Previously: When compiling for host, our constructed call graph went
*through* kernel calls. This meant that if we had
host calls kernel calls HD
we would incorrectly mark the HD function as known-emitted on the host
side, and thus perform host-side checks on it.
Fixing this exposed another issue, wherein when marking a function as
known-emitted, we also need to traverse the callgraph of its template,
because non-dependent calls are attached to a function's template, not
its instantiation.
llvm-svn: 284355
Although the itanium environment uses the itanium layout for C++, treat the
dllexport semantics more similarly to the MSVC specification. This preserves
the existing behaviour for the use of the itanium ABI on non-windows-itanium
environments. Export the inline definitions too.
llvm-svn: 284288
1) Merge and demote variable definitions when we find a redefinition in
MergeVarDecls, not only when we find one in AddInitializerToDecl (we only reach
the second case if it's the addition of the initializer itself that converts an
existing declaration into a definition).
2) When rebuilding a redeclaration chain for a variable, if we merge two
definitions together, mark the definitions as merged so the retained definition
is made visible whenever the demoted definition would have been.
Original commit message (from r283882):
[modules] PR28752: Do not instantiate variable declarations which are not visible.
Original patch by Vassil Vassilev! Changes listed above are mine.
llvm-svn: 284284
This commit combines a couple of redundant functions that do availability
attribute context checking into a more correct/simpler one.
Differential revision: https://reviews.llvm.org/D25283
llvm-svn: 284265
There was a bug in the implementation of captured statements. If it has
a lambda expression in it and the same lambda expression is used outside
the captured region, clang produced an error:
```
error: definition with same mangled name as another definition
```
Here is an example:
```
struct A {
template <typename L>
void g(const L&) { }
};
template<typename T>
void f() {
{
A().g([](){});
}
A().g([](){});
}
int main() {
f<void>();
}
```
Error report:
```
main.cpp:3:10: error: definition with same mangled name as another
definition
void g(const L&) { }
^
main.cpp:3:10: note: previous definition is here
```
Patch fixes this bug.
llvm-svn: 284229
Summary:
Emitting deferred diagnostics during codegen was a hack. It did work,
but usability was poor, both for us as compiler devs and for users. We
don't codegen if there are any sema errors, so for users this meant that
they wouldn't see deferred errors if there were any non-deferred errors.
For devs, this meant that we had to carefully split up our tests so that
when we tested deferred errors, we didn't emit any non-deferred errors.
This change moves checking for deferred errors into Sema. See the big
comment in SemaCUDA.cpp for an overview of the idea.
This checking adds overhead to compilation, because we have to maintain
a partial call graph. As a result, this change makes deferred errors a
CUDA-only concept (whereas before they were a general concept). If
anyone else wants to use this framework for something other than CUDA,
we can generalize at that time.
This patch makes the minimal set of test changes -- after this lands,
I'll go back through and do a cleanup of the tests that we no longer
have to split up.
Reviewers: rnk
Subscribers: cfe-commits, rsmith, tra
Differential Revision: https://reviews.llvm.org/D25541
llvm-svn: 284158
Summary:
Together these let you easily create diagnostics that
- are never emitted for host code
- are always emitted for __device__ and __global__ functions, and
- are emitted for __host__ __device__ functions iff these functions are
codegen'ed.
At the moment there are only three diagnostics that need this treatment,
but I have more to add, and it's not sustainable to write code for emitting
every such diagnostic twice, and from a special wrapper in SemaCUDA.cpp.
While we're at it, don't emit the function name in
err_cuda_device_exceptions: It's not necessary to print it, and making
this work in the new framework in the face of a null value for
dyn_cast<FunctionDecl>(CurContext) isn't worth the effort.
Reviewers: rnk
Subscribers: cfe-commits, tra
Differential Revision: https://reviews.llvm.org/D25139
llvm-svn: 284143
In r276159, we started to defer emitting initializers for VarDecls, but
forgot to add the initializers for non-C++ language.
rdar://28740482
llvm-svn: 284142
Original message:
"[modules] PR28752: Do not instantiate variable declarations which are not visible.
https://reviews.llvm.org/D24508
Patch developed in collaboration with Richard Smith!"
llvm-svn: 284008
This commit improves code completion for properties that are declared in
Objective-C protocols by making sure that properties show up in completions
when they are accessed through a qualified id.
rdar://24426041
Differential Revision: https://reviews.llvm.org/D25436
llvm-svn: 284007
Previously, this was an immediate, don't pass go, don't collect $200
error. But this precludes us from writing code like
__host__ __device__ void launch_kernel() {
kernel<<<...>>>();
}
Such code isn't wrong, following our notions of right and wrong in CUDA,
unless it's codegen'ed.
llvm-svn: 283963
swift_error/swift_context parameter
We need to be able to decelare witness functions which append the self type and
the self witness tables at the end of the parameter list.
rdar://28720996
llvm-svn: 283933
This reverts commit r283802. It introduces temporarily static
initializers, because StringRef ctor isn't (yet) constexpr for
string literals.
I plan to get there this week, but apparently GCC is so terrible
with these static initializer right now (10 min+ extra codegen
time was reported) that I'll hold on to this patch till the
constexpr one is ready, and land these at the same time.
llvm-svn: 283920
Currently Clang allows partial initializer for C99 but not for OpenCL, e.g.
float a[16][16] = {1.0f, 2.0f};
is allowed in C99 but not allowed in OpenCL.
This patch fixes that.
Differential Revision: https://reviews.llvm.org/D25335
llvm-svn: 283891
match other CUDA preference orders, per discussion with jlebar. We now model
this in an attempt to match overload resolution as closely as possible:
- First, we throw out all non-callable (due to CUDA host/device mismatch)
operator delete functions.
- Then we apply sizedness / alignedness preferences based on whether the type
is overaligned and whether the deallocation function is a member.
- Finally, we use the CUDA callability preference as a tiebreaker.
llvm-svn: 283830
Summary:
This is possible now that MapVector supports move-only values.
Depends on D25404.
Reviewers: timshen
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D25405
llvm-svn: 283766
Summary:
Once a base class has been made invalid (by a static_assert for example) all using-member declarations in the derived classes will result in a "not a base class" diagnostic. This diagnostic is very misleading and should not be emitted.
This change is needed to help libc++ produce reasonable diagnostics in `std::optional` and `std::variant`.
Reviewers: rsmith, majnemer, aaron.ballman
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D25430
llvm-svn: 283755
Summary:
These cause us to consider all functions in-between to be __host__
__device__.
You can nest these pragmas; you just can't have more 'end's than
'begin's.
Reviewers: rsmith
Subscribers: tra, jhen, cfe-commits
Differential Revision: https://reviews.llvm.org/D24975
llvm-svn: 283677
Summary:
Move CheckCUDACall from ActOnCallExpr and BuildDeclRefExpr to
DiagnoseUseOfDecl. This lets us catch some edge cases we were missing,
specifically around class operators.
This necessitates a few other changes:
- Avoid emitting duplicate deferred diags in CheckCUDACall.
Previously we'd carefully placed our call to CheckCUDACall such that
it would only ever run once for a particular callsite. But now this
isn't the case.
- Emit deferred diagnostics from a template
specialization/instantiation's primary template, in addition to from
the specialization/instantiation itself. DiagnoseUseOfDecl ends up
putting the deferred diagnostics on the template, rather than the
specialization, so we need to check both.
Reviewers: rsmith
Subscribers: cfe-commits, tra
Differential Revision: https://reviews.llvm.org/D24573
llvm-svn: 283637
CheckSingleAssignmentConstraints. These no longer produce ExprError() when they
have not emitted an error, and reliably inform the caller when they *have*
emitted an error.
This fixes some serious issues where we would fail to emit any diagnostic for
invalid code and then attempt to emit code for an invalid AST, and conversely
some issues where we would emit two diagnostics for the same problem.
llvm-svn: 283508
Summary:
Previously the statement `co_return {42}` would be transformed into `P.return_void()`, since the type of `{42}` is represented as `void` by Clang.
This patch fixes the bug by checking for `InitListExpr` arguments and transforming them accordingly.
Reviewers: majnemer, GorNishanov, rsmith
Subscribers: mehdi_amini, cfe-commits
Differential Revision: https://reviews.llvm.org/D25296
llvm-svn: 283495
Looks like the smart quote was copy/pasted from the C++ standard.
The smart quote was not encoded as valid UTF-8 (?), even though vim was
detecting the file as UTF-8. This broke the clang-format Python script,
which tried to read the file using the same encoding as vim detected.
llvm-svn: 283487
This commit fixes a crash that happens when clang is analyzing a
transparent_union attribute on a union which has a field with incomplete type.
rdar://28630028
Differential Revision: https://reviews.llvm.org/D25273
llvm-svn: 283432
new expression, distinguish between the case of a constant and non-constant
initializer. In the former case, if the bound is erroneous (too many
initializer elements, bound is negative, or allocated size overflows), reject,
and take the bound into account when determining whether we need to
default-construct any elements. In the remanining cases, move the logic to
check for default-constructibility of trailing elements into the initialization
code rather than inventing a bogus array bound, to cope with cases where the
number of initialized elements is not the same as the number of initializer
list elements (this can happen due to string literal initialization or brace
elision).
This also fixes rejects-valid and crash-on-valid errors when initializing a
new'd array of character type from a braced string literal.
llvm-svn: 283406
This commit improves the packed member warning by showing the name of the
anonymous structure/union when it was defined within a typedef declaration.
rdar://28498901
Differential Revision: https://reviews.llvm.org/D25106
llvm-svn: 283304
Treating large 0x*LL literals as signed instead of unsigned is not a
conforming language extension, so move it out of -fms-extensions.
Came up in PR30605
llvm-svn: 283227
Previously if a file-level function was defined inside befriending
template class, it always was treated as defined. For instance, the code like:
```
int func(int x);
template<typename T> class C1 {
friend int func(int x) { return x; }
};
template<typename T> class C2 {
friend int func(int x) { return x; }
};
```
could not be compiled due to function redefinition, although not of the templates
is instantiated. Moreover, the body of friend function can contain use of template
parameters, attempt to get definition of such function outside any instantiation
causes compiler abnormal termination.
Other compilers (gcc, icc) follow viewpoint that the body of the function defined
in friend declaration becomes available when corresponding class is instantiated.
This patch implements this viewpoint in clang.
Definitions introduced by friend declarations in template classes are not added
to the redeclaration chain of corresponding function. Only when the template is
instantiated, instantiation of the function definition is placed to the chain.
The fix was made in collaboration with Richard Smith.
This change fixes PR8035, PR17923, PR22307 and PR25848.
Differential Revision: http://reviews.llvm.org/D16989
llvm-svn: 283207
Summary:
We'd attempted to allow this, but turns out we were doing a very bad
job. :)
Making this work properly would be a giant change in clang. For
example, we'd need to make CXXRecordDecl::getDestructor()
context-sensitive, because the destructor you end up with depends on
where you're calling it from.
For now (and hopefully for ever), just disallow overloading of
destructors in CUDA.
Reviewers: rsmith
Subscribers: cfe-commits, tra
Differential Revision: https://reviews.llvm.org/D24571
llvm-svn: 283120
Summary:
Also makes -fcoroutines_ts to be both a Driver and CC1 flag.
Patch mostly by EricWF.
Reviewers: rnk, cfe-commits, rsmith, EricWF
Subscribers: mehdi_amini
Differential Revision: https://reviews.llvm.org/D25130
llvm-svn: 283064
Also add a test that we disallow
__constant__ __shared__ int x;
because it's possible to break this without breaking
__shared__ __constant__ int x;
Reviewers: rnk
Subscribers: cfe-commits, tra
Differential Revision: https://reviews.llvm.org/D25125
llvm-svn: 282985
assume that ::operator new provides no more alignment than is necessary for any
primitive type, except when we're on a GNU OS, where glibc's malloc guarantees
to provide 64-bit alignment on 32-bit systems and 128-bit alignment on 64-bit
systems. This can be controlled by the command-line -fnew-alignment flag.
llvm-svn: 282974
Summary: The title says it all.
Reviewers: rsmith, GorNishanov
Subscribers: mehdi_amini, cfe-commits
Differential Revision: https://reviews.llvm.org/D25078
llvm-svn: 282973
Support lax convertions on compound assignment expressions like:
typedef __attribute__((vector_size(8))) double float64x1_t;
typedef __attribute__((vector_size(16))) double float64x2_t;
float64x1_t vget_low_f64(float64x2_t __p0);
double c = 3.0;
float64x2_t v = {0.0, 1.0};
c += vget_low_f64(v);
This restores one more valid behavior pre r266366, and is a incremental
follow up from work committed in r274646.
While here, make the check more strict, add FIXMEs, clean up variable
names to match what they can actually be and update testcases to reflect
that. We now reject:
typedef float float2 __attribute__ ((vector_size (8)));
double d;
f2 += d;
which doesn't fit as a direct bitcast anyway.
Differential Revision: https://reviews.llvm.org/D24472
rdar://problem/28033929
llvm-svn: 282968
I'd said that nvcc doesn't allow you to add __host__ or __device__
attributes on lambdas in all circumstances, but I believe this was user
error on my part. I can't reproduce these warnings/errors if I pass
--expt-extended-lambda to nvcc.
llvm-svn: 282912
Summary:
This lets people link against LLVM and their own version of the UTF
library.
I determined this only affects llvm, clang, lld, and lldb by running
$ git grep -wl 'UTF[0-9]\+\|\bConvertUTF\bisLegalUTF\|getNumBytesFor' | cut -f 1 -d '/' | sort | uniq
clang
lld
lldb
llvm
Tested with
ninja lldb
ninja check-clang check-llvm check-lld
(ninja check-lldb doesn't complete for me with or without this patch.)
Reviewers: rnk
Subscribers: klimek, beanz, mgorny, llvm-commits
Differential Revision: https://reviews.llvm.org/D24996
llvm-svn: 282822
In some cases, non-special member functions were being marked as being defaulted
in templated classes. This can cause interactions with later code that expects
the default function to be one of the specific member functions. Fix the check
so that templated class members are checked the same way as non-templated class
members are.
llvm-svn: 282547
__attribute__((amdgpu_flat_work_group_size(<min>, <max>))) - request minimum and maximum flat work group size
__attribute__((amdgpu_waves_per_eu(<min>[, <max>]))) - request minimum and/or maximum waves per execution unit
Differential Revision: https://reviews.llvm.org/D24513
llvm-svn: 282371
The backend can't encode all possible values of the argument and will fail isel. Checking in the frontend presents a friendlier experience to the user.
I started with builtins that can only take _MM_CUR_DIRECTION or _MM_NO_EXC. More builtins coming in the future.
llvm-svn: 282228
Summary:
The diagnostic did not handle ~ well. An expression such as ~0 is often used when 'all ones' is needed.
Differential Revision: https://reviews.llvm.org/D24232
llvm-svn: 282156
Summary:
Offset was doubled in size, but the assignment was missing. We just need
to reassign to the original variable in this case to fix it.
Reviewers: cfe-commits, echristo
Subscribers: meikeb
Differential Revision: https://reviews.llvm.org/D24648
llvm-svn: 281706
Summary:
The warning for a format string not being a string literal and therefore
being potentially insecure is overly strict for indices into string
literals. This fix checks if the index into the string literal is
precomputable. If that's the case it will check if the suffix of that
string literal is a valid format string string literal. It will still
issue the aforementioned warning for out of range indices into the
string literal.
Patch by Meike Baumgärtner (meikeb)
Reviewers: rsmith
Subscribers: srhines, cfe-commits
Differential Revision: https://reviews.llvm.org/D24584
llvm-svn: 281686
r278501 inadvertently introduced a bug in which it disallowed shifting
scalar operands by vector operands when not compiling for OpenCL. This
commit fixes it.
Patch by Vladimir Yakovlev.
Differential Revision: https://reviews.llvm.org/D24467
llvm-svn: 281669
The underlying type for an enumeration in C is either char, signed int, or unsigned int. In the case the underlying type is chosen to be char (such as when passing -fshort-enums or using __attribute__((packed)) on the enum declaration), the enumeration can result in undefined behavior. However, when the underlying type is signed int or unsigned int (or long long as an extension), there is no undefined behavior because the types are compatible. This patch silences diagnostics for the latter while retaining the diagnostics for the former.
This patch addresses PR29140.
llvm-svn: 281632
Summary: This reverts r281527 because I messed up the attribution.
Reviewers: srhines
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D24579
llvm-svn: 281530
Summary:
The warning for a format string not being a sting literal and therefore
being potentially insecure is overly strict for indecies into sting
literals. This fix checks if the index into the string literal is
precomputable. If thats the case it will check if the suffix of that
sting literal is a valid format string string literal. It will still
issue the aforementioned warning for out of range indecies into the
string literal.
Reviewers: rsmith
Subscribers: srhines, cfe-commits
Differential Revision: https://reviews.llvm.org/D23820
llvm-svn: 281527
CUDA target attributes are used for function overloading and must not be merged.
This fixes a bug where attributes were inherited during function template
specialization in CUDA and made it impossible for specialized function
to provide its own target attributes.
Differential Revision: https://reviews.llvm.org/D24522
llvm-svn: 281406
This mostly behaves cl.exe's behavior, even though clang-cl is stricter in some
corner cases and more lenient in others (see the included test).
To make the uuid declared previously here diagnostic work correctly, tweak
stripTypeAttributesOffDeclSpec() to keep attributes in the right order.
https://reviews.llvm.org/D24469
llvm-svn: 281367
For ObjC type parameter, we used to have TypedefType that is canonicalized to
id or the bound type. We can't represent "T <protocol>" and thus will lose
the type information in the following example:
@interface MyMutableDictionary<KeyType, ObjectType> : NSObject
- (void)setObject:(ObjectType)obj forKeyedSubscript:(KeyType <NSCopying>)key;
@end
MyMutableDictionary<NSString *, NSString *> *stringsByString;
NSNumber *n1, *n2;
stringsByString[n1] = n2;
--> no warning on type mismatch of the key.
To fix the problem, we introduce a new type ObjCTypeParamType that supports
a list of protocol qualifiers.
We create ObjCTypeParamType for ObjCTypeParamDecl when we create
ObjCTypeParamDecl. We also substitute ObjCTypeParamType instead of TypedefType
on an ObjCTypeParamDecl.
rdar://24619481
rdar://25060179
Differential Revision: http://reviews.llvm.org/D23080
llvm-svn: 281358
We also need to add ObjCTypeParamTypeLoc. ObjCTypeParamType supports the
representation of "T <protocol>" where T is a type parameter. Before this,
we use TypedefType to represent the type parameter for ObjC.
ObjCTypeParamType has "ObjCTypeParamDecl *OTPDecl" and it extends from
ObjCProtocolQualifiers. It is a non-canonical type and is canonicalized
to the underlying type with the protocol qualifiers.
rdar://24619481
rdar://25060179
Differential Revision: http://reviews.llvm.org/D23079
llvm-svn: 281355
To construct the canonical type of ObjCTypeParamType, we need to apply
qualifiers on ObjCObjectPointerType. The updated applyObjCProtocolQualifiers
handles this case by merging the protocol lists, constructing a new
ObjCObjectType, then a new ObjCObjectPointerType.
rdar://24619481
rdar://25060179
Differential Revision: http://reviews.llvm.org/D24059
llvm-svn: 281353
Vassil Vassilev