Fix MemoryBuffer::getSTDIN to *not* return null if stdin is empty, this is a lame API.

Also, Stringrefify some more MemoryBuffer functions, and add two performance FIXMEs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86630 91177308-0d34-0410-b5e6-96231b3b80d8

Allow targets to specify register classes whose member registers should not be renamed to break anti-dependencies.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86628 91177308-0d34-0410-b5e6-96231b3b80d8

pull a bunch of logic out of instcombine into instsimplify for compare
simplification, this handles the foldable fcmp x,x cases among many others.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86627 91177308-0d34-0410-b5e6-96231b3b80d8

Pass the (optional) TargetData object to ConstantFoldInstOperands
and ConstantFoldCompareInstOperands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86626 91177308-0d34-0410-b5e6-96231b3b80d8

inline a simple function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86625 91177308-0d34-0410-b5e6-96231b3b80d8

rename SimplifyCompare -> SimplifyCmpInst and split it into
Simplify[IF]Cmp pieces.  Add some predicates to CmpInst to
determine whether a predicate is fp or int.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86624 91177308-0d34-0410-b5e6-96231b3b80d8

Now that the default is 'enabled,' a separate command line option for ARM is
not necessary.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86621 91177308-0d34-0410-b5e6-96231b3b80d8

Add testcase for recent checkin.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86620 91177308-0d34-0410-b5e6-96231b3b80d8

fix ConstantFoldCompareInstOperands to take the LHS/RHS as
individual operands instead of taking a temporary array

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86619 91177308-0d34-0410-b5e6-96231b3b80d8

Add StringSwitch::Cases overloads, for matching multiple strings to a single
value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86618 91177308-0d34-0410-b5e6-96231b3b80d8

use instructionsimplify instead of a weak clone of ad-hoc folding stuff.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86616 91177308-0d34-0410-b5e6-96231b3b80d8

Update test

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86614 91177308-0d34-0410-b5e6-96231b3b80d8

stub out a new libanalysis "instruction simplify" interface that
takes decimated instructions and applies identities to them.  This
is pretty minimal at this point, but I plan to pull some instcombine
logic out into these and similar routines.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86613 91177308-0d34-0410-b5e6-96231b3b80d8

Remove dlsym stubs, with Nate Begeman's permission.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86606 91177308-0d34-0410-b5e6-96231b3b80d8

Enable dynamic stack realignment by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86604 91177308-0d34-0410-b5e6-96231b3b80d8

stub out a new form of BasicBlock::RemovePredecessorAndSimplify which
simplifies instruction users of PHIs when the phi is eliminated.  This
will be moved to transforms/utils after some other refactoring.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86603 91177308-0d34-0410-b5e6-96231b3b80d8

Set dynamic stack realignment to real values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86602 91177308-0d34-0410-b5e6-96231b3b80d8

Remove an unneeded #include.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86601 91177308-0d34-0410-b5e6-96231b3b80d8

Fix for 64-bit builds.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86600 91177308-0d34-0410-b5e6-96231b3b80d8

Similar to r86588, but for Darwin this time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86592 91177308-0d34-0410-b5e6-96231b3b80d8

The jump table was being generated before the end label for exception handling
was generated. This caused code like this:

## The asm code for the function
        .section        __TEXT,__const
        .align  2
lJTI11_0:
LJTI11_0:
        .long    LBB11_16
        .long    LBB11_4
        .long    LBB11_5
        .long    LBB11_6
        .long    LBB11_7
        .long    LBB11_8
        .long    LBB11_9
        .long    LBB11_10
        .long    LBB11_11
        .long    LBB11_12
        .long    LBB11_13
        .long    LBB11_14
Leh_func_end11:   ## <---now in the wrong section!

The `Leh_func_end11' would then end up in the wrong section, causing the
resulting EH frame information to be wrong:

__ZL11CheckRightsjPKcbRbRP6NSData.eh:
    .set    Lset500eh,Leh_frame_end11-Leh_frame_begin11
    .long   Lset500eh  ; Length of Frame Information Entry
Leh_frame_begin11:
    .long   Leh_frame_begin11-Leh_frame_common
    .long   Leh_func_begin11-.
    .set    Lset501eh,Leh_func_end11-Leh_func_begin11
    .long   Lset501eh                                   ; FDE address range
`Lset501eh' is now something huge instead of the real value.

The X86 back-end generates the jump table after the EH information is
emitted. Do the same here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86588 91177308-0d34-0410-b5e6-96231b3b80d8

Print "..." instead of all the uninteresting register clobbers on call
instructions. This makes CodeGen dumps significantly less noisy.

Example before:
  BL <ga:@bar>, %R0<imp-def>, %R1<imp-def,dead>, %R2<imp-def,dead>, %R3<imp-def,dead>, %R12<imp-def,dead>, %LR<imp-def,dead>, %D0<imp-def,dead>, %D1<imp-def,dead>, %D2<imp-def,dead>, %D3<imp-def,dead>, %D4<imp-def,dead>, %D5<imp-def,dead>, %D6<imp-def,dead>, %D7<imp-def,dead>, %D16<imp-def,dead>, %D17<imp-def,dead>, %D18<imp-def,dead>, %D19<imp-def,dead>, %D20<imp-def,dead>, %D21<imp-def,dead>, %D22<imp-def,dead>, %D23<imp-def,dead>, %D24<imp-def,dead>, %D25<imp-def,dead>, %D26<imp-def,dead>, %D27<imp-def,dead>, %D28<imp-def,dead>, %D29<imp-def,dead>, %D30<imp-def,dead>, %D31<imp-def,dead>, %CPSR<imp-def,dead>, %FPSCR<imp-def,dead>

Same example after:
  BL <ga:@bar>, %R0<imp-def>, %R1<imp-def,dead>, %LR<imp-def,dead>, %CPSR<imp-def,dead>, ...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86583 91177308-0d34-0410-b5e6-96231b3b80d8

Default-addressspace null pointers don't alias anything. This allows
GVN to be more aggressive. Patch by Hans Wennborg! (with a comment added by me)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86582 91177308-0d34-0410-b5e6-96231b3b80d8

Fix dependencies added to model memory aliasing for post-RA scheduling. The dependencies were overly conservative for memory access that are known not to alias.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86580 91177308-0d34-0410-b5e6-96231b3b80d8

The inbounds keyword isn't relevant to overindexing of
static array types. Thanks to Duncan for pointing this out!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86576 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a comment in a typo that Duncan noticed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86575 91177308-0d34-0410-b5e6-96231b3b80d8

Remove the "special case" for zero-length arrays, and rephrase this
paragraph to be more precise.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86572 91177308-0d34-0410-b5e6-96231b3b80d8

Generalize LCSSA to handle loops with exits with predecessors outside
the loop. This is needed because with indirectbr it may not be possible
for LoopSimplify to guarantee that all loop exit predecessors are
inside the loop. This fixes PR5437.

LCCSA no longer actually requires LoopSimplify form, but for now it
must still have the dependency because the PassManager doesn't know
how to schedule LoopSimplify otherwise.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86569 91177308-0d34-0410-b5e6-96231b3b80d8

Fix an 80-column violation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86567 91177308-0d34-0410-b5e6-96231b3b80d8

Minor tidiness fixes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86565 91177308-0d34-0410-b5e6-96231b3b80d8

Constify MachineFunctionAnalysis' TargetMachine reference.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86564 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86558 91177308-0d34-0410-b5e6-96231b3b80d8

Suppress implicit copy ctor and copy assignment for MachineFunction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86557 91177308-0d34-0410-b5e6-96231b3b80d8

Use ',' separation in XFAILs, lit doesn't evaluate them as regexs (easy to add,
but might as well use the more standard syntax).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86553 91177308-0d34-0410-b5e6-96231b3b80d8

add zextOrTrunc and sextOrTrunc methods, that are similar to the ones in APInt

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86549 91177308-0d34-0410-b5e6-96231b3b80d8

Work around assembler not recognizing #0.0 form immediate for vmcp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86548 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Detect gv, circo, twopi, neato, fdo, dot and dotty.

Patch by Arnaud Allard de Grandmaison!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86547 91177308-0d34-0410-b5e6-96231b3b80d8

Make lib/Support/Debug.cpp SetCurrentDebugType implementation part of llvm namespace to match function declaration in Debug.h.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86544 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR5149.

http://llvm.org/bugs/show_bug.cgi?id=5149

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86543 91177308-0d34-0410-b5e6-96231b3b80d8

make this handle redefinition of malloc with different prototype correctly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86525 91177308-0d34-0410-b5e6-96231b3b80d8

if a 'with overflow' intrinsic just has the normal result used, simplify
it to a normal binop.  Patch by Alastair Lynn, testcase by me.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86524 91177308-0d34-0410-b5e6-96231b3b80d8

Hide a couple of options.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86522 91177308-0d34-0410-b5e6-96231b3b80d8

80 col.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86521 91177308-0d34-0410-b5e6-96231b3b80d8

fix PR5104: when printing a single character, return the result of
putchar in case there is an error.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86515 91177308-0d34-0410-b5e6-96231b3b80d8

fix some bogus asserts, PR5049

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86514 91177308-0d34-0410-b5e6-96231b3b80d8

random tidy

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86511 91177308-0d34-0410-b5e6-96231b3b80d8

remove a redundant printout, LinkInArchive prints this as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86510 91177308-0d34-0410-b5e6-96231b3b80d8

enhance PHI slicing to handle the case when a slicable PHI is begin
used by a chain of other PHIs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86503 91177308-0d34-0410-b5e6-96231b3b80d8

Small cleanups.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86499 91177308-0d34-0410-b5e6-96231b3b80d8

Revert my previous patch to ABCD and fix things the right way.  There are two problems addressed
here:

1) We need to avoid processing sigma nodes as phi nodes for constraint generation.
2) We need to generate constraints for comparisons against constants properly.

This includes our first working ABCD test!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86498 91177308-0d34-0410-b5e6-96231b3b80d8

comment typos pointed out by Duncan

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86497 91177308-0d34-0410-b5e6-96231b3b80d8

Use Unified Assembly Syntax for the ARM backend.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86494 91177308-0d34-0410-b5e6-96231b3b80d8

Fix an issue where the ordering of blocks within a function could lead to different constraint
graphs being produced.  The cause was that we were incorrectly marking sigma instructions as
processed after handling the sigma-specific constraints for them, potentially neglecting to
process them as normal instructions as well.

Unfortunately, the testcase that inspired this still doesn't work because of a bug in the solver,
which is next on the list to debug.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86486 91177308-0d34-0410-b5e6-96231b3b80d8

Add a 'zkill' script, which is more-or-less a fancy (although not necessarily
very robust) version of killall. Because I like making shiny new wheels out of
spare parts.

For use by buildbots when people insist on making cc1 infinite loop. :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86484 91177308-0d34-0410-b5e6-96231b3b80d8

Teach an instcombine to not pull trunc instructions through PHI nodes
when both the source and dest are illegal types, since it would cause
the phi to grow (for example, we shouldn't transform test14b's phi to
a phi on i320).  This fixes an infinite loop on i686 bootstrap with
phi slicing turned on, so turn it back on.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86483 91177308-0d34-0410-b5e6-96231b3b80d8

Revert commit 81144, and add a comment.  It caused bugpoint timeouts
not to work any more on linux.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86481 91177308-0d34-0410-b5e6-96231b3b80d8

reapply r8644[3-5] with only the scary part
(SliceUpIllegalIntegerPHI) disabled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86480 91177308-0d34-0410-b5e6-96231b3b80d8

Speculatively revert r8644[3-5], they seem to be leading to infinite loops in
llvm-gcc bootstrap.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86478 91177308-0d34-0410-b5e6-96231b3b80d8

Add and-not (bic) patterns. Based heavily on patch by Brian Lucas!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86471 91177308-0d34-0410-b5e6-96231b3b80d8

Move OR patterns upper to all logical stuff. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86470 91177308-0d34-0410-b5e6-96231b3b80d8

Some nice peephole patterns. Based on patch by Brian Lucas!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86469 91177308-0d34-0410-b5e6-96231b3b80d8

Print tab before operand of jcc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86468 91177308-0d34-0410-b5e6-96231b3b80d8

Fix invalid operand updates & implement post-inc memory operands

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86466 91177308-0d34-0410-b5e6-96231b3b80d8

Throw an error when stack realignment stuff fails instead of silent
code miscompilation

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86463 91177308-0d34-0410-b5e6-96231b3b80d8

It is invalid to infer the value type from the result #0 of the node
since the instruction might use the other result of different type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86462 91177308-0d34-0410-b5e6-96231b3b80d8

Remove ByteswapSCANFResults, it is dead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86458 91177308-0d34-0410-b5e6-96231b3b80d8

NNT: Remove DejaGNU test from NewNightlyTest reports, this aspect of testing is
handled by buildbots now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86454 91177308-0d34-0410-b5e6-96231b3b80d8

Two small fixes for site.exp for cmake.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86453 91177308-0d34-0410-b5e6-96231b3b80d8

Derive the right paths to use during testing instead of passing it in via make.

Also, fix a few other details of the cmake test target and rename it to
'check'. CMake tests now work for the most part, but there are a handful of
failures left due to missing site.exp bits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86452 91177308-0d34-0410-b5e6-96231b3b80d8

Switch to using 'lit.site.cfg.in' for the site config template for Unit tests,
and generate it for CMake builds as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86451 91177308-0d34-0410-b5e6-96231b3b80d8

Cleanup some unused RUN lines.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86450 91177308-0d34-0410-b5e6-96231b3b80d8

lit: Hardcode whence seek value, os.SEEK_END isn't always available.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86449 91177308-0d34-0410-b5e6-96231b3b80d8

lit: Warn when a test suite contains no tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86448 91177308-0d34-0410-b5e6-96231b3b80d8

lit: Drop require_and_and support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86447 91177308-0d34-0410-b5e6-96231b3b80d8

Moved some ManagedStatics out of the SlotIndexes header.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86446 91177308-0d34-0410-b5e6-96231b3b80d8

another more interesting test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86445 91177308-0d34-0410-b5e6-96231b3b80d8

feature test for the new transformation in r86443

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86444 91177308-0d34-0410-b5e6-96231b3b80d8

teach a couple of instcombine transformations involving PHIs to
not turn a PHI in a legal type into a PHI of an illegal type, and
add a new optimization that breaks up insane integer PHI nodes into
small pieces (PR3451).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86443 91177308-0d34-0410-b5e6-96231b3b80d8

We don't need to byteswap, the interpreter assumes the program is running
native anyways. This fixes a crash using %d and similar in a scanf statement.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86440 91177308-0d34-0410-b5e6-96231b3b80d8

lit: Workaround a Win32/subprocess bug when appending.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86437 91177308-0d34-0410-b5e6-96231b3b80d8

lit: Preserve the PATHEXT variable when running subcommands, this is important on Win32

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86436 91177308-0d34-0410-b5e6-96231b3b80d8

Make TargetData::getStringRepresentation spit out native integer types,
this gives llvm-gcc generated modules the right data.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86435 91177308-0d34-0410-b5e6-96231b3b80d8

Remove test. Execution tests are slow and generally not worth it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86434 91177308-0d34-0410-b5e6-96231b3b80d8

Fix run line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86429 91177308-0d34-0410-b5e6-96231b3b80d8

Fix the interpreter to not crash due to zeroext/signext

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86428 91177308-0d34-0410-b5e6-96231b3b80d8

Prevent warning spew about -fPIC when using CMake generated Xcode project files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86427 91177308-0d34-0410-b5e6-96231b3b80d8

Use aligned load/store instructions for spilling Q registers when we know the stack slot is 128 bit aligned

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86425 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86423 91177308-0d34-0410-b5e6-96231b3b80d8

Fix CMake reporting of target triple.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86419 91177308-0d34-0410-b5e6-96231b3b80d8

Stop running get_target_triple more than we need to.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86418 91177308-0d34-0410-b5e6-96231b3b80d8

Fix MSVC warning ( | with bool and unsigned int).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86417 91177308-0d34-0410-b5e6-96231b3b80d8

Fix class -> struct tag.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86416 91177308-0d34-0410-b5e6-96231b3b80d8

x86 vector shuffle cleanup/fixes:

1. rename the movhp patfrag to movlhps, since thats what it actually matches
2. eliminate the bogus movhps load and store patterns, they were incorrect.  The load transforms are already handled (correctly) by shufps/unpack.
3. revert a recent test change to its correct form.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86415 91177308-0d34-0410-b5e6-96231b3b80d8

80-column cleanup of file header comments

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86408 91177308-0d34-0410-b5e6-96231b3b80d8

Support alignment specifier for NEON vld/vst instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86404 91177308-0d34-0410-b5e6-96231b3b80d8

Improve tail call elimination to handle the switch statement.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86403 91177308-0d34-0410-b5e6-96231b3b80d8

t2ldrpci_pic can be used for blockaddress as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86400 91177308-0d34-0410-b5e6-96231b3b80d8

temporarily remove these tests, as they are breaking in the buildbot,
Eric, please investigate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86399 91177308-0d34-0410-b5e6-96231b3b80d8

make instcombine only rewrite a chain of computation
(eliminating some extends) if the new type of the
computation is legal or if both the source and dest
are illegal.  This prevents instcombine from changing big
chains of computation into i64 on 32-bit targets for
example.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86398 91177308-0d34-0410-b5e6-96231b3b80d8

indicate what the native integer types for the target are.
Please verify.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86397 91177308-0d34-0410-b5e6-96231b3b80d8