1/* 2 * Copyright (C) 2021 Collabora, Ltd. 3 * Copyright (C) 2019 Ryan Houdek <Sonicadvance1@gmail.com> 4 * Copyright (C) 2014 Rob Clark <robclark@freedesktop.org> 5 * Copyright © 2015 Red Hat 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a 8 * copy of this software and associated documentation files (the "Software"), 9 * to deal in the Software without restriction, including without limitation 10 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 11 * and/or sell copies of the Software, and to permit persons to whom the 12 * Software is furnished to do so, subject to the following conditions: 13 * 14 * The above copyright notice and this permission notice (including the next 15 * paragraph) shall be included in all copies or substantial portions of the 16 * Software. 17 * 18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 21 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 23 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 24 * SOFTWARE. 25 */ 26 27#include <getopt.h> 28#include <string.h> 29#include "disassemble.h" 30#include "compiler.h" 31 32#include "main/mtypes.h" 33#include "compiler/glsl/standalone.h" 34#include "compiler/glsl/glsl_to_nir.h" 35#include "compiler/glsl/gl_nir.h" 36#include "compiler/nir_types.h" 37#include "util/u_dynarray.h" 38#include "bifrost_compile.h" 39 40unsigned gpu_id = 0x7212; 41int verbose = 0; 42 43static gl_shader_stage 44filename_to_stage(const char *stage) 45{ 46 const char *ext = strrchr(stage, '.'); 47 48 if (ext == NULL) { 49 fprintf(stderr, "No extension found in %s\n", stage); 50 exit(1); 51 } 52 53 if (!strcmp(ext, ".cs") || !strcmp(ext, ".comp")) 54 return MESA_SHADER_COMPUTE; 55 else if (!strcmp(ext, ".vs") || !strcmp(ext, ".vert")) 56 return MESA_SHADER_VERTEX; 57 else if (!strcmp(ext, ".fs") || !strcmp(ext, ".frag")) 58 return MESA_SHADER_FRAGMENT; 59 else { 60 fprintf(stderr, "Invalid extension %s\n", ext); 61 exit(1); 62 } 63 64 unreachable("Should've returned or bailed"); 65} 66 67static int 68st_packed_uniforms_type_size(const struct glsl_type *type, bool bindless) 69{ 70 return glsl_count_dword_slots(type, bindless); 71} 72 73static int 74glsl_type_size(const struct glsl_type *type, bool bindless) 75{ 76 return glsl_count_attribute_slots(type, false); 77} 78 79static void 80insert_sorted(struct exec_list *var_list, nir_variable *new_var) 81{ 82 nir_foreach_variable_in_list (var, var_list) { 83 if (var->data.location > new_var->data.location) { 84 exec_node_insert_node_before(&var->node, &new_var->node); 85 return; 86 } 87 } 88 exec_list_push_tail(var_list, &new_var->node); 89} 90 91static void 92sort_varyings(nir_shader *nir, nir_variable_mode mode) 93{ 94 struct exec_list new_list; 95 exec_list_make_empty(&new_list); 96 nir_foreach_variable_with_modes_safe (var, nir, mode) { 97 exec_node_remove(&var->node); 98 insert_sorted(&new_list, var); 99 } 100 exec_list_append(&nir->variables, &new_list); 101} 102 103static void 104fixup_varying_slots(nir_shader *nir, nir_variable_mode mode) 105{ 106 nir_foreach_variable_with_modes (var, nir, mode) { 107 if (var->data.location >= VARYING_SLOT_VAR0) { 108 var->data.location += 9; 109 } else if ((var->data.location >= VARYING_SLOT_TEX0) && 110 (var->data.location <= VARYING_SLOT_TEX7)) { 111 var->data.location += VARYING_SLOT_VAR0 - VARYING_SLOT_TEX0; 112 } 113 } 114} 115 116static void 117compile_shader(int stages, char **files) 118{ 119 struct gl_shader_program *prog; 120 nir_shader *nir[MESA_SHADER_COMPUTE + 1]; 121 unsigned shader_types[MESA_SHADER_COMPUTE + 1]; 122 123 if (stages > MESA_SHADER_COMPUTE) { 124 fprintf(stderr, "Too many stages"); 125 exit(1); 126 } 127 128 for (unsigned i = 0; i < stages; ++i) 129 shader_types[i] = filename_to_stage(files[i]); 130 131 struct standalone_options options = { 132 .glsl_version = 300, /* ES - needed for precision */ 133 .do_link = true, 134 .lower_precision = true 135 }; 136 137 static struct gl_context local_ctx; 138 139 prog = standalone_compile_shader(&options, stages, files, &local_ctx); 140 141 for (unsigned i = 0; i < stages; ++i) { 142 gl_shader_stage stage = shader_types[i]; 143 prog->_LinkedShaders[stage]->Program->info.stage = stage; 144 } 145 146 struct util_dynarray binary; 147 148 util_dynarray_init(&binary, NULL); 149 150 for (unsigned i = 0; i < stages; ++i) { 151 nir[i] = glsl_to_nir(&local_ctx, prog, shader_types[i], &bifrost_nir_options); 152 153 if (shader_types[i] == MESA_SHADER_VERTEX) { 154 nir_assign_var_locations(nir[i], nir_var_shader_in, &nir[i]->num_inputs, 155 glsl_type_size); 156 sort_varyings(nir[i], nir_var_shader_out); 157 nir_assign_var_locations(nir[i], nir_var_shader_out, &nir[i]->num_outputs, 158 glsl_type_size); 159 fixup_varying_slots(nir[i], nir_var_shader_out); 160 } else if (shader_types[i] == MESA_SHADER_FRAGMENT) { 161 sort_varyings(nir[i], nir_var_shader_in); 162 nir_assign_var_locations(nir[i], nir_var_shader_in, &nir[i]->num_inputs, 163 glsl_type_size); 164 fixup_varying_slots(nir[i], nir_var_shader_in); 165 nir_assign_var_locations(nir[i], nir_var_shader_out, &nir[i]->num_outputs, 166 glsl_type_size); 167 } 168 169 nir_assign_var_locations(nir[i], nir_var_uniform, &nir[i]->num_uniforms, 170 glsl_type_size); 171 172 NIR_PASS_V(nir[i], nir_lower_global_vars_to_local); 173 NIR_PASS_V(nir[i], nir_lower_io_to_temporaries, nir_shader_get_entrypoint(nir[i]), true, i == 0); 174 NIR_PASS_V(nir[i], nir_opt_copy_prop_vars); 175 NIR_PASS_V(nir[i], nir_opt_combine_stores, nir_var_all); 176 177 NIR_PASS_V(nir[i], nir_lower_system_values); 178 NIR_PASS_V(nir[i], gl_nir_lower_samplers, prog); 179 NIR_PASS_V(nir[i], nir_split_var_copies); 180 NIR_PASS_V(nir[i], nir_lower_var_copies); 181 182 NIR_PASS_V(nir[i], nir_lower_io, nir_var_uniform, 183 st_packed_uniforms_type_size, 184 (nir_lower_io_options)0); 185 NIR_PASS_V(nir[i], nir_lower_uniforms_to_ubo, true, false); 186 187 /* before buffers and vars_to_ssa */ 188 NIR_PASS_V(nir[i], gl_nir_lower_images, true); 189 190 NIR_PASS_V(nir[i], gl_nir_lower_buffers, prog); 191 NIR_PASS_V(nir[i], nir_opt_constant_folding); 192 193 struct panfrost_compile_inputs inputs = { 194 .gpu_id = gpu_id, 195 }; 196 struct pan_shader_info info = { 0 }; 197 198 util_dynarray_clear(&binary); 199 bifrost_compile_shader_nir(nir[i], &inputs, &binary, &info); 200 201 char *fn = NULL; 202 asprintf(&fn, "shader_%u.bin", i); 203 assert(fn != NULL); 204 FILE *fp = fopen(fn, "wb"); 205 fwrite(binary.data, 1, binary.size, fp); 206 fclose(fp); 207 free(fn); 208 } 209 210 util_dynarray_fini(&binary); 211} 212 213#define BI_FOURCC(ch0, ch1, ch2, ch3) ( \ 214 (uint32_t)(ch0) | (uint32_t)(ch1) << 8 | \ 215 (uint32_t)(ch2) << 16 | (uint32_t)(ch3) << 24) 216 217static void 218disassemble(const char *filename) 219{ 220 FILE *fp = fopen(filename, "rb"); 221 assert(fp); 222 223 fseek(fp, 0, SEEK_END); 224 unsigned filesize = ftell(fp); 225 rewind(fp); 226 227 uint32_t *code = malloc(filesize); 228 unsigned res = fread(code, 1, filesize, fp); 229 if (res != filesize) { 230 printf("Couldn't read full file\n"); 231 } 232 233 fclose(fp); 234 235 if (filesize && code[0] == BI_FOURCC('M', 'B', 'S', '2')) { 236 for (int i = 0; i < filesize / 4; ++i) { 237 if (code[i] != BI_FOURCC('O', 'B', 'J', 'C')) 238 continue; 239 240 unsigned size = code[i + 1]; 241 unsigned offset = i + 2; 242 243 disassemble_bifrost(stdout, (uint8_t*)(code + offset), size, verbose); 244 } 245 } else { 246 disassemble_bifrost(stdout, (uint8_t*)code, filesize, verbose); 247 } 248 249 free(code); 250} 251 252int 253main(int argc, char **argv) 254{ 255 int c; 256 257 if (argc < 2) { 258 printf("Pass a command\n"); 259 exit(1); 260 } 261 262 static struct option longopts[] = { 263 { "id", optional_argument, NULL, 'i' }, 264 { "gpu", optional_argument, NULL, 'g' }, 265 { "verbose", no_argument, &verbose, 'v' }, 266 { NULL, 0, NULL, 0 } 267 }; 268 269 static struct { 270 const char *name; 271 unsigned major, minor; 272 } gpus[] = { 273 { "G71", 6, 0 }, 274 { "G72", 6, 2 }, 275 { "G51", 7, 0 }, 276 { "G76", 7, 1 }, 277 { "G52", 7, 2 }, 278 { "G31", 7, 3 }, 279 { "G77", 9, 0 }, 280 { "G57", 9, 1 }, 281 { "G78", 9, 2 }, 282 { "G57", 9, 3 }, 283 { "G68", 9, 4 }, 284 { "G78AE", 9, 5 }, 285 }; 286 287 while ((c = getopt_long(argc, argv, "v:", longopts, NULL)) != -1) { 288 289 switch (c) { 290 case 'i': 291 gpu_id = atoi(optarg); 292 293 if (!gpu_id) { 294 fprintf(stderr, "Expected GPU ID, got %s\n", optarg); 295 return 1; 296 } 297 298 break; 299 case 'g': 300 gpu_id = 0; 301 302 /* Compatibility with the Arm compiler */ 303 if (strncmp(optarg, "Mali-", 5) == 0) optarg += 5; 304 305 for (unsigned i = 0; i < ARRAY_SIZE(gpus); ++i) { 306 if (strcmp(gpus[i].name, optarg)) continue; 307 308 unsigned major = gpus[i].major; 309 unsigned minor = gpus[i].minor; 310 311 gpu_id = (major << 12) | (minor << 8); 312 break; 313 } 314 315 if (!gpu_id) { 316 fprintf(stderr, "Unknown GPU %s\n", optarg); 317 return 1; 318 } 319 320 break; 321 default: 322 break; 323 } 324 } 325 326 if (strcmp(argv[optind], "compile") == 0) 327 compile_shader(argc - optind - 1, &argv[optind + 1]); 328 else if (strcmp(argv[optind], "disasm") == 0) 329 disassemble(argv[optind + 1]); 330 else { 331 fprintf(stderr, "Unknown command. Valid: compile/disasm\n"); 332 return 1; 333 } 334 335 return 0; 336} 337