This patch fixes a bug where we would assume all value-dependent
enable_if conditions give successful results.
Instead, we consider value-dependent enable_if conditions to always
fail. While this isn't ideal, this is the best we can realistically do
without changing both enable_if's semantics and large parts of Sema
(specifically, all of the parts that don't expect type dependence to
come out of nowhere, and that may interact with overload resolution).
Reid Kleckner [Tue, 10 May 2016 21:00:03 +0000 (21:00 +0000)]
Add -Wcast-calling-convention to warn when casting away calling conventions
Summary:
This only warns on casts of the address of a function defined in the
current TU. In this case, the fix is likely to be local and the warning
useful.
Here are some things we could experiment with in the future:
- Fire on declarations as well as definitions
- Limit the warning to non-void function prototypes
- Limit the warning to mismatches of caller and callee cleanup CCs
This warning is currently off by default while we study its usefulness.
[VFS] Reconstruct the VFS overlay tree for more accurate lookup
The way we currently build the internal VFS overlay representation leads
to inefficient path search and might yield wrong answers when asked for
recursive or regular directory iteration.
Currently, when reading an YAML file, each YAML root entry is placed
inside a new root in the filesystem overlay. In the crash reproducer, a
simple "@import Foundation" currently maps to 43 roots, and when looking
up paths, we traverse a directory tree for each of these different
roots, until we find a match (or don't). This has two consequences:
- It's slow.
- Directory iteration gives incomplete results since it only return
results within one root - since contents of the same directory can be
declared inside different roots, the result isn't accurate.
This is in part fault of the way we currently write out the YAML file
when emitting the crash reproducer - we could generate only one root and
that would make it fast and correct again. However, we should not rely
on how the client writes the YAML, but provide a good internal
representation regardless.
This patch builds a proper virtual directory tree out of the YAML
representation, allowing faster search and proper iteration. Besides the
crash reproducer, this potentially benefits other VFS clients.
Derek Schuff [Tue, 10 May 2016 17:44:46 +0000 (17:44 +0000)]
Do not register incompatible C++ destructors with __cxa_atexit
Summary:
For a static object with a nontrivial destructor, clang generates an
initializer function (__cxx_global_var_init) which registers that
object's destructor using __cxa_atexit. However some ABIs (ARM,
WebAssembly) use destructors that return 'this' instead of having void
return (which does not match the signature of function pointers passed
to __cxa_atexit). This results in undefined behavior when the destructors are
called. All the calling conventions I know of on ARM can tolerate this,
but WebAssembly requires the signatures of indirect calls to match the
called function.
This patch disables that direct registration of destructors for ABIs
that have this-returning destructors.
Alexey Bataev [Tue, 10 May 2016 09:57:36 +0000 (09:57 +0000)]
[OPENMP 4.5] Add codegen support in runtime for '[non]monotonic'
schedule modifiers.
Runtime library expects some additional data in schedule argument for
loop-based directives, that have additional schedule modifiers
'monotonic|nonmonotonic'.
[Sema] Fix an overload resolution bug with enable_if.
Currently, if clang::isBetterOverloadCandidate encounters an enable_if
attribute on either candidate that it's inspecting, it will ignore all
lower priority attributes (e.g. pass_object_size). This is problematic
in cases like:
Artem Belevich [Mon, 9 May 2016 22:09:56 +0000 (22:09 +0000)]
[CUDA] Restrict init of local __shared__ variables to empty constructors only.
Allow only empty constructors for local __shared__ variables in a way
identical to restrictions imposed on dynamic initializers for global
variables on device.
Artem Belevich [Mon, 9 May 2016 19:36:08 +0000 (19:36 +0000)]
[CUDA] Only __shared__ variables can be static local on device side.
According to CUDA programming guide (v7.5):
> E.2.9.4: Within the body of a device or global function, only
> shared variables may be declared with static storage class.
Proper detection and handling of RHEL and variants.
- Don't consider "/etc/lsb-release" to be Ubuntu only.
- Detect SL, too.
- Only add "--no-add-needed" for RHEL7 (or Fedora), not for RHEL6
(that's what the compilers shipped with RHEL do).
- removed RHEL4 which is now four years past EOL and certainly incapable
of building or running any recent version of llvm/clang.
RHEL: Look in more places to find g++ headers and runtime.
Some distros with ten years of support ship an old gcc but later offer
more recent versions for installation in parallel. These versions are
typically not only needed for the compilation of llvm/clang, but also to
properly use the clang binary that comes out.
Clang already searches /usr at runtime for the most recent installation
of gcc. This patch appends paths for add-on installations of gcc in
RHEL.
Enable support for __float128 in Clang and enable it on pertinent platforms
This patch corresponds to reviews:
http://reviews.llvm.org/D15120
http://reviews.llvm.org/D19125
It adds support for the __float128 keyword, literals and target feature to
enable it. Based on the latter of the two aforementioned reviews, this feature
is enabled on Linux on i386/X86 as well as SystemZ.
This is also the second attempt in commiting this feature. The first attempt
did not enable it on required platforms which caused failures when compiling
type_traits with -std=gnu++11.
If you see failures with compiling this header on your platform after this
commit, it is likely that your platform needs to have this feature enabled.
Samuel Antao [Sun, 8 May 2016 06:43:56 +0000 (06:43 +0000)]
[OpenMP] Check for associated statements with hasAssociatedStmt() when scanning for device code.
Summary: `getAssociatedStmt()` contains an assertion that assumes the statement always exists. In device code scanning, we need to look into the associated statement therefore we check its existence. This patch replaces `getAssociatedStmt` by `hasAssociatedStmt` so that we do not trigger the assertion for directives that happen not to have an associated statement (e.g target enter/exit data).
[CrashReproducer] Always use realpath for destination
When running reproducer scripts we need that original symlinks from the
source filesystem are reproduced in the VFS so that different virtual
paths can map to the same file, allowing the FileManager to share the
same UID between these virtual entries. This avoids all sorts of module
redefinition errors when using frameworks.
Use a StringRef instead of a FileEntry in the moduleMapAddHeader
callback to allow more flexibility on what to collect on further
patches. This changes the interface I introduced in r264971.
Richard Smith [Fri, 6 May 2016 23:14:07 +0000 (23:14 +0000)]
[modules] Attempt to improve performance for declaration merging without a Sema
object in C. Rather than using the DeclContext (which is very slow because it
triggers us to build a lookup table for the DeclContext), use a separate map
from identifiers to decls for this case, containing only the ones we've
actually deserialized. For convenience, this map is implemented as an
IdentifierResolver.
Manman Ren [Fri, 6 May 2016 19:57:16 +0000 (19:57 +0000)]
Availability: attach the note to the declaration with the attributes.
Sometimes, the declaration we found has inherited availability
attributes, attaching the note to it does not tell us where the
availability attributes are in the source.
Go through the redecl chain to find the declaration with actual
availability attributes.
Replace use of /dev/null with /var/empty. lit will substitute the /dev/null
include path resulting in failures. Use a path under /var/empty which is
supposed to be empty to ensure that we can successfully test.
-I dir
... If dir begins with =, then the = will be replaced by the sysroot prefix;
see --sysroot and -isysroot.
Add support to expand the `=` as a prefix of the include path with the sysroot
if specified. `-isysroot` takes precedence over `--sysroot` as the normal
argument behaviour occurs. The ordering of the `-isysroot` is relevant to the
path substituted. If no `--sysroot=` or `-isysroot` option is present, the = is
not expanded.
Ben Craig [Fri, 6 May 2016 13:29:46 +0000 (13:29 +0000)]
Adding omitted column to invalid loc diagnostic.
note_fe_backend_invalid_loc expects three arguments (file, line, column),
and will assert when only given two. The other two places in this file that
use note_fe_backend_invalid_loc already supply the Column for the third
parameter.
Alexey Bataev [Fri, 6 May 2016 09:40:08 +0000 (09:40 +0000)]
[OPENMP 4.0] Codegen for 'declare simd' directive.
OpenMP 4.0 adds support for elemental functions using declarative
directive '#pragma omp declare simd'. Patch adds mangling for simd
functions in accordance with
https://sourceware.org/glibc/wiki/libmvec?action=AttachFile&do=view&target=VectorABI.txt
Akira Hatanaka [Thu, 5 May 2016 01:41:07 +0000 (01:41 +0000)]
Do not add uwtable attribute by default for MachO targets.
r217178 changed clang to add function attribute uwtable by default on
Win64, which caused the __eh_frame section to be emitted by default.
This commit restores the previous behavior for MachO targets.
Richard Smith [Thu, 5 May 2016 00:56:12 +0000 (00:56 +0000)]
[modules] Enforce the rules that an explicit or partial specialization must be
declared before it is used. Because we don't use normal name lookup to find
these, the normal code to filter out non-visible names from name lookup results
does not apply.
Fix the doc extraction script to work with hasAnyName and with equalsNode.
The change from llvm::VariadicFunction to internal::VariadicFunction
broke the extraction of hasAnyName().
equalsNode was broken because the argument type is 'const XXXX*' and
the internal space caused a failure on the regex.
Akira Hatanaka [Wed, 4 May 2016 18:07:20 +0000 (18:07 +0000)]
[ObjC] Enter a new evaluation context before calling
BuildBlockForLambdaConversion.
Previously, clang would build an incorrect AST for the following code:
id test() {
return @{@"a": [](){}, @"b": [](){}};
}
ReturnStmt 0x10d080448
`-ExprWithCleanups 0x10d080428
|-cleanup Block 0x10d0801f0 // points to the second BlockDecl
...
-BlockDecl 0x10d07f150 // First block
...
-BlockDecl 0x10d0801f0 // Second block
...
`-ExprWithCleanups 0x10d0801d0
|-cleanup Block 0x10d07f150 // points to the first BlockDecl
To fix the bug, this commit enters a new evaluation context to reset
ExprNeedsCleanups before each block is parsed.
Pete Cooper [Wed, 4 May 2016 17:37:27 +0000 (17:37 +0000)]
Update test CHECK lines after r268509. NFC.
r268509 causes this test case to be fully unrolled, so checking for an icmp is
no longer valid. Given that this test is for ARC anyway, checking for the icmp seems
unecessary.
Do not disable completely loop unroll when optimizing for size.
Let the loop unroll pass handle /Os. It already checks that option and adjust its thresholds accordingly. Also, will allow the #pragma unroll to have an effect in /Os.
Gabor Horvath [Wed, 4 May 2016 11:59:39 +0000 (11:59 +0000)]
[ASTMatchers] New matcher forFunction
Summary: Matcher proposed in the review of checker misc-assign-operator (name pending). Its goal is to find the direct enclosing function declaration of a statement and run the inner matcher on it. Two version is attached in this patch (thus it will not compile), to be decided which approach to take. The second one always chooses one single parent while the first one does a depth-first search upwards (thus a height-first search) and returns the first positive match of the inner matcher (thus it always returns zero or one matches, not more). Further questions: is it enough to implement it in-place, or ASTMatchersInternals or maybe ASTMatchFinder should be involved?
projects / clang / log