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
;
35 * Modify the swizzle make to move one component to another
37 * \param m IR swizzle to be modified
38 * \param from Component in the RHS that is to be swizzled
39 * \param to Desired swizzle location of \c from
42 update_rhs_swizzle(ir_swizzle_mask
&m
, unsigned from
, unsigned to
)
45 case 0: m
.x
= from
; break;
46 case 1: m
.y
= from
; break;
47 case 2: m
.z
= from
; break;
48 case 3: m
.w
= from
; break;
49 default: assert(!"Should not get here.");
52 m
.num_components
= MAX2(m
.num_components
, (to
+ 1));
56 ir_assignment::set_lhs(ir_rvalue
*lhs
)
59 bool swizzled
= false;
62 ir_swizzle
*swiz
= lhs
->as_swizzle();
67 unsigned write_mask
= 0;
68 ir_swizzle_mask rhs_swiz
= { 0, 0, 0, 0, 0, 0 };
70 for (unsigned i
= 0; i
< swiz
->mask
.num_components
; i
++) {
74 case 0: c
= swiz
->mask
.x
; break;
75 case 1: c
= swiz
->mask
.y
; break;
76 case 2: c
= swiz
->mask
.z
; break;
77 case 3: c
= swiz
->mask
.w
; break;
78 default: assert(!"Should not get here.");
81 write_mask
|= (((this->write_mask
>> i
) & 1) << c
);
82 update_rhs_swizzle(rhs_swiz
, i
, c
);
85 this->write_mask
= write_mask
;
88 this->rhs
= new(mem_ctx
) ir_swizzle(this->rhs
, rhs_swiz
);
93 /* Now, RHS channels line up with the LHS writemask. Collapse it
94 * to just the channels that will be written.
96 ir_swizzle_mask rhs_swiz
= { 0, 0, 0, 0, 0, 0 };
98 for (int i
= 0; i
< 4; i
++) {
99 if (write_mask
& (1 << i
))
100 update_rhs_swizzle(rhs_swiz
, i
, rhs_chan
++);
102 this->rhs
= new(mem_ctx
) ir_swizzle(this->rhs
, rhs_swiz
);
105 assert((lhs
== NULL
) || lhs
->as_dereference());
107 this->lhs
= (ir_dereference
*) lhs
;
111 ir_assignment::whole_variable_written()
113 ir_variable
*v
= this->lhs
->whole_variable_referenced();
118 if (v
->type
->is_scalar())
121 if (v
->type
->is_vector()) {
122 const unsigned mask
= (1U << v
->type
->vector_elements
) - 1;
124 if (mask
!= this->write_mask
)
128 /* Either all the vector components are assigned or the variable is some
129 * composite type (and the whole thing is assigned.
134 ir_assignment::ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
,
135 ir_rvalue
*condition
, unsigned write_mask
)
137 this->ir_type
= ir_type_assignment
;
138 this->condition
= condition
;
141 this->write_mask
= write_mask
;
143 if (lhs
->type
->is_scalar() || lhs
->type
->is_vector()) {
144 int lhs_components
= 0;
145 for (int i
= 0; i
< 4; i
++) {
146 if (write_mask
& (1 << i
))
150 assert(lhs_components
== this->rhs
->type
->vector_elements
);
154 ir_assignment::ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
,
155 ir_rvalue
*condition
)
157 this->ir_type
= ir_type_assignment
;
158 this->condition
= condition
;
161 /* If the RHS is a vector type, assume that all components of the vector
162 * type are being written to the LHS. The write mask comes from the RHS
163 * because we can have a case where the LHS is a vec4 and the RHS is a
164 * vec3. In that case, the assignment is:
166 * (assign (...) (xyz) (var_ref lhs) (var_ref rhs))
168 if (rhs
->type
->is_vector())
169 this->write_mask
= (1U << rhs
->type
->vector_elements
) - 1;
170 else if (rhs
->type
->is_scalar())
171 this->write_mask
= 1;
173 this->write_mask
= 0;
179 ir_expression::ir_expression(int op
, const struct glsl_type
*type
,
180 ir_rvalue
*op0
, ir_rvalue
*op1
)
182 this->ir_type
= ir_type_expression
;
184 this->operation
= ir_expression_operation(op
);
185 this->operands
[0] = op0
;
186 this->operands
[1] = op1
;
190 ir_expression::get_num_operands(ir_expression_operation op
)
192 /* Update ir_print_visitor.cpp when updating this list. */
193 const int num_operands
[] = {
194 1, /* ir_unop_bit_not */
195 1, /* ir_unop_logic_not */
198 1, /* ir_unop_sign */
201 1, /* ir_unop_sqrt */
204 1, /* ir_unop_exp2 */
205 1, /* ir_unop_log2 */
215 1, /* ir_unop_trunc */
216 1, /* ir_unop_ceil */
217 1, /* ir_unop_floor */
218 1, /* ir_unop_fract */
219 1, /* ir_unop_round_even */
224 1, /* ir_unop_dFdx */
225 1, /* ir_unop_dFdy */
227 1, /* ir_unop_noise */
229 2, /* ir_binop_add */
230 2, /* ir_binop_sub */
231 2, /* ir_binop_mul */
232 2, /* ir_binop_div */
233 2, /* ir_binop_mod */
235 2, /* ir_binop_less */
236 2, /* ir_binop_greater */
237 2, /* ir_binop_lequal */
238 2, /* ir_binop_gequal */
239 2, /* ir_binop_equal */
240 2, /* ir_binop_nequal */
241 2, /* ir_binop_all_equal */
242 2, /* ir_binop_any_nequal */
244 2, /* ir_binop_lshift */
245 2, /* ir_binop_rshift */
246 2, /* ir_binop_bit_and */
247 2, /* ir_binop_bit_xor */
248 2, /* ir_binop_bit_or */
250 2, /* ir_binop_logic_and */
251 2, /* ir_binop_logic_xor */
252 2, /* ir_binop_logic_or */
254 2, /* ir_binop_dot */
255 2, /* ir_binop_min */
256 2, /* ir_binop_max */
258 2, /* ir_binop_pow */
261 assert(sizeof(num_operands
) / sizeof(num_operands
[0]) == ir_binop_pow
+ 1);
263 return num_operands
[op
];
266 static const char *const operator_strs
[] = {
324 const char *ir_expression::operator_string(ir_expression_operation op
)
326 assert((unsigned int) op
< Elements(operator_strs
));
327 assert(Elements(operator_strs
) == (ir_binop_pow
+ 1));
328 return operator_strs
[op
];
331 const char *ir_expression::operator_string()
333 return operator_string(this->operation
);
336 ir_expression_operation
337 ir_expression::get_operator(const char *str
)
339 const int operator_count
= sizeof(operator_strs
) / sizeof(operator_strs
[0]);
340 for (int op
= 0; op
< operator_count
; op
++) {
341 if (strcmp(str
, operator_strs
[op
]) == 0)
342 return (ir_expression_operation
) op
;
344 return (ir_expression_operation
) -1;
347 ir_constant::ir_constant()
349 this->ir_type
= ir_type_constant
;
352 ir_constant::ir_constant(const struct glsl_type
*type
,
353 const ir_constant_data
*data
)
355 assert((type
->base_type
>= GLSL_TYPE_UINT
)
356 && (type
->base_type
<= GLSL_TYPE_BOOL
));
358 this->ir_type
= ir_type_constant
;
360 memcpy(& this->value
, data
, sizeof(this->value
));
363 ir_constant::ir_constant(float f
)
365 this->ir_type
= ir_type_constant
;
366 this->type
= glsl_type::float_type
;
367 this->value
.f
[0] = f
;
368 for (int i
= 1; i
< 16; i
++) {
369 this->value
.f
[i
] = 0;
373 ir_constant::ir_constant(unsigned int u
)
375 this->ir_type
= ir_type_constant
;
376 this->type
= glsl_type::uint_type
;
377 this->value
.u
[0] = u
;
378 for (int i
= 1; i
< 16; i
++) {
379 this->value
.u
[i
] = 0;
383 ir_constant::ir_constant(int i
)
385 this->ir_type
= ir_type_constant
;
386 this->type
= glsl_type::int_type
;
387 this->value
.i
[0] = i
;
388 for (int i
= 1; i
< 16; i
++) {
389 this->value
.i
[i
] = 0;
393 ir_constant::ir_constant(bool b
)
395 this->ir_type
= ir_type_constant
;
396 this->type
= glsl_type::bool_type
;
397 this->value
.b
[0] = b
;
398 for (int i
= 1; i
< 16; i
++) {
399 this->value
.b
[i
] = false;
403 ir_constant::ir_constant(const ir_constant
*c
, unsigned i
)
405 this->ir_type
= ir_type_constant
;
406 this->type
= c
->type
->get_base_type();
408 switch (this->type
->base_type
) {
409 case GLSL_TYPE_UINT
: this->value
.u
[0] = c
->value
.u
[i
]; break;
410 case GLSL_TYPE_INT
: this->value
.i
[0] = c
->value
.i
[i
]; break;
411 case GLSL_TYPE_FLOAT
: this->value
.f
[0] = c
->value
.f
[i
]; break;
412 case GLSL_TYPE_BOOL
: this->value
.b
[0] = c
->value
.b
[i
]; break;
413 default: assert(!"Should not get here."); break;
417 ir_constant::ir_constant(const struct glsl_type
*type
, exec_list
*value_list
)
419 this->ir_type
= ir_type_constant
;
422 assert(type
->is_scalar() || type
->is_vector() || type
->is_matrix()
423 || type
->is_record() || type
->is_array());
425 if (type
->is_array()) {
426 this->array_elements
= talloc_array(this, ir_constant
*, type
->length
);
428 foreach_list(node
, value_list
) {
429 ir_constant
*value
= (ir_constant
*) node
;
430 assert(value
->as_constant() != NULL
);
432 this->array_elements
[i
++] = value
;
437 /* If the constant is a record, the types of each of the entries in
438 * value_list must be a 1-for-1 match with the structure components. Each
439 * entry must also be a constant. Just move the nodes from the value_list
440 * to the list in the ir_constant.
442 /* FINISHME: Should there be some type checking and / or assertions here? */
443 /* FINISHME: Should the new constant take ownership of the nodes from
444 * FINISHME: value_list, or should it make copies?
446 if (type
->is_record()) {
447 value_list
->move_nodes_to(& this->components
);
451 for (unsigned i
= 0; i
< 16; i
++) {
452 this->value
.u
[i
] = 0;
455 ir_constant
*value
= (ir_constant
*) (value_list
->head
);
457 /* Constructors with exactly one scalar argument are special for vectors
458 * and matrices. For vectors, the scalar value is replicated to fill all
459 * the components. For matrices, the scalar fills the components of the
460 * diagonal while the rest is filled with 0.
462 if (value
->type
->is_scalar() && value
->next
->is_tail_sentinel()) {
463 if (type
->is_matrix()) {
464 /* Matrix - fill diagonal (rest is already set to 0) */
465 assert(type
->base_type
== GLSL_TYPE_FLOAT
);
466 for (unsigned i
= 0; i
< type
->matrix_columns
; i
++)
467 this->value
.f
[i
* type
->vector_elements
+ i
] = value
->value
.f
[0];
469 /* Vector or scalar - fill all components */
470 switch (type
->base_type
) {
473 for (unsigned i
= 0; i
< type
->components(); i
++)
474 this->value
.u
[i
] = value
->value
.u
[0];
476 case GLSL_TYPE_FLOAT
:
477 for (unsigned i
= 0; i
< type
->components(); i
++)
478 this->value
.f
[i
] = value
->value
.f
[0];
481 for (unsigned i
= 0; i
< type
->components(); i
++)
482 this->value
.b
[i
] = value
->value
.b
[0];
485 assert(!"Should not get here.");
492 if (type
->is_matrix() && value
->type
->is_matrix()) {
493 assert(value
->next
->is_tail_sentinel());
495 /* From section 5.4.2 of the GLSL 1.20 spec:
496 * "If a matrix is constructed from a matrix, then each component
497 * (column i, row j) in the result that has a corresponding component
498 * (column i, row j) in the argument will be initialized from there."
500 unsigned cols
= MIN2(type
->matrix_columns
, value
->type
->matrix_columns
);
501 unsigned rows
= MIN2(type
->vector_elements
, value
->type
->vector_elements
);
502 for (unsigned i
= 0; i
< cols
; i
++) {
503 for (unsigned j
= 0; j
< rows
; j
++) {
504 const unsigned src
= i
* value
->type
->vector_elements
+ j
;
505 const unsigned dst
= i
* type
->vector_elements
+ j
;
506 this->value
.f
[dst
] = value
->value
.f
[src
];
510 /* "All other components will be initialized to the identity matrix." */
511 for (unsigned i
= cols
; i
< type
->matrix_columns
; i
++)
512 this->value
.f
[i
* type
->vector_elements
+ i
] = 1.0;
517 /* Use each component from each entry in the value_list to initialize one
518 * component of the constant being constructed.
520 for (unsigned i
= 0; i
< type
->components(); /* empty */) {
521 assert(value
->as_constant() != NULL
);
522 assert(!value
->is_tail_sentinel());
524 for (unsigned j
= 0; j
< value
->type
->components(); j
++) {
525 switch (type
->base_type
) {
527 this->value
.u
[i
] = value
->get_uint_component(j
);
530 this->value
.i
[i
] = value
->get_int_component(j
);
532 case GLSL_TYPE_FLOAT
:
533 this->value
.f
[i
] = value
->get_float_component(j
);
536 this->value
.b
[i
] = value
->get_bool_component(j
);
539 /* FINISHME: What to do? Exceptions are not the answer.
545 if (i
>= type
->components())
549 value
= (ir_constant
*) value
->next
;
554 ir_constant::zero(void *mem_ctx
, const glsl_type
*type
)
556 assert(type
->is_numeric() || type
->is_boolean());
558 ir_constant
*c
= new(mem_ctx
) ir_constant
;
560 memset(&c
->value
, 0, sizeof(c
->value
));
566 ir_constant::get_bool_component(unsigned i
) const
568 switch (this->type
->base_type
) {
569 case GLSL_TYPE_UINT
: return this->value
.u
[i
] != 0;
570 case GLSL_TYPE_INT
: return this->value
.i
[i
] != 0;
571 case GLSL_TYPE_FLOAT
: return ((int)this->value
.f
[i
]) != 0;
572 case GLSL_TYPE_BOOL
: return this->value
.b
[i
];
573 default: assert(!"Should not get here."); break;
576 /* Must return something to make the compiler happy. This is clearly an
583 ir_constant::get_float_component(unsigned i
) const
585 switch (this->type
->base_type
) {
586 case GLSL_TYPE_UINT
: return (float) this->value
.u
[i
];
587 case GLSL_TYPE_INT
: return (float) this->value
.i
[i
];
588 case GLSL_TYPE_FLOAT
: return this->value
.f
[i
];
589 case GLSL_TYPE_BOOL
: return this->value
.b
[i
] ? 1.0 : 0.0;
590 default: assert(!"Should not get here."); break;
593 /* Must return something to make the compiler happy. This is clearly an
600 ir_constant::get_int_component(unsigned i
) const
602 switch (this->type
->base_type
) {
603 case GLSL_TYPE_UINT
: return this->value
.u
[i
];
604 case GLSL_TYPE_INT
: return this->value
.i
[i
];
605 case GLSL_TYPE_FLOAT
: return (int) this->value
.f
[i
];
606 case GLSL_TYPE_BOOL
: return this->value
.b
[i
] ? 1 : 0;
607 default: assert(!"Should not get here."); break;
610 /* Must return something to make the compiler happy. This is clearly an
617 ir_constant::get_uint_component(unsigned i
) const
619 switch (this->type
->base_type
) {
620 case GLSL_TYPE_UINT
: return this->value
.u
[i
];
621 case GLSL_TYPE_INT
: return this->value
.i
[i
];
622 case GLSL_TYPE_FLOAT
: return (unsigned) this->value
.f
[i
];
623 case GLSL_TYPE_BOOL
: return this->value
.b
[i
] ? 1 : 0;
624 default: assert(!"Should not get here."); break;
627 /* Must return something to make the compiler happy. This is clearly an
634 ir_constant::get_array_element(unsigned i
) const
636 assert(this->type
->is_array());
638 /* From page 35 (page 41 of the PDF) of the GLSL 1.20 spec:
640 * "Behavior is undefined if a shader subscripts an array with an index
641 * less than 0 or greater than or equal to the size the array was
644 * Most out-of-bounds accesses are removed before things could get this far.
645 * There are cases where non-constant array index values can get constant
650 else if (i
>= this->type
->length
)
651 i
= this->type
->length
- 1;
653 return array_elements
[i
];
657 ir_constant::get_record_field(const char *name
)
659 int idx
= this->type
->field_index(name
);
664 if (this->components
.is_empty())
667 exec_node
*node
= this->components
.head
;
668 for (int i
= 0; i
< idx
; i
++) {
671 /* If the end of the list is encountered before the element matching the
672 * requested field is found, return NULL.
674 if (node
->is_tail_sentinel())
678 return (ir_constant
*) node
;
683 ir_constant::has_value(const ir_constant
*c
) const
685 if (this->type
!= c
->type
)
688 if (this->type
->is_array()) {
689 for (unsigned i
= 0; i
< this->type
->length
; i
++) {
690 if (this->array_elements
[i
]->has_value(c
->array_elements
[i
]))
696 if (this->type
->base_type
== GLSL_TYPE_STRUCT
) {
697 const exec_node
*a_node
= this->components
.head
;
698 const exec_node
*b_node
= c
->components
.head
;
700 while (!a_node
->is_tail_sentinel()) {
701 assert(!b_node
->is_tail_sentinel());
703 const ir_constant
*const a_field
= (ir_constant
*) a_node
;
704 const ir_constant
*const b_field
= (ir_constant
*) b_node
;
706 if (!a_field
->has_value(b_field
))
709 a_node
= a_node
->next
;
710 b_node
= b_node
->next
;
716 for (unsigned i
= 0; i
< this->type
->components(); i
++) {
717 switch (this->type
->base_type
) {
719 if (this->value
.u
[i
] != c
->value
.u
[i
])
723 if (this->value
.i
[i
] != c
->value
.i
[i
])
726 case GLSL_TYPE_FLOAT
:
727 if (this->value
.f
[i
] != c
->value
.f
[i
])
731 if (this->value
.b
[i
] != c
->value
.b
[i
])
735 assert(!"Should not get here.");
744 ir_constant::is_zero() const
746 if (!this->type
->is_scalar() && !this->type
->is_vector())
749 for (unsigned c
= 0; c
< this->type
->vector_elements
; c
++) {
750 switch (this->type
->base_type
) {
751 case GLSL_TYPE_FLOAT
:
752 if (this->value
.f
[c
] != 0.0)
756 if (this->value
.i
[c
] != 0)
760 if (this->value
.u
[c
] != 0)
764 if (this->value
.b
[c
] != false)
768 /* The only other base types are structures, arrays, and samplers.
769 * Samplers cannot be constants, and the others should have been
770 * filtered out above.
772 assert(!"Should not get here.");
781 ir_constant::is_one() const
783 if (!this->type
->is_scalar() && !this->type
->is_vector())
786 for (unsigned c
= 0; c
< this->type
->vector_elements
; c
++) {
787 switch (this->type
->base_type
) {
788 case GLSL_TYPE_FLOAT
:
789 if (this->value
.f
[c
] != 1.0)
793 if (this->value
.i
[c
] != 1)
797 if (this->value
.u
[c
] != 1)
801 if (this->value
.b
[c
] != true)
805 /* The only other base types are structures, arrays, and samplers.
806 * Samplers cannot be constants, and the others should have been
807 * filtered out above.
809 assert(!"Should not get here.");
819 this->ir_type
= ir_type_loop
;
820 this->cmp
= ir_unop_neg
;
823 this->increment
= NULL
;
824 this->counter
= NULL
;
828 ir_dereference_variable::ir_dereference_variable(ir_variable
*var
)
830 this->ir_type
= ir_type_dereference_variable
;
832 this->type
= (var
!= NULL
) ? var
->type
: glsl_type::error_type
;
836 ir_dereference_array::ir_dereference_array(ir_rvalue
*value
,
837 ir_rvalue
*array_index
)
839 this->ir_type
= ir_type_dereference_array
;
840 this->array_index
= array_index
;
841 this->set_array(value
);
845 ir_dereference_array::ir_dereference_array(ir_variable
*var
,
846 ir_rvalue
*array_index
)
848 void *ctx
= talloc_parent(var
);
850 this->ir_type
= ir_type_dereference_array
;
851 this->array_index
= array_index
;
852 this->set_array(new(ctx
) ir_dereference_variable(var
));
857 ir_dereference_array::set_array(ir_rvalue
*value
)
860 this->type
= glsl_type::error_type
;
862 if (this->array
!= NULL
) {
863 const glsl_type
*const vt
= this->array
->type
;
865 if (vt
->is_array()) {
866 type
= vt
->element_type();
867 } else if (vt
->is_matrix()) {
868 type
= vt
->column_type();
869 } else if (vt
->is_vector()) {
870 type
= vt
->get_base_type();
876 ir_dereference_record::ir_dereference_record(ir_rvalue
*value
,
879 this->ir_type
= ir_type_dereference_record
;
880 this->record
= value
;
881 this->field
= talloc_strdup(this, field
);
882 this->type
= (this->record
!= NULL
)
883 ? this->record
->type
->field_type(field
) : glsl_type::error_type
;
887 ir_dereference_record::ir_dereference_record(ir_variable
*var
,
890 void *ctx
= talloc_parent(var
);
892 this->ir_type
= ir_type_dereference_record
;
893 this->record
= new(ctx
) ir_dereference_variable(var
);
894 this->field
= talloc_strdup(this, field
);
895 this->type
= (this->record
!= NULL
)
896 ? this->record
->type
->field_type(field
) : glsl_type::error_type
;
899 bool type_contains_sampler(const glsl_type
*type
)
901 if (type
->is_array()) {
902 return type_contains_sampler(type
->fields
.array
);
903 } else if (type
->is_record()) {
904 for (unsigned int i
= 0; i
< type
->length
; i
++) {
905 if (type_contains_sampler(type
->fields
.structure
[i
].type
))
910 return type
->is_sampler();
915 ir_dereference::is_lvalue()
917 ir_variable
*var
= this->variable_referenced();
919 /* Every l-value derference chain eventually ends in a variable.
921 if ((var
== NULL
) || var
->read_only
)
924 if (this->type
->is_array() && !var
->array_lvalue
)
927 /* From page 17 (page 23 of the PDF) of the GLSL 1.20 spec:
929 * "Samplers cannot be treated as l-values; hence cannot be used
930 * as out or inout function parameters, nor can they be
933 if (type_contains_sampler(this->type
))
940 const char *tex_opcode_strs
[] = { "tex", "txb", "txl", "txd", "txf" };
942 const char *ir_texture::opcode_string()
944 assert((unsigned int) op
<=
945 sizeof(tex_opcode_strs
) / sizeof(tex_opcode_strs
[0]));
946 return tex_opcode_strs
[op
];
950 ir_texture::get_opcode(const char *str
)
952 const int count
= sizeof(tex_opcode_strs
) / sizeof(tex_opcode_strs
[0]);
953 for (int op
= 0; op
< count
; op
++) {
954 if (strcmp(str
, tex_opcode_strs
[op
]) == 0)
955 return (ir_texture_opcode
) op
;
957 return (ir_texture_opcode
) -1;
962 ir_texture::set_sampler(ir_dereference
*sampler
)
964 assert(sampler
!= NULL
);
965 this->sampler
= sampler
;
967 switch (sampler
->type
->sampler_type
) {
968 case GLSL_TYPE_FLOAT
:
969 this->type
= glsl_type::vec4_type
;
972 this->type
= glsl_type::ivec4_type
;
975 this->type
= glsl_type::uvec4_type
;
982 ir_swizzle::init_mask(const unsigned *comp
, unsigned count
)
984 assert((count
>= 1) && (count
<= 4));
986 memset(&this->mask
, 0, sizeof(this->mask
));
987 this->mask
.num_components
= count
;
989 unsigned dup_mask
= 0;
992 assert(comp
[3] <= 3);
993 dup_mask
|= (1U << comp
[3])
994 & ((1U << comp
[0]) | (1U << comp
[1]) | (1U << comp
[2]));
995 this->mask
.w
= comp
[3];
998 assert(comp
[2] <= 3);
999 dup_mask
|= (1U << comp
[2])
1000 & ((1U << comp
[0]) | (1U << comp
[1]));
1001 this->mask
.z
= comp
[2];
1004 assert(comp
[1] <= 3);
1005 dup_mask
|= (1U << comp
[1])
1006 & ((1U << comp
[0]));
1007 this->mask
.y
= comp
[1];
1010 assert(comp
[0] <= 3);
1011 this->mask
.x
= comp
[0];
1014 this->mask
.has_duplicates
= dup_mask
!= 0;
1016 /* Based on the number of elements in the swizzle and the base type
1017 * (i.e., float, int, unsigned, or bool) of the vector being swizzled,
1018 * generate the type of the resulting value.
1020 type
= glsl_type::get_instance(val
->type
->base_type
, mask
.num_components
, 1);
1023 ir_swizzle::ir_swizzle(ir_rvalue
*val
, unsigned x
, unsigned y
, unsigned z
,
1024 unsigned w
, unsigned count
)
1027 const unsigned components
[4] = { x
, y
, z
, w
};
1028 this->ir_type
= ir_type_swizzle
;
1029 this->init_mask(components
, count
);
1032 ir_swizzle::ir_swizzle(ir_rvalue
*val
, const unsigned *comp
,
1036 this->ir_type
= ir_type_swizzle
;
1037 this->init_mask(comp
, count
);
1040 ir_swizzle::ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
)
1042 this->ir_type
= ir_type_swizzle
;
1045 this->type
= glsl_type::get_instance(val
->type
->base_type
,
1046 mask
.num_components
, 1);
1055 ir_swizzle::create(ir_rvalue
*val
, const char *str
, unsigned vector_length
)
1057 void *ctx
= talloc_parent(val
);
1059 /* For each possible swizzle character, this table encodes the value in
1060 * \c idx_map that represents the 0th element of the vector. For invalid
1061 * swizzle characters (e.g., 'k'), a special value is used that will allow
1062 * detection of errors.
1064 static const unsigned char base_idx
[26] = {
1065 /* a b c d e f g h i j k l m */
1066 R
, R
, I
, I
, I
, I
, R
, I
, I
, I
, I
, I
, I
,
1067 /* n o p q r s t u v w x y z */
1068 I
, I
, S
, S
, R
, S
, S
, I
, I
, X
, X
, X
, X
1071 /* Each valid swizzle character has an entry in the previous table. This
1072 * table encodes the base index encoded in the previous table plus the actual
1073 * index of the swizzle character. When processing swizzles, the first
1074 * character in the string is indexed in the previous table. Each character
1075 * in the string is indexed in this table, and the value found there has the
1076 * value form the first table subtracted. The result must be on the range
1079 * For example, the string "wzyx" will get X from the first table. Each of
1080 * the charcaters will get X+3, X+2, X+1, and X+0 from this table. After
1081 * subtraction, the swizzle values are { 3, 2, 1, 0 }.
1083 * The string "wzrg" will get X from the first table. Each of the characters
1084 * will get X+3, X+2, R+0, and R+1 from this table. After subtraction, the
1085 * swizzle values are { 3, 2, 4, 5 }. Since 4 and 5 are outside the range
1086 * [0,3], the error is detected.
1088 static const unsigned char idx_map
[26] = {
1089 /* a b c d e f g h i j k l m */
1090 R
+3, R
+2, 0, 0, 0, 0, R
+1, 0, 0, 0, 0, 0, 0,
1091 /* n o p q r s t u v w x y z */
1092 0, 0, S
+2, S
+3, R
+0, S
+0, S
+1, 0, 0, X
+3, X
+0, X
+1, X
+2
1095 int swiz_idx
[4] = { 0, 0, 0, 0 };
1099 /* Validate the first character in the swizzle string and look up the base
1100 * index value as described above.
1102 if ((str
[0] < 'a') || (str
[0] > 'z'))
1105 const unsigned base
= base_idx
[str
[0] - 'a'];
1108 for (i
= 0; (i
< 4) && (str
[i
] != '\0'); i
++) {
1109 /* Validate the next character, and, as described above, convert it to a
1112 if ((str
[i
] < 'a') || (str
[i
] > 'z'))
1115 swiz_idx
[i
] = idx_map
[str
[i
] - 'a'] - base
;
1116 if ((swiz_idx
[i
] < 0) || (swiz_idx
[i
] >= (int) vector_length
))
1123 return new(ctx
) ir_swizzle(val
, swiz_idx
[0], swiz_idx
[1], swiz_idx
[2],
1133 ir_swizzle::variable_referenced()
1135 return this->val
->variable_referenced();
1139 ir_variable::ir_variable(const struct glsl_type
*type
, const char *name
,
1140 ir_variable_mode mode
)
1141 : max_array_access(0), read_only(false), centroid(false), invariant(false),
1142 mode(mode
), interpolation(ir_var_smooth
), array_lvalue(false)
1144 this->ir_type
= ir_type_variable
;
1146 this->name
= talloc_strdup(this, name
);
1147 this->explicit_location
= false;
1148 this->location
= -1;
1149 this->warn_extension
= NULL
;
1150 this->constant_value
= NULL
;
1151 this->origin_upper_left
= false;
1152 this->pixel_center_integer
= false;
1154 if (type
&& type
->base_type
== GLSL_TYPE_SAMPLER
)
1155 this->read_only
= true;
1160 ir_variable::interpolation_string() const
1162 switch (this->interpolation
) {
1163 case ir_var_smooth
: return "smooth";
1164 case ir_var_flat
: return "flat";
1165 case ir_var_noperspective
: return "noperspective";
1168 assert(!"Should not get here.");
1174 ir_variable::component_slots() const
1176 /* FINISHME: Sparsely accessed arrays require fewer slots. */
1177 return this->type
->component_slots();
1181 ir_function_signature::ir_function_signature(const glsl_type
*return_type
)
1182 : return_type(return_type
), is_defined(false), _function(NULL
)
1184 this->ir_type
= ir_type_function_signature
;
1185 this->is_builtin
= false;
1190 ir_function_signature::qualifiers_match(exec_list
*params
)
1192 exec_list_iterator iter_a
= parameters
.iterator();
1193 exec_list_iterator iter_b
= params
->iterator();
1195 /* check that the qualifiers match. */
1196 while (iter_a
.has_next()) {
1197 ir_variable
*a
= (ir_variable
*)iter_a
.get();
1198 ir_variable
*b
= (ir_variable
*)iter_b
.get();
1200 if (a
->read_only
!= b
->read_only
||
1201 a
->mode
!= b
->mode
||
1202 a
->interpolation
!= b
->interpolation
||
1203 a
->centroid
!= b
->centroid
) {
1205 /* parameter a's qualifiers don't match */
1217 ir_function_signature::replace_parameters(exec_list
*new_params
)
1219 /* Destroy all of the previous parameter information. If the previous
1220 * parameter information comes from the function prototype, it may either
1221 * specify incorrect parameter names or not have names at all.
1223 foreach_iter(exec_list_iterator
, iter
, parameters
) {
1224 assert(((ir_instruction
*) iter
.get())->as_variable() != NULL
);
1229 new_params
->move_nodes_to(¶meters
);
1233 ir_function::ir_function(const char *name
)
1235 this->ir_type
= ir_type_function
;
1236 this->name
= talloc_strdup(this, name
);
1241 ir_function::has_user_signature()
1243 foreach_list(n
, &this->signatures
) {
1244 ir_function_signature
*const sig
= (ir_function_signature
*) n
;
1245 if (!sig
->is_builtin
)
1253 ir_call::get_error_instruction(void *ctx
)
1255 ir_call
*call
= new(ctx
) ir_call
;
1257 call
->type
= glsl_type::error_type
;
1262 ir_call::set_callee(ir_function_signature
*sig
)
1264 assert((this->type
== NULL
) || (this->type
== sig
->return_type
));
1270 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
)
1272 foreach_iter(exec_list_iterator
, iter
, *list
) {
1273 ((ir_instruction
*)iter
.get())->accept(visitor
);
1279 steal_memory(ir_instruction
*ir
, void *new_ctx
)
1281 ir_variable
*var
= ir
->as_variable();
1282 ir_constant
*constant
= ir
->as_constant();
1283 if (var
!= NULL
&& var
->constant_value
!= NULL
)
1284 steal_memory(var
->constant_value
, ir
);
1286 /* The components of aggregate constants are not visited by the normal
1287 * visitor, so steal their values by hand.
1289 if (constant
!= NULL
) {
1290 if (constant
->type
->is_record()) {
1291 foreach_iter(exec_list_iterator
, iter
, constant
->components
) {
1292 ir_constant
*field
= (ir_constant
*)iter
.get();
1293 steal_memory(field
, ir
);
1295 } else if (constant
->type
->is_array()) {
1296 for (unsigned int i
= 0; i
< constant
->type
->length
; i
++) {
1297 steal_memory(constant
->array_elements
[i
], ir
);
1302 talloc_steal(new_ctx
, ir
);
1307 reparent_ir(exec_list
*list
, void *mem_ctx
)
1309 foreach_list(node
, list
) {
1310 visit_tree((ir_instruction
*) node
, steal_memory
, mem_ctx
);