function definition can produce a constant expression. This also provides the
last few checks for [dcl.constexpr]p3 and [dcl.constexpr]p4.
llvm-svn: 149108
for it to be used in converted constant expression checking, and fix a couple
of issues:
- Conversion operators implicitly invoked prior to the narrowing conversion
were not being correctly handled when determining whether a constant value
was narrowed.
- For conversions from floating-point to integral types, the diagnostic text
incorrectly always claimed that the source expression was not a constant
expression.
llvm-svn: 148381
- Add atomic-to/from-nonatomic cast types
- Emit atomic operations for arithmetic on atomic types
- Emit non-atomic stores for initialisation of atomic types, but atomic stores and loads for every other store / load
- Add a __atomic_init() intrinsic which does a non-atomic store to an _Atomic() type. This is needed for the corresponding C11 stdatomic.h function.
- Enables the relevant __has_feature() checks. The feature isn't 100% complete yet, but it's done enough that we want people testing it.
Still to do:
- Make the arithmetic operations on atomic types (e.g. Atomic(int) foo = 1; foo++;) use the correct LLVM intrinsic if one exists, not a loop with a cmpxchg.
- Add a signal fence builtin
- Properly set the fenv state in atomic operations on floating point values
- Correctly handle things like _Atomic(_Complex double) which are too large for an atomic cmpxchg on some platforms (this requires working out what 'correctly' means in this context)
- Fix the many remaining corner cases
llvm-svn: 148242
APValue::Array and APValue::MemberPointer. All APValue values can now be emitted
as constants.
Add new CGCXXABI entry point for emitting an APValue MemberPointer. The other
entrypoints dealing with constant member pointers are no longer necessary and
will be removed in a later change.
Switch codegen from using EvaluateAsRValue/EvaluateAsLValue to
VarDecl::evaluateValue. This performs caching and deals with the nasty cases in
C++11 where a non-const object's initializer can refer indirectly to
previously-initialized fields within the same object.
Building the intermediate APValue object incurs a measurable performance hit on
pathological testcases with huge initializer lists, so we continue to build IR
directly from the Expr nodes for array and record types outside of C++11.
llvm-svn: 148178
zero-initialize the first union member. Also fix a bug where initializing an
array of types compatible with wchar_t from a wide string literal failed in C,
and fortify the C++ tests in this area. This part can't be tested without a code
change to enable array evaluation in C (where an existing test fails).
llvm-svn: 148035
pointer-arithmetic-related undefined behavior and unspecified results. We
continue to fold such values, but now notice they aren't constant expressions.
llvm-svn: 147659
With that done, remove a bunch of buggy code from CGExprConstant for handling scalar expressions which is no longer necessary.
Fixes PR11705.
llvm-svn: 147561
Also temporarily remove the assumption from IR gen that we can emit IR for every
constant we can fold, since it isn't currently true in C++11, to fix PR11676.
Original comment from r147271:
constexpr: perform zero-initialization prior to / instead of performing a
constructor call when appropriate. Thanks to Eli for spotting this.
llvm-svn: 147384
variable is initialized by a non-constant expression, and pass in the variable
being declared so that earlier-initialized fields' values can be used.
Rearrange VarDecl init evaluation to make this possible, and in so doing fix a
long-standing issue in our C++ constant expression handling, where we would
mishandle cases like:
extern const int a;
const int n = a;
const int a = 5;
int arr[n];
Here, n is not initialized by a constant expression, so can't be used in an ICE,
even though the initialization expression would be an ICE if it appeared later
in the TU. This requires computing whether the initializer is an ICE eagerly,
and saving that information in PCH files.
llvm-svn: 146856
floating literal value does not fit into the destination type. Such casts have
undefined behavior at translation time; treating them as non-ICE matches the
behavior of modern gcc versions.
llvm-svn: 146842
fails within a call to a constexpr function. Add -fconstexpr-backtrace-limit
argument to driver and frontend, to control the maximum number of notes so
produced (default 10). Fix APValue printing to be able to pretty-print all
APValue types, and move the testing for this functionality from a unittest to
a -verify test now that it's visible in clang's output.
llvm-svn: 146749
whether an expression is a (core) constant expression as a side-effect of
evaluation. This takes us from accepting far too few expressions as ICEs to
accepting slightly too many -- fixes for the remaining cases are coming next.
The diagnostics produced when an expression is found to be non-constant are
currently quite poor (with generic wording but reasonable source locations),
and will be improved in subsequent commits.
llvm-svn: 146289
documentation) with one based on what GCC's __builtin_constant_p is actually
intended to do (discovered by asking a friendly GCC developer).
In particular, an expression which folds to a pointer is now only considered to
be a "constant" by this builtin if it refers to the first character in a string
literal.
This fixes a rather subtle wrong-code issue when building with glibc. Given:
const char cs[4] = "abcd";
int f(const char *p) { return strncmp(p, cs, 4); }
... the macro magic for strncmp produces a (potentially crashing) call to
strlen(cs), because it expands to an expression starting with:
__builtin_constant_p(cs) && strlen(cs) < 4 ? /* ... */
Under the secret true meaning of __builtin_constant_p, this is guaranteed to be
safe!
llvm-svn: 146236