platform/upstream/llvm.git
9 years agoUse a more accurate implementation for exp
Tom Stellard [Wed, 13 May 2015 03:55:09 +0000 (03:55 +0000)]
Use a more accurate implementation for exp

Using exp2(x * M_LOG2E_F) does not give us accurate enough results for
OpenCL.  If you look at the new exp implementation you'll see that
it does multiply the input by M_LOG2E_F, but it still uses the original
input in part of the calculation.

This exp implementation was ported from the AMD builtin library
and has been tested with piglit, OpenCV, and the ocl conformance tests.

llvm-svn: 237229

9 years agoImplement exp2 using OpenCL C rather than using an intrinsic
Tom Stellard [Wed, 13 May 2015 03:55:07 +0000 (03:55 +0000)]
Implement exp2 using OpenCL C rather than using an intrinsic

Not all targets support the intrinsic, so it's better to have a
generic implementation which does not use it.

This exp2 implementation was ported from the AMD builtin library
and has been tested with piglit, OpenCV, and the ocl conformance tests.

llvm-svn: 237228

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

llvm-svn: 237227

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.

llvm-svn: 237226

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

llvm-svn: 237225

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

llvm-svn: 237224

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

llvm-svn: 237223

9 years ago[LLD] Properly relocate the LSDA field of MachO eh-frames.
Lang Hames [Wed, 13 May 2015 00:44:47 +0000 (00:44 +0000)]
[LLD] Properly relocate the LSDA field of MachO eh-frames.

Previously the LSDA field was not being relocated during linking, leading to
failures for some EH tests.

llvm-svn: 237222

9 years agoDon't change the selected platform when creating the dummy target.
Robert Flack [Wed, 13 May 2015 00:39:24 +0000 (00:39 +0000)]
Don't change the selected platform when creating the dummy target.

A dummy target is used by TargetList::CreateTargetInternal to prime newly
created targets. the first time this is done it creates the dummy target. The
dummy target is created with the host platform (See
TargetList::CreateDummyTarget) which results in switching the selected platform
back to the host platform even when creating a target for a different platform.
This change avoids changing the selected platform while creating the dummy
target to prevent this side effect.

Test Plan:
./dotest.py $DOTEST_OPTS -t -p TestCreateAfterAttach.py
Tests using process attach (e.g. TestCreateAfterAttach.py, TestHelloWorld.py)
now run successfully mac -> linux.

Differential Revision: http://reviews.llvm.org/D9712

llvm-svn: 237221

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

llvm-svn: 237220

9 years ago[LLD] Add support for MachO ripRel32MinusNAnon relocations.
Lang Hames [Wed, 13 May 2015 00:29:43 +0000 (00:29 +0000)]
[LLD] Add support for MachO ripRel32MinusNAnon relocations.

llvm-svn: 237219

9 years agoFixed a ton of gcc compile warnings
Vince Harron [Wed, 13 May 2015 00:25:54 +0000 (00:25 +0000)]
Fixed a ton of gcc compile warnings

Removed some unused variables, added some consts, changed some casts
to const_cast. I don't think any of these changes are very
controversial.

Differential Revision: http://reviews.llvm.org/D9674

llvm-svn: 237218

9 years ago[LLD] Add a mutex to prevent concurrent modification of the dylib maps in
Lang Hames [Wed, 13 May 2015 00:17:08 +0000 (00:17 +0000)]
[LLD] Add a mutex to prevent concurrent modification of the dylib maps in
MachOLinkingContext.

llvm-svn: 237217

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

llvm-svn: 237216

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

llvm-svn: 237214

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.

llvm-svn: 237212

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.

llvm-svn: 237211

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.

llvm-svn: 237210

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).

llvm-svn: 237209

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.

llvm-svn: 237208

9 years agodosep can run tests in a specific subfolder
Vince Harron [Tue, 12 May 2015 23:12:19 +0000 (23:12 +0000)]
dosep can run tests in a specific subfolder

For example:

./dosep.py -o "$DOTEST_OPTS" lang/c/

Differential Revision: http://reviews.llvm.org/D9724

llvm-svn: 237207

9 years agoAdded support for XTIMEOUT to dosep
Vince Harron [Tue, 12 May 2015 23:10:36 +0000 (23:10 +0000)]
Added support for XTIMEOUT to dosep

Ideally, this would be put in the individual test files.

Unfortunately, I'm not sure how to do that quickly/easily.
I'm open to suggestions.

In the meantime, I'll submit this to stabilze the build server.

llvm-svn: 237206

9 years agoDocument a known build issue on OS X 10.8 and later.
Eric Fiselier [Tue, 12 May 2015 22:55:30 +0000 (22:55 +0000)]
Document a known build issue on OS X 10.8 and later.

llvm-svn: 237205

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.

llvm-svn: 237204

9 years agoChanged renaming of local symbols by inserting a dot before the numeric suffix.
Sunil Srivastava [Tue, 12 May 2015 22:44:24 +0000 (22:44 +0000)]
Changed renaming of local symbols by inserting a dot before the numeric suffix.
Modified two test cases to adjust to the above change in renaming.
These two files were causing the buildbot failure in Polly, #30204 for example.
Details in http://reviews.llvm.org/D9483
This checkin goes with r237150 and r237151

llvm-svn: 237203

9 years agoHave '__have_extension(cxx_variadic_templates)' return true for any C++ standard.
Eric Fiselier [Tue, 12 May 2015 22:37:23 +0000 (22:37 +0000)]
Have '__have_extension(cxx_variadic_templates)' return true for any C++ standard.

llvm-svn: 237202

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

llvm-svn: 237201

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

llvm-svn: 237200

9 years agobenchcomp: Add macho_symsizes mode for comparing Mach-O object symbol sizes.
Peter Collingbourne [Tue, 12 May 2015 22:14:26 +0000 (22:14 +0000)]
benchcomp: Add macho_symsizes mode for comparing Mach-O object symbol sizes.

llvm-svn: 237199

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

llvm-svn: 237198

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.

llvm-svn: 237197

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.

llvm-svn: 237196

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.

llvm-svn: 237195

9 years agoAdd missing #includes, found by modules build.
Richard Smith [Tue, 12 May 2015 21:48:00 +0000 (21:48 +0000)]
Add missing #includes, found by modules build.

llvm-svn: 237194

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

llvm-svn: 237193

9 years agoAdd a new error for unexpected semi-colon before closing delimiter.
Richard Trieu [Tue, 12 May 2015 21:36:35 +0000 (21:36 +0000)]
Add a new error for unexpected semi-colon before closing delimiter.

Previously, if a semi-colon is unexpectedly added before a closing ')', ']' or
'}', two errors and one note would emitted, and the parsing would get confused
to which scope it was in.  This change consumes the semi-colon, recovers
parsing better, and emits only one error with a fix-it.

llvm-svn: 237192

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

llvm-svn: 237191

9 years agoInclude missing 'sanitizer_platform_limits_posix.h'
Ismail Pazarbasi [Tue, 12 May 2015 21:30:16 +0000 (21:30 +0000)]
Include missing 'sanitizer_platform_limits_posix.h'

Unbreak Mac builds. `#include` seems to be forgotten during staging.

llvm-svn: 237190

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.

llvm-svn: 237189

9 years agoInstrProf: Update name of compiler-rt routine for setting filename
Justin Bogner [Tue, 12 May 2015 21:23:16 +0000 (21:23 +0000)]
InstrProf: Update name of compiler-rt routine for setting filename

Patch by Teresa Johnson.

llvm-svn: 237187

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.

llvm-svn: 237186

9 years ago[PlaceSafepoints] Switch to being a FunctionPass
Philip Reames [Tue, 12 May 2015 21:21:18 +0000 (21:21 +0000)]
[PlaceSafepoints] Switch to being a FunctionPass

The pass doesn't actually modify the module outside of the function being processed. The only confusing piece is that it both inserts calls and then inlines the resulting calls. Given that, it definitely invalidates module level analysis results, but many FunctionPasses do that.

Differential Revision: http://reviews.llvm.org/D9590

llvm-svn: 237185

9 years agoRevert accidental file checkin
Vince Harron [Tue, 12 May 2015 21:19:39 +0000 (21:19 +0000)]
Revert accidental file checkin

llvm-svn: 237184

9 years agoclang-format AllocateTarget. NFC
Douglas Katzman [Tue, 12 May 2015 21:18:10 +0000 (21:18 +0000)]
clang-format AllocateTarget. NFC

llvm-svn: 237183

9 years agoXFAIL'd TestPersistObjCPointeeType.py
Vince Harron [Tue, 12 May 2015 21:16:51 +0000 (21:16 +0000)]
XFAIL'd TestPersistObjCPointeeType.py

llvm-svn: 237182

9 years agofix formatting; NFC
Sanjay Patel [Tue, 12 May 2015 21:14:24 +0000 (21:14 +0000)]
fix formatting; NFC

llvm-svn: 237181

9 years ago[PlaceSafepoints] Make internal helper pass a FunctionPass
Philip Reames [Tue, 12 May 2015 21:09:36 +0000 (21:09 +0000)]
[PlaceSafepoints] Make internal helper pass a FunctionPass

Switch from using a LoopPass to using a FunctionPass for the internal helper analysis pass. The next step is going to be to make this a true analysis pass which is required by the PlaceSafepoints pass itself.

p.s. The interesting semantic part here is that we're changing the iteration order over the loops. It shouldn't matter, but that's the reason to separate this into it's own distinct patch.

Differential Revision: http://reviews.llvm.org/D9588

llvm-svn: 237180

9 years agoARM: Remove Itineraries for swift CPU
Matthias Braun [Tue, 12 May 2015 21:07:54 +0000 (21:07 +0000)]
ARM: Remove Itineraries for swift CPU

They do more harm than good when used in the MachineScheduler as they
tend to take preference to register pressure minimsation which is more
important for swift.

Differential Revision: http://reviews.llvm.org/D9718

llvm-svn: 237179

9 years agoPlatform settings doesn't need permission options.
Chaoren Lin [Tue, 12 May 2015 21:06:48 +0000 (21:06 +0000)]
Platform settings doesn't need permission options.

Summary:
Also removes conflict error message for '-w' between
--user-write and --working-dir.

Reviewers: flackr, clayborg

Reviewed By: clayborg

Subscribers: lldb-commits

Differential Revision: http://reviews.llvm.org/D9711

llvm-svn: 237178

9 years agoCall system's sigfillset and sigprocmask functions from sanitizers
Ismail Pazarbasi [Tue, 12 May 2015 20:56:44 +0000 (20:56 +0000)]
Call system's sigfillset and sigprocmask functions from sanitizers

Reviewers: kcc, glider, dvyukov

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9637

llvm-svn: 237177

9 years ago[PlaceSafepoints] Use analysis infrastructure to get dominator tree
Philip Reames [Tue, 12 May 2015 20:56:33 +0000 (20:56 +0000)]
[PlaceSafepoints] Use analysis infrastructure to get dominator tree

The old code computed dominators for every loop. This was terribly slow with no good reason. Just use the standard infrastructure for analysis passes.

Differential Revision: http://reviews.llvm.org/D9586

llvm-svn: 237176

9 years ago[X86] Always return the sret parameter in eax/rax, even on 32-bit
Reid Kleckner [Tue, 12 May 2015 20:56:32 +0000 (20:56 +0000)]
[X86] Always return the sret parameter in eax/rax, even on 32-bit

Summary:
This rule was always in the old SysV i386 ABI docs and the new ones that
H.J. Lu has put together, but we never noticed:

  EAX   scratch register; also used to return integer and pointer values
        from functions; also stores the address of a returned struct or union

Fixes PR23491.

Reviewers: majnemer

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9715

llvm-svn: 237175

9 years agoAdded minor comment
Vince Harron [Tue, 12 May 2015 20:55:43 +0000 (20:55 +0000)]
Added minor comment

llvm-svn: 237174

9 years agoSanitizers: Implement `GetRSS` on Mac OS X
Ismail Pazarbasi [Tue, 12 May 2015 20:47:21 +0000 (20:47 +0000)]
Sanitizers: Implement `GetRSS` on Mac OS X

Reviewers: kcc, glider, dvyukov

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D9636

llvm-svn: 237173

9 years ago[PlaceSafepoints] Remove dependence on LoopSimplify
Philip Reames [Tue, 12 May 2015 20:43:48 +0000 (20:43 +0000)]
[PlaceSafepoints] Remove dependence on LoopSimplify

As a step towards getting rid of internal pass manager hack entirely, remove the need for loop simplify to run in the inner pass manager. The new code does produce slightly different loop structures, so this isn't technically NFC.

Differential Revision: http://reviews.llvm.org/D9585

llvm-svn: 237172

9 years agoList Alexey Bataev as code owner for Clang's OpenMP support, as discussed offline.
Richard Smith [Tue, 12 May 2015 20:29:41 +0000 (20:29 +0000)]
List Alexey Bataev as code owner for Clang's OpenMP support, as discussed offline.

Thanks for all your hard work getting us OpenMP feature-complete, Alexey!

llvm-svn: 237171

9 years ago[PATCH] [HEXAGON] Add a test program to verify calling convention
Sundeep Kushwaha [Tue, 12 May 2015 20:13:10 +0000 (20:13 +0000)]
[PATCH] [HEXAGON] Add a test program to verify calling convention
for large struct return by value.

Differential Revision: http://reviews.llvm.org/D9709

llvm-svn: 237170

9 years agoConvert PHI getIncomingValue() to foreach over incoming_values(). NFC.
Pete Cooper [Tue, 12 May 2015 20:05:31 +0000 (20:05 +0000)]
Convert PHI getIncomingValue() to foreach over incoming_values().  NFC.

We already had a method to iterate over all the incoming values of a PHI.  This just changes all eligible code to use it.

Ineligible code included anything which cared about the index, or was also trying to get the i'th incoming BB.

llvm-svn: 237169

9 years agoAdd a const method to iterator over PHI incoming values.
Pete Cooper [Tue, 12 May 2015 20:05:23 +0000 (20:05 +0000)]
Add a const method to iterator over PHI incoming values.

This will be used in the next commit which converts code to use incoming_values().

llvm-svn: 237168

9 years agoConstify method. NFC
Pete Cooper [Tue, 12 May 2015 20:05:20 +0000 (20:05 +0000)]
Constify method.  NFC

llvm-svn: 237167

9 years ago[Statepoints] Split the calling convention and statepoint flags operand to STATEPOINT...
Pat Gavlin [Tue, 12 May 2015 19:50:19 +0000 (19:50 +0000)]
[Statepoints] Split the calling convention and statepoint flags operand to STATEPOINT into two separate operands.

Differential Revision: http://reviews.llvm.org/D9623

llvm-svn: 237166

9 years agoStrip trailing whitespace. NFC
Douglas Katzman [Tue, 12 May 2015 19:42:31 +0000 (19:42 +0000)]
Strip trailing whitespace. NFC

llvm-svn: 237165

9 years agoR600/SI: Fix bug in VGPR spilling
Tom Stellard [Tue, 12 May 2015 18:59:17 +0000 (18:59 +0000)]
R600/SI: Fix bug in VGPR spilling

AMDGPU::SI_SPILL_V96_RESTORE was missing from a switch statement, which
caused the srsrc and soffset register to not be set correctly.

This commit replaces the switch statement with a SITargetInfo query
to make sure all spill instructions are covered.

Differential Revision: http://reviews.llvm.org/D9582

llvm-svn: 237164

9 years ago[lib/Fuzzer] guess the right number of workers if -jobs=N is given but -workers=M...
Kostya Serebryany [Tue, 12 May 2015 18:51:57 +0000 (18:51 +0000)]
[lib/Fuzzer] guess the right number of workers if -jobs=N is given but -workers=M is not. Update the docs.

llvm-svn: 237163

9 years agoWindows: fix bug in getcwd() and add chdir().
Ted Woodward [Tue, 12 May 2015 18:47:33 +0000 (18:47 +0000)]
Windows: fix bug in getcwd() and add chdir().

Summary:
GetCurrentDirectory() returns the number of characters copied; 0 is a failure, not a success.

Add implementation for chdir().

Reviewers: zturner

Reviewed By: zturner

Subscribers: lldb-commits

Differential Revision: http://reviews.llvm.org/D9300

llvm-svn: 237162

9 years ago[Builtins] Implement half-precision conversions.
Ahmed Bougacha [Tue, 12 May 2015 18:33:42 +0000 (18:33 +0000)]
[Builtins] Implement half-precision conversions.

Mostly uninteresting, except:
- in __extendXfYf2, when checking if the number is normal, the old
  code relied on the unsignedness of src_rep_t, which is a problem
  when sizeof(src_rep_t) < sizeof(int): the result gets promoted to
  int, the signedness of which breaks the comparison.
  I added an explicit cast; it shouldn't affect other types.
- we can't pass __fp16, so src_t and src_rep_t are the same.
- the gnu_*_ieee symbols are simply duplicated definitions, as aliases
  are problematic on mach-o (where only weak aliases are supported;
  that's not what we want).

Differential Revision: http://reviews.llvm.org/D9693

llvm-svn: 237161

9 years agoRevert r237157, "YAML: Fix typos. NFC".
Alex Lorenz [Tue, 12 May 2015 17:44:32 +0000 (17:44 +0000)]
Revert r237157, "YAML: Fix typos. NFC".

'Iff' isn't a typo, it's a shorthand for 'if and only if'.

llvm-svn: 237160

9 years agoFixed double-free in case of module loading error.
Artem Belevich [Tue, 12 May 2015 17:44:15 +0000 (17:44 +0000)]
Fixed double-free in case of module loading error.

GetOutputStream() owns the stream it returns pointer to and the
pointer should never be freed by us. When we fail to load and exit
early, unique_ptr still holds the pointer and frees it which leads to
compiler crash when CompilerInstance attempts to free it again.

Added regression test for failed bitcode linking.

Differential Revision: http://reviews.llvm.org/D9625

llvm-svn: 237159

9 years ago[mips][microMIPSr6] Implement SELEQZ and SELNEZ instructions
Jozef Kolek [Tue, 12 May 2015 17:39:32 +0000 (17:39 +0000)]
[mips][microMIPSr6] Implement SELEQZ and SELNEZ instructions

This patch implements SELEQZ and SELNEZ instructions using mapping.

Differential Revision: http://reviews.llvm.org/D8497

llvm-svn: 237158

9 years agoYAML: Fix typos. NFC.
Alex Lorenz [Tue, 12 May 2015 17:31:17 +0000 (17:31 +0000)]
YAML: Fix typos. NFC.

llvm-svn: 237157

9 years agoReimplement heuristic for estimating complete-unroll optimization effects.
Michael Zolotukhin [Tue, 12 May 2015 17:20:03 +0000 (17:20 +0000)]
Reimplement heuristic for estimating complete-unroll optimization effects.

Summary:
This patch reimplements heuristic that tries to estimate optimization beneftis
from complete loop unrolling.

In this patch I kept the minimal changes - e.g. I removed code handling
branches and folding compares. That's a promising area, but now there
are too many questions to discuss before we can enable it.

Test Plan: Tests are included in the patch.

Reviewers: hfinkel, chandlerc

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D8816

llvm-svn: 237156

9 years agoImplement sin for double types
Tom Stellard [Tue, 12 May 2015 17:18:47 +0000 (17:18 +0000)]
Implement sin for double types

This implementation was ported from the AMD builtin library
and has been tested with piglit, OpenCV, and the ocl conformance tests.

llvm-svn: 237155

9 years agoImplement cos for double types
Tom Stellard [Tue, 12 May 2015 17:18:46 +0000 (17:18 +0000)]
Implement cos for double types

This implementation was ported from the AMD builtin library
and has been tested with piglit, OpenCV, and the ocl conformance tests.

llvm-svn: 237154

9 years ago[Mips] Return false for isFPCloseToIncomingSP()
Petar Jovanovic [Tue, 12 May 2015 17:14:05 +0000 (17:14 +0000)]
[Mips] Return false for isFPCloseToIncomingSP()

On Mips, frame pointer points to the same side of the frame as the stack
pointer. This function is used to decide where to put register scavenging
spill slot. So far, it was put on the wrong side of the frame, and thus it
was too far away from $fp when frame was larger than 2^15 bytes.

Patch by Vladimir Radosavljevic.

http://reviews.llvm.org/D8895

llvm-svn: 237153

9 years agoR600/SI: add pass to mark CF live ranges as non-spillable
Tom Stellard [Tue, 12 May 2015 17:13:02 +0000 (17:13 +0000)]
R600/SI: add pass to mark CF live ranges as non-spillable

Spilling can insert instructions almost anywhere, and this can mess
up control flow lowering in a multitude of ways, due to instruction
reordering. Let's sort this out the easy way: never spill registers
involved with control flow, i.e. saved EXEC masks.

Unfortunately, this does not work at all with optimizations disabled,
as the register allocator ignores spill weights. This should be
addressed in a future commit.

The test was reduced from the "stacks" shader of [1]. Some issues
trigger the machine verifier while another one is checked manually.

[1] http://madebyevan.com/webgl-path-tracing/

v2: only insert pass with optimizations enabled, merge test runs.

Patch by: Grigori Goronzy

llvm-svn: 237152

9 years agoChanged renaming of local symbols by inserting a dot vefore the numeric suffix
Sunil Srivastava [Tue, 12 May 2015 16:48:43 +0000 (16:48 +0000)]
Changed renaming of local symbols by inserting a dot vefore the numeric suffix
details in http://reviews.llvm.org/D9483
goes with llvm checkin r237150

llvm-svn: 237151

9 years agoChanged renaming of local symbols by inserting a dot vefore the numeric suffix.
Sunil Srivastava [Tue, 12 May 2015 16:47:30 +0000 (16:47 +0000)]
Changed renaming of local symbols by inserting a dot vefore the numeric suffix.
One code change and several test changes to match that
details in http://reviews.llvm.org/D9481

llvm-svn: 237150

9 years ago[sanitizer] Update "sancov.py missing" to allow __sanitizer_cov_with_check().
Sergey Matveev [Tue, 12 May 2015 16:46:54 +0000 (16:46 +0000)]
[sanitizer] Update "sancov.py missing" to allow __sanitizer_cov_with_check().

llvm-svn: 237149

9 years agoDelete dead code. NFC
Jonathan Roelofs [Tue, 12 May 2015 16:45:53 +0000 (16:45 +0000)]
Delete dead code. NFC

llvm-svn: 237148

9 years ago[sanitizer] Override pipefail in coverage-missing test.
Sergey Matveev [Tue, 12 May 2015 16:32:41 +0000 (16:32 +0000)]
[sanitizer] Override pipefail in coverage-missing test.

This should finally fix it.

llvm-svn: 237147

9 years agoComment disambiguation.
Douglas Katzman [Tue, 12 May 2015 15:45:56 +0000 (15:45 +0000)]
Comment disambiguation.

llvm-svn: 237146

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

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

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

llvm-svn: 237145

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

llvm-svn: 237144

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

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

llvm-svn: 237143

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

It is no longer used.

llvm-svn: 237142

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

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

llvm-svn: 237141

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

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

llvm-svn: 237140

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

llvm-svn: 237139

9 years agoImplement atan2pi builtin
Tom Stellard [Tue, 12 May 2015 14:48:26 +0000 (14:48 +0000)]
Implement atan2pi builtin

This implementation was ported from the AMD builtin library
and has been tested with piglit, OpenCV, and the ocl conformance tests.

llvm-svn: 237138

9 years ago[sanitizer] Re-enable a broken test with debug output.
Sergey Matveev [Tue, 12 May 2015 14:43:31 +0000 (14:43 +0000)]
[sanitizer] Re-enable a broken test with debug output.

llvm-svn: 237137

9 years agoFix clang-format build from the solution; the underlying path has changed to include...
Manuel Klimek [Tue, 12 May 2015 14:41:39 +0000 (14:41 +0000)]
Fix clang-format build from the solution; the underlying path has changed to include the VS directory structure.

llvm-svn: 237136

9 years agoXFAIL and Skip flaky test case on linux
Tamas Berghammer [Tue, 12 May 2015 14:25:43 +0000 (14:25 +0000)]
XFAIL and Skip flaky test case on linux

llvm-svn: 237135

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

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

llvm-svn: 237134

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

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

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

llvm-svn: 237133

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

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

llvm-svn: 237132

9 years agoImplement atan2 for doubles
Tom Stellard [Tue, 12 May 2015 13:48:51 +0000 (13:48 +0000)]
Implement atan2 for doubles

This implementation was ported from the AMD builtin library
and has been tested with piglit, OpenCV, and the ocl conformance tests.

llvm-svn: 237131

9 years agopatch to allow lib name at configure time (from C.Bergstrom)
Andrey Churbanov [Tue, 12 May 2015 13:31:57 +0000 (13:31 +0000)]
patch to allow lib name at configure time (from C.Bergstrom)

llvm-svn: 237130

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

llvm-svn: 237129

9 years agoCall remote platform GetSupportedArchitectureAtIndex if connected to remote.
Robert Flack [Tue, 12 May 2015 13:22:37 +0000 (13:22 +0000)]
Call remote platform GetSupportedArchitectureAtIndex if connected to remote.

Updated PlatformLinux::GetSupportedArchitectureAtIndex to call the
PlatformRemoteGdbServer::GetSupportedArchitectureAtIndex if connected remotely.
This should return the correct thing for android (to fix those failing tests),
and is also working for mac to linux.

Test Plan:
./dotest.py $DOTEST_OPTS -t -p TestCallStdStringFunction.py
The above still passes when running mac->linux indicating it successfully
identified PlatformLinux as the target platform and converted the mmap options
correctly.

Differential Revision: http://reviews.llvm.org/D9672

llvm-svn: 237128

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

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

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

llvm-svn: 237127