Register masks will be used as a compact representation of large clobber
lists.  Currently, an x86 call instruction has some 40 operands
representing call-clobbered registers.  That's more than 1kB of useless
operands per call site.

A register mask operand references a bit mask of call-preserved
registers, everything else is clobbered.  The bit mask will typically
come from TargetRegisterInfo::getCallPreservedMask().

By abandoning ImplicitDefs for call-clobbered registers, it also becomes
possible to share call instruction descriptions between calling
conventions, and we can get rid of the WINCALL* instructions.

This patch introduces the new operand kind.  Future patches will add
RegMask support to target-independent passes before finally the fixed
clobber lists can be removed from call instruction descriptions.

llvm-svn: 148250

Make the auto-detection hack for the iOS simulator set the target triple correctly.  Getting the target triple wrong mostly appears to work, but messes up in subtle cases; for example, we incorrectly conclude that fwrite is actually named fwrite$UNIX2003.  Also shuffles around the auto-detection code a bit to try and make it a bit more reliable.  Fixes <rdar://problem/10664848>.

llvm-svn: 148249

llvm-svn: 148248

non-constant-folded-switch containing a constant-folded switch.

llvm-svn: 148247

One more small optimization:  Where possible, for loops that do a search and then try to break out of the loop early, eliminate the attempt to break out of the loop after the last search.  And with that, I'm declaring __dynamic_cast done.  Though if anyone sees any problems, has suggestions for improvements, or wants to contribute some test cases, that is certainly welcome feedback.

llvm-svn: 148246

llvm-svn: 148245

llvm-svn: 148244

statement which has an unscoped case inside it.
Patch by Aaron Ballman

llvm-svn: 148243

- Add atomic-to/from-nonatomic cast types
- Emit atomic operations for arithmetic on atomic types
- Emit non-atomic stores for initialisation of atomic types, but atomic stores and loads for every other store / load
- Add a __atomic_init() intrinsic which does a non-atomic store to an _Atomic() type.  This is needed for the corresponding C11 stdatomic.h function.
- Enables the relevant __has_feature() checks.  The feature isn't 100% complete yet, but it's done enough that we want people testing it.

Still to do:

- Make the arithmetic operations on atomic types (e.g. Atomic(int) foo = 1; foo++;) use the correct LLVM intrinsic if one exists, not a loop with a cmpxchg.
- Add a signal fence builtin
- Properly set the fenv state in atomic operations on floating point values
- Correctly handle things like _Atomic(_Complex double) which are too large for an atomic cmpxchg on some platforms (this requires working out what 'correctly' means in this context)
- Fix the many remaining corner cases

llvm-svn: 148242

llvm-svn: 148241

llvm-svn: 148240

llvm-svn: 148239

llvm-svn: 148238

Committed as obvious

llvm-svn: 148237

llvm-svn: 148236

llvm-svn: 148235

llvm-svn: 148234

llvm-svn: 148233

A fix for the previous commit: "integer constant is too large for ‘long’ type" error on some 32-bit bots

llvm-svn: 148232

Adding a basic ELF dynamic loader and MC-JIT for ELF. Functionality is currently basic and will be enhanced with future patches.
Patch developed by Andy Kaylor and Daniel Malea. Reviewed on llvm-commits.

llvm-svn: 148231

llvm-svn: 148230

llvm-svn: 148229

Changed intrinsic ID operand to a target constant as its not used in any arithmetic so should not be checked in legalisation

llvm-svn: 148228

I think this is getting close on __dynamic_cast.  There's been quite a bit of code rearrangement, renaming, and better commenting.  This exercise has exposed and fixed a few more bugs.  I've also added several more tests (there's definitely a need for more tests here).

llvm-svn: 148227

We know that the blend instructions only use the MSB, so if the mask is
sign-extended then we can convert it into a SHL instruction. This is a
common pattern because the type-legalizer sign-extends the i1 type which
is used by the LLVM-IR for the condition.

Added a new optimization in SimplifyDemandedBits for SIGN_EXTEND_INREG -> SHL.

llvm-svn: 148225

class/Objective-C protocol suffices get all of the redeclarations of
that declaration wired to the definition, we no longer need to record
the identity of the definition in every declaration. Instead, just
record a bit to indicate whether a particular declaration is the
definition.

llvm-svn: 148224

protocol, record the definition pointer in the canonical declaration
for that entity, and then propagate that definition pointer from the
canonical declaration to all other deserialized declarations. This
approach works well even when deserializing declarations that didn't
know about the original definition, which can occur with modules.

A nice bonus from this definition-deserialization approach is that we
no longer need update records when a definition is added, because the
redeclaration chains ensure that the if any declaration is loaded, the
definition will also get loaded.

llvm-svn: 148223

chains, again. The prior implementation was very linked-list oriented, and
the list-splicing logic was both fairly convoluted (when loading from
multiple modules) and failed to preserve a reasonable ordering for the
redeclaration chains.

This new implementation uses a simpler strategy, where we store the
ordered redeclaration chains in an array-like structure (indexed based
on the first declaration), and use that ordering to add individual
deserialized declarations to the end of the existing chain. That way,
the chain mimics the ordering from its modules, and a bug somewhere is
far less likely to result in a broken linked list.

llvm-svn: 148222

llvm-svn: 148221

llvm-svn: 148220

llvm-svn: 148219

llvm-svn: 148218

llvm-svn: 148217

llvm-svn: 148216

Fixup for r148132. Type replacement for LoopsProperties: from DenseMap to std::map, since we need to keep a valid pointer to properties of current loop.

Message for r148132:
LoopUnswitch: All helper data that is collected during loop-unswitch iterations was moved to separated class (LUAnalysisCache).

llvm-svn: 148215

we're loading from the global array, not how it is spelled in the asm.
This should fix the MSVC bots.

llvm-svn: 148214

removal desired.

llvm-svn: 148213

re-computed rather than the variables be re-used just after the assert.
Just use the variables since we have them already. Fixes an unused
variable warning.

Also fix an 80-column violation.

llvm-svn: 148212

or Clang is using this, and it would be hard to use it correctly given
the thread hostility of the function. Also, it never checked the return
which is rather dangerous with chdir. If someone was in fact using this,
please let me know, as well as what the usecase actually is so that
I can add it back and make it more correct and secure to use. (That
said, it's never going to be "safe" per-se, but we could at least
document the risks...)

llvm-svn: 148211

Don't consider decltype(e) for an instantiation-dependent, but not
type-dependent, e to be non-type-dependent but canonical(!).

llvm-svn: 148210