[PR40778][Sema] Adjust addr space of operands in builtin operators.
Adjust address space for references and pointer operands of builtin operators.
Currently this change only fixes addr space in assignment (= and |=) operator,
that is needed for the test case reported in the bug. Wider support for all
other operations will follow.
Gabor Marton [Thu, 7 Mar 2019 13:01:51 +0000 (13:01 +0000)]
[ASTImporter] Handle redecl chain of FunctionTemplateDecls
Summary:
Redecl chains of function templates are not handled well currently. We
want to handle them similarly to functions, i.e. try to keep the
structure of the original AST as much as possible. The aim is to not
squash a prototype with a definition, rather we create both and put them
in a redecl chain.
David Stenberg [Thu, 7 Mar 2019 13:01:17 +0000 (13:01 +0000)]
[analyzer] Handle comparison between non-default AS symbol and constant
Summary:
When comparing a symbolic region and a constant, the constant would be
widened or truncated to the width of a void pointer, meaning that the
constant could be incorrectly truncated when handling symbols for
non-default address spaces. In the attached test case this resulted in a
false positive since the constant was truncated to zero. To fix this,
widen/truncate the constant to the width of the symbol expression's
type.
This commit does not consider non-symbolic regions as I'm not sure how
to generalize getting the type there.
Since
commit 56f548bbbb7e4387a69708f70724d00e9e076153
[modules] Round-trip -Werror flag through explicit module build.
the behavior of CXTranslationUnit_KeepGoing changed:
Unresolved #includes are fatal errors again. As a consequence, some
templates are not instantiated and lead to confusing errors.
Revert to the old behavior: With CXTranslationUnit_KeepGoing fatal
errors are mapped to errors.
Michael Liao [Wed, 6 Mar 2019 21:16:27 +0000 (21:16 +0000)]
[CUDA][HIP][DebugInfo] Skip reference device function
Summary:
- A device functions could be used as a non-type template parameter in a
global/host function template. However, we should not try to retrieve that
device function and reference it in the host-side debug info as it's
only valid at device side.
This reverts rL355522 (https://reviews.llvm.org/D57335).
Kills buildbots that use '-Werror' with the following error:
/var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/llvm/lib/IR/Value.cpp:657:7: error: default label in switch which covers all enumeration values [-Werror,-Wcovered-switch-default]
See buildbots http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fast/builds/30200/steps/check-llvm%20asan/logs/stdio for more information.
Reland "[Remarks] Refactor remark diagnostic emission in a RemarkStreamer"
This allows us to store more info about where we're emitting the remarks
without cluttering LLVMContext. This is needed for future support for
the remark section.
[Remarks] Refactor remark diagnostic emission in a RemarkStreamer
This allows us to store more info about where we're emitting the remarks
without cluttering LLVMContext. This is needed for future support for
the remark section.
Hans Wennborg [Wed, 6 Mar 2019 10:26:19 +0000 (10:26 +0000)]
Inline asm constraints: allow ICE-like pointers for the "n" constraint (PR40890)
Apparently GCC allows this, and there's code relying on it (see bug).
The idea is to allow expression that would have been allowed if they
were cast to int. So I based the code on how such a cast would be done
(the CK_PointerToIntegral case in IntExprEvaluator::VisitCastExpr()).
Roman Lebedev [Wed, 6 Mar 2019 07:45:10 +0000 (07:45 +0000)]
[clang][OpenMP] Revert "OMPFlushClause is synthetic, no such clause exists"
Summary:
This reverts rL352390 / D57280.
As discussed in https://reviews.llvm.org/D57112#inline-506781,
'flush' clause does not exist in the OpenMP spec, it can not be
specified, and `OMPFlushClause` class is just a helper class.
Now, here's the caveat. I have read @ABataev's
> Well, I think it would be good to filter out OMPC_flush somehow
> because there is no such clause actually, it is a pseudo clause
> for better handling of the flush directive.
as if that clause is pseudo clause that only exists for the sole
purpose of simplifying the parser. As in, it never reaches AST.
I did not however try to verify that. Too bad, i was wrong.
It absolutely *does* reach AST. Therefore my understanding/justification
for the change was flawed, which makes the patch a regression which **must** be reverted.
@gribozavr has brought that up again in https://reviews.llvm.org/D57112#inline-521238
> > ...
> Sorry to be late for this discussion, but I don't think this conclusion
> follows. ASTMatchers are supposed to match the AST as it is.
> Even if OMPC_flush is synthetic, it exists in the AST, and users might
> want to match it. I think users would find anything else (trying to filter
> out AST nodes that are not in the source code) to be surprising. For example,
> there's a matcher materializeTemporaryExpr even though this AST node is a
> Clang invention and is not a part of the C++ spec.
>
> Matching only constructs that appear in the source code is not feasible with
> ASTMatchers, because they are based on Clang's AST that exposes tons of semantic
> information, and its design is dictated by the structure of the semantic information.
> See "RFC: Tree-based refactorings with Clang" in cfe-dev for a library that will
> focus on representing source code as faithfully as possible.
>
> Not to even mention that this code is in ASTTypeTraits, a general library for
> handling AST nodes, not specifically for AST Matchers...
Add a remark for importing modules. Depending on whether this is a
direct import (into the TU being built by this compiler instance) or
transitive import (into an already-imported module), the diagnostic has
two forms:
importing module 'Foo' from 'path/to/Foo.pcm'
importing module 'Foo' into 'Bar' from 'path/to/Foo.pcm'
Also drop a redundant FileCheck invocation in Rmodule-build.m that was
using -Reverything, since the notes from -Rmodule-import were confusing
it.
Leonard Chan [Wed, 6 Mar 2019 00:28:43 +0000 (00:28 +0000)]
[Fixed Point Arithmetic] Fixed Point and Integer Conversions
This patch includes the necessary code for converting between a fixed point type and integer.
This also includes constant expression evaluation for conversions with these types.
Yaxun Liu [Tue, 5 Mar 2019 18:19:35 +0000 (18:19 +0000)]
[CUDA][HIP][Sema] Fix template kernel with function as template parameter
If a kernel template has a function as its template parameter, a device function should be
allowed as template argument since a kernel can call a device function. However,
currently if the kernel template is instantiated in a host function, clang will emit an error
message saying the device function is an invalid candidate for the template parameter.
This happens because clang checks the reference to the device function during parsing
the template arguments. At this point, the template is not instantiated yet. Clang incorrectly
assumes the device function is called by the host function and emits the error message.
This patch fixes the issue by disabling checking of device function during parsing template
arguments and deferring the check to the instantion of the template. At that point, the
template decl is already available, therefore the check can be done against the instantiated
function template decl.
Yaxun Liu [Tue, 5 Mar 2019 17:52:32 +0000 (17:52 +0000)]
Allow bundle size to be 0 in clang-offload-bundler
HIP uses clang-offload-bundler to create fat binary. The bundle for host is empty.
Currently clang-offload-bundler checks if the bundle size is 0 when unbundling.
If so it will exit without unbundling the remaining bundles. This causes
clang-offload-bundler not being able to unbundle fat binaries generated for HIP.
This patch allows bundles size to be 0 when clang-offload-bundler unbundles
input files.
Alexey Bataev [Tue, 5 Mar 2019 17:47:18 +0000 (17:47 +0000)]
[OPENMP]Target region: emit const firstprivates as globals with constant
memory.
If the variable with the constant non-scalar type is firstprivatized in
the target region, the local copy is created with the data copying.
Instead, we allocate the copy in the constant memory and avoid extra
copying in the outlined target regions. This global copy is used in the
target regions without loss of the performance.
Yaxun Liu [Tue, 5 Mar 2019 16:07:56 +0000 (16:07 +0000)]
[HIP] Do not unbundle object files for -fno-gpu-rdc
When -fno-gpu-rdc is set, device code is compiled, linked, and assembled into fat binary
and embedded as string in object files. The object files are normal object files which
can be linked by host linker. In the linking stage, the object files should not be unbundled
when -fno-gpu-rdc is set since they are normal object files, not bundles. The object files
only need to be unbundled when -fgpu-rdc is set.
Currently clang always unbundles object files, disregarding -fgpu-rdc option.
Gabor Marton [Tue, 5 Mar 2019 11:23:24 +0000 (11:23 +0000)]
[ASTImporter] Fix redecl failures of Class and ClassTemplate
Summary:
Redecl chains of classes and class templates are not handled well
currently. We want to handle them similarly to functions, i.e. try to
keep the structure of the original AST as much as possible. The aim is
to not squash a prototype with a definition, rather we create both and
put them in a redecl chain.
Shoaib Meenai [Tue, 5 Mar 2019 00:38:32 +0000 (00:38 +0000)]
[cmake] Create exports for umbrella library targets
When using the umbrella llvm-libraries and clang-libraries targets, we
should export all library targets, otherwise they'll be part of our
distribution but not usable from the CMake package.
Shoaib Meenai [Mon, 4 Mar 2019 21:19:53 +0000 (21:19 +0000)]
[build] Rename clang-headers to clang-resource-headers
Summary:
The current install-clang-headers target installs clang's resource
directory headers. This is different from the install-llvm-headers
target, which installs LLVM's API headers. We want to introduce the
corresponding target to clang, and the natural name for that new target
would be install-clang-headers. Rename the existing target to
install-clang-resource-headers to free up the install-clang-headers name
for the new target, following the discussion on cfe-dev [1].
I didn't find any bots on zorg referencing install-clang-headers. I'll
send out another PSA to cfe-dev to accompany this rename.
Shafik Yaghmour [Mon, 4 Mar 2019 20:25:54 +0000 (20:25 +0000)]
[ASTImporter] Handle built-in when importing SourceLocation and FileID
Summary:
Currently when we see a built-in we try and import the include location. Instead what we do now is find the buffer like we do for the invalid case and copy that over to the to context.
Rong Xu [Mon, 4 Mar 2019 20:21:31 +0000 (20:21 +0000)]
[PGO] Clang part of change for context-sensitive PGO (part1)
Part 1 of CSPGO change in Clang. This includes changes in clang options
and calls to llvm PassManager. Tests will be committed in part2.
This change needs the PassManager change in llvm.
Erich Keane [Mon, 4 Mar 2019 14:54:52 +0000 (14:54 +0000)]
Give builtins and alloc/dealloc operators the default calling convention.
On SPIR targets, the default calling convention is SpirFunction.
However, operator new/delete and builtins were being created with CC_C.
The result is indirect references to new/delete (or builtins that are permitted
to be called indirectly have a mismatched type, as well as questionable codegen
in some cases.
This patch sets both to the default calling convention, so that it
properly matches the calling convention of the target.
Kristof Umann [Mon, 4 Mar 2019 00:28:16 +0000 (00:28 +0000)]
[analyzer] Enable subcheckers to possess checker options
Under the term "subchecker", I mean checkers that do not have a checker class on
their own, like unix.MallocChecker to unix.DynamicMemoryModeling.
Since a checker object was required in order to retrieve checker options,
subcheckers couldn't possess options on their own.
This patch is also an excuse to change the argument order of getChecker*Option,
it always bothered me, now it resembles the actual command line argument
(checkername:option=value).
Michal Gorny [Sun, 3 Mar 2019 10:06:34 +0000 (10:06 +0000)]
[clang] [ToolChains/NetBSD] Support relative libc++ header path
Support locating the libc++ header files relatively to the clang
executable, in addition to the default system path. This is meant
to cover two use cases: running just-built clang from the install
directory, and running installed clang from non-standard location
(e.g. /usr/local).
This is the first step towards ensuring that tests of more LLVM projects
can work out-of-the-box within the build tree, and use the correct set
of headers (rather than e.g. mixing just-built clang+libcxx with system
install of libcxx). It avoids requiring the user to hack around missing
include paths, or LLVM build system to replicate system-specific C++
library defaults in order to append appropriate paths implicitly.
James Y Knight [Sat, 2 Mar 2019 20:22:48 +0000 (20:22 +0000)]
Make the new SanitizerMask code added in r355190 constexpr.
Then, as a consequence, remove the complex set of workarounds for
initialization order -- which are apparently not 100% reliable.
The only downside is that some of the member functions are now
specific to kNumElem == 2, and will need to be updated if that
constant is increased in the future.
Unfortunately, the current code caused an initialization-order runtime
failure for me in some compilation modes. It appears that in a
toolchain without init-array enabled, the order of initialization of
static data members of a template can be reversed w.r.t. the order
within a file.
This caused e.g. SanitizerKind::CFI to be initialized to 0.
I'm not quite sure if that is an allowable ordering variation, or
nonconforming behavior, but in any case, making everything constexpr
eliminates the possibility of such an issue.
Tom Stellard [Sat, 2 Mar 2019 00:50:13 +0000 (00:50 +0000)]
lib/Header: Simplify CMakeLists.txt
Summary:
Replace cut and pasted code with cmake macros and reduce the number of
install commands. This fixes an issue where the headers were being
installed twice.
This clean up should also make future modifications easier, like
adding a cmake option to install header files into a custom resource
directory.
Thomas Lively [Sat, 2 Mar 2019 00:18:09 +0000 (00:18 +0000)]
[WebAssembly] Temporarily disable bulk-memory with -pthread
Summary:
To prevent the instability of bulk-memory in the wasm backend from
blocking separate pthread testing, temporarily remove the logic that
adds -mbulk-memory in the presence of -pthread. Since browsers will
ship bulk memory before or alongside threads, this change will be
reverted as soon as bulk memory has stabilized in the backend.
Paul Robinson [Fri, 1 Mar 2019 20:58:04 +0000 (20:58 +0000)]
[DWARF] Make -g with empty assembler source work better.
This was sometimes causing clang or llvm-mc to crash, and in other
cases could emit a bogus DWARF line-table header. I did an interim
patch in r352541; this patch should be a cleaner and more complete
fix, and retains the test.
Pierre Gousseau [Fri, 1 Mar 2019 10:05:15 +0000 (10:05 +0000)]
[Driver] Allow enum SanitizerOrdinal to represent more than 64 different sanitizer checks, NFC.
enum SanitizerOrdinal has reached maximum capacity, this change extends the capacity to 128 sanitizer checks.
This can eventually allow us to add gcc 8's options "-fsanitize=pointer-substract" and "-fsanitize=pointer-compare".
This is a recommit of r354873 but with a fix for unqualified lookup error in lldb cmake build bot.
Paul Hoad [Fri, 1 Mar 2019 09:09:54 +0000 (09:09 +0000)]
[clang-format] [NFC] clang-format the Format library
Previously revisions commited non-clang-formatted changes to the Format library, this means submitting any revision e.g. {D55170} can cause additional whitespace changes to potentially be included in a revision.
Commit a non functional change using latest build Windows clang-format r351376 with no other changes, to remove these differences
All FormatTests
pass [==========] 652 tests from 20 test cases ran.
CodeGen: Fix PR40605 by splitting constant struct initializers
When emitting initializers for local structures for code built with
-ftrivial-auto-var-init, replace constant structures with sequences of
stores.
This appears to greatly help removing dead initialization stores to those
locals that are later overwritten by other data.
This also removes a lot of .rodata constants (see PR40605), replacing most
of them with immediate values (for Linux kernel the .rodata size is
reduced by ~1.9%)
Akira Hatanaka [Fri, 1 Mar 2019 06:43:20 +0000 (06:43 +0000)]
[Sema][ObjC] Allow silencing -Wobjc-designated-initializers warnings by
declaring an unavailable method in the subclass's extension that
overrides the designated initializer in the base class.
r243676 made changes to allow declaring the unavailable method in the
subclass interface to silence the warning. This commit additionally
allows declaring the unavailable method in the class extension.
[CodeGen] Fix calling llvm.var.annotation outside of a basic block.
When we have an annotated local variable after a function returns, we
generate IR that fails verification with the error
> Instruction referencing instruction not embedded in a basic block!
And it means that bitcast referencing alloca doesn't have a parent basic
block.
Fix by checking if we are at an unreachable point and skip emitting
annotations. This approach is similar to the way we emit variable
initializer and debug info.
[clang-format][TableGen] Don't add spaces around items in square braces.
Summary:
clang-formatting wants to add spaces around items in square braces, e.g. [1, 2] -> [ 1, 2 ]. Based on a quick check [1], it seems like most cases are using the [1, 2] format, so make that the consistent one.
[1] in llvm `.td` files, the regex `\[[^ ]` (bracket followed by not-a-space) shows up ~400 times, but `\[\s[^ ]` (bracket followed by one space and one not-a-space) shows up ~40 times => ~90% uses this format.
Jordan Rupprecht [Thu, 28 Feb 2019 19:16:45 +0000 (19:16 +0000)]
[clang-format][NFC] Allow getLLVMStyle() to take a language
Summary:
getLLVMStyle() sets the default style, but doesn't take the language as a parameter, so can't set default parameters when they differ from C++. This change adds LanguageKind as an input to getLLVMStyle so that we can start doing that.
See D55964 as a motivation for this, where we want Tablegen to be formatted differently than C++.
Thomas Lively [Thu, 28 Feb 2019 18:39:08 +0000 (18:39 +0000)]
[WebAssembly] Remove uses of ThreadModel
Summary:
In the clang UI, replaces -mthread-model posix with -matomics as the
source of truth on threading. In the backend, replaces
-thread-model=posix with the atomics target feature, which is now
collected on the WebAssemblyTargetMachine along with all other used
features. These collected features will also be used to emit the
target features section in the future.
The default configuration for the backend is thread-model=posix and no
atomics, which was previously an invalid configuration. This change
makes the default valid because the thread model is ignored.
A side effect of this change is that objects are never emitted with
passive segments. It will instead be up to the linker to decide
whether sections should be active or passive based on whether atomics
are used in the final link.
Gabor Marton [Thu, 28 Feb 2019 15:24:59 +0000 (15:24 +0000)]
[CTU] Do not allow different CPP dialects in CTU
Summary:
If CPP dialects are different then return with error.
Consider this STL code:
template<typename _Alloc>
struct __alloc_traits
#if __cplusplus >= 201103L
: std::allocator_traits<_Alloc>
#endif
{ // ...
};
This class template would create ODR errors during merging the two units,
since in one translation unit the class template has a base class, however
in the other unit it has none.
Erik Pilkington [Thu, 28 Feb 2019 00:47:55 +0000 (00:47 +0000)]
[CodeGen] Fix some broken IR generated by -fsanitize=unsigned-integer-overflow
I think the author of the function assumed that `GetInsertBlock()`
wouldn't change from where `atomicPHI` was created, but this isn't
true when `-fsanitize=unsigned-integer-overflow` is enabled (we
generate an overflow/continuation label). Fix by keeping track of the
block we want to return to to complete the cmpxchg loop.
Reid Kleckner [Wed, 27 Feb 2019 23:38:44 +0000 (23:38 +0000)]
[InstrProf] Use separate comdat group for data and counters
Summary:
I hadn't realized that instrumentation runs before inlining, so we can't
use the function as the comdat group. Doing so can create relocations
against discarded sections when references to discarded __profc_
variables are inlined into functions outside the function's comdat
group.
In the future, perhaps we should consider standardizing the comdat group
names that ELF and COFF use. It will save object file size, since
__profv_$sym won't appear in the symbol table again.
Jan Korous [Wed, 27 Feb 2019 21:47:40 +0000 (21:47 +0000)]
[clang][index-while-building] FileIndexRecord
Basic data structures for index
Tests are missing from this patch - will be covered properly by tests for the whole feature.
I'm just trying to split into smaller patches to make it easier for reviewers.
Akira Hatanaka [Wed, 27 Feb 2019 18:17:16 +0000 (18:17 +0000)]
Avoid needlessly copying a block to the heap when a block literal
initializes a local auto variable or is assigned to a local auto
variable that is declared in the scope that introduced the block
literal.
Thus the completion code must map the input format of <UIKit/> to
the path of UIKit.framework/Headers as well as strip the
".framework" suffix when auto-completing the framework name.
Balazs Keri [Wed, 27 Feb 2019 16:31:48 +0000 (16:31 +0000)]
[ASTImporter] Improve import of FileID.
Summary:
Even if the content cache has a directory and filename, it may be a virtual file.
The old code returned with error in this case, but it is worth to try to handle
the file as it were a memory buffer.
Yaxun Liu [Wed, 27 Feb 2019 02:02:52 +0000 (02:02 +0000)]
[HIP] change kernel stub name
Add .stub to kernel stub function name so that it is different from kernel
name in device code. This is necessary to let debugger find correct symbol
for kernel.
The current constraint logic is both too lax and too strict. It fails
for input outside the [INT_MIN..INT_MAX] range, but it also implicitly
accepts 0 as value when it should not. Adjust logic to handle both
correctly.
Yaxun Liu [Tue, 26 Feb 2019 22:24:49 +0000 (22:24 +0000)]
[CUDA][HIP] Check calling convention based on function target
MSVC header files using vectorcall to differentiate overloaded functions, which
causes failure for AMDGPU target. This is because clang does not check function
calling convention based on function target.
This patch checks calling convention using the proper target info.
Alexey Bataev [Tue, 26 Feb 2019 21:51:16 +0000 (21:51 +0000)]
[OPENMP][CUDA]Do not emit warnings for variables in late-reported asm
statements.
If the assembler instruction is not generated and the delayed diagnostic
is emitted, we may end up with extra warning message for variables used
in the asm statement. Since the asm statement is not built, the
variables may be left non-referenced and it may produce a warning about
a use of the non-initialized variables.
Reid Kleckner [Tue, 26 Feb 2019 20:42:52 +0000 (20:42 +0000)]
[MS] Don't emit coverage for deleting dtors
Summary:
The MS C++ ABI has no constructor variants, but it has destructor
variants, so we should move the deleting destructor variant check
outside the check for "does the ABI have constructor variants".
Fixes PR37561, so basic code coverage works on Windows with C++.
Tom Roeder [Tue, 26 Feb 2019 19:26:41 +0000 (19:26 +0000)]
[ASTImporter] Add support for importing ChooseExpr AST nodes.
Summary:
This allows ASTs to be merged when they contain ChooseExpr (the GNU
__builtin_choose_expr construction). This is needed, for example, for
cross-CTU analysis of C code that makes use of __builtin_choose_expr.
The node is already supported in the AST, but it didn't have a matcher
in ASTMatchers. So, this change adds the matcher and adds support to
ASTImporter.
This was originally reviewed and approved in
https://reviews.llvm.org/D58292 and submitted as r354832. It was
reverted in r354839 due to failures on the Windows CI builds.
This version fixes the test failures on Windows, which were caused by
differences in template expansion between versions of clang on different
OSes. The version of clang built with MSVC and running on Windows never
expands the template in the C++ test in ImportExpr.ImportChooseExpr in
clang/unittests/AST/ASTImporter.cpp, but the version on Linux does for
the empty arguments and -fms-compatibility.
So, this version of the patch drops the C++ test for
__builtin_choose_expr, since that version was written to catch
regressions of the logic for isConditionTrue() in the AST import code
for ChooseExpr, and those regressions are also caught by
ASTImporterOptionSpecificTestBase.ImportChooseExpr, which does work on
Windows.
1) AMD family 17h "znver2" tune flag (-march, -mcpu).
2) ISAs that are enabled for "znver2" architecture.
3) For the time being, it uses the znver1 scheduler model.
4) Tests are updated.
5) This patch is the clang counterpart to D58343
Yaxun Liu [Tue, 26 Feb 2019 16:20:41 +0000 (16:20 +0000)]
[OpenCL] Fix assertion due to blocks
A recent change caused assertion in CodeGenFunction::EmitBlockCallExpr when a block is called.
There is code
Func = CGM.getOpenCLRuntime().getInvokeFunction(E->getCallee());
getCalleeDecl calls Expr::getReferencedDeclOfCallee, which does not handle
BlockExpr and returns nullptr, which causes isa to assert.
projects / clang / log