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