[CUDA] Remove incorrect comment in CUDASetLambdaAttrs.
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.
[CUDA] Emit a warning if a CUDA host/device/global attribute is placed after '(...)'.
Summary:
This is probably the sane place for the attribute to go, but nvcc
specifically rejects it. Other GNU-style attributes are allowed in this
position (although judging from the warning it emits for
host/device/global, those attributes are applied to the lambda's
anonymous struct, not to the function itself).
It would be nice to have a FixIt message here, but doing so, or even
just getting the correct range for the attribute, including its '((' and
'))'s, is apparently Hard.
Martin Storsjo [Fri, 30 Sep 2016 19:13:46 +0000 (19:13 +0000)]
[MS] Implement __iso_volatile loads/stores as builtins
These are supposed to produce the same as normal volatile
pointer loads/stores. When -volatile:ms is specified,
normal volatile pointers are forced to have atomic semantics
(as is the default on x86 in MSVC mode). In that case,
these builtins should still produce non-atomic volatile
loads/stores without acquire/release semantics, which
the new test verifies.
These are only available on ARM (and on AArch64,
although clang doesn't support AArch64/Windows yet).
This implements what is missing for PR30394, making it possible
to compile C++ for ARM in MSVC mode with MSVC headers.
Samuel Antao [Fri, 30 Sep 2016 15:34:19 +0000 (15:34 +0000)]
[CUDA][OpenMP] Add a generic offload action builder
Summary:
This patch proposes a new class to generate and record action dependences related with offloading. The builder provides three main functionalities:
- Add device dependences to host actions.
- Add host dependence to device actions.
- Register device top-level actions.
The constructor of the builder detect the programming models that should be supported, and generates a specialized builder for each. If a new programming model is to be added in the future, only a new specialized builder has to be implemented.
When the specialized builder is generated, it produces programming-model-specific diagnostics.
A CUDA specialized builder is proposed in the patch that mostly consists of the partition of the current `buildCudaAction` by the three different functionalities.
Richard Smith [Thu, 29 Sep 2016 21:30:12 +0000 (21:30 +0000)]
Switch to a different workaround for unimplementability of P0145R3 in MS ABIs.
Instead of ignoring the evaluation order rule, ignore the "destroy parameters
in reverse construction order" rule for the small number of problematic cases.
This only causes incorrect behavior in the rare case where both parameters to
an overloaded operator <<, >>, ->*, &&, ||, or comma are of class type with
non-trivial destructor, and the program is depending on those parameters being
destroyed in reverse construction order.
We could do a little better here by reversing the order of parameter
destruction for those functions (and reversing the argument evaluation order
for all direct calls, not just those with operator syntax), but that is not a
complete solution to the problem, as the same situation can be reached by an
indirect function call.
[sanitizer-coverage/libFuzzer] make the guards for trace-pc 32-bit; create one array of guards per function, instead of one guard per BB. reorganize the code so that trace-pc-guard does not create unneeded globals
Jonas Hahnfeld [Thu, 29 Sep 2016 07:43:08 +0000 (07:43 +0000)]
XFAIL Driver/darwin-stdlib.cpp if CLANG_DEFAULT_CXX_STDLIB is set
Until someone rewrites the stdlib logic for Darwin so that we don't need
to pass down the -stdlib argument to cc1.
(see https://llvm.org/bugs/show_bug.cgi?id=30548)
Richard Smith [Wed, 28 Sep 2016 22:08:38 +0000 (22:08 +0000)]
Fix bug where template argument deduction of a non-type template parameter used
as a template argument in a template-id, from a null non-type template
argument, failed.
Richard Smith [Wed, 28 Sep 2016 21:01:37 +0000 (21:01 +0000)]
cxx_status: use HTML5 details/summary elements to hide implementation status
tables for fully-implemented language modes by default. Also add some missing
elements to TS support table.
Richard Smith [Wed, 28 Sep 2016 20:42:56 +0000 (20:42 +0000)]
Add a couple more tentative names for upcoming SD-6 feature checks. These might
not reflect the final chosen names, but supporting them now seems to have
little downside.
Richard Smith [Wed, 28 Sep 2016 19:09:10 +0000 (19:09 +0000)]
Re-commit r282556, reverted in r282564, with a fix to CallArgList::addFrom to
function correctly when targeting MS ABIs (this appears to have never mattered
prior to this change).
Update test case to always cover both 32-bit and 64-bit Windows ABIs, since
they behave somewhat differently from each other here.
Update test case to also cover operators , && and ||, which it appears are also
affected by P0145R3 (they're not explicitly called out by the design document,
but this is the emergent behavior of the existing wording).
Original commit message:
P0145R3 (C++17 evaluation order tweaks): evaluate the right-hand side of
assignment and compound-assignment operators before the left-hand side. (Even
if it's an overloaded operator.)
This completes the implementation of P0145R3 + P0400R0 for all targets except
Windows, where the evaluation order guarantees for <<, >>, and ->* are
unimplementable as the ABI requires the function arguments are evaluated from
right to left (because parameter destructors are run from left to right in the
callee).
[X86] Remove the mm_malloc.h include guard hack from the X86 builtins tests
The X86 clang/test/CodeGen/*builtins.c tests define the mm_malloc.h include
guard as a hack for avoiding its inclusion (mm_malloc.h requires a hosted
environment since it expects stdlib.h to be available - which is not the case
in these internal clang codegen tests).
This patch removes this hack and instead passes -ffreestanding to clang cc1.
Eric Liu [Wed, 28 Sep 2016 11:02:16 +0000 (11:02 +0000)]
Merge conflicting replacements when they are order-independent.
Summary:
Now two replacements are considered order-independent if applying them in
either order produces the same result. These include (but not restricted
to) replacements that:
- don't overlap (being directly adjacent is fine) and
- are overlapping deletions.
- are insertions at the same offset and applying them in either order
has the same effect, i.e. X + Y = Y + X if one inserts text X and the
other inserts text Y.
Discussion about this design can be found in D24717
Richard Smith [Tue, 27 Sep 2016 23:44:22 +0000 (23:44 +0000)]
P0145R3 (C++17 evaluation order tweaks): evaluate the right-hand side of
assignment and compound-assignment operators before the left-hand side. (Even
if it's an overloaded operator.)
This completes the implementation of P0145R3 + P0400R0 for all targets except
Windows, where the evaluation order guarantees for <<, >>, and ->* are
unimplementable as the ABI requires the function arguments are evaluated from
right to left (because parameter destructors are run from left to right in the
callee).
Alex Lorenz [Tue, 27 Sep 2016 23:30:36 +0000 (23:30 +0000)]
[Coverage] The coverage region for switch covers the code after the switch.
This patch fixes a regression introduced in r262697 that changed the way the
coverage regions for switches are constructed. The PGO instrumentation counter
for a switch statement refers to the counter at the exit of the switch.
Therefore, the coverage region for the switch statement should cover the code
that comes after the switch, and not the switch statement itself.
Richard Trieu [Tue, 27 Sep 2016 22:28:59 +0000 (22:28 +0000)]
Fix defaulted member functions for templated classes.
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.
Richard Smith [Mon, 26 Sep 2016 23:49:47 +0000 (23:49 +0000)]
P0145R3 (C++17 evaluation order tweaks): consistently emit the LHS of array
subscripting before the RHS, regardless of which is the base and which is the
index.
Daniel Jasper [Mon, 26 Sep 2016 22:19:08 +0000 (22:19 +0000)]
[clang-format] Don't allow newline after uppercase Obj-C block return types
Fixes the following:
BOOL (^aaa)(void) = ^BOOL {
};
The first BOOL's token was getting set to TT_FunctionAnnotationRParen
incorrectly, which was causing an unexpected newline after (^aaa). This
was introduced in r245846.
Avoid failing in the backend when the rewrite map does not exist. Rather check
that the map exists in the frontend before handing it off to the backend. Add
the missing rewrite maps that the tests were referencing.
[AMDGPU] Expose flat work group size, register and wave control attributes
__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
[analyzer] Fix crash in RetainCountChecker::checkEndFunction
The class BodyFarm creates bodies for
OSAtomicCompareAndSwap*, objc_atomicCompareAndSwap*, dispatch_sync*, dispatch_once*
and for them the flag isBodyAutosynthesized is set to true.
This diff
1. makes AnalysisConsumer::HandleCode skip the autosynthesized code
2. replaces assert(LCtx->getParent()) in RetainCountChecker::checkEndFunction
by assert(!LCtx->inTopFrame()) (minor cleanup)
Teresa Johnson [Fri, 23 Sep 2016 20:38:09 +0000 (20:38 +0000)]
[LTO] Add -flto-jobs=N to control backend parallelism
Summary:
Currently, a linker option must be used to control the backend
parallelism of ThinLTO. The linker option varies depending on the
linker (e.g. gold vs ld64). Add a new clang option -flto-jobs=N
to control this.
I've added in the wiring to pass this to the gold plugin. I also
added in the logic to pass this down in the form I understand that
ld64 uses on MacOS, for the darwin target.
Sebastian Pop [Fri, 23 Sep 2016 16:16:25 +0000 (16:16 +0000)]
set the underlying value of “#pragma STDC FP_CONTRACT” on by default
Clang has the default FP contraction setting of “-ffp-contract=on”, which
doesn't really mean “on” in the conventional sense of the word, but rather
really means “according to the per-statement effective value of the relevant
pragma”.
Before this patch, Clang has that pragma defaulting to “off”. Since the
“-ffp-contract=on” mode is really an AND of two booleans and the second of them
defaults to “off”, the whole thing effectively defaults to “off”. This patch
changes the default value of the pragma to “on”, thus making the default pair of
booleans (on, on) rather than (on, off). This makes FP optimization slightly
more aggressive than before when not using either “-Ofast”, “-ffast-math”, or
“-ffp-contract=fast”. Even with this patch the compiler still respects
“-ffp-contract=off”.
As per a suggestion by Steve Canon, the added code does _not_ require “-O3” or
higher. This is so as to try our best to preserve identical floating-point
results for unchanged source code compiling for an unchanged target when only
changing from any optimization level in the set (“-O0”, “-O1”, “-O2”, “-O3”) to
any other optimization level in that set. “-Os” and “-Oz” seem to be behaving
identically, i.e. should probably be considered a part of the aforementioned
set, but I have not reviewed this rigorously. “-Ofast” is explicitly _not_ a
member of that set.
Patch authored by Abe Skolnik [a.skolnik@samsung.com] and Stephen Canon [scanon@apple.com].
Eric Liu [Fri, 23 Sep 2016 15:10:56 +0000 (15:10 +0000)]
[clang-format] support header deletion in cleanupAroundReplacemnts.
Summary:
- If a replacement has offset UINT_MAX, length 0, and a replacement text
that is an #include directive, this will insert the #include into the
correct block in the \p Code.
- If a replacement has offset UINT_MAX, length 1, and a replacement text
that is the name of the header to be removed, the header will be removed
from \p Code if it exists.
[asan] Fix incorrect SEH symbol mangling on win64.
Summary:
The ASAN unittests are failing (check-asan-dynamic) due to an incorrect symbol name:
```
LINK : error LNK2001: unresolved external symbol ___asan_seh_interceptor
```
On win64, the linker is not adding an extra underscore. This was correctly fixed in the same file for other uses.
After that patch, most of the unittests are passing, but some related to SEH needs to be fixed.
```
Failing Tests (4):
AddressSanitizer-x86_64-windows-dynamic :: TestCases/Windows/dll_intercept_memchr.cc
AddressSanitizer-x86_64-windows-dynamic :: TestCases/Windows/dll_intercept_memcpy_indirect.cc
AddressSanitizer-x86_64-windows-dynamic :: TestCases/Windows/dll_seh.cc
AddressSanitizer-x86_64-windows-dynamic :: TestCases/Windows/seh.cc
[AVX-512] Add initial support for checking rounding mode arguments of builtins.
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.
Like NetBSD, OpenBSD prefers having a consistent set of typedefs
across the architectures it supports over strictly following the ARM
ABIs. The diff below makes sure that clang's view of those types
matches OpenBSD's system header files. It also adds a test that
checks the relevant types on all OpenBSD platforms that clang works
on. Hopefully we can add mips64 and powerpc to that list in the
future.
projects / clang / log