Yaxun Liu [Thu, 15 Feb 2018 16:39:19 +0000 (16:39 +0000)]
[OpenCL] Fix __enqueue_block for block with captures
The following test case causes issue with codegen of __enqueue_block
void (^block)(void) = ^{ callee(id, out); };
enqueue_kernel(queue, 0, ndrange, block);
Clang first does codegen for block expression in the first line and deletes its block info.
Clang then tries to do codegen for the same block expression again for the second line,
and fails because the block info is gone.
The fix is to do normal codegen for both lines. Introduce an API to OpenCL runtime to
record llvm block invoke function and llvm block literal emitted for each AST block
expression, and use the recorded information for generating the wrapper kernel.
The EmitBlockLiteral APIs are cleaned up to minimize changes to the normal codegen
of blocks.
Another minor issue is that some clean up AST expression is generated for block
with captures, which can be stripped by IgnoreImplicit.
Aaron Ballman [Thu, 15 Feb 2018 16:20:20 +0000 (16:20 +0000)]
NFC; clean up this file based on our coding standards. The impetus was considerable use of a type name as an identifier for an object.
Changed identifier names (especially function parameters) to not clash with type names and to follow the proper naming conventions. Use of explicit type names changed to use auto where appropriate. Removed unused parameters that should have never been added in the first place. Minor formatting cleanups.
The changes were mostly mechanical and should have no functional impact.
Artem Dergachev [Thu, 15 Feb 2018 03:26:43 +0000 (03:26 +0000)]
[analyzer] Inline constructors for destroyable temporaries.
Since r325210, in cfg-temporary-dtors mode, we can rely on the CFG to tell us
that we're indeed constructing a temporary, so we can trivially construct a
temporary region and inline the constructor.
Much like r325202, this is only done under the off-by-default
cfg-temporary-dtors flag because the temporary destructor, even if available,
will not be inlined and won't have the correct object value (target region).
Unless this is fixed, it is quite unsafe to inline the constructor.
If the temporary is lifetime-extended, the destructor would be an automatic
destructor, which would be evaluated with a "correct" target region - modulo
the series of incorrect relocations performed during the lifetime extension.
It means that at least, values within the object are guaranteed to be properly
escaped or invalidated.
Artem Dergachev [Thu, 15 Feb 2018 03:13:36 +0000 (03:13 +0000)]
[CFG] Provide construction contexts for temproary objects.
Constructors of C++ temporary objects that have destructors now can be queried
to discover that they're indeed constructing temporary objects.
The respective CXXBindTemporaryExpr, which is also repsonsible for destroying
the temporary at the end of full-expression, is now available at the
construction site in the CFG. This is all the context we need to provide for
temporary objects that are not lifetime extended. For lifetime-extended
temporaries, more context is necessary.
Artem Dergachev [Thu, 15 Feb 2018 02:51:58 +0000 (02:51 +0000)]
[analyzer] Decide on inlining destructors via EvalCallOptions.
EvalCallOptions were introduced in r324018 for allowing various parts of
ExprEngine to notify the inlining mechanism, while preparing for evaluating a
function call, of possible difficulties with evaluating the call that they
foresee. Then mayInlineCall() would still be a single place for making the
decision.
Use that mechanism for destructors as well - pass the necessary flags from the
CFG-element-specific destructor handlers.
Part of this patch accidentally leaked into r324018, which led into a change in
tests; this change is reverted now, because even though the change looked
correct, the underlying behavior wasn't. Both of these commits were not intended
to introduce any function changes otherwise.
Artem Dergachev [Thu, 15 Feb 2018 02:32:32 +0000 (02:32 +0000)]
[analyzer] Allow inlining constructors into return values.
This only affects the cfg-temporary-dtors mode - in this mode we begin inlining
constructors that are constructing function return values. These constructors
have a correct construction context since r324952.
Because temporary destructors are not only never inlined, but also don't have
the correct target region yet, this change is not entirely safe. But this
will be fixed in the subsequent commits, while this stays off behind the
cfg-temporary-dtors flag.
Lifetime extension for return values is still not modeled correctly.
[clang-format] Recognize percents as format specifiers in protos
Summary:
Frequently, a percent in protos denotes a formatting specifier for string replacement.
Thus it is desirable to keep the percent together with what follows after it.
Alexey Bataev [Wed, 14 Feb 2018 17:38:47 +0000 (17:38 +0000)]
[OpenMP] Fix trailing space when printing pragmas, by Joel. E. Denny
Summary:
-ast-print prints omp pragmas with a trailing space. While this
behavior is likely of little concern to most users, surely it's
unintentional, and it's annoying for some source-level work I'm
pursuing. This patch focuses on omp pragmas, but it also fixes
init_seg and loop hint pragmas because they share implementation.
The testing strategy here is to add usually just one '{{$}}' per
relevant -ast-print test file. This seems to achieve good code
coverage. However, this strategy is probably easy to forget as the
tests evolve. That's probably fine as this fix is far from critical.
The main goal of the testing is to aid the initial review.
This patch also adds a fixme for "#pragma unroll", which prints as
"#pragma unroll (enable)", which is invalid syntax.
Jonas Hahnfeld [Wed, 14 Feb 2018 16:04:03 +0000 (16:04 +0000)]
[CUDA] Allow external variables in separate compilation
According to the CUDA Programming Guide this is prohibited in
whole program compilation mode. This makes sense because external
references cannot be satisfied in that mode anyway. However,
such variables are allowed in separate compilation mode which
is a valid use case.
Ivan A. Kosarev [Wed, 14 Feb 2018 13:10:35 +0000 (13:10 +0000)]
[AST] Refine the condition for element-dependent array fillers
This patch fixes clang to not consider braced initializers for
aggregate elements of arrays to be potentially dependent on the
indices of the initialized elements. Resolves bug 18978:
initialize a large static array = clang oom?
https://bugs.llvm.org/show_bug.cgi?id=18978
Mikhail Maltsev [Wed, 14 Feb 2018 11:34:25 +0000 (11:34 +0000)]
[Sema] Fix decltype of static data members
Summary:
According to the C++11 standard [dcl.type.simple]p4:
The type denoted by decltype(e) is defined as follows:
- if e is an unparenthesized id-expression or an unparenthesized
class member access (5.2.5), decltype(e) is the type of the entity
named by e.
Currently Clang handles the 'member access' case incorrectly for
static data members (decltype returns T& instead of T). This patch
fixes the issue.
Erich Keane [Wed, 14 Feb 2018 00:14:07 +0000 (00:14 +0000)]
Implement function attribute artificial
Added support in clang for GCC function attribute 'artificial'. This attribute
is used to control stepping behavior of debugger with respect to inline
functions.
Artem Dergachev [Mon, 12 Feb 2018 22:36:36 +0000 (22:36 +0000)]
[CFG] Provide construction contexts for return value constructors.
When the current function returns a C++ object by value, CFG elements for
constructors that construct the return values can now be queried to discover
that they're indeed participating in construction of the respective return value
at the respective return statement.
John Baldwin [Mon, 12 Feb 2018 22:22:01 +0000 (22:22 +0000)]
Look for 32-bit libraries in /usr/lib32 for MIPS O32 on FreeBSD.
Summary:
FreeBSD N64 MIPS systems can include 32-bit libraries for O32 in
/usr/lib32 similar to the 32-bit compatibility libraries provided
for FreeBSD/amd64 and FreeBSD/powerpc64.
Fangrui Song [Mon, 12 Feb 2018 17:42:09 +0000 (17:42 +0000)]
[libclang] Add `CXSymbolRole role` to CXIdxEntityRefInfo
Summary:
CXIdxEntityRefInfo contains the member `CXIdxEntityRefKind kind;` to
differentiate implicit and direct calls. However, there are more roles
defined in SymbolRole. Among them, `Read/Write` are probably the most
useful ones as they can be used to differentiate Read/Write occurrences
of a symbol for document highlight in a text document.
See `export namespace DocumentHighlightKind`
on https://microsoft.github.io/language-server-protocol/specification
Reid Kleckner [Mon, 12 Feb 2018 17:37:06 +0000 (17:37 +0000)]
[Sema] Don't mark plain MS enums as fixed
Summary:
This fixes a flaw in our AST: PR27098
MSVC always gives plain enums the underlying type 'int'. Clang does this
as well, but we claim the enum is "fixed", as if the user actually wrote
': int'. It means we end up emitting spurious -Wsign-compare warnings on
code like this:
Erich Keane [Mon, 12 Feb 2018 17:01:41 +0000 (17:01 +0000)]
Make attribute-target on a Definition-after-use update the LLVM attributes
As reported here: https://bugs.llvm.org/show_bug.cgi?id=36301
The issue is that the 'use' causes the plain declaration to emit
the attributes to LLVM-IR. However, if the definition added it
later, these would silently disappear.
This commit extracts that logic to its own function in CodeGenModule,
and has the attribute-applications done during 'definition' update
the attributes properly.
[clang-format] Fix comment indentation in text protos
Summary: This patch fixes a bug where the comment indent of comments in text protos gets messed up because by default paren states get created with AlignColons = true (which makes snese for ObjC).
Aaron Ballman [Mon, 12 Feb 2018 13:38:25 +0000 (13:38 +0000)]
Allow the NS, CF, and ObjC attributes to be used with -fdouble-square-bracket-attributes. The syntactic locations for such attributes on ObjC constructs have been specifically chosen to follow the GNU attribute syntactic locations.
ASan+operator new[]: Add an option for more thorough operator new[] cookie poisoning
Summary:
Right now clang is skipping array cookie poisoning for any operator
new[] which is not part of the set of replaceable global allocation
functions.
This commit adds a flag to tell clang to poison all operator new[]
cookies.
A previous review was poisoning all array cookies unconditionally, but
there is an edge case which would stop working under ASan (a custom
operator new[] saves whatever pointer it returned, and then accesses
it).
This newer revision adds a command line argument to toggle this feature.
Original revision: https://reviews.llvm.org/D41301
Compiler-rt test revision with an explanation of the edge case: https://reviews.llvm.org/D41664
Jonas Hahnfeld [Mon, 12 Feb 2018 10:46:45 +0000 (10:46 +0000)]
[CUDA] Add option to generate relocatable device code
As a first step, pass '-c/--compile-only' to ptxas so that it
doesn't complain about references to external function. This
will successfully generate object files, but they won't work
at runtime because the registration routines need to adapted.
Aaron Smith [Sat, 10 Feb 2018 21:28:55 +0000 (21:28 +0000)]
Fix test clang-diff-json.cpp
Summary:
This test would fail if the python path had spaces. Add a quote around the path to fix this problem and update some test values changed by the addition of quotes around the path.
Gabor Horvath [Sat, 10 Feb 2018 14:04:45 +0000 (14:04 +0000)]
[Templight] Template Instantiation Observer
This patch adds a base-class called TemplateInstantiationObserver which gets
notified whenever a template instantiation is entered or exited during
semantic analysis. This is a base class used to implement the template
profiling and debugging tool called
Templight (https://github.com/mikael-s-persson/templight).
The patch also makes a few more changes:
* ActiveTemplateInstantiation class is moved out of the Sema class (so it can be used with inclusion of Sema.h).
* CreateFrontendAction function in front-end utilities is given external linkage (not longer a hidden static function).
* TemplateInstObserverChain data member added to Sema class to hold the list of template-inst observers.
* Notifications to the template-inst observer are added at the key places where templates are instantiated.
Artem Dergachev [Sat, 10 Feb 2018 03:14:22 +0000 (03:14 +0000)]
[analyzer] NFC: Assert that our fix for noreturn destructors keeps working.
Massive false positives were known to be caused by continuing the analysis
after a destructor with a noreturn attribute has been executed in the program
but not modeled in the analyzer due to being missing in the CFG.
Now that work is being done on enabling the modeling of temporary constructors
and destructors in the CFG, we need to make sure that the heuristic that
suppresses these false positives keeps working when such modeling is disabled.
In particular, different code paths open up when the corresponding constructor
is being inlined during analysis.
Artem Dergachev [Sat, 10 Feb 2018 03:04:59 +0000 (03:04 +0000)]
[analyzer] Fix a merge error in -analyzer-config tests.
It was introduced when two -analyzer-config options were added almost
simultaneously in r324793 and r324668 and the option count was not
rebased correctly in the tests.
Artem Dergachev [Sat, 10 Feb 2018 02:55:08 +0000 (02:55 +0000)]
[analyzer] NFC: Use CFG construction contexts instead of homemade lookahead.
The analyzer was relying on peeking the next CFG element during analysis
whenever it was trying to figure out what object is being constructed
by a given constructor. This information is now available in the current CFG
element in all cases that were previously supported by the analyzer,
so no complicated lookahead is necessary anymore.
No functional change intended - the context in the CFG should for now be
available if and only if it was previously discoverable via CFG lookahead.
Artem Dergachev [Sat, 10 Feb 2018 02:46:14 +0000 (02:46 +0000)]
[CFG] Provide construction contexts when constructors have cleanups.
Now that we make it possible to query the CFG constructor element to find
information about the construction site, possible cleanup work represented by
ExprWithCleanups should not prevent us from providing this information.
This allows us to have a correct construction context for variables initialized
"by value" via elidable copy-constructors, such as 'i' in
Artem Dergachev [Sat, 10 Feb 2018 02:18:04 +0000 (02:18 +0000)]
[CFG] Add construction context for constructor initializers.
CFG elements for constructors of fields and base classes that are being
initialized before the body of the whole-class constructor starts can now be
queried to discover that they're indeed participating in initialization of their
respective fields or bases before the whole-class constructor kicks in.
CFG construction contexts are now capable of representing CXXCtorInitializer
triggers, which aren't considered to be statements in the Clang AST.
Artem Dergachev [Sat, 10 Feb 2018 00:55:49 +0000 (00:55 +0000)]
[analyzer] Add missing pre-post-statement callbacks for OffsetOfExpr.
This expression may or may not be evaluated in compile time, so tracking the
result symbol is of potential interest. However, run-time offsetof is not yet
supported by the analyzer, so for now this callback is only there to assist
future implementation.
[NFC] Extract method to SourceManager for traversing the macro "stack"
The code for going up the macro arg expansion is duplicated in many
places (and we need it for the analyzer as well, so I did not want to
duplicate it two more times).
This patch is an NFC, so the semantics should remain the same.
Matt Davis [Fri, 9 Feb 2018 22:10:09 +0000 (22:10 +0000)]
[CodeGen] Use the zero initializer instead of storing an all zero representation.
Summary:
This change avoids the overhead of storing, and later crawling,
an initializer list of all zeros for arrays. When LLVM
visits this (llvm/IR/Constants.cpp) ConstantArray::getImpl()
it will scan the list looking for an array of all zero.
We can avoid the store, and short-cut the scan, by detecting
all zeros when clang builds-up the initialization representation.
This was brought to my attention when investigating PR36030
Adrian Prantl [Fri, 9 Feb 2018 18:43:10 +0000 (18:43 +0000)]
Introduce an API for LLDB to compute the default module cache path
LLDB creates Clang modules and had an incomplete copy of the clang
Driver code that compute the -fmodule-cache-path. This patch makes the
clang driver code accessible to LLDB.
This happens because the colon is placed at `ContinuationIndent +
LongestObjCSelectorName`, so that any selector can be wrapped. This is
however not needed in case the longest selector is the firstone, and
not wrapped.
To overcome this, this patch does not include the first selector in
`LongestObjCSelectorName` computation (in TokenAnnotator), and lets
`ContinuationIndenter` decide how to account for the first selector
when wrapping. (Note this was already partly the case, see line 521
of ContinuationIndenter.cpp)
This way, the code gets properly aligned whenever possible without
breaking the continuation indent.
Adjust the ObjC protocol conformance workaround to be more extensible.
Use a synthetic type for the protocol (`struct Protocol`). Embed this
within a reserved namespace to permit extending the extended pointer
type qualifiers similarly for ObjC lifetime qualifiers.
Introduce additional special handling for `__autoreleasing`, `__strong`,
and `__weak` Objective C lifetime qualifiers. We decorate these by
creating an artificial template type `Autoreleasing`, `Strong`, or
`Weak` in the `__ObjC` namespace. These are only considered in the
template type specialization and not the function parameter.
Richard Smith [Fri, 9 Feb 2018 01:15:13 +0000 (01:15 +0000)]
[modules] Fix incorrect diagnostic mapping computation when a module changes
diagnostic settings using _Pragma within a macro.
The AST writer had previously been assuming that all diagnostic state
transitions would occur within a FileID corresponding to a file. When a
diagnostic state change occured within a macro, it was unable to form a
location for that state change and would instead corrupt the diagnostic state
of the "root" node (and thus that of the main compilation).
Also introduce a "#pragma clang __debug diag_mapping" debugging utility
that I added to track this issue down.
Erich Keane [Thu, 8 Feb 2018 23:49:40 +0000 (23:49 +0000)]
Add size to constexpr Arrays
What seems to be a bug in older versions of MSVC, constexpr
member arrays with a redefinition (to force emission) require
their initial definition to have the size between the brackets.
Artem Dergachev [Thu, 8 Feb 2018 23:28:29 +0000 (23:28 +0000)]
[analyzer] MallocChecker: Fix one more bug category.
Even though most of the inconsistencies in MallocChecker's bug categories were
fixed in r302016, one more was introduced in r301913 which was later missed.
projects / clang / log