#include <util/symtab/context.h>
#include <util/irep/irep.h>
#include <string>

Include dependency graph for builtin_names.h:

This graph shows which files directly or indirectly include this file:

Enumerations
enum class	ieee_float_builtin { none , nearbyint , remainder , fma }
	The float library calls that have an IEEE node of their own. More...

Functions
bool	compare_float_suffix (const irep_idt &identifier, const std::string &name)
	Callee-name matching shared by clang_c_adjust::do_special_functions and its IREP2 counterpart.
bool	compare_unscore_builtin (const irep_idt &identifier, const std::string &name)
	name and its __- and __builtin_-prefixed spellings, each with the float suffixes: isnan, __isnanf, __builtin_isnanl, ...
bool	is_abs_builtin_name (const irep_idt &identifier)
	True for the abs builtins that may be lowered to an abs node.
ieee_float_builtin	ieee_float_builtin_of (const irep_idt &identifier)
bool	is_name_matched_builtin (const irep_idt &identifier)
	The lowerings that match a callee's base name, so a program that defines one of these names itself would have its body discarded and the builtin verified in its place (#6904).
bool	builtin_shadows_user_definition (const contextt &context, const irep_idt &base_name, const irep_idt &symbol_id)
	True when lowering this call would throw away a definition the program supplies.

Enumeration Type Documentation

◆ ieee_float_builtin

enum class ieee_float_builtin

strong

The float library calls that have an IEEE node of their own.

Both passes lower these, so the spelling set is shared; the arity each node takes is not, because legacy splices whatever arguments the call has into a fixed-arity node (see clang_c_adjust_irep2.cpp).

Enumerator
none
nearbyint
remainder
fma

Function Documentation

◆ builtin_shadows_user_definition()

bool builtin_shadows_user_definition	(	const contextt &	context,
		const irep_idt &	base_name,
		const irep_idt &	symbol_id )

True when lowering this call would throw away a definition the program supplies.

Libc's own declarations are bodiless, and the <cmath> overloads forward to a spelling that lowers in their place – the __builtin_ one for most, the f/l suffixes for fma (src/cpp/library/cmath) – so a std:: call still reaches a node, one frame further in.

Parameters

base_name	the spelling the lowerings match on
symbol_id	the callee's linkage identifier, which is what the symbol table is keyed by

◆ compare_float_suffix()

bool compare_float_suffix	(	const irep_idt &	identifier,
		const std::string &	name )

Callee-name matching shared by clang_c_adjust::do_special_functions and its IREP2 counterpart.

Both lower the same builtins, so both must agree on which spellings are one; a second copy of these predicates is exactly the shape of divergence unit/util/c_typecast.test.cpp exists to catch elsewhere. name, namef, named, namel.

◆ compare_unscore_builtin()

bool compare_unscore_builtin	(	const irep_idt &	identifier,
		const std::string &	name )

name and its __- and __builtin_-prefixed spellings, each with the float suffixes: isnan, __isnanf, __builtin_isnanl, ...

◆ ieee_float_builtin_of()

ieee_float_builtin ieee_float_builtin_of ( const irep_idt & identifier )

◆ is_abs_builtin_name()

bool is_abs_builtin_name ( const irep_idt & identifier )

True for the abs builtins that may be lowered to an abs node.

That node becomes (x >= 0) ? x : -x, ill-typed for anything but an arithmetic argument, so a program overloading the name for a class type – std::abs(complex) is why <complex> ships without it – keeps its call.

◆ is_name_matched_builtin()

bool is_name_matched_builtin ( const irep_idt & identifier )

The lowerings that match a callee's base name, so a program that defines one of these names itself would have its body discarded and the builtin verified in its place (#6904).

These are all spellings a program is free to reuse – mylib::abs, mylib::isinf – unlike the __builtin_-prefixed and CPROVER-prefixed entries, which are reserved.

Enumerations

Functions