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