llvm-svn: 239657

Reverting until it's easier to use this in a real-world build, e.g.
when the linker can handle it.

llvm-svn: 239656

llvm-svn: 239655

This patch fixes LWG issue 2422 by removing the DECAY_COPY from call once.
The review can be found here: http://reviews.llvm.org/D10191

llvm-svn: 239654

Until GCC 5.1 the __is_trivially* intrinsics were not provided. Enable use of
the builtins for GCC 5.1.

Also enable Reference qualified member functions for GCC 4.9 and greater.

This patch also defines _GNUC_VER to 0 when __GNUC__ is not defined because
libc++ assumes _GNUC_VER is always defined.

llvm-svn: 239653

Fix a few typos and run-on sentences in the clang POD documentation.

Patch by Brian R. Gaeke!

llvm-svn: 239652

Instead, just EvaluateAsInt().

Follow-up to r239549: rsmith points out that isICE() is expensive;
seems like it's not the right concept anyway, as it fails on
`static const' in C, and will actually trigger the assert below on:
    test/Sema/inline-asm-validate-x86.c

llvm-svn: 239651

llvm-svn: 239650

Replacing the dependancy on __member_function_traits with is_function allows
is_member_function_pointer to work more often. In particular it allows it to
work when we don't have variadic templates but the function has an arity > 3.

llvm-svn: 239649

The __atomic_is_lock_free(...) function sometimes requires linkage to libatomic
if it cannot be evaluated at compile time. Remove __c11_atomic_is_lock_free
and use __atomic_is_lock_free(sizeof(Tp)) directly so that it can be evaluated
at compile time.

llvm-svn: 239648

Far better than an assertion about expecting ELF.

llvm-svn: 239647

On error, the temporary output stream wouldn't be flushed and therefore the
caller would see an empty error message.

Patch by Antoine Pitrou

Differential Revision: http://reviews.llvm.org/D10241 

llvm-svn: 239646

NFC.

llvm-svn: 239645

Summary:
Scalarizer has two data structures that hold information about changes
to the function, Gathered and Scattered. These are cleared in finish()
at the end of runOnFunction() if finish() detects any changes to the
function. 

However, finish() was checking for changes by only checking if
Gathered was non-empty. The function visitStore() only modifies
Scattered without touching Gathered. As a result, Scattered could have
ended up having stale data if Scalarizer only scalarized store
instructions. Since the data in Scattered is used during the execution
of the pass, this introduced dangling pointer errors. 

The fix is to check whether both Scattered and Gathered are empty
before deciding what to do in finish().

Reviewers: srhines

Reviewed By: srhines

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D10422

llvm-svn: 239644

Summary:
In addition to easier syntax, IRBuilder makes sure to set correct
debug locations for newly added instructions (bitcast and
llvm.lifetime itself). This restores the original behavior, which
was modified by r234581 (reapplied as r235553).

Extend one of the tests to check for debug locations.

Test Plan: regression test suite

Reviewers: aadg, dblaikie

Subscribers: cfe-commits, majnemer

Differential Revision: http://reviews.llvm.org/D10418

llvm-svn: 239643

NFC.

llvm-svn: 239642

llvm-svn: 239641

into partitions. Also, add an option to clone stub definitions (not just decls)
into partitions: these definitions could be inlined in some places to avoid the
overhead of calling via the stub.

Found by inspection - no test case yet, although I plan to add a unit test for
this once the CompileOnDemand layer refactoring settles down.

llvm-svn: 239640

The underlying problem in PR23823 already existed before my recent change
in r239558, but that change made it worse (failing not only for undeclared
symbols, but also failed overload resolution). This makes Clang not try to
delay the lookup in SFINAE context. I assume no current code is relying on
SFINAE working with lookups that need to be delayed, because that never
seems to have worked.

Differential Revision: http://reviews.llvm.org/D10417

llvm-svn: 239639

If llvm.lifetime.end turns out to be the first instruction in the last
basic block, we can decrement the iterator twice, going past rend.
At the moment, this can never happen because llvm.lifetime.end always
goes immediately after bitcast, but relying on this is very brittle.

llvm-svn: 239638

- Add glc, slc, and tfe operands to flat instructions
- Add missing flat instructions
- Fix the encoding of flat_load_dwordx3 and flat_store_dwordx3.

llvm-svn: 239637

llvm-svn: 239636

tests accordingly.

llvm-svn: 239635

[Hexagon] Making intrinsic tests agnostic to register allocation.  Narrowing intrinsic parameters to appropriate width.

llvm-svn: 239634

We were adding an extra "-mlinker-version" argument to the invocation
based on a value inferred from "ld -v". This is set by the build
systems to either a sane value or an empty string (e.g. for custom
built ld), which we don't want to pass on.

No test really possible because the value depends on both host system
and how CMake was invoked.

llvm-svn: 239633

As suggested by jroelofs in a prior review (D9752),
it makes sense to generally prefer multi-line format.

llvm-svn: 239632

Based on ArchType, Clang's driver can select a non-Clang compiler.
String parsing in Clang would have sufficed if it were only that,
however this change anticipates true llvm support.

Differential Revision: http://reviews.llvm.org/D10413

llvm-svn: 239631

Like we do for the various __*tf* tests, check that long double is the
128bit type we expect directly in the header. The latter is now used by
unrelated tests (__*hf* since r237161), and those tests will break for
no reason if uint128_t doesn't exist, and long double isn't fp128.

llvm-svn: 239630

In the glorious future of opaque pointer types, it won't be possible to
retrieve the pointee type of a pointer type which is what's being done
in this GEP loop - but the first iteration is always a pointer type and
the loop doesn't care about that case, except whether or not the index
is a constant.

So pull that special case out before the loop and start at the second
iteration (index 1) instead.

Originally committed in r236670 and reverted with a test case in
r239015. This change keeps the test case working while also avoiding
depending on pointee types.

llvm-svn: 239629

llvm-svn: 239628

llvm-svn: 239627

MCSymbol.h already forwards declares MCExpr and only uses MCExpr* so doesn't
need to include the header.

llvm-svn: 239626

We would get this right in the case where an explicit cast was formed
but not when we were performing an implicit conversion.

This fixes PR23828.

llvm-svn: 239625

This came up in the patch review for http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20150608/281362.html.

llvm-svn: 239624

For hung off uses, we need a Use* to tell use where the operands are.
This was User::OperandList but we want to remove that to save space
of all subclasses which aren't making use of 'hung off uses'.

Hung off uses now allocate their own 'OperandList' Use* in the
User::new which they call.

getOperandList() now uses the hung off uses bit to work out where the
Use* for the OperandList lives.  If a User has hung off uses, then this
bit tells them to go back a single Use* from the User* and use that
value as the OperandList.

If a User has no hung off uses, then we get the first operand by
subtracting (NumOperands * sizeof(Use)) from the User this pointer.

This saves a pointer from User and all subclasses.  Given the average
size of a subclass of User is 112 or 128 bytes, this saves around 7% of space
With malloc tending to align to 16-bytes the real saving is typically more like 3.5%.

On 'opt -O2 verify-uselistorder.lto.bc', peak memory usage prior to this change
is 149MB and after is 143MB so the savings are around 2.5% of peak.

Looking at some passes which allocate many Instructions and Values, parseIR drops
from 54.25MB to 52.21MB while the Inliner calls to Instruction::clone() drops
from 28.20MB to 27.05MB.

Reviewed by Duncan Exon Smith.

llvm-svn: 239623

There are now 2 versions of User::new.  The first takes a size_t and is the current
implementation for subclasses which need 0 or more Use's allocated for their operands.

The new version takes no extra arguments to say that this subclass needs 'hung off uses'.
The HungOffUses bool is now set in this version of User::new and we can assert in
allocHungOffUses that we are allowed to have hung off uses.
This ensures we call the correct version of User::new for subclasses which need hung off uses.

A future commit will then allocate space for a single Use* which will be used
in place of User::OperandList once that field has been removed.

Reviewed by Duncan Exon Smith.

llvm-svn: 239622

This is to try make it very clear that subclasses shouldn't be changing
the value directly.  Now that OperandList for normal instructions is computed
using the NumOperands, its critical that the NumOperands is accurate or we
could compute the wrong offset to the first operand.

I looked over all places which update NumOperands and they are all safe.
Hung off use User's don't use NumOperands to compute the OperandList so they
are safe to continue to manipulate it.  The only other User which changed it
was GlobalVariable which has an optional init list but always allocated space
for a single Use.  It was correctly setting NumOperands to 1 before setting an
initializer, and setting it to 0 after clearing the init list, so the order was safe.

Added some comments to that code to make sure that this isn't changed in future
without being aware of this constraint.

Reviewed by Duncan Exon Smith.

llvm-svn: 239621

We don't want anyone to access OperandList directly as its going to be removed
and computed instead.  This uses getter's and setter's instead in which we
can later change the underlying implementation of OperandList.

Reviewed by Duncan Exon Smith.

llvm-svn: 239620

GCC allows case-insensitive values for -mcpu, -march and -mtune options.
This patch implements the same behaviour for the -mcpu option for the AArch64 target.

llvm-svn: 239619

The underlaying issues is that this code can't really know if an OS specific or
processor specific section number should return true or false.

One option would be to assert or return an error, but that looks like over
engineering since extensions are not that common.

It seems better to have these be direct implementation of the ELF spec so that
they are natural for someone familiar with ELF reading the code.

Code that does have to handle OS/Architecture specific values can do it at
a higher level.

llvm-svn: 239618