Revert r73790, and replace it with a significantly less ugly solution.  Rather than trying to make the global reader-writer lock work,
create separate recursive mutexes for each value map.  The recursive-ness fixes the double-acquiring issue, which having one per ValueMap
lets us continue to maintain some concurrency.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73801 91177308-0d34-0410-b5e6-96231b3b80d8

stub out some hacky code for wiring up the new asmprinter interfaces
on X86.  Not useful yet.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73799 91177308-0d34-0410-b5e6-96231b3b80d8

allow clients of the asmprinter to opt-out of the boilerplate with a #define.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73798 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a typo in a comment that Frits von Bommel noticed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73796 91177308-0d34-0410-b5e6-96231b3b80d8

Make GetMinTrailingZeros a member function of ScalarEvolution,
so that it can access the TargetData member (when available) and
use ValueTracking.h information to compute information for
SCEVUnknown Values.

Also add GetMinLeadingZeros and GetMinSignBits functions,
with minimal implementations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73794 91177308-0d34-0410-b5e6-96231b3b80d8

Re-apply r73718, now that the fix in r73787 is in, and add a
hand-crafted testcase which demonstrates the bug that was exposed
in 254.gap.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73793 91177308-0d34-0410-b5e6-96231b3b80d8

DwarfWriter is used to emit EH info also.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73792 91177308-0d34-0410-b5e6-96231b3b80d8

Enable arm pre-allocation load / store multiple optimization pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73791 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a serious bug that would cause deadlock during abstract type refinement.  The constant creation
gets involved, and we end up trying to recursively acquire a writer lock.  The fix for this is slightly horrible,
and involves passing a boolean "locked" parameter around in Constants.cpp, but it's better than having locked and
unlocked versions of most of the code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73790 91177308-0d34-0410-b5e6-96231b3b80d8

Fix LSR's OptimizeSMax to ignore max operators with more than 2 operands,
which it isn't prepared to handle.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73787 91177308-0d34-0410-b5e6-96231b3b80d8

mv CodeGen/DebugLoc.h Support/DebugLoc.h

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73786 91177308-0d34-0410-b5e6-96231b3b80d8

Move up dwarf writer initialization in common AsmPrinter class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73784 91177308-0d34-0410-b5e6-96231b3b80d8

Revert 73718. It's breaking 254.gap.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73783 91177308-0d34-0410-b5e6-96231b3b80d8

Unbreak cyclic deps

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73781 91177308-0d34-0410-b5e6-96231b3b80d8

Fix bad paste-o in the MSVC atomics.

Patch by Ð\98ван Ð¡Ð¾Ñ\80окин.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73779 91177308-0d34-0410-b5e6-96231b3b80d8

Forgot to remove some explicit locking when it became implicit in the ValueMap.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73778 91177308-0d34-0410-b5e6-96231b3b80d8

Fences are expensive.  Avoid doing them when we're not in multithreaded mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73777 91177308-0d34-0410-b5e6-96231b3b80d8

Make ScalarEvolution::print print getSCEVAtScope values for
expressions, when interesting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73776 91177308-0d34-0410-b5e6-96231b3b80d8

Move the memory fences out of the path for single-threaded mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73775 91177308-0d34-0410-b5e6-96231b3b80d8

Don't (unconditionally) use getSCEVAtScope to simplify the step
expression in IVUsers, because in the case of a use of a non-linear
addrec outside of a loop, this causes the addrec to be evaluated as
a linear addrec.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73774 91177308-0d34-0410-b5e6-96231b3b80d8

Fix incorrect comment pointed out by Duncan.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73773 91177308-0d34-0410-b5e6-96231b3b80d8

Clean up nightly tester to use Perl's File::Find.
Patch by Benjamin Kramer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73772 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a major typo.

Nicolas, this is probably the cause of the failures you were seeing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73770 91177308-0d34-0410-b5e6-96231b3b80d8

make jump threading handle lexically identical compare instructions
as if they were multiple uses of the same instruction.  This interacts
well with the existing loadpre that j-t does to open up many new jump
threads earlier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73768 91177308-0d34-0410-b5e6-96231b3b80d8

fix the sparc codegen to not depend on the sparc asmprinter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73767 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for AuroraUX.  Patch by evocallaghan.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73766 91177308-0d34-0410-b5e6-96231b3b80d8

Since -Wno-long-long is ignored without -pedantic,
place it with -pedantic.  Remove -Wunused since it
is implied by -Wall.  Group -Wno-unused-parameter
with -Wall -W since it is the combination of these
two that turns on -Wunused-parameter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73765 91177308-0d34-0410-b5e6-96231b3b80d8

Add comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73761 91177308-0d34-0410-b5e6-96231b3b80d8

Fix for PR2484: add an SSE1 pattern for a shuffle we normally prefer to
handle with an SSE2 instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73760 91177308-0d34-0410-b5e6-96231b3b80d8

Should be using Bcc (average) latency to determine if-conversion threshold, not BL.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73759 91177308-0d34-0410-b5e6-96231b3b80d8

Minor cleanup; fixes review comments for a previous patch.  Sorry for
taking so long to get to this!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73757 91177308-0d34-0410-b5e6-96231b3b80d8

Teach jump threading to look at comparisons between phi nodes and non-constants.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73755 91177308-0d34-0410-b5e6-96231b3b80d8

Misc tweaks to Intel asm printing to make it more compatible with MASM.
Patch by Benedict Gaster.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73753 91177308-0d34-0410-b5e6-96231b3b80d8

Improve tail call elim to move loads above readonly calls
when it allows forming a tail call.  Patch by Frits van
Bommel.  This implements PR4323.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73752 91177308-0d34-0410-b5e6-96231b3b80d8

part of PR4405: disable a contentious optimization for
strcmp -> memcmp when the lengths of the strings are unknown.

Patch by Nick Lewycky!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73751 91177308-0d34-0410-b5e6-96231b3b80d8

More VNInfo tweaking, plus a little progress on intra-block splitting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73750 91177308-0d34-0410-b5e6-96231b3b80d8

Transfer dead markers when a ldrd is changed into a ldm or a pair of ldr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73749 91177308-0d34-0410-b5e6-96231b3b80d8

Include DataTypes.h for 'uint64_t'.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73748 91177308-0d34-0410-b5e6-96231b3b80d8

Latency information for ARM v6. It's rough and not yet hooked up. Right now  we are only using branch latency to determine if-conversion limits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73747 91177308-0d34-0410-b5e6-96231b3b80d8

Mark a few Thumb instructions commutable; just happened to spot this
while experimenting.  I'm reasonably sure this is correct, but please
tell me if these instructions have some strange property which makes this
change unsafe.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73746 91177308-0d34-0410-b5e6-96231b3b80d8

Fix weird class-size-being-different problems.  At some level this is being caused by config.h not being
included everywhere.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73745 91177308-0d34-0410-b5e6-96231b3b80d8

add a file I missed, this goes with r73743

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73744 91177308-0d34-0410-b5e6-96231b3b80d8

Add some scaffolding for a new experimental asmprinter
implementation.  The idea is that we want asmprinting to
work by converting MachineInstrs into a new MCInst class,
then the per-instruction asmprinter works on MCInst.  MCInst
and the new asmprinters will not depend on most of the
llvm code generators.  This allows building diassemblers
that don't link in the whole llvm code generator.  This is
step #1 of many.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73743 91177308-0d34-0410-b5e6-96231b3b80d8

merge the common darwin settings from the X86/PPC/ARM targets
into DarwinTargetAsmInfo.cpp.  The remaining differences should
be evaluated.  It seems strange that x86/arm has .zerofill but ppc
doesn't, etc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73742 91177308-0d34-0410-b5e6-96231b3b80d8

fix build problem pointed out by John Thompson!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73739 91177308-0d34-0410-b5e6-96231b3b80d8

move mangler quote handling from asm printers to TargetAsmInfo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73738 91177308-0d34-0410-b5e6-96231b3b80d8

minor cleanup

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73737 91177308-0d34-0410-b5e6-96231b3b80d8

simplify macro debug info directive handling.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73736 91177308-0d34-0410-b5e6-96231b3b80d8

Remove UseThumbBacktraces. Just check if subtarget is darwin.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73734 91177308-0d34-0410-b5e6-96231b3b80d8

fix file header

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73733 91177308-0d34-0410-b5e6-96231b3b80d8

Add a skeleton driver for new machine code level fun.  llvm-mc is meant
to be a test driver of other components in the system, which will develop
over time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73732 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase for rdar://6983634

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73729 91177308-0d34-0410-b5e6-96231b3b80d8

Add a parent pointer to SCEV, in preparation for getting rid of the global uniquing tables.  No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73728 91177308-0d34-0410-b5e6-96231b3b80d8

Improved PHI def marking, replaced some gotos with breaks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73727 91177308-0d34-0410-b5e6-96231b3b80d8

Allow user to specify separate SVN url for llvm and the test suite.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73726 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73725 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73723 91177308-0d34-0410-b5e6-96231b3b80d8

Add a SmartScopedLock, and use it to simplify code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73722 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify with SmartRWMutex.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73721 91177308-0d34-0410-b5e6-96231b3b80d8

On Darwin, ams printer should output a second label before a jump table so the linker knows it's a new atom. But this is only needed if the jump table is put in a separate section from the function body.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73720 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify by using no-op-when-not-multithreaded locks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73719 91177308-0d34-0410-b5e6-96231b3b80d8

Generalize LSR's OptimizeSMax to handle unsigned max tests as well
as signed max tests. Along with r73717, this helps CodeGen avoid
emitting code for a maximum operation for this class of loop.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73718 91177308-0d34-0410-b5e6-96231b3b80d8

Recognize n != 0 ? n : 1 as umax(n, 1). Previously only ULT/UGT/ULE/UGE
comparisons were recognized for umax, but instcombine canonicalizes
unsigned comparisons with zero to this simpler form.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73717 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify using mutexes that become no-ops when not in multithreaded mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73716 91177308-0d34-0410-b5e6-96231b3b80d8

Revert IRBuilder CC propagation. Fix SimplifyLibCalls instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73715 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify a lot of code by using a R/W mutex that becomes a no-op when multithreading is disabled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73714 91177308-0d34-0410-b5e6-96231b3b80d8

Use SmartMutex to simplify.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73712 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify the SmartMutex implementation a bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73711 91177308-0d34-0410-b5e6-96231b3b80d8

Give RWMutex the SmartRWMutex treatment too.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73710 91177308-0d34-0410-b5e6-96231b3b80d8

Insert a SmartMutex templated class into the class hierarchy, which takes a template parameter specifying whether this mutex
should become a no-op when not running in multithreaded mode.  Make sys::Mutex a typedef of SmartMutex<false>, to preserve source compatibility.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73709 91177308-0d34-0410-b5e6-96231b3b80d8

Update CMake files to account for new location of Threading.cpp.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73708 91177308-0d34-0410-b5e6-96231b3b80d8

Move Threading.[h|cpp] from Support to System.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73707 91177308-0d34-0410-b5e6-96231b3b80d8

Remove the code from IVUsers that attempted to handle
casted induction variables in cases where the cast
isn't foldable. It ended up being a pessimization in
many cases. This could be fixed, but it would require
a bunch of complicated code in IVUsers' clients. The
advantages of this approach aren't visible enough to
justify it at this time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73706 91177308-0d34-0410-b5e6-96231b3b80d8

Generalize the zext(trunc(t) & C) instcombine to work even with
C is not a low-bits mask, and add a similar instcombine for
zext((trunc(t) & C) ^ C).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73705 91177308-0d34-0410-b5e6-96231b3b80d8

Update a comment to reflect the code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73704 91177308-0d34-0410-b5e6-96231b3b80d8

Fix the double checked locking in this file too.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73703 91177308-0d34-0410-b5e6-96231b3b80d8

As pointed out by Duncan, I accidentally dropped the first MemoryFence of the
double-checked locking pattern here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73701 91177308-0d34-0410-b5e6-96231b3b80d8

Forgot to commit this hunk

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73693 91177308-0d34-0410-b5e6-96231b3b80d8

Propagate calling conv for invokes too

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73692 91177308-0d34-0410-b5e6-96231b3b80d8

Don't use int8_t, that requires DataTypes.h

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73682 91177308-0d34-0410-b5e6-96231b3b80d8

Fix build when disable thread support.

Patch by Howard Su.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73680 91177308-0d34-0410-b5e6-96231b3b80d8

include header for uint8_t.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73675 91177308-0d34-0410-b5e6-96231b3b80d8

Add braces to clarify if/else structure and remove warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73673 91177308-0d34-0410-b5e6-96231b3b80d8

- Update register allocation hint after coalescing. This is done by the target since the hint is target dependent. This is important for ARM register pair hints.
- Register allocator should resolve the second part of the hint (register number) before passing it to the target since it knows virtual register to physical register mapping.
- More fixes to get ARM load / store double word working.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73671 91177308-0d34-0410-b5e6-96231b3b80d8

It looks like nobody is working on PR 4158, so I'm
adding a check to catch this case at compile time
instead of quietly generating incorrect code.
That will at least let us identify CBE failures
that are not due to this problem.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73668 91177308-0d34-0410-b5e6-96231b3b80d8

Fix trailing whitespace from ScalarEvolution::print.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73666 91177308-0d34-0410-b5e6-96231b3b80d8

ARM unified syntax is not specific to ELF; use it for Darwin, too.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73665 91177308-0d34-0410-b5e6-96231b3b80d8

Teach ScalarEvolution how to recognize another xor(and(x, C), C) case.
If C is a single bit and the and gets analyzed as a truncate and
zero-extend, the xor can be represnted as an add.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73664 91177308-0d34-0410-b5e6-96231b3b80d8

Delete unnecessary braces.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73663 91177308-0d34-0410-b5e6-96231b3b80d8

Reapply r73647 in a non-broken form.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73662 91177308-0d34-0410-b5e6-96231b3b80d8

Fix asm string from MOVi16

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73661 91177308-0d34-0410-b5e6-96231b3b80d8

Thumb2 instructions are enabled only in unified assembler mode.
Emit switch directive for it. I have no idea whether this is
requirement for Darwin or not.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73660 91177308-0d34-0410-b5e6-96231b3b80d8

Instcombine zext(trunc(x) & mask) to x&mask, even if the trunc has
multiple users.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73656 91177308-0d34-0410-b5e6-96231b3b80d8

Use double-checked locking for this lazy initialization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73653 91177308-0d34-0410-b5e6-96231b3b80d8

Protect the GC table in Function.cpp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73647 91177308-0d34-0410-b5e6-96231b3b80d8

Honour calling convention and attributes of Callee by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73646 91177308-0d34-0410-b5e6-96231b3b80d8

Use atomic increment here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73643 91177308-0d34-0410-b5e6-96231b3b80d8

Thread-safe (and ManagedStatic-ized) LeakDetector!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73642 91177308-0d34-0410-b5e6-96231b3b80d8

Guard mutation of the timing info global.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73639 91177308-0d34-0410-b5e6-96231b3b80d8

We need to use double-checked locking for lazy initialization in this case when running multithreaded.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73636 91177308-0d34-0410-b5e6-96231b3b80d8