oota-llvm.git
8 years agoRefactor FunctionImporter::importFunctions with a helper function to process the...
Mehdi Amini [Thu, 3 Dec 2015 02:37:33 +0000 (02:37 +0000)]
Refactor FunctionImporter::importFunctions with a helper function to process the Worklist (NFC)

This precludes some more functional changes to perform bulk imports.

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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254583 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAdapt comment and rename variable in ModuleLinker to describe more accurately the...
Mehdi Amini [Thu, 3 Dec 2015 02:37:30 +0000 (02:37 +0000)]
Adapt comment and rename variable in ModuleLinker to describe more accurately the actual use.

Thanks Sean Silva for the suggestion.

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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254582 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRemove "ExportingModule" from ThinLTO Index (NFC)
Mehdi Amini [Thu, 3 Dec 2015 02:37:23 +0000 (02:37 +0000)]
Remove "ExportingModule" from ThinLTO Index (NFC)

There is no real reason the index has to have the concept of an
exporting Module. We should be able to have one single unique
instance of the Index, and it should be read-only after creation
for the whole ThinLTO processing.
The linker plugin should be able to process multiple modules (in
parallel or in sequence) with the same index.

The only reason the ExportingModule was present seems to be to
implement hasExportedFunctions() that is used by the Module linker
to decide what to do with the current Module.
For now I replaced it with a query to the map of Modules path to
see if this module was declared in the Index and consider that if
it is the case then it is probably exporting function.
On the long term the Linker interface needs to evolve and this
call should not be needed anymore.

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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254581 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAdd a TODO item that the nop handling before FP conditional branches is
Joerg Sonnenberger [Thu, 3 Dec 2015 02:35:24 +0000 (02:35 +0000)]
Add a TODO item that the nop handling before FP conditional branches is
not enough for SPARCv7.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254580 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoScheduleDAGInstrs: Rework schedule graph builder.
Matthias Braun [Thu, 3 Dec 2015 02:05:27 +0000 (02:05 +0000)]
ScheduleDAGInstrs: Rework schedule graph builder.

The new algorithm remembers the uses encountered while walking backwards
until a matching def is found. Contrary to the previous version this:
- Works without LiveIntervals being available
- Allows to increase the precision to subregisters/lanemasks
  (not used for now)

The changes in the AMDGPU tests are necessary because the R600 scheduler
is not stable with respect to the order of nodes in the ready queues.

Differential Revision: http://reviews.llvm.org/D9068

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254577 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRegisterPressure: Use range based for, fix else style; NFC
Matthias Braun [Thu, 3 Dec 2015 01:44:45 +0000 (01:44 +0000)]
RegisterPressure: Use range based for, fix else style; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254575 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[PGO] Add v2 format compatibility test
Xinliang David Li [Thu, 3 Dec 2015 01:05:31 +0000 (01:05 +0000)]
[PGO] Add v2 format compatibility test

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254572 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoMC: Make sure to clear *all* of MCMachOStreamer's state
Justin Bogner [Thu, 3 Dec 2015 00:52:20 +0000 (00:52 +0000)]
MC: Make sure to clear *all* of MCMachOStreamer's state

The CreatedADWARFSection flag was added in r232842, but isn't cleared
properly when resetting the streamer's state. Fix that.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254571 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[WebAssembly] Add a test for wasm-store-results pass
Derek Schuff [Thu, 3 Dec 2015 00:50:30 +0000 (00:50 +0000)]
[WebAssembly] Add a test for wasm-store-results pass

Differential Revision: http://reviews.llvm.org/D15167

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254570 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[WebAssembly] Assert that byval and nest are not used for return types.
Dan Gohman [Wed, 2 Dec 2015 23:40:03 +0000 (23:40 +0000)]
[WebAssembly] Assert that byval and nest are not used for return types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254567 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRename a header guard to be more appropriate
David Majnemer [Wed, 2 Dec 2015 23:28:27 +0000 (23:28 +0000)]
Rename a header guard to be more appropriate

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254566 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoForgot to add this file with r254562.
David Majnemer [Wed, 2 Dec 2015 23:09:05 +0000 (23:09 +0000)]
Forgot to add this file with r254562.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254565 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[Hexagon] Improve lowering of instructions to the MC layer
Krzysztof Parzyszek [Wed, 2 Dec 2015 23:08:29 +0000 (23:08 +0000)]
[Hexagon] Improve lowering of instructions to the MC layer

- Add extenders when necessary.
- Handle some basic relocations.

This should fix the failure in tools/clang/test/CodeGenCXX/crash.cpp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254564 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoMove EH-specific helper functions to a more appropriate place
David Majnemer [Wed, 2 Dec 2015 23:06:39 +0000 (23:06 +0000)]
Move EH-specific helper functions to a more appropriate place

No functionality change is intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254562 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFixup for r254547: use format_hex() to simplify code.
Alexey Samsonov [Wed, 2 Dec 2015 22:59:22 +0000 (22:59 +0000)]
Fixup for r254547: use format_hex() to simplify code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254560 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoSwitch the linker to having a whitelist of GVs.
Rafael Espindola [Wed, 2 Dec 2015 22:59:04 +0000 (22:59 +0000)]
Switch the linker to having a whitelist of GVs.

This replaces DoNotLinkFromSource with ValuesToLink. It also moves the
computation of ValuesToLink earlier.

It is a bit simpler and an important step in slitting the linker into an
ir mover and a linker proper.

The test change is because we now avoid creating dead declarations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254559 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoLibfuzzer: do not pass null into user function
Mike Aizatsky [Wed, 2 Dec 2015 22:43:53 +0000 (22:43 +0000)]
Libfuzzer: do not pass null into user function

Differential Revision: http://reviews.llvm.org/D15098

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254558 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoUse std::string instead of strdup() and free() in WinCodeViewLineTables
Reid Kleckner [Wed, 2 Dec 2015 22:34:30 +0000 (22:34 +0000)]
Use std::string instead of strdup() and free() in WinCodeViewLineTables

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254557 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoDelete what is now duplicated code.
Rafael Espindola [Wed, 2 Dec 2015 22:22:24 +0000 (22:22 +0000)]
Delete what is now duplicated code.

Having to import an alias as declaration is not thinlto specific.

The test difference are because when we already have a decl and we are
not importing it, we just leave the decl alone.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254556 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[llvm-dwp] Include only the non-empty columns in the cu_index
David Blaikie [Wed, 2 Dec 2015 22:01:56 +0000 (22:01 +0000)]
[llvm-dwp] Include only the non-empty columns in the cu_index

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254555 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[PGO] Allow input value node list to be null
Xinliang David Li [Wed, 2 Dec 2015 21:47:43 +0000 (21:47 +0000)]
[PGO] Allow input value node list to be null

This is to handle the case when vp node linked
list array is laziliy initialized at runtime

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254551 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix a typo in LoopVectorize.cpp. NFC.
Cong Hou [Wed, 2 Dec 2015 21:33:47 +0000 (21:33 +0000)]
Fix a typo in LoopVectorize.cpp. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254549 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[PowerPC] Remove wild call to RegScavenger::initRegState().
Alexey Samsonov [Wed, 2 Dec 2015 21:25:28 +0000 (21:25 +0000)]
[PowerPC] Remove wild call to RegScavenger::initRegState().

This call should in fact be made by RegScavenger::enterBasicBlock()
called below. The first call does nothing except for triggering UB,
indicated by UBSan (passing nullptr to memset()).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254548 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[Hexagon] Remove std::hex in favor of format().
Alexey Samsonov [Wed, 2 Dec 2015 21:13:43 +0000 (21:13 +0000)]
[Hexagon] Remove std::hex in favor of format().

std::hex is not used anywhere in LLVM code base except for this place,
and it has a known undefined behavior (at least in libstdc++ 4.9.3):
https://llvm.org/bugs/show_bug.cgi?id=18156, which fires in UBSan
bootstrap of LLVM.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254547 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoTests: PPC: remove unnecessary metadata. NFC
Kyle Butt [Wed, 2 Dec 2015 21:08:03 +0000 (21:08 +0000)]
Tests: PPC: remove unnecessary metadata. NFC

Remove unnecessary metadata from a test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254544 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAlso copy private linkage globals when needed.
Rafael Espindola [Wed, 2 Dec 2015 20:57:33 +0000 (20:57 +0000)]
Also copy private linkage globals when needed.

This was an omission when handling COFF style comdats with local keys.
Should fix the sanitizer-windows bot.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254543 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRe-enable UBSan tests for SystemZ: PR20980 was fixed.
Alexey Samsonov [Wed, 2 Dec 2015 20:46:51 +0000 (20:46 +0000)]
Re-enable UBSan tests for SystemZ: PR20980 was fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254542 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoDon't copy information from aliasee to alias.
Rafael Espindola [Wed, 2 Dec 2015 20:03:17 +0000 (20:03 +0000)]
Don't copy information from aliasee to alias.

They are independent.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254541 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAMDGPU/SI: Correctly emit agent global segment variables when targeting HSA
Tom Stellard [Wed, 2 Dec 2015 19:47:57 +0000 (19:47 +0000)]
AMDGPU/SI: Correctly emit agent global segment variables when targeting HSA

Differential Revision: http://reviews.llvm.org/D14508

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254540 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[Hexagon] Remove TFRI_V4 instruction, use existing A2_tfrsi instead
Krzysztof Parzyszek [Wed, 2 Dec 2015 19:44:35 +0000 (19:44 +0000)]
[Hexagon] Remove TFRI_V4 instruction, use existing A2_tfrsi instead

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254539 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix linking when we copy over only a decl.
Rafael Espindola [Wed, 2 Dec 2015 19:30:52 +0000 (19:30 +0000)]
Fix linking when we copy over only a decl.

We were failing to copy the fact that the GV is weak and in the case of
an alias, producing invalid IR.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254538 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[CodeGen]: Fix bad interaction with AntiDep breaking and inline asm.
Kyle Butt [Wed, 2 Dec 2015 18:58:51 +0000 (18:58 +0000)]
[CodeGen]: Fix bad interaction with AntiDep breaking and inline asm.

AggressiveAntiDepBreaker was renaming registers specified by the user
for inline assembly. While this will work for compiler-specified
registers, it won't work for user-specified registers, and at the time
this runs, I don't currently see a way to distinguish them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254532 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoTest Commit: iteratee
Kyle Butt [Wed, 2 Dec 2015 18:53:33 +0000 (18:53 +0000)]
Test Commit: iteratee

Remove whitespace from blank lines. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254531 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix accidental off by one change
Fiona Glaser [Wed, 2 Dec 2015 18:46:23 +0000 (18:46 +0000)]
Fix accidental off by one change

Didn't break any tests, but did unnecessary extra work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254529 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAMDGPU: Fix msan test failure
Tom Stellard [Wed, 2 Dec 2015 18:35:23 +0000 (18:35 +0000)]
AMDGPU: Fix msan test failure

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254527 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoScheduler / Regalloc: use unique_ptr[] instead of std::vector
Fiona Glaser [Wed, 2 Dec 2015 18:32:59 +0000 (18:32 +0000)]
Scheduler / Regalloc: use unique_ptr[] instead of std::vector

vector.resize() is significantly slower than memset in many STLs
and the cost of initializing these vectors is significant on targets
with many registers. Since we don't need the overhead of a vector,
use a simple unique_ptr instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254526 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[llvm-profdata] Change instr prof counter overflow to saturate rather than discard
Nathan Slingerland [Wed, 2 Dec 2015 18:19:24 +0000 (18:19 +0000)]
[llvm-profdata] Change instr prof counter overflow to saturate rather than discard

Summary: This changes overflow handling during instrumentation profile merge. Rathar than throwing away records that would result in counter overflow, merged counts are instead clamped to the maximum representable value. A warning about counter overflow is still surfaced to the user as before.

Reviewers: dnovillo, davidxl, silvas

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D14893

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254525 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAArch64: use ldxp/stxp pair to implement 128-bit atomic loads.
Tim Northover [Wed, 2 Dec 2015 18:12:57 +0000 (18:12 +0000)]
AArch64: use ldxp/stxp pair to implement 128-bit atomic loads.

The ARM ARM is clear that 128-bit loads are only guaranteed to have been atomic
if there has been a corresponding successful stxp. It's less clear for AArch32, so
I'm leaving that alone for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254524 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[WebAssembly] Fix comments to say "LIFO" instead of "FIFO" when describing a stack.
Dan Gohman [Wed, 2 Dec 2015 18:08:49 +0000 (18:08 +0000)]
[WebAssembly] Fix comments to say "LIFO" instead of "FIFO" when describing a stack.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254523 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAMDGPU/SI: Don't emit group segment global variables
Tom Stellard [Wed, 2 Dec 2015 17:00:42 +0000 (17:00 +0000)]
AMDGPU/SI: Don't emit group segment global variables

Summary: Only global or readonly segment variables should appear in object files.

Reviewers: arsenm

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D15111

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254519 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoDo (A == C1 || A == C2) -> (A & ~(C1 ^ C2)) == C1 rather than (A == C1 || A == C2...
David Majnemer [Wed, 2 Dec 2015 16:15:07 +0000 (16:15 +0000)]
Do (A == C1 || A == C2) -> (A & ~(C1 ^ C2)) == C1 rather than (A == C1 || A == C2) -> (A | (C1 ^ C2)) == C2 when C1 ^ C2 is a power of 2.

Differential Revision: http://reviews.llvm.org/D14223

Patch by Amaury SECHET!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254518 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAdd an interesting case we already get right.
Rafael Espindola [Wed, 2 Dec 2015 15:02:43 +0000 (15:02 +0000)]
Add an interesting case we already get right.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254514 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRollback r254508 and r254511 to fix buildbots
Andy Gibbs [Wed, 2 Dec 2015 14:36:48 +0000 (14:36 +0000)]
Rollback r254508 and r254511 to fix buildbots

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254513 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoBy intel spec
Michael Zuckerman [Wed, 2 Dec 2015 14:34:34 +0000 (14:34 +0000)]
By intel spec

|9B DD /7| FSTSW m2byte| Valid Valid Store FPU status word at m2byteafter checking for pending unmasked floating-point exceptions.|
|9B DF E0| FSTSW AX| Valid Valid Store FPU status word in AX register after checking for pending unmasked floating-point exceptions.|
|DD /7 |FNSTSW *m2byte| Valid Valid Store FPU status word at m2bytewithout checking for pending unmasked floating-point exceptions.|
|DF E0 |FNSTSW *AX| Valid Valid Store FPU status word in AX register without checking for pending unmasked floating-point exceptions|

m2byte is word register, and therefor instruction operand need to be change from f32mem to i16mem.

Differential Revision: http://reviews.llvm.org/D14953

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254512 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix buildbots broken by r254508
Andy Gibbs [Wed, 2 Dec 2015 14:22:18 +0000 (14:22 +0000)]
Fix buildbots broken by r254508

g++ 4.7 does not allow an inline defaulted virtual destructor to be overridden,
giving the error "looser throw specifier for ... overridding ~SCEVPredicate()
noexcept (true)" (see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=53613).

The work-around given in the bug report above has been utilised here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254511 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix class SCEVPredicate has virtual functions and accessible non-virtual destructor
Andy Gibbs [Wed, 2 Dec 2015 13:41:24 +0000 (13:41 +0000)]
Fix class SCEVPredicate has virtual functions and accessible non-virtual destructor

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254508 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[AArch64]: Add support for Cortex-A35
Christof Douma [Wed, 2 Dec 2015 11:53:44 +0000 (11:53 +0000)]
[AArch64]: Add support for Cortex-A35

Adds support for the new Cortex-A35 ARMv8-A core.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254503 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoPatch to fix a crash in the PowerPC back end due to ISD::ROTL and ISD::ROTR
Nemanja Ivanovic [Wed, 2 Dec 2015 10:36:24 +0000 (10:36 +0000)]
Patch to fix a crash in the PowerPC back end due to ISD::ROTL and ISD::ROTR
not being expanded. Test case included.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254501 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[mips][microMIPS] Implement PREPEND, RADDU.W.QB, RDDSP, REPL.PH, REPL.QB, REPLV.PH...
Hrvoje Varga [Wed, 2 Dec 2015 09:31:24 +0000 (09:31 +0000)]
[mips][microMIPS] Implement PREPEND, RADDU.W.QB, RDDSP, REPL.PH, REPL.QB, REPLV.PH, REPLV.QB and MTHLIP instructions
Differential Revision: http://reviews.llvm.org/D14527

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254496 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86][FMA] Optimize FNEG(FMUL) Patterns
Simon Pilgrim [Wed, 2 Dec 2015 09:07:55 +0000 (09:07 +0000)]
[X86][FMA] Optimize FNEG(FMUL) Patterns

On FMA targets, we can avoid having to load a constant to negate a float/double multiply by instead using a FNMSUB (-(X*Y)-0)

Fix for PR24366

Differential Revision: http://reviews.llvm.org/D14909

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254495 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAVX-512: Updated cost of FP/SINT/UINT conversion operations
Elena Demikhovsky [Wed, 2 Dec 2015 08:59:47 +0000 (08:59 +0000)]
AVX-512: Updated cost of FP/SINT/UINT conversion operations

I checked and updated the cost of AVX-512 conversion operations. Added cost of conversion operations in DQ mode.
Conversion of illegal types that requires vector split is not calculated right now (like for other X86 targets).

Differential Revision: http://reviews.llvm.org/D15074

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254494 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86][AVX512] add comi with Sae
Asaf Badouh [Wed, 2 Dec 2015 08:17:51 +0000 (08:17 +0000)]
[X86][AVX512] add comi with Sae
add builtin_ia32_vcomisd and builtin_ia32_vcomisd

Differential Revision: http://reviews.llvm.org/D14331

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254493 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[llvm-dwp] Don't rely on implicit move assignment operator (MSVC won't synthesize...
David Blaikie [Wed, 2 Dec 2015 07:09:26 +0000 (07:09 +0000)]
[llvm-dwp] Don't rely on implicit move assignment operator (MSVC won't synthesize one)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254492 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[AttributeSet] Overload AttributeSet::addAttribute to reduce compile
Akira Hatanaka [Wed, 2 Dec 2015 06:58:49 +0000 (06:58 +0000)]
[AttributeSet] Overload AttributeSet::addAttribute to reduce compile
time.

The new overloaded function is used when an attribute is added to a
large number of slots of an AttributeSet (for example, to function
parameters). This is much faster than calling AttributeSet::addAttribute
once per slot, because AttributeSet::getImpl (which calls
FoldingSet::FIndNodeOrInsertPos) is called only once per function
instead of once per slot.

With this commit, clang compiles a file which used to take over 22
minutes in just 13 seconds.

rdar://problem/23581000

Differential Revision: http://reviews.llvm.org/D15085

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254491 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Change getZeroVector to take an MVT instead of EVT. One minor change needed...
Craig Topper [Wed, 2 Dec 2015 06:39:19 +0000 (06:39 +0000)]
[X86] Change getZeroVector to take an MVT instead of EVT. One minor change needed to only try to perform 256-it shuffle combines on legal vector types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254490 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[llvm-dwp] Emit a rather fictional debug_cu_index
David Blaikie [Wed, 2 Dec 2015 06:21:34 +0000 (06:21 +0000)]
[llvm-dwp] Emit a rather fictional debug_cu_index

This is very rudimentary support for debug_cu_index, but it is enough to
allow llvm-dwarfdump to find the offsets for  contributions and
correctly dump debug_info.

It will need to actually find the real signature of the unit and build
the real hash table with the right number of buckets, as per the DWP
specification.

It will also need to be expanded to cover the tu_index as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254489 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoDebugInfo\DWARF: Privatize some accidentally public members
David Blaikie [Wed, 2 Dec 2015 06:21:28 +0000 (06:21 +0000)]
DebugInfo\DWARF: Privatize some accidentally public members

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254488 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Fix weird identation. NFC
Craig Topper [Wed, 2 Dec 2015 05:24:38 +0000 (05:24 +0000)]
[X86] Fix weird identation. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254487 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoChange ModuleLinker to take a set of GlobalValues to import instead of a single one
Mehdi Amini [Wed, 2 Dec 2015 04:34:28 +0000 (04:34 +0000)]
Change ModuleLinker to take a set of GlobalValues to import instead of a single one

For efficiency reason, when importing multiple functions for the same Module,
we can avoid reparsing it every time.

Differential Revision: http://reviews.llvm.org/D15102

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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254486 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[libFuzzer] add a test that is built with -fsanitize-coverage=trace-bb
Kostya Serebryany [Wed, 2 Dec 2015 02:49:37 +0000 (02:49 +0000)]
[libFuzzer] add a test that is built with -fsanitize-coverage=trace-bb

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254484 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[sanitizer coverage] when adding a bb trace instrumentation, do it instead, not in...
Kostya Serebryany [Wed, 2 Dec 2015 02:37:13 +0000 (02:37 +0000)]
[sanitizer coverage] when adding a bb trace instrumentation, do it instead, not in addition to, regular coverage. Do the regular coverage in the run-time instead

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254482 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Fix a think-o when checking if the eflags needs to be preserved.
Quentin Colombet [Wed, 2 Dec 2015 02:07:00 +0000 (02:07 +0000)]
[X86] Fix a think-o when checking if the eflags needs to be preserved.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254480 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoModify FunctionImport to take a callback to load modules
Mehdi Amini [Wed, 2 Dec 2015 02:00:29 +0000 (02:00 +0000)]
Modify FunctionImport to take a callback to load modules

When linking static archive, there is no individual module files to
load. Instead they can be mmap'ed and could be initialized from a
buffer directly. The callback provide flexibility to override the
scheme for loading module from the summary.

Differential Revision: http://reviews.llvm.org/D15101

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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254479 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Make sure the prologue does not clobber EFLAGS when it lives accross it.
Quentin Colombet [Wed, 2 Dec 2015 01:22:54 +0000 (01:22 +0000)]
[X86] Make sure the prologue does not clobber EFLAGS when it lives accross it.

This is a superset of the fix done in r254448.

This fixes PR25607.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254478 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAArch64: fix 128-bit shifts
Tim Northover [Wed, 2 Dec 2015 00:33:54 +0000 (00:33 +0000)]
AArch64: fix 128-bit shifts

We mustn't introduce a shift of exactly 64-bits for any inputs, since that's an
UNDEF value (and worse, it's not what you want with the natural Arch64
implementation).

The generated code is pretty horrific, but I couldn't come up with an obviously
better alternative (if the amount is constant EXTR could help). Turns out
128-bit shifts are just nasty.

rdar://22491037

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254475 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoUse default member initializers.
Rafael Espindola [Tue, 1 Dec 2015 23:06:26 +0000 (23:06 +0000)]
Use default member initializers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254473 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoDefine member operator delete
Xinliang David Li [Tue, 1 Dec 2015 23:05:27 +0000 (23:05 +0000)]
Define member operator delete

For the struct with trailing objects, define
a member operator delete. Without this, the program
will fail when -fsized-deallocation option is used
where the wrong size will be passed to the global
delete operator.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254471 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAMDGPU: Error on addrspacecasts that aren't actually implemented
Matt Arsenault [Tue, 1 Dec 2015 23:04:05 +0000 (23:04 +0000)]
AMDGPU: Error on addrspacecasts that aren't actually implemented

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254469 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAMDGPU: Implement isNoopAddrSpaceCast
Matt Arsenault [Tue, 1 Dec 2015 23:04:00 +0000 (23:04 +0000)]
AMDGPU: Implement isNoopAddrSpaceCast

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254468 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoRemove unnecessary getter.
Rafael Espindola [Tue, 1 Dec 2015 23:01:51 +0000 (23:01 +0000)]
Remove unnecessary getter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254466 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoPass down the dst GV to linkGlobalValueBody. NFC.
Rafael Espindola [Tue, 1 Dec 2015 22:40:40 +0000 (22:40 +0000)]
Pass down the dst GV to linkGlobalValueBody. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254465 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix a bug in IfConversion.cpp.
Cong Hou [Tue, 1 Dec 2015 21:50:20 +0000 (21:50 +0000)]
Fix a bug in IfConversion.cpp.

The bug is introduced in r254377 which failed some tests on ARM, where a new
probability is assigned to a successor but the provided BB may not be a
successor.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254463 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoARM: Change ArchCheck field to uint64_t
Matthias Braun [Tue, 1 Dec 2015 21:48:52 +0000 (21:48 +0000)]
ARM: Change ArchCheck field to uint64_t

The values in this field are compared against getAvailableFeatures()
which returns an uint64_t. This was causing problems in an internal
branch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254462 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAMDGPU: Disallow flat_scr in SI assembler
Matt Arsenault [Tue, 1 Dec 2015 20:31:08 +0000 (20:31 +0000)]
AMDGPU: Disallow flat_scr in SI assembler

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254459 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[PGO] Add support for reading multiple versions of indexed profile format profile...
Xinliang David Li [Tue, 1 Dec 2015 20:26:26 +0000 (20:26 +0000)]
[PGO] Add support for reading multiple versions of indexed profile format profile data

Profile readers using incompatible on-disk hash table format can now share the same
implementation and interfaces.

Differential Revision: http://reviews.llvm.org/D15100

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254458 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoDelete unused includes.
Rafael Espindola [Tue, 1 Dec 2015 20:23:19 +0000 (20:23 +0000)]
Delete unused includes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254457 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoIR: Clean up some duplicated code in ConstantDataSequential creation. NFC
Justin Bogner [Tue, 1 Dec 2015 20:20:49 +0000 (20:20 +0000)]
IR: Clean up some duplicated code in ConstantDataSequential creation. NFC

ConstantDataArray::getImpl and ConstantDataVector::getImpl had a lot
of copy pasta in how they handled sequences of constants. Break that
out into a couple of simple functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254456 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoclang-format LinkModules.cpp.
Rafael Espindola [Tue, 1 Dec 2015 20:11:43 +0000 (20:11 +0000)]
clang-format LinkModules.cpp.

Most of the file has been changed recently and was already clang-format
clean.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254454 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agouse range-based for loops; NFCI
Sanjay Patel [Tue, 1 Dec 2015 19:57:43 +0000 (19:57 +0000)]
use range-based for loops; NFCI

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254453 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAMDGPU: Optimize VOP2 operand legalization
Matt Arsenault [Tue, 1 Dec 2015 19:57:17 +0000 (19:57 +0000)]
AMDGPU: Optimize VOP2 operand legalization

Don't use commuteInstruction, and don't commute if
doing so will not improve legality. Skip the more
complex checks for literal operands and constant bus restrictions,
which are not a concern for VOP2 instructions because src1
does not accept SGPRs or constants and few implicitly
read vcc.

This gets called quite a few times and the
attempts at commuting are a significant fraction
of the time spent in SIFixSGPRCopies, so it's
somewhat worthwhile to optimize. With this patch and others
leading up to it, this reduces the compile time of SIFixSGPRCopies
on some of the LuxMark 2 kernels from ~8ms to ~5ms on my system.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254452 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoUse references now that it is natural to do so.
Rafael Espindola [Tue, 1 Dec 2015 19:50:54 +0000 (19:50 +0000)]
Use references now that it is natural to do so.

The linker never takes ownership of a module or changes which module it
is refering to, making it natural to use references.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254449 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[X86] Make sure the prologue does not clobber EFLAGS when it lives accross it.
Quentin Colombet [Tue, 1 Dec 2015 19:49:31 +0000 (19:49 +0000)]
[X86] Make sure the prologue does not clobber EFLAGS when it lives accross it.

This fixes PR25629.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254448 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoUse nullptr (NFC)
Xinliang David Li [Tue, 1 Dec 2015 19:47:32 +0000 (19:47 +0000)]
Use nullptr (NFC)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254447 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agodon't repeat function/variable names in comments; NFC
Sanjay Patel [Tue, 1 Dec 2015 19:32:35 +0000 (19:32 +0000)]
don't repeat function/variable names in comments; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254445 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoFix Thumb1 epilogue generation
Artyom Skrobov [Tue, 1 Dec 2015 19:25:11 +0000 (19:25 +0000)]
Fix Thumb1 epilogue generation

Summary:
This had been broken for a very long time, but nobody noticed until
D14357 enabled shrink-wrapping by default.

Reviewers: jroelofs, qcolombet

Subscribers: tyomitch, llvm-commits, rengolin

Differential Revision: http://reviews.llvm.org/D14986

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254444 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agofix typo; NFC
Sanjay Patel [Tue, 1 Dec 2015 19:19:18 +0000 (19:19 +0000)]
fix typo; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254442 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[llvm-dwp] Deduplicate strings in the debug_str.dwo section
David Blaikie [Tue, 1 Dec 2015 19:17:58 +0000 (19:17 +0000)]
[llvm-dwp] Deduplicate strings in the debug_str.dwo section

Also, ensure that references to those strings in debug_str_offsets.dwo
correctly refer to the deduplicated strings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254441 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[AArch64] Fix a corner case in BitFeild select
Weiming Zhao [Tue, 1 Dec 2015 19:17:49 +0000 (19:17 +0000)]
[AArch64] Fix a corner case in BitFeild select

Summary:
When not useful bits, BitWidth becomes 0 and APInt will not be happy.

See https://llvm.org/bugs/show_bug.cgi?id=25571

We can just mark the operand as IMPLICIT_DEF is none bits of it is used.

Reviewers: t.p.northover, jmolloy

Subscribers: gberry, jmolloy, mgrang, aemerson, llvm-commits, rengolin

Differential Revision: http://reviews.llvm.org/D14803

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254440 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAMDGPU: Report extractelement as free in cost model
Matt Arsenault [Tue, 1 Dec 2015 19:08:39 +0000 (19:08 +0000)]
AMDGPU: Report extractelement as free in cost model

The cost for scalarized operations is computed as N * (scalar operation
cost + 1 extractelement + 1 insertelement). This partially fixes
inflating the cost of scalarized operations since every operation is
scalarized and free. I don't think we want any cost asociated with
scalarization, but for now insertelement is still counted. I'm not sure
if we should pretend that insertelement is also free, or add a way
to compute a custom scalarization cost.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254438 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[Verifier] Improve error for cross-module refs
Keno Fischer [Tue, 1 Dec 2015 19:06:36 +0000 (19:06 +0000)]
[Verifier] Improve error for cross-module refs

By including the module name in the error message.
This makes the error message much more useful and
saves a trip to the debugger.

Reviewers: dexonsmith

Subscribers: dexonsmith, llvm-commits

Differential Revision: http://reviews.llvm.org/D14473

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254437 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoDelete dead code.
Rafael Espindola [Tue, 1 Dec 2015 18:50:35 +0000 (18:50 +0000)]
Delete dead code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254436 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoUse a forwarding constructor instead of an init method.
Rafael Espindola [Tue, 1 Dec 2015 18:46:19 +0000 (18:46 +0000)]
Use a forwarding constructor instead of an init method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254435 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoDelete the setModule method from the Linker.
Rafael Espindola [Tue, 1 Dec 2015 18:41:30 +0000 (18:41 +0000)]
Delete the setModule method from the Linker.

It was only used from LTO for a debug feature, and LTO can just create
another linker.

It is pretty odd to have a method to reset the module in the middle of a
link. It would make IdentifiedStructTypes inconsistent with the Module
for example.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254434 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[llvm-dwp] Correctly update debug_str_offsets.dwo when linking dwo files
David Blaikie [Tue, 1 Dec 2015 18:07:07 +0000 (18:07 +0000)]
[llvm-dwp] Correctly update debug_str_offsets.dwo when linking dwo files

This doesn't deduplicate strings in the debug_str section, nor does it
properly wire up the index so that debug_info can /find/ these strings,
but it does correct the str_offsets specifically.

Follow up patches to address those related/next issues.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254431 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAMDGPU/SI: Remove REGISTER_STORE/REGISTER_LOAD code which is now dead
Tom Stellard [Tue, 1 Dec 2015 17:45:22 +0000 (17:45 +0000)]
AMDGPU/SI: Remove REGISTER_STORE/REGISTER_LOAD code which is now dead

Reviewers: arsenm

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D15050

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254427 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoAMDGPU: Use the default strings for data emission directives
Tom Stellard [Tue, 1 Dec 2015 17:45:17 +0000 (17:45 +0000)]
AMDGPU: Use the default strings for data emission directives

Summary:
This makes the assembly output look nicer and there is no reason to
have custom strings for these.

Reviewers: arsenm

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D14671

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254426 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[x86] add a convenience method to check for FMA capability; NFCI
Sanjay Patel [Tue, 1 Dec 2015 17:27:55 +0000 (17:27 +0000)]
[x86] add a convenience method to check for FMA capability; NFCI

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254425 91177308-0d34-0410-b5e6-96231b3b80d8

8 years agoMake appending var linking less of a special case.
Rafael Espindola [Tue, 1 Dec 2015 17:17:04 +0000 (17:17 +0000)]
Make appending var linking less of a special case.

It has to be a bit special because:
* materializeInitFor is not really supposed to call replaceAllUsesWith.
  The caller has a plain variable with Dst and expects just the
  initializer to be set, not for it to be removed.
* Calling mutateType as we used to do before gets some type
  inconsistency which breaks the bitcode writer.
* If linkAppendingVarProto create a dest decl with the correct type to
  avoid the above problems, it needs to put the original dst init in
  some side table for materializeInitFor to use.

In the end the simplest solution seems to be to just have
linkAppendingVarProto do all the work and set ValueMap[SrcGV to avoid
recursion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254424 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[ThinLTO] Wrap dbgs() output in DEBUG macro
Teresa Johnson [Tue, 1 Dec 2015 17:12:10 +0000 (17:12 +0000)]
[ThinLTO] Wrap dbgs() output in DEBUG macro

Missed in a couple places.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254422 91177308-0d34-0410-b5e6-96231b3b80d8

8 years ago[ThinLTO] Remove stale comment (NFC)
Teresa Johnson [Tue, 1 Dec 2015 16:45:23 +0000 (16:45 +0000)]
[ThinLTO] Remove stale comment (NFC)

Stale as of r254036 which added basic profitability check.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254421 91177308-0d34-0410-b5e6-96231b3b80d8