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/ir3_compiler.h"
  41 #include "ir3/ir3_gallium.h"
  42 #include "ir3/ir3_nir.h"
  43 #include "ir3/instr-a3xx.h"
  44 #include "ir3/ir3.h"
  45
  46 #include "compiler/glsl/standalone.h"
  47 #include "compiler/glsl/glsl_to_nir.h"
  48 #include "compiler/glsl/gl_nir.h"
  49 #include "compiler/nir_types.h"
  50 #include "compiler/spirv/nir_spirv.h"
  51
  52 static void dump_info(struct ir3_shader_variant *so, const char *str)
  53 {
  54         uint32_t *bin;
  55         const char *type = ir3_shader_stage(so->shader);
  56         bin = ir3_shader_assemble(so, so->shader->compiler->gpu_id);
  57         debug_printf("; %s: %s\n", type, str);
  58         ir3_shader_disasm(so, bin, stdout);
  59         free(bin);
  60 }
  61
  62 static void
  63 insert_sorted(struct exec_list *var_list, nir_variable *new_var)
  64 {
  65         nir_foreach_variable(var, var_list) {
  66                 if (var->data.location > new_var->data.location) {
  67                         exec_node_insert_node_before(&var->node, &new_var->node);
  68                         return;
  69                 }
  70         }
  71         exec_list_push_tail(var_list, &new_var->node);
  72 }
  73
  74 static void
  75 sort_varyings(struct exec_list *var_list)
  76 {
  77         struct exec_list new_list;
  78         exec_list_make_empty(&new_list);
  79         nir_foreach_variable_safe(var, var_list) {
  80                 exec_node_remove(&var->node);
  81                 insert_sorted(&new_list, var);
  82         }
  83         exec_list_move_nodes_to(&new_list, var_list);
  84 }
  85
  86 static void
  87 fixup_varying_slots(struct exec_list *var_list)
  88 {
  89         nir_foreach_variable(var, var_list) {
  90                 if (var->data.location >= VARYING_SLOT_VAR0) {
  91                         var->data.location += 9;
  92                 } else if ((var->data.location >= VARYING_SLOT_TEX0) &&
  93                                 (var->data.location <= VARYING_SLOT_TEX7)) {
  94                         var->data.location += VARYING_SLOT_VAR0 - VARYING_SLOT_TEX0;
  95                 }
  96         }
  97 }
  98
  99 static struct ir3_compiler *compiler;
 100
 101 static nir_shader *
 102 load_glsl(unsigned num_files, char* const* files, gl_shader_stage stage)
 103 {
 104         static const struct standalone_options options = {
 105                         .glsl_version = 460,
 106                         .do_link = true,
 107         };
 108         struct gl_shader_program *prog;
 109         const nir_shader_compiler_options *nir_options =
 110                         ir3_get_compiler_options(compiler);
 111
 112         prog = standalone_compile_shader(&options, num_files, files);
 113         if (!prog)
 114                 errx(1, "couldn't parse `%s'", files[0]);
 115
 116         nir_shader *nir = glsl_to_nir(prog, stage, nir_options);
 117
 118         /* required NIR passes: */
 119         if (nir_options->lower_all_io_to_temps ||
 120                         nir->info.stage == MESA_SHADER_VERTEX ||
 121                         nir->info.stage == MESA_SHADER_GEOMETRY) {
 122                 NIR_PASS_V(nir, nir_lower_io_to_temporaries,
 123                                 nir_shader_get_entrypoint(nir),
 124                                 true, true);
 125         } else if (nir->info.stage == MESA_SHADER_FRAGMENT) {
 126                 NIR_PASS_V(nir, nir_lower_io_to_temporaries,
 127                                 nir_shader_get_entrypoint(nir),
 128                                 true, false);
 129         }
 130
 131         NIR_PASS_V(nir, nir_lower_global_vars_to_local);
 132         NIR_PASS_V(nir, nir_split_var_copies);
 133         NIR_PASS_V(nir, nir_lower_var_copies);
 134
 135         NIR_PASS_V(nir, nir_split_var_copies);
 136         NIR_PASS_V(nir, nir_lower_var_copies);
 137         nir_print_shader(nir, stdout);
 138         NIR_PASS_V(nir, gl_nir_lower_atomics, prog, true);
 139         NIR_PASS_V(nir, nir_lower_atomics_to_ssbo, 8);
 140         nir_print_shader(nir, stdout);
 141
 142         switch (stage) {
 143         case MESA_SHADER_VERTEX:
 144                 nir_assign_var_locations(&nir->inputs,
 145                                 &nir->num_inputs,
 146                                 ir3_glsl_type_size);
 147
 148                 /* Re-lower global vars, to deal with any dead VS inputs. */
 149                 NIR_PASS_V(nir, nir_lower_global_vars_to_local);
 150
 151                 sort_varyings(&nir->outputs);
 152                 nir_assign_var_locations(&nir->outputs,
 153                                 &nir->num_outputs,
 154                                 ir3_glsl_type_size);
 155                 fixup_varying_slots(&nir->outputs);
 156                 break;
 157         case MESA_SHADER_FRAGMENT:
 158                 sort_varyings(&nir->inputs);
 159                 nir_assign_var_locations(&nir->inputs,
 160                                 &nir->num_inputs,
 161                                 ir3_glsl_type_size);
 162                 fixup_varying_slots(&nir->inputs);
 163                 nir_assign_var_locations(&nir->outputs,
 164                                 &nir->num_outputs,
 165                                 ir3_glsl_type_size);
 166                 break;
 167         case MESA_SHADER_COMPUTE:
 168                 break;
 169         default:
 170                 errx(1, "unhandled shader stage: %d", stage);
 171         }
 172
 173         nir_assign_var_locations(&nir->uniforms,
 174                         &nir->num_uniforms,
 175                         ir3_glsl_type_size);
 176
 177         NIR_PASS_V(nir, nir_lower_system_values);
 178         NIR_PASS_V(nir, nir_lower_io, nir_var_all, ir3_glsl_type_size, 0);
 179         NIR_PASS_V(nir, gl_nir_lower_samplers, prog);
 180
 181         return nir;
 182 }
 183
 184 static int
 185 read_file(const char *filename, void **ptr, size_t *size)
 186 {
 187         int fd, ret;
 188         struct stat st;
 189
 190         *ptr = MAP_FAILED;
 191
 192         fd = open(filename, O_RDONLY);
 193         if (fd == -1) {
 194                 warnx("couldn't open `%s'", filename);
 195                 return 1;
 196         }
 197
 198         ret = fstat(fd, &st);
 199         if (ret)
 200                 errx(1, "couldn't stat `%s'", filename);
 201
 202         *size = st.st_size;
 203         *ptr = mmap(NULL, st.st_size, PROT_READ, MAP_SHARED, fd, 0);
 204         if (*ptr == MAP_FAILED)
 205                 errx(1, "couldn't map `%s'", filename);
 206
 207         close(fd);
 208
 209         return 0;
 210 }
 211
 212 static void debug_func(void *priv, enum nir_spirv_debug_level level,
 213                 size_t spirv_offset, const char *message)
 214 {
 215 //      printf("%s\n", message);
 216 }
 217
 218 static nir_shader *
 219 load_spirv(const char *filename, const char *entry, gl_shader_stage stage)
 220 {
 221         const struct spirv_to_nir_options spirv_options = {
 222                 /* these caps are just make-believe */
 223                 .caps = {
 224                         .draw_parameters = true,
 225                         .float64 = true,
 226                         .image_read_without_format = true,
 227                         .image_write_without_format = true,
 228                         .int64 = true,
 229                         .variable_pointers = true,
 230                 },
 231                 .lower_workgroup_access_to_offsets = true,
 232                 .debug = {
 233                         .func = debug_func,
 234                 }
 235         };
 236         nir_function *entry_point;
 237         void *buf;
 238         size_t size;
 239
 240         read_file(filename, &buf, &size);
 241
 242         entry_point = spirv_to_nir(buf, size / 4,
 243                         NULL, 0, /* spec_entries */
 244                         stage, entry,
 245                         &spirv_options,
 246                         ir3_get_compiler_options(compiler));
 247
 248         nir_print_shader(entry_point->shader, stdout);
 249
 250         return entry_point->shader;
 251 }
 252
 253 static void print_usage(void)
 254 {
 255         printf("Usage: ir3_compiler [OPTIONS]... <file.tgsi | file.spv entry_point | (file.vert | file.frag)*>\n");
 256         printf("    --verbose         - verbose compiler/debug messages\n");
 257         printf("    --binning-pass    - generate binning pass shader (VERT)\n");
 258         printf("    --color-two-side  - emulate two-sided color (FRAG)\n");
 259         printf("    --half-precision  - use half-precision\n");
 260         printf("    --saturate-s MASK - bitmask of samplers to saturate S coord\n");
 261         printf("    --saturate-t MASK - bitmask of samplers to saturate T coord\n");
 262         printf("    --saturate-r MASK - bitmask of samplers to saturate R coord\n");
 263         printf("    --astc-srgb MASK  - bitmask of samplers to enable astc-srgb workaround\n");
 264         printf("    --stream-out      - enable stream-out (aka transform feedback)\n");
 265         printf("    --ucp MASK        - bitmask of enabled user-clip-planes\n");
 266         printf("    --gpu GPU_ID      - specify gpu-id (default 320)\n");
 267         printf("    --help            - show this message\n");
 268 }
 269
 270 int main(int argc, char **argv)
 271 {
 272         int ret = 0, n = 1;
 273         char *filenames[2];
 274         int num_files = 0;
 275         unsigned stage = 0;
 276         struct ir3_shader_variant v;
 277         struct ir3_shader s;
 278         struct ir3_shader_key key = {};
 279         /* TODO cmdline option to target different gpus: */
 280         unsigned gpu_id = 320;
 281         const char *info;
 282         const char *entry;
 283         void *ptr;
 284         bool from_spirv = false;
 285         size_t size;
 286
 287         memset(&s, 0, sizeof(s));
 288         memset(&v, 0, sizeof(v));
 289
 290         /* cmdline args which impact shader variant get spit out in a
 291          * comment on the first line..  a quick/dirty way to preserve
 292          * that info so when ir3test recompiles the shader with a new
 293          * compiler version, we use the same shader-key settings:
 294          */
 295         debug_printf("; options:");
 296
 297         while (n < argc) {
 298                 if (!strcmp(argv[n], "--verbose")) {
 299                         ir3_shader_debug |= IR3_DBG_OPTMSGS | IR3_DBG_DISASM;
 300                         n++;
 301                         continue;
 302                 }
 303
 304                 if (!strcmp(argv[n], "--binning-pass")) {
 305                         debug_printf(" %s", argv[n]);
 306                         v.binning_pass = true;
 307                         n++;
 308                         continue;
 309                 }
 310
 311                 if (!strcmp(argv[n], "--color-two-side")) {
 312                         debug_printf(" %s", argv[n]);
 313                         key.color_two_side = true;
 314                         n++;
 315                         continue;
 316                 }
 317
 318                 if (!strcmp(argv[n], "--half-precision")) {
 319                         debug_printf(" %s", argv[n]);
 320                         key.half_precision = true;
 321                         n++;
 322                         continue;
 323                 }
 324
 325                 if (!strcmp(argv[n], "--saturate-s")) {
 326                         debug_printf(" %s %s", argv[n], argv[n+1]);
 327                         key.vsaturate_s = key.fsaturate_s = strtol(argv[n+1], NULL, 0);
 328                         n += 2;
 329                         continue;
 330                 }
 331
 332                 if (!strcmp(argv[n], "--saturate-t")) {
 333                         debug_printf(" %s %s", argv[n], argv[n+1]);
 334                         key.vsaturate_t = key.fsaturate_t = strtol(argv[n+1], NULL, 0);
 335                         n += 2;
 336                         continue;
 337                 }
 338
 339                 if (!strcmp(argv[n], "--saturate-r")) {
 340                         debug_printf(" %s %s", argv[n], argv[n+1]);
 341                         key.vsaturate_r = key.fsaturate_r = strtol(argv[n+1], NULL, 0);
 342                         n += 2;
 343                         continue;
 344                 }
 345
 346                 if (!strcmp(argv[n], "--astc-srgb")) {
 347                         debug_printf(" %s %s", argv[n], argv[n+1]);
 348                         key.vastc_srgb = key.fastc_srgb = strtol(argv[n+1], NULL, 0);
 349                         n += 2;
 350                         continue;
 351                 }
 352
 353                 if (!strcmp(argv[n], "--stream-out")) {
 354                         struct ir3_stream_output_info *so = &s.stream_output;
 355                         debug_printf(" %s", argv[n]);
 356                         /* TODO more dynamic config based on number of outputs, etc
 357                          * rather than just hard-code for first output:
 358                          */
 359                         so->num_outputs = 1;
 360                         so->stride[0] = 4;
 361                         so->output[0].register_index = 0;
 362                         so->output[0].start_component = 0;
 363                         so->output[0].num_components = 4;
 364                         so->output[0].output_buffer = 0;
 365                         so->output[0].dst_offset = 2;
 366                         so->output[0].stream = 0;
 367                         n++;
 368                         continue;
 369                 }
 370
 371                 if (!strcmp(argv[n], "--ucp")) {
 372                         debug_printf(" %s %s", argv[n], argv[n+1]);
 373                         key.ucp_enables = strtol(argv[n+1], NULL, 0);
 374                         n += 2;
 375                         continue;
 376                 }
 377
 378                 if (!strcmp(argv[n], "--gpu")) {
 379                         debug_printf(" %s %s", argv[n], argv[n+1]);
 380                         gpu_id = strtol(argv[n+1], NULL, 0);
 381                         n += 2;
 382                         continue;
 383                 }
 384
 385                 if (!strcmp(argv[n], "--help")) {
 386                         print_usage();
 387                         return 0;
 388                 }
 389
 390                 break;
 391         }
 392         debug_printf("\n");
 393
 394         while (n < argc) {
 395                 char *filename = argv[n];
 396                 char *ext = strrchr(filename, '.');
 397
 398                 if (strcmp(ext, ".tgsi") == 0) {
 399                         if (num_files != 0)
 400                                 errx(1, "in TGSI mode, only a single file may be specified");
 401                         s.from_tgsi = true;
 402                 } else if (strcmp(ext, ".spv") == 0) {
 403                         if (num_files != 0)
 404                                 errx(1, "in SPIR-V mode, only a single file may be specified");
 405                         stage = MESA_SHADER_COMPUTE;
 406                         from_spirv = true;
 407                         filenames[num_files++] = filename;
 408                         n++;
 409                         if (n == argc)
 410                                 errx(1, "in SPIR-V mode, an entry point must be specified");
 411                         entry = argv[n];
 412                         n++;
 413                 } else if (strcmp(ext, ".comp") == 0) {
 414                         if (s.from_tgsi || from_spirv)
 415                                 errx(1, "cannot mix GLSL/TGSI/SPIRV");
 416                         if (num_files >= ARRAY_SIZE(filenames))
 417                                 errx(1, "too many GLSL files");
 418                         stage = MESA_SHADER_COMPUTE;
 419                 } else if (strcmp(ext, ".frag") == 0) {
 420                         if (s.from_tgsi || from_spirv)
 421                                 errx(1, "cannot mix GLSL/TGSI/SPIRV");
 422                         if (num_files >= ARRAY_SIZE(filenames))
 423                                 errx(1, "too many GLSL files");
 424                         stage = MESA_SHADER_FRAGMENT;
 425                 } else if (strcmp(ext, ".vert") == 0) {
 426                         if (s.from_tgsi)
 427                                 errx(1, "cannot mix GLSL and TGSI");
 428                         if (num_files >= ARRAY_SIZE(filenames))
 429                                 errx(1, "too many GLSL files");
 430                         stage = MESA_SHADER_VERTEX;
 431                 } else {
 432                         print_usage();
 433                         return -1;
 434                 }
 435
 436                 filenames[num_files++] = filename;
 437
 438                 n++;
 439         }
 440
 441         nir_shader *nir;
 442
 443         compiler = ir3_compiler_create(NULL, gpu_id);
 444
 445         if (s.from_tgsi) {
 446                 struct tgsi_token toks[65536];
 447
 448                 ret = read_file(filenames[0], &ptr, &size);
 449                 if (ret) {
 450                         print_usage();
 451                         return ret;
 452                 }
 453
 454                 if (ir3_shader_debug & IR3_DBG_OPTMSGS)
 455                         debug_printf("%s\n", (char *)ptr);
 456
 457                 if (!tgsi_text_translate(ptr, toks, ARRAY_SIZE(toks)))
 458                         errx(1, "could not parse `%s'", filenames[0]);
 459
 460                 if (ir3_shader_debug & IR3_DBG_OPTMSGS)
 461                         tgsi_dump(toks, 0);
 462
 463                 nir = ir3_tgsi_to_nir(compiler, toks);
 464                 NIR_PASS_V(nir, nir_lower_global_vars_to_local);
 465         } else if (from_spirv) {
 466                 nir = load_spirv(filenames[0], entry, stage);
 467
 468                 NIR_PASS_V(nir, nir_lower_io, nir_var_all, ir3_glsl_type_size,
 469                                 (nir_lower_io_options)0);
 470
 471                 /* TODO do this somewhere else */
 472                 nir_lower_int64(nir, ~0);
 473                 nir_lower_system_values(nir);
 474         } else if (num_files > 0) {
 475                 nir = load_glsl(num_files, filenames, stage);
 476         } else {
 477                 print_usage();
 478                 return -1;
 479         }
 480
 481         s.compiler = compiler;
 482         s.nir = ir3_optimize_nir(&s, nir, NULL);
 483
 484         v.key = key;
 485         v.shader = &s;
 486         s.type = v.type = nir->info.stage;
 487
 488         info = "NIR compiler";
 489         ret = ir3_compile_shader_nir(s.compiler, &v);
 490         if (ret) {
 491                 fprintf(stderr, "compiler failed!\n");
 492                 return ret;
 493         }
 494         dump_info(&v, info);
 495 }