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.
Hans Wennborg [Tue, 17 Jun 2014 00:00:18 +0000 (00:00 +0000)]
MS static locals mangling: don't count enum scopes
We may not have the mangling for static locals vs. enums completely figured out,
but at least for my simple test cases, enums should not increment the mangling
number.
Sylvestre Ledru [Mon, 16 Jun 2014 20:31:15 +0000 (20:31 +0000)]
Check that the directory does not exist.
Otherwise, it could allows local users to obtain sensitive information or
overwrite arbitrary files via a symlink attack on temporary directories with
predictable names.
Reported as CVE-2014-2893 ( https://security-tracker.debian.org/tracker/CVE-2014-2893 )
Found by Jakub Wilk
Richard Smith [Mon, 16 Jun 2014 20:26:19 +0000 (20:26 +0000)]
[modules] When we merge redecl chains or mark a decl used with an update
record, mark all subsequent decls as 'used' too, to maintain the AST invariant
that getPreviousDecl()->Used implies this->Used.
David Majnemer [Mon, 16 Jun 2014 18:46:51 +0000 (18:46 +0000)]
MS ABI: Implement x86_64 RTTI
Summary:
The RTTI scheme for x86_64 is largely the same as the one for i386.
Differences are largely limited to avoiding load-time relocations by
replacing pointers to RTTI metadata with the difference of that data
relative to the load address of the module.
Interestingly, this precludes the possibility of successfully using RTTI
data from another DLL. The ImageBase reference is always relative to
the current DLL.
Alp Toker [Mon, 16 Jun 2014 13:56:47 +0000 (13:56 +0000)]
Use the ShowInSystemHeader bit consistently for all diagnostics
By describing system header suppressions directly in tablegen we eliminate
special cases in getDiagnosticSeverity().
Dropping the reliance on builtin diagnostic classes when mapping also gets us
closer to the goal of reusing the diagnostic machinery for custom diagnostics.
Alp Toker [Sun, 15 Jun 2014 23:30:39 +0000 (23:30 +0000)]
Hide the concept of diagnostic levels from lex, parse and sema
The compilation pipeline doesn't actually need to know about the high-level
concept of diagnostic mappings, and hiding the final computed level presents
several simplifications and other potential benefits.
The only exceptions are opportunistic checks to see whether expensive code
paths can be avoided for diagnostics that are guaranteed to be ignored at a
certain SourceLocation.
This commit formalizes that invariant by introducing and using
DiagnosticsEngine::isIgnored() in place of individual level checks throughout
lex, parse and sema.
This improves conformance with ACLE 6.4.1. Define additional macros that
indicate support for the ARM and Thumb instruction set architecture. This
includes the following set of macros:
Sylvestre Ledru [Sat, 14 Jun 2014 08:45:32 +0000 (08:45 +0000)]
With the option '-analyzer-config stable-report-filename=true',
instead of report-XXXXXX.html, scan-build/clang analyzer generate
report-<filename>-<function, method name>-<function position>-<id>.html.
(id = i++ for several issues found in the same function/method)
Richard Smith [Fri, 13 Jun 2014 23:04:49 +0000 (23:04 +0000)]
A non-trivial array-fill expression isn't necessarily a CXXConstructExpr. It
could be an InitListExpr that runs constructors in C++11 onwards. Fixes a
recent regression (introduced in r210091).
Tim Northover [Fri, 13 Jun 2014 19:43:04 +0000 (19:43 +0000)]
Atomics: emit "cmpxchg weak" where possible
Most builtins date from before the "cmpxchg weak" was a gleam in the
C++ committee's eye, so fortunately not much needs to change. But a
few of them *do* acknowledge that failure is possible.
For these, we'll emit the usual cartesian product of cmpxchg
operations if we can't statically determine weakness. CodeGen can
sort it out later if the function gets inlined.
The only other non-trivial aspect of this is (I think) that we emit
the scalar expression for "IsWeak" once, at the beginning, and
propagate its value through the successive blocks. There's not much in
it, but it's slightly more consistent with the existing handling of
FailureOrder.
Bill Schmidt [Fri, 13 Jun 2014 18:30:06 +0000 (18:30 +0000)]
[PPC64LE] Run some existing Altivec tests on powerpc64le as well
There are several Altivec tests that formerly ran only on big-endian
targets (and in some cases only on 32-bit targets). It is useful to
verify these on little-endian targets as well.
While testing these, I discovered a typo in <altivec.h>. This is also
fixed by this patch.
Tyler Nowicki [Fri, 13 Jun 2014 17:57:25 +0000 (17:57 +0000)]
Adds a Pragma spelling for attributes to tablegen and makes use of it for loop
hint attributes. Includes tests for pragma printing and for attribute order
which is incorrectly reversed by ParsedAttributes.
Alexey Samsonov [Fri, 13 Jun 2014 17:53:44 +0000 (17:53 +0000)]
Remove top-level Clang -fsanitize= flags for optional ASan features.
Init-order and use-after-return modes can currently be enabled
by runtime flags. use-after-scope mode is not really working at the
moment.
The only problem I see is that users won't be able to disable extra
instrumentation for init-order and use-after-scope by a top-level Clang flag.
But this instrumentation was implicitly enabled for quite a while and
we didn't hear from users hurt by it.
Tim Northover [Fri, 13 Jun 2014 14:24:59 +0000 (14:24 +0000)]
IR-change: cmpxchg operations now return { iN, i1 }.
This is a minimal fix for clang. I'll soon add support for generating
weak variants when requested, but that's not really necessary for the
LLVM change in isolation.
David Majnemer [Fri, 13 Jun 2014 06:43:46 +0000 (06:43 +0000)]
MS ABI: Fix inheritance model calculation in CRTP
CRTP-like patterns involve a class which inherits from another class
using itself as a template parameter.
However, the base class itself may try to create a pointer-to-member
which involves the derived class. This is problematic because we
may not have finished parsing the most derived classes' base specifiers
yet.
It turns out that MSVC simply uses the unspecified inheritance model
instead of doing anything fancy.
Reid Kleckner [Thu, 12 Jun 2014 23:03:48 +0000 (23:03 +0000)]
Recover from missing 'typename' in sizeof(T::InnerType)
Summary:
'sizeof' is a UnaryExprOrTypeTrait, and it can contain either a type or
an expression. This change threads a RecoveryTSI parameter through the
layers between TransformUnaryExprOrTypeTrait the point at which we look
up the type. If lookup finds a single type result after instantiation,
we now build TypeSourceInfo for it just like a normal transformation
would.
This fixes the last error in the hello world ATL app that I've been
working with, and it now links and runs with clang. Please try it and
file bugs!
Nico Weber [Thu, 12 Jun 2014 21:15:10 +0000 (21:15 +0000)]
Tweak documentation.
1. Having "get started", "get involved", and "hacking" makes it hard to find
how to send patches, so add a link from "get involved" to "hacking".
2. Remove an almost 5 year old note on the test running meachanism changing
soon.
3. Let "hacking" link to the LLVM developer policy.
Objective-C ARC. Blocks that strongly capture themselves
to call themselves will get the warning:
"Capturing <itself> strongly in this block is likely to
lead to a retain cycle". Cut down on the amount of noise
by noticing that user at some point sets the captured variable
to null in order to release it (and break the cycle).
// rdar://16944538
Reid Kleckner [Thu, 12 Jun 2014 19:49:17 +0000 (19:49 +0000)]
MS ABI: Fix forming pointers to members of a base class
Previously we would calculate the inheritance model of a class when
requiring a pointer to member type of that class to be complete. The
inheritance model is used to figure out how many fields are used by the
member pointer.
However, once we require a pointer to member of a derived class type to
be complete, we can form pointers to members of bases without
calculating the inheritance model for those bases. This was causing
crashes on this simple test case:
struct A {
void f();
void f(int);
};
struct B : public A {};
void g() { void (B::*a)() = &B::f; }
Now we calculate the inheritance models of all base classes when
completing a member pointer type.
Thanks to David Blakie and Richard Smith for pointing out that we can retain the
-Wswitch coverage while avoiding the warning from GCC by pushing the unreachable
outside of the switch!
tools/clang/lib/Basic/DiagnosticIDs.cpp: In function ‘clang::DiagnosticIDs::Level toLevel(clang::diag::Severity)’:
tools/clang/lib/Basic/DiagnosticIDs.cpp:382:1: warning: control reaches end of non-void function [-Wreturn-type]
tools/clang/lib/Format/Format.cpp: In member function ‘virtual std::string clang::format::ParseErrorCategory::message(int) const’:
tools/clang/lib/Format/Format.cpp:282:1: warning: control reaches end of non-void function [-Wreturn-type]
Add a default cases that asserts that we handle the severity, parse error.
Richard Smith [Thu, 12 Jun 2014 00:32:32 +0000 (00:32 +0000)]
[modules] Add first-draft module maps for Clang. These don't cover everything
yet, but they're enough for a clean bootstrap (with a few local patches that
are yet to be committed).
Richard Smith [Thu, 12 Jun 2014 00:01:45 +0000 (00:01 +0000)]
PR19966: don't crash/assert when __underlying_type is applied to an incomplete
enumeration type. I've also filed a LWG issue pointing out that this should be
ill-formed.
Hans Wennborg [Wed, 11 Jun 2014 22:44:39 +0000 (22:44 +0000)]
Don't inherit dllimport to inline move assignment operators
Current MSVC versions don't have move assignment operators, so we
can't rely on them being available in the dll. If we have the
definition, we can just use that directly. This breaks pointer
equality, but should work fine otherwise.
When there is an MSVC version that supports move assignment,
we can key this off the -fmsc-ver option.
Aaron Ballman [Wed, 11 Jun 2014 19:11:24 +0000 (19:11 +0000)]
The automated server-side process isn't quite right yet, so this is a newly-generated attribute reference. Tanya is looking into the server process when she gets the chance.
Kaelyn Takata [Wed, 11 Jun 2014 18:33:46 +0000 (18:33 +0000)]
Add some comments to TypoCorrectionConsumer.
Also move the constructor for NamespaceSpecifierSet out of line to
improve the class' readability. I meant to do these two things while
cleaning up the previous TypoCorrectionConsumer changes and have them
folded into those changes.
Kaelyn Takata [Wed, 11 Jun 2014 18:07:08 +0000 (18:07 +0000)]
Sink SpecifierInfo into the only class that uses it.
SpecifierInfo is not used outside of NamespaceSpecifierSet except
indirectly through NamespaceSpecifierSet's iterator, so clean up the
code a bit by moving SpecifierInfo into NamespaceSpecifierSet. Also drop
SpecifierInfo's trivial yet verbose constructor since brace
initiialization is sufficient in the only two places the constructor was
being explicitly called.
Kaelyn Takata [Wed, 11 Jun 2014 18:07:05 +0000 (18:07 +0000)]
Make a couple of helper classes private to TypoCorrectionConsumer.
The SpecifierInfo and NamespaceSpecifierSet are now only used by
TypoCorrectionConsumer, so treat them as the implementation details of
TypoCorrectionConsumer that they are. Also make NamespaceSpecifierSet's
method names more style guide compliant.
projects / clang / log