Manman Ren [Wed, 5 Feb 2014 20:40:15 +0000 (20:40 +0000)]
PGO: instrumentation based profiling sets function attributes.
We collect a maximal function count among all functions in the pgo data file.
For functions that are hot, we set its InlineHint attribute. For functions that
are cold, we set its Cold attribute.
We currently treat functions with >= 30% of the maximal function count as hot
and functions with <= 1% of the maximal function count are treated as cold.
These two numbers are from preliminary tuning on SPEC.
This commit should not affect non-PGO builds and should boost performance on
instrumentation based PGO.
Kaelyn Uhrain [Wed, 5 Feb 2014 18:57:51 +0000 (18:57 +0000)]
Don't consider records with a NULL identifier as a name for typo correction.
Because in C++, "anonymous" doesn't mean "nameless" for records. In
other words, RecordDecl::isAnonymousStructOrUnion only returns true if
the record lacks a name *and* is not used as the type in an object's
declaration.
Reid Kleckner [Wed, 5 Feb 2014 17:27:08 +0000 (17:27 +0000)]
MS ABI: Mangle member pointer template arguments
Member pointers are mangled as they would be represented at runtime.
They can be a single integer literal, single decl, or a tuple with some
more numbers tossed in. With Clang today, most of those numbers will be
zero because we reject pointers to members of virtual bases.
This change required moving VTableContextBase ownership from
CodeGenVTables to ASTContext, because mangling now depends on vtable
layout.
I also hoisted the inheritance model helpers up to be inline static
methods of MSInheritanceAttr. This makes the AST code that deals with
member pointers much more readable.
MSVC doesn't appear to have stable manglings of null member pointers:
- Null data memptrs in function templates have a mangling collision with
the first field of a non-polymorphic single inheritance class.
- The mangling of null data memptrs changes if you add casts.
- Large null function memptrs in class templates crash MSVC.
Clang uses the class template mangling for null data memptrs and the
function template mangling for null function memptrs to deal with this.
Added the hasLoopVariable sub-matcher for forRangeStmt.
Summary:
This sub-matcher makes it possible to access directly the range-based for
loop variable: forRangeStmt(hasLoopVariable(anything()).bind(...)).
I've tried to re-generate the docs, but the diffs seem to include much more than
this change could cause, so I'd better leave docs update to someone who knows
the intended changes in the contents better.
John McCall [Tue, 4 Feb 2014 23:58:19 +0000 (23:58 +0000)]
Tighten lax vector-conversion rules and enforce them consistently.
When a lax conversion featured a vector and a non-vector, we were
only requiring the non-vector to be a scalar type, but really it
needs to be a real type (i.e. integral or real floating); it is
not reasonable to allow a pointer, member pointer, or complex
type here.
r198474 required lax conversions to match in "data size", i.e.
element size * element count, forbidding matches that happen
only because a vector is rounded up to the nearest power of two
in size. Unfortunately, the erroneous logic was repeated in
several different places; unify them to use the new condition,
so that it triggers for arbitrary conversions and not just
those performed as part of binary operator checking.
Justin Bogner [Tue, 4 Feb 2014 19:18:37 +0000 (19:18 +0000)]
Clean up whitespace checks
In TokenLexer::ExpandFunctionArguments(), CurTok.hasLeadingSpace() is
checked in multiple locations, each time subtly differently. Checking it
early, when the token is seen, and using NextTokGetsSpace exclusively
after that makes the code simpler.
Justin Bogner [Tue, 4 Feb 2014 19:18:35 +0000 (19:18 +0000)]
Fix whitespace handling in empty macro expansions
When a macro expansion does not result in any tokens, and the macro name
is preceded by whitespace, the whitespace should be passed to the first
token that follows the macro expansion. Similarly when a macro expansion
ends with a placemarker token, and that placemarker token is preceded by
whitespace. This worked already for top-level macro expansions, but is
now extended to also work for nested macro expansions.
Justin Bogner [Tue, 4 Feb 2014 19:18:32 +0000 (19:18 +0000)]
Fix whitespace handling in empty macro arguments
When a function-like macro definition ends with one of the macro's
parameters, and the argument is empty, any whitespace before the
parameter name in the macro definition needs to be preserved. Promoting
the existing NextTokGetsSpace to a preserved bit-field and checking it
at the end of the macro expansion allows it to be moved to the first
token following the macro expansion result.
Justin Bogner [Tue, 4 Feb 2014 19:18:28 +0000 (19:18 +0000)]
Fix whitespace handling in ## operator
In x ## y, where x and y are regular tokens, whitespace between x and ##
is ignored, and whitespace between ## and y is also ignored. When either
x or y is a function argument, whitespace was preserved, but it should
not be. This patch removes the checks for whitespace before ## and
before y, and in the special case where x is an empty macro argument and
y is a regular token, actively removes whitespace before y.
One existing test is affected by that change, but as clang's output now
matches the standard's requirements and that of GCC, I've tweaked the
testcase.
Tim Northover [Tue, 4 Feb 2014 14:55:52 +0000 (14:55 +0000)]
ARM & AArch64: combine implementation of vcaXYZ intrinsics
Now that the back-end intrinsics are more regular, there's no need for the
special handling these got in the front-end, so they can be moved to
EmitCommonNeonBuiltinExpr.
Richard Smith [Tue, 4 Feb 2014 01:14:30 +0000 (01:14 +0000)]
Add implicit declarations of allocation functions when looking them up for
redeclaration, not just when looking them up for a use -- we need the implicit
declaration to appropriately check various properties of them (notably, whether
they're deleted).
Allow specifying a custom PathDiagnosticConsumer for use with the static analyzer.
Summary:
Make objects returned by CreateAnalysisConsumer expose an interface,
that allows providing a custom PathDiagnosticConsumer, so that users can have
raw data in a form easily usable from the code (unlike plist/HTML in a file).
David Majnemer [Mon, 3 Feb 2014 00:29:57 +0000 (00:29 +0000)]
MS ABI: Fix some layout tests
Some lines intended to be used for testing x86_64 ABI compatibility were
not firing because lines were annotated with the wrong FileCheck prefix:
X64 vs x64
N.B. Changes beyond just changing x64 to X64 were made, presumably
because other parts of the layout engine have changed. I've verified
the changes to make sure that MSVC creates a compatible layout.
Bob Wilson [Sat, 1 Feb 2014 21:06:21 +0000 (21:06 +0000)]
Fix an assertion failure when building for the iOS simulator. rdar://15959009
When building for i386 or x86_64 with IPHONEOS_DEPLOYMENT_TARGET set in the
environment, the toolchain correctly recognizes that the target platform is
the iOS simulator. The code in Darwin::addMinVersionArgs was not updated for
svn 197148, where isTargetIPhoneOS() was widely replaced by isTargetIOSBased().
This is kind of a strange case, though, because we probably ought to be
passing -ios_simulator_version_min to the linker, but according to the FIXME
in the code, we intentionally avoid that unless the -mios-simulator-version-min
option was used. I don't know whether it is safe to change that yet, so
for now, I am just fixing the assertion failure.
Reid Kleckner [Sat, 1 Feb 2014 00:04:45 +0000 (00:04 +0000)]
[ms-cxxabi] Use inalloca on win32 when passing non-trivial C++ objects
When a non-trivial parameter is present, clang now gathers up all the
parameters that lack inreg and puts them into a packed struct. MSVC
always aligns each parameter to 4 bytes and no more, so this is a pretty
simple struct to lay out.
On win64, non-trivial records are passed indirectly. Prior to this
change, clang was incorrectly using byval on win64.
I'm able to self-host a working clang with this change and additional
LLVM patches.
Reid Kleckner [Fri, 31 Jan 2014 22:54:50 +0000 (22:54 +0000)]
[ms-cxxabi] Use x86_cdeclmethodcc for __cdecl methods on win32
This fixes PR15768, where the sret parameter and the 'this' parameter
are in the wrong order.
Instance methods compiled by MSVC never return records in registers,
they always return indirectly through an sret pointer. That sret
pointer always comes after the 'this' parameter, for both __cdecl and
__thiscall methods.
Unfortunately, the same is true for other calling conventions, so we'll
have to change the overall approach here relatively soon.
Richard Smith [Fri, 31 Jan 2014 20:47:44 +0000 (20:47 +0000)]
Track the currently-being-built submodule inside the preprocessor (rather than
just storing a flag indicating if there was one), and include it in the 'end of
module' annotation. No functionality change.
Tim Northover [Fri, 31 Jan 2014 10:46:52 +0000 (10:46 +0000)]
ARM & AArch64: unify the rest of the completely shared NEON implementations
This should be the last routine patch: AArch64 does still delegate to
EmitARMBuiltinExpr, but the remaining instances have complications of
one sort or another so some more cunning thought will be needed.
Ben Langmuir [Fri, 31 Jan 2014 01:06:56 +0000 (01:06 +0000)]
Fix autolinking when modules are imported in pch files
Add the ImportDecl to the set of interesting delcarations that are
deserialized eagerly when an AST file is loaded (rather than lazily like
most decls). This is required to get auto linking to work when there is
no explicit import in the main file. Also resolve a FIXME to rename
'ExternalDefinitions', since that is only one of the things that need eager
deserialization. The new name is 'EagerlyDeserializedDecls'. The corresponding
AST bitcode is also renamed.
Richard Smith [Thu, 30 Jan 2014 22:24:05 +0000 (22:24 +0000)]
PR14995: Allow a dependent type as the second parameter of operator++ and
operator--, since it might instantiate as 'int' (or, if it's a pack, it
might instantiate as an empty pack).
Richard Smith [Thu, 30 Jan 2014 22:05:38 +0000 (22:05 +0000)]
Don't produce a 'returning reference to local' warning if a lambda returns a
reference (or pointer) to a variable from the closure object or from the
surrounding function scope.
projects / clang / log