6 years agoFix a bug in MachineBlockPlacement that may cause assertion failure during BranchProb...
Cong Hou [Tue, 1 Dec 2015 00:55:42 +0000 (00:55 +0000)]
Fix a bug in MachineBlockPlacement that may cause assertion failure during BranchProbability construction.

The root cause is the rounding behavior in BranchProbability construction. We may consider to use truncation instead in the future.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254356 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-dwp] Initial partial prototype
David Blaikie [Tue, 1 Dec 2015 00:48:39 +0000 (00:48 +0000)]
[llvm-dwp] Initial partial prototype

This just concatenates the common DWP sections without doing any of the
fancy DWP things like:

1) update str_offsets
2) deduplicating strings
3) merging/creating cu/tu_index

Patches for these will follow shortly.

(also not sure about target triple/object file type for this tool - do I
really need a whole triple just to write an object file that contains
purely static/hardcoded bytes in each section? & I guess I should just
pick it based on the first input, maybe, rather than hardcoding for now
- but we only produce .dwo on ELF platforms with objcopy for now anyway)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254355 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agollvm-dwp: Initial layout
David Blaikie [Tue, 1 Dec 2015 00:48:34 +0000 (00:48 +0000)]
llvm-dwp: Initial layout

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254354 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[safestack] Protect byval function arguments.
Evgeniy Stepanov [Tue, 1 Dec 2015 00:40:05 +0000 (00:40 +0000)]
[safestack] Protect byval function arguments.

Detect unsafe byval function arguments and move them to the unsafe

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254353 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoExtend debug info for function parameters in SDAG.
Evgeniy Stepanov [Tue, 1 Dec 2015 00:34:30 +0000 (00:34 +0000)]
Extend debug info for function parameters in SDAG.

SDAG currently can emit debug location for function parameters when
an llvm.dbg.declare points to either a function argument SSA temp,
or to an AllocaInst. This change extends this logic by adding a
fallback case when neither of the above is true.

This is required for SafeStack, which may copy the contents of a
byval function argument into something that is not an alloca, and
then describe the target as the new location of the said argument.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254352 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[safestack] Fix handling of array allocas.
Evgeniy Stepanov [Tue, 1 Dec 2015 00:06:13 +0000 (00:06 +0000)]
[safestack] Fix handling of array allocas.

The current code does not take alloca array size into account and,
as a result, considers any access past the first array element to be

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254350 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoReplace all weight-based interfaces in MBB with probability-based interfaces, and...
Cong Hou [Tue, 1 Dec 2015 00:02:51 +0000 (00:02 +0000)]
Replace all weight-based interfaces in MBB with probability-based interfaces, and update all uses of old interfaces.

The patch in http://reviews.llvm.org/D13745 is broken into four parts:

1. New interfaces without functional changes (http://reviews.llvm.org/D13908).
2. Use new interfaces in SelectionDAG, while in other passes treat probabilities
as weights (http://reviews.llvm.org/D14361).
3. Use new interfaces in all other passes.
4. Remove old interfaces.

This patch is 3+4 above. In this patch, MBB won't provide weight-based
interfaces any more, which are totally replaced by probability-based ones.
The interface addSuccessor() is redesigned so that the default probability is
unknown. We allow unknown probabilities but don't allow using it together
with known probabilities in successor list. That is to say, we either have a
list of successors with all known probabilities, or all unknown
probabilities. In the latter case, we assume each successor has 1/N
probability where N is the number of successors. An assertion checks if the
user is attempting to add a successor with the disallowed mixed use as stated
above. This can help us catch many misuses.

All uses of weight-based interfaces are now updated to use probability-based

Differential revision: http://reviews.llvm.org/D14973

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254348 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoThis reverts commit r254336 and r254344.
Rafael Espindola [Mon, 30 Nov 2015 23:54:19 +0000 (23:54 +0000)]
This reverts commit r254336 and r254344.

They broke a bot and I am debugging why.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254347 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoDisable a consistency check.
Rafael Espindola [Mon, 30 Nov 2015 23:05:25 +0000 (23:05 +0000)]
Disable a consistency check.

Trying to figure out why it fails on a bot but passes locally.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254344 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[InstCombine] add tests to show potential vector IR shuffle transforms
Sanjay Patel [Mon, 30 Nov 2015 22:39:36 +0000 (22:39 +0000)]
[InstCombine] add tests to show potential vector IR shuffle transforms

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254342 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][FMA4] Prefer FMA4 to FMA
Simon Pilgrim [Mon, 30 Nov 2015 22:22:06 +0000 (22:22 +0000)]
[X86][FMA4] Prefer FMA4 to FMA

We currently output FMA instructions on targets which support both FMA4 + FMA (i.e. later Bulldozer CPUS bdver2/bdver3/bdver4).

This patch flips this so FMA4 is preferred; this is for several reasons:

1 - FMA4 is non-destructive reducing the need for mov instructions.
2 - Its more straighforward to commute and fold inputs (although the recent work on FMA has reduced this difference).
3 - All supported targets have FMA4 performance equal or better to FMA - Piledriver (bdver2) in particular has half the throughput when executing FMA instructions.

Its looks like no future AMD processor lines will support FMA4 after the Bulldozer series so we're not causing problems for later CPUs.

Differential Revision: http://reviews.llvm.org/D14997

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254339 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoStart deciding earlier what to link.
Rafael Espindola [Mon, 30 Nov 2015 22:01:43 +0000 (22:01 +0000)]
Start deciding earlier what to link.

A traditional linker is roughly split in symbol resolution and "copying

The two tasks are badly mixed in lib/Linker.

This starts splitting them apart.

With this patch there are no direct call to linkGlobalValueBody or
linkGlobalValueProto. Everything is linked via WapValue.

This also includes a few fixes:
* A GV goes undefined if the comdat is dropped (comdat11.ll).
* We error if an internal GV goes undefined (comdat13.ll).
* We don't link an unused comdat.

The first two match the behavior of an ELF linker. The second one is
equivalent to running globaldce on the input.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254336 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoHave 'optnone' respect the -fast-isel=false option.
Paul Robinson [Mon, 30 Nov 2015 21:56:16 +0000 (21:56 +0000)]
Have 'optnone' respect the -fast-isel=false option.

This is primarily useful for debugging optnone v. ISel issues.

Differential Revision: http://reviews.llvm.org/D14792

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254335 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Update test/CodeGen/X86/avg.ll with the help of update_llc_test_checks.py....
Cong Hou [Mon, 30 Nov 2015 21:46:08 +0000 (21:46 +0000)]
[X86] Update test/CodeGen/X86/avg.ll with the help of update_llc_test_checks.py. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254334 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Fix unused function
Matt Arsenault [Mon, 30 Nov 2015 21:32:10 +0000 (21:32 +0000)]
AMDGPU: Fix unused function

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254333 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Error if too many user SGPRs used
Matt Arsenault [Mon, 30 Nov 2015 21:16:07 +0000 (21:16 +0000)]
AMDGPU: Error if too many user SGPRs used

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254332 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Rework how private buffer passed for HSA
Matt Arsenault [Mon, 30 Nov 2015 21:16:03 +0000 (21:16 +0000)]
AMDGPU: Rework how private buffer passed for HSA

If we know we have stack objects, we reserve the registers
that the private buffer resource and wave offset are passed
and use them directly.

If not, reserve the last 5 SGPRs just in case we need to spill.
After register allocation, try to pick the next available registers
instead of the last SGPRs, and then insert copies from the inputs
to the reserved registers in the progloue.

This also only selectively enables all of the input registers
which are really required instead of always enabling them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254331 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Rename enums to be consistent with HSA code object terminology
Matt Arsenault [Mon, 30 Nov 2015 21:15:57 +0000 (21:15 +0000)]
AMDGPU: Rename enums to be consistent with HSA code object terminology

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254330 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Remove SIPrepareScratchRegs
Matt Arsenault [Mon, 30 Nov 2015 21:15:53 +0000 (21:15 +0000)]
AMDGPU: Remove SIPrepareScratchRegs

It does not work because of emergency stack slots.
This pass was supposed to eliminate dummy registers for the
spill instructions, but the register scavenger can introduce
more during PrologEpilogInserter, so some would end up
left behind if they were needed.

The potential for spilling the scratch resource descriptor
and offset register makes doing something like this
overly complicated. Reserve registers to use for the resource
descriptor and use them directly in eliminateFrameIndex.

Also removes creating another scratch resource descriptor
when directly selecting scratch MUBUF instructions.

The choice of which registers are reserved is temporary.
For now it attempts to pick the next available registers
after the user and system SGPRs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254329 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Use assert zext for workgroup sizes
Matt Arsenault [Mon, 30 Nov 2015 21:15:45 +0000 (21:15 +0000)]
AMDGPU: Use assert zext for workgroup sizes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254328 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ARM] For old thumb ISA like v4t, we cannot use PC directly in pop.
Quentin Colombet [Mon, 30 Nov 2015 20:37:58 +0000 (20:37 +0000)]
[ARM] For old thumb ISA like v4t, we cannot use PC directly in pop.

Fix the epilogue emission to account for that.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254325 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAvoid writing to source directory of tests
Reid Kleckner [Mon, 30 Nov 2015 20:36:23 +0000 (20:36 +0000)]
Avoid writing to source directory of tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254324 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SimplifyLibCalls] Remove useless bits of this tests.
Davide Italiano [Mon, 30 Nov 2015 19:38:35 +0000 (19:38 +0000)]
[SimplifyLibCalls] Remove useless bits of this tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254318 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SimplifyLibCalls] Transform log(exp2(y)) to y*log(2) under fast-math.
Davide Italiano [Mon, 30 Nov 2015 19:36:35 +0000 (19:36 +0000)]
[SimplifyLibCalls] Transform log(exp2(y)) to y*log(2) under fast-math.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254317 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add RIP to GR64_TCW64
David Majnemer [Mon, 30 Nov 2015 19:04:19 +0000 (19:04 +0000)]
[X86] Add RIP to GR64_TCW64

The MachineVerifier wants to check that the register operands of an
instruction belong to the instruction's register class.  RIP-relative
control flow instructions violated this by referencing RIP.  While this
was fixed for SysV, it was never fixed for Win64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254315 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoEnable shrink wrapping for PPC64
Kit Barton [Mon, 30 Nov 2015 18:59:41 +0000 (18:59 +0000)]
Enable shrink wrapping for PPC64

Re-enable shrink wrapping for PPC64 Little Endian.

One minor modification to PPCFrameLowering::findScratchRegister was necessary to handle fall-thru blocks (blocks with no terminator) correctly.

Tested with all LLVM test, clang tests, and the self-hosting build, with no problems found.

PHabricator: http://reviews.llvm.org/D14778

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254314 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix another llvm.ctors merging bug.
Rafael Espindola [Mon, 30 Nov 2015 18:54:24 +0000 (18:54 +0000)]
Fix another llvm.ctors merging bug.

We were not looking past casts to see if an element should be included
or not.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254313 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Fix a few minor compiler warnings. NFC.
Dan Gohman [Mon, 30 Nov 2015 18:42:08 +0000 (18:42 +0000)]
[WebAssembly] Fix a few minor compiler warnings. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254311 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agofix formatting; NFC
Sanjay Patel [Mon, 30 Nov 2015 17:52:02 +0000 (17:52 +0000)]
fix formatting; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254310 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] NFC Reordering headers.
Colin LeMahieu [Mon, 30 Nov 2015 17:32:34 +0000 (17:32 +0000)]
[Hexagon] NFC Reordering headers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254307 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Don't reserve SCRATCH_PTR input register
Matt Arsenault [Mon, 30 Nov 2015 15:46:47 +0000 (15:46 +0000)]
AMDGPU: Don't reserve SCRATCH_PTR input register

This hasn't been doing anything since using relocations was added.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254304 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoSilencing a 32-bit to 64-bit implicit conversion warning; NFC.
Aaron Ballman [Mon, 30 Nov 2015 14:52:33 +0000 (14:52 +0000)]
Silencing a 32-bit to 64-bit implicit conversion warning; NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254302 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips][microMIPS] Implement LBUX, LHX, LWX, MAQ_S[A].W.PHL, MAQ_S[A].W.PHR, MFHI...
Hrvoje Varga [Mon, 30 Nov 2015 12:58:39 +0000 (12:58 +0000)]
[mips][microMIPS] Implement LBUX, LHX, LWX, MAQ_S[A].W.PHL, MAQ_S[A].W.PHR, MFHI, MFLO, MTHI and MTLO instructions
Differential Revision: http://reviews.llvm.org/D14436

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254297 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips][microMIPS] Fix issue with offset operand of BALC and BC instructions
Zoran Jovanovic [Mon, 30 Nov 2015 12:56:18 +0000 (12:56 +0000)]
[mips][microMIPS] Fix issue with offset operand of BALC and BC instructions
Value of offset operand for microMIPS BALC and BC instructions is currently shifted 2 bits, but it should be 1 bit.
Differential Revision: http://reviews.llvm.org/D14770

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254296 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAVX512: regenerate avx512bw intrincics tests results.
Igor Breger [Mon, 30 Nov 2015 10:40:52 +0000 (10:40 +0000)]
AVX512: regenerate avx512bw intrincics tests results.

Differential Revision: http://reviews.llvm.org/D15069

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254295 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips][ias] Removed MSA instructions from base architecture valid-xfail.s's.
Daniel Sanders [Mon, 30 Nov 2015 09:52:00 +0000 (09:52 +0000)]
[mips][ias] Removed MSA instructions from base architecture valid-xfail.s's.

valid-xfail.s is for instructions that should be valid in the given ISA but
incorrectly fail. MSA instructions are correct to fail since MSA is not enabled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254293 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips][microMIPS] Implement PRECR.QB.PH, PRECR_SRA[_R].PH.W, PRECRQ.PH.W, PRECRQ...
Zlatko Buljan [Mon, 30 Nov 2015 08:37:38 +0000 (08:37 +0000)]
Differential Revision: http://reviews.llvm.org/D14605

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254291 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert r254279 "[X86] Use ArrayRef. NFC". It seems to have upset an MSVC build bot.
Craig Topper [Mon, 30 Nov 2015 02:28:19 +0000 (02:28 +0000)]
Revert r254279 "[X86] Use ArrayRef. NFC". It seems to have upset an MSVC build bot.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254280 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Use ArrayRef. NFC
Craig Topper [Mon, 30 Nov 2015 02:08:05 +0000 (02:08 +0000)]
[X86] Use ArrayRef. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254279 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ADT] Fix typo in comment
Sanjoy Das [Mon, 30 Nov 2015 01:24:17 +0000 (01:24 +0000)]
[ADT] Fix typo in comment

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254278 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AVX512] The vpermi2 instructions require an integer vector for the index vector...
Craig Topper [Mon, 30 Nov 2015 00:13:24 +0000 (00:13 +0000)]
[AVX512] The vpermi2 instructions require an integer vector for the index vector. This is reflected correctly in the intrinsics, but was not refelected in the isel patterns.

For the floating point types, this requires adding a bitcast to the index vector when its passed through to the output.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254277 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SCEV] Use lambda instead of std::bind; NFC
Sanjoy Das [Sun, 29 Nov 2015 23:40:57 +0000 (23:40 +0000)]
[SCEV] Use lambda instead of std::bind; NFC

The lambda is more readable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254276 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SCEV] Use range version of all_of; NFC
Sanjoy Das [Sun, 29 Nov 2015 23:40:53 +0000 (23:40 +0000)]
[SCEV] Use range version of all_of; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254275 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Remove duplicate entries from intrinsics tables and add asserts to verify there...
Craig Topper [Sun, 29 Nov 2015 23:18:32 +0000 (23:18 +0000)]
[X86] Remove duplicate entries from intrinsics tables and add asserts to verify there are no others.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254274 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix out of bounds access in hasStructRetAttr
Sanjoy Das [Sun, 29 Nov 2015 23:15:43 +0000 (23:15 +0000)]
Fix out of bounds access in hasStructRetAttr

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254273 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Delete an obsolete TODO comment.
Dan Gohman [Sun, 29 Nov 2015 23:09:41 +0000 (23:09 +0000)]
[WebAssembly] Delete an obsolete TODO comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254272 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Set several MCInstrDesc flags.
Dan Gohman [Sun, 29 Nov 2015 22:59:19 +0000 (22:59 +0000)]
[WebAssembly] Set several MCInstrDesc flags.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254271 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] int_x86_avx2_permps and X86ISD::VPERMV should take an integer vector for its...
Craig Topper [Sun, 29 Nov 2015 22:53:22 +0000 (22:53 +0000)]
[X86] int_x86_avx2_permps and X86ISD::VPERMV should take an integer vector for its shuffle indices.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254269 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Delete unused functions. NFC.
Dan Gohman [Sun, 29 Nov 2015 22:48:57 +0000 (22:48 +0000)]
[WebAssembly] Delete unused functions. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254268 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Minor clang-format and selected clang-tidy cleanups. NFC.
Dan Gohman [Sun, 29 Nov 2015 22:32:02 +0000 (22:32 +0000)]
[WebAssembly] Minor clang-format and selected clang-tidy cleanups. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254267 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agofix typos in comments; NFC
Sanjay Patel [Sun, 29 Nov 2015 22:09:34 +0000 (22:09 +0000)]
fix typos in comments; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254266 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SimplifyLibCalls] Don't crash if the function doesn't have a name.
Davide Italiano [Sun, 29 Nov 2015 21:58:56 +0000 (21:58 +0000)]
[SimplifyLibCalls] Don't crash if the function doesn't have a name.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254265 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SimplifyLibCalls] Cross out implemented transformations.
Davide Italiano [Sun, 29 Nov 2015 21:00:43 +0000 (21:00 +0000)]
[SimplifyLibCalls] Cross out implemented transformations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254264 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SimplifyLibCalls] Tranform log(pow(x, y)) -> y*log(x).
Davide Italiano [Sun, 29 Nov 2015 20:58:04 +0000 (20:58 +0000)]
[SimplifyLibCalls] Tranform log(pow(x, y)) -> y*log(x).

This one is enabled only under -ffast-math. There are cases where the
difference between the value computed and the correct value is huge
even for ffast-math, e.g. as Steven pointed out:

x = -1, y = -4
log(pow(-1), 4) = 0
4*log(-1) = NaN

I checked what GCC does and apparently they do the same optimization
(which result in the dramatic difference). Future work might try to
make this (slightly) less worse.

Differential Revision: http://reviews.llvm.org/D14400

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254263 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoSamplePGO - Do not use std::to_string in diagnostics.
Diego Novillo [Sun, 29 Nov 2015 18:23:26 +0000 (18:23 +0000)]
SamplePGO - Do not use std::to_string in diagnostics.

This fixes buildbots in systems that std::to_string is not present. It
also tidies the output of the diagnostic to render doubles a bit better
(thanks Ben Kramer for help with string streams and format).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254261 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoUse a lambda instead of std::bind and std::mem_fn I introduced in r254242. NFC
Craig Topper [Sun, 29 Nov 2015 18:05:22 +0000 (18:05 +0000)]
Use a lambda instead of std::bind and std::mem_fn I introduced in r254242. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254260 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][SSE] Added support for lowering to ADDSUBPS/ADDSUBPD with commuted inputs
Simon Pilgrim [Sun, 29 Nov 2015 16:41:04 +0000 (16:41 +0000)]
[X86][SSE] Added support for lowering to ADDSUBPS/ADDSUBPD with commuted inputs

We could already recognise shuffle(FSUB, FADD) -> ADDSUB, this allow us to recognise shuffle(FADD, FSUB) -> ADDSUB by commuting the shuffle mask prior to matching.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254259 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAdd a passing test.
Rafael Espindola [Sun, 29 Nov 2015 15:52:12 +0000 (15:52 +0000)]
Add a passing test.

When a comdat is discarded, any globals defined in it become undefined.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254258 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoDon't depend on the order the IR is copied.
Rafael Espindola [Sun, 29 Nov 2015 15:22:49 +0000 (15:22 +0000)]
Don't depend on the order the IR is copied.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254257 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoDon't depend on the order the IR is copied.
Rafael Espindola [Sun, 29 Nov 2015 15:08:39 +0000 (15:08 +0000)]
Don't depend on the order the IR is copied.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254256 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoMake this test less strict.
Rafael Espindola [Sun, 29 Nov 2015 14:53:06 +0000 (14:53 +0000)]
Make this test less strict.

We just want to test what is copied, no the order.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254255 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoSimplify. NFC.
Rafael Espindola [Sun, 29 Nov 2015 14:33:06 +0000 (14:33 +0000)]
Simplify. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254254 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAVX512:Implemented encoding for the vmovq.s instruction.
Igor Breger [Sun, 29 Nov 2015 07:41:26 +0000 (07:41 +0000)]
AVX512:Implemented encoding for the vmovq.s instruction.

Differential Revision: http://reviews.llvm.org/D14810

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254248 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRemove an intermediate lambda. NFC
Craig Topper [Sun, 29 Nov 2015 05:38:08 +0000 (05:38 +0000)]
Remove an intermediate lambda. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254246 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago Minor code cleanups
Xinliang David Li [Sun, 29 Nov 2015 04:52:34 +0000 (04:52 +0000)]
 Minor code cleanups

 - Add const keyword
 - fix code comments
 - move forward decl to the common file

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254244 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRemove unnecessary intermediate lambda. NFC
Craig Topper [Sun, 29 Nov 2015 04:37:14 +0000 (04:37 +0000)]
Remove unnecessary intermediate lambda. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254243 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SelectionDAG] Use std::any_of instead of a manually coded loop. NFC
Craig Topper [Sun, 29 Nov 2015 04:37:11 +0000 (04:37 +0000)]
[SelectionDAG] Use std::any_of instead of a manually coded loop. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254242 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoCorrectly handle llvm.global_ctors merging.
Rafael Espindola [Sun, 29 Nov 2015 03:29:42 +0000 (03:29 +0000)]
Correctly handle llvm.global_ctors merging.

We were not handling the case where an entry must be dropped and the
destination module has no llvm.global_ctors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254241 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix a crash when writing merged bitcode.
Rafael Espindola [Sun, 29 Nov 2015 03:21:30 +0000 (03:21 +0000)]
Fix a crash when writing merged bitcode.

Playing with mutateType in here was making getValueType and getType

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254240 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SimplifyLibCalls] Use any_of(). Suggested by David Blaikie!
Davide Italiano [Sat, 28 Nov 2015 22:27:48 +0000 (22:27 +0000)]
[SimplifyLibCalls] Use any_of(). Suggested by David Blaikie!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254239 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SimplifyLibCalls] Fix inverted condition that lead to an uninitialized memory read...
Benjamin Kramer [Sat, 28 Nov 2015 21:43:12 +0000 (21:43 +0000)]
[SimplifyLibCalls] Fix inverted condition that lead to an uninitialized memory read below.

Found by msan!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254238 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][AVX] Regenerate ADDSUB tests
Simon Pilgrim [Sat, 28 Nov 2015 19:20:49 +0000 (19:20 +0000)]
[X86][AVX] Regenerate ADDSUB tests

Tidied up triple and regenerate tests using update_llc_test_checks.py

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254237 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[PGO] Move value profile format related structures and APIs to common file
Xinliang David Li [Sat, 28 Nov 2015 19:07:09 +0000 (19:07 +0000)]
[PGO] Move value profile format related structures and APIs to common file

This is the last step to enable profile runtime to share the same value prof
data format and reader/writer code with llvm host tools. The VP related
data structures are moved to a section in InstrProfData.inc enabled with macro
INSTR_PROF_VALUE_PROF_DATA, and common API implementations are enabled with
INSTR_PROF_COMMON_API_IMPL. There should be no functional change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254235 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "[ARM] Generate ABI_optimization_goals build attribute, as described in the...
Renato Golin [Sat, 28 Nov 2015 17:23:46 +0000 (17:23 +0000)]
Revert "[ARM] Generate ABI_optimization_goals build attribute, as described in the ARM ARM."

This reverts commit r254201 and r254202, as it broke test-suite,
self-hosting and sanitizer tests on ARM buildbots.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254234 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][FMA] Added 512-bit tests to match 128/256-bit tests coverage
Simon Pilgrim [Sat, 28 Nov 2015 16:04:24 +0000 (16:04 +0000)]
[X86][FMA] Added 512-bit tests to match 128/256-bit tests coverage

As discussed on D14909

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254233 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][FMA] More thorough FMA tests
Simon Pilgrim [Sat, 28 Nov 2015 14:28:44 +0000 (14:28 +0000)]
[X86][FMA] More thorough FMA tests

Added FMADD/FMSUB/FNMADD/FNMSUB tests for all types

Added load folding tests for 512-bit vectors

NOTE: Many of the AVX512 FMA instructions don't yet commute/fold correctly

As discussed on D14909

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254232 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][AVX2] Tidied up PBROADCAST tests
Simon Pilgrim [Sat, 28 Nov 2015 14:15:40 +0000 (14:15 +0000)]
[X86][AVX2] Tidied up PBROADCAST tests

Tidied up triple and regenerate tests using update_llc_test_checks.py

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254231 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agollvm/test/CodeGen/SystemZ/alloca-04.ll REQUIRES asserts due to -debug-pass.
NAKAMURA Takumi [Sat, 28 Nov 2015 13:05:49 +0000 (13:05 +0000)]
llvm/test/CodeGen/SystemZ/alloca-04.ll REQUIRES asserts due to -debug-pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254230 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Stack realignment] Handling of aligned allocas.
Jonas Paulsson [Sat, 28 Nov 2015 11:02:32 +0000 (11:02 +0000)]
[Stack realignment] Handling of aligned allocas.

This patch implements dynamic realignment of stack objects for targets
with a non-realigned stack pointer. Behaviour in FunctionLoweringInfo
is changed so that for a target that has StackRealignable set to
false, over-aligned static allocas are considered to be variable-sized
objects and are handled with DYNAMIC_STACKALLOC nodes.

It would be good to group aligned allocas into a single big alloca as
an optimization, but this is yet todo.

SystemZ benefits from this, due to its stack frame layout.

New tests SystemZ/alloca-03.ll for aligned allocas, and
SystemZ/alloca-04.ll for "no-realign-stack" attribute on functions.

Review and help from Ulrich Weigand and Hal Finkel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254227 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoUse range-based for loops. NFC
Craig Topper [Sat, 28 Nov 2015 08:23:04 +0000 (08:23 +0000)]
Use range-based for loops. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254222 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[TableGen] Use SmallString instead of std::string to build up a string to avoid heap...
Craig Topper [Sat, 28 Nov 2015 08:23:02 +0000 (08:23 +0000)]
[TableGen] Use SmallString instead of std::string to build up a string to avoid heap allocations. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254221 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[PGO] Add return code for vp rt record init routine to indicate error condition
Xinliang David Li [Sat, 28 Nov 2015 05:47:34 +0000 (05:47 +0000)]
[PGO] Add return code for vp rt record init routine to indicate error condition

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254220 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[PGO] Allow value profile writer interface to allocated target buffer
Xinliang David Li [Sat, 28 Nov 2015 05:37:01 +0000 (05:37 +0000)]
[PGO] Allow value profile writer interface to allocated target buffer

Raw profile writer needs to write all data of one kind in one continuous block,
so the buffer needs to be pre-allocated and passed to the writer method in
pieces for function profile data. The change adds the support for raw value data

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254219 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFunction name cleanup (NFC)
Xinliang David Li [Sat, 28 Nov 2015 05:06:00 +0000 (05:06 +0000)]
Function name cleanup (NFC)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254218 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[PGO] Extract VP data integrity check code into a helper function (NFC)
Xinliang David Li [Sat, 28 Nov 2015 04:56:07 +0000 (04:56 +0000)]
[PGO] Extract VP data integrity check code into a helper function (NFC)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254217 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[autoconf] Fix MinGW build
Keno Fischer [Sat, 28 Nov 2015 00:54:12 +0000 (00:54 +0000)]
[autoconf] Fix MinGW build

This is the autoconf analog of r251201. I realize autoconf is
deprecated, but while it's in tree, it should at least be kept working.

Also add the deprecation message to configure.ac such that AutoRegen
actually picks ip up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254215 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoPass .ll directly to llvm-link.
Rafael Espindola [Fri, 27 Nov 2015 23:47:15 +0000 (23:47 +0000)]
Pass .ll directly to llvm-link.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254214 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoPass .ll directly to llvm-link
Rafael Espindola [Fri, 27 Nov 2015 23:21:45 +0000 (23:21 +0000)]
Pass .ll directly to llvm-link

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254213 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoSamplePGO - Add initial support for inliner annotations.
Diego Novillo [Fri, 27 Nov 2015 23:14:51 +0000 (23:14 +0000)]
SamplePGO - Add initial support for inliner annotations.

This adds two thresholds to the sample profiler to affect inlining
decisions: the concept of global hotness and coldness.

Functions that have accumulated more than a certain fraction of samples at
runtime, are annotated with the InlineHint attribute. Conversely,
functions that accumulate less than a certain fraction of samples, are
annotated with the Cold attribute.

This is very similar to the hints emitted by Clang when using
instrumentation profiles.

Notice that this is a very blunt instrument. A function may have
globally collected a significant fraction of samples, but that does not
necessarily mean that every callsite for that function is hot.

Ideally, we would annotate each callsite with the samples collected at
that callsite. This way, the inliner can incorporate all these weights
into its cost model.

Once the inliner offers this functionality, we can change the hints
emitted here to a more precise per-callsite annotation. For now, this is
providing some measure of speedups with our internal benchmarks. I've
observed speedups of up to 23% (though the geo mean is about 3%). I expect
these numbers to improve as the inliner gets better annotations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254212 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoSamplePGO - Fix default threshold for hot callsites.
Diego Novillo [Fri, 27 Nov 2015 23:14:49 +0000 (23:14 +0000)]
SamplePGO - Fix default threshold for hot callsites.

Based on testing of internal benchmarks, I'm lowering this threshold to
a value of 0.1%.  This means that SamplePGO will respect 99.9% of the
original inline decisions when following a profile.

The performance difference is noticeable in some tests. With the
previous threshold, the speedups over baseline -O2 was about 0.63%. With
the new default, the speedups are around 3% on average.

The point of this threshold is not to do more aggressive inlining. When
an inlined callsite crosses this threshold, SamplePGO will redo the
inline decision so that it can better apply the input profile.

By respecting most original inline decisions, we can apply more of the
input profile because the shape of the code follows the profile more

In the next series, I'll be looking at adding some inline hints for the
cold callsites and for toplevel functions that are hot/cold as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254211 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoModernize the test a bit
Rafael Espindola [Fri, 27 Nov 2015 23:13:17 +0000 (23:13 +0000)]
Modernize the test a bit

Remove out of date comment.
Pass .ll files to llvm-link.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254210 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoSimplify the linking of recursive data.
Rafael Espindola [Fri, 27 Nov 2015 20:28:19 +0000 (20:28 +0000)]
Simplify the linking of recursive data.

Now the ValueMapper has two callbacks. The first one maps the
declaration. The ValueMapper records the mapping and then materializes
the body/initializer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254209 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFollow-up fix for r254201
Artyom Skrobov [Fri, 27 Nov 2015 16:20:34 +0000 (16:20 +0000)]
Follow-up fix for r254201

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254202 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ARM] Generate ABI_optimization_goals build attribute, as described in the ARM ARM.
Artyom Skrobov [Fri, 27 Nov 2015 15:30:51 +0000 (15:30 +0000)]
[ARM] Generate ABI_optimization_goals build attribute, as described in the ARM ARM.

Since this build attribute corresponds to a whole module, and
different functions in a module may differ in the optimizations
enabled for them, this attribute is emitted after all functions,
and only in the case that the optimization goals for all
functions match.

Reviewers: logan, hans

Subscribers: aemerson, rengolin, llvm-commits

Differential Revision: http://reviews.llvm.org/D14934

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254201 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64] Add ARMv8.2-A FP16 scalar instructions
Oliver Stannard [Fri, 27 Nov 2015 13:04:48 +0000 (13:04 +0000)]
[AArch64] Add ARMv8.2-A FP16 scalar instructions

ARMv8.2-A adds 16-bit floating point versions of all existing VFP
floating-point instructions. This is an optional extension, so all of
these instructions require the FeatureFullFP16 subtarget feature.

Most of these instructions are the same as the 32- and 64-bit versions,
but with the type field (bits 23-22) set to 0b11. Previously the top bit
of the size field was always 0, so the instruction classes only provided
a 1-bit size field, which I have widened to 2 bits.

Differential Revision: http://reviews.llvm.org/D15014

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254198 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[sanitizer] [dfsan] Unify aarch64 mapping
Adhemerval Zanella [Fri, 27 Nov 2015 12:42:39 +0000 (12:42 +0000)]
[sanitizer] [dfsan] Unify aarch64 mapping

This patch changes the DFSan instrumentation for aarch64 to instead
of using fixes application mask defined by SANITIZER_AARCH64_VMA
to read the application shadow mask value from compiler-rt. The value
is initialized based on runtime VAM detection.

Along with this patch a compiler-rt one will also be added to export
the shadow mask variable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254196 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SimplifyLibCalls] Use range-based loop. NFC.
Davide Italiano [Fri, 27 Nov 2015 08:05:40 +0000 (08:05 +0000)]
[SimplifyLibCalls] Use range-based loop. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254193 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[TableGen] Sort pattern predicates before concatenating into a string so that differe...
Craig Topper [Fri, 27 Nov 2015 05:44:04 +0000 (05:44 +0000)]
[TableGen] Sort pattern predicates before concatenating into a string so that different orders of the same set will produce the same string. This can reduce the number of unique predicates in the isel tables. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254192 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Pair a NoVLX with HasAVX512 to match the others and remove a unique predicate...
Craig Topper [Fri, 27 Nov 2015 05:44:02 +0000 (05:44 +0000)]
[X86] Pair a NoVLX with HasAVX512 to match the others and remove a unique predicate check in the isel tables. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254191 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agotest: bail early if tool_path is None
Andrew Wilkins [Fri, 27 Nov 2015 05:07:26 +0000 (05:07 +0000)]
test: bail early if tool_path is None

tool_path will be None for llvm-go if Go cannot be found

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254190 91177308-0d34-0410-b5e6-96231b3b80d8