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
49 bool ir_rvalue::is_basis() const
55 * Modify the swizzle make to move one component to another
57 * \param m IR swizzle to be modified
58 * \param from Component in the RHS that is to be swizzled
59 * \param to Desired swizzle location of \c from
62 update_rhs_swizzle(ir_swizzle_mask
&m
, unsigned from
, unsigned to
)
65 case 0: m
.x
= from
; break;
66 case 1: m
.y
= from
; break;
67 case 2: m
.z
= from
; break;
68 case 3: m
.w
= from
; break;
69 default: assert(!"Should not get here.");
72 m
.num_components
= MAX2(m
.num_components
, (to
+ 1));
76 ir_assignment::set_lhs(ir_rvalue
*lhs
)
79 bool swizzled
= false;
82 ir_swizzle
*swiz
= lhs
->as_swizzle();
87 unsigned write_mask
= 0;
88 ir_swizzle_mask rhs_swiz
= { 0, 0, 0, 0, 0, 0 };
90 for (unsigned i
= 0; i
< swiz
->mask
.num_components
; i
++) {
94 case 0: c
= swiz
->mask
.x
; break;
95 case 1: c
= swiz
->mask
.y
; break;
96 case 2: c
= swiz
->mask
.z
; break;
97 case 3: c
= swiz
->mask
.w
; break;
98 default: assert(!"Should not get here.");
101 write_mask
|= (((this->write_mask
>> i
) & 1) << c
);
102 update_rhs_swizzle(rhs_swiz
, i
, c
);
105 this->write_mask
= write_mask
;
108 this->rhs
= new(mem_ctx
) ir_swizzle(this->rhs
, rhs_swiz
);
113 /* Now, RHS channels line up with the LHS writemask. Collapse it
114 * to just the channels that will be written.
116 ir_swizzle_mask rhs_swiz
= { 0, 0, 0, 0, 0, 0 };
118 for (int i
= 0; i
< 4; i
++) {
119 if (write_mask
& (1 << i
))
120 update_rhs_swizzle(rhs_swiz
, i
, rhs_chan
++);
122 this->rhs
= new(mem_ctx
) ir_swizzle(this->rhs
, rhs_swiz
);
125 assert((lhs
== NULL
) || lhs
->as_dereference());
127 this->lhs
= (ir_dereference
*) lhs
;
131 ir_assignment::whole_variable_written()
133 ir_variable
*v
= this->lhs
->whole_variable_referenced();
138 if (v
->type
->is_scalar())
141 if (v
->type
->is_vector()) {
142 const unsigned mask
= (1U << v
->type
->vector_elements
) - 1;
144 if (mask
!= this->write_mask
)
148 /* Either all the vector components are assigned or the variable is some
149 * composite type (and the whole thing is assigned.
154 ir_assignment::ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
,
155 ir_rvalue
*condition
, unsigned write_mask
)
157 this->ir_type
= ir_type_assignment
;
158 this->condition
= condition
;
161 this->write_mask
= write_mask
;
163 if (lhs
->type
->is_scalar() || lhs
->type
->is_vector()) {
164 int lhs_components
= 0;
165 for (int i
= 0; i
< 4; i
++) {
166 if (write_mask
& (1 << i
))
170 assert(lhs_components
== this->rhs
->type
->vector_elements
);
174 ir_assignment::ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
,
175 ir_rvalue
*condition
)
177 this->ir_type
= ir_type_assignment
;
178 this->condition
= condition
;
181 /* If the RHS is a vector type, assume that all components of the vector
182 * type are being written to the LHS. The write mask comes from the RHS
183 * because we can have a case where the LHS is a vec4 and the RHS is a
184 * vec3. In that case, the assignment is:
186 * (assign (...) (xyz) (var_ref lhs) (var_ref rhs))
188 if (rhs
->type
->is_vector())
189 this->write_mask
= (1U << rhs
->type
->vector_elements
) - 1;
190 else if (rhs
->type
->is_scalar())
191 this->write_mask
= 1;
193 this->write_mask
= 0;
198 ir_expression::ir_expression(int op
, const struct glsl_type
*type
,
199 ir_rvalue
*op0
, ir_rvalue
*op1
,
200 ir_rvalue
*op2
, ir_rvalue
*op3
)
202 this->ir_type
= ir_type_expression
;
204 this->operation
= ir_expression_operation(op
);
205 this->operands
[0] = op0
;
206 this->operands
[1] = op1
;
207 this->operands
[2] = op2
;
208 this->operands
[3] = op3
;
210 int num_operands
= get_num_operands(this->operation
);
211 for (int i
= num_operands
; i
< 4; i
++) {
212 assert(this->operands
[i
] == NULL
);
217 ir_expression::ir_expression(int op
, ir_rvalue
*op0
)
219 this->ir_type
= ir_type_expression
;
221 this->operation
= ir_expression_operation(op
);
222 this->operands
[0] = op0
;
223 this->operands
[1] = NULL
;
224 this->operands
[2] = NULL
;
225 this->operands
[3] = NULL
;
227 assert(op
<= ir_last_unop
);
229 switch (this->operation
) {
230 case ir_unop_bit_not
:
231 case ir_unop_logic_not
:
246 case ir_unop_round_even
:
249 case ir_unop_sin_reduced
:
250 case ir_unop_cos_reduced
:
253 case ir_unop_bitfield_reverse
:
254 this->type
= op0
->type
;
260 case ir_unop_bitcast_f2i
:
261 case ir_unop_bit_count
:
262 case ir_unop_find_msb
:
263 case ir_unop_find_lsb
:
264 this->type
= glsl_type::get_instance(GLSL_TYPE_INT
,
265 op0
->type
->vector_elements
, 1);
271 case ir_unop_bitcast_i2f
:
272 case ir_unop_bitcast_u2f
:
273 this->type
= glsl_type::get_instance(GLSL_TYPE_FLOAT
,
274 op0
->type
->vector_elements
, 1);
279 this->type
= glsl_type::get_instance(GLSL_TYPE_BOOL
,
280 op0
->type
->vector_elements
, 1);
285 case ir_unop_bitcast_f2u
:
286 this->type
= glsl_type::get_instance(GLSL_TYPE_UINT
,
287 op0
->type
->vector_elements
, 1);
291 case ir_unop_unpack_half_2x16_split_x
:
292 case ir_unop_unpack_half_2x16_split_y
:
293 this->type
= glsl_type::float_type
;
297 this->type
= glsl_type::bool_type
;
300 case ir_unop_pack_snorm_2x16
:
301 case ir_unop_pack_snorm_4x8
:
302 case ir_unop_pack_unorm_2x16
:
303 case ir_unop_pack_unorm_4x8
:
304 case ir_unop_pack_half_2x16
:
305 this->type
= glsl_type::uint_type
;
308 case ir_unop_unpack_snorm_2x16
:
309 case ir_unop_unpack_unorm_2x16
:
310 case ir_unop_unpack_half_2x16
:
311 this->type
= glsl_type::vec2_type
;
314 case ir_unop_unpack_snorm_4x8
:
315 case ir_unop_unpack_unorm_4x8
:
316 this->type
= glsl_type::vec4_type
;
320 assert(!"not reached: missing automatic type setup for ir_expression");
321 this->type
= op0
->type
;
326 ir_expression::ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
)
328 this->ir_type
= ir_type_expression
;
330 this->operation
= ir_expression_operation(op
);
331 this->operands
[0] = op0
;
332 this->operands
[1] = op1
;
333 this->operands
[2] = NULL
;
334 this->operands
[3] = NULL
;
336 assert(op
> ir_last_unop
);
338 switch (this->operation
) {
339 case ir_binop_all_equal
:
340 case ir_binop_any_nequal
:
341 this->type
= glsl_type::bool_type
;
352 if (op0
->type
->is_scalar()) {
353 this->type
= op1
->type
;
354 } else if (op1
->type
->is_scalar()) {
355 this->type
= op0
->type
;
357 /* FINISHME: matrix types */
358 assert(!op0
->type
->is_matrix() && !op1
->type
->is_matrix());
359 assert(op0
->type
== op1
->type
);
360 this->type
= op0
->type
;
364 case ir_binop_logic_and
:
365 case ir_binop_logic_xor
:
366 case ir_binop_logic_or
:
367 case ir_binop_bit_and
:
368 case ir_binop_bit_xor
:
369 case ir_binop_bit_or
:
370 assert(!op0
->type
->is_matrix());
371 assert(!op1
->type
->is_matrix());
372 if (op0
->type
->is_scalar()) {
373 this->type
= op1
->type
;
374 } else if (op1
->type
->is_scalar()) {
375 this->type
= op0
->type
;
377 assert(op0
->type
->vector_elements
== op1
->type
->vector_elements
);
378 this->type
= op0
->type
;
383 case ir_binop_nequal
:
384 case ir_binop_lequal
:
385 case ir_binop_gequal
:
387 case ir_binop_greater
:
388 assert(op0
->type
== op1
->type
);
389 this->type
= glsl_type::get_instance(GLSL_TYPE_BOOL
,
390 op0
->type
->vector_elements
, 1);
394 this->type
= glsl_type::float_type
;
397 case ir_binop_pack_half_2x16_split
:
398 this->type
= glsl_type::uint_type
;
401 case ir_binop_lshift
:
402 case ir_binop_rshift
:
403 this->type
= op0
->type
;
406 case ir_binop_vector_extract
:
407 this->type
= op0
->type
->get_scalar_type();
411 assert(!"not reached: missing automatic type setup for ir_expression");
412 this->type
= glsl_type::float_type
;
416 ir_expression::ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
,
419 this->ir_type
= ir_type_expression
;
421 this->operation
= ir_expression_operation(op
);
422 this->operands
[0] = op0
;
423 this->operands
[1] = op1
;
424 this->operands
[2] = op2
;
425 this->operands
[3] = NULL
;
427 assert(op
> ir_last_binop
&& op
<= ir_last_triop
);
429 switch (this->operation
) {
432 case ir_triop_bitfield_extract
:
433 case ir_triop_vector_insert
:
434 this->type
= op0
->type
;
438 this->type
= op1
->type
;
442 assert(!"not reached: missing automatic type setup for ir_expression");
443 this->type
= glsl_type::float_type
;
448 ir_expression::get_num_operands(ir_expression_operation op
)
450 assert(op
<= ir_last_opcode
);
452 if (op
<= ir_last_unop
)
455 if (op
<= ir_last_binop
)
458 if (op
<= ir_last_triop
)
461 if (op
<= ir_last_quadop
)
468 static const char *const operator_strs
[] = {
517 "unpackHalf2x16_split_x",
518 "unpackHalf2x16_split_y",
549 "packHalf2x16_split",
562 const char *ir_expression::operator_string(ir_expression_operation op
)
564 assert((unsigned int) op
< Elements(operator_strs
));
565 assert(Elements(operator_strs
) == (ir_quadop_vector
+ 1));
566 return operator_strs
[op
];
569 const char *ir_expression::operator_string()
571 return operator_string(this->operation
);
575 depth_layout_string(ir_depth_layout layout
)
578 case ir_depth_layout_none
: return "";
579 case ir_depth_layout_any
: return "depth_any";
580 case ir_depth_layout_greater
: return "depth_greater";
581 case ir_depth_layout_less
: return "depth_less";
582 case ir_depth_layout_unchanged
: return "depth_unchanged";
590 ir_expression_operation
591 ir_expression::get_operator(const char *str
)
593 const int operator_count
= sizeof(operator_strs
) / sizeof(operator_strs
[0]);
594 for (int op
= 0; op
< operator_count
; op
++) {
595 if (strcmp(str
, operator_strs
[op
]) == 0)
596 return (ir_expression_operation
) op
;
598 return (ir_expression_operation
) -1;
601 ir_constant::ir_constant()
603 this->ir_type
= ir_type_constant
;
606 ir_constant::ir_constant(const struct glsl_type
*type
,
607 const ir_constant_data
*data
)
609 assert((type
->base_type
>= GLSL_TYPE_UINT
)
610 && (type
->base_type
<= GLSL_TYPE_BOOL
));
612 this->ir_type
= ir_type_constant
;
614 memcpy(& this->value
, data
, sizeof(this->value
));
617 ir_constant::ir_constant(float f
)
619 this->ir_type
= ir_type_constant
;
620 this->type
= glsl_type::float_type
;
621 this->value
.f
[0] = f
;
622 for (int i
= 1; i
< 16; i
++) {
623 this->value
.f
[i
] = 0;
627 ir_constant::ir_constant(unsigned int u
)
629 this->ir_type
= ir_type_constant
;
630 this->type
= glsl_type::uint_type
;
631 this->value
.u
[0] = u
;
632 for (int i
= 1; i
< 16; i
++) {
633 this->value
.u
[i
] = 0;
637 ir_constant::ir_constant(int i
)
639 this->ir_type
= ir_type_constant
;
640 this->type
= glsl_type::int_type
;
641 this->value
.i
[0] = i
;
642 for (int i
= 1; i
< 16; i
++) {
643 this->value
.i
[i
] = 0;
647 ir_constant::ir_constant(bool b
)
649 this->ir_type
= ir_type_constant
;
650 this->type
= glsl_type::bool_type
;
651 this->value
.b
[0] = b
;
652 for (int i
= 1; i
< 16; i
++) {
653 this->value
.b
[i
] = false;
657 ir_constant::ir_constant(const ir_constant
*c
, unsigned i
)
659 this->ir_type
= ir_type_constant
;
660 this->type
= c
->type
->get_base_type();
662 switch (this->type
->base_type
) {
663 case GLSL_TYPE_UINT
: this->value
.u
[0] = c
->value
.u
[i
]; break;
664 case GLSL_TYPE_INT
: this->value
.i
[0] = c
->value
.i
[i
]; break;
665 case GLSL_TYPE_FLOAT
: this->value
.f
[0] = c
->value
.f
[i
]; break;
666 case GLSL_TYPE_BOOL
: this->value
.b
[0] = c
->value
.b
[i
]; break;
667 default: assert(!"Should not get here."); break;
671 ir_constant::ir_constant(const struct glsl_type
*type
, exec_list
*value_list
)
673 this->ir_type
= ir_type_constant
;
676 assert(type
->is_scalar() || type
->is_vector() || type
->is_matrix()
677 || type
->is_record() || type
->is_array());
679 if (type
->is_array()) {
680 this->array_elements
= ralloc_array(this, ir_constant
*, type
->length
);
682 foreach_list(node
, value_list
) {
683 ir_constant
*value
= (ir_constant
*) node
;
684 assert(value
->as_constant() != NULL
);
686 this->array_elements
[i
++] = value
;
691 /* If the constant is a record, the types of each of the entries in
692 * value_list must be a 1-for-1 match with the structure components. Each
693 * entry must also be a constant. Just move the nodes from the value_list
694 * to the list in the ir_constant.
696 /* FINISHME: Should there be some type checking and / or assertions here? */
697 /* FINISHME: Should the new constant take ownership of the nodes from
698 * FINISHME: value_list, or should it make copies?
700 if (type
->is_record()) {
701 value_list
->move_nodes_to(& this->components
);
705 for (unsigned i
= 0; i
< 16; i
++) {
706 this->value
.u
[i
] = 0;
709 ir_constant
*value
= (ir_constant
*) (value_list
->head
);
711 /* Constructors with exactly one scalar argument are special for vectors
712 * and matrices. For vectors, the scalar value is replicated to fill all
713 * the components. For matrices, the scalar fills the components of the
714 * diagonal while the rest is filled with 0.
716 if (value
->type
->is_scalar() && value
->next
->is_tail_sentinel()) {
717 if (type
->is_matrix()) {
718 /* Matrix - fill diagonal (rest is already set to 0) */
719 assert(type
->base_type
== GLSL_TYPE_FLOAT
);
720 for (unsigned i
= 0; i
< type
->matrix_columns
; i
++)
721 this->value
.f
[i
* type
->vector_elements
+ i
] = value
->value
.f
[0];
723 /* Vector or scalar - fill all components */
724 switch (type
->base_type
) {
727 for (unsigned i
= 0; i
< type
->components(); i
++)
728 this->value
.u
[i
] = value
->value
.u
[0];
730 case GLSL_TYPE_FLOAT
:
731 for (unsigned i
= 0; i
< type
->components(); i
++)
732 this->value
.f
[i
] = value
->value
.f
[0];
735 for (unsigned i
= 0; i
< type
->components(); i
++)
736 this->value
.b
[i
] = value
->value
.b
[0];
739 assert(!"Should not get here.");
746 if (type
->is_matrix() && value
->type
->is_matrix()) {
747 assert(value
->next
->is_tail_sentinel());
749 /* From section 5.4.2 of the GLSL 1.20 spec:
750 * "If a matrix is constructed from a matrix, then each component
751 * (column i, row j) in the result that has a corresponding component
752 * (column i, row j) in the argument will be initialized from there."
754 unsigned cols
= MIN2(type
->matrix_columns
, value
->type
->matrix_columns
);
755 unsigned rows
= MIN2(type
->vector_elements
, value
->type
->vector_elements
);
756 for (unsigned i
= 0; i
< cols
; i
++) {
757 for (unsigned j
= 0; j
< rows
; j
++) {
758 const unsigned src
= i
* value
->type
->vector_elements
+ j
;
759 const unsigned dst
= i
* type
->vector_elements
+ j
;
760 this->value
.f
[dst
] = value
->value
.f
[src
];
764 /* "All other components will be initialized to the identity matrix." */
765 for (unsigned i
= cols
; i
< type
->matrix_columns
; i
++)
766 this->value
.f
[i
* type
->vector_elements
+ i
] = 1.0;
771 /* Use each component from each entry in the value_list to initialize one
772 * component of the constant being constructed.
774 for (unsigned i
= 0; i
< type
->components(); /* empty */) {
775 assert(value
->as_constant() != NULL
);
776 assert(!value
->is_tail_sentinel());
778 for (unsigned j
= 0; j
< value
->type
->components(); j
++) {
779 switch (type
->base_type
) {
781 this->value
.u
[i
] = value
->get_uint_component(j
);
784 this->value
.i
[i
] = value
->get_int_component(j
);
786 case GLSL_TYPE_FLOAT
:
787 this->value
.f
[i
] = value
->get_float_component(j
);
790 this->value
.b
[i
] = value
->get_bool_component(j
);
793 /* FINISHME: What to do? Exceptions are not the answer.
799 if (i
>= type
->components())
803 value
= (ir_constant
*) value
->next
;
808 ir_constant::zero(void *mem_ctx
, const glsl_type
*type
)
810 assert(type
->is_scalar() || type
->is_vector() || type
->is_matrix()
811 || type
->is_record() || type
->is_array());
813 ir_constant
*c
= new(mem_ctx
) ir_constant
;
815 memset(&c
->value
, 0, sizeof(c
->value
));
817 if (type
->is_array()) {
818 c
->array_elements
= ralloc_array(c
, ir_constant
*, type
->length
);
820 for (unsigned i
= 0; i
< type
->length
; i
++)
821 c
->array_elements
[i
] = ir_constant::zero(c
, type
->element_type());
824 if (type
->is_record()) {
825 for (unsigned i
= 0; i
< type
->length
; i
++) {
826 ir_constant
*comp
= ir_constant::zero(mem_ctx
, type
->fields
.structure
[i
].type
);
827 c
->components
.push_tail(comp
);
835 ir_constant::get_bool_component(unsigned i
) const
837 switch (this->type
->base_type
) {
838 case GLSL_TYPE_UINT
: return this->value
.u
[i
] != 0;
839 case GLSL_TYPE_INT
: return this->value
.i
[i
] != 0;
840 case GLSL_TYPE_FLOAT
: return ((int)this->value
.f
[i
]) != 0;
841 case GLSL_TYPE_BOOL
: return this->value
.b
[i
];
842 default: assert(!"Should not get here."); break;
845 /* Must return something to make the compiler happy. This is clearly an
852 ir_constant::get_float_component(unsigned i
) const
854 switch (this->type
->base_type
) {
855 case GLSL_TYPE_UINT
: return (float) this->value
.u
[i
];
856 case GLSL_TYPE_INT
: return (float) this->value
.i
[i
];
857 case GLSL_TYPE_FLOAT
: return this->value
.f
[i
];
858 case GLSL_TYPE_BOOL
: return this->value
.b
[i
] ? 1.0f
: 0.0f
;
859 default: assert(!"Should not get here."); break;
862 /* Must return something to make the compiler happy. This is clearly an
869 ir_constant::get_int_component(unsigned i
) const
871 switch (this->type
->base_type
) {
872 case GLSL_TYPE_UINT
: return this->value
.u
[i
];
873 case GLSL_TYPE_INT
: return this->value
.i
[i
];
874 case GLSL_TYPE_FLOAT
: return (int) this->value
.f
[i
];
875 case GLSL_TYPE_BOOL
: return this->value
.b
[i
] ? 1 : 0;
876 default: assert(!"Should not get here."); break;
879 /* Must return something to make the compiler happy. This is clearly an
886 ir_constant::get_uint_component(unsigned i
) const
888 switch (this->type
->base_type
) {
889 case GLSL_TYPE_UINT
: return this->value
.u
[i
];
890 case GLSL_TYPE_INT
: return this->value
.i
[i
];
891 case GLSL_TYPE_FLOAT
: return (unsigned) this->value
.f
[i
];
892 case GLSL_TYPE_BOOL
: return this->value
.b
[i
] ? 1 : 0;
893 default: assert(!"Should not get here."); break;
896 /* Must return something to make the compiler happy. This is clearly an
903 ir_constant::get_array_element(unsigned i
) const
905 assert(this->type
->is_array());
907 /* From page 35 (page 41 of the PDF) of the GLSL 1.20 spec:
909 * "Behavior is undefined if a shader subscripts an array with an index
910 * less than 0 or greater than or equal to the size the array was
913 * Most out-of-bounds accesses are removed before things could get this far.
914 * There are cases where non-constant array index values can get constant
919 else if (i
>= this->type
->length
)
920 i
= this->type
->length
- 1;
922 return array_elements
[i
];
926 ir_constant::get_record_field(const char *name
)
928 int idx
= this->type
->field_index(name
);
933 if (this->components
.is_empty())
936 exec_node
*node
= this->components
.head
;
937 for (int i
= 0; i
< idx
; i
++) {
940 /* If the end of the list is encountered before the element matching the
941 * requested field is found, return NULL.
943 if (node
->is_tail_sentinel())
947 return (ir_constant
*) node
;
951 ir_constant::copy_offset(ir_constant
*src
, int offset
)
953 switch (this->type
->base_type
) {
956 case GLSL_TYPE_FLOAT
:
957 case GLSL_TYPE_BOOL
: {
958 unsigned int size
= src
->type
->components();
959 assert (size
<= this->type
->components() - offset
);
960 for (unsigned int i
=0; i
<size
; i
++) {
961 switch (this->type
->base_type
) {
963 value
.u
[i
+offset
] = src
->get_uint_component(i
);
966 value
.i
[i
+offset
] = src
->get_int_component(i
);
968 case GLSL_TYPE_FLOAT
:
969 value
.f
[i
+offset
] = src
->get_float_component(i
);
972 value
.b
[i
+offset
] = src
->get_bool_component(i
);
974 default: // Shut up the compiler
981 case GLSL_TYPE_STRUCT
: {
982 assert (src
->type
== this->type
);
983 this->components
.make_empty();
984 foreach_list(node
, &src
->components
) {
985 ir_constant
*const orig
= (ir_constant
*) node
;
987 this->components
.push_tail(orig
->clone(this, NULL
));
992 case GLSL_TYPE_ARRAY
: {
993 assert (src
->type
== this->type
);
994 for (unsigned i
= 0; i
< this->type
->length
; i
++) {
995 this->array_elements
[i
] = src
->array_elements
[i
]->clone(this, NULL
);
1001 assert(!"Should not get here.");
1007 ir_constant::copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
)
1009 assert (!type
->is_array() && !type
->is_record());
1011 if (!type
->is_vector() && !type
->is_matrix()) {
1017 for (int i
=0; i
<4; i
++) {
1018 if (mask
& (1 << i
)) {
1019 switch (this->type
->base_type
) {
1020 case GLSL_TYPE_UINT
:
1021 value
.u
[i
+offset
] = src
->get_uint_component(id
++);
1024 value
.i
[i
+offset
] = src
->get_int_component(id
++);
1026 case GLSL_TYPE_FLOAT
:
1027 value
.f
[i
+offset
] = src
->get_float_component(id
++);
1029 case GLSL_TYPE_BOOL
:
1030 value
.b
[i
+offset
] = src
->get_bool_component(id
++);
1033 assert(!"Should not get here.");
1041 ir_constant::has_value(const ir_constant
*c
) const
1043 if (this->type
!= c
->type
)
1046 if (this->type
->is_array()) {
1047 for (unsigned i
= 0; i
< this->type
->length
; i
++) {
1048 if (!this->array_elements
[i
]->has_value(c
->array_elements
[i
]))
1054 if (this->type
->base_type
== GLSL_TYPE_STRUCT
) {
1055 const exec_node
*a_node
= this->components
.head
;
1056 const exec_node
*b_node
= c
->components
.head
;
1058 while (!a_node
->is_tail_sentinel()) {
1059 assert(!b_node
->is_tail_sentinel());
1061 const ir_constant
*const a_field
= (ir_constant
*) a_node
;
1062 const ir_constant
*const b_field
= (ir_constant
*) b_node
;
1064 if (!a_field
->has_value(b_field
))
1067 a_node
= a_node
->next
;
1068 b_node
= b_node
->next
;
1074 for (unsigned i
= 0; i
< this->type
->components(); i
++) {
1075 switch (this->type
->base_type
) {
1076 case GLSL_TYPE_UINT
:
1077 if (this->value
.u
[i
] != c
->value
.u
[i
])
1081 if (this->value
.i
[i
] != c
->value
.i
[i
])
1084 case GLSL_TYPE_FLOAT
:
1085 if (this->value
.f
[i
] != c
->value
.f
[i
])
1088 case GLSL_TYPE_BOOL
:
1089 if (this->value
.b
[i
] != c
->value
.b
[i
])
1093 assert(!"Should not get here.");
1102 ir_constant::is_zero() const
1104 if (!this->type
->is_scalar() && !this->type
->is_vector())
1107 for (unsigned c
= 0; c
< this->type
->vector_elements
; c
++) {
1108 switch (this->type
->base_type
) {
1109 case GLSL_TYPE_FLOAT
:
1110 if (this->value
.f
[c
] != 0.0)
1114 if (this->value
.i
[c
] != 0)
1117 case GLSL_TYPE_UINT
:
1118 if (this->value
.u
[c
] != 0)
1121 case GLSL_TYPE_BOOL
:
1122 if (this->value
.b
[c
] != false)
1126 /* The only other base types are structures, arrays, and samplers.
1127 * Samplers cannot be constants, and the others should have been
1128 * filtered out above.
1130 assert(!"Should not get here.");
1139 ir_constant::is_one() const
1141 if (!this->type
->is_scalar() && !this->type
->is_vector())
1144 for (unsigned c
= 0; c
< this->type
->vector_elements
; c
++) {
1145 switch (this->type
->base_type
) {
1146 case GLSL_TYPE_FLOAT
:
1147 if (this->value
.f
[c
] != 1.0)
1151 if (this->value
.i
[c
] != 1)
1154 case GLSL_TYPE_UINT
:
1155 if (this->value
.u
[c
] != 1)
1158 case GLSL_TYPE_BOOL
:
1159 if (this->value
.b
[c
] != true)
1163 /* The only other base types are structures, arrays, and samplers.
1164 * Samplers cannot be constants, and the others should have been
1165 * filtered out above.
1167 assert(!"Should not get here.");
1176 ir_constant::is_negative_one() const
1178 if (!this->type
->is_scalar() && !this->type
->is_vector())
1181 if (this->type
->is_boolean())
1184 for (unsigned c
= 0; c
< this->type
->vector_elements
; c
++) {
1185 switch (this->type
->base_type
) {
1186 case GLSL_TYPE_FLOAT
:
1187 if (this->value
.f
[c
] != -1.0)
1191 if (this->value
.i
[c
] != -1)
1194 case GLSL_TYPE_UINT
:
1195 if (int(this->value
.u
[c
]) != -1)
1199 /* The only other base types are structures, arrays, samplers, and
1200 * booleans. Samplers cannot be constants, and the others should
1201 * have been filtered out above.
1203 assert(!"Should not get here.");
1212 ir_constant::is_basis() const
1214 if (!this->type
->is_scalar() && !this->type
->is_vector())
1217 if (this->type
->is_boolean())
1221 for (unsigned c
= 0; c
< this->type
->vector_elements
; c
++) {
1222 switch (this->type
->base_type
) {
1223 case GLSL_TYPE_FLOAT
:
1224 if (this->value
.f
[c
] == 1.0)
1226 else if (this->value
.f
[c
] != 0.0)
1230 if (this->value
.i
[c
] == 1)
1232 else if (this->value
.i
[c
] != 0)
1235 case GLSL_TYPE_UINT
:
1236 if (int(this->value
.u
[c
]) == 1)
1238 else if (int(this->value
.u
[c
]) != 0)
1242 /* The only other base types are structures, arrays, samplers, and
1243 * booleans. Samplers cannot be constants, and the others should
1244 * have been filtered out above.
1246 assert(!"Should not get here.");
1256 this->ir_type
= ir_type_loop
;
1257 this->cmp
= ir_unop_neg
;
1260 this->increment
= NULL
;
1261 this->counter
= NULL
;
1265 ir_dereference_variable::ir_dereference_variable(ir_variable
*var
)
1267 assert(var
!= NULL
);
1269 this->ir_type
= ir_type_dereference_variable
;
1271 this->type
= var
->type
;
1275 ir_dereference_array::ir_dereference_array(ir_rvalue
*value
,
1276 ir_rvalue
*array_index
)
1278 this->ir_type
= ir_type_dereference_array
;
1279 this->array_index
= array_index
;
1280 this->set_array(value
);
1284 ir_dereference_array::ir_dereference_array(ir_variable
*var
,
1285 ir_rvalue
*array_index
)
1287 void *ctx
= ralloc_parent(var
);
1289 this->ir_type
= ir_type_dereference_array
;
1290 this->array_index
= array_index
;
1291 this->set_array(new(ctx
) ir_dereference_variable(var
));
1296 ir_dereference_array::set_array(ir_rvalue
*value
)
1298 assert(value
!= NULL
);
1300 this->array
= value
;
1302 const glsl_type
*const vt
= this->array
->type
;
1304 if (vt
->is_array()) {
1305 type
= vt
->element_type();
1306 } else if (vt
->is_matrix()) {
1307 type
= vt
->column_type();
1308 } else if (vt
->is_vector()) {
1309 type
= vt
->get_base_type();
1314 ir_dereference_record::ir_dereference_record(ir_rvalue
*value
,
1317 assert(value
!= NULL
);
1319 this->ir_type
= ir_type_dereference_record
;
1320 this->record
= value
;
1321 this->field
= ralloc_strdup(this, field
);
1322 this->type
= this->record
->type
->field_type(field
);
1326 ir_dereference_record::ir_dereference_record(ir_variable
*var
,
1329 void *ctx
= ralloc_parent(var
);
1331 this->ir_type
= ir_type_dereference_record
;
1332 this->record
= new(ctx
) ir_dereference_variable(var
);
1333 this->field
= ralloc_strdup(this, field
);
1334 this->type
= this->record
->type
->field_type(field
);
1338 ir_dereference::is_lvalue() const
1340 ir_variable
*var
= this->variable_referenced();
1342 /* Every l-value derference chain eventually ends in a variable.
1344 if ((var
== NULL
) || var
->read_only
)
1347 /* From page 17 (page 23 of the PDF) of the GLSL 1.20 spec:
1349 * "Samplers cannot be treated as l-values; hence cannot be used
1350 * as out or inout function parameters, nor can they be
1353 if (this->type
->contains_sampler())
1360 static const char *tex_opcode_strs
[] = { "tex", "txb", "txl", "txd", "txf", "txf_ms", "txs", "lod" };
1362 const char *ir_texture::opcode_string()
1364 assert((unsigned int) op
<=
1365 sizeof(tex_opcode_strs
) / sizeof(tex_opcode_strs
[0]));
1366 return tex_opcode_strs
[op
];
1370 ir_texture::get_opcode(const char *str
)
1372 const int count
= sizeof(tex_opcode_strs
) / sizeof(tex_opcode_strs
[0]);
1373 for (int op
= 0; op
< count
; op
++) {
1374 if (strcmp(str
, tex_opcode_strs
[op
]) == 0)
1375 return (ir_texture_opcode
) op
;
1377 return (ir_texture_opcode
) -1;
1382 ir_texture::set_sampler(ir_dereference
*sampler
, const glsl_type
*type
)
1384 assert(sampler
!= NULL
);
1385 assert(type
!= NULL
);
1386 this->sampler
= sampler
;
1389 if (this->op
== ir_txs
) {
1390 assert(type
->base_type
== GLSL_TYPE_INT
);
1391 } else if (this->op
== ir_lod
) {
1392 assert(type
->vector_elements
== 2);
1393 assert(type
->base_type
== GLSL_TYPE_FLOAT
);
1395 assert(sampler
->type
->sampler_type
== (int) type
->base_type
);
1396 if (sampler
->type
->sampler_shadow
)
1397 assert(type
->vector_elements
== 4 || type
->vector_elements
== 1);
1399 assert(type
->vector_elements
== 4);
1405 ir_swizzle::init_mask(const unsigned *comp
, unsigned count
)
1407 assert((count
>= 1) && (count
<= 4));
1409 memset(&this->mask
, 0, sizeof(this->mask
));
1410 this->mask
.num_components
= count
;
1412 unsigned dup_mask
= 0;
1415 assert(comp
[3] <= 3);
1416 dup_mask
|= (1U << comp
[3])
1417 & ((1U << comp
[0]) | (1U << comp
[1]) | (1U << comp
[2]));
1418 this->mask
.w
= comp
[3];
1421 assert(comp
[2] <= 3);
1422 dup_mask
|= (1U << comp
[2])
1423 & ((1U << comp
[0]) | (1U << comp
[1]));
1424 this->mask
.z
= comp
[2];
1427 assert(comp
[1] <= 3);
1428 dup_mask
|= (1U << comp
[1])
1429 & ((1U << comp
[0]));
1430 this->mask
.y
= comp
[1];
1433 assert(comp
[0] <= 3);
1434 this->mask
.x
= comp
[0];
1437 this->mask
.has_duplicates
= dup_mask
!= 0;
1439 /* Based on the number of elements in the swizzle and the base type
1440 * (i.e., float, int, unsigned, or bool) of the vector being swizzled,
1441 * generate the type of the resulting value.
1443 type
= glsl_type::get_instance(val
->type
->base_type
, mask
.num_components
, 1);
1446 ir_swizzle::ir_swizzle(ir_rvalue
*val
, unsigned x
, unsigned y
, unsigned z
,
1447 unsigned w
, unsigned count
)
1450 const unsigned components
[4] = { x
, y
, z
, w
};
1451 this->ir_type
= ir_type_swizzle
;
1452 this->init_mask(components
, count
);
1455 ir_swizzle::ir_swizzle(ir_rvalue
*val
, const unsigned *comp
,
1459 this->ir_type
= ir_type_swizzle
;
1460 this->init_mask(comp
, count
);
1463 ir_swizzle::ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
)
1465 this->ir_type
= ir_type_swizzle
;
1468 this->type
= glsl_type::get_instance(val
->type
->base_type
,
1469 mask
.num_components
, 1);
1478 ir_swizzle::create(ir_rvalue
*val
, const char *str
, unsigned vector_length
)
1480 void *ctx
= ralloc_parent(val
);
1482 /* For each possible swizzle character, this table encodes the value in
1483 * \c idx_map that represents the 0th element of the vector. For invalid
1484 * swizzle characters (e.g., 'k'), a special value is used that will allow
1485 * detection of errors.
1487 static const unsigned char base_idx
[26] = {
1488 /* a b c d e f g h i j k l m */
1489 R
, R
, I
, I
, I
, I
, R
, I
, I
, I
, I
, I
, I
,
1490 /* n o p q r s t u v w x y z */
1491 I
, I
, S
, S
, R
, S
, S
, I
, I
, X
, X
, X
, X
1494 /* Each valid swizzle character has an entry in the previous table. This
1495 * table encodes the base index encoded in the previous table plus the actual
1496 * index of the swizzle character. When processing swizzles, the first
1497 * character in the string is indexed in the previous table. Each character
1498 * in the string is indexed in this table, and the value found there has the
1499 * value form the first table subtracted. The result must be on the range
1502 * For example, the string "wzyx" will get X from the first table. Each of
1503 * the charcaters will get X+3, X+2, X+1, and X+0 from this table. After
1504 * subtraction, the swizzle values are { 3, 2, 1, 0 }.
1506 * The string "wzrg" will get X from the first table. Each of the characters
1507 * will get X+3, X+2, R+0, and R+1 from this table. After subtraction, the
1508 * swizzle values are { 3, 2, 4, 5 }. Since 4 and 5 are outside the range
1509 * [0,3], the error is detected.
1511 static const unsigned char idx_map
[26] = {
1512 /* a b c d e f g h i j k l m */
1513 R
+3, R
+2, 0, 0, 0, 0, R
+1, 0, 0, 0, 0, 0, 0,
1514 /* n o p q r s t u v w x y z */
1515 0, 0, S
+2, S
+3, R
+0, S
+0, S
+1, 0, 0, X
+3, X
+0, X
+1, X
+2
1518 int swiz_idx
[4] = { 0, 0, 0, 0 };
1522 /* Validate the first character in the swizzle string and look up the base
1523 * index value as described above.
1525 if ((str
[0] < 'a') || (str
[0] > 'z'))
1528 const unsigned base
= base_idx
[str
[0] - 'a'];
1531 for (i
= 0; (i
< 4) && (str
[i
] != '\0'); i
++) {
1532 /* Validate the next character, and, as described above, convert it to a
1535 if ((str
[i
] < 'a') || (str
[i
] > 'z'))
1538 swiz_idx
[i
] = idx_map
[str
[i
] - 'a'] - base
;
1539 if ((swiz_idx
[i
] < 0) || (swiz_idx
[i
] >= (int) vector_length
))
1546 return new(ctx
) ir_swizzle(val
, swiz_idx
[0], swiz_idx
[1], swiz_idx
[2],
1556 ir_swizzle::variable_referenced() const
1558 return this->val
->variable_referenced();
1562 ir_variable::ir_variable(const struct glsl_type
*type
, const char *name
,
1563 ir_variable_mode mode
)
1564 : max_array_access(0), read_only(false), centroid(false), invariant(false),
1565 mode(mode
), interpolation(INTERP_QUALIFIER_NONE
)
1567 this->ir_type
= ir_type_variable
;
1569 this->name
= ralloc_strdup(this, name
);
1570 this->explicit_location
= false;
1571 this->has_initializer
= false;
1572 this->location
= -1;
1573 this->location_frac
= 0;
1574 this->warn_extension
= NULL
;
1575 this->constant_value
= NULL
;
1576 this->constant_initializer
= NULL
;
1577 this->origin_upper_left
= false;
1578 this->pixel_center_integer
= false;
1579 this->depth_layout
= ir_depth_layout_none
;
1582 if (type
&& type
->base_type
== GLSL_TYPE_SAMPLER
)
1583 this->read_only
= true;
1588 ir_variable::interpolation_string() const
1590 switch (this->interpolation
) {
1591 case INTERP_QUALIFIER_NONE
: return "no";
1592 case INTERP_QUALIFIER_SMOOTH
: return "smooth";
1593 case INTERP_QUALIFIER_FLAT
: return "flat";
1594 case INTERP_QUALIFIER_NOPERSPECTIVE
: return "noperspective";
1597 assert(!"Should not get here.");
1602 glsl_interp_qualifier
1603 ir_variable::determine_interpolation_mode(bool flat_shade
)
1605 if (this->interpolation
!= INTERP_QUALIFIER_NONE
)
1606 return (glsl_interp_qualifier
) this->interpolation
;
1607 int location
= this->location
;
1609 location
== VARYING_SLOT_COL0
|| location
== VARYING_SLOT_COL1
;
1610 if (flat_shade
&& is_gl_Color
)
1611 return INTERP_QUALIFIER_FLAT
;
1613 return INTERP_QUALIFIER_SMOOTH
;
1617 ir_function_signature::ir_function_signature(const glsl_type
*return_type
,
1618 builtin_available_predicate b
)
1619 : return_type(return_type
), is_defined(false), builtin_info(b
),
1622 this->ir_type
= ir_type_function_signature
;
1623 this->origin
= NULL
;
1628 ir_function_signature::is_builtin() const
1630 return builtin_info
!= NULL
;
1635 ir_function_signature::is_builtin_available(const _mesa_glsl_parse_state
*state
) const
1637 /* We can't call the predicate without a state pointer, so just say that
1638 * the signature is available. At compile time, we need the filtering,
1639 * but also receive a valid state pointer. At link time, we're resolving
1640 * imported built-in prototypes to their definitions, which will always
1641 * be an exact match. So we can skip the filtering.
1646 assert(builtin_info
!= NULL
);
1647 return builtin_info(state
);
1652 modes_match(unsigned a
, unsigned b
)
1657 /* Accept "in" vs. "const in" */
1658 if ((a
== ir_var_const_in
&& b
== ir_var_function_in
) ||
1659 (b
== ir_var_const_in
&& a
== ir_var_function_in
))
1667 ir_function_signature::qualifiers_match(exec_list
*params
)
1669 exec_list_iterator iter_a
= parameters
.iterator();
1670 exec_list_iterator iter_b
= params
->iterator();
1672 /* check that the qualifiers match. */
1673 while (iter_a
.has_next()) {
1674 ir_variable
*a
= (ir_variable
*)iter_a
.get();
1675 ir_variable
*b
= (ir_variable
*)iter_b
.get();
1677 if (a
->read_only
!= b
->read_only
||
1678 !modes_match(a
->mode
, b
->mode
) ||
1679 a
->interpolation
!= b
->interpolation
||
1680 a
->centroid
!= b
->centroid
) {
1682 /* parameter a's qualifiers don't match */
1694 ir_function_signature::replace_parameters(exec_list
*new_params
)
1696 /* Destroy all of the previous parameter information. If the previous
1697 * parameter information comes from the function prototype, it may either
1698 * specify incorrect parameter names or not have names at all.
1700 foreach_iter(exec_list_iterator
, iter
, parameters
) {
1701 assert(((ir_instruction
*) iter
.get())->as_variable() != NULL
);
1706 new_params
->move_nodes_to(¶meters
);
1710 ir_function::ir_function(const char *name
)
1712 this->ir_type
= ir_type_function
;
1713 this->name
= ralloc_strdup(this, name
);
1718 ir_function::has_user_signature()
1720 foreach_list(n
, &this->signatures
) {
1721 ir_function_signature
*const sig
= (ir_function_signature
*) n
;
1722 if (!sig
->is_builtin())
1730 ir_rvalue::error_value(void *mem_ctx
)
1732 ir_rvalue
*v
= new(mem_ctx
) ir_rvalue
;
1734 v
->type
= glsl_type::error_type
;
1740 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
)
1742 foreach_iter(exec_list_iterator
, iter
, *list
) {
1743 ((ir_instruction
*)iter
.get())->accept(visitor
);
1749 steal_memory(ir_instruction
*ir
, void *new_ctx
)
1751 ir_variable
*var
= ir
->as_variable();
1752 ir_constant
*constant
= ir
->as_constant();
1753 if (var
!= NULL
&& var
->constant_value
!= NULL
)
1754 steal_memory(var
->constant_value
, ir
);
1756 if (var
!= NULL
&& var
->constant_initializer
!= NULL
)
1757 steal_memory(var
->constant_initializer
, ir
);
1759 /* The components of aggregate constants are not visited by the normal
1760 * visitor, so steal their values by hand.
1762 if (constant
!= NULL
) {
1763 if (constant
->type
->is_record()) {
1764 foreach_iter(exec_list_iterator
, iter
, constant
->components
) {
1765 ir_constant
*field
= (ir_constant
*)iter
.get();
1766 steal_memory(field
, ir
);
1768 } else if (constant
->type
->is_array()) {
1769 for (unsigned int i
= 0; i
< constant
->type
->length
; i
++) {
1770 steal_memory(constant
->array_elements
[i
], ir
);
1775 ralloc_steal(new_ctx
, ir
);
1780 reparent_ir(exec_list
*list
, void *mem_ctx
)
1782 foreach_list(node
, list
) {
1783 visit_tree((ir_instruction
*) node
, steal_memory
, mem_ctx
);
1789 try_min_one(ir_rvalue
*ir
)
1791 ir_expression
*expr
= ir
->as_expression();
1793 if (!expr
|| expr
->operation
!= ir_binop_min
)
1796 if (expr
->operands
[0]->is_one())
1797 return expr
->operands
[1];
1799 if (expr
->operands
[1]->is_one())
1800 return expr
->operands
[0];
1806 try_max_zero(ir_rvalue
*ir
)
1808 ir_expression
*expr
= ir
->as_expression();
1810 if (!expr
|| expr
->operation
!= ir_binop_max
)
1813 if (expr
->operands
[0]->is_zero())
1814 return expr
->operands
[1];
1816 if (expr
->operands
[1]->is_zero())
1817 return expr
->operands
[0];
1823 ir_rvalue::as_rvalue_to_saturate()
1825 ir_expression
*expr
= this->as_expression();
1830 ir_rvalue
*max_zero
= try_max_zero(expr
);
1832 return try_min_one(max_zero
);
1834 ir_rvalue
*min_one
= try_min_one(expr
);
1836 return try_max_zero(min_one
);
1845 vertices_per_prim(GLenum prim
)
1854 case GL_LINES_ADJACENCY
:
1856 case GL_TRIANGLES_ADJACENCY
:
1859 assert(!"Bad primitive");