Benjamin Kramer [Fri, 23 Oct 2015 09:04:55 +0000 (09:04 +0000)]
[AST] Re-add TypeLocs and NestedNameSpecifierLocs to the ParentMap.
This relands r250831 after some fixes to shrink the ParentMap overall
with one addtional tweak: nodes with pointer identity (e.g. Decl* and
friends) can be store more efficiently so I put them in a separate map.
All other nodes (so far only TypeLoc and NNSLoc) go in a different map
keyed on DynTypedNode. This further uglifies the code but significantly
reduces memory overhead.
Overall this change still make ParentMap significantly larger but it's
nowhere as bad as before. I see about 25 MB over baseline (pre-r251008)
on X86ISelLowering.cpp. If this becomes an issue we could consider
splitting the maps further as DynTypedNode is still larger (32 bytes)
than a single TypeLoc (16 bytes) but I didn't want to introduce even
more complexity now.
John McCall [Thu, 22 Oct 2015 18:38:17 +0000 (18:38 +0000)]
Define weak and __weak to mean ARC-style weak references, even in MRC.
Previously, __weak was silently accepted and ignored in MRC mode.
That makes this a potentially source-breaking change that we have to
roll out cautiously. Accordingly, for the time being, actual support
for __weak references in MRC is experimental, and the compiler will
reject attempts to actually form such references. The intent is to
eventually enable the feature by default in all non-GC modes.
(It is, of course, incompatible with ObjC GC's interpretation of
__weak.)
If you like, you can enable this feature with
-Xclang -fobjc-weak
but like any -Xclang option, this option may be removed at any point,
e.g. if/when it is eventually enabled by default.
This patch also enables the use of the ARC __unsafe_unretained qualifier
in MRC. Unlike __weak, this is being enabled immediately. Since
variables are essentially __unsafe_unretained by default in MRC,
the only practical uses are (1) communication and (2) changing the
default behavior of by-value block capture.
As an implementation matter, this means that the ObjC ownership
qualifiers may appear in any ObjC language mode, and so this patch
removes a number of checks for getLangOpts().ObjCAutoRefCount
that were guarding the processing of these qualifiers. I don't
expect this to be a significant drain on performance; it may even
be faster to just check for these qualifiers directly on a type
(since it's probably in a register anyway) than to do N dependent
loads to grab the LangOptions.
David Majnemer [Thu, 22 Oct 2015 18:04:22 +0000 (18:04 +0000)]
[MS ABI] Don't crash when inheriting from base with trailing empty array member
We got this right for Itanium but not MSVC because CGRecordLayoutBuilder
was checking if the base's size was zero when it should have been
checking the non-virtual size.
Benjamin Kramer [Thu, 22 Oct 2015 15:04:10 +0000 (15:04 +0000)]
[Tooling] Add a utility function to replace one nested name with another.
One problem in clang-tidy and other clang tools face is that there is no
way to lookup an arbitrary name in the AST, that's buried deep inside Sema
and might not even be what the user wants as the new name may be freshly
inserted and not available in the AST.
A common use case for lookups is replacing one nested name with another
while minimizing namespace qualifications, so replacing 'ns::foo' with
'ns::bar' will use just 'bar' if we happen to be inside the namespace 'ns'.
This adds a little helper utility for exactly that use case.
Gabor Horvath [Thu, 22 Oct 2015 11:53:04 +0000 (11:53 +0000)]
[analyzer] Bug identification
This patch adds hashes to the plist and html output to be able to identfy bugs
for suppressing false positives or diff results against a baseline. This hash
aims to be resilient for code evolution and is usable to identify bugs in two
different snapshots of the same software. One missing piece however is a
permanent unique identifier of the checker that produces the warning. Once that
issue is resolved, the hashes generated are going to change. Until that point
this feature is marked experimental, but it is suitable for early adoption.
Benjamin Kramer [Thu, 22 Oct 2015 11:21:40 +0000 (11:21 +0000)]
[AST] Store Decl* and Stmt* directly into the ParentMap.
These are by far the most common types to be parents in the AST so it makes
sense to optimize for them. Put them directly into the value of the map.
This currently saves 32 bytes per parent in the map and a pointer
indirection at the cost of some additional complexity in the code.
Sadly this means we cannot return an ArrayRef from getParents anymore, add
a proxy class that can own a single DynTypedNode and otherwise behaves
exactly the same as ArrayRef.
For example on a random large file (X86ISelLowering.cpp) this reduces the
size of the parent map by 24 MB.
In this patch, the file static method addProfileRT is
moved to be a virtual member function of base ToolChain class.
This allows derived toolchain to override the default behavior
easily and make it consistent with Darwin toolchain (a TODO was
added for this refactoring - now removed). A new helper method
is also introduced to test if instrumentation profile option
is turned on or not.
Richard Smith [Thu, 22 Oct 2015 03:52:15 +0000 (03:52 +0000)]
[coroutines] Add lexer support for co_await, co_yield, and co_return keywords.
Add -fcoroutines flag (just for -cc1 for now) to enable the feature. Early
indications are that this will be part of -std=c++1z.
Craig Topper [Thu, 22 Oct 2015 03:13:10 +0000 (03:13 +0000)]
Fix a couple places where InsertText was being called with a pointer and size when it really expects a StringRef and a normally optional bool argument.
The pointer was being implicitly converted to a StringRef and the size was being passed into the bool. Since the bool has a default value normally, no one noticed that the wrong number of arguments was given.
Craig Topper [Thu, 22 Oct 2015 01:56:18 +0000 (01:56 +0000)]
Change SortAndUniqueProtocols to operate directly on a SmallVector rather than taking a pointer and element count that it modifies. This paves the way to directly convert the small vector into an ArrayRef without needing to explicitly pass the modified size. No functional change intended.
While there also use a range-based for loop and use append instead of insert to copy elements into the empty SmallVector."
Ben Langmuir [Wed, 21 Oct 2015 23:12:45 +0000 (23:12 +0000)]
Fix use-after-free in ModuleManager
When removing out-of-date modules we might have left behind a VisitOrder
that contains pointers to freed ModuleFiles. This was very rarely seen,
because it only happens when modules go out of date and the VisitOrder
happens to have the right size to not be recomputed.
John McCall [Wed, 21 Oct 2015 22:06:03 +0000 (22:06 +0000)]
Enable ARC on the fragile runtime.
This is almost entirely a matter of just flipping a switch. 99% of
the runtime support is available all the way back to when it was
implemented in the non-fragile runtime, i.e. in Lion. However,
fragile runtimes do not recognize ARC-style ivar layout strings,
which means that accessing __strong or __weak ivars reflectively
(e.g. via object_setIvar) will end up accessing the ivar as if it
were __unsafe_unretained. Therefore, when using reflective
technologies like KVC, be sure that your paths always refer to a
property.
Reid Kleckner [Wed, 21 Oct 2015 22:01:02 +0000 (22:01 +0000)]
[Driver] Alias -fvisibility=internal to -fvisibility=hidden
The ELF symbol visibilities are:
- internal: Not visibile across DSOs, cannot pass address across DSOs
- hidden: Not visibile across DSOs, can be called indirectly
- default: Usually visible across DSOs, possibly interposable
- protected: Visible across DSOs, not interposable
LLVM only supports the latter 3 visibilities. Internal visibility is in
theory useful, as it allows you to assume that the caller is maintaining
a PIC register for you in %ebx, or in some other pre-arranged location.
As far as LLVM is concerned, this isn't worth the trouble. Using hidden
visibility is always correct, so we can just do that.
John McCall [Wed, 21 Oct 2015 18:06:47 +0000 (18:06 +0000)]
Fix and stylize the emission of GC/ARC ivar and GC block layout strings.
Specifically, handle under-aligned object references (by explicitly
ignoring them, because this just isn't representable in the format;
yes, this means that GC silently ignores such references), descend
into anonymous structs and unions, stop classifying fields of
pointer-to-strong/weak type as strong/weak in ARC mode, and emit
skips to cover the entirety of block layouts in GC mode. As a
cleanup, extract this code into a helper class, avoid a number of
unnecessary copies and layout queries, generate skips implicitly
instead of explicitly tracking them, and clarify the bitmap-creation
logic.
Nico Weber [Wed, 21 Oct 2015 17:13:45 +0000 (17:13 +0000)]
clang-format: Teach --sort-includes to interleave #include and #import.
clang accepts both #include and #import for includes (the latter having an
implicit header guard). Let clang-format interleave both types if
--sort-includes is passed. #import is used frequently in Objective-C code.
Benjamin Kramer [Wed, 21 Oct 2015 16:33:15 +0000 (16:33 +0000)]
Shrink DynTypedNode by one pointer from 40 to 32 bytes (on x86_64).
The MemoizationData cache was introduced to avoid a series of enum
compares at the cost of making DynTypedNode bigger. This change reverts
to using an enum compare but instead of building a chain of comparison
the enum values are reordered so the check can be performed with a
simple greater than. The alternative would be to steal a bit from the
enum but I think that's a more complex solution and not really needed
here.
I tried this on several large .cpp files with clang-tidy and didn't
notice any performance difference. The test change is due to matchers
being sorted by their node kind.
Benjamin Kramer [Wed, 21 Oct 2015 10:07:26 +0000 (10:07 +0000)]
Revert "[AST] Put TypeLocs and NestedNameSpecifierLocs into the ParentMap."
Putting DynTypedNode in the ParentMap bloats its memory foot print.
Before the void* key had 8 bytes, now we're at 40 bytes per key which
can mean multiple gigabytes increase for large ASTs and this count
doesn't even include all the added TypeLoc nodes. Revert until I come
up with a better data structure.
Richard Barton [Wed, 21 Oct 2015 10:03:55 +0000 (10:03 +0000)]
Fix __ARM_FP value for sp-only FPUs with Half-precision
The logic for parsing FP capabilities to set __ARM_FP was mistakenly removing
the Half-Precision capability when handling fp-only-sp resulting in a value
of 0x4. Section 6.5.1 of ACLE states that for such FP architectures the value
should be 0x6
Richard Smith [Wed, 21 Oct 2015 07:13:52 +0000 (07:13 +0000)]
[modules] libstdc++ defines some static inline functions in its internal
headers. If those headers end up being textually included twice into the same
module, we get ambiguity errors.
Work around this by downgrading the ambiguity error to a warning if multiple
identical internal-linkage functions appear in an overload set, and just pick
one of those functions as the lookup result.
Craig Topper [Wed, 21 Oct 2015 04:52:36 +0000 (04:52 +0000)]
Parse into an unsigned type instead of a signed type and then checking for positive and casting to unsigned. Since we know the string starts with a digit it couldn't be negative anyway. NFCI
Reid Kleckner [Tue, 20 Oct 2015 21:04:13 +0000 (21:04 +0000)]
Re-land r250592 without rejecting field refs in unevaluated contexts
This time, I went with the first approach from
http://reviews.llvm.org/D6700, where clang actually attempts to form an
implicit member reference from an UnresolvedLookupExpr. We know that
there are only two possible outcomes at this point, a DeclRefExpr of the
FieldDecl or an error, but its safer to reuse the existing machinery for
this.
Reid Kleckner [Tue, 20 Oct 2015 18:12:08 +0000 (18:12 +0000)]
Look through using decls when classifying implicit member access
Clang will now accept this valid C++11 code:
struct A { int field; };
struct B : A {
using A::field;
enum { TheSize = sizeof(field) };
};
Previously we would classify the 'field' reference as something other
than a field, and then forget to apply the C++11 rule to allow
non-static data member references in unevaluated contexts.
This usually arises in class templates that want to reference fields of
a dependent base in an unevaluated context outside of an instance
method. Such contexts do not allow references to 'this', so the only way
to access the field is with a using decl and an implicit member
reference.
Chris Bieneman [Tue, 20 Oct 2015 16:39:25 +0000 (16:39 +0000)]
[CMake] Make external compiler-rt install scripts relative to CMAKE_INSTALL_PREFIX.
This change makes LLVM_BUILD_EXTERNAL_COMPILER_RT work correctly when overriding CMAKE_INSTALL_PREFIX on the install action (which is how LLVM_CREATE_XCODE_TOOLCHAIN works).
This fix is two parts:
(1) Pass CMAKE_INSTALL_PREFIX in as a variable from the parent install to the child install
(2) When passing COMPILER_RT_INSTALL_PATH into the external project make sure it is passed as a string not a path.
Not specifying the full path for COMPILER_RT_INSTALL_PATH isn't enough to fix the issue because relative paths specified on the CMake command line are expanded relative to the working directory before the cache is populated. Forcing this to a string allows it to remain a relative path through to the install() calls. Relative paths specified in install() calls are expanded relative to CMAKE_INSTALL_PREFIX at install time.
Benjamin Kramer [Tue, 20 Oct 2015 15:08:46 +0000 (15:08 +0000)]
[AST] Put TypeLocs and NestedNameSpecifierLocs into the ParentMap.
Firstly this changes the type of parent map to be keyed on DynTypedNode to
simplify the following changes. This comes with a DenseMapInfo for
DynTypedNode, which is a bit incomplete still and will probably only work
for parentmap right now.
Then the RecursiveASTVisitor in ASTContext is updated and finally
ASTMatchers hasParent and hasAncestor learn about the new functionality.
Now ParentMap is only missing TemplateArgumentLocs and CXXCtorInitializers.
Andrea Di Biagio [Tue, 20 Oct 2015 11:19:54 +0000 (11:19 +0000)]
[x86] Fix maskload/store intrinsic definitions in avxintrin.h
According to the Intel documentation, the mask operand of a maskload and
maskstore intrinsics is always a vector of packed integer/long integer values.
This patch introduces the following two changes:
1. It fixes the avx maskload/store intrinsic definitions in avxintrin.h.
2. It changes BuiltinsX86.def to match the correct gcc definitions for avx
maskload/store (see D13861 for more details).
Alexey Bataev [Tue, 20 Oct 2015 04:24:12 +0000 (04:24 +0000)]
[DEBUG INFO] Emit debug info for type used in explicit cast only.
Currently debug info for types used in explicit cast only is not emitted. It happened after a patch for better alignment handling. This patch fixes this bug.
Differential Revision: http://reviews.llvm.org/D13582
Reid Kleckner [Tue, 20 Oct 2015 00:31:42 +0000 (00:31 +0000)]
Revert "Diagnose UnresolvedLookupExprs that resolve to instance members in static methods"
This reverts commit r250592.
It has issues around unevaluated contexts, like this:
template <class T> struct A { T i; };
template <class T>
struct B : A<T> {
using A<T>::i;
typedef decltype(i) U;
};
template struct B<int>;
David Majnemer [Mon, 19 Oct 2015 23:22:49 +0000 (23:22 +0000)]
[MS ABI] Give linkonce_odr, instead of external_linkage, to certain kinds of static data members
Out-of-line definitions of static data members which have an inline
initializer must get GVA_DiscardableODR linkage instead of
GVA_StrongExternal linkage.
MSVC 2013's behavior is different with respect to this and would cause
link errors if one TU provided a definition while another did not.
MSVC 2015 fixed this bug, making this OK. Note that the 2015 behavior
is always compatible with 2013: it never produces a strong definition.
Nico Weber [Mon, 19 Oct 2015 16:21:29 +0000 (16:21 +0000)]
clang-format: Use pipes instead of temporary files for most lit tests.
This makes the format tests look more like most other FileCheck tests in clang.
The multiple-inputs tests still use temp files, to make sure that the file
input code in clang-format stays tested.
Stop stripping out the comment lines in style-on-command-line.cpp as they don't
get in the way and it makes the test simpler. Also remove 2>&1s on the tests in
that file that don't need it.
Marek Kurdej [Mon, 19 Oct 2015 10:08:35 +0000 (10:08 +0000)]
Added new options to ClangFormat VSIX package.
Summary:
Added new options to ClangFormat VSIX package:
* fallback-style
* assume-filename
* sort-includes.
Changed version to 1.1 (otherwise one couldn't update).
Fixed clang-format escaping of XML reserved characters.
Reviewers: hans, aaron.ballman, klimek, rnk, zturner
The Intel MCU psABI requires floating-point values to be passed in-reg.
This makes the x86-32 ABI code respect "-mfloat-abi soft" and generate float inreg arguments.
Alexey Bataev [Mon, 19 Oct 2015 06:40:17 +0000 (06:40 +0000)]
[OPENMP] Fix for http://llvm.org/PR25221: Infinite loop while parsing OpenMP directive
Clang skipped annot_pragma_openmp token, while it should be considered as a stop token while skipping tokens.
Since the attribute documentation is now auto-generated, the previous references
are no longer valid. This prevented the docs build from completing
successfully.
Nico Weber [Mon, 19 Oct 2015 01:08:30 +0000 (01:08 +0000)]
Update `clang-format -help` output in clang-format docs.
-assume-filename, -fallback-style, and -sort-includes are new. (They're also
longer than the previous options, so all descriptions shift over by some amount,
making this diff look larger than it is.)
It looks like someone renamed "General options" to "Generic Options" too.
Nico Weber [Mon, 19 Oct 2015 01:03:19 +0000 (01:03 +0000)]
Update list of languages advertised in OVERVIEW: A tool to format C/C++/Java/JavaScript/Objective-C/Protobuf code.
If no arguments are specified, it formats the code from standard input
and writes the result to the standard output.
If <file>s are given, it reformats the files. If -i is specified
together with <file>s, the files are edited in-place. Otherwise, the
result is written to the standard output.
USAGE: clang-format [options] [<file> ...]
OPTIONS:
-assume-filename=<string> - When reading from stdin, clang-format assumes this
filename to look for a style config file (with
-style=file) and to determine the language.
-cursor=<uint> - The position of the cursor when invoking
clang-format from an editor integration
-dump-config - Dump configuration options to stdout and exit.
Can be used with -style option.
-fallback-style=<string> - The name of the predefined style used as a
fallback in case clang-format is invoked with
-style=file, but can not find the .clang-format
file to use.
Use -fallback-style=none to skip formatting.
-help - Display available options (-help-hidden for more)
-i - Inplace edit <file>s, if specified.
-length=<uint> - Format a range of this length (in bytes).
Multiple ranges can be formatted by specifying
several -offset and -length pairs.
When only a single -offset is specified without
-length, clang-format will format up to the end
of the file.
Can only be used with one input file.
-lines=<string> - <start line>:<end line> - format a range of
lines (both 1-based).
Multiple ranges can be formatted by specifying
several -lines arguments.
Can't be used with -offset and -length.
Can only be used with one input file.
-offset=<uint> - Format a range starting at this byte offset.
Multiple ranges can be formatted by specifying
several -offset and -length pairs.
Can only be used with one input file.
-output-replacements-xml - Output replacements as XML.
-sort-includes - Sort touched include lines
-style=<string> - Coding style, currently supports:
LLVM, Google, Chromium, Mozilla, WebKit.
Use -style=file to load style configuration from
.clang-format file located in one of the parent
directories of the source file (or current
directory for stdin).
Use -style="{key: value, ...}" to set specific
parameters, e.g.:
-style="{BasedOnStyle: llvm, IndentWidth: 8}"
-version - Display the version of this program output.
Nick Lewycky [Sun, 18 Oct 2015 20:32:12 +0000 (20:32 +0000)]
No functionality change, just fix whitespace, a typo and remove an unnecessary
emacs mode marker. (Changes left behind from another patch that ended up not
working out.)
Dimitry Andric [Sun, 18 Oct 2015 13:32:20 +0000 (13:32 +0000)]
Support linking against OpenMP runtime on FreeBSD.
Summary:
Similar to rL248426 (which was a followup to rL248379 and rL248424), add the
required libraries for OpenMP on the linker command line, and update the test
case.
Daniel Jasper [Sun, 18 Oct 2015 07:02:28 +0000 (07:02 +0000)]
clang-format: [JS] Handle string literals spanning character classes.
If a RegExp contains a character group with a quote (/["]/), the
trailing end of it is first tokenized as a string literal, which leads
to the merging code seeing an unbalanced bracket.
This change parses regex literals from the left hand side. That
simplifies the parsing code and also allows correctly handling escapes
and character classes, hopefully correctly parsing all regex literals.
Patch by Martin Probst, thank you.
Review: http://reviews.llvm.org/D13765
projects / clang / log