oota-llvm.git
11 years agoRemove MachineInstr::setIsInsideBundle().
Jakob Stoklund Olesen [Tue, 18 Dec 2012 23:40:14 +0000 (23:40 +0000)]
Remove MachineInstr::setIsInsideBundle().

The bundle flags are now maintained by the slightly higher-level
functions bundleWithPred() / bundleWithSucc() which enforce consistent
bundle flags between neighboring instructions.

See also MIBundleBuilder for an even higher-level approach to building
bundles.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170475 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUse bidirectional bundle flags to simplify important functions.
Jakob Stoklund Olesen [Tue, 18 Dec 2012 23:21:49 +0000 (23:21 +0000)]
Use bidirectional bundle flags to simplify important functions.

The bundle_iterator::operator++ function now doesn't need to dig out the
basic block and check against end(). It can use the isBundledWithSucc()
flag to find the last bundled instruction safely.

Similarly, MachineInstr::isBundled() no longer needs to look at
iterators etc. It only has to look at flags.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170473 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agordar://12801297
Shuxin Yang [Tue, 18 Dec 2012 23:10:12 +0000 (23:10 +0000)]
rdar://12801297

 InstCombine for unsafe floating-point add/sub.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170471 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoEnable the loop vectorizer in clang and not in the pass manager, so that we can disab...
Nadav Rotem [Tue, 18 Dec 2012 23:09:44 +0000 (23:09 +0000)]
Enable the loop vectorizer in clang and not in the pass manager, so that we can disable it in clang.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170470 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoVerify bundle flag consistency when setting them.
Jakob Stoklund Olesen [Tue, 18 Dec 2012 23:00:28 +0000 (23:00 +0000)]
Verify bundle flag consistency when setting them.

Now that the bundle flag aware APIs are all in place, it is possible to
continuously verify the flag consistency.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170465 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoReverse order of checking SSE level when calculating compare cost, so we check
Jakub Staszak [Tue, 18 Dec 2012 22:57:56 +0000 (22:57 +0000)]
Reverse order of checking SSE level when calculating compare cost, so we check
AVX2 before AVX.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170464 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoVerify bundle flags for consistency in MachineVerifier.
Jakob Stoklund Olesen [Tue, 18 Dec 2012 22:55:07 +0000 (22:55 +0000)]
Verify bundle flags for consistency in MachineVerifier.

The new bidirectional bundle flags are redundant, so inadvertent bundle
tearing can be detected in the machine code verifier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170463 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDisable ARM partial flag dependency optimization at -Oz
Quentin Colombet [Tue, 18 Dec 2012 22:47:16 +0000 (22:47 +0000)]
Disable ARM partial flag dependency optimization at -Oz

To not over constrain the scheduler for ARM in thumb mode, some optimizations  for code size reduction, specific to ARM thumb, are blocked when they add a dependency (like write after read dependency).

Disables this check when code size is the priority, i.e., code is compiled with -Oz.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170462 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDon't allow the automatically updated MI flags to be set directly.
Jakob Stoklund Olesen [Tue, 18 Dec 2012 21:36:05 +0000 (21:36 +0000)]
Don't allow the automatically updated MI flags to be set directly.

The bundle-related MI flags need to be kept in sync with the neighboring
instructions. Don't allow the bulk flag-setting setFlags() function to
change them.

Also don't copy MI flags when cloning an instruction. The clone's bundle
flags will be set when it is explicitly inserted into a bundle.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170459 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoTighten up the splice() API for bundled instructions.
Jakob Stoklund Olesen [Tue, 18 Dec 2012 20:59:41 +0000 (20:59 +0000)]
Tighten up the splice() API for bundled instructions.

Remove the instr_iterator versions of the splice() functions. It doesn't
seem useful to be able to splice sequences of instructions that don't
consist of full bundles.

The normal splice functions that take MBB::iterator arguments are not
changed, and they can move whole bundles around without any problems.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170456 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMISched: add dependence to ExitSU to model live-out latency.
Andrew Trick [Tue, 18 Dec 2012 20:53:01 +0000 (20:53 +0000)]
MISched: add dependence to ExitSU to model live-out latency.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170454 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMISched: Cleanup, redundant statement.
Andrew Trick [Tue, 18 Dec 2012 20:52:58 +0000 (20:52 +0000)]
MISched: Cleanup, redundant statement.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170453 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMISched: Heuristics, compare latency more precisely. It matters more for some targets.
Andrew Trick [Tue, 18 Dec 2012 20:52:56 +0000 (20:52 +0000)]
MISched: Heuristics, compare latency more precisely. It matters more for some targets.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170452 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMISched: Remove SchedRemainder::IsResourceLimited. I don't know how to compute it.
Andrew Trick [Tue, 18 Dec 2012 20:52:54 +0000 (20:52 +0000)]
MISched: Remove SchedRemainder::IsResourceLimited. I don't know how to compute it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170451 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMISched: cleanup, use the proper iterator type.
Andrew Trick [Tue, 18 Dec 2012 20:52:52 +0000 (20:52 +0000)]
MISched: cleanup, use the proper iterator type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170450 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMISched: minor improvement, initialize remaining resources before the first schedulin...
Andrew Trick [Tue, 18 Dec 2012 20:52:49 +0000 (20:52 +0000)]
MISched: minor improvement, initialize remaining resources before the first scheduling decision.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170449 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd an assertion for a likely ilist::splice() contract violation.
Jakob Stoklund Olesen [Tue, 18 Dec 2012 19:28:37 +0000 (19:28 +0000)]
Add an assertion for a likely ilist::splice() contract violation.

The single-element ilist::splice() function supports a noop move:

  List.splice(I, List, I);

The corresponding std::list function doesn't allow that, so add a unit
test to document that behavior.

This also means that

  List.splice(I, List, F);

is somewhat surprisingly not equivalent to

  List.splice(I, List, F, next(F));

This patch adds an assertion to catch the illegal case I == F above.
Alternatively, we could make I == F a legal noop, but that would make
ilist differ even more from std::list.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170443 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLoopVectorize: Emit reductions as log2(vectorsize) shuffles + vector ops instead...
Benjamin Kramer [Tue, 18 Dec 2012 18:40:20 +0000 (18:40 +0000)]
LoopVectorize: Emit reductions as log2(vectorsize) shuffles + vector ops instead of scalar operations.

For example on x86 with SSE4.2 a <8 x i8> add reduction becomes
movdqa %xmm0, %xmm1
movhlps %xmm1, %xmm1            ## xmm1 = xmm1[1,1]
paddw %xmm0, %xmm1
pshufd $1, %xmm1, %xmm0        ## xmm0 = xmm1[1,0,0,0]
paddw %xmm1, %xmm0
phaddw %xmm0, %xmm0
pextrb $0, %xmm0, %edx

instead of
pextrb $2, %xmm0, %esi
pextrb $0, %xmm0, %edx
addb %sil, %dl
pextrb $4, %xmm0, %esi
addb %dl, %sil
pextrb $6, %xmm0, %edx
addb %sil, %dl
pextrb $8, %xmm0, %esi
addb %dl, %sil
pextrb $10, %xmm0, %edi
pextrb $14, %xmm0, %edx
addb %sil, %dil
pextrb $12, %xmm0, %esi
addb %dil, %sil
addb %sil, %dl

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170439 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoGet rid of the pesky -Woverloaded-virtual warning. No change in functionality.
Eli Bendersky [Tue, 18 Dec 2012 18:21:29 +0000 (18:21 +0000)]
Get rid of the pesky -Woverloaded-virtual warning. No change in functionality.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170438 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoTighten the insert() API for bundled instructions.
Jakob Stoklund Olesen [Tue, 18 Dec 2012 17:54:53 +0000 (17:54 +0000)]
Tighten the insert() API for bundled instructions.

The normal insert() function takes an MBB::iterator position, and
inserts a stand-alone MachineInstr as before.

The insert() function that takes an MBB::instr_iterator position can
insert instructions inside a bundle, and will now update the bundle
flags correctly when that happens.

When the insert position is between two bundles, it is unclear whether
the instruction should be appended to the previous bundle, prepended to
the next bundle, or stand on its own. The MBB::insert() function doesn't
bundle the instruction in that case, use the MIBundleBuilder class for
that.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170437 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoCheck multiple register classes for inline asm tied registers
Hal Finkel [Tue, 18 Dec 2012 17:50:58 +0000 (17:50 +0000)]
Check multiple register classes for inline asm tied registers

A register can be associated with several distinct register classes.
For example, on PPC, the floating point registers are each associated with
both F4RC (which holds f32) and F8RC (which holds f64). As a result, this code
would fail when provided with a floating point register and an f64 operand
because it would happen to find the register in the F4RC class first and
return that. From the F4RC class, SDAG would extract f32 as the register
type and then assert because of the invalid implied conversion between
the f64 value and the f32 register.

Instead, search all register classes. If a register class containing the
the requested register has the requested type, then return that register
class. Otherwise, as before, return the first register class found that
contains the requested register.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170436 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoEnable the loop vectorizer.
Nadav Rotem [Tue, 18 Dec 2012 06:37:12 +0000 (06:37 +0000)]
Enable the loop vectorizer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170416 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRename the test so that we can add additional vectors-of-pointers tests
Nadav Rotem [Tue, 18 Dec 2012 05:50:54 +0000 (05:50 +0000)]
Rename the test so that we can add additional vectors-of-pointers tests
into the same file in the future.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170414 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSROA: Replace calls to getScalarSizeInBits to DataLayout's API because
Nadav Rotem [Tue, 18 Dec 2012 05:23:31 +0000 (05:23 +0000)]
SROA: Replace calls to getScalarSizeInBits to DataLayout's API because
getScalarSizeInBits could not handle vectors of pointers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170412 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agollvm/test/MC/ELF/comp-dir.s: Appease MSYS Bash.
NAKAMURA Takumi [Tue, 18 Dec 2012 05:08:12 +0000 (05:08 +0000)]
llvm/test/MC/ELF/comp-dir.s: Appease MSYS Bash.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170410 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoInitialize NoRedZone and remove unused default values.
Rafael Espindola [Tue, 18 Dec 2012 03:35:05 +0000 (03:35 +0000)]
Initialize NoRedZone and remove unused default values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170404 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoCleanup comment and formatting
Eli Bendersky [Tue, 18 Dec 2012 00:53:36 +0000 (00:53 +0000)]
Cleanup comment and formatting

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170398 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRepair bundles that were broken by removing and reinserting the first
Jakob Stoklund Olesen [Tue, 18 Dec 2012 00:46:39 +0000 (00:46 +0000)]
Repair bundles that were broken by removing and reinserting the first
instruction.

This isn't strictly necessary at the moment because Thumb2SizeReduction
also copies all MI flags from the old instruction to the new. However, a
future patch will make that kind of direct flag tampering illegal.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170395 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFormatting.
Eric Christopher [Tue, 18 Dec 2012 00:42:26 +0000 (00:42 +0000)]
Formatting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170394 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd support for passing -main-file-name all the way through to
Eric Christopher [Tue, 18 Dec 2012 00:31:01 +0000 (00:31 +0000)]
Add support for passing -main-file-name all the way through to
the assembler.

Part of PR14624

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170390 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoCleanup formatting and whitespace.
Eric Christopher [Tue, 18 Dec 2012 00:30:54 +0000 (00:30 +0000)]
Cleanup formatting and whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170389 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoExtract a method, no functional change intended.
Jakob Stoklund Olesen [Tue, 18 Dec 2012 00:13:11 +0000 (00:13 +0000)]
Extract a method, no functional change intended.

Sadly, this costs us a perfectly good opportunity to use 'goto'.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170385 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoTighten up the erase/remove API for bundled instructions.
Jakob Stoklund Olesen [Mon, 17 Dec 2012 23:55:38 +0000 (23:55 +0000)]
Tighten up the erase/remove API for bundled instructions.

Most code is oblivious to bundles and uses the MBB::iterator which only
visits whole bundles. MBB::erase() operates on whole bundles at a time
as before.

MBB::remove() now refuses to remove bundled instructions. It is not safe
to remove all instructions in a bundle without deleting them since there
is no way of returning pointers to all the removed instructions.

MBB::remove_instr() and MBB::erase_instr() will now update bundle flags
correctly, lifting individual instructions out of bundles while leaving
the remaining bundle intact.

The MachineInstr convenience functions are updated so

  eraseFromParent() erases a whole bundle as before
  eraseFromBundle() erases a single instruction, leaving the rest of its bundle.
  removeFromParent() refuses to operate on bundled instructions, and
  removeFromBundle() lifts a single instruction out of its bundle.

These functions will no longer accidentally split or coalesce bundles -
bundle flags are updated to preserve the existing bundling, and explicit
bundleWith* / unbundleFrom* functions should be used to change the
instruction bundling.

This API update is still a work in progress. I am going to update APIs
first so they maintain bundle flags automatically when possible. Then
I'll add stricter verification of the bundle flags.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170384 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoEmitDebugLabel should by default be the same as EmitLabel everywhere.
Reed Kotler [Mon, 17 Dec 2012 23:41:45 +0000 (23:41 +0000)]
EmitDebugLabel should by default be the same as EmitLabel everywhere.
It must be explicity set in MCPureStreamer because otherwise it will
inherit incorrectly from the parent.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170383 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agofix indentation
Eli Bendersky [Mon, 17 Dec 2012 22:50:56 +0000 (22:50 +0000)]
fix indentation

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170381 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[arm fast-isel] Minor cleanup. No functional change intended.
Chad Rosier [Mon, 17 Dec 2012 22:35:29 +0000 (22:35 +0000)]
[arm fast-isel] Minor cleanup. No functional change intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170379 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix some integer constant warnings by using a suffix
Nick Kledzik [Mon, 17 Dec 2012 22:11:17 +0000 (22:11 +0000)]
Fix some integer constant warnings by using a suffix

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170376 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd a triple to this test -- it has to be an ELF platform...
Chandler Carruth [Mon, 17 Dec 2012 21:44:50 +0000 (21:44 +0000)]
Add a triple to this test -- it has to be an ELF platform...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170374 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoPrepare LLVM to fix PR14625, exposing a hook in MCContext to manage the
Chandler Carruth [Mon, 17 Dec 2012 21:32:42 +0000 (21:32 +0000)]
Prepare LLVM to fix PR14625, exposing a hook in MCContext to manage the
compilation directory.

This defaults to the current working directory, just as it always has,
but now an assembler can choose to override it with a custom directory.
I've taught llvm-mc about this option and added a test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170371 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agore-enable test cases now that traits work with g++. Fix some g++ warnings
Nick Kledzik [Mon, 17 Dec 2012 20:43:53 +0000 (20:43 +0000)]
re-enable test cases now that traits work with g++.  Fix some g++ warnings

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170369 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove trailing whitespace
Michael Ilseman [Mon, 17 Dec 2012 20:40:14 +0000 (20:40 +0000)]
Remove trailing whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170368 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemoved trailing whitespace
Michael Ilseman [Mon, 17 Dec 2012 20:37:55 +0000 (20:37 +0000)]
Removed trailing whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170367 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[arm fast-isel] Fast-isel only handles simple VTs, so make sure the necessary
Chad Rosier [Mon, 17 Dec 2012 19:59:43 +0000 (19:59 +0000)]
[arm fast-isel] Fast-isel only handles simple VTs, so make sure the necessary
checks are in place.  Some minor cleanup as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170360 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUse different trait techniques to be compatible with g++
Nick Kledzik [Mon, 17 Dec 2012 19:02:05 +0000 (19:02 +0000)]
Use different trait techniques to be compatible with g++

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170355 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix another SROA crasher, PR14601.
Chandler Carruth [Mon, 17 Dec 2012 18:48:07 +0000 (18:48 +0000)]
Fix another SROA crasher, PR14601.

This was a silly oversight, we weren't pruning allocas which were used
by variable-length memory intrinsics from the set that could be widened
and promoted as integers. Fix that.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170353 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoQuery section for whether it should be executable.
Tim Northover [Mon, 17 Dec 2012 17:59:35 +0000 (17:59 +0000)]
Query section for whether it should be executable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170350 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoTeach MachO which sections contain code
Tim Northover [Mon, 17 Dec 2012 17:59:32 +0000 (17:59 +0000)]
Teach MachO which sections contain code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170349 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[msan] Fix lint warning.
Evgeniy Stepanov [Mon, 17 Dec 2012 16:30:05 +0000 (16:30 +0000)]
[msan] Fix lint warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170347 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd instruction encodings / disassembly support for l2r instructions.
Richard Osborne [Mon, 17 Dec 2012 16:28:02 +0000 (16:28 +0000)]
Add instruction encodings / disassembly support for l2r instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170345 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: enable S_*N2_* instructions
Tom Stellard [Mon, 17 Dec 2012 15:14:56 +0000 (15:14 +0000)]
R600: enable S_*N2_* instructions

They seem to work fine.

Patch by: Christian König

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Tested-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Christian König <deathsimple@vodafone.de>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170343 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: BB operand support for SI
Tom Stellard [Mon, 17 Dec 2012 15:14:54 +0000 (15:14 +0000)]
R600: BB operand support for SI

Patch by: Christian König

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Tested-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Christian König <deathsimple@vodafone.de>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170342 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoR600: remove nonsense setPrefLoopAlignment
Tom Stellard [Mon, 17 Dec 2012 15:14:53 +0000 (15:14 +0000)]
R600: remove nonsense setPrefLoopAlignment

The Align parameter is a power of two, so 16 results in 64K
alignment. Additional to that even 16 byte alignment doesn't
make any sense, so just remove it.

Patch by: Christian König

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Tested-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Christian König <deathsimple@vodafone.de>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170341 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoTeach the rewriting of memcpy calls to support subvector copies.
Chandler Carruth [Mon, 17 Dec 2012 14:51:24 +0000 (14:51 +0000)]
Teach the rewriting of memcpy calls to support subvector copies.

This also cleans up a bit of the memcpy call rewriting by sinking some
irrelevant code further down and making the call-emitting code a bit
more concrete.

Previously, memcpy of a subvector would actually miscompile (!!!) the
copy into a single vector element copy. I have no idea how this ever
worked. =/ This is the memcpy half of PR14478 which we probably weren't
noticing previously because it didn't actually assert.

The rewrite relies on the newly refactored insert- and extractVector
functions to do the heavy lifting, and those are the same as used for
loads and stores which makes the test coverage a bit more meaningful
here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170338 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRevert/correct some FastISel changes in r170104 (EVT->MVT for
Patrik Hagglund [Mon, 17 Dec 2012 14:30:06 +0000 (14:30 +0000)]
Revert/correct some FastISel changes in r170104 (EVT->MVT for
TargetLowering::getRegClassFor).

Some isSimple() guards were missing, or getSimpleVT() were hoisted too
far, resulting in asserts on valid LLVM assembly input.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170336 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoOptimize tree walking in markAliveBlocks.
Evgeniy Stepanov [Mon, 17 Dec 2012 14:28:00 +0000 (14:28 +0000)]
Optimize tree walking in markAliveBlocks.

Check whether a BB is known as reachable before adding it to the worklist.
This way BB's with multiple predecessors are added to the list no more than
once.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170335 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd instruction encodings for PEEK and ENDIN.
Richard Osborne [Mon, 17 Dec 2012 14:23:54 +0000 (14:23 +0000)]
Add instruction encodings for PEEK and ENDIN.

Previously these were marked with the wrong format.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170334 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix a secondary bug I introduced while fixing the first part of PR14478.
Chandler Carruth [Mon, 17 Dec 2012 14:03:01 +0000 (14:03 +0000)]
Fix a secondary bug I introduced while fixing the first part of PR14478.

The first half of fixing this bug was actually in r170328, but was
entirely coincidental. It did however get me to realize the nature of
the bug, and adapt the test case to test more interesting behavior. In
turn, that uncovered the rest of the bug which I've fixed here.

This should fix two new asserts that showed up in the vectorize nightly
tester.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170333 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix parameter name in prototypes in XCoreDisassembler.
Richard Osborne [Mon, 17 Dec 2012 13:55:49 +0000 (13:55 +0000)]
Fix parameter name in prototypes in XCoreDisassembler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170332 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoHoist a convertValue call to the two paths where it is needed.
Chandler Carruth [Mon, 17 Dec 2012 13:51:03 +0000 (13:51 +0000)]
Hoist a convertValue call to the two paths where it is needed.

I noticed this while looking at r170328. We only ever do a vector
rewrite when the alloca *is* the vector type, so it's good to not paper
over bugs here by doing a convertValue that isn't needed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170331 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd instruction encodings / disassembly support for rus instructions.
Richard Osborne [Mon, 17 Dec 2012 13:50:04 +0000 (13:50 +0000)]
Add instruction encodings / disassembly support for rus instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170330 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoHoist the insertVector helper to be a static helper.
Chandler Carruth [Mon, 17 Dec 2012 13:41:21 +0000 (13:41 +0000)]
Hoist the insertVector helper to be a static helper.

This will allow its use inside of memcpy rewriting as well. This routine
is more complex than extractVector, and some of its uses are not 100%
where I want them to be so there is still some work to do here.

While this can technically change the output in some cases, it shouldn't
be a change that matters -- IE, it can leave some dead code lying around
that prior versions did not, etc.

Yet another step in the refactorings leading up to the solution to the
last component of PR14478.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170328 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd instruction encodings for ZEXT and SEXT.
Richard Osborne [Mon, 17 Dec 2012 13:20:37 +0000 (13:20 +0000)]
Add instruction encodings for ZEXT and SEXT.

Previously these were marked with the wrong format.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170327 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLift the extractVector helper all the way out to a static helper function.
Chandler Carruth [Mon, 17 Dec 2012 13:07:30 +0000 (13:07 +0000)]
Lift the extractVector helper all the way out to a static helper function.

The method helpers all implicitly act upon the alloca, and what we
really want is a fully generic helper. Doing memcpy rewrites is more
special than all other rewrites because we are at times rewriting
instructions which touch pointers *other* than the alloca. As
a consequence all of the helpers needed by memcpy rewriting of
sub-vector copies will need to be generalized fully.

Note that all of these helpers ({insert,extract}{Integer,Vector}) are
woefully uncommented. I'm going to go back through and document them
once I get the factoring correct.

No functionality changed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170325 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFactor the vector load rewriting into a more generic form.
Chandler Carruth [Mon, 17 Dec 2012 12:50:21 +0000 (12:50 +0000)]
Factor the vector load rewriting into a more generic form.

This makes it suitable for use in rewriting memcpy in the presence of
subvector memcpy intrinsics.

No functionality changed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170324 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd instruction encodings / disassembly support for 2r instructions.
Richard Osborne [Mon, 17 Dec 2012 12:29:31 +0000 (12:29 +0000)]
Add instruction encodings / disassembly support for 2r instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170323 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd instruction encodings / disassembly support for 0r instructions.
Richard Osborne [Mon, 17 Dec 2012 12:26:29 +0000 (12:26 +0000)]
Add instruction encodings / disassembly support for 0r instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170322 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSimplify assertion in XCoreInstPrinter.
Richard Osborne [Mon, 17 Dec 2012 12:13:46 +0000 (12:13 +0000)]
Simplify assertion in XCoreInstPrinter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170321 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUpdate comments to match recommended doxygen style.
Richard Osborne [Mon, 17 Dec 2012 12:13:41 +0000 (12:13 +0000)]
Update comments to match recommended doxygen style.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170320 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove unnecessary include.
Richard Osborne [Mon, 17 Dec 2012 12:13:32 +0000 (12:13 +0000)]
Remove unnecessary include.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170319 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix typo that results in new landing pads not getting a name, fixing PR14617.
Duncan Sands [Mon, 17 Dec 2012 12:02:36 +0000 (12:02 +0000)]
Fix typo that results in new landing pads not getting a name, fixing PR14617.
Patch by Chris Toshok.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170318 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix comment typo.
Duncan Sands [Mon, 17 Dec 2012 11:43:15 +0000 (11:43 +0000)]
Fix comment typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170317 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove EFLAGS from the BLSI/BLSMSK/BLSR patterns. The nodes created by DAG combine...
Craig Topper [Mon, 17 Dec 2012 06:13:48 +0000 (06:13 +0000)]
Remove EFLAGS from the BLSI/BLSMSK/BLSR patterns. The nodes created by DAG combine don't contain an EFLAGS def.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170308 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSimplify BMI ANDN matching to use patterns instead of a DAG combine. Also add ANDN...
Craig Topper [Mon, 17 Dec 2012 05:12:30 +0000 (05:12 +0000)]
Simplify BMI ANDN matching to use patterns instead of a DAG combine. Also add ANDN to isDefConvertible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170305 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd rest of BMI/BMI2 instructions to the folding tables as well as popcnt and lzcnt.
Craig Topper [Mon, 17 Dec 2012 05:02:29 +0000 (05:02 +0000)]
Add rest of BMI/BMI2 instructions to the folding tables as well as popcnt and lzcnt.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170304 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove store forms of DEC/INC from isDefConvertible. Since they are stores they don...
Craig Topper [Mon, 17 Dec 2012 04:55:07 +0000 (04:55 +0000)]
Remove store forms of DEC/INC from isDefConvertible. Since they are stores they don't have a register def.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170303 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix the first part of PR14478: memset now works.
Chandler Carruth [Mon, 17 Dec 2012 04:07:37 +0000 (04:07 +0000)]
Fix the first part of PR14478: memset now works.

PR14478 highlights a serious problem in SROA that simply wasn't being
exercised due to a lack of vector input code mixed with C-library
function calls. Part of SROA was written carefully to handle subvector
accesses via memset and memcpy, but the rewriter never grew support for
this. Fixing it required refactoring the subvector access code in other
parts of SROA so it could be shared, and then fixing the splat formation
logic and using subvector insertion (this patch).

The PR isn't quite fixed yet, as memcpy is still broken in the same way.
I'm starting on that series of patches now.

Hopefully this will be enough to bring the bullet benchmark back to life
with the bb-vectorizer enabled, but that may require fixing memcpy as
well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170301 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoExtract the logic for inserting a subvector into a vector alloca.
Chandler Carruth [Mon, 17 Dec 2012 04:07:35 +0000 (04:07 +0000)]
Extract the logic for inserting a subvector into a vector alloca.

No functionality changed. Another step of refactoring toward solving
PR14487.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170300 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLift the integer splat computation into a helper function.
Chandler Carruth [Mon, 17 Dec 2012 04:07:30 +0000 (04:07 +0000)]
Lift the integer splat computation into a helper function.

No functionality changed. Refactoring leading up to the fix for PR14478
which requires some significant changes to the memset and memcpy
rewriting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170299 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd debug prints for when optimizeLoadInstr folds a load.
Craig Topper [Mon, 17 Dec 2012 03:56:00 +0000 (03:56 +0000)]
Add debug prints for when optimizeLoadInstr folds a load.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170298 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd tests for disassembly of 1r XCore instructions.
Richard Osborne [Sun, 16 Dec 2012 18:06:30 +0000 (18:06 +0000)]
Add tests for disassembly of 1r XCore instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170295 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd instruction encodings and disassembly for 1r instructions.
Richard Osborne [Sun, 16 Dec 2012 17:37:34 +0000 (17:37 +0000)]
Add instruction encodings and disassembly for 1r instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170293 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd XCore disassembler.
Richard Osborne [Sun, 16 Dec 2012 17:29:14 +0000 (17:29 +0000)]
Add XCore disassembler.

Currently there is no instruction encoding info and
XCoreDisassembler::getInstruction() always returns Fail. I intend to add
instruction encodings and tests in follow on commits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170292 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove invalid instruction encodings.
Richard Osborne [Sun, 16 Dec 2012 16:46:31 +0000 (16:46 +0000)]
Remove invalid instruction encodings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170291 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMark anything deriving from PseudoInstXCore as a pseudo instruction.
Richard Osborne [Sun, 16 Dec 2012 16:46:28 +0000 (16:46 +0000)]
Mark anything deriving from PseudoInstXCore as a pseudo instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170290 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSet instruction size correctly in XCoreInstrFormats.td
Richard Osborne [Sun, 16 Dec 2012 16:46:24 +0000 (16:46 +0000)]
Set instruction size correctly in XCoreInstrFormats.td

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170289 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoChange XCoreAsmPrinter to lower MachineInstrs to MCInsts before emission.
Richard Osborne [Sun, 16 Dec 2012 16:20:48 +0000 (16:20 +0000)]
Change XCoreAsmPrinter to lower MachineInstrs to MCInsts before emission.

This change adds XCoreMCInstLower to do the lowering to MCInst and
XCoreInstPrinter to print the MCInsts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170288 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoReplace ${:comment} with the comment symbol.
Richard Osborne [Sun, 16 Dec 2012 15:59:02 +0000 (15:59 +0000)]
Replace ${:comment} with the comment symbol.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170286 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDeclare class DwarfDebug before use instead of relying on a forward declaration
Dmitri Gribenko [Sun, 16 Dec 2012 12:57:36 +0000 (12:57 +0000)]
Declare class DwarfDebug before use instead of relying on a forward declaration
from some other unrelated header.

Patch by Kai.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170284 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMCPureStreamer.cpp: Try to fix build, pruning EmitDebugLabel().
NAKAMURA Takumi [Sun, 16 Dec 2012 04:23:20 +0000 (04:23 +0000)]
MCPureStreamer.cpp: Try to fix build, pruning EmitDebugLabel().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170280 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoThis patch is needed to make c++ exceptions work for mips16.
Reed Kotler [Sun, 16 Dec 2012 04:00:45 +0000 (04:00 +0000)]
This patch is needed to make c++ exceptions work for mips16.

Mips16 is really a processor decoding mode (ala thumb 1) and in the same
program, mips16 and mips32 functions can exist and can call each other.

If a jal type instruction encounters an address with the lower bit set, then
the processor switches to mips16 mode (if it is not already in it). If the
lower bit is not set, then it switches to mips32 mode.

The linker knows which functions are mips16 and which are mips32.
When relocation is performed on code labels, this lower order bit is
set if the code label is a mips16 code label.

In general this works just fine, however when creating exception handling
tables and dwarf, there are cases where you don't want this lower order
bit added in.

This has been traditionally distinguished in gas assembly source by using a
different syntax for the label.

lab1:      ; this will cause the lower order bit to be added
lab2=.     ; this will not cause the lower order bit to be added

In some cases, it does not matter because in dwarf and debug tables
the difference of two labels is used and in that case the lower order
bits subtract each other out.

To fix this, I have added to mcstreamer the notion of a debuglabel.
The default is for label and debug label to be the same. So calling
EmitLabel and EmitDebugLabel produce the same result.

For various reasons, there is only one set of labels that needs to be
modified for the mips exceptions to work. These are the "$eh_func_beginXXX"
labels.

Mips overrides the debug label suffix from ":" to "=." .

This initial patch fixes exceptions. More changes most likely
will be needed to DwarfCFException to make all of this work
for actual debugging. These changes will be to emit debug labels in some
places where a simple label is emitted now.

Some historical discussion on this from gcc can be found at:
http://gcc.gnu.org/ml/gcc-patches/2008-08/msg00623.html
http://gcc.gnu.org/ml/gcc-patches/2008-11/msg01273.html

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170279 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoX86: Add a couple of target-specific dag combines that turn VSELECTS into psubus...
Benjamin Kramer [Sat, 15 Dec 2012 16:47:44 +0000 (16:47 +0000)]
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible.

We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases
if y is a constant. DAGCombiner canonicalizes those so we first have to undo the
canonicalization for those cases. The pattern occurs in gzip when the loop
vectorizer is enabled. Part of PR14613.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170273 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd a corollary test for PR14572. We got this code path correct already.
Chandler Carruth [Sat, 15 Dec 2012 09:31:54 +0000 (09:31 +0000)]
Add a corollary test for PR14572. We got this code path correct already.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170271 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRelax an overly aggressive assert to fix PR14572.
Chandler Carruth [Sat, 15 Dec 2012 09:26:06 +0000 (09:26 +0000)]
Relax an overly aggressive assert to fix PR14572.

The alloca width is based on the alloc size, not the type size.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170270 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMake '-mtune=x86_64' assume fast unaligned memory accesses.
Chandler Carruth [Sat, 15 Dec 2012 09:01:13 +0000 (09:01 +0000)]
Make '-mtune=x86_64' assume fast unaligned memory accesses.

Not all chips targeted by x86_64 have this feature, but a dramatically
increasing number do. Specifying a chip-specific tuning parameter will
continue to turn the feature on or off as appropriate for that
particular chip, but the generic flag should try to achieve the best
performance on the most widely available hardware. Today, the number of
chips with fast UA access dwarfs those without in the x86-64 space.

Note that this also brings LLVM's code generation for this '-march' flag
more in line with that of modern GCCs. Reviewed by Dan Gohman.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170269 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoActually update the grammar of this sentence to reflect the removal of CellSPU.
Chandler Carruth [Sat, 15 Dec 2012 08:56:20 +0000 (08:56 +0000)]
Actually update the grammar of this sentence to reflect the removal of CellSPU.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170268 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRevert r170246, "Enable the loop vectorizer by default."
NAKAMURA Takumi [Sat, 15 Dec 2012 06:11:13 +0000 (06:11 +0000)]
Revert r170246, "Enable the loop vectorizer by default."

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170267 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoThis code implements most of mips16 hardfloat as it is done by gcc.
Reed Kotler [Sat, 15 Dec 2012 00:20:05 +0000 (00:20 +0000)]
This code implements most of mips16 hardfloat as it is done by gcc.
In this case, essentially it is soft float with different library routines.
The next step will be to make this fully interoperational with mips32 floating
point and that requires creating stubs for functions with signatures that
contain floating point types.

I have a more sophisticated design for mips16 hardfloat which I hope to
implement at a later time that directly does floating point without the need
for function calls.

The mips16 encoding has no floating point instructions so one needs to
switch to mips32 mode to execute floating point instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170259 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoTo simplify some code move the unit emission into the holders.
Eric Christopher [Sat, 15 Dec 2012 00:04:07 +0000 (00:04 +0000)]
To simplify some code move the unit emission into the holders.
Make emitDIE public accordingly. No functional change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170258 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUse begin and end label names from the section for info.
Eric Christopher [Sat, 15 Dec 2012 00:04:04 +0000 (00:04 +0000)]
Use begin and end label names from the section for info.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170257 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMake sure the alternate PC+imm syntax of LDR instruction with a small
Kevin Enderby [Fri, 14 Dec 2012 23:04:25 +0000 (23:04 +0000)]
Make sure the alternate PC+imm syntax of LDR instruction with a small
immediate generates the narrow version.  Needed when doing round-trip
assemble/disassemble testing using the alternate syntax that specifies
'pc' directly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170255 91177308-0d34-0410-b5e6-96231b3b80d8