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
  50 static void dump_info(struct ir3_shader_variant *so, const char *str)
  51 {
  52         uint32_t *bin;
  53         const char *type = ir3_shader_stage(so->shader);
  54         bin = ir3_shader_assemble(so, so->shader->compiler->gpu_id);
  55         debug_printf("; %s: %s\n", type, str);
  56         ir3_shader_disasm(so, bin);
  57         free(bin);
  58 }
  59
  60 int st_glsl_type_size(const struct glsl_type *type);
  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 = 140,
 106                         .do_link = true,
 107         };
 108         struct gl_shader_program *prog;
 109
 110         prog = standalone_compile_shader(&options, num_files, files);
 111         if (!prog)
 112                 errx(1, "couldn't parse `%s'", files[0]);
 113
 114         nir_shader *nir = glsl_to_nir(prog, stage, ir3_get_compiler_options(compiler));
 115
 116         /* required NIR passes: */
 117         /* TODO cmdline args for some of the conditional lowering passes? */
 118
 119         NIR_PASS_V(nir, nir_lower_io_to_temporaries,
 120                         nir_shader_get_entrypoint(nir),
 121                         true, true);
 122         NIR_PASS_V(nir, nir_lower_global_vars_to_local);
 123         NIR_PASS_V(nir, nir_split_var_copies);
 124         NIR_PASS_V(nir, nir_lower_var_copies);
 125
 126         NIR_PASS_V(nir, nir_split_var_copies);
 127         NIR_PASS_V(nir, nir_lower_var_copies);
 128         NIR_PASS_V(nir, nir_lower_io_types);
 129
 130         switch (stage) {
 131         case MESA_SHADER_VERTEX:
 132                 nir_assign_var_locations(&nir->inputs,
 133                                 &nir->num_inputs,
 134                                 st_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                                 st_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                                 st_glsl_type_size);
 150                 fixup_varying_slots(&nir->inputs);
 151                 nir_assign_var_locations(&nir->outputs,
 152                                 &nir->num_outputs,
 153                                 st_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                         st_glsl_type_size);
 162
 163         NIR_PASS_V(nir, nir_lower_system_values);
 164         NIR_PASS_V(nir, nir_lower_io, nir_var_all, st_glsl_type_size, 0);
 165         NIR_PASS_V(nir, 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 print_usage(void)
 199 {
 200         printf("Usage: ir3_compiler [OPTIONS]... <file.tgsi | (file.vert | file.frag)*>\n");
 201         printf("    --verbose         - verbose compiler/debug messages\n");
 202         printf("    --binning-pass    - generate binning pass shader (VERT)\n");
 203         printf("    --color-two-side  - emulate two-sided color (FRAG)\n");
 204         printf("    --half-precision  - use half-precision\n");
 205         printf("    --saturate-s MASK - bitmask of samplers to saturate S coord\n");
 206         printf("    --saturate-t MASK - bitmask of samplers to saturate T coord\n");
 207         printf("    --saturate-r MASK - bitmask of samplers to saturate R coord\n");
 208         printf("    --astc-srgb MASK  - bitmask of samplers to enable astc-srgb workaround\n");
 209         printf("    --stream-out      - enable stream-out (aka transform feedback)\n");
 210         printf("    --ucp MASK        - bitmask of enabled user-clip-planes\n");
 211         printf("    --gpu GPU_ID      - specify gpu-id (default 320)\n");
 212         printf("    --help            - show this message\n");
 213 }
 214
 215 int main(int argc, char **argv)
 216 {
 217         int ret = 0, n = 1;
 218         char *filenames[2];
 219         int num_files = 0;
 220         unsigned stage = 0;
 221         struct ir3_shader_variant v;
 222         struct ir3_shader s;
 223         struct ir3_shader_key key = {};
 224         /* TODO cmdline option to target different gpus: */
 225         unsigned gpu_id = 320;
 226         const char *info;
 227         void *ptr;
 228         size_t size;
 229
 230         memset(&s, 0, sizeof(s));
 231         memset(&v, 0, sizeof(v));
 232
 233         /* cmdline args which impact shader variant get spit out in a
 234          * comment on the first line..  a quick/dirty way to preserve
 235          * that info so when ir3test recompiles the shader with a new
 236          * compiler version, we use the same shader-key settings:
 237          */
 238         debug_printf("; options:");
 239
 240         while (n < argc) {
 241                 if (!strcmp(argv[n], "--verbose")) {
 242                         fd_mesa_debug |= FD_DBG_MSGS | FD_DBG_OPTMSGS | FD_DBG_DISASM;
 243                         n++;
 244                         continue;
 245                 }
 246
 247                 if (!strcmp(argv[n], "--binning-pass")) {
 248                         debug_printf(" %s", argv[n]);
 249                         key.binning_pass = true;
 250                         n++;
 251                         continue;
 252                 }
 253
 254                 if (!strcmp(argv[n], "--color-two-side")) {
 255                         debug_printf(" %s", argv[n]);
 256                         key.color_two_side = true;
 257                         n++;
 258                         continue;
 259                 }
 260
 261                 if (!strcmp(argv[n], "--half-precision")) {
 262                         debug_printf(" %s", argv[n]);
 263                         key.half_precision = true;
 264                         n++;
 265                         continue;
 266                 }
 267
 268                 if (!strcmp(argv[n], "--saturate-s")) {
 269                         debug_printf(" %s %s", argv[n], argv[n+1]);
 270                         key.vsaturate_s = key.fsaturate_s = strtol(argv[n+1], NULL, 0);
 271                         n += 2;
 272                         continue;
 273                 }
 274
 275                 if (!strcmp(argv[n], "--saturate-t")) {
 276                         debug_printf(" %s %s", argv[n], argv[n+1]);
 277                         key.vsaturate_t = key.fsaturate_t = strtol(argv[n+1], NULL, 0);
 278                         n += 2;
 279                         continue;
 280                 }
 281
 282                 if (!strcmp(argv[n], "--saturate-r")) {
 283                         debug_printf(" %s %s", argv[n], argv[n+1]);
 284                         key.vsaturate_r = key.fsaturate_r = strtol(argv[n+1], NULL, 0);
 285                         n += 2;
 286                         continue;
 287                 }
 288
 289                 if (!strcmp(argv[n], "--astc-srgb")) {
 290                         debug_printf(" %s %s", argv[n], argv[n+1]);
 291                         key.vastc_srgb = key.fastc_srgb = strtol(argv[n+1], NULL, 0);
 292                         n += 2;
 293                         continue;
 294                 }
 295
 296                 if (!strcmp(argv[n], "--stream-out")) {
 297                         struct pipe_stream_output_info *so = &s.stream_output;
 298                         debug_printf(" %s", argv[n]);
 299                         /* TODO more dynamic config based on number of outputs, etc
 300                          * rather than just hard-code for first output:
 301                          */
 302                         so->num_outputs = 1;
 303                         so->stride[0] = 4;
 304                         so->output[0].register_index = 0;
 305                         so->output[0].start_component = 0;
 306                         so->output[0].num_components = 4;
 307                         so->output[0].output_buffer = 0;
 308                         so->output[0].dst_offset = 2;
 309                         so->output[0].stream = 0;
 310                         n++;
 311                         continue;
 312                 }
 313
 314                 if (!strcmp(argv[n], "--ucp")) {
 315                         debug_printf(" %s %s", argv[n], argv[n+1]);
 316                         key.ucp_enables = strtol(argv[n+1], NULL, 0);
 317                         n += 2;
 318                         continue;
 319                 }
 320
 321                 if (!strcmp(argv[n], "--gpu")) {
 322                         debug_printf(" %s %s", argv[n], argv[n+1]);
 323                         gpu_id = strtol(argv[n+1], NULL, 0);
 324                         n += 2;
 325                         continue;
 326                 }
 327
 328                 if (!strcmp(argv[n], "--help")) {
 329                         print_usage();
 330                         return 0;
 331                 }
 332
 333                 break;
 334         }
 335         debug_printf("\n");
 336
 337         while (n < argc) {
 338                 char *filename = argv[n];
 339                 char *ext = rindex(filename, '.');
 340
 341                 if (strcmp(ext, ".tgsi") == 0) {
 342                         if (num_files != 0)
 343                                 errx(1, "in TGSI mode, only a single file may be specified");
 344                         s.from_tgsi = true;
 345                 } else if (strcmp(ext, ".frag") == 0) {
 346                         if (s.from_tgsi)
 347                                 errx(1, "cannot mix GLSL and TGSI");
 348                         if (num_files >= ARRAY_SIZE(filenames))
 349                                 errx(1, "too many GLSL files");
 350                         stage = MESA_SHADER_FRAGMENT;
 351                 } else if (strcmp(ext, ".vert") == 0) {
 352                         if (s.from_tgsi)
 353                                 errx(1, "cannot mix GLSL and TGSI");
 354                         if (num_files >= ARRAY_SIZE(filenames))
 355                                 errx(1, "too many GLSL files");
 356                         stage = MESA_SHADER_VERTEX;
 357                 } else {
 358                         print_usage();
 359                         return -1;
 360                 }
 361
 362                 filenames[num_files++] = filename;
 363
 364                 n++;
 365         }
 366
 367         nir_shader *nir;
 368
 369         compiler = ir3_compiler_create(NULL, gpu_id);
 370
 371         if (s.from_tgsi) {
 372                 struct tgsi_token toks[65536];
 373
 374                 ret = read_file(filenames[0], &ptr, &size);
 375                 if (ret) {
 376                         print_usage();
 377                         return ret;
 378                 }
 379
 380                 if (fd_mesa_debug & FD_DBG_OPTMSGS)
 381                         debug_printf("%s\n", (char *)ptr);
 382
 383                 if (!tgsi_text_translate(ptr, toks, ARRAY_SIZE(toks)))
 384                         errx(1, "could not parse `%s'", filenames[0]);
 385
 386                 if (fd_mesa_debug & FD_DBG_OPTMSGS)
 387                         tgsi_dump(toks, 0);
 388
 389                 nir = ir3_tgsi_to_nir(toks);
 390         } else if (num_files > 0) {
 391                 nir = load_glsl(num_files, filenames, stage);
 392         } else {
 393                 print_usage();
 394                 return -1;
 395         }
 396
 397         s.compiler = compiler;
 398         s.nir = ir3_optimize_nir(&s, nir, NULL);
 399
 400         v.key = key;
 401         v.shader = &s;
 402
 403         switch (nir->stage) {
 404         case MESA_SHADER_FRAGMENT:
 405                 s.type = v.type = SHADER_FRAGMENT;
 406                 break;
 407         case MESA_SHADER_VERTEX:
 408                 s.type = v.type = SHADER_VERTEX;
 409                 break;
 410         case MESA_SHADER_COMPUTE:
 411                 s.type = v.type = SHADER_COMPUTE;
 412                 break;
 413         default:
 414                 errx(1, "unhandled shader stage: %d", nir->stage);
 415         }
 416
 417         info = "NIR compiler";
 418         ret = ir3_compile_shader_nir(s.compiler, &v);
 419         if (ret) {
 420                 fprintf(stderr, "compiler failed!\n");
 421                 return ret;
 422         }
 423         dump_info(&v, info);
 424 }