Reid Kleckner [Tue, 25 Feb 2014 00:59:14 +0000 (00:59 +0000)]
MS ABI: Return sret parameters when using inalloca
Previously the X86 backend would look for the sret attribute and handle
this for us. inalloca takes that all away, so we have to do the return
ourselves now.
David Majnemer [Mon, 24 Feb 2014 23:34:17 +0000 (23:34 +0000)]
Sema: Make getPreferredTypeAlign respect alignments specified with an aligned attribute on a typedef
When calculating the preferred alignment of a type, consider if a alignment
attribute came from a typedef declaration. If one did, do not naturally align
the type.
Patch by Stephan Tolksdorf, with a little tweaking and an additional testcase by me.
Richard Smith [Mon, 24 Feb 2014 20:50:36 +0000 (20:50 +0000)]
If preprocessing results in a token with leading whitespace that was expanded
from a macro in column 0, ensure that we print whitespace before it in the -E
output. Patch by Harald van Dijk!
Richard Smith [Mon, 24 Feb 2014 20:45:00 +0000 (20:45 +0000)]
If the first token in a macro that appears at the start of a line expands to
nothing, be sure to inform the *next* token expanded from the macro that it is
now at the start of a line. Patch by Harald van Dijk!
Hans Wennborg [Mon, 24 Feb 2014 15:58:24 +0000 (15:58 +0000)]
[Win32 ABI] Defer operator delete checks until vtable is marked used
We were previously checking at every destructor declaration, but that was a bit
excessive. Since the deleting destructor is emitted with the vtable, do the
check when the vtable is marked used.
Manuel Klimek [Mon, 24 Feb 2014 10:28:36 +0000 (10:28 +0000)]
Fix docs generation for the AST matchers:
1. Move internal functions into ASTMatchersInternal.
2. Adapt dump_ast_matchers.py to the new VariadicOperatorMatcherFunc
signature.
3. Update the actual docs with the updated tool / code.
Kevin Qin [Mon, 24 Feb 2014 02:45:03 +0000 (02:45 +0000)]
[AArch64] Change int64_t from 'long long int' to 'long int' for AArch64 target.
Most 64-bit targets define int64_t as long int, and AArch64 should
make same definition to follow LP64 model. In GNU tool chain, int64_t
is defined as long int for 64-bit target. So to get consistent with GNU,
it's better Changing int64_t from 'long long int' to 'long int',
otherwise clang will get different name mangling suffix compared with g++.
- Only include offsets with local (in function scope) symbols, where we don't encode scoping
- Only include the filename with non-system symbols. Presumably the system headers will not provide conflicting definitions.
Aaron Ballman [Sun, 23 Feb 2014 16:27:21 +0000 (16:27 +0000)]
Fixing the indentation of the RST content in the release notes. Fixes two warnings when building the HTML content and improves the format of the resulting content.
Now that the integrated assembler is considered a first class feature of the
compiler and has a proper feature flag, document the change in the compiler
flags. Ensure that we indicate that the legacy flags are still available, but,
encourage users to switch to the feature flags.
clang: add -f{no-,}integrate-as as consistent parameters
The integrated assembler is a feature. This makes the new flags the default
option, and the previous versions aliases. Ideally, at some point the aliases
would be entirely removed.
Bob Wilson [Sun, 23 Feb 2014 00:11:56 +0000 (00:11 +0000)]
Remove support for the QA_OVERRIDE_GCC3_OPTIONS environment variable.
In r199283 I switched the name of this variable to CCC_OVERRIDE_OPTIONS, but
I kept some code to continue recognizing the old name temporarily. As far as
I know, the only use of this was for some internal testing at Apple, and we've
now switched to use the new name. If anyone else is still using this and needs
more time to switch names, I guess we'll find out! <rdar://problem/15821425>
Forward the -no-integrated-as option to -cc1 rather than simply invoking the
appropriate tool. This is useful since this option has been overloaded to
permit disabling of parsing inline assembly at the MC layer.
This re-applies the previous version of the patch with a renaming of the driver
option to the public name rather than the internal name (-target vs -triple).
The actual failure is fixed separately of an overly aggressive negative pattern
match in the MIPS driver tests. It also fixes the incorrect test for targets
that have the integrated assembler disabled by default.
The tests attempt to validate the invocation of the assembler program with the
integrated assembler disabled. However, the match pattern for the negative
tests are lax and will match both the driver invocation as well as the assembler
invocation. Make the tests more strict by ensuring that we only match the
assembler invocation.
Forward the -no-integrated-as option to -cc1 rather than simply invoking the
appropriate tool. This is useful since this option has been overloaded to
permit disabling of parsing inline assembly at the MC layer.
The language forbids defining enums in prototypes, so this check is normally
redundant, but if an enum is defined during template instantiation it should
not be added to the prototype scope.
While at it, clean up the code that deals with tag definitions in prototype
scope and expand the visibility warning to cover the case where an anonymous
enum is defined.
Warren Hunt [Sat, 22 Feb 2014 00:47:24 +0000 (00:47 +0000)]
[_mm_prefetch] Returning previously deleted comment.
No functional change. It's unclear if the word FIXME is relevant given
that the macro behaves as intended.
Warren Hunt [Sat, 22 Feb 2014 00:40:37 +0000 (00:40 +0000)]
[MS-ABI] Fix MSRecordLayout to handel MultiDimensionalArrays
A recent change caused multi-dimensional arrays not to be handled
correctly, this patch fixes that. Also, it adds a lit test for
multi-dimensional arrays.
Ted Kremenek [Sat, 22 Feb 2014 00:02:03 +0000 (00:02 +0000)]
[ObjC] Make attribute 'objc_protocol_requires_explicit_implementation' behave correctly with default property synthesis.
In particular, if we see an @property within the @interface of a class
conforming to a protocol with this attribute, we treat that as
if the implementation were available, per the rules of default
property synthesis.
Warren Hunt [Fri, 21 Feb 2014 23:49:50 +0000 (23:49 +0000)]
Complete Rewrite of CGRecordLayoutBuilder
CGRecordLayoutBuilder was aging, complex, multi-pass, and shows signs of
existing before ASTRecordLayoutBuilder. It redundantly performed many
layout operations that are now performed by ASTRecordLayoutBuilder and
asserted that the results were the same. With the addition of support
for the MS-ABI, such as placement of vbptrs, vtordisps, different
bitfield layout and a variety of other features, CGRecordLayoutBuilder
was growing unwieldy in its redundancy.
This patch re-architects CGRecordLayoutBuilder to not perform any
redundant layout but rather, as directly as possible, lower an
ASTRecordLayout to an llvm::type. The new architecture is significantly
smaller and simpler than the CGRecordLayoutBuilder and contains fewer
ABI-specific code paths. It's also one pass.
The architecture of the new system is described in the comments. For the
most part, the new system simply takes all of the fields and bases from
an ASTRecordLayout, sorts them, inserts padding and dumps a record.
Bitfields, unions and primary virtual bases make this process a bit more
complicated. See the inline comments.
In addition, this patch updates a few lit tests due to the fact that the
new system computes more accurate llvm types than CGRecordLayoutBuilder.
Each change is commented individually in the review.
Ben Langmuir [Fri, 21 Feb 2014 23:39:37 +0000 (23:39 +0000)]
Add a VFSFromYAML class and a parser to create it
Provides a way to create a virtual file system using a YAML file that
supports mapping a file to a path on an 'external' file system. The
external file system will typically be the 'real' file system, but for
testing it can be changed.
A future patch will add a clang option to allow the user to specify such
a file and overlay it, but for now this code is only exercised by the
unit tests.
Warren Hunt [Fri, 21 Feb 2014 23:08:53 +0000 (23:08 +0000)]
Reapply 201734 but with appropriate gcc compatibility
Because GCC incorrectly defines _mm_prefetch to take anything that casts
to void*, people have started using that behavior. The previous patch
that made _mm_prefetch actually take a const char * broke compatibility
with existing code. This update to the patch leaves the macro that
defines _mm_prefetch with the (void*) cast when _MSC_VER is not defined.
Aaron Ballman [Fri, 21 Feb 2014 21:05:14 +0000 (21:05 +0000)]
Adding role-based capability attributes that allow you to express role management: asserting a capability is held, acquiring a capability and releasing a capability. Also includes some skeleton documentation for these new attributes.
Aaron Ballman [Fri, 21 Feb 2014 14:14:04 +0000 (14:14 +0000)]
Moving the documentation for the type safety checking attributes into AttrDocs. If a custom heading is provided, do not automatically generate the alternate spelling list. This is necessary because some attributes have distinct semantic spellings and meanings, but use the same semantic attribute internally. Such attributes should have multiple elements in their documentation list, but not show all spellings. At some point, it would be nice to have a way to attach the documentation element to a specific spelling for these cases.
Tim Northover [Fri, 21 Feb 2014 11:57:24 +0000 (11:57 +0000)]
ARM & AArch64: use table for EmitCommonNeonBuiltinExpr
This extends the intrinsic lookup table format slightly, and adds
entries for use the shared ARM/AArch64 definitions. The benefit is
currently smaller than for the SISD intrinsics (there's more custom
code implementing this set), but a few lines are saved and there's
scope for future expansion.
NAKAMURA Takumi [Fri, 21 Feb 2014 07:59:59 +0000 (07:59 +0000)]
[CMake] Get rid of explicit dependencies to include/clang/*.inc and introduce CLANG_TABLEGEN_TARGETS.
This does;
- clang_tablegen() adds each tblgen'd target to global property CLANG_TABLEGEN_TARGETS as list.
- List of targets is added to LLVM_COMMON_DEPENDS.
- all clang libraries and targets depend on generated headers.
You might wonder this would be regression, but in fact, this is little loss.
- Almost all of clang libraries depend on tblgen'd files and clang-tblgen.
- clang-tblgen may cause short stall-out but doesn't cause unconditional rebuild.
- Each library's dependencies to tblgen'd files might vary along headers' structure.
It made hard to track and update *really optimal* dependencies.
Each dependency to intrinsics_gen and ClangSACheckers is left as DEPENDS.
Reid Kleckner [Fri, 21 Feb 2014 02:27:32 +0000 (02:27 +0000)]
MS ABI: Include the vfptr offset in memptrs to virtual methods
Virtual methods expect 'this' to point to the vfptr containing the
virtual method, and this extends to virtual member pointer thunks. The
relevant vfptr is always at offset zero on entry to the thunk, and no
this adjustment is needed.
Previously we would not include the vfptr adjustment in the member
pointer, and we'd look at the vfptr offset when loading from the vftable
in the thunk.
Warren Hunt [Fri, 21 Feb 2014 01:40:35 +0000 (01:40 +0000)]
[MS-ABI] Update to zero-sized padding algorithm
Slight change to the way zero-sized sub-objects are tracked in the
presence of virtual bases.
In addition we correctly distinguish between dsize and nvsize.
addresses http://llvm.org/bugs/show_bug.cgi?id=18826
Unit tests are included.
Bob Wilson [Fri, 21 Feb 2014 00:20:07 +0000 (00:20 +0000)]
Fix an assertion failure when invoking dsymutil.
There is no bound architecture for the dsymutil action in the driver. Trying
to check various properties of the target will cause an assertion failure
because the target doesn't get initialized without a bound architecture.
<rdar://problem/16111555>
Reid Kleckner [Thu, 20 Feb 2014 23:37:45 +0000 (23:37 +0000)]
Revert the removal of PPCallbacks::PragmaComment() in r201821
The pp-trace clang tool was using it successfully. We can still delete
the callbacks in Frontend/PrintPreprocessedOutput.cpp because they were
effectively dead.
David Majnemer [Thu, 20 Feb 2014 23:22:07 +0000 (23:22 +0000)]
Sema: Do not assert when dereferencing member pointer using virtual inheritance with an incomplete class type
The MS ABI requires that we determine the vbptr offset if have a
virtual inheritance model. Instead, raise an error pointing to the
diagnostic when this happens.
Reid Kleckner [Thu, 20 Feb 2014 23:07:29 +0000 (23:07 +0000)]
Revert "Enable MSRecordLayout in the presence of external sources."
This reverts commit r201810.
It was failing these tests on my workstation:
Clang :: CodeGen/override-layout.c
Clang :: CodeGenCXX/override-layout.cpp
Clang :: PCH/check-deserializations.cpp
Reid Kleckner [Thu, 20 Feb 2014 22:59:51 +0000 (22:59 +0000)]
Expand macros in pragmas with -fms-extensions and -E
gcc never expands macros in pragmas and MSVC always expands macros
before processing pragmas. Clang usually allows macro expansion, except
in a handful of pragmas, most of which are handled by the lexer.
Also remove PPCallbacks for pragmas that are currently handled in the
parser. Without a Parser, such as with clang -E, these callbacks would
never be called.
Ben Langmuir [Thu, 20 Feb 2014 21:59:23 +0000 (21:59 +0000)]
Recommit virtual file system
Previously reverted in r201755 due to causing an assertion failure.
I've removed the offending assertion, and taught the CompilerInstance to
create a default virtual file system inside createFileManager. In the
future, we should be able to reach into the CompilerInvocation to
customize this behaviour without breaking clients that don't care.
Warren Hunt [Thu, 20 Feb 2014 20:14:29 +0000 (20:14 +0000)]
Enable MSRecordLayout in the presence of external sources.
External sources shouldn't prevent the layout engine from using
MSLayout. If lldb were to support debugging in microsoft mode, some
code will need to be added to MSRecordLayoutBuilder to handel external
layouts.
Jordan Rose [Thu, 20 Feb 2014 17:43:31 +0000 (17:43 +0000)]
[CMake] Fix installation without CLANG_BUILD_EXAMPLES
When CLANG_BUILD_EXAMPLES is not on we set the EXCLUDE_FROM_ALL
directory property for the examples/ directory to tell CMake not to
build them by default. The AddLLVM.cmake APIs are not aware of this and
try to install targets that are not built. This does not cause an
install-time error because CMake excludes the directory from the default
installation. However, now that installation attaches targets to the
LLVMExports export set CMake-based applications that find_package(LLVM)
fail because the example plugin binary is not available.
Tell the AddLLVM.cmake APIs to exclude the examples from installation by
setting the EXCLUDE_FROM_ALL variable they check.
projects / clang / log