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 "freedreno_util.h"
  41
  42 #include "ir3_compiler.h"
  43 #include "ir3_nir.h"
  44 #include "instr-a3xx.h"
  45 #include "ir3.h"
  46
  47 #include "compiler/glsl/standalone.h"
  48 #include "compiler/glsl/glsl_to_nir.h"
  49 #include "compiler/glsl/gl_nir.h"
  50 #include "compiler/nir_types.h"
  51 #include "compiler/spirv/nir_spirv.h"
  52
  53 static void dump_info(struct ir3_shader_variant *so, const char *str)
  54 {
  55         uint32_t *bin;
  56         const char *type = ir3_shader_stage(so->shader);
  57         bin = ir3_shader_assemble(so, so->shader->compiler->gpu_id);
  58         debug_printf("; %s: %s\n", type, str);
  59         ir3_shader_disasm(so, bin, stdout);
  60         free(bin);
  61 }
  62
  63 static void
  64 insert_sorted(struct exec_list *var_list, nir_variable *new_var)
  65 {
  66         nir_foreach_variable(var, var_list) {
  67                 if (var->data.location > new_var->data.location) {
  68                         exec_node_insert_node_before(&var->node, &new_var->node);
  69                         return;
  70                 }
  71         }
  72         exec_list_push_tail(var_list, &new_var->node);
  73 }
  74
  75 static void
  76 sort_varyings(struct exec_list *var_list)
  77 {
  78         struct exec_list new_list;
  79         exec_list_make_empty(&new_list);
  80         nir_foreach_variable_safe(var, var_list) {
  81                 exec_node_remove(&var->node);
  82                 insert_sorted(&new_list, var);
  83         }
  84         exec_list_move_nodes_to(&new_list, var_list);
  85 }
  86
  87 static void
  88 fixup_varying_slots(struct exec_list *var_list)
  89 {
  90         nir_foreach_variable(var, var_list) {
  91                 if (var->data.location >= VARYING_SLOT_VAR0) {
  92                         var->data.location += 9;
  93                 } else if ((var->data.location >= VARYING_SLOT_TEX0) &&
  94                                 (var->data.location <= VARYING_SLOT_TEX7)) {
  95                         var->data.location += VARYING_SLOT_VAR0 - VARYING_SLOT_TEX0;
  96                 }
  97         }
  98 }
  99
 100 static struct ir3_compiler *compiler;
 101
 102 static nir_shader *
 103 load_glsl(unsigned num_files, char* const* files, gl_shader_stage stage)
 104 {
 105         static const struct standalone_options options = {
 106                         .glsl_version = 460,
 107                         .do_link = true,
 108         };
 109         struct gl_shader_program *prog;
 110         const nir_shader_compiler_options *nir_options =
 111                         ir3_get_compiler_options(compiler);
 112
 113         prog = standalone_compile_shader(&options, num_files, files);
 114         if (!prog)
 115                 errx(1, "couldn't parse `%s'", files[0]);
 116
 117         nir_shader *nir = glsl_to_nir(prog, stage, nir_options);
 118
 119         /* required NIR passes: */
 120         if (nir_options->lower_all_io_to_temps ||
 121                         nir->info.stage == MESA_SHADER_VERTEX ||
 122                         nir->info.stage == MESA_SHADER_GEOMETRY) {
 123                 NIR_PASS_V(nir, nir_lower_io_to_temporaries,
 124                                 nir_shader_get_entrypoint(nir),
 125                                 true, true);
 126         } else if (nir->info.stage == MESA_SHADER_FRAGMENT) {
 127                 NIR_PASS_V(nir, nir_lower_io_to_temporaries,
 128                                 nir_shader_get_entrypoint(nir),
 129                                 true, false);
 130         }
 131
 132         NIR_PASS_V(nir, nir_lower_global_vars_to_local);
 133         NIR_PASS_V(nir, nir_split_var_copies);
 134         NIR_PASS_V(nir, nir_lower_var_copies);
 135
 136         NIR_PASS_V(nir, nir_split_var_copies);
 137         NIR_PASS_V(nir, nir_lower_var_copies);
 138         nir_print_shader(nir, stdout);
 139         NIR_PASS_V(nir, gl_nir_lower_atomics, prog, true);
 140         NIR_PASS_V(nir, nir_lower_atomics_to_ssbo, 8);
 141         nir_print_shader(nir, stdout);
 142
 143         switch (stage) {
 144         case MESA_SHADER_VERTEX:
 145                 nir_assign_var_locations(&nir->inputs,
 146                                 &nir->num_inputs,
 147                                 ir3_glsl_type_size);
 148
 149                 /* Re-lower global vars, to deal with any dead VS inputs. */
 150                 NIR_PASS_V(nir, nir_lower_global_vars_to_local);
 151
 152                 sort_varyings(&nir->outputs);
 153                 nir_assign_var_locations(&nir->outputs,
 154                                 &nir->num_outputs,
 155                                 ir3_glsl_type_size);
 156                 fixup_varying_slots(&nir->outputs);
 157                 break;
 158         case MESA_SHADER_FRAGMENT:
 159                 sort_varyings(&nir->inputs);
 160                 nir_assign_var_locations(&nir->inputs,
 161                                 &nir->num_inputs,
 162                                 ir3_glsl_type_size);
 163                 fixup_varying_slots(&nir->inputs);
 164                 nir_assign_var_locations(&nir->outputs,
 165                                 &nir->num_outputs,
 166                                 ir3_glsl_type_size);
 167                 break;
 168         case MESA_SHADER_COMPUTE:
 169                 break;
 170         default:
 171                 errx(1, "unhandled shader stage: %d", stage);
 172         }
 173
 174         nir_assign_var_locations(&nir->uniforms,
 175                         &nir->num_uniforms,
 176                         ir3_glsl_type_size);
 177
 178         NIR_PASS_V(nir, nir_lower_system_values);
 179         NIR_PASS_V(nir, nir_lower_io, nir_var_all, ir3_glsl_type_size, 0);
 180         NIR_PASS_V(nir, gl_nir_lower_samplers, prog);
 181
 182         return nir;
 183 }
 184
 185 static int
 186 read_file(const char *filename, void **ptr, size_t *size)
 187 {
 188         int fd, ret;
 189         struct stat st;
 190
 191         *ptr = MAP_FAILED;
 192
 193         fd = open(filename, O_RDONLY);
 194         if (fd == -1) {
 195                 warnx("couldn't open `%s'", filename);
 196                 return 1;
 197         }
 198
 199         ret = fstat(fd, &st);
 200         if (ret)
 201                 errx(1, "couldn't stat `%s'", filename);
 202
 203         *size = st.st_size;
 204         *ptr = mmap(NULL, st.st_size, PROT_READ, MAP_SHARED, fd, 0);
 205         if (*ptr == MAP_FAILED)
 206                 errx(1, "couldn't map `%s'", filename);
 207
 208         close(fd);
 209
 210         return 0;
 211 }
 212
 213 static void debug_func(void *priv, enum nir_spirv_debug_level level,
 214                 size_t spirv_offset, const char *message)
 215 {
 216 //      printf("%s\n", message);
 217 }
 218
 219 static nir_shader *
 220 load_spirv(const char *filename, const char *entry, gl_shader_stage stage)
 221 {
 222         const struct spirv_to_nir_options spirv_options = {
 223                 /* these caps are just make-believe */
 224                 .caps = {
 225                         .draw_parameters = true,
 226                         .float64 = true,
 227                         .image_read_without_format = true,
 228                         .image_write_without_format = true,
 229                         .int64 = true,
 230                         .variable_pointers = true,
 231                 },
 232                 .lower_workgroup_access_to_offsets = true,
 233                 .debug = {
 234                         .func = debug_func,
 235                 }
 236         };
 237         nir_function *entry_point;
 238         void *buf;
 239         size_t size;
 240
 241         read_file(filename, &buf, &size);
 242
 243         entry_point = spirv_to_nir(buf, size / 4,
 244                         NULL, 0, /* spec_entries */
 245                         stage, entry,
 246                         &spirv_options,
 247                         ir3_get_compiler_options(compiler));
 248
 249         nir_print_shader(entry_point->shader, stdout);
 250
 251         return entry_point->shader;
 252 }
 253
 254 static void print_usage(void)
 255 {
 256         printf("Usage: ir3_compiler [OPTIONS]... <file.tgsi | file.spv entry_point | (file.vert | file.frag)*>\n");
 257         printf("    --verbose         - verbose compiler/debug messages\n");
 258         printf("    --binning-pass    - generate binning pass shader (VERT)\n");
 259         printf("    --color-two-side  - emulate two-sided color (FRAG)\n");
 260         printf("    --half-precision  - use half-precision\n");
 261         printf("    --saturate-s MASK - bitmask of samplers to saturate S coord\n");
 262         printf("    --saturate-t MASK - bitmask of samplers to saturate T coord\n");
 263         printf("    --saturate-r MASK - bitmask of samplers to saturate R coord\n");
 264         printf("    --astc-srgb MASK  - bitmask of samplers to enable astc-srgb workaround\n");
 265         printf("    --stream-out      - enable stream-out (aka transform feedback)\n");
 266         printf("    --ucp MASK        - bitmask of enabled user-clip-planes\n");
 267         printf("    --gpu GPU_ID      - specify gpu-id (default 320)\n");
 268         printf("    --help            - show this message\n");
 269 }
 270
 271 int main(int argc, char **argv)
 272 {
 273         int ret = 0, n = 1;
 274         char *filenames[2];
 275         int num_files = 0;
 276         unsigned stage = 0;
 277         struct ir3_shader_variant v;
 278         struct ir3_shader s;
 279         struct ir3_shader_key key = {};
 280         /* TODO cmdline option to target different gpus: */
 281         unsigned gpu_id = 320;
 282         const char *info;
 283         const char *entry;
 284         void *ptr;
 285         bool from_spirv = false;
 286         size_t size;
 287
 288         memset(&s, 0, sizeof(s));
 289         memset(&v, 0, sizeof(v));
 290
 291         /* cmdline args which impact shader variant get spit out in a
 292          * comment on the first line..  a quick/dirty way to preserve
 293          * that info so when ir3test recompiles the shader with a new
 294          * compiler version, we use the same shader-key settings:
 295          */
 296         debug_printf("; options:");
 297
 298         while (n < argc) {
 299                 if (!strcmp(argv[n], "--verbose")) {
 300                         fd_mesa_debug |= FD_DBG_MSGS | FD_DBG_OPTMSGS | FD_DBG_DISASM;
 301                         n++;
 302                         continue;
 303                 }
 304
 305                 if (!strcmp(argv[n], "--binning-pass")) {
 306                         debug_printf(" %s", argv[n]);
 307                         v.binning_pass = true;
 308                         n++;
 309                         continue;
 310                 }
 311
 312                 if (!strcmp(argv[n], "--color-two-side")) {
 313                         debug_printf(" %s", argv[n]);
 314                         key.color_two_side = true;
 315                         n++;
 316                         continue;
 317                 }
 318
 319                 if (!strcmp(argv[n], "--half-precision")) {
 320                         debug_printf(" %s", argv[n]);
 321                         key.half_precision = true;
 322                         n++;
 323                         continue;
 324                 }
 325
 326                 if (!strcmp(argv[n], "--saturate-s")) {
 327                         debug_printf(" %s %s", argv[n], argv[n+1]);
 328                         key.vsaturate_s = key.fsaturate_s = strtol(argv[n+1], NULL, 0);
 329                         n += 2;
 330                         continue;
 331                 }
 332
 333                 if (!strcmp(argv[n], "--saturate-t")) {
 334                         debug_printf(" %s %s", argv[n], argv[n+1]);
 335                         key.vsaturate_t = key.fsaturate_t = strtol(argv[n+1], NULL, 0);
 336                         n += 2;
 337                         continue;
 338                 }
 339
 340                 if (!strcmp(argv[n], "--saturate-r")) {
 341                         debug_printf(" %s %s", argv[n], argv[n+1]);
 342                         key.vsaturate_r = key.fsaturate_r = strtol(argv[n+1], NULL, 0);
 343                         n += 2;
 344                         continue;
 345                 }
 346
 347                 if (!strcmp(argv[n], "--astc-srgb")) {
 348                         debug_printf(" %s %s", argv[n], argv[n+1]);
 349                         key.vastc_srgb = key.fastc_srgb = strtol(argv[n+1], NULL, 0);
 350                         n += 2;
 351                         continue;
 352                 }
 353
 354                 if (!strcmp(argv[n], "--stream-out")) {
 355                         struct pipe_stream_output_info *so = &s.stream_output;
 356                         debug_printf(" %s", argv[n]);
 357                         /* TODO more dynamic config based on number of outputs, etc
 358                          * rather than just hard-code for first output:
 359                          */
 360                         so->num_outputs = 1;
 361                         so->stride[0] = 4;
 362                         so->output[0].register_index = 0;
 363                         so->output[0].start_component = 0;
 364                         so->output[0].num_components = 4;
 365                         so->output[0].output_buffer = 0;
 366                         so->output[0].dst_offset = 2;
 367                         so->output[0].stream = 0;
 368                         n++;
 369                         continue;
 370                 }
 371
 372                 if (!strcmp(argv[n], "--ucp")) {
 373                         debug_printf(" %s %s", argv[n], argv[n+1]);
 374                         key.ucp_enables = strtol(argv[n+1], NULL, 0);
 375                         n += 2;
 376                         continue;
 377                 }
 378
 379                 if (!strcmp(argv[n], "--gpu")) {
 380                         debug_printf(" %s %s", argv[n], argv[n+1]);
 381                         gpu_id = strtol(argv[n+1], NULL, 0);
 382                         n += 2;
 383                         continue;
 384                 }
 385
 386                 if (!strcmp(argv[n], "--help")) {
 387                         print_usage();
 388                         return 0;
 389                 }
 390
 391                 break;
 392         }
 393         debug_printf("\n");
 394
 395         while (n < argc) {
 396                 char *filename = argv[n];
 397                 char *ext = strrchr(filename, '.');
 398
 399                 if (strcmp(ext, ".tgsi") == 0) {
 400                         if (num_files != 0)
 401                                 errx(1, "in TGSI mode, only a single file may be specified");
 402                         s.from_tgsi = true;
 403                 } else if (strcmp(ext, ".spv") == 0) {
 404                         if (num_files != 0)
 405                                 errx(1, "in SPIR-V mode, only a single file may be specified");
 406                         stage = MESA_SHADER_COMPUTE;
 407                         from_spirv = true;
 408                         filenames[num_files++] = filename;
 409                         n++;
 410                         if (n == argc)
 411                                 errx(1, "in SPIR-V mode, an entry point must be specified");
 412                         entry = argv[n];
 413                         n++;
 414                 } else if (strcmp(ext, ".comp") == 0) {
 415                         if (s.from_tgsi || from_spirv)
 416                                 errx(1, "cannot mix GLSL/TGSI/SPIRV");
 417                         if (num_files >= ARRAY_SIZE(filenames))
 418                                 errx(1, "too many GLSL files");
 419                         stage = MESA_SHADER_COMPUTE;
 420                 } else if (strcmp(ext, ".frag") == 0) {
 421                         if (s.from_tgsi || from_spirv)
 422                                 errx(1, "cannot mix GLSL/TGSI/SPIRV");
 423                         if (num_files >= ARRAY_SIZE(filenames))
 424                                 errx(1, "too many GLSL files");
 425                         stage = MESA_SHADER_FRAGMENT;
 426                 } else if (strcmp(ext, ".vert") == 0) {
 427                         if (s.from_tgsi)
 428                                 errx(1, "cannot mix GLSL and TGSI");
 429                         if (num_files >= ARRAY_SIZE(filenames))
 430                                 errx(1, "too many GLSL files");
 431                         stage = MESA_SHADER_VERTEX;
 432                 } else {
 433                         print_usage();
 434                         return -1;
 435                 }
 436
 437                 filenames[num_files++] = filename;
 438
 439                 n++;
 440         }
 441
 442         nir_shader *nir;
 443
 444         compiler = ir3_compiler_create(NULL, gpu_id);
 445
 446         if (s.from_tgsi) {
 447                 struct tgsi_token toks[65536];
 448
 449                 ret = read_file(filenames[0], &ptr, &size);
 450                 if (ret) {
 451                         print_usage();
 452                         return ret;
 453                 }
 454
 455                 if (fd_mesa_debug & FD_DBG_OPTMSGS)
 456                         debug_printf("%s\n", (char *)ptr);
 457
 458                 if (!tgsi_text_translate(ptr, toks, ARRAY_SIZE(toks)))
 459                         errx(1, "could not parse `%s'", filenames[0]);
 460
 461                 if (fd_mesa_debug & FD_DBG_OPTMSGS)
 462                         tgsi_dump(toks, 0);
 463
 464                 nir = ir3_tgsi_to_nir(toks);
 465                 NIR_PASS_V(nir, nir_lower_global_vars_to_local);
 466         } else if (from_spirv) {
 467                 nir = load_spirv(filenames[0], entry, stage);
 468
 469                 NIR_PASS_V(nir, nir_lower_io, nir_var_all, ir3_glsl_type_size,
 470                                 (nir_lower_io_options)0);
 471
 472                 /* TODO do this somewhere else */
 473                 nir_lower_int64(nir, ~0);
 474                 nir_lower_system_values(nir);
 475         } else if (num_files > 0) {
 476                 nir = load_glsl(num_files, filenames, stage);
 477         } else {
 478                 print_usage();
 479                 return -1;
 480         }
 481
 482         s.compiler = compiler;
 483         s.nir = ir3_optimize_nir(&s, nir, NULL);
 484
 485         v.key = key;
 486         v.shader = &s;
 487         s.type = v.type = nir->info.stage;
 488
 489         info = "NIR compiler";
 490         ret = ir3_compile_shader_nir(s.compiler, &v);
 491         if (ret) {
 492                 fprintf(stderr, "compiler failed!\n");
 493                 return ret;
 494         }
 495         dump_info(&v, info);
 496 }