• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2021 Collabora, Ltd.
3  * Copyright (C) 2019 Ryan Houdek <Sonicadvance1@gmail.com>
4  * Copyright (C) 2014 Rob Clark <robclark@freedesktop.org>
5  * Copyright © 2015 Red Hat
6  *
7  * Permission is hereby granted, free of charge, to any person obtaining a
8  * copy of this software and associated documentation files (the "Software"),
9  * to deal in the Software without restriction, including without limitation
10  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
11  * and/or sell copies of the Software, and to permit persons to whom the
12  * Software is furnished to do so, subject to the following conditions:
13  *
14  * The above copyright notice and this permission notice (including the next
15  * paragraph) shall be included in all copies or substantial portions of the
16  * Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
24  * SOFTWARE.
25  */
26 
27 #include <getopt.h>
28 #include <string.h>
29 #include "disassemble.h"
30 #include "valhall/disassemble.h"
31 #include "compiler.h"
32 
33 #include "main/mtypes.h"
34 #include "compiler/glsl/standalone.h"
35 #include "compiler/glsl/glsl_to_nir.h"
36 #include "compiler/glsl/gl_nir.h"
37 #include "compiler/nir_types.h"
38 #include "util/u_dynarray.h"
39 #include "bifrost_compile.h"
40 
41 unsigned gpu_id = 0x7212;
42 int verbose = 0;
43 
44 static gl_shader_stage
filename_to_stage(const char * stage)45 filename_to_stage(const char *stage)
46 {
47         const char *ext = strrchr(stage, '.');
48 
49         if (ext == NULL) {
50                 fprintf(stderr, "No extension found in %s\n", stage);
51                 exit(1);
52         }
53 
54         if (!strcmp(ext, ".cs") || !strcmp(ext, ".comp"))
55                 return MESA_SHADER_COMPUTE;
56         else if (!strcmp(ext, ".vs") || !strcmp(ext, ".vert"))
57                 return MESA_SHADER_VERTEX;
58         else if (!strcmp(ext, ".fs") || !strcmp(ext, ".frag"))
59                 return MESA_SHADER_FRAGMENT;
60         else {
61                 fprintf(stderr, "Invalid extension %s\n", ext);
62                 exit(1);
63         }
64 
65         unreachable("Should've returned or bailed");
66 }
67 
68 static int
st_packed_uniforms_type_size(const struct glsl_type * type,bool bindless)69 st_packed_uniforms_type_size(const struct glsl_type *type, bool bindless)
70 {
71    return glsl_count_dword_slots(type, bindless);
72 }
73 
74 static int
glsl_type_size(const struct glsl_type * type,bool bindless)75 glsl_type_size(const struct glsl_type *type, bool bindless)
76 {
77    return glsl_count_attribute_slots(type, false);
78 }
79 
80 static void
insert_sorted(struct exec_list * var_list,nir_variable * new_var)81 insert_sorted(struct exec_list *var_list, nir_variable *new_var)
82 {
83    nir_foreach_variable_in_list (var, var_list) {
84       if (var->data.location > new_var->data.location) {
85          exec_node_insert_node_before(&var->node, &new_var->node);
86          return;
87       }
88    }
89    exec_list_push_tail(var_list, &new_var->node);
90 }
91 
92 static void
sort_varyings(nir_shader * nir,nir_variable_mode mode)93 sort_varyings(nir_shader *nir, nir_variable_mode mode)
94 {
95    struct exec_list new_list;
96    exec_list_make_empty(&new_list);
97    nir_foreach_variable_with_modes_safe (var, nir, mode) {
98       exec_node_remove(&var->node);
99       insert_sorted(&new_list, var);
100    }
101    exec_list_append(&nir->variables, &new_list);
102 }
103 
104 static void
fixup_varying_slots(nir_shader * nir,nir_variable_mode mode)105 fixup_varying_slots(nir_shader *nir, nir_variable_mode mode)
106 {
107    nir_foreach_variable_with_modes (var, nir, mode) {
108       if (var->data.location >= VARYING_SLOT_VAR0) {
109          var->data.location += 9;
110       } else if ((var->data.location >= VARYING_SLOT_TEX0) &&
111                  (var->data.location <= VARYING_SLOT_TEX7)) {
112          var->data.location += VARYING_SLOT_VAR0 - VARYING_SLOT_TEX0;
113       }
114    }
115 }
116 
117 static void
compile_shader(int stages,char ** files)118 compile_shader(int stages, char **files)
119 {
120         struct gl_shader_program *prog;
121         nir_shader *nir[MESA_SHADER_COMPUTE + 1];
122         unsigned shader_types[MESA_SHADER_COMPUTE + 1];
123 
124         if (stages > MESA_SHADER_COMPUTE) {
125                 fprintf(stderr, "Too many stages");
126                 exit(1);
127         }
128 
129         for (unsigned i = 0; i < stages; ++i)
130                 shader_types[i] = filename_to_stage(files[i]);
131 
132         struct standalone_options options = {
133                 .glsl_version = 300, /* ES - needed for precision */
134                 .do_link = true,
135                 .lower_precision = true
136         };
137 
138         static struct gl_context local_ctx;
139 
140         prog = standalone_compile_shader(&options, stages, files, &local_ctx);
141 
142         for (unsigned i = 0; i < stages; ++i) {
143                 gl_shader_stage stage = shader_types[i];
144                 prog->_LinkedShaders[stage]->Program->info.stage = stage;
145         }
146 
147         struct util_dynarray binary;
148 
149         util_dynarray_init(&binary, NULL);
150 
151         for (unsigned i = 0; i < stages; ++i) {
152                 nir[i] = glsl_to_nir(&local_ctx.Const, prog, shader_types[i], &bifrost_nir_options);
153 
154                 if (shader_types[i] == MESA_SHADER_VERTEX) {
155                         nir_assign_var_locations(nir[i], nir_var_shader_in, &nir[i]->num_inputs,
156                                         glsl_type_size);
157                         sort_varyings(nir[i], nir_var_shader_out);
158                         nir_assign_var_locations(nir[i], nir_var_shader_out, &nir[i]->num_outputs,
159                                         glsl_type_size);
160                         fixup_varying_slots(nir[i], nir_var_shader_out);
161                 } else if (shader_types[i] == MESA_SHADER_FRAGMENT) {
162                       sort_varyings(nir[i], nir_var_shader_in);
163                       nir_assign_var_locations(nir[i], nir_var_shader_in, &nir[i]->num_inputs,
164                                       glsl_type_size);
165                       fixup_varying_slots(nir[i], nir_var_shader_in);
166                       nir_assign_var_locations(nir[i], nir_var_shader_out, &nir[i]->num_outputs,
167                                       glsl_type_size);
168                 }
169 
170                 nir_assign_var_locations(nir[i], nir_var_uniform, &nir[i]->num_uniforms,
171                                 glsl_type_size);
172 
173                 NIR_PASS_V(nir[i], nir_lower_global_vars_to_local);
174                 NIR_PASS_V(nir[i], nir_lower_io_to_temporaries, nir_shader_get_entrypoint(nir[i]), true, i == 0);
175                 NIR_PASS_V(nir[i], nir_opt_copy_prop_vars);
176                 NIR_PASS_V(nir[i], nir_opt_combine_stores, nir_var_all);
177 
178                 NIR_PASS_V(nir[i], nir_lower_system_values);
179                 NIR_PASS_V(nir[i], gl_nir_lower_samplers, prog);
180                 NIR_PASS_V(nir[i], nir_split_var_copies);
181                 NIR_PASS_V(nir[i], nir_lower_var_copies);
182 
183                 NIR_PASS_V(nir[i], nir_lower_io, nir_var_uniform,
184                                 st_packed_uniforms_type_size,
185                                 (nir_lower_io_options)0);
186                 NIR_PASS_V(nir[i], nir_lower_uniforms_to_ubo, true, false);
187 
188                 /* before buffers and vars_to_ssa */
189                 NIR_PASS_V(nir[i], gl_nir_lower_images, true);
190 
191                 NIR_PASS_V(nir[i], gl_nir_lower_buffers, prog);
192                 NIR_PASS_V(nir[i], nir_opt_constant_folding);
193 
194                 struct panfrost_compile_inputs inputs = {
195                         .gpu_id = gpu_id,
196                         .fixed_sysval_ubo = -1,
197                 };
198                 struct pan_shader_info info = { 0 };
199 
200                 util_dynarray_clear(&binary);
201                 bifrost_compile_shader_nir(nir[i], &inputs, &binary, &info);
202 
203                 char *fn = NULL;
204                 asprintf(&fn, "shader_%u.bin", i);
205                 assert(fn != NULL);
206                 FILE *fp = fopen(fn, "wb");
207                 fwrite(binary.data, 1, binary.size, fp);
208                 fclose(fp);
209                 free(fn);
210         }
211 
212         util_dynarray_fini(&binary);
213 }
214 
215 #define BI_FOURCC(ch0, ch1, ch2, ch3) ( \
216   (uint32_t)(ch0)        | (uint32_t)(ch1) << 8 | \
217   (uint32_t)(ch2) << 16  | (uint32_t)(ch3) << 24)
218 
219 static void
disassemble(const char * filename)220 disassemble(const char *filename)
221 {
222         FILE *fp = fopen(filename, "rb");
223         assert(fp);
224 
225         fseek(fp, 0, SEEK_END);
226         unsigned filesize = ftell(fp);
227         rewind(fp);
228 
229         uint32_t *code = malloc(filesize);
230         unsigned res = fread(code, 1, filesize, fp);
231         if (res != filesize) {
232                 printf("Couldn't read full file\n");
233         }
234 
235         fclose(fp);
236 
237         void *entrypoint = code;
238 
239         if (filesize && code[0] == BI_FOURCC('M', 'B', 'S', '2')) {
240                 for (int i = 0; i < filesize / 4; ++i) {
241                         if (code[i] != BI_FOURCC('O', 'B', 'J', 'C'))
242                                 continue;
243 
244                         unsigned size = code[i + 1];
245                         unsigned offset = i + 2;
246 
247                         entrypoint = code + offset;
248                         filesize = size;
249                 }
250         }
251 
252         if ((gpu_id >> 12) >= 9)
253                 disassemble_valhall(stdout, entrypoint, filesize, verbose);
254         else
255                 disassemble_bifrost(stdout, entrypoint, filesize, verbose);
256 
257         free(code);
258 }
259 
260 int
main(int argc,char ** argv)261 main(int argc, char **argv)
262 {
263         int c;
264 
265         if (argc < 2) {
266                 printf("Pass a command\n");
267                 exit(1);
268         }
269 
270         static struct option longopts[] = {
271                 { "id", optional_argument, NULL, 'i' },
272                 { "gpu", optional_argument, NULL, 'g' },
273                 { "verbose", no_argument, &verbose, 'v' },
274                 { NULL, 0, NULL, 0 }
275         };
276 
277         static struct {
278                 const char *name;
279                 unsigned major, minor;
280         } gpus[] = {
281                 { "G71",   6, 0 },
282                 { "G72",   6, 2 },
283                 { "G51",   7, 0 },
284                 { "G76",   7, 1 },
285                 { "G52",   7, 2 },
286                 { "G31",   7, 3 },
287                 { "G77",   9, 0 },
288                 { "G57",   9, 1 },
289                 { "G78",   9, 2 },
290                 { "G57",   9, 3 },
291                 { "G68",   9, 4 },
292                 { "G78AE", 9, 5 },
293         };
294 
295         while ((c = getopt_long(argc, argv, "v:", longopts, NULL)) != -1) {
296 
297                 switch (c) {
298                 case 'i':
299                         gpu_id = atoi(optarg);
300 
301                         if (!gpu_id) {
302                                 fprintf(stderr, "Expected GPU ID, got %s\n", optarg);
303                                 return 1;
304                         }
305 
306                         break;
307                 case 'g':
308                         gpu_id = 0;
309 
310                         /* Compatibility with the Arm compiler */
311                         if (strncmp(optarg, "Mali-", 5) == 0) optarg += 5;
312 
313                         for (unsigned i = 0; i < ARRAY_SIZE(gpus); ++i) {
314                                 if (strcmp(gpus[i].name, optarg)) continue;
315 
316                                 unsigned major = gpus[i].major;
317                                 unsigned minor = gpus[i].minor;
318 
319                                 gpu_id = (major << 12) | (minor << 8);
320                                 break;
321                         }
322 
323                         if (!gpu_id) {
324                                 fprintf(stderr, "Unknown GPU %s\n", optarg);
325                                 return 1;
326                         }
327 
328                         break;
329                 default:
330                         break;
331                 }
332         }
333 
334         if (strcmp(argv[optind], "compile") == 0)
335                 compile_shader(argc - optind - 1, &argv[optind + 1]);
336         else if (strcmp(argv[optind], "disasm") == 0)
337                 disassemble(argv[optind + 1]);
338         else {
339                 fprintf(stderr, "Unknown command. Valid: compile/disasm\n");
340                 return 1;
341         }
342 
343         return 0;
344 }
345