oota-llvm.git
8 years ago[DWARF] Add CIE header fields address_size and segment_size when generating dwarf-4
Keith Walker [Tue, 12 May 2015 15:25:08 +0000 (15:25 +0000)]
[DWARF] Add CIE header fields address_size and segment_size when generating dwarf-4

The DWARF-4 specification added 2 new fields in the CIE header called
address_size and segment_size.
Create these 2 new fields when generating dwarf-4 CIE entries, print out
the new fields when dumping the CIE and update tests

Differential Revision: http://reviews.llvm.org/D9558

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237145 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agouse 'auto' to improve readability; NFC
Sanjay Patel [Tue, 12 May 2015 15:15:55 +0000 (15:15 +0000)]
use 'auto' to improve readability; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237144 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoR600/SI: Update tablegen defs to avoid restoring spilled sgprs to m0
Tom Stellard [Tue, 12 May 2015 15:00:53 +0000 (15:00 +0000)]
R600/SI: Update tablegen defs to avoid restoring spilled sgprs to m0

We had code to do this in SIRegisterInfo::eliminateFrameIndex(), but
it is easier to just change the definition of SI_SPILL_S32_RESTORE to
only allow numbered sgprs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237143 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoR600/SI: Remove M0Reg register class
Tom Stellard [Tue, 12 May 2015 15:00:52 +0000 (15:00 +0000)]
R600/SI: Remove M0Reg register class

It is no longer used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237142 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoR600/SI: Remove explicit m0 operand from DS instructions
Tom Stellard [Tue, 12 May 2015 15:00:49 +0000 (15:00 +0000)]
R600/SI: Remove explicit m0 operand from DS instructions

Instead add m0 as an implicit operand.  This helps avoid spills
of the m0 register in some cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237141 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoR600/SI: Remove explicit m0 operand from v_interp instructions
Tom Stellard [Tue, 12 May 2015 15:00:46 +0000 (15:00 +0000)]
R600/SI: Remove explicit m0 operand from v_interp instructions

Instead add m0 as an implicit operand.  This helps avoid spills
of the m0 register in some cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237140 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agofix typos, grammar; NFC
Sanjay Patel [Tue, 12 May 2015 14:52:42 +0000 (14:52 +0000)]
fix typos, grammar; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237139 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoR600/SI: Make sendmsg test more strict
Tom Stellard [Tue, 12 May 2015 14:18:16 +0000 (14:18 +0000)]
R600/SI: Make sendmsg test more strict

We want to make sure that the m0 copies are being cse'd.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237134 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoR600/SI: Remove explicit m0 operand from s_sendmsg
Tom Stellard [Tue, 12 May 2015 14:18:14 +0000 (14:18 +0000)]
R600/SI: Remove explicit m0 operand from s_sendmsg

Instead add m0 as an implicit operand.  This allows us to avoid using
the M0Reg register class and eliminates a number of unnecessary spills
when using s_sendmsg instructions.  This impacts one shader in the
shader-db:

SGPRS: 48 -> 40 (-16.67 %)
VGPRS: 112 -> 108 (-3.57 %)
Code Size: 40132 -> 38796 (-3.33 %) bytes
LDS: 0 -> 0 (0.00 %) blocks
Scratch: 2048 -> 0 (-100.00 %) bytes per wave

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237133 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoR600/SI: Replace TRI->getRegClass(Reg) with TRI->getPhysRegClass(Reg)
Tom Stellard [Tue, 12 May 2015 14:18:11 +0000 (14:18 +0000)]
R600/SI: Replace TRI->getRegClass(Reg) with TRI->getPhysRegClass(Reg)

TRI->getRegClass() takes a register class ID, not a register.  We were
using this incorrectly in a few places.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237132 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAVX-512, X86: Added lowering for shift operations for SKX.
Elena Demikhovsky [Tue, 12 May 2015 13:25:46 +0000 (13:25 +0000)]
AVX-512, X86: Added lowering for shift operations for SKX.
The other changes in the LowerShift() are not functional,
just to make the code more convenient.
So, the functional changes for SKX only.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237129 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[ARM] Use AEABI aligned function variants
John Brawn [Tue, 12 May 2015 13:13:38 +0000 (13:13 +0000)]
[ARM] Use AEABI aligned function variants

AEABI defines aligned variants of memcpy etc. that can be faster than
the default version due to not having to do alignment checks. When
emitting target code for these functions make use of these aligned
variants if possible. Also convert memset to memclr if possible.

Differential Revision: http://reviews.llvm.org/D8060

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237127 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoReverse ordering of base and derived pointer during safepoint lowering.
Igor Laevsky [Tue, 12 May 2015 13:12:14 +0000 (13:12 +0000)]
Reverse ordering of base and derived pointer during safepoint lowering.

According to the documentation in StackMap section for the safepoint we should have:
"The first Location in each pair describes the base pointer for the object. The second is the derived pointer actually being relocated."
But before this change we emitted them in reverse order - derived pointer first, base pointer second.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237126 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Remove useless target specific combine on TRUNCATE dag nodes.
Andrea Di Biagio [Tue, 12 May 2015 12:34:22 +0000 (12:34 +0000)]
[X86] Remove useless target specific combine on TRUNCATE dag nodes.

Before revision 171146, function 'PerformTruncateCombine' used to perform
a premature lowering of TRUNCATE dag nodes.
Revision 171146 then moved all the logic implemented by PerformTruncateCombine
to a custom lowering hook. However, that revision forgot to delete
function PerformTruncateCombine from the code.

This patch removes function 'PerformTruncateCombine' since it has no effect
on the SelectionDAG. No functional change intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237122 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[mips][FastISel] Handle calls with non legal types i8 and i16.
Vasileios Kalintiris [Tue, 12 May 2015 12:29:17 +0000 (12:29 +0000)]
[mips][FastISel] Handle calls with non legal types i8 and i16.

Summary: Allow calls with non legal integer types based on i8 and i16 to be processed by mips fast-isel.

Based on a patch by Reed Kotler.

Test Plan:
"Make check" test forthcoming.
Test-suite passes at O0/O2 and with mips32 r1/r2

Reviewers: rkotler, dsanders

Subscribers: llvm-commits, rfuhler

Differential Revision: http://reviews.llvm.org/D6770

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237121 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[mips][FastISel] Simplify callabi.ll by using multiple check prefixes.
Vasileios Kalintiris [Tue, 12 May 2015 12:17:11 +0000 (12:17 +0000)]
[mips][FastISel] Simplify callabi.ll by using multiple check prefixes.

Reviewers: dsanders

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9635

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237119 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[mips][FastISel] Allow computation of addresses from constant expressions.
Vasileios Kalintiris [Tue, 12 May 2015 12:08:31 +0000 (12:08 +0000)]
[mips][FastISel] Allow computation of addresses from constant expressions.

Summary:
Try to compute addresses when the offset from a memory location is a constant
expression.

Based on a patch by Reed Kotler.

Test Plan:
Passes test-suite for -O0/O2 and mips 32 r1/r2

Reviewers: rkotler, dsanders

Subscribers: llvm-commits, aemerson, rfuhler

Differential Revision: http://reviews.llvm.org/D6767

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237117 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoChange TargetParser enum names to avoid macro conflicts (llvm)
Renato Golin [Tue, 12 May 2015 10:33:58 +0000 (10:33 +0000)]
Change TargetParser enum names to avoid macro conflicts (llvm)

sys/time.h on Solaris (and possibly other systems) defines "SEC" as "1"
using a cpp macro.  The result is that this fails to compile.

Fixes https://llvm.org/PR23482

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237112 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAVX-512: asm parser errors check
Elena Demikhovsky [Tue, 12 May 2015 09:47:23 +0000 (09:47 +0000)]
AVX-512: asm parser errors check
I reverted the error check that was removed in 236416.
I put the it in a separate file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237107 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAVX-512: select operation for i1 vectors
Elena Demikhovsky [Tue, 12 May 2015 09:36:52 +0000 (09:36 +0000)]
AVX-512: select operation for i1 vectors
like: select i1 %cond, <16 x i1> %a, <16 x i1> %b.
I added pseudo-CMOV patterns to resolve the "select".
Added tests for KNL and SKX.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237106 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] DAGCombine should not assume arbitrary vector types are simple
Michael Kuperstein [Tue, 12 May 2015 07:33:07 +0000 (07:33 +0000)]
[X86] DAGCombine should not assume arbitrary vector types are simple

The X86-specific DAGCombine for stores should not assume vector types are always simple.
This fixes PR23476.

Differential Revision: http://reviews.llvm.org/D9659

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237097 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRemove unnecessary forward declaration.
Craig Topper [Tue, 12 May 2015 06:09:57 +0000 (06:09 +0000)]
Remove unnecessary forward declaration.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237093 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRemove unnecessary variables by folding calls into for loop header. NFC.
Craig Topper [Tue, 12 May 2015 05:25:10 +0000 (05:25 +0000)]
Remove unnecessary variables by folding calls into for loop header. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237090 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[lib/Fuzzer] remove the -dfsan=1 flag, just use -use_traces=1 (w/ or w/o dfsan)
Kostya Serebryany [Tue, 12 May 2015 01:58:34 +0000 (01:58 +0000)]
[lib/Fuzzer] remove the -dfsan=1 flag, just use -use_traces=1 (w/ or w/o dfsan)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237083 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[lib/Fuzzer] detach the pulse thread instad of joining it
Kostya Serebryany [Tue, 12 May 2015 01:43:20 +0000 (01:43 +0000)]
[lib/Fuzzer] detach the pulse thread instad of joining it

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237082 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoMigrate existing backends that care about software floating point
Eric Christopher [Tue, 12 May 2015 01:26:05 +0000 (01:26 +0000)]
Migrate existing backends that care about software floating point
to use the information in the module rather than TargetOptions.

We've had and clang has used the use-soft-float attribute for some
time now so have the backends set a subtarget feature based on
a particular function now that subtargets are created based on
functions and function attributes.

For the one middle end soft float check go ahead and create
an overloadable TargetLowering::useSoftFloat function that
just checks the TargetSubtargetInfo in all cases.

Also remove the command line option that hard codes whether or
not soft-float is set by using the attribute for all of the
target specific test cases - for the generic just go ahead and
add the attribute in the one case that showed up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237079 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFixing memory leak
Andrew Kaylor [Tue, 12 May 2015 00:13:51 +0000 (00:13 +0000)]
Fixing memory leak

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237072 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRefactoring gc_relocate related code in CodeGenPrepare.cpp
Sanjoy Das [Mon, 11 May 2015 23:47:30 +0000 (23:47 +0000)]
Refactoring gc_relocate related code in CodeGenPrepare.cpp

Summary:
The original code inserted new instructions by following a
Create->Remove->ReInsert flow. This patch removes the unnecessary
Remove->ReInsert part by setting up the InsertPoint correctly at the
very beginning. This change does not introduce any functionality change.

Patch by Chen Li!

Reviewers: reames, AndyAyers, sanjoy

Reviewed By: sanjoy

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9687

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237070 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRename variables in gc_relocate related functions to follow LLVM's naming conventions.
Sanjoy Das [Mon, 11 May 2015 23:47:27 +0000 (23:47 +0000)]
Rename variables in gc_relocate related functions to follow LLVM's naming conventions.

Summary:
This patch is to rename some variables to CamelCase in gc_relocate
related functions. There is no functionality change.

Patch by Chen Li!

Reviewers: reames, AndyAyers, sanjoy

Reviewed By: sanjoy

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9681

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237069 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[lib/Fuzzer] don't record traces when trace collection is off
Kostya Serebryany [Mon, 11 May 2015 23:25:28 +0000 (23:25 +0000)]
[lib/Fuzzer] don't record traces when trace collection is off

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237067 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[MemCpyOpt] Look at any dependency -not just source- for memset+memcpy.
Ahmed Bougacha [Mon, 11 May 2015 23:09:46 +0000 (23:09 +0000)]
[MemCpyOpt] Look at any dependency -not just source- for memset+memcpy.

This fixes another miscompile introduced by r235232: when there was a
dependency on the memcpy destination other than the memset, we would
ignore it, because we only looked at the source dependency.

It was a mistake to use SrcDepInfo.  Instead, just use DepInfo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237066 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoSimplify a return expression and an access to an alloca's allocated type
David Blaikie [Mon, 11 May 2015 23:09:25 +0000 (23:09 +0000)]
Simplify a return expression and an access to an alloca's allocated type

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237065 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[WinEH] Handle nested landing pads that return directly to the parent function.
Andrew Kaylor [Mon, 11 May 2015 23:06:02 +0000 (23:06 +0000)]
[WinEH] Handle nested landing pads that return directly to the parent function.

Differential Revision: http://reviews.llvm.org/D9684

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237063 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAdd more missing #includes, found by modules build.
Richard Smith [Mon, 11 May 2015 22:41:07 +0000 (22:41 +0000)]
Add more missing #includes, found by modules build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237062 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAdd missing #include, found by modules build.
Richard Smith [Mon, 11 May 2015 22:32:06 +0000 (22:32 +0000)]
Add missing #include, found by modules build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237061 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAdd missing #includes, found by modules build.
Richard Smith [Mon, 11 May 2015 22:31:40 +0000 (22:31 +0000)]
Add missing #includes, found by modules build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237060 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoReaddress r236990, use of static members on a non-static variable.
David Blaikie [Mon, 11 May 2015 22:20:48 +0000 (22:20 +0000)]
Readdress r236990, use of static members on a non-static variable.

The TargetRegistry is just a namespace-like class, instantiated in one
place to use a range-based for loop. Instead, expose access to the
registry via a range-based 'targets()' function instead. This makes most
uses a bit awkward/more verbose - but eventually we should just add a
range-based find_if function which will streamline these functions. I'm
happy to mkae them a bit awkward in the interim as encouragement to
improve the algorithms in time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237059 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix tablegen's PrintFatalError function to run registered file
James Y Knight [Mon, 11 May 2015 22:17:13 +0000 (22:17 +0000)]
Fix tablegen's PrintFatalError function to run registered file
cleanups.

Also, change code in tablegen which printed a message and then called
"exit(1)" to use PrintFatalError, instead.

This fixes instances where an empty output file was left behind after
a failed tablegen invocation, which would confuse subsequent ninja
runs into not attempting to rebuild.

Differential Revision: http://reviews.llvm.org/D9608

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237058 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[lib/Fuzzer] when running multiple fuzzing processes, print something every 10 minute...
Kostya Serebryany [Mon, 11 May 2015 21:31:51 +0000 (21:31 +0000)]
[lib/Fuzzer] when running multiple fuzzing processes, print something every 10 minutes to avoid buildbot timeouts

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237054 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix input validation issues in llvm-as/llvm-dis
Alexey Samsonov [Mon, 11 May 2015 21:20:20 +0000 (21:20 +0000)]
Fix input validation issues in llvm-as/llvm-dis

Summary:
1. llvm-as/llvm-dis tools do not check for input filename length.
2. llvm-dis does not verify the `Streamer` variable against `nullptr` properly, so the `M` variable could be uninitialized (e.g. if the input file does not exist) leading to null dref.

Patch by Lenar Safin!

Reviewers: samsonov

Reviewed By: samsonov

Subscribers: samsonov, llvm-commits

Differential Revision: http://reviews.llvm.org/D9584

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237051 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[lib/Fuzzer] rename FuzzerDFSan.cpp to FuzzerTraceState.cpp; update comments. NFC...
Kostya Serebryany [Mon, 11 May 2015 21:16:27 +0000 (21:16 +0000)]
[lib/Fuzzer] rename FuzzerDFSan.cpp to FuzzerTraceState.cpp; update comments. NFC expected

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237050 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agopropagate IR-level fast-math-flags to DAG nodes; 2nd try; NFC
Sanjay Patel [Mon, 11 May 2015 21:07:09 +0000 (21:07 +0000)]
propagate IR-level fast-math-flags to DAG nodes; 2nd try; NFC

This is a less ambitious version of:
http://reviews.llvm.org/rL236546

because that was reverted in:
http://reviews.llvm.org/rL236600

because it caused memory corruption that wasn't related to FMF
but was actually due to making nodes with 2 operands derive from a
plain SDNode rather than a BinarySDNode.

This patch adds the minimum plumbing necessary to use IR-level
fast-math-flags (FMF) in the backend without actually using
them for anything yet. This is a follow-on to:
http://reviews.llvm.org/rL235997

...which split the existing nsw / nuw / exact flags and FMF
into their own struct.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237046 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[LoopIdiomRecognize] Transform backedge-taken count check into an assertion.
Davide Italiano [Mon, 11 May 2015 21:02:34 +0000 (21:02 +0000)]
[LoopIdiomRecognize] Transform backedge-taken count check into an assertion.

runOnCountable() allowed the caller to call on a loop without a
predictable backedge-taken count. Change the code so that only loops
with computable backdge-count can call this function, in order to catch
abuses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237044 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[lib/Fuzzer] add a trace-based mutatation logic. Same idea as with DFSan-based mutato...
Kostya Serebryany [Mon, 11 May 2015 20:51:19 +0000 (20:51 +0000)]
[lib/Fuzzer] add a trace-based mutatation logic. Same idea as with DFSan-based mutator, but instead of relying on taint  tracking, try to find the data directly in the input. More (logic and comments) to go.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237043 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFixing build warnings
Andrew Kaylor [Mon, 11 May 2015 20:45:11 +0000 (20:45 +0000)]
Fixing build warnings

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237042 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[WinEH] Update exception numbering to give handlers their own base state.
Andrew Kaylor [Mon, 11 May 2015 19:41:19 +0000 (19:41 +0000)]
[WinEH] Update exception numbering to give handlers their own base state.

Differential Revision: http://reviews.llvm.org/D9512

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237014 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agogroup getNode() variants by purpose and add comments; NFC
Sanjay Patel [Mon, 11 May 2015 19:34:10 +0000 (19:34 +0000)]
group getNode() variants by purpose and add comments; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237013 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[RewriteStatepointsForGC] Fix a bug on creating gc_relocate for pointer to vector...
Sanjoy Das [Mon, 11 May 2015 18:49:34 +0000 (18:49 +0000)]
[RewriteStatepointsForGC] Fix a bug on creating gc_relocate for pointer to vector of pointers

Summary:
In RewriteStatepointsForGC pass, we create a gc_relocate intrinsic for
each relocated pointer, and the gc_relocate has the same type with the
pointer. During the creation of gc_relocate intrinsic, llvm requires to
mangle its type. However, llvm does not support mangling of all possible
types. RewriteStatepointsForGC will hit an assertion failure when it
tries to create a gc_relocate for pointer to vector of pointers because
mangling for vector of pointers is not supported.

This patch changes the way RewriteStatepointsForGC pass creates
gc_relocate. For each relocated pointer, we erase the type of pointers
and create an unified gc_relocate of type i8 addrspace(1)*. Then a
bitcast is inserted to convert the gc_relocate to the correct type. In
this way, gc_relocate does not need to deal with different types of
pointers and the unsupported type mangling is no longer a problem. This
change would also ease further merge when LLVM erases types of pointers
and introduces an unified pointer type.

Some minor changes are also introduced to gc_relocate related part in
InstCombineCalls, CodeGenPrepare, and Verifier accordingly.

Patch by Chen Li!

Reviewers: reames, AndyAyers, sanjoy

Reviewed By: sanjoy

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9592

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237009 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoLiveRangeCalc: Improve error messages on malformed IR
Matthias Braun [Mon, 11 May 2015 18:47:47 +0000 (18:47 +0000)]
LiveRangeCalc: Improve error messages on malformed IR

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237008 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoCodeGen: Make MachineInstr::untieRegOperand() a public function
Tom Stellard [Mon, 11 May 2015 17:40:54 +0000 (17:40 +0000)]
CodeGen: Make MachineInstr::untieRegOperand() a public function

This makes it easier to update in place instructions with tied operands.

Differential Revision: http://reviews.llvm.org/D9231

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237005 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Updates to X86 backend for f16 promotion
Pirama Arumuga Nainar [Mon, 11 May 2015 17:14:39 +0000 (17:14 +0000)]
[X86] Updates to X86 backend for f16 promotion

Summary:
r235215 adds support for f16 to be considered as a load/store type and
promote f16 operations to f32.

This patch has miscellaneous fixes for the X86 backend so all f16
operations are handled:
1. Set loadextaction for f16 vectors to expand.
2. Handle FP_EXTEND in a switch statement when handling v2f32
3. Do not fold (FP_TO_SINT (load f16)) into FP_TO_INT*_IN_MEM or
(store (SINT_TO_FP )) to a FILD.

Tests included.

Reviewers: ab, srhines, delena

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9092

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237004 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRip min/max pattern matching out of InstCombine and into
James Molloy [Mon, 11 May 2015 14:42:20 +0000 (14:42 +0000)]
Rip min/max pattern matching out of InstCombine and into
 ValueTracking.

This matching functionality is useful in more than just InstCombine, so
make it available in ValueTracking.

NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236998 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAmends r236990, because I failed at hitting "save" before commit.
Aaron Ballman [Mon, 11 May 2015 13:11:38 +0000 (13:11 +0000)]
Amends r236990, because I failed at hitting "save" before commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236991 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoReplacing a range-based for loop with an old-style for loop. This code was previously...
Aaron Ballman [Mon, 11 May 2015 13:10:17 +0000 (13:10 +0000)]
Replacing a range-based for loop with an old-style for loop. This code was previously causing a warning with MSVC about a compiler-generated local variable because TargetRegistry::begin() and end() are static member functions. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236990 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoSilencing an MSVC warning: '<<' : result of 32-bit shift implicitly converted to...
Aaron Ballman [Mon, 11 May 2015 12:45:53 +0000 (12:45 +0000)]
Silencing an MSVC warning: '<<' : result of 32-bit shift implicitly converted to 64 bits (was 64-bit shift intended?); NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236987 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[Testsuite] Renumber metadata in ScopedNoAliasAA test to match CHECK lines
Adam Nemet [Mon, 11 May 2015 09:10:14 +0000 (09:10 +0000)]
[Testsuite] Renumber metadata in ScopedNoAliasAA test to match CHECK lines

Summary:
Now it's much easier to follow what's happening in this test.

Also removed some unused metadata entries.

Reviewers: hfinkel

Reviewed By: hfinkel

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9601

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236981 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAVX-512: Changed CC parameter in "cmp" intrinsic
Elena Demikhovsky [Mon, 11 May 2015 09:03:14 +0000 (09:03 +0000)]
AVX-512: Changed CC parameter in "cmp" intrinsic
from i8 to i32 according to the Intel Spec

by Igor Breger (igor.breger@intel.com)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236979 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[Docs] Fix scoped noalias example
Adam Nemet [Mon, 11 May 2015 08:30:28 +0000 (08:30 +0000)]
[Docs] Fix scoped noalias example

Summary:
As far as I understand the entire point of this example is to show that
if noalias is not a superset/equal to the alias.scope list on a scope
domain then load could reference locations that the store is not known
to not-alias i.e may alias.

Reviewers: hfinkel

Reviewed By: hfinkel

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9598

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236977 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[InstCombine/PowerPC] Fix single-precision QPX load/store replacement
Hal Finkel [Mon, 11 May 2015 06:37:03 +0000 (06:37 +0000)]
[InstCombine/PowerPC] Fix single-precision QPX load/store replacement

The QPX single-precision load/store intrinsics have implied
truncation/extension from/to the declared value type of <4 x double> to the
memory type of <4 x float>. When we can prove the alignment of the pointer
argument, and thus replace the intrinsic with a regular load or store, we need
to load or store the correct data type (<4 x float>) instead of (<4 x double>).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236973 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFixed compilation warning, NFC.
Elena Demikhovsky [Mon, 11 May 2015 06:23:41 +0000 (06:23 +0000)]
Fixed compilation warning, NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236972 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAVX-512: Added SKX instructions and intrinsics:
Elena Demikhovsky [Mon, 11 May 2015 06:05:05 +0000 (06:05 +0000)]
AVX-512: Added SKX instructions and intrinsics:
{add/sub/mul/div/} x {ps/pd} x {128/256} 2. max/min with sae

By Asaf Badouh (asaf.badouh@intel.com)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236971 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoMake buildbots happy
David Majnemer [Mon, 11 May 2015 05:33:27 +0000 (05:33 +0000)]
Make buildbots happy

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236970 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[InstCombine] Canonicalize single element array store
David Majnemer [Mon, 11 May 2015 05:04:27 +0000 (05:04 +0000)]
[InstCombine] Canonicalize single element array store

Use the element type instead of the aggregate type.

Differential Revision: http://reviews.llvm.org/D9591

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236969 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[InstCombine] Canonicalize single element array load
David Majnemer [Mon, 11 May 2015 05:04:22 +0000 (05:04 +0000)]
[InstCombine] Canonicalize single element array load

Use the element type instead of the aggregate type.

Differential Revision: http://reviews.llvm.org/D9596

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236968 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAVX-512: fixed UINT_TO_FP operation for 512-bit types.
Elena Demikhovsky [Sun, 10 May 2015 14:23:52 +0000 (14:23 +0000)]
AVX-512: fixed UINT_TO_FP operation for 512-bit types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236955 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[SelectionDAG] Fixed constant folding issue when legalised types are smaller then...
Simon Pilgrim [Sun, 10 May 2015 14:14:51 +0000 (14:14 +0000)]
[SelectionDAG] Fixed constant folding issue when legalised types are smaller then the folded type.

Found when testing with llvm-stress on i686 targets.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236954 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoSanitizerCoverage: Use `createSanitizerCtor` to create ctor and call init
Ismail Pazarbasi [Sun, 10 May 2015 13:45:05 +0000 (13:45 +0000)]
SanitizerCoverage: Use `createSanitizerCtor` to create ctor and call init

Second attempt; instead of using a named local variable, passing
arguments directly to `createSanitizerCtorAndInitFunctions` worked
on Windows.

Reviewers: kcc, samsonov

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D8780

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236951 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAVX-512: fixed a bug in i1 vectors lowering
Elena Demikhovsky [Sun, 10 May 2015 10:33:32 +0000 (10:33 +0000)]
AVX-512: fixed a bug in i1 vectors lowering

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236947 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoSystemZ: silence a GCC warning
Saleem Abdulrasool [Sun, 10 May 2015 00:53:41 +0000 (00:53 +0000)]
SystemZ: silence a GCC warning

warning: enumeral and non-enumeral type in conditional expression

Cast the 0 to the appropriate type.  NFC.  Identified by GCC 4.9.2

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236942 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAdd polly support to sort_includes.py
Tobias Grosser [Sat, 9 May 2015 09:08:56 +0000 (09:08 +0000)]
Add polly support to sort_includes.py

Changes:
 -  Add "isl/" as a system library prefix. Even though isl is regularly
    imported into polly, it is still used like an external library.
 -  Add "json/" as a system library prefix. Polly uses json-cpp as external
    library.
 -  Distinguish between llvm and subproject libraries. Always sort subprojects
    before LLVM. This was already the case with clang, as 'clang' comes before
    'llvm', but we also want 'polly' to be sorted before 'llvm'.

The sorting of headers that are not part of Polly or isl remains unchanged.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236929 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agollvm/test/CodeGen/AArch64/tailcall_misched_graph.ll: s/REQUIRE/REQUIRES/
NAKAMURA Takumi [Sat, 9 May 2015 05:59:00 +0000 (05:59 +0000)]
llvm/test/CodeGen/AArch64/tailcall_misched_graph.ll: s/REQUIRE/REQUIRES/

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236928 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix MergeConsecutiveStore for non-byte-sized memory accesses.
James Y Knight [Sat, 9 May 2015 03:13:37 +0000 (03:13 +0000)]
Fix MergeConsecutiveStore for non-byte-sized memory accesses.

The bug showed up as a compile-time assertion failure:
  Assertion `NumBits >= MIN_INT_BITS && "bitwidth too small"' failed
when building msan tests on x86-64.

Prior to r236850, this bug was masked due to a bogus alignment check,
which also accidentally rejected non-byte-sized accesses. Afterwards,
an invalid ElementSizeBytes == 0 got further into the function, and
triggered the assertion failure.

It would probably be a good idea to allow it to handle merging stores
of unusual widths as well, but for now, to un-break it, I'm just
making the minimal fix.

Differential Revision: http://reviews.llvm.org/D9626

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236927 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoMachineCSE: Add a target query for the LookAheadLimit heurisitic
Tom Stellard [Sat, 9 May 2015 00:56:07 +0000 (00:56 +0000)]
MachineCSE: Add a target query for the LookAheadLimit heurisitic

This is used to determine whether or not to CSE physical register
defs.

Differential Revision: http://reviews.llvm.org/D9472

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236923 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[Fast-ISel] Don't mark the first use of a remat constant as killed.
Pete Cooper [Sat, 9 May 2015 00:51:03 +0000 (00:51 +0000)]
[Fast-ISel] Don't mark the first use of a remat constant as killed.

When emitting something like 'add x, 1000' if we remat the 1000 then we should be able to
mark the vreg containing 1000 as killed.  Given that we go bottom up in fast-isel, a later
use of 1000 will be higher up in the BB and won't kill it, or be impacted by the lower kill.

However, rematerialised constant expressions aren't generated bottom up.  The local value save area
grows downwards.  This means that if you remat 2 constant expressions which both use 1000 then the
first will kill it, then the second, which is *lower* in the BB will read a killed register.

This is the case in the attached test where the 2 GEPs both need to generate 'add x, 6680' for the constant offset.

Note that this commit only makes kill flag generation conservative.  There's nothing else obviously wrong with
the local value save area growing downwards, and in fact it needs to for handling arbitrarily complex constant expressions.

However, it would be nice if there was a solution which would let us generate more accurate kill flags, or just kill flags completely.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236922 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix compile error
Arnold Schwaighofer [Sat, 9 May 2015 00:10:25 +0000 (00:10 +0000)]
Fix compile error

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236921 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRevert r236912.
Quentin Colombet [Sat, 9 May 2015 00:02:06 +0000 (00:02 +0000)]
Revert r236912.

Author: dblaikie
Date: Fri May  8 17:47:50 2015
New Revision: 236912

URL: http://llvm.org/viewvc/llvm-project?rev=236912&view=rev
Log:
[opaque pointer type] Cleanup a few references to pointee types using nearby non-pointee types of the same value

& cleanup a convoluted return expression while I'm here

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236919 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[Target/ARM] Remove unused 'private' from class.
Davide Italiano [Fri, 8 May 2015 23:58:28 +0000 (23:58 +0000)]
[Target/ARM] Remove unused 'private' from class.

Differential Revision: http://reviews.llvm.org/D9611
Reviewed by: rengolin

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236918 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoScheduleDAGInstrs: In functions with tail calls PseudoSourceValues are not non-aliasi...
Arnold Schwaighofer [Fri, 8 May 2015 23:52:00 +0000 (23:52 +0000)]
ScheduleDAGInstrs: In functions with tail calls PseudoSourceValues are not non-aliasing distinct objects

The code that builds the dependence graph assumes that two PseudoSourceValues
don't alias. In a tail calling function two FixedStackObjects might refer to the
same location. Worse 'immutable' fixed stack objects like function arguments are
not immutable and will be clobbered.

Change this so that a load from a FixedStackObject is not invariant in a tail
calling function and don't return a PseudoSourceValue for an instruction in tail
calling functions when building the dependence graph so that we handle function
arguments conservatively.

Fix for PR23459.

rdar://20740035

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236916 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[opaque pointer type] Cleanup a few references to pointee types using nearby non...
David Blaikie [Fri, 8 May 2015 22:47:50 +0000 (22:47 +0000)]
[opaque pointer type] Cleanup a few references to pointee types using nearby non-pointee types of the same value

& cleanup a convoluted return expression while I'm here

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236912 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[lib/Fuzzer] build tests that work well with dfsan also w/o dfsan
Kostya Serebryany [Fri, 8 May 2015 21:45:19 +0000 (21:45 +0000)]
[lib/Fuzzer] build tests that work well with dfsan also w/o dfsan

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236909 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[lib/Fuzzer] use -fsanitize-coverage=trace-cmp when building LLVM with LLVM_USE_SANIT...
Kostya Serebryany [Fri, 8 May 2015 21:30:55 +0000 (21:30 +0000)]
[lib/Fuzzer] use -fsanitize-coverage=trace-cmp when building LLVM with LLVM_USE_SANITIZE_COVERAGE; in lib/Fuzzer try to reload the corpus to pick up new units from other processes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236906 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoSwitch lowering: cluster adjacent fall-through cases even at -O0
Hans Wennborg [Fri, 8 May 2015 21:23:39 +0000 (21:23 +0000)]
Switch lowering: cluster adjacent fall-through cases even at -O0

It's cheap to do, and codegen is much faster if cases can be merged
into clusters.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236905 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAdd a comment about DepthFirstIterator's skipchildren behavior
Daniel Berlin [Fri, 8 May 2015 21:17:24 +0000 (21:17 +0000)]
Add a comment about DepthFirstIterator's skipchildren behavior

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236904 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoTargetParser: FPU/ARCH/EXT parsing refactory - NFC
Renato Golin [Fri, 8 May 2015 21:04:27 +0000 (21:04 +0000)]
TargetParser: FPU/ARCH/EXT parsing refactory - NFC

This new class in a global context contain arch-specific knowledge in order
to provide LLVM libraries, tools and projects with the ability to understand
the architectures. For now, only FPU, ARCH and ARCH extensions on ARM are
supported.

Current behaviour it to parse from free-text to enum values and back, so that
all users can share the same parser and codes. This simplifies a lot both the
ASM/Obj streamers in the back-end (where this came from), and the front-end
parsers for command line arguments (where this is going to be used next).

The previous implementation, using .def/.h includes is deprecated due to its
inflexibility to be built without the backend support and for being too
cumbersome. As more architectures join this scheme, and as more features of
such architectures are added (such as hardware features, type sizes, etc) into
a full blown TargetDescription class, having a set of classes is the most
sane implementation.

The ultimate goal of this refactor both LLVM's and Clang's target description
classes into one unique interface, so that we can de-duplicate and standardise
the descriptions, as well as make it available for other front-ends, tools,
etc.

The FPU parsing for command line options in Clang has been converted to use
this new library and a number of aliases were added for compatibility:
 * A bogus neon-vfpv3 alias (neon defaults to vfp3)
 * armv5/v6
 * {fp4/fp5}-{sp/dp}-d16

Next steps:
 * Port Clang's ARCH/EXT parsing to use this library.
 * Create a TableGen back-end to generate this information.
 * Run this TableGen process regardless of which back-ends are built.
 * Expose more information and rename it to TargetDescription.
 * Continue re-factoring Clang to use as much of it as possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236900 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoUpdate Function::getContext documentation. NFC.
Diego Novillo [Fri, 8 May 2015 21:01:29 +0000 (21:01 +0000)]
Update Function::getContext documentation. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236899 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[Fast-ISel] Clear kill flags on registers replaced by updateValueMap.
Pete Cooper [Fri, 8 May 2015 20:46:54 +0000 (20:46 +0000)]
[Fast-ISel] Clear kill flags on registers replaced by updateValueMap.

When selecting an extract instruction, we don't actually generate code but instead work out which register we are reading, and rewrite uses of the extract def to the source register.  This is done via updateValueMap,.

However, its possible that the source register we are rewriting *to* to also have uses.  If those uses are after a kill of the value we are rewriting *from* then we have uses after a kill and the verifier fails.

This code checks for the case where the to register is also used, and if so it clears all kill on the from register.  This is conservative, but better that always clearing kills on the from register.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236897 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[Hexagon] Generate more hardware loops
Brendon Cahoon [Fri, 8 May 2015 20:18:21 +0000 (20:18 +0000)]
[Hexagon] Generate more hardware loops

Refactored parts of the hardware loop pass to generate
more. Also, added more tests.

Differential Revision: http://reviews.llvm.org/D9568

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236896 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[BasicAA] Fix zext & sext handling
Sanjoy Das [Fri, 8 May 2015 18:58:55 +0000 (18:58 +0000)]
[BasicAA] Fix zext & sext handling

Summary:

There are several unhandled edge cases in BasicAA's GetLinearExpression
method. This changes fixes outstanding issues, including zext / sext of
a constant with the sign bit set, and the refusal to decompose zexts or
sexts of wrapping arithmetic.

Test Plan: Unit tests added in //q.ext.ll//.

Patch by Nick White.

Reviewers: hfinkel, sanjoy

Reviewed By: hfinkel, sanjoy

Subscribers: sanjoy, llvm-commits, hfinkel

Differential Revision: http://reviews.llvm.org/D6682

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236894 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoReplace branch-to-unreachable with assertion.
David Blaikie [Fri, 8 May 2015 18:52:28 +0000 (18:52 +0000)]
Replace branch-to-unreachable with assertion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236893 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix a docs build break introduced by rL236888.
Pat Gavlin [Fri, 8 May 2015 18:37:49 +0000 (18:37 +0000)]
Fix a docs build break introduced by rL236888.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236891 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Fast-ISel was incorrectly always killing the source of a truncate.
Pete Cooper [Fri, 8 May 2015 18:29:42 +0000 (18:29 +0000)]
[X86] Fast-ISel was incorrectly always killing the source of a truncate.

A trunc from i32 to i1 on x86_64 generates an instruction such as

%vreg19<def> = COPY %vreg9:sub_8bit<kill>; GR8:%vreg19 GR32:%vreg9

However, the copy here should only have the kill flag on the 32-bit path, not the 64-bit one.
Otherwise, we are killing the source of the truncate which could be used later in the program.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236890 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoExtend the statepoint intrinsic to allow statepoints to be marked as transitions...
Pat Gavlin [Fri, 8 May 2015 18:07:42 +0000 (18:07 +0000)]
Extend the statepoint intrinsic to allow statepoints to be marked as transitions from GC-aware code to code that is not GC-aware.

This changes the shape of the statepoint intrinsic from:

  @llvm.experimental.gc.statepoint(anyptr target, i32 # call args, i32 unused, ...call args, i32 # deopt args, ...deopt args, ...gc args)

to:

  @llvm.experimental.gc.statepoint(anyptr target, i32 # call args, i32 flags, ...call args, i32 # transition args, ...transition args, i32 # deopt args, ...deopt args, ...gc args)

This extension offers the backend the opportunity to insert (somewhat) arbitrary code to manage the transition from GC-aware code to code that is not GC-aware and back.

In order to support the injection of transition code, this extension wraps the STATEPOINT ISD node generated by the usual lowering lowering with two additional nodes: GC_TRANSITION_START and GC_TRANSITION_END. The transition arguments that were passed passed to the intrinsic (if any) are lowered and provided as operands to these nodes and may be used by the backend during code generation.

Eventually, the lowering of the GC_TRANSITION_{START,END} nodes should be informed by the GC strategy in use for the function containing the intrinsic call; for now, these nodes are instead replaced with no-ops.

Differential Revision: http://reviews.llvm.org/D9501

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236888 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[NoTTI] reject negative scale in addressing mode
Jingyue Wu [Fri, 8 May 2015 18:07:24 +0000 (18:07 +0000)]
[NoTTI] reject negative scale in addressing mode

Summary:
I noticed this bug when deubging a WIP on LSR. I wonder whether and how we
should add a regression test for this.

Test Plan: no tests failed.

Reviewers: atrick

Subscribers: hfinkel, llvm-commits

Differential Revision: http://reviews.llvm.org/D9536

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236887 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoClear kill flags on all used registers when sinking instructions.
Pete Cooper [Fri, 8 May 2015 17:54:32 +0000 (17:54 +0000)]
Clear kill flags on all used registers when sinking instructions.

The test here was sinking the AND here to a lower BB:

%vreg7<def> = ANDWri %vreg8, 0; GPR32common:%vreg7,%vreg8
TBNZW %vreg8<kill>, 0, <BB#1>; GPR32common:%vreg8

which meant that vreg8 was read after it was killed.

This commit changes the code from clearing kill flags on the AND to clearing flags on all registers used by the AND.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236886 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago80 cols fix since i'm looking at this function anyway. NFC
Pete Cooper [Fri, 8 May 2015 17:54:29 +0000 (17:54 +0000)]
80 cols fix since i'm looking at this function anyway.  NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236885 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRemove duplicate cmake target I added in r236792.
Pete Cooper [Fri, 8 May 2015 16:59:53 +0000 (16:59 +0000)]
Remove duplicate cmake target I added in r236792.

Thanks to Daniel Jasper for pointing out the mistake.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236881 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoUnbreak build: Makefile must have the same change as CMakeLists.txt
Douglas Katzman [Fri, 8 May 2015 16:39:59 +0000 (16:39 +0000)]
Unbreak build: Makefile must have the same change as CMakeLists.txt

This was omitted from http://reviews.llvm.org/D9441
(the irony is that that was to detect omissions in something else)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236878 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[Hexagon] Update AnalyzeBranch, etc target hooks
Brendon Cahoon [Fri, 8 May 2015 16:16:29 +0000 (16:16 +0000)]
[Hexagon] Update AnalyzeBranch, etc target hooks

Improved the AnalyzeBranch, InsertBranch, and RemoveBranch
functions in order to handle more of our branch instructions.
This requires changes to analyzeCompare and PredicateInstructions.
Specifically, we've added support for new value compare jumps,
improved handling of endloop, added more compare instructions,
and improved support for predicate instructions.

Differential Revision: http://reviews.llvm.org/D9559

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236876 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoPrevent further errors of omission when adding backend names.
Douglas Katzman [Fri, 8 May 2015 15:34:12 +0000 (15:34 +0000)]
Prevent further errors of omission when adding backend names.

Differential Revision: http://reviews.llvm.org/D9441

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236865 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Teach 'getTargetShuffleMask' how to look through ISD::WrapperRIP when decoding...
Andrea Di Biagio [Fri, 8 May 2015 15:11:07 +0000 (15:11 +0000)]
[X86] Teach 'getTargetShuffleMask' how to look through ISD::WrapperRIP when decoding a PSHUFB mask.

The function 'getTargetShuffleMask' already knows how to deal with PSHUFB nodes
where the mask node is a load from constant pool, and the constant pool node
is wrapped by a X86ISD::Wrapper node. This patch extends that logic by teaching
it how to also look through X86ISD::WrapperRIP.

This helps function combineX86ShufflesRecusively to combine more shuffle
sequences containing PSHUFB nodes if we are in RIPRel PIC mode.

Before this change, llc (with -relocation-model=pic -march=x86-64) was unable
to decode a pshufb where the mask was loaded from a constant pool. For example,
the no-op shuffle from test 'x86-fold-pshufb.ll' was not folded into its
operand, so instead of generating a single 'movaps' the backend always
generated a sub-optimal 'movdqa + pshufb' sequence.

Added test x86-fold-pshufb.ll.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236863 91177308-0d34-0410-b5e6-96231b3b80d8