Vlad Tsyrklevich [Thu, 17 Jan 2019 17:53:45 +0000 (17:53 +0000)]
TLS: Respect visibility for thread_local variables on Darwin (PR40327)
Summary:
Teach clang to mark thread wrappers for thread_local variables with
hidden visibility when the original variable is marked with hidden
visibility. This is necessary on Darwin which exposes the thread wrapper
instead of the thread variable. The thread wrapper would previously
always be created with default visibility unless it had
linkonce*/weak_odr linkage.
CodeGen: Cast llvm.flt.rounds result to match __builtin_flt_rounds
llvm.flt.rounds returns an i32, but the builtin expects an integer.
On targets where integers are not 32-bits clang tries to bitcast the result, causing an assertion failure.
Richard Smith [Wed, 16 Jan 2019 22:01:39 +0000 (22:01 +0000)]
PR40329: [adl] Fix determination of associated classes when searching a
member enum and then its enclosing class.
There are situations where ADL will collect a class but not the complete
set of associated classes / namespaces of that class. When that
happened, and we later tried to collect those associated classes /
namespaces, we would previously short-circuit the lookup and not find
them. Eg, for:
struct A : B { enum E; };
if we first looked for associated classes/namespaces of A::E, we'd find
only A. But if we then tried to also collect associated
classes/namespaces of A (which should include the base class B), we
would not add B because we had already visited A.
This also fixes a minor issue where we would fail to collect associated
classes from an overloaded class member access expression naming a
static member function.
Leonard Chan [Wed, 16 Jan 2019 19:53:50 +0000 (19:53 +0000)]
[Fixed Point Arithmetic] Fixed Point Subtraction
This patch covers subtraction between fixed point types and other fixed point
types or integers, using the conversion rules described in 4.1.4 of N1169.
Leonard Chan [Wed, 16 Jan 2019 18:13:59 +0000 (18:13 +0000)]
[Fixed Point Arithmetic] Fixed Point Addition
This patch covers addition between fixed point types and other fixed point
types or integers, using the conversion rules described in 4.1.4 of N1169.
Usual arithmetic rules do not apply to binary operations when one of the
operands is a fixed point type, and the result of the operation must be
calculated with the full precision of the operands, so we should not perform
any casting to a common type.
This patch does not include constant expression evaluation for addition of
fixed point types. That will be addressed in another patch since I think this
one is already big enough.
Jeremy Morse [Wed, 16 Jan 2019 17:41:29 +0000 (17:41 +0000)]
Add a REQUIRES: darwin line for a mac test.
This test, apparently for macs, fails on Windows as lit can't emulate
the shell subprocess $(which...) correctly. Some other netbsd and linux
buildbots also fail here. Limit to macs as a temporary workaround.
* Accept as an argument constants in range 0..63 (aligned with TI headers and linker scripts provided with TI GCC toolchain).
* Emit function attribute 'interrupt'='xx' instead of aliases (used in the backend to create a section for particular interrupt vector).
* Add more diagnostics.
Ilya Biryukov [Wed, 16 Jan 2019 13:18:59 +0000 (13:18 +0000)]
Set '-target' flag in the test checking the MacOS include dir
To fix a buildbot failure on PS4, see
http://lab.llvm.org:8011/builders/llvm-clang-lld-x86_64-scei-ps4-ubuntu-fast/builds/42251
The test was added in r351222 and aims to check only a particular
Mac configuration. However it relied on the default compiler target
by default, therefore unintentionally failing on PS4.
Pavel Labath [Wed, 16 Jan 2019 09:55:32 +0000 (09:55 +0000)]
[Support] Remove error return value from one overload of fs::make_absolute
Summary:
The version of make_absolute which accepted a specific directory to use
as the "base" for the computation could never fail, even though it
returned a std::error_code. The reason for that seems to be historical
-- the CWD flavour (which can fail due to failure to retrieve CWD) was
there first, and the new version was implemented by extending that.
This removes the error return value from the non-CWD overload and
reimplements the CWD version on top of that. This enables us to remove
some dead code where people were pessimistically trying to handle the
errors returned from this function.
Philip Pfaffe [Wed, 16 Jan 2019 09:28:01 +0000 (09:28 +0000)]
[NewPM][TSan] Reiterate the TSan port
Summary:
Second iteration of D56433 which got reverted in rL350719. The problem
in the previous version was that we dropped the thunk calling the tsan init
function. The new version keeps the thunk which should appease dyld, but is not
actually OK wrt. the current semantics of function passes. Hence, add a
helper to insert the functions only on the first time. The helper
allows hooking into the insertion to be able to append them to the
global ctors list.
Craig Topper [Wed, 16 Jan 2019 07:17:14 +0000 (07:17 +0000)]
[X86] Correct the type string for __builtin_ia32_gathersiv16sf to make the indices an integer type not an FP type.
The element count and width remain the same. This went unnoticed because default conversion from builtin to intrinsic will generate a bitcast if the types don't match.
Eric Fiselier [Wed, 16 Jan 2019 02:34:36 +0000 (02:34 +0000)]
[SemaCXX] Unconfuse Clang when std::align_val_t is unscoped in C++03
When -faligned-allocation is specified in C++03 libc++ defines
std::align_val_t as an unscoped enumeration type (because Clang didn't
provide scoped enumerations as an extension until 8.0).
Unfortunately Clang confuses the `align_val_t` overloads of delete with
the sized deallocation overloads which aren't enabled. This caused Clang
to call the aligned deallocation function as if it were the sized
deallocation overload.
Reid Kleckner [Tue, 15 Jan 2019 21:24:55 +0000 (21:24 +0000)]
[clang-cl] Alias /Zc:alignedNew[-] to -f[no-]aligned-allocation
Implements PR40180.
clang-cl has one minor behavior difference with cl with this change.
Clang allows the user to enable the C++17 feature of aligned allocation
without enabling all of C++17, but MSVC will not call the aligned
allocation overloads unless -std:c++17 is passed. While our behavior is
technically incompatible, it would require making driver mode specific
changes to match MSVC precisely, and clang's behavior is useful because
it allows people to experiment with new C++17 features individually.
Therefore, I plan to leave it as is.
Craig Topper [Tue, 15 Jan 2019 19:58:36 +0000 (19:58 +0000)]
[Nios2] Remove Nios2 backend
As mentioned here http://lists.llvm.org/pipermail/llvm-dev/2019-January/129121.html This backend is incomplete and has not been maintained in several months.
Haojian Wu [Tue, 15 Jan 2019 19:05:50 +0000 (19:05 +0000)]
[Tooling] Make clang-tool find libc++ dir on mac when running on a file without compilation database.
Summary:
This is a regression of r348365.
When clang-tools run on a file without a complation database (`clang-check /tmp/t.cc`),
we will use fixed compilation database as a fallback. However the actual compiler
path in the fallback complation command is just `clang-tool` which is
insufficient to detect the libc++ dir.
Erich Keane [Tue, 15 Jan 2019 17:51:09 +0000 (17:51 +0000)]
Fix cpu-dispatch MV regression caused by r347812
r347812 permitted forward declarations for cpu-dispatch functions, which
are occassionally useful as exposition in header files. However, this inadvertently
permitted this function to become multiversioned after a usage. This
patch ensures that the "CausesMV" checks are still run in the
forward-declaration case.
The BoundNodesTreeBuilder class is used both directly and indirectly as a local
variable in matchesAncestorOfRecursively, memoizedMatchesAncestorOfRecursively
and other functions that happen to be on long recursive call paths. By reducing
the inline storage size of the SmallVector we dramatically reduce the stack
requirements of ASTMatchers. Running clang-tidy with a large number of checks
enabled on a few arbitrarily chosen files show no performance regression.
Roman Lebedev [Tue, 15 Jan 2019 11:20:02 +0000 (11:20 +0000)]
[OpenCL] opencl-c.h: read_image*(): sampler-less, and image{1,2}d_array_t variants are OpenCL-1.2+, mark them as such
Summary:
Refer to [[ https://www.khronos.org/registry/OpenCL/specs/opencl-1.1.pdf#page=242 | `6.11.13.2 Built-in Image Functions` ]],
and [[ https://www.khronos.org/registry/OpenCL/specs/opencl-1.1.pdf#page=306 | `9.6.8 Image Read and Write Functions` ]] of the OpenCL 1.1 spec.
* There is no mention of `image1d_array_t` and `image2d_array_t` anywhere in the OpenCL 1.1 spec.
* All the `read_image{f,i,ui,h}()` functions, as of OpenCL 1.1 spec, have a second required parameter `sampler_t sampler`
Should have prevented the following regression:
https://redmine.darktable.org/issues/12493
Roman Lebedev [Tue, 15 Jan 2019 09:44:25 +0000 (09:44 +0000)]
[clang][UBSan] Sanitization for alignment assumptions.
Summary:
UB isn't nice. It's cool and powerful, but not nice.
Having a way to detect it is nice though.
[[ https://wg21.link/p1007r3 | P1007R3: std::assume_aligned ]] / http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2018/p1007r2.pdf says:
```
We propose to add this functionality via a library function instead of a core language attribute.
...
If the pointer passed in is not aligned to at least N bytes, calling assume_aligned results in undefined behaviour.
```
This differential teaches clang to sanitize all the various variants of this assume-aligned attribute.
Requires D54588 for LLVM IRBuilder changes.
The compiler-rt part is D54590.
This is a second commit, the original one was r351105,
which was mass-reverted in r351159 because 2 compiler-rt tests were failing.
Dan Gohman [Tue, 15 Jan 2019 06:58:16 +0000 (06:58 +0000)]
[WebAssembly] Support multilibs for wasm32 and add a wasm OS that uses it
This adds support for multilib paths for wasm32 targets, following
[Debian's Multiarch conventions], and also adds an experimental OS name in
order to test it.
Craig Topper [Tue, 15 Jan 2019 05:03:18 +0000 (05:03 +0000)]
[X86] Make _xgetbv/_xsetbv on non-windows platforms
Summary:
This patch attempts to redo what was tried in r278783, but was reverted.
These intrinsics should be available on non-windows platforms with "xsave" feature check. But on Windows platforms they shouldn't have feature check since that's how MSVC behaves.
To accomplish this I've added a MS builtin with no feature check. And a normal gcc builtin with a feature check. When _MSC_VER is not defined _xgetbv/_xsetbv will be macros pointing to the gcc builtin name.
I've moved the forward declarations from intrin.h to immintrin.h to match the MSDN documentation and used that as the header file for the MS builtin.
I'm not super happy with this implementation, and I'm open to suggestions for better ways to do it.
Petr Hosek [Tue, 15 Jan 2019 01:15:00 +0000 (01:15 +0000)]
[CMake][Fuchsia] Synchronize first and second stage builds
This reorders options between the first and second stage builds to make
them better lined up. The change also re-enables tests for first stage
which is useful e.g. for cross-compiling when we cannot run tests for
second stage directly (i.e. without emulation).
Roman Lebedev [Mon, 14 Jan 2019 19:09:27 +0000 (19:09 +0000)]
[clang][UBSan] Sanitization for alignment assumptions.
Summary:
UB isn't nice. It's cool and powerful, but not nice.
Having a way to detect it is nice though.
[[ https://wg21.link/p1007r3 | P1007R3: std::assume_aligned ]] / http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2018/p1007r2.pdf says:
```
We propose to add this functionality via a library function instead of a core language attribute.
...
If the pointer passed in is not aligned to at least N bytes, calling assume_aligned results in undefined behaviour.
```
This differential teaches clang to sanitize all the various variants of this assume-aligned attribute.
Requires D54588 for LLVM IRBuilder changes.
The compiler-rt part is D54590.
Kristina Brooks [Mon, 14 Jan 2019 18:16:51 +0000 (18:16 +0000)]
[Sema] Expose a control flag for integer to pointer ext warning
While building openJDK11u, it seems that some of the code in the
native core libraries make liberal use of integer to pointer
comparisons. We currently have no flag to disabled this warning.
This add such a flag.
Sam McCall [Mon, 14 Jan 2019 17:16:00 +0000 (17:16 +0000)]
[AST] Fix double-traversal of code in top-level lambdas in RAV(implicit = yes).
Summary:
Prior to r351069, lambda classes were traversed or not depending on the
{Function, Class, Namespace, TU} DeclContext containing them.
If it was a function (common case) they were not traversed.
If it was a namespace or TU (top-level lambda) they were traversed as part of
that DeclContext traversal.
r351069 "fixed" RAV to traverse these as part of the LambdaExpr, which is the
right place. But top-level lambdas are now traversed twice.
We fix that as blocks and block captures were apparently fixed in the past.
Maybe it would be nicer to avoid adding the lambda classes to the DeclContext
in the first place, but I can't work out the implications of that.
[OpenCL] Set generic addr space of 'this' in special class members.
Set address spaces of 'this' param correctly for implicit special
class members.
This also changes initialization conversion sequence to separate
address space conversion from other qualifiers in case of binding
reference to a temporary. In this case address space conversion
should happen after the binding (unlike for other quals). This is
needed to materialize it correctly in the alloca address space.
Sam McCall [Mon, 14 Jan 2019 10:31:42 +0000 (10:31 +0000)]
[AST] RecursiveASTVisitor visits lambda classes when implicit visitation is on.
Summary:
This fixes ASTContext's parent map for nodes in such classes (e.g. operator()).
https://bugs.llvm.org/show_bug.cgi?id=39949
This also changes the observed shape of the AST for implicit RAVs.
- this includes AST MatchFinder: cxxRecordDecl() now matches lambda classes,
functionDecl() matches the call operator, and the parent chain is body -> call
operator -> lambda class -> lambdaexpr rather than body -> lambdaexpr.
- this appears not to matter for the ASTImporterLookupTable builder
- this doesn't matter for the other RAVs in-tree.
In order to do this, we remove the TraverseLambdaBody hook. The problem is it's
hard/weird to ensure this hook is called when traversing via the implicit class.
There were just two users of this hook in-tree, who use it to skip bodies.
I replaced these with explicitly traversing the captures only. Another approach
would be recording the bodies when the lambda is visited, and then recognizing
them later.
I'd be open to suggestion on how to preserve this hook, instead.
Steven Wu [Fri, 11 Jan 2019 21:16:04 +0000 (21:16 +0000)]
[Darwin][Driver] Don't pass a file as object_path_lto during ThinLTO
Summary:
After r327851, Driver::GetTemporaryPath will create the file rather than
just create a potientially unqine filename. If clang driver pass the
file as parameter as -object_path_lto, ld64 will pass it back to libLTO
as GeneratedObjectsDirectory, which is going to cause a LLVM ERROR if it
is not a directory.
Now during thinLTO, pass a temp directory path to linker instread.
Stephen Kelly [Fri, 11 Jan 2019 19:16:01 +0000 (19:16 +0000)]
Implement Attr dumping in terms of visitors
Remove now-vestigial dumpType and dumpBareDeclRef methods. The old
tablegen generated code used to expect them to be present, but the new
generated code has no such requirement.
Teresa Johnson [Fri, 11 Jan 2019 18:32:07 +0000 (18:32 +0000)]
[LTO] Add option to enable LTOUnit splitting, and disable unless needed
Summary:
Adds a new -f[no]split-lto-unit flag that is disabled by default to
control module splitting during ThinLTO. It is automatically enabled
for -fsanitize=cfi and -fwhole-program-vtables.
The new EnableSplitLTOUnit codegen flag is passed down to llvm
via a new module flag of the same name.
projects / clang / log