oota-llvm.git
9 years agoAdd another InstCombine pass after LoopUnroll.
Wei Mi [Thu, 14 May 2015 22:02:54 +0000 (22:02 +0000)]
Add another InstCombine pass after LoopUnroll.

This is to cleanup some redundency generated by LoopUnroll pass. Such redundency may not be cleaned up by existing passes after LoopUnroll.

Differential Revision: http://reviews.llvm.org/D9777

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237395 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDon't rely on implicit pointerness of 'auto'.
Davide Italiano [Thu, 14 May 2015 21:52:12 +0000 (21:52 +0000)]
Don't rely on implicit pointerness of 'auto'.

This ends up being a copy. Pointy hat to me.
Reported by: dexonsmith, dblaikie

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237394 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix memory leak introduced in r237314.
Alex Lorenz [Thu, 14 May 2015 20:46:12 +0000 (20:46 +0000)]
Fix memory leak introduced in r237314.

The commit r237314 that implements YAML block parsing
introduced a leak that was caught by the ASAN linux buildbot.
YAML Parser stores its tokens in an ilist, and allocates
tokens using a BumpPtrAllocator, but doesn't call the
destructor for the allocated tokens. R237314 added an
std::string field to a Token which leaked as the Token's
destructor wasn't called. This commit fixes this leak
by calling the Token's destructor when a Token is being
removed from an ilist of tokens.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237389 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Hexagon] Generate hardware loop for a vectorized loop
Brendon Cahoon [Thu, 14 May 2015 20:36:19 +0000 (20:36 +0000)]
[Hexagon] Generate hardware loop for a vectorized loop

The induction variable in the vectorized loop wasn't
recognized properly, so a hardware loop wasn't generated.

Differential Revision: http://reviews.llvm.org/D9722

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237388 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoTurn effective assert(0) into llvm_unreachable
Matthias Braun [Thu, 14 May 2015 18:33:29 +0000 (18:33 +0000)]
Turn effective assert(0) into llvm_unreachable

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237379 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUpdate obsolete comments, fix typo, delete trailing space.
Douglas Katzman [Thu, 14 May 2015 18:07:04 +0000 (18:07 +0000)]
Update obsolete comments, fix typo, delete trailing space.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237378 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[ConstantFolding] Fix wrong folding of intrinsic 'convert.from.fp16'.
Andrea Di Biagio [Thu, 14 May 2015 18:01:48 +0000 (18:01 +0000)]
[ConstantFolding] Fix wrong folding of intrinsic 'convert.from.fp16'.

Function 'ConstantFoldScalarCall' (in ConstantFolding.cpp) works under the
wrong assumption that a call to 'convert.from.fp16' returns a value of
type 'float'.
However, intrinsic 'convert.from.fp16' can be overloaded; for example, we
can call 'convert.from.fp16.f64' to convert from half to double; etc.

Before this patch, the following example would have triggered an assertion
failure in opt (with -constprop):

```
define double @foo() {
entry:
  %0 = call double @llvm.convert.from.fp16.f64(i16 0)
  ret double %0
}
```

This patch fixes the problem in ConstantFolding.cpp. When folding a call to
convert.from.fp16, we perform a different kind of conversion based on the call
return type.

Added test 'Transform/ConstProp/convert-from-fp16.ll'.

Differential Revision: http://reviews.llvm.org/D9771

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237377 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoTargetSchedule: factor out common code; NFC
Matthias Braun [Thu, 14 May 2015 18:01:13 +0000 (18:01 +0000)]
TargetSchedule: factor out common code; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237376 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove MCInstrItineraries includes in parts that don't use them anymore
Matthias Braun [Thu, 14 May 2015 18:01:11 +0000 (18:01 +0000)]
Remove MCInstrItineraries includes in parts that don't use them anymore

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237375 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Hexagon] Remove dead constant assignment in hardware loop pass
Brendon Cahoon [Thu, 14 May 2015 17:31:40 +0000 (17:31 +0000)]
[Hexagon] Remove dead constant assignment in hardware loop pass

After converting a loop to a hardware loop, the pass should remove
any unnecessary instructions from the old compare-and-branch
code. This patch removes a dead constant assignment that was
used in the compare instruction.

Differential Revision: http://reviews.llvm.org/D9720

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237373 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoEnable solid lzma compression for cpack, decreases setup size by ~30%
Ismail Donmez [Thu, 14 May 2015 17:07:41 +0000 (17:07 +0000)]
Enable solid lzma compression for cpack, decreases setup size by ~30%

Reviewed by Hans Wennborg

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237372 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoReflow long lines of some LLVMBuild files
Douglas Katzman [Thu, 14 May 2015 15:38:27 +0000 (15:38 +0000)]
Reflow long lines of some LLVMBuild files

Differential Revision: http://reviews.llvm.org/D9752

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237367 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips] [IAS] Enforce .set nomacro.
Toma Tabacu [Thu, 14 May 2015 14:51:32 +0000 (14:51 +0000)]
[mips] [IAS] Enforce .set nomacro.

Summary: When used, ".set nomacro" causes warning messages to be reported when we expand pseudo-instructions to multiple machine instructions.

Reviewers: dsanders

Reviewed By: dsanders

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9564

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237366 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Hexagon] Check for underflow/wrap in hardware loop pass
Brendon Cahoon [Thu, 14 May 2015 14:15:08 +0000 (14:15 +0000)]
[Hexagon] Check for underflow/wrap in hardware loop pass

If the loop trip count may underflow or wrap, the compiler should
not generate a hardware loop since the trip count will be
incorrect.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237365 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips] [IAS] Emit .set macro/nomacro.
Toma Tabacu [Thu, 14 May 2015 13:42:10 +0000 (13:42 +0000)]
[mips] [IAS] Emit .set macro/nomacro.

Reviewers: dsanders

Reviewed By: dsanders

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9563

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237363 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips] Do not place users of $ra in the delay slot of call instructions.
Vasileios Kalintiris [Thu, 14 May 2015 13:17:56 +0000 (13:17 +0000)]
[mips] Do not place users of $ra in the delay slot of call instructions.

Summary:
When we are trying to fill the delay slot of a call instruction, we must avoid
filler instructions that use the $ra register. This fixes the test
MultiSource/Applications/JM/lencod when we enable the forward delay slot filler.

Reviewers: dsanders

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9670

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237362 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRe-apply r237247 - [AArch64] Codegen VMAX/VMIN for safe math cases
Artyom Skrobov [Thu, 14 May 2015 12:59:46 +0000 (12:59 +0000)]
Re-apply r237247 - [AArch64] Codegen VMAX/VMIN for safe math cases

No longer breaks SPEC2000/2006

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237361 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAttempt to fix MSVC bots
Adam Nemet [Thu, 14 May 2015 12:33:32 +0000 (12:33 +0000)]
Attempt to fix MSVC bots

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237359 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoNew Loop Distribution pass
Adam Nemet [Thu, 14 May 2015 12:05:18 +0000 (12:05 +0000)]
New Loop Distribution pass

Summary:
This implements the initial version as was proposed earlier this year
(http://lists.cs.uiuc.edu/pipermail/llvmdev/2015-January/080462.html).
Since then Loop Access Analysis was split out from the Loop Vectorizer
and was made into a separate analysis pass.  Loop Distribution becomes
the second user of this analysis.

The pass is off by default and can be enabled
with -enable-loop-distribution.  There is currently no notion of
profitability; if there is a loop with dependence cycles, the pass will
try to split them off from other memory operations into a separate loop.

I decided to remove the control-dependence calculation from this first
version.  This and the issues with the PDT are actively discussed so it
probably makes sense to treat it separately.  Right now I just mark all
terminator instruction required which keeps identical CFGs for each
distributed loop.  This seems to be working pretty well for 456.hmmer
where even though there is an empty if-then block in the distributed
loop initially, it gets completely removed.

The pass keeps DominatorTree and LoopInfo updated.  I've tested this
with -loop-distribute-verify with the testsuite where we distribute ~90
loops.  SimplifyLoop is violated in some cases and I have a FIXME
covering this.

Reviewers: hfinkel, nadav, aschwaighofer

Reviewed By: aschwaighofer

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D8831

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237358 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFixed some typos and broken links in source level debugging docs.
Michael Kuperstein [Thu, 14 May 2015 10:58:59 +0000 (10:58 +0000)]
Fixed some typos and broken links in source level debugging docs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237357 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips] [IAS] Warn when LA is used with a 64-bit symbol.
Toma Tabacu [Thu, 14 May 2015 10:53:40 +0000 (10:53 +0000)]
[mips] [IAS] Warn when LA is used with a 64-bit symbol.

Reviewers: dsanders

Reviewed By: dsanders

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9295

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237356 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips] [IAS] Give expandLoadAddressSym() more specific arguments. NFC.
Toma Tabacu [Thu, 14 May 2015 10:02:58 +0000 (10:02 +0000)]
[mips] [IAS] Give expandLoadAddressSym() more specific arguments. NFC.

Summary:
If we only pass the necessary operands, we don't have to determine the position of the symbol operand when entering expandLoadAddressSym().
This simplifies the expandLoadAddressSym() code.

Reviewers: dsanders

Reviewed By: dsanders

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9291

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237355 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[AArch64] Slight naming changes and comments for AArch64NamedImmMapper
Vladimir Sukharev [Thu, 14 May 2015 09:50:14 +0000 (09:50 +0000)]
[AArch64] Slight naming changes and comments for AArch64NamedImmMapper

Reviewers: echristo

Subscribers: llvm-commits

Follow-up to: http://reviews.llvm.org/D8496#158595

Relates to: http://reviews.llvm.org/rL235089

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237354 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAVX-512: Added i1 type handling for calling conventions.
Elena Demikhovsky [Thu, 14 May 2015 09:04:45 +0000 (09:04 +0000)]
AVX-512: Added i1 type handling for calling conventions.
i1 type is a legal type on AVX-512 and can be passed as parameter or return value.
i1 is promoted to i8 on return and to i32 for call arguments (i8 is also promoted to i32 here).
The result code is similar to the previous X86 targets, where i1 is allways promoted to i8.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237350 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoTableGen: Avoid undefined behaviour by doing this shift in int64
Justin Bogner [Thu, 14 May 2015 06:47:02 +0000 (06:47 +0000)]
TableGen: Avoid undefined behaviour by doing this shift in int64

Found by ubsan. This was taking a bool and left shifting by 32 - the
result is 64 bit, so we should really do the math in a type it fits
in.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237345 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[TableGen] Remove an unnecessary outer 'if' around 3 separate inner ifs. No functiona...
Craig Topper [Thu, 14 May 2015 05:54:02 +0000 (05:54 +0000)]
[TableGen] Remove an unnecessary outer 'if' around 3 separate inner ifs. No functional change intended.

The outer if had 3 separate conditions ORed together and then the inner ifs detected which of the three conditions it was by using only a portion of the specific condition. Just put the whole condition in each inner if and remove the outer if.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237343 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[TableGen] Simplify some code. NFC
Craig Topper [Thu, 14 May 2015 05:53:59 +0000 (05:53 +0000)]
[TableGen] Simplify some code. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237342 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[TableGen] Remove ListInit::size() in favor of getSize() which does the same thing...
Craig Topper [Thu, 14 May 2015 05:53:56 +0000 (05:53 +0000)]
[TableGen] Remove ListInit::size() in favor of getSize() which does the same thing and is already used in most places. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237341 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[TableGen] Replace some calls to ListInit::getSize() with ListInit::empty() if it...
Craig Topper [Thu, 14 May 2015 05:53:53 +0000 (05:53 +0000)]
[TableGen] Replace some calls to ListInit::getSize() with ListInit::empty() if it was just comparing to 0. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237340 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agocmake: Use -fno-sanitize-recover=all - the old spelling is deprecated
Justin Bogner [Thu, 14 May 2015 04:52:57 +0000 (04:52 +0000)]
cmake: Use -fno-sanitize-recover=all - the old spelling is deprecated

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237334 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDon't omit the constant when computing a cross-section relative relocation.
Andy Ayers [Thu, 14 May 2015 01:10:41 +0000 (01:10 +0000)]
Don't omit the constant when computing a cross-section relative relocation.

Differential Revision: http://reviews.llvm.org/D9692

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237327 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[CodeGen] Use standard -not gnueabi- naming for f16 libcalls on Darwin.
Ahmed Bougacha [Thu, 14 May 2015 01:00:51 +0000 (01:00 +0000)]
[CodeGen] Use standard -not gnueabi- naming for f16 libcalls on Darwin.

Other targets probably should as well.  Since r237161, compiler-rt has
both, but I don't see why anything other than gnueabi would use a
gnueabi naming scheme.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237324 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert r237046. See the testcase on the thread where r237046 was committed.
Nick Lewycky [Wed, 13 May 2015 23:41:47 +0000 (23:41 +0000)]
Revert r237046. See the testcase on the thread where r237046 was committed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237317 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoYAML: Implement block scalar parsing.
Alex Lorenz [Wed, 13 May 2015 23:10:51 +0000 (23:10 +0000)]
YAML: Implement block scalar parsing.

This commit implements the parsing of YAML block scalars.
Some code existed for it before, but it couldn't parse block
scalars.

This commit adds a new yaml node type to represent the block
scalar values.

This commit also deletes the 'spec-09-27' and 'spec-09-28' tests
as they are identical to the test file 'spec-09-26'.

This commit introduces 3 new utility functions to the YAML scanner
class: `skip_s_space`, `advanceWhile` and `consumeLineBreakIfPresent`.

Reviewers: Duncan P. N. Exon Smith

Differential Revision: http://reviews.llvm.org/D9503

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237314 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[opaque pointer type] Use the value type of the GlobalVariable rather than accessing...
David Blaikie [Wed, 13 May 2015 22:55:01 +0000 (22:55 +0000)]
[opaque pointer type] Use the value type of the GlobalVariable rather than accessing it through the pointee's type

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237312 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[opaque pointer type] Use GlobalVariable::getValueType rather than accessing it throu...
David Blaikie [Wed, 13 May 2015 22:54:54 +0000 (22:54 +0000)]
[opaque pointer type] Use GlobalVariable::getValueType rather than accessing it through the GV's pointee type

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237311 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Fix PR23271 - RIP-relative decoding bug in disassembler.
Douglas Katzman [Wed, 13 May 2015 22:44:52 +0000 (22:44 +0000)]
[X86] Fix PR23271 - RIP-relative decoding bug in disassembler.

Differential Revision: http://reviews.llvm.org/D9110

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237310 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoConstruct ArrayRef<const T*> from vector<T>
Pete Cooper [Wed, 13 May 2015 22:43:09 +0000 (22:43 +0000)]
Construct ArrayRef<const T*> from vector<T>

ArrayRef already has a SFINAE constructor which can construct ArrayRef<const T*> from ArrayRef<T*>.

This adds methods to do the same directly from SmallVector and std::vector.  This avoids an intermediate step through the use of makeArrayRef.

Also update the users of this in LICM and SROA to remove the now unnecessary makeArrayRef call.

Reviewed by David Blaikie.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237309 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[lib/Fuzzer] update docs about test corpuses in git
Kostya Serebryany [Wed, 13 May 2015 22:42:28 +0000 (22:42 +0000)]
[lib/Fuzzer] update docs about test corpuses in git

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237308 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoInstrProf: Fix display of large numbers in llvm-cov
Justin Bogner [Wed, 13 May 2015 22:41:48 +0000 (22:41 +0000)]
InstrProf: Fix display of large numbers in llvm-cov

llvm-cov was truncating numbers that were larger than a particular
fixed width, which is as confusing as it is useless. Instead, we use
engineering notation with SI prefix for magnitude.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237307 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd llvm::all_of which wraps std::all_of.
Pete Cooper [Wed, 13 May 2015 22:19:13 +0000 (22:19 +0000)]
Add llvm::all_of which wraps std::all_of.

This version doesn't need begin/end but can instead just take a type which has begin/end methods.

Use this to replace an eligible foreach loop in LoopInfo found by David Blaikie in r237224.

Reviewed by David Blaikie.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237301 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix typo on Instruction::insert{After,Before}(). NFC
Jonathan Roelofs [Wed, 13 May 2015 22:14:43 +0000 (22:14 +0000)]
Fix typo on Instruction::insert{After,Before}(). NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237300 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoInstrProf: Treat functions with a coverage map but no profile as unreached
Justin Bogner [Wed, 13 May 2015 22:03:04 +0000 (22:03 +0000)]
InstrProf: Treat functions with a coverage map but no profile as unreached

If we have a coverage mapping but no profile data for a function,
calling it mismatched is misleading. This can just as easily be
unreachable code that was stripped from the binary. Instead, treat
these the same as functions where we have an explicit "zero" coverage
map by setting the count to zero for each mapped region.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237298 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMC: clang-format MCSymbol.h, NFC
Duncan P. N. Exon Smith [Wed, 13 May 2015 21:41:14 +0000 (21:41 +0000)]
MC: clang-format MCSymbol.h, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237297 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoARM: remove possible vestiges of the legacy JIT???
Tim Northover [Wed, 13 May 2015 20:28:41 +0000 (20:28 +0000)]
ARM: remove possible vestiges of the legacy JIT???

There's no need to manually pass modifier strings around to tell an operand how
to print now, that information is encoded in the operand itself since the MC
layer came along.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237295 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoARM: remove custom jump table UID
Tim Northover [Wed, 13 May 2015 20:28:38 +0000 (20:28 +0000)]
ARM: remove custom jump table UID

We were creating and propagating two separate indices for each jump table (from
back in the mists of time). However, the generic index used by other backends
is sufficient to emit a unique symbol so this was unneeded.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237294 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoARM: refactor optimizeThumb2JumpTables.
Tim Northover [Wed, 13 May 2015 20:28:32 +0000 (20:28 +0000)]
ARM: refactor optimizeThumb2JumpTables.

The previous logic mixed 2 separate questions:
  + Can we form a TBB/TBH instruction?
  + Can we remove the jump-table calculation before it?

It then performed a bunch of random tests on the instructions earlier in the
basic block, which were probably sufficient to answer 2 but only because of the
very limited ways in which a t2BR_JT can actually be created.

For example there's no reason to expect the LeaInst to define the same base
register as the following indexing calulation. In practice this means we might
have missed opportunities to form TBB/TBH, in theory you could end up
misidentifying a sequence and removing the wrong LEA:

     %R1 = t2LEApcrelJT ...
     %R2 = t2LEApcrelJT ...
     <... using and killing %R2 ...>
     %R2 = t2ADDr %R1, $Ridx

Before we would have looked for an LEA defining %R2 and found the wrong one. We
just got lucky that jump table setup was (almost?) always confined to a single
basic block and there was only one jump table per block.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237293 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Statepoints][Docs] Fix a couple of out of date examples.
Sanjoy Das [Wed, 13 May 2015 20:20:10 +0000 (20:20 +0000)]
[Statepoints][Docs] Fix a couple of out of date examples.

Things I had missed in r237285.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237290 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Statepoints][Docs] Fix typo: change a period to a comma.
Sanjoy Das [Wed, 13 May 2015 20:19:51 +0000 (20:19 +0000)]
[Statepoints][Docs] Fix typo: change a period to a comma.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237289 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Safepoints][Verifier] Fix a tautological Assert.
Sanjoy Das [Wed, 13 May 2015 20:11:59 +0000 (20:11 +0000)]
[Safepoints][Verifier] Fix a tautological Assert.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237287 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[PlaceSafepoints] New attributes for patchable statepoints.
Sanjoy Das [Wed, 13 May 2015 20:11:31 +0000 (20:11 +0000)]
[PlaceSafepoints] New attributes for patchable statepoints.

Summary:
This patch teaches the PlaceSafepoints pass about two `CallSite`
function attributes:

 * "statepoint-id": if the string value of this attribute can be parsed
   as an integer, then it is propagated to the ID parameter of the
   statepoint created.

 * "statepoint-num-patch-bytes": if the string value of this attribute
   can be parsed as an integer, then it is propagated to the `num patch
   bytes` parameter of the statepoint created.

This change intentionally does not assert on a malformed value for these
attributes, given that they're not "official" attributes.

Reviewers: reames, pgavlin

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9735

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237286 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[PlaceSafepoints] Update docs for r237214.
Sanjoy Das [Wed, 13 May 2015 20:11:24 +0000 (20:11 +0000)]
[PlaceSafepoints] Update docs for r237214.

Show the two new ID and NumPatchBytes fields in the PlaceSafepoint
examples in Statepoints.rst to avoid confusion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237285 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[LoopIdiomRecognize] Use auto + range-based loop. NFC intended.
Davide Italiano [Wed, 13 May 2015 19:51:21 +0000 (19:51 +0000)]
[LoopIdiomRecognize] Use auto + range-based loop. NFC intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237284 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[llvm-bcanalyzer] Add -show-binary-blobs option.
Jordan Rose [Wed, 13 May 2015 18:51:49 +0000 (18:51 +0000)]
[llvm-bcanalyzer] Add -show-binary-blobs option.

-dump mode normally omits blob data that contains unprintable characters.
When -show-binary-blobs is passed, it unilaterally escapes all blobs,
allowing those with binary data to be displayed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237276 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMC: Modernize MCOperand API naming. NFC.
Jim Grosbach [Wed, 13 May 2015 18:37:00 +0000 (18:37 +0000)]
MC: Modernize MCOperand API naming. NFC.

MCOperand::Create*() methods renamed to MCOperand::create*().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237275 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[opaque pointer type] Constant Folding: Use GEPOperator to access the pointee source...
David Blaikie [Wed, 13 May 2015 18:35:29 +0000 (18:35 +0000)]
[opaque pointer type] Constant Folding: Use GEPOperator to access the pointee source type rather than going through the first operand's pointer type

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237274 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[opaque pointer type] Pass the explicit function type down to the instruction constru...
David Blaikie [Wed, 13 May 2015 18:35:26 +0000 (18:35 +0000)]
[opaque pointer type] Pass the explicit function type down to the instruction constructor when parsing invoke instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237273 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[lib/Fuzzer] enable -use_counters=1 by default
Kostya Serebryany [Wed, 13 May 2015 18:31:46 +0000 (18:31 +0000)]
[lib/Fuzzer] enable -use_counters=1 by default

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237272 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[NaryReassociate] avoid running forever
Jingyue Wu [Wed, 13 May 2015 18:12:24 +0000 (18:12 +0000)]
[NaryReassociate] avoid running forever

Avoid running forever by checking we are not reassociating an expression into
the same form.

Tested with @avoid_infinite_loops in nary-add.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237269 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Hexagon] Generate loop1 instruction for nested loops
Brendon Cahoon [Wed, 13 May 2015 17:56:03 +0000 (17:56 +0000)]
[Hexagon] Generate loop1 instruction for nested loops

loop1 is for the outer loop and loop0 is for the inner loop.

Differential Revision: http://reviews.llvm.org/D9680

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237266 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd function entry counts from sample profiles.
Diego Novillo [Wed, 13 May 2015 17:04:29 +0000 (17:04 +0000)]
Add function entry counts from sample profiles.

This patch uses the new function profile metadata "function_entry_count"
to annotate entry counts from sample profiles.

In a sampling profile, the total samples collected at the function entry
are an approximation for the number of times that function was invoked.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237265 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips] [IAS] Preemptively fix warning introduced by r237255. NFC.
Toma Tabacu [Wed, 13 May 2015 16:02:41 +0000 (16:02 +0000)]
[mips] [IAS] Preemptively fix warning introduced by r237255. NFC.

Some compilers warn about using the ternary operator with an unsigned variable
and enum.
I haven't seen this trigger in the llvm.org buildbots yet, but it probably will
at some point.

Reported by Daniel Sanders.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237262 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUpdate ELFObjectWriter::reset() following r236255.
Yaron Keren [Wed, 13 May 2015 15:17:19 +0000 (15:17 +0000)]
Update ELFObjectWriter::reset() following r236255.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237261 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd function entry count metadata.
Diego Novillo [Wed, 13 May 2015 15:13:45 +0000 (15:13 +0000)]
Add function entry count metadata.

Summary:
This adds three Function methods to handle function entry counts:
setEntryCount() and getEntryCount().

Entry counts are stored under the MD_prof metadata node with the name
"function_entry_count". They are unsigned 64 bit values set by profilers
(instrumentation and sample profiler changes coming up).

Added documentation for new profile metadata and tests.

Reviewers: dexonsmith, bogner

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9628

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237260 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoTest commit: Remove unnecessary spaces.
Teresa Johnson [Wed, 13 May 2015 15:04:14 +0000 (15:04 +0000)]
Test commit: Remove unnecessary spaces.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237259 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Hexagon] Generate hardware loop when loop has a critical edge
Brendon Cahoon [Wed, 13 May 2015 14:54:24 +0000 (14:54 +0000)]
[Hexagon] Generate hardware loop when loop has a critical edge

The hardware loop pass should try to generate a hardware loop
instruction when the original loop has a critical edge.

Differential Revision: http://reviews.llvm.org/D9678

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237258 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips][microMIPSr6] Implement CLO and CLZ instructions
Jozef Kolek [Wed, 13 May 2015 14:18:11 +0000 (14:18 +0000)]
[mips][microMIPSr6] Implement CLO and CLZ instructions

This patch implements CLO and CLZ instructions using mapping.

Differential Revision: http://reviews.llvm.org/D8553

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237257 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert r237247 - [AArch64] Codegen VMAX/VMIN.. as it is causing failures in SPEC2000...
Silviu Baranga [Wed, 13 May 2015 14:03:18 +0000 (14:03 +0000)]
Revert r237247 - [AArch64] Codegen VMAX/VMIN.. as it is causing failures in SPEC2000/2006

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237256 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips] [IAS] Unify common functionality of LA and LI.
Toma Tabacu [Wed, 13 May 2015 13:56:16 +0000 (13:56 +0000)]
[mips] [IAS] Unify common functionality of LA and LI.

Summary: A side-effect of this is that LA gains proper handling of unsigned and positive signed 16-bit immediates and more accurate error messages.

Reviewers: dsanders

Reviewed By: dsanders

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9290

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237255 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[AArch64] Codegen VMAX/VMIN for safe math cases
Artyom Skrobov [Wed, 13 May 2015 12:01:09 +0000 (12:01 +0000)]
[AArch64] Codegen VMAX/VMIN for safe math cases

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237247 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoReverting r237234, "Use std::bitset for SubtargetFeatures"
Michael Kuperstein [Wed, 13 May 2015 10:28:46 +0000 (10:28 +0000)]
Reverting r237234, "Use std::bitset for SubtargetFeatures"

The buildbots are still not satisfied.
MIPS and ARM are failing (even though at least MIPS was expected to pass).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237245 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips] [IAS] Merge the micromips-expressions.s test into expr1.s. NFC.
Toma Tabacu [Wed, 13 May 2015 09:53:53 +0000 (09:53 +0000)]
[mips] [IAS] Merge the micromips-expressions.s test into expr1.s. NFC.

Summary: Also did some minor reformatting in the resulting test.

Reviewers: dsanders

Reviewed By: dsanders

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9702

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237242 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[DebugInfo] Debug locations for constant SD nodes
Sergey Dmitrouk [Wed, 13 May 2015 08:58:03 +0000 (08:58 +0000)]
[DebugInfo] Debug locations for constant SD nodes

Several updates for [DebugInfo] Add debug locations to constant SD nodes (r235989).
Includes:

 *  re-enabling the change (disabled recently);
 *  missing change for FP constants;
 *  resetting debug location of constant node if it's used more than at one place
    to prevent emission of wrong locations in case of coalesced constants;
 *  a couple of additional tests.

Now all look ups in CSEMap are wrapped by additional method.

Comment in D9084 suggests that debug locations aren't useful for "target constants",
so there might be one more change related to this API (namely, dropping debug
locations for getTarget*Constant methods).

Differential Revision: http://reviews.llvm.org/D9604

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237237 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse std::bitset for SubtargetFeatures
Michael Kuperstein [Wed, 13 May 2015 08:27:08 +0000 (08:27 +0000)]
Use std::bitset for SubtargetFeatures

Previously, subtarget features were a bitfield with the underlying type being uint64_t.
Since several targets (X86 and ARM, in particular) have hit or were very close to hitting this bound, switching the features to use a bitset.
No functional change.

The first two times this was committed (r229831, r233055), it caused several buildbot failures.
At least some of the ARM and MIPS ones were due to gcc/binutils issues, and should now be fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237234 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAVX-512: fixed a bug in encoding of VPSRAQ instrcution,
Elena Demikhovsky [Wed, 13 May 2015 07:35:05 +0000 (07:35 +0000)]
AVX-512: fixed a bug in encoding of VPSRAQ instrcution,
added a bunch of encoding tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237232 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse ArrayRef::slice instead of manually constructing an ArrayRef from ArrayRef iterat...
Craig Topper [Wed, 13 May 2015 06:57:51 +0000 (06:57 +0000)]
Use ArrayRef::slice instead of manually constructing an ArrayRef from ArrayRef iterators. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237231 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoConstify arguments to methods in LICM. NFC
Pete Cooper [Wed, 13 May 2015 01:12:18 +0000 (01:12 +0000)]
Constify arguments to methods in LICM.  NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237227 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoChange LoadAndStorePromoter to take ArrayRef instead of SmallVectorImpl&.
Pete Cooper [Wed, 13 May 2015 01:12:16 +0000 (01:12 +0000)]
Change LoadAndStorePromoter to take ArrayRef instead of SmallVectorImpl&.

The array passed to LoadAndStorePromoter's constructor was a constant reference to a SmallVectorImpl, which is just the same as passing an ArrayRef.

Also, the data in the array can be 'const Instruction*' instead of 'Instruction*'.  Its not possible to convert a SmallVectorImpl<T*> to SmallVectorImpl<const T*>, but ArrayRef does provide such a method.

Currently this added calls to makeArrayRef which should be a nop, but i'm going to kick off a discussion about improving ArrayRef to not need these.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237226 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoConstify arguments in AliasSetTracker methods. NFC
Pete Cooper [Wed, 13 May 2015 01:12:12 +0000 (01:12 +0000)]
Constify arguments in AliasSetTracker methods.  NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237225 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoChange a loop in LoopInfo to foreach. NFC
Pete Cooper [Wed, 13 May 2015 01:12:09 +0000 (01:12 +0000)]
Change a loop in LoopInfo to foreach.  NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237224 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoConstify arguments to methods in LoopInfo. NFC
Pete Cooper [Wed, 13 May 2015 01:12:06 +0000 (01:12 +0000)]
Constify arguments to methods in LoopInfo.  NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237223 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[PlaceSafepoints] Reduce dominator tree recalculation
Philip Reames [Wed, 13 May 2015 00:32:23 +0000 (00:32 +0000)]
[PlaceSafepoints] Reduce dominator tree recalculation

Reduce recalculation of the dominator tree by identifying all sites that will need a safepoint poll before doing any of the insertion. This allows us to invalidate the dominator info once, rather than once per safepoint poll inserted.

While I'm at it, update findLocationForEntrySafepoint to properly update the dom tree now that the interface has been made easy. When first written, it wasn't per comment in the code.

Differential Revision: http://reviews.llvm.org/D9727

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237220 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[SLSR] handles non-canonicalized Mul candidates
Jingyue Wu [Wed, 13 May 2015 00:03:17 +0000 (00:03 +0000)]
[SLSR] handles non-canonicalized Mul candidates

such as (2 + B) * S.

Tested by @non_canonicalized in slsr-mul.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237216 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Statepoints] Support for "patchable" statepoints.
Sanjoy Das [Tue, 12 May 2015 23:52:24 +0000 (23:52 +0000)]
[Statepoints] Support for "patchable" statepoints.

Summary:
This change adds two new parameters to the statepoint intrinsic, `i64 id`
and `i32 num_patch_bytes`.  `id` gets propagated to the ID field
in the generated StackMap section.  If the `num_patch_bytes` is
non-zero then the statepoint is lowered to `num_patch_bytes` bytes of
nops instead of a call (the spill and reload code remains unchanged).
A non-zero `num_patch_bytes` is useful in situations where a language
runtime requires complete control over how a call is lowered.

This change brings statepoints one step closer to patchpoints.  With
some additional work (that is not part of this patch) it should be
possible to get rid of `TargetOpcode::STATEPOINT` altogether.

PlaceSafepoints generates `statepoint` wrappers with `id` set to
`0xABCDEF00` (the old default value for the ID reported in the stackmap)
and `num_patch_bytes` set to `0`.  This can be made more sophisticated
later.

Reviewers: reames, pgavlin, swaroop.sridhar, AndyAyers

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9546

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237214 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[PlaceSafepoints] Followup to commit L237172
Philip Reames [Tue, 12 May 2015 23:39:23 +0000 (23:39 +0000)]
[PlaceSafepoints] Followup to commit L237172

Responding to review feedback from http://reviews.llvm.org/D9585

1) Remove a variable shadow by converting the outer loop to a range for loop.  We never really used the 'i' variable which was being shadowed.
2) Reduce DominatorTree recalculations by passing the DT to SplitEdge.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237212 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCodeGen: ignore DEBUG_VALUE nodes in KILL tagging
Saleem Abdulrasool [Tue, 12 May 2015 23:36:18 +0000 (23:36 +0000)]
CodeGen: ignore DEBUG_VALUE nodes in KILL tagging

DEBUG_VALUE nodes do not take part in code generation.  Ignore them when
performing KILL updates.  Addresses PR23486.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237211 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert r237175: [X86] Always return the sret parameter in eax/rax ...
Chandler Carruth [Tue, 12 May 2015 23:34:27 +0000 (23:34 +0000)]
Revert r237175: [X86] Always return the sret parameter in eax/rax ...

This commit broke an x86 test and the bots have been broken for well
over an hour now so I'm just reverting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237210 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Unrolling] Refactor the start and step offsets to simplify overflow
Chandler Carruth [Tue, 12 May 2015 23:32:56 +0000 (23:32 +0000)]
[Unrolling] Refactor the start and step offsets to simplify overflow
checking and make the cache faster and smaller.

I had thought that using an APInt here would be useful, but I think
I was just wrong. Notably, we don't have to do any fancy overflow
checking, we can just bound the values as quite small and do the math in
a higher precision integer. I've switched to a signed integer so that
UBSan will even point out if we ever have integer overflow. I've added
various asserts to try to catch things as well and hoisted the overflow
checks so that we just leave the too-large offsets out of the SCEV-GEP
cache. This makes the value in the cache quite a bit smaller which is
probably worthwhile.

No functionality changed here (for trip counts under 1 billion).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237209 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[lib/Fuzzer] A simple script to synchronise a fuzz test corpus with an external git...
Kostya Serebryany [Tue, 12 May 2015 23:19:12 +0000 (23:19 +0000)]
[lib/Fuzzer] A simple script to synchronise a fuzz test corpus with an external git repository.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237208 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAllow the CMake option 'LLVM_ABI_BREAKING_CHECKS' to be undefined.
Eric Fiselier [Tue, 12 May 2015 22:49:18 +0000 (22:49 +0000)]
Allow the CMake option 'LLVM_ABI_BREAKING_CHECKS' to be undefined.

When building libc++abi in a standalone configuration the CMake option
'LLVM_ABI_BREAKING_CHECKS` will not be defined.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237204 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCVP: Improve handling of Selects used as incoming PHI values
Bjorn Steinbrink [Tue, 12 May 2015 22:31:47 +0000 (22:31 +0000)]
CVP: Improve handling of Selects used as incoming PHI values

Summary:
If the branch that leads to the PHI node and the Select instruction
depend on correlated conditions, we might be able to directly use the
corresponding value from the Select instruction as the incoming value
for the PHI node, allowing later removal of the select instruction.

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9051

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237201 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[RewriteStatepointsForGC] Extend base pointer to handle more cases w/vectors
Philip Reames [Tue, 12 May 2015 22:19:52 +0000 (22:19 +0000)]
[RewriteStatepointsForGC] Extend base pointer to handle more cases w/vectors

When relocating a pointer, we need to determine a base pointer for the derived pointer being relocated. We have limited support for handling a pointer extracted from a vector; the current code only handled the case where the entire vector was known to contain base pointers. This patch extends the reasoning to handle chains of insertelements where the indices are constants. This case turns out to be fairly common in vectorized code. We can now handle vectors which contains mixtures of base and derived pointers provided the insertelements use constant indices.

Note that this doesn't solve the general problem. To handle variable indexed insertelements, we'd need to scalarize and introduce conditional branching based on the index. Alternatively, we could eagerly scalarize, but the code structure doesn't currently make either fix easy. The patch also doesn't handle shufflevector or other vector manipulation for much the same reasons. I plan to defer this work until I have a motivating test case.

Differential Revision: http://reviews.llvm.org/D9676

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237200 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[lib/Fuzzer] use sha1sum for the file hash
Kostya Serebryany [Tue, 12 May 2015 22:03:34 +0000 (22:03 +0000)]
[lib/Fuzzer] use sha1sum for the file hash

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237198 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoTidy comments in SampleProfile header. NFC.
Diego Novillo [Tue, 12 May 2015 22:03:00 +0000 (22:03 +0000)]
Tidy comments in SampleProfile header. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237197 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[PlaceSafepoints] Add missing "override" to PlaceBackedgeSafepointsImpl::runOnFunction
Justin Bogner [Tue, 12 May 2015 21:49:47 +0000 (21:49 +0000)]
[PlaceSafepoints] Add missing "override" to PlaceBackedgeSafepointsImpl::runOnFunction

Pointed out by -Winconsistent-missing-override.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237196 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd missing #include and forward decl, found by modules build.
Richard Smith [Tue, 12 May 2015 21:49:03 +0000 (21:49 +0000)]
Add missing #include and forward decl, found by modules build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237195 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMergeFunctions: Two different sized allocas are *not* the same
Arnold Schwaighofer [Tue, 12 May 2015 21:42:22 +0000 (21:42 +0000)]
MergeFunctions: Two different sized allocas are *not* the same

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237193 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Statepoints] Clean up statepoint argument accessors.
Pat Gavlin [Tue, 12 May 2015 21:33:48 +0000 (21:33 +0000)]
[Statepoints] Clean up statepoint argument accessors.

Differential Revision: http://reviews.llvm.org/D9622

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237191 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert "ARM: Remove Itineraries for swift CPU"
Matthias Braun [Tue, 12 May 2015 21:28:39 +0000 (21:28 +0000)]
Revert "ARM: Remove Itineraries for swift CPU"

Reverting until I figure out the new lit failures.

This reverts commit r237179.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237189 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoInstrProf: Update name of compiler-rt routine for setting filename
Justin Bogner [Tue, 12 May 2015 21:23:09 +0000 (21:23 +0000)]
InstrProf: Update name of compiler-rt routine for setting filename

Patch by Teresa Johnson.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237186 91177308-0d34-0410-b5e6-96231b3b80d8