freedreno/ir3: add ir3_finalize_nir()
[mesa.git] / 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 "nir/tgsi_to_nir.h"
37 #include "tgsi/tgsi_parse.h"
38 #include "tgsi/tgsi_text.h"
39 #include "tgsi/tgsi_dump.h"
40
41 #include "ir3/ir3_compiler.h"
42 #include "ir3/ir3_gallium.h"
43 #include "ir3/ir3_nir.h"
44 #include "ir3/instr-a3xx.h"
45 #include "ir3/ir3.h"
46
47 #include "main/mtypes.h"
48
49 #include "compiler/glsl/standalone.h"
50 #include "compiler/glsl/glsl_to_nir.h"
51 #include "compiler/glsl/gl_nir.h"
52 #include "compiler/nir_types.h"
53 #include "compiler/spirv/nir_spirv.h"
54
55 #include "pipe/p_context.h"
56
57 static void dump_info(struct ir3_shader_variant *so, const char *str)
58 {
59 uint32_t *bin;
60 const char *type = ir3_shader_stage(so);
61 bin = ir3_shader_assemble(so);
62 debug_printf("; %s: %s\n", type, str);
63 ir3_shader_disasm(so, bin, stdout);
64 free(bin);
65 }
66
67 static void
68 insert_sorted(struct exec_list *var_list, nir_variable *new_var)
69 {
70 nir_foreach_variable(var, var_list) {
71 if (var->data.location > new_var->data.location) {
72 exec_node_insert_node_before(&var->node, &new_var->node);
73 return;
74 }
75 }
76 exec_list_push_tail(var_list, &new_var->node);
77 }
78
79 static void
80 sort_varyings(struct exec_list *var_list)
81 {
82 struct exec_list new_list;
83 exec_list_make_empty(&new_list);
84 nir_foreach_variable_safe(var, var_list) {
85 exec_node_remove(&var->node);
86 insert_sorted(&new_list, var);
87 }
88 exec_list_move_nodes_to(&new_list, var_list);
89 }
90
91 static void
92 fixup_varying_slots(struct exec_list *var_list)
93 {
94 nir_foreach_variable(var, var_list) {
95 if (var->data.location >= VARYING_SLOT_VAR0) {
96 var->data.location += 9;
97 } else if ((var->data.location >= VARYING_SLOT_TEX0) &&
98 (var->data.location <= VARYING_SLOT_TEX7)) {
99 var->data.location += VARYING_SLOT_VAR0 - VARYING_SLOT_TEX0;
100 }
101 }
102 }
103
104 static struct ir3_compiler *compiler;
105
106 static nir_shader *
107 load_glsl(unsigned num_files, char* const* files, gl_shader_stage stage)
108 {
109 static const struct standalone_options options = {
110 .glsl_version = 460,
111 .do_link = true,
112 };
113 struct gl_shader_program *prog;
114 const nir_shader_compiler_options *nir_options =
115 ir3_get_compiler_options(compiler);
116 static struct gl_context local_ctx;
117
118 prog = standalone_compile_shader(&options, num_files, files, &local_ctx);
119 if (!prog)
120 errx(1, "couldn't parse `%s'", files[0]);
121
122 nir_shader *nir = glsl_to_nir(&local_ctx, prog, stage, nir_options);
123
124 /* required NIR passes: */
125 if (nir_options->lower_all_io_to_temps ||
126 nir->info.stage == MESA_SHADER_VERTEX ||
127 nir->info.stage == MESA_SHADER_GEOMETRY) {
128 NIR_PASS_V(nir, nir_lower_io_to_temporaries,
129 nir_shader_get_entrypoint(nir),
130 true, true);
131 } else if (nir->info.stage == MESA_SHADER_FRAGMENT) {
132 NIR_PASS_V(nir, nir_lower_io_to_temporaries,
133 nir_shader_get_entrypoint(nir),
134 true, false);
135 }
136
137 NIR_PASS_V(nir, nir_lower_global_vars_to_local);
138 NIR_PASS_V(nir, nir_split_var_copies);
139 NIR_PASS_V(nir, nir_lower_var_copies);
140
141 NIR_PASS_V(nir, nir_split_var_copies);
142 NIR_PASS_V(nir, nir_lower_var_copies);
143 nir_print_shader(nir, stdout);
144 NIR_PASS_V(nir, gl_nir_lower_atomics, prog, true);
145 NIR_PASS_V(nir, nir_lower_atomics_to_ssbo);
146 nir_print_shader(nir, stdout);
147
148 switch (stage) {
149 case MESA_SHADER_VERTEX:
150 nir_assign_var_locations(&nir->inputs,
151 &nir->num_inputs,
152 ir3_glsl_type_size);
153
154 /* Re-lower global vars, to deal with any dead VS inputs. */
155 NIR_PASS_V(nir, nir_lower_global_vars_to_local);
156
157 sort_varyings(&nir->outputs);
158 nir_assign_var_locations(&nir->outputs,
159 &nir->num_outputs,
160 ir3_glsl_type_size);
161 fixup_varying_slots(&nir->outputs);
162 break;
163 case MESA_SHADER_FRAGMENT:
164 sort_varyings(&nir->inputs);
165 nir_assign_var_locations(&nir->inputs,
166 &nir->num_inputs,
167 ir3_glsl_type_size);
168 fixup_varying_slots(&nir->inputs);
169 nir_assign_var_locations(&nir->outputs,
170 &nir->num_outputs,
171 ir3_glsl_type_size);
172 break;
173 case MESA_SHADER_COMPUTE:
174 case MESA_SHADER_KERNEL:
175 break;
176 default:
177 errx(1, "unhandled shader stage: %d", stage);
178 }
179
180 nir_assign_var_locations(&nir->uniforms,
181 &nir->num_uniforms,
182 ir3_glsl_type_size);
183
184 NIR_PASS_V(nir, nir_lower_system_values);
185 NIR_PASS_V(nir, nir_lower_frexp);
186 NIR_PASS_V(nir, nir_lower_io, nir_var_all, ir3_glsl_type_size, 0);
187 NIR_PASS_V(nir, gl_nir_lower_samplers, prog);
188
189 return nir;
190 }
191
192 static int
193 read_file(const char *filename, void **ptr, size_t *size)
194 {
195 int fd, ret;
196 struct stat st;
197
198 *ptr = MAP_FAILED;
199
200 fd = open(filename, O_RDONLY);
201 if (fd == -1) {
202 warnx("couldn't open `%s'", filename);
203 return 1;
204 }
205
206 ret = fstat(fd, &st);
207 if (ret)
208 errx(1, "couldn't stat `%s'", filename);
209
210 *size = st.st_size;
211 *ptr = mmap(NULL, st.st_size, PROT_READ, MAP_SHARED, fd, 0);
212 if (*ptr == MAP_FAILED)
213 errx(1, "couldn't map `%s'", filename);
214
215 close(fd);
216
217 return 0;
218 }
219
220 static void debug_func(void *priv, enum nir_spirv_debug_level level,
221 size_t spirv_offset, const char *message)
222 {
223 // printf("%s\n", message);
224 }
225
226 static nir_shader *
227 load_spirv(const char *filename, const char *entry, gl_shader_stage stage)
228 {
229 const struct spirv_to_nir_options spirv_options = {
230 /* these caps are just make-believe */
231 .caps = {
232 .draw_parameters = true,
233 .float64 = true,
234 .image_read_without_format = true,
235 .image_write_without_format = true,
236 .int64 = true,
237 .variable_pointers = true,
238 },
239 .lower_ubo_ssbo_access_to_offsets = true,
240 .debug = {
241 .func = debug_func,
242 }
243 };
244 nir_shader *nir;
245 void *buf;
246 size_t size;
247
248 read_file(filename, &buf, &size);
249
250 nir = spirv_to_nir(buf, size / 4,
251 NULL, 0, /* spec_entries */
252 stage, entry,
253 &spirv_options,
254 ir3_get_compiler_options(compiler));
255
256 nir_print_shader(nir, stdout);
257
258 return nir;
259 }
260
261 static void print_usage(void)
262 {
263 printf("Usage: ir3_compiler [OPTIONS]... <file.tgsi | file.spv entry_point | (file.vert | file.frag)*>\n");
264 printf(" --verbose - verbose compiler/debug messages\n");
265 printf(" --binning-pass - generate binning pass shader (VERT)\n");
266 printf(" --color-two-side - emulate two-sided color (FRAG)\n");
267 printf(" --half-precision - use half-precision\n");
268 printf(" --saturate-s MASK - bitmask of samplers to saturate S coord\n");
269 printf(" --saturate-t MASK - bitmask of samplers to saturate T coord\n");
270 printf(" --saturate-r MASK - bitmask of samplers to saturate R coord\n");
271 printf(" --astc-srgb MASK - bitmask of samplers to enable astc-srgb workaround\n");
272 printf(" --stream-out - enable stream-out (aka transform feedback)\n");
273 printf(" --ucp MASK - bitmask of enabled user-clip-planes\n");
274 printf(" --gpu GPU_ID - specify gpu-id (default 320)\n");
275 printf(" --help - show this message\n");
276 }
277
278 int main(int argc, char **argv)
279 {
280 int ret = 0, n = 1;
281 char *filenames[2];
282 int num_files = 0;
283 unsigned stage = 0;
284 struct ir3_shader_variant v;
285 struct ir3_shader s;
286 struct ir3_shader_key key = {};
287 /* TODO cmdline option to target different gpus: */
288 unsigned gpu_id = 320;
289 const char *info;
290 const char *entry;
291 void *ptr;
292 bool from_spirv = false;
293 bool from_tgsi = false;
294 size_t size;
295
296 memset(&s, 0, sizeof(s));
297 memset(&v, 0, sizeof(v));
298
299 /* cmdline args which impact shader variant get spit out in a
300 * comment on the first line.. a quick/dirty way to preserve
301 * that info so when ir3test recompiles the shader with a new
302 * compiler version, we use the same shader-key settings:
303 */
304 debug_printf("; options:");
305
306 while (n < argc) {
307 if (!strcmp(argv[n], "--verbose")) {
308 ir3_shader_debug |= IR3_DBG_OPTMSGS | IR3_DBG_DISASM;
309 n++;
310 continue;
311 }
312
313 if (!strcmp(argv[n], "--binning-pass")) {
314 debug_printf(" %s", argv[n]);
315 v.binning_pass = true;
316 n++;
317 continue;
318 }
319
320 if (!strcmp(argv[n], "--color-two-side")) {
321 debug_printf(" %s", argv[n]);
322 key.color_two_side = true;
323 n++;
324 continue;
325 }
326
327 if (!strcmp(argv[n], "--half-precision")) {
328 debug_printf(" %s", argv[n]);
329 n++;
330 continue;
331 }
332
333 if (!strcmp(argv[n], "--saturate-s")) {
334 debug_printf(" %s %s", argv[n], argv[n+1]);
335 key.vsaturate_s = key.fsaturate_s = strtol(argv[n+1], NULL, 0);
336 n += 2;
337 continue;
338 }
339
340 if (!strcmp(argv[n], "--saturate-t")) {
341 debug_printf(" %s %s", argv[n], argv[n+1]);
342 key.vsaturate_t = key.fsaturate_t = strtol(argv[n+1], NULL, 0);
343 n += 2;
344 continue;
345 }
346
347 if (!strcmp(argv[n], "--saturate-r")) {
348 debug_printf(" %s %s", argv[n], argv[n+1]);
349 key.vsaturate_r = key.fsaturate_r = strtol(argv[n+1], NULL, 0);
350 n += 2;
351 continue;
352 }
353
354 if (!strcmp(argv[n], "--astc-srgb")) {
355 debug_printf(" %s %s", argv[n], argv[n+1]);
356 key.vastc_srgb = key.fastc_srgb = strtol(argv[n+1], NULL, 0);
357 n += 2;
358 continue;
359 }
360
361 if (!strcmp(argv[n], "--stream-out")) {
362 struct ir3_stream_output_info *so = &s.stream_output;
363 debug_printf(" %s", argv[n]);
364 /* TODO more dynamic config based on number of outputs, etc
365 * rather than just hard-code for first output:
366 */
367 so->num_outputs = 1;
368 so->stride[0] = 4;
369 so->output[0].register_index = 0;
370 so->output[0].start_component = 0;
371 so->output[0].num_components = 4;
372 so->output[0].output_buffer = 0;
373 so->output[0].dst_offset = 2;
374 so->output[0].stream = 0;
375 n++;
376 continue;
377 }
378
379 if (!strcmp(argv[n], "--ucp")) {
380 debug_printf(" %s %s", argv[n], argv[n+1]);
381 key.ucp_enables = strtol(argv[n+1], NULL, 0);
382 n += 2;
383 continue;
384 }
385
386 if (!strcmp(argv[n], "--gpu")) {
387 debug_printf(" %s %s", argv[n], argv[n+1]);
388 gpu_id = strtol(argv[n+1], NULL, 0);
389 n += 2;
390 continue;
391 }
392
393 if (!strcmp(argv[n], "--help")) {
394 print_usage();
395 return 0;
396 }
397
398 break;
399 }
400 debug_printf("\n");
401
402 while (n < argc) {
403 char *filename = argv[n];
404 char *ext = strrchr(filename, '.');
405
406 if (strcmp(ext, ".tgsi") == 0) {
407 if (num_files != 0)
408 errx(1, "in TGSI mode, only a single file may be specified");
409 from_tgsi = true;
410 } else if (strcmp(ext, ".spv") == 0) {
411 if (num_files != 0)
412 errx(1, "in SPIR-V mode, only a single file may be specified");
413 stage = MESA_SHADER_COMPUTE;
414 from_spirv = true;
415 filenames[num_files++] = filename;
416 n++;
417 if (n == argc)
418 errx(1, "in SPIR-V mode, an entry point must be specified");
419 entry = argv[n];
420 n++;
421 } else if (strcmp(ext, ".comp") == 0) {
422 if (from_tgsi || from_spirv)
423 errx(1, "cannot mix GLSL/TGSI/SPIRV");
424 if (num_files >= ARRAY_SIZE(filenames))
425 errx(1, "too many GLSL files");
426 stage = MESA_SHADER_COMPUTE;
427 } else if (strcmp(ext, ".frag") == 0) {
428 if (from_tgsi || from_spirv)
429 errx(1, "cannot mix GLSL/TGSI/SPIRV");
430 if (num_files >= ARRAY_SIZE(filenames))
431 errx(1, "too many GLSL files");
432 stage = MESA_SHADER_FRAGMENT;
433 } else if (strcmp(ext, ".vert") == 0) {
434 if (from_tgsi)
435 errx(1, "cannot mix GLSL and TGSI");
436 if (num_files >= ARRAY_SIZE(filenames))
437 errx(1, "too many GLSL files");
438 stage = MESA_SHADER_VERTEX;
439 } else {
440 print_usage();
441 return -1;
442 }
443
444 filenames[num_files++] = filename;
445
446 n++;
447 }
448
449 nir_shader *nir;
450
451 compiler = ir3_compiler_create(NULL, gpu_id);
452
453 if (from_tgsi) {
454 struct tgsi_token toks[65536];
455 const nir_shader_compiler_options *nir_options =
456 ir3_get_compiler_options(compiler);
457
458 ret = read_file(filenames[0], &ptr, &size);
459 if (ret) {
460 print_usage();
461 return ret;
462 }
463
464 if (ir3_shader_debug & IR3_DBG_OPTMSGS)
465 debug_printf("%s\n", (char *)ptr);
466
467 if (!tgsi_text_translate(ptr, toks, ARRAY_SIZE(toks)))
468 errx(1, "could not parse `%s'", filenames[0]);
469
470 if (ir3_shader_debug & IR3_DBG_OPTMSGS)
471 tgsi_dump(toks, 0);
472
473 nir = tgsi_to_nir_noscreen(toks, nir_options);
474 NIR_PASS_V(nir, nir_lower_global_vars_to_local);
475 } else if (from_spirv) {
476 nir = load_spirv(filenames[0], entry, stage);
477
478 NIR_PASS_V(nir, nir_lower_io, nir_var_all, ir3_glsl_type_size,
479 (nir_lower_io_options)0);
480
481 /* TODO do this somewhere else */
482 nir_lower_int64(nir, ~0);
483 nir_lower_system_values(nir);
484 } else if (num_files > 0) {
485 nir = load_glsl(num_files, filenames, stage);
486 } else {
487 print_usage();
488 return -1;
489 }
490
491 s.compiler = compiler;
492 s.nir = nir;
493
494 ir3_finalize_nir(compiler, nir);
495
496 v.key = key;
497 v.shader = &s;
498 s.type = v.type = nir->info.stage;
499
500 ir3_nir_lower_variant(&v, nir);
501
502 info = "NIR compiler";
503 ret = ir3_compile_shader_nir(s.compiler, &v);
504 if (ret) {
505 fprintf(stderr, "compiler failed!\n");
506 return ret;
507 }
508 dump_info(&v, info);
509 }