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 "ir3/ir3_compiler.h"
41#include "ir3/ir3_gallium.h"
42#include "ir3/ir3_nir.h"
43#include "ir3/instr-a3xx.h"
44#include "ir3/ir3.h"
45
46#include "main/mtypes.h"
47
48#include "compiler/glsl/standalone.h"
49#include "compiler/glsl/glsl_to_nir.h"
50#include "compiler/glsl/gl_nir.h"
51#include "compiler/nir_types.h"
52#include "compiler/spirv/nir_spirv.h"
53
54#include "pipe/p_context.h"
55
56static void dump_info(struct ir3_shader_variant *so, const char *str)
57{
58	uint32_t *bin;
59	const char *type = ir3_shader_stage(so->shader);
60	bin = ir3_shader_assemble(so, so->shader->compiler->gpu_id);
61	debug_printf("; %s: %s\n", type, str);
62	ir3_shader_disasm(so, bin, stdout);
63	free(bin);
64}
65
66static void
67insert_sorted(struct exec_list *var_list, nir_variable *new_var)
68{
69	nir_foreach_variable(var, var_list) {
70		if (var->data.location > new_var->data.location) {
71			exec_node_insert_node_before(&var->node, &new_var->node);
72			return;
73		}
74	}
75	exec_list_push_tail(var_list, &new_var->node);
76}
77
78static void
79sort_varyings(struct exec_list *var_list)
80{
81	struct exec_list new_list;
82	exec_list_make_empty(&new_list);
83	nir_foreach_variable_safe(var, var_list) {
84		exec_node_remove(&var->node);
85		insert_sorted(&new_list, var);
86	}
87	exec_list_move_nodes_to(&new_list, var_list);
88}
89
90static void
91fixup_varying_slots(struct exec_list *var_list)
92{
93	nir_foreach_variable(var, var_list) {
94		if (var->data.location >= VARYING_SLOT_VAR0) {
95			var->data.location += 9;
96		} else if ((var->data.location >= VARYING_SLOT_TEX0) &&
97				(var->data.location <= VARYING_SLOT_TEX7)) {
98			var->data.location += VARYING_SLOT_VAR0 - VARYING_SLOT_TEX0;
99		}
100	}
101}
102
103static struct ir3_compiler *compiler;
104
105static nir_shader *
106load_glsl(unsigned num_files, char* const* files, gl_shader_stage stage)
107{
108	static const struct standalone_options options = {
109			.glsl_version = 460,
110			.do_link = true,
111	};
112	struct gl_shader_program *prog;
113	const nir_shader_compiler_options *nir_options =
114			ir3_get_compiler_options(compiler);
115	static struct gl_context local_ctx;
116
117	prog = standalone_compile_shader(&options, num_files, files, &local_ctx);
118	if (!prog)
119		errx(1, "couldn't parse `%s'", files[0]);
120
121	nir_shader *nir = glsl_to_nir(&local_ctx, prog, stage, nir_options);
122
123	/* required NIR passes: */
124	if (nir_options->lower_all_io_to_temps ||
125			nir->info.stage == MESA_SHADER_VERTEX ||
126			nir->info.stage == MESA_SHADER_GEOMETRY) {
127		NIR_PASS_V(nir, nir_lower_io_to_temporaries,
128				nir_shader_get_entrypoint(nir),
129				true, true);
130	} else if (nir->info.stage == MESA_SHADER_FRAGMENT) {
131		NIR_PASS_V(nir, nir_lower_io_to_temporaries,
132				nir_shader_get_entrypoint(nir),
133				true, false);
134	}
135
136	NIR_PASS_V(nir, nir_lower_global_vars_to_local);
137	NIR_PASS_V(nir, nir_split_var_copies);
138	NIR_PASS_V(nir, nir_lower_var_copies);
139
140	NIR_PASS_V(nir, nir_split_var_copies);
141	NIR_PASS_V(nir, nir_lower_var_copies);
142	nir_print_shader(nir, stdout);
143	NIR_PASS_V(nir, gl_nir_lower_atomics, prog, true);
144	NIR_PASS_V(nir, nir_lower_atomics_to_ssbo, 8);
145	nir_print_shader(nir, stdout);
146
147	switch (stage) {
148	case MESA_SHADER_VERTEX:
149		nir_assign_var_locations(&nir->inputs,
150				&nir->num_inputs,
151				ir3_glsl_type_size);
152
153		/* Re-lower global vars, to deal with any dead VS inputs. */
154		NIR_PASS_V(nir, nir_lower_global_vars_to_local);
155
156		sort_varyings(&nir->outputs);
157		nir_assign_var_locations(&nir->outputs,
158				&nir->num_outputs,
159				ir3_glsl_type_size);
160		fixup_varying_slots(&nir->outputs);
161		break;
162	case MESA_SHADER_FRAGMENT:
163		sort_varyings(&nir->inputs);
164		nir_assign_var_locations(&nir->inputs,
165				&nir->num_inputs,
166				ir3_glsl_type_size);
167		fixup_varying_slots(&nir->inputs);
168		nir_assign_var_locations(&nir->outputs,
169				&nir->num_outputs,
170				ir3_glsl_type_size);
171		break;
172	case MESA_SHADER_COMPUTE:
173	case MESA_SHADER_KERNEL:
174		break;
175	default:
176		errx(1, "unhandled shader stage: %d", stage);
177	}
178
179	nir_assign_var_locations(&nir->uniforms,
180			&nir->num_uniforms,
181			ir3_glsl_type_size);
182
183	NIR_PASS_V(nir, nir_lower_system_values);
184	NIR_PASS_V(nir, nir_lower_frexp);
185	NIR_PASS_V(nir, nir_lower_io, nir_var_all, ir3_glsl_type_size, 0);
186	NIR_PASS_V(nir, gl_nir_lower_samplers, prog);
187
188	return nir;
189}
190
191static int
192read_file(const char *filename, void **ptr, size_t *size)
193{
194	int fd, ret;
195	struct stat st;
196
197	*ptr = MAP_FAILED;
198
199	fd = open(filename, O_RDONLY);
200	if (fd == -1) {
201		warnx("couldn't open `%s'", filename);
202		return 1;
203	}
204
205	ret = fstat(fd, &st);
206	if (ret)
207		errx(1, "couldn't stat `%s'", filename);
208
209	*size = st.st_size;
210	*ptr = mmap(NULL, st.st_size, PROT_READ, MAP_SHARED, fd, 0);
211	if (*ptr == MAP_FAILED)
212		errx(1, "couldn't map `%s'", filename);
213
214	close(fd);
215
216	return 0;
217}
218
219static void debug_func(void *priv, enum nir_spirv_debug_level level,
220		size_t spirv_offset, const char *message)
221{
222//	printf("%s\n", message);
223}
224
225static nir_shader *
226load_spirv(const char *filename, const char *entry, gl_shader_stage stage)
227{
228	const struct spirv_to_nir_options spirv_options = {
229		/* these caps are just make-believe */
230		.caps = {
231			.draw_parameters = true,
232			.float64 = true,
233			.image_read_without_format = true,
234			.image_write_without_format = true,
235			.int64 = true,
236			.variable_pointers = true,
237		},
238		.lower_workgroup_access_to_offsets = true,
239		.lower_ubo_ssbo_access_to_offsets = true,
240		.debug = {
241			.func = debug_func,
242		}
243	};
244	nir_function *entry_point;
245	void *buf;
246	size_t size;
247
248	read_file(filename, &buf, &size);
249
250	entry_point = 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(entry_point->shader, stdout);
257
258	return entry_point->shader;
259}
260
261static 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
278int 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	size_t size;
294
295	memset(&s, 0, sizeof(s));
296	memset(&v, 0, sizeof(v));
297
298	/* cmdline args which impact shader variant get spit out in a
299	 * comment on the first line..  a quick/dirty way to preserve
300	 * that info so when ir3test recompiles the shader with a new
301	 * compiler version, we use the same shader-key settings:
302	 */
303	debug_printf("; options:");
304
305	while (n < argc) {
306		if (!strcmp(argv[n], "--verbose")) {
307			ir3_shader_debug |= IR3_DBG_OPTMSGS | IR3_DBG_DISASM;
308			n++;
309			continue;
310		}
311
312		if (!strcmp(argv[n], "--binning-pass")) {
313			debug_printf(" %s", argv[n]);
314			v.binning_pass = true;
315			n++;
316			continue;
317		}
318
319		if (!strcmp(argv[n], "--color-two-side")) {
320			debug_printf(" %s", argv[n]);
321			key.color_two_side = true;
322			n++;
323			continue;
324		}
325
326		if (!strcmp(argv[n], "--half-precision")) {
327			debug_printf(" %s", argv[n]);
328			key.half_precision = true;
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			s.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 (s.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 (s.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 (s.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 (s.from_tgsi) {
454		struct tgsi_token toks[65536];
455
456		ret = read_file(filenames[0], &ptr, &size);
457		if (ret) {
458			print_usage();
459			return ret;
460		}
461
462		if (ir3_shader_debug & IR3_DBG_OPTMSGS)
463			debug_printf("%s\n", (char *)ptr);
464
465		if (!tgsi_text_translate(ptr, toks, ARRAY_SIZE(toks)))
466			errx(1, "could not parse `%s'", filenames[0]);
467
468		if (ir3_shader_debug & IR3_DBG_OPTMSGS)
469			tgsi_dump(toks, 0);
470
471		nir = ir3_tgsi_to_nir(compiler, toks, NULL);
472		NIR_PASS_V(nir, nir_lower_global_vars_to_local);
473	} else if (from_spirv) {
474		nir = load_spirv(filenames[0], entry, stage);
475
476		NIR_PASS_V(nir, nir_lower_io, nir_var_all, ir3_glsl_type_size,
477				(nir_lower_io_options)0);
478
479		/* TODO do this somewhere else */
480		nir_lower_int64(nir, ~0);
481		nir_lower_system_values(nir);
482	} else if (num_files > 0) {
483		nir = load_glsl(num_files, filenames, stage);
484	} else {
485		print_usage();
486		return -1;
487	}
488
489	s.compiler = compiler;
490	s.nir = ir3_optimize_nir(&s, nir, NULL);
491
492	v.key = key;
493	v.shader = &s;
494	s.type = v.type = nir->info.stage;
495
496	info = "NIR compiler";
497	ret = ir3_compile_shader_nir(s.compiler, &v);
498	if (ret) {
499		fprintf(stderr, "compiler failed!\n");
500		return ret;
501	}
502	dump_info(&v, info);
503}
504