OSDN Git Service

android-x86/external-llvm.git
6 years agofix 80 column violation.
Adrian Prantl [Fri, 29 Sep 2017 22:46:22 +0000 (22:46 +0000)]
fix 80 column violation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314564 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert 314561 due to debug build assertion failure
Xinliang David Li [Fri, 29 Sep 2017 22:30:34 +0000 (22:30 +0000)]
Revert 314561 due to debug build assertion failure

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314563 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-rc] Serialize MENU resources to .res files (serialization, pt 3).
Marek Sokolowski [Fri, 29 Sep 2017 22:25:05 +0000 (22:25 +0000)]
[llvm-rc] Serialize MENU resources to .res files (serialization, pt 3).

This allows MENU resources to be serialized.

MENU resource statement doc:
msdn.microsoft.com/en-us/library/windows/desktop/aa381025.aspx
POPUP sub-statement doc:
msdn.microsoft.com/en-us/library/windows/desktop/aa381030.aspx
MENUITEM sub-statement doc:
msdn.microsoft.com/en-us/library/windows/desktop/aa381024.aspx
MENUHEADER structure:
msdn.microsoft.com/en-us/library/windows/desktop/ms648018.aspx (and
NORMALMENUITEM, POPUPMENUITEM structs).

Thanks for Nico Weber for his original work in this area.

Differential Revision: https://reviews.llvm.org/D37828

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314562 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoEliminate PHI (int typed) which has only one use by intptr
Xinliang David Li [Fri, 29 Sep 2017 22:10:15 +0000 (22:10 +0000)]
Eliminate PHI (int typed) which has only one use by intptr

This patch will eliminate redundant intptr/ptrtoint that pessimizes
analyses such as SCEV, AA and will make optimization passes such
as auto-vectorization more powerful.

Differential revision: http://reviews.llvm.org/D37832

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314561 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "Use the basic cost if a GEP is not used as addressing mode"
Alex Shlyapnikov [Fri, 29 Sep 2017 22:04:45 +0000 (22:04 +0000)]
Revert "Use the basic cost if a GEP is not used as addressing mode"

This reverts commit r314517.

This commit crashes sanitizer bots, for example:
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux/builds/4167

Stack snippet:
...
/mnt/b/sanitizer-buildbot1/sanitizer-x86_64-linux/build/llvm/include/llvm/Support/Casting.h:255:0
llvm::TargetTransformInfoImplCRTPBase<llvm::X86TTIImpl>::getGEPCost(llvm::GEPOperator const*, llvm::ArrayRef<llvm::Value const*>)
/mnt/b/sanitizer-buildbot1/sanitizer-x86_64-linux/build/llvm/include/llvm/Analysis/TargetTransformInfoImpl.h:742:0
llvm::TargetTransformInfoImplCRTPBase<llvm::X86TTIImpl>::getUserCost(llvm::User const*, llvm::ArrayRef<llvm::Value const*>)
/mnt/b/sanitizer-buildbot1/sanitizer-x86_64-linux/build/llvm/include/llvm/Analysis/TargetTransformInfoImpl.h:782:0
/mnt/b/sanitizer-buildbot1/sanitizer-x86_64-linux/build/llvm/lib/Analysis/TargetTransformInfo.cpp:116:0
/mnt/b/sanitizer-buildbot1/sanitizer-x86_64-linux/build/llvm/include/llvm/ADT/SmallVector.h:116:0
/mnt/b/sanitizer-buildbot1/sanitizer-x86_64-linux/build/llvm/include/llvm/ADT/SmallVector.h:343:0
/mnt/b/sanitizer-buildbot1/sanitizer-x86_64-linux/build/llvm/include/llvm/ADT/SmallVector.h:864:0
/mnt/b/sanitizer-buildbot1/sanitizer-x86_64-linux/build/llvm/include/llvm/Analysis/TargetTransformInfo.h:285:0
...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314560 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[CodeGen] Fix some Clang-tidy modernize-use-using and Include What You Use warnings...
Eugene Zelenko [Fri, 29 Sep 2017 21:55:49 +0000 (21:55 +0000)]
[CodeGen] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314559 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "[CMake] Remove `CMAKE_.*_OUTPUT_DIRECTORY` (NFCI)"
Brian Gesiak [Fri, 29 Sep 2017 19:50:41 +0000 (19:50 +0000)]
Revert "[CMake] Remove `CMAKE_.*_OUTPUT_DIRECTORY` (NFCI)"

Summary:
It appears polly makes use of the `CMAKE_RUNTIME_OUTPUT_DIRECTORY` variable
when configuring its lit test suite. Reverting this for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314551 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[CMake] Remove `CMAKE_.*_OUTPUT_DIRECTORY` (NFCI)
Brian Gesiak [Fri, 29 Sep 2017 19:34:57 +0000 (19:34 +0000)]
[CMake] Remove `CMAKE_.*_OUTPUT_DIRECTORY` (NFCI)

Summary:
Three `CMAKE_.*_OUTPUT_DIRECTORY` variables used to be set in CMake and
referenced in various other parts of the project. However, in r198205
chapuni added a note to "don't set them anymore", and any remaining
references to them were subsequently removed in r198316 and r199592.

Now that the variables are no longer used anywhere, remove them, along
with the comments advising against using them any longer.

Test Plan:
I ran `check-all` and confirmed the tests built and passed.

Reviewers: beanz, chapuni

Reviewed By: beanz

Subscribers: mgorny

Differential Revision: https://reviews.llvm.org/D38389

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314550 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-rc] Serialize ACCELERATORS to .res files (serialization, pt 2).
Marek Sokolowski [Fri, 29 Sep 2017 19:07:44 +0000 (19:07 +0000)]
[llvm-rc] Serialize ACCELERATORS to .res files (serialization, pt 2).

This allows llvm-rc to serialize ACCELERATORS resources.

Additionally, as this is the first type of resource to support basic
optional resource statements (LANGUAGE, CHARACTERISTICS, VERSION),

ACCELERATORS statement documentation:
msdn.microsoft.com/en-us/library/windows/desktop/aa380610.aspx
Accelerator table structure documentation:
msdn.microsoft.com/en-us/library/windows/desktop/ms648010.aspx
Optional resource statement fields are described in:
msdn.microsoft.com/en-us/library/windows/desktop/ms648027.aspx

Thanks for Nico Weber for his original work in this area.

Differential Revision: https://reviews.llvm.org/D37824

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314549 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[LV] Use correct insertion point when type shrinking reductions
Matthew Simpson [Fri, 29 Sep 2017 18:07:39 +0000 (18:07 +0000)]
[LV] Use correct insertion point when type shrinking reductions

When type shrinking reductions, we should insert the truncations and extends at
the end of the loop latch block. Previously, these instructions were inserted
at the end of the loop header block. The difference is only a problem for loops
with predicated instructions (e.g., conditional stores and instructions that
may divide by zero). For these instructions, we create new basic blocks inside
the vectorized loop, which cause the loop header and latch to no longer be the
same block. This should fix PR34687.

Reference: https://bugs.llvm.org/show_bug.cgi?id=34687

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314542 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-rc] Refactoring needed for ACCELERATORS and MENU resources.
Marek Sokolowski [Fri, 29 Sep 2017 17:46:32 +0000 (17:46 +0000)]
[llvm-rc] Refactoring needed for ACCELERATORS and MENU resources.

This is a part of llvm-rc serialization patch set (serialization, pt 1.5).

This:

* Unifies the internal representation of flags in ACCELERATORS and MENU
   with the corresponding representation in .res files (noticed in
   https://reviews.llvm.org/D37828#inline-329828).
* Creates an RCResource subclass, OptStatementsRCResource, describing
   resource statements that can declare resource-local optional statements
   (proposed in https://reviews.llvm.org/D37824#inline-329775).

These modifications don't fit to any of the current patches, so I'm
submitting them as a separate patch.

Differential Revision: https://reviews.llvm.org/D37841

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314541 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoUse LLVM_ENABLE_ABI_BREAKING_CHECKS correctly
Sanjoy Das [Fri, 29 Sep 2017 17:17:54 +0000 (17:17 +0000)]
Use LLVM_ENABLE_ABI_BREAKING_CHECKS correctly

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314539 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-rc] Serialize HTML resources to .res files (serialization, pt 1).
Marek Sokolowski [Fri, 29 Sep 2017 17:14:09 +0000 (17:14 +0000)]
[llvm-rc] Serialize HTML resources to .res files (serialization, pt 1).

This allows to process HTML resources defined in .rc scripts and output
them to resulting .res files. Additionally, some infrastructure allowing
to output these files is created.

This is the first resource type we can operate on.

Thanks to Nico Weber for his original work in this area.

Differential Revision: reviews.llvm.org/D37283

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314538 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoDisplay relative hotness with two decimal digits after the decimal point
Adam Nemet [Fri, 29 Sep 2017 16:56:54 +0000 (16:56 +0000)]
Display relative hotness with two decimal digits after the decimal point

I've seen cases where tiny inlined functions have such a high execution count
that most everything would show up with a relative of hotness of 0%.  Since
the inlined functions effectively disappear you need to tune in the lower
range, thus we need more precision.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314537 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix Wmismatched-tags warning.
Simon Pilgrim [Fri, 29 Sep 2017 16:52:27 +0000 (16:52 +0000)]
Fix Wmismatched-tags warning.

InlineAsmIdentifierInfo was declared a class in some places and a struct in others.

Partial reversion of rL314508

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314536 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[test] Enable LeakSanitizer on 64-bit Darwin ASan llvm builds
Francis Ricci [Fri, 29 Sep 2017 16:51:50 +0000 (16:51 +0000)]
[test] Enable LeakSanitizer on 64-bit Darwin ASan llvm builds

Summary:
Also disables leak checking on lto tests, due to many leaks reported
in the system's ld64.

Reviewers: kcc, pcc, bogner, kubamracek

Subscribers: mehdi_amini, llvm-commits

Differential Revision: https://reviews.llvm.org/D37781

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314535 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Allow each data segment to specify its own alignment
Sam Clegg [Fri, 29 Sep 2017 16:50:08 +0000 (16:50 +0000)]
[WebAssembly] Allow each data segment to specify its own alignment

Also, add a flags field as we will almost certainly
be needing that soon too.

Differential Revision: https://reviews.llvm.org/D38296

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314534 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SimplifyIndVar] Do not fail when we constant fold an IV user to ConstantPointerNull
Hongbin Zheng [Fri, 29 Sep 2017 16:32:12 +0000 (16:32 +0000)]
[SimplifyIndVar] Do not fail when we constant fold an IV user to ConstantPointerNull

The type of a SCEVConstant may not match the corresponding LLVM Value.
In this case, we skip the constant folding for now.

TODO: Replace ConstantInt Zero by ConstantPointerNull

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314531 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: fix bad test exposed by r314522
Nicolai Haehnle [Fri, 29 Sep 2017 16:07:05 +0000 (16:07 +0000)]
AMDGPU: fix bad test exposed by r314522

The test attempts to use -1 as carry-in for v_addc_*.

Before writing r314522, I did actually test this on real hardware,
and found that it doesn't work. So r314522 is correct in restricting
the carry-in operand: just remove those tests to make things pass
again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314530 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ThinLTO] Use decimal suffix for promoted values to match demanglers
Teresa Johnson [Fri, 29 Sep 2017 15:55:42 +0000 (15:55 +0000)]
[ThinLTO] Use decimal suffix for promoted values to match demanglers

Summary:
Demanglers such as libiberty know how to strip suffixes of the form
\.[a-zA-Z]+\.\d+, but our current promoted value suffixes are
.llvm.${modulehash}, where the module hash is in hex. Change the
module hash to decimal to allow demanglers to handle this.

Reviewers: danielcdh

Subscribers: llvm-commits, inglorion

Differential Revision: https://reviews.llvm.org/D38405

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314527 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[dwarfdump][NFC] Consistent printing of address ranges
Jonas Devlieghere [Fri, 29 Sep 2017 15:41:22 +0000 (15:41 +0000)]
[dwarfdump][NFC] Consistent printing of address ranges

This implement the insertion operator for DWARF address ranges so they
are consistently printed as [LowPC, HighPC).

While a dump method might have felt more consistent, it is used
exclusively for printing error messages in the verifier and never used
for actual dumping. Hence this approach is more intuitive and creates
less clutter at the call sites.

Differential revision: https://reviews.llvm.org/D38395

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314523 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: VALU carry-in and v_cndmask condition cannot be EXEC
Nicolai Haehnle [Fri, 29 Sep 2017 15:37:31 +0000 (15:37 +0000)]
AMDGPU: VALU carry-in and v_cndmask condition cannot be EXEC

The hardware will only forward EXEC_LO; the high 32 bits will be zero.

Additionally, inline constants do not work. At least,

   v_addc_u32_e64 v0, vcc, v0, v1, -1

which could conceivably be used to combine (v0 + v1 + 1) into a single
instruction, acts as if all carry-in bits are zero.

The llvm.amdgcn.ps.live test is adjusted; it would be nice to combine

   s_mov_b64 s[0:1], exec
   v_cndmask_b32_e64 v0, v1, v2, s[0:1]

into

   v_mov_b32 v0, v3

but it's not particularly high priority.

Fixes dEQP-GLES31.functional.shaders.helper_invocation.value.*

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314522 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoUse the basic cost if a GEP is not used as addressing mode
Jun Bum Lim [Fri, 29 Sep 2017 14:50:16 +0000 (14:50 +0000)]
Use the basic cost if a GEP is not used as addressing mode

Summary:
Currently, getGEPCost() returns TCC_FREE whenever a GEP is a legal addressing mode in the target.
However, since it doesn't check its actual users, it will return FREE even in cases
where the GEP cannot be folded away as a part of actual addressing mode.
For example, if an user of the GEP is a call instruction taking the GEP as a parameter,
then the GEP may not be folded in isel.

Reviewers: hfinkel, efriedma, mcrosier, jingyue, haicheng

Reviewed By: hfinkel

Subscribers: javed.absar, llvm-commits

Differential Revision: https://reviews.llvm.org/D38085

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314517 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SystemZ] implement shouldCoalesce()
Jonas Paulsson [Fri, 29 Sep 2017 14:31:39 +0000 (14:31 +0000)]
[SystemZ]  implement shouldCoalesce()

Implement shouldCoalesce() to help regalloc avoid running out of GR128
registers.

If a COPY involving a subreg of a GR128 is coalesced, the live range of the
GR128 virtual register will be extended. If this happens where there are
enough phys-reg clobbers present, regalloc will run out of registers (if
there is not a single GR128 allocatable register available).

This patch tries to allow coalescing only when it can prove that this will be
safe by checking the (local) interval in question.

Review: Ulrich Weigand, Quentin Colombet
https://reviews.llvm.org/D37899
https://bugs.llvm.org/show_bug.cgi?id=34610

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314516 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix spelling in comments. NFCI.
Simon Pilgrim [Fri, 29 Sep 2017 14:13:47 +0000 (14:13 +0000)]
Fix spelling in comments. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314515 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Improve codegen for inverted overflow checking intrinsics.
Amara Emerson [Fri, 29 Sep 2017 13:53:44 +0000 (13:53 +0000)]
[X86] Improve codegen for inverted overflow checking intrinsics.

Adds a new combine for: xor(setcc cc, val), 1 --> setcc (invert(cc), val)

Differential Revision: https://reviews.llvm.org/D38161

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314514 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ARM] v8.3-a complex number support
Sam Parker [Fri, 29 Sep 2017 13:11:33 +0000 (13:11 +0000)]
[ARM] v8.3-a complex number support

New instructions are added to AArch32 and AArch64 to aid
floating-point multiplication and addition of complex numbers, where
the complex numbers are packed in a vector register as a pair of
elements. The Imaginary part of the number is placed in the more
significant element, and the Real part of the number is placed in the
less significant element.

This patch adds assembler for the ARM target.

Differential Revision: https://reviews.llvm.org/D36789

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314511 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoSmall modification <NFC>
Michael Zuckerman [Fri, 29 Sep 2017 12:45:54 +0000 (12:45 +0000)]
Small modification <NFC>

Change-Id: I360abccee12cae29bd2ac4f8399c9ecc92eb7f13

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314510 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix Wmismatched-tags warning.
Simon Pilgrim [Fri, 29 Sep 2017 11:42:05 +0000 (11:42 +0000)]
Fix Wmismatched-tags warning.

InlineAsmIdentifierInfo was declared a class in some places and a class in others.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314508 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips] Reordering callseq* nodes to be linear
Aleksandar Beserminji [Fri, 29 Sep 2017 11:05:02 +0000 (11:05 +0000)]
[mips] Reordering callseq* nodes to be linear

Fix nested callseq* nodes by moving callseq_start after the
arguments calculation to temporary registers, so that callseq* nodes
in resulting DAG are linear.

Recommitting r314497. This version does not contain test which fails
when compiler is not build in debug mode.

Differential Revision: https://reviews.llvm.org/D37328

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314507 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "[mips] Reordering callseq* nodes to be linear"
Aleksandar Beserminji [Fri, 29 Sep 2017 10:52:03 +0000 (10:52 +0000)]
Revert "[mips] Reordering callseq* nodes to be linear"

Added test relies on the compiler being built in debug mode,
which may not be the case.

This reverts commit r314497.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314506 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips] Add missing license info, formatting changes. NFCI
Simon Dardis [Fri, 29 Sep 2017 10:08:06 +0000 (10:08 +0000)]
[mips] Add missing license info, formatting changes. NFCI

Add missing license information to MicroMipsInstrFPU.td and
fix most of the formatting errors present. Others will be
addressed in a follow up commits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314505 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][SSE] Added more tests for vector multiplications as utility for D37896
Simon Pilgrim [Fri, 29 Sep 2017 10:02:01 +0000 (10:02 +0000)]
[X86][SSE] Added more tests for vector multiplications as utility for D37896

Added additional tests for vector multiplications with multipliers that are:
 * powers of 2 displaced by 1,
 * product of a power of 2 displaced by one with another power of 2.

Patch by @pacxx (Michael Haidl)

Differential Revision: https://reviews.llvm.org/D38350

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314504 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips] Add test cases for dext/dins family of instructions
Aleksandar Beserminji [Fri, 29 Sep 2017 09:53:24 +0000 (09:53 +0000)]
[mips] Add test cases for dext/dins family of instructions

Add missing test cases for dext, dextm, dextu, dins, dinsm and
dinsu instructions.

Differential Revision: https://reviews.llvm.org/D37741

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314503 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AMDGPU] calling conventions for AMDPAL OS type
Tim Renouf [Fri, 29 Sep 2017 09:51:22 +0000 (09:51 +0000)]
[AMDGPU] calling conventions for AMDPAL OS type

Summary:
This commit adds comments on how the AMDPAL OS type overloads the
existing AMDGPU_ calling conventions used by Mesa, and adds a couple of
new ones.

Reviewers: arsenm, nhaehnle, dstuttard

Subscribers: mehdi_amini, kzhuravl, wdng, yaxunl, t-tye, llvm-commits

Differential Revision: https://reviews.llvm.org/D37752

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314502 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AMDGPU] AMDPAL scratch buffer support
Tim Renouf [Fri, 29 Sep 2017 09:49:35 +0000 (09:49 +0000)]
[AMDGPU] AMDPAL scratch buffer support

Summary:
Added support for scratch (including spilling) for OS type amdpal:
generates code to set up the scratch descriptor if it is needed.

With amdpal, the scratch resource descriptor is loaded from offset 0 of
the global information table. The low 32 bits of the address of the
global information table is passed in s0.

Added amdgpu-git-ptr-high function attribute to hard-wire the high 32
bits of the address of the global information table. If the function
attribute is not specified, or is 0xffffffff, then the backend generates
code to use the high 32 bits of pc.

The documentation for the AMDPAL ABI will be added in a later commit.

Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye

Differential Revision: https://reviews.llvm.org/D37483

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314501 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Triple] Add AMDPAL operating system type
Tim Renouf [Fri, 29 Sep 2017 09:48:12 +0000 (09:48 +0000)]
[Triple] Add AMDPAL operating system type

Summary:
This operating system type represents the AMDGPU PAL runtime, and will
be required by the AMDGPU backend in order to generate correct code for
this runtime.

Currently it generates the same code as not specifying an OS at all.
That will change in future commits.

Patch from Tim Corringham.

Subscribers: arsenm, nhaehnle

Differential Revision: https://reviews.llvm.org/D37380

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314500 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[dwarfdump][NFC] Consistent errors and warnings with --verify
Jonas Devlieghere [Fri, 29 Sep 2017 09:33:31 +0000 (09:33 +0000)]
[dwarfdump][NFC] Consistent errors and warnings with --verify

This patch introduces 3 helper functions: error(), warn() and note() to
make printing  during verification more consistent. When supported, the
respective prefixes are printed in color using the same color scheme as
clang.

Differential revision: https://reviews.llvm.org/D38368

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314498 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips] Reordering callseq* nodes to be linear
Aleksandar Beserminji [Fri, 29 Sep 2017 09:32:14 +0000 (09:32 +0000)]
[mips] Reordering callseq* nodes to be linear

Fix nested callseq* nodes by moving callseq_start after the
arguments calculation to temporary registers, so that callseq* nodes
in resulting DAG are linear.

Differential Revision: https://reviews.llvm.org/D37328

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314497 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][MS-InlineAsm] Extended support for variables / identifiers on memory / immediat...
Coby Tayree [Fri, 29 Sep 2017 07:02:46 +0000 (07:02 +0000)]
[X86][MS-InlineAsm] Extended support for variables / identifiers on memory / immediate expressions

Allow the proper recognition of Enum values and global variables inside ms inline-asm memory / immediate expressions, as they require some additional overhead and treated incorrect if doesn't early recognized.
supersedes D33278, D35774

Differential Revision: https://reviews.llvm.org/D37412

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314493 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoMake find_opt_files vararg
Adam Nemet [Fri, 29 Sep 2017 05:20:53 +0000 (05:20 +0000)]
Make find_opt_files vararg

This is slightly less verbose for the common case of a single build directory
and more intuitive when using this API directly from the interpreter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314491 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ORC] Replace decltype with a concrete type to make MSVC happy.
Lang Hames [Fri, 29 Sep 2017 05:03:43 +0000 (05:03 +0000)]
[ORC] Replace decltype with a concrete type to make MSVC happy.

This should fix some build failures on windows bots due to r314486.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314490 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[CMake] Fix typo "Wraning" (NFC)
Brian Gesiak [Fri, 29 Sep 2017 02:48:07 +0000 (02:48 +0000)]
[CMake] Fix typo "Wraning" (NFC)

Summary:
The typo was added in https://reviews.llvm.org/rL247151.
It should be "warning", not "wraning".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314486 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agollvm-readobj: fix a few typos (NFC)
Saleem Abdulrasool [Fri, 29 Sep 2017 02:45:44 +0000 (02:45 +0000)]
llvm-readobj: fix a few typos (NFC)

Correct the spelling of multiple in a couple of sites.

Patch by Alex Langford!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314485 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "[BypassSlowDivision] Improve our handling of divisions by constants"
Sanjoy Das [Fri, 29 Sep 2017 00:54:16 +0000 (00:54 +0000)]
Revert "[BypassSlowDivision] Improve our handling of divisions by constants"

This reverts commit r314253.  It causes a miscompile on P100 in an internal
benchmark.  Reverting while I investigate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314482 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agollvm-dwarfdump: support .apple-namespaces in --find
Adrian Prantl [Fri, 29 Sep 2017 00:52:33 +0000 (00:52 +0000)]
llvm-dwarfdump: support .apple-namespaces in --find

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314481 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-rc] Import all make_unique invocations from llvm namespace.
Marek Sokolowski [Fri, 29 Sep 2017 00:33:57 +0000 (00:33 +0000)]
[llvm-rc] Import all make_unique invocations from llvm namespace.

Previous patch fixed one of LLVM buildbots (lld-x86_64-win7).
However, some others have already been failing because of make_unique
compilation error (llvm-clang-x86_64-expensive-checks-win).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314480 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agollvm-dwarfdump: add support for .apple_types in --find
Adrian Prantl [Fri, 29 Sep 2017 00:33:22 +0000 (00:33 +0000)]
llvm-dwarfdump: add support for .apple_types in --find

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314479 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-rc] Add user-defined resources parsing ability. [8/8]
Marek Sokolowski [Fri, 29 Sep 2017 00:14:18 +0000 (00:14 +0000)]
[llvm-rc] Add user-defined resources parsing ability. [8/8]

This allows llvm-rc to parse user-defined resources (ref:
msdn.microsoft.com/en-us/library/windows/desktop/aa381054.aspx).
These statements either import files, or put the specified raw data in
the resulting resource file.

Thanks to Nico Weber for his original work in this area.

Differential Revision: https://reviews.llvm.org/D37033

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314478 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-rc] Add integer expressions parsing ability. [7/8]
Marek Sokolowski [Thu, 28 Sep 2017 23:53:25 +0000 (23:53 +0000)]
[llvm-rc] Add integer expressions parsing ability. [7/8]

This allows the ints to be written as integer expressions evaluating to
unsigned 16-bit/32-bit integers.

All the expressions may use the following operators: + - & | ~, and
parentheses. Minus token - can be also unary. There is no precedence of
the operators other than the unary operators binding stronger than their
binary counterparts.

Differential Revision: https://reviews.llvm.org/D37022

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314477 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MachineOutliner][NFC] Simplify logic in pruneCandidates
Jessica Paquette [Thu, 28 Sep 2017 23:39:36 +0000 (23:39 +0000)]
[MachineOutliner][NFC] Simplify logic in pruneCandidates

This commit yanks out the repeated sections of code in pruneCandidates into
two lambdas: ShouldSkipCandidate and Prune. This simplifies the logic in
pruneCandidates significantly, and reduces the chance of introducing bugs by
folding all of the shared logic into one place.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314475 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Don't select (cmp (and, imm), 0) to testw
Craig Topper [Thu, 28 Sep 2017 23:35:36 +0000 (23:35 +0000)]
[X86] Don't select (cmp (and, imm), 0) to testw

Summary:
X86ISelDAGToDAG tries to analyze ANDs compared with 0 to optimize to narrower immediates using subregisters.

I don't think we should be optimizing to 16-bit test instructions. It goes against our normal behavior of promoting i16 operations to i32. It only saves one byte due to the need to add a 0x66 prefix. I think it would also be subject to a length changing prefix penalty in the decoders on Intel CPUs.

Reviewers: RKSimon, zvi, spatel

Reviewed By: spatel

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D38273

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314474 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-rc] Fix-up for r314468 (argument-dependent lookup in make_unique).
Marek Sokolowski [Thu, 28 Sep 2017 23:12:53 +0000 (23:12 +0000)]
[llvm-rc] Fix-up for r314468 (argument-dependent lookup in make_unique).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314472 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoARM: Fix cases where CSI Restored bit is not cleared
Matthias Braun [Thu, 28 Sep 2017 23:12:06 +0000 (23:12 +0000)]
ARM: Fix cases where CSI Restored bit is not cleared

LR is an untypical callee saved register in that it is restored into a
different register (PC) and thus does not live-out of the return block.
This case requires the `Restored` flag in CalleeSavedInfo to be cleared.

This fixes a number of cases where this wasn't handled correctly yet.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314471 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agobpf: fix a bug for disassembling ld_pseudo inst
Yonghong Song [Thu, 28 Sep 2017 22:47:34 +0000 (22:47 +0000)]
bpf: fix a bug for disassembling ld_pseudo inst

Signed-off-by: Yonghong Song <yhs@fb.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314469 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-rc] Add VERSIONINFO parsing ability. [6/8]
Marek Sokolowski [Thu, 28 Sep 2017 22:41:38 +0000 (22:41 +0000)]
[llvm-rc] Add VERSIONINFO parsing ability. [6/8]

This extends the set of llvm-rc parser's available resources by
another one, VERSIONINFO.

Ref: msdn.microsoft.com/en-us/library/windows/desktop/aa381058.aspx

Thanks to Nico Weber for his original work in this area.

Differential Revision: https://reviews.llvm.org/D37021

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314468 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Fix some Clang-tidy modernize and Include What You Use warnings; other...
Eugene Zelenko [Thu, 28 Sep 2017 22:27:31 +0000 (22:27 +0000)]
[Hexagon] Fix some Clang-tidy modernize and Include What You Use warnings; other minor fixes (NFC).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314467 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[x86] add tests for possible insertelement to shuffle transform; NFC
Sanjay Patel [Thu, 28 Sep 2017 22:27:25 +0000 (22:27 +0000)]
[x86] add tests for possible insertelement to shuffle transform; NFC

See PR34716 and D38316 for more discussion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314466 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SystemZ] Fix fall-out from r314428
Ulrich Weigand [Thu, 28 Sep 2017 22:08:25 +0000 (22:08 +0000)]
[SystemZ] Fix fall-out from r314428

The expensive-checks build bot found a problem with the r314428 commit:
if CC is live after a ATOMIC_CMP_SWAPW instruction, it needs to be
marked as live-in to the block after the loop the pseudo gets expanded
to.  This actually fixes a code-gen bug as well, since if the CC isn't
live, the CR and JLH are merged to a CRJLH which doesn't actually set
the condition code any more.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314465 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Make use of vpmovwb when possible in LowerMULH
Craig Topper [Thu, 28 Sep 2017 20:10:34 +0000 (20:10 +0000)]
[X86] Make use of vpmovwb when possible in LowerMULH

If we have BWI, we can truncate in a much simpler way by using vpmovwb. This even works without VLX by using the wider zmm->ymm truncate with a subvector extract.

Differential Revision: https://reviews.llvm.org/D38375

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314457 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix -Werror build.
Evgeniy Stepanov [Thu, 28 Sep 2017 19:43:53 +0000 (19:43 +0000)]
Fix -Werror build.

/code/llvm-project/llvm/unittests/ExecutionEngine/Orc/RTDyldObjectLinkingLayerTest.cpp:260:38: error: lambda capture 'this' is not used [-Werror,-Wunused-lambda-capture]
                                    [this](decltype(ObjLayer)::ObjHandleT,

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314454 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ARM] Restore the right frame pointer register in Int_eh_sjlj_longjmp
Martin Storsjo [Thu, 28 Sep 2017 19:04:30 +0000 (19:04 +0000)]
[ARM] Restore the right frame pointer register in Int_eh_sjlj_longjmp

In setupEntryBlockAndCallSites in CodeGen/SjLjEHPrepare.cpp,
we fetch and store the actual frame pointer, but on return via
the longjmp intrinsic, it always was restored into the r7 variable.

On windows, the frame pointer should be restored into r11 instead of r7.

On Darwin (where sjlj exception handling is used by default), the frame
pointer is always r7, both in arm and thumb mode, and likewise, on
windows, the frame pointer always is r11.

On linux however, if sjlj exception handling is enabled (which it isn't
by default), libcxxabi and the user code can be built in differing modes
using different registers as frame pointer. Therefore, when restoring
registers on a platform where we don't always use the same register
depending on code mode, restore both r7 and r11.

Differential Revision: https://reviews.llvm.org/D38253

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314451 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ARM] Fix SJLJ exception handling when manually chosen on a platform where it isn...
Martin Storsjo [Thu, 28 Sep 2017 19:04:14 +0000 (19:04 +0000)]
[ARM] Fix SJLJ exception handling when manually chosen on a platform where it isn't default

Differential Revision: https://reviews.llvm.org/D38252

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314450 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoMIR: Serialize CaleeSavedInfo Restored flag
Matthias Braun [Thu, 28 Sep 2017 18:52:14 +0000 (18:52 +0000)]
MIR: Serialize CaleeSavedInfo Restored flag

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314449 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AVX512] Add avx512bw command lines to 128-bit idiv tests.
Craig Topper [Thu, 28 Sep 2017 18:45:29 +0000 (18:45 +0000)]
[AVX512] Add avx512bw command lines to 128-bit idiv tests.

The multiply lowering on some of the tests can take advantage of the vpmovwb to simplify the truncate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314448 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Use target independent ZERO_EXTEND/SIGN_EXTEND nodes were possible in LowerMULH
Craig Topper [Thu, 28 Sep 2017 18:45:28 +0000 (18:45 +0000)]
[X86] Use target independent ZERO_EXTEND/SIGN_EXTEND nodes were possible in LowerMULH

We aren't do any in register extends here so we should be able to just the target independent nodes directly and allow them to be lowered as necessary.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314447 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Move a setOperation action for ISD::TRUNCATE near another one in the same if...
Craig Topper [Thu, 28 Sep 2017 18:45:27 +0000 (18:45 +0000)]
[X86] Move a setOperation action for ISD::TRUNCATE near another one in the same if. Remove one that is redundant with another subtarget features.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314446 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAddress further review feedback. (NFC)
Adrian Prantl [Thu, 28 Sep 2017 18:31:51 +0000 (18:31 +0000)]
Address further review feedback. (NFC)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314443 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agotry and appease gcc
Adrian Prantl [Thu, 28 Sep 2017 18:27:00 +0000 (18:27 +0000)]
try and appease gcc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314442 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agollvm-dwarfdump: implement --find for .apple_names
Adrian Prantl [Thu, 28 Sep 2017 18:10:52 +0000 (18:10 +0000)]
llvm-dwarfdump: implement --find for .apple_names

This patch implements the dwarfdump option --find=<name>.  This option
looks for a DIE in the accelerator tables and dumps it if found.  This
initial patch only adds support for .apple_names to keep the review
small, adding the other sections and pubnames support should be
trivial though.

Differential Revision: https://reviews.llvm.org/D38282

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314439 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ORC] Fix the type of RTDyldObjectLinkingLayer::NotifyLoadedFtor.
Lang Hames [Thu, 28 Sep 2017 17:43:07 +0000 (17:43 +0000)]
[ORC] Fix the type of RTDyldObjectLinkingLayer::NotifyLoadedFtor.

Bug found by Stefan Granitz. Thanks Stefan!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314436 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[JumpThreading] Preserve DT and LVI across the pass
Evandro Menezes [Thu, 28 Sep 2017 17:24:40 +0000 (17:24 +0000)]
[JumpThreading] Preserve DT and LVI across the pass

JumpThreading now preserves dominance and lazy value information across the
entire pass.  The pass manager is also informed of this preservation with
the goal of DT and LVI being recalculated fewer times overall during
compilation.

This change prepares JumpThreading for enhanced opportunities; particularly
those across loop boundaries.

Patch by: Brian Rzycki <b.rzycki@samsung.com>,
          Sebastian Pop <s.pop@samsung.com>

Differential revision: https://reviews.llvm.org/D37528

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314435 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Use BWI instructions to improve lowering of v32i8 MULHU/S
Craig Topper [Thu, 28 Sep 2017 17:00:21 +0000 (17:00 +0000)]
[X86] Use BWI instructions to improve lowering of v32i8 MULHU/S

Summary: If we have BWI instructions we can widen to v32i16 to do the multiply instead of splitting.

Reviewers: RKSimon, spatel, zvi

Reviewed By: zvi

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D38305

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314432 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Remove dead code from X86ISelDAGToDAG.cpp multiply handling
Craig Topper [Thu, 28 Sep 2017 16:56:36 +0000 (16:56 +0000)]
[X86] Remove dead code from X86ISelDAGToDAG.cpp multiply handling

Summary:
Lowering never creates X86ISD::UMUL for 8-bit types. X86ISD::UMUL8 is used instead. If X86ISD::UMUL 8-bit were ever used it would crash.

DAGCombiner replaces UMUL_LOHI/SMUL_LOHI with a wider MUL and a shift if the type twice as wide is legal. So we should never see i8 UMUL_LOHI/SMUL_LOHI. In fact I think there was a bug in part of the i8 code. Similar is true for i16 though without the bug.

Reviewers: RKSimon, spatel, zvi

Reviewed By: zvi

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D38276

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314430 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Use correct subvector index when combining two insert subvectors featuring...
Craig Topper [Thu, 28 Sep 2017 16:53:16 +0000 (16:53 +0000)]
[X86] Use correct subvector index when combining two insert subvectors featuring zero vectors.

Previously we were using one of the subvector indices twice. The included test case causes an assert without this change.

Thanks to Simon Pilgrim for catching this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314429 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SystemZ] Custom-expand ATOMIC_CMP_AND_SWAP_WITH_SUCCESS
Ulrich Weigand [Thu, 28 Sep 2017 16:22:54 +0000 (16:22 +0000)]
[SystemZ] Custom-expand ATOMIC_CMP_AND_SWAP_WITH_SUCCESS

The SystemZ compare-and-swap instructions already provide the "success"
indication via a condition-code value, so the default expansion of those
operations generates an unnecessary extra comparsion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314428 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[dwarfdump] Verify that CUs have a unit DIE.
Jonas Devlieghere [Thu, 28 Sep 2017 15:57:50 +0000 (15:57 +0000)]
[dwarfdump] Verify that CUs have a unit DIE.

This patch adds a check to the DWARF verifier to detect CUs without a
unit DIE.

Differential revision: https://reviews.llvm.org/D38363

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314426 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoUse SDValue::getConstantOperandVal helper. NFCI.
Simon Pilgrim [Thu, 28 Sep 2017 15:53:27 +0000 (15:53 +0000)]
Use SDValue::getConstantOperandVal helper. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314425 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips] Remove codegen support for branch likely instructions.
Simon Dardis [Thu, 28 Sep 2017 15:24:07 +0000 (15:24 +0000)]
[mips] Remove codegen support for branch likely instructions.

This patch disables codegen support for branch likely instructions to
address a potential bug. These branches were unselectable as
they had the same patterns as the normal branches but came after them
when ISel was concerned.

The branch likely instructions were marked as having no delay
slots when they have annulling delay slots. The delay slot filler
does not currently handle annulling delay slot branches, so this
would lead to wrong codegen if these branches were generated.

Reviewers: atanasyan, nitesh.jain

Differential Revision: https://reviews.llvm.org/D38169

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314421 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoDocs: fix link to Debugger intrinsic functions
Hans Wennborg [Thu, 28 Sep 2017 15:16:37 +0000 (15:16 +0000)]
Docs: fix link to Debugger intrinsic functions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314420 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[LoopUnroll] Fix use after poison.
Benjamin Kramer [Thu, 28 Sep 2017 14:47:39 +0000 (14:47 +0000)]
[LoopUnroll] Fix use after poison.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314418 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add overflow intrinsic test in preparation for D38161.
Amara Emerson [Thu, 28 Sep 2017 13:43:48 +0000 (13:43 +0000)]
[X86] Add overflow intrinsic test in preparation for D38161.

This commit adds the test file before codegen changes as requested in
D38161 to make it easier to see the difference.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314416 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DebugInfo] Do not extend range for physreg in LiveDebugVariables
Bjorn Pettersson [Thu, 28 Sep 2017 13:10:06 +0000 (13:10 +0000)]
[DebugInfo] Do not extend range for physreg in LiveDebugVariables

Summary:
A DBG_VALUE that is referring to a physical register is
valid up until the next def of the register, or the end
of the basic block that it belongs to.

LiveDebugVariables is computing live intervals (slot index
ranges) for DBG_VALUE instructions, before regalloc, in order
to be able to re-insert DBG_VALUE instructions again after
regalloc. When the DBG_VALUE is mapping a variable to a
physical register we do not need to compute the range. We
should simply re-insert the DBG_VALUE at the start position.

The problem that was found, resulting in this patch, was a
situation when the DBG_VALUE was the last real use of the
physical register. The computeIntervals/extendDef methods
extended the range to cover the whole basic block, even though
the physical register very well could be allocated to some
virtual register inside the basic block. So the extended
range could not be trusted.

This patch is a preparation for https://reviews.llvm.org/D38229,
where the goal is to insert DBG_VALUE after each new definition
of a variable, even if the virtual registers that the variable
was connected to has been coalesced into using the same physical
register (e.g. due to two address instructions). For more info
see https://bugs.llvm.org/show_bug.cgi?id=34545

Reviewers: aprantl, rnk, echristo

Reviewed By: aprantl

Subscribers: Ka-Ka, llvm-commits

Differential Revision: https://reviews.llvm.org/D38140

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314414 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[LoopInfo] Don't poison random memory regions.
Benjamin Kramer [Thu, 28 Sep 2017 12:53:20 +0000 (12:53 +0000)]
[LoopInfo] Don't poison random memory regions.

The second argument for Allocator::Deallocate is the number of elements,
not the size of a single element. In asan mode specifying a large number
of elements poisoned random memory regions, leading to crashes
everywhere.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314413 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[LVI] Move LVILatticeVal class to separate header file (NFC).
Florian Hahn [Thu, 28 Sep 2017 11:09:22 +0000 (11:09 +0000)]
[LVI] Move LVILatticeVal class to separate header file (NFC).

Summary:
This allows sharing the lattice value code between LVI and SCCP (D36656).

It also adds a `satisfiesPredicate` function, used by D36656.

Reviewers: davide, sanjoy, efriedma

Reviewed By: sanjoy

Subscribers: mgorny, llvm-commits

Differential Revision: https://reviews.llvm.org/D37591

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314411 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[x86][AsmParser] Allow some more MS size directives
Coby Tayree [Thu, 28 Sep 2017 11:04:08 +0000 (11:04 +0000)]
[x86][AsmParser] Allow some more MS size directives
MS allows the following size directives: float/double and long as synonymous to dword/qword and dword, respectively.
Differential Revision: https://reviews.llvm.org/D37190

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314410 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-cov] Create directory structure when filtering using -name*= options
Sean Eveson [Thu, 28 Sep 2017 10:07:30 +0000 (10:07 +0000)]
[llvm-cov] Create directory structure when filtering using -name*= options

Before this change using any of the -name*= command line options with an output
directory would result in a single file (functions.txt/functions.html)
containing the coverage for those specific functions. Now you get the same
directory structure as when not using any -name*= options.

Differential Revision: https://reviews.llvm.org/D38280

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314396 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoTeach TargetInstrInfo::getInlineAsmLength to parse .space directives with integer...
Alex Bradbury [Thu, 28 Sep 2017 09:31:46 +0000 (09:31 +0000)]
Teach TargetInstrInfo::getInlineAsmLength to parse .space directives with integer arguments

It's currently quite difficult to test passes like branch relaxation, which
requires branches with large displacement to be generated. The .space assembler
directive makes it easy to create arbitrarily large basic blocks, but
getInlineAsmLength is not able to parse it and so the size of the block is not
correctly estimated. Other backends (AArch64, AMDGPU) introduce options just
for testing that artificially restrict the ranges of branch instructions (e.g.
aarch64-tbz-offset-bits). Although parsing a single form of the .space
directive feels inelegant, it does allow a more direct testing approach.

This patch adapts the .space parsing code from
Mips16InstrInfo::getInlineAsmLength and removes it now the extra functionality
is provided by the base implementation. I want to move this functionality to
the generic getInlineAsmLength as 1) I need the same for RISC-V, and 2) I feel
other backends will benefit from more direct testing of large branch
displacements.

Differential Revision: https://reviews.llvm.org/D37798

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314393 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[PowerPC] eliminate partially redundant compare instruction
Hiroshi Inoue [Thu, 28 Sep 2017 08:38:19 +0000 (08:38 +0000)]
[PowerPC] eliminate partially redundant compare instruction

This is a follow-on of D37211.
D37211 eliminates a compare instruction if two conditional branches can be made based on the one compare instruction, e.g.
if (a == 0) { ... }
else if (a < 0) { ... }

This patch extends this optimization to support partially redundant cases, which often happen in while loops.
For example, one compare instruction is moved from the loop body into the preheader by this optimization in the following example.
do {
  if (a == 0) dummy1();
  a = func(a);
} while (a > 0);

Differential Revision: https://reviews.llvm.org/D38236

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314390 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[RISCV] Add common fixups and relocations
Alex Bradbury [Thu, 28 Sep 2017 08:26:24 +0000 (08:26 +0000)]
[RISCV] Add common fixups and relocations

%lo(), %hi(), and %pcrel_hi() are supported and test cases have been added to
ensure the appropriate fixups and relocations are generated. I've added an
instruction format field which is used in RISCVMCCodeEmitter to, for
instance, tell whether it should emit a lo12_i fixup or a lo12_s fixup
(RISC-V has two 12-bit immediate encodings depending on the instruction
type).

Differential Revision: https://reviews.llvm.org/D23568

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314389 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[RegAllocGreedy]: Allow recoloring of done register if it's non-tied
Mikael Holmen [Thu, 28 Sep 2017 08:22:35 +0000 (08:22 +0000)]
[RegAllocGreedy]: Allow recoloring of done register if it's non-tied

Summary:
If we have a non-allocated register, we allow us to try recoloring of an
already allocated and "Done" register, even if they are of the same
register class, if the non-allocated register has at least one tied def
and the allocated one has none.

It should be easier to recolor the non-tied register than the tied one, so
it might be an improvement even if they use the same regclasses.

Reviewers: qcolombet

Reviewed By: qcolombet

Subscribers: llvm-commits, MatzeB

Differential Revision: https://reviews.llvm.org/D38309

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314388 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[RISCV] Define RISC-V specific e_flags
Alex Bradbury [Thu, 28 Sep 2017 07:54:01 +0000 (07:54 +0000)]
[RISCV] Define RISC-V specific e_flags

Add RISC-V e_flags as defined in the ABI document:

https://github.com/riscv/riscv-elf-psabi-doc/blob/master/riscv-elf.md#file-header

Differential Revision: https://reviews.llvm.org/D38310
Patch by Chih-Mao Chen.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314386 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Adding more cases to horizontal [f]add/[f]sub for avx512.
Jatin Bhateja [Thu, 28 Sep 2017 07:40:52 +0000 (07:40 +0000)]
[X86] Adding more cases to horizontal [f]add/[f]sub for avx512.

Reviewers: jbhateja

Reviewed By: jbhateja

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D38344

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314385 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DAGCombiner] Fix an off-by-one error in vector logic
George Burgess IV [Thu, 28 Sep 2017 06:17:19 +0000 (06:17 +0000)]
[DAGCombiner] Fix an off-by-one error in vector logic

Without this, we could end up trying to get the Nth (0-indexed) element
from a subvector of size N.

Differential Revision: https://reviews.llvm.org/D37880

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314380 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agobpf: add new insns for bswap_to_le and negation
Yonghong Song [Thu, 28 Sep 2017 02:46:11 +0000 (02:46 +0000)]
bpf: add new insns for bswap_to_le and negation

This patch adds new insn, "reg = be16/be32/be64 reg",
for bswap to little endian for big-endian target (bpfeb).
It also adds new insn for negation "reg = -reg".

Currently, for source code, e.g.,
  b = -a
LLVM still prefers to generate:
  b = 0 - a
But "reg = -reg" format can be used in assembly code.

Signed-off-by: Yonghong Song <yhs@fb.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314376 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoUse a BumpPtrAllocator for Loop objects
Sanjoy Das [Thu, 28 Sep 2017 02:45:42 +0000 (02:45 +0000)]
Use a BumpPtrAllocator for Loop objects

Summary:
And now that we no longer have to explicitly free() the Loop instances, we can
(with more ease) use the destructor of LoopBase to do what LoopBase::clear() was
doing.

Reviewers: chandlerc

Subscribers: mehdi_amini, mcrosier, llvm-commits

Differential Revision: https://reviews.llvm.org/D38201

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314375 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ORC] Update the GlobalMappingLayer interface to fit the error-ized layer
Lang Hames [Thu, 28 Sep 2017 02:17:35 +0000 (02:17 +0000)]
[ORC] Update the GlobalMappingLayer interface to fit the error-ized layer
concept.

Add a unit-test to make sure we don't backslide, and tweak the MockBaseLayer
utility to make it easier to test this kind of thing in the future.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314374 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix a UBsan bot.
Rui Ueyama [Thu, 28 Sep 2017 00:27:39 +0000 (00:27 +0000)]
Fix a UBsan bot.

If we do not initialize Prefix here, Prefix.data() returns a nullptr.
Later, it is passed to memcpy. memcpy's behavior is undefined if src (or
dst) is a nullptr even if a given size is 0. That's why this code
triggered UBsan.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314368 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[CodeGen] Fix some Clang-tidy modernize-use-default-member-init and Include What...
Eugene Zelenko [Wed, 27 Sep 2017 23:26:01 +0000 (23:26 +0000)]
[CodeGen] Fix some Clang-tidy modernize-use-default-member-init and Include What You Use warnings; other minor fixes (NFC).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314363 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoCheck for overflows when calculating the offset in GetGEPCost.
Justin Lebar [Wed, 27 Sep 2017 23:16:56 +0000 (23:16 +0000)]
Check for overflows when calculating the offset in GetGEPCost.

Summary:
This avoids C++ UB if the GEP is weird and the calculation overflows
int64_t, and it's also observable in the cost model's results.

Such GEPs are almost surely not valid pointers, but LLVM nonetheless
generates them sometimes.

Reviewers: sanjoy

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D38337

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314362 91177308-0d34-0410-b5e6-96231b3b80d8