Driver: Stop linking to C++ when using sanitizers on Darwin
Sanitizers on Darwin are built as dynamic libraries, not static libraries.
Sanitizers will have their C++ dependency satisfied internally (LC_LOAD_DYLIB)
in the libclang_rt dylib. As long as the sanitizers stay dynamic and not static,
linking against C++ when enabling a sanitizer becomes over linkage.
[CUDA] Check that our CUDA install supports the requested architectures.
Summary:
Raise an error if you're using a CUDA installation that's too old for
the requested architectures. In practice, this means that you need a
CUDA 8 install to compile for sm_6*.
[OPENMP] Do not create helper expressions in dependent contexts, NFC.
OpenMP relies on some helper expressions generated during semantic
analysis. But they are required only for codegen and not required in
dependent contexts. Patch removes generation of some of helper
expressions.
[analyzer] Suppress false positives in std::shared_ptr
The analyzer does not model C++ temporary destructors completely and so
reports false alarms about leaks of memory allocated by the internals of
shared_ptr:
std::shared_ptr<int> p(new int(1));
p = nullptr; // 'Potential leak of memory pointed to by field __cntrl_'
This patch suppresses all diagnostics where the end of the path is inside
a method in std::shared_ptr.
It also reorganizes the tests for suppressions in the C++ standard library
to use a separate simulated header for library functions with bugs
that were deliberately inserted to test suppression. This will prevent
other tests from using these as models.
[CUDA] Add utility functions for dealing with CUDA versions / architectures.
Summary:
Currently our handling of CUDA architectures is scattered all around
clang. This patch centralizes it.
A key advantage of this centralization is that you can now write a C++
switch on e.g. CudaArch and get a compile error if you don't handle one
of the enum values.
David Majnemer [Wed, 6 Jul 2016 21:07:53 +0000 (21:07 +0000)]
[CodeGen, DebugInfo] Use hasLocalLinkage instead of hasInternalLinkage
For the purpose of emitting debug info, entities with private linkage
should be treated the same as internal linkage. While this doesn't
change anything in practice, it makes the code a little less confusing.
This patch: (a) revivies the behavior of `3.0 + vget_low_f64(v)` prior to
r266366, (b) add support for `vget_low_f64(v) + 3.0` and (c) add support for
assignments.
These vector semantics have never really been tied up but it seems
odd that we used to support some binop froms but do not support
assignment. If we did support scalar for the purposes of arithmetic, we
should probably be able to reinterpret as scalar for the purposes of
assignment too.
Adrian McCarthy [Wed, 6 Jul 2016 14:46:42 +0000 (14:46 +0000)]
Include debug info for nested structs and classes
This includes nested types in the member list, even if there are no members of that type. Note that structs and classes have themselves as an "implicit struct" as the first member, so we skip implicit ones.
[X86] Use native IR for immediate values 0-7 of packed fp cmp builtins. This makes them the same as what is done when using the SSE builtins for these same encodings.
Erik Pilkington [Tue, 5 Jul 2016 17:57:24 +0000 (17:57 +0000)]
[Sema] Fix a bug where pack expansion was not expanded in type alias
The problem is that the parameter pack in a function type type alias is not
reexpanded after being transformed. Also remove an incorrect comment in a
similar function. Fixes PR26017.
Re-apply "test: Use add_lit_testsuites so that subsets of tests can be specified"
This version should actually remove the empty directories I removed
all of the files from. Thanks to tstellar for pointing out git-svn's
--rmdir flag.
Original message:
This creates make/ninja targets like check-clang-codegen and
check-clang-unit, much like LLVM already has. I had to move some input
files into Input directories so they weren't picked up as test
directories.
test: Use add_lit_testsuites so that subsets of tests can be specified
This creates make/ninja targets like check-clang-codegen and
check-clang-unit, much like LLVM already has. I had to move some input
files into Input directories so they weren't picked up as test
directories.
[X86][AVX512F] add float/double abs intrinsics
add abs intrinsics that use native LLVM-IR.
change _mm512_mask[z]_and_epi{32|64} to use select intrinsic
[OpenCL] An implementation of device side enqueue (DSE) from OpenCL v2.0 s6.13.17.
- Added new Builtins: enqueue_kernel, get_kernel_work_group_size
and get_kernel_preferred_work_group_size_multiple.
These Builtins use custom check to diagnose parameters of the passed Blocks
i. e. variable number of 'local void*' type params, and check different
overloads specified in Table 6.31 of OpenCL v2.0.
- IR is generated as an internal library call for each OpenCL Builtin,
reusing ObjC Block implementation.
ntrinsics _mm256_permutexvar_epi64 doesn't accept three parameters as specify bellow.
I deleted the extra mask parameter.
__m256i _mm256_permutexvar_epi64 (__m256i idx, __m256i a)
#include "immintrin.h"
Instruction: vpermq
CPUID Flags: AVX512VL + AVX512F
Description
Shuffle 64-bit integers in a across lanes using the corresponding index in idx, and store the results in dst.
Operation
FOR j := 0 to 3
i := j*64
id := idx[i+1:i]*64
dst[i+63:i] := a[id+63:id]
ENDFOR
dst[MAX:256] := 0
dst[MAX:256] := 0
[X86] Use undefined instead of setzero in shufflevector based intrinsics when the second source is unused. Rewrite immediate extractions in shuffle intrinsics to be in ((c >> x) & y) form instead of ((c & z) >> x). This way only x varies between each use instead of having to vary x and z.
[Sparc] Allow LEON cpu models to be selected with -mcpu
Summary: This change exposes the recently added LEON CPUs (D19359) in the LLVM Sparc backend to Clang, allowing the cpu's to be selected using the -mcpu flag.
[AVX512] Modify what indices we emit for the zero vector we use for zero extension of the result of a v2i1 or v4i1 masked compare. This way we emit something that the backend easily interprets as a concatenation rather than a true shuffle. This delivers slightly better codegen with the current backend capabilities.
[Refactor NFC] Rename the (non-CCE, fold-failure) Diag during constant expression evaluation as FFDiag.
Currently, we have CCEDiags (C++11 core constant expression diags) and Fold failure diagnostics [I don't claim to yet fully understand exactly why we need the difference]. This patch explicitly replaces Info.Diag (whose use always represents a fold failure diag within the file) with Info.FFDiag. This makes it more easily greppable in the file, and just like the name Info.CCEDiag, it gives the reader slight further insight into the nature of the diagnostic (as opposed to Info.Diag).
This patch is a preliminary refactoring step in an effort to allow support for compatibility-warnings and extensions (such as constexpr lambda) during constant expression evaluation.
Richard Smith [Sat, 2 Jul 2016 01:32:16 +0000 (01:32 +0000)]
PR28394: For compatibility with c++11 and c++14, if a static constexpr data
member is redundantly redeclared outside the class definition in code built in
c++17 mode, ensure we emit a non-discardable definition of the data member for
c++11 and c++14 compilations to use.
Summary:
The TargetInfo for 'renderscript32' and 'renderscript64' ArchTypes are
subclasses of ARMleTargetInfo and AArch64leTargetInfo respectively.
RenderScript32TargetInfo modifies the ARM ABI to set LongWidth and
LongAlign to be 64-bits. Other than this modification, the underlying
TargetInfo base classes is initialized as if they have "armv7" and
"aarch64" architecture type respectively.
David Majnemer [Fri, 1 Jul 2016 23:12:54 +0000 (23:12 +0000)]
[CodeView] Include MSVC style names for unnamed types
The CodeView printer expects to be able to generate fully qualified
names from the debug info graph. This means that we need to include the
MSVC-style name in the debug info for anonymous types.
C++14 init-capture: error out instead of crashing.
When we have template arguments, we have a function and a pattern, the variable
in init-capture belongs to the pattern decl when checking if the lhs of
"max = current" is modifiable:
auto find = [max = init](auto current) {
max = current;
};
In function isReferenceToNonConstCapture, we handle the case where the decl
context for the variable is not part of the current context.
Instead of crashing, we emit an error message:
cannot assign to a variable captured by copy in a non-mutable lambda
Matt Arsenault [Fri, 1 Jul 2016 17:38:14 +0000 (17:38 +0000)]
Emit more intrinsics for builtin functions
This is important for building libclc. Since r273039 tests are failing
due to now emitting calls to these functions instead of emitting the
DAG node. The libm function names are implemented for OpenCL, and should
call the locally defined versions, so -fno-builtin is used. The IR
Some functions use the __builtins and expect the intrinsics to be
emitted. Without this we end up with nobuiltin calls to intrinsics
or to unsupported library calls.
Eric Fiselier [Fri, 1 Jul 2016 01:24:09 +0000 (01:24 +0000)]
[Feature] Add a builtin for indexing into parameter packs. Patch by Louis Dionne.
This patch adds a __nth_element builtin that allows fetching the n-th type of a
parameter pack with very little compile-time overhead. The patch was inspired by
r252036 and r252115 by David Majnemer, which add a similar __make_integer_seq
builtin for efficiently creating a std::integer_sequence.
Reviewed as D15421. http://reviews.llvm.org/D15421
Samuel Antao [Thu, 30 Jun 2016 21:22:08 +0000 (21:22 +0000)]
[OpenMP] Use fopenmp prefix for all options introduced by the offloading implementation.
Summary: This patch changes the options used by offloading to start with -fopenmp instead of -fomp. This makes the option naming more consistent and materializes a suggestion by Richard Smith in http://reviews.llvm.org/D9888.
Justin Lebar [Thu, 30 Jun 2016 20:29:29 +0000 (20:29 +0000)]
Fix ASTMatchersNodeTest to work on Windows.
It was failing because it had an explicit check for whether we're on
Windows.
There are a few other similar explicit checks in this file which I
didn't remove because they serve as reasonable documentation that the
test doesn't work with a Windows triple.
Richard Smith [Thu, 30 Jun 2016 20:24:30 +0000 (20:24 +0000)]
Fix typo-correction crash if a typo occurs within the operand of a
function-style cast to a non-dependent type which is then used in an invalid
way. We'd lose the "type dependent" bit here, and downstream Sema processing
would then discard the expression if it was used in a context where its type
rendered it invalid.
Summary:
This lets LLVM perform IPO over these functions. In particular, it
allows LLVM to emit ld.global.nc for loads to __restrict pointers in
kernels that are never written to.
Justin Lebar [Thu, 30 Jun 2016 18:12:25 +0000 (18:12 +0000)]
Don't instantiate a full host toolchain in ASTMatchersTest.
Summary:
This test was stat()'ing large swaths of /usr/lib hundreds of times, as
every invocation of matchesConditionally*() created a new Linux
toolchain.
In addition to being slow, perf indicated this was causing substantial
contention in the kernel.
Something is...interesting in the kernel, as without this patch I
sometimes see ~11m spent in the kernel, and sometimes ~5m. This
corresponds to bimodal ninja check-clang times of ~30s and ~20s.
It's not clear to me exactly what causes the bimodality. In any case,
this change makes this test run in 2.5s, down from 17s, and it seems to
cause us to get the 20s ninja check-clang time unconditionally.
Reid Kleckner [Thu, 30 Jun 2016 17:41:31 +0000 (17:41 +0000)]
[codeview] Emit qualified display names if -gline-tables-only is on
When -gmlt is on, we don't emit namespace or class scope information,
and the CodeView emission code in LLVM can't compute the fully qualified
name. If we know LLVM won't be able to get the name right, go ahead and
emit the qualified name in the frontend.
We could change our -gmlt emission strategy to include those scopes when
emitting codeview, but that would increase memory usage and slow down
LTO and add more complexity to debug info emission.
The same problem exists when you debug a -gmlt binary with GDB, so we
should consider removing '&& EmitCodeView' from the condition here at
some point in the future after evaluating the impact on object file
size.