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 */
38 #include "util/rounding.h" /* for _mesa_roundeven */
39 #include "util/half_float.h"
41 #include "compiler/glsl_types.h"
42 #include "program/hash_table.h"
45 dot_f(ir_constant
*op0
, ir_constant
*op1
)
47 assert(op0
->type
->is_float() && op1
->type
->is_float());
50 for (unsigned c
= 0; c
< op0
->type
->components(); c
++)
51 result
+= op0
->value
.f
[c
] * op1
->value
.f
[c
];
57 dot_d(ir_constant
*op0
, ir_constant
*op1
)
59 assert(op0
->type
->is_double() && op1
->type
->is_double());
62 for (unsigned c
= 0; c
< op0
->type
->components(); c
++)
63 result
+= op0
->value
.d
[c
] * op1
->value
.d
[c
];
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.
74 bitcast_u2f(unsigned int u
)
76 assert(sizeof(float) == sizeof(unsigned int));
78 memcpy(&f
, &u
, sizeof(f
));
85 assert(sizeof(float) == sizeof(unsigned int));
87 memcpy(&u
, &f
, sizeof(f
));
92 * Evaluate one component of a floating-point 4x8 unpacking function.
95 (*pack_1x8_func_t
)(float);
98 * Evaluate one component of a floating-point 2x16 unpacking function.
101 (*pack_1x16_func_t
)(float);
104 * Evaluate one component of a floating-point 4x8 unpacking function.
107 (*unpack_1x8_func_t
)(uint8_t);
110 * Evaluate one component of a floating-point 2x16 unpacking function.
113 (*unpack_1x16_func_t
)(uint16_t);
116 * Evaluate a 2x16 floating-point packing function.
119 pack_2x16(pack_1x16_func_t pack_1x16
,
122 /* From section 8.4 of the GLSL ES 3.00 spec:
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.
130 * The specifications for the other packing functions contain similar
134 u
|= ((uint32_t) pack_1x16(x
) << 0);
135 u
|= ((uint32_t) pack_1x16(y
) << 16);
140 * Evaluate a 4x8 floating-point packing function.
143 pack_4x8(pack_1x8_func_t pack_1x8
,
144 float x
, float y
, float z
, float w
)
146 /* From section 8.4 of the GLSL 4.30 spec:
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.
154 * The specifications for the other packing functions contain similar
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);
166 * Evaluate a 2x16 floating-point unpacking function.
169 unpack_2x16(unpack_1x16_func_t unpack_1x16
,
173 /* From section 8.4 of the GLSL ES 3.00 spec:
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.
181 * The specifications for the other unpacking functions contain similar
184 *x
= unpack_1x16((uint16_t) (u
& 0xffff));
185 *y
= unpack_1x16((uint16_t) (u
>> 16));
189 * Evaluate a 4x8 floating-point unpacking function.
192 unpack_4x8(unpack_1x8_func_t unpack_1x8
, uint32_t u
,
193 float *x
, float *y
, float *z
, float *w
)
195 /* From section 8.4 of the GLSL 4.30 spec:
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.
203 * The specifications for the other unpacking functions contain similar
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));
213 * Evaluate one component of packSnorm4x8.
216 pack_snorm_1x8(float x
)
218 /* From section 8.4 of the GLSL 4.30 spec:
222 * The conversion for component c of v to fixed point is done as
225 * packSnorm4x8: round(clamp(c, -1, +1) * 127.0)
228 _mesa_lroundevenf(CLAMP(x
, -1.0f
, +1.0f
) * 127.0f
);
232 * Evaluate one component of packSnorm2x16.
235 pack_snorm_1x16(float x
)
237 /* From section 8.4 of the GLSL ES 3.00 spec:
241 * The conversion for component c of v to fixed point is done as
244 * packSnorm2x16: round(clamp(c, -1, +1) * 32767.0)
247 _mesa_lroundevenf(CLAMP(x
, -1.0f
, +1.0f
) * 32767.0f
);
251 * Evaluate one component of unpackSnorm4x8.
254 unpack_snorm_1x8(uint8_t u
)
256 /* From section 8.4 of the GLSL 4.30 spec:
260 * The conversion for unpacked fixed-point value f to floating point is
263 * unpackSnorm4x8: clamp(f / 127.0, -1, +1)
265 return CLAMP((int8_t) u
/ 127.0f
, -1.0f
, +1.0f
);
269 * Evaluate one component of unpackSnorm2x16.
272 unpack_snorm_1x16(uint16_t u
)
274 /* From section 8.4 of the GLSL ES 3.00 spec:
278 * The conversion for unpacked fixed-point value f to floating point is
281 * unpackSnorm2x16: clamp(f / 32767.0, -1, +1)
283 return CLAMP((int16_t) u
/ 32767.0f
, -1.0f
, +1.0f
);
287 * Evaluate one component packUnorm4x8.
290 pack_unorm_1x8(float x
)
292 /* From section 8.4 of the GLSL 4.30 spec:
296 * The conversion for component c of v to fixed point is done as
299 * packUnorm4x8: round(clamp(c, 0, +1) * 255.0)
301 return (uint8_t) (int) _mesa_roundevenf(CLAMP(x
, 0.0f
, 1.0f
) * 255.0f
);
305 * Evaluate one component packUnorm2x16.
308 pack_unorm_1x16(float x
)
310 /* From section 8.4 of the GLSL ES 3.00 spec:
314 * The conversion for component c of v to fixed point is done as
317 * packUnorm2x16: round(clamp(c, 0, +1) * 65535.0)
319 return (uint16_t) (int)
320 _mesa_roundevenf(CLAMP(x
, 0.0f
, 1.0f
) * 65535.0f
);
324 * Evaluate one component of unpackUnorm4x8.
327 unpack_unorm_1x8(uint8_t u
)
329 /* From section 8.4 of the GLSL 4.30 spec:
333 * The conversion for unpacked fixed-point value f to floating point is
336 * unpackUnorm4x8: f / 255.0
338 return (float) u
/ 255.0f
;
342 * Evaluate one component of unpackUnorm2x16.
345 unpack_unorm_1x16(uint16_t u
)
347 /* From section 8.4 of the GLSL ES 3.00 spec:
351 * The conversion for unpacked fixed-point value f to floating point is
354 * unpackUnorm2x16: f / 65535.0
356 return (float) u
/ 65535.0f
;
360 * Evaluate one component of packHalf2x16.
363 pack_half_1x16(float x
)
365 return _mesa_float_to_half(x
);
369 * Evaluate one component of unpackHalf2x16.
372 unpack_half_1x16(uint16_t u
)
374 return _mesa_half_to_float(u
);
378 * Get the constant that is ultimately referenced by an r-value, in a constant
379 * expression evaluation context.
381 * The offset is used when the reference is to a specific column of a matrix.
384 constant_referenced(const ir_dereference
*deref
,
385 struct hash_table
*variable_context
,
386 ir_constant
*&store
, int &offset
)
391 if (variable_context
== NULL
)
394 switch (deref
->ir_type
) {
395 case ir_type_dereference_array
: {
396 const ir_dereference_array
*const da
=
397 (const ir_dereference_array
*) deref
;
399 ir_constant
*const index_c
=
400 da
->array_index
->constant_expression_value(variable_context
);
402 if (!index_c
|| !index_c
->type
->is_scalar() || !index_c
->type
->is_integer())
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);
409 ir_constant
*substore
;
412 const ir_dereference
*const deref
= da
->array
->as_dereference();
416 if (!constant_referenced(deref
, variable_context
, substore
, suboffset
))
419 const glsl_type
*const vt
= da
->array
->type
;
420 if (vt
->is_array()) {
421 store
= substore
->get_array_element(index
);
423 } else if (vt
->is_matrix()) {
425 offset
= index
* vt
->vector_elements
;
426 } else if (vt
->is_vector()) {
428 offset
= suboffset
+ index
;
434 case ir_type_dereference_record
: {
435 const ir_dereference_record
*const dr
=
436 (const ir_dereference_record
*) deref
;
438 const ir_dereference
*const deref
= dr
->record
->as_dereference();
442 ir_constant
*substore
;
445 if (!constant_referenced(deref
, variable_context
, substore
, suboffset
))
448 /* Since we're dropping it on the floor...
450 assert(suboffset
== 0);
452 store
= substore
->get_record_field(dr
->field
);
456 case ir_type_dereference_variable
: {
457 const ir_dereference_variable
*const dv
=
458 (const ir_dereference_variable
*) deref
;
460 store
= (ir_constant
*) hash_table_find(variable_context
, dv
->var
);
465 assert(!"Should not get here.");
469 return store
!= NULL
;
474 ir_rvalue::constant_expression_value(struct hash_table
*)
476 assert(this->type
->is_error());
481 ir_expression::constant_expression_value(struct hash_table
*variable_context
)
483 if (this->type
->is_error())
486 ir_constant
*op
[ARRAY_SIZE(this->operands
)] = { NULL
, };
487 ir_constant_data data
;
489 memset(&data
, 0, sizeof(data
));
491 for (unsigned operand
= 0; operand
< this->get_num_operands(); operand
++) {
492 op
[operand
] = this->operands
[operand
]->constant_expression_value(variable_context
);
498 switch (this->operation
) {
499 case ir_binop_lshift
:
500 case ir_binop_rshift
:
502 case ir_binop_interpolate_at_offset
:
503 case ir_binop_interpolate_at_sample
:
504 case ir_binop_vector_extract
:
506 case ir_triop_bitfield_extract
:
510 assert(op
[0]->type
->base_type
== op
[1]->type
->base_type
);
514 bool op0_scalar
= op
[0]->type
->is_scalar();
515 bool op1_scalar
= op
[1] != NULL
&& op
[1]->type
->is_scalar();
517 /* When iterating over a vector or matrix's components, we want to increase
518 * the loop counter. However, for scalars, we want to stay at 0.
520 unsigned c0_inc
= op0_scalar
? 0 : 1;
521 unsigned c1_inc
= op1_scalar
? 0 : 1;
523 if (op1_scalar
|| !op
[1]) {
524 components
= op
[0]->type
->components();
526 components
= op
[1]->type
->components();
529 void *ctx
= ralloc_parent(this);
531 /* Handle array operations here, rather than below. */
532 if (op
[0]->type
->is_array()) {
533 assert(op
[1] != NULL
&& op
[1]->type
->is_array());
534 switch (this->operation
) {
535 case ir_binop_all_equal
:
536 return new(ctx
) ir_constant(op
[0]->has_value(op
[1]));
537 case ir_binop_any_nequal
:
538 return new(ctx
) ir_constant(!op
[0]->has_value(op
[1]));
545 switch (this->operation
) {
546 case ir_unop_bit_not
:
547 switch (op
[0]->type
->base_type
) {
549 for (unsigned c
= 0; c
< components
; c
++)
550 data
.i
[c
] = ~ op
[0]->value
.i
[c
];
553 for (unsigned c
= 0; c
< components
; c
++)
554 data
.u
[c
] = ~ op
[0]->value
.u
[c
];
561 case ir_unop_logic_not
:
562 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
563 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++)
564 data
.b
[c
] = !op
[0]->value
.b
[c
];
568 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
569 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
570 data
.i
[c
] = (int) op
[0]->value
.f
[c
];
574 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
575 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
576 data
.i
[c
] = (unsigned) op
[0]->value
.f
[c
];
580 assert(op
[0]->type
->base_type
== GLSL_TYPE_INT
);
581 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
582 data
.f
[c
] = (float) op
[0]->value
.i
[c
];
586 assert(op
[0]->type
->base_type
== GLSL_TYPE_UINT
);
587 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
588 data
.f
[c
] = (float) op
[0]->value
.u
[c
];
592 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
593 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
594 data
.f
[c
] = op
[0]->value
.b
[c
] ? 1.0F
: 0.0F
;
598 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
599 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
600 data
.b
[c
] = op
[0]->value
.f
[c
] != 0.0F
? true : false;
604 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
605 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
606 data
.i
[c
] = op
[0]->value
.b
[c
] ? 1 : 0;
610 assert(op
[0]->type
->is_integer());
611 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
612 data
.b
[c
] = op
[0]->value
.u
[c
] ? true : false;
616 assert(op
[0]->type
->base_type
== GLSL_TYPE_UINT
);
617 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
618 data
.i
[c
] = op
[0]->value
.u
[c
];
622 assert(op
[0]->type
->base_type
== GLSL_TYPE_INT
);
623 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
624 data
.u
[c
] = op
[0]->value
.i
[c
];
627 case ir_unop_bitcast_i2f
:
628 assert(op
[0]->type
->base_type
== GLSL_TYPE_INT
);
629 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
630 data
.f
[c
] = bitcast_u2f(op
[0]->value
.i
[c
]);
633 case ir_unop_bitcast_f2i
:
634 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
635 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
636 data
.i
[c
] = bitcast_f2u(op
[0]->value
.f
[c
]);
639 case ir_unop_bitcast_u2f
:
640 assert(op
[0]->type
->base_type
== GLSL_TYPE_UINT
);
641 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
642 data
.f
[c
] = bitcast_u2f(op
[0]->value
.u
[c
]);
645 case ir_unop_bitcast_f2u
:
646 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
647 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
648 data
.u
[c
] = bitcast_f2u(op
[0]->value
.f
[c
]);
652 assert(op
[0]->type
->base_type
== GLSL_TYPE_DOUBLE
);
653 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
654 data
.f
[c
] = op
[0]->value
.d
[c
];
658 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
659 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
660 data
.d
[c
] = op
[0]->value
.f
[c
];
664 assert(op
[0]->type
->base_type
== GLSL_TYPE_DOUBLE
);
665 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
666 data
.i
[c
] = op
[0]->value
.d
[c
];
670 assert(op
[0]->type
->base_type
== GLSL_TYPE_INT
);
671 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
672 data
.d
[c
] = op
[0]->value
.i
[c
];
676 assert(op
[0]->type
->base_type
== GLSL_TYPE_DOUBLE
);
677 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
678 data
.u
[c
] = op
[0]->value
.d
[c
];
682 assert(op
[0]->type
->base_type
== GLSL_TYPE_UINT
);
683 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
684 data
.d
[c
] = op
[0]->value
.u
[c
];
688 assert(op
[0]->type
->base_type
== GLSL_TYPE_DOUBLE
);
689 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
690 data
.b
[c
] = op
[0]->value
.d
[c
] != 0.0;
694 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
695 if (op
[0]->type
->base_type
== GLSL_TYPE_DOUBLE
)
696 data
.d
[c
] = trunc(op
[0]->value
.d
[c
]);
698 data
.f
[c
] = truncf(op
[0]->value
.f
[c
]);
702 case ir_unop_round_even
:
703 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
704 if (op
[0]->type
->base_type
== GLSL_TYPE_DOUBLE
)
705 data
.d
[c
] = _mesa_roundeven(op
[0]->value
.d
[c
]);
707 data
.f
[c
] = _mesa_roundevenf(op
[0]->value
.f
[c
]);
712 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
713 if (op
[0]->type
->base_type
== GLSL_TYPE_DOUBLE
)
714 data
.d
[c
] = ceil(op
[0]->value
.d
[c
]);
716 data
.f
[c
] = ceilf(op
[0]->value
.f
[c
]);
721 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
722 if (op
[0]->type
->base_type
== GLSL_TYPE_DOUBLE
)
723 data
.d
[c
] = floor(op
[0]->value
.d
[c
]);
725 data
.f
[c
] = floorf(op
[0]->value
.f
[c
]);
730 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
731 switch (this->type
->base_type
) {
732 case GLSL_TYPE_FLOAT
:
733 data
.f
[c
] = op
[0]->value
.f
[c
] - floor(op
[0]->value
.f
[c
]);
735 case GLSL_TYPE_DOUBLE
:
736 data
.d
[c
] = op
[0]->value
.d
[c
] - floor(op
[0]->value
.d
[c
]);
745 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
746 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
747 data
.f
[c
] = sinf(op
[0]->value
.f
[c
]);
752 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
753 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
754 data
.f
[c
] = cosf(op
[0]->value
.f
[c
]);
759 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
760 switch (this->type
->base_type
) {
762 data
.u
[c
] = -((int) op
[0]->value
.u
[c
]);
765 data
.i
[c
] = -op
[0]->value
.i
[c
];
767 case GLSL_TYPE_FLOAT
:
768 data
.f
[c
] = -op
[0]->value
.f
[c
];
770 case GLSL_TYPE_DOUBLE
:
771 data
.d
[c
] = -op
[0]->value
.d
[c
];
780 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
781 switch (this->type
->base_type
) {
783 data
.i
[c
] = op
[0]->value
.i
[c
];
785 data
.i
[c
] = -data
.i
[c
];
787 case GLSL_TYPE_FLOAT
:
788 data
.f
[c
] = fabs(op
[0]->value
.f
[c
]);
790 case GLSL_TYPE_DOUBLE
:
791 data
.d
[c
] = fabs(op
[0]->value
.d
[c
]);
800 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
801 switch (this->type
->base_type
) {
803 data
.i
[c
] = (op
[0]->value
.i
[c
] > 0) - (op
[0]->value
.i
[c
] < 0);
805 case GLSL_TYPE_FLOAT
:
806 data
.f
[c
] = float((op
[0]->value
.f
[c
] > 0)-(op
[0]->value
.f
[c
] < 0));
808 case GLSL_TYPE_DOUBLE
:
809 data
.d
[c
] = double((op
[0]->value
.d
[c
] > 0)-(op
[0]->value
.d
[c
] < 0));
818 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
819 switch (this->type
->base_type
) {
820 case GLSL_TYPE_FLOAT
:
821 if (op
[0]->value
.f
[c
] != 0.0)
822 data
.f
[c
] = 1.0F
/ op
[0]->value
.f
[c
];
824 case GLSL_TYPE_DOUBLE
:
825 if (op
[0]->value
.d
[c
] != 0.0)
826 data
.d
[c
] = 1.0 / op
[0]->value
.d
[c
];
835 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
836 if (op
[0]->type
->base_type
== GLSL_TYPE_DOUBLE
)
837 data
.d
[c
] = 1.0 / sqrt(op
[0]->value
.d
[c
]);
839 data
.f
[c
] = 1.0F
/ sqrtf(op
[0]->value
.f
[c
]);
844 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
845 if (op
[0]->type
->base_type
== GLSL_TYPE_DOUBLE
)
846 data
.d
[c
] = sqrt(op
[0]->value
.d
[c
]);
848 data
.f
[c
] = sqrtf(op
[0]->value
.f
[c
]);
853 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
854 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
855 data
.f
[c
] = expf(op
[0]->value
.f
[c
]);
860 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
861 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
862 data
.f
[c
] = exp2f(op
[0]->value
.f
[c
]);
867 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
868 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
869 data
.f
[c
] = logf(op
[0]->value
.f
[c
]);
874 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
875 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
876 data
.f
[c
] = log2f(op
[0]->value
.f
[c
]);
881 case ir_unop_dFdx_coarse
:
882 case ir_unop_dFdx_fine
:
884 case ir_unop_dFdy_coarse
:
885 case ir_unop_dFdy_fine
:
886 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
887 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
892 case ir_unop_pack_snorm_2x16
:
893 assert(op
[0]->type
== glsl_type::vec2_type
);
894 data
.u
[0] = pack_2x16(pack_snorm_1x16
,
898 case ir_unop_pack_snorm_4x8
:
899 assert(op
[0]->type
== glsl_type::vec4_type
);
900 data
.u
[0] = pack_4x8(pack_snorm_1x8
,
906 case ir_unop_unpack_snorm_2x16
:
907 assert(op
[0]->type
== glsl_type::uint_type
);
908 unpack_2x16(unpack_snorm_1x16
,
910 &data
.f
[0], &data
.f
[1]);
912 case ir_unop_unpack_snorm_4x8
:
913 assert(op
[0]->type
== glsl_type::uint_type
);
914 unpack_4x8(unpack_snorm_1x8
,
916 &data
.f
[0], &data
.f
[1], &data
.f
[2], &data
.f
[3]);
918 case ir_unop_pack_unorm_2x16
:
919 assert(op
[0]->type
== glsl_type::vec2_type
);
920 data
.u
[0] = pack_2x16(pack_unorm_1x16
,
924 case ir_unop_pack_unorm_4x8
:
925 assert(op
[0]->type
== glsl_type::vec4_type
);
926 data
.u
[0] = pack_4x8(pack_unorm_1x8
,
932 case ir_unop_unpack_unorm_2x16
:
933 assert(op
[0]->type
== glsl_type::uint_type
);
934 unpack_2x16(unpack_unorm_1x16
,
936 &data
.f
[0], &data
.f
[1]);
938 case ir_unop_unpack_unorm_4x8
:
939 assert(op
[0]->type
== glsl_type::uint_type
);
940 unpack_4x8(unpack_unorm_1x8
,
942 &data
.f
[0], &data
.f
[1], &data
.f
[2], &data
.f
[3]);
944 case ir_unop_pack_half_2x16
:
945 assert(op
[0]->type
== glsl_type::vec2_type
);
946 data
.u
[0] = pack_2x16(pack_half_1x16
,
950 case ir_unop_unpack_half_2x16
:
951 assert(op
[0]->type
== glsl_type::uint_type
);
952 unpack_2x16(unpack_half_1x16
,
954 &data
.f
[0], &data
.f
[1]);
957 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
);
958 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
959 data
.f
[c
] = powf(op
[0]->value
.f
[c
], op
[1]->value
.f
[c
]);
964 if (op
[0]->type
->base_type
== GLSL_TYPE_DOUBLE
)
965 data
.d
[0] = dot_d(op
[0], op
[1]);
967 data
.f
[0] = dot_f(op
[0], op
[1]);
971 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
972 for (unsigned c
= 0, c0
= 0, c1
= 0;
974 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
976 switch (op
[0]->type
->base_type
) {
978 data
.u
[c
] = MIN2(op
[0]->value
.u
[c0
], op
[1]->value
.u
[c1
]);
981 data
.i
[c
] = MIN2(op
[0]->value
.i
[c0
], op
[1]->value
.i
[c1
]);
983 case GLSL_TYPE_FLOAT
:
984 data
.f
[c
] = MIN2(op
[0]->value
.f
[c0
], op
[1]->value
.f
[c1
]);
986 case GLSL_TYPE_DOUBLE
:
987 data
.d
[c
] = MIN2(op
[0]->value
.d
[c0
], op
[1]->value
.d
[c1
]);
996 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
997 for (unsigned c
= 0, c0
= 0, c1
= 0;
999 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1001 switch (op
[0]->type
->base_type
) {
1002 case GLSL_TYPE_UINT
:
1003 data
.u
[c
] = MAX2(op
[0]->value
.u
[c0
], op
[1]->value
.u
[c1
]);
1006 data
.i
[c
] = MAX2(op
[0]->value
.i
[c0
], op
[1]->value
.i
[c1
]);
1008 case GLSL_TYPE_FLOAT
:
1009 data
.f
[c
] = MAX2(op
[0]->value
.f
[c0
], op
[1]->value
.f
[c1
]);
1011 case GLSL_TYPE_DOUBLE
:
1012 data
.d
[c
] = MAX2(op
[0]->value
.d
[c0
], op
[1]->value
.d
[c1
]);
1021 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
1022 for (unsigned c
= 0, c0
= 0, c1
= 0;
1024 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1026 switch (op
[0]->type
->base_type
) {
1027 case GLSL_TYPE_UINT
:
1028 data
.u
[c
] = op
[0]->value
.u
[c0
] + op
[1]->value
.u
[c1
];
1031 data
.i
[c
] = op
[0]->value
.i
[c0
] + op
[1]->value
.i
[c1
];
1033 case GLSL_TYPE_FLOAT
:
1034 data
.f
[c
] = op
[0]->value
.f
[c0
] + op
[1]->value
.f
[c1
];
1036 case GLSL_TYPE_DOUBLE
:
1037 data
.d
[c
] = op
[0]->value
.d
[c0
] + op
[1]->value
.d
[c1
];
1046 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
1047 for (unsigned c
= 0, c0
= 0, c1
= 0;
1049 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1051 switch (op
[0]->type
->base_type
) {
1052 case GLSL_TYPE_UINT
:
1053 data
.u
[c
] = op
[0]->value
.u
[c0
] - op
[1]->value
.u
[c1
];
1056 data
.i
[c
] = op
[0]->value
.i
[c0
] - op
[1]->value
.i
[c1
];
1058 case GLSL_TYPE_FLOAT
:
1059 data
.f
[c
] = op
[0]->value
.f
[c0
] - op
[1]->value
.f
[c1
];
1061 case GLSL_TYPE_DOUBLE
:
1062 data
.d
[c
] = op
[0]->value
.d
[c0
] - op
[1]->value
.d
[c1
];
1071 /* Check for equal types, or unequal types involving scalars */
1072 if ((op
[0]->type
== op
[1]->type
&& !op
[0]->type
->is_matrix())
1073 || op0_scalar
|| op1_scalar
) {
1074 for (unsigned c
= 0, c0
= 0, c1
= 0;
1076 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1078 switch (op
[0]->type
->base_type
) {
1079 case GLSL_TYPE_UINT
:
1080 data
.u
[c
] = op
[0]->value
.u
[c0
] * op
[1]->value
.u
[c1
];
1083 data
.i
[c
] = op
[0]->value
.i
[c0
] * op
[1]->value
.i
[c1
];
1085 case GLSL_TYPE_FLOAT
:
1086 data
.f
[c
] = op
[0]->value
.f
[c0
] * op
[1]->value
.f
[c1
];
1088 case GLSL_TYPE_DOUBLE
:
1089 data
.d
[c
] = op
[0]->value
.d
[c0
] * op
[1]->value
.d
[c1
];
1096 assert(op
[0]->type
->is_matrix() || op
[1]->type
->is_matrix());
1098 /* Multiply an N-by-M matrix with an M-by-P matrix. Since either
1099 * matrix can be a GLSL vector, either N or P can be 1.
1101 * For vec*mat, the vector is treated as a row vector. This
1102 * means the vector is a 1-row x M-column matrix.
1104 * For mat*vec, the vector is treated as a column vector. Since
1105 * matrix_columns is 1 for vectors, this just works.
1107 const unsigned n
= op
[0]->type
->is_vector()
1108 ? 1 : op
[0]->type
->vector_elements
;
1109 const unsigned m
= op
[1]->type
->vector_elements
;
1110 const unsigned p
= op
[1]->type
->matrix_columns
;
1111 for (unsigned j
= 0; j
< p
; j
++) {
1112 for (unsigned i
= 0; i
< n
; i
++) {
1113 for (unsigned k
= 0; k
< m
; k
++) {
1114 if (op
[0]->type
->base_type
== GLSL_TYPE_DOUBLE
)
1115 data
.d
[i
+n
*j
] += op
[0]->value
.d
[i
+n
*k
]*op
[1]->value
.d
[k
+m
*j
];
1117 data
.f
[i
+n
*j
] += op
[0]->value
.f
[i
+n
*k
]*op
[1]->value
.f
[k
+m
*j
];
1125 /* FINISHME: Emit warning when division-by-zero is detected. */
1126 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
1127 for (unsigned c
= 0, c0
= 0, c1
= 0;
1129 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1131 switch (op
[0]->type
->base_type
) {
1132 case GLSL_TYPE_UINT
:
1133 if (op
[1]->value
.u
[c1
] == 0) {
1136 data
.u
[c
] = op
[0]->value
.u
[c0
] / op
[1]->value
.u
[c1
];
1140 if (op
[1]->value
.i
[c1
] == 0) {
1143 data
.i
[c
] = op
[0]->value
.i
[c0
] / op
[1]->value
.i
[c1
];
1146 case GLSL_TYPE_FLOAT
:
1147 data
.f
[c
] = op
[0]->value
.f
[c0
] / op
[1]->value
.f
[c1
];
1149 case GLSL_TYPE_DOUBLE
:
1150 data
.d
[c
] = op
[0]->value
.d
[c0
] / op
[1]->value
.d
[c1
];
1159 /* FINISHME: Emit warning when division-by-zero is detected. */
1160 assert(op
[0]->type
== op
[1]->type
|| op0_scalar
|| op1_scalar
);
1161 for (unsigned c
= 0, c0
= 0, c1
= 0;
1163 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1165 switch (op
[0]->type
->base_type
) {
1166 case GLSL_TYPE_UINT
:
1167 if (op
[1]->value
.u
[c1
] == 0) {
1170 data
.u
[c
] = op
[0]->value
.u
[c0
] % op
[1]->value
.u
[c1
];
1174 if (op
[1]->value
.i
[c1
] == 0) {
1177 data
.i
[c
] = op
[0]->value
.i
[c0
] % op
[1]->value
.i
[c1
];
1180 case GLSL_TYPE_FLOAT
:
1181 /* We don't use fmod because it rounds toward zero; GLSL specifies
1184 data
.f
[c
] = op
[0]->value
.f
[c0
] - op
[1]->value
.f
[c1
]
1185 * floorf(op
[0]->value
.f
[c0
] / op
[1]->value
.f
[c1
]);
1187 case GLSL_TYPE_DOUBLE
:
1188 /* We don't use fmod because it rounds toward zero; GLSL specifies
1191 data
.d
[c
] = op
[0]->value
.d
[c0
] - op
[1]->value
.d
[c1
]
1192 * floor(op
[0]->value
.d
[c0
] / op
[1]->value
.d
[c1
]);
1201 case ir_binop_logic_and
:
1202 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
1203 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++)
1204 data
.b
[c
] = op
[0]->value
.b
[c
] && op
[1]->value
.b
[c
];
1206 case ir_binop_logic_xor
:
1207 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
1208 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++)
1209 data
.b
[c
] = op
[0]->value
.b
[c
] ^ op
[1]->value
.b
[c
];
1211 case ir_binop_logic_or
:
1212 assert(op
[0]->type
->base_type
== GLSL_TYPE_BOOL
);
1213 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++)
1214 data
.b
[c
] = op
[0]->value
.b
[c
] || op
[1]->value
.b
[c
];
1218 assert(op
[0]->type
== op
[1]->type
);
1219 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
1220 switch (op
[0]->type
->base_type
) {
1221 case GLSL_TYPE_UINT
:
1222 data
.b
[c
] = op
[0]->value
.u
[c
] < op
[1]->value
.u
[c
];
1225 data
.b
[c
] = op
[0]->value
.i
[c
] < op
[1]->value
.i
[c
];
1227 case GLSL_TYPE_FLOAT
:
1228 data
.b
[c
] = op
[0]->value
.f
[c
] < op
[1]->value
.f
[c
];
1230 case GLSL_TYPE_DOUBLE
:
1231 data
.b
[c
] = op
[0]->value
.d
[c
] < op
[1]->value
.d
[c
];
1238 case ir_binop_greater
:
1239 assert(op
[0]->type
== op
[1]->type
);
1240 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
1241 switch (op
[0]->type
->base_type
) {
1242 case GLSL_TYPE_UINT
:
1243 data
.b
[c
] = op
[0]->value
.u
[c
] > op
[1]->value
.u
[c
];
1246 data
.b
[c
] = op
[0]->value
.i
[c
] > op
[1]->value
.i
[c
];
1248 case GLSL_TYPE_FLOAT
:
1249 data
.b
[c
] = op
[0]->value
.f
[c
] > op
[1]->value
.f
[c
];
1251 case GLSL_TYPE_DOUBLE
:
1252 data
.b
[c
] = op
[0]->value
.d
[c
] > op
[1]->value
.d
[c
];
1259 case ir_binop_lequal
:
1260 assert(op
[0]->type
== op
[1]->type
);
1261 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
1262 switch (op
[0]->type
->base_type
) {
1263 case GLSL_TYPE_UINT
:
1264 data
.b
[c
] = op
[0]->value
.u
[c
] <= op
[1]->value
.u
[c
];
1267 data
.b
[c
] = op
[0]->value
.i
[c
] <= op
[1]->value
.i
[c
];
1269 case GLSL_TYPE_FLOAT
:
1270 data
.b
[c
] = op
[0]->value
.f
[c
] <= op
[1]->value
.f
[c
];
1272 case GLSL_TYPE_DOUBLE
:
1273 data
.b
[c
] = op
[0]->value
.d
[c
] <= op
[1]->value
.d
[c
];
1280 case ir_binop_gequal
:
1281 assert(op
[0]->type
== op
[1]->type
);
1282 for (unsigned c
= 0; c
< op
[0]->type
->components(); c
++) {
1283 switch (op
[0]->type
->base_type
) {
1284 case GLSL_TYPE_UINT
:
1285 data
.b
[c
] = op
[0]->value
.u
[c
] >= op
[1]->value
.u
[c
];
1288 data
.b
[c
] = op
[0]->value
.i
[c
] >= op
[1]->value
.i
[c
];
1290 case GLSL_TYPE_FLOAT
:
1291 data
.b
[c
] = op
[0]->value
.f
[c
] >= op
[1]->value
.f
[c
];
1293 case GLSL_TYPE_DOUBLE
:
1294 data
.b
[c
] = op
[0]->value
.d
[c
] >= op
[1]->value
.d
[c
];
1301 case ir_binop_equal
:
1302 assert(op
[0]->type
== op
[1]->type
);
1303 for (unsigned c
= 0; c
< components
; c
++) {
1304 switch (op
[0]->type
->base_type
) {
1305 case GLSL_TYPE_UINT
:
1306 data
.b
[c
] = op
[0]->value
.u
[c
] == op
[1]->value
.u
[c
];
1309 data
.b
[c
] = op
[0]->value
.i
[c
] == op
[1]->value
.i
[c
];
1311 case GLSL_TYPE_FLOAT
:
1312 data
.b
[c
] = op
[0]->value
.f
[c
] == op
[1]->value
.f
[c
];
1314 case GLSL_TYPE_BOOL
:
1315 data
.b
[c
] = op
[0]->value
.b
[c
] == op
[1]->value
.b
[c
];
1317 case GLSL_TYPE_DOUBLE
:
1318 data
.b
[c
] = op
[0]->value
.d
[c
] == op
[1]->value
.d
[c
];
1325 case ir_binop_nequal
:
1326 assert(op
[0]->type
== op
[1]->type
);
1327 for (unsigned c
= 0; c
< components
; c
++) {
1328 switch (op
[0]->type
->base_type
) {
1329 case GLSL_TYPE_UINT
:
1330 data
.b
[c
] = op
[0]->value
.u
[c
] != op
[1]->value
.u
[c
];
1333 data
.b
[c
] = op
[0]->value
.i
[c
] != op
[1]->value
.i
[c
];
1335 case GLSL_TYPE_FLOAT
:
1336 data
.b
[c
] = op
[0]->value
.f
[c
] != op
[1]->value
.f
[c
];
1338 case GLSL_TYPE_BOOL
:
1339 data
.b
[c
] = op
[0]->value
.b
[c
] != op
[1]->value
.b
[c
];
1341 case GLSL_TYPE_DOUBLE
:
1342 data
.b
[c
] = op
[0]->value
.d
[c
] != op
[1]->value
.d
[c
];
1349 case ir_binop_all_equal
:
1350 data
.b
[0] = op
[0]->has_value(op
[1]);
1352 case ir_binop_any_nequal
:
1353 data
.b
[0] = !op
[0]->has_value(op
[1]);
1356 case ir_binop_lshift
:
1357 for (unsigned c
= 0, c0
= 0, c1
= 0;
1359 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1361 if (op
[0]->type
->base_type
== GLSL_TYPE_INT
&&
1362 op
[1]->type
->base_type
== GLSL_TYPE_INT
) {
1363 data
.i
[c
] = op
[0]->value
.i
[c0
] << op
[1]->value
.i
[c1
];
1365 } else if (op
[0]->type
->base_type
== GLSL_TYPE_INT
&&
1366 op
[1]->type
->base_type
== GLSL_TYPE_UINT
) {
1367 data
.i
[c
] = op
[0]->value
.i
[c0
] << op
[1]->value
.u
[c1
];
1369 } else if (op
[0]->type
->base_type
== GLSL_TYPE_UINT
&&
1370 op
[1]->type
->base_type
== GLSL_TYPE_INT
) {
1371 data
.u
[c
] = op
[0]->value
.u
[c0
] << op
[1]->value
.i
[c1
];
1373 } else if (op
[0]->type
->base_type
== GLSL_TYPE_UINT
&&
1374 op
[1]->type
->base_type
== GLSL_TYPE_UINT
) {
1375 data
.u
[c
] = op
[0]->value
.u
[c0
] << op
[1]->value
.u
[c1
];
1380 case ir_binop_rshift
:
1381 for (unsigned c
= 0, c0
= 0, c1
= 0;
1383 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1385 if (op
[0]->type
->base_type
== GLSL_TYPE_INT
&&
1386 op
[1]->type
->base_type
== GLSL_TYPE_INT
) {
1387 data
.i
[c
] = op
[0]->value
.i
[c0
] >> op
[1]->value
.i
[c1
];
1389 } else if (op
[0]->type
->base_type
== GLSL_TYPE_INT
&&
1390 op
[1]->type
->base_type
== GLSL_TYPE_UINT
) {
1391 data
.i
[c
] = op
[0]->value
.i
[c0
] >> op
[1]->value
.u
[c1
];
1393 } else if (op
[0]->type
->base_type
== GLSL_TYPE_UINT
&&
1394 op
[1]->type
->base_type
== GLSL_TYPE_INT
) {
1395 data
.u
[c
] = op
[0]->value
.u
[c0
] >> op
[1]->value
.i
[c1
];
1397 } else if (op
[0]->type
->base_type
== GLSL_TYPE_UINT
&&
1398 op
[1]->type
->base_type
== GLSL_TYPE_UINT
) {
1399 data
.u
[c
] = op
[0]->value
.u
[c0
] >> op
[1]->value
.u
[c1
];
1404 case ir_binop_bit_and
:
1405 for (unsigned c
= 0, c0
= 0, c1
= 0;
1407 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1409 switch (op
[0]->type
->base_type
) {
1411 data
.i
[c
] = op
[0]->value
.i
[c0
] & op
[1]->value
.i
[c1
];
1413 case GLSL_TYPE_UINT
:
1414 data
.u
[c
] = op
[0]->value
.u
[c0
] & op
[1]->value
.u
[c1
];
1422 case ir_binop_bit_or
:
1423 for (unsigned c
= 0, c0
= 0, c1
= 0;
1425 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1427 switch (op
[0]->type
->base_type
) {
1429 data
.i
[c
] = op
[0]->value
.i
[c0
] | op
[1]->value
.i
[c1
];
1431 case GLSL_TYPE_UINT
:
1432 data
.u
[c
] = op
[0]->value
.u
[c0
] | op
[1]->value
.u
[c1
];
1440 case ir_binop_vector_extract
: {
1441 const int c
= CLAMP(op
[1]->value
.i
[0], 0,
1442 (int) op
[0]->type
->vector_elements
- 1);
1444 switch (op
[0]->type
->base_type
) {
1445 case GLSL_TYPE_UINT
:
1446 data
.u
[0] = op
[0]->value
.u
[c
];
1449 data
.i
[0] = op
[0]->value
.i
[c
];
1451 case GLSL_TYPE_FLOAT
:
1452 data
.f
[0] = op
[0]->value
.f
[c
];
1454 case GLSL_TYPE_DOUBLE
:
1455 data
.d
[0] = op
[0]->value
.d
[c
];
1457 case GLSL_TYPE_BOOL
:
1458 data
.b
[0] = op
[0]->value
.b
[c
];
1466 case ir_binop_bit_xor
:
1467 for (unsigned c
= 0, c0
= 0, c1
= 0;
1469 c0
+= c0_inc
, c1
+= c1_inc
, c
++) {
1471 switch (op
[0]->type
->base_type
) {
1473 data
.i
[c
] = op
[0]->value
.i
[c0
] ^ op
[1]->value
.i
[c1
];
1475 case GLSL_TYPE_UINT
:
1476 data
.u
[c
] = op
[0]->value
.u
[c0
] ^ op
[1]->value
.u
[c1
];
1484 case ir_unop_bitfield_reverse
:
1485 /* http://graphics.stanford.edu/~seander/bithacks.html#BitReverseObvious */
1486 for (unsigned c
= 0; c
< components
; c
++) {
1487 unsigned int v
= op
[0]->value
.u
[c
]; // input bits to be reversed
1488 unsigned int r
= v
; // r will be reversed bits of v; first get LSB of v
1489 int s
= sizeof(v
) * CHAR_BIT
- 1; // extra shift needed at end
1491 for (v
>>= 1; v
; v
>>= 1) {
1496 r
<<= s
; // shift when v's highest bits are zero
1502 case ir_unop_bit_count
:
1503 for (unsigned c
= 0; c
< components
; c
++) {
1505 unsigned v
= op
[0]->value
.u
[c
];
1507 for (; v
; count
++) {
1514 case ir_unop_find_msb
:
1515 for (unsigned c
= 0; c
< components
; c
++) {
1516 int v
= op
[0]->value
.i
[c
];
1518 if (v
== 0 || (op
[0]->type
->base_type
== GLSL_TYPE_INT
&& v
== -1))
1522 unsigned top_bit
= op
[0]->type
->base_type
== GLSL_TYPE_UINT
1523 ? 0 : v
& (1u << 31);
1525 while (((v
& (1u << 31)) == top_bit
) && count
!= 32) {
1530 data
.i
[c
] = 31 - count
;
1535 case ir_unop_find_lsb
:
1536 for (unsigned c
= 0; c
< components
; c
++) {
1537 if (op
[0]->value
.i
[c
] == 0)
1541 unsigned v
= op
[0]->value
.u
[c
];
1543 for (; !(v
& 1); v
>>= 1) {
1551 case ir_unop_saturate
:
1552 for (unsigned c
= 0; c
< components
; c
++) {
1553 data
.f
[c
] = CLAMP(op
[0]->value
.f
[c
], 0.0f
, 1.0f
);
1556 case ir_unop_pack_double_2x32
:
1557 /* XXX needs to be checked on big-endian */
1558 memcpy(&data
.d
[0], &op
[0]->value
.u
[0], sizeof(double));
1560 case ir_unop_unpack_double_2x32
:
1561 /* XXX needs to be checked on big-endian */
1562 memcpy(&data
.u
[0], &op
[0]->value
.d
[0], sizeof(double));
1565 case ir_triop_bitfield_extract
: {
1566 for (unsigned c
= 0; c
< components
; c
++) {
1567 int offset
= op
[1]->value
.i
[c
];
1568 int bits
= op
[2]->value
.i
[c
];
1572 else if (offset
< 0 || bits
< 0)
1573 data
.u
[c
] = 0; /* Undefined, per spec. */
1574 else if (offset
+ bits
> 32)
1575 data
.u
[c
] = 0; /* Undefined, per spec. */
1577 if (op
[0]->type
->base_type
== GLSL_TYPE_INT
) {
1578 /* int so that the right shift will sign-extend. */
1579 int value
= op
[0]->value
.i
[c
];
1580 value
<<= 32 - bits
- offset
;
1581 value
>>= 32 - bits
;
1584 unsigned value
= op
[0]->value
.u
[c
];
1585 value
<<= 32 - bits
- offset
;
1586 value
>>= 32 - bits
;
1594 case ir_binop_ldexp
:
1595 for (unsigned c
= 0; c
< components
; c
++) {
1596 if (op
[0]->type
->base_type
== GLSL_TYPE_DOUBLE
) {
1597 data
.d
[c
] = ldexp(op
[0]->value
.d
[c
], op
[1]->value
.i
[c
]);
1598 /* Flush subnormal values to zero. */
1599 if (!isnormal(data
.d
[c
]))
1600 data
.d
[c
] = copysign(0.0, op
[0]->value
.d
[c
]);
1602 data
.f
[c
] = ldexpf(op
[0]->value
.f
[c
], op
[1]->value
.i
[c
]);
1603 /* Flush subnormal values to zero. */
1604 if (!isnormal(data
.f
[c
]))
1605 data
.f
[c
] = copysignf(0.0f
, op
[0]->value
.f
[c
]);
1611 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
||
1612 op
[0]->type
->base_type
== GLSL_TYPE_DOUBLE
);
1613 assert(op
[1]->type
->base_type
== GLSL_TYPE_FLOAT
||
1614 op
[1]->type
->base_type
== GLSL_TYPE_DOUBLE
);
1615 assert(op
[2]->type
->base_type
== GLSL_TYPE_FLOAT
||
1616 op
[2]->type
->base_type
== GLSL_TYPE_DOUBLE
);
1618 for (unsigned c
= 0; c
< components
; c
++) {
1619 if (op
[0]->type
->base_type
== GLSL_TYPE_DOUBLE
)
1620 data
.d
[c
] = op
[0]->value
.d
[c
] * op
[1]->value
.d
[c
]
1621 + op
[2]->value
.d
[c
];
1623 data
.f
[c
] = op
[0]->value
.f
[c
] * op
[1]->value
.f
[c
]
1624 + op
[2]->value
.f
[c
];
1628 case ir_triop_lrp
: {
1629 assert(op
[0]->type
->base_type
== GLSL_TYPE_FLOAT
||
1630 op
[0]->type
->base_type
== GLSL_TYPE_DOUBLE
);
1631 assert(op
[1]->type
->base_type
== GLSL_TYPE_FLOAT
||
1632 op
[1]->type
->base_type
== GLSL_TYPE_DOUBLE
);
1633 assert(op
[2]->type
->base_type
== GLSL_TYPE_FLOAT
||
1634 op
[2]->type
->base_type
== GLSL_TYPE_DOUBLE
);
1636 unsigned c2_inc
= op
[2]->type
->is_scalar() ? 0 : 1;
1637 for (unsigned c
= 0, c2
= 0; c
< components
; c2
+= c2_inc
, c
++) {
1638 if (op
[0]->type
->base_type
== GLSL_TYPE_DOUBLE
)
1639 data
.d
[c
] = op
[0]->value
.d
[c
] * (1.0 - op
[2]->value
.d
[c2
]) +
1640 (op
[1]->value
.d
[c
] * op
[2]->value
.d
[c2
]);
1642 data
.f
[c
] = op
[0]->value
.f
[c
] * (1.0f
- op
[2]->value
.f
[c2
]) +
1643 (op
[1]->value
.f
[c
] * op
[2]->value
.f
[c2
]);
1649 for (unsigned c
= 0; c
< components
; c
++) {
1650 if (op
[1]->type
->base_type
== GLSL_TYPE_DOUBLE
)
1651 data
.d
[c
] = op
[0]->value
.b
[c
] ? op
[1]->value
.d
[c
]
1652 : op
[2]->value
.d
[c
];
1654 data
.u
[c
] = op
[0]->value
.b
[c
] ? op
[1]->value
.u
[c
]
1655 : op
[2]->value
.u
[c
];
1659 case ir_triop_vector_insert
: {
1660 const unsigned idx
= op
[2]->value
.u
[0];
1662 memcpy(&data
, &op
[0]->value
, sizeof(data
));
1664 switch (this->type
->base_type
) {
1666 data
.i
[idx
] = op
[1]->value
.i
[0];
1668 case GLSL_TYPE_UINT
:
1669 data
.u
[idx
] = op
[1]->value
.u
[0];
1671 case GLSL_TYPE_FLOAT
:
1672 data
.f
[idx
] = op
[1]->value
.f
[0];
1674 case GLSL_TYPE_BOOL
:
1675 data
.b
[idx
] = op
[1]->value
.b
[0];
1677 case GLSL_TYPE_DOUBLE
:
1678 data
.d
[idx
] = op
[1]->value
.d
[0];
1681 assert(!"Should not get here.");
1687 case ir_quadop_bitfield_insert
: {
1688 for (unsigned c
= 0; c
< components
; c
++) {
1689 int offset
= op
[2]->value
.i
[c
];
1690 int bits
= op
[3]->value
.i
[c
];
1693 data
.u
[c
] = op
[0]->value
.u
[c
];
1694 else if (offset
< 0 || bits
< 0)
1695 data
.u
[c
] = 0; /* Undefined, per spec. */
1696 else if (offset
+ bits
> 32)
1697 data
.u
[c
] = 0; /* Undefined, per spec. */
1699 unsigned insert_mask
= ((1ull << bits
) - 1) << offset
;
1701 unsigned insert
= op
[1]->value
.u
[c
];
1703 insert
&= insert_mask
;
1705 unsigned base
= op
[0]->value
.u
[c
];
1706 base
&= ~insert_mask
;
1708 data
.u
[c
] = base
| insert
;
1714 case ir_quadop_vector
:
1715 for (unsigned c
= 0; c
< this->type
->vector_elements
; c
++) {
1716 switch (this->type
->base_type
) {
1718 data
.i
[c
] = op
[c
]->value
.i
[0];
1720 case GLSL_TYPE_UINT
:
1721 data
.u
[c
] = op
[c
]->value
.u
[0];
1723 case GLSL_TYPE_FLOAT
:
1724 data
.f
[c
] = op
[c
]->value
.f
[0];
1726 case GLSL_TYPE_DOUBLE
:
1727 data
.d
[c
] = op
[c
]->value
.d
[0];
1736 /* FINISHME: Should handle all expression types. */
1740 return new(ctx
) ir_constant(this->type
, &data
);
1745 ir_texture::constant_expression_value(struct hash_table
*)
1747 /* texture lookups aren't constant expressions */
1753 ir_swizzle::constant_expression_value(struct hash_table
*variable_context
)
1755 ir_constant
*v
= this->val
->constant_expression_value(variable_context
);
1758 ir_constant_data data
= { { 0 } };
1760 const unsigned swiz_idx
[4] = {
1761 this->mask
.x
, this->mask
.y
, this->mask
.z
, this->mask
.w
1764 for (unsigned i
= 0; i
< this->mask
.num_components
; i
++) {
1765 switch (v
->type
->base_type
) {
1766 case GLSL_TYPE_UINT
:
1767 case GLSL_TYPE_INT
: data
.u
[i
] = v
->value
.u
[swiz_idx
[i
]]; break;
1768 case GLSL_TYPE_FLOAT
: data
.f
[i
] = v
->value
.f
[swiz_idx
[i
]]; break;
1769 case GLSL_TYPE_BOOL
: data
.b
[i
] = v
->value
.b
[swiz_idx
[i
]]; break;
1770 case GLSL_TYPE_DOUBLE
:data
.d
[i
] = v
->value
.d
[swiz_idx
[i
]]; break;
1771 default: assert(!"Should not get here."); break;
1775 void *ctx
= ralloc_parent(this);
1776 return new(ctx
) ir_constant(this->type
, &data
);
1783 ir_dereference_variable::constant_expression_value(struct hash_table
*variable_context
)
1787 /* Give priority to the context hashtable, if it exists */
1788 if (variable_context
) {
1789 ir_constant
*value
= (ir_constant
*)hash_table_find(variable_context
, var
);
1794 /* The constant_value of a uniform variable is its initializer,
1795 * not the lifetime constant value of the uniform.
1797 if (var
->data
.mode
== ir_var_uniform
)
1800 if (!var
->constant_value
)
1803 return var
->constant_value
->clone(ralloc_parent(var
), NULL
);
1808 ir_dereference_array::constant_expression_value(struct hash_table
*variable_context
)
1810 ir_constant
*array
= this->array
->constant_expression_value(variable_context
);
1811 ir_constant
*idx
= this->array_index
->constant_expression_value(variable_context
);
1813 if ((array
!= NULL
) && (idx
!= NULL
)) {
1814 void *ctx
= ralloc_parent(this);
1815 if (array
->type
->is_matrix()) {
1816 /* Array access of a matrix results in a vector.
1818 const unsigned column
= idx
->value
.u
[0];
1820 const glsl_type
*const column_type
= array
->type
->column_type();
1822 /* Offset in the constant matrix to the first element of the column
1825 const unsigned mat_idx
= column
* column_type
->vector_elements
;
1827 ir_constant_data data
= { { 0 } };
1829 switch (column_type
->base_type
) {
1830 case GLSL_TYPE_UINT
:
1832 for (unsigned i
= 0; i
< column_type
->vector_elements
; i
++)
1833 data
.u
[i
] = array
->value
.u
[mat_idx
+ i
];
1837 case GLSL_TYPE_FLOAT
:
1838 for (unsigned i
= 0; i
< column_type
->vector_elements
; i
++)
1839 data
.f
[i
] = array
->value
.f
[mat_idx
+ i
];
1843 case GLSL_TYPE_DOUBLE
:
1844 for (unsigned i
= 0; i
< column_type
->vector_elements
; i
++)
1845 data
.d
[i
] = array
->value
.d
[mat_idx
+ i
];
1850 assert(!"Should not get here.");
1854 return new(ctx
) ir_constant(column_type
, &data
);
1855 } else if (array
->type
->is_vector()) {
1856 const unsigned component
= idx
->value
.u
[0];
1858 return new(ctx
) ir_constant(array
, component
);
1860 const unsigned index
= idx
->value
.u
[0];
1861 return array
->get_array_element(index
)->clone(ctx
, NULL
);
1869 ir_dereference_record::constant_expression_value(struct hash_table
*)
1871 ir_constant
*v
= this->record
->constant_expression_value();
1873 return (v
!= NULL
) ? v
->get_record_field(this->field
) : NULL
;
1878 ir_assignment::constant_expression_value(struct hash_table
*)
1880 /* FINISHME: Handle CEs involving assignment (return RHS) */
1886 ir_constant::constant_expression_value(struct hash_table
*)
1893 ir_call::constant_expression_value(struct hash_table
*variable_context
)
1895 return this->callee
->constant_expression_value(&this->actual_parameters
, variable_context
);
1899 bool ir_function_signature::constant_expression_evaluate_expression_list(const struct exec_list
&body
,
1900 struct hash_table
*variable_context
,
1901 ir_constant
**result
)
1903 foreach_in_list(ir_instruction
, inst
, &body
) {
1904 switch(inst
->ir_type
) {
1906 /* (declare () type symbol) */
1907 case ir_type_variable
: {
1908 ir_variable
*var
= inst
->as_variable();
1909 hash_table_insert(variable_context
, ir_constant::zero(this, var
->type
), var
);
1913 /* (assign [condition] (write-mask) (ref) (value)) */
1914 case ir_type_assignment
: {
1915 ir_assignment
*asg
= inst
->as_assignment();
1916 if (asg
->condition
) {
1917 ir_constant
*cond
= asg
->condition
->constant_expression_value(variable_context
);
1920 if (!cond
->get_bool_component(0))
1924 ir_constant
*store
= NULL
;
1927 if (!constant_referenced(asg
->lhs
, variable_context
, store
, offset
))
1930 ir_constant
*value
= asg
->rhs
->constant_expression_value(variable_context
);
1935 store
->copy_masked_offset(value
, offset
, asg
->write_mask
);
1939 /* (return (expression)) */
1940 case ir_type_return
:
1942 *result
= inst
->as_return()->value
->constant_expression_value(variable_context
);
1943 return *result
!= NULL
;
1945 /* (call name (ref) (params))*/
1946 case ir_type_call
: {
1947 ir_call
*call
= inst
->as_call();
1949 /* Just say no to void functions in constant expressions. We
1950 * don't need them at that point.
1953 if (!call
->return_deref
)
1956 ir_constant
*store
= NULL
;
1959 if (!constant_referenced(call
->return_deref
, variable_context
,
1963 ir_constant
*value
= call
->constant_expression_value(variable_context
);
1968 store
->copy_offset(value
, offset
);
1972 /* (if condition (then-instructions) (else-instructions)) */
1974 ir_if
*iif
= inst
->as_if();
1976 ir_constant
*cond
= iif
->condition
->constant_expression_value(variable_context
);
1977 if (!cond
|| !cond
->type
->is_boolean())
1980 exec_list
&branch
= cond
->get_bool_component(0) ? iif
->then_instructions
: iif
->else_instructions
;
1983 if (!constant_expression_evaluate_expression_list(branch
, variable_context
, result
))
1986 /* If there was a return in the branch chosen, drop out now. */
1993 /* Every other expression type, we drop out. */
1999 /* Reaching the end of the block is not an error condition */
2007 ir_function_signature::constant_expression_value(exec_list
*actual_parameters
, struct hash_table
*variable_context
)
2009 const glsl_type
*type
= this->return_type
;
2010 if (type
== glsl_type::void_type
)
2013 /* From the GLSL 1.20 spec, page 23:
2014 * "Function calls to user-defined functions (non-built-in functions)
2015 * cannot be used to form constant expressions."
2017 if (!this->is_builtin())
2021 * Of the builtin functions, only the texture lookups and the noise
2022 * ones must not be used in constant expressions. They all include
2023 * specific opcodes so they don't need to be special-cased at this
2027 /* Initialize the table of dereferencable names with the function
2028 * parameters. Verify their const-ness on the way.
2030 * We expect the correctness of the number of parameters to have
2031 * been checked earlier.
2033 hash_table
*deref_hash
= hash_table_ctor(8, hash_table_pointer_hash
,
2034 hash_table_pointer_compare
);
2036 /* If "origin" is non-NULL, then the function body is there. So we
2037 * have to use the variable objects from the object with the body,
2038 * but the parameter instanciation on the current object.
2040 const exec_node
*parameter_info
= origin
? origin
->parameters
.get_head_raw() : parameters
.get_head_raw();
2042 foreach_in_list(ir_rvalue
, n
, actual_parameters
) {
2043 ir_constant
*constant
= n
->constant_expression_value(variable_context
);
2044 if (constant
== NULL
) {
2045 hash_table_dtor(deref_hash
);
2050 ir_variable
*var
= (ir_variable
*)parameter_info
;
2051 hash_table_insert(deref_hash
, constant
, var
);
2053 parameter_info
= parameter_info
->next
;
2056 ir_constant
*result
= NULL
;
2058 /* Now run the builtin function until something non-constant
2059 * happens or we get the result.
2061 if (constant_expression_evaluate_expression_list(origin
? origin
->body
: body
, deref_hash
, &result
) && result
)
2062 result
= result
->clone(ralloc_parent(this), NULL
);
2064 hash_table_dtor(deref_hash
);