1 /*
2 * Copyright © 2010 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
22 */
23
24 /**
25 * \file ir_constant_expression.cpp
26 * Evaluate and process constant valued expressions
27 *
28 * In GLSL, constant valued expressions are used in several places. These
29 * must be processed and evaluated very early in the compilation process.
30 *
31 * * Sizes of arrays
32 * * Initializers for uniforms
33 * * Initializers for \c const variables
34 */
35
36 #include <math.h>
37 #include "main/core.h" /* for MAX2, MIN2, CLAMP */
38 #include "util/rounding.h" /* for _mesa_roundeven */
39 #include "util/half_float.h"
40 #include "ir.h"
41 #include "compiler/glsl_types.h"
42 #include "util/hash_table.h"
43
44 static float
dot_f(ir_constant * op0,ir_constant * op1)45 dot_f(ir_constant *op0, ir_constant *op1)
46 {
47 assert(op0->type->is_float() && op1->type->is_float());
48
49 float result = 0;
50 for (unsigned c = 0; c < op0->type->components(); c++)
51 result += op0->value.f[c] * op1->value.f[c];
52
53 return result;
54 }
55
56 static double
dot_d(ir_constant * op0,ir_constant * op1)57 dot_d(ir_constant *op0, ir_constant *op1)
58 {
59 assert(op0->type->is_double() && op1->type->is_double());
60
61 double result = 0;
62 for (unsigned c = 0; c < op0->type->components(); c++)
63 result += op0->value.d[c] * op1->value.d[c];
64
65 return result;
66 }
67
68 /* This method is the only one supported by gcc. Unions in particular
69 * are iffy, and read-through-converted-pointer is killed by strict
70 * aliasing. OTOH, the compiler sees through the memcpy, so the
71 * resulting asm is reasonable.
72 */
73 static float
bitcast_u2f(unsigned int u)74 bitcast_u2f(unsigned int u)
75 {
76 assert(sizeof(float) == sizeof(unsigned int));
77 float f;
78 memcpy(&f, &u, sizeof(f));
79 return f;
80 }
81
82 static unsigned int
bitcast_f2u(float f)83 bitcast_f2u(float f)
84 {
85 assert(sizeof(float) == sizeof(unsigned int));
86 unsigned int u;
87 memcpy(&u, &f, sizeof(f));
88 return u;
89 }
90
91 /**
92 * Evaluate one component of a floating-point 4x8 unpacking function.
93 */
94 typedef uint8_t
95 (*pack_1x8_func_t)(float);
96
97 /**
98 * Evaluate one component of a floating-point 2x16 unpacking function.
99 */
100 typedef uint16_t
101 (*pack_1x16_func_t)(float);
102
103 /**
104 * Evaluate one component of a floating-point 4x8 unpacking function.
105 */
106 typedef float
107 (*unpack_1x8_func_t)(uint8_t);
108
109 /**
110 * Evaluate one component of a floating-point 2x16 unpacking function.
111 */
112 typedef float
113 (*unpack_1x16_func_t)(uint16_t);
114
115 /**
116 * Evaluate a 2x16 floating-point packing function.
117 */
118 static uint32_t
pack_2x16(pack_1x16_func_t pack_1x16,float x,float y)119 pack_2x16(pack_1x16_func_t pack_1x16,
120 float x, float y)
121 {
122 /* From section 8.4 of the GLSL ES 3.00 spec:
123 *
124 * packSnorm2x16
125 * -------------
126 * The first component of the vector will be written to the least
127 * significant bits of the output; the last component will be written to
128 * the most significant bits.
129 *
130 * The specifications for the other packing functions contain similar
131 * language.
132 */
133 uint32_t u = 0;
134 u |= ((uint32_t) pack_1x16(x) << 0);
135 u |= ((uint32_t) pack_1x16(y) << 16);
136 return u;
137 }
138
139 /**
140 * Evaluate a 4x8 floating-point packing function.
141 */
142 static uint32_t
pack_4x8(pack_1x8_func_t pack_1x8,float x,float y,float z,float w)143 pack_4x8(pack_1x8_func_t pack_1x8,
144 float x, float y, float z, float w)
145 {
146 /* From section 8.4 of the GLSL 4.30 spec:
147 *
148 * packSnorm4x8
149 * ------------
150 * The first component of the vector will be written to the least
151 * significant bits of the output; the last component will be written to
152 * the most significant bits.
153 *
154 * The specifications for the other packing functions contain similar
155 * language.
156 */
157 uint32_t u = 0;
158 u |= ((uint32_t) pack_1x8(x) << 0);
159 u |= ((uint32_t) pack_1x8(y) << 8);
160 u |= ((uint32_t) pack_1x8(z) << 16);
161 u |= ((uint32_t) pack_1x8(w) << 24);
162 return u;
163 }
164
165 /**
166 * Evaluate a 2x16 floating-point unpacking function.
167 */
168 static void
unpack_2x16(unpack_1x16_func_t unpack_1x16,uint32_t u,float * x,float * y)169 unpack_2x16(unpack_1x16_func_t unpack_1x16,
170 uint32_t u,
171 float *x, float *y)
172 {
173 /* From section 8.4 of the GLSL ES 3.00 spec:
174 *
175 * unpackSnorm2x16
176 * ---------------
177 * The first component of the returned vector will be extracted from
178 * the least significant bits of the input; the last component will be
179 * extracted from the most significant bits.
180 *
181 * The specifications for the other unpacking functions contain similar
182 * language.
183 */
184 *x = unpack_1x16((uint16_t) (u & 0xffff));
185 *y = unpack_1x16((uint16_t) (u >> 16));
186 }
187
188 /**
189 * Evaluate a 4x8 floating-point unpacking function.
190 */
191 static void
unpack_4x8(unpack_1x8_func_t unpack_1x8,uint32_t u,float * x,float * y,float * z,float * w)192 unpack_4x8(unpack_1x8_func_t unpack_1x8, uint32_t u,
193 float *x, float *y, float *z, float *w)
194 {
195 /* From section 8.4 of the GLSL 4.30 spec:
196 *
197 * unpackSnorm4x8
198 * --------------
199 * The first component of the returned vector will be extracted from
200 * the least significant bits of the input; the last component will be
201 * extracted from the most significant bits.
202 *
203 * The specifications for the other unpacking functions contain similar
204 * language.
205 */
206 *x = unpack_1x8((uint8_t) (u & 0xff));
207 *y = unpack_1x8((uint8_t) (u >> 8));
208 *z = unpack_1x8((uint8_t) (u >> 16));
209 *w = unpack_1x8((uint8_t) (u >> 24));
210 }
211
212 /**
213 * Evaluate one component of packSnorm4x8.
214 */
215 static uint8_t
pack_snorm_1x8(float x)216 pack_snorm_1x8(float x)
217 {
218 /* From section 8.4 of the GLSL 4.30 spec:
219 *
220 * packSnorm4x8
221 * ------------
222 * The conversion for component c of v to fixed point is done as
223 * follows:
224 *
225 * packSnorm4x8: round(clamp(c, -1, +1) * 127.0)
226 */
227 return (uint8_t)
228 _mesa_lroundevenf(CLAMP(x, -1.0f, +1.0f) * 127.0f);
229 }
230
231 /**
232 * Evaluate one component of packSnorm2x16.
233 */
234 static uint16_t
pack_snorm_1x16(float x)235 pack_snorm_1x16(float x)
236 {
237 /* From section 8.4 of the GLSL ES 3.00 spec:
238 *
239 * packSnorm2x16
240 * -------------
241 * The conversion for component c of v to fixed point is done as
242 * follows:
243 *
244 * packSnorm2x16: round(clamp(c, -1, +1) * 32767.0)
245 */
246 return (uint16_t)
247 _mesa_lroundevenf(CLAMP(x, -1.0f, +1.0f) * 32767.0f);
248 }
249
250 /**
251 * Evaluate one component of unpackSnorm4x8.
252 */
253 static float
unpack_snorm_1x8(uint8_t u)254 unpack_snorm_1x8(uint8_t u)
255 {
256 /* From section 8.4 of the GLSL 4.30 spec:
257 *
258 * unpackSnorm4x8
259 * --------------
260 * The conversion for unpacked fixed-point value f to floating point is
261 * done as follows:
262 *
263 * unpackSnorm4x8: clamp(f / 127.0, -1, +1)
264 */
265 return CLAMP((int8_t) u / 127.0f, -1.0f, +1.0f);
266 }
267
268 /**
269 * Evaluate one component of unpackSnorm2x16.
270 */
271 static float
unpack_snorm_1x16(uint16_t u)272 unpack_snorm_1x16(uint16_t u)
273 {
274 /* From section 8.4 of the GLSL ES 3.00 spec:
275 *
276 * unpackSnorm2x16
277 * ---------------
278 * The conversion for unpacked fixed-point value f to floating point is
279 * done as follows:
280 *
281 * unpackSnorm2x16: clamp(f / 32767.0, -1, +1)
282 */
283 return CLAMP((int16_t) u / 32767.0f, -1.0f, +1.0f);
284 }
285
286 /**
287 * Evaluate one component packUnorm4x8.
288 */
289 static uint8_t
pack_unorm_1x8(float x)290 pack_unorm_1x8(float x)
291 {
292 /* From section 8.4 of the GLSL 4.30 spec:
293 *
294 * packUnorm4x8
295 * ------------
296 * The conversion for component c of v to fixed point is done as
297 * follows:
298 *
299 * packUnorm4x8: round(clamp(c, 0, +1) * 255.0)
300 */
301 return (uint8_t) (int) _mesa_roundevenf(CLAMP(x, 0.0f, 1.0f) * 255.0f);
302 }
303
304 /**
305 * Evaluate one component packUnorm2x16.
306 */
307 static uint16_t
pack_unorm_1x16(float x)308 pack_unorm_1x16(float x)
309 {
310 /* From section 8.4 of the GLSL ES 3.00 spec:
311 *
312 * packUnorm2x16
313 * -------------
314 * The conversion for component c of v to fixed point is done as
315 * follows:
316 *
317 * packUnorm2x16: round(clamp(c, 0, +1) * 65535.0)
318 */
319 return (uint16_t) (int)
320 _mesa_roundevenf(CLAMP(x, 0.0f, 1.0f) * 65535.0f);
321 }
322
323 /**
324 * Evaluate one component of unpackUnorm4x8.
325 */
326 static float
unpack_unorm_1x8(uint8_t u)327 unpack_unorm_1x8(uint8_t u)
328 {
329 /* From section 8.4 of the GLSL 4.30 spec:
330 *
331 * unpackUnorm4x8
332 * --------------
333 * The conversion for unpacked fixed-point value f to floating point is
334 * done as follows:
335 *
336 * unpackUnorm4x8: f / 255.0
337 */
338 return (float) u / 255.0f;
339 }
340
341 /**
342 * Evaluate one component of unpackUnorm2x16.
343 */
344 static float
unpack_unorm_1x16(uint16_t u)345 unpack_unorm_1x16(uint16_t u)
346 {
347 /* From section 8.4 of the GLSL ES 3.00 spec:
348 *
349 * unpackUnorm2x16
350 * ---------------
351 * The conversion for unpacked fixed-point value f to floating point is
352 * done as follows:
353 *
354 * unpackUnorm2x16: f / 65535.0
355 */
356 return (float) u / 65535.0f;
357 }
358
359 /**
360 * Evaluate one component of packHalf2x16.
361 */
362 static uint16_t
pack_half_1x16(float x)363 pack_half_1x16(float x)
364 {
365 return _mesa_float_to_half(x);
366 }
367
368 /**
369 * Evaluate one component of unpackHalf2x16.
370 */
371 static float
unpack_half_1x16(uint16_t u)372 unpack_half_1x16(uint16_t u)
373 {
374 return _mesa_half_to_float(u);
375 }
376
377 /**
378 * Get the constant that is ultimately referenced by an r-value, in a constant
379 * expression evaluation context.
380 *
381 * The offset is used when the reference is to a specific column of a matrix.
382 */
383 static bool
constant_referenced(const ir_dereference * deref,struct hash_table * variable_context,ir_constant * & store,int & offset)384 constant_referenced(const ir_dereference *deref,
385 struct hash_table *variable_context,
386 ir_constant *&store, int &offset)
387 {
388 store = NULL;
389 offset = 0;
390
391 if (variable_context == NULL)
392 return false;
393
394 switch (deref->ir_type) {
395 case ir_type_dereference_array: {
396 const ir_dereference_array *const da =
397 (const ir_dereference_array *) deref;
398
399 ir_constant *const index_c =
400 da->array_index->constant_expression_value(variable_context);
401
402 if (!index_c || !index_c->type->is_scalar() || !index_c->type->is_integer())
403 break;
404
405 const int index = index_c->type->base_type == GLSL_TYPE_INT ?
406 index_c->get_int_component(0) :
407 index_c->get_uint_component(0);
408
409 ir_constant *substore;
410 int suboffset;
411
412 const ir_dereference *const deref = da->array->as_dereference();
413 if (!deref)
414 break;
415
416 if (!constant_referenced(deref, variable_context, substore, suboffset))
417 break;
418
419 const glsl_type *const vt = da->array->type;
420 if (vt->is_array()) {
421 store = substore->get_array_element(index);
422 offset = 0;
423 } else if (vt->is_matrix()) {
424 store = substore;
425 offset = index * vt->vector_elements;
426 } else if (vt->is_vector()) {
427 store = substore;
428 offset = suboffset + index;
429 }
430
431 break;
432 }
433
434 case ir_type_dereference_record: {
435 const ir_dereference_record *const dr =
436 (const ir_dereference_record *) deref;
437
438 const ir_dereference *const deref = dr->record->as_dereference();
439 if (!deref)
440 break;
441
442 ir_constant *substore;
443 int suboffset;
444
445 if (!constant_referenced(deref, variable_context, substore, suboffset))
446 break;
447
448 /* Since we're dropping it on the floor...
449 */
450 assert(suboffset == 0);
451
452 store = substore->get_record_field(dr->field);
453 break;
454 }
455
456 case ir_type_dereference_variable: {
457 const ir_dereference_variable *const dv =
458 (const ir_dereference_variable *) deref;
459
460 hash_entry *entry = _mesa_hash_table_search(variable_context, dv->var);
461 if (entry)
462 store = (ir_constant *) entry->data;
463 break;
464 }
465
466 default:
467 assert(!"Should not get here.");
468 break;
469 }
470
471 return store != NULL;
472 }
473
474
475 ir_constant *
constant_expression_value(struct hash_table *)476 ir_rvalue::constant_expression_value(struct hash_table *)
477 {
478 assert(this->type->is_error());
479 return NULL;
480 }
481
482 static uint32_t
bitfield_reverse(uint32_t v)483 bitfield_reverse(uint32_t v)
484 {
485 /* http://graphics.stanford.edu/~seander/bithacks.html#BitReverseObvious */
486 uint32_t r = v; // r will be reversed bits of v; first get LSB of v
487 int s = sizeof(v) * CHAR_BIT - 1; // extra shift needed at end
488
489 for (v >>= 1; v; v >>= 1) {
490 r <<= 1;
491 r |= v & 1;
492 s--;
493 }
494 r <<= s; // shift when v's highest bits are zero
495
496 return r;
497 }
498
499 static int
find_msb_uint(uint32_t v)500 find_msb_uint(uint32_t v)
501 {
502 int count = 0;
503
504 /* If v == 0, then the loop will terminate when count == 32. In that case
505 * 31-count will produce the -1 result required by GLSL findMSB().
506 */
507 while (((v & (1u << 31)) == 0) && count != 32) {
508 count++;
509 v <<= 1;
510 }
511
512 return 31 - count;
513 }
514
515 static int
find_msb_int(int32_t v)516 find_msb_int(int32_t v)
517 {
518 /* If v is signed, findMSB() returns the position of the most significant
519 * zero bit.
520 */
521 return find_msb_uint(v < 0 ? ~v : v);
522 }
523
524 static float
ldexpf_flush_subnormal(float x,int exp)525 ldexpf_flush_subnormal(float x, int exp)
526 {
527 const float result = ldexpf(x, exp);
528
529 /* Flush subnormal values to zero. */
530 return !isnormal(result) ? copysignf(0.0f, x) : result;
531 }
532
533 static double
ldexp_flush_subnormal(double x,int exp)534 ldexp_flush_subnormal(double x, int exp)
535 {
536 const double result = ldexp(x, exp);
537
538 /* Flush subnormal values to zero. */
539 return !isnormal(result) ? copysign(0.0, x) : result;
540 }
541
542 static uint32_t
bitfield_extract_uint(uint32_t value,int offset,int bits)543 bitfield_extract_uint(uint32_t value, int offset, int bits)
544 {
545 if (bits == 0)
546 return 0;
547 else if (offset < 0 || bits < 0)
548 return 0; /* Undefined, per spec. */
549 else if (offset + bits > 32)
550 return 0; /* Undefined, per spec. */
551 else {
552 value <<= 32 - bits - offset;
553 value >>= 32 - bits;
554 return value;
555 }
556 }
557
558 static int32_t
bitfield_extract_int(int32_t value,int offset,int bits)559 bitfield_extract_int(int32_t value, int offset, int bits)
560 {
561 if (bits == 0)
562 return 0;
563 else if (offset < 0 || bits < 0)
564 return 0; /* Undefined, per spec. */
565 else if (offset + bits > 32)
566 return 0; /* Undefined, per spec. */
567 else {
568 value <<= 32 - bits - offset;
569 value >>= 32 - bits;
570 return value;
571 }
572 }
573
574 static uint32_t
bitfield_insert(uint32_t base,uint32_t insert,int offset,int bits)575 bitfield_insert(uint32_t base, uint32_t insert, int offset, int bits)
576 {
577 if (bits == 0)
578 return base;
579 else if (offset < 0 || bits < 0)
580 return 0; /* Undefined, per spec. */
581 else if (offset + bits > 32)
582 return 0; /* Undefined, per spec. */
583 else {
584 unsigned insert_mask = ((1ull << bits) - 1) << offset;
585
586 insert <<= offset;
587 insert &= insert_mask;
588 base &= ~insert_mask;
589
590 return base | insert;
591 }
592 }
593
594 ir_constant *
constant_expression_value(struct hash_table * variable_context)595 ir_expression::constant_expression_value(struct hash_table *variable_context)
596 {
597 if (this->type->is_error())
598 return NULL;
599
600 ir_constant *op[ARRAY_SIZE(this->operands)] = { NULL, };
601 ir_constant_data data;
602
603 memset(&data, 0, sizeof(data));
604
605 for (unsigned operand = 0; operand < this->get_num_operands(); operand++) {
606 op[operand] = this->operands[operand]->constant_expression_value(variable_context);
607 if (!op[operand])
608 return NULL;
609 }
610
611 if (op[1] != NULL)
612 switch (this->operation) {
613 case ir_binop_lshift:
614 case ir_binop_rshift:
615 case ir_binop_ldexp:
616 case ir_binop_interpolate_at_offset:
617 case ir_binop_interpolate_at_sample:
618 case ir_binop_vector_extract:
619 case ir_triop_csel:
620 case ir_triop_bitfield_extract:
621 break;
622
623 default:
624 assert(op[0]->type->base_type == op[1]->type->base_type);
625 break;
626 }
627
628 bool op0_scalar = op[0]->type->is_scalar();
629 bool op1_scalar = op[1] != NULL && op[1]->type->is_scalar();
630
631 /* When iterating over a vector or matrix's components, we want to increase
632 * the loop counter. However, for scalars, we want to stay at 0.
633 */
634 unsigned c0_inc = op0_scalar ? 0 : 1;
635 unsigned c1_inc = op1_scalar ? 0 : 1;
636 unsigned components;
637 if (op1_scalar || !op[1]) {
638 components = op[0]->type->components();
639 } else {
640 components = op[1]->type->components();
641 }
642
643 void *ctx = ralloc_parent(this);
644
645 /* Handle array operations here, rather than below. */
646 if (op[0]->type->is_array()) {
647 assert(op[1] != NULL && op[1]->type->is_array());
648 switch (this->operation) {
649 case ir_binop_all_equal:
650 return new(ctx) ir_constant(op[0]->has_value(op[1]));
651 case ir_binop_any_nequal:
652 return new(ctx) ir_constant(!op[0]->has_value(op[1]));
653 default:
654 break;
655 }
656 return NULL;
657 }
658
659 #include "ir_expression_operation_constant.h"
660
661 return new(ctx) ir_constant(this->type, &data);
662 }
663
664
665 ir_constant *
constant_expression_value(struct hash_table *)666 ir_texture::constant_expression_value(struct hash_table *)
667 {
668 /* texture lookups aren't constant expressions */
669 return NULL;
670 }
671
672
673 ir_constant *
constant_expression_value(struct hash_table * variable_context)674 ir_swizzle::constant_expression_value(struct hash_table *variable_context)
675 {
676 ir_constant *v = this->val->constant_expression_value(variable_context);
677
678 if (v != NULL) {
679 ir_constant_data data = { { 0 } };
680
681 const unsigned swiz_idx[4] = {
682 this->mask.x, this->mask.y, this->mask.z, this->mask.w
683 };
684
685 for (unsigned i = 0; i < this->mask.num_components; i++) {
686 switch (v->type->base_type) {
687 case GLSL_TYPE_UINT:
688 case GLSL_TYPE_INT: data.u[i] = v->value.u[swiz_idx[i]]; break;
689 case GLSL_TYPE_FLOAT: data.f[i] = v->value.f[swiz_idx[i]]; break;
690 case GLSL_TYPE_BOOL: data.b[i] = v->value.b[swiz_idx[i]]; break;
691 case GLSL_TYPE_DOUBLE:data.d[i] = v->value.d[swiz_idx[i]]; break;
692 default: assert(!"Should not get here."); break;
693 }
694 }
695
696 void *ctx = ralloc_parent(this);
697 return new(ctx) ir_constant(this->type, &data);
698 }
699 return NULL;
700 }
701
702
703 ir_constant *
constant_expression_value(struct hash_table * variable_context)704 ir_dereference_variable::constant_expression_value(struct hash_table *variable_context)
705 {
706 assert(var);
707
708 /* Give priority to the context hashtable, if it exists */
709 if (variable_context) {
710 hash_entry *entry = _mesa_hash_table_search(variable_context, var);
711
712 if(entry)
713 return (ir_constant *) entry->data;
714 }
715
716 /* The constant_value of a uniform variable is its initializer,
717 * not the lifetime constant value of the uniform.
718 */
719 if (var->data.mode == ir_var_uniform)
720 return NULL;
721
722 if (!var->constant_value)
723 return NULL;
724
725 return var->constant_value->clone(ralloc_parent(var), NULL);
726 }
727
728
729 ir_constant *
constant_expression_value(struct hash_table * variable_context)730 ir_dereference_array::constant_expression_value(struct hash_table *variable_context)
731 {
732 ir_constant *array = this->array->constant_expression_value(variable_context);
733 ir_constant *idx = this->array_index->constant_expression_value(variable_context);
734
735 if ((array != NULL) && (idx != NULL)) {
736 void *ctx = ralloc_parent(this);
737 if (array->type->is_matrix()) {
738 /* Array access of a matrix results in a vector.
739 */
740 const unsigned column = idx->value.u[0];
741
742 const glsl_type *const column_type = array->type->column_type();
743
744 /* Offset in the constant matrix to the first element of the column
745 * to be extracted.
746 */
747 const unsigned mat_idx = column * column_type->vector_elements;
748
749 ir_constant_data data = { { 0 } };
750
751 switch (column_type->base_type) {
752 case GLSL_TYPE_UINT:
753 case GLSL_TYPE_INT:
754 for (unsigned i = 0; i < column_type->vector_elements; i++)
755 data.u[i] = array->value.u[mat_idx + i];
756
757 break;
758
759 case GLSL_TYPE_FLOAT:
760 for (unsigned i = 0; i < column_type->vector_elements; i++)
761 data.f[i] = array->value.f[mat_idx + i];
762
763 break;
764
765 case GLSL_TYPE_DOUBLE:
766 for (unsigned i = 0; i < column_type->vector_elements; i++)
767 data.d[i] = array->value.d[mat_idx + i];
768
769 break;
770
771 default:
772 assert(!"Should not get here.");
773 break;
774 }
775
776 return new(ctx) ir_constant(column_type, &data);
777 } else if (array->type->is_vector()) {
778 const unsigned component = idx->value.u[0];
779
780 return new(ctx) ir_constant(array, component);
781 } else {
782 const unsigned index = idx->value.u[0];
783 return array->get_array_element(index)->clone(ctx, NULL);
784 }
785 }
786 return NULL;
787 }
788
789
790 ir_constant *
constant_expression_value(struct hash_table *)791 ir_dereference_record::constant_expression_value(struct hash_table *)
792 {
793 ir_constant *v = this->record->constant_expression_value();
794
795 return (v != NULL) ? v->get_record_field(this->field) : NULL;
796 }
797
798
799 ir_constant *
constant_expression_value(struct hash_table *)800 ir_assignment::constant_expression_value(struct hash_table *)
801 {
802 /* FINISHME: Handle CEs involving assignment (return RHS) */
803 return NULL;
804 }
805
806
807 ir_constant *
constant_expression_value(struct hash_table *)808 ir_constant::constant_expression_value(struct hash_table *)
809 {
810 return this;
811 }
812
813
814 ir_constant *
constant_expression_value(struct hash_table * variable_context)815 ir_call::constant_expression_value(struct hash_table *variable_context)
816 {
817 return this->callee->constant_expression_value(&this->actual_parameters, variable_context);
818 }
819
820
constant_expression_evaluate_expression_list(const struct exec_list & body,struct hash_table * variable_context,ir_constant ** result)821 bool ir_function_signature::constant_expression_evaluate_expression_list(const struct exec_list &body,
822 struct hash_table *variable_context,
823 ir_constant **result)
824 {
825 foreach_in_list(ir_instruction, inst, &body) {
826 switch(inst->ir_type) {
827
828 /* (declare () type symbol) */
829 case ir_type_variable: {
830 ir_variable *var = inst->as_variable();
831 _mesa_hash_table_insert(variable_context, var, ir_constant::zero(this, var->type));
832 break;
833 }
834
835 /* (assign [condition] (write-mask) (ref) (value)) */
836 case ir_type_assignment: {
837 ir_assignment *asg = inst->as_assignment();
838 if (asg->condition) {
839 ir_constant *cond = asg->condition->constant_expression_value(variable_context);
840 if (!cond)
841 return false;
842 if (!cond->get_bool_component(0))
843 break;
844 }
845
846 ir_constant *store = NULL;
847 int offset = 0;
848
849 if (!constant_referenced(asg->lhs, variable_context, store, offset))
850 return false;
851
852 ir_constant *value = asg->rhs->constant_expression_value(variable_context);
853
854 if (!value)
855 return false;
856
857 store->copy_masked_offset(value, offset, asg->write_mask);
858 break;
859 }
860
861 /* (return (expression)) */
862 case ir_type_return:
863 assert (result);
864 *result = inst->as_return()->value->constant_expression_value(variable_context);
865 return *result != NULL;
866
867 /* (call name (ref) (params))*/
868 case ir_type_call: {
869 ir_call *call = inst->as_call();
870
871 /* Just say no to void functions in constant expressions. We
872 * don't need them at that point.
873 */
874
875 if (!call->return_deref)
876 return false;
877
878 ir_constant *store = NULL;
879 int offset = 0;
880
881 if (!constant_referenced(call->return_deref, variable_context,
882 store, offset))
883 return false;
884
885 ir_constant *value = call->constant_expression_value(variable_context);
886
887 if(!value)
888 return false;
889
890 store->copy_offset(value, offset);
891 break;
892 }
893
894 /* (if condition (then-instructions) (else-instructions)) */
895 case ir_type_if: {
896 ir_if *iif = inst->as_if();
897
898 ir_constant *cond = iif->condition->constant_expression_value(variable_context);
899 if (!cond || !cond->type->is_boolean())
900 return false;
901
902 exec_list &branch = cond->get_bool_component(0) ? iif->then_instructions : iif->else_instructions;
903
904 *result = NULL;
905 if (!constant_expression_evaluate_expression_list(branch, variable_context, result))
906 return false;
907
908 /* If there was a return in the branch chosen, drop out now. */
909 if (*result)
910 return true;
911
912 break;
913 }
914
915 /* Every other expression type, we drop out. */
916 default:
917 return false;
918 }
919 }
920
921 /* Reaching the end of the block is not an error condition */
922 if (result)
923 *result = NULL;
924
925 return true;
926 }
927
928 ir_constant *
constant_expression_value(exec_list * actual_parameters,struct hash_table * variable_context)929 ir_function_signature::constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context)
930 {
931 const glsl_type *type = this->return_type;
932 if (type == glsl_type::void_type)
933 return NULL;
934
935 /* From the GLSL 1.20 spec, page 23:
936 * "Function calls to user-defined functions (non-built-in functions)
937 * cannot be used to form constant expressions."
938 */
939 if (!this->is_builtin())
940 return NULL;
941
942 /*
943 * Of the builtin functions, only the texture lookups and the noise
944 * ones must not be used in constant expressions. They all include
945 * specific opcodes so they don't need to be special-cased at this
946 * point.
947 */
948
949 /* Initialize the table of dereferencable names with the function
950 * parameters. Verify their const-ness on the way.
951 *
952 * We expect the correctness of the number of parameters to have
953 * been checked earlier.
954 */
955 hash_table *deref_hash = _mesa_hash_table_create(NULL, _mesa_hash_pointer,
956 _mesa_key_pointer_equal);
957
958 /* If "origin" is non-NULL, then the function body is there. So we
959 * have to use the variable objects from the object with the body,
960 * but the parameter instanciation on the current object.
961 */
962 const exec_node *parameter_info = origin ? origin->parameters.get_head_raw() : parameters.get_head_raw();
963
964 foreach_in_list(ir_rvalue, n, actual_parameters) {
965 ir_constant *constant = n->constant_expression_value(variable_context);
966 if (constant == NULL) {
967 _mesa_hash_table_destroy(deref_hash, NULL);
968 return NULL;
969 }
970
971
972 ir_variable *var = (ir_variable *)parameter_info;
973 _mesa_hash_table_insert(deref_hash, var, constant);
974
975 parameter_info = parameter_info->next;
976 }
977
978 ir_constant *result = NULL;
979
980 /* Now run the builtin function until something non-constant
981 * happens or we get the result.
982 */
983 if (constant_expression_evaluate_expression_list(origin ? origin->body : body, deref_hash, &result) && result)
984 result = result->clone(ralloc_parent(this), NULL);
985
986 _mesa_hash_table_destroy(deref_hash, NULL);
987
988 return result;
989 }
990