David Majnemer [Fri, 10 Apr 2015 04:52:06 +0000 (04:52 +0000)]
[Sema] Don't assume that an initializer list has an initializer
Given something like 'int({}, 1)', we would try to emit a diagnostic
regarding the excess element in the scalar initializer. However, we
assumed that the initializer list had an element in it.
David Majnemer [Thu, 9 Apr 2015 19:53:25 +0000 (19:53 +0000)]
[Sema] Diagnose references to unbound arrays in function definitions
A [*] is only allowed in a declaration for a function, not in its
definition. We didn't correctly recurse on reference types while
looking for it, causing us to crash in CodeGen instead of rejecting it.
Properly implement warn_unused_result checking for classes/structs.
The previous implementation would copy the attribute from the class to
functions that have the class as their return type when the functions
are first declared. This proved to have two flaws:
1) if the class is forward-declared without the attribute and a
function or method with the class as a its return type is declared,
and afterward the class is defined with warn_unused_result, the
function or method would never inherit the attribute, and
2) the check simply failed for functions and methods that are part of
a template instantiation, regardless of whether the class with
warn_unused_result is part of a specific instantiation or part of
the template itself (presumably because those function/method
declaration does not hit the same code path as a non-template one
and so never inherits the attribute).
The new approach is to instead modify the two places where a function or
method call is checked for the warn_unused_result attribute on the decl
by extending the checks to also look for the attribute on the decl's
return type.
Additionally, the check for return types that have the warn_unused_result
now excludes pointers and references to such types, as such return types do
not necessarily imply a transfer of ownership for the underlying object
being referred to by the return value. This does not change the behavior
of functions that are directly given the warn_unused_result attribute.
Process the -freciprocal-math optimization flag (PR20912)
The driver currently accepts but ignores the -freciprocal-math flag.
This patch passes the flag through and enables 'arcp' fast-math-flag
generation in IR.
Note that this change does not actually enable the optimization for
any target. The reassociation optimization that this flag specifies
was implemented by http://reviews.llvm.org/D6334 :
http://llvm.org/viewvc/llvm-project?view=revision&revision=222510
Because the optimization is done in the backend rather than IR,
the backend must be modified to understand instruction-level
fast-math-flags or a new function-level attribute must be created.
Also note that -freciprocal-math is independent of any target-specific
usage of reciprocal estimate hardware instructions. That requires
its own flag ('-mrecip').
Adds ARM Cortex-R4 and R4F support and tests in Clang. Though Cortex-R4
support was present, the support for hwdiv in thumb-mode was not defined
or tested properly. This has also been added.
Hans Wennborg [Wed, 8 Apr 2015 22:55:09 +0000 (22:55 +0000)]
clang-cl: Support the /fp options (PR23112)
This hooks up the /fp options as aliases for -f[no-]fast-math and
-f[no]-trapping-math. It probably doesn't match cl.exe's behaviour
completely (e.g. LLVM is currently never as precise as /fp:precise),
but it's close enough.
_CxxFrameHandler3 calls terminate if a cleanup action throws, regardless
of what bits you put in the xdata tables. There's no need to model this
in the IR, since we just have to take it out later.
Reland "[SEH] Implement filter capturing in CodeGen"
The test should be fixed. It was failing in NDEBUG builds due to a
missing '*' character in a regex. In asserts builds, the pattern matched
a single digit value, which became a double digit value in NDEBUG
builds. Go figure.
[Objective-C Sema] Use canonical type of properties when comparing
redeclaration of property in class extension and to avoid
bogus error. rdar://20469452
Use the most recent previous decl to check if inline is added after a definition
This affects this test case:
void foo();
template <typename T> class C {
friend inline void foo();
};
inline void foo() {}
C<int> c;
Here, we instantiate the foo friend decl and add it to foo's redecl
chain. However, our previous decl pointer happens to reference the first
declaration of foo, which is not marked inline. When we check to see if
foo was already defined, we implicitly search all previous decls. We
should do the same for the inline check, instead of just checking this
particular previous decl.
David Majnemer [Tue, 7 Apr 2015 22:08:51 +0000 (22:08 +0000)]
[Sema] Correctly recurse when looking for [*] in function definitions
A [*] is only allowed in a declaration for a function, not in its
definition. We didn't correctly recurse while looking for it, causing
us to crash in CodeGen instead of rejecting it.
Revert "Mark instantiated function decls as inline specified if any pattern is"
It breaks down on this test case:
void foo();
template <typename T> class C {
friend void foo();
};
inline void foo() {}
C<int> c;
We shouldn't be marking the instantiation of the friend decl of foo as
inline-specified. It may be possible to fix this by determining if the
full definition is part of the current template, but it seems better to
rever tot green until we come up with a full solution.
This reverts commit r233817, as well as follow-ups r233820 and r233821.
[Objective-C Sema] Patch to not issue unavailbility/deprecated
warning when multiple method declarations are found in global pool
with differing types and some are available.
rdar://20408445
DebugInfo: LLVM API change in r234326 for array-like tuple wrappers
Update a few calls to `DIBuilder` now that `MDTuple` array-wrappers
don't have implicit conversions to `MDTuple*`. I may circle back and
update `DIBuilder` to take arrays here, to make it easier for the
callers.
Error message was:
CGDebugInfo.cpp(1047) : error C2666: 'llvm::MDTypeRefArray::operator []' : 2 overloads have similar conversions
DebugInfoMetadata.h(106): could be 'llvm::MDTypeRef llvm::MDTypeRefArray::operator [](unsigned int) const'
while trying to match the argument list '(llvm::DITypeArray, int)'
David Majnemer [Tue, 7 Apr 2015 02:37:09 +0000 (02:37 +0000)]
[Sema] Don't permit dependent alignments on non-dependent typedef-names
A dependent alignment attribute (like __attribute__((aligned(...))) or
__declspec(align(...))) on a non-dependent typedef or using declaration
poses a considerable challenge: the type is _not_ dependent, the size
_may_ be dependent if the type is used as an array type, the alignment
_is_ dependent.
It is reasonable for a compiler to be able to query the size and
alignment of a complete type. Let's help that become an invariant.
Bob Wilson [Tue, 7 Apr 2015 01:03:35 +0000 (01:03 +0000)]
Report an error when -m<os>-version-min= does not specify a version.
Currently if you use -mmacosx-version-min or -mios-version-min without
specifying a version number, clang silently sets the minimum version to
"0.0.0". This is almost certainly not what was intended, so it is better
to report it as an error. rdar://problem/20433945
[WinEH] Don't create an alloca for unnamed catch parameters
The catch object parameter to llvm.eh.begincatch is optional, and can be
null. We can save some ourselves the stack space, copy ctor, and dtor
calls if we pass null.
While capturing filters aren't very common, we'd like to outline
__finally blocks in the frontend to simplify -O0 EH preparation and
reduce code size. Finally blocks are usually have captures, and this is
the first step towards that.
Currently we don't support capturing 'this' or VLAs.
DebugInfo: Use DILexicalBlockFile::getContext() over getScope()
`getScope()` passes the scope back through a `DILexicalBlock` even
though the underlying pointer may be an incompatible `MDSubprogram`.
Just use `getContext()` directly.
Benjamin Kramer [Mon, 6 Apr 2015 20:01:49 +0000 (20:01 +0000)]
MSan told me that we actually dump the entire scratch buffer into PCH files, initialize it.
Writing 4k of zeros is preferrable to 4k of random memory. Document that. While
there remove the initialization of the first byte of the buffer and start at
index zero. It was writing a literal '0' instead of a null byte at the
beginning anyways, which didn't matter since we never read it.
[Objective-C patch] Patch to fix a crash in IRGen because
of incorrect AST when a compound literal of Objective-C
property access is used to initialize a vertor of floats.
rdar://20407999
Gating clang-fuzzer on the same conditions required to build the LLVMFuzzer library. Otherwise, we can run into a situation where clang-fuzzer attempts to build, but its dependency was never built.
Benjamin Kramer [Mon, 6 Apr 2015 12:30:43 +0000 (12:30 +0000)]
Update our list of distros a bit.
- Debian jessie will be released this month, add the next testing version to the list.
- RHEL7 was released last june.
- Ubuntu utopic was released last october, vivid will follow later this month.
David Majnemer [Sun, 5 Apr 2015 05:32:54 +0000 (05:32 +0000)]
[AST] String literal operator templates have two template args, not one
StmtPrinter assumed that the first template arg was the pack and
attempted to iterate it. However, the GNU extension (which is really
just N3599), has two template arguments. In this case, the second
argument is the pack containing the string contents.
Handle this by desugaring the call to the explicit operator.
For example:
"qux" _zombocom will be shown as
operator "" _zombocom<char, 'q', 'u', 'x'>() in diagnostics and AST
dumps.
N.B. It is actually impossible to render the arguments back to the
source form without storing more information in the AST. For example,
we cannot tell if the user wrote u8"qux" or "qux". We also lose
fidelity when it comes to non-char types for this exact reason (e.g. it
is hard to render a list of wchar_t back to something that can be
printed to the screen even if you don't have to consider surrogate
pairs).
David Majnemer [Sat, 4 Apr 2015 05:37:48 +0000 (05:37 +0000)]
[MS ABI] A pointer-to-function cannot be caught as a pointer-to-void
Don't assume that all pointers are convertible to void pointer.
Instead correctly respect [conv.ptr]p2; only allow pointer types with an
object pointee type to be caught as pointer-to-void.
[i386 ABI] expand small C like structs in C++, just like how we handle small
C structs.
This comes up when we have a function that takes a struct and is defined in a
C++ file and used in a C file.
Before this commit, we will generate byval for C++ and will expand the struct
for C, thus causing difference at IR level. We will use bitcast of function type
at the callsite, which causes the inliner to not inline the function.
This commit changes how we handle small C like structs at IR level, but at
backend, we should generate the same argument passing before and after the
commit.
Note that the condition for expanding is still over conservative. We should be
able to expand type that is spelled with “class” and types that are not C-like.
But this commit fixes the inconsistent argument passing between C/C++.
David Blaikie [Thu, 2 Apr 2015 18:55:21 +0000 (18:55 +0000)]
[opaque pointer type] Update for GEP API changes in LLVM
Now the GEP constant utility functions require the type to be explicitly
passed (since eventually the pointer type will be opaque and not convey
the required type information). For now callers can still pass nullptr
(though none were needed here in Clang, which is nice) if
convenienc/necessary, but eventually that will be disallowed as well.
Benjamin Kramer [Thu, 2 Apr 2015 16:19:54 +0000 (16:19 +0000)]
Lower the default alignment on ASTContext's operator new.
It was documented as 8 and operator new[] defaults to 8, but the normal
operator new was never updated and happily wasted bytes on every other
allocation.
We still have to allocate all Types with 16 byte alignment, update the
allocation calls for Types that were missing explicit alignment.
Benjamin Kramer [Thu, 2 Apr 2015 15:29:07 +0000 (15:29 +0000)]
[ast] Put the Stmt hierarchy on a diet for 64 bit targets.
Previously we would waste 32 bits on alignment, use LLVM_ALIGNAS to
free that space for derived classes an place. Sadly still have to #ifdef
out MSVC 2013 because it can't align based on a sizeof expr.
No intended functionality change. New byte counts:
sizeof(before) | sizeof(after)
This is a tool for checking consistency of code generation with different
compiler options (such as -g or outputting to .s). This tool has found a number
of code generation issues. The script acts as a wrapper to clang or clang++
performing 2 (or more) compiles then comparing the object files. Instructions
for use are in check_cfc.py including how to use with LNT.
[UBSan] Embed UBSan into ASan runtime (Clang part).
Summary:
Change the way we use ASan and UBSan together. Instead of keeping two
separate runtimes (libclang_rt.asan and libclang_rt.ubsan), embed UBSan
into ASan and get rid of libclang_rt.ubsan. If UBSan is not supported on
a platform, all UBSan sources are just compiled into dummy empty object
files. UBSan initialization code (e.g. flag parsing) is directly called
from ASan initialization, so we are able to enforce correct
initialization order.
This mirrors the approach we already use for ASan+LSan. This change doesn't
modify the way we use standalone UBSan.
projects / clang / log