Justin Lebar [Thu, 13 Oct 2016 20:52:12 +0000 (20:52 +0000)]
[CUDA] Emit deferred diagnostics during Sema rather than during codegen.
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.
CodeGen: ensure that the runtime calling convention matches
Incorrect specification of the calling convention results in UB which can cause
the code path to be eliminated. Simplify the existing code by using the
RuntimeCall constructor in `CodeGenFunction`.
Justin Lebar [Thu, 13 Oct 2016 18:45:08 +0000 (18:45 +0000)]
[CUDA] Add Sema::CUDADiagBuilder and Sema::CUDADiagIf{Device,Host}Code().
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.
Artem Dergachev [Thu, 13 Oct 2016 11:41:12 +0000 (11:41 +0000)]
[analyzer] Link libStaticAnalyzerCheckers to libASTMatchers.
AST matchers are useful for the analyzer's checkers.
More patches on particular checkers shall follow.
This is the first time clang binary gets linked to ASTMatchers.
The binary size increase for the clang executable would be
+0.5% in release mode, +2% in debug mode.
Alexey Bataev [Thu, 13 Oct 2016 09:52:46 +0000 (09:52 +0000)]
Fix for PR30639: CGDebugInfo Null dereference with OpenMP array
access, by Erich Keane
OpenMP creates a variable array type with a a null size-expr. The Debug
generation failed to due to this. This patch corrects the openmp
implementation, updates the tests, and adds a new one for this
condition.
Devin Coughlin [Wed, 12 Oct 2016 23:57:05 +0000 (23:57 +0000)]
[analyzer] DeallocChecker: Don't warn about directly-set IBOutlet ivars on macOS
On macOS (but not iOS), if an ObjC property has no setter, the nib-loading code
for an IBOutlet is documented as directly setting the backing ivar without
retaining the value -- even if the property is 'retain'.
This resulted in false positives from the DeallocChecker for code that did not
release such ivars in -dealloc.
To avoid these false positives, treat IBOutlet ivars that back a property
without a setter as having an unknown release requirement in macOS.
Albert Gutowski [Wed, 12 Oct 2016 22:01:05 +0000 (22:01 +0000)]
Implement MS _BitScan intrinsics
Summary: _BitScan intrinsics (and some others, for example _Interlocked and _bittest) are supposed to work on both ARM and x86. This is an attempt to isolate them, avoiding repeating their code or writing separate function for each builtin.
Alex Lorenz [Wed, 12 Oct 2016 11:40:15 +0000 (11:40 +0000)]
[CodeCompletion] Show protocol properties that are accessed through qualified id
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.
Alex Lorenz [Wed, 12 Oct 2016 09:36:35 +0000 (09:36 +0000)]
[Sema] Handle transparent_union attributes in C mode only
This commit marks the transparent_union attributes as C only because clang
doesn't support them in C++ mode. Prior to this commit, clang still tried to
verify these attributes in C++, leading to crashes when analyzing templated
transparent_union unions that have dependent field types. This commit ensures
that such crashes won't happen again.
As a result of this commit clang now displays a warning every time it encounters
a transparent_union attribute in C++ mode.
Mehdi Amini [Tue, 11 Oct 2016 19:04:24 +0000 (19:04 +0000)]
Revert "Change Builtins name to be stored as StringRef instead of raw pointers (NFC)"
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.
Reapply [Driver][Diagnostics] Make 'show option names' default for driver warnings
Reapply r283827 by fixing the tests to not be target specific
Currently, driver level warnings do not show option names (e.g. warning:
complain about foo [-Woption-name]) in a diagnostic unless
-fdiagnostics-show-option is explictly specified. OTOH, the driver by
default turn this option on for CC1. Change the logic to show option
names by default in the driver as well.
Justin Lebar [Tue, 11 Oct 2016 17:36:03 +0000 (17:36 +0000)]
[CUDA] Re-land support for <complex> (r283683 and r283680).
These were reverted in r283753 and r283747.
The first patch added a header to the root 'Headers' install directory,
instead of into 'Headers/cuda_wrappers'. This was fixed in the second
patch, but by then the damage was done: The bad header stayed in the
'Headers' directory, continuing to break the build.
We reverted both patches in an attempt to fix things, but that still
didn't get rid of the header, so the Windows boostrap build remained
broken.
It's probably worth fixing up our cmake logic to remove things from the
install dirs, but in the meantime, re-land these patches, since we
believe they no longer have this bug.
Manuel Klimek [Tue, 11 Oct 2016 09:25:34 +0000 (09:25 +0000)]
Delete clang-completion-mode.el.
It has been unmaintained for a while (last change was more than four
years ago), and it appears not widely used.
By now there are multiple well-maintained alternatives (emacs-ycmd,
atuo-complete-clang), and if users try to make this work they'll likely
have a bad user experience.
Reasoning and problems pointed out by Philipp Stephani.
Hal Finkel [Tue, 11 Oct 2016 00:26:09 +0000 (00:26 +0000)]
Add an option to save the backend-produced YAML optimization record to a file
The backend now has the capability to save information from optimizations, the
same information that can be used to generate optimization diagnostics but in
machine-consumable form, into an output file. This can be enabled when using
opt (see r282539), and this change enables it when using clang. The idea is
that other tools will be able to consume these files, and perhaps in
combination with the original source code, produce various kinds of
optimization reports for users (and for compiler developers).
We now have at-least two tools that can consume these files:
* tools/llvm-opt-report
* utils/opt-viewer
Using the flag -fsave-optimization-record will cause the YAML file to be
generated; the file name will be based on the output file name (if we're using
-c or -S and have an output name), or the input file name. When we're using
CUDA, or some other offloading mechanism, separate files are generated for each
backend target. The output file name can be specified by the user using
-foptimization-record-file=filename.
Richard Smith [Tue, 11 Oct 2016 00:21:10 +0000 (00:21 +0000)]
Aligned allocation versus CUDA: make deallocation function preference order
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.
[Driver][Diagnostics] Make 'show option names' default for driver warnings
Currently, driver level warnings do not show option names (e.g. warning:
complain about foo [-Woption-name]) in a diagnostic unless
-fdiagnostics-show-option is explictly specified. OTOH, the driver by
default turn this option on for CC1. Change the logic to show option
names by default in the driver as well.
Adrian Prantl [Mon, 10 Oct 2016 21:56:20 +0000 (21:56 +0000)]
[Driver] Let -gline-tables-only win when it comes after -gmodules.
The -gmodules option is all about putting debug type info into clang
modules and for line tables the type information is irrelevant, so
combining these two options makes no sense.
This commmit fixes the behavior to match the one documented on the
clang man page: the last -g... option wins.
Nick Lewycky [Mon, 10 Oct 2016 20:07:13 +0000 (20:07 +0000)]
Make the LValue created in EmitValueForIvarAtOffset have the same Qualifiers in the LValue as the QualType in the LValue. No functionality change intended.
Eric Fiselier [Mon, 10 Oct 2016 14:26:40 +0000 (14:26 +0000)]
[Sema] Prevent using member declaration diagnostic if the base class is invalid.
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`.
Nico Weber [Mon, 10 Oct 2016 14:10:00 +0000 (14:10 +0000)]
Revert r283680.
Breaks bootstrap builds on (at least) Windows:
In file included from D:\buildslave\clang-x64-ninja-win7\llvm\lib\Support\Allocator.cpp:14:
In file included from D:\buildslave\clang-x64-ninja-win7\llvm\include\llvm/Support/Allocator.h:24:
In file included from D:\buildslave\clang-x64-ninja-win7\llvm\include\llvm/ADT/SmallVector.h:20:
In file included from D:\buildslave\clang-x64-ninja-win7\llvm\include\llvm/Support/MathExtras.h:19:
D:\buildslave\clang-x64-ninja-win7\stage1.install\bin\..\lib\clang\4.0.0\include\algorithm(63,8) :
error: unknown type name '__device__'
inline __device__ const __T &
Michal Gorny [Mon, 10 Oct 2016 12:23:40 +0000 (12:23 +0000)]
[Driver] Make -print-libgcc-file-name print compiler-rt lib when used
Make the -print-libgcc-file-name option print an appropriate compiler
runtime library, that is libgcc.a if gcc runtime is used
and an appropriate compiler-rt library if that runtime is used.
The main use for this is to allow linking executables built with
-nodefaultlibs (e.g. to avoid linking to the standard C++ library) to
the compiler runtime library, e.g. using:
in which case currently a program built like this linked to the gcc
runtime unconditionally. The patch fixes it to use compiler-rt libraries
instead when compiler-rt is the active runtime.
1.'v' constraint for (x86) non-avx arch imitates the already implemented 'x' constraint, i.e. allows XMM{0-15} & YMM{0-15} depending on the apparent arch & mode (32/64).
2.for the avx512 arch it allows [X,Y,Z]MM{0-31} (mode dependent)
This patch applies the needed changes to clang
LLVM patch: https://reviews.llvm.org/D25005
Hal Finkel [Sun, 9 Oct 2016 03:06:31 +0000 (03:06 +0000)]
When optimizing for size, enable loop rerolling by default
We have a loop-rerolling optimization which can be enabled by using
-freroll-loops. While sometimes loops are hand-unrolled for performance
reasons, when optimizing for size, we should always undo this manual
optimization to produce smaller code (our optimizer's unroller will still
unroll the rerolled loops if it thinks that is a good idea).
Justin Lebar [Sat, 8 Oct 2016 22:16:03 +0000 (22:16 +0000)]
[CUDA] Declare our __device__ math functions in the same inline namespace as our standard library.
Summary:
Currently we declare our inline __device__ math functions in namespace
std. But libstdc++ and libc++ declare these functions in an inline
namespace inside namespace std. We need to match this because, in a
later patch, we want to get e.g. <complex> to use our device overloads,
and it only will if those overloads are in the right inline namespace.