Add the TLS initialization functions to a list of initialization functions. The
back-end takes this list and places the function pointers into the correct
section. This way they're called before `main().'
<rdar://problem/13733006>
llvm-svn: 180739
This includes the following fixes:
- Implement 4 subtly different variants of qualifier mangling and use them
in what I believe are the right places.
- Fix handling of array types. Previously we were always decaying them,
which is wrong if the type appears as a template argument, pointee,
referent etc.
Fixes PR13182.
Differential Revision: http://llvm-reviews.chandlerc.com/D709
llvm-svn: 180250
Add a CXXDefaultInitExpr, analogous to CXXDefaultArgExpr, and use it both in
CXXCtorInitializers and in InitListExprs to represent a default initializer.
There's an additional complication here: because the default initializer can
refer to the initialized object via its 'this' pointer, we need to make sure
that 'this' points to the right thing within the evaluation.
llvm-svn: 179958
Also,
- abstract out the indirect/in memory/in registers decisions into the CGCXXABI
- fix handling of empty struct arguments for '-cxx-abi microsoft'
- add/fix tests
llvm-svn: 179681
non-constant constructors or non-trivial destructors. Plus bugfixes for
thread_local references bound to temporaries (the temporaries themselves are
lifetime-extended to become thread_local), and the corresponding case for
std::initializer_list.
llvm-svn: 179496
Summary:
Handles all inheritance models for both data and function member
pointers.
Also implements isZeroInitializable() and refactors some of the null
member pointer code.
MSVC supports converting member pointers through virtual bases, which
clang does not (yet?) support. Implementing that extension is covered
by http://llvm.org/15713
Reviewers: rjmccall
CC: cfe-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D613
llvm-svn: 179305
linkage specification, and is marked as __attribute__((used)), try to also give
it the unmangled name (by emitting an internal linkage alias) if nothing else
within the translation unit would use that name. This allows inline asm in that
translation unit to use the entity via its unmangled name, which people
apparently rely on.
llvm-svn: 178950
This reverts commit r178497 since the backend has been fixed.
Also add a test to ensure that we're emitting template information for unions.
llvm-svn: 178587
when we actually end a lexical block.
* Added new test for line table / block cleanup.
* Follow-up to r177819 / rdar://problem/13115369
llvm-svn: 178490
* Store the .block_descriptor (instead of self) in the alloca so we
can guarantee that all captured variables are available at -O0.
* Add the missing OpDeref for the alloca.
rdar://problem/12767564
llvm-svn: 178361
This reverts commit 5035c483b7fcbf0fa2a7afba24fa35a10995d195.
This schema change wasn't necessary after all. I'm going ith a different
solution that will hopefully use space more conservatively.
llvm-svn: 178213
Summary:
The only vector types a user can pass from MSVC code to clang code are
the ones from *mmintrin.h, so we only have to match the MSVC mangling
for these types. MSVC mangles the __m128 family of types as tag types,
which we match. For other vector types, we emit a unique tag type
mangling that won't match anything produced by MSVC.
Reviewers: rjmccall
CC: chandlerc, timurrrr, cfe-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D576
llvm-svn: 178036
Summary:
For non-dynamic classes (no virtual bases), member data pointers are
simple offsets from the base of the record. Dynamic classes use an
aggregate for member data pointers and are therefore currently
unsupported.
Unlike Itanium, the ms ABI uses 0 to represent null for polymorphic
classes. Non-polymorphic classes use -1 like Itanium, since 0 is a
valid field offset.
Reviewers: rjmccall
CC: timurrrr, cfe-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D558
llvm-svn: 177753
picking up cleanups from earlier in the statement. Also fix a
crash-on-invalid where a reference to an invalid decl from an
enclosing scope was causing an expression to fail to build, but
only *after* a cleanup was registered from that statement,
causing an assertion downstream.
The crash-on-valid is rdar://13459289.
llvm-svn: 177692
This isn't necessary & with the next change to LLVM the DW_TAG_file_type entry
won't be emitted at all - only the raw filename/directory pair, so match on
that directly instead.
llvm-svn: 177609
The #line directive is mostly for backend testing (keeping these files matching
should simplify maintenance somewhat) though the corresponding backend test
improvement/update doesn't verify the file information directly just yet.
Coming in a later iteration.
llvm-svn: 177559
For constructors/desctructors that return 'this', if there exists a callsite
that returns 'this' and is immediately before the return instruction, make
sure we are using the return value from the callsite.
We don't need to keep 'this' alive through the callsite. It also enables
optimizations in the backend, such as tail call optimization.
Updated from r177211.
rdar://12818789
llvm-svn: 177541
Mostly, try to depend on the annotation comments more so these tests are more
legible, brief, and agnostic to schema changes in the future (sure, they're not
agnostic to changes to the comment annotations but since they're easier to read
they should be easier to update if that happens).
llvm-svn: 177457
Checking for the annotation comment rather than the metadata values makes these
tests resilient to a coming refactor that will pull these fields out into a
separate metadata node.
llvm-svn: 177237
For constructors/desctructors that return 'this', if there exists a callsite
that returns 'this' and is immediately before the return instruction, make
sure we are using the return value from the callsite.
We don't need to keep 'this' alive through the callsite. It also enables
optimizations in the backend, such as tail call optimization.
rdar://12818789
llvm-svn: 177211
the balance between expected behavior and compatibility with the gdb
testsuite.
(GDB gets confused if we break an expression into multiple debug
stmts so we enable this behavior only for inlined functions. For the
full experience people can still use -gcolumn-info.)
llvm-svn: 177164
This fixes a crash in
namespace {
struct X {};
}
extern "C" X test2_b;
X test2_b
before we would assign different linkages to each of the test2_b decls.
llvm-svn: 176869
field to be memcpy'd, rather instead of ASTContext::getTypeAlign(<Field Type>).
For packed structs the alignment of a field may be less than the alignment of
the field's type.
<rdar://problem/13338585>
llvm-svn: 176512
[[noreturn]] function are not required to also be [[noreturn]]. We still emit
calls to virtual __attribute__((noreturn)) functions as noreturn; unlike GCC,
we do require overriders to also be noreturn for that attribute.
llvm-svn: 176476
calls and declarations.
LLVM has a default CC determined by the target triple. This is
not always the actual default CC for the ABI we've been asked to
target, and so we sometimes find ourselves annotating all user
functions with an explicit calling convention. Since these
calling conventions usually agree for the simple set of argument
types passed to most runtime functions, using the LLVM-default CC
in principle has no effect. However, the LLVM optimizer goes
into histrionics if it sees this kind of formal CC mismatch,
since it has no concept of CC compatibility. Therefore, if this
module happens to define the "runtime" function, or got LTO'ed
with such a definition, we can miscompile; so it's quite
important to get this right.
Defining runtime functions locally is quite common in embedded
applications.
llvm-svn: 176286
Since r175326 an implicitly hidden template argument can cause a template
installation to become hidden, even if the template itself has an explicit
default visibility. This requires that we keep track of "late" additions
of the visibility attribute.
This is hopefully the last followup change. It just removes the caching of
visibilities from types so that we can see new attributes even after a type has
been used.
llvm-svn: 176164
bitfield. CGBitField::StorageAlignment holds the alignment in chars, but
emitMemcpy had been treating it as if it were held in bits, leading to
underaligned memcpys.
Related to PR15348.
Thanks very much to Chandler for the diagnosis.
llvm-svn: 176163
Now that implicitly hidden template arguments can make an instantiation hidden,
it is important to look at more than just the canonical decl of the argument
in order to see if an attribute is available in a more recent decl.
This has the disadvantage of exposing when getExplicitVisibility is called,
but lets us handle cases like
template <typename T>
struct __attribute__((visibility("default"))) barT {
static void zed() {}
};
class foo;
class __attribute__((visibility("default"))) foo;
template struct barT<foo>;
llvm-svn: 176112
GCC applies a pragma weak to a decl if it matches the mangled name. We used
to apply if it matched the plain name.
This patch is a compromise: we apply the pragma only if it matches the name
and the decl has C language linkage.
llvm-svn: 176110
Weather we should give C language linkage to functions and variables with
internal linkage probably depends on how much code assumes it. The standard
says they should have no language linkage, but gcc and msvc assign them
C language linkage.
This commit removes the hack that was preventing the mangling on static
functions declare in extern C contexts. It is an experiment to see if we
can implement the rules in the standard.
If it turns out that many users depend on these functions and variables
having C language linkage, we should change isExternC instead and try
to convince the CWG to change the standard.
llvm-svn: 175937
control the visibility of a type for the purposes of RTTI
and template argument restrictions independently of how
visibility propagates to its non-type member declarations.
Also fix r175326 to not ignore template argument visibility
on a template explicit instantiation when a member has
an explicit attribute but the instantiation does not.
The type_visibility work is rdar://11880378
llvm-svn: 175587
We treat this as an alternative to -fvisibility=<?>
which changes the default value visibility to "hidden"
and the default type visibility to "default".
Expose a -cc1 option for changing the default type
visibility, repurposing -fvisibility as the default
value visibility option (also setting type visibility
from it in the absence of a specific option).
rdar://13079314
llvm-svn: 175480
bitfield related issues.
The original commit broke Takumi's builder. The bug was caused by bitfield sizes
being determined by their underlying type, rather than the field info. A similar
issue with bitfield alignments showed up on closer testing. Both have been fixed
in this patch.
llvm-svn: 175389
linkonce_odr. Emit construction vtables as internal in this case, since the ABI
does not guarantee that they will be availble externally.
llvm-svn: 175330
for distinguishing type vs. value visibility.
The changes to the visibility of explicit specializations
are intentional. The change to the "ugly" test case is
a consequence of a sensible implementation, and I am happy
to argue that this is better behavior. Other changes may
or may not be intended; it is quite difficult to divine
intent from some of the code I altered.
I've left behind a comment which I hope explains the
philosophy behind visibility computation.
llvm-svn: 175326
some cases where functions with no language linkage were being treated as having
C language linkage. In particular, don't warn in
extern "C" {
static NonPod foo();
}
Since getLanguageLinkage checks the language linkage, the linkage computation
cannot use the language linkage. Break the loop by checking just the context
in the linkage computation.
llvm-svn: 175117
move-constructors and move-assignment operators, use memcpy to copy adjacent
POD members.
Previously, classes with one or more Non-POD members would fall back on
element-wise copies for all members, including POD members. This often
generated a lot of IR. Without padding metadata, it wasn't often possible
for the LLVM optimizers to turn the element-wise copies into a memcpy.
This code hasn't yet received any serious tuning. I didn't see any serious
regressions on a self-hosted clang build, or any of the nightly tests, but
I think it's important to get this out in the wild to get more testing.
Insights, feedback and comments welcome.
Many thanks to David Blaikie, Richard Smith, and especially John McCall for
their help and feedback on this work.
llvm-svn: 174919
This test was written to make sure *something* sane is generated, not
to test any ABI's signedness semantics.
This should allow the test to pass if AArch64 is the default target.
llvm-svn: 174618
This can yield dramatic speedups of dynamic_cast for simple inheritance trees,
at least with libsupc++. Neither libcxxabi nor libcxxrt make use of this
hint currently, it was never implemented because clang didn't support it.
There was some concern about the number of class hierarchy walks this change
introduces. If it turns out to be an issue we can add caching either at the cast
pair level or even deeper, but we also do a lot of walks in Sema so this
codepath is probably fairly optimized already.
llvm-svn: 174293
- Relax a expression for arm-gnueabi.
- Exclude msvc to limit target triplets to add limited a few targets.
Feel free to remove actions if guys thought they redundant.
llvm-svn: 174278
designator" diagnostic with more correct and more human-friendly "cannot take
address of rvalue of type 'T'".
For the case of & &T::f, provide a custom diagnostic, rather than unhelpfully
saying "cannot take address of rvalue of type '<overloaded function type>'".
For the case of &array_temporary, treat it just like a class temporary
(including allowing it as an extension); the existing diagnostic wording
for the class temporary case works fine.
llvm-svn: 174262
This addresses several (not all) debug info tests that use explicit metadata
numbers. Wherever the same number appeared more than once in a test I used
a named match to ensure the same number appeared in all those cases (this may
still be overly constraining test cases as they may not have actually cared
about that relationship). For one-off numbers I just replaced them with an
unnamed regex.
This may underconstrain poorly written test cases that were interested in
checking that certain metadata nodes were related but didn't actually match
on all the related nodes numbers.
llvm-svn: 174247
r173593 made us a little too eager to associate all code at the end of a
function with the user-written 'return' line. This caused problems with
breakpoints as they'd be set in exception handling code preceeding the
actual non-exception return handling code, leading to the breakpoint never
being hit in non-exceptional execution.
This change restores the pre-r173593 exception handling line information where
the cleanup code is associated with the '}' not the return line.
llvm-svn: 174206
In cooperation with the LLVM patch, this should implement all scalar front-end
parts of the C and C++ ABIs for AArch64.
This patch excludes the NEON support also reviewed due to an outbreak of
batshit insanity in our legal department. That will be committed soon bringing
the changes to precisely what has been approved.
Further reviews would be gratefully received.
llvm-svn: 174055
Title: [PR9027] volatile struct bug: member is not loaded at -O;
This is caused by last flag passed to @llvm.memcpy being false,
not honoring that aggregate has at least one 'volatile' data member
(even though aggregate itself has not been qualified as 'volatile'.
As a result, optimization optimizes away the memcpy altogether.
Patch review by John MaCall (I still need to fix up a test though).
llvm-svn: 173535
never key functions. We did not implement that rule for the
iOS ABI, which was driven by what was implemented in gcc-4.2.
However, implement it now for other ARM-based platforms.
llvm-svn: 173515
with -target i686-win32, you will see;
debug-info-static-member.cpp:11:22: error: in-class initializer for static data member of type 'const float' requires 'constexpr' specifier [-Wstatic-float-init]
const static float const_b = 3.14;
^ ~~~~
constexpr
llvm-svn: 173418
Looks like r161368 fixed this for one case but not all. This change generalizes
the solution over all the unwrapping cases. Now that preserving the qualifiers
is done independent of the particular type being unwrapped I won't bother
adding test cases for each one but at least demonstrate that this change was
necessary & sufficient to fix the bug.
llvm-svn: 173002
Adding the pseudo first parameter to a member function pointer's function type
and mark it as artificial.
Combined with a fix to GDB ( http://sourceware.org/bugzilla/show_bug.cgi?id=14998 )
this fixes gdb.cp/member-ptr.exp with Clang.
llvm-svn: 172911
it apart from [[gnu::noreturn]] / __attribute__((noreturn)), since their
semantics are not equivalent (for instance, we treat [[gnu::noreturn]] as
affecting the function type, whereas [[noreturn]] does not).
llvm-svn: 172691
The testcase in pr14929 shows that this is extremely hard to do. If we choose
to apply the attribute, that causes the visibility of some decls to change and
that can happen really late (during codegen).
Current gcc warns and ignores the attribute in this testcase with a warning.
This suggest that the correct solution is to find a point in the compilation
where we can compute the visibility and
* assert it was never computed before
* reject any attempts to compute it again in the future (with warnings).
llvm-svn: 172305
storage and thus is implicitly zero-initialized, no need to
do C++11 memory model. This patch unconditionally detects
such condition and zeroinitializer's the variable.
Patch has been commented on and OKed by Doug off-line.
// rdar://12897704
llvm-svn: 172144
Using added LLVM functionality in r171698. This works in GDB for member
variable pointers but not member function pointers. See the LLVM commit and
GDB bug 14998 for details.
Un-xfailing cases in the GDB 7.5 test suite will follow.
llvm-svn: 171699
Catch some cases I'd missed in r171605 related to unnamed parameters of record
type. This resolves all remaining cases of PR14573 suppression in the GDB 7.5
test suite. Fix to the test suite to follow.
llvm-svn: 171633
Referring back to the original commit (r115090) which was a frontend only test
I adjusted this test to verify the frontend change that was made, to emit the
protected access value in the flags metadata field.
llvm-svn: 171604
the body of a functions. The problem was that hasBody looks at the entire chain
and causes problems to -fvisibility-inlines-hidden if the cache was not
invalidated.
Original message:
Cache visibility of decls.
This unifies the linkage and visibility caching. I first implemented this when
working on pr13844, but the previous fixes removed the performance advantage of
this one.
This is still a step in the right direction for making linkage and visibility
cheap to use.
llvm-svn: 171053
of a member function with parenthesized declarator.
Like this test case:
class Foo {
const char *(baz)() {
return __PRETTY_FUNCTION__;
}
};
llvm-svn: 170233
I wasn't sure where to put the test case for this, but this seemed like as good
a place as any. I had to reorder the tests here to make them legible while
still matching the order of metadata output in the IR file (for some reason
making it virtual changed the ordering).
Relevant commit to fix up LLVM to actually respect 'artificial' member
variables is coming once I write up a test case for it.
llvm-svn: 170154
both LE and BE targets.
AFAICT, Clang get's this correct for PPC64. I've compared it to GCC 4.8
output for PPC64 (thanks Roman!) and to my limited ability to read power
assembly, it looks functionally equivalent. It would be really good to
fill in the assertions on this test case for x86-32, PPC32, ARM, etc.,
but I've reached the limit of my time and energy... Hopefully other
folks can chip in as it would be good to have this in place to test any
subsequent changes.
To those who care about PPC64 performance, a side note: there is some
*obnoxiously* bad code generated for these test cases. It would be worth
someone's time to sit down and teach the PPC backend to pattern match
these IR constructs better. It appears that things like '(shr %foo,
<imm>)' turn into 'rldicl R, R, 64-<imm>, <imm>' or some such. They
don't even get combined with other 'rldicl' instructions *immediately
adjacent*. I'll add a couple of these patterns to the README, but
I think it would be better to look at all the patterns produced by this
and other bitfield access code, and systematically build up a collection
of patterns that efficiently reduce them to the minimal code.
llvm-svn: 169693
This was an egregious bug due to the several iterations of refactorings
that took place. Size no longer meant what it original did by the time
I finished, but this line of code never got updated. Unfortunately we
had essentially zero tests for this in the regression test suite. =[
I've added a PPC64 run over the bitfield test case I've been primarily
using. I'm still looking at adding more tests and making sure this is
the *correct* bitfield access code on PPC64 linux, but it looks pretty
close to me, and it is *worlds* better than before this patch as it no
longer asserts! =] More commits to follow with at least additional tests
and maybe more fixes.
Sorry for the long breakage due to this....
llvm-svn: 169691
generally support the C++11 memory model requirements for bitfield
accesses by relying more heavily on LLVM's memory model.
The primary change this introduces is to move from a manually aligned
and strided access pattern across the bits of the bitfield to a much
simpler lump access of all bits in the bitfield followed by math to
extract the bits relevant for the particular field.
This simplifies the code significantly, but relies on LLVM to
intelligently lowering these integers.
I have tested LLVM's lowering both synthetically and in benchmarks. The
lowering appears to be functional, and there are no really significant
performance regressions. Different code patterns accessing bitfields
will vary in how this impacts them. The only real regressions I'm seeing
are a few patterns where the LLVM code generation for loads that feed
directly into a mask operation don't take advantage of the x86 ability
to do a smaller load and a cheap zero-extension. This doesn't regress
any benchmark in the nightly test suite on my box past the noise
threshold, but my box is quite noisy. I'll be watching the LNT numbers,
and will look into further improvements to the LLVM lowering as needed.
llvm-svn: 169489
The count attribute is more accurate with regards to the size of an array. It
also obviates the upper bound attribute in the subrange. We can also better
handle an unbound array by setting the count to -1 instead of the lower bound to
1 and upper bound to 0.
llvm-svn: 169311
The count field is necessary because there isn't a difference between the 'lo'
and 'hi' attributes for a one-element array and a zero-element array. When the
count is '0', we know that this is a zero-element array. When it's >=1, then
it's a normal constant sized array. When it's -1, then the array is unbounded.
llvm-svn: 169219
non-trivial if they would not call a move operation, even if they would in fact
call a trivial copy operation. A proper fix is to follow, but this small
directed fix is intended for porting to the 3.2 release branch.
llvm-svn: 167920
This corrects the mangling and linkage of classes (& their member functions) in
cases like this:
struct foo {
struct {
void func() { ... }
} x;
};
we were accidentally giving this nested unnamed struct 'no' linkage where it
should've had the linkage of the outer class. The mangling was incorrecty too,
mangling as TU-wide unnamed type mangling of $_X rather than class-scoped
mangling of UtX_.
This also fixes -Wunused-member-function which would incorrectly diagnose
'func' as unused due to it having no linkage & thus appearing to be TU-local
when in fact it might be correctly used in another TU.
Similar mangling should be applied to function local classes in similar cases
but I've deferred that for a subsequent patch.
Review/discussion by Richard Smith, John McCall, & especially Eli Friedman.
llvm-svn: 167906
If we have a type 'int a[1]' and a type 'int b[0]', the generated DWARF is the
same for both of them because we use the 'upper_bound' attribute. Instead use
the 'count' attrbute, which gives the correct number of elements in the array.
<rdar://problem/12566646>
llvm-svn: 167807
- In C++11, perform overload resolution over all assignment operators, rather than just looking for copy/move assignment operators.
- Clean up after temporaries produced by operator= immediately, rather than accumulating them until the end of the function.
llvm-svn: 167798
Do this by making the mangleNumber(APSInt) overload look like
the int64_t version. (The latter should probably just delegate
to the former).
Test from Evgeny Eltsin!
llvm-svn: 167599
at whether the *selected* constructor would be trivial rather than considering
whether the array's element type has *any* non-trivial constructors of the
relevant kind.
llvm-svn: 167562
checks to enable. Remove frontend support for -fcatch-undefined-behavior,
-faddress-sanitizer and -fthread-sanitizer now that they don't do anything.
llvm-svn: 167413
test/CodeGenCXX/member-alignment.cpp. The test succeeds for
powerpc64-unknown-linux-gnu. If other flavors of powerpc are
shown by buildbots to still be broken, we can adjust the test
at that time.
llvm-svn: 167143
doesn't include padding up to the alignment of the record, take this
as a cue that the alignment of the record should (conservatively) be
set to 1. This is similar to other the other cues we use to determine
that the record has a lower alignment, e.g., that the
externally-supplied layout places fields at lower offsets than we
would. Fixes <rdar://problem/12582052>; test case in LLDB.
llvm-svn: 166824
Bill Wendling