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_shader *
  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_ubo_ssbo_access_to_offsets = true,
  42       .caps = { false },
  43    };
  44    const nir_shader_compiler_options *nir_options =
  45       ir3_get_compiler_options(compiler);
  46
  47    /* convert VkSpecializationInfo */
  48    struct nir_spirv_specialization *spec = NULL;
  49    uint32_t num_spec = 0;
  50    if (spec_info && spec_info->mapEntryCount) {
  51       spec = malloc(sizeof(*spec) * spec_info->mapEntryCount);
  52       if (!spec)
  53          return NULL;
  54
  55       for (uint32_t i = 0; i < spec_info->mapEntryCount; i++) {
  56          const VkSpecializationMapEntry *entry = &spec_info->pMapEntries[i];
  57          const void *data = spec_info->pData + entry->offset;
  58          assert(data + entry->size <= spec_info->pData + spec_info->dataSize);
  59          spec[i].id = entry->constantID;
  60          if (entry->size == 8)
  61             spec[i].data64 = *(const uint64_t *) data;
  62          else
  63             spec[i].data32 = *(const uint32_t *) data;
  64          spec[i].defined_on_module = false;
  65       }
  66
  67       num_spec = spec_info->mapEntryCount;
  68    }
  69
  70    nir_shader *nir =
  71       spirv_to_nir(words, word_count, spec, num_spec, stage, entry_point_name,
  72                    &spirv_options, nir_options);
  73
  74    free(spec);
  75
  76    assert(nir->info.stage == stage);
  77    nir_validate_shader(nir, "after spirv_to_nir");
  78
  79    return nir;
  80 }
  81
  82 static void
  83 tu_sort_variables_by_location(struct exec_list *variables)
  84 {
  85    struct exec_list sorted;
  86    exec_list_make_empty(&sorted);
  87
  88    nir_foreach_variable_safe(var, variables)
  89    {
  90       exec_node_remove(&var->node);
  91
  92       /* insert the variable into the sorted list */
  93       nir_variable *next = NULL;
  94       nir_foreach_variable(tmp, &sorted)
  95       {
  96          if (var->data.location < tmp->data.location) {
  97             next = tmp;
  98             break;
  99          }
 100       }
 101       if (next)
 102          exec_node_insert_node_before(&next->node, &var->node);
 103       else
 104          exec_list_push_tail(&sorted, &var->node);
 105    }
 106
 107    exec_list_move_nodes_to(&sorted, variables);
 108 }
 109
 110 struct tu_shader *
 111 tu_shader_create(struct tu_device *dev,
 112                  gl_shader_stage stage,
 113                  const VkPipelineShaderStageCreateInfo *stage_info,
 114                  const VkAllocationCallbacks *alloc)
 115 {
 116    const struct tu_shader_module *module =
 117       tu_shader_module_from_handle(stage_info->module);
 118    struct tu_shader *shader;
 119
 120    const uint32_t max_variant_count = (stage == MESA_SHADER_VERTEX) ? 2 : 1;
 121    shader = vk_zalloc2(
 122       &dev->alloc, alloc,
 123       sizeof(*shader) + sizeof(struct ir3_shader_variant) * max_variant_count,
 124       8, VK_SYSTEM_ALLOCATION_SCOPE_COMMAND);
 125    if (!shader)
 126       return NULL;
 127
 128    /* translate SPIR-V to NIR */
 129    assert(module->code_size % 4 == 0);
 130    nir_shader *nir = tu_spirv_to_nir(
 131       dev->compiler, (const uint32_t *) module->code, module->code_size / 4,
 132       stage, stage_info->pName, stage_info->pSpecializationInfo);
 133    if (!nir) {
 134       vk_free2(&dev->alloc, alloc, shader);
 135       return NULL;
 136    }
 137
 138    if (unlikely(dev->physical_device->instance->debug_flags & TU_DEBUG_NIR)) {
 139       fprintf(stderr, "translated nir:\n");
 140       nir_print_shader(nir, stderr);
 141    }
 142
 143    /* TODO what needs to happen? */
 144
 145    switch (stage) {
 146    case MESA_SHADER_VERTEX:
 147       tu_sort_variables_by_location(&nir->outputs);
 148       break;
 149    case MESA_SHADER_TESS_CTRL:
 150    case MESA_SHADER_TESS_EVAL:
 151    case MESA_SHADER_GEOMETRY:
 152       tu_sort_variables_by_location(&nir->inputs);
 153       tu_sort_variables_by_location(&nir->outputs);
 154       break;
 155    case MESA_SHADER_FRAGMENT:
 156       tu_sort_variables_by_location(&nir->inputs);
 157       break;
 158    case MESA_SHADER_COMPUTE:
 159       break;
 160    default:
 161       unreachable("invalid gl_shader_stage");
 162       break;
 163    }
 164
 165    nir_assign_var_locations(&nir->inputs, &nir->num_inputs,
 166                             ir3_glsl_type_size);
 167    nir_assign_var_locations(&nir->outputs, &nir->num_outputs,
 168                             ir3_glsl_type_size);
 169    nir_assign_var_locations(&nir->uniforms, &nir->num_uniforms,
 170                             ir3_glsl_type_size);
 171
 172    NIR_PASS_V(nir, nir_lower_system_values);
 173    NIR_PASS_V(nir, nir_lower_frexp);
 174    NIR_PASS_V(nir, nir_lower_io, nir_var_all, ir3_glsl_type_size, 0);
 175
 176    nir_shader_gather_info(nir, nir_shader_get_entrypoint(nir));
 177
 178    shader->ir3_shader.compiler = dev->compiler;
 179    shader->ir3_shader.type = stage;
 180    shader->ir3_shader.nir = nir;
 181
 182    return shader;
 183 }
 184
 185 void
 186 tu_shader_destroy(struct tu_device *dev,
 187                   struct tu_shader *shader,
 188                   const VkAllocationCallbacks *alloc)
 189 {
 190    if (shader->ir3_shader.nir)
 191       ralloc_free(shader->ir3_shader.nir);
 192
 193    for (uint32_t i = 0; i < 1 + shader->has_binning_pass; i++) {
 194       if (shader->variants[i].ir)
 195          ir3_destroy(shader->variants[i].ir);
 196    }
 197
 198    if (shader->ir3_shader.const_state.immediates)
 199            free(shader->ir3_shader.const_state.immediates);
 200    if (shader->binary)
 201       free(shader->binary);
 202    if (shader->binning_binary)
 203       free(shader->binning_binary);
 204
 205    vk_free2(&dev->alloc, alloc, shader);
 206 }
 207
 208 void
 209 tu_shader_compile_options_init(
 210    struct tu_shader_compile_options *options,
 211    const VkGraphicsPipelineCreateInfo *pipeline_info)
 212 {
 213    *options = (struct tu_shader_compile_options) {
 214       /* TODO ir3_key */
 215
 216       .optimize = !(pipeline_info->flags &
 217                     VK_PIPELINE_CREATE_DISABLE_OPTIMIZATION_BIT),
 218       .include_binning_pass = true,
 219    };
 220 }
 221
 222 static uint32_t *
 223 tu_compile_shader_variant(struct ir3_shader *shader,
 224                           const struct ir3_shader_key *key,
 225                           struct ir3_shader_variant *nonbinning,
 226                           struct ir3_shader_variant *variant)
 227 {
 228    variant->shader = shader;
 229    variant->type = shader->type;
 230    variant->key = *key;
 231    variant->binning_pass = !!nonbinning;
 232    variant->nonbinning = nonbinning;
 233
 234    int ret = ir3_compile_shader_nir(shader->compiler, variant);
 235    if (ret)
 236       return NULL;
 237
 238    /* when assemble fails, we rely on tu_shader_destroy to clean up the
 239     * variant
 240     */
 241    return ir3_shader_assemble(variant, shader->compiler->gpu_id);
 242 }
 243
 244 VkResult
 245 tu_shader_compile(struct tu_device *dev,
 246                   struct tu_shader *shader,
 247                   const struct tu_shader *next_stage,
 248                   const struct tu_shader_compile_options *options,
 249                   const VkAllocationCallbacks *alloc)
 250 {
 251    if (options->optimize) {
 252       /* ignore the key for the first pass of optimization */
 253       ir3_optimize_nir(&shader->ir3_shader, shader->ir3_shader.nir, NULL);
 254
 255       if (unlikely(dev->physical_device->instance->debug_flags &
 256                    TU_DEBUG_NIR)) {
 257          fprintf(stderr, "optimized nir:\n");
 258          nir_print_shader(shader->ir3_shader.nir, stderr);
 259       }
 260    }
 261
 262    shader->binary = tu_compile_shader_variant(
 263       &shader->ir3_shader, &options->key, NULL, &shader->variants[0]);
 264    if (!shader->binary)
 265       return VK_ERROR_OUT_OF_HOST_MEMORY;
 266
 267    /* compile another variant for the binning pass */
 268    if (options->include_binning_pass &&
 269        shader->ir3_shader.type == MESA_SHADER_VERTEX) {
 270       shader->binning_binary = tu_compile_shader_variant(
 271          &shader->ir3_shader, &options->key, &shader->variants[0],
 272          &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 }