Alexey Bader [Tue, 19 Feb 2019 15:19:06 +0000 (15:19 +0000)]
[OpenCL] Change type of block pointer for OpenCL
Summary:
For some reason OpenCL blocks in LLVM IR are represented as function pointers.
These pointers do not point to any real function and never get called. Actually
they point to some structure, which in turn contains pointer to the real block
invoke function.
This patch changes represntation of OpenCL blocks in LLVM IR from function
pointers to pointers to `%struct.__block_literal_generic`.
Such representation allows to avoid unnecessary bitcasts and simplifies
further processing (e.g. translation to SPIR-V ) of the module for targets
which do not support function pointers.
Gabor Marton [Mon, 18 Feb 2019 11:09:56 +0000 (11:09 +0000)]
[ASTImporter] Unify redecl chain tests as type parameterized tests
Summary:
This patch unifies all those tests which check the correctness of the
redecl chains. Previously we had several structurally very similar test
cases for each language construct (class, function, variable, function
template, ...).
We still use value-parameterized tests for the different AST
compatibility switches (-fdelayed-template-parsing, -fms-compatibility).
Gtest makes it possible to have either value-parameterized or
type-parameterized fixtures. However, we cannot have both value- and
type-parameterized test fixtures. So we use a value-parameterized test
fixture in the gtest sense. We intend to mimic gtest's type-parameters
via the type template parameter. We manually instantiate the different
tests with the each types.
After this patch I am planning to put the "generic redecl chain" related
tests into their own separate test file (in another patch).
Bruno Ricci [Sun, 17 Feb 2019 18:50:51 +0000 (18:50 +0000)]
Recommit "[AST] Factor out the logic of the various Expr::Ignore*"
(Originally commited in r354215 and reverted in r354216 because of a
missed failing clang-tidy test (fix in r354228))
Now that the implementation of all of the Expr::Ignore* is in Expr.cpp
we can try to remove some duplication. Do this by separating the logic
of the Expr::Ignore* from the iterative loop.
This is NFC, except for one change: IgnoreParenImpCasts now skips,
among other things, everything that IgnoreImpCasts skips. This means
FullExpr are now skipped by IgnoreParenImpCasts. This was likely an
oversight when FullExpr was added to the nodes skipped by IgnoreImpCasts.
Bruno Ricci [Sun, 17 Feb 2019 13:32:39 +0000 (13:32 +0000)]
[AST] Factor out the logic of the various Expr::Ignore*
Now that the implementation of all of the Expr::Ignore* is in Expr.cpp
we can try to remove some duplication. Do this by separating the logic of
the Expr::Ignore* from the iterative loop.
This is NFC, except for one change: IgnoreParenImpCasts now skips, among
other things, everything that IgnoreImpCasts skips. This means FullExpr
are now skipped by IgnoreParenImpCasts. This was likely an oversight when
FullExpr was added to the nodes skipped by IgnoreImpCasts.
Petr Hosek [Sun, 17 Feb 2019 01:02:40 +0000 (01:02 +0000)]
[Driver][Fuchsia] Support -nolibc flag
This can be used to disable libc linking. This flag is supported by
GCC since version 9 as well as some Clang target toolchains. This
change also includes tests for all -no* flags which previously weren't
covered.
David Green [Sat, 16 Feb 2019 11:19:04 +0000 (11:19 +0000)]
Move multiline raw string literal out of macro. NFC
Certain combinations of gcc and ccache fail when the raw
string literal is preprocessed. This just moves the string
out as is done elsewhere in the same file.
Erik Pilkington [Sat, 16 Feb 2019 01:11:47 +0000 (01:11 +0000)]
[Sema] Diagnose floating point conversions based on target semantics
...instead of just comparing rank. Also, fix a bad warning about
_Float16, since its declared out of order in BuiltinTypes.def,
meaning comparing rank using BuiltinType::getKind() is incorrect.
Volodymyr Sapsai [Sat, 16 Feb 2019 01:01:08 +0000 (01:01 +0000)]
[ObjC generics] Fix applying `__kindof` to the type parameter.
Fixes the warning about incompatible pointer types on assigning to a
subclass of type argument an expression of type `__kindof TypeParam`.
We already have a mechanism in `ASTContext::canAssignObjCInterfaces`
that handles `ObjCObjectType` with `__kindof`. But it wasn't triggered
because during type substitution `__kindof TypeParam` was represented as
`AttributedType` with attribute `ObjCKindOf` and equivalent type
`TypeArg`. For assignment type checking we use canonical types so
attributed type was desugared and the attribute was ignored.
The fix is in checking transformed `AttributedType` and pushing
`__kindof` down into `ObjCObjectType` when necessary.
clang-format with UseTab: Always sometimes doesn't insert the right amount of tabs.
Trailing comments are not always aligned properly when UseTab is set to Always.
Consider:
int a; // x
int bbbbbbbb; // x
With .clang-format:
---
Language: Cpp
BasedOnStyle: LLVM
UseTab: Always
...
The trailing comments of this code block should be aligned, but aren't
To align the first trailing comment it needs to insert 8 spaces. This should be
one tab plus six spaces. It skips the logic of the first partial tab in
FirstTabWidth (=2) + Style.TabWidth (=8) <= Spaces (=8) and only inserts one
tab. Proposed fix and test is attached.
Volodymyr Sapsai [Fri, 15 Feb 2019 22:14:58 +0000 (22:14 +0000)]
[ObjC] For type substitution in generics use a regular recursive type visitor.
Switch to the inheritance-based visitor from the lambda-based visitor to
allow both preorder and postorder customizations during type
transformation. NFC intended.
Richard Smith [Fri, 15 Feb 2019 21:53:07 +0000 (21:53 +0000)]
Fix implementation of [temp.local]p4.
When a template-name is looked up, we need to give injected-class-name
declarations of class templates special treatment, as they denote a
template rather than a type.
Previously we achieved this by applying a filter to the lookup results
after completing name lookup, but that is incorrect in various ways, not
least of which is that it lost all information about access and how
members were named, and the filtering caused us to generally lose
all ambiguity errors between templates and non-templates.
We now preserve the lookup results exactly, and the few places that need
to map from a declaration found by name lookup into a declaration of a
template do so explicitly. Deduplication of repeated lookup results of
the same injected-class-name declaration is done by name lookup instead
of after the fact.
This reinstates r354091, which was previously reverted in r354097
because it exposed bugs in lldb and compiler-rt. Those bugs were fixed
in r354173 and r354174 respectively.
Dan Albert [Fri, 15 Feb 2019 20:31:54 +0000 (20:31 +0000)]
[Driver] Default all Android ARM targets to NEON.
Summary:
There are an insignificant number of ARM Android devices that don't
support NEON. Default to using NEON since that will improve
performance on the majority of devices. Users that need to target
non-NEON devices can still explicitly disable NEON.
Reid Kleckner [Fri, 15 Feb 2019 19:59:45 +0000 (19:59 +0000)]
[MSVC] Recognize `static_assert` keyword in C and C++98
Summary:
The main effect is that clang now accepts the following conforming C11
code with MSVC headers:
#include <assert.h>
static_assert(1, "true");
This is a non-conforming extension (the keyword is outside the
implementer's namespace), so it is placed under -fms-compatibility
instead of -fms-extensions like most MSVC-specific keyword extensions.
Normally, in C11, the compiler is supposed to provide the _Static_assert
keyword, and assert.h should define static_assert to _Static_assert.
However, that is not what MSVC does, and MSVC doesn't even provide
_Static_assert.
This also has the less important side effect of enabling static_assert
in C++98 mode with -fms-compatibility. It's exceptionally difficult to
use modern MSVC headers without C++14 even, so this is relatively
unimportant.
Bruno Ricci [Fri, 15 Feb 2019 18:12:58 +0000 (18:12 +0000)]
[Sema][NFC] SequenceChecker: Add tests for references/members, and prepare for the C++17 tests
Add some tests for unsequenced operations with members and references.
For now most of it is unhandled but it shows what work needs to be done.
Also merge the tests for the C++17 sequencing rules in warn-unsequenced.cpp
since we want to make sure that the appropriate warnings are still present
in C++17 without duplicating the whole content of warn-unsequenced.cpp.
Shoaib Meenai [Fri, 15 Feb 2019 15:59:04 +0000 (15:59 +0000)]
[clang] Add build and install targets for clang libraries
This is modeled after the existing llvm-libraries target. It's a
convenient way to include all clang libraries in a distribution.
This differs slightly from the llvm-libraries target in that it adds any
library added via add_clang_library, whereas llvm-libraries only
includes targets added via add_llvm_library that didn't use the MODULE
or BUILDTREE_ONLY arguments. add_clang_library doesn't appear to have
any equivalents of those arguments, so the conditions don't apply.
Shoaib Meenai [Fri, 15 Feb 2019 15:59:02 +0000 (15:59 +0000)]
[clang] Create install targets for non-shared libraries
I don't see a reason for these to not have install targets created,
which in turn allows them to be bundled in distributions. This doesn't
affect the "install" target, since that just runs all CMake install
rules (and we were already creating install rules for these).
Sam McCall [Fri, 15 Feb 2019 07:16:11 +0000 (07:16 +0000)]
[Analysis] -Wunreachable-code shouldn't fire on the increment of a foreach loop
Summary:
The idea is that the code here isn't written, so doesn't indicate a bug.
Similar to code expanded from macros.
This means the warning no longer fires on this code:
for (auto C : collection) {
process(C);
return;
}
handleEmptyCollection();
Unclear whether this is more often a bug or not in practice, I think it's a
reasonable idiom in some cases.
Either way, if we want to warn on "loop that doesn't loop", I think it should be
a separate warning, and catch `while(1) break;`
Richard Smith [Fri, 15 Feb 2019 00:29:04 +0000 (00:29 +0000)]
Fix implementation of [temp.local]p4.
When a template-name is looked up, we need to give injected-class-name
declarations of class templates special treatment, as they denote a
template rather than a type.
Previously we achieved this by applying a filter to the lookup results
after completing name lookup, but that is incorrect in various ways, not
least of which is that it lost all information about access and how
members were named, and the filtering caused us to generally lose
all ambiguity errors between templates and non-templates.
We now preserve the lookup results exactly, and the few places that need
to map from a declaration found by name lookup into a declaration of a
template do so explicitly. Deduplication of repeated lookup results of
the same injected-class-name declaration is done by name lookup instead
of after the fact.
Richard Smith [Fri, 15 Feb 2019 00:27:53 +0000 (00:27 +0000)]
PR40642: Fix determination of whether the final statement of a statement
expression is a discarded-value expression.
Summary:
We used to get this wrong in three ways:
1) During parsing, an expression-statement followed by the }) ending a
statement expression was always treated as producing the value of the
statement expression. That's wrong for ({ if (1) expr; })
2) During template instantiation, various kinds of statement (most
statements not appearing directly in a compound-statement) were not
treated as discarded-value expressions, resulting in missing volatile
loads (etc).
3) In all contexts, an expression-statement with attributes was not
treated as producing the value of the statement expression, eg
({ [[attr]] expr; }).
Also fix incorrect enforcement of OpenMP rule that directives can "only
be placed in the program at a position where ignoring or deleting the
directive would result in a program with correct syntax". In particular,
a label (be it goto, case, or default) should not affect whether
directives are permitted.
Jan Korous [Thu, 14 Feb 2019 23:02:35 +0000 (23:02 +0000)]
[clang][FileManager] fillRealPathName even if we aren't opening the file
The pathname wasn't previously filled when the getFile() method was called with openFile = false.
We are caching FileEntry-s in ParsedAST::Includes in clangd and this caused the problem.
This fixes an internal test failure in clangd - ClangdTests.GoToInclude.All
Erik Pilkington [Thu, 14 Feb 2019 22:48:01 +0000 (22:48 +0000)]
[Sema] Fix-up a -Wfloat-conversion diagnostic
We were warning on valid ObjC property reference exprs, and passing
in the wrong arguments to DiagnoseFloatingImpCast (leading to a badly
worded diagnostic).
Bruno Ricci [Thu, 14 Feb 2019 15:43:17 +0000 (15:43 +0000)]
[Sema] Fix a regression introduced in "[AST][Sema] Remove CallExpr::setNumArgs"
D54902 removed CallExpr::setNumArgs in preparation of tail-allocating the
arguments of CallExpr. It did this by allocating storage for
max(number of arguments, number of parameters in the prototype). The
temporarily nulled arguments however causes issues in BuildResolvedCallExpr
when typo correction is done just after the creation of the call expression.
This was unfortunately missed by the tests /:
To fix this, delay setting the number of arguments to
max(number of arguments, number of parameters in the prototype) until we are
ready for it. It would be nice to have this encapsulated in CallExpr but this
is the best I can come up with under the constraint that we cannot add
anything the CallExpr.
Gabor Marton [Thu, 14 Feb 2019 13:07:03 +0000 (13:07 +0000)]
[ASTImporter] Check visibility/linkage of functions and variables
Summary:
During import of a global variable with external visibility the lookup
will find variables (with the same name) but with static visibility.
Clearly, we cannot put them into the same redecl chain. The same is
true in case of functions. In this fix we filter the lookup results and
consider only those which have the same visibility as the decl we
currently import.
We consider two decls in two anonymous namsepaces to have the same
visibility only if they are imported from the very same translation
unit.
Yaxun Liu [Thu, 14 Feb 2019 02:00:09 +0000 (02:00 +0000)]
[CUDA][HIP] Use device side kernel and variable names when registering them
__hipRegisterFunction and __hipRegisterVar need to accept device side kernel and variable names
so that HIP runtime can associate kernel stub functions in host code with kernel symbols in fat binaries,
and associate shadow variables in host code with device variables in fat binaries.
Currently, clang assumes kernel functions and device variables have the same name as the kernel
stub functions and shadow variables. However, when host is compiled in windows with MSVC C++
ABI and device is compiled with Itanium C++ ABI (e.g. AMDGPU), kernels and device symbols in fat
binary are mangled differently than host.
This patch gets the device side kernel and variable name by mangling them in the mangle context
of aux target.
Douglas Yung [Thu, 14 Feb 2019 01:11:32 +0000 (01:11 +0000)]
Relax restriction in tests to where "-emit-llvm-bc" and "-emit-obj" must appear.
The CHECK lines as structured were requiring them to appear only in a certain
position while all that is really needed is to check that they are present.
Leonard Chan [Wed, 13 Feb 2019 22:22:48 +0000 (22:22 +0000)]
[NewPM] Second attempt at porting ASan
This is the second attempt to port ASan to new PM after D52739. This takes the
initialization requried by ASan from the Module by moving it into a separate
class with it's own analysis that the new PM ASan can use.
Changes:
- Split AddressSanitizer into 2 passes: 1 for the instrumentation on the
function, and 1 for the pass itself which creates an instance of the first
during it's run. The same is done for AddressSanitizerModule.
- Add new PM AddressSanitizer and AddressSanitizerModule.
- Add legacy and new PM analyses for reading data needed to initialize ASan with.
- Removed DominatorTree dependency from ASan since it was unused.
- Move GlobalsMetadata and ShadowMapping out of anonymous namespace since the
new PM analysis holds these 2 classes and will need to expose them.
Robert Widmann [Wed, 13 Feb 2019 22:22:23 +0000 (22:22 +0000)]
Restore Check for Unreachable Exit Block in -Winfinite-recursion
Summary:
When this was rewritten in D43737, the logic changed to better explore infinite loops. The check for a reachable exit block was deleted which accidentally introduced false positives in case the exit node was unreachable.
We were testing for cases like this, but @steven_wu provided an additional test case that I've included in the regression tests for this patch.
Erik Pilkington [Wed, 13 Feb 2019 20:32:37 +0000 (20:32 +0000)]
[Sema] Delay checking whether objc_designated_initializer is being applied to an init method
This fixes a regression that was caused by r335084, which reversed
the order that attributes are applied. objc_method_family can change
whether a method is an init method, so the order that these
attributes are applied matters. The commit fixes this by delaying the
init check until after all attributes have been applied.
Hubert Tong [Wed, 13 Feb 2019 20:17:13 +0000 (20:17 +0000)]
[PowerPC] Stop defining _ARCH_PWR6X on POWER7 and up
Summary:
The predefined macro `_ARCH_PWR6X` is associated with GCC's
`-mcpu=power6x` option, which enables generation of P6 "raw mode"
instructions such as `mftgpr`.
Later POWER processors build upon the "architected mode", not the raw
one. `_ARCH_PWR6X` should not be defined for these later processors.
Nico Weber [Wed, 13 Feb 2019 19:04:26 +0000 (19:04 +0000)]
Re-enable the test disabled in r353836 and hopefully make it pass in gcc builds
Argument evaluation order is different between gcc and clang, so pull out
the Builder calls to make the generated IR independent of the host compiler's
argument evaluation order. Thanks to rnk for reminding me of this clang/gcc
difference.
Simon Atanasyan [Wed, 13 Feb 2019 18:27:09 +0000 (18:27 +0000)]
[Headers][mips] Add `__attribute__((__mode__(__unwind_word__)))` to the _Unwind_Word / _Unwind_SWord definitions
The rationale of this change is to fix _Unwind_Word / _Unwind_SWord
definitions for MIPS N32 ABI. This ABI uses 32-bit pointers,
but _Unwind_Word and _Unwind_SWord types are eight bytes long.
# The __attribute__((__mode__(__unwind_word__))) is added to the type
definitions. It makes them equal to the corresponding definitions used
by GCC and allows to override types using `getUnwindWordWidth` function.
# The `getUnwindWordWidth` virtual function override in the `MipsTargetInfo`
class and provides correct type size values.
Petr Hosek [Wed, 13 Feb 2019 17:28:47 +0000 (17:28 +0000)]
[AArch64] Support reserving arbitrary general purpose registers
This is a follow up to D48580 and D48581 which allows reserving
arbitrary general purpose registers with the exception of registers
with special purpose (X8, X16-X18, X29, X30) and registers used by LLVM
(X0, X19). This change also generalizes some of the existing logic to
rely entirely on values generated from tablegen.
Aaron Enye Shi [Wed, 13 Feb 2019 16:12:16 +0000 (16:12 +0000)]
[HIP] Handle compile -m options and propagate into LLC
Allow the compile options for -m such as -mxnack/-mno-xnack, -msram-ecc/-mno-sram-ecc, -mcode-object-v3/-mno-code-object-v3 to propagate into LLC args. Fix an issue where -mattr was pushed even when it was empty.
Also add lit tests to verify features are properly passed.
Adam Balogh [Wed, 13 Feb 2019 12:25:47 +0000 (12:25 +0000)]
[Analyzer] Crash fix for FindLastStoreBRVisitor
FindLastStoreBRVisitor tries to find the first node in the exploded graph where
the current value was assigned to a region. This node is called the "store
site". It is identified by a pair of Pred and Succ nodes where Succ already has
the binding for the value while Pred does not have it. However the visitor
mistakenly identifies a node pair as the store site where the value is a
`LazyCompoundVal` and `Pred` does not have a store yet but `Succ` has it. In
this case the `LazyCompoundVal` is different in the `Pred` node because it also
contains the store which is different in the two nodes. This error may lead to
crashes (a declaration is cast to a parameter declaration without check) or
misleading bug path notes.
In this patch we fix this problem by checking for unequal `LazyCompoundVals`: if
their region is equal, and their store is the same as the store of their nodes
we consider them as equal when looking for the "store site". This is an
approximation because we do not check for differences of the subvalues
(structure members or array elements) in the stores.
- Remove most of the discussion of the x86_64 implementation;
link to an older version of the documentation for details of
that implementation.
- Add description of the compatibility and security issues discovered
during the development of the aarch64 implementation for Android.
Craig Topper [Tue, 12 Feb 2019 22:12:19 +0000 (22:12 +0000)]
[X86] Follow up to r353878, add MSVC compatibility command lines to other intrinsic tests that uses packed structs to control alignment.
r353878 fixed a bug in _mm_loadu_ps and added a command line to catch it. Adding additional command lines to prevent breaking other intrinsics in the future.
Make ModuleDependencyCollector's method virtual (NFC)
For reproducers in LLDB we want to hook up into the existing clang
infrastructure. To make that happen we need to be able to override the
ModuleDependencyCollector's methods.
The alternative was to inherit from the DependencyCollector directly,
but that would mean re-implementing the ModuleDependencyListener and the
ModuleDependencyPPCallbacks and ModuleDependencyMMCallbacks.
Aaron Enye Shi [Tue, 12 Feb 2019 21:21:06 +0000 (21:21 +0000)]
[HIP] Handle compile -m options and propagate into LLC
Allow the compile options for -m such as -mxnack/-mno-xnack, -msram-ecc/-mno-sram-ecc, -mcode-object-v3/-mno-code-object-v3 to propagate into LLC args.
Also add lit tests to verify features are properly passed.
Reid Kleckner [Tue, 12 Feb 2019 21:04:21 +0000 (21:04 +0000)]
[X86] Use __m128_u for _mm_loadu_ps after r353555
Add secondary triple to existing SSE test for it. I audited other uses
of __attribute__((__packed__)) in the intrinsic headers, and this seemed
to be the only missing one.
Erik Pilkington [Mon, 11 Feb 2019 23:21:39 +0000 (23:21 +0000)]
Add a new attribute, fortify_stdlib
This attribute applies to declarations of C stdlib functions
(sprintf, memcpy...) that have known fortified variants
(__sprintf_chk, __memcpy_chk, ...). When applied, clang will emit
calls to the fortified variant functions instead of calls to the
defaults.
In GCC, this is done by adding gnu_inline-style wrapper functions,
but that doesn't work for us for variadic functions because we don't
support __builtin_va_arg_pack (and have no intention to).
This attribute takes two arguments, the first is 'type' argument
passed through to __builtin_object_size, and the second is a flag
argument that gets passed through to the variadic checking variants.
Eli Friedman [Mon, 11 Feb 2019 22:54:27 +0000 (22:54 +0000)]
[Sema] Mark GNU compound literal array init as an rvalue.
Basically the same issue as string init, except it didn't really have
any visible consequences before I removed the implicit lvalue-to-rvalue
conversion from CodeGen.
While I'm here, a couple minor drive-by cleanups: IgnoreParens never
returns a ConstantExpr, and there was a potential crash with string init
involving a ChooseExpr.
The analyzer test change maybe indicates we could simplify the analyzer
code a little with this fix? Apparently a hack was added to support
lvalues in initializers in r315750, but I'm not really familiar with the
relevant code.
Fixes regression reported in the kernel build at
https://bugs.llvm.org/show_bug.cgi?id=40430#c6 .
Heejin Ahn [Mon, 11 Feb 2019 22:47:50 +0000 (22:47 +0000)]
[WebAssembly] Make thread-related options consistent
Summary:
There have been three options related to threads and users had to set
all three of them separately to get the correct compilation results.
This makes sure the relationship between the options makes sense and
sets necessary options for users if only part of the necessary options
are specified. This does:
- Remove `-matomics`; this option alone does not enable anything, so
removed it to not confuse users.
- `-mthread-model posix` sets `-target-feature +atomics`
- `-pthread` sets both `-target-feature +atomics` and
`-mthread-model posix`
Also errors out when explicitly given options don't match, such as
`-pthread` is given with `-mthread-model single`.
Petr Hosek [Mon, 11 Feb 2019 20:13:42 +0000 (20:13 +0000)]
[CodeGen] Set construction vtable visibility after creating initializer
We must only set the construction vtable visibility after we create the
vtable initializer, otherwise the global value will be treated as
declaration rather than definition and the visibility won't be set.
Tom Tan [Mon, 11 Feb 2019 20:04:02 +0000 (20:04 +0000)]
[COFF, ARM64] Remove definitions for _byteswap library functions
_byteswap_* functions are are implemented in below file as normal function
from libucrt.lib and declared in stdlib.h. Define them in intrin.h triggers
lld error "conflicting comdat type" and "duplicate symbols" which was just
added to LLD (https://reviews.llvm.org/D57324).
projects / clang / log