oota-llvm.git
9 years ago[SCEV] Remove a test case added in r249168
Sanjoy Das [Thu, 22 Oct 2015 19:57:41 +0000 (19:57 +0000)]
[SCEV] Remove a test case added in r249168

The test case wasn't testing what it was commented to be testing; and
when I tried to fix the test I noticed that SCEV does not support the
simplification that the test was supposed to test.

This change removes the test case to avoid confusion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251053 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SCEV] Commute zero extends through <nuw> additions
Sanjoy Das [Thu, 22 Oct 2015 19:57:38 +0000 (19:57 +0000)]
[SCEV] Commute zero extends through <nuw> additions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251052 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SCEV] Opportunistically interpret unsigned constraints as signed
Sanjoy Das [Thu, 22 Oct 2015 19:57:34 +0000 (19:57 +0000)]
[SCEV] Opportunistically interpret unsigned constraints as signed

Summary:
An unsigned comparision is equivalent to is corresponding signed version
if both the operands being compared are positive.  Teach SCEV to use
this fact when profitable.

Reviewers: atrick, hfinkel, reames, nlewycky

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D13687

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251051 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SCEV] Teach SCEV some axioms about non-wrapping arithmetic
Sanjoy Das [Thu, 22 Oct 2015 19:57:29 +0000 (19:57 +0000)]
[SCEV] Teach SCEV some axioms about non-wrapping arithmetic

Summary:
 - A s<  (A + C)<nsw> if C >  0
 - A s<= (A + C)<nsw> if C >= 0
 - (A + C)<nsw> s<  A if C <  0
 - (A + C)<nsw> s<= A if C <= 0

Right now `C` needs to be a constant, but we can later generalize it to
be a non-constant if needed.

Reviewers: atrick, hfinkel, reames, nlewycky

Subscribers: sanjoy, llvm-commits

Differential Revision: http://reviews.llvm.org/D13686

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251050 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SCEV] Commute sign extends through nsw additions
Sanjoy Das [Thu, 22 Oct 2015 19:57:25 +0000 (19:57 +0000)]
[SCEV] Commute sign extends through nsw additions

Summary: Depends on D13613.

Reviewers: atrick, hfinkel, reames, nlewycky

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D13685

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251049 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SCEV] Mark AddExprs as nsw or nuw if legal
Sanjoy Das [Thu, 22 Oct 2015 19:57:19 +0000 (19:57 +0000)]
[SCEV] Mark AddExprs as nsw or nuw if legal

Summary:
This uses `ScalarEvolution::getRange` and not potentially control
dependent `nsw` and `nuw` bits on the arithmetic instruction.

Reviewers: atrick, hfinkel, nlewycky

Subscribers: llvm-commits, sanjoy

Differential Revision: http://reviews.llvm.org/D13613

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251048 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[ASan] Minor fixes to dynamic allocas handling:
Alexey Samsonov [Thu, 22 Oct 2015 19:51:59 +0000 (19:51 +0000)]
[ASan] Minor fixes to dynamic allocas handling:

* Don't instrument promotable dynamic allocas:
  We already have a test that checks that promotable dynamic allocas are
  ignored, as well as static promotable allocas. Make sure this test will
  still pass if/when we enable dynamic alloca instrumentation by default.

* Handle lifetime intrinsics before handling dynamic allocas:
  lifetime intrinsics may refer to dynamic allocas, so we need to emit
  instrumentation before these dynamic allocas would be replaced.

Differential Revision: http://reviews.llvm.org/D12704

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251045 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[ExecutionEngine] Garbage collect some dead (and unsafe) code.
Davide Italiano [Thu, 22 Oct 2015 18:46:27 +0000 (18:46 +0000)]
[ExecutionEngine] Garbage collect some dead (and unsafe) code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251042 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[IntrinsicEmitter] Remove GET_INTRINSIC_MODREF_BEHAVIOR table
Igor Laevsky [Thu, 22 Oct 2015 18:35:34 +0000 (18:35 +0000)]
[IntrinsicEmitter] Remove GET_INTRINSIC_MODREF_BEHAVIOR table

There is no need to generate separate table for intrinsics mod ref behaviour.
It can now be determined purely from function attributes.

Differential Revision: http://reviews.llvm.org/D13917

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251040 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAvoid storing a second copy of each string in StringTableBuilder.
Rafael Espindola [Thu, 22 Oct 2015 18:32:06 +0000 (18:32 +0000)]
Avoid storing a second copy of each string in StringTableBuilder.

This was only use in the extremely uncommon case of @@@ symbols on ELF.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251039 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAArch64: Disable the latency heuristic
Matthias Braun [Thu, 22 Oct 2015 18:07:38 +0000 (18:07 +0000)]
AArch64: Disable the latency heuristic

It turned out not to improve any of our benchmarks but occasionally led
to increased register pressure and spilling.

Only enabling for the Cyclone CPU as the results on the cortex CPUs
give mixed results.

Differential Revision: http://reviews.llvm.org/D13708

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251038 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMachineScheduler: Add a way to disable the 'ReduceLatency' heuristic
Matthias Braun [Thu, 22 Oct 2015 18:07:31 +0000 (18:07 +0000)]
MachineScheduler: Add a way to disable the 'ReduceLatency' heuristic

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251037 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove the last traces of X86CompilationCallback as it is completely
Eric Christopher [Thu, 22 Oct 2015 17:55:35 +0000 (17:55 +0000)]
Remove the last traces of X86CompilationCallback as it is completely
unused.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251035 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCodeGen: increase bits allocated for LegalizeActions
Tim Northover [Thu, 22 Oct 2015 17:20:48 +0000 (17:20 +0000)]
CodeGen: increase bits allocated for LegalizeActions

The array handling CondCodes only allocated 2 bits to describe the
desired action for each type. The new addition of a "LibCall" option
overflowed this and caused corruption for Custom actions.

No in-tree targets have a Custom CondCodeAction, so unfortunately it
can't be tested.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251033 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoChange makeLibCall to take an ArrayRef<SDValue> instead of pointer and size. This...
Craig Topper [Thu, 22 Oct 2015 17:05:00 +0000 (17:05 +0000)]
Change makeLibCall to take an ArrayRef<SDValue> instead of pointer and size. This removes the need to pass a hardcoded size in many places. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251032 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix broken build under MSVC.
Zachary Turner [Thu, 22 Oct 2015 16:42:31 +0000 (16:42 +0000)]
Fix broken build under MSVC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251030 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse ArrayRef instead of pointer and size. NFC
Craig Topper [Thu, 22 Oct 2015 16:35:56 +0000 (16:35 +0000)]
Use ArrayRef instead of pointer and size. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251029 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] - Catch extra combine opportunities for redundant imuls.
Zia Ansari [Thu, 22 Oct 2015 16:14:45 +0000 (16:14 +0000)]
[X86] - Catch extra combine opportunities for redundant imuls.

When we fold "mul ((add x, c1), c1)" -> "add ((mul x, c2), c1*c2)", we bail if (add x, c1) has multiple
users which would result in an extra add instruction.
In such cases, this patch adds a check to see if we can eliminate a multiply instruction in exchange for the extra add.

I also added the capability of doing the existing optimization with non-splatted vectors (splatted also works).

Differential Revision: http://reviews.llvm.org/D13740

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251028 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[PPC] Fix PR24686 by failing assembly for an invalid relocation
Bill Schmidt [Thu, 22 Oct 2015 15:53:44 +0000 (15:53 +0000)]
[PPC] Fix PR24686 by failing assembly for an invalid relocation

PR24686 identifies a problem where a relocation expression is invalid
when not all of the symbols in the expression can be locally
resolved.  This causes the compiler to request a PC-relative half16ds
relocation, which is nonsensical for PowerPC.  This patch recognizes
this situation and ensures we fail the assembly cleanly.

Test case provided by Anton Blanchard.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251027 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAvoid hash lookups when finalizing StringTableBuilder. NFC.
Rafael Espindola [Thu, 22 Oct 2015 15:26:35 +0000 (15:26 +0000)]
Avoid hash lookups when finalizing StringTableBuilder. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251024 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse array_pod_sort. NFC.
Rafael Espindola [Thu, 22 Oct 2015 15:15:44 +0000 (15:15 +0000)]
Use array_pod_sort. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251023 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix incorrect target triple in fp16-promote.ll
Pirama Arumuga Nainar [Thu, 22 Oct 2015 14:15:00 +0000 (14:15 +0000)]
Fix incorrect target triple in fp16-promote.ll

Summary:
Hyphens were missing from the triple, causing it to be parsed
incorrectly.  This patch updates the triple and makes necessary
changes to the expected output.

Patch is from Vinicius Tinti.

Reviewers: ab, tinti

Subscribers: srhines, llvm-commits

Differential Revision: http://reviews.llvm.org/D13792

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251020 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips][mips16] Fix typo in FileCheck directive.
Daniel Sanders [Thu, 22 Oct 2015 14:01:52 +0000 (14:01 +0000)]
[mips][mips16] Fix typo in FileCheck directive.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251019 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86][AVX512] extend vcvtph2ps to support xmm/ymm and sae versions
Asaf Badouh [Thu, 22 Oct 2015 14:01:16 +0000 (14:01 +0000)]
[X86][AVX512] extend vcvtph2ps to support xmm/ymm and sae versions

Differential Revision: http://reviews.llvm.org/D13945

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251018 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[GlobalsAA] Loosen an overly conservative bailout
James Molloy [Thu, 22 Oct 2015 13:44:26 +0000 (13:44 +0000)]
[GlobalsAA] Loosen an overly conservative bailout

Instead of bailing out when we see loads, analyze them. If we can prove that the loaded-from address must escape, then we can conclude that a load from that address must escape too and therefore cannot alias a non-addr-taken global.

When checking if a Value can alias a non-addr-taken global, if the Value is a LoadInst of a non-global, recurse instead of bailing.

If we can follow a trail of loads up to some base that is captured, we know by inference that all the loads we followed are also captured.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251017 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[BasicAA] Non-equal indices in a GEP of a SequentialType don't overlap
James Molloy [Thu, 22 Oct 2015 13:28:18 +0000 (13:28 +0000)]
[BasicAA] Non-equal indices in a GEP of a SequentialType don't overlap

If the final indices of two GEPs can be proven to not be equal, and
the GEP is of a SequentialType (not a StructType), then the two GEPs
do not alias.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251016 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[ValueTracking] Add a new predicate: isKnownNonEqual()
James Molloy [Thu, 22 Oct 2015 13:18:42 +0000 (13:18 +0000)]
[ValueTracking] Add a new predicate: isKnownNonEqual()

isKnownNonEqual(A, B) returns true if it can be determined that A != B.

At the moment it only knows two facts, that a non-wrapping add of nonzero to a value cannot be that value:

A + B != A [where B != 0, addition is nsw or nuw]

and that contradictory known bits imply two values are not equal.

This patch also hooks this up to InstSimplify; InstSimplify had a peephole for the first fact but not the second so this teaches InstSimplify a new trick too (alas no measured performance impact!)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251012 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix add_llvm_external_project.
Manuel Klimek [Thu, 22 Oct 2015 08:31:46 +0000 (08:31 +0000)]
Fix add_llvm_external_project.

r250835 unintentionally discarded the optional parameter to the
add_llvm_external_project() macro that may point to a path when the said
path is different from ${name}. This should fix it by passing ${ARGN} on
to add_llvm_subdirectory(). The problem manifests itself with e.g.
add_llvm_external_project(clang-tools-extra extra) from
clang/tools/CMakeLists.txt

Patch by Luchesar V. Iliev.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251001 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse range-based for loop in sys::path::append(). NFC.
Pawel Bylica [Thu, 22 Oct 2015 08:12:15 +0000 (08:12 +0000)]
Use range-based for loop in sys::path::append(). NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250999 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAVX-512: Fixed a bug in select_cc for i1 type
Elena Demikhovsky [Thu, 22 Oct 2015 07:10:29 +0000 (07:10 +0000)]
AVX-512: Fixed a bug in select_cc for i1 type
Fixed faiure:
LLVM ERROR: Cannot select: t33: i1 = select_cc t25, Constant:i32<0>, t45, t42, seteq:ch

added a test

Differential Revision: http://reviews.llvm.org/D13943

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250996 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoPartially reverted changes from r250686
Elena Demikhovsky [Thu, 22 Oct 2015 06:20:29 +0000 (06:20 +0000)]
Partially reverted changes from r250686
Clang runtime failure was reported.
   Assertion failed: (isExtended() && "Type is not extended!"), function getTypeForEVT
I'll need to add a proper handling for PointerType in masked load/store intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250995 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[IR] Add a `makeNoWrapRegion` method to `ConstantRange`
Sanjoy Das [Thu, 22 Oct 2015 03:12:57 +0000 (03:12 +0000)]
[IR] Add a `makeNoWrapRegion` method to `ConstantRange`

Summary: This will be used in a future change to ScalarEvolution.

Reviewers: hfinkel, reames, nlewycky

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D13612

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250975 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[OperandBundles] Teach AliasAnalysis about operand bundles
Sanjoy Das [Thu, 22 Oct 2015 03:12:51 +0000 (03:12 +0000)]
[OperandBundles] Teach AliasAnalysis about operand bundles

Summary:
If a `CallSite` has operand bundles, then do not peek into the called
function to get a more precise `ModRef` answer.

This is tested using `argmemonly`, `-basicaa` and `-gvn`; but the
functionality is not specific to any of these.

Depends on D13961

Reviewers: reames, chandlerc

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D13962

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250974 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[OperandBundles] Make function attributes conservatively correct
Sanjoy Das [Thu, 22 Oct 2015 03:12:22 +0000 (03:12 +0000)]
[OperandBundles] Make function attributes conservatively correct

Summary:
This makes attribute accessors on `CallInst` and `InvokeInst` do the
(conservatively) right thing.  This essentially involves, in some
cases, *not* falling back querying the attributes on the called
`llvm::Function` when operand bundles are present.

Attributes locally present on the `CallInst` or `InvokeInst` will still
override operand bundle semantics.  The LangRef has been amended to
reflect this.  Note: this change does not do anything prevent
`-function-attrs` from inferring `CallSite` local attributes after
inspecting the called function -- that will be done as a separate
change.

I've used `-adce` and `-early-cse` to test these changes.  There is
nothing special about these passes (and they did not require any
changes) except that they seemed be the easiest way to write the tests.

This change does not add deal with `argmemonly`.  That's a later change
because alias analysis requires a related fix before `argmemonly` can be
tested.

Reviewers: reames, chandlerc

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D13961

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250973 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoWebAssembly: fix more syntax
JF Bastien [Thu, 22 Oct 2015 02:32:50 +0000 (02:32 +0000)]
WebAssembly: fix more syntax

br_if shouldn't start with a dot.
div and rem went from prefix u/s to suffix.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250972 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd missing load/store flags to thumb2 instructions.
Pete Cooper [Thu, 22 Oct 2015 01:48:57 +0000 (01:48 +0000)]
Add missing load/store flags to thumb2 instructions.

These were the cause of a verifier error when building 7zip with
-verify-machineinstrs.  Running 'make check' with the verifier
triggered the same error on the test here so i've updated the test
to run the verifier on one of its runs instead of adding a new one.

While looking at this code, there was a stale comment that these
instructions were only used for disassembly.  This probably used to
be the case, but they are now used in the 'ARM load / store optimization pass' too.

This reapplies r242300 which was reverted in r242428 due to bot failures.

Ultimately those failures were spurious and completely unrelated to this commit.  I reverted this
at the time because it was thought to be at fault.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250969 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAttempt to fix formatting for operand bundles
David Majnemer [Thu, 22 Oct 2015 01:46:38 +0000 (01:46 +0000)]
Attempt to fix formatting for operand bundles

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250968 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUpdate LangRef for changes to the alias textual IR format to support typeless pointer...
David Blaikie [Thu, 22 Oct 2015 01:17:29 +0000 (01:17 +0000)]
Update LangRef for changes to the alias textual IR format to support typeless pointer work

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250967 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[WinEH] Remove extraneous call to emitEHRegistrationOffsetLabel
David Majnemer [Wed, 21 Oct 2015 23:20:39 +0000 (23:20 +0000)]
[WinEH] Remove extraneous call to emitEHRegistrationOffsetLabel

It's a relic from the earlier implementation, let's remove it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250964 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[PM]: Fix a doc typo. NFC
Justin Bogner [Wed, 21 Oct 2015 22:51:59 +0000 (22:51 +0000)]
[PM]: Fix a doc typo. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250962 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAMDGPU: Fix adding redundant m0 uses
Matt Arsenault [Wed, 21 Oct 2015 22:37:51 +0000 (22:37 +0000)]
AMDGPU: Fix adding redundant m0 uses

BuildMI already adds these since they are defined correctly now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250961 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAMDGPU: Fix verifier error in SIFoldOperands
Matt Arsenault [Wed, 21 Oct 2015 22:37:50 +0000 (22:37 +0000)]
AMDGPU: Fix verifier error in SIFoldOperands

There may be other use operands that also need their kill flags cleared.

This happens in a few tests when SIFoldOperands is moved after
PeepholeOptimizer.

PeepholeOptimizer rewrites cases that look like:
%vreg0 = ...
%vreg1 = COPY %vreg0
use %vreg1<kill>
%vreg2 = COPY %vreg0
use %vreg2<kill>

to use the earlier source to
%vreg0 = ...
use %vreg0
use %vreg0

Currently SIFoldOperands sees the copied registers, so there is
only one use. So far I haven't managed to come up with a test
that currently has multiple uses of a foldable VGPR -> VGPR copy.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250960 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAMDGPU: Split DiagnosticInfoUnsupported into its own file
Matt Arsenault [Wed, 21 Oct 2015 22:37:46 +0000 (22:37 +0000)]
AMDGPU: Split DiagnosticInfoUnsupported into its own file

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250959 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[JIT] Towards a working small memory model.
Davide Italiano [Wed, 21 Oct 2015 22:12:03 +0000 (22:12 +0000)]
[JIT] Towards a working small memory model.

This commit introduces an option, --preallocate, so that we can get memory
upfront and use it in small memory model tests (in order to get
reliable results).

Differential Revision: http://reviews.llvm.org/D13630

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250956 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAMDGPU: Simplify VOP3 operand legalization.
Matt Arsenault [Wed, 21 Oct 2015 21:51:02 +0000 (21:51 +0000)]
AMDGPU: Simplify VOP3 operand legalization.

This was checking for a variety of situations that should
never happen. This saves a tiny bit of compile time.

We should not be selecting instructions with invalid operands in the
first place. Most of the time for registers copys are inserted
to the correct operand register class.

For VOP3, since all operand types are supported and literal
constants never are, we just need to verify the constant bus
requirements (all immediates should be legal inline ones).

The only possibly tricky case to maybe worry about is if when
legalizing operands in moveToVALU with s_add_i32 and similar
instructions. If the original s_add_i32 had a literal constant
and we need to replace it with v_add_i32_e64 we would have an
unsupported literal operand.  However, I don't think we should worry
about that because SIFoldOperands should handle folding literal
constant operands into the SALU instructions based on the uses.
At SIFoldOperands time, the legality and profitability of
operand types is a bit different.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250951 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[InstCombine] Revise the test case to match full sequene
Michael Liao [Wed, 21 Oct 2015 21:50:58 +0000 (21:50 +0000)]
[InstCombine] Revise the test case to match full sequene

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250950 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAMDGPU: Fix not checking implicit operands in verifyInstruction
Matt Arsenault [Wed, 21 Oct 2015 21:15:01 +0000 (21:15 +0000)]
AMDGPU: Fix not checking implicit operands in verifyInstruction

When verifying constant bus restrictions, this wasn't catching
uses in implicit operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250948 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse numeric_limits instead of LLONG_MAX
Matt Arsenault [Wed, 21 Oct 2015 21:10:12 +0000 (21:10 +0000)]
Use numeric_limits instead of LLONG_MAX

This is a build fix for configurations where LLONG_MAX is
not defined in system headers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250946 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoLegalizeDAG: Implement promote for build_vector
Matt Arsenault [Wed, 21 Oct 2015 21:10:10 +0000 (21:10 +0000)]
LegalizeDAG: Implement promote for build_vector

This will be used in future commits for AMDGPU to promote
operations on i64 vectors into operations on 32-bit vector
components.

This will be used / tested in future AMDGPU commits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250945 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Verifier] Minor comment update, NFC
Vedant Kumar [Wed, 21 Oct 2015 20:33:31 +0000 (20:33 +0000)]
[Verifier] Minor comment update, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250943 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[RuntimeDyld] Ignore ST_FILE symbols when constructing GlobalSymbolTable
Keno Fischer [Wed, 21 Oct 2015 20:22:04 +0000 (20:22 +0000)]
[RuntimeDyld] Ignore ST_FILE symbols when constructing GlobalSymbolTable

Summary: ELF's STT_File symbols may overlap with regular globals in
other files, so we should ignore them here in order to avoid having
bogus entries in the symbol table that confuse us when resolving relocations.

Reviewers: lhames

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D13888

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250942 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Orc] Clean up a comment.
Lang Hames [Wed, 21 Oct 2015 20:13:41 +0000 (20:13 +0000)]
[Orc] Clean up a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250940 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDrop assert that a call with struct return goes to a function with sret
Joerg Sonnenberger [Wed, 21 Oct 2015 20:05:01 +0000 (20:05 +0000)]
Drop assert that a call with struct return goes to a function with sret
attribute. Clang incorrectly misses it on __muldc3 and friends and the
type system doesn't include it properly either.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250938 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[WinEH] Add test for llvm.va.start in catchpad
Reid Kleckner [Wed, 21 Oct 2015 19:54:40 +0000 (19:54 +0000)]
[WinEH] Add test for llvm.va.start in catchpad

It already works, but we should have a test for it.

This used to be PR23094 in the old model.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250936 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSilence Visual C++ warning in function summary parsing code (NFC)
Teresa Johnson [Wed, 21 Oct 2015 19:25:14 +0000 (19:25 +0000)]
Silence Visual C++ warning in function summary parsing code (NFC)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250929 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[x86] move recursive add match for LEA to helper function; NFCI
Sanjay Patel [Wed, 21 Oct 2015 18:56:06 +0000 (18:56 +0000)]
[x86] move recursive add match for LEA to helper function; NFCI

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250926 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert r250923 as config.h is not an installed header.
Yaron Keren [Wed, 21 Oct 2015 18:36:52 +0000 (18:36 +0000)]
Revert r250923 as config.h is not an installed header.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250924 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoInclude llvm/Config/config.h in FileSystem.h as it depends upon HAVE_SYS_STAT_H which...
Yaron Keren [Wed, 21 Oct 2015 18:28:35 +0000 (18:28 +0000)]
Include llvm/Config/config.h in FileSystem.h as it depends upon HAVE_SYS_STAT_H which is defined (or not) in config.h.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250923 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SimplifyCFG] Don't use-after-free an SSA value
David Majnemer [Wed, 21 Oct 2015 18:22:24 +0000 (18:22 +0000)]
[SimplifyCFG] Don't use-after-free an SSA value

SimplifyTerminatorOnSelect didn't consider the possibility that the
condition might be related to one of PHI nodes.

This fixes PR25267.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250922 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[OCaml] Expose Llvm.{set_,}unnamed_addr.
Peter Zotov [Wed, 21 Oct 2015 17:43:02 +0000 (17:43 +0000)]
[OCaml] Expose Llvm.{set_,}unnamed_addr.

Patch by Jacques-Pascal Deplaix <jp.deplaix@gmail.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250912 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Add AMD mwaitx, monitorx, and clzero instructions to the assembly parser and...
Craig Topper [Wed, 21 Oct 2015 17:26:45 +0000 (17:26 +0000)]
[X86] Add AMD mwaitx, monitorx, and clzero instructions to the assembly parser and disassembler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250911 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[x86] add test case that shows holes in LEA isel
Sanjay Patel [Wed, 21 Oct 2015 17:24:00 +0000 (17:24 +0000)]
[x86] add test case that shows holes in LEA isel

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250910 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoBacking out commit r250906 as it broke lld.
Kevin Enderby [Wed, 21 Oct 2015 17:13:20 +0000 (17:13 +0000)]
Backing out commit r250906 as it broke lld.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250908 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoThis removes the eating of the error in Archive::Child::getSize() when the characters
Kevin Enderby [Wed, 21 Oct 2015 16:59:24 +0000 (16:59 +0000)]
This removes the eating of the error in Archive::Child::getSize() when the characters
in the size field in the archive header for the member is not a number.  To do this we
have all of the needed methods return ErrorOr to push them up until we get out of lib.
Then the tools and can handle the error in whatever way is appropriate for that tool.

So the solution is to plumb all the ErrorOr stuff through everything that touches archives.
This include its iterators as one can create an Archive object but the first or any other
Child object may fail to be created due to a bad size field in its header.

Thanks to Lang Hames on the changes making child_iterator contain an
ErrorOr<Child> instead of a Child and the needed changes to ErrorOr.h to add
operator overloading for * and -> .

We don’t want to use llvm_unreachable() as it calls abort() and is produces a “crash”
and using report_fatal_error() to move the error checking will cause the program to
stop, neither of which are really correct in library code. There are still some uses of
these that should be cleaned up in this library code for other than the size field.

Also corrected the code where the size gets us to the “at the end of the archive”
which is OK but past the end of the archive will return object_error::parse_failed now.

The test cases use archives with text files so one can see the non-digit character,
in this case a ‘%’, in the size field.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250906 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Option] Use an ArrayRef to store the Option Infos in OptTable. NFC
Craig Topper [Wed, 21 Oct 2015 16:30:42 +0000 (16:30 +0000)]
[Option] Use an ArrayRef to store the Option Infos in OptTable. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250901 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[llvm-cov] Adjust column widths for function and file reports
Vedant Kumar [Wed, 21 Oct 2015 16:03:32 +0000 (16:03 +0000)]
[llvm-cov] Adjust column widths for function and file reports

Previously, we only expanded function and filename column widths when
rendering file reports. This commit makes the change for function
reports as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250900 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips][mips16] Re-work the inline assembly stubs to work with IAS. NFC.
Daniel Sanders [Wed, 21 Oct 2015 12:44:14 +0000 (12:44 +0000)]
[mips][mips16] Re-work the inline assembly stubs to work with IAS. NFC.

Summary:
Previously, we were inserting an InlineAsm statement for each line of the
inline assembly. This works for GAS but it triggers prologue/epilogue
emission when IAS is in use. This caused:
    .set noreorder
    .cpload $25
to be emitted as:
    .set push
    .set reorder
    .set noreorder
    .set pop
    .set push
    .set reorder
    .cpload $25
    .set pop
which led to assembler errors and caused the test to fail.

The whitespace-after-comma changes included in this patch are necessary to
match the output when IAS is in use.

Reviewers: vkalintiris

Subscribers: rkotler, llvm-commits, dsanders

Differential Revision: http://reviews.llvm.org/D13653

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250895 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[AA] Enhance the new AliasAnalysis infrastructure with an optional
Chandler Carruth [Wed, 21 Oct 2015 12:15:19 +0000 (12:15 +0000)]
[AA] Enhance the new AliasAnalysis infrastructure with an optional
"external" AA wrapper pass.

This is a generic hook that can be used to thread custom code into the
primary AAResultsWrapperPass for the legacy pass manager in order to
allow it to merge external AA results into the AA results it is
building. It does this by threading in a raw callback and so it is
*very* powerful and should serve almost any use case I have come up with
for extending the set of alias analyses used. The only thing not well
supported here is using a *different order* of alias analyses. That form
of extension *is* supportable with the new pass manager, and I can make
the callback structure here more elaborate to support it in the legacy
pass manager if this is a critical use case that people are already
depending on, but the only use cases I have heard of thus far should be
reasonably satisfied by this simpler extension mechanism.

It is hard to test this using normal facilities (the built-in AAs don't
use this for obvious reasons) so I've written a fairly extensive set of
custom passes in the alias analysis unit test that should be an
excellent test case because it models the out-of-tree users: it adds
a totally custom AA to the system. This should also serve as
a reasonably good example and guide for out-of-tree users to follow in
order to rig up their existing alias analyses.

No support in opt for commandline control is provided here however. I'm
really unhappy with the kind of contortions that would be required to
support that. It would fully re-introduce the analysis group
self-recursion kind of patterns. =/

I've heard from out-of-tree users that this will unblock their use cases
with extending AAs on top of the new infrastructure and let us retain
the new analysis-group-free-world.

Differential Revision: http://reviews.llvm.org/D13418

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250894 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMasked Load/Store optimization for scalar code
Elena Demikhovsky [Wed, 21 Oct 2015 11:50:54 +0000 (11:50 +0000)]
Masked Load/Store optimization for scalar code
When we have to convert the masked.load, masked.store to scalar code, we generate a chain of conditional basic blocks.
I added optimization for constant mask vector.

Differential Revision: http://reviews.llvm.org/D13855

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250893 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips][msa] Remove copy_u.d and move copy_u.w to MSA64.
Daniel Sanders [Wed, 21 Oct 2015 09:58:54 +0000 (09:58 +0000)]
[mips][msa] Remove copy_u.d and move copy_u.w to MSA64.

Summary:
The forwards compatibility strategy employed by MIPS is to consider registers
to be infinitely sign-extended. Then on ISA's with a wider register, the result
of existing instructions are sign-extended to register width and zero-extended
counterparts are added. copy_u.w on MSA32 and copy_u.w on MSA64 violate this
strategy and we have therefore corrected the MSA specs to fix this.

We still keep track of sign/zero-extension during legalization but we now
match copy_s.[wd] where required.

No change required to clang since __builtin_msa_copy_u_[wd] will map to
copy_s.[wd] where appropriate for the target.

Reviewers: vkalintiris

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D13472

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250887 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoLet MachineVerifier be aware of mem-to-mem instructions.
Jonas Paulsson [Wed, 21 Oct 2015 07:39:47 +0000 (07:39 +0000)]
Let MachineVerifier be aware of mem-to-mem instructions.

A mem-to-mem instruction (that both loads and stores), which store to an
FI, cannot pass the verifier since it thinks it is loading from the FI.

For the mem-to-mem instruction, do a looser check in visitMachineOperand()
and only check liveness at the reg-slot while analyzing a frame index operand.

Needed to make CodeGen/SystemZ/xor-01.ll pass with -verify-machineinstrs,
which now runs with this flag.

Reviewed by Evan Cheng and Quentin Colombet.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250885 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDo not use `dyn_cast<X>` after `isa<X>` (NFC)
Mehdi Amini [Wed, 21 Oct 2015 06:11:01 +0000 (06:11 +0000)]
Do not use `dyn_cast<X>` after `isa<X>` (NFC)

From: Mehdi Amini <mehdi.amini@apple.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250883 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert "Add missing #include, found by modules build."
Mehdi Amini [Wed, 21 Oct 2015 06:10:55 +0000 (06:10 +0000)]
Revert "Add missing #include, found by modules build."

This reverts commit r250239.
It seems unwanted changes got committed here, and part of
the patch does not seem correct.
For instance RoundUpToAlignment() is called without its returned
value actually used.

From: Mehdi Amini <mehdi.amini@apple.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250882 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoTail duplication can mix incompatible registers in phi nodes
Krzysztof Parzyszek [Wed, 21 Oct 2015 02:40:06 +0000 (02:40 +0000)]
Tail duplication can mix incompatible registers in phi nodes

Do not tail duplicate blocks where the successor has a phi node,
and the corresponding value in that phi node uses a subregister.

http://reviews.llvm.org/D13922

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250877 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoWebAssembly: support imports
JF Bastien [Wed, 21 Oct 2015 02:23:09 +0000 (02:23 +0000)]
WebAssembly: support imports

C/C++ code can declare an extern function, which will show up as an import in WebAssembly's output. It's expected that the linker will resolve these, and mark unresolved imports as call_import (I have a patch which does this in wasmate).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250875 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoTolerate negative offset when matching sample profile.
Dehao Chen [Wed, 21 Oct 2015 01:22:27 +0000 (01:22 +0000)]
Tolerate negative offset when matching sample profile.

In some cases (as illustrated in the unittest), lineno can be less than the heade_lineno because the function body are included from some other files. In this case, offset will be negative. This patch makes clang still able to match the profile to IR in this situation.

http://reviews.llvm.org/D13914

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250873 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Hexagon] Bit-based instruction simplification
Krzysztof Parzyszek [Tue, 20 Oct 2015 22:57:13 +0000 (22:57 +0000)]
[Hexagon] Bit-based instruction simplification

Analyze bit patterns of operands and values of instructions to perform
various simplifications, dead/redundant code elimination, etc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250868 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Hexagon] Fix isNVStorable flag in .td files
Krzysztof Parzyszek [Tue, 20 Oct 2015 22:40:57 +0000 (22:40 +0000)]
[Hexagon] Fix isNVStorable flag in .td files

An upper half and a double word cannot be used as value sources in a
new-value store.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250867 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[MemorySanitizer] NFC. Do not use GET_INTRINSIC_MODREF_BEHAVIOR table.
Igor Laevsky [Tue, 20 Oct 2015 21:33:30 +0000 (21:33 +0000)]
[MemorySanitizer] NFC. Do not use GET_INTRINSIC_MODREF_BEHAVIOR table.

It is now possible to infer intrinsic modref behaviour purely from intrinsic attributes.
This change will allow to completely remove GET_INTRINSIC_MODREF_BEHAVIOR table.

Differential Revision: http://reviews.llvm.org/D13907

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250860 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86][SSE] Add 256-bit vector bit rotation tests.
Simon Pilgrim [Tue, 20 Oct 2015 20:27:23 +0000 (20:27 +0000)]
[X86][SSE] Add 256-bit vector bit rotation tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250853 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agobugpoint: Remove implicit ilist iterator conversions, NFC
Duncan P. N. Exon Smith [Tue, 20 Oct 2015 19:36:39 +0000 (19:36 +0000)]
bugpoint: Remove implicit ilist iterator conversions, NFC

This is the last of the implicit ilist iterator conversions in LLVM.
Still up for debate whether we let these bitrot back:
http://lists.llvm.org/pipermail/llvm-dev/2015-October/091617.html

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250852 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Hexagon] Capture aggregate variables by reference, not value
Krzysztof Parzyszek [Tue, 20 Oct 2015 19:33:46 +0000 (19:33 +0000)]
[Hexagon] Capture aggregate variables by reference, not value

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250851 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Hexagon] Do not fall-through if there is no CFG edge
Krzysztof Parzyszek [Tue, 20 Oct 2015 19:30:21 +0000 (19:30 +0000)]
[Hexagon] Do not fall-through if there is no CFG edge

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250850 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Hexagon] Use symbolic name for subregister instead of hardcoded number
Krzysztof Parzyszek [Tue, 20 Oct 2015 19:26:36 +0000 (19:26 +0000)]
[Hexagon] Use symbolic name for subregister instead of hardcoded number

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250849 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Hexagon] Change Based->Base in getBasedWithImmOffset
Krzysztof Parzyszek [Tue, 20 Oct 2015 19:21:05 +0000 (19:21 +0000)]
[Hexagon] Change Based->Base in getBasedWithImmOffset

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250848 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Hexagon] Remove the remnants of isConstExtProfitable
Krzysztof Parzyszek [Tue, 20 Oct 2015 19:04:53 +0000 (19:04 +0000)]
[Hexagon] Remove the remnants of isConstExtProfitable

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250845 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agounittests: Remove implicit ilist iterator conversions, NFC
Duncan P. N. Exon Smith [Tue, 20 Oct 2015 18:30:20 +0000 (18:30 +0000)]
unittests: Remove implicit ilist iterator conversions, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250843 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agollvm-diff: Remove implicit ilist iterator conversions, NFC
Duncan P. N. Exon Smith [Tue, 20 Oct 2015 18:17:05 +0000 (18:17 +0000)]
llvm-diff: Remove implicit ilist iterator conversions, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250842 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[CMake] All the checks for if LLVM_VERSION_* variables are set need to be if(DEFINED...
Chris Bieneman [Tue, 20 Oct 2015 18:16:37 +0000 (18:16 +0000)]
[CMake] All the checks for if LLVM_VERSION_* variables are set need to be if(DEFINED ...)

This is because if you set one of the variables to 0, if(NOT ...) is true, which isn't what you actually want. Should have thought that through better the first time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250841 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[CMake] Refactor subdirectory inclusion code to take a project name.
Chris Bieneman [Tue, 20 Oct 2015 16:42:58 +0000 (16:42 +0000)]
[CMake] Refactor subdirectory inclusion code to take a project name.

Summary:
This refactoring makes some of the code used to control including subdirectories parameterized so it can be re-used elsewhere.

Specifically I want to re-use this code in clang to be able to turn off specific tool subdirectories.

Reviewers: chapuni, filcab, bogner, Bigcheese

Subscribers: emaste, llvm-commits

Differential Revision: http://reviews.llvm.org/D13783

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250835 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoTwo switch blocks in VectorLegalizer::LegalizeOp already have a
Artyom Skrobov [Tue, 20 Oct 2015 15:06:37 +0000 (15:06 +0000)]
Two switch blocks in VectorLegalizer::LegalizeOp already have a

  default: llvm_unreachable("This action is not supported yet!");

-- so I'm adding one to the third switch block, too.

This is a follow-up fix for http://reviews.llvm.org/D13862

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250830 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SystemZ] Use LivePhysRegs helper class in SystemZShortenInst.cpp.
Jonas Paulsson [Tue, 20 Oct 2015 15:05:58 +0000 (15:05 +0000)]
[SystemZ] Use LivePhysRegs helper class in SystemZShortenInst.cpp.

Don't use home brewed liveness tracking code for phys regs, since
this class does the job.

Reviewed by Ulrich Weigand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250829 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SystemZ] Comment fix in test/CodeGen/SystemZ/fp-cmp-05.ll
Jonas Paulsson [Tue, 20 Oct 2015 15:05:54 +0000 (15:05 +0000)]
[SystemZ] Comment fix in test/CodeGen/SystemZ/fp-cmp-05.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250828 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdding support for TargetLoweringBase::LibCall
Artyom Skrobov [Tue, 20 Oct 2015 13:14:52 +0000 (13:14 +0000)]
Adding support for TargetLoweringBase::LibCall

Summary:
TargetLoweringBase::Expand is defined as "Try to expand this to other ops,
otherwise use a libcall." For ISD::UDIV and ISD::SDIV, the choice between
the two possibilities was defined in a rather convoluted way:

- if DIVREM is legal, expand to DIVREM
- if DIVREM has a custom lowering, expand to DIVREM
- if DIVREM libcall is defined and a remainder from the same division is
  computed elsewhere, expand to a DIVREM libcall
- else, expand to a DIV libcall

This had the undesirable effect that if both DIV and DIVREM are implemented
as libcalls, then ISD::UDIV and ISD::SDIV are expanded to the heavier DIVREM
libcall, even when the remainder isn't used.

The new code adds a new LegalizeAction, TargetLoweringBase::LibCall, so that
backends can directly control whether they prefer an expansion or a conversion
to a libcall. This makes the generic lowering code even more generic,
allowing its reuse in a wider range of target-specific configurations.

The useful effect is that ARM backend will now generate a call
to __aeabi_{i,u}div rather than __aeabi_{i,u}divmod in cases where
it doesn't need the remainder. There's no functional change outside
the ARM backend.

Reviewers: t.p.northover, rengolin

Subscribers: t.p.northover, llvm-commits, aemerson

Differential Revision: http://reviews.llvm.org/D13862

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250826 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCombining DIV+REM->DIVREM doesn't belong in LegalizeDAG; move it over into DAGCombiner.
Artyom Skrobov [Tue, 20 Oct 2015 13:06:02 +0000 (13:06 +0000)]
Combining DIV+REM->DIVREM doesn't belong in LegalizeDAG; move it over into DAGCombiner.

Summary:
In addition to moving the code over, this patch amends the DIV,REM -> DIVREM
combining to run on all affected nodes at once: if the nodes are converted
to DIVREM one at a time, then the resulting DIVREM may get legalized by the
backend into something target-specific that we won't be able to recognize
and correlate with the remaining nodes.

The motivation is to "prepare terrain" for D13862: when we set DIV and REM
to be legalized to libcalls, instead of the DIVREM, we otherwise lose the
ability to combine them together. To prevent this, we need to take the
DIV,REM -> DIVREM combining out of the lowering stage.

Reviewers: RKSimon, eli.friedman, rengolin

Subscribers: john.brawn, rengolin, llvm-commits

Differential Revision: http://reviews.llvm.org/D13733

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250825 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAVX512: Implemented encoding and intrinsics for VPBROADCASTB/W/D/Q instructions.
Igor Breger [Tue, 20 Oct 2015 11:56:42 +0000 (11:56 +0000)]
AVX512: Implemented encoding and intrinsics for VPBROADCASTB/W/D/Q instructions.

Differential Revision: http://reviews.llvm.org/D13884

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250819 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[x86] Fix AVX maskload/store intrinsic prototypes.
Andrea Di Biagio [Tue, 20 Oct 2015 11:20:13 +0000 (11:20 +0000)]
[x86] Fix AVX maskload/store intrinsic prototypes.

The mask value type for maskload/maskstore GCC builtins is never a vector of
packed floats/doubles.

This patch fixes the following issues:
1. The mask argument for builtin_ia32_maskloadpd and builtin_ia32_maskstorepd
   should be of type llvm_v2i64_ty and not llvm_v2f64_ty.
2. The mask argument for builtin_ia32_maskloadpd256 and
   builtin_ia32_maskstorepd256 should be of type llvm_v4i64_ty and not
   llvm_v4f64_ty.
3. The mask argument for builtin_ia32_maskloadps and builtin_ia32_maskstoreps
   should be of type llvm_v4i32_ty and not llvm_v4f32_ty.
4. The mask argument for builtin_ia32_maskloadps256 and
   builtin_ia32_maskstoreps256 should be of type llvm_v8i32_ty and not
   llvm_v8f32_ty.

Differential Revision: http://reviews.llvm.org/D13776

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250817 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix missing INITIALIZE_PASS_DEPENDENCY for AddressSanitizer
Keno Fischer [Tue, 20 Oct 2015 10:13:55 +0000 (10:13 +0000)]
Fix missing INITIALIZE_PASS_DEPENDENCY for AddressSanitizer

Summary: In r231241, TargetLibraryInfoWrapperPass was added to
`getAnalysisUsage` for `AddressSanitizer`, but the corresponding
`INITIALIZE_PASS_DEPENDENCY` was not added.

Reviewers: dvyukov, chandlerc, kcc

Subscribers: kcc, llvm-commits

Differential Revision: http://reviews.llvm.org/D13629

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250813 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMake class final to pacify -Wnon-virtual-dtor.
Manuel Klimek [Tue, 20 Oct 2015 08:21:01 +0000 (08:21 +0000)]
Make class final to pacify -Wnon-virtual-dtor.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250805 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAMDGPU: Add MachineInstr overloads for instruction format tests
Matt Arsenault [Tue, 20 Oct 2015 04:35:43 +0000 (04:35 +0000)]
AMDGPU: Add MachineInstr overloads for instruction format tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@250797 91177308-0d34-0410-b5e6-96231b3b80d8