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.
24 #include "main/core.h" /* for MAX2 */
26 #include "ir_visitor.h"
27 #include "glsl_types.h"
29 ir_rvalue::ir_rvalue()
31 this->type
= glsl_type::error_type
;
34 bool ir_rvalue::is_zero() const
39 bool ir_rvalue::is_one() const
44 bool ir_rvalue::is_negative_one() const
50 * Modify the swizzle make to move one component to another
52 * \param m IR swizzle to be modified
53 * \param from Component in the RHS that is to be swizzled
54 * \param to Desired swizzle location of \c from
57 update_rhs_swizzle(ir_swizzle_mask
&m
, unsigned from
, unsigned to
)
60 case 0: m
.x
= from
; break;
61 case 1: m
.y
= from
; break;
62 case 2: m
.z
= from
; break;
63 case 3: m
.w
= from
; break;
64 default: assert(!"Should not get here.");
67 m
.num_components
= MAX2(m
.num_components
, (to
+ 1));
71 ir_assignment::set_lhs(ir_rvalue
*lhs
)
74 bool swizzled
= false;
77 ir_swizzle
*swiz
= lhs
->as_swizzle();
82 unsigned write_mask
= 0;
83 ir_swizzle_mask rhs_swiz
= { 0, 0, 0, 0, 0, 0 };
85 for (unsigned i
= 0; i
< swiz
->mask
.num_components
; i
++) {
89 case 0: c
= swiz
->mask
.x
; break;
90 case 1: c
= swiz
->mask
.y
; break;
91 case 2: c
= swiz
->mask
.z
; break;
92 case 3: c
= swiz
->mask
.w
; break;
93 default: assert(!"Should not get here.");
96 write_mask
|= (((this->write_mask
>> i
) & 1) << c
);
97 update_rhs_swizzle(rhs_swiz
, i
, c
);
100 this->write_mask
= write_mask
;
103 this->rhs
= new(mem_ctx
) ir_swizzle(this->rhs
, rhs_swiz
);
108 /* Now, RHS channels line up with the LHS writemask. Collapse it
109 * to just the channels that will be written.
111 ir_swizzle_mask rhs_swiz
= { 0, 0, 0, 0, 0, 0 };
113 for (int i
= 0; i
< 4; i
++) {
114 if (write_mask
& (1 << i
))
115 update_rhs_swizzle(rhs_swiz
, i
, rhs_chan
++);
117 this->rhs
= new(mem_ctx
) ir_swizzle(this->rhs
, rhs_swiz
);
120 assert((lhs
== NULL
) || lhs
->as_dereference());
122 this->lhs
= (ir_dereference
*) lhs
;
126 ir_assignment::whole_variable_written()
128 ir_variable
*v
= this->lhs
->whole_variable_referenced();
133 if (v
->type
->is_scalar())
136 if (v
->type
->is_vector()) {
137 const unsigned mask
= (1U << v
->type
->vector_elements
) - 1;
139 if (mask
!= this->write_mask
)
143 /* Either all the vector components are assigned or the variable is some
144 * composite type (and the whole thing is assigned.
149 ir_assignment::ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
,
150 ir_rvalue
*condition
, unsigned write_mask
)
152 this->ir_type
= ir_type_assignment
;
153 this->condition
= condition
;
156 this->write_mask
= write_mask
;
158 if (lhs
->type
->is_scalar() || lhs
->type
->is_vector()) {
159 int lhs_components
= 0;
160 for (int i
= 0; i
< 4; i
++) {
161 if (write_mask
& (1 << i
))
165 assert(lhs_components
== this->rhs
->type
->vector_elements
);
169 ir_assignment::ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
,
170 ir_rvalue
*condition
)
172 this->ir_type
= ir_type_assignment
;
173 this->condition
= condition
;
176 /* If the RHS is a vector type, assume that all components of the vector
177 * type are being written to the LHS. The write mask comes from the RHS
178 * because we can have a case where the LHS is a vec4 and the RHS is a
179 * vec3. In that case, the assignment is:
181 * (assign (...) (xyz) (var_ref lhs) (var_ref rhs))
183 if (rhs
->type
->is_vector())
184 this->write_mask
= (1U << rhs
->type
->vector_elements
) - 1;
185 else if (rhs
->type
->is_scalar())
186 this->write_mask
= 1;
188 this->write_mask
= 0;
194 ir_expression::ir_expression(int op
, const struct glsl_type
*type
,
197 assert(get_num_operands(ir_expression_operation(op
)) == 1);
198 this->ir_type
= ir_type_expression
;
200 this->operation
= ir_expression_operation(op
);
201 this->operands
[0] = op0
;
202 this->operands
[1] = NULL
;
203 this->operands
[2] = NULL
;
204 this->operands
[3] = NULL
;
207 ir_expression::ir_expression(int op
, const struct glsl_type
*type
,
208 ir_rvalue
*op0
, ir_rvalue
*op1
)
210 assert(((op1
== NULL
) && (get_num_operands(ir_expression_operation(op
)) == 1))
211 || (get_num_operands(ir_expression_operation(op
)) == 2));
212 this->ir_type
= ir_type_expression
;
214 this->operation
= ir_expression_operation(op
);
215 this->operands
[0] = op0
;
216 this->operands
[1] = op1
;
217 this->operands
[2] = NULL
;
218 this->operands
[3] = NULL
;
221 ir_expression::ir_expression(int op
, const struct glsl_type
*type
,
222 ir_rvalue
*op0
, ir_rvalue
*op1
,
223 ir_rvalue
*op2
, ir_rvalue
*op3
)
225 this->ir_type
= ir_type_expression
;
227 this->operation
= ir_expression_operation(op
);
228 this->operands
[0] = op0
;
229 this->operands
[1] = op1
;
230 this->operands
[2] = op2
;
231 this->operands
[3] = op3
;
234 ir_expression::ir_expression(int op
, ir_rvalue
*op0
)
236 this->ir_type
= ir_type_expression
;
238 this->operation
= ir_expression_operation(op
);
239 this->operands
[0] = op0
;
240 this->operands
[1] = NULL
;
241 this->operands
[2] = NULL
;
242 this->operands
[3] = NULL
;
244 assert(op
<= ir_last_unop
);
246 switch (this->operation
) {
247 case ir_unop_bit_not
:
248 case ir_unop_logic_not
:
263 case ir_unop_round_even
:
266 case ir_unop_sin_reduced
:
267 case ir_unop_cos_reduced
:
270 this->type
= op0
->type
;
275 this->type
= glsl_type::get_instance(GLSL_TYPE_INT
,
276 op0
->type
->vector_elements
, 1);
282 this->type
= glsl_type::get_instance(GLSL_TYPE_FLOAT
,
283 op0
->type
->vector_elements
, 1);
288 this->type
= glsl_type::get_instance(GLSL_TYPE_BOOL
,
289 op0
->type
->vector_elements
, 1);
293 this->type
= glsl_type::float_type
;
297 this->type
= glsl_type::bool_type
;
301 assert(!"not reached: missing automatic type setup for ir_expression");
302 this->type
= op0
->type
;
307 ir_expression::ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
)
309 this->ir_type
= ir_type_expression
;
311 this->operation
= ir_expression_operation(op
);
312 this->operands
[0] = op0
;
313 this->operands
[1] = op1
;
314 this->operands
[2] = NULL
;
315 this->operands
[3] = NULL
;
317 assert(op
> ir_last_unop
);
319 switch (this->operation
) {
320 case ir_binop_all_equal
:
321 case ir_binop_any_nequal
:
322 this->type
= glsl_type::bool_type
;
333 if (op0
->type
->is_scalar()) {
334 this->type
= op1
->type
;
335 } else if (op1
->type
->is_scalar()) {
336 this->type
= op0
->type
;
338 /* FINISHME: matrix types */
339 assert(!op0
->type
->is_matrix() && !op1
->type
->is_matrix());
340 assert(op0
->type
== op1
->type
);
341 this->type
= op0
->type
;
345 case ir_binop_logic_and
:
346 case ir_binop_logic_xor
:
347 case ir_binop_logic_or
:
348 case ir_binop_bit_and
:
349 case ir_binop_bit_xor
:
350 case ir_binop_bit_or
:
351 if (op0
->type
->is_scalar()) {
352 this->type
= op1
->type
;
353 } else if (op1
->type
->is_scalar()) {
354 this->type
= op0
->type
;
359 case ir_binop_nequal
:
360 case ir_binop_lequal
:
361 case ir_binop_gequal
:
363 case ir_binop_greater
:
364 assert(op0
->type
== op1
->type
);
365 this->type
= glsl_type::get_instance(GLSL_TYPE_BOOL
,
366 op0
->type
->vector_elements
, 1);
370 this->type
= glsl_type::float_type
;
373 case ir_binop_lshift
:
374 case ir_binop_rshift
:
375 this->type
= op0
->type
;
379 assert(!"not reached: missing automatic type setup for ir_expression");
380 this->type
= glsl_type::float_type
;
385 ir_expression::get_num_operands(ir_expression_operation op
)
387 assert(op
<= ir_last_opcode
);
389 if (op
<= ir_last_unop
)
392 if (op
<= ir_last_binop
)
395 if (op
== ir_quadop_vector
)
402 static const char *const operator_strs
[] = {
463 const char *ir_expression::operator_string(ir_expression_operation op
)
465 assert((unsigned int) op
< Elements(operator_strs
));
466 assert(Elements(operator_strs
) == (ir_quadop_vector
+ 1));
467 return operator_strs
[op
];
470 const char *ir_expression::operator_string()
472 return operator_string(this->operation
);
475 ir_expression_operation
476 ir_expression::get_operator(const char *str
)
478 const int operator_count
= sizeof(operator_strs
) / sizeof(operator_strs
[0]);
479 for (int op
= 0; op
< operator_count
; op
++) {
480 if (strcmp(str
, operator_strs
[op
]) == 0)
481 return (ir_expression_operation
) op
;
483 return (ir_expression_operation
) -1;
486 ir_constant::ir_constant()
488 this->ir_type
= ir_type_constant
;
491 ir_constant::ir_constant(const struct glsl_type
*type
,
492 const ir_constant_data
*data
)
494 assert((type
->base_type
>= GLSL_TYPE_UINT
)
495 && (type
->base_type
<= GLSL_TYPE_BOOL
));
497 this->ir_type
= ir_type_constant
;
499 memcpy(& this->value
, data
, sizeof(this->value
));
502 ir_constant::ir_constant(float f
)
504 this->ir_type
= ir_type_constant
;
505 this->type
= glsl_type::float_type
;
506 this->value
.f
[0] = f
;
507 for (int i
= 1; i
< 16; i
++) {
508 this->value
.f
[i
] = 0;
512 ir_constant::ir_constant(unsigned int u
)
514 this->ir_type
= ir_type_constant
;
515 this->type
= glsl_type::uint_type
;
516 this->value
.u
[0] = u
;
517 for (int i
= 1; i
< 16; i
++) {
518 this->value
.u
[i
] = 0;
522 ir_constant::ir_constant(int i
)
524 this->ir_type
= ir_type_constant
;
525 this->type
= glsl_type::int_type
;
526 this->value
.i
[0] = i
;
527 for (int i
= 1; i
< 16; i
++) {
528 this->value
.i
[i
] = 0;
532 ir_constant::ir_constant(bool b
)
534 this->ir_type
= ir_type_constant
;
535 this->type
= glsl_type::bool_type
;
536 this->value
.b
[0] = b
;
537 for (int i
= 1; i
< 16; i
++) {
538 this->value
.b
[i
] = false;
542 ir_constant::ir_constant(const ir_constant
*c
, unsigned i
)
544 this->ir_type
= ir_type_constant
;
545 this->type
= c
->type
->get_base_type();
547 switch (this->type
->base_type
) {
548 case GLSL_TYPE_UINT
: this->value
.u
[0] = c
->value
.u
[i
]; break;
549 case GLSL_TYPE_INT
: this->value
.i
[0] = c
->value
.i
[i
]; break;
550 case GLSL_TYPE_FLOAT
: this->value
.f
[0] = c
->value
.f
[i
]; break;
551 case GLSL_TYPE_BOOL
: this->value
.b
[0] = c
->value
.b
[i
]; break;
552 default: assert(!"Should not get here."); break;
556 ir_constant::ir_constant(const struct glsl_type
*type
, exec_list
*value_list
)
558 this->ir_type
= ir_type_constant
;
561 assert(type
->is_scalar() || type
->is_vector() || type
->is_matrix()
562 || type
->is_record() || type
->is_array());
564 if (type
->is_array()) {
565 this->array_elements
= talloc_array(this, ir_constant
*, type
->length
);
567 foreach_list(node
, value_list
) {
568 ir_constant
*value
= (ir_constant
*) node
;
569 assert(value
->as_constant() != NULL
);
571 this->array_elements
[i
++] = value
;
576 /* If the constant is a record, the types of each of the entries in
577 * value_list must be a 1-for-1 match with the structure components. Each
578 * entry must also be a constant. Just move the nodes from the value_list
579 * to the list in the ir_constant.
581 /* FINISHME: Should there be some type checking and / or assertions here? */
582 /* FINISHME: Should the new constant take ownership of the nodes from
583 * FINISHME: value_list, or should it make copies?
585 if (type
->is_record()) {
586 value_list
->move_nodes_to(& this->components
);
590 for (unsigned i
= 0; i
< 16; i
++) {
591 this->value
.u
[i
] = 0;
594 ir_constant
*value
= (ir_constant
*) (value_list
->head
);
596 /* Constructors with exactly one scalar argument are special for vectors
597 * and matrices. For vectors, the scalar value is replicated to fill all
598 * the components. For matrices, the scalar fills the components of the
599 * diagonal while the rest is filled with 0.
601 if (value
->type
->is_scalar() && value
->next
->is_tail_sentinel()) {
602 if (type
->is_matrix()) {
603 /* Matrix - fill diagonal (rest is already set to 0) */
604 assert(type
->base_type
== GLSL_TYPE_FLOAT
);
605 for (unsigned i
= 0; i
< type
->matrix_columns
; i
++)
606 this->value
.f
[i
* type
->vector_elements
+ i
] = value
->value
.f
[0];
608 /* Vector or scalar - fill all components */
609 switch (type
->base_type
) {
612 for (unsigned i
= 0; i
< type
->components(); i
++)
613 this->value
.u
[i
] = value
->value
.u
[0];
615 case GLSL_TYPE_FLOAT
:
616 for (unsigned i
= 0; i
< type
->components(); i
++)
617 this->value
.f
[i
] = value
->value
.f
[0];
620 for (unsigned i
= 0; i
< type
->components(); i
++)
621 this->value
.b
[i
] = value
->value
.b
[0];
624 assert(!"Should not get here.");
631 if (type
->is_matrix() && value
->type
->is_matrix()) {
632 assert(value
->next
->is_tail_sentinel());
634 /* From section 5.4.2 of the GLSL 1.20 spec:
635 * "If a matrix is constructed from a matrix, then each component
636 * (column i, row j) in the result that has a corresponding component
637 * (column i, row j) in the argument will be initialized from there."
639 unsigned cols
= MIN2(type
->matrix_columns
, value
->type
->matrix_columns
);
640 unsigned rows
= MIN2(type
->vector_elements
, value
->type
->vector_elements
);
641 for (unsigned i
= 0; i
< cols
; i
++) {
642 for (unsigned j
= 0; j
< rows
; j
++) {
643 const unsigned src
= i
* value
->type
->vector_elements
+ j
;
644 const unsigned dst
= i
* type
->vector_elements
+ j
;
645 this->value
.f
[dst
] = value
->value
.f
[src
];
649 /* "All other components will be initialized to the identity matrix." */
650 for (unsigned i
= cols
; i
< type
->matrix_columns
; i
++)
651 this->value
.f
[i
* type
->vector_elements
+ i
] = 1.0;
656 /* Use each component from each entry in the value_list to initialize one
657 * component of the constant being constructed.
659 for (unsigned i
= 0; i
< type
->components(); /* empty */) {
660 assert(value
->as_constant() != NULL
);
661 assert(!value
->is_tail_sentinel());
663 for (unsigned j
= 0; j
< value
->type
->components(); j
++) {
664 switch (type
->base_type
) {
666 this->value
.u
[i
] = value
->get_uint_component(j
);
669 this->value
.i
[i
] = value
->get_int_component(j
);
671 case GLSL_TYPE_FLOAT
:
672 this->value
.f
[i
] = value
->get_float_component(j
);
675 this->value
.b
[i
] = value
->get_bool_component(j
);
678 /* FINISHME: What to do? Exceptions are not the answer.
684 if (i
>= type
->components())
688 value
= (ir_constant
*) value
->next
;
693 ir_constant::zero(void *mem_ctx
, const glsl_type
*type
)
695 assert(type
->is_numeric() || type
->is_boolean());
697 ir_constant
*c
= new(mem_ctx
) ir_constant
;
699 memset(&c
->value
, 0, sizeof(c
->value
));
705 ir_constant::get_bool_component(unsigned i
) const
707 switch (this->type
->base_type
) {
708 case GLSL_TYPE_UINT
: return this->value
.u
[i
] != 0;
709 case GLSL_TYPE_INT
: return this->value
.i
[i
] != 0;
710 case GLSL_TYPE_FLOAT
: return ((int)this->value
.f
[i
]) != 0;
711 case GLSL_TYPE_BOOL
: return this->value
.b
[i
];
712 default: assert(!"Should not get here."); break;
715 /* Must return something to make the compiler happy. This is clearly an
722 ir_constant::get_float_component(unsigned i
) const
724 switch (this->type
->base_type
) {
725 case GLSL_TYPE_UINT
: return (float) this->value
.u
[i
];
726 case GLSL_TYPE_INT
: return (float) this->value
.i
[i
];
727 case GLSL_TYPE_FLOAT
: return this->value
.f
[i
];
728 case GLSL_TYPE_BOOL
: return this->value
.b
[i
] ? 1.0 : 0.0;
729 default: assert(!"Should not get here."); break;
732 /* Must return something to make the compiler happy. This is clearly an
739 ir_constant::get_int_component(unsigned i
) const
741 switch (this->type
->base_type
) {
742 case GLSL_TYPE_UINT
: return this->value
.u
[i
];
743 case GLSL_TYPE_INT
: return this->value
.i
[i
];
744 case GLSL_TYPE_FLOAT
: return (int) this->value
.f
[i
];
745 case GLSL_TYPE_BOOL
: return this->value
.b
[i
] ? 1 : 0;
746 default: assert(!"Should not get here."); break;
749 /* Must return something to make the compiler happy. This is clearly an
756 ir_constant::get_uint_component(unsigned i
) const
758 switch (this->type
->base_type
) {
759 case GLSL_TYPE_UINT
: return this->value
.u
[i
];
760 case GLSL_TYPE_INT
: return this->value
.i
[i
];
761 case GLSL_TYPE_FLOAT
: return (unsigned) this->value
.f
[i
];
762 case GLSL_TYPE_BOOL
: return this->value
.b
[i
] ? 1 : 0;
763 default: assert(!"Should not get here."); break;
766 /* Must return something to make the compiler happy. This is clearly an
773 ir_constant::get_array_element(unsigned i
) const
775 assert(this->type
->is_array());
777 /* From page 35 (page 41 of the PDF) of the GLSL 1.20 spec:
779 * "Behavior is undefined if a shader subscripts an array with an index
780 * less than 0 or greater than or equal to the size the array was
783 * Most out-of-bounds accesses are removed before things could get this far.
784 * There are cases where non-constant array index values can get constant
789 else if (i
>= this->type
->length
)
790 i
= this->type
->length
- 1;
792 return array_elements
[i
];
796 ir_constant::get_record_field(const char *name
)
798 int idx
= this->type
->field_index(name
);
803 if (this->components
.is_empty())
806 exec_node
*node
= this->components
.head
;
807 for (int i
= 0; i
< idx
; i
++) {
810 /* If the end of the list is encountered before the element matching the
811 * requested field is found, return NULL.
813 if (node
->is_tail_sentinel())
817 return (ir_constant
*) node
;
822 ir_constant::has_value(const ir_constant
*c
) const
824 if (this->type
!= c
->type
)
827 if (this->type
->is_array()) {
828 for (unsigned i
= 0; i
< this->type
->length
; i
++) {
829 if (!this->array_elements
[i
]->has_value(c
->array_elements
[i
]))
835 if (this->type
->base_type
== GLSL_TYPE_STRUCT
) {
836 const exec_node
*a_node
= this->components
.head
;
837 const exec_node
*b_node
= c
->components
.head
;
839 while (!a_node
->is_tail_sentinel()) {
840 assert(!b_node
->is_tail_sentinel());
842 const ir_constant
*const a_field
= (ir_constant
*) a_node
;
843 const ir_constant
*const b_field
= (ir_constant
*) b_node
;
845 if (!a_field
->has_value(b_field
))
848 a_node
= a_node
->next
;
849 b_node
= b_node
->next
;
855 for (unsigned i
= 0; i
< this->type
->components(); i
++) {
856 switch (this->type
->base_type
) {
858 if (this->value
.u
[i
] != c
->value
.u
[i
])
862 if (this->value
.i
[i
] != c
->value
.i
[i
])
865 case GLSL_TYPE_FLOAT
:
866 if (this->value
.f
[i
] != c
->value
.f
[i
])
870 if (this->value
.b
[i
] != c
->value
.b
[i
])
874 assert(!"Should not get here.");
883 ir_constant::is_zero() const
885 if (!this->type
->is_scalar() && !this->type
->is_vector())
888 for (unsigned c
= 0; c
< this->type
->vector_elements
; c
++) {
889 switch (this->type
->base_type
) {
890 case GLSL_TYPE_FLOAT
:
891 if (this->value
.f
[c
] != 0.0)
895 if (this->value
.i
[c
] != 0)
899 if (this->value
.u
[c
] != 0)
903 if (this->value
.b
[c
] != false)
907 /* The only other base types are structures, arrays, and samplers.
908 * Samplers cannot be constants, and the others should have been
909 * filtered out above.
911 assert(!"Should not get here.");
920 ir_constant::is_one() const
922 if (!this->type
->is_scalar() && !this->type
->is_vector())
925 for (unsigned c
= 0; c
< this->type
->vector_elements
; c
++) {
926 switch (this->type
->base_type
) {
927 case GLSL_TYPE_FLOAT
:
928 if (this->value
.f
[c
] != 1.0)
932 if (this->value
.i
[c
] != 1)
936 if (this->value
.u
[c
] != 1)
940 if (this->value
.b
[c
] != true)
944 /* The only other base types are structures, arrays, and samplers.
945 * Samplers cannot be constants, and the others should have been
946 * filtered out above.
948 assert(!"Should not get here.");
957 ir_constant::is_negative_one() const
959 if (!this->type
->is_scalar() && !this->type
->is_vector())
962 if (this->type
->is_boolean())
965 for (unsigned c
= 0; c
< this->type
->vector_elements
; c
++) {
966 switch (this->type
->base_type
) {
967 case GLSL_TYPE_FLOAT
:
968 if (this->value
.f
[c
] != -1.0)
972 if (this->value
.i
[c
] != -1)
976 if (int(this->value
.u
[c
]) != -1)
980 /* The only other base types are structures, arrays, samplers, and
981 * booleans. Samplers cannot be constants, and the others should
982 * have been filtered out above.
984 assert(!"Should not get here.");
994 this->ir_type
= ir_type_loop
;
995 this->cmp
= ir_unop_neg
;
998 this->increment
= NULL
;
999 this->counter
= NULL
;
1003 ir_dereference_variable::ir_dereference_variable(ir_variable
*var
)
1005 this->ir_type
= ir_type_dereference_variable
;
1007 this->type
= (var
!= NULL
) ? var
->type
: glsl_type::error_type
;
1011 ir_dereference_array::ir_dereference_array(ir_rvalue
*value
,
1012 ir_rvalue
*array_index
)
1014 this->ir_type
= ir_type_dereference_array
;
1015 this->array_index
= array_index
;
1016 this->set_array(value
);
1020 ir_dereference_array::ir_dereference_array(ir_variable
*var
,
1021 ir_rvalue
*array_index
)
1023 void *ctx
= talloc_parent(var
);
1025 this->ir_type
= ir_type_dereference_array
;
1026 this->array_index
= array_index
;
1027 this->set_array(new(ctx
) ir_dereference_variable(var
));
1032 ir_dereference_array::set_array(ir_rvalue
*value
)
1034 this->array
= value
;
1035 this->type
= glsl_type::error_type
;
1037 if (this->array
!= NULL
) {
1038 const glsl_type
*const vt
= this->array
->type
;
1040 if (vt
->is_array()) {
1041 type
= vt
->element_type();
1042 } else if (vt
->is_matrix()) {
1043 type
= vt
->column_type();
1044 } else if (vt
->is_vector()) {
1045 type
= vt
->get_base_type();
1051 ir_dereference_record::ir_dereference_record(ir_rvalue
*value
,
1054 this->ir_type
= ir_type_dereference_record
;
1055 this->record
= value
;
1056 this->field
= talloc_strdup(this, field
);
1057 this->type
= (this->record
!= NULL
)
1058 ? this->record
->type
->field_type(field
) : glsl_type::error_type
;
1062 ir_dereference_record::ir_dereference_record(ir_variable
*var
,
1065 void *ctx
= talloc_parent(var
);
1067 this->ir_type
= ir_type_dereference_record
;
1068 this->record
= new(ctx
) ir_dereference_variable(var
);
1069 this->field
= talloc_strdup(this, field
);
1070 this->type
= (this->record
!= NULL
)
1071 ? this->record
->type
->field_type(field
) : glsl_type::error_type
;
1074 bool type_contains_sampler(const glsl_type
*type
)
1076 if (type
->is_array()) {
1077 return type_contains_sampler(type
->fields
.array
);
1078 } else if (type
->is_record()) {
1079 for (unsigned int i
= 0; i
< type
->length
; i
++) {
1080 if (type_contains_sampler(type
->fields
.structure
[i
].type
))
1085 return type
->is_sampler();
1090 ir_dereference::is_lvalue()
1092 ir_variable
*var
= this->variable_referenced();
1094 /* Every l-value derference chain eventually ends in a variable.
1096 if ((var
== NULL
) || var
->read_only
)
1099 if (this->type
->is_array() && !var
->array_lvalue
)
1102 /* From page 17 (page 23 of the PDF) of the GLSL 1.20 spec:
1104 * "Samplers cannot be treated as l-values; hence cannot be used
1105 * as out or inout function parameters, nor can they be
1108 if (type_contains_sampler(this->type
))
1115 const char *tex_opcode_strs
[] = { "tex", "txb", "txl", "txd", "txf" };
1117 const char *ir_texture::opcode_string()
1119 assert((unsigned int) op
<=
1120 sizeof(tex_opcode_strs
) / sizeof(tex_opcode_strs
[0]));
1121 return tex_opcode_strs
[op
];
1125 ir_texture::get_opcode(const char *str
)
1127 const int count
= sizeof(tex_opcode_strs
) / sizeof(tex_opcode_strs
[0]);
1128 for (int op
= 0; op
< count
; op
++) {
1129 if (strcmp(str
, tex_opcode_strs
[op
]) == 0)
1130 return (ir_texture_opcode
) op
;
1132 return (ir_texture_opcode
) -1;
1137 ir_texture::set_sampler(ir_dereference
*sampler
)
1139 assert(sampler
!= NULL
);
1140 this->sampler
= sampler
;
1142 switch (sampler
->type
->sampler_type
) {
1143 case GLSL_TYPE_FLOAT
:
1144 this->type
= glsl_type::vec4_type
;
1147 this->type
= glsl_type::ivec4_type
;
1149 case GLSL_TYPE_UINT
:
1150 this->type
= glsl_type::uvec4_type
;
1157 ir_swizzle::init_mask(const unsigned *comp
, unsigned count
)
1159 assert((count
>= 1) && (count
<= 4));
1161 memset(&this->mask
, 0, sizeof(this->mask
));
1162 this->mask
.num_components
= count
;
1164 unsigned dup_mask
= 0;
1167 assert(comp
[3] <= 3);
1168 dup_mask
|= (1U << comp
[3])
1169 & ((1U << comp
[0]) | (1U << comp
[1]) | (1U << comp
[2]));
1170 this->mask
.w
= comp
[3];
1173 assert(comp
[2] <= 3);
1174 dup_mask
|= (1U << comp
[2])
1175 & ((1U << comp
[0]) | (1U << comp
[1]));
1176 this->mask
.z
= comp
[2];
1179 assert(comp
[1] <= 3);
1180 dup_mask
|= (1U << comp
[1])
1181 & ((1U << comp
[0]));
1182 this->mask
.y
= comp
[1];
1185 assert(comp
[0] <= 3);
1186 this->mask
.x
= comp
[0];
1189 this->mask
.has_duplicates
= dup_mask
!= 0;
1191 /* Based on the number of elements in the swizzle and the base type
1192 * (i.e., float, int, unsigned, or bool) of the vector being swizzled,
1193 * generate the type of the resulting value.
1195 type
= glsl_type::get_instance(val
->type
->base_type
, mask
.num_components
, 1);
1198 ir_swizzle::ir_swizzle(ir_rvalue
*val
, unsigned x
, unsigned y
, unsigned z
,
1199 unsigned w
, unsigned count
)
1202 const unsigned components
[4] = { x
, y
, z
, w
};
1203 this->ir_type
= ir_type_swizzle
;
1204 this->init_mask(components
, count
);
1207 ir_swizzle::ir_swizzle(ir_rvalue
*val
, const unsigned *comp
,
1211 this->ir_type
= ir_type_swizzle
;
1212 this->init_mask(comp
, count
);
1215 ir_swizzle::ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
)
1217 this->ir_type
= ir_type_swizzle
;
1220 this->type
= glsl_type::get_instance(val
->type
->base_type
,
1221 mask
.num_components
, 1);
1230 ir_swizzle::create(ir_rvalue
*val
, const char *str
, unsigned vector_length
)
1232 void *ctx
= talloc_parent(val
);
1234 /* For each possible swizzle character, this table encodes the value in
1235 * \c idx_map that represents the 0th element of the vector. For invalid
1236 * swizzle characters (e.g., 'k'), a special value is used that will allow
1237 * detection of errors.
1239 static const unsigned char base_idx
[26] = {
1240 /* a b c d e f g h i j k l m */
1241 R
, R
, I
, I
, I
, I
, R
, I
, I
, I
, I
, I
, I
,
1242 /* n o p q r s t u v w x y z */
1243 I
, I
, S
, S
, R
, S
, S
, I
, I
, X
, X
, X
, X
1246 /* Each valid swizzle character has an entry in the previous table. This
1247 * table encodes the base index encoded in the previous table plus the actual
1248 * index of the swizzle character. When processing swizzles, the first
1249 * character in the string is indexed in the previous table. Each character
1250 * in the string is indexed in this table, and the value found there has the
1251 * value form the first table subtracted. The result must be on the range
1254 * For example, the string "wzyx" will get X from the first table. Each of
1255 * the charcaters will get X+3, X+2, X+1, and X+0 from this table. After
1256 * subtraction, the swizzle values are { 3, 2, 1, 0 }.
1258 * The string "wzrg" will get X from the first table. Each of the characters
1259 * will get X+3, X+2, R+0, and R+1 from this table. After subtraction, the
1260 * swizzle values are { 3, 2, 4, 5 }. Since 4 and 5 are outside the range
1261 * [0,3], the error is detected.
1263 static const unsigned char idx_map
[26] = {
1264 /* a b c d e f g h i j k l m */
1265 R
+3, R
+2, 0, 0, 0, 0, R
+1, 0, 0, 0, 0, 0, 0,
1266 /* n o p q r s t u v w x y z */
1267 0, 0, S
+2, S
+3, R
+0, S
+0, S
+1, 0, 0, X
+3, X
+0, X
+1, X
+2
1270 int swiz_idx
[4] = { 0, 0, 0, 0 };
1274 /* Validate the first character in the swizzle string and look up the base
1275 * index value as described above.
1277 if ((str
[0] < 'a') || (str
[0] > 'z'))
1280 const unsigned base
= base_idx
[str
[0] - 'a'];
1283 for (i
= 0; (i
< 4) && (str
[i
] != '\0'); i
++) {
1284 /* Validate the next character, and, as described above, convert it to a
1287 if ((str
[i
] < 'a') || (str
[i
] > 'z'))
1290 swiz_idx
[i
] = idx_map
[str
[i
] - 'a'] - base
;
1291 if ((swiz_idx
[i
] < 0) || (swiz_idx
[i
] >= (int) vector_length
))
1298 return new(ctx
) ir_swizzle(val
, swiz_idx
[0], swiz_idx
[1], swiz_idx
[2],
1308 ir_swizzle::variable_referenced()
1310 return this->val
->variable_referenced();
1314 ir_variable::ir_variable(const struct glsl_type
*type
, const char *name
,
1315 ir_variable_mode mode
)
1316 : max_array_access(0), read_only(false), centroid(false), invariant(false),
1317 mode(mode
), interpolation(ir_var_smooth
), array_lvalue(false)
1319 this->ir_type
= ir_type_variable
;
1321 this->name
= talloc_strdup(this, name
);
1322 this->explicit_location
= false;
1323 this->location
= -1;
1324 this->warn_extension
= NULL
;
1325 this->constant_value
= NULL
;
1326 this->origin_upper_left
= false;
1327 this->pixel_center_integer
= false;
1330 if (type
&& type
->base_type
== GLSL_TYPE_SAMPLER
)
1331 this->read_only
= true;
1336 ir_variable::interpolation_string() const
1338 switch (this->interpolation
) {
1339 case ir_var_smooth
: return "smooth";
1340 case ir_var_flat
: return "flat";
1341 case ir_var_noperspective
: return "noperspective";
1344 assert(!"Should not get here.");
1350 ir_variable::component_slots() const
1352 /* FINISHME: Sparsely accessed arrays require fewer slots. */
1353 return this->type
->component_slots();
1357 ir_function_signature::ir_function_signature(const glsl_type
*return_type
)
1358 : return_type(return_type
), is_defined(false), _function(NULL
)
1360 this->ir_type
= ir_type_function_signature
;
1361 this->is_builtin
= false;
1366 ir_function_signature::qualifiers_match(exec_list
*params
)
1368 exec_list_iterator iter_a
= parameters
.iterator();
1369 exec_list_iterator iter_b
= params
->iterator();
1371 /* check that the qualifiers match. */
1372 while (iter_a
.has_next()) {
1373 ir_variable
*a
= (ir_variable
*)iter_a
.get();
1374 ir_variable
*b
= (ir_variable
*)iter_b
.get();
1376 if (a
->read_only
!= b
->read_only
||
1377 a
->mode
!= b
->mode
||
1378 a
->interpolation
!= b
->interpolation
||
1379 a
->centroid
!= b
->centroid
) {
1381 /* parameter a's qualifiers don't match */
1393 ir_function_signature::replace_parameters(exec_list
*new_params
)
1395 /* Destroy all of the previous parameter information. If the previous
1396 * parameter information comes from the function prototype, it may either
1397 * specify incorrect parameter names or not have names at all.
1399 foreach_iter(exec_list_iterator
, iter
, parameters
) {
1400 assert(((ir_instruction
*) iter
.get())->as_variable() != NULL
);
1405 new_params
->move_nodes_to(¶meters
);
1409 ir_function::ir_function(const char *name
)
1411 this->ir_type
= ir_type_function
;
1412 this->name
= talloc_strdup(this, name
);
1417 ir_function::has_user_signature()
1419 foreach_list(n
, &this->signatures
) {
1420 ir_function_signature
*const sig
= (ir_function_signature
*) n
;
1421 if (!sig
->is_builtin
)
1429 ir_call::get_error_instruction(void *ctx
)
1431 ir_call
*call
= new(ctx
) ir_call
;
1433 call
->type
= glsl_type::error_type
;
1438 ir_call::set_callee(ir_function_signature
*sig
)
1440 assert((this->type
== NULL
) || (this->type
== sig
->return_type
));
1446 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
)
1448 foreach_iter(exec_list_iterator
, iter
, *list
) {
1449 ((ir_instruction
*)iter
.get())->accept(visitor
);
1455 steal_memory(ir_instruction
*ir
, void *new_ctx
)
1457 ir_variable
*var
= ir
->as_variable();
1458 ir_constant
*constant
= ir
->as_constant();
1459 if (var
!= NULL
&& var
->constant_value
!= NULL
)
1460 steal_memory(var
->constant_value
, ir
);
1462 /* The components of aggregate constants are not visited by the normal
1463 * visitor, so steal their values by hand.
1465 if (constant
!= NULL
) {
1466 if (constant
->type
->is_record()) {
1467 foreach_iter(exec_list_iterator
, iter
, constant
->components
) {
1468 ir_constant
*field
= (ir_constant
*)iter
.get();
1469 steal_memory(field
, ir
);
1471 } else if (constant
->type
->is_array()) {
1472 for (unsigned int i
= 0; i
< constant
->type
->length
; i
++) {
1473 steal_memory(constant
->array_elements
[i
], ir
);
1478 talloc_steal(new_ctx
, ir
);
1483 reparent_ir(exec_list
*list
, void *mem_ctx
)
1485 foreach_list(node
, list
) {
1486 visit_tree((ir_instruction
*) node
, steal_memory
, mem_ctx
);
1492 try_min_one(ir_rvalue
*ir
)
1494 ir_expression
*expr
= ir
->as_expression();
1496 if (!expr
|| expr
->operation
!= ir_binop_min
)
1499 if (expr
->operands
[0]->is_one())
1500 return expr
->operands
[1];
1502 if (expr
->operands
[1]->is_one())
1503 return expr
->operands
[0];
1509 try_max_zero(ir_rvalue
*ir
)
1511 ir_expression
*expr
= ir
->as_expression();
1513 if (!expr
|| expr
->operation
!= ir_binop_max
)
1516 if (expr
->operands
[0]->is_zero())
1517 return expr
->operands
[1];
1519 if (expr
->operands
[1]->is_zero())
1520 return expr
->operands
[0];
1526 ir_rvalue::as_rvalue_to_saturate()
1528 ir_expression
*expr
= this->as_expression();
1533 ir_rvalue
*max_zero
= try_max_zero(expr
);
1535 return try_min_one(max_zero
);
1537 ir_rvalue
*min_one
= try_min_one(expr
);
1539 return try_max_zero(min_one
);