2 * Copyright © 2010 Intel Corporation
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:
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
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.
25 * \file ir_constant_expression.cpp
26 * Evaluate and process constant valued expressions
28 * In GLSL, constant valued expressions are used in several places. These
29 * must be processed and evaluated very early in the compilation process.
32 * * Initializers for uniforms
33 * * Initializers for \c const variables
37 #include "main/core.h" /* for MAX2, MIN2, CLAMP */
39 #include "ir_visitor.h"
40 #include "glsl_types.h"
41 #include "program/hash_table.h"
43 #if defined(_MSC_VER) && (_MSC_VER < 1800)
44 static int isnormal(double x
)
46 return _fpclass(x
) == _FPCLASS_NN
|| _fpclass(x
) == _FPCLASS_PN
;
48 #elif defined(__SUNPRO_CC)
50 static int isnormal(double x
)
52 return fpclass(x
) == FP_NORMAL
;
57 static double copysign(double x
, double y
)
59 return _copysign(x
, y
);
64 dot(ir_constant
*op0
, ir_constant
*op1
)
66 assert(op0
->type
->is_float() && op1
->type
->is_float());
69 for (unsigned c
= 0; c
< op0
->type
->components(); c
++)
70 result
+= op0
->value
.f
[c
] * op1
->value
.f
[c
];
75 /* This method is the only one supported by gcc. Unions in particular
76 * are iffy, and read-through-converted-pointer is killed by strict
77 * aliasing. OTOH, the compiler sees through the memcpy, so the
78 * resulting asm is reasonable.
81 bitcast_u2f(unsigned int u
)
83 assert(sizeof(float) == sizeof(unsigned int));
85 memcpy(&f
, &u
, sizeof(f
));
92 assert(sizeof(float) == sizeof(unsigned int));
94 memcpy(&u
, &f
, sizeof(f
));
99 * Evaluate one component of a floating-point 4x8 unpacking function.
102 (*pack_1x8_func_t
)(float);
105 * Evaluate one component of a floating-point 2x16 unpacking function.
108 (*pack_1x16_func_t
)(float);
111 * Evaluate one component of a floating-point 4x8 unpacking function.
114 (*unpack_1x8_func_t
)(uint8_t);
117 * Evaluate one component of a floating-point 2x16 unpacking function.
120 (*unpack_1x16_func_t
)(uint16_t);
123 * Evaluate a 2x16 floating-point packing function.
126 pack_2x16(pack_1x16_func_t pack_1x16
,
129 /* From section 8.4 of the GLSL ES 3.00 spec:
133 * The first component of the vector will be written to the least
134 * significant bits of the output; the last component will be written to
135 * the most significant bits.
137 * The specifications for the other packing functions contain similar
141 u
|= ((uint32_t) pack_1x16(x
) << 0);
142 u
|= ((uint32_t) pack_1x16(y
) << 16);
147 * Evaluate a 4x8 floating-point packing function.
150 pack_4x8(pack_1x8_func_t pack_1x8
,
151 float x
, float y
, float z
, float w
)
153 /* From section 8.4 of the GLSL 4.30 spec:
157 * The first component of the vector will be written to the least
158 * significant bits of the output; the last component will be written to
159 * the most significant bits.
161 * The specifications for the other packing functions contain similar
165 u
|= ((uint32_t) pack_1x8(x
) << 0);
166 u
|= ((uint32_t) pack_1x8(y
) << 8);
167 u
|= ((uint32_t) pack_1x8(z
) << 16);
168 u
|= ((uint32_t) pack_1x8(w
) << 24);
173 * Evaluate a 2x16 floating-point unpacking function.
176 unpack_2x16(unpack_1x16_func_t unpack_1x16
,
180 /* From section 8.4 of the GLSL ES 3.00 spec:
184 * The first component of the returned vector will be extracted from
185 * the least significant bits of the input; the last component will be
186 * extracted from the most significant bits.
188 * The specifications for the other unpacking functions contain similar
191 *x
= unpack_1x16((uint16_t) (u
& 0xffff));
192 *y
= unpack_1x16((uint16_t) (u
>> 16));
196 * Evaluate a 4x8 floating-point unpacking function.
199 unpack_4x8(unpack_1x8_func_t unpack_1x8
, uint32_t u
,
200 float *x
, float *y
, float *z
, float *w
)
202 /* From section 8.4 of the GLSL 4.30 spec:
206 * The first component of the returned vector will be extracted from
207 * the least significant bits of the input; the last component will be
208 * extracted from the most significant bits.
210 * The specifications for the other unpacking functions contain similar
213 *x
= unpack_1x8((uint8_t) (u
& 0xff));
214 *y
= unpack_1x8((uint8_t) (u
>> 8));
215 *z
= unpack_1x8((uint8_t) (u
>> 16));
216 *w
= unpack_1x8((uint8_t) (u
>> 24));
220 * Evaluate one component of packSnorm4x8.
223 pack_snorm_1x8(float x
)
225 /* From section 8.4 of the GLSL 4.30 spec:
229 * The conversion for component c of v to fixed point is done as
232 * packSnorm4x8: round(clamp(c, -1, +1) * 127.0)
234 * We must first cast the float to an int, because casting a negative
235 * float to a uint is undefined.
237 return (uint8_t) (int8_t)
238 _mesa_round_to_even(CLAMP(x
, -1.0f
, +1.0f
) * 127.0f
);
242 * Evaluate one component of packSnorm2x16.
245 pack_snorm_1x16(float x
)
247 /* From section 8.4 of the GLSL ES 3.00 spec:
251 * The conversion for component c of v to fixed point is done as
254 * packSnorm2x16: round(clamp(c, -1, +1) * 32767.0)
256 * We must first cast the float to an int, because casting a negative
257 * float to a uint is undefined.
259 return (uint16_t) (int16_t)
260 _mesa_round_to_even(CLAMP(x
, -1.0f
, +1.0f
) * 32767.0f
);
264 * Evaluate one component of unpackSnorm4x8.
267 unpack_snorm_1x8(uint8_t u
)
269 /* From section 8.4 of the GLSL 4.30 spec:
273 * The conversion for unpacked fixed-point value f to floating point is
276 * unpackSnorm4x8: clamp(f / 127.0, -1, +1)
278 return CLAMP((int8_t) u
/ 127.0f
, -1.0f
, +1.0f
);
282 * Evaluate one component of unpackSnorm2x16.
285 unpack_snorm_1x16(uint16_t u
)
287 /* From section 8.4 of the GLSL ES 3.00 spec:
291 * The conversion for unpacked fixed-point value f to floating point is
294 * unpackSnorm2x16: clamp(f / 32767.0, -1, +1)
296 return CLAMP((int16_t) u
/ 32767.0f
, -1.0f
, +1.0f
);
300 * Evaluate one component packUnorm4x8.
303 pack_unorm_1x8(float x
)
305 /* From section 8.4 of the GLSL 4.30 spec:
309 * The conversion for component c of v to fixed point is done as
312 * packUnorm4x8: round(clamp(c, 0, +1) * 255.0)
314 return (uint8_t) _mesa_round_to_even(CLAMP(x
, 0.0f
, 1.0f
) * 255.0f
);
318 * Evaluate one component packUnorm2x16.
321 pack_unorm_1x16(float x
)
323 /* From section 8.4 of the GLSL ES 3.00 spec:
327 * The conversion for component c of v to fixed point is done as
330 * packUnorm2x16: round(clamp(c, 0, +1) * 65535.0)
332 return (uint16_t) _mesa_round_to_even(CLAMP(x
, 0.0f
, 1.0f
) * 65535.0f
);
336 * Evaluate one component of unpackUnorm4x8.
339 unpack_unorm_1x8(uint8_t u
)
341 /* From section 8.4 of the GLSL 4.30 spec:
345 * The conversion for unpacked fixed-point value f to floating point is
348 * unpackUnorm4x8: f / 255.0
350 return (float) u
/ 255.0f
;
354 * Evaluate one component of unpackUnorm2x16.
357 unpack_unorm_1x16(uint16_t u
)
359 /* From section 8.4 of the GLSL ES 3.00 spec:
363 * The conversion for unpacked fixed-point value f to floating point is
366 * unpackUnorm2x16: f / 65535.0
368 return (float) u
/ 65535.0f
;
372 * Evaluate one component of packHalf2x16.
375 pack_half_1x16(float x
)
377 return _mesa_float_to_half(x
);
381 * Evaluate one component of unpackHalf2x16.
384 unpack_half_1x16(uint16_t u
)
386 return _mesa_half_to_float(u
);
390 ir_rvalue::constant_expression_value(struct hash_table
*variable_context
)
392 assert(this->type
->is_error());
397 ir_expression::constant_expression_value(struct hash_table
*variable_context
)
399 if (this->type
->is_error())
402 ir_constant
*op
[Elements(this->operands
)] = { NULL
, };
403 ir_constant_data data
;
405 memset(&data
, 0, sizeof(data
));
407 for (unsigned operand
= 0; operand
< this->get_num_operands(); operand
++) {
408 op
[operand
] = this->operands
[operand
]->constant_expression_value(variable_context
);
414 switch (this->operation
) {
415 case ir_binop_lshift
:
416 case ir_binop_rshift
:
418 case ir_binop_vector_extract
:
420 case ir_triop_bitfield_extract
:
424 assert(op
[0]->type
->base_type
== op
[1]->type
->base_type
);
428 bool op0_scalar
= op
[0]->type
->is_scalar();
429 bool op1_scalar
= op
[1] != NULL
&& op
[1]->type
->is_scalar();
431 /* When iterating over a vector or matrix's components, we want to increase
432 * the loop counter. However, for scalars, we want to stay at 0.
434 unsigned c0_inc
= op0_scalar
? 0 : 1;
435 unsigned c1_inc
= op1_scalar
? 0 : 1;
437 if (op1_scalar
|| !op
[1]) {
438 components
= op
[0]->type
->components();
440 components
= op
[1]->type
->components();
443 void *ctx
= ralloc_parent(this);
445 /* Handle array operations here, rather than below. */
446 if (op
[0]->type
->is_array()) {
447 assert(op
[1] != NULL
&& op
[1]->type
->is_array());
448 switch (this->operation
) {
449 case ir_binop_all_equal
:
450 return new(ctx
) ir_constant(op
[0]->has_value(op
[1]));
451 case ir_binop_any_nequal
:
452 return new(ctx
) ir_constant(!op
[0]->has_value(op
[1]));
459 switch (this->operation
) {
460 case ir_unop_bit_not
:
461 switch (op
[0]->type
->base_type
) {
463 for (unsigned c
= 0; c
< components
; c
++)
464 data
.i
[c
] = ~ op
[0]->value
.i
[c
];
467 for (unsigned c
= 0; c
< components
; c
++)
468 data
.u
[c
] = ~ op
[0]->value
.u
[c
];
475 case ir_unop_logic_not
:
476 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
477 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++)
478 data
.b
[c
] = !op
[0]->value
.b
[c
];
482 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
483 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
484 data
.i
[c
] = (int) op
[0]->value
.f
[c
];
488 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
489 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
490 data
.i
[c
] = (unsigned) op
[0]->value
.f
[c
];
494 assert(op
[0]->type
->base_type
== GLSL_TYPE_INT
);
495 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
496 data
.f
[c
] = (float) op
[0]->value
.i
[c
];
500 assert(op
[0]->type
->base_type
== GLSL_TYPE_UINT
);
501 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
502 data
.f
[c
] = (float) op
[0]->value
.u
[c
];
506 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
507 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
508 data
.f
[c
] = op
[0]->value
.b
[c
] ? 1.0F
: 0.0F
;
512 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
513 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
514 data
.b
[c
] = op
[0]->value
.f
[c
] != 0.0F
? true : false;
518 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
519 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
520 data
.u
[c
] = op
[0]->value
.b
[c
] ? 1 : 0;
524 assert(op
[0]->type
->is_integer());
525 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
526 data
.b
[c
] = op
[0]->value
.u
[c
] ? true : false;
530 assert(op
[0]->type
->base_type
== GLSL_TYPE_UINT
);
531 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
532 data
.i
[c
] = op
[0]->value
.u
[c
];
536 assert(op
[0]->type
->base_type
== GLSL_TYPE_INT
);
537 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
538 data
.u
[c
] = op
[0]->value
.i
[c
];
541 case ir_unop_bitcast_i2f
:
542 assert(op
[0]->type
->base_type
== GLSL_TYPE_INT
);
543 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
544 data
.f
[c
] = bitcast_u2f(op
[0]->value
.i
[c
]);
547 case ir_unop_bitcast_f2i
:
548 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
549 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
550 data
.i
[c
] = bitcast_f2u(op
[0]->value
.f
[c
]);
553 case ir_unop_bitcast_u2f
:
554 assert(op
[0]->type
->base_type
== GLSL_TYPE_UINT
);
555 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
556 data
.f
[c
] = bitcast_u2f(op
[0]->value
.u
[c
]);
559 case ir_unop_bitcast_f2u
:
560 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
561 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
562 data
.u
[c
] = bitcast_f2u(op
[0]->value
.f
[c
]);
566 assert(op
[0]->type
->is_boolean());
568 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
569 if (op
[0]->value
.b
[c
])
575 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
576 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
577 data
.f
[c
] = truncf(op
[0]->value
.f
[c
]);
581 case ir_unop_round_even
:
582 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
583 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
584 data
.f
[c
] = _mesa_round_to_even(op
[0]->value
.f
[c
]);
589 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
590 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
591 data
.f
[c
] = ceilf(op
[0]->value
.f
[c
]);
596 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
597 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
598 data
.f
[c
] = floorf(op
[0]->value
.f
[c
]);
603 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
604 switch (this->type
->base_type
) {
611 case GLSL_TYPE_FLOAT
:
612 data
.f
[c
] = op
[0]->value
.f
[c
] - floor(op
[0]->value
.f
[c
]);
621 case ir_unop_sin_reduced
:
622 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
623 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
624 data
.f
[c
] = sinf(op
[0]->value
.f
[c
]);
629 case ir_unop_cos_reduced
:
630 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
631 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
632 data
.f
[c
] = cosf(op
[0]->value
.f
[c
]);
637 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
638 switch (this->type
->base_type
) {
640 data
.u
[c
] = -((int) op
[0]->value
.u
[c
]);
643 data
.i
[c
] = -op
[0]->value
.i
[c
];
645 case GLSL_TYPE_FLOAT
:
646 data
.f
[c
] = -op
[0]->value
.f
[c
];
655 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
656 switch (this->type
->base_type
) {
658 data
.u
[c
] = op
[0]->value
.u
[c
];
661 data
.i
[c
] = op
[0]->value
.i
[c
];
663 data
.i
[c
] = -data
.i
[c
];
665 case GLSL_TYPE_FLOAT
:
666 data
.f
[c
] = fabs(op
[0]->value
.f
[c
]);
675 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
676 switch (this->type
->base_type
) {
678 data
.u
[c
] = op
[0]->value
.i
[c
] > 0;
681 data
.i
[c
] = (op
[0]->value
.i
[c
] > 0) - (op
[0]->value
.i
[c
] < 0);
683 case GLSL_TYPE_FLOAT
:
684 data
.f
[c
] = float((op
[0]->value
.f
[c
] > 0)-(op
[0]->value
.f
[c
] < 0));
693 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
694 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
695 switch (this->type
->base_type
) {
697 if (op
[0]->value
.u
[c
] != 0.0)
698 data
.u
[c
] = 1 / op
[0]->value
.u
[c
];
701 if (op
[0]->value
.i
[c
] != 0.0)
702 data
.i
[c
] = 1 / op
[0]->value
.i
[c
];
704 case GLSL_TYPE_FLOAT
:
705 if (op
[0]->value
.f
[c
] != 0.0)
706 data
.f
[c
] = 1.0F
/ op
[0]->value
.f
[c
];
715 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
716 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
717 data
.f
[c
] = 1.0F
/ sqrtf(op
[0]->value
.f
[c
]);
722 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
723 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
724 data
.f
[c
] = sqrtf(op
[0]->value
.f
[c
]);
729 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
730 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
731 data
.f
[c
] = expf(op
[0]->value
.f
[c
]);
736 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
737 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
738 data
.f
[c
] = exp2f(op
[0]->value
.f
[c
]);
743 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
744 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
745 data
.f
[c
] = logf(op
[0]->value
.f
[c
]);
750 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
751 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
752 data
.f
[c
] = log2f(op
[0]->value
.f
[c
]);
758 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
759 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
764 case ir_unop_pack_snorm_2x16
:
765 assert(op
[0]->type
== glsl_type::vec2_type
);
766 data
.u
[0] = pack_2x16(pack_snorm_1x16
,
770 case ir_unop_pack_snorm_4x8
:
771 assert(op
[0]->type
== glsl_type::vec4_type
);
772 data
.u
[0] = pack_4x8(pack_snorm_1x8
,
778 case ir_unop_unpack_snorm_2x16
:
779 assert(op
[0]->type
== glsl_type::uint_type
);
780 unpack_2x16(unpack_snorm_1x16
,
782 &data
.f
[0], &data
.f
[1]);
784 case ir_unop_unpack_snorm_4x8
:
785 assert(op
[0]->type
== glsl_type::uint_type
);
786 unpack_4x8(unpack_snorm_1x8
,
788 &data
.f
[0], &data
.f
[1], &data
.f
[2], &data
.f
[3]);
790 case ir_unop_pack_unorm_2x16
:
791 assert(op
[0]->type
== glsl_type::vec2_type
);
792 data
.u
[0] = pack_2x16(pack_unorm_1x16
,
796 case ir_unop_pack_unorm_4x8
:
797 assert(op
[0]->type
== glsl_type::vec4_type
);
798 data
.u
[0] = pack_4x8(pack_unorm_1x8
,
804 case ir_unop_unpack_unorm_2x16
:
805 assert(op
[0]->type
== glsl_type::uint_type
);
806 unpack_2x16(unpack_unorm_1x16
,
808 &data
.f
[0], &data
.f
[1]);
810 case ir_unop_unpack_unorm_4x8
:
811 assert(op
[0]->type
== glsl_type::uint_type
);
812 unpack_4x8(unpack_unorm_1x8
,
814 &data
.f
[0], &data
.f
[1], &data
.f
[2], &data
.f
[3]);
816 case ir_unop_pack_half_2x16
:
817 assert(op
[0]->type
== glsl_type::vec2_type
);
818 data
.u
[0] = pack_2x16(pack_half_1x16
,
822 case ir_unop_unpack_half_2x16
:
823 assert(op
[0]->type
== glsl_type::uint_type
);
824 unpack_2x16(unpack_half_1x16
,
826 &data
.f
[0], &data
.f
[1]);
829 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
830 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
831 data
.f
[c
] = powf(op
[0]->value
.f
[c
], op
[1]->value
.f
[c
]);
836 data
.f
[0] = dot(op
[0], op
[1]);
840 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
841 for (unsigned c
= 0, c0
= 0, c1
= 0;
843 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
845 switch (op
[0]->type
->base_type
) {
847 data
.u
[c
] = MIN2(op
[0]->value
.u
[c0
], op
[1]->value
.u
[c1
]);
850 data
.i
[c
] = MIN2(op
[0]->value
.i
[c0
], op
[1]->value
.i
[c1
]);
852 case GLSL_TYPE_FLOAT
:
853 data
.f
[c
] = MIN2(op
[0]->value
.f
[c0
], op
[1]->value
.f
[c1
]);
862 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
863 for (unsigned c
= 0, c0
= 0, c1
= 0;
865 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
867 switch (op
[0]->type
->base_type
) {
869 data
.u
[c
] = MAX2(op
[0]->value
.u
[c0
], op
[1]->value
.u
[c1
]);
872 data
.i
[c
] = MAX2(op
[0]->value
.i
[c0
], op
[1]->value
.i
[c1
]);
874 case GLSL_TYPE_FLOAT
:
875 data
.f
[c
] = MAX2(op
[0]->value
.f
[c0
], op
[1]->value
.f
[c1
]);
884 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
885 for (unsigned c
= 0, c0
= 0, c1
= 0;
887 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
889 switch (op
[0]->type
->base_type
) {
891 data
.u
[c
] = op
[0]->value
.u
[c0
] + op
[1]->value
.u
[c1
];
894 data
.i
[c
] = op
[0]->value
.i
[c0
] + op
[1]->value
.i
[c1
];
896 case GLSL_TYPE_FLOAT
:
897 data
.f
[c
] = op
[0]->value
.f
[c0
] + op
[1]->value
.f
[c1
];
906 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
907 for (unsigned c
= 0, c0
= 0, c1
= 0;
909 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
911 switch (op
[0]->type
->base_type
) {
913 data
.u
[c
] = op
[0]->value
.u
[c0
] - op
[1]->value
.u
[c1
];
916 data
.i
[c
] = op
[0]->value
.i
[c0
] - op
[1]->value
.i
[c1
];
918 case GLSL_TYPE_FLOAT
:
919 data
.f
[c
] = op
[0]->value
.f
[c0
] - op
[1]->value
.f
[c1
];
928 /* Check for equal types, or unequal types involving scalars */
929 if ((op
[0]->type
== op
[1]->type
&& !op
[0]->type
->is_matrix())
930 || op0_scalar
|| op1_scalar
) {
931 for (unsigned c
= 0, c0
= 0, c1
= 0;
933 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
935 switch (op
[0]->type
->base_type
) {
937 data
.u
[c
] = op
[0]->value
.u
[c0
] * op
[1]->value
.u
[c1
];
940 data
.i
[c
] = op
[0]->value
.i
[c0
] * op
[1]->value
.i
[c1
];
942 case GLSL_TYPE_FLOAT
:
943 data
.f
[c
] = op
[0]->value
.f
[c0
] * op
[1]->value
.f
[c1
];
950 assert(op
[0]->type
->is_matrix() || op
[1]->type
->is_matrix());
952 /* Multiply an N-by-M matrix with an M-by-P matrix. Since either
953 * matrix can be a GLSL vector, either N or P can be 1.
955 * For vec*mat, the vector is treated as a row vector. This
956 * means the vector is a 1-row x M-column matrix.
958 * For mat*vec, the vector is treated as a column vector. Since
959 * matrix_columns is 1 for vectors, this just works.
961 const unsigned n
= op
[0]->type
->is_vector()
962 ? 1 : op
[0]->type
->vector_elements
;
963 const unsigned m
= op
[1]->type
->vector_elements
;
964 const unsigned p
= op
[1]->type
->matrix_columns
;
965 for (unsigned j
= 0; j
< p
; j
++) {
966 for (unsigned i
= 0; i
< n
; i
++) {
967 for (unsigned k
= 0; k
< m
; k
++) {
968 data
.f
[i
+n
*j
] += op
[0]->value
.f
[i
+n
*k
]*op
[1]->value
.f
[k
+m
*j
];
976 /* FINISHME: Emit warning when division-by-zero is detected. */
977 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
978 for (unsigned c
= 0, c0
= 0, c1
= 0;
980 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
982 switch (op
[0]->type
->base_type
) {
984 if (op
[1]->value
.u
[c1
] == 0) {
987 data
.u
[c
] = op
[0]->value
.u
[c0
] / op
[1]->value
.u
[c1
];
991 if (op
[1]->value
.i
[c1
] == 0) {
994 data
.i
[c
] = op
[0]->value
.i
[c0
] / op
[1]->value
.i
[c1
];
997 case GLSL_TYPE_FLOAT
:
998 data
.f
[c
] = op
[0]->value
.f
[c0
] / op
[1]->value
.f
[c1
];
1007 /* FINISHME: Emit warning when division-by-zero is detected. */
1008 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
1009 for (unsigned c
= 0, c0
= 0, c1
= 0;
1011 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1013 switch (op
[0]->type
->base_type
) {
1014 case GLSL_TYPE_UINT
:
1015 if (op
[1]->value
.u
[c1
] == 0) {
1018 data
.u
[c
] = op
[0]->value
.u
[c0
] % op
[1]->value
.u
[c1
];
1022 if (op
[1]->value
.i
[c1
] == 0) {
1025 data
.i
[c
] = op
[0]->value
.i
[c0
] % op
[1]->value
.i
[c1
];
1028 case GLSL_TYPE_FLOAT
:
1029 /* We don't use fmod because it rounds toward zero; GLSL specifies
1032 data
.f
[c
] = op
[0]->value
.f
[c0
] - op
[1]->value
.f
[c1
]
1033 * floorf(op
[0]->value
.f
[c0
] / op
[1]->value
.f
[c1
]);
1042 case ir_binop_logic_and
:
1043 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
1044 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++)
1045 data
.b
[c
] = op
[0]->value
.b
[c
] && op
[1]->value
.b
[c
];
1047 case ir_binop_logic_xor
:
1048 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
1049 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++)
1050 data
.b
[c
] = op
[0]->value
.b
[c
] ^ op
[1]->value
.b
[c
];
1052 case ir_binop_logic_or
:
1053 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
1054 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++)
1055 data
.b
[c
] = op
[0]->value
.b
[c
] || op
[1]->value
.b
[c
];
1059 assert(op
[0]->type
== op
[1]->type
);
1060 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
1061 switch (op
[0]->type
->base_type
) {
1062 case GLSL_TYPE_UINT
:
1063 data
.b
[c
] = op
[0]->value
.u
[c
] < op
[1]->value
.u
[c
];
1066 data
.b
[c
] = op
[0]->value
.i
[c
] < op
[1]->value
.i
[c
];
1068 case GLSL_TYPE_FLOAT
:
1069 data
.b
[c
] = op
[0]->value
.f
[c
] < op
[1]->value
.f
[c
];
1076 case ir_binop_greater
:
1077 assert(op
[0]->type
== op
[1]->type
);
1078 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
1079 switch (op
[0]->type
->base_type
) {
1080 case GLSL_TYPE_UINT
:
1081 data
.b
[c
] = op
[0]->value
.u
[c
] > op
[1]->value
.u
[c
];
1084 data
.b
[c
] = op
[0]->value
.i
[c
] > op
[1]->value
.i
[c
];
1086 case GLSL_TYPE_FLOAT
:
1087 data
.b
[c
] = op
[0]->value
.f
[c
] > op
[1]->value
.f
[c
];
1094 case ir_binop_lequal
:
1095 assert(op
[0]->type
== op
[1]->type
);
1096 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
1097 switch (op
[0]->type
->base_type
) {
1098 case GLSL_TYPE_UINT
:
1099 data
.b
[c
] = op
[0]->value
.u
[c
] <= op
[1]->value
.u
[c
];
1102 data
.b
[c
] = op
[0]->value
.i
[c
] <= op
[1]->value
.i
[c
];
1104 case GLSL_TYPE_FLOAT
:
1105 data
.b
[c
] = op
[0]->value
.f
[c
] <= op
[1]->value
.f
[c
];
1112 case ir_binop_gequal
:
1113 assert(op
[0]->type
== op
[1]->type
);
1114 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
1115 switch (op
[0]->type
->base_type
) {
1116 case GLSL_TYPE_UINT
:
1117 data
.b
[c
] = op
[0]->value
.u
[c
] >= op
[1]->value
.u
[c
];
1120 data
.b
[c
] = op
[0]->value
.i
[c
] >= op
[1]->value
.i
[c
];
1122 case GLSL_TYPE_FLOAT
:
1123 data
.b
[c
] = op
[0]->value
.f
[c
] >= op
[1]->value
.f
[c
];
1130 case ir_binop_equal
:
1131 assert(op
[0]->type
== op
[1]->type
);
1132 for (unsigned c
= 0; c
< components
; c
++) {
1133 switch (op
[0]->type
->base_type
) {
1134 case GLSL_TYPE_UINT
:
1135 data
.b
[c
] = op
[0]->value
.u
[c
] == op
[1]->value
.u
[c
];
1138 data
.b
[c
] = op
[0]->value
.i
[c
] == op
[1]->value
.i
[c
];
1140 case GLSL_TYPE_FLOAT
:
1141 data
.b
[c
] = op
[0]->value
.f
[c
] == op
[1]->value
.f
[c
];
1143 case GLSL_TYPE_BOOL
:
1144 data
.b
[c
] = op
[0]->value
.b
[c
] == op
[1]->value
.b
[c
];
1151 case ir_binop_nequal
:
1152 assert(op
[0]->type
== op
[1]->type
);
1153 for (unsigned c
= 0; c
< components
; c
++) {
1154 switch (op
[0]->type
->base_type
) {
1155 case GLSL_TYPE_UINT
:
1156 data
.b
[c
] = op
[0]->value
.u
[c
] != op
[1]->value
.u
[c
];
1159 data
.b
[c
] = op
[0]->value
.i
[c
] != op
[1]->value
.i
[c
];
1161 case GLSL_TYPE_FLOAT
:
1162 data
.b
[c
] = op
[0]->value
.f
[c
] != op
[1]->value
.f
[c
];
1164 case GLSL_TYPE_BOOL
:
1165 data
.b
[c
] = op
[0]->value
.b
[c
] != op
[1]->value
.b
[c
];
1172 case ir_binop_all_equal
:
1173 data
.b
[0] = op
[0]->has_value(op
[1]);
1175 case ir_binop_any_nequal
:
1176 data
.b
[0] = !op
[0]->has_value(op
[1]);
1179 case ir_binop_lshift
:
1180 for (unsigned c
= 0, c0
= 0, c1
= 0;
1182 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1184 if (op
[0]->type
->base_type
== GLSL_TYPE_INT
&&
1185 op
[1]->type
->base_type
== GLSL_TYPE_INT
) {
1186 data
.i
[c
] = op
[0]->value
.i
[c0
] << op
[1]->value
.i
[c1
];
1188 } else if (op
[0]->type
->base_type
== GLSL_TYPE_INT
&&
1189 op
[1]->type
->base_type
== GLSL_TYPE_UINT
) {
1190 data
.i
[c
] = op
[0]->value
.i
[c0
] << op
[1]->value
.u
[c1
];
1192 } else if (op
[0]->type
->base_type
== GLSL_TYPE_UINT
&&
1193 op
[1]->type
->base_type
== GLSL_TYPE_INT
) {
1194 data
.u
[c
] = op
[0]->value
.u
[c0
] << op
[1]->value
.i
[c1
];
1196 } else if (op
[0]->type
->base_type
== GLSL_TYPE_UINT
&&
1197 op
[1]->type
->base_type
== GLSL_TYPE_UINT
) {
1198 data
.u
[c
] = op
[0]->value
.u
[c0
] << op
[1]->value
.u
[c1
];
1203 case ir_binop_rshift
:
1204 for (unsigned c
= 0, c0
= 0, c1
= 0;
1206 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1208 if (op
[0]->type
->base_type
== GLSL_TYPE_INT
&&
1209 op
[1]->type
->base_type
== GLSL_TYPE_INT
) {
1210 data
.i
[c
] = op
[0]->value
.i
[c0
] >> op
[1]->value
.i
[c1
];
1212 } else if (op
[0]->type
->base_type
== GLSL_TYPE_INT
&&
1213 op
[1]->type
->base_type
== GLSL_TYPE_UINT
) {
1214 data
.i
[c
] = op
[0]->value
.i
[c0
] >> op
[1]->value
.u
[c1
];
1216 } else if (op
[0]->type
->base_type
== GLSL_TYPE_UINT
&&
1217 op
[1]->type
->base_type
== GLSL_TYPE_INT
) {
1218 data
.u
[c
] = op
[0]->value
.u
[c0
] >> op
[1]->value
.i
[c1
];
1220 } else if (op
[0]->type
->base_type
== GLSL_TYPE_UINT
&&
1221 op
[1]->type
->base_type
== GLSL_TYPE_UINT
) {
1222 data
.u
[c
] = op
[0]->value
.u
[c0
] >> op
[1]->value
.u
[c1
];
1227 case ir_binop_bit_and
:
1228 for (unsigned c
= 0, c0
= 0, c1
= 0;
1230 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1232 switch (op
[0]->type
->base_type
) {
1234 data
.i
[c
] = op
[0]->value
.i
[c0
] & op
[1]->value
.i
[c1
];
1236 case GLSL_TYPE_UINT
:
1237 data
.u
[c
] = op
[0]->value
.u
[c0
] & op
[1]->value
.u
[c1
];
1245 case ir_binop_bit_or
:
1246 for (unsigned c
= 0, c0
= 0, c1
= 0;
1248 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1250 switch (op
[0]->type
->base_type
) {
1252 data
.i
[c
] = op
[0]->value
.i
[c0
] | op
[1]->value
.i
[c1
];
1254 case GLSL_TYPE_UINT
:
1255 data
.u
[c
] = op
[0]->value
.u
[c0
] | op
[1]->value
.u
[c1
];
1263 case ir_binop_vector_extract
: {
1264 const int c
= CLAMP(op
[1]->value
.i
[0], 0,
1265 (int) op
[0]->type
->vector_elements
- 1);
1267 switch (op
[0]->type
->base_type
) {
1268 case GLSL_TYPE_UINT
:
1269 data
.u
[0] = op
[0]->value
.u
[c
];
1272 data
.i
[0] = op
[0]->value
.i
[c
];
1274 case GLSL_TYPE_FLOAT
:
1275 data
.f
[0] = op
[0]->value
.f
[c
];
1277 case GLSL_TYPE_BOOL
:
1278 data
.b
[0] = op
[0]->value
.b
[c
];
1286 case ir_binop_bit_xor
:
1287 for (unsigned c
= 0, c0
= 0, c1
= 0;
1289 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1291 switch (op
[0]->type
->base_type
) {
1293 data
.i
[c
] = op
[0]->value
.i
[c0
] ^ op
[1]->value
.i
[c1
];
1295 case GLSL_TYPE_UINT
:
1296 data
.u
[c
] = op
[0]->value
.u
[c0
] ^ op
[1]->value
.u
[c1
];
1304 case ir_unop_bitfield_reverse
:
1305 /* http://graphics.stanford.edu/~seander/bithacks.html#BitReverseObvious */
1306 for (unsigned c
= 0; c
< components
; c
++) {
1307 unsigned int v
= op
[0]->value
.u
[c
]; // input bits to be reversed
1308 unsigned int r
= v
; // r will be reversed bits of v; first get LSB of v
1309 int s
= sizeof(v
) * CHAR_BIT
- 1; // extra shift needed at end
1311 for (v
>>= 1; v
; v
>>= 1) {
1316 r
<<= s
; // shift when v's highest bits are zero
1322 case ir_unop_bit_count
:
1323 for (unsigned c
= 0; c
< components
; c
++) {
1325 unsigned v
= op
[0]->value
.u
[c
];
1327 for (; v
; count
++) {
1334 case ir_unop_find_msb
:
1335 for (unsigned c
= 0; c
< components
; c
++) {
1336 int v
= op
[0]->value
.i
[c
];
1338 if (v
== 0 || (op
[0]->type
->base_type
== GLSL_TYPE_INT
&& v
== -1))
1342 int top_bit
= op
[0]->type
->base_type
== GLSL_TYPE_UINT
1343 ? 0 : v
& (1 << 31);
1345 while (((v
& (1 << 31)) == top_bit
) && count
!= 32) {
1350 data
.i
[c
] = 31 - count
;
1355 case ir_unop_find_lsb
:
1356 for (unsigned c
= 0; c
< components
; c
++) {
1357 if (op
[0]->value
.i
[c
] == 0)
1361 unsigned v
= op
[0]->value
.u
[c
];
1363 for (; !(v
& 1); v
>>= 1) {
1371 case ir_triop_bitfield_extract
: {
1372 int offset
= op
[1]->value
.i
[0];
1373 int bits
= op
[2]->value
.i
[0];
1375 for (unsigned c
= 0; c
< components
; c
++) {
1378 else if (offset
< 0 || bits
< 0)
1379 data
.u
[c
] = 0; /* Undefined, per spec. */
1380 else if (offset
+ bits
> 32)
1381 data
.u
[c
] = 0; /* Undefined, per spec. */
1383 if (op
[0]->type
->base_type
== GLSL_TYPE_INT
) {
1384 /* int so that the right shift will sign-extend. */
1385 int value
= op
[0]->value
.i
[c
];
1386 value
<<= 32 - bits
- offset
;
1387 value
>>= 32 - bits
;
1390 unsigned value
= op
[0]->value
.u
[c
];
1391 value
<<= 32 - bits
- offset
;
1392 value
>>= 32 - bits
;
1400 case ir_binop_bfm
: {
1401 int bits
= op
[0]->value
.i
[0];
1402 int offset
= op
[1]->value
.i
[0];
1404 for (unsigned c
= 0; c
< components
; c
++) {
1406 data
.u
[c
] = op
[0]->value
.u
[c
];
1407 else if (offset
< 0 || bits
< 0)
1408 data
.u
[c
] = 0; /* Undefined for bitfieldInsert, per spec. */
1409 else if (offset
+ bits
> 32)
1410 data
.u
[c
] = 0; /* Undefined for bitfieldInsert, per spec. */
1412 data
.u
[c
] = ((1 << bits
) - 1) << offset
;
1417 case ir_binop_ldexp
:
1418 for (unsigned c
= 0; c
< components
; c
++) {
1419 data
.f
[c
] = ldexp(op
[0]->value
.f
[c
], op
[1]->value
.i
[c
]);
1420 /* Flush subnormal values to zero. */
1421 if (!isnormal(data
.f
[c
]))
1422 data
.f
[c
] = copysign(0.0f
, op
[0]->value
.f
[c
]);
1427 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
1428 assert(op
[1]->type
->base_type
== GLSL_TYPE_FLOAT
);
1429 assert(op
[2]->type
->base_type
== GLSL_TYPE_FLOAT
);
1431 for (unsigned c
= 0; c
< components
; c
++) {
1432 data
.f
[c
] = op
[0]->value
.f
[c
] * op
[1]->value
.f
[c
]
1433 + op
[2]->value
.f
[c
];
1437 case ir_triop_lrp
: {
1438 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
1439 assert(op
[1]->type
->base_type
== GLSL_TYPE_FLOAT
);
1440 assert(op
[2]->type
->base_type
== GLSL_TYPE_FLOAT
);
1442 unsigned c2_inc
= op
[2]->type
->is_scalar() ? 0 : 1;
1443 for (unsigned c
= 0, c2
= 0; c
< components
; c2
+= c2_inc
, c
++) {
1444 data
.f
[c
] = op
[0]->value
.f
[c
] * (1.0f
- op
[2]->value
.f
[c2
]) +
1445 (op
[1]->value
.f
[c
] * op
[2]->value
.f
[c2
]);
1451 for (unsigned c
= 0; c
< components
; c
++) {
1452 data
.u
[c
] = op
[0]->value
.b
[c
] ? op
[1]->value
.u
[c
]
1453 : op
[2]->value
.u
[c
];
1457 case ir_triop_vector_insert
: {
1458 const unsigned idx
= op
[2]->value
.u
[0];
1460 memcpy(&data
, &op
[0]->value
, sizeof(data
));
1462 switch (this->type
->base_type
) {
1464 data
.i
[idx
] = op
[1]->value
.i
[0];
1466 case GLSL_TYPE_UINT
:
1467 data
.u
[idx
] = op
[1]->value
.u
[0];
1469 case GLSL_TYPE_FLOAT
:
1470 data
.f
[idx
] = op
[1]->value
.f
[0];
1472 case GLSL_TYPE_BOOL
:
1473 data
.b
[idx
] = op
[1]->value
.b
[0];
1476 assert(!"Should not get here.");
1482 case ir_quadop_bitfield_insert
: {
1483 int offset
= op
[2]->value
.i
[0];
1484 int bits
= op
[3]->value
.i
[0];
1486 for (unsigned c
= 0; c
< components
; c
++) {
1488 data
.u
[c
] = op
[0]->value
.u
[c
];
1489 else if (offset
< 0 || bits
< 0)
1490 data
.u
[c
] = 0; /* Undefined, per spec. */
1491 else if (offset
+ bits
> 32)
1492 data
.u
[c
] = 0; /* Undefined, per spec. */
1494 unsigned insert_mask
= ((1 << bits
) - 1) << offset
;
1496 unsigned insert
= op
[1]->value
.u
[c
];
1498 insert
&= insert_mask
;
1500 unsigned base
= op
[0]->value
.u
[c
];
1501 base
&= ~insert_mask
;
1503 data
.u
[c
] = base
| insert
;
1509 case ir_quadop_vector
:
1510 for (unsigned c
= 0; c
< this->type
->vector_elements
; c
++) {
1511 switch (this->type
->base_type
) {
1513 data
.i
[c
] = op
[c
]->value
.i
[0];
1515 case GLSL_TYPE_UINT
:
1516 data
.u
[c
] = op
[c
]->value
.u
[0];
1518 case GLSL_TYPE_FLOAT
:
1519 data
.f
[c
] = op
[c
]->value
.f
[0];
1528 /* FINISHME: Should handle all expression types. */
1532 return new(ctx
) ir_constant(this->type
, &data
);
1537 ir_texture::constant_expression_value(struct hash_table
*variable_context
)
1539 /* texture lookups aren't constant expressions */
1545 ir_swizzle::constant_expression_value(struct hash_table
*variable_context
)
1547 ir_constant
*v
= this->val
->constant_expression_value(variable_context
);
1550 ir_constant_data data
= { { 0 } };
1552 const unsigned swiz_idx
[4] = {
1553 this->mask
.x
, this->mask
.y
, this->mask
.z
, this->mask
.w
1556 for (unsigned i
= 0; i
< this->mask
.num_components
; i
++) {
1557 switch (v
->type
->base_type
) {
1558 case GLSL_TYPE_UINT
:
1559 case GLSL_TYPE_INT
: data
.u
[i
] = v
->value
.u
[swiz_idx
[i
]]; break;
1560 case GLSL_TYPE_FLOAT
: data
.f
[i
] = v
->value
.f
[swiz_idx
[i
]]; break;
1561 case GLSL_TYPE_BOOL
: data
.b
[i
] = v
->value
.b
[swiz_idx
[i
]]; break;
1562 default: assert(!"Should not get here."); break;
1566 void *ctx
= ralloc_parent(this);
1567 return new(ctx
) ir_constant(this->type
, &data
);
1574 ir_dereference_variable::constant_referenced(struct hash_table
*variable_context
,
1575 ir_constant
*&store
, int &offset
) const
1577 if (variable_context
) {
1578 store
= (ir_constant
*)hash_table_find(variable_context
, var
);
1587 ir_dereference_variable::constant_expression_value(struct hash_table
*variable_context
)
1589 /* This may occur during compile and var->type is glsl_type::error_type */
1593 /* Give priority to the context hashtable, if it exists */
1594 if (variable_context
) {
1595 ir_constant
*value
= (ir_constant
*)hash_table_find(variable_context
, var
);
1600 /* The constant_value of a uniform variable is its initializer,
1601 * not the lifetime constant value of the uniform.
1603 if (var
->data
.mode
== ir_var_uniform
)
1606 if (!var
->constant_value
)
1609 return var
->constant_value
->clone(ralloc_parent(var
), NULL
);
1614 ir_dereference_array::constant_referenced(struct hash_table
*variable_context
,
1615 ir_constant
*&store
, int &offset
) const
1617 ir_constant
*index_c
= array_index
->constant_expression_value(variable_context
);
1619 if (!index_c
|| !index_c
->type
->is_scalar() || !index_c
->type
->is_integer()) {
1625 int index
= index_c
->type
->base_type
== GLSL_TYPE_INT
?
1626 index_c
->get_int_component(0) :
1627 index_c
->get_uint_component(0);
1629 ir_constant
*substore
;
1631 const ir_dereference
*deref
= array
->as_dereference();
1638 deref
->constant_referenced(variable_context
, substore
, suboffset
);
1646 const glsl_type
*vt
= array
->type
;
1647 if (vt
->is_array()) {
1648 store
= substore
->get_array_element(index
);
1652 if (vt
->is_matrix()) {
1654 offset
= index
* vt
->vector_elements
;
1657 if (vt
->is_vector()) {
1659 offset
= suboffset
+ index
;
1668 ir_dereference_array::constant_expression_value(struct hash_table
*variable_context
)
1670 ir_constant
*array
= this->array
->constant_expression_value(variable_context
);
1671 ir_constant
*idx
= this->array_index
->constant_expression_value(variable_context
);
1673 if ((array
!= NULL
) && (idx
!= NULL
)) {
1674 void *ctx
= ralloc_parent(this);
1675 if (array
->type
->is_matrix()) {
1676 /* Array access of a matrix results in a vector.
1678 const unsigned column
= idx
->value
.u
[0];
1680 const glsl_type
*const column_type
= array
->type
->column_type();
1682 /* Offset in the constant matrix to the first element of the column
1685 const unsigned mat_idx
= column
* column_type
->vector_elements
;
1687 ir_constant_data data
= { { 0 } };
1689 switch (column_type
->base_type
) {
1690 case GLSL_TYPE_UINT
:
1692 for (unsigned i
= 0; i
< column_type
->vector_elements
; i
++)
1693 data
.u
[i
] = array
->value
.u
[mat_idx
+ i
];
1697 case GLSL_TYPE_FLOAT
:
1698 for (unsigned i
= 0; i
< column_type
->vector_elements
; i
++)
1699 data
.f
[i
] = array
->value
.f
[mat_idx
+ i
];
1704 assert(!"Should not get here.");
1708 return new(ctx
) ir_constant(column_type
, &data
);
1709 } else if (array
->type
->is_vector()) {
1710 const unsigned component
= idx
->value
.u
[0];
1712 return new(ctx
) ir_constant(array
, component
);
1714 const unsigned index
= idx
->value
.u
[0];
1715 return array
->get_array_element(index
)->clone(ctx
, NULL
);
1723 ir_dereference_record::constant_referenced(struct hash_table
*variable_context
,
1724 ir_constant
*&store
, int &offset
) const
1726 ir_constant
*substore
;
1728 const ir_dereference
*deref
= record
->as_dereference();
1735 deref
->constant_referenced(variable_context
, substore
, suboffset
);
1743 store
= substore
->get_record_field(field
);
1748 ir_dereference_record::constant_expression_value(struct hash_table
*variable_context
)
1750 ir_constant
*v
= this->record
->constant_expression_value();
1752 return (v
!= NULL
) ? v
->get_record_field(this->field
) : NULL
;
1757 ir_assignment::constant_expression_value(struct hash_table
*variable_context
)
1759 /* FINISHME: Handle CEs involving assignment (return RHS) */
1765 ir_constant::constant_expression_value(struct hash_table
*variable_context
)
1772 ir_call::constant_expression_value(struct hash_table
*variable_context
)
1774 return this->callee
->constant_expression_value(&this->actual_parameters
, variable_context
);
1778 bool ir_function_signature::constant_expression_evaluate_expression_list(const struct exec_list
&body
,
1779 struct hash_table
*variable_context
,
1780 ir_constant
**result
)
1782 foreach_list(n
, &body
) {
1783 ir_instruction
*inst
= (ir_instruction
*)n
;
1784 switch(inst
->ir_type
) {
1786 /* (declare () type symbol) */
1787 case ir_type_variable
: {
1788 ir_variable
*var
= inst
->as_variable();
1789 hash_table_insert(variable_context
, ir_constant::zero(this, var
->type
), var
);
1793 /* (assign [condition] (write-mask) (ref) (value)) */
1794 case ir_type_assignment
: {
1795 ir_assignment
*asg
= inst
->as_assignment();
1796 if (asg
->condition
) {
1797 ir_constant
*cond
= asg
->condition
->constant_expression_value(variable_context
);
1800 if (!cond
->get_bool_component(0))
1804 ir_constant
*store
= NULL
;
1806 asg
->lhs
->constant_referenced(variable_context
, store
, offset
);
1811 ir_constant
*value
= asg
->rhs
->constant_expression_value(variable_context
);
1816 store
->copy_masked_offset(value
, offset
, asg
->write_mask
);
1820 /* (return (expression)) */
1821 case ir_type_return
:
1823 *result
= inst
->as_return()->value
->constant_expression_value(variable_context
);
1824 return *result
!= NULL
;
1826 /* (call name (ref) (params))*/
1827 case ir_type_call
: {
1828 ir_call
*call
= inst
->as_call();
1830 /* Just say no to void functions in constant expressions. We
1831 * don't need them at that point.
1834 if (!call
->return_deref
)
1837 ir_constant
*store
= NULL
;
1839 call
->return_deref
->constant_referenced(variable_context
, store
, offset
);
1844 ir_constant
*value
= call
->constant_expression_value(variable_context
);
1849 store
->copy_offset(value
, offset
);
1853 /* (if condition (then-instructions) (else-instructions)) */
1855 ir_if
*iif
= inst
->as_if();
1857 ir_constant
*cond
= iif
->condition
->constant_expression_value(variable_context
);
1858 if (!cond
|| !cond
->type
->is_boolean())
1861 exec_list
&branch
= cond
->get_bool_component(0) ? iif
->then_instructions
: iif
->else_instructions
;
1864 if (!constant_expression_evaluate_expression_list(branch
, variable_context
, result
))
1867 /* If there was a return in the branch chosen, drop out now. */
1874 /* Every other expression type, we drop out. */
1880 /* Reaching the end of the block is not an error condition */
1888 ir_function_signature::constant_expression_value(exec_list
*actual_parameters
, struct hash_table
*variable_context
)
1890 const glsl_type
*type
= this->return_type
;
1891 if (type
== glsl_type::void_type
)
1894 /* From the GLSL 1.20 spec, page 23:
1895 * "Function calls to user-defined functions (non-built-in functions)
1896 * cannot be used to form constant expressions."
1898 if (!this->is_builtin())
1902 * Of the builtin functions, only the texture lookups and the noise
1903 * ones must not be used in constant expressions. They all include
1904 * specific opcodes so they don't need to be special-cased at this
1908 /* Initialize the table of dereferencable names with the function
1909 * parameters. Verify their const-ness on the way.
1911 * We expect the correctness of the number of parameters to have
1912 * been checked earlier.
1914 hash_table
*deref_hash
= hash_table_ctor(8, hash_table_pointer_hash
,
1915 hash_table_pointer_compare
);
1917 /* If "origin" is non-NULL, then the function body is there. So we
1918 * have to use the variable objects from the object with the body,
1919 * but the parameter instanciation on the current object.
1921 const exec_node
*parameter_info
= origin
? origin
->parameters
.head
: parameters
.head
;
1923 foreach_list(n
, actual_parameters
) {
1924 ir_constant
*constant
= ((ir_rvalue
*) n
)->constant_expression_value(variable_context
);
1925 if (constant
== NULL
) {
1926 hash_table_dtor(deref_hash
);
1931 ir_variable
*var
= (ir_variable
*)parameter_info
;
1932 hash_table_insert(deref_hash
, constant
, var
);
1934 parameter_info
= parameter_info
->next
;
1937 ir_constant
*result
= NULL
;
1939 /* Now run the builtin function until something non-constant
1940 * happens or we get the result.
1942 if (constant_expression_evaluate_expression_list(origin
? origin
->body
: body
, deref_hash
, &result
) && result
)
1943 result
= result
->clone(ralloc_parent(this), NULL
);
1945 hash_table_dtor(deref_hash
);