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 * Get the constant that is ultimately referenced by an r-value, in a constant
391 * expression evaluation context.
393 * The offset is used when the reference is to a specific column of a matrix.
396 constant_referenced(const ir_dereference
*deref
,
397 struct hash_table
*variable_context
,
398 ir_constant
*&store
, int &offset
)
403 if (variable_context
== NULL
)
406 switch (deref
->ir_type
) {
407 case ir_type_dereference_array
: {
408 const ir_dereference_array
*const da
=
409 (const ir_dereference_array
*) deref
;
411 ir_constant
*index_c
=
412 da
->array_index
->constant_expression_value(variable_context
);
414 if (!index_c
|| !index_c
->type
->is_scalar() || !index_c
->type
->is_integer())
417 int index
= index_c
->type
->base_type
== GLSL_TYPE_INT
?
418 index_c
->get_int_component(0) :
419 index_c
->get_uint_component(0);
421 ir_constant
*substore
;
424 const ir_dereference
*deref
= da
->array
->as_dereference();
428 if (!constant_referenced(deref
, variable_context
, substore
, suboffset
))
431 const glsl_type
*vt
= da
->array
->type
;
432 if (vt
->is_array()) {
433 store
= substore
->get_array_element(index
);
437 if (vt
->is_matrix()) {
439 offset
= index
* vt
->vector_elements
;
442 if (vt
->is_vector()) {
444 offset
= suboffset
+ index
;
451 case ir_type_dereference_record
: {
452 const ir_dereference_record
*const dr
=
453 (const ir_dereference_record
*) deref
;
455 const ir_dereference
*const deref
= dr
->record
->as_dereference();
459 ir_constant
*substore
;
462 if (!constant_referenced(deref
, variable_context
, substore
, suboffset
))
465 /* Since we're dropping it on the floor...
467 assert(suboffset
== 0);
469 store
= substore
->get_record_field(dr
->field
);
473 case ir_type_dereference_variable
: {
474 const ir_dereference_variable
*const dv
=
475 (const ir_dereference_variable
*) deref
;
477 store
= (ir_constant
*) hash_table_find(variable_context
, dv
->var
);
482 assert(!"Should not get here.");
486 return store
!= NULL
;
491 ir_rvalue::constant_expression_value(struct hash_table
*variable_context
)
493 assert(this->type
->is_error());
498 ir_expression::constant_expression_value(struct hash_table
*variable_context
)
500 if (this->type
->is_error())
503 ir_constant
*op
[Elements(this->operands
)] = { NULL
, };
504 ir_constant_data data
;
506 memset(&data
, 0, sizeof(data
));
508 for (unsigned operand
= 0; operand
< this->get_num_operands(); operand
++) {
509 op
[operand
] = this->operands
[operand
]->constant_expression_value(variable_context
);
515 switch (this->operation
) {
516 case ir_binop_lshift
:
517 case ir_binop_rshift
:
519 case ir_binop_vector_extract
:
521 case ir_triop_bitfield_extract
:
525 assert(op
[0]->type
->base_type
== op
[1]->type
->base_type
);
529 bool op0_scalar
= op
[0]->type
->is_scalar();
530 bool op1_scalar
= op
[1] != NULL
&& op
[1]->type
->is_scalar();
532 /* When iterating over a vector or matrix's components, we want to increase
533 * the loop counter. However, for scalars, we want to stay at 0.
535 unsigned c0_inc
= op0_scalar
? 0 : 1;
536 unsigned c1_inc
= op1_scalar
? 0 : 1;
538 if (op1_scalar
|| !op
[1]) {
539 components
= op
[0]->type
->components();
541 components
= op
[1]->type
->components();
544 void *ctx
= ralloc_parent(this);
546 /* Handle array operations here, rather than below. */
547 if (op
[0]->type
->is_array()) {
548 assert(op
[1] != NULL
&& op
[1]->type
->is_array());
549 switch (this->operation
) {
550 case ir_binop_all_equal
:
551 return new(ctx
) ir_constant(op
[0]->has_value(op
[1]));
552 case ir_binop_any_nequal
:
553 return new(ctx
) ir_constant(!op
[0]->has_value(op
[1]));
560 switch (this->operation
) {
561 case ir_unop_bit_not
:
562 switch (op
[0]->type
->base_type
) {
564 for (unsigned c
= 0; c
< components
; c
++)
565 data
.i
[c
] = ~ op
[0]->value
.i
[c
];
568 for (unsigned c
= 0; c
< components
; c
++)
569 data
.u
[c
] = ~ op
[0]->value
.u
[c
];
576 case ir_unop_logic_not
:
577 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
578 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++)
579 data
.b
[c
] = !op
[0]->value
.b
[c
];
583 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
584 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
585 data
.i
[c
] = (int) 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
.i
[c
] = (unsigned) op
[0]->value
.f
[c
];
595 assert(op
[0]->type
->base_type
== GLSL_TYPE_INT
);
596 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
597 data
.f
[c
] = (float) op
[0]->value
.i
[c
];
601 assert(op
[0]->type
->base_type
== GLSL_TYPE_UINT
);
602 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
603 data
.f
[c
] = (float) op
[0]->value
.u
[c
];
607 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
608 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
609 data
.f
[c
] = op
[0]->value
.b
[c
] ? 1.0F
: 0.0F
;
613 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
614 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
615 data
.b
[c
] = op
[0]->value
.f
[c
] != 0.0F
? true : false;
619 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
620 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
621 data
.u
[c
] = op
[0]->value
.b
[c
] ? 1 : 0;
625 assert(op
[0]->type
->is_integer());
626 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
627 data
.b
[c
] = op
[0]->value
.u
[c
] ? true : false;
631 assert(op
[0]->type
->base_type
== GLSL_TYPE_UINT
);
632 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
633 data
.i
[c
] = op
[0]->value
.u
[c
];
637 assert(op
[0]->type
->base_type
== GLSL_TYPE_INT
);
638 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
639 data
.u
[c
] = op
[0]->value
.i
[c
];
642 case ir_unop_bitcast_i2f
:
643 assert(op
[0]->type
->base_type
== GLSL_TYPE_INT
);
644 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
645 data
.f
[c
] = bitcast_u2f(op
[0]->value
.i
[c
]);
648 case ir_unop_bitcast_f2i
:
649 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
650 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
651 data
.i
[c
] = bitcast_f2u(op
[0]->value
.f
[c
]);
654 case ir_unop_bitcast_u2f
:
655 assert(op
[0]->type
->base_type
== GLSL_TYPE_UINT
);
656 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
657 data
.f
[c
] = bitcast_u2f(op
[0]->value
.u
[c
]);
660 case ir_unop_bitcast_f2u
:
661 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
662 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
663 data
.u
[c
] = bitcast_f2u(op
[0]->value
.f
[c
]);
667 assert(op
[0]->type
->is_boolean());
669 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
670 if (op
[0]->value
.b
[c
])
676 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
677 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
678 data
.f
[c
] = truncf(op
[0]->value
.f
[c
]);
682 case ir_unop_round_even
:
683 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
684 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
685 data
.f
[c
] = _mesa_round_to_even(op
[0]->value
.f
[c
]);
690 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
691 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
692 data
.f
[c
] = ceilf(op
[0]->value
.f
[c
]);
697 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
698 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
699 data
.f
[c
] = floorf(op
[0]->value
.f
[c
]);
704 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
705 switch (this->type
->base_type
) {
712 case GLSL_TYPE_FLOAT
:
713 data
.f
[c
] = op
[0]->value
.f
[c
] - floor(op
[0]->value
.f
[c
]);
722 case ir_unop_sin_reduced
:
723 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
724 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
725 data
.f
[c
] = sinf(op
[0]->value
.f
[c
]);
730 case ir_unop_cos_reduced
:
731 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
732 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
733 data
.f
[c
] = cosf(op
[0]->value
.f
[c
]);
738 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
739 switch (this->type
->base_type
) {
741 data
.u
[c
] = -((int) op
[0]->value
.u
[c
]);
744 data
.i
[c
] = -op
[0]->value
.i
[c
];
746 case GLSL_TYPE_FLOAT
:
747 data
.f
[c
] = -op
[0]->value
.f
[c
];
756 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
757 switch (this->type
->base_type
) {
759 data
.u
[c
] = op
[0]->value
.u
[c
];
762 data
.i
[c
] = op
[0]->value
.i
[c
];
764 data
.i
[c
] = -data
.i
[c
];
766 case GLSL_TYPE_FLOAT
:
767 data
.f
[c
] = fabs(op
[0]->value
.f
[c
]);
776 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
777 switch (this->type
->base_type
) {
779 data
.u
[c
] = op
[0]->value
.i
[c
] > 0;
782 data
.i
[c
] = (op
[0]->value
.i
[c
] > 0) - (op
[0]->value
.i
[c
] < 0);
784 case GLSL_TYPE_FLOAT
:
785 data
.f
[c
] = float((op
[0]->value
.f
[c
] > 0)-(op
[0]->value
.f
[c
] < 0));
794 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
795 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
796 switch (this->type
->base_type
) {
798 if (op
[0]->value
.u
[c
] != 0.0)
799 data
.u
[c
] = 1 / op
[0]->value
.u
[c
];
802 if (op
[0]->value
.i
[c
] != 0.0)
803 data
.i
[c
] = 1 / op
[0]->value
.i
[c
];
805 case GLSL_TYPE_FLOAT
:
806 if (op
[0]->value
.f
[c
] != 0.0)
807 data
.f
[c
] = 1.0F
/ op
[0]->value
.f
[c
];
816 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
817 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
818 data
.f
[c
] = 1.0F
/ sqrtf(op
[0]->value
.f
[c
]);
823 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
824 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
825 data
.f
[c
] = sqrtf(op
[0]->value
.f
[c
]);
830 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
831 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
832 data
.f
[c
] = expf(op
[0]->value
.f
[c
]);
837 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
838 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
839 data
.f
[c
] = exp2f(op
[0]->value
.f
[c
]);
844 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
845 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
846 data
.f
[c
] = logf(op
[0]->value
.f
[c
]);
851 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
852 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
853 data
.f
[c
] = log2f(op
[0]->value
.f
[c
]);
859 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
860 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
865 case ir_unop_pack_snorm_2x16
:
866 assert(op
[0]->type
== glsl_type::vec2_type
);
867 data
.u
[0] = pack_2x16(pack_snorm_1x16
,
871 case ir_unop_pack_snorm_4x8
:
872 assert(op
[0]->type
== glsl_type::vec4_type
);
873 data
.u
[0] = pack_4x8(pack_snorm_1x8
,
879 case ir_unop_unpack_snorm_2x16
:
880 assert(op
[0]->type
== glsl_type::uint_type
);
881 unpack_2x16(unpack_snorm_1x16
,
883 &data
.f
[0], &data
.f
[1]);
885 case ir_unop_unpack_snorm_4x8
:
886 assert(op
[0]->type
== glsl_type::uint_type
);
887 unpack_4x8(unpack_snorm_1x8
,
889 &data
.f
[0], &data
.f
[1], &data
.f
[2], &data
.f
[3]);
891 case ir_unop_pack_unorm_2x16
:
892 assert(op
[0]->type
== glsl_type::vec2_type
);
893 data
.u
[0] = pack_2x16(pack_unorm_1x16
,
897 case ir_unop_pack_unorm_4x8
:
898 assert(op
[0]->type
== glsl_type::vec4_type
);
899 data
.u
[0] = pack_4x8(pack_unorm_1x8
,
905 case ir_unop_unpack_unorm_2x16
:
906 assert(op
[0]->type
== glsl_type::uint_type
);
907 unpack_2x16(unpack_unorm_1x16
,
909 &data
.f
[0], &data
.f
[1]);
911 case ir_unop_unpack_unorm_4x8
:
912 assert(op
[0]->type
== glsl_type::uint_type
);
913 unpack_4x8(unpack_unorm_1x8
,
915 &data
.f
[0], &data
.f
[1], &data
.f
[2], &data
.f
[3]);
917 case ir_unop_pack_half_2x16
:
918 assert(op
[0]->type
== glsl_type::vec2_type
);
919 data
.u
[0] = pack_2x16(pack_half_1x16
,
923 case ir_unop_unpack_half_2x16
:
924 assert(op
[0]->type
== glsl_type::uint_type
);
925 unpack_2x16(unpack_half_1x16
,
927 &data
.f
[0], &data
.f
[1]);
930 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
931 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
932 data
.f
[c
] = powf(op
[0]->value
.f
[c
], op
[1]->value
.f
[c
]);
937 data
.f
[0] = dot(op
[0], op
[1]);
941 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
942 for (unsigned c
= 0, c0
= 0, c1
= 0;
944 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
946 switch (op
[0]->type
->base_type
) {
948 data
.u
[c
] = MIN2(op
[0]->value
.u
[c0
], op
[1]->value
.u
[c1
]);
951 data
.i
[c
] = MIN2(op
[0]->value
.i
[c0
], op
[1]->value
.i
[c1
]);
953 case GLSL_TYPE_FLOAT
:
954 data
.f
[c
] = MIN2(op
[0]->value
.f
[c0
], op
[1]->value
.f
[c1
]);
963 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
964 for (unsigned c
= 0, c0
= 0, c1
= 0;
966 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
968 switch (op
[0]->type
->base_type
) {
970 data
.u
[c
] = MAX2(op
[0]->value
.u
[c0
], op
[1]->value
.u
[c1
]);
973 data
.i
[c
] = MAX2(op
[0]->value
.i
[c0
], op
[1]->value
.i
[c1
]);
975 case GLSL_TYPE_FLOAT
:
976 data
.f
[c
] = MAX2(op
[0]->value
.f
[c0
], op
[1]->value
.f
[c1
]);
985 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
986 for (unsigned c
= 0, c0
= 0, c1
= 0;
988 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
990 switch (op
[0]->type
->base_type
) {
992 data
.u
[c
] = op
[0]->value
.u
[c0
] + op
[1]->value
.u
[c1
];
995 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 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
1008 for (unsigned c
= 0, c0
= 0, c1
= 0;
1010 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1012 switch (op
[0]->type
->base_type
) {
1013 case GLSL_TYPE_UINT
:
1014 data
.u
[c
] = op
[0]->value
.u
[c0
] - op
[1]->value
.u
[c1
];
1017 data
.i
[c
] = op
[0]->value
.i
[c0
] - op
[1]->value
.i
[c1
];
1019 case GLSL_TYPE_FLOAT
:
1020 data
.f
[c
] = op
[0]->value
.f
[c0
] - op
[1]->value
.f
[c1
];
1029 /* Check for equal types, or unequal types involving scalars */
1030 if ((op
[0]->type
== op
[1]->type
&& !op
[0]->type
->is_matrix())
1031 || op0_scalar
|| op1_scalar
) {
1032 for (unsigned c
= 0, c0
= 0, c1
= 0;
1034 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1036 switch (op
[0]->type
->base_type
) {
1037 case GLSL_TYPE_UINT
:
1038 data
.u
[c
] = op
[0]->value
.u
[c0
] * op
[1]->value
.u
[c1
];
1041 data
.i
[c
] = op
[0]->value
.i
[c0
] * op
[1]->value
.i
[c1
];
1043 case GLSL_TYPE_FLOAT
:
1044 data
.f
[c
] = op
[0]->value
.f
[c0
] * op
[1]->value
.f
[c1
];
1051 assert(op
[0]->type
->is_matrix() || op
[1]->type
->is_matrix());
1053 /* Multiply an N-by-M matrix with an M-by-P matrix. Since either
1054 * matrix can be a GLSL vector, either N or P can be 1.
1056 * For vec*mat, the vector is treated as a row vector. This
1057 * means the vector is a 1-row x M-column matrix.
1059 * For mat*vec, the vector is treated as a column vector. Since
1060 * matrix_columns is 1 for vectors, this just works.
1062 const unsigned n
= op
[0]->type
->is_vector()
1063 ? 1 : op
[0]->type
->vector_elements
;
1064 const unsigned m
= op
[1]->type
->vector_elements
;
1065 const unsigned p
= op
[1]->type
->matrix_columns
;
1066 for (unsigned j
= 0; j
< p
; j
++) {
1067 for (unsigned i
= 0; i
< n
; i
++) {
1068 for (unsigned k
= 0; k
< m
; k
++) {
1069 data
.f
[i
+n
*j
] += op
[0]->value
.f
[i
+n
*k
]*op
[1]->value
.f
[k
+m
*j
];
1077 /* FINISHME: Emit warning when division-by-zero is detected. */
1078 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
1079 for (unsigned c
= 0, c0
= 0, c1
= 0;
1081 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1083 switch (op
[0]->type
->base_type
) {
1084 case GLSL_TYPE_UINT
:
1085 if (op
[1]->value
.u
[c1
] == 0) {
1088 data
.u
[c
] = op
[0]->value
.u
[c0
] / op
[1]->value
.u
[c1
];
1092 if (op
[1]->value
.i
[c1
] == 0) {
1095 data
.i
[c
] = op
[0]->value
.i
[c0
] / op
[1]->value
.i
[c1
];
1098 case GLSL_TYPE_FLOAT
:
1099 data
.f
[c
] = op
[0]->value
.f
[c0
] / op
[1]->value
.f
[c1
];
1108 /* FINISHME: Emit warning when division-by-zero is detected. */
1109 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
1110 for (unsigned c
= 0, c0
= 0, c1
= 0;
1112 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1114 switch (op
[0]->type
->base_type
) {
1115 case GLSL_TYPE_UINT
:
1116 if (op
[1]->value
.u
[c1
] == 0) {
1119 data
.u
[c
] = op
[0]->value
.u
[c0
] % op
[1]->value
.u
[c1
];
1123 if (op
[1]->value
.i
[c1
] == 0) {
1126 data
.i
[c
] = op
[0]->value
.i
[c0
] % op
[1]->value
.i
[c1
];
1129 case GLSL_TYPE_FLOAT
:
1130 /* We don't use fmod because it rounds toward zero; GLSL specifies
1133 data
.f
[c
] = op
[0]->value
.f
[c0
] - op
[1]->value
.f
[c1
]
1134 * floorf(op
[0]->value
.f
[c0
] / op
[1]->value
.f
[c1
]);
1143 case ir_binop_logic_and
:
1144 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
1145 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++)
1146 data
.b
[c
] = op
[0]->value
.b
[c
] && op
[1]->value
.b
[c
];
1148 case ir_binop_logic_xor
:
1149 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
1150 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++)
1151 data
.b
[c
] = op
[0]->value
.b
[c
] ^ op
[1]->value
.b
[c
];
1153 case ir_binop_logic_or
:
1154 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
1155 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++)
1156 data
.b
[c
] = op
[0]->value
.b
[c
] || op
[1]->value
.b
[c
];
1160 assert(op
[0]->type
== op
[1]->type
);
1161 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
1162 switch (op
[0]->type
->base_type
) {
1163 case GLSL_TYPE_UINT
:
1164 data
.b
[c
] = op
[0]->value
.u
[c
] < op
[1]->value
.u
[c
];
1167 data
.b
[c
] = op
[0]->value
.i
[c
] < op
[1]->value
.i
[c
];
1169 case GLSL_TYPE_FLOAT
:
1170 data
.b
[c
] = op
[0]->value
.f
[c
] < op
[1]->value
.f
[c
];
1177 case ir_binop_greater
:
1178 assert(op
[0]->type
== op
[1]->type
);
1179 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
1180 switch (op
[0]->type
->base_type
) {
1181 case GLSL_TYPE_UINT
:
1182 data
.b
[c
] = op
[0]->value
.u
[c
] > op
[1]->value
.u
[c
];
1185 data
.b
[c
] = op
[0]->value
.i
[c
] > op
[1]->value
.i
[c
];
1187 case GLSL_TYPE_FLOAT
:
1188 data
.b
[c
] = op
[0]->value
.f
[c
] > op
[1]->value
.f
[c
];
1195 case ir_binop_lequal
:
1196 assert(op
[0]->type
== op
[1]->type
);
1197 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
1198 switch (op
[0]->type
->base_type
) {
1199 case GLSL_TYPE_UINT
:
1200 data
.b
[c
] = op
[0]->value
.u
[c
] <= op
[1]->value
.u
[c
];
1203 data
.b
[c
] = op
[0]->value
.i
[c
] <= op
[1]->value
.i
[c
];
1205 case GLSL_TYPE_FLOAT
:
1206 data
.b
[c
] = op
[0]->value
.f
[c
] <= op
[1]->value
.f
[c
];
1213 case ir_binop_gequal
:
1214 assert(op
[0]->type
== op
[1]->type
);
1215 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
1216 switch (op
[0]->type
->base_type
) {
1217 case GLSL_TYPE_UINT
:
1218 data
.b
[c
] = op
[0]->value
.u
[c
] >= op
[1]->value
.u
[c
];
1221 data
.b
[c
] = op
[0]->value
.i
[c
] >= op
[1]->value
.i
[c
];
1223 case GLSL_TYPE_FLOAT
:
1224 data
.b
[c
] = op
[0]->value
.f
[c
] >= op
[1]->value
.f
[c
];
1231 case ir_binop_equal
:
1232 assert(op
[0]->type
== op
[1]->type
);
1233 for (unsigned c
= 0; c
< components
; c
++) {
1234 switch (op
[0]->type
->base_type
) {
1235 case GLSL_TYPE_UINT
:
1236 data
.b
[c
] = op
[0]->value
.u
[c
] == op
[1]->value
.u
[c
];
1239 data
.b
[c
] = op
[0]->value
.i
[c
] == op
[1]->value
.i
[c
];
1241 case GLSL_TYPE_FLOAT
:
1242 data
.b
[c
] = op
[0]->value
.f
[c
] == op
[1]->value
.f
[c
];
1244 case GLSL_TYPE_BOOL
:
1245 data
.b
[c
] = op
[0]->value
.b
[c
] == op
[1]->value
.b
[c
];
1252 case ir_binop_nequal
:
1253 assert(op
[0]->type
== op
[1]->type
);
1254 for (unsigned c
= 0; c
< components
; c
++) {
1255 switch (op
[0]->type
->base_type
) {
1256 case GLSL_TYPE_UINT
:
1257 data
.b
[c
] = op
[0]->value
.u
[c
] != op
[1]->value
.u
[c
];
1260 data
.b
[c
] = op
[0]->value
.i
[c
] != op
[1]->value
.i
[c
];
1262 case GLSL_TYPE_FLOAT
:
1263 data
.b
[c
] = op
[0]->value
.f
[c
] != op
[1]->value
.f
[c
];
1265 case GLSL_TYPE_BOOL
:
1266 data
.b
[c
] = op
[0]->value
.b
[c
] != op
[1]->value
.b
[c
];
1273 case ir_binop_all_equal
:
1274 data
.b
[0] = op
[0]->has_value(op
[1]);
1276 case ir_binop_any_nequal
:
1277 data
.b
[0] = !op
[0]->has_value(op
[1]);
1280 case ir_binop_lshift
:
1281 for (unsigned c
= 0, c0
= 0, c1
= 0;
1283 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1285 if (op
[0]->type
->base_type
== GLSL_TYPE_INT
&&
1286 op
[1]->type
->base_type
== GLSL_TYPE_INT
) {
1287 data
.i
[c
] = op
[0]->value
.i
[c0
] << op
[1]->value
.i
[c1
];
1289 } else if (op
[0]->type
->base_type
== GLSL_TYPE_INT
&&
1290 op
[1]->type
->base_type
== GLSL_TYPE_UINT
) {
1291 data
.i
[c
] = op
[0]->value
.i
[c0
] << op
[1]->value
.u
[c1
];
1293 } else if (op
[0]->type
->base_type
== GLSL_TYPE_UINT
&&
1294 op
[1]->type
->base_type
== GLSL_TYPE_INT
) {
1295 data
.u
[c
] = op
[0]->value
.u
[c0
] << op
[1]->value
.i
[c1
];
1297 } else if (op
[0]->type
->base_type
== GLSL_TYPE_UINT
&&
1298 op
[1]->type
->base_type
== GLSL_TYPE_UINT
) {
1299 data
.u
[c
] = op
[0]->value
.u
[c0
] << op
[1]->value
.u
[c1
];
1304 case ir_binop_rshift
:
1305 for (unsigned c
= 0, c0
= 0, c1
= 0;
1307 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1309 if (op
[0]->type
->base_type
== GLSL_TYPE_INT
&&
1310 op
[1]->type
->base_type
== GLSL_TYPE_INT
) {
1311 data
.i
[c
] = op
[0]->value
.i
[c0
] >> op
[1]->value
.i
[c1
];
1313 } else if (op
[0]->type
->base_type
== GLSL_TYPE_INT
&&
1314 op
[1]->type
->base_type
== GLSL_TYPE_UINT
) {
1315 data
.i
[c
] = op
[0]->value
.i
[c0
] >> op
[1]->value
.u
[c1
];
1317 } else if (op
[0]->type
->base_type
== GLSL_TYPE_UINT
&&
1318 op
[1]->type
->base_type
== GLSL_TYPE_INT
) {
1319 data
.u
[c
] = op
[0]->value
.u
[c0
] >> op
[1]->value
.i
[c1
];
1321 } else if (op
[0]->type
->base_type
== GLSL_TYPE_UINT
&&
1322 op
[1]->type
->base_type
== GLSL_TYPE_UINT
) {
1323 data
.u
[c
] = op
[0]->value
.u
[c0
] >> op
[1]->value
.u
[c1
];
1328 case ir_binop_bit_and
:
1329 for (unsigned c
= 0, c0
= 0, c1
= 0;
1331 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1333 switch (op
[0]->type
->base_type
) {
1335 data
.i
[c
] = op
[0]->value
.i
[c0
] & op
[1]->value
.i
[c1
];
1337 case GLSL_TYPE_UINT
:
1338 data
.u
[c
] = op
[0]->value
.u
[c0
] & op
[1]->value
.u
[c1
];
1346 case ir_binop_bit_or
:
1347 for (unsigned c
= 0, c0
= 0, c1
= 0;
1349 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1351 switch (op
[0]->type
->base_type
) {
1353 data
.i
[c
] = op
[0]->value
.i
[c0
] | op
[1]->value
.i
[c1
];
1355 case GLSL_TYPE_UINT
:
1356 data
.u
[c
] = op
[0]->value
.u
[c0
] | op
[1]->value
.u
[c1
];
1364 case ir_binop_vector_extract
: {
1365 const int c
= CLAMP(op
[1]->value
.i
[0], 0,
1366 (int) op
[0]->type
->vector_elements
- 1);
1368 switch (op
[0]->type
->base_type
) {
1369 case GLSL_TYPE_UINT
:
1370 data
.u
[0] = op
[0]->value
.u
[c
];
1373 data
.i
[0] = op
[0]->value
.i
[c
];
1375 case GLSL_TYPE_FLOAT
:
1376 data
.f
[0] = op
[0]->value
.f
[c
];
1378 case GLSL_TYPE_BOOL
:
1379 data
.b
[0] = op
[0]->value
.b
[c
];
1387 case ir_binop_bit_xor
:
1388 for (unsigned c
= 0, c0
= 0, c1
= 0;
1390 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1392 switch (op
[0]->type
->base_type
) {
1394 data
.i
[c
] = op
[0]->value
.i
[c0
] ^ op
[1]->value
.i
[c1
];
1396 case GLSL_TYPE_UINT
:
1397 data
.u
[c
] = op
[0]->value
.u
[c0
] ^ op
[1]->value
.u
[c1
];
1405 case ir_unop_bitfield_reverse
:
1406 /* http://graphics.stanford.edu/~seander/bithacks.html#BitReverseObvious */
1407 for (unsigned c
= 0; c
< components
; c
++) {
1408 unsigned int v
= op
[0]->value
.u
[c
]; // input bits to be reversed
1409 unsigned int r
= v
; // r will be reversed bits of v; first get LSB of v
1410 int s
= sizeof(v
) * CHAR_BIT
- 1; // extra shift needed at end
1412 for (v
>>= 1; v
; v
>>= 1) {
1417 r
<<= s
; // shift when v's highest bits are zero
1423 case ir_unop_bit_count
:
1424 for (unsigned c
= 0; c
< components
; c
++) {
1426 unsigned v
= op
[0]->value
.u
[c
];
1428 for (; v
; count
++) {
1435 case ir_unop_find_msb
:
1436 for (unsigned c
= 0; c
< components
; c
++) {
1437 int v
= op
[0]->value
.i
[c
];
1439 if (v
== 0 || (op
[0]->type
->base_type
== GLSL_TYPE_INT
&& v
== -1))
1443 int top_bit
= op
[0]->type
->base_type
== GLSL_TYPE_UINT
1444 ? 0 : v
& (1 << 31);
1446 while (((v
& (1 << 31)) == top_bit
) && count
!= 32) {
1451 data
.i
[c
] = 31 - count
;
1456 case ir_unop_find_lsb
:
1457 for (unsigned c
= 0; c
< components
; c
++) {
1458 if (op
[0]->value
.i
[c
] == 0)
1462 unsigned v
= op
[0]->value
.u
[c
];
1464 for (; !(v
& 1); v
>>= 1) {
1472 case ir_triop_bitfield_extract
: {
1473 int offset
= op
[1]->value
.i
[0];
1474 int bits
= op
[2]->value
.i
[0];
1476 for (unsigned c
= 0; c
< components
; c
++) {
1479 else if (offset
< 0 || bits
< 0)
1480 data
.u
[c
] = 0; /* Undefined, per spec. */
1481 else if (offset
+ bits
> 32)
1482 data
.u
[c
] = 0; /* Undefined, per spec. */
1484 if (op
[0]->type
->base_type
== GLSL_TYPE_INT
) {
1485 /* int so that the right shift will sign-extend. */
1486 int value
= op
[0]->value
.i
[c
];
1487 value
<<= 32 - bits
- offset
;
1488 value
>>= 32 - bits
;
1491 unsigned value
= op
[0]->value
.u
[c
];
1492 value
<<= 32 - bits
- offset
;
1493 value
>>= 32 - bits
;
1501 case ir_binop_bfm
: {
1502 int bits
= op
[0]->value
.i
[0];
1503 int offset
= op
[1]->value
.i
[0];
1505 for (unsigned c
= 0; c
< components
; c
++) {
1507 data
.u
[c
] = op
[0]->value
.u
[c
];
1508 else if (offset
< 0 || bits
< 0)
1509 data
.u
[c
] = 0; /* Undefined for bitfieldInsert, per spec. */
1510 else if (offset
+ bits
> 32)
1511 data
.u
[c
] = 0; /* Undefined for bitfieldInsert, per spec. */
1513 data
.u
[c
] = ((1 << bits
) - 1) << offset
;
1518 case ir_binop_ldexp
:
1519 for (unsigned c
= 0; c
< components
; c
++) {
1520 data
.f
[c
] = ldexp(op
[0]->value
.f
[c
], op
[1]->value
.i
[c
]);
1521 /* Flush subnormal values to zero. */
1522 if (!isnormal(data
.f
[c
]))
1523 data
.f
[c
] = copysign(0.0f
, op
[0]->value
.f
[c
]);
1528 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
1529 assert(op
[1]->type
->base_type
== GLSL_TYPE_FLOAT
);
1530 assert(op
[2]->type
->base_type
== GLSL_TYPE_FLOAT
);
1532 for (unsigned c
= 0; c
< components
; c
++) {
1533 data
.f
[c
] = op
[0]->value
.f
[c
] * op
[1]->value
.f
[c
]
1534 + op
[2]->value
.f
[c
];
1538 case ir_triop_lrp
: {
1539 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
1540 assert(op
[1]->type
->base_type
== GLSL_TYPE_FLOAT
);
1541 assert(op
[2]->type
->base_type
== GLSL_TYPE_FLOAT
);
1543 unsigned c2_inc
= op
[2]->type
->is_scalar() ? 0 : 1;
1544 for (unsigned c
= 0, c2
= 0; c
< components
; c2
+= c2_inc
, c
++) {
1545 data
.f
[c
] = op
[0]->value
.f
[c
] * (1.0f
- op
[2]->value
.f
[c2
]) +
1546 (op
[1]->value
.f
[c
] * op
[2]->value
.f
[c2
]);
1552 for (unsigned c
= 0; c
< components
; c
++) {
1553 data
.u
[c
] = op
[0]->value
.b
[c
] ? op
[1]->value
.u
[c
]
1554 : op
[2]->value
.u
[c
];
1558 case ir_triop_vector_insert
: {
1559 const unsigned idx
= op
[2]->value
.u
[0];
1561 memcpy(&data
, &op
[0]->value
, sizeof(data
));
1563 switch (this->type
->base_type
) {
1565 data
.i
[idx
] = op
[1]->value
.i
[0];
1567 case GLSL_TYPE_UINT
:
1568 data
.u
[idx
] = op
[1]->value
.u
[0];
1570 case GLSL_TYPE_FLOAT
:
1571 data
.f
[idx
] = op
[1]->value
.f
[0];
1573 case GLSL_TYPE_BOOL
:
1574 data
.b
[idx
] = op
[1]->value
.b
[0];
1577 assert(!"Should not get here.");
1583 case ir_quadop_bitfield_insert
: {
1584 int offset
= op
[2]->value
.i
[0];
1585 int bits
= op
[3]->value
.i
[0];
1587 for (unsigned c
= 0; c
< components
; c
++) {
1589 data
.u
[c
] = op
[0]->value
.u
[c
];
1590 else if (offset
< 0 || bits
< 0)
1591 data
.u
[c
] = 0; /* Undefined, per spec. */
1592 else if (offset
+ bits
> 32)
1593 data
.u
[c
] = 0; /* Undefined, per spec. */
1595 unsigned insert_mask
= ((1 << bits
) - 1) << offset
;
1597 unsigned insert
= op
[1]->value
.u
[c
];
1599 insert
&= insert_mask
;
1601 unsigned base
= op
[0]->value
.u
[c
];
1602 base
&= ~insert_mask
;
1604 data
.u
[c
] = base
| insert
;
1610 case ir_quadop_vector
:
1611 for (unsigned c
= 0; c
< this->type
->vector_elements
; c
++) {
1612 switch (this->type
->base_type
) {
1614 data
.i
[c
] = op
[c
]->value
.i
[0];
1616 case GLSL_TYPE_UINT
:
1617 data
.u
[c
] = op
[c
]->value
.u
[0];
1619 case GLSL_TYPE_FLOAT
:
1620 data
.f
[c
] = op
[c
]->value
.f
[0];
1629 /* FINISHME: Should handle all expression types. */
1633 return new(ctx
) ir_constant(this->type
, &data
);
1638 ir_texture::constant_expression_value(struct hash_table
*variable_context
)
1640 /* texture lookups aren't constant expressions */
1646 ir_swizzle::constant_expression_value(struct hash_table
*variable_context
)
1648 ir_constant
*v
= this->val
->constant_expression_value(variable_context
);
1651 ir_constant_data data
= { { 0 } };
1653 const unsigned swiz_idx
[4] = {
1654 this->mask
.x
, this->mask
.y
, this->mask
.z
, this->mask
.w
1657 for (unsigned i
= 0; i
< this->mask
.num_components
; i
++) {
1658 switch (v
->type
->base_type
) {
1659 case GLSL_TYPE_UINT
:
1660 case GLSL_TYPE_INT
: data
.u
[i
] = v
->value
.u
[swiz_idx
[i
]]; break;
1661 case GLSL_TYPE_FLOAT
: data
.f
[i
] = v
->value
.f
[swiz_idx
[i
]]; break;
1662 case GLSL_TYPE_BOOL
: data
.b
[i
] = v
->value
.b
[swiz_idx
[i
]]; break;
1663 default: assert(!"Should not get here."); break;
1667 void *ctx
= ralloc_parent(this);
1668 return new(ctx
) ir_constant(this->type
, &data
);
1675 ir_dereference_variable::constant_expression_value(struct hash_table
*variable_context
)
1677 /* This may occur during compile and var->type is glsl_type::error_type */
1681 /* Give priority to the context hashtable, if it exists */
1682 if (variable_context
) {
1683 ir_constant
*value
= (ir_constant
*)hash_table_find(variable_context
, var
);
1688 /* The constant_value of a uniform variable is its initializer,
1689 * not the lifetime constant value of the uniform.
1691 if (var
->data
.mode
== ir_var_uniform
)
1694 if (!var
->constant_value
)
1697 return var
->constant_value
->clone(ralloc_parent(var
), NULL
);
1702 ir_dereference_array::constant_expression_value(struct hash_table
*variable_context
)
1704 ir_constant
*array
= this->array
->constant_expression_value(variable_context
);
1705 ir_constant
*idx
= this->array_index
->constant_expression_value(variable_context
);
1707 if ((array
!= NULL
) && (idx
!= NULL
)) {
1708 void *ctx
= ralloc_parent(this);
1709 if (array
->type
->is_matrix()) {
1710 /* Array access of a matrix results in a vector.
1712 const unsigned column
= idx
->value
.u
[0];
1714 const glsl_type
*const column_type
= array
->type
->column_type();
1716 /* Offset in the constant matrix to the first element of the column
1719 const unsigned mat_idx
= column
* column_type
->vector_elements
;
1721 ir_constant_data data
= { { 0 } };
1723 switch (column_type
->base_type
) {
1724 case GLSL_TYPE_UINT
:
1726 for (unsigned i
= 0; i
< column_type
->vector_elements
; i
++)
1727 data
.u
[i
] = array
->value
.u
[mat_idx
+ i
];
1731 case GLSL_TYPE_FLOAT
:
1732 for (unsigned i
= 0; i
< column_type
->vector_elements
; i
++)
1733 data
.f
[i
] = array
->value
.f
[mat_idx
+ i
];
1738 assert(!"Should not get here.");
1742 return new(ctx
) ir_constant(column_type
, &data
);
1743 } else if (array
->type
->is_vector()) {
1744 const unsigned component
= idx
->value
.u
[0];
1746 return new(ctx
) ir_constant(array
, component
);
1748 const unsigned index
= idx
->value
.u
[0];
1749 return array
->get_array_element(index
)->clone(ctx
, NULL
);
1757 ir_dereference_record::constant_expression_value(struct hash_table
*variable_context
)
1759 ir_constant
*v
= this->record
->constant_expression_value();
1761 return (v
!= NULL
) ? v
->get_record_field(this->field
) : NULL
;
1766 ir_assignment::constant_expression_value(struct hash_table
*variable_context
)
1768 /* FINISHME: Handle CEs involving assignment (return RHS) */
1774 ir_constant::constant_expression_value(struct hash_table
*variable_context
)
1781 ir_call::constant_expression_value(struct hash_table
*variable_context
)
1783 return this->callee
->constant_expression_value(&this->actual_parameters
, variable_context
);
1787 bool ir_function_signature::constant_expression_evaluate_expression_list(const struct exec_list
&body
,
1788 struct hash_table
*variable_context
,
1789 ir_constant
**result
)
1791 foreach_list(n
, &body
) {
1792 ir_instruction
*inst
= (ir_instruction
*)n
;
1793 switch(inst
->ir_type
) {
1795 /* (declare () type symbol) */
1796 case ir_type_variable
: {
1797 ir_variable
*var
= inst
->as_variable();
1798 hash_table_insert(variable_context
, ir_constant::zero(this, var
->type
), var
);
1802 /* (assign [condition] (write-mask) (ref) (value)) */
1803 case ir_type_assignment
: {
1804 ir_assignment
*asg
= inst
->as_assignment();
1805 if (asg
->condition
) {
1806 ir_constant
*cond
= asg
->condition
->constant_expression_value(variable_context
);
1809 if (!cond
->get_bool_component(0))
1813 ir_constant
*store
= NULL
;
1816 if (!constant_referenced(asg
->lhs
, variable_context
, store
, offset
))
1819 ir_constant
*value
= asg
->rhs
->constant_expression_value(variable_context
);
1824 store
->copy_masked_offset(value
, offset
, asg
->write_mask
);
1828 /* (return (expression)) */
1829 case ir_type_return
:
1831 *result
= inst
->as_return()->value
->constant_expression_value(variable_context
);
1832 return *result
!= NULL
;
1834 /* (call name (ref) (params))*/
1835 case ir_type_call
: {
1836 ir_call
*call
= inst
->as_call();
1838 /* Just say no to void functions in constant expressions. We
1839 * don't need them at that point.
1842 if (!call
->return_deref
)
1845 ir_constant
*store
= NULL
;
1848 if (!constant_referenced(call
->return_deref
, variable_context
,
1852 ir_constant
*value
= call
->constant_expression_value(variable_context
);
1857 store
->copy_offset(value
, offset
);
1861 /* (if condition (then-instructions) (else-instructions)) */
1863 ir_if
*iif
= inst
->as_if();
1865 ir_constant
*cond
= iif
->condition
->constant_expression_value(variable_context
);
1866 if (!cond
|| !cond
->type
->is_boolean())
1869 exec_list
&branch
= cond
->get_bool_component(0) ? iif
->then_instructions
: iif
->else_instructions
;
1872 if (!constant_expression_evaluate_expression_list(branch
, variable_context
, result
))
1875 /* If there was a return in the branch chosen, drop out now. */
1882 /* Every other expression type, we drop out. */
1888 /* Reaching the end of the block is not an error condition */
1896 ir_function_signature::constant_expression_value(exec_list
*actual_parameters
, struct hash_table
*variable_context
)
1898 const glsl_type
*type
= this->return_type
;
1899 if (type
== glsl_type::void_type
)
1902 /* From the GLSL 1.20 spec, page 23:
1903 * "Function calls to user-defined functions (non-built-in functions)
1904 * cannot be used to form constant expressions."
1906 if (!this->is_builtin())
1910 * Of the builtin functions, only the texture lookups and the noise
1911 * ones must not be used in constant expressions. They all include
1912 * specific opcodes so they don't need to be special-cased at this
1916 /* Initialize the table of dereferencable names with the function
1917 * parameters. Verify their const-ness on the way.
1919 * We expect the correctness of the number of parameters to have
1920 * been checked earlier.
1922 hash_table
*deref_hash
= hash_table_ctor(8, hash_table_pointer_hash
,
1923 hash_table_pointer_compare
);
1925 /* If "origin" is non-NULL, then the function body is there. So we
1926 * have to use the variable objects from the object with the body,
1927 * but the parameter instanciation on the current object.
1929 const exec_node
*parameter_info
= origin
? origin
->parameters
.head
: parameters
.head
;
1931 foreach_list(n
, actual_parameters
) {
1932 ir_constant
*constant
= ((ir_rvalue
*) n
)->constant_expression_value(variable_context
);
1933 if (constant
== NULL
) {
1934 hash_table_dtor(deref_hash
);
1939 ir_variable
*var
= (ir_variable
*)parameter_info
;
1940 hash_table_insert(deref_hash
, constant
, var
);
1942 parameter_info
= parameter_info
->next
;
1945 ir_constant
*result
= NULL
;
1947 /* Now run the builtin function until something non-constant
1948 * happens or we get the result.
1950 if (constant_expression_evaluate_expression_list(origin
? origin
->body
: body
, deref_hash
, &result
) && result
)
1951 result
= result
->clone(ralloc_parent(this), NULL
);
1953 hash_table_dtor(deref_hash
);