Richard Smith [Mon, 5 Oct 2015 20:05:21 +0000 (20:05 +0000)]
Revert r107690 (for PR7417) and add a testcase that it breaks. The approach of
that change turns out to not be reasonable: mutating the AST of a parsed
template during instantiation is not a sound thing to do, does not work across
chained PCH / modules builds, and is in any case a special-case workaround to a
more general problem that should be solved centrally.
Adrian Prantl [Mon, 5 Oct 2015 17:41:16 +0000 (17:41 +0000)]
Undo the unique_ptr'fication of CodeGenABITypes::CGM introduced in r248762.
include/clang/CodeGenABITypes.h is in meant to be included by external
users, but using a unique_ptr on the private CodeGenModule introduces a
dependency on the type definition that prevents such a use.
Benjamin Kramer [Mon, 5 Oct 2015 13:55:20 +0000 (13:55 +0000)]
[VFS] Add working directories to every virtual file system.
For RealFileSystem this is getcwd()/chdir(), the synthetic file systems can
make up one for themselves. OverlayFileSystem now synchronizes the working
directories when a new FS is added to the overlay or the overlay working
directory is set. This allows purely artificial file systems that have zero
ties to the underlying disks.
Benjamin Kramer [Mon, 5 Oct 2015 13:55:14 +0000 (13:55 +0000)]
[VFS] Add an in-memory file system implementation.
This is a simple file system tree of memory buffers that can be filled by a
client. In conjunction with an OverlayFS it can be used to make virtual
files accessible right next to physical files. This can be used as a
replacement for the virtual file handling in FileManager and which I intend
to remove eventually.
Re-commit "Add support for the new mips-mti-linux toolchain."
r249137 added support for the new mips-mti-linux toolchain. However,
the new tests of that commit, broke some buildbots because they didn't use
the correct regular expressions to capture the filename of Clang & LLD.
This commit re-applies the changes of r249137 and fixes the tests in
r249137 in order to match the filenames of the Clang and LLD executable.
Daniel Jasper [Mon, 5 Oct 2015 07:36:40 +0000 (07:36 +0000)]
clang-format: Remove sentence in option cost threshold that weakens the
statement. Specifically, we don't want people that have already written
a patch to just write a style guide for their 1-person project.
Keno Fischer [Sun, 4 Oct 2015 18:51:04 +0000 (18:51 +0000)]
Fix makefile build on OSX when ARM targets are not enabled
Summary: When LLVM/Clang is built without ARM support, the ios_kext runtime
library is not built, but without this patch, the Makefile still tries to
copy it. This is a recent regression, because the ios_kext library used
to also be built on x86_64.
Add -f[no-]declspec to control recognition of __declspec as a keyword
In versions of clang prior to r238238, __declspec was recognized as a keyword in
all modes. It was then changed to only be enabled when Microsoft or Borland
extensions were enabled (and for CUDA, as a temporary measure). There is a
desire to support __declspec in Playstation code, and possibly other
environments. This commit adds a command-line switch to allow explicit
enabling/disabling of the recognition of __declspec as a keyword. Recognition
is enabled by default in Microsoft, Borland, CUDA, and PS4 environments, and
disabled in all other environments.
Eric Fiselier [Sun, 4 Oct 2015 00:11:02 +0000 (00:11 +0000)]
Diagnose const atomics in __atomic builtins.
Diagnose when a pointer to const T is used as the first argument in at atomic
builtin unless that builtin is a load operation. This is already checked for
C11 atomics builtins but not for __atomic ones.
This patch was given the LGTM by rsmith when it was part
of a larger review. (See http://reviews.llvm.org/D10407)
The default target is ARM on the ARM self host bots. This is problematic since
the behaviour on x86, x64 is different from ARM. Explicitly pass the target.
This should hopefully fix the ARM bots.
The Windows on ARM ABI recommends that FPO be disabled. This is since the
Windows on ARM ABI uses the FP for fast stack walking. By paying the slight
cost of the loss of registers, a much faster backtrace is possible by using the
frame pointer since the pdata need not be consulted. Furthermore, even if pdata
is not available, you can still more easily reconstruct the stack.
Ahmed Bougacha [Fri, 2 Oct 2015 22:41:59 +0000 (22:41 +0000)]
Don't nil check non-nil class receiver of AArch64 stret calls.
I randomly came across this difference between AArch64 and other targets:
on the latter, we don't emit nil checks for known non-nil class method
calls thanks to r247350, but we still do for AArch64 stret calls.
They use different code paths, because those are special, as they go
through the regular msgSend, not the msgSend*_stret variants.
Piotr Padlewski [Fri, 2 Oct 2015 22:12:40 +0000 (22:12 +0000)]
Emiting invariant.group.barrier for ctors bugfix
Ensure that the vptr store in the most-derived constructor is not behind
an invariant group barrier. Previously, the base-most vptr store would
be the one behind no barrier, and that could result in the creator of
the object thinking it had the base-most vtable.
This bug caused clang call pure virtual functions when called from
constructor body.
Samuel Antao [Fri, 2 Oct 2015 17:14:03 +0000 (17:14 +0000)]
[OpenMP] Capture global variables in target regions.
All global variables that are not enclosed in a declare target region
must be captured in the target region as local variables do. Currently,
there is no support for declare target, so this patch adds support for
capturing all the global variables used in a the target region.
Samuel Antao [Fri, 2 Oct 2015 16:14:20 +0000 (16:14 +0000)]
[OpenMP] Target directive host codegen.
This patch implements the outlining for offloading functions for code
annotated with the OpenMP target directive. It uses a temporary naming
of the outlined functions that will have to be updated later on once
target side codegen and registration of offloading libraries is
implemented - the naming needs to be made unique in the produced
library.
[DarwinDriver] Reapply: Use -lto_library to specify the path for libLTO.dylib
Reapply r248935.
Usually, when using LTO with a clang installation newer than the
system's one, there's a libLTO.dylib version mismatch and LTO fails. One
solution to this is to make ld point to the right libLTO.dylib by
changing DYLD_LIBRARY_PATH.
However, ld64 supports specifying the complete path to the desired
libLTO.dylib through the -lto_library option. This commit adds support
for the clang driver to use this option whenever it's capable of finding
a libLTO.dylib in clang's installed library directory. This way, we
don't need to rely on DYLD_LIBRARY_PATH nor get caught by version
mismatches.
Make test more resilient to FastIsel changes. NFC.
Currently FastISel doesn't know how to select vector bitcasts.
During instruction selection, fast-isel always falls back to SelectionDAG
every time it encounters a vector bitcast.
As a consequence of this, all the 'packed vector shift by immedate count'
test cases in avx2-builtins.c are optimized by the DAGCombiner.
In particular, the DAGCombiner would always fold trivial stack loads of
constant shift counts into the operands of packed shift builtins.
This behavior would start changing as soon as I reapply revision 249121.
That revision would teach x86 fast-isel how to select bitcasts between vector
types of the same size.
As a consequence of that change, fast-isel would less often fall back to
SelectionDAG. More importantly, DAGCombiner would no longer be able to
simplify the code by folding the stack reload of a constant.
Summary:
This new toolchain uses primarily LLVM-based tools, eg. compiler-rt, lld,
libcxx, etc. Because of this, it doesn't require neither an existing GCC
installation nor a GNU environment. Ideally, in a follow-up patch we
would like to add a new --{llvm|clang}-toolchain option (similar to
--gcc-toolchain) in order to allow the use of this toolchain with
independent Clang builds. For the time being, we use the --sysroot
option just to test the correctness of the paths generated by the
driver.
James Y Knight [Fri, 2 Oct 2015 13:41:04 +0000 (13:41 +0000)]
Move functions declared in Stmt{ObjC,CXX}.h and OpenMPClause.h into
their associated .cpp file.
Previous refactorings long long ago had split out the above categories
of classes from Stmt.h into their own header files, but failed to also
split the Stmt.cpp implementation file similarly. Do so for
readability's sake.
Richard Smith [Fri, 2 Oct 2015 00:49:37 +0000 (00:49 +0000)]
PR24921: checking explicitly-specified template arguments when matching a
partial specialization can perform conversions on the argument. Be sure we
start again from the original argument when checking each possible template.
Fix the SSE4 byte sign extension in a cleaner way, and more thoroughly
test that our intrinsics behave the same under -fsigned-char and
-funsigned-char.
This further testing uncovered that AVX-2 has a broken cmpgt for 8-bit
elements, and has for a long time. This is fixed in the same way as
SSE4 handles the case.
The other ISA extensions currently work correctly because they use
specific instruction intrinsics. As soon as they are rewritten in terms
of generic IR, they will need to add these special casts. I've added the
necessary testing to catch this however, so we shouldn't have to chase
it down again.
I considered changing the core typedef to be signed, but that seems like
a bad idea. Notably, it would be an ABI break if anyone is reaching into
the innards of the intrinsic headers and passing __v16qi on an API
boundary. I can't be completely confident that this wouldn't happen due
to a macro expanding in a lambda, etc., so it seems much better to leave
it alone. It also matches GCC's behavior exactly.
A fun side note is that for both GCC and Clang, -funsigned-char really
does change the semantics of __v16qi. To observe this, consider:
Kaelyn Takata [Thu, 1 Oct 2015 22:38:51 +0000 (22:38 +0000)]
Don't try to get a CXXRecordDecl from a non-class TemplateSpecializationType.
With -fms-extensions it is possible to have a non-class record that is a
template specialization cause an assertion failure via the call to
Type::getAsCXXRecordDecl. Fixes PR 24246.
Douglas Gregor [Thu, 1 Oct 2015 20:20:47 +0000 (20:20 +0000)]
Perform Objective-C lifetime adjustments before comparing deduced lambda result types.
Objective-C ARC lifetime qualifiers are dropped when canonicalizing
function types. Perform the same adjustment before comparing the
deduced result types of lambdas. Fixes rdar://problem/22344904.
This commit supports Sean Eveson's work on loop widening. It is NFC for now.
It adds a new TK_EntireMemSpace invalidation trait that, when applied to a
MemSpaceRegion, indicates that the entire memory space should be invalidated.
Clients can add this trait before invalidating. For example:
This commit updates the existing logic invalidating global memspace regions for
calls to additionally handle arbitrary memspaces. When generating initial
clusters during cluster analysis we now add a cluster to the worklist if
the memspace for its base is marked with TK_EntireMemSpace.
This also moves the logic for invalidating globals from ClusterAnalysis to
invalidateRegionsWorker so that it is not shared with removeDeadBindingsWorker.
There are no explicit tests with this patch -- but when applied to Sean's patch
for loop widening in http://reviews.llvm.org/D12358 and after updating his code
to set the trait, the failing tests in that patch now pass.
Prior to this patch, -Wtautological-overlap-compare would only warn us
if there was a sketchy logical comparison between variables and
IntegerLiterals. This patch makes -Wtautological-overlap-compare aware
of EnumConstantDecls, so it can apply the same logic to them.
Allow a ToolChain to compute the path of a compiler-rt's component.
Summary:
This patch moves getCompilerRT() from the clang::driver::tools namespace to
the ToolChain class. This is needed for multilib toolchains that need to
place their libraries in Clang's resource directory with a layout that is
different from the default one.
Hal Finkel [Thu, 1 Oct 2015 13:39:49 +0000 (13:39 +0000)]
[PowerPC] Define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_* macros on all PPC cores
We support all __sync_val_compare_and_swap_* builtins (only 64-bit on 64-bit
targets) on all cores, and should define the corresponding
__GCC_HAVE_SYNC_COMPARE_AND_SWAP_* macros, just as GCC does. As it turns out,
this is really important because they're needed to prevent a bad ODR violation
with libstdc++'s std::shared_ptr (this is well explained in PR12730).
We were doing this only for P8, but this is necessary on all PPC systems.
Daniel Jasper [Thu, 1 Oct 2015 10:06:54 +0000 (10:06 +0000)]
[clang-format] Add support of consecutive declarations alignment
This allows clang-format to align identifiers in consecutive
declarations. This is useful for increasing the readability of the code
in the same way the alignment of assignations is.
The code is a slightly modified version of the consecutive assignment
alignment code. Currently only the identifiers are aligned, and there is
no support of alignment of the pointer star or reference symbol.
The patch also solve the issue of alignments not being possible due to
the ColumnLimit for both the existing AlignConsecutiveAligments and the
new AlignConsecutiveDeclarations.
Patch over a really horrible bug in our vector builtins that showed up
recently when we started using direct conversion to model sign
extension. The __v16qi type we use for SSE v16i8 vectors is defined in
terms of 'char' which may or may not be signed! This causes us to
generate pmovsx and pmovzx depending on the setting of -funsigned-char.
This patch just forms an explicitly signed type and uses that to
formulate the sign extension. While this gets the correct behavior
(which we now verify with the enhanced test) this is just the tip of the
ice berg. Now that I know what to look for, I have found errors of this
sort *throughout* our vector code. Fortunately, this is the only
specific place where I know of users actively having their code
miscompiled by Clang due to this, so I'm keeping the fix for those users
minimal and targeted.
I'll be sending a proper email for discussion of how to fix these
systematically, what the implications are, and just how widely broken
this is... From what I can tell, we have never shipped a correct set of
builtin headers for x86 when users rely on -funsigned-char. Oops.
David Majnemer [Wed, 30 Sep 2015 22:07:43 +0000 (22:07 +0000)]
[Sema] Don't crash when friending an unqualified templated constructor
Unqualified templated constructors cannot be friended and our lack of a
diagnostic led to violated invariants. Instead, raise a diagnostic when
processing the friend declaration.
Douglas Gregor [Wed, 30 Sep 2015 21:27:42 +0000 (21:27 +0000)]
Don't inherit availability information when implementing a protocol requirement.
When an Objective-C method implements a protocol requirement, do not
inherit any availability information from the protocol
requirement. Rather, check that the implementation is not less
available than the protocol requirement, as we do when overriding a
method that has availability. Fixes rdar://problem/22734745.
[DarwinDriver] Use -lto_library to specify the path for libLTO.dylib
Usually, when using LTO with a clang installation newer than the
system's one, there's a libLTO.dylib version mismatch and LTO fails. One
solution to this is to make ld point to the right libLTO.dylib by
changing DYLD_LIBRARY_PATH.
However, ld64 supports specifying the complete path to the desired
libLTO.dylib through the -lto_library option. This commit adds support
for the clang driver to use this option whenever it's capable of finding
a libLTO.dylib in clang's installed library directory. This way, we
don't need to rely on DYLD_LIBRARY_PATH nor get caught by version
mismatches.
[Sema] Avoid crashing during this-> insertion recovery
We get into this bad state when someone defines a new member function
for a class but forgets to add the declaration to the class body.
Calling the new member function from a member function template of the
class will crash during instantiation.
[OpenCL 2.0] Enable program scope variables, Section 6.5.1.
- Remove virtual SC_OpenCLWorkGroupLocal storage type specifier
as it conflicts with static local variables now and prevents
diagnosing static local address space variables correctly.
- Allow static local and global variables (OpenCL2.0 s6.8 and s6.5.1).
- Improve diagnostics of allowed ASes for variables in different scopes:
(i) Global or static local variables have to be in global
or constant ASes (OpenCL1.2 s6.5, OpenCL2.0 s6.5.1);
(ii) Non-kernel function variables can't be declared in local
or constant ASes (OpenCL1.1 s6.5.2 and s6.5.3).
James Y Knight [Wed, 30 Sep 2015 14:04:23 +0000 (14:04 +0000)]
Fix FunctionParmPackExpr::Create() to take a ParmVarDecl* array.
FunctionParmPackExpr actually stores an array of ParmVarDecl* (and
accessors return that). But, the FunctionParmPackExpr::Create()
constructor accepted an array of Decl *s instead.
It was easy for this mismatch to occur without any obvious sign of
something wrong, since both the store and the access used independent
'reinterpet_cast<XX>(this+1)' calls.
[ARM][NEON] Use address space in vld([1234]|[234]lane) and vst([1234]|[234]lane) instructions
This is the clang commit associated with llvm r248887.
This commit changes the interface of the vld[1234], vld[234]lane, and vst[1234],
vst[234]lane ARM neon intrinsics and associates an address space with the
pointer that these intrinsics take. This changes, e.g.,
<2 x i32> @llvm.arm.neon.vld1.v2i32(i8*, i32)
to
<2 x i32> @llvm.arm.neon.vld1.v2i32.p0i8(i8*, i32)
This change ensures that address spaces are fully taken into account in the ARM
target during lowering of interleaved loads and stores.
projects / clang / log