3 * Copyright © 2010 Intel Corporation
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22 * DEALINGS IN THE SOFTWARE.
33 #include "glsl_types.h"
35 #include "ir_visitor.h"
36 #include "ir_hierarchical_visitor.h"
37 #include "main/mtypes.h"
40 * \defgroup IR Intermediate representation nodes
48 * Each concrete class derived from \c ir_instruction has a value in this
49 * enumerant. The value for the type is stored in \c ir_instruction::ir_type
50 * by the constructor. While using type tags is not very C++, it is extremely
51 * convenient. For example, during debugging you can simply inspect
52 * \c ir_instruction::ir_type to find out the actual type of the object.
54 * In addition, it is possible to use a switch-statement based on \c
55 * \c ir_instruction::ir_type to select different behavior for different object
56 * types. For functions that have only slight differences for several object
57 * types, this allows writing very straightforward, readable code.
61 * Zero is unused so that the IR validator can detect cases where
62 * \c ir_instruction::ir_type has not been initialized.
69 ir_type_dereference_array,
70 ir_type_dereference_record,
71 ir_type_dereference_variable,
75 ir_type_function_signature,
82 ir_type_max /**< maximum ir_type enum number, for validation */
86 * Base class of all IR instructions
88 class ir_instruction : public exec_node {
90 enum ir_node_type ir_type;
93 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
94 * there's a virtual destructor present. Because we almost
95 * universally use ralloc for our memory management of
96 * ir_instructions, the destructor doesn't need to do any work.
98 virtual ~ir_instruction()
102 /** ir_print_visitor helper for debugging. */
103 void print(void) const;
105 virtual void accept(ir_visitor *) = 0;
106 virtual ir_visitor_status accept(ir_hierarchical_visitor *) = 0;
107 virtual ir_instruction *clone(void *mem_ctx,
108 struct hash_table *ht) const = 0;
111 * \name IR instruction downcast functions
113 * These functions either cast the object to a derived class or return
114 * \c NULL if the object's type does not match the specified derived class.
115 * Additional downcast functions will be added as needed.
118 virtual class ir_variable * as_variable() { return NULL; }
119 virtual class ir_function * as_function() { return NULL; }
120 virtual class ir_dereference * as_dereference() { return NULL; }
121 virtual class ir_dereference_array * as_dereference_array() { return NULL; }
122 virtual class ir_dereference_variable *as_dereference_variable() { return NULL; }
123 virtual class ir_expression * as_expression() { return NULL; }
124 virtual class ir_rvalue * as_rvalue() { return NULL; }
125 virtual class ir_loop * as_loop() { return NULL; }
126 virtual class ir_assignment * as_assignment() { return NULL; }
127 virtual class ir_call * as_call() { return NULL; }
128 virtual class ir_return * as_return() { return NULL; }
129 virtual class ir_if * as_if() { return NULL; }
130 virtual class ir_swizzle * as_swizzle() { return NULL; }
131 virtual class ir_constant * as_constant() { return NULL; }
132 virtual class ir_discard * as_discard() { return NULL; }
138 ir_type = ir_type_unset;
144 * The base class for all "values"/expression trees.
146 class ir_rvalue : public ir_instruction {
148 const struct glsl_type *type;
150 virtual ir_rvalue *clone(void *mem_ctx, struct hash_table *) const;
152 virtual void accept(ir_visitor *v)
157 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
159 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
161 virtual ir_rvalue * as_rvalue()
166 ir_rvalue *as_rvalue_to_saturate();
168 virtual bool is_lvalue() const
174 * Get the variable that is ultimately referenced by an r-value
176 virtual ir_variable *variable_referenced() const
183 * If an r-value is a reference to a whole variable, get that variable
186 * Pointer to a variable that is completely dereferenced by the r-value. If
187 * the r-value is not a dereference or the dereference does not access the
188 * entire variable (i.e., it's just one array element, struct field), \c NULL
191 virtual ir_variable *whole_variable_referenced()
197 * Determine if an r-value has the value zero
199 * The base implementation of this function always returns \c false. The
200 * \c ir_constant class over-rides this function to return \c true \b only
201 * for vector and scalar types that have all elements set to the value
202 * zero (or \c false for booleans).
204 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one
206 virtual bool is_zero() const;
209 * Determine if an r-value has the value one
211 * The base implementation of this function always returns \c false. The
212 * \c ir_constant class over-rides this function to return \c true \b only
213 * for vector and scalar types that have all elements set to the value
214 * one (or \c true for booleans).
216 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one
218 virtual bool is_one() const;
221 * Determine if an r-value has the value negative one
223 * The base implementation of this function always returns \c false. The
224 * \c ir_constant class over-rides this function to return \c true \b only
225 * for vector and scalar types that have all elements set to the value
226 * negative one. For boolean times, the result is always \c false.
228 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
230 virtual bool is_negative_one() const;
234 * Return a generic value of error_type.
236 * Allocation will be performed with 'mem_ctx' as ralloc owner.
238 static ir_rvalue *error_value(void *mem_ctx);
246 * Variable storage classes
248 enum ir_variable_mode {
249 ir_var_auto = 0, /**< Function local variables and globals. */
250 ir_var_uniform, /**< Variable declared as a uniform. */
254 ir_var_const_in, /**< "in" param that must be a constant expression */
255 ir_var_system_value, /**< Ex: front-face, instance-id, etc. */
256 ir_var_temporary /**< Temporary variable generated during compilation. */
260 * \brief Layout qualifiers for gl_FragDepth.
262 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
263 * with a layout qualifier.
265 enum ir_depth_layout {
266 ir_depth_layout_none, /**< No depth layout is specified. */
268 ir_depth_layout_greater,
269 ir_depth_layout_less,
270 ir_depth_layout_unchanged
274 * \brief Convert depth layout qualifier to string.
277 depth_layout_string(ir_depth_layout layout);
280 * Description of built-in state associated with a uniform
282 * \sa ir_variable::state_slots
284 struct ir_state_slot {
289 class ir_variable : public ir_instruction {
291 ir_variable(const struct glsl_type *, const char *, ir_variable_mode);
293 virtual ir_variable *clone(void *mem_ctx, struct hash_table *ht) const;
295 virtual ir_variable *as_variable()
300 virtual void accept(ir_visitor *v)
305 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
309 * Get the string value for the interpolation qualifier
311 * \return The string that would be used in a shader to specify \c
312 * mode will be returned.
314 * This function is used to generate error messages of the form "shader
315 * uses %s interpolation qualifier", so in the case where there is no
316 * interpolation qualifier, it returns "no".
318 * This function should only be used on a shader input or output variable.
320 const char *interpolation_string() const;
323 * Determine how this variable should be interpolated based on its
324 * interpolation qualifier (if present), whether it is gl_Color or
325 * gl_SecondaryColor, and whether flatshading is enabled in the current GL
328 * The return value will always be either INTERP_QUALIFIER_SMOOTH,
329 * INTERP_QUALIFIER_NOPERSPECTIVE, or INTERP_QUALIFIER_FLAT.
331 glsl_interp_qualifier determine_interpolation_mode(bool flat_shade);
334 * Declared type of the variable
336 const struct glsl_type *type;
339 * Delcared name of the variable
344 * Highest element accessed with a constant expression array index
346 * Not used for non-array variables.
348 unsigned max_array_access;
351 * Is the variable read-only?
353 * This is set for variables declared as \c const, shader inputs,
356 unsigned read_only:1;
358 unsigned invariant:1;
361 * Has this variable been used for reading or writing?
363 * Several GLSL semantic checks require knowledge of whether or not a
364 * variable has been used. For example, it is an error to redeclare a
365 * variable as invariant after it has been used.
367 * This is only maintained in the ast_to_hir.cpp path, not in
368 * Mesa's fixed function or ARB program paths.
373 * Has this variable been statically assigned?
375 * This answers whether the variable was assigned in any path of
376 * the shader during ast_to_hir. This doesn't answer whether it is
377 * still written after dead code removal, nor is it maintained in
378 * non-ast_to_hir.cpp (GLSL parsing) paths.
383 * Storage class of the variable.
385 * \sa ir_variable_mode
390 * Interpolation mode for shader inputs / outputs
392 * \sa ir_variable_interpolation
394 unsigned interpolation:2;
397 * \name ARB_fragment_coord_conventions
400 unsigned origin_upper_left:1;
401 unsigned pixel_center_integer:1;
405 * Was the location explicitly set in the shader?
407 * If the location is explicitly set in the shader, it \b cannot be changed
408 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
411 unsigned explicit_location:1;
412 unsigned explicit_index:1;
415 * Does this variable have an initializer?
417 * This is used by the linker to cross-validiate initializers of global
420 unsigned has_initializer:1;
423 * \brief Layout qualifier for gl_FragDepth.
425 * This is not equal to \c ir_depth_layout_none if and only if this
426 * variable is \c gl_FragDepth and a layout qualifier is specified.
428 ir_depth_layout depth_layout;
431 * Storage location of the base of this variable
433 * The precise meaning of this field depends on the nature of the variable.
435 * - Vertex shader input: one of the values from \c gl_vert_attrib.
436 * - Vertex shader output: one of the values from \c gl_vert_result.
437 * - Fragment shader input: one of the values from \c gl_frag_attrib.
438 * - Fragment shader output: one of the values from \c gl_frag_result.
439 * - Uniforms: Per-stage uniform slot number.
440 * - Other: This field is not currently used.
442 * If the variable is a uniform, shader input, or shader output, and the
443 * slot has not been assigned, the value will be -1.
448 * output index for dual source blending.
453 * Built-in state that backs this uniform
455 * Once set at variable creation, \c state_slots must remain invariant.
456 * This is because, ideally, this array would be shared by all clones of
457 * this variable in the IR tree. In other words, we'd really like for it
458 * to be a fly-weight.
460 * If the variable is not a uniform, \c num_state_slots will be zero and
461 * \c state_slots will be \c NULL.
464 unsigned num_state_slots; /**< Number of state slots used */
465 ir_state_slot *state_slots; /**< State descriptors. */
469 * Emit a warning if this variable is accessed.
471 const char *warn_extension;
474 * Value assigned in the initializer of a variable declared "const"
476 ir_constant *constant_value;
479 * Constant expression assigned in the initializer of the variable
482 * This field and \c ::constant_value are distinct. Even if the two fields
483 * refer to constants with the same value, they must point to separate
486 ir_constant *constant_initializer;
492 * The representation of a function instance; may be the full definition or
493 * simply a prototype.
495 class ir_function_signature : public ir_instruction {
496 /* An ir_function_signature will be part of the list of signatures in
500 ir_function_signature(const glsl_type *return_type);
502 virtual ir_function_signature *clone(void *mem_ctx,
503 struct hash_table *ht) const;
504 ir_function_signature *clone_prototype(void *mem_ctx,
505 struct hash_table *ht) const;
507 virtual void accept(ir_visitor *v)
512 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
515 * Attempt to evaluate this function as a constant expression,
516 * given a list of the actual parameters and the variable context.
517 * Returns NULL for non-built-ins.
519 ir_constant *constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context);
522 * Get the name of the function for which this is a signature
524 const char *function_name() const;
527 * Get a handle to the function for which this is a signature
529 * There is no setter function, this function returns a \c const pointer,
530 * and \c ir_function_signature::_function is private for a reason. The
531 * only way to make a connection between a function and function signature
532 * is via \c ir_function::add_signature. This helps ensure that certain
533 * invariants (i.e., a function signature is in the list of signatures for
534 * its \c _function) are met.
536 * \sa ir_function::add_signature
538 inline const class ir_function *function() const
540 return this->_function;
544 * Check whether the qualifiers match between this signature's parameters
545 * and the supplied parameter list. If not, returns the name of the first
546 * parameter with mismatched qualifiers (for use in error messages).
548 const char *qualifiers_match(exec_list *params);
551 * Replace the current parameter list with the given one. This is useful
552 * if the current information came from a prototype, and either has invalid
553 * or missing parameter names.
555 void replace_parameters(exec_list *new_params);
558 * Function return type.
560 * \note This discards the optional precision qualifier.
562 const struct glsl_type *return_type;
565 * List of ir_variable of function parameters.
567 * This represents the storage. The paramaters passed in a particular
568 * call will be in ir_call::actual_paramaters.
570 struct exec_list parameters;
572 /** Whether or not this function has a body (which may be empty). */
573 unsigned is_defined:1;
575 /** Whether or not this function signature is a built-in. */
576 unsigned is_builtin:1;
578 /** Body of instructions in the function. */
579 struct exec_list body;
582 /** Function of which this signature is one overload. */
583 class ir_function *_function;
585 /** Function signature of which this one is a prototype clone */
586 const ir_function_signature *origin;
588 friend class ir_function;
591 * Helper function to run a list of instructions for constant
592 * expression evaluation.
594 * The hash table represents the values of the visible variables.
595 * There are no scoping issues because the table is indexed on
596 * ir_variable pointers, not variable names.
598 * Returns false if the expression is not constant, true otherwise,
599 * and the value in *result if result is non-NULL.
601 bool constant_expression_evaluate_expression_list(const struct exec_list &body,
602 struct hash_table *variable_context,
603 ir_constant **result);
608 * Header for tracking multiple overloaded functions with the same name.
609 * Contains a list of ir_function_signatures representing each of the
612 class ir_function : public ir_instruction {
614 ir_function(const char *name);
616 virtual ir_function *clone(void *mem_ctx, struct hash_table *ht) const;
618 virtual ir_function *as_function()
623 virtual void accept(ir_visitor *v)
628 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
630 void add_signature(ir_function_signature *sig)
632 sig->_function = this;
633 this->signatures.push_tail(sig);
637 * Get an iterator for the set of function signatures
639 exec_list_iterator iterator()
641 return signatures.iterator();
645 * Find a signature that matches a set of actual parameters, taking implicit
646 * conversions into account. Also flags whether the match was exact.
648 ir_function_signature *matching_signature(const exec_list *actual_param,
649 bool *match_is_exact);
652 * Find a signature that matches a set of actual parameters, taking implicit
653 * conversions into account.
655 ir_function_signature *matching_signature(const exec_list *actual_param);
658 * Find a signature that exactly matches a set of actual parameters without
659 * any implicit type conversions.
661 ir_function_signature *exact_matching_signature(const exec_list *actual_ps);
664 * Name of the function.
668 /** Whether or not this function has a signature that isn't a built-in. */
669 bool has_user_signature();
672 * List of ir_function_signature for each overloaded function with this name.
674 struct exec_list signatures;
677 inline const char *ir_function_signature::function_name() const
679 return this->_function->name;
685 * IR instruction representing high-level if-statements
687 class ir_if : public ir_instruction {
689 ir_if(ir_rvalue *condition)
690 : condition(condition)
692 ir_type = ir_type_if;
695 virtual ir_if *clone(void *mem_ctx, struct hash_table *ht) const;
697 virtual ir_if *as_if()
702 virtual void accept(ir_visitor *v)
707 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
709 ir_rvalue *condition;
710 /** List of ir_instruction for the body of the then branch */
711 exec_list then_instructions;
712 /** List of ir_instruction for the body of the else branch */
713 exec_list else_instructions;
718 * IR instruction representing a high-level loop structure.
720 class ir_loop : public ir_instruction {
724 virtual ir_loop *clone(void *mem_ctx, struct hash_table *ht) const;
726 virtual void accept(ir_visitor *v)
731 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
733 virtual ir_loop *as_loop()
739 * Get an iterator for the instructions of the loop body
741 exec_list_iterator iterator()
743 return body_instructions.iterator();
746 /** List of ir_instruction that make up the body of the loop. */
747 exec_list body_instructions;
750 * \name Loop counter and controls
752 * Represents a loop like a FORTRAN \c do-loop.
755 * If \c from and \c to are the same value, the loop will execute once.
758 ir_rvalue *from; /** Value of the loop counter on the first
759 * iteration of the loop.
761 ir_rvalue *to; /** Value of the loop counter on the last
762 * iteration of the loop.
764 ir_rvalue *increment;
765 ir_variable *counter;
768 * Comparison operation in the loop terminator.
770 * If any of the loop control fields are non-\c NULL, this field must be
771 * one of \c ir_binop_less, \c ir_binop_greater, \c ir_binop_lequal,
772 * \c ir_binop_gequal, \c ir_binop_equal, or \c ir_binop_nequal.
779 class ir_assignment : public ir_instruction {
781 ir_assignment(ir_rvalue *lhs, ir_rvalue *rhs, ir_rvalue *condition = NULL);
784 * Construct an assignment with an explicit write mask
787 * Since a write mask is supplied, the LHS must already be a bare
788 * \c ir_dereference. The cannot be any swizzles in the LHS.
790 ir_assignment(ir_dereference *lhs, ir_rvalue *rhs, ir_rvalue *condition,
791 unsigned write_mask);
793 virtual ir_assignment *clone(void *mem_ctx, struct hash_table *ht) const;
795 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
797 virtual void accept(ir_visitor *v)
802 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
804 virtual ir_assignment * as_assignment()
810 * Get a whole variable written by an assignment
812 * If the LHS of the assignment writes a whole variable, the variable is
813 * returned. Otherwise \c NULL is returned. Examples of whole-variable
816 * - Assigning to a scalar
817 * - Assigning to all components of a vector
818 * - Whole array (or matrix) assignment
819 * - Whole structure assignment
821 ir_variable *whole_variable_written();
824 * Set the LHS of an assignment
826 void set_lhs(ir_rvalue *lhs);
829 * Left-hand side of the assignment.
831 * This should be treated as read only. If you need to set the LHS of an
832 * assignment, use \c ir_assignment::set_lhs.
837 * Value being assigned
842 * Optional condition for the assignment.
844 ir_rvalue *condition;
848 * Component mask written
850 * For non-vector types in the LHS, this field will be zero. For vector
851 * types, a bit will be set for each component that is written. Note that
852 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
854 * A partially-set write mask means that each enabled channel gets
855 * the value from a consecutive channel of the rhs. For example,
856 * to write just .xyw of gl_FrontColor with color:
858 * (assign (constant bool (1)) (xyw)
859 * (var_ref gl_FragColor)
860 * (swiz xyw (var_ref color)))
862 unsigned write_mask:4;
865 /* Update ir_expression::num_operands() and operator_strs when
866 * updating this list.
868 enum ir_expression_operation {
877 ir_unop_exp, /**< Log base e on gentype */
878 ir_unop_log, /**< Natural log on gentype */
881 ir_unop_f2i, /**< Float-to-integer conversion. */
882 ir_unop_i2f, /**< Integer-to-float conversion. */
883 ir_unop_f2b, /**< Float-to-boolean conversion */
884 ir_unop_b2f, /**< Boolean-to-float conversion */
885 ir_unop_i2b, /**< int-to-boolean conversion */
886 ir_unop_b2i, /**< Boolean-to-int conversion */
887 ir_unop_u2f, /**< Unsigned-to-float conversion. */
888 ir_unop_i2u, /**< Integer-to-unsigned conversion. */
889 ir_unop_u2i, /**< Unsigned-to-integer conversion. */
890 ir_unop_bitcast_i2f, /**< Bit-identical int-to-float "conversion" */
891 ir_unop_bitcast_f2i, /**< Bit-identical float-to-int "conversion" */
892 ir_unop_bitcast_u2f, /**< Bit-identical uint-to-float "conversion" */
893 ir_unop_bitcast_f2u, /**< Bit-identical float-to-uint "conversion" */
897 * \name Unary floating-point rounding operations.
908 * \name Trigonometric operations.
913 ir_unop_sin_reduced, /**< Reduced range sin. [-pi, pi] */
914 ir_unop_cos_reduced, /**< Reduced range cos. [-pi, pi] */
918 * \name Partial derivatives.
928 * A sentinel marking the last of the unary operations.
930 ir_last_unop = ir_unop_noise,
938 * Takes one of two combinations of arguments:
943 * Does not take integer types.
948 * \name Binary comparison operators which return a boolean vector.
949 * The type of both operands must be equal.
959 * Returns single boolean for whether all components of operands[0]
960 * equal the components of operands[1].
964 * Returns single boolean for whether any component of operands[0]
965 * is not equal to the corresponding component of operands[1].
971 * \name Bit-wise binary operations.
992 * A sentinel marking the last of the binary operations.
994 ir_last_binop = ir_binop_pow,
999 * A sentinel marking the last of all operations.
1001 ir_last_opcode = ir_last_binop
1004 class ir_expression : public ir_rvalue {
1007 * Constructor for unary operation expressions
1009 ir_expression(int op, const struct glsl_type *type, ir_rvalue *);
1010 ir_expression(int op, ir_rvalue *);
1013 * Constructor for binary operation expressions
1015 ir_expression(int op, const struct glsl_type *type,
1016 ir_rvalue *, ir_rvalue *);
1017 ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1);
1020 * Constructor for quad operator expressions
1022 ir_expression(int op, const struct glsl_type *type,
1023 ir_rvalue *, ir_rvalue *, ir_rvalue *, ir_rvalue *);
1025 virtual ir_expression *as_expression()
1030 virtual ir_expression *clone(void *mem_ctx, struct hash_table *ht) const;
1033 * Attempt to constant-fold the expression
1035 * The "variable_context" hash table links ir_variable * to ir_constant *
1036 * that represent the variables' values. \c NULL represents an empty
1039 * If the expression cannot be constant folded, this method will return
1042 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1045 * Determine the number of operands used by an expression
1047 static unsigned int get_num_operands(ir_expression_operation);
1050 * Determine the number of operands used by an expression
1052 unsigned int get_num_operands() const
1054 return (this->operation == ir_quadop_vector)
1055 ? this->type->vector_elements : get_num_operands(operation);
1059 * Return a string representing this expression's operator.
1061 const char *operator_string();
1064 * Return a string representing this expression's operator.
1066 static const char *operator_string(ir_expression_operation);
1070 * Do a reverse-lookup to translate the given string into an operator.
1072 static ir_expression_operation get_operator(const char *);
1074 virtual void accept(ir_visitor *v)
1079 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1081 ir_expression_operation operation;
1082 ir_rvalue *operands[4];
1087 * HIR instruction representing a high-level function call, containing a list
1088 * of parameters and returning a value in the supplied temporary.
1090 class ir_call : public ir_instruction {
1092 ir_call(ir_function_signature *callee,
1093 ir_dereference_variable *return_deref,
1094 exec_list *actual_parameters)
1095 : return_deref(return_deref), callee(callee)
1097 ir_type = ir_type_call;
1098 assert(callee->return_type != NULL);
1099 actual_parameters->move_nodes_to(& this->actual_parameters);
1100 this->use_builtin = callee->is_builtin;
1103 virtual ir_call *clone(void *mem_ctx, struct hash_table *ht) const;
1105 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1107 virtual ir_call *as_call()
1112 virtual void accept(ir_visitor *v)
1117 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1120 * Get an iterator for the set of acutal parameters
1122 exec_list_iterator iterator()
1124 return actual_parameters.iterator();
1128 * Get the name of the function being called.
1130 const char *callee_name() const
1132 return callee->function_name();
1136 * Generates an inline version of the function before @ir,
1137 * storing the return value in return_deref.
1139 void generate_inline(ir_instruction *ir);
1142 * Storage for the function's return value.
1143 * This must be NULL if the return type is void.
1145 ir_dereference_variable *return_deref;
1148 * The specific function signature being called.
1150 ir_function_signature *callee;
1152 /* List of ir_rvalue of paramaters passed in this call. */
1153 exec_list actual_parameters;
1155 /** Should this call only bind to a built-in function? */
1161 * \name Jump-like IR instructions.
1163 * These include \c break, \c continue, \c return, and \c discard.
1166 class ir_jump : public ir_instruction {
1170 ir_type = ir_type_unset;
1174 class ir_return : public ir_jump {
1179 this->ir_type = ir_type_return;
1182 ir_return(ir_rvalue *value)
1185 this->ir_type = ir_type_return;
1188 virtual ir_return *clone(void *mem_ctx, struct hash_table *) const;
1190 virtual ir_return *as_return()
1195 ir_rvalue *get_value() const
1200 virtual void accept(ir_visitor *v)
1205 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1212 * Jump instructions used inside loops
1214 * These include \c break and \c continue. The \c break within a loop is
1215 * different from the \c break within a switch-statement.
1217 * \sa ir_switch_jump
1219 class ir_loop_jump : public ir_jump {
1226 ir_loop_jump(jump_mode mode)
1228 this->ir_type = ir_type_loop_jump;
1233 virtual ir_loop_jump *clone(void *mem_ctx, struct hash_table *) const;
1235 virtual void accept(ir_visitor *v)
1240 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1242 bool is_break() const
1244 return mode == jump_break;
1247 bool is_continue() const
1249 return mode == jump_continue;
1252 /** Mode selector for the jump instruction. */
1253 enum jump_mode mode;
1255 /** Loop containing this break instruction. */
1260 * IR instruction representing discard statements.
1262 class ir_discard : public ir_jump {
1266 this->ir_type = ir_type_discard;
1267 this->condition = NULL;
1270 ir_discard(ir_rvalue *cond)
1272 this->ir_type = ir_type_discard;
1273 this->condition = cond;
1276 virtual ir_discard *clone(void *mem_ctx, struct hash_table *ht) const;
1278 virtual void accept(ir_visitor *v)
1283 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1285 virtual ir_discard *as_discard()
1290 ir_rvalue *condition;
1296 * Texture sampling opcodes used in ir_texture
1298 enum ir_texture_opcode {
1299 ir_tex, /**< Regular texture look-up */
1300 ir_txb, /**< Texture look-up with LOD bias */
1301 ir_txl, /**< Texture look-up with explicit LOD */
1302 ir_txd, /**< Texture look-up with partial derivatvies */
1303 ir_txf, /**< Texel fetch with explicit LOD */
1304 ir_txs /**< Texture size */
1309 * IR instruction to sample a texture
1311 * The specific form of the IR instruction depends on the \c mode value
1312 * selected from \c ir_texture_opcodes. In the printed IR, these will
1315 * Texel offset (0 or an expression)
1316 * | Projection divisor
1317 * | | Shadow comparitor
1320 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1321 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1322 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1323 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1324 * (txf <type> <sampler> <coordinate> 0 <lod>)
1325 * (txs <type> <sampler> <lod>)
1327 class ir_texture : public ir_rvalue {
1329 ir_texture(enum ir_texture_opcode op)
1330 : op(op), projector(NULL), shadow_comparitor(NULL), offset(NULL)
1332 this->ir_type = ir_type_texture;
1335 virtual ir_texture *clone(void *mem_ctx, struct hash_table *) const;
1337 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1339 virtual void accept(ir_visitor *v)
1344 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1347 * Return a string representing the ir_texture_opcode.
1349 const char *opcode_string();
1351 /** Set the sampler and type. */
1352 void set_sampler(ir_dereference *sampler, const glsl_type *type);
1355 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1357 static ir_texture_opcode get_opcode(const char *);
1359 enum ir_texture_opcode op;
1361 /** Sampler to use for the texture access. */
1362 ir_dereference *sampler;
1364 /** Texture coordinate to sample */
1365 ir_rvalue *coordinate;
1368 * Value used for projective divide.
1370 * If there is no projective divide (the common case), this will be
1371 * \c NULL. Optimization passes should check for this to point to a constant
1372 * of 1.0 and replace that with \c NULL.
1374 ir_rvalue *projector;
1377 * Coordinate used for comparison on shadow look-ups.
1379 * If there is no shadow comparison, this will be \c NULL. For the
1380 * \c ir_txf opcode, this *must* be \c NULL.
1382 ir_rvalue *shadow_comparitor;
1384 /** Texel offset. */
1388 ir_rvalue *lod; /**< Floating point LOD */
1389 ir_rvalue *bias; /**< Floating point LOD bias */
1391 ir_rvalue *dPdx; /**< Partial derivative of coordinate wrt X */
1392 ir_rvalue *dPdy; /**< Partial derivative of coordinate wrt Y */
1398 struct ir_swizzle_mask {
1405 * Number of components in the swizzle.
1407 unsigned num_components:3;
1410 * Does the swizzle contain duplicate components?
1412 * L-value swizzles cannot contain duplicate components.
1414 unsigned has_duplicates:1;
1418 class ir_swizzle : public ir_rvalue {
1420 ir_swizzle(ir_rvalue *, unsigned x, unsigned y, unsigned z, unsigned w,
1423 ir_swizzle(ir_rvalue *val, const unsigned *components, unsigned count);
1425 ir_swizzle(ir_rvalue *val, ir_swizzle_mask mask);
1427 virtual ir_swizzle *clone(void *mem_ctx, struct hash_table *) const;
1429 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1431 virtual ir_swizzle *as_swizzle()
1437 * Construct an ir_swizzle from the textual representation. Can fail.
1439 static ir_swizzle *create(ir_rvalue *, const char *, unsigned vector_length);
1441 virtual void accept(ir_visitor *v)
1446 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1448 bool is_lvalue() const
1450 return val->is_lvalue() && !mask.has_duplicates;
1454 * Get the variable that is ultimately referenced by an r-value
1456 virtual ir_variable *variable_referenced() const;
1459 ir_swizzle_mask mask;
1463 * Initialize the mask component of a swizzle
1465 * This is used by the \c ir_swizzle constructors.
1467 void init_mask(const unsigned *components, unsigned count);
1471 class ir_dereference : public ir_rvalue {
1473 virtual ir_dereference *clone(void *mem_ctx, struct hash_table *) const = 0;
1475 virtual ir_dereference *as_dereference()
1480 bool is_lvalue() const;
1483 * Get the variable that is ultimately referenced by an r-value
1485 virtual ir_variable *variable_referenced() const = 0;
1488 * Get the constant that is ultimately referenced by an r-value,
1489 * in a constant expression evaluation context.
1491 * The offset is used when the reference is to a specific column of
1494 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const = 0;
1498 class ir_dereference_variable : public ir_dereference {
1500 ir_dereference_variable(ir_variable *var);
1502 virtual ir_dereference_variable *clone(void *mem_ctx,
1503 struct hash_table *) const;
1505 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1507 virtual ir_dereference_variable *as_dereference_variable()
1513 * Get the variable that is ultimately referenced by an r-value
1515 virtual ir_variable *variable_referenced() const
1521 * Get the constant that is ultimately referenced by an r-value,
1522 * in a constant expression evaluation context.
1524 * The offset is used when the reference is to a specific column of
1527 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1529 virtual ir_variable *whole_variable_referenced()
1531 /* ir_dereference_variable objects always dereference the entire
1532 * variable. However, if this dereference is dereferenced by anything
1533 * else, the complete deferefernce chain is not a whole-variable
1534 * dereference. This method should only be called on the top most
1535 * ir_rvalue in a dereference chain.
1540 virtual void accept(ir_visitor *v)
1545 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1548 * Object being dereferenced.
1554 class ir_dereference_array : public ir_dereference {
1556 ir_dereference_array(ir_rvalue *value, ir_rvalue *array_index);
1558 ir_dereference_array(ir_variable *var, ir_rvalue *array_index);
1560 virtual ir_dereference_array *clone(void *mem_ctx,
1561 struct hash_table *) const;
1563 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1565 virtual ir_dereference_array *as_dereference_array()
1571 * Get the variable that is ultimately referenced by an r-value
1573 virtual ir_variable *variable_referenced() const
1575 return this->array->variable_referenced();
1579 * Get the constant that is ultimately referenced by an r-value,
1580 * in a constant expression evaluation context.
1582 * The offset is used when the reference is to a specific column of
1585 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1587 virtual void accept(ir_visitor *v)
1592 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1595 ir_rvalue *array_index;
1598 void set_array(ir_rvalue *value);
1602 class ir_dereference_record : public ir_dereference {
1604 ir_dereference_record(ir_rvalue *value, const char *field);
1606 ir_dereference_record(ir_variable *var, const char *field);
1608 virtual ir_dereference_record *clone(void *mem_ctx,
1609 struct hash_table *) const;
1611 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1614 * Get the variable that is ultimately referenced by an r-value
1616 virtual ir_variable *variable_referenced() const
1618 return this->record->variable_referenced();
1622 * Get the constant that is ultimately referenced by an r-value,
1623 * in a constant expression evaluation context.
1625 * The offset is used when the reference is to a specific column of
1628 virtual void constant_referenced(struct hash_table *variable_context, ir_constant *&store, int &offset) const;
1630 virtual void accept(ir_visitor *v)
1635 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1643 * Data stored in an ir_constant
1645 union ir_constant_data {
1653 class ir_constant : public ir_rvalue {
1655 ir_constant(const struct glsl_type *type, const ir_constant_data *data);
1656 ir_constant(bool b);
1657 ir_constant(unsigned int u);
1659 ir_constant(float f);
1662 * Construct an ir_constant from a list of ir_constant values
1664 ir_constant(const struct glsl_type *type, exec_list *values);
1667 * Construct an ir_constant from a scalar component of another ir_constant
1669 * The new \c ir_constant inherits the type of the component from the
1673 * In the case of a matrix constant, the new constant is a scalar, \b not
1676 ir_constant(const ir_constant *c, unsigned i);
1679 * Return a new ir_constant of the specified type containing all zeros.
1681 static ir_constant *zero(void *mem_ctx, const glsl_type *type);
1683 virtual ir_constant *clone(void *mem_ctx, struct hash_table *) const;
1685 virtual ir_constant *constant_expression_value(struct hash_table *variable_context = NULL);
1687 virtual ir_constant *as_constant()
1692 virtual void accept(ir_visitor *v)
1697 virtual ir_visitor_status accept(ir_hierarchical_visitor *);
1700 * Get a particular component of a constant as a specific type
1702 * This is useful, for example, to get a value from an integer constant
1703 * as a float or bool. This appears frequently when constructors are
1704 * called with all constant parameters.
1707 bool get_bool_component(unsigned i) const;
1708 float get_float_component(unsigned i) const;
1709 int get_int_component(unsigned i) const;
1710 unsigned get_uint_component(unsigned i) const;
1713 ir_constant *get_array_element(unsigned i) const;
1715 ir_constant *get_record_field(const char *name);
1718 * Copy the values on another constant at a given offset.
1720 * The offset is ignored for array or struct copies, it's only for
1721 * scalars or vectors into vectors or matrices.
1723 * With identical types on both sides and zero offset it's clone()
1724 * without creating a new object.
1727 void copy_offset(ir_constant *src, int offset);
1730 * Copy the values on another constant at a given offset and
1731 * following an assign-like mask.
1733 * The mask is ignored for scalars.
1735 * Note that this function only handles what assign can handle,
1736 * i.e. at most a vector as source and a column of a matrix as
1740 void copy_masked_offset(ir_constant *src, int offset, unsigned int mask);
1743 * Determine whether a constant has the same value as another constant
1745 * \sa ir_constant::is_zero, ir_constant::is_one,
1746 * ir_constant::is_negative_one
1748 bool has_value(const ir_constant *) const;
1750 virtual bool is_zero() const;
1751 virtual bool is_one() const;
1752 virtual bool is_negative_one() const;
1755 * Value of the constant.
1757 * The field used to back the values supplied by the constant is determined
1758 * by the type associated with the \c ir_instruction. Constants may be
1759 * scalars, vectors, or matrices.
1761 union ir_constant_data value;
1763 /* Array elements */
1764 ir_constant **array_elements;
1766 /* Structure fields */
1767 exec_list components;
1771 * Parameterless constructor only used by the clone method
1779 * Apply a visitor to each IR node in a list
1782 visit_exec_list(exec_list *list, ir_visitor *visitor);
1785 * Validate invariants on each IR node in a list
1787 void validate_ir_tree(exec_list *instructions);
1789 struct _mesa_glsl_parse_state;
1790 struct gl_shader_program;
1793 * Detect whether an unlinked shader contains static recursion
1795 * If the list of instructions is determined to contain static recursion,
1796 * \c _mesa_glsl_error will be called to emit error messages for each function
1797 * that is in the recursion cycle.
1800 detect_recursion_unlinked(struct _mesa_glsl_parse_state *state,
1801 exec_list *instructions);
1804 * Detect whether a linked shader contains static recursion
1806 * If the list of instructions is determined to contain static recursion,
1807 * \c link_error_printf will be called to emit error messages for each function
1808 * that is in the recursion cycle. In addition,
1809 * \c gl_shader_program::LinkStatus will be set to false.
1812 detect_recursion_linked(struct gl_shader_program *prog,
1813 exec_list *instructions);
1816 * Make a clone of each IR instruction in a list
1818 * \param in List of IR instructions that are to be cloned
1819 * \param out List to hold the cloned instructions
1822 clone_ir_list(void *mem_ctx, exec_list *out, const exec_list *in);
1825 _mesa_glsl_initialize_variables(exec_list *instructions,
1826 struct _mesa_glsl_parse_state *state);
1829 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state *state);
1832 _mesa_glsl_release_functions(void);
1835 reparent_ir(exec_list *list, void *mem_ctx);
1837 struct glsl_symbol_table;
1840 import_prototypes(const exec_list *source, exec_list *dest,
1841 struct glsl_symbol_table *symbols, void *mem_ctx);
1844 ir_has_call(ir_instruction *ir);
1847 do_set_program_inouts(exec_list *instructions, struct gl_program *prog,
1848 bool is_fragment_shader);
1851 prototype_string(const glsl_type *return_type, const char *name,
1852 exec_list *parameters);