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 #include "pipe/p_context.h"
  53
  54 static void dump_info(struct ir3_shader_variant *so, const char *str)
  55 {
  56         uint32_t *bin;
  57         const char *type = ir3_shader_stage(so->shader);
  58         bin = ir3_shader_assemble(so, so->shader->compiler->gpu_id);
  59         debug_printf("; %s: %s\n", type, str);
  60         ir3_shader_disasm(so, bin, stdout);
  61         free(bin);
  62 }
  63
  64 static void
  65 insert_sorted(struct exec_list *var_list, nir_variable *new_var)
  66 {
  67         nir_foreach_variable(var, var_list) {
  68                 if (var->data.location > new_var->data.location) {
  69                         exec_node_insert_node_before(&var->node, &new_var->node);
  70                         return;
  71                 }
  72         }
  73         exec_list_push_tail(var_list, &new_var->node);
  74 }
  75
  76 static void
  77 sort_varyings(struct exec_list *var_list)
  78 {
  79         struct exec_list new_list;
  80         exec_list_make_empty(&new_list);
  81         nir_foreach_variable_safe(var, var_list) {
  82                 exec_node_remove(&var->node);
  83                 insert_sorted(&new_list, var);
  84         }
  85         exec_list_move_nodes_to(&new_list, var_list);
  86 }
  87
  88 static void
  89 fixup_varying_slots(struct exec_list *var_list)
  90 {
  91         nir_foreach_variable(var, var_list) {
  92                 if (var->data.location >= VARYING_SLOT_VAR0) {
  93                         var->data.location += 9;
  94                 } else if ((var->data.location >= VARYING_SLOT_TEX0) &&
  95                                 (var->data.location <= VARYING_SLOT_TEX7)) {
  96                         var->data.location += VARYING_SLOT_VAR0 - VARYING_SLOT_TEX0;
  97                 }
  98         }
  99 }
 100
 101 static struct ir3_compiler *compiler;
 102
 103 static nir_shader *
 104 load_glsl(unsigned num_files, char* const* files, gl_shader_stage stage)
 105 {
 106         static const struct standalone_options options = {
 107                         .glsl_version = 460,
 108                         .do_link = true,
 109         };
 110         struct gl_shader_program *prog;
 111         const nir_shader_compiler_options *nir_options =
 112                         ir3_get_compiler_options(compiler);
 113         static struct gl_context local_ctx;
 114
 115         prog = standalone_compile_shader(&options, num_files, files, &local_ctx);
 116         if (!prog)
 117                 errx(1, "couldn't parse `%s'", files[0]);
 118
 119         nir_shader *nir = glsl_to_nir(prog, stage, nir_options);
 120
 121         /* required NIR passes: */
 122         if (nir_options->lower_all_io_to_temps ||
 123                         nir->info.stage == MESA_SHADER_VERTEX ||
 124                         nir->info.stage == MESA_SHADER_GEOMETRY) {
 125                 NIR_PASS_V(nir, nir_lower_io_to_temporaries,
 126                                 nir_shader_get_entrypoint(nir),
 127                                 true, true);
 128         } else if (nir->info.stage == MESA_SHADER_FRAGMENT) {
 129                 NIR_PASS_V(nir, nir_lower_io_to_temporaries,
 130                                 nir_shader_get_entrypoint(nir),
 131                                 true, false);
 132         }
 133
 134         NIR_PASS_V(nir, nir_lower_global_vars_to_local);
 135         NIR_PASS_V(nir, nir_split_var_copies);
 136         NIR_PASS_V(nir, nir_lower_var_copies);
 137
 138         NIR_PASS_V(nir, nir_split_var_copies);
 139         NIR_PASS_V(nir, nir_lower_var_copies);
 140         nir_print_shader(nir, stdout);
 141         NIR_PASS_V(nir, gl_nir_lower_atomics, prog, true);
 142         NIR_PASS_V(nir, nir_lower_atomics_to_ssbo, 8);
 143         nir_print_shader(nir, stdout);
 144
 145         switch (stage) {
 146         case MESA_SHADER_VERTEX:
 147                 nir_assign_var_locations(&nir->inputs,
 148                                 &nir->num_inputs,
 149                                 ir3_glsl_type_size);
 150
 151                 /* Re-lower global vars, to deal with any dead VS inputs. */
 152                 NIR_PASS_V(nir, nir_lower_global_vars_to_local);
 153
 154                 sort_varyings(&nir->outputs);
 155                 nir_assign_var_locations(&nir->outputs,
 156                                 &nir->num_outputs,
 157                                 ir3_glsl_type_size);
 158                 fixup_varying_slots(&nir->outputs);
 159                 break;
 160         case MESA_SHADER_FRAGMENT:
 161                 sort_varyings(&nir->inputs);
 162                 nir_assign_var_locations(&nir->inputs,
 163                                 &nir->num_inputs,
 164                                 ir3_glsl_type_size);
 165                 fixup_varying_slots(&nir->inputs);
 166                 nir_assign_var_locations(&nir->outputs,
 167                                 &nir->num_outputs,
 168                                 ir3_glsl_type_size);
 169                 break;
 170         case MESA_SHADER_COMPUTE:
 171         case MESA_SHADER_KERNEL:
 172                 break;
 173         default:
 174                 errx(1, "unhandled shader stage: %d", stage);
 175         }
 176
 177         nir_assign_var_locations(&nir->uniforms,
 178                         &nir->num_uniforms,
 179                         ir3_glsl_type_size);
 180
 181         NIR_PASS_V(nir, nir_lower_system_values);
 182         NIR_PASS_V(nir, nir_lower_io, nir_var_all, ir3_glsl_type_size, 0);
 183         NIR_PASS_V(nir, gl_nir_lower_samplers, prog);
 184
 185         return nir;
 186 }
 187
 188 static int
 189 read_file(const char *filename, void **ptr, size_t *size)
 190 {
 191         int fd, ret;
 192         struct stat st;
 193
 194         *ptr = MAP_FAILED;
 195
 196         fd = open(filename, O_RDONLY);
 197         if (fd == -1) {
 198                 warnx("couldn't open `%s'", filename);
 199                 return 1;
 200         }
 201
 202         ret = fstat(fd, &st);
 203         if (ret)
 204                 errx(1, "couldn't stat `%s'", filename);
 205
 206         *size = st.st_size;
 207         *ptr = mmap(NULL, st.st_size, PROT_READ, MAP_SHARED, fd, 0);
 208         if (*ptr == MAP_FAILED)
 209                 errx(1, "couldn't map `%s'", filename);
 210
 211         close(fd);
 212
 213         return 0;
 214 }
 215
 216 static void debug_func(void *priv, enum nir_spirv_debug_level level,
 217                 size_t spirv_offset, const char *message)
 218 {
 219 //      printf("%s\n", message);
 220 }
 221
 222 static nir_shader *
 223 load_spirv(const char *filename, const char *entry, gl_shader_stage stage)
 224 {
 225         const struct spirv_to_nir_options spirv_options = {
 226                 /* these caps are just make-believe */
 227                 .caps = {
 228                         .draw_parameters = true,
 229                         .float64 = true,
 230                         .image_read_without_format = true,
 231                         .image_write_without_format = true,
 232                         .int64 = true,
 233                         .variable_pointers = true,
 234                 },
 235                 .lower_workgroup_access_to_offsets = true,
 236                 .lower_ubo_ssbo_access_to_offsets = true,
 237                 .debug = {
 238                         .func = debug_func,
 239                 }
 240         };
 241         nir_function *entry_point;
 242         void *buf;
 243         size_t size;
 244
 245         read_file(filename, &buf, &size);
 246
 247         entry_point = spirv_to_nir(buf, size / 4,
 248                         NULL, 0, /* spec_entries */
 249                         stage, entry,
 250                         &spirv_options,
 251                         ir3_get_compiler_options(compiler));
 252
 253         nir_print_shader(entry_point->shader, stdout);
 254
 255         return entry_point->shader;
 256 }
 257
 258 static void print_usage(void)
 259 {
 260         printf("Usage: ir3_compiler [OPTIONS]... <file.tgsi | file.spv entry_point | (file.vert | file.frag)*>\n");
 261         printf("    --verbose         - verbose compiler/debug messages\n");
 262         printf("    --binning-pass    - generate binning pass shader (VERT)\n");
 263         printf("    --color-two-side  - emulate two-sided color (FRAG)\n");
 264         printf("    --half-precision  - use half-precision\n");
 265         printf("    --saturate-s MASK - bitmask of samplers to saturate S coord\n");
 266         printf("    --saturate-t MASK - bitmask of samplers to saturate T coord\n");
 267         printf("    --saturate-r MASK - bitmask of samplers to saturate R coord\n");
 268         printf("    --astc-srgb MASK  - bitmask of samplers to enable astc-srgb workaround\n");
 269         printf("    --stream-out      - enable stream-out (aka transform feedback)\n");
 270         printf("    --ucp MASK        - bitmask of enabled user-clip-planes\n");
 271         printf("    --gpu GPU_ID      - specify gpu-id (default 320)\n");
 272         printf("    --help            - show this message\n");
 273 }
 274
 275 int main(int argc, char **argv)
 276 {
 277         int ret = 0, n = 1;
 278         char *filenames[2];
 279         int num_files = 0;
 280         unsigned stage = 0;
 281         struct ir3_shader_variant v;
 282         struct ir3_shader s;
 283         struct ir3_shader_key key = {};
 284         /* TODO cmdline option to target different gpus: */
 285         unsigned gpu_id = 320;
 286         const char *info;
 287         const char *entry;
 288         void *ptr;
 289         bool from_spirv = false;
 290         size_t size;
 291
 292         memset(&s, 0, sizeof(s));
 293         memset(&v, 0, sizeof(v));
 294
 295         /* cmdline args which impact shader variant get spit out in a
 296          * comment on the first line..  a quick/dirty way to preserve
 297          * that info so when ir3test recompiles the shader with a new
 298          * compiler version, we use the same shader-key settings:
 299          */
 300         debug_printf("; options:");
 301
 302         while (n < argc) {
 303                 if (!strcmp(argv[n], "--verbose")) {
 304                         ir3_shader_debug |= IR3_DBG_OPTMSGS | IR3_DBG_DISASM;
 305                         n++;
 306                         continue;
 307                 }
 308
 309                 if (!strcmp(argv[n], "--binning-pass")) {
 310                         debug_printf(" %s", argv[n]);
 311                         v.binning_pass = true;
 312                         n++;
 313                         continue;
 314                 }
 315
 316                 if (!strcmp(argv[n], "--color-two-side")) {
 317                         debug_printf(" %s", argv[n]);
 318                         key.color_two_side = true;
 319                         n++;
 320                         continue;
 321                 }
 322
 323                 if (!strcmp(argv[n], "--half-precision")) {
 324                         debug_printf(" %s", argv[n]);
 325                         key.half_precision = true;
 326                         n++;
 327                         continue;
 328                 }
 329
 330                 if (!strcmp(argv[n], "--saturate-s")) {
 331                         debug_printf(" %s %s", argv[n], argv[n+1]);
 332                         key.vsaturate_s = key.fsaturate_s = strtol(argv[n+1], NULL, 0);
 333                         n += 2;
 334                         continue;
 335                 }
 336
 337                 if (!strcmp(argv[n], "--saturate-t")) {
 338                         debug_printf(" %s %s", argv[n], argv[n+1]);
 339                         key.vsaturate_t = key.fsaturate_t = strtol(argv[n+1], NULL, 0);
 340                         n += 2;
 341                         continue;
 342                 }
 343
 344                 if (!strcmp(argv[n], "--saturate-r")) {
 345                         debug_printf(" %s %s", argv[n], argv[n+1]);
 346                         key.vsaturate_r = key.fsaturate_r = strtol(argv[n+1], NULL, 0);
 347                         n += 2;
 348                         continue;
 349                 }
 350
 351                 if (!strcmp(argv[n], "--astc-srgb")) {
 352                         debug_printf(" %s %s", argv[n], argv[n+1]);
 353                         key.vastc_srgb = key.fastc_srgb = strtol(argv[n+1], NULL, 0);
 354                         n += 2;
 355                         continue;
 356                 }
 357
 358                 if (!strcmp(argv[n], "--stream-out")) {
 359                         struct ir3_stream_output_info *so = &s.stream_output;
 360                         debug_printf(" %s", argv[n]);
 361                         /* TODO more dynamic config based on number of outputs, etc
 362                          * rather than just hard-code for first output:
 363                          */
 364                         so->num_outputs = 1;
 365                         so->stride[0] = 4;
 366                         so->output[0].register_index = 0;
 367                         so->output[0].start_component = 0;
 368                         so->output[0].num_components = 4;
 369                         so->output[0].output_buffer = 0;
 370                         so->output[0].dst_offset = 2;
 371                         so->output[0].stream = 0;
 372                         n++;
 373                         continue;
 374                 }
 375
 376                 if (!strcmp(argv[n], "--ucp")) {
 377                         debug_printf(" %s %s", argv[n], argv[n+1]);
 378                         key.ucp_enables = strtol(argv[n+1], NULL, 0);
 379                         n += 2;
 380                         continue;
 381                 }
 382
 383                 if (!strcmp(argv[n], "--gpu")) {
 384                         debug_printf(" %s %s", argv[n], argv[n+1]);
 385                         gpu_id = strtol(argv[n+1], NULL, 0);
 386                         n += 2;
 387                         continue;
 388                 }
 389
 390                 if (!strcmp(argv[n], "--help")) {
 391                         print_usage();
 392                         return 0;
 393                 }
 394
 395                 break;
 396         }
 397         debug_printf("\n");
 398
 399         while (n < argc) {
 400                 char *filename = argv[n];
 401                 char *ext = strrchr(filename, '.');
 402
 403                 if (strcmp(ext, ".tgsi") == 0) {
 404                         if (num_files != 0)
 405                                 errx(1, "in TGSI mode, only a single file may be specified");
 406                         s.from_tgsi = true;
 407                 } else if (strcmp(ext, ".spv") == 0) {
 408                         if (num_files != 0)
 409                                 errx(1, "in SPIR-V mode, only a single file may be specified");
 410                         stage = MESA_SHADER_COMPUTE;
 411                         from_spirv = true;
 412                         filenames[num_files++] = filename;
 413                         n++;
 414                         if (n == argc)
 415                                 errx(1, "in SPIR-V mode, an entry point must be specified");
 416                         entry = argv[n];
 417                         n++;
 418                 } else if (strcmp(ext, ".comp") == 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_COMPUTE;
 424                 } else if (strcmp(ext, ".frag") == 0) {
 425                         if (s.from_tgsi || from_spirv)
 426                                 errx(1, "cannot mix GLSL/TGSI/SPIRV");
 427                         if (num_files >= ARRAY_SIZE(filenames))
 428                                 errx(1, "too many GLSL files");
 429                         stage = MESA_SHADER_FRAGMENT;
 430                 } else if (strcmp(ext, ".vert") == 0) {
 431                         if (s.from_tgsi)
 432                                 errx(1, "cannot mix GLSL and TGSI");
 433                         if (num_files >= ARRAY_SIZE(filenames))
 434                                 errx(1, "too many GLSL files");
 435                         stage = MESA_SHADER_VERTEX;
 436                 } else {
 437                         print_usage();
 438                         return -1;
 439                 }
 440
 441                 filenames[num_files++] = filename;
 442
 443                 n++;
 444         }
 445
 446         nir_shader *nir;
 447
 448         compiler = ir3_compiler_create(NULL, gpu_id);
 449
 450         if (s.from_tgsi) {
 451                 struct tgsi_token toks[65536];
 452
 453                 ret = read_file(filenames[0], &ptr, &size);
 454                 if (ret) {
 455                         print_usage();
 456                         return ret;
 457                 }
 458
 459                 if (ir3_shader_debug & IR3_DBG_OPTMSGS)
 460                         debug_printf("%s\n", (char *)ptr);
 461
 462                 if (!tgsi_text_translate(ptr, toks, ARRAY_SIZE(toks)))
 463                         errx(1, "could not parse `%s'", filenames[0]);
 464
 465                 if (ir3_shader_debug & IR3_DBG_OPTMSGS)
 466                         tgsi_dump(toks, 0);
 467
 468                 nir = ir3_tgsi_to_nir(compiler, toks, NULL);
 469                 NIR_PASS_V(nir, nir_lower_global_vars_to_local);
 470         } else if (from_spirv) {
 471                 nir = load_spirv(filenames[0], entry, stage);
 472
 473                 NIR_PASS_V(nir, nir_lower_io, nir_var_all, ir3_glsl_type_size,
 474                                 (nir_lower_io_options)0);
 475
 476                 /* TODO do this somewhere else */
 477                 nir_lower_int64(nir, ~0);
 478                 nir_lower_system_values(nir);
 479         } else if (num_files > 0) {
 480                 nir = load_glsl(num_files, filenames, stage);
 481         } else {
 482                 print_usage();
 483                 return -1;
 484         }
 485
 486         s.compiler = compiler;
 487         s.nir = ir3_optimize_nir(&s, nir, NULL);
 488
 489         v.key = key;
 490         v.shader = &s;
 491         s.type = v.type = nir->info.stage;
 492
 493         info = "NIR compiler";
 494         ret = ir3_compile_shader_nir(s.compiler, &v);
 495         if (ret) {
 496                 fprintf(stderr, "compiler failed!\n");
 497                 return ret;
 498         }
 499         dump_info(&v, info);
 500 }