Merge branch 'blender2.7'
[blender.git] / source / blender / gpu / intern / gpu_codegen.c
1 /*
2  * This program is free software; you can redistribute it and/or
3  * modify it under the terms of the GNU General Public License
4  * as published by the Free Software Foundation; either version 2
5  * of the License, or (at your option) any later version.
6  *
7  * This program is distributed in the hope that it will be useful,
8  * but WITHOUT ANY WARRANTY; without even the implied warranty of
9  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10  * GNU General Public License for more details.
11  *
12  * You should have received a copy of the GNU General Public License
13  * along with this program; if not, write to the Free Software Foundation,
14  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
15  *
16  * The Original Code is Copyright (C) 2005 Blender Foundation.
17  * All rights reserved.
18  */
19
20 /** \file
21  * \ingroup gpu
22  *
23  * Convert material node-trees to GLSL.
24  */
25
26 #include "MEM_guardedalloc.h"
27
28 #include "DNA_customdata_types.h"
29 #include "DNA_image_types.h"
30 #include "DNA_material_types.h"
31 #include "DNA_node_types.h"
32
33 #include "BLI_blenlib.h"
34 #include "BLI_hash_mm2a.h"
35 #include "BLI_link_utils.h"
36 #include "BLI_utildefines.h"
37 #include "BLI_dynstr.h"
38 #include "BLI_ghash.h"
39 #include "BLI_threads.h"
40
41 #include "PIL_time.h"
42
43 #include "GPU_extensions.h"
44 #include "GPU_glew.h"
45 #include "GPU_material.h"
46 #include "GPU_shader.h"
47 #include "GPU_texture.h"
48 #include "GPU_uniformbuffer.h"
49
50 #include "BLI_sys_types.h" /* for intptr_t support */
51
52 #include "gpu_codegen.h"
53
54 #include <string.h>
55 #include <stdarg.h>
56
57 extern char datatoc_gpu_shader_material_glsl[];
58 extern char datatoc_gpu_shader_geometry_glsl[];
59
60 static char *glsl_material_library = NULL;
61
62 /* -------------------- GPUPass Cache ------------------ */
63 /**
64  * Internal shader cache: This prevent the shader recompilation / stall when
65  * using undo/redo AND also allows for GPUPass reuse if the Shader code is the
66  * same for 2 different Materials. Unused GPUPasses are free by Garbage collection.
67  */
68
69 /* Only use one linklist that contains the GPUPasses grouped by hash. */
70 static GPUPass *pass_cache = NULL;
71 static SpinLock pass_cache_spin;
72
73 static uint32_t gpu_pass_hash(const char *frag_gen, const char *defs, GPUVertAttrLayers *attrs)
74 {
75         BLI_HashMurmur2A hm2a;
76         BLI_hash_mm2a_init(&hm2a, 0);
77         BLI_hash_mm2a_add(&hm2a, (uchar *)frag_gen, strlen(frag_gen));
78         if (attrs) {
79                 for (int att_idx = 0; att_idx < attrs->totlayer; att_idx++) {
80                         char *name = attrs->layer[att_idx].name;
81                         BLI_hash_mm2a_add(&hm2a, (uchar *)name, strlen(name));
82                 }
83         }
84         if (defs)
85                 BLI_hash_mm2a_add(&hm2a, (uchar *)defs, strlen(defs));
86
87         return BLI_hash_mm2a_end(&hm2a);
88 }
89
90 /* Search by hash only. Return first pass with the same hash.
91  * There is hash collision if (pass->next && pass->next->hash == hash) */
92 static GPUPass *gpu_pass_cache_lookup(uint32_t hash)
93 {
94         BLI_spin_lock(&pass_cache_spin);
95         /* Could be optimized with a Lookup table. */
96         for (GPUPass *pass = pass_cache; pass; pass = pass->next) {
97                 if (pass->hash == hash) {
98                         BLI_spin_unlock(&pass_cache_spin);
99                         return pass;
100                 }
101         }
102         BLI_spin_unlock(&pass_cache_spin);
103         return NULL;
104 }
105
106 /* Check all possible passes with the same hash. */
107 static GPUPass *gpu_pass_cache_resolve_collision(
108         GPUPass *pass, const char *vert, const char *geom, const char *frag, const char *defs, uint32_t hash)
109 {
110         BLI_spin_lock(&pass_cache_spin);
111         /* Collision, need to strcmp the whole shader. */
112         for (; pass && (pass->hash == hash); pass = pass->next) {
113                 if ((defs != NULL) && (strcmp(pass->defines, defs) != 0)) { /* Pass */ }
114                 else if ((geom != NULL) && (strcmp(pass->geometrycode, geom) != 0)) { /* Pass */ }
115                 else if ((strcmp(pass->fragmentcode, frag) == 0) &&
116                          (strcmp(pass->vertexcode, vert) == 0))
117                 {
118                         BLI_spin_unlock(&pass_cache_spin);
119                         return pass;
120                 }
121         }
122         BLI_spin_unlock(&pass_cache_spin);
123         return NULL;
124 }
125
126 /* -------------------- GPU Codegen ------------------ */
127
128 /* type definitions and constants */
129
130 #define MAX_FUNCTION_NAME    64
131 #define MAX_PARAMETER        32
132
133 typedef enum {
134         FUNCTION_QUAL_IN,
135         FUNCTION_QUAL_OUT,
136         FUNCTION_QUAL_INOUT
137 } GPUFunctionQual;
138
139 typedef struct GPUFunction {
140         char name[MAX_FUNCTION_NAME];
141         eGPUType paramtype[MAX_PARAMETER];
142         GPUFunctionQual paramqual[MAX_PARAMETER];
143         int totparam;
144 } GPUFunction;
145
146 /* Indices match the eGPUType enum */
147 static const char *GPU_DATATYPE_STR[17] = {
148         "", "float", "vec2", "vec3", "vec4",
149         NULL, NULL, NULL, NULL, "mat3", NULL, NULL, NULL, NULL, NULL, NULL, "mat4",
150 };
151
152 /* GLSL code parsing for finding function definitions.
153  * These are stored in a hash for lookup when creating a material. */
154
155 static GHash *FUNCTION_HASH = NULL;
156 #if 0
157 static char *FUNCTION_PROTOTYPES = NULL;
158 static GPUShader *FUNCTION_LIB = NULL;
159 #endif
160
161 static int gpu_str_prefix(const char *str, const char *prefix)
162 {
163         while (*str && *prefix) {
164                 if (*str != *prefix)
165                         return 0;
166
167                 str++;
168                 prefix++;
169         }
170
171         return (*prefix == '\0');
172 }
173
174 static char *gpu_str_skip_token(char *str, char *token, int max)
175 {
176         int len = 0;
177
178         /* skip a variable/function name */
179         while (*str) {
180                 if (ELEM(*str, ' ', '(', ')', ',', ';', '\t', '\n', '\r'))
181                         break;
182                 else {
183                         if (token && len < max - 1) {
184                                 *token = *str;
185                                 token++;
186                                 len++;
187                         }
188                         str++;
189                 }
190         }
191
192         if (token)
193                 *token = '\0';
194
195         /* skip the next special characters:
196          * note the missing ')' */
197         while (*str) {
198                 if (ELEM(*str, ' ', '(', ',', ';', '\t', '\n', '\r'))
199                         str++;
200                 else
201                         break;
202         }
203
204         return str;
205 }
206
207 static void gpu_parse_functions_string(GHash *hash, char *code)
208 {
209         GPUFunction *function;
210         eGPUType type;
211         GPUFunctionQual qual;
212         int i;
213
214         while ((code = strstr(code, "void "))) {
215                 function = MEM_callocN(sizeof(GPUFunction), "GPUFunction");
216
217                 code = gpu_str_skip_token(code, NULL, 0);
218                 code = gpu_str_skip_token(code, function->name, MAX_FUNCTION_NAME);
219
220                 /* get parameters */
221                 while (*code && *code != ')') {
222                         /* test if it's an input or output */
223                         qual = FUNCTION_QUAL_IN;
224                         if (gpu_str_prefix(code, "out "))
225                                 qual = FUNCTION_QUAL_OUT;
226                         if (gpu_str_prefix(code, "inout "))
227                                 qual = FUNCTION_QUAL_INOUT;
228                         if ((qual != FUNCTION_QUAL_IN) || gpu_str_prefix(code, "in "))
229                                 code = gpu_str_skip_token(code, NULL, 0);
230
231                         /* test for type */
232                         type = GPU_NONE;
233                         for (i = 1; i < ARRAY_SIZE(GPU_DATATYPE_STR); i++) {
234                                 if (GPU_DATATYPE_STR[i] && gpu_str_prefix(code, GPU_DATATYPE_STR[i])) {
235                                         type = i;
236                                         break;
237                                 }
238                         }
239
240                         if (!type && gpu_str_prefix(code, "samplerCube")) {
241                                 type = GPU_TEXCUBE;
242                         }
243                         if (!type && gpu_str_prefix(code, "sampler2DShadow")) {
244                                 type = GPU_SHADOW2D;
245                         }
246                         if (!type && gpu_str_prefix(code, "sampler1DArray")) {
247                                 type = GPU_TEX1D_ARRAY;
248                         }
249                         if (!type && gpu_str_prefix(code, "sampler2D")) {
250                                 type = GPU_TEX2D;
251                         }
252                         if (!type && gpu_str_prefix(code, "sampler3D")) {
253                                 type = GPU_TEX3D;
254                         }
255
256                         if (!type && gpu_str_prefix(code, "Closure")) {
257                                 type = GPU_CLOSURE;
258                         }
259
260                         if (type) {
261                                 /* add parameter */
262                                 code = gpu_str_skip_token(code, NULL, 0);
263                                 code = gpu_str_skip_token(code, NULL, 0);
264                                 function->paramqual[function->totparam] = qual;
265                                 function->paramtype[function->totparam] = type;
266                                 function->totparam++;
267                         }
268                         else {
269                                 fprintf(stderr, "GPU invalid function parameter in %s.\n", function->name);
270                                 break;
271                         }
272                 }
273
274                 if (function->name[0] == '\0' || function->totparam == 0) {
275                         fprintf(stderr, "GPU functions parse error.\n");
276                         MEM_freeN(function);
277                         break;
278                 }
279
280                 BLI_ghash_insert(hash, function->name, function);
281         }
282 }
283
284 #if 0
285 static char *gpu_generate_function_prototyps(GHash *hash)
286 {
287         DynStr *ds = BLI_dynstr_new();
288         GHashIterator *ghi;
289         GPUFunction *function;
290         char *name, *prototypes;
291         int a;
292
293         /* automatically generate function prototypes to add to the top of the
294          * generated code, to avoid have to add the actual code & recompile all */
295         ghi = BLI_ghashIterator_new(hash);
296
297         for (; !BLI_ghashIterator_done(ghi); BLI_ghashIterator_step(ghi)) {
298                 name = BLI_ghashIterator_getValue(ghi);
299                 function = BLI_ghashIterator_getValue(ghi);
300
301                 BLI_dynstr_appendf(ds, "void %s(", name);
302                 for (a = 0; a < function->totparam; a++) {
303                         if (function->paramqual[a] == FUNCTION_QUAL_OUT)
304                                 BLI_dynstr_append(ds, "out ");
305                         else if (function->paramqual[a] == FUNCTION_QUAL_INOUT)
306                                 BLI_dynstr_append(ds, "inout ");
307
308                         if (function->paramtype[a] == GPU_TEX2D)
309                                 BLI_dynstr_append(ds, "sampler2D");
310                         else if (function->paramtype[a] == GPU_SHADOW2D)
311                                 BLI_dynstr_append(ds, "sampler2DShadow");
312                         else
313                                 BLI_dynstr_append(ds, GPU_DATATYPE_STR[function->paramtype[a]]);
314 #  if 0
315                         BLI_dynstr_appendf(ds, " param%d", a);
316 #  endif
317
318                         if (a != function->totparam - 1)
319                                 BLI_dynstr_append(ds, ", ");
320                 }
321                 BLI_dynstr_append(ds, ");\n");
322         }
323
324         BLI_dynstr_append(ds, "\n");
325
326         prototypes = BLI_dynstr_get_cstring(ds);
327         BLI_dynstr_free(ds);
328
329         return prototypes;
330 }
331 #endif
332
333 static GPUFunction *gpu_lookup_function(const char *name)
334 {
335         if (!FUNCTION_HASH) {
336                 FUNCTION_HASH = BLI_ghash_str_new("GPU_lookup_function gh");
337                 gpu_parse_functions_string(FUNCTION_HASH, glsl_material_library);
338         }
339
340         return BLI_ghash_lookup(FUNCTION_HASH, (const void *)name);
341 }
342
343 void gpu_codegen_init(void)
344 {
345         GPU_code_generate_glsl_lib();
346 }
347
348 void gpu_codegen_exit(void)
349 {
350         extern Material defmaterial; /* render module abuse... */
351
352         if (defmaterial.gpumaterial.first)
353                 GPU_material_free(&defmaterial.gpumaterial);
354
355         if (FUNCTION_HASH) {
356                 BLI_ghash_free(FUNCTION_HASH, NULL, MEM_freeN);
357                 FUNCTION_HASH = NULL;
358         }
359
360         GPU_shader_free_builtin_shaders();
361
362         if (glsl_material_library) {
363                 MEM_freeN(glsl_material_library);
364                 glsl_material_library = NULL;
365         }
366
367 #if 0
368         if (FUNCTION_PROTOTYPES) {
369                 MEM_freeN(FUNCTION_PROTOTYPES);
370                 FUNCTION_PROTOTYPES = NULL;
371         }
372         if (FUNCTION_LIB) {
373                 GPU_shader_free(FUNCTION_LIB);
374                 FUNCTION_LIB = NULL;
375         }
376 #endif
377 }
378
379 /* GLSL code generation */
380
381 static void codegen_convert_datatype(DynStr *ds, int from, int to, const char *tmp, int id)
382 {
383         char name[1024];
384
385         BLI_snprintf(name, sizeof(name), "%s%d", tmp, id);
386
387         if (from == to) {
388                 BLI_dynstr_append(ds, name);
389         }
390         else if (to == GPU_FLOAT) {
391                 if (from == GPU_VEC4)
392                         BLI_dynstr_appendf(ds, "convert_rgba_to_float(%s)", name);
393                 else if (from == GPU_VEC3)
394                         BLI_dynstr_appendf(ds, "(%s.r + %s.g + %s.b) / 3.0", name, name, name);
395                 else if (from == GPU_VEC2)
396                         BLI_dynstr_appendf(ds, "%s.r", name);
397         }
398         else if (to == GPU_VEC2) {
399                 if (from == GPU_VEC4)
400                         BLI_dynstr_appendf(ds, "vec2((%s.r + %s.g + %s.b) / 3.0, %s.a)", name, name, name, name);
401                 else if (from == GPU_VEC3)
402                         BLI_dynstr_appendf(ds, "vec2((%s.r + %s.g + %s.b) / 3.0, 1.0)", name, name, name);
403                 else if (from == GPU_FLOAT)
404                         BLI_dynstr_appendf(ds, "vec2(%s, 1.0)", name);
405         }
406         else if (to == GPU_VEC3) {
407                 if (from == GPU_VEC4)
408                         BLI_dynstr_appendf(ds, "%s.rgb", name);
409                 else if (from == GPU_VEC2)
410                         BLI_dynstr_appendf(ds, "vec3(%s.r, %s.r, %s.r)", name, name, name);
411                 else if (from == GPU_FLOAT)
412                         BLI_dynstr_appendf(ds, "vec3(%s, %s, %s)", name, name, name);
413         }
414         else if (to == GPU_VEC4) {
415                 if (from == GPU_VEC3)
416                         BLI_dynstr_appendf(ds, "vec4(%s, 1.0)", name);
417                 else if (from == GPU_VEC2)
418                         BLI_dynstr_appendf(ds, "vec4(%s.r, %s.r, %s.r, %s.g)", name, name, name, name);
419                 else if (from == GPU_FLOAT)
420                         BLI_dynstr_appendf(ds, "vec4(%s, %s, %s, 1.0)", name, name, name);
421         }
422         else if (to == GPU_CLOSURE) {
423                 if (from == GPU_VEC4)
424                         BLI_dynstr_appendf(ds, "closure_emission(%s.rgb)", name);
425                 else if (from == GPU_VEC3)
426                         BLI_dynstr_appendf(ds, "closure_emission(%s.rgb)", name);
427                 else if (from == GPU_VEC2)
428                         BLI_dynstr_appendf(ds, "closure_emission(%s.rrr)", name);
429                 else if (from == GPU_FLOAT)
430                         BLI_dynstr_appendf(ds, "closure_emission(vec3(%s, %s, %s))", name, name, name);
431         }
432         else {
433                 BLI_dynstr_append(ds, name);
434         }
435 }
436
437 static void codegen_print_datatype(DynStr *ds, const eGPUType type, float *data)
438 {
439         int i;
440
441         BLI_dynstr_appendf(ds, "%s(", GPU_DATATYPE_STR[type]);
442
443         for (i = 0; i < type; i++) {
444                 BLI_dynstr_appendf(ds, "%.12f", data[i]);
445                 if (i == type - 1)
446                         BLI_dynstr_append(ds, ")");
447                 else
448                         BLI_dynstr_append(ds, ", ");
449         }
450 }
451
452 static int codegen_input_has_texture(GPUInput *input)
453 {
454         if (input->link)
455                 return 0;
456         else
457                 return (input->source == GPU_SOURCE_TEX);
458 }
459
460 const char *GPU_builtin_name(eGPUBuiltin builtin)
461 {
462         if (builtin == GPU_VIEW_MATRIX)
463                 return "unfviewmat";
464         else if (builtin == GPU_OBJECT_MATRIX)
465                 return "unfobmat";
466         else if (builtin == GPU_INVERSE_VIEW_MATRIX)
467                 return "unfinvviewmat";
468         else if (builtin == GPU_INVERSE_OBJECT_MATRIX)
469                 return "unfinvobmat";
470         else if (builtin == GPU_LOC_TO_VIEW_MATRIX)
471                 return "unflocaltoviewmat";
472         else if (builtin == GPU_INVERSE_LOC_TO_VIEW_MATRIX)
473                 return "unfinvlocaltoviewmat";
474         else if (builtin == GPU_VIEW_POSITION)
475                 return "varposition";
476         else if (builtin == GPU_VIEW_NORMAL)
477                 return "varnormal";
478         else if (builtin == GPU_OBCOLOR)
479                 return "unfobcolor";
480         else if (builtin == GPU_AUTO_BUMPSCALE)
481                 return "unfobautobumpscale";
482         else if (builtin == GPU_CAMERA_TEXCO_FACTORS)
483                 return "unfcameratexfactors";
484         else if (builtin == GPU_PARTICLE_SCALAR_PROPS)
485                 return "unfparticlescalarprops";
486         else if (builtin == GPU_PARTICLE_LOCATION)
487                 return "unfparticleco";
488         else if (builtin == GPU_PARTICLE_VELOCITY)
489                 return "unfparticlevel";
490         else if (builtin == GPU_PARTICLE_ANG_VELOCITY)
491                 return "unfparticleangvel";
492         else if (builtin == GPU_OBJECT_INFO)
493                 return "unfobjectinfo";
494         else if (builtin == GPU_VOLUME_DENSITY)
495                 return "sampdensity";
496         else if (builtin == GPU_VOLUME_FLAME)
497                 return "sampflame";
498         else if (builtin == GPU_VOLUME_TEMPERATURE)
499                 return "unftemperature";
500         else if (builtin == GPU_BARYCENTRIC_TEXCO)
501                 return "unfbarycentrictex";
502         else if (builtin == GPU_BARYCENTRIC_DIST)
503                 return "unfbarycentricdist";
504         else
505                 return "";
506 }
507
508 /* assign only one texid per buffer to avoid sampling the same texture twice */
509 static void codegen_set_texid(GHash *bindhash, GPUInput *input, int *texid, void *key)
510 {
511         if (BLI_ghash_haskey(bindhash, key)) {
512                 /* Reuse existing texid */
513                 input->texid = POINTER_AS_INT(BLI_ghash_lookup(bindhash, key));
514         }
515         else {
516                 /* Allocate new texid */
517                 input->texid = *texid;
518                 (*texid)++;
519                 input->bindtex = true;
520                 BLI_ghash_insert(bindhash, key, POINTER_FROM_INT(input->texid));
521         }
522 }
523
524 static void codegen_set_unique_ids(ListBase *nodes)
525 {
526         GHash *bindhash;
527         GPUNode *node;
528         GPUInput *input;
529         GPUOutput *output;
530         int id = 1, texid = 0;
531
532         bindhash = BLI_ghash_ptr_new("codegen_set_unique_ids1 gh");
533
534         for (node = nodes->first; node; node = node->next) {
535                 for (input = node->inputs.first; input; input = input->next) {
536                         /* set id for unique names of uniform variables */
537                         input->id = id++;
538
539                         /* set texid used for settings texture slot */
540                         if (codegen_input_has_texture(input)) {
541                                 input->bindtex = false;
542                                 if (input->ima) {
543                                         /* input is texture from image */
544                                         codegen_set_texid(bindhash, input, &texid, input->ima);
545                                 }
546                                 else if (input->coba) {
547                                         /* input is color band texture, check coba pointer */
548                                         codegen_set_texid(bindhash, input, &texid, input->coba);
549                                 }
550                                 else {
551                                         /* Either input->ima or input->coba should be non-NULL. */
552                                         BLI_assert(0);
553                                 }
554                         }
555                 }
556
557                 for (output = node->outputs.first; output; output = output->next) {
558                         /* set id for unique names of tmp variables storing output */
559                         output->id = id++;
560                 }
561         }
562
563         BLI_ghash_free(bindhash, NULL, NULL);
564 }
565
566 /**
567  * It will create an UBO for GPUMaterial if there is any GPU_DYNAMIC_UBO.
568  */
569 static int codegen_process_uniforms_functions(GPUMaterial *material, DynStr *ds, ListBase *nodes)
570 {
571         GPUNode *node;
572         GPUInput *input;
573         const char *name;
574         int builtins = 0;
575         ListBase ubo_inputs = {NULL, NULL};
576
577         /* print uniforms */
578         for (node = nodes->first; node; node = node->next) {
579                 for (input = node->inputs.first; input; input = input->next) {
580                         if (input->source == GPU_SOURCE_TEX) {
581                                 /* create exactly one sampler for each texture */
582                                 if (codegen_input_has_texture(input) && input->bindtex) {
583                                         BLI_dynstr_appendf(
584                                                 ds, "uniform %s samp%d;\n",
585                                                 (input->coba) ? "sampler1DArray" : "sampler2D",
586                                                 input->texid);
587                                 }
588                         }
589                         else if (input->source == GPU_SOURCE_BUILTIN) {
590                                 /* only define each builtin uniform/varying once */
591                                 if (!(builtins & input->builtin)) {
592                                         builtins |= input->builtin;
593                                         name = GPU_builtin_name(input->builtin);
594
595                                         if (gpu_str_prefix(name, "samp")) {
596                                                 if ((input->builtin == GPU_VOLUME_DENSITY) ||
597                                                     (input->builtin == GPU_VOLUME_FLAME))
598                                                 {
599                                                         BLI_dynstr_appendf(ds, "uniform sampler3D %s;\n", name);
600                                                 }
601                                         }
602                                         else if (gpu_str_prefix(name, "unf")) {
603                                                 BLI_dynstr_appendf(
604                                                         ds, "uniform %s %s;\n",
605                                                         GPU_DATATYPE_STR[input->type], name);
606                                         }
607                                         else {
608                                                 BLI_dynstr_appendf(
609                                                         ds, "in %s %s;\n",
610                                                         GPU_DATATYPE_STR[input->type], name);
611                                         }
612                                 }
613                         }
614                         else if (input->source == GPU_SOURCE_STRUCT) {
615                                 /* Add other struct here if needed. */
616                                 BLI_dynstr_appendf(ds, "Closure strct%d = CLOSURE_DEFAULT;\n", input->id);
617                         }
618                         else if (input->source == GPU_SOURCE_UNIFORM) {
619                                 if (!input->link) {
620                                         /* We handle the UBOuniforms separately. */
621                                         BLI_addtail(&ubo_inputs, BLI_genericNodeN(input));
622                                 }
623                         }
624                         else if (input->source == GPU_SOURCE_CONSTANT) {
625                                 BLI_dynstr_appendf(
626                                         ds, "const %s cons%d = ",
627                                         GPU_DATATYPE_STR[input->type], input->id);
628                                 codegen_print_datatype(ds, input->type, input->vec);
629                                 BLI_dynstr_append(ds, ";\n");
630                         }
631                         else if (input->source == GPU_SOURCE_ATTR && input->attr_first) {
632                                 BLI_dynstr_appendf(
633                                         ds, "in %s var%d;\n",
634                                         GPU_DATATYPE_STR[input->type], input->attr_id);
635                         }
636                 }
637         }
638
639         /* Handle the UBO block separately. */
640         if ((material != NULL) && !BLI_listbase_is_empty(&ubo_inputs)) {
641                 GPU_material_uniform_buffer_create(material, &ubo_inputs);
642
643                 /* Inputs are sorted */
644                 BLI_dynstr_appendf(ds, "\nlayout (std140) uniform %s {\n", GPU_UBO_BLOCK_NAME);
645
646                 for (LinkData *link = ubo_inputs.first; link; link = link->next) {
647                         input = link->data;
648                         BLI_dynstr_appendf(
649                                 ds, "\t%s unf%d;\n",
650                                 GPU_DATATYPE_STR[input->type], input->id);
651                 }
652                 BLI_dynstr_append(ds, "};\n");
653                 BLI_freelistN(&ubo_inputs);
654         }
655
656         BLI_dynstr_append(ds, "\n");
657
658         return builtins;
659 }
660
661 static void codegen_declare_tmps(DynStr *ds, ListBase *nodes)
662 {
663         GPUNode *node;
664         GPUOutput *output;
665
666         for (node = nodes->first; node; node = node->next) {
667                 /* declare temporary variables for node output storage */
668                 for (output = node->outputs.first; output; output = output->next) {
669                         if (output->type == GPU_CLOSURE) {
670                                 BLI_dynstr_appendf(
671                                         ds, "\tClosure tmp%d;\n", output->id);
672                         }
673                         else {
674                                 BLI_dynstr_appendf(
675                                         ds, "\t%s tmp%d;\n",
676                                         GPU_DATATYPE_STR[output->type], output->id);
677                         }
678                 }
679         }
680
681         BLI_dynstr_append(ds, "\n");
682 }
683
684 static void codegen_call_functions(DynStr *ds, ListBase *nodes, GPUOutput *finaloutput)
685 {
686         GPUNode *node;
687         GPUInput *input;
688         GPUOutput *output;
689
690         for (node = nodes->first; node; node = node->next) {
691                 BLI_dynstr_appendf(ds, "\t%s(", node->name);
692
693                 for (input = node->inputs.first; input; input = input->next) {
694                         if (input->source == GPU_SOURCE_TEX) {
695                                 BLI_dynstr_appendf(ds, "samp%d", input->texid);
696                         }
697                         else if (input->source == GPU_SOURCE_OUTPUT) {
698                                 codegen_convert_datatype(
699                                         ds, input->link->output->type, input->type,
700                                         "tmp", input->link->output->id);
701                         }
702                         else if (input->source == GPU_SOURCE_BUILTIN) {
703                                 /* TODO(fclem) get rid of that. */
704                                 if (input->builtin == GPU_INVERSE_VIEW_MATRIX)
705                                         BLI_dynstr_append(ds, "viewinv");
706                                 else if (input->builtin == GPU_VIEW_MATRIX)
707                                         BLI_dynstr_append(ds, "viewmat");
708                                 else if (input->builtin == GPU_CAMERA_TEXCO_FACTORS)
709                                         BLI_dynstr_append(ds, "camtexfac");
710                                 else if (input->builtin == GPU_LOC_TO_VIEW_MATRIX)
711                                         BLI_dynstr_append(ds, "localtoviewmat");
712                                 else if (input->builtin == GPU_INVERSE_LOC_TO_VIEW_MATRIX)
713                                         BLI_dynstr_append(ds, "invlocaltoviewmat");
714                                 else if (input->builtin == GPU_BARYCENTRIC_DIST)
715                                         BLI_dynstr_append(ds, "barycentricDist");
716                                 else if (input->builtin == GPU_BARYCENTRIC_TEXCO)
717                                         BLI_dynstr_append(ds, "barytexco");
718                                 else if (input->builtin == GPU_OBJECT_MATRIX)
719                                         BLI_dynstr_append(ds, "objmat");
720                                 else if (input->builtin == GPU_INVERSE_OBJECT_MATRIX)
721                                         BLI_dynstr_append(ds, "objinv");
722                                 else if (input->builtin == GPU_VIEW_POSITION)
723                                         BLI_dynstr_append(ds, "viewposition");
724                                 else if (input->builtin == GPU_VIEW_NORMAL)
725                                         BLI_dynstr_append(ds, "facingnormal");
726                                 else
727                                         BLI_dynstr_append(ds, GPU_builtin_name(input->builtin));
728                         }
729                         else if (input->source == GPU_SOURCE_STRUCT) {
730                                 BLI_dynstr_appendf(ds, "strct%d", input->id);
731                         }
732                         else if (input->source == GPU_SOURCE_UNIFORM) {
733                                 BLI_dynstr_appendf(ds, "unf%d", input->id);
734                         }
735                         else if (input->source == GPU_SOURCE_CONSTANT) {
736                                 BLI_dynstr_appendf(ds, "cons%d", input->id);
737                         }
738                         else if (input->source == GPU_SOURCE_ATTR) {
739                                 BLI_dynstr_appendf(ds, "var%d", input->attr_id);
740                         }
741
742                         BLI_dynstr_append(ds, ", ");
743                 }
744
745                 for (output = node->outputs.first; output; output = output->next) {
746                         BLI_dynstr_appendf(ds, "tmp%d", output->id);
747                         if (output->next)
748                                 BLI_dynstr_append(ds, ", ");
749                 }
750
751                 BLI_dynstr_append(ds, ");\n");
752         }
753
754         BLI_dynstr_appendf(ds, "\n\treturn tmp%d", finaloutput->id);
755         BLI_dynstr_append(ds, ";\n");
756 }
757
758 static char *code_generate_fragment(GPUMaterial *material, ListBase *nodes, GPUOutput *output, int *rbuiltins)
759 {
760         DynStr *ds = BLI_dynstr_new();
761         char *code;
762         int builtins;
763
764 #if 0
765         BLI_dynstr_append(ds, FUNCTION_PROTOTYPES);
766 #endif
767
768         codegen_set_unique_ids(nodes);
769         *rbuiltins = builtins = codegen_process_uniforms_functions(material, ds, nodes);
770
771         if (builtins & GPU_BARYCENTRIC_TEXCO)
772                 BLI_dynstr_append(ds, "in vec2 barycentricTexCo;\n");
773
774         if (builtins & GPU_BARYCENTRIC_DIST)
775                 BLI_dynstr_append(ds, "flat in vec3 barycentricDist;\n");
776
777         BLI_dynstr_append(ds, "Closure nodetree_exec(void)\n{\n");
778
779         if (builtins & GPU_BARYCENTRIC_TEXCO) {
780                 BLI_dynstr_append(ds, "#ifdef HAIR_SHADER\n");
781                 BLI_dynstr_append(ds, "\tvec2 barytexco = vec2((fract(barycentricTexCo.y) != 0.0)\n"
782                                       "\t                      ? barycentricTexCo.x\n"
783                                       "\t                      : 1.0 - barycentricTexCo.x,\n"
784                                       "\t                      0.0);\n");
785                 BLI_dynstr_append(ds, "#else\n");
786                 BLI_dynstr_append(ds, "\tvec2 barytexco = barycentricTexCo;\n");
787                 BLI_dynstr_append(ds, "#endif\n");
788         }
789         /* TODO(fclem) get rid of that. */
790         if (builtins & GPU_VIEW_MATRIX)
791                 BLI_dynstr_append(ds, "\t#define viewmat ViewMatrix\n");
792         if (builtins & GPU_CAMERA_TEXCO_FACTORS)
793                 BLI_dynstr_append(ds, "\t#define camtexfac CameraTexCoFactors\n");
794         if (builtins & GPU_OBJECT_MATRIX)
795                 BLI_dynstr_append(ds, "\t#define objmat ModelMatrix\n");
796         if (builtins & GPU_INVERSE_OBJECT_MATRIX)
797                 BLI_dynstr_append(ds, "\t#define objinv ModelMatrixInverse\n");
798         if (builtins & GPU_INVERSE_VIEW_MATRIX)
799                 BLI_dynstr_append(ds, "\t#define viewinv ViewMatrixInverse\n");
800         if (builtins & GPU_LOC_TO_VIEW_MATRIX)
801                 BLI_dynstr_append(ds, "\t#define localtoviewmat ModelViewMatrix\n");
802         if (builtins & GPU_INVERSE_LOC_TO_VIEW_MATRIX)
803                 BLI_dynstr_append(ds, "\t#define invlocaltoviewmat ModelViewMatrixInverse\n");
804         if (builtins & GPU_VIEW_NORMAL)
805                 BLI_dynstr_append(ds, "\tvec3 facingnormal = gl_FrontFacing? viewNormal: -viewNormal;\n");
806         if (builtins & GPU_VIEW_POSITION)
807                 BLI_dynstr_append(ds, "\t#define viewposition viewPosition\n");
808
809         codegen_declare_tmps(ds, nodes);
810         codegen_call_functions(ds, nodes, output);
811
812         BLI_dynstr_append(ds, "}\n");
813
814         /* XXX This cannot go into gpu_shader_material.glsl because main() would be parsed and generate error */
815         /* Old glsl mode compat. */
816         BLI_dynstr_append(ds, "#ifndef NODETREE_EXEC\n");
817         BLI_dynstr_append(ds, "out vec4 fragColor;\n");
818         BLI_dynstr_append(ds, "void main()\n");
819         BLI_dynstr_append(ds, "{\n");
820         BLI_dynstr_append(ds, "\tClosure cl = nodetree_exec();\n");
821         BLI_dynstr_append(ds, "\tfragColor = vec4(cl.radiance, cl.opacity);\n");
822         BLI_dynstr_append(ds, "}\n");
823         BLI_dynstr_append(ds, "#endif\n\n");
824
825         /* create shader */
826         code = BLI_dynstr_get_cstring(ds);
827         BLI_dynstr_free(ds);
828
829 #if 0
830         if (G.debug & G_DEBUG) printf("%s\n", code);
831 #endif
832
833         return code;
834 }
835
836 static const char *attr_prefix_get(CustomDataType type)
837 {
838         switch (type) {
839                 case CD_ORCO:           return "orco";
840                 case CD_MTFACE:         return "u";
841                 case CD_TANGENT:        return "t";
842                 case CD_MCOL:           return "c";
843                 case CD_AUTO_FROM_NAME: return "a";
844                 default: BLI_assert(false && "GPUVertAttr Prefix type not found : This should not happen!"); return "";
845         }
846 }
847
848 static char *code_generate_vertex(ListBase *nodes, const char *vert_code, bool use_geom)
849 {
850         DynStr *ds = BLI_dynstr_new();
851         GPUNode *node;
852         GPUInput *input;
853         char *code;
854         int builtins = 0;
855
856         /* Hairs uv and col attributes are passed by bufferTextures. */
857         BLI_dynstr_append(
858                 ds,
859                 "#ifdef HAIR_SHADER\n"
860                 "#define DEFINE_ATTR(type, attr) uniform samplerBuffer attr\n"
861                 "#else\n"
862                 "#define DEFINE_ATTR(type, attr) in type attr\n"
863                 "#endif\n"
864         );
865
866         for (node = nodes->first; node; node = node->next) {
867                 for (input = node->inputs.first; input; input = input->next) {
868                         if (input->source == GPU_SOURCE_BUILTIN) {
869                                 builtins |= input->builtin;
870                         }
871                         if (input->source == GPU_SOURCE_ATTR && input->attr_first) {
872                                 /* XXX FIXME : see notes in mesh_render_data_create() */
873                                 /* NOTE : Replicate changes to mesh_render_data_create() in draw_cache_impl_mesh.c */
874                                 if (input->attr_type == CD_ORCO) {
875                                         /* orco is computed from local positions, see below */
876                                         BLI_dynstr_appendf(ds, "uniform vec3 OrcoTexCoFactors[2];\n");
877                                 }
878                                 else if (input->attr_name[0] == '\0') {
879                                         BLI_dynstr_appendf(ds, "DEFINE_ATTR(%s, %s);\n", GPU_DATATYPE_STR[input->type], attr_prefix_get(input->attr_type));
880                                         BLI_dynstr_appendf(ds, "#define att%d %s\n", input->attr_id, attr_prefix_get(input->attr_type));
881                                 }
882                                 else {
883                                         uint hash = BLI_ghashutil_strhash_p(input->attr_name);
884                                         BLI_dynstr_appendf(
885                                                 ds, "DEFINE_ATTR(%s, %s%u);\n",
886                                                 GPU_DATATYPE_STR[input->type], attr_prefix_get(input->attr_type), hash);
887                                         BLI_dynstr_appendf(
888                                                 ds, "#define att%d %s%u\n",
889                                                 input->attr_id, attr_prefix_get(input->attr_type), hash);
890                                         /* Auto attribute can be vertex color byte buffer.
891                                          * We need to know and convert them to linear space in VS. */
892                                         if (input->attr_type == CD_AUTO_FROM_NAME) {
893                                                 BLI_dynstr_appendf(ds, "uniform bool ba%u;\n", hash);
894                                                 BLI_dynstr_appendf(ds, "#define att%d_is_srgb ba%u\n", input->attr_id, hash);
895                                         }
896                                 }
897                                 BLI_dynstr_appendf(
898                                         ds, "out %s var%d%s;\n",
899                                         GPU_DATATYPE_STR[input->type], input->attr_id, use_geom ? "g" : "");
900                         }
901                 }
902         }
903
904         if (builtins & GPU_BARYCENTRIC_TEXCO) {
905                 BLI_dynstr_append(ds, "#ifdef HAIR_SHADER\n");
906                 BLI_dynstr_appendf(
907                         ds, "out vec2 barycentricTexCo%s;\n",
908                         use_geom ? "g" : "");
909                 BLI_dynstr_append(ds, "#endif\n");
910         }
911
912         if (builtins & GPU_BARYCENTRIC_DIST) {
913                 BLI_dynstr_appendf(ds, "out vec3 barycentricPosg;\n");
914         }
915
916
917         BLI_dynstr_append(ds, "\n");
918
919         BLI_dynstr_append(
920                 ds,
921                 "#define USE_ATTR\n"
922                 "uniform mat3 NormalMatrix;\n"
923                 "uniform mat4 ModelMatrixInverse;\n"
924                 "uniform mat4 ModelMatrix;\n"
925                 "vec3 srgb_to_linear_attr(vec3 c) {\n"
926                 "\tc = max(c, vec3(0.0));\n"
927                 "\tvec3 c1 = c * (1.0 / 12.92);\n"
928                 "\tvec3 c2 = pow((c + 0.055) * (1.0 / 1.055), vec3(2.4));\n"
929                 "\treturn mix(c1, c2, step(vec3(0.04045), c));\n"
930                 "}\n\n"
931         );
932
933         /* Prototype because defined later. */
934         BLI_dynstr_append(
935                 ds,
936                 "vec2 hair_get_customdata_vec2(const samplerBuffer);\n"
937                 "vec3 hair_get_customdata_vec3(const samplerBuffer);\n"
938                 "vec4 hair_get_customdata_vec4(const samplerBuffer);\n"
939                 "vec3 hair_get_strand_pos(void);\n"
940                 "int hair_get_base_id(void);\n"
941                 "\n"
942         );
943
944         BLI_dynstr_append(ds, "void pass_attr(in vec3 position) {\n");
945
946         BLI_dynstr_append(ds, "#ifdef HAIR_SHADER\n");
947
948         if (builtins & GPU_BARYCENTRIC_TEXCO) {
949                 /* To match cycles without breaking into individual segment we encode if we need to invert
950                  * the first component into the second component. We invert if the barycentricTexCo.y
951                  * is NOT 0.0 or 1.0. */
952                 BLI_dynstr_appendf(
953                         ds, "\tint _base_id = hair_get_base_id();\n");
954                 BLI_dynstr_appendf(
955                         ds, "\tbarycentricTexCo%s.x = float((_base_id %% 2) == 1);\n",
956                         use_geom ? "g" : "");
957                 BLI_dynstr_appendf(
958                         ds, "\tbarycentricTexCo%s.y = float(((_base_id %% 4) %% 3) > 0);\n",
959                         use_geom ? "g" : "");
960         }
961
962         if (builtins & GPU_BARYCENTRIC_DIST) {
963                 BLI_dynstr_appendf(ds, "\tbarycentricPosg = position;\n");
964         }
965
966         for (node = nodes->first; node; node = node->next) {
967                 for (input = node->inputs.first; input; input = input->next) {
968                         if (input->source == GPU_SOURCE_ATTR && input->attr_first) {
969                                 if (input->attr_type == CD_TANGENT) {
970                                         /* Not supported by hairs */
971                                         BLI_dynstr_appendf(
972                                                 ds, "\tvar%d%s = vec4(0.0);\n",
973                                                 input->attr_id, use_geom ? "g" : "");
974                                 }
975                                 else if (input->attr_type == CD_ORCO) {
976                                         BLI_dynstr_appendf(
977                                                 ds, "\tvar%d%s = OrcoTexCoFactors[0] + (ModelMatrixInverse * vec4(hair_get_strand_pos(), 1.0)).xyz * OrcoTexCoFactors[1];\n",
978                                                 input->attr_id, use_geom ? "g" : "");
979                                 }
980                                 else {
981                                         BLI_dynstr_appendf(
982                                                 ds, "\tvar%d%s = hair_get_customdata_%s(att%d);\n",
983                                                 input->attr_id, use_geom ? "g" : "", GPU_DATATYPE_STR[input->type], input->attr_id);
984                                 }
985                         }
986                 }
987         }
988
989         BLI_dynstr_append(ds, "#else /* MESH_SHADER */\n");
990
991         /* GPU_BARYCENTRIC_TEXCO cannot be computed based on gl_VertexID
992          * for MESH_SHADER because of indexed drawing. In this case a
993          * geometry shader is needed. */
994
995         if (builtins & GPU_BARYCENTRIC_DIST) {
996                 BLI_dynstr_appendf(ds, "\tbarycentricPosg = (ModelMatrix * vec4(position, 1.0)).xyz;\n");
997         }
998
999         for (node = nodes->first; node; node = node->next) {
1000                 for (input = node->inputs.first; input; input = input->next) {
1001                         if (input->source == GPU_SOURCE_ATTR && input->attr_first) {
1002                                 if (input->attr_type == CD_TANGENT) { /* silly exception */
1003                                         BLI_dynstr_appendf(
1004                                                 ds, "\tvar%d%s.xyz = NormalMatrix * att%d.xyz;\n",
1005                                                 input->attr_id, use_geom ? "g" : "", input->attr_id);
1006                                         BLI_dynstr_appendf(
1007                                                 ds, "\tvar%d%s.w = att%d.w;\n",
1008                                                 input->attr_id, use_geom ? "g" : "", input->attr_id);
1009                                 }
1010                                 else if (input->attr_type == CD_ORCO) {
1011                                         BLI_dynstr_appendf(
1012                                                 ds, "\tvar%d%s = OrcoTexCoFactors[0] + position * OrcoTexCoFactors[1];\n",
1013                                                 input->attr_id, use_geom ? "g" : "");
1014                                 }
1015                                 else if (input->attr_type == CD_MCOL) {
1016                                         BLI_dynstr_appendf(
1017                                                 ds, "\tvar%d%s = srgb_to_linear_attr(att%d);\n",
1018                                                 input->attr_id, use_geom ? "g" : "", input->attr_id);
1019                                 }
1020                                 else if (input->attr_type == CD_AUTO_FROM_NAME) {
1021                                         BLI_dynstr_appendf(
1022                                                 ds, "\tvar%d%s = (att%d_is_srgb) ? srgb_to_linear_attr(att%d) : att%d;\n",
1023                                                 input->attr_id, use_geom ? "g" : "",
1024                                                 input->attr_id, input->attr_id, input->attr_id);
1025                                 }
1026                                 else {
1027                                         BLI_dynstr_appendf(
1028                                                 ds, "\tvar%d%s = att%d;\n",
1029                                                 input->attr_id, use_geom ? "g" : "", input->attr_id);
1030                                 }
1031                         }
1032                 }
1033         }
1034         BLI_dynstr_append(ds, "#endif /* HAIR_SHADER */\n");
1035
1036         BLI_dynstr_append(ds, "}\n");
1037
1038         if (use_geom) {
1039                 /* XXX HACK: Eevee specific. */
1040                 char *vert_new, *vert_new2;
1041                 vert_new = BLI_str_replaceN(vert_code, "worldPosition", "worldPositiong");
1042                 vert_new2 = vert_new;
1043                 vert_new = BLI_str_replaceN(vert_new2, "viewPosition", "viewPositiong");
1044                 MEM_freeN(vert_new2);
1045                 vert_new2 = vert_new;
1046                 vert_new = BLI_str_replaceN(vert_new2, "worldNormal", "worldNormalg");
1047                 MEM_freeN(vert_new2);
1048                 vert_new2 = vert_new;
1049                 vert_new = BLI_str_replaceN(vert_new2, "viewNormal", "viewNormalg");
1050                 MEM_freeN(vert_new2);
1051
1052                 BLI_dynstr_append(ds, vert_new);
1053
1054                 MEM_freeN(vert_new);
1055         }
1056         else {
1057                 BLI_dynstr_append(ds, vert_code);
1058         }
1059
1060         code = BLI_dynstr_get_cstring(ds);
1061
1062         BLI_dynstr_free(ds);
1063
1064 #if 0
1065         if (G.debug & G_DEBUG) printf("%s\n", code);
1066 #endif
1067
1068         return code;
1069 }
1070
1071 static char *code_generate_geometry(ListBase *nodes, const char *geom_code, const char *defines)
1072 {
1073         DynStr *ds = BLI_dynstr_new();
1074         GPUNode *node;
1075         GPUInput *input;
1076         char *code;
1077         int builtins = 0;
1078
1079         /* XXX we should not make specific eevee cases here. */
1080         bool is_hair_shader = (strstr(defines, "HAIR_SHADER") != NULL);
1081
1082         /* Create prototype because attributes cannot be declared before layout. */
1083         BLI_dynstr_appendf(ds, "void pass_attr(in int vert);\n");
1084         BLI_dynstr_appendf(ds, "void calc_barycentric_distances(vec3 pos0, vec3 pos1, vec3 pos2);\n");
1085         BLI_dynstr_append(ds, "#define USE_ATTR\n");
1086
1087         /* Generate varying declarations. */
1088         for (node = nodes->first; node; node = node->next) {
1089                 for (input = node->inputs.first; input; input = input->next) {
1090                         if (input->source == GPU_SOURCE_BUILTIN) {
1091                                 builtins |= input->builtin;
1092                         }
1093                         if (input->source == GPU_SOURCE_ATTR && input->attr_first) {
1094                                 BLI_dynstr_appendf(
1095                                         ds, "in %s var%dg[];\n",
1096                                         GPU_DATATYPE_STR[input->type],
1097                                         input->attr_id);
1098                                 BLI_dynstr_appendf(
1099                                         ds, "out %s var%d;\n",
1100                                         GPU_DATATYPE_STR[input->type],
1101                                         input->attr_id);
1102                         }
1103                 }
1104         }
1105
1106         if (builtins & GPU_BARYCENTRIC_TEXCO) {
1107                 BLI_dynstr_appendf(ds, "#ifdef HAIR_SHADER\n");
1108                 BLI_dynstr_appendf(ds, "in vec2 barycentricTexCog[];\n");
1109                 BLI_dynstr_appendf(ds, "#endif\n");
1110
1111                 BLI_dynstr_appendf(ds, "out vec2 barycentricTexCo;\n");
1112         }
1113
1114         if (builtins & GPU_BARYCENTRIC_DIST) {
1115                 BLI_dynstr_appendf(ds, "in vec3 barycentricPosg[];\n");
1116                 BLI_dynstr_appendf(ds, "flat out vec3 barycentricDist;\n");
1117         }
1118
1119         if (geom_code == NULL) {
1120                 /* Force geometry usage if GPU_BARYCENTRIC_DIST or GPU_BARYCENTRIC_TEXCO are used.
1121                  * Note: GPU_BARYCENTRIC_TEXCO only requires it if the shader is not drawing hairs. */
1122                 if ((builtins & (GPU_BARYCENTRIC_DIST | GPU_BARYCENTRIC_TEXCO)) == 0 || is_hair_shader) {
1123                         /* Early out */
1124                         BLI_dynstr_free(ds);
1125                         return NULL;
1126                 }
1127                 else {
1128                         /* Force geom shader usage */
1129                         /* TODO put in external file. */
1130                         BLI_dynstr_appendf(ds, "layout(triangles) in;\n");
1131                         BLI_dynstr_appendf(ds, "layout(triangle_strip, max_vertices=3) out;\n");
1132
1133                         BLI_dynstr_appendf(ds, "in vec3 worldPositiong[];\n");
1134                         BLI_dynstr_appendf(ds, "in vec3 viewPositiong[];\n");
1135                         BLI_dynstr_appendf(ds, "in vec3 worldNormalg[];\n");
1136                         BLI_dynstr_appendf(ds, "in vec3 viewNormalg[];\n");
1137
1138                         BLI_dynstr_appendf(ds, "out vec3 worldPosition;\n");
1139                         BLI_dynstr_appendf(ds, "out vec3 viewPosition;\n");
1140                         BLI_dynstr_appendf(ds, "out vec3 worldNormal;\n");
1141                         BLI_dynstr_appendf(ds, "out vec3 viewNormal;\n");
1142
1143                         BLI_dynstr_appendf(ds, "void main(){\n");
1144
1145                         if (builtins & GPU_BARYCENTRIC_DIST) {
1146                                 BLI_dynstr_appendf(ds, "\tcalc_barycentric_distances(barycentricPosg[0], barycentricPosg[1], barycentricPosg[2]);\n");
1147                         }
1148
1149                         BLI_dynstr_appendf(ds, "\tgl_Position = gl_in[0].gl_Position;\n");
1150                         BLI_dynstr_appendf(ds, "\tpass_attr(0);\n");
1151                         BLI_dynstr_appendf(ds, "\tEmitVertex();\n");
1152
1153                         BLI_dynstr_appendf(ds, "\tgl_Position = gl_in[1].gl_Position;\n");
1154                         BLI_dynstr_appendf(ds, "\tpass_attr(1);\n");
1155                         BLI_dynstr_appendf(ds, "\tEmitVertex();\n");
1156
1157                         BLI_dynstr_appendf(ds, "\tgl_Position = gl_in[2].gl_Position;\n");
1158                         BLI_dynstr_appendf(ds, "\tpass_attr(2);\n");
1159                         BLI_dynstr_appendf(ds, "\tEmitVertex();\n");
1160                         BLI_dynstr_appendf(ds, "};\n");
1161                 }
1162         }
1163         else {
1164                 BLI_dynstr_append(ds, geom_code);
1165         }
1166
1167         if (builtins & GPU_BARYCENTRIC_DIST) {
1168                 BLI_dynstr_appendf(ds, "void calc_barycentric_distances(vec3 pos0, vec3 pos1, vec3 pos2) {\n");
1169                 BLI_dynstr_appendf(ds, "\tvec3 edge21 = pos2 - pos1;\n");
1170                 BLI_dynstr_appendf(ds, "\tvec3 edge10 = pos1 - pos0;\n");
1171                 BLI_dynstr_appendf(ds, "\tvec3 edge02 = pos0 - pos2;\n");
1172                 BLI_dynstr_appendf(ds, "\tvec3 d21 = normalize(edge21);\n");
1173                 BLI_dynstr_appendf(ds, "\tvec3 d10 = normalize(edge10);\n");
1174                 BLI_dynstr_appendf(ds, "\tvec3 d02 = normalize(edge02);\n");
1175
1176                 BLI_dynstr_appendf(ds, "\tfloat d = dot(d21, edge02);\n");
1177                 BLI_dynstr_appendf(ds, "\tbarycentricDist.x = sqrt(dot(edge02, edge02) - d * d);\n");
1178                 BLI_dynstr_appendf(ds, "\td = dot(d02, edge10);\n");
1179                 BLI_dynstr_appendf(ds, "\tbarycentricDist.y = sqrt(dot(edge10, edge10) - d * d);\n");
1180                 BLI_dynstr_appendf(ds, "\td = dot(d10, edge21);\n");
1181                 BLI_dynstr_appendf(ds, "\tbarycentricDist.z = sqrt(dot(edge21, edge21) - d * d);\n");
1182                 BLI_dynstr_append(ds, "}\n");
1183         }
1184
1185         /* Generate varying assignments. */
1186         BLI_dynstr_appendf(ds, "void pass_attr(in int vert) {\n");
1187
1188         /* XXX HACK: Eevee specific. */
1189         if (geom_code == NULL) {
1190                 BLI_dynstr_appendf(ds, "\tworldPosition = worldPositiong[vert];\n");
1191                 BLI_dynstr_appendf(ds, "\tviewPosition = viewPositiong[vert];\n");
1192                 BLI_dynstr_appendf(ds, "\tworldNormal = worldNormalg[vert];\n");
1193                 BLI_dynstr_appendf(ds, "\tviewNormal = viewNormalg[vert];\n");
1194         }
1195
1196         if (builtins & GPU_BARYCENTRIC_TEXCO) {
1197                 BLI_dynstr_appendf(ds, "#ifdef HAIR_SHADER\n");
1198                 BLI_dynstr_appendf(ds, "\tbarycentricTexCo = barycentricTexCog[vert];\n");
1199                 BLI_dynstr_appendf(ds, "#else\n");
1200                 BLI_dynstr_appendf(ds, "\tbarycentricTexCo.x = float((vert %% 3) == 0);\n");
1201                 BLI_dynstr_appendf(ds, "\tbarycentricTexCo.y = float((vert %% 3) == 1);\n");
1202                 BLI_dynstr_appendf(ds, "#endif\n");
1203         }
1204
1205         for (node = nodes->first; node; node = node->next) {
1206                 for (input = node->inputs.first; input; input = input->next) {
1207                         if (input->source == GPU_SOURCE_ATTR && input->attr_first) {
1208                                 /* TODO let shader choose what to do depending on what the attribute is. */
1209                                 BLI_dynstr_appendf(ds, "\tvar%d = var%dg[vert];\n", input->attr_id, input->attr_id);
1210                         }
1211                 }
1212         }
1213         BLI_dynstr_append(ds, "}\n");
1214
1215         code = BLI_dynstr_get_cstring(ds);
1216         BLI_dynstr_free(ds);
1217
1218         return code;
1219 }
1220
1221 void GPU_code_generate_glsl_lib(void)
1222 {
1223         DynStr *ds;
1224
1225         /* only initialize the library once */
1226         if (glsl_material_library)
1227                 return;
1228
1229         ds = BLI_dynstr_new();
1230
1231         BLI_dynstr_append(ds, datatoc_gpu_shader_material_glsl);
1232
1233
1234         glsl_material_library = BLI_dynstr_get_cstring(ds);
1235
1236         BLI_dynstr_free(ds);
1237 }
1238
1239
1240 /* GPU pass binding/unbinding */
1241
1242 GPUShader *GPU_pass_shader_get(GPUPass *pass)
1243 {
1244         return pass->shader;
1245 }
1246
1247 void GPU_nodes_extract_dynamic_inputs(GPUShader *shader, ListBase *inputs, ListBase *nodes)
1248 {
1249         GPUNode *node;
1250         GPUInput *next, *input;
1251
1252         BLI_listbase_clear(inputs);
1253
1254         if (!shader)
1255                 return;
1256
1257         for (node = nodes->first; node; node = node->next) {
1258                 int z = 0;
1259                 for (input = node->inputs.first; input; input = next, z++) {
1260                         next = input->next;
1261
1262                         /* attributes don't need to be bound, they already have
1263                          * an id that the drawing functions will use. Builtins have
1264                          * constant names. */
1265                         if (ELEM(input->source, GPU_SOURCE_ATTR, GPU_SOURCE_BUILTIN)) {
1266                                 continue;
1267                         }
1268
1269                         if (input->source == GPU_SOURCE_TEX)
1270                                 BLI_snprintf(input->shadername, sizeof(input->shadername), "samp%d", input->texid);
1271                         else {
1272                                 BLI_snprintf(input->shadername, sizeof(input->shadername), "unf%d", input->id);
1273                         }
1274
1275                         if (input->source == GPU_SOURCE_TEX) {
1276                                 if (input->bindtex) {
1277                                         input->shaderloc = GPU_shader_get_uniform_ensure(shader, input->shadername);
1278                                         /* extract nodes */
1279                                         BLI_remlink(&node->inputs, input);
1280                                         BLI_addtail(inputs, input);
1281                                 }
1282                         }
1283                 }
1284         }
1285 }
1286
1287 /* Node Link Functions */
1288
1289 static GPUNodeLink *GPU_node_link_create(void)
1290 {
1291         GPUNodeLink *link = MEM_callocN(sizeof(GPUNodeLink), "GPUNodeLink");
1292         link->users++;
1293
1294         return link;
1295 }
1296
1297 static void gpu_node_link_free(GPUNodeLink *link)
1298 {
1299         link->users--;
1300
1301         if (link->users < 0)
1302                 fprintf(stderr, "GPU_node_link_free: negative refcount\n");
1303
1304         if (link->users == 0) {
1305                 if (link->output)
1306                         link->output->link = NULL;
1307                 MEM_freeN(link);
1308         }
1309 }
1310
1311 /* Node Functions */
1312
1313 static GPUNode *GPU_node_begin(const char *name)
1314 {
1315         GPUNode *node = MEM_callocN(sizeof(GPUNode), "GPUNode");
1316
1317         node->name = name;
1318
1319         return node;
1320 }
1321
1322 static void gpu_node_input_link(GPUNode *node, GPUNodeLink *link, const eGPUType type)
1323 {
1324         GPUInput *input;
1325         GPUNode *outnode;
1326         const char *name;
1327
1328         if (link->link_type == GPU_NODE_LINK_OUTPUT) {
1329                 outnode = link->output->node;
1330                 name = outnode->name;
1331                 input = outnode->inputs.first;
1332
1333                 if ((STREQ(name, "set_value") || STREQ(name, "set_rgb") || STREQ(name, "set_rgba")) &&
1334                     (input->type == type))
1335                 {
1336                         input = MEM_dupallocN(outnode->inputs.first);
1337                         if (input->link)
1338                                 input->link->users++;
1339                         BLI_addtail(&node->inputs, input);
1340                         return;
1341                 }
1342         }
1343
1344         input = MEM_callocN(sizeof(GPUInput), "GPUInput");
1345         input->node = node;
1346         input->type = type;
1347
1348         switch (link->link_type) {
1349                 case GPU_NODE_LINK_BUILTIN:
1350                         input->source = GPU_SOURCE_BUILTIN;
1351                         input->builtin = link->builtin;
1352                         break;
1353                 case GPU_NODE_LINK_OUTPUT:
1354                         input->source = GPU_SOURCE_OUTPUT;
1355                         input->link = link;
1356                         link->users++;
1357                         break;
1358                 case GPU_NODE_LINK_COLORBAND:
1359                         input->source = GPU_SOURCE_TEX;
1360                         input->coba = link->coba;
1361                         break;
1362                 case GPU_NODE_LINK_IMAGE_BLENDER:
1363                         input->source = GPU_SOURCE_TEX;
1364                         input->ima = link->ima;
1365                         input->iuser = link->iuser;
1366                         input->image_isdata = link->image_isdata;
1367                         break;
1368                 case GPU_NODE_LINK_ATTR:
1369                         input->source = GPU_SOURCE_ATTR;
1370                         input->attr_type = link->attr_type;
1371                         BLI_strncpy(input->attr_name, link->attr_name, sizeof(input->attr_name));
1372                         break;
1373                 case GPU_NODE_LINK_CONSTANT:
1374                         input->source = (type == GPU_CLOSURE) ? GPU_SOURCE_STRUCT : GPU_SOURCE_CONSTANT;
1375                         break;
1376                 case GPU_NODE_LINK_UNIFORM:
1377                         input->source = GPU_SOURCE_UNIFORM;
1378                         break;
1379                 default:
1380                         break;
1381         }
1382
1383         if (ELEM(input->source, GPU_SOURCE_CONSTANT, GPU_SOURCE_UNIFORM)) {
1384                 memcpy(input->vec, link->data, type * sizeof(float));
1385         }
1386
1387         if (link->link_type != GPU_NODE_LINK_OUTPUT) {
1388                 MEM_freeN(link);
1389         }
1390         BLI_addtail(&node->inputs, input);
1391 }
1392
1393
1394 static const char *gpu_uniform_set_function_from_type(eNodeSocketDatatype type)
1395 {
1396         switch (type) {
1397                 /* For now INT is supported as float. */
1398                 case SOCK_INT:
1399                 case SOCK_FLOAT:
1400                         return "set_value";
1401                 case SOCK_VECTOR:
1402                         return "set_rgb";
1403                 case SOCK_RGBA:
1404                         return "set_rgba";
1405                 default:
1406                         BLI_assert(!"No gpu function for non-supported eNodeSocketDatatype");
1407                         return NULL;
1408         }
1409 }
1410
1411 /**
1412  * Link stack uniform buffer.
1413  * This is called for the input/output sockets that are note connected.
1414  */
1415 static GPUNodeLink *gpu_uniformbuffer_link(
1416         GPUMaterial *mat, bNode *node, GPUNodeStack *stack, const int index, const eNodeSocketInOut in_out)
1417 {
1418         bNodeSocket *socket;
1419
1420         if (in_out == SOCK_IN) {
1421                 socket = BLI_findlink(&node->inputs, index);
1422         }
1423         else {
1424                 socket = BLI_findlink(&node->outputs, index);
1425         }
1426
1427         BLI_assert(socket != NULL);
1428         BLI_assert(socket->in_out == in_out);
1429
1430         if ((socket->flag & SOCK_HIDE_VALUE) == 0) {
1431                 GPUNodeLink *link;
1432                 switch (socket->type) {
1433                         case SOCK_FLOAT:
1434                         {
1435                                 bNodeSocketValueFloat *socket_data = socket->default_value;
1436                                 link = GPU_uniform(&socket_data->value);
1437                                 break;
1438                         }
1439                         case SOCK_VECTOR:
1440                         {
1441                                 bNodeSocketValueVector *socket_data = socket->default_value;
1442                                 link = GPU_uniform(socket_data->value);
1443                                 break;
1444                         }
1445                         case SOCK_RGBA:
1446                         {
1447                                 bNodeSocketValueRGBA *socket_data = socket->default_value;
1448                                 link = GPU_uniform(socket_data->value);
1449                                 break;
1450                         }
1451                         default:
1452                                 return NULL;
1453                                 break;
1454                 }
1455
1456                 if (in_out == SOCK_IN) {
1457                         GPU_link(mat, gpu_uniform_set_function_from_type(socket->type), link, &stack->link);
1458                 }
1459                 return link;
1460         }
1461         return NULL;
1462 }
1463
1464 static void gpu_node_input_socket(GPUMaterial *material, bNode *bnode, GPUNode *node, GPUNodeStack *sock, const int index)
1465 {
1466         if (sock->link) {
1467                 gpu_node_input_link(node, sock->link, sock->type);
1468         }
1469         else if ((material != NULL) && (gpu_uniformbuffer_link(material, bnode, sock, index, SOCK_IN) != NULL)) {
1470                 gpu_node_input_link(node, sock->link, sock->type);
1471         }
1472         else {
1473                 gpu_node_input_link(node, GPU_constant(sock->vec), sock->type);
1474         }
1475 }
1476
1477 static void gpu_node_output(GPUNode *node, const eGPUType type, GPUNodeLink **link)
1478 {
1479         GPUOutput *output = MEM_callocN(sizeof(GPUOutput), "GPUOutput");
1480
1481         output->type = type;
1482         output->node = node;
1483
1484         if (link) {
1485                 *link = output->link = GPU_node_link_create();
1486                 output->link->link_type = GPU_NODE_LINK_OUTPUT;
1487                 output->link->output = output;
1488
1489                 /* note: the caller owns the reference to the link, GPUOutput
1490                  * merely points to it, and if the node is destroyed it will
1491                  * set that pointer to NULL */
1492         }
1493
1494         BLI_addtail(&node->outputs, output);
1495 }
1496
1497 void GPU_inputs_free(ListBase *inputs)
1498 {
1499         GPUInput *input;
1500
1501         for (input = inputs->first; input; input = input->next) {
1502                 if (input->link)
1503                         gpu_node_link_free(input->link);
1504         }
1505
1506         BLI_freelistN(inputs);
1507 }
1508
1509 static void gpu_node_free(GPUNode *node)
1510 {
1511         GPUOutput *output;
1512
1513         GPU_inputs_free(&node->inputs);
1514
1515         for (output = node->outputs.first; output; output = output->next)
1516                 if (output->link) {
1517                         output->link->output = NULL;
1518                         gpu_node_link_free(output->link);
1519                 }
1520
1521         BLI_freelistN(&node->outputs);
1522         MEM_freeN(node);
1523 }
1524
1525 static void gpu_nodes_free(ListBase *nodes)
1526 {
1527         GPUNode *node;
1528
1529         while ((node = BLI_pophead(nodes))) {
1530                 gpu_node_free(node);
1531         }
1532 }
1533
1534 /* vertex attributes */
1535
1536 void GPU_nodes_get_vertex_attrs(ListBase *nodes, GPUVertAttrLayers *attrs)
1537 {
1538         GPUNode *node;
1539         GPUInput *input;
1540         int a;
1541
1542         /* convert attributes requested by node inputs to an array of layers,
1543          * checking for duplicates and assigning id's starting from zero. */
1544
1545         memset(attrs, 0, sizeof(*attrs));
1546
1547         for (node = nodes->first; node; node = node->next) {
1548                 for (input = node->inputs.first; input; input = input->next) {
1549                         if (input->source == GPU_SOURCE_ATTR) {
1550                                 for (a = 0; a < attrs->totlayer; a++) {
1551                                         if (attrs->layer[a].type == input->attr_type &&
1552                                             STREQ(attrs->layer[a].name, input->attr_name))
1553                                         {
1554                                                 break;
1555                                         }
1556                                 }
1557
1558                                 if (a < GPU_MAX_ATTR) {
1559                                         if (a == attrs->totlayer) {
1560                                                 input->attr_id = attrs->totlayer++;
1561                                                 input->attr_first = true;
1562
1563                                                 attrs->layer[a].type = input->attr_type;
1564                                                 attrs->layer[a].attr_id = input->attr_id;
1565                                                 BLI_strncpy(
1566                                                         attrs->layer[a].name, input->attr_name,
1567                                                         sizeof(attrs->layer[a].name));
1568                                         }
1569                                         else {
1570                                                 input->attr_id = attrs->layer[a].attr_id;
1571                                         }
1572                                 }
1573                         }
1574                 }
1575         }
1576 }
1577
1578 /* varargs linking  */
1579
1580 GPUNodeLink *GPU_attribute(const CustomDataType type, const char *name)
1581 {
1582         GPUNodeLink *link = GPU_node_link_create();
1583         link->link_type = GPU_NODE_LINK_ATTR;
1584         link->attr_name = name;
1585         /* Fall back to the UV layer, which matches old behavior. */
1586         if (type == CD_AUTO_FROM_NAME && name[0] == '\0') {
1587                 link->attr_type = CD_MTFACE;
1588         }
1589         else {
1590                 link->attr_type = type;
1591         }
1592         return link;
1593 }
1594
1595 GPUNodeLink *GPU_constant(float *num)
1596 {
1597         GPUNodeLink *link = GPU_node_link_create();
1598         link->link_type = GPU_NODE_LINK_CONSTANT;
1599         link->data = num;
1600         return link;
1601 }
1602
1603 GPUNodeLink *GPU_uniform(float *num)
1604 {
1605         GPUNodeLink *link = GPU_node_link_create();
1606         link->link_type = GPU_NODE_LINK_UNIFORM;
1607         link->data = num;
1608         return link;
1609 }
1610
1611 GPUNodeLink *GPU_image(Image *ima, ImageUser *iuser, bool is_data)
1612 {
1613         GPUNodeLink *link = GPU_node_link_create();
1614         link->link_type = GPU_NODE_LINK_IMAGE_BLENDER;
1615         link->ima = ima;
1616         link->iuser = iuser;
1617         link->image_isdata = is_data;
1618         return link;
1619 }
1620
1621 GPUNodeLink *GPU_color_band(GPUMaterial *mat, int size, float *pixels, float *row)
1622 {
1623         GPUNodeLink *link = GPU_node_link_create();
1624         link->link_type = GPU_NODE_LINK_COLORBAND;
1625         link->coba = gpu_material_ramp_texture_row_set(mat, size, pixels, row);
1626         MEM_freeN(pixels);
1627         return link;
1628 }
1629
1630 GPUNodeLink *GPU_builtin(eGPUBuiltin builtin)
1631 {
1632         GPUNodeLink *link = GPU_node_link_create();
1633         link->link_type = GPU_NODE_LINK_BUILTIN;
1634         link->builtin = builtin;
1635         return link;
1636 }
1637
1638 bool GPU_link(GPUMaterial *mat, const char *name, ...)
1639 {
1640         GPUNode *node;
1641         GPUFunction *function;
1642         GPUNodeLink *link, **linkptr;
1643         va_list params;
1644         int i;
1645
1646         function = gpu_lookup_function(name);
1647         if (!function) {
1648                 fprintf(stderr, "GPU failed to find function %s\n", name);
1649                 return false;
1650         }
1651
1652         node = GPU_node_begin(name);
1653
1654         va_start(params, name);
1655         for (i = 0; i < function->totparam; i++) {
1656                 if (function->paramqual[i] != FUNCTION_QUAL_IN) {
1657                         linkptr = va_arg(params, GPUNodeLink **);
1658                         gpu_node_output(node, function->paramtype[i], linkptr);
1659                 }
1660                 else {
1661                         link = va_arg(params, GPUNodeLink *);
1662                         gpu_node_input_link(node, link, function->paramtype[i]);
1663                 }
1664         }
1665         va_end(params);
1666
1667         gpu_material_add_node(mat, node);
1668
1669         return true;
1670 }
1671
1672 bool GPU_stack_link(GPUMaterial *material, bNode *bnode, const char *name, GPUNodeStack *in, GPUNodeStack *out, ...)
1673 {
1674         GPUNode *node;
1675         GPUFunction *function;
1676         GPUNodeLink *link, **linkptr;
1677         va_list params;
1678         int i, totin, totout;
1679
1680         function = gpu_lookup_function(name);
1681         if (!function) {
1682                 fprintf(stderr, "GPU failed to find function %s\n", name);
1683                 return false;
1684         }
1685
1686         node = GPU_node_begin(name);
1687         totin = 0;
1688         totout = 0;
1689
1690         if (in) {
1691                 for (i = 0; !in[i].end; i++) {
1692                         if (in[i].type != GPU_NONE) {
1693                                 gpu_node_input_socket(material, bnode, node, &in[i], i);
1694                                 totin++;
1695                         }
1696                 }
1697         }
1698
1699         if (out) {
1700                 for (i = 0; !out[i].end; i++) {
1701                         if (out[i].type != GPU_NONE) {
1702                                 gpu_node_output(node, out[i].type, &out[i].link);
1703                                 totout++;
1704                         }
1705                 }
1706         }
1707
1708         va_start(params, out);
1709         for (i = 0; i < function->totparam; i++) {
1710                 if (function->paramqual[i] != FUNCTION_QUAL_IN) {
1711                         if (totout == 0) {
1712                                 linkptr = va_arg(params, GPUNodeLink **);
1713                                 gpu_node_output(node, function->paramtype[i], linkptr);
1714                         }
1715                         else
1716                                 totout--;
1717                 }
1718                 else {
1719                         if (totin == 0) {
1720                                 link = va_arg(params, GPUNodeLink *);
1721                                 if (link->socket)
1722                                         gpu_node_input_socket(NULL, NULL, node, link->socket, -1);
1723                                 else
1724                                         gpu_node_input_link(node, link, function->paramtype[i]);
1725                         }
1726                         else
1727                                 totin--;
1728                 }
1729         }
1730         va_end(params);
1731
1732         gpu_material_add_node(material, node);
1733
1734         return true;
1735 }
1736
1737 GPUNodeLink *GPU_uniformbuffer_link_out(GPUMaterial *mat, bNode *node, GPUNodeStack *stack, const int index)
1738 {
1739         return gpu_uniformbuffer_link(mat, node, stack, index, SOCK_OUT);
1740 }
1741
1742 /* Pass create/free */
1743
1744 static void gpu_nodes_tag(GPUNodeLink *link)
1745 {
1746         GPUNode *node;
1747         GPUInput *input;
1748
1749         if (!link->output)
1750                 return;
1751
1752         node = link->output->node;
1753         if (node->tag)
1754                 return;
1755
1756         node->tag = true;
1757         for (input = node->inputs.first; input; input = input->next)
1758                 if (input->link)
1759                         gpu_nodes_tag(input->link);
1760 }
1761
1762 void GPU_nodes_prune(ListBase *nodes, GPUNodeLink *outlink)
1763 {
1764         GPUNode *node, *next;
1765
1766         for (node = nodes->first; node; node = node->next)
1767                 node->tag = false;
1768
1769         gpu_nodes_tag(outlink);
1770
1771         for (node = nodes->first; node; node = next) {
1772                 next = node->next;
1773
1774                 if (!node->tag) {
1775                         BLI_remlink(nodes, node);
1776                         gpu_node_free(node);
1777                 }
1778         }
1779 }
1780
1781 static bool gpu_pass_is_valid(GPUPass *pass)
1782 {
1783         /* Shader is not null if compilation is successful. */
1784         return (pass->compiled == false || pass->shader != NULL);
1785 }
1786
1787 GPUPass *GPU_generate_pass(
1788         GPUMaterial *material,
1789         GPUNodeLink *frag_outlink,
1790         struct GPUVertAttrLayers *attrs,
1791         ListBase *nodes,
1792         int *builtins,
1793         const char *vert_code,
1794         const char *geom_code,
1795         const char *frag_lib,
1796         const char *defines)
1797 {
1798         char *vertexcode, *geometrycode, *fragmentcode;
1799         GPUPass *pass = NULL, *pass_hash = NULL;
1800
1801         /* prune unused nodes */
1802         GPU_nodes_prune(nodes, frag_outlink);
1803
1804         GPU_nodes_get_vertex_attrs(nodes, attrs);
1805
1806         /* generate code */
1807         char *fragmentgen = code_generate_fragment(material, nodes, frag_outlink->output, builtins);
1808
1809         /* Cache lookup: Reuse shaders already compiled */
1810         uint32_t hash = gpu_pass_hash(fragmentgen, defines, attrs);
1811         pass_hash = gpu_pass_cache_lookup(hash);
1812
1813         if (pass_hash && (pass_hash->next == NULL || pass_hash->next->hash != hash)) {
1814                 /* No collision, just return the pass. */
1815                 MEM_freeN(fragmentgen);
1816                 if (!gpu_pass_is_valid(pass_hash)) {
1817                         /* Shader has already been created but failed to compile. */
1818                         return NULL;
1819                 }
1820                 pass_hash->refcount += 1;
1821                 return pass_hash;
1822         }
1823
1824         /* Either the shader is not compiled or there is a hash collision...
1825          * continue generating the shader strings. */
1826         char *tmp = BLI_strdupcat(frag_lib, glsl_material_library);
1827
1828         geometrycode = code_generate_geometry(nodes, geom_code, defines);
1829         vertexcode = code_generate_vertex(nodes, vert_code, (geometrycode != NULL));
1830         fragmentcode = BLI_strdupcat(tmp, fragmentgen);
1831
1832         MEM_freeN(fragmentgen);
1833         MEM_freeN(tmp);
1834
1835         if (pass_hash) {
1836                 /* Cache lookup: Reuse shaders already compiled */
1837                 pass = gpu_pass_cache_resolve_collision(pass_hash, vertexcode, geometrycode, fragmentcode, defines, hash);
1838         }
1839
1840         if (pass) {
1841                 /* Cache hit. Reuse the same GPUPass and GPUShader. */
1842                 if (!gpu_pass_is_valid(pass)) {
1843                         /* Shader has already been created but failed to compile. */
1844                         return NULL;
1845                 }
1846
1847                 MEM_SAFE_FREE(vertexcode);
1848                 MEM_SAFE_FREE(fragmentcode);
1849                 MEM_SAFE_FREE(geometrycode);
1850
1851                 pass->refcount += 1;
1852         }
1853         else {
1854                 /* We still create a pass even if shader compilation
1855                  * fails to avoid trying to compile again and again. */
1856                 pass = MEM_callocN(sizeof(GPUPass), "GPUPass");
1857                 pass->shader = NULL;
1858                 pass->refcount = 1;
1859                 pass->hash = hash;
1860                 pass->vertexcode = vertexcode;
1861                 pass->fragmentcode = fragmentcode;
1862                 pass->geometrycode = geometrycode;
1863                 pass->defines = (defines) ? BLI_strdup(defines) : NULL;
1864                 pass->compiled = false;
1865
1866                 BLI_spin_lock(&pass_cache_spin);
1867                 if (pass_hash != NULL) {
1868                         /* Add after the first pass having the same hash. */
1869                         pass->next = pass_hash->next;
1870                         pass_hash->next = pass;
1871                 }
1872                 else {
1873                         /* No other pass have same hash, just prepend to the list. */
1874                         BLI_LINKS_PREPEND(pass_cache, pass);
1875                 }
1876                 BLI_spin_unlock(&pass_cache_spin);
1877         }
1878
1879         return pass;
1880 }
1881
1882 static int count_active_texture_sampler(GPUShader *shader, char *source)
1883 {
1884         char *code = source;
1885         int samplers_id[64]; /* Remember this is per stage. */
1886         int sampler_len = 0;
1887
1888         while ((code = strstr(code, "uniform "))) {
1889                 /* Move past "uniform". */
1890                 code += 7;
1891                 /* Skip following spaces. */
1892                 while (*code == ' ') { code++; }
1893                 /* Skip "i" from potential isamplers. */
1894                 if (*code == 'i') { code++; }
1895                 /* Skip following spaces. */
1896                 if (gpu_str_prefix(code, "sampler")) {
1897                         /* Move past "uniform". */
1898                         code += 7;
1899                         /* Skip sampler type suffix. */
1900                         while (*code != ' ' && *code != '\0') { code++; }
1901                         /* Skip following spaces. */
1902                         while (*code == ' ') { code++; }
1903
1904                         if (*code != '\0') {
1905                                 char sampler_name[64];
1906                                 code = gpu_str_skip_token(code, sampler_name, sizeof(sampler_name));
1907                                 int id = GPU_shader_get_uniform_ensure(shader, sampler_name);
1908
1909                                 if (id == -1) {
1910                                         continue;
1911                                 }
1912                                 /* Catch duplicates. */
1913                                 bool is_duplicate = false;
1914                                 for (int i = 0; i < sampler_len; ++i) {
1915                                         if (samplers_id[i] == id) {
1916                                                 is_duplicate = true;
1917                                         }
1918                                 }
1919
1920                                 if (!is_duplicate) {
1921                                         samplers_id[sampler_len] = id;
1922                                         sampler_len++;
1923                                 }
1924                         }
1925                 }
1926         }
1927
1928         return sampler_len;
1929 }
1930
1931 static bool gpu_pass_shader_validate(GPUPass *pass)
1932 {
1933         if (pass->shader == NULL) {
1934                 return false;
1935         }
1936
1937         /* NOTE: The only drawback of this method is that it will count a sampler
1938          * used in the fragment shader and only declared (but not used) in the vertex
1939          * shader as used by both. But this corner case is not happening for now. */
1940         int vert_samplers_len = count_active_texture_sampler(pass->shader, pass->vertexcode);
1941         int frag_samplers_len = count_active_texture_sampler(pass->shader, pass->fragmentcode);
1942
1943         int total_samplers_len = vert_samplers_len + frag_samplers_len;
1944
1945         /* Validate against opengl limit. */
1946         if ((frag_samplers_len > GPU_max_textures_frag()) ||
1947             (frag_samplers_len > GPU_max_textures_vert()))
1948         {
1949                 return false;
1950         }
1951
1952         if (pass->geometrycode) {
1953                 int geom_samplers_len = count_active_texture_sampler(pass->shader, pass->geometrycode);
1954                 total_samplers_len += geom_samplers_len;
1955                 if (geom_samplers_len > GPU_max_textures_geom()) {
1956                         return false;
1957                 }
1958         }
1959
1960         return (total_samplers_len <= GPU_max_textures());
1961 }
1962
1963 void GPU_pass_compile(GPUPass *pass, const char *shname)
1964 {
1965         if (!pass->compiled) {
1966                 pass->shader = GPU_shader_create(
1967                         pass->vertexcode,
1968                         pass->fragmentcode,
1969                         pass->geometrycode,
1970                         NULL,
1971                         pass->defines,
1972                         shname);
1973
1974                 /* NOTE: Some drivers / gpu allows more active samplers than the opengl limit.
1975                  * We need to make sure to count active samplers to avoid undefined behavior. */
1976                 if (!gpu_pass_shader_validate(pass)) {
1977                         if (pass->shader != NULL) {
1978                                 fprintf(stderr, "GPUShader: error: too many samplers in shader.\n");
1979                                 GPU_shader_free(pass->shader);
1980                         }
1981                         pass->shader = NULL;
1982                 }
1983                 else if (!BLI_thread_is_main()) {
1984                         /* For some Intel drivers, you must use the program at least once
1985                          * in the rendering context that it is linked. */
1986                         glUseProgram(GPU_shader_get_program(pass->shader));
1987                         glUseProgram(0);
1988                 }
1989
1990                 pass->compiled = true;
1991         }
1992 }
1993
1994 void GPU_pass_release(GPUPass *pass)
1995 {
1996         BLI_assert(pass->refcount > 0);
1997         pass->refcount--;
1998 }
1999
2000 static void gpu_pass_free(GPUPass *pass)
2001 {
2002         BLI_assert(pass->refcount == 0);
2003         if (pass->shader) {
2004                 GPU_shader_free(pass->shader);
2005         }
2006         MEM_SAFE_FREE(pass->fragmentcode);
2007         MEM_SAFE_FREE(pass->geometrycode);
2008         MEM_SAFE_FREE(pass->vertexcode);
2009         MEM_SAFE_FREE(pass->defines);
2010         MEM_freeN(pass);
2011 }
2012
2013 void GPU_pass_free_nodes(ListBase *nodes)
2014 {
2015         gpu_nodes_free(nodes);
2016 }
2017
2018 void GPU_pass_cache_garbage_collect(void)
2019 {
2020         static int lasttime = 0;
2021         const int shadercollectrate = 60; /* hardcoded for now. */
2022         int ctime = (int)PIL_check_seconds_timer();
2023
2024         if (ctime < shadercollectrate + lasttime)
2025                 return;
2026
2027         lasttime = ctime;
2028
2029         BLI_spin_lock(&pass_cache_spin);
2030         GPUPass *next, **prev_pass = &pass_cache;
2031         for (GPUPass *pass = pass_cache; pass; pass = next) {
2032                 next = pass->next;
2033                 if (pass->refcount == 0) {
2034                         /* Remove from list */
2035                         *prev_pass = next;
2036                         gpu_pass_free(pass);
2037                 }
2038                 else {
2039                         prev_pass = &pass->next;
2040                 }
2041         }
2042         BLI_spin_unlock(&pass_cache_spin);
2043 }
2044
2045 void GPU_pass_cache_init(void)
2046 {
2047         BLI_spin_init(&pass_cache_spin);
2048 }
2049
2050 void GPU_pass_cache_free(void)
2051 {
2052         BLI_spin_lock(&pass_cache_spin);
2053         while (pass_cache) {
2054                 GPUPass *next = pass_cache->next;
2055                 gpu_pass_free(pass_cache);
2056                 pass_cache = next;
2057         }
2058         BLI_spin_unlock(&pass_cache_spin);
2059
2060         BLI_spin_end(&pass_cache_spin);
2061 }