NAKAMURA Takumi [Sun, 22 Jun 2014 12:42:29 +0000 (12:42 +0000)]
Revert r211402 (and r211408,r211410), "CodeGen: Refactor dynamic_cast and typeid" It crashes msvc codegen in clang/test/SemaCXX/undefined-internal.cpp.
Alp Toker [Sun, 22 Jun 2014 04:31:15 +0000 (04:31 +0000)]
Enable WindowsToolChain on all native Windows builds
Make binaries built by MSVC, mingw and clang functionally equivalent. The
checks are trivially performed at runtime to eliminate functional differences
between supported configurations that used to be hard-coded.
Alp Toker [Sun, 22 Jun 2014 03:27:52 +0000 (03:27 +0000)]
WindowsToolChain: decouple build environment from the toolchain
Don't try to find the MSVC version that the binaries were built with. Doing so
defeats testing by causing invalid test passes on the build servers.
Whichever Visual Studio (or clang-cl.exe) edition was used to build the clang
package, it's strictly orthogonal and has no relation to software versions
available on the user's PC.
Alp Toker [Sat, 21 Jun 2014 23:31:52 +0000 (23:31 +0000)]
TextDiagnostic: print remark level diagnostics in bold too
The purpose of bolding these is to make them visually distinct from
continuations (supplemental note diagnostics). Therefore, the bolding applies
to all severities _including_ remarks -- it's not in any way an indicator of
priority. Also simplify and comment.
David Majnemer [Sat, 21 Jun 2014 18:46:07 +0000 (18:46 +0000)]
Lex: Use the correct types for MS integer suffixes
Something went wrong with r211426, it is an older version of this code
and should not have been committed. It was reverted with r211434.
Original commit message:
We didn't properly implement support for the sized integer suffixes.
Suffixes like i16 were essentially ignored instead of mapping them to
the appropriately sized integer type.
Rafael Espindola [Sat, 21 Jun 2014 12:39:25 +0000 (12:39 +0000)]
Revert "Lex: Use the correct types for MS integer suffixes"
This reverts commit r211426.
This broke the arm bots. The crash can be reproduced on X86 by running.
./bin/clang -cc1 -fsyntax-only -verify -fms-extensions ~/llvm/clang/test/Lexer/ms-extensions.c -triple arm-linux
David Majnemer [Sat, 21 Jun 2014 00:51:59 +0000 (00:51 +0000)]
Lex: Use the correct types for MS integer suffixes
We didn't properly implement support for the sized integer suffixes.
Suffixes like i16 were essentially ignored instead of mapping them to
the appropriately sized integer type.
The version information for Visual Studio is spread over multiple variables.
The newer Windows SDK has started making use of some of the extended versioning
variables that were previously undefined. Enhance our compatibility definitions
for these cases.
_MSC_VER is defined to be the Major * 100 + Minor. _MSC_FULL_VER is defined to
be Major * 10000000 + Minor * 100000 + Build. And _MSC_BUILD is the build
revision of the compiler.
Extend the -fmsc-version option in a compatible manner. If the value is the
previous form of MMmm, then we assume that the build number is 0. Otherwise, a
specific build number may be passed by using the form MMmmbbbbb. Due to
bitwidth limitations of the option, it is currently not possible to define a
revision value.
The version information can be passed as either the decimal encoded value
(_MSC_FULL_VER or _MSC_VER) or as a dot-delimited value.
The change to the TextDiagnostic is to deal with the updated encoding of the
version information.
Justin Bogner [Fri, 20 Jun 2014 22:16:00 +0000 (22:16 +0000)]
Driver: Record that we're in crashdump and push flags to ConstructJob
It's more flexible and arguably better layering to set flags to modify
compiling for diagnostics in the CC1 job themselves, rather than
tweaking the driver flags and letting them propagate.
There is one visible change this causes: crash report files will now
get preprocessed names (.i and friends).
David Majnemer [Fri, 20 Jun 2014 21:11:00 +0000 (21:11 +0000)]
CodeGen: Refactor dynamic_cast and typeid
This refactors the emission of dynamic_cast and typeid expressions so
that ABI specific knowledge lives in appropriate places. There are
quite a few benefits for having the two implementations share a common
core like sharing logic for optimization opportunities.
Ulrich Weigand [Fri, 20 Jun 2014 16:37:40 +0000 (16:37 +0000)]
[PowerPC] Fix small argument stack slot offset for LE
When small arguments (structures < 8 bytes or "float") are passed in a
stack slot in the ppc64 SVR4 ABI, they must reside in the least
significant part of that slot. On BE, this means that an offset needs
to be added to the stack address of the parameter, but on LE, the least
significant part of the slot has the same address as the slot itself.
For the most part, this is handled in the LLVM back-end, where I just
fixed the LE case in commit r211368.
However, there is one piece of the clang front-end that is also aware of
these stack-slot offsets: PPC64_SVR4_ABIInfo::EmitVAArg. This patch
updates that routine to take endianness into account.
James Molloy [Fri, 20 Jun 2014 14:35:13 +0000 (14:35 +0000)]
The ability to use vector initializer lists is a GNU vector extension
and is unrelated to the NEON intrinsics in arm_neon.h. On little
endian machines it works fine, however on big endian machines it
exhibits surprising behaviour:
uint32x2_t x = {42, 64};
return vget_lane_u32(x, 0); // Will return 64.
Because of this, explicitly call out that it is unsupported on big
endian machines.
This patch will emit the following warning in big-endian mode:
test.c:3:15: warning: vector initializers are a GNU extension and are not compatible with NEON intrinsics [-Wgnu]
int32x4_t x = {0, 1, 2, 3};
^
test.c:3:15: note: consider using vld1q_s32() to initialize a vector from memory, or vcombine_s32(vcreate_s32(), vcreate_s32()) to initialize from integer constants
1 warning generated.
Ulrich Weigand [Fri, 20 Jun 2014 13:41:24 +0000 (13:41 +0000)]
[PowerPC] Add MultiarchIncludeDirs for powerpc64le
There was already partial support for multi-arch on powerpc64le,
but the MultiarchIncludeDirs setting was missing. This patch
adds the appropriate definition, and also extends the
linux-header-search.cpp test case to verify an Ubuntu 14.04
powerpc64le tree.
Daniel Jasper [Fri, 20 Jun 2014 08:44:22 +0000 (08:44 +0000)]
Fix/Improve SourceRange of explicitly defaulted members
When adding the implicit compound statement (required for Codegen?), the
end location was previously overridden by the start location, probably
based on the assumptions:
* The location of the compound statement should be the member's location
* The compound statement if present is the last element of a FunctionDecl
This patch changes the location of the compound statement to the
member's end location.
Justin Bogner [Fri, 20 Jun 2014 03:28:46 +0000 (03:28 +0000)]
Frontend: Disentangle removePathTraversal from concatenating paths
This reimplements part of r211303 in a bit of a cleaner way. Doing so
allows us to use a proper absolute path when calling addFileMapping
rather than relying on a substring being one, which should fix the
tests on Windows.
Successfully loaded module files may be referenced in other
ModuleManagers, so don't invalidate them. Two related things are fixed:
1) I thought the last module in the manager was always the one that
failed, but it isn't. So check explicitly against the list of
vetted modules from ReadASTCore.
2) We now keep the file descriptor of pcm file open, which avoids the
possibility of having two different pcms for the same module loaded when
building in parallel with headers being modified during a build.
Objective-C qoi. When Objective-C pointer mismatches with
a qualified-id type because pointer is object of a forward
class declaration, include this info in a diagnostic note.
// rdar://10751015
Justin Bogner [Thu, 19 Jun 2014 19:36:03 +0000 (19:36 +0000)]
Frontend: Add a CC1 flag to dump module dependencies to a directory
This adds the -module-dependency-dir to clang -cc1, which specifies a
directory to copy all of a module's dependencies into in a form
suitable to be used as a VFS using -ivfsoverlay with the generated
vfs.yaml.
This is useful for crashdumps that involve modules, so that the module
dependencies will be intact when a crash report script is used to
reproduce a problem on another machine.
We currently encode the absolute path to the dump directory, due to
limitations in the VFS system. Until we can handle relative paths in
the VFS, users of the VFS map may need to run a simple search and
replace in the file.
Eli Bendersky [Thu, 19 Jun 2014 18:30:15 +0000 (18:30 +0000)]
Fix PR20069: bad loop pragma arguments crash FE
This patch fixes a crash when handling malformed arguments to loop pragmas such
as: "#pragma clang loop vectorize(()". Essentially any argument which is not an
identifier or constant resulted in a crash. This patch also changes a couple of
the error messages which weren't quite correct. New behavior with this patch vs
old behavior:
#pragma clang loop vectorize(1)
OLD: error: missing keyword; expected 'enable' or 'disable'
NEW: error: invalid argument; expected 'enable' or 'disable'
Reid Kleckner [Thu, 19 Jun 2014 01:23:22 +0000 (01:23 +0000)]
DiagnoseUnknownTypename always emits a diagnostic and returns true
Make it return void and delete the dead code in the parser that handled
the case where it might return false. This has been dead since 2010
when John deleted Action.h.
Objective-C ARC. Allow conversion of (void*) pointers to
retainable ObjC pointers without requiring a bridge-cast
in the context of pointer comparison as this is in effect
a +0 context. // rdar://16627903
Objective-C ARC. Allow conversion of (void*) pointers to
retainable ObjC pointers without requiring a bridge-cast
by recognizing this as a +0 context. // rdar://16627903
Add support for _InterlockedCompareExchangePointer, _InterlockExchangePointer,
_InterlockExchange. These are available as a compiler intrinsic on ARM and x86.
These are used directly by the Windows SDK headers without use of the intrin
header.
Jordan Rose [Wed, 18 Jun 2014 19:23:30 +0000 (19:23 +0000)]
[analyzer] Don't create new PostStmt nodes if we don't have to.
Doing this caused us to mistakenly think we'd seen a particular state before
when we actually hadn't, which resulted in false negatives. Credit to
Rafael Auler for discovering this issue!
Tim Northover [Wed, 18 Jun 2014 08:37:28 +0000 (08:37 +0000)]
AArch64: re-enable tests that were looking for a non-existent backend.
In the final phase of the merge, I managed to disable a bunch of Clang
tests accidentally. Fortunately none of them seem to have broken in
the interim.
Hans Wennborg [Wed, 18 Jun 2014 01:21:33 +0000 (01:21 +0000)]
Fix bug in code for avoiding dynamic initialization of dllimport globals
When instantiating dllimport variables with dynamic initializers, don't
bail out of Sema::InstantiateVariableInitializer without calling
PopExpressionEvaluationContext().
This was causing a stale object to stay on the ExprEvalContexts stack,
causing subsequent calls to getCurrentMangleNumberContext() to fail,
resulting in incorrect numbering of static locals (and probably other
broken things).
Objective-C ARC. Do not warn about properties with both
IBOutlet and weak attributes when accessed being
unpredictably set to nil because usage of such properties
are always single threaded and its ivar cannot be set
to nil asynchronously. // rdar://15885642
Ben Langmuir [Tue, 17 Jun 2014 22:35:27 +0000 (22:35 +0000)]
Retry building modules that were compiled by other instances and are out-of-date
When another clang instance builds a module, it may still be considered
"out of date" for the current instance in a couple of cases*. This
patch prevents us from giving spurious errors when compilers race to
build a module by allowing the module load to fail when the pcm was
built by a different compiler instance.
* Cases where a module can be out of date despite just having been
built:
1) There are different -I paths between invocations that result in
finding a different module map file for some dependent module. This is
not an error, and should never be diagnosed.
2) There are file system races where the headers making up a module are
touched or moved. Although this can sometimes mean trouble, diagnosing
it only during a build-race is worse than useless and we cannot detect
this in general. It is more robust to just rebuild. This was causing
spurious issues in some setups where only the modtime of headers was
bumped during a build.
Diego Novillo [Tue, 17 Jun 2014 20:01:51 +0000 (20:01 +0000)]
Remove dead code.
The parsing for -Rpass= had been factored into the function
GenerateOptimizationRemarkRegex, but at the time I forgot to remove
the original code that just handled OPT_Rpass_EQ.
Zachary Turner [Tue, 17 Jun 2014 19:57:15 +0000 (19:57 +0000)]
Change libclang initialization to use std::call_once instead of
hand rolled once-initialization, and rename the mutex to be more
descriptive of its actual purpose.
James Molloy [Tue, 17 Jun 2014 13:11:27 +0000 (13:11 +0000)]
Rewrite ARM NEON intrinsic emission completely.
There comes a time in the life of any amateur code generator when dumb string
concatenation just won't cut it any more. For NeonEmitter.cpp, that time has
come.
There were a bunch of magic type codes which meant different things depending on
the context. There were a bunch of special cases that really had no reason to be
there but the whole thing was so creaky that removing them would cause something
weird to fall over. There was a 1000 line switch statement for code generation
involving string concatenation, which actually did lexical scoping to an extent
(!!) with a bunch of semi-repeated cases.
I tried to refactor this three times in three different ways without
success. The only way forward was to rewrite the entire thing. Luckily the
testing coverage on this stuff is absolutely massive, both with regression tests
and the "emperor" random test case generator.
The main change is that previously, in arm_neon.td a bunch of "Operation"s were
defined with special names. NeonEmitter.cpp knew about these Operations and
would emit code based on a huge switch. Actually this doesn't make much sense -
the type information was held as strings, so type checking was impossible. Also
TableGen's DAG type actually suits this sort of code generation very well
(surprising that...)
So now every operation is defined in terms of TableGen DAGs. There are a bunch
of operators to use, including "op" (a generic unary or binary operator), "call"
(to call other intrinsics) and "shuffle" (take a guess...). One of the main
advantages of this apart from making it more obvious what is going on, is that
we have proper type inference. This has two obvious advantages:
1) TableGen can error on bad intrinsic definitions easier, instead of just
generating wrong code.
2) Calls to other intrinsics are typechecked too. So
we no longer need to work out whether the thing we call needs to be the Q-lane
version or the D-lane version - TableGen knows that itself!
Here's an example: before:
case OpAbdl: {
std::string abd = MangleName("vabd", typestr, ClassS) + "(__a, __b)";
if (typestr[0] != 'U') {
// vabd results are always unsigned and must be zero-extended.
std::string utype = "U" + typestr.str();
s += "(" + TypeString(proto[0], typestr) + ")";
abd = "(" + TypeString('d', utype) + ")" + abd;
s += Extend(utype, abd) + ";";
} else {
s += Extend(typestr, abd) + ";";
}
break;
}
As an example of what happens if you do something wrong now, here's what happens
if you make $p0 unsigned before the call to "vabd" - that is, $p0 -> (cast "U",
$p0):
arm_neon.td:574:1: error: No compatible intrinsic found - looking up intrinsic 'vabd(uint8x8_t, int8x8_t)'
Available overloads:
- float64x2_t vabdq_v(float64x2_t, float64x2_t)
- float64x1_t vabd_v(float64x1_t, float64x1_t)
- float64_t vabdd_f64(float64_t, float64_t)
- float32_t vabds_f32(float32_t, float32_t)
... snip ...
This makes it seriously easy to work out what you've done wrong in fairly nasty
intrinsics.
As part of this I've massively beefed up the documentation in arm_neon.td too.
Things still to do / on the radar:
- Testcase generation. This was implemented in the previous version and not in
the new one, because
- Autogenerated tests are not being run. The testcase in test/ differs from
the autogenerated version.
- There were a whole slew of special cases in the testcase generation that just
felt (and looked) like hacks.
If someone really feels strongly about this, I can try and reimplement it too.
- Big endian. That's coming soon and should be a very small diff on top of this one.
projects / clang / log