    /* TODO what needs to happen? */
 147
 148    switch (stage) {
 149    case MESA_SHADER_VERTEX:
 150       tu_sort_variables_by_location(&nir->outputs);
 151       break;
 152    case MESA_SHADER_TESS_CTRL:
 153    case MESA_SHADER_TESS_EVAL:
 154    case MESA_SHADER_GEOMETRY:
 155       tu_sort_variables_by_location(&nir->inputs);
 156       tu_sort_variables_by_location(&nir->outputs);
 157       break;
 158    case MESA_SHADER_FRAGMENT:
 159       tu_sort_variables_by_location(&nir->inputs);
 160       break;
 161    case MESA_SHADER_COMPUTE:
 162       break;
 163    default:
 164       unreachable("invalid gl_shader_stage");
 165       break;
 166    }
 167
 168    nir_assign_var_locations(&nir->inputs, &nir->num_inputs,
 169                             ir3_glsl_type_size);
 170    nir_assign_var_locations(&nir->outputs, &nir->num_outputs,
 171                             ir3_glsl_type_size);
 172    nir_assign_var_locations(&nir->uniforms, &nir->num_uniforms,
 173                             ir3_glsl_type_size);
 174
 175    NIR_PASS_V(nir, nir_lower_system_values);
 176    NIR_PASS_V(nir, nir_lower_io, nir_var_all, ir3_glsl_type_size, 0);
 177
 178    nir_shader_gather_info(nir, entry_point->impl);
 179
 180    shader->ir3_shader.compiler = dev->compiler;
 181    shader->ir3_shader.type = stage;
 182    shader->ir3_shader.nir = nir;
 183
 184    return shader;
 185 }
 186
 187 void
 188 tu_shader_destroy(struct tu_device *dev,
 189                   struct tu_shader *shader,
 190                   const VkAllocationCallbacks *alloc)
 191 {
 192    if (shader->ir3_shader.nir)
 193       ralloc_free(shader->ir3_shader.nir);
 194
 195    for (uint32_t i = 0; i < 1 + shader->has_binning_pass; i++) {
 196       if (shader->variants[i].ir)
 197          ir3_destroy(shader->variants[i].ir);
 198       if (shader->variants[i].immediates)
 199          free(shader->variants[i].immediates);
 200    }
 201
 202    if (shader->binary)
 203       free(shader->binary);
 204    if (shader->binning_binary)
 205       free(shader->binning_binary);
 206
 207    vk_free2(&dev->alloc, alloc, shader);
 208 }
 209
 210 void
 211 tu_shader_compile_options_init(
 212    struct tu_shader_compile_options *options,
 213    const VkGraphicsPipelineCreateInfo *pipeline_info)
 214 {
 215    *options = (struct tu_shader_compile_options) {
 216       /* TODO ir3_key */
 217
 218       .optimize = !(pipeline_info->flags &
 219                     VK_PIPELINE_CREATE_DISABLE_OPTIMIZATION_BIT),
 220       .include_binning_pass = true,
 221    };
 222 }
 223
 224 static uint32_t *
 225 tu_compile_shader_variant(struct ir3_shader *shader,
 226                           const struct ir3_shader_key *key,
 227                           bool binning_pass,
 228                           struct ir3_shader_variant *variant)
 229 {
 230    variant->shader = shader;
 231    variant->type = shader->type;
 232    variant->key = *key;
 233    variant->binning_pass = binning_pass;
 234
 235    int ret = ir3_compile_shader_nir(shader->compiler, variant);
 236    if (ret)
 237       return NULL;
 238
 239    /* when assemble fails, we rely on tu_shader_destroy to clean up the
 240     * variant
 241     */
 242    return ir3_shader_assemble(variant, shader->compiler->gpu_id);
 243 }
 244
 245 VkResult
 246 tu_shader_compile(struct tu_device *dev,
 247                   struct tu_shader *shader,
 248                   const struct tu_shader *next_stage,
 249                   const struct tu_shader_compile_options *options,
 250                   const VkAllocationCallbacks *alloc)
 251 {
 252    if (options->optimize) {
 253       /* ignore the key for the first pass of optimization */
 254       ir3_optimize_nir(&shader->ir3_shader, shader->ir3_shader.nir, NULL);
 255
 256       if (unlikely(dev->physical_device->instance->debug_flags &
 257                    TU_DEBUG_NIR)) {
 258          fprintf(stderr, "optimized nir:\n");
 259          nir_print_shader(shader->ir3_shader.nir, stderr);
 260       }
 261    }
 262
 263    shader->binary = tu_compile_shader_variant(
 264       &shader->ir3_shader, &options->key, false, &shader->variants[0]);
 265    if (!shader->binary)
 266       return VK_ERROR_OUT_OF_HOST_MEMORY;
 267
 268    /* compile another variant for the binning pass */
 269    if (options->include_binning_pass &&
 270        shader->ir3_shader.type == MESA_SHADER_VERTEX) {
 271       shader->binning_binary = tu_compile_shader_variant(
 272          &shader->ir3_shader, &options->key, true, &shader->variants[1]);
 273       if (!shader->binning_binary)
 274          return VK_ERROR_OUT_OF_HOST_MEMORY;
 275
 276       shader->has_binning_pass = true;
 277    }
 278
 279    if (unlikely(dev->physical_device->instance->debug_flags & TU_DEBUG_IR3)) {
 280       fprintf(stderr, "disassembled ir3:\n");
 281       fprintf(stderr, "shader: %s\n",
 282               gl_shader_stage_name(shader->ir3_shader.type));
 283       ir3_shader_disasm(&shader->variants[0], shader->binary, stderr);
 284
 285       if (shader->has_binning_pass) {
 286          fprintf(stderr, "disassembled ir3:\n");
 287          fprintf(stderr, "shader: %s (binning)\n",
 288                  gl_shader_stage_name(shader->ir3_shader.type));
 289          ir3_shader_disasm(&shader->variants[1], shader->binning_binary,
 290                            stderr);
 291       }
 292    }
 293
 294    return VK_SUCCESS;
 295 }
 296
 297 VkResult
 298 tu_CreateShaderModule(VkDevice _device,
 299                       const VkShaderModuleCreateInfo *pCreateInfo,
 300                       const VkAllocationCallbacks *pAllocator,
 301                       VkShaderModule *pShaderModule)
 302 {
 303    TU_FROM_HANDLE(tu_device, device, _device);
 304    struct tu_shader_module *module;
 305
 306    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO);
 307    assert(pCreateInfo->flags == 0);
 308    assert(pCreateInfo->codeSize % 4 == 0);
 309
 310    module = vk_alloc2(&device->alloc, pAllocator,
 311                       sizeof(*module) + pCreateInfo->codeSize, 8,
 312                       VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 313    if (module == NULL)
 314       return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 315
 316    module->code_size = pCreateInfo->codeSize;
 317    memcpy(module->code, pCreateInfo->pCode, pCreateInfo->codeSize);
 318
 319    _mesa_sha1_compute(module->code, module->code_size, module->sha1);
 320
 321    *pShaderModule = tu_shader_module_to_handle(module);
 322
 323    return VK_SUCCESS;
 324 }
 325
 326 void
 327 tu_DestroyShaderModule(VkDevice _device,
 328                        VkShaderModule _module,
 329                        const VkAllocationCallbacks *pAllocator)
 330 {
 331    TU_FROM_HANDLE(tu_device, device, _device);
 332    TU_FROM_HANDLE(tu_shader_module, module, _module);
 333
 334    if (!module)
 335       return;
 336
 337    vk_free2(&device->alloc, pAllocator, module);
 338 }