oota-llvm.git
9 years ago[X86][SSE] Added additional vector sign/zero load extension tests.
Simon Pilgrim [Sat, 25 Jul 2015 14:07:20 +0000 (14:07 +0000)]
[X86][SSE] Added additional vector sign/zero load extension tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243216 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86][SSE] Added additional vector sign/zero extension tests.
Simon Pilgrim [Sat, 25 Jul 2015 11:17:35 +0000 (11:17 +0000)]
[X86][SSE] Added additional vector sign/zero extension tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243212 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[LoopUnswitch] Improve loop unswitch pass to find trivial unswitch conditions more...
Chen Li [Sat, 25 Jul 2015 03:21:06 +0000 (03:21 +0000)]
[LoopUnswitch] Improve loop unswitch pass to find trivial unswitch conditions more effectively

Summary:
This patch improves trivial loop unswitch.

The current trivial loop unswitch only checks if loop header's terminator contains a trivial unswitch condition. But if the loop header only has one reachable successor (due to intentionally or unintentionally missed code simplification), we should consider the successor as part of the loop header. Therefore, instead of stopping at loop header's terminator, we should keep traversing its successors within loop until reach a *real* conditional branch or switch (whose condition can not be constant folded). This change will enable a single -loop-unswitch pass to unswitch multiple trivial conditions (unswitch one trivial condition could open opportunity to unswitch another one in the same loop), while the old implementation can unswitch only one per pass.

Reviewers: reames, broune

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D11481

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243203 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[AArch64][FastISel] Always use an AND instruction when truncating to non-legal types.
Juergen Ributzka [Sat, 25 Jul 2015 02:16:53 +0000 (02:16 +0000)]
[AArch64][FastISel] Always use an AND instruction when truncating to non-legal types.

When truncating to non-legal types (such as i16, i8 and i1) always use an AND
instruction to mask out the upper bits. This was only done when the source type
was an i64, but not when the source type was an i32.

This commit fixes this and adds the missing i32 truncate tests.

This fixes rdar://problem/21990703.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243198 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix PPCMaterializeInt to check the size of the integer based on the
Eric Christopher [Sat, 25 Jul 2015 00:48:08 +0000 (00:48 +0000)]
Fix PPCMaterializeInt to check the size of the integer based on the
extension property we're requesting - zero or sign extended.

This fixes cases where we want to return a zero extended 32-bit -1
and not be sign extended for the entire register. Also updated the
already out of date comment with the current behavior.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243192 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoPPCMaterializeInt should only take a ConstantInt so represent this in the prototype
Eric Christopher [Sat, 25 Jul 2015 00:48:06 +0000 (00:48 +0000)]
PPCMaterializeInt should only take a ConstantInt so represent this in the prototype
and fix up all uses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243191 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[AArch64] Define subtarget feature "reserve-x18", which is used to decide
Akira Hatanaka [Sat, 25 Jul 2015 00:18:31 +0000 (00:18 +0000)]
[AArch64] Define subtarget feature "reserve-x18", which is used to decide
whether register x18 should be reserved.

This change is needed because we cannot use a backend option to set
cl::opt "aarch64-reserve-x18" when doing LTO.

Out-of-tree projects currently using cl::opt option "-aarch64-reserve-x18"
to reserve x18 should make changes to add subtarget feature "reserve-x18"
to the IR.

rdar://problem/21529937

Differential Revision: http://reviews.llvm.org/D11463

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243186 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDI/Verifier: Fix argument bitrot in DILocalVariable
Duncan P. N. Exon Smith [Fri, 24 Jul 2015 23:59:25 +0000 (23:59 +0000)]
DI/Verifier: Fix argument bitrot in DILocalVariable

Add a verifier check that `DILocalVariable`s of tag
`DW_TAG_arg_variable` always have a non-zero 'arg:' field, and those of
tag `DW_TAG_auto_variable` always have a zero 'arg:' field.  These are
the only configurations that are properly understood by the backend.

(Also, fix the bad examples in LangRef and test/Assembler, and fix the
bug in Kaleidoscope Ch8.)

A large number of testcases seem to have bitrotted their way forward
from some ancient version of the debug info hierarchy that didn't have
`arg:` parameters.  If you have out-of-tree testcases that start failing
in the verifier and you don't care enough to get the `arg:` right, you
may have some luck just calling:

    sed -e 's/, arg: 0/, arg: 1/'

or some such, but I hand-updated the ones in tree.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243183 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMIR Serialization: Serialize MachineFrameInfo's callee saved information.
Alex Lorenz [Fri, 24 Jul 2015 22:22:50 +0000 (22:22 +0000)]
MIR Serialization: Serialize MachineFrameInfo's callee saved information.

This commit serializes the callee saved information from the class
'MachineFrameInfo'. This commit extends the YAML mappings for the fixed and
the ordinary stack objects and adds an optional 'callee-saved-register'
attribute. This attribute is used to serialize the callee save information.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243173 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoHandle loop with negtive induction variable increment
Lawrence Hu [Fri, 24 Jul 2015 22:01:49 +0000 (22:01 +0000)]
Handle loop with negtive induction variable increment

This patch extend LoopReroll pass to hand the loops which
is similar to the following:

      while (len > 1) {
            sum4 += buf[len];
            sum4 += buf[len-1];
            len -= 2;
        }

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243171 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove unnecessary null check. NFC.
Pete Cooper [Fri, 24 Jul 2015 21:38:01 +0000 (21:38 +0000)]
Remove unnecessary null check.  NFC.

Since both places which set this variable do so with dyn_cast, and not
dyn_cast_or_null, its impossible to get a nullptr here, so we can remove
the check.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243167 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse make_range(rbegin(), rend()) to allow foreach loops. NFC.
Pete Cooper [Fri, 24 Jul 2015 21:13:43 +0000 (21:13 +0000)]
Use make_range(rbegin(), rend()) to allow foreach loops.  NFC.

Instead of the pattern

for (auto I = x.rbegin(), E = x.end(); I != E; ++I)

we can use make_range to construct the reverse range and iterate using
that instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243163 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDI: Fix unit tests after r243160
Duncan P. N. Exon Smith [Fri, 24 Jul 2015 21:11:06 +0000 (21:11 +0000)]
DI: Fix unit tests after r243160

These always empty fields are gone, so don't test that they're empty.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243162 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDI: Remove unnecessary DICompositeTypeBase
Duncan P. N. Exon Smith [Fri, 24 Jul 2015 20:56:36 +0000 (20:56 +0000)]
DI: Remove unnecessary DICompositeTypeBase

Remove unnecessary and confusing common base class for `DICompositeType`
and `DISubroutineType`.

While at a high-level `DISubroutineType` is a sort of composite of other
types, it has no shared code paths, and its fields are completely
disjoint.  This relationship was left over from the old debug info
hierarchy.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243160 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDI: Simplify DebugInfoFinder::processType(), NFC
Duncan P. N. Exon Smith [Fri, 24 Jul 2015 20:56:10 +0000 (20:56 +0000)]
DI: Simplify DebugInfoFinder::processType(), NFC

Handle `DISubroutineType` up-front rather than as part of a branch for
`DICompositeTypeBase`.  The only shared code path was looking through
the base type, but `DISubroutineType` can never have a base type.

This also removes the last use of `DICompositeTypeBase`, since we can
strengthen the cast to `DICompositeType`.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243159 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDI: Remove dead code: getDICompositeType()
Duncan P. N. Exon Smith [Fri, 24 Jul 2015 20:46:46 +0000 (20:46 +0000)]
DI: Remove dead code: getDICompositeType()

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243158 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAsmPrinter: Use DICompositeType in updateAcceleratorTables(), NFC
Duncan P. N. Exon Smith [Fri, 24 Jul 2015 20:45:26 +0000 (20:45 +0000)]
AsmPrinter: Use DICompositeType in updateAcceleratorTables(), NFC

`DISubroutineType` is impossible at this `dyn_cast` site, since we're
only dealing with named types and `DISubroutineType` cannot be named.
Strengthen the `dyn_cast` to `DICompositeType`.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243157 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMIR Serialization: Serialize the simple virtual register allocation hints.
Alex Lorenz [Fri, 24 Jul 2015 20:35:40 +0000 (20:35 +0000)]
MIR Serialization: Serialize the simple virtual register allocation hints.

This commit serializes the virtual register allocations hints of type 0.
These hints specify the preferred physical registers for allocations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243156 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDI: Remove DIDerivedTypeBase
Duncan P. N. Exon Smith [Fri, 24 Jul 2015 20:16:36 +0000 (20:16 +0000)]
DI: Remove DIDerivedTypeBase

Remove an unnecessary (and confusing) common subclass for
`DIDerivedType` and `DICompositeType`.  These classes aren't really
related, and even in the old debug info hierarchy, there was a
long-standing FIXME to separate them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243152 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoVerifier: Sink filename check into visitMDCompositeType(), NFC
Duncan P. N. Exon Smith [Fri, 24 Jul 2015 19:57:19 +0000 (19:57 +0000)]
Verifier: Sink filename check into visitMDCompositeType(), NFC

We really only want to check this for unions and classes (all the other
tags have been ruled out), so simplify the check and move it to the
right place.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243150 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoVerifier: Remove unnecessary references to DW_TAG_subroutine_type, NFC
Duncan P. N. Exon Smith [Fri, 24 Jul 2015 19:52:18 +0000 (19:52 +0000)]
Verifier: Remove unnecessary references to DW_TAG_subroutine_type, NFC

Remove unnecessary references to `DW_TAG_subroutine_type` in
`visitDICompositeType()` and `visitDIDerivedTypeBase()`, since
`visitDISubroutineType()` doesn't call either of those (and shouldn't,
since subroutine types are really quite special).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243149 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDI: Clarify isUnsignedDIType(), NFC
Duncan P. N. Exon Smith [Fri, 24 Jul 2015 19:42:12 +0000 (19:42 +0000)]
DI: Clarify isUnsignedDIType(), NFC

Refactor `isUnsignedDIType()` to deal with `DICompositeType` explicitly.
Since `DW_TAG_subroutine_type` isn't handled here (the assertions about
tags rule it out), this allows strengthening the `dyn_cast` to
`DIDerivedType`.

Besides making the code clearer, this it removes a use of
`DIDerivedTypeBase`.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243148 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd const to some Type* parameters which didn't need to be mutable. NFC.
Pete Cooper [Fri, 24 Jul 2015 19:19:26 +0000 (19:19 +0000)]
Add const to some Type* parameters which didn't need to be mutable.  NFC.

We were only getting the size of the type which doesn't need to modify
the type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243146 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove unused variable. NFC.
Diego Novillo [Fri, 24 Jul 2015 19:18:32 +0000 (19:18 +0000)]
Remove unused variable. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243145 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDI: Strengthen some dyn_casts to DIDerivedType, NFC
Duncan P. N. Exon Smith [Fri, 24 Jul 2015 19:17:20 +0000 (19:17 +0000)]
DI: Strengthen some dyn_casts to DIDerivedType, NFC

The surrounding code proves in both cases that these must be
`DIDerivedType` if they're `DIDerivedTypeBase`, so strengthen the
`dyn_cast`s to the more specific type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243143 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove the user-count threshold when analyzing read attributes
Jingyue Wu [Fri, 24 Jul 2015 19:05:53 +0000 (19:05 +0000)]
Remove the user-count threshold when analyzing read attributes

Summary:
This threshold limited FunctionAttrs ability to prove arguments to be read-only.
In NVPTX, a specialized instruction ld.global.nc can be used to load memory
with non-coherent texture cache. We notice that in SHOC [1] benchmark, some
function arguments are not marked with readonly because FunctionAttrs reaches
a hardcoded threshold when analysis uses.

Removing this threshold won't cause significant regression in compilation time, because the worst-case time complexity of the algorithm is still O(# of instructions) for each parameter.

Patched by Xuetian Weng.

[1] https://github.com/vetter/shoc

Reviewers: nlewycky, jingyue, nicholas

Subscribers: nicholas, test, llvm-commits

Differential Revision: http://reviews.llvm.org/D11311

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243141 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[RewriteStatepointsForGC] Adjust naming scheme to be more stable
Philip Reames [Fri, 24 Jul 2015 19:01:39 +0000 (19:01 +0000)]
[RewriteStatepointsForGC] Adjust naming scheme to be more stable

The names for instructions inserted were previous dependent on iteration order.  By deriving the names from the original instructions, we can avoid instability in tests without resorting to ordered traversals.  It also makes the IR mildly easier to read at large scale.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243140 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDI: Strengthen block-byref cast to DIDerivedType, NFC
Duncan P. N. Exon Smith [Fri, 24 Jul 2015 18:58:32 +0000 (18:58 +0000)]
DI: Strengthen block-byref cast to DIDerivedType, NFC

This code is visiting the members of a block-byref, and we know those
are all `DIDerivedType`.  Strengthen the cast.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243138 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse foreach loops for StructType::elements(). NFC.
Pete Cooper [Fri, 24 Jul 2015 18:55:49 +0000 (18:55 +0000)]
Use foreach loops for StructType::elements().  NFC.

We had a few places where we did

for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {

but those could instead do

for (auto *EltTy : STy->elements()) {

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243136 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd const to a bunch of Type* in DataLayout. NFC.
Pete Cooper [Fri, 24 Jul 2015 18:29:09 +0000 (18:29 +0000)]
Add const to a bunch of Type* in DataLayout.  NFC.

Almost all methods in DataLayout took mutable pointers but didn't need to.
These were only accessing constant methods of the types, or using the Type*
to key a map.  Neither of these needs a mutable pointer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243135 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDI: Only DICompositeType has getElements(), NFC
Duncan P. N. Exon Smith [Fri, 24 Jul 2015 18:17:17 +0000 (18:17 +0000)]
DI: Only DICompositeType has getElements(), NFC

There is an assertion inside `DICompositeTypeBase::getElements()` that
`this` is not a `DISubroutineType`, leaving only `DICompositeType`.
Make that clear at the call sites.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243134 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMIR Parser: Run the machine verifier after initializing machine functions.
Alex Lorenz [Fri, 24 Jul 2015 17:44:49 +0000 (17:44 +0000)]
MIR Parser: Run the machine verifier after initializing machine functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243128 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[RuntimeDyld] MachO: Add support for ARM scattered vanilla relocations.
Lang Hames [Fri, 24 Jul 2015 17:40:04 +0000 (17:40 +0000)]
[RuntimeDyld] MachO: Add support for ARM scattered vanilla relocations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243126 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMIR Tests: Add liveins and successors to make tests pass with machine verifier.
Alex Lorenz [Fri, 24 Jul 2015 17:36:55 +0000 (17:36 +0000)]
MIR Tests: Add liveins and successors to make tests pass with machine verifier.

This commit adds the liveins and successors properties to machine basic blocks
in some of the MIR tests to ensure that the tests will pass when the MIR parser
will run the machine verifier after initializing a machine function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243124 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMIR Tests: Make the basic block successor test an X86 specific test.
Alex Lorenz [Fri, 24 Jul 2015 17:31:55 +0000 (17:31 +0000)]
MIR Tests: Make the basic block successor test an X86 specific test.

This commit moves and transforms the generic test
'CodeGen/MIR/successor-basic-blocks.mir' into an X86 specific test
'CodeGen/MIR/X86/successor-basic-blocks.mir'. This change is required in order
to enable the machine verifier for the MIR parser, as the machine verifier
verifies that the machine basic blocks contain instructions that actually
determine the machine basic block successors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243123 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAVX-512: Implemented encoding , DAG lowering and intrinsics for Integer Truncate...
Igor Breger [Fri, 24 Jul 2015 17:24:15 +0000 (17:24 +0000)]
AVX-512: Implemented encoding , DAG lowering and intrinsics for Integer Truncate with/without saturation
Added tests for DAG lowering ,encoding and intrinsic

Differential Revision: http://reviews.llvm.org/D11218

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243122 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUpdate for r243115 which changed the DataLayout API on TargetMachine but
Chandler Carruth [Fri, 24 Jul 2015 17:23:09 +0000 (17:23 +0000)]
Update for r243115 which changed the DataLayout API on TargetMachine but
didn't update the gold-plugin.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243121 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agotest-release.sh: Defer test errors until the end
Hans Wennborg [Fri, 24 Jul 2015 16:16:09 +0000 (16:16 +0000)]
test-release.sh: Defer test errors until the end

This makes the script run to the end and produce tarballs even on test
failures, and then highlights any errors afterwards.

(I first tried just storing the errors in a global variable, but that
didn't work as the "test_llvmCore" function invocation is actually
running as a sub-shell.)

Differential Revision: http://reviews.llvm.org/D11478

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243116 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove access to the DataLayout in the TargetMachine
Mehdi Amini [Fri, 24 Jul 2015 16:04:22 +0000 (16:04 +0000)]
Remove access to the DataLayout in the TargetMachine

Summary:
Replace getDataLayout() with a createDataLayout() method to make
explicit that it is intended to create a DataLayout only and not
accessing it for other purpose.

This change is the last of a series of commits dedicated to have a
single DataLayout during compilation by using always the one owned
by the module.

Reviewers: echristo

Subscribers: jholewinski, llvm-commits, rafael, yaron.keren

Differential Revision: http://reviews.llvm.org/D11103

(cherry picked from commit 5609fc56bca971e5a7efeaa6ca4676638eaec5ea)

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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243114 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agofix wrong comment; NFC
Sanjay Patel [Fri, 24 Jul 2015 16:02:14 +0000 (16:02 +0000)]
fix wrong comment; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243113 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agollvm/test/tools/dsymutil/ARM/lit.local.cfg: Fix possibly typo, s/X86/ARM/.
NAKAMURA Takumi [Fri, 24 Jul 2015 11:55:11 +0000 (11:55 +0000)]
llvm/test/tools/dsymutil/ARM/lit.local.cfg: Fix possibly typo, s/X86/ARM/.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243106 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[ARM] - Fix lowering of shufflevectors in AArch32
Luke Cheeseman [Fri, 24 Jul 2015 09:57:05 +0000 (09:57 +0000)]
[ARM] - Fix lowering of shufflevectors in AArch32

Some shufflevectors are currently being incorrectly lowered in the AArch32
backend as the existing checks for detecting the NEON operations from the
shufflevector instruction expects the shuffle mask and the vector operands to be
of the same length.

This is not always the case as the mask may be twice as long as the operand;
here only the lower half of the shufflemask gets checked, so provided the lower
half of the shufflemask looks like a vector transpose (or even is just all -1
for undef) then the intrinsics may get incorrectly lowered into a vector
transpose (VTRN) instruction.

This patch fixes this by accommodating for both cases and adds regression tests.

Differential Revision: http://reviews.llvm.org/D11407

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243103 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoWhen lowering vector shifts a check is performed to see if the value to shift by
Luke Cheeseman [Fri, 24 Jul 2015 09:31:48 +0000 (09:31 +0000)]
When lowering vector shifts a check is performed to see if the value to shift by
is an immediate, in this check the value is negated and stored in and int64_t.
The value can be -2^63 yet the result cannot be stored in an int64_t and this
gives some undefined behaviour causing failures. The negation is only necessary
when the values is within a certain range and so it should not need to negate
-2^63, this patch introduces this and also a regression test.

Differential Revision: http://reviews.llvm.org/D11408

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243100 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[dsymutil] Implement support for universal mach-o object files.
Frederic Riss [Fri, 24 Jul 2015 06:41:11 +0000 (06:41 +0000)]
[dsymutil] Implement support for universal mach-o object files.

This patch allows llvm-dsymutil to read universal (aka fat) macho object
files and archives. The patch touches nearly everything in the BinaryHolder,
but it is fairly mechinical: the methods that returned MemoryBufferRefs or
ObjectFiles now return a vector of those, and the high-level access function
takes a triple argument to select the architecture.

There is no support yet for handling fat executables and thus no support for
writing fat object files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243096 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[dsymutil] Make the triple detection more strict.
Frederic Riss [Fri, 24 Jul 2015 06:41:04 +0000 (06:41 +0000)]
[dsymutil] Make the triple detection more strict.

MachOObjectFile offers a method for detecting the correct triple, use
it instead of the previous approximation. This doesn't matter right
now, but it will become important for mach-o universal (fat) binaries.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243095 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[dsymutil] Refactor BinaryHolder internals. NFC
Frederic Riss [Fri, 24 Jul 2015 06:40:59 +0000 (06:40 +0000)]
[dsymutil] Refactor BinaryHolder internals. NFC

Call a helper that resets all the internal state of the BinaryHolder
when we change the underlying memory buffer. Makes a followup patch
a tiny bit smaller.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243094 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert "Remove access to the DataLayout in the TargetMachine"
Mehdi Amini [Fri, 24 Jul 2015 03:36:55 +0000 (03:36 +0000)]
Revert "Remove access to the DataLayout in the TargetMachine"

This reverts commit 0f720d984f419c747709462f7476dff962c0bc41.

It breaks clang too badly, I need to prepare a proper patch for clang
first.

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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243089 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[bpf] initial support for debug_info
Alexei Starovoitov [Fri, 24 Jul 2015 03:17:08 +0000 (03:17 +0000)]
[bpf] initial support for debug_info

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243087 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[llvm-reaobj] Display COFF-specific sections/tables only if the object is COFF.
Davide Italiano [Fri, 24 Jul 2015 02:14:20 +0000 (02:14 +0000)]
[llvm-reaobj] Display COFF-specific sections/tables only if the object is COFF.

Just skip them otherwise.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243086 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoHandle resolvable branches in complete loop unroll heuristic.
Michael Zolotukhin [Fri, 24 Jul 2015 01:53:04 +0000 (01:53 +0000)]
Handle resolvable branches in complete loop unroll heuristic.

Summary:
Resolving a branch allows us to ignore blocks that won't be executed, and thus make our estimate more accurate.
This patch is intended to be applied after D10205 (though it could be applied independently).

Reviewers: chandlerc

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D10206

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243084 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove access to the DataLayout in the TargetMachine
Mehdi Amini [Fri, 24 Jul 2015 01:44:39 +0000 (01:44 +0000)]
Remove access to the DataLayout in the TargetMachine

Summary:
Replace getDataLayout() with a createDataLayout() method to make
explicit that it is intended to create a DataLayout only and not
accessing it for other purpose.

This change is the last of a series of commits dedicated to have a
single DataLayout during compilation by using always the one owned
by the module.

Reviewers: echristo

Subscribers: jholewinski, llvm-commits, rafael, yaron.keren

Differential Revision: http://reviews.llvm.org/D11103

(cherry picked from commit 5609fc56bca971e5a7efeaa6ca4676638eaec5ea)

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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243083 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agobuild: fix small typo in cmake doxygen build
Saleem Abdulrasool [Fri, 24 Jul 2015 01:14:25 +0000 (01:14 +0000)]
build: fix small typo in cmake doxygen build

A search word spelled as "searhc" in the LLVM_DOXYGEN_SEARCHENGINE_URL cmake
variable docstring.

Patch by Daniel Otero!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243082 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMIRParser/LLVMBuild.txt: Add MC for MCRegisterInfo::getDwarfRegNum().
NAKAMURA Takumi [Fri, 24 Jul 2015 01:12:36 +0000 (01:12 +0000)]
MIRParser/LLVMBuild.txt: Add MC for MCRegisterInfo::getDwarfRegNum().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243081 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoReorder alphabetically.
NAKAMURA Takumi [Fri, 24 Jul 2015 01:12:28 +0000 (01:12 +0000)]
Reorder alphabetically.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243080 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoClean up function attributes on PPC fast-isel tests.
Eric Christopher [Fri, 24 Jul 2015 01:07:50 +0000 (01:07 +0000)]
Clean up function attributes on PPC fast-isel tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243079 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[libFuzzer] allow users to supply their own implementation of rand
Kostya Serebryany [Fri, 24 Jul 2015 01:06:40 +0000 (01:06 +0000)]
[libFuzzer] allow users to supply their own implementation of rand

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243078 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[RewriteStatepointsForGC] Fix release build warning
Philip Reames [Fri, 24 Jul 2015 00:42:55 +0000 (00:42 +0000)]
[RewriteStatepointsForGC] Fix release build warning

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243076 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd missing underlines for a docs section. NFC
Jonathan Roelofs [Fri, 24 Jul 2015 00:29:50 +0000 (00:29 +0000)]
Add missing underlines for a docs section. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243075 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[RewriteStatepointsForGC] Use a worklist algorithm for first part of base pointer...
Philip Reames [Fri, 24 Jul 2015 00:02:11 +0000 (00:02 +0000)]
[RewriteStatepointsForGC] Use a worklist algorithm for first part of base pointer algorithm [NFC]

The new code should hopefully be equivalent to the old code; it just uses a worklist to track instructions which need to visited rather than iterating over all instructions visited each time. This should be faster, but the primary benefit is that the purpose should be more clear and the diff of adding another instruction type (forthcoming) much more obvious.

Differential Revision: http://reviews.llvm.org/D11480

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243071 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agotest commit, only added one space
Lawrence Hu [Thu, 23 Jul 2015 23:55:28 +0000 (23:55 +0000)]
test commit, only added one space

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243070 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[NaryReassociate] remove redundant code
Jingyue Wu [Thu, 23 Jul 2015 23:13:37 +0000 (23:13 +0000)]
[NaryReassociate] remove redundant code

This check is already done by findClosestMatchingDominator.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243065 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMIR Serialization: Serialize the '.cfi_offset' CFI instruction.
Alex Lorenz [Thu, 23 Jul 2015 23:09:07 +0000 (23:09 +0000)]
MIR Serialization: Serialize the '.cfi_offset' CFI instruction.

Reviewers: Duncan P. N. Exon Smith

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243062 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoWebAssembly: test that valid -mcpu flags are accepted.
JF Bastien [Thu, 23 Jul 2015 23:00:04 +0000 (23:00 +0000)]
WebAssembly: test that valid -mcpu flags are accepted.

Summary: AArch64 has a similar test.

Subscribers: sunfish, aemerson, llvm-commits, jfb

Differential Revision: http://reviews.llvm.org/D11479

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243058 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agofix crash in machine trace metrics due to processing dbg_value instructions (PR24199)
Sanjay Patel [Thu, 23 Jul 2015 22:56:53 +0000 (22:56 +0000)]
fix crash in machine trace metrics due to processing dbg_value instructions (PR24199)

The test in PR24199 ( https://llvm.org/bugs/show_bug.cgi?id=24199 ) crashes because machine
trace metrics was not ignoring dbg_value instructions when calculating data dependencies.

The machine-combiner pass asks machine trace metrics to calculate an instruction trace,
does some reassociations, and calls MachineInstr::eraseFromParentAndMarkDBGValuesForRemoval()
along with MachineTraceMetrics::invalidate(). The dbg_value instructions have their operands
invalidated, but the instructions are not expected to be deleted.

On a subsequent loop iteration of the machine-combiner pass, machine trace metrics would be
called again and die while accessing the invalid debug instructions.

Differential Revision: http://reviews.llvm.org/D11423

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243057 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[RewriteStatepointsForGC] Rename PhiState to reflect that it's associated w/more...
Philip Reames [Thu, 23 Jul 2015 22:49:14 +0000 (22:49 +0000)]
[RewriteStatepointsForGC] Rename PhiState to reflect that it's associated w/more than just PHIs

Today, Select instructions also have associated PhiStates.  In the near future, so will ExtractElement and SuffleVector.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243056 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[RewriteStatepointsForGC] Use idomatic mechanisms for debug tracing [NFC]
Philip Reames [Thu, 23 Jul 2015 22:25:26 +0000 (22:25 +0000)]
[RewriteStatepointsForGC] Use idomatic mechanisms for debug tracing [NFC]

Deleting much of the code using trace-rewrite-statepoints and use idiomatic DEBUG statements instead.  This includes adding operator<< to a helper class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243054 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[ARM] Register (existing) ARMLoadStoreOpt pass with LLVM pass manager.
David Gross [Thu, 23 Jul 2015 22:12:46 +0000 (22:12 +0000)]
[ARM] Register (existing) ARMLoadStoreOpt pass with LLVM pass manager.

Summary: Among other things, this allows -print-after-all/-print-before-all to dump IR around this pass.

Subscribers: aemerson, llvm-commits, rengolin

Differential Revision: http://reviews.llvm.org/D11373

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243052 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMoving tests in to X86 directory.
Colin LeMahieu [Thu, 23 Jul 2015 21:55:26 +0000 (21:55 +0000)]
Moving tests in to X86 directory.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243049 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoTest commit.
David Gross [Thu, 23 Jul 2015 21:46:09 +0000 (21:46 +0000)]
Test commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243046 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[RewriteStatepointsForGC] Simplify code around meet of PhiStates [NFC]
Philip Reames [Thu, 23 Jul 2015 21:41:27 +0000 (21:41 +0000)]
[RewriteStatepointsForGC] Simplify code around meet of PhiStates [NFC]

We don't need to pass in the map from BDV to PhiStates; we can instead handle that externally and let the MeetPhiStates helper class just meet PhiStates.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243045 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUsing an input object file instead of trying to generate an object file.
Colin LeMahieu [Thu, 23 Jul 2015 21:40:19 +0000 (21:40 +0000)]
Using an input object file instead of trying to generate an object file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243044 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSpecifying a test triple.
Colin LeMahieu [Thu, 23 Jul 2015 21:24:52 +0000 (21:24 +0000)]
Specifying a test triple.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243042 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[llvm-objdump] Add -D and --disassemble-all flags that attempt disassembly on all...
Colin LeMahieu [Thu, 23 Jul 2015 20:58:49 +0000 (20:58 +0000)]
[llvm-objdump] Add -D and --disassemble-all flags that attempt disassembly on all sections instead of just text sections.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243041 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Scalarizer] Fix potential for stale data in Scattered across invocations
Matt Wala [Thu, 23 Jul 2015 20:53:46 +0000 (20:53 +0000)]
[Scalarizer] Fix potential for stale data in Scattered across invocations

Summary:
Scalarizer has two data structures that hold information about changes
to the function, Gathered and Scattered. These are cleared in finish()
at the end of runOnFunction() if finish() detects any changes to the
function.

However, finish() was checking for changes by only checking if
Gathered was non-empty. The function visitStore() only modifies
Scattered without touching Gathered. As a result, Scattered could have
ended up having stale data if Scalarizer only scalarized store
instructions. Since the data in Scattered is used during the execution
of the pass, this introduced dangling pointer errors.

The fix is to check whether both Scattered and Gathered are empty
before deciding what to do in finish(). This also fixes a problem
where the Function can be modified although the pass returns false.

Reviewers: rnk

Subscribers: rnk, srhines, llvm-commits

Differential Revision: http://reviews.llvm.org/D10459

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243040 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoX86: Use dyn_cast instead of isa+cast, NFC
Duncan P. N. Exon Smith [Thu, 23 Jul 2015 19:27:07 +0000 (19:27 +0000)]
X86: Use dyn_cast instead of isa+cast, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243034 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoThis patch eanble register coalescing to coalesce the following:
Weiming Zhao [Thu, 23 Jul 2015 19:24:53 +0000 (19:24 +0000)]
This patch eanble register coalescing to coalesce the following:

  %vreg2<def> = MOVi32imm 1; GPR32:%vreg2
  %W1<def> = COPY %vreg2; GPR32:%vreg2
into:
  %W1<def> = MOVi32imm 1
Patched by Lawrence Hu (lawrence@codeaurora.org)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243033 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[libFuzzer] dump long running units to disk
Kostya Serebryany [Thu, 23 Jul 2015 18:37:22 +0000 (18:37 +0000)]
[libFuzzer] dump long running units to disk

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243031 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoNFC. Explicitly specify attributes in BasicAA/cs-cs.ll test.
Igor Laevsky [Thu, 23 Jul 2015 14:31:18 +0000 (14:31 +0000)]
NFC. Explicitly specify attributes in BasicAA/cs-cs.ll test.
This will simplify verifying correctness for a changes which modify attributes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243016 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDelete dead code.
Rafael Espindola [Thu, 23 Jul 2015 13:42:16 +0000 (13:42 +0000)]
Delete dead code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243015 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse helper function. NFC.
Rafael Espindola [Thu, 23 Jul 2015 12:51:44 +0000 (12:51 +0000)]
Use helper function. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243012 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd a version of getSymbol with an explicit symbol table. Use it. NFC.
Rafael Espindola [Thu, 23 Jul 2015 12:49:40 +0000 (12:49 +0000)]
Add a version of getSymbol with an explicit symbol table. Use it. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243011 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Allow load folding into PUSH instructions
Michael Kuperstein [Thu, 23 Jul 2015 12:23:45 +0000 (12:23 +0000)]
[X86] Allow load folding into PUSH instructions

Adds pushes to the folding tables.
This also required a fix to the TD definition, since the memory forms of
the push instructions did not have the right mayLoad/mayStore flags.

Differential Revision: http://reviews.llvm.org/D11340

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243010 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[x86] change FP scalar builtin naming convention
Asaf Badouh [Thu, 23 Jul 2015 12:10:51 +0000 (12:10 +0000)]
[x86] change FP scalar builtin naming convention

Differential Revision: http://reviews.llvm.org/D11454

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243008 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[asan] Rename the ABI versioning symbol to '__asan_version_mismatch_check' instead...
Kuba Brecka [Thu, 23 Jul 2015 10:54:06 +0000 (10:54 +0000)]
[asan] Rename the ABI versioning symbol to '__asan_version_mismatch_check' instead of abusing '__asan_init'

We currently version `__asan_init` and when the ABI version doesn't match, the linker gives a `undefined reference to '__asan_init_v5'` message. From this, it might not be obvious that it's actually a version mismatch error. This patch makes the error message much clearer by changing the name of the undefined symbol to be `__asan_version_mismatch_check_xxx` (followed by the version string). We obviously don't want the initializer to be named like that, so it's a separate symbol that is used only for the purpose of version checking.

Reviewed at http://reviews.llvm.org/D11004

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243003 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Fix order of operands for ins and outs instructions when parsing intel syntax
Michael Kuperstein [Thu, 23 Jul 2015 10:23:48 +0000 (10:23 +0000)]
[X86] Fix order of operands for ins and outs instructions when parsing intel syntax

Patch by: marina.yatsina@intel.com
Differential Revision: http://reviews.llvm.org/D11337

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243001 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[GMR] Add a late run of GlobalsModRef to the main pass pipeline behind
Chandler Carruth [Thu, 23 Jul 2015 09:34:01 +0000 (09:34 +0000)]
[GMR] Add a late run of GlobalsModRef to the main pass pipeline behind
the general GMR-in-non-LTO flag.

Without this, we have the global information during the CGSCC pipeline
for GVN and such, but don't have it available during the late loop
optimizations such as the vectorizer. Moreover, after the CGSCC pipeline
has finished we have substantially more accurate and refined call graph
information, function annotations, etc, which will make GMR even more
powerful than it is early in the pipelien.

Note that we have to play silly games with preserving AliasAnalysis
(which is now trivially preserved) in order to let a module analysis
magically be preserved into the entire function pass pipeline.
Simultaneously we have to not make GMR an immutable pass in order to be
able to re-run it and collect fresh data on the final call graph.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@242999 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSupport printing relocations in files with no section table.
Rafael Espindola [Thu, 23 Jul 2015 09:11:05 +0000 (09:11 +0000)]
Support printing relocations in files with no section table.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@242998 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse typdef to simplify the code. NFC.
Rafael Espindola [Thu, 23 Jul 2015 08:48:14 +0000 (08:48 +0000)]
Use typdef to simplify the code. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@242995 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoX86: Fixed assertion failure in 32-bit mode
Elena Demikhovsky [Thu, 23 Jul 2015 08:25:23 +0000 (08:25 +0000)]
X86: Fixed assertion failure in 32-bit mode
The DAG Node "SCALAR_TO_VECTOR" may be created if the type of the scalar element is legal.
Added a check for the scalar type before creating this node.
Added a test that fails with assertion on the current version.

Differential Revision: http://reviews.llvm.org/D11413

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@242994 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAvoid using -Wl,-z,defs on Cygwin.
Yaron Keren [Thu, 23 Jul 2015 08:06:12 +0000 (08:06 +0000)]
Avoid using -Wl,-z,defs on Cygwin.

Prior to CMAKE 2.8.4 that was covered by the WIN32 conditional but
from 2.8.4 CMAKE no longer defined WIN32 when running under Cygwin
and it needs its own test.

Patch by Martell Malone!

http://reviews.llvm.org/D11347

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@242993 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert r242990: "AVX-512: Implemented encoding , DAG lowering and ..."
Chandler Carruth [Thu, 23 Jul 2015 08:03:44 +0000 (08:03 +0000)]
Revert r242990: "AVX-512: Implemented encoding , DAG lowering and ..."

This commit broke the build. Numerous build bots broken, and it was
blocking my progress so reverting.

It should be trivial to reproduce -- enable the BPF backend and it
should fail when running llvm-tblgen.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@242992 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[GMR] Switch the function info we store for every function to be a much
Chandler Carruth [Thu, 23 Jul 2015 07:50:52 +0000 (07:50 +0000)]
[GMR] Switch the function info we store for every function to be a much
more dense datastructure. We actually only have 3 bits of information
and an often-null pointer here. This fits very nicely into a
pointer-size value in the DenseMap from Function -> Info. Then we take
one more pointer hop to get to a secondary DenseMap from GlobalValue ->
ModRefInfo when we actually have precise info for particular globals.

This is more code than I would really like to do this packing, but it
ended up reasonably cleanly laid out. It should ensure we don't hit
scaling limitations with more widespread use of GMR.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@242991 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAVX-512: Implemented encoding , DAG lowering and intrinsics for Integer Truncate...
Igor Breger [Thu, 23 Jul 2015 07:39:21 +0000 (07:39 +0000)]
AVX-512: Implemented encoding , DAG lowering and intrinsics for Integer Truncate with/without saturation
Added tests for DAG lowering ,encoding and intrinsic

Differential Revision: http://reviews.llvm.org/D11218

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@242990 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[ScalarEvolution] Change addRequired to addRequiredTransitive on two passes where...
Craig Topper [Thu, 23 Jul 2015 07:33:48 +0000 (07:33 +0000)]
[ScalarEvolution] Change addRequired to addRequiredTransitive on two passes where ScalarEvolution stores long lived raw pointers to objects those passes own.

This prevents the pointers from dangling when those passes are freed.

http://reviews.llvm.org/D11236
Patch by Steve King.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@242989 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoReplace a 0 with nullptr. NFC
Craig Topper [Thu, 23 Jul 2015 07:17:53 +0000 (07:17 +0000)]
Replace a 0 with nullptr. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@242988 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAVX : Fix ISA disabling in case AVX512VL , some instructions should be disabled only...
Igor Breger [Thu, 23 Jul 2015 07:11:14 +0000 (07:11 +0000)]
AVX : Fix ISA disabling in case AVX512VL , some instructions should be disabled only if AVX512BW and AVX512VL present.
Tests added.

Differential Revision: http://reviews.llvm.org/D11414

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@242987 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove unnecessary in C++11 c_str() calls
Yaron Keren [Thu, 23 Jul 2015 05:49:29 +0000 (05:49 +0000)]
Remove unnecessary in C++11 c_str() calls

While theoratically required in pre-C++11 to avoid re-allocation upon call,
C++11 guarantees that c_str() returns a pointer to the internal array so
pre-calling c_str() is no longer required.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@242983 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[NVPTX] run LSR before straight-line optimizations
Jingyue Wu [Thu, 23 Jul 2015 04:59:07 +0000 (04:59 +0000)]
[NVPTX] run LSR before straight-line optimizations

Summary:
Straight-line optimizations can simplify the loop body and make LSR's
cost analysis more precise. This significantly improves several Eigen3
CUDA benchmarks.

With this change, EigenContractionKernel runs up to 40% faster
(https://bitbucket.org/eigen/eigen/src/753ceee5f206ff7dde9f6a41a5a420749fc9406f/unsupported/Eigen/CXX11/src/Tensor/TensorContractionCuda.h?at=default#cl-502).
EigenConvolutionKernel2D runs up to 10% faster
(https://bitbucket.org/eigen/eigen/src/753ceee5f206ff7dde9f6a41a5a420749fc9406f/unsupported/Eigen/CXX11/src/Tensor/TensorConvolution.h?at=default#cl-605).

I have some difficulties writing small tests that benefit from this
reordering due to a seemingly issue with LSR (being discussed at
http://lists.cs.uiuc.edu/pipermail/llvmdev/2015-July/088244.html).

See the review thread for the compilation time impact of GVN.

Reviewers: eliben, jholewinski

Subscribers: llvm-commits, jholewinski

Differential Revision: http://reviews.llvm.org/D11304

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@242982 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRefactor duplicated code and check for invalid symbol table size.
Rafael Espindola [Thu, 23 Jul 2015 03:24:22 +0000 (03:24 +0000)]
Refactor duplicated code and check for invalid symbol table size.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@242981 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[GMR] Further improve the FunctionInfo API inside of GlobalsModRef, NFC.
Chandler Carruth [Thu, 23 Jul 2015 00:12:32 +0000 (00:12 +0000)]
[GMR] Further improve the FunctionInfo API inside of GlobalsModRef, NFC.

This takes the operation of merging a callee's information into the
current information and embeds it into the FunctionInfo type itself.
This is much cleaner as now we don't need to expose iteration of the
globals, etc.

Also, switched all the uses of a raw integer two maintain the mod/ref
info during the SCC walk into just directly manipulating it in the
FunctionInfo object.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@242976 91177308-0d34-0410-b5e6-96231b3b80d8