1/* 2 * Copyright © Microsoft Corporation 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#include "spirv_to_dxil.h" 25#include "nir_to_dxil.h" 26#include "dxil_nir.h" 27#include "shader_enums.h" 28#include "spirv/nir_spirv.h" 29#include "util/blob.h" 30 31#include "git_sha1.h" 32#include "vulkan/vulkan.h" 33 34static void 35shared_var_info(const struct glsl_type* type, unsigned* size, unsigned* align) 36{ 37 assert(glsl_type_is_vector_or_scalar(type)); 38 39 uint32_t comp_size = glsl_type_is_boolean(type) ? 4 : glsl_get_bit_size(type) / 8; 40 unsigned length = glsl_get_vector_elements(type); 41 *size = comp_size * length; 42 *align = comp_size; 43} 44 45static nir_variable * 46add_runtime_data_var(nir_shader *nir, unsigned desc_set, unsigned binding) 47{ 48 unsigned runtime_data_size = 49 nir->info.stage == MESA_SHADER_COMPUTE 50 ? sizeof(struct dxil_spirv_compute_runtime_data) 51 : sizeof(struct dxil_spirv_vertex_runtime_data); 52 53 const struct glsl_type *array_type = 54 glsl_array_type(glsl_uint_type(), runtime_data_size / sizeof(unsigned), 55 sizeof(unsigned)); 56 const struct glsl_struct_field field = {array_type, "arr"}; 57 nir_variable *var = nir_variable_create( 58 nir, nir_var_mem_ubo, 59 glsl_struct_type(&field, 1, "runtime_data", false), "runtime_data"); 60 var->data.descriptor_set = desc_set; 61 // Check that desc_set fits on descriptor_set 62 assert(var->data.descriptor_set == desc_set); 63 var->data.binding = binding; 64 var->data.how_declared = nir_var_hidden; 65 return var; 66} 67 68struct lower_system_values_data { 69 nir_address_format ubo_format; 70 unsigned desc_set; 71 unsigned binding; 72}; 73 74static bool 75lower_shader_system_values(struct nir_builder *builder, nir_instr *instr, 76 void *cb_data) 77{ 78 if (instr->type != nir_instr_type_intrinsic) { 79 return false; 80 } 81 82 nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr); 83 84 /* All the intrinsics we care about are loads */ 85 if (!nir_intrinsic_infos[intrin->intrinsic].has_dest) 86 return false; 87 88 assert(intrin->dest.is_ssa); 89 90 int offset = 0; 91 switch (intrin->intrinsic) { 92 case nir_intrinsic_load_num_workgroups: 93 offset = 94 offsetof(struct dxil_spirv_compute_runtime_data, group_count_x); 95 break; 96 case nir_intrinsic_load_first_vertex: 97 offset = offsetof(struct dxil_spirv_vertex_runtime_data, first_vertex); 98 break; 99 case nir_intrinsic_load_is_indexed_draw: 100 offset = 101 offsetof(struct dxil_spirv_vertex_runtime_data, is_indexed_draw); 102 break; 103 case nir_intrinsic_load_base_instance: 104 offset = offsetof(struct dxil_spirv_vertex_runtime_data, base_instance); 105 break; 106 default: 107 return false; 108 } 109 110 struct lower_system_values_data *data = 111 (struct lower_system_values_data *)cb_data; 112 113 builder->cursor = nir_after_instr(instr); 114 nir_address_format ubo_format = data->ubo_format; 115 116 nir_ssa_def *index = nir_vulkan_resource_index( 117 builder, nir_address_format_num_components(ubo_format), 118 nir_address_format_bit_size(ubo_format), 119 nir_imm_int(builder, 0), 120 .desc_set = data->desc_set, .binding = data->binding, 121 .desc_type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER); 122 123 nir_ssa_def *load_desc = nir_load_vulkan_descriptor( 124 builder, nir_address_format_num_components(ubo_format), 125 nir_address_format_bit_size(ubo_format), 126 index, .desc_type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER); 127 128 nir_ssa_def *load_data = build_load_ubo_dxil( 129 builder, nir_channel(builder, load_desc, 0), 130 nir_imm_int(builder, offset), 131 nir_dest_num_components(intrin->dest), nir_dest_bit_size(intrin->dest)); 132 133 nir_ssa_def_rewrite_uses(&intrin->dest.ssa, load_data); 134 nir_instr_remove(instr); 135 return true; 136} 137 138static bool 139dxil_spirv_nir_lower_shader_system_values(nir_shader *shader, 140 nir_address_format ubo_format, 141 unsigned desc_set, unsigned binding) 142{ 143 struct lower_system_values_data data = { 144 .ubo_format = ubo_format, 145 .desc_set = desc_set, 146 .binding = binding, 147 }; 148 return nir_shader_instructions_pass(shader, lower_shader_system_values, 149 nir_metadata_block_index | 150 nir_metadata_dominance | 151 nir_metadata_loop_analysis, 152 &data); 153} 154 155bool 156spirv_to_dxil(const uint32_t *words, size_t word_count, 157 struct dxil_spirv_specialization *specializations, 158 unsigned int num_specializations, dxil_spirv_shader_stage stage, 159 const char *entry_point_name, 160 const struct dxil_spirv_runtime_conf *conf, 161 struct dxil_spirv_object *out_dxil) 162{ 163 if (stage == MESA_SHADER_NONE || stage == MESA_SHADER_KERNEL) 164 return false; 165 166 struct spirv_to_nir_options spirv_opts = { 167 .ubo_addr_format = nir_address_format_32bit_index_offset, 168 .ssbo_addr_format = nir_address_format_32bit_index_offset, 169 .shared_addr_format = nir_address_format_32bit_offset_as_64bit, 170 171 // use_deref_buffer_array_length + nir_lower_explicit_io force 172 // get_ssbo_size to take in the return from load_vulkan_descriptor 173 // instead of vulkan_resource_index. This makes it much easier to 174 // get the DXIL handle for the SSBO. 175 .use_deref_buffer_array_length = true 176 }; 177 178 glsl_type_singleton_init_or_ref(); 179 180 struct nir_shader_compiler_options nir_options = *dxil_get_nir_compiler_options(); 181 // We will manually handle base_vertex when vertex_id and instance_id have 182 // have been already converted to zero-base. 183 nir_options.lower_base_vertex = !conf->zero_based_vertex_instance_id; 184 185 nir_shader *nir = spirv_to_nir( 186 words, word_count, (struct nir_spirv_specialization *)specializations, 187 num_specializations, (gl_shader_stage)stage, entry_point_name, 188 &spirv_opts, &nir_options); 189 if (!nir) { 190 glsl_type_singleton_decref(); 191 return false; 192 } 193 194 nir_validate_shader(nir, 195 "Validate before feeding NIR to the DXIL compiler"); 196 197 const struct nir_lower_sysvals_to_varyings_options sysvals_to_varyings = { 198 .frag_coord = true, 199 .point_coord = true, 200 }; 201 NIR_PASS_V(nir, nir_lower_sysvals_to_varyings, &sysvals_to_varyings); 202 203 NIR_PASS_V(nir, nir_lower_system_values); 204 205 if (conf->zero_based_vertex_instance_id) { 206 // vertex_id and instance_id should have already been transformed to 207 // base zero before spirv_to_dxil was called. Therefore, we can zero out 208 // base/firstVertex/Instance. 209 gl_system_value system_values[] = {SYSTEM_VALUE_FIRST_VERTEX, 210 SYSTEM_VALUE_BASE_VERTEX, 211 SYSTEM_VALUE_BASE_INSTANCE}; 212 NIR_PASS_V(nir, dxil_nir_lower_system_values_to_zero, system_values, 213 ARRAY_SIZE(system_values)); 214 } 215 216 bool requires_runtime_data = false; 217 NIR_PASS(requires_runtime_data, nir, 218 dxil_spirv_nir_lower_shader_system_values, 219 spirv_opts.ubo_addr_format, 220 conf->runtime_data_cbv.register_space, 221 conf->runtime_data_cbv.base_shader_register); 222 if (requires_runtime_data) { 223 add_runtime_data_var(nir, conf->runtime_data_cbv.register_space, 224 conf->runtime_data_cbv.base_shader_register); 225 } 226 227 NIR_PASS_V(nir, nir_split_per_member_structs); 228 229 NIR_PASS_V(nir, nir_lower_explicit_io, nir_var_mem_ubo | nir_var_mem_ssbo, 230 nir_address_format_32bit_index_offset); 231 232 if (!nir->info.shared_memory_explicit_layout) { 233 NIR_PASS_V(nir, nir_lower_vars_to_explicit_types, nir_var_mem_shared, 234 shared_var_info); 235 } 236 NIR_PASS_V(nir, nir_lower_explicit_io, nir_var_mem_shared, 237 nir_address_format_32bit_offset_as_64bit); 238 239 nir_variable_mode nir_var_function_temp = 240 nir_var_shader_in | nir_var_shader_out; 241 NIR_PASS_V(nir, nir_lower_variable_initializers, 242 nir_var_function_temp); 243 NIR_PASS_V(nir, nir_opt_deref); 244 NIR_PASS_V(nir, nir_lower_returns); 245 NIR_PASS_V(nir, nir_inline_functions); 246 NIR_PASS_V(nir, nir_lower_variable_initializers, 247 ~nir_var_function_temp); 248 249 // Pick off the single entrypoint that we want. 250 nir_function *entrypoint; 251 foreach_list_typed_safe(nir_function, func, node, &nir->functions) { 252 if (func->is_entrypoint) 253 entrypoint = func; 254 else 255 exec_node_remove(&func->node); 256 } 257 assert(exec_list_length(&nir->functions) == 1); 258 259 NIR_PASS_V(nir, nir_lower_clip_cull_distance_arrays); 260 NIR_PASS_V(nir, nir_lower_io_to_temporaries, entrypoint->impl, true, true); 261 NIR_PASS_V(nir, nir_lower_global_vars_to_local); 262 NIR_PASS_V(nir, nir_split_var_copies); 263 NIR_PASS_V(nir, nir_lower_var_copies); 264 NIR_PASS_V(nir, nir_lower_io_arrays_to_elements_no_indirects, false); 265 266 NIR_PASS_V(nir, nir_lower_alu_to_scalar, NULL, NULL); 267 NIR_PASS_V(nir, nir_opt_dce); 268 NIR_PASS_V(nir, dxil_nir_lower_double_math); 269 270 { 271 bool progress; 272 do 273 { 274 progress = false; 275 NIR_PASS(progress, nir, nir_copy_prop); 276 NIR_PASS(progress, nir, nir_opt_copy_prop_vars); 277 NIR_PASS(progress, nir, nir_opt_deref); 278 NIR_PASS(progress, nir, nir_opt_dce); 279 NIR_PASS(progress, nir, nir_opt_undef); 280 NIR_PASS(progress, nir, nir_opt_constant_folding); 281 NIR_PASS(progress, nir, nir_opt_cse); 282 if (nir_opt_trivial_continues(nir)) { 283 progress = true; 284 NIR_PASS(progress, nir, nir_copy_prop); 285 NIR_PASS(progress, nir, nir_opt_dce); 286 } 287 NIR_PASS(progress, nir, nir_lower_vars_to_ssa); 288 NIR_PASS(progress, nir, nir_opt_algebraic); 289 } while (progress); 290 } 291 292 NIR_PASS_V(nir, nir_lower_readonly_images_to_tex, true); 293 nir_lower_tex_options lower_tex_options = {0}; 294 NIR_PASS_V(nir, nir_lower_tex, &lower_tex_options); 295 296 NIR_PASS_V(nir, dxil_nir_split_clip_cull_distance); 297 NIR_PASS_V(nir, dxil_nir_lower_loads_stores_to_dxil); 298 NIR_PASS_V(nir, dxil_nir_create_bare_samplers); 299 NIR_PASS_V(nir, dxil_nir_lower_bool_input); 300 301 nir_shader_gather_info(nir, nir_shader_get_entrypoint(nir)); 302 303 nir->info.inputs_read = 304 dxil_reassign_driver_locations(nir, nir_var_shader_in, 0); 305 306 if (stage != MESA_SHADER_FRAGMENT) { 307 nir->info.outputs_written = 308 dxil_reassign_driver_locations(nir, nir_var_shader_out, 0); 309 } else { 310 dxil_sort_ps_outputs(nir); 311 } 312 313 struct nir_to_dxil_options opts = {.vulkan_environment = true}; 314 315 struct blob dxil_blob; 316 if (!nir_to_dxil(nir, &opts, &dxil_blob)) { 317 if (dxil_blob.allocated) 318 blob_finish(&dxil_blob); 319 glsl_type_singleton_decref(); 320 return false; 321 } 322 323 out_dxil->metadata.requires_runtime_data = requires_runtime_data; 324 blob_finish_get_buffer(&dxil_blob, &out_dxil->binary.buffer, 325 &out_dxil->binary.size); 326 327 glsl_type_singleton_decref(); 328 return true; 329} 330 331void 332spirv_to_dxil_free(struct dxil_spirv_object *dxil) 333{ 334 free(dxil->binary.buffer); 335} 336 337uint64_t 338spirv_to_dxil_get_version() 339{ 340 const char sha1[] = MESA_GIT_SHA1; 341 const char* dash = strchr(sha1, '-'); 342 if (dash) { 343 return strtoull(dash + 1, NULL, 16); 344 } 345 return 0; 346} 347