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