src/gallium/drivers/freedreno/ir3/ir3_cmdline.c

   1 /*
   2  * Copyright (C) 2014 Rob Clark <robclark@freedesktop.org>
   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 FROM,
  20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  21  * SOFTWARE.
  22  *
  23  * Authors:
  24  *    Rob Clark <robclark@freedesktop.org>
  25  */
  26
  27 #include <sys/types.h>
  28 #include <sys/stat.h>
  29 #include <sys/mman.h>
  30 #include <fcntl.h>
  31 #include <stdint.h>
  32 #include <stdlib.h>
  33 #include <stdio.h>
  34 #include <err.h>
  35
  36 #include "tgsi/tgsi_parse.h"
  37 #include "tgsi/tgsi_text.h"
  38 #include "tgsi/tgsi_dump.h"
  39
  40 #include "ir3_compiler.h"
  41 #include "ir3_nir.h"
  42 #include "instr-a3xx.h"
  43 #include "ir3.h"
  44
  45 #include "compiler/glsl/standalone.h"
  46 #include "compiler/glsl/glsl_to_nir.h"
  47 #include "compiler/glsl/gl_nir.h"
  48 #include "compiler/nir_types.h"
  49 #include "compiler/spirv/nir_spirv.h"
  50
  51 static void dump_info(struct ir3_shader_variant *so, const char *str)
  52 {
  53         uint32_t *bin;
  54         const char *type = ir3_shader_stage(so->shader);
  55         bin = ir3_shader_assemble(so, so->shader->compiler->gpu_id);
  56         debug_printf("; %s: %s\n", type, str);
  57         ir3_shader_disasm(so, bin, stdout);
  58         free(bin);
  59 }
  60
  61 static void
  62 insert_sorted(struct exec_list *var_list, nir_variable *new_var)
  63 {
  64         nir_foreach_variable(var, var_list) {
  65                 if (var->data.location > new_var->data.location) {
  66                         exec_node_insert_node_before(&var->node, &new_var->node);
  67                         return;
  68                 }
  69         }
  70         exec_list_push_tail(var_list, &new_var->node);
  71 }
  72
  73 static void
  74 sort_varyings(struct exec_list *var_list)
  75 {
  76         struct exec_list new_list;
  77         exec_list_make_empty(&new_list);
  78         nir_foreach_variable_safe(var, var_list) {
  79                 exec_node_remove(&var->node);
  80                 insert_sorted(&new_list, var);
  81         }
  82         exec_list_move_nodes_to(&new_list, var_list);
  83 }
  84
  85 static void
  86 fixup_varying_slots(struct exec_list *var_list)
  87 {
  88         nir_foreach_variable(var, var_list) {
  89                 if (var->data.location >= VARYING_SLOT_VAR0) {
  90                         var->data.location += 9;
  91                 } else if ((var->data.location >= VARYING_SLOT_TEX0) &&
  92                                 (var->data.location <= VARYING_SLOT_TEX7)) {
  93                         var->data.location += VARYING_SLOT_VAR0 - VARYING_SLOT_TEX0;
  94                 }
  95         }
  96 }
  97
  98 static struct ir3_compiler *compiler;
  99
 100 static nir_shader *
 101 load_glsl(unsigned num_files, char* const* files, gl_shader_stage stage)
 102 {
 103         static const struct standalone_options options = {
 104                         .glsl_version = 460,
 105                         .do_link = true,
 106         };
 107         struct gl_shader_program *prog;
 108         const nir_shader_compiler_options *nir_options =
 109                         ir3_get_compiler_options(compiler);
 110
 111         prog = standalone_compile_shader(&options, num_files, files);
 112         if (!prog)
 113                 errx(1, "couldn't parse `%s'", files[0]);
 114
 115         nir_shader *nir = glsl_to_nir(prog, stage, nir_options);
 116
 117         /* required NIR passes: */
 118         if (nir_options->lower_all_io_to_temps ||
 119                         nir->info.stage == MESA_SHADER_VERTEX ||
 120                         nir->info.stage == MESA_SHADER_GEOMETRY) {
 121                 NIR_PASS_V(nir, nir_lower_io_to_temporaries,
 122                                 nir_shader_get_entrypoint(nir),
 123                                 true, true);
 124         } else if (nir->info.stage == MESA_SHADER_FRAGMENT) {
 125                 NIR_PASS_V(nir, nir_lower_io_to_temporaries,
 126                                 nir_shader_get_entrypoint(nir),
 127                                 true, false);
 128         }
 129
 130         NIR_PASS_V(nir, nir_lower_global_vars_to_local);
 131         NIR_PASS_V(nir, nir_split_var_copies);
 132         NIR_PASS_V(nir, nir_lower_var_copies);
 133
 134         NIR_PASS_V(nir, nir_split_var_copies);
 135         NIR_PASS_V(nir, nir_lower_var_copies);
 136         nir_print_shader(nir, stdout);
 137         NIR_PASS_V(nir, gl_nir_lower_atomics, prog, true);
 138         NIR_PASS_V(nir, nir_lower_atomics_to_ssbo, 8);
 139         nir_print_shader(nir, stdout);
 140
 141         switch (stage) {
 142         case MESA_SHADER_VERTEX:
 143                 nir_assign_var_locations(&nir->inputs,
 144                                 &nir->num_inputs,
 145                                 ir3_glsl_type_size);
 146
 147                 /* Re-lower global vars, to deal with any dead VS inputs. */
 148                 NIR_PASS_V(nir, nir_lower_global_vars_to_local);
 149
 150                 sort_varyings(&nir->outputs);
 151                 nir_assign_var_locations(&nir->outputs,
 152                                 &nir->num_outputs,
 153                                 ir3_glsl_type_size);
 154                 fixup_varying_slots(&nir->outputs);
 155                 break;
 156         case MESA_SHADER_FRAGMENT:
 157                 sort_varyings(&nir->inputs);
 158                 nir_assign_var_locations(&nir->inputs,
 159                                 &nir->num_inputs,
 160                                 ir3_glsl_type_size);
 161                 fixup_varying_slots(&nir->inputs);
 162                 nir_assign_var_locations(&nir->outputs,
 163                                 &nir->num_outputs,
 164                                 ir3_glsl_type_size);
 165                 break;
 166         case MESA_SHADER_COMPUTE:
 167                 break;
 168         default:
 169                 errx(1, "unhandled shader stage: %d", stage);
 170         }
 171
 172         nir_assign_var_locations(&nir->uniforms,
 173                         &nir->num_uniforms,
 174                         ir3_glsl_type_size);
 175
 176         NIR_PASS_V(nir, nir_lower_system_values);
 177         NIR_PASS_V(nir, nir_lower_io, nir_var_all, ir3_glsl_type_size, 0);
 178         NIR_PASS_V(nir, gl_nir_lower_samplers, prog);
 179
 180         return nir;
 181 }
 182
 183 static int
 184 read_file(const char *filename, void **ptr, size_t *size)
 185 {
 186         int fd, ret;
 187         struct stat st;
 188
 189         *ptr = MAP_FAILED;
 190
 191         fd = open(filename, O_RDONLY);
 192         if (fd == -1) {
 193                 warnx("couldn't open `%s'", filename);
 194                 return 1;
 195         }
 196
 197         ret = fstat(fd, &st);
 198         if (ret)
 199                 errx(1, "couldn't stat `%s'", filename);
 200
 201         *size = st.st_size;
 202         *ptr = mmap(NULL, st.st_size, PROT_READ, MAP_SHARED, fd, 0);
 203         if (*ptr == MAP_FAILED)
 204                 errx(1, "couldn't map `%s'", filename);
 205
 206         close(fd);
 207
 208         return 0;
 209 }
 210
 211 static void debug_func(void *priv, enum nir_spirv_debug_level level,
 212                 size_t spirv_offset, const char *message)
 213 {
 214 //      printf("%s\n", message);
 215 }
 216
 217 static nir_shader *
 218 load_spirv(const char *filename, const char *entry, gl_shader_stage stage)
 219 {
 220         const struct spirv_to_nir_options spirv_options = {
 221                 /* these caps are just make-believe */
 222                 .caps = {
 223                         .draw_parameters = true,
 224                         .float64 = true,
 225                         .image_read_without_format = true,
 226                         .image_write_without_format = true,
 227                         .int64 = true,
 228                         .variable_pointers = true,
 229                 },
 230                 .lower_workgroup_access_to_offsets = true,
 231                 .debug = {
 232                         .func = debug_func,
 233                 }
 234         };
 235         nir_function *entry_point;
 236         void *buf;
 237         size_t size;
 238
 239         read_file(filename, &buf, &size);
 240
 241         entry_point = spirv_to_nir(buf, size / 4,
 242                         NULL, 0, /* spec_entries */
 243                         stage, entry,
 244                         &spirv_options,
 245                         ir3_get_compiler_options(compiler));
 246
 247         nir_print_shader(entry_point->shader, stdout);
 248
 249         return entry_point->shader;
 250 }
 251
 252 static void print_usage(void)
 253 {
 254         printf("Usage: ir3_compiler [OPTIONS]... <file.tgsi | file.spv entry_point | (file.vert | file.frag)*>\n");
 255         printf("    --verbose         - verbose compiler/debug messages\n");
 256         printf("    --binning-pass    - generate binning pass shader (VERT)\n");
 257         printf("    --color-two-side  - emulate two-sided color (FRAG)\n");
 258         printf("    --half-precision  - use half-precision\n");
 259         printf("    --saturate-s MASK - bitmask of samplers to saturate S coord\n");
 260         printf("    --saturate-t MASK - bitmask of samplers to saturate T coord\n");
 261         printf("    --saturate-r MASK - bitmask of samplers to saturate R coord\n");
 262         printf("    --astc-srgb MASK  - bitmask of samplers to enable astc-srgb workaround\n");
 263         printf("    --stream-out      - enable stream-out (aka transform feedback)\n");
 264         printf("    --ucp MASK        - bitmask of enabled user-clip-planes\n");
 265         printf("    --gpu GPU_ID      - specify gpu-id (default 320)\n");
 266         printf("    --help            - show this message\n");
 267 }
 268
 269 int main(int argc, char **argv)
 270 {
 271         int ret = 0, n = 1;
 272         char *filenames[2];
 273         int num_files = 0;
 274         unsigned stage = 0;
 275         struct ir3_shader_variant v;
 276         struct ir3_shader s;
 277         struct ir3_shader_key key = {};
 278         /* TODO cmdline option to target different gpus: */
 279         unsigned gpu_id = 320;
 280         const char *info;
 281         const char *entry;
 282         void *ptr;
 283         bool from_spirv = false;
 284         size_t size;
 285
 286         memset(&s, 0, sizeof(s));
 287         memset(&v, 0, sizeof(v));
 288
 289         /* cmdline args which impact shader variant get spit out in a
 290          * comment on the first line..  a quick/dirty way to preserve
 291          * that info so when ir3test recompiles the shader with a new
 292          * compiler version, we use the same shader-key settings:
 293          */
 294         debug_printf("; options:");
 295
 296         while (n < argc) {
 297                 if (!strcmp(argv[n], "--verbose")) {
 298                         ir3_shader_debug |= IR3_DBG_OPTMSGS | IR3_DBG_DISASM;
 299                         n++;
 300                         continue;
 301                 }
 302
 303                 if (!strcmp(argv[n], "--binning-pass")) {
 304                         debug_printf(" %s", argv[n]);
 305                         v.binning_pass = true;
 306                         n++;
 307                         continue;
 308                 }
 309
 310                 if (!strcmp(argv[n], "--color-two-side")) {
 311                         debug_printf(" %s", argv[n]);
 312                         key.color_two_side = true;
 313                         n++;
 314                         continue;
 315                 }
 316
 317                 if (!strcmp(argv[n], "--half-precision")) {
 318                         debug_printf(" %s", argv[n]);
 319                         key.half_precision = true;
 320                         n++;
 321                         continue;
 322                 }
 323
 324                 if (!strcmp(argv[n], "--saturate-s")) {
 325                         debug_printf(" %s %s", argv[n], argv[n+1]);
 326                         key.vsaturate_s = key.fsaturate_s = strtol(argv[n+1], NULL, 0);
 327                         n += 2;
 328                         continue;
 329                 }
 330
 331                 if (!strcmp(argv[n], "--saturate-t")) {
 332                         debug_printf(" %s %s", argv[n], argv[n+1]);
 333                         key.vsaturate_t = key.fsaturate_t = strtol(argv[n+1], NULL, 0);
 334                         n += 2;
 335                         continue;
 336                 }
 337
 338                 if (!strcmp(argv[n], "--saturate-r")) {
 339                         debug_printf(" %s %s", argv[n], argv[n+1]);
 340                         key.vsaturate_r = key.fsaturate_r = strtol(argv[n+1], NULL, 0);
 341                         n += 2;
 342                         continue;
 343                 }
 344
 345                 if (!strcmp(argv[n], "--astc-srgb")) {
 346                         debug_printf(" %s %s", argv[n], argv[n+1]);
 347                         key.vastc_srgb = key.fastc_srgb = strtol(argv[n+1], NULL, 0);
 348                         n += 2;
 349                         continue;
 350                 }
 351
 352                 if (!strcmp(argv[n], "--stream-out")) {
 353                         struct ir3_stream_output_info *so = &s.stream_output;
 354                         debug_printf(" %s", argv[n]);
 355                         /* TODO more dynamic config based on number of outputs, etc
 356                          * rather than just hard-code for first output:
 357                          */
 358                         so->num_outputs = 1;
 359                         so->stride[0] = 4;
 360                         so->output[0].register_index = 0;
 361                         so->output[0].start_component = 0;
 362                         so->output[0].num_components = 4;
 363                         so->output[0].output_buffer = 0;
 364                         so->output[0].dst_offset = 2;
 365                         so->output[0].stream = 0;
 366                         n++;
 367                         continue;
 368                 }
 369
 370                 if (!strcmp(argv[n], "--ucp")) {
 371                         debug_printf(" %s %s", argv[n], argv[n+1]);
 372                         key.ucp_enables = strtol(argv[n+1], NULL, 0);
 373                         n += 2;
 374                         continue;
 375                 }
 376
 377                 if (!strcmp(argv[n], "--gpu")) {
 378                         debug_printf(" %s %s", argv[n], argv[n+1]);
 379                         gpu_id = strtol(argv[n+1], NULL, 0);
 380                         n += 2;
 381                         continue;
 382                 }
 383
 384                 if (!strcmp(argv[n], "--help")) {
 385                         print_usage();
 386                         return 0;
 387                 }
 388
 389                 break;
 390         }
 391         debug_printf("\n");
 392
 393         while (n < argc) {
 394                 char *filename = argv[n];
 395                 char *ext = strrchr(filename, '.');
 396
 397                 if (strcmp(ext, ".tgsi") == 0) {
 398                         if (num_files != 0)
 399                                 errx(1, "in TGSI mode, only a single file may be specified");
 400                         s.from_tgsi = true;
 401                 } else if (strcmp(ext, ".spv") == 0) {
 402                         if (num_files != 0)
 403                                 errx(1, "in SPIR-V mode, only a single file may be specified");
 404                         stage = MESA_SHADER_COMPUTE;
 405                         from_spirv = true;
 406                         filenames[num_files++] = filename;
 407                         n++;
 408                         if (n == argc)
 409                                 errx(1, "in SPIR-V mode, an entry point must be specified");
 410                         entry = argv[n];
 411                         n++;
 412                 } else if (strcmp(ext, ".comp") == 0) {
 413                         if (s.from_tgsi || from_spirv)
 414                                 errx(1, "cannot mix GLSL/TGSI/SPIRV");
 415                         if (num_files >= ARRAY_SIZE(filenames))
 416                                 errx(1, "too many GLSL files");
 417                         stage = MESA_SHADER_COMPUTE;
 418                 } else if (strcmp(ext, ".frag") == 0) {
 419                         if (s.from_tgsi || from_spirv)
 420                                 errx(1, "cannot mix GLSL/TGSI/SPIRV");
 421                         if (num_files >= ARRAY_SIZE(filenames))
 422                                 errx(1, "too many GLSL files");
 423                         stage = MESA_SHADER_FRAGMENT;
 424                 } else if (strcmp(ext, ".vert") == 0) {
 425                         if (s.from_tgsi)
 426                                 errx(1, "cannot mix GLSL and TGSI");
 427                         if (num_files >= ARRAY_SIZE(filenames))
 428                                 errx(1, "too many GLSL files");
 429                         stage = MESA_SHADER_VERTEX;
 430                 } else {
 431                         print_usage();
 432                         return -1;
 433                 }
 434
 435                 filenames[num_files++] = filename;
 436
 437                 n++;
 438         }
 439
 440         nir_shader *nir;
 441
 442         compiler = ir3_compiler_create(NULL, gpu_id);
 443
 444         if (s.from_tgsi) {
 445                 struct tgsi_token toks[65536];
 446
 447                 ret = read_file(filenames[0], &ptr, &size);
 448                 if (ret) {
 449                         print_usage();
 450                         return ret;
 451                 }
 452
 453                 if (ir3_shader_debug & IR3_DBG_OPTMSGS)
 454                         debug_printf("%s\n", (char *)ptr);
 455
 456                 if (!tgsi_text_translate(ptr, toks, ARRAY_SIZE(toks)))
 457                         errx(1, "could not parse `%s'", filenames[0]);
 458
 459                 if (ir3_shader_debug & IR3_DBG_OPTMSGS)
 460                         tgsi_dump(toks, 0);
 461
 462                 nir = ir3_tgsi_to_nir(toks);
 463                 NIR_PASS_V(nir, nir_lower_global_vars_to_local);
 464         } else if (from_spirv) {
 465                 nir = load_spirv(filenames[0], entry, stage);
 466
 467                 NIR_PASS_V(nir, nir_lower_io, nir_var_all, ir3_glsl_type_size,
 468                                 (nir_lower_io_options)0);
 469
 470                 /* TODO do this somewhere else */
 471                 nir_lower_int64(nir, ~0);
 472                 nir_lower_system_values(nir);
 473         } else if (num_files > 0) {
 474                 nir = load_glsl(num_files, filenames, stage);
 475         } else {
 476                 print_usage();
 477                 return -1;
 478         }
 479
 480         s.compiler = compiler;
 481         s.nir = ir3_optimize_nir(&s, nir, NULL);
 482
 483         v.key = key;
 484         v.shader = &s;
 485         s.type = v.type = nir->info.stage;
 486
 487         info = "NIR compiler";
 488         ret = ir3_compile_shader_nir(s.compiler, &v);
 489         if (ret) {
 490                 fprintf(stderr, "compiler failed!\n");
 491                 return ret;
 492         }
 493         dump_info(&v, info);
 494 }