OSDN Git Service

android-x86/external-llvm.git
6 years ago[CostModel] Recognise BROADCAST shuffle mask if the elements come from the second src
Simon Pilgrim [Wed, 13 Jun 2018 16:52:02 +0000 (16:52 +0000)]
[CostModel] Recognise BROADCAST shuffle mask if the elements come from the second src

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334620 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert: [llvm-mca] Flush the output stream before we start the analysis of a new...
Andrea Di Biagio [Wed, 13 Jun 2018 16:33:52 +0000 (16:33 +0000)]
Revert: [llvm-mca] Flush the output stream before we start the analysis of a new code region. NFC

Not sure why, but it breaks buildbot clang-cmake-armv8-full.
It causes a failure in TEST 'Xray-armhf-linux :: TestCases/Posix/profiling-single-threaded.cc'.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334617 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[CostModel][X86] Test showing failure to recognise BROADCAST shuffle mask if the...
Simon Pilgrim [Wed, 13 Jun 2018 16:33:42 +0000 (16:33 +0000)]
[CostModel][X86] Test showing failure to recognise BROADCAST shuffle mask if the elements come from the second src

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334616 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-mca] Flush the output stream before we start the analysis of a new code region...
Andrea Di Biagio [Wed, 13 Jun 2018 15:43:56 +0000 (15:43 +0000)]
[llvm-mca] Flush the output stream before we start the analysis of a new code region. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334610 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AMDGPU][MC][GFX8][GFX9] Allow LDS direct reads for BUFFER_LOAD_DWORDX2/X3/X4
Dmitry Preobrazhensky [Wed, 13 Jun 2018 15:32:46 +0000 (15:32 +0000)]
[AMDGPU][MC][GFX8][GFX9] Allow LDS direct reads for BUFFER_LOAD_DWORDX2/X3/X4

See bug 37653: https://bugs.llvm.org/show_bug.cgi?id=37653

Reviewers: artem.tamazov, arsenm

Differential Revision: https://reviews.llvm.org/D47885

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334609 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DAGCombiner] remove hasOneUse() check from fadd constants transform
Sanjay Patel [Wed, 13 Jun 2018 15:22:48 +0000 (15:22 +0000)]
[DAGCombiner] remove hasOneUse() check from fadd constants transform

We're constant folding here, so we shouldn't check uses. This matches
the IR optimizer behavior.

The x86 test shows the expected win. The AArch64 test shows something
else. This only seems to happen if the "generic" AArch64 CPU model is
used by MachineCombiner, so I'll file a bug report to follow-up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334608 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Move isSDNodeSourceOfDivergence() implementation to SITargetLowering
Tom Stellard [Wed, 13 Jun 2018 15:06:37 +0000 (15:06 +0000)]
AMDGPU: Move isSDNodeSourceOfDivergence() implementation to SITargetLowering

Summary:
The code that handles ISD:Register and ISD::CopyFromReg assumes
the target is amdgcn, so this is broken on r600.  We don't
need this analysis on r600 anyway so we can safely move
it to SITargetLowering.

Reviewers: alex-t, arsenm, nhaehnle

Reviewed By: arsenm

Subscribers: msearles, kzhuravl, wdng, yaxunl, dstuttard, tpr, t-tye, llvm-commits

Differential Revision: https://reviews.llvm.org/D46298

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334607 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[x86] add test for fadd with more than one use; NFC
Sanjay Patel [Wed, 13 Jun 2018 15:01:07 +0000 (15:01 +0000)]
[x86] add test for fadd with more than one use; NFC

The equivalent AArch64 test added at rL334556 isn't showing
the expected output from the DAGCombiner code change that
would fix this example. That's a machine combiner bug from
what I see.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334605 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[FPEnv] Expand constrained FP operations
Cameron McInally [Wed, 13 Jun 2018 14:32:12 +0000 (14:32 +0000)]
[FPEnv] Expand constrained FP operations

Add a helper function to expand constrained FP operations as needed.
Note that the Strict POWI operation is not handled in this patch since
the format is slightly different from the others.

Differential Revision: https://reviews.llvm.org/D47491

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334603 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoDo not enforce absolute path argv0 in windows
Hans Wennborg [Wed, 13 Jun 2018 14:29:26 +0000 (14:29 +0000)]
Do not enforce absolute path argv0 in windows

Even if we support no-canonical-prefix on
clang-cl(https://reviews.llvm.org/D47480), argv0 becomes absolute path
in clang-cl and that embeds absolute path in /showIncludes.

This patch removes such full path normalization from InitLLVM on
windows, and that removes absolute path from clang-cl output
(obj/stdout/stderr) when debug flag is disabled.

Patch by Takuto Ikuta!

Differential Revision https://reviews.llvm.org/D47578

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334602 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-exegesis] Fix buildbot - power was using native target for X86.
Guillaume Chatelet [Wed, 13 Jun 2018 14:07:36 +0000 (14:07 +0000)]
[llvm-exegesis] Fix buildbot - power was using native target for X86.

Reviewers: courbet

Reviewed By: courbet

Subscribers: tschuett, llvm-commits

Differential Revision: https://reviews.llvm.org/D48125

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334601 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-exegesis] Fix failing assert when creating Snippet for LAHF.
Guillaume Chatelet [Wed, 13 Jun 2018 13:53:56 +0000 (13:53 +0000)]
[llvm-exegesis] Fix failing assert when creating Snippet for LAHF.

Reviewers: courbet

Subscribers: tschuett, llvm-commits

Differential Revision: https://reviews.llvm.org/D48123

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334599 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "Improve handling of COPY instructions with identical value numbers"
Krzysztof Parzyszek [Wed, 13 Jun 2018 13:49:06 +0000 (13:49 +0000)]
Revert "Improve handling of COPY instructions with identical value numbers"

This reverts r334594, it breaks buildbots and fails with expensive checks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334598 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-exegesis] Cleaner design without mutable data.
Guillaume Chatelet [Wed, 13 Jun 2018 13:24:41 +0000 (13:24 +0000)]
[llvm-exegesis] Cleaner design without mutable data.

Summary: Previous design was relying on the 'mutate' keyword and was quite confusing. This version separate mutable from immutable data and makes it clearer what changes and what doesn't.

Reviewers: courbet

Subscribers: tschuett, llvm-commits

Differential Revision: https://reviews.llvm.org/D48020

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334596 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips][microMIPS] Extending size reduction pass with LWP and SWP
Zoran Jovanovic [Wed, 13 Jun 2018 12:51:37 +0000 (12:51 +0000)]
[mips][microMIPS] Extending size reduction pass with LWP and SWP
Author: milena.vujosevic.janicic
Reviewers: sdardis
The patch extends size reduction pass for MicroMIPS.
It introduces reduction of two instructions into one instruction:
Two SW instructions are transformed into one SWP instrucition.
Two LW instructions are transformed into one LWP instrucition.
Differential Revision: https://reviews.llvm.org/D39115

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334595 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoImprove handling of COPY instructions with identical value numbers
Krzysztof Parzyszek [Wed, 13 Jun 2018 12:47:17 +0000 (12:47 +0000)]
Improve handling of COPY instructions with identical value numbers

Differential Revision: https://reviews.llvm.org/D48102

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334594 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[x86] eliminate even more sign-bit tests with vector select
Sanjay Patel [Wed, 13 Jun 2018 12:28:32 +0000 (12:28 +0000)]
[x86] eliminate even more sign-bit tests with vector select

This shortcoming was noted in D47330, and the test diffs show we already
had other examples where we failed to fold to a SHRUNKBLEND:

/// Dynamic (non-constant condition) vector blend where only the sign bits
/// of the condition elements are used. This is used to enforce that the
/// condition mask is not valid for generic VSELECT optimizations.

This patch implements an idea from D48043 and would obsolete that patch
because it catches more cases (notable the AVX1 case that was missed there).
All we're doing is allowing the existing transform to fire more often by
removing the post-legalize constraint. All of the relevant feature checks
and other predicates are left as-is.

Differential Revision: https://reviews.llvm.org/D48078

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334592 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[RISCV] Add codegen support for atomic load/stores with RV32A
Alex Bradbury [Wed, 13 Jun 2018 12:04:51 +0000 (12:04 +0000)]
[RISCV] Add codegen support for atomic load/stores with RV32A

Fences are inserted according to table A.6 in the current draft of version 2.3
of the RISC-V Instruction Set Manual, which incorporates the memory model
changes and definitions contributed by the RISC-V Memory Consistency Model
task group.

Instruction selection failures will now occur for 8/16/32-bit atomicrmw and
cmpxchg operations when targeting RV32IA until lowering for these operations
is added in a follow-on patch.

Differential Revision: https://reviews.llvm.org/D47589

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334591 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[RISCV] Codegen support for atomic operations on RV32I
Alex Bradbury [Wed, 13 Jun 2018 11:58:46 +0000 (11:58 +0000)]
[RISCV] Codegen support for atomic operations on RV32I

This patch adds lowering for atomic fences and relies on AtomicExpandPass to
lower atomic loads/stores, atomic rmw, and cmpxchg to __atomic_* libcalls.

test/CodeGen/RISCV/atomic-* are modelled on the exhaustive
test/CodeGen/PPC/atomics-regression.ll, and will prove more useful once RV32A
codegen support is introduced.

Fence mappings are taken from table A.6 in the current draft of version 2.3 of
the RISC-V Instruction Set Manual, which incorporates the memory model changes
and definitions contributed by the RISC-V Memory Consistency Model task group.

Differential Revision: https://reviews.llvm.org/D47587

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334590 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SLPVectorizer] getSameOpcode - remove useless cast [NFC]
Simon Pilgrim [Wed, 13 Jun 2018 10:49:24 +0000 (10:49 +0000)]
[SLPVectorizer] getSameOpcode - remove useless cast [NFC]

There's no need to cast the base Value to an Instruction

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334588 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SLPVectorizer] getSameOpcode - remove unusued alternate code [NFC]
Simon Pilgrim [Wed, 13 Jun 2018 10:14:27 +0000 (10:14 +0000)]
[SLPVectorizer] getSameOpcode - remove unusued alternate code [NFC]

We early-out for the case where we don't use alternate opcodes, so no need to check for it later.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334587 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[TableGen] Emit a fatal error on inconsistencies in resource units vs cycles.
Clement Courbet [Wed, 13 Jun 2018 09:41:49 +0000 (09:41 +0000)]
[TableGen] Emit a fatal error on inconsistencies in resource units vs cycles.

Summary:
For targets I'm not familiar with, I've automatically made the "default to 1 for each resource" behaviour explicit in the td files.
For more obvious cases, I've ventured a fix.

Some notes:
 - Exynos is especially fishy.
 - AArch64SchedThunderX2T99.td had some truncated entries. If I understand correctly, the person who wrote that interpreted the ResourceCycle as a range. I made the decision to use the upper/lower bound for consistency with the 'Latency' value. I'm sure there is a better choice.
 - The change to X86ScheduleBtVer2.td is an NFC, it just makes values more explicit.

Also see PR37310.

Reviewers: RKSimon, craig.topper, javed.absar

Subscribers: kristof.beyls, llvm-commits

Differential Revision: https://reviews.llvm.org/D46356

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334586 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoGettingStarted.rst: Fix 'If you you' typo (PR37787)
Hans Wennborg [Wed, 13 Jun 2018 09:11:10 +0000 (09:11 +0000)]
GettingStarted.rst: Fix 'If you you' typo (PR37787)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334584 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[PowerPC] fix trivial typos in comment, NFC
Hiroshi Inoue [Wed, 13 Jun 2018 08:54:13 +0000 (08:54 +0000)]
[PowerPC] fix trivial typos in comment, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334583 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix -DLLVM_ENABLE_THREADS=OFF build after r334537
Hans Wennborg [Wed, 13 Jun 2018 08:43:03 +0000 (08:43 +0000)]
Fix -DLLVM_ENABLE_THREADS=OFF build after r334537

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334582 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix "Optional" is ambiguous error on some bots
Pavel Labath [Wed, 13 Jun 2018 08:29:19 +0000 (08:29 +0000)]
Fix "Optional" is ambiguous error on some bots

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334580 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[PowerPC] avoid verification failure due to PowerPC VSX Swap Removal pass
Hiroshi Inoue [Wed, 13 Jun 2018 08:25:14 +0000 (08:25 +0000)]
[PowerPC] avoid verification failure due to PowerPC VSX Swap Removal pass

This patch fixes a failure in lnt tests with -verify-machineinstrs option.
When VSX Swap Removal pass swaps two register operands, it did not maintain kill flags associated with operands. This patch swaps flags as well as register number to avoid inconsistent kill flags information.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334579 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DWARF/AccelTable] Remove getDIESectionOffset for DWARF v5 entries
Pavel Labath [Wed, 13 Jun 2018 08:14:27 +0000 (08:14 +0000)]
[DWARF/AccelTable] Remove getDIESectionOffset for DWARF v5 entries

Summary:
This method was not correct for entries in DWO files as it assumed it
could just add up the CU and DIE offsets to get the absolute DIE offset.
This is not correct for the DWO files, as here the CU offset will
reference the skeleton unit, whereas the DIE offset will be the offset
in the full unit in the DWO file.

Unfortunately, this means that we are not able to determine the absolute
DIE offset using the information in the .debug_names section alone,
which means we have to offload some of this work to the users of this
class.

To demonstrate how this can be done, I've added/fixed the ability to
lookup entries using accelerator tables in DWO files in llvm-dwarfdump.
To make this happen, I've needed to make two extra changes in other
classes:
- made the DWARFContext method to lookup a CU based on the section
  offset public. I've needed this functionality to lookup a CU, and this
  seems like a useful thing in general.
- made DWARFUnit::getDWOId call extractDIEsIfNeeded. Before this, the
  DWOId was filled in only if the root DIE happened to be parsed
  before we called the accessor. Since the lazy parsing is supposed to
  happen under the hood, calling extractDIEsIfNeeded seems appropriate.

Reviewers: JDevlieghere, aprantl, dblaikie

Subscribers: mgrang, llvm-commits

Differential Revision: https://reviews.llvm.org/D48009

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334578 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use...
Craig Topper [Wed, 13 Jun 2018 07:19:21 +0000 (07:19 +0000)]
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334576 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoSet the code model when specified.
Bill Wendling [Wed, 13 Jun 2018 05:53:59 +0000 (05:53 +0000)]
Set the code model when specified.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334571 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoutils/release: Add merge-git.sh
Tom Stellard [Wed, 13 Jun 2018 05:14:10 +0000 (05:14 +0000)]
utils/release: Add merge-git.sh

Summary:
This script allows you to use git to backport a commit to a stable
branch while generating the exact same commit message (ignoring
whitespace) that you would get from using the merge.sh script with svn.

Reviewers: hansw

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D47760

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334568 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SimplifyIndVars] Ignore dead users
Max Kazantsev [Wed, 13 Jun 2018 02:25:32 +0000 (02:25 +0000)]
[SimplifyIndVars] Ignore dead users

IndVarSimplify sometimes makes transforms basing on users that are trivially dead. In particular,
if DCE wasn't run before it, there may be a dead `sext/zext` in loop that will trigger widening
transforms, however it makes no sense to do it.

This patch teaches IndVarsSimplify ignore the mist trivial cases of that.

Differential Revision: https://reviews.llvm.org/D47974
Reviewed By: sanjoy

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334567 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "Fix how LLVMOPTIONALCOMPONENTS is passed to llvm-build"
Ahmed Bougacha [Wed, 13 Jun 2018 02:16:01 +0000 (02:16 +0000)]
Revert "Fix how LLVMOPTIONALCOMPONENTS is passed to llvm-build"

This reverts commit r334543.

My understanding is, that commit is intended to make the llvm-build
invocation have a correct "--enable-optional-components" value, but:
- it already has a value: it's quoted in the command line a few lines
  below, and, if I hack llvm-build to print sys.argv, it does look correct:
    -- llvm-build output: ['.../utils/llvm-build/llvm-build',
      '--native-target', 'X86', '--enable-targets', 'X86;ARM;AArch64',
      '--enable-optional-components', '',
      '--write-library-table',
      '.../build/tools/llvm-config/LibraryDependencies.inc',
      '--write-cmake-fragment', '.../build/LLVMBuild.cmake']
- the " " string seems to evaluate to TRUE in CMake (*sigh*), so this
  basically force-enables LLVM_USE_INTEL_JITEVENTS, regardless of the
  value of the option.
  On Darwin, JITEvents is not supported, so this bypasses that OS check
  but is guaranteed to fail later.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334566 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Mark all instructions that have masked store semantics with NotMemoryFoldable...
Craig Topper [Wed, 13 Jun 2018 00:04:08 +0000 (00:04 +0000)]
[X86] Mark all instructions that have masked store semantics with NotMemoryFoldable. Remove dependency on SchedRW from memory table autogenerator.

Previously we were whitelisting in instructions based on their SchedRW value. With the masked store instructions explicitly removed via NotMemoryFoldable, we don't seem to need this check anymore.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334563 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Remove VPCOMPRESSB/W from the autogenerated load folding table.
Craig Topper [Wed, 13 Jun 2018 00:04:04 +0000 (00:04 +0000)]
[X86] Remove VPCOMPRESSB/W from the autogenerated load folding table.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334562 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AMDGPU] DAG combine to produce V_PERM_B32
Stanislav Mekhanoshin [Tue, 12 Jun 2018 23:50:37 +0000 (23:50 +0000)]
[AMDGPU] DAG combine to produce V_PERM_B32

Differential Revision: https://reviews.llvm.org/D48099

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334559 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64] add tests for fadd with more than one use; NFC
Sanjay Patel [Tue, 12 Jun 2018 22:50:37 +0000 (22:50 +0000)]
[AArch64] add tests for fadd with more than one use; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334556 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAdded missing include to AMDHSAKernelDescriptor.h
Raphael Isemann [Tue, 12 Jun 2018 22:40:20 +0000 (22:40 +0000)]
Added missing include to AMDHSAKernelDescriptor.h

We use size_t in this header, so we also need to include
cstddef to make it compile. Fixes the module builds.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334555 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DAGCombiner] Recognize more patterns for ABS
Krzysztof Parzyszek [Tue, 12 Jun 2018 21:51:49 +0000 (21:51 +0000)]
[DAGCombiner] Recognize more patterns for ABS

Differential Revision: https://reviews.llvm.org/D47831

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334553 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] add avx512 tests for potentially miscompiling cvttp2si/cvttp2ui (PR37751).
Craig Topper [Tue, 12 Jun 2018 21:42:42 +0000 (21:42 +0000)]
[X86] add avx512 tests for potentially miscompiling cvttp2si/cvttp2ui (PR37751).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334551 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRemove malloc.h include from Intel JIT events code
Reid Kleckner [Tue, 12 Jun 2018 21:15:27 +0000 (21:15 +0000)]
Remove malloc.h include from Intel JIT events code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334547 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAdd null check to Intel JIT event listener
Reid Kleckner [Tue, 12 Jun 2018 20:54:11 +0000 (20:54 +0000)]
Add null check to Intel JIT event listener

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334544 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix how LLVMOPTIONALCOMPONENTS is passed to llvm-build
Reid Kleckner [Tue, 12 Jun 2018 20:53:46 +0000 (20:53 +0000)]
Fix how LLVMOPTIONALCOMPONENTS is passed to llvm-build

Patch by Force.Charlie-I

If LLVM_USE_INTEL_JITEVENTS and LLVM_USE_OPROFILE  not set,
"${LLVMOPTIONALCOMPONENTS}" is empty, but
**--enable-optional-components** need arg, Cause
**--write-library-table** to be skipped parsed.

Differential Revision: https://reviews.llvm.org/D47982

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334543 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ORC] Add a fallback definition generator for VSOs.
Lang Hames [Tue, 12 Jun 2018 20:43:18 +0000 (20:43 +0000)]
[ORC] Add a fallback definition generator for VSOs.

If a VSO has a fallback definition generator attached it will be called during
lookup (and lookupFlags) for any unresolved symbols. The definition generator
can add new definitions to the VSO for any unresolved symbol. This allows VSOs
to generate new definitions on demand.

The immediate use case for this code is supporting VSOs that can import
definitions found via dlsym on demand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334538 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ORC] Refactor blocking lookup logic into the blockingLookup function, and
Lang Hames [Tue, 12 Jun 2018 20:43:17 +0000 (20:43 +0000)]
[ORC] Refactor blocking lookup logic into the blockingLookup function, and
implement existing blocking lookups (the lookup function) and
JITSymbolResolverAdapter on top of that.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334537 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[RuntimeDyld] Add an assert to catch misbehaving symbol resolvers.
Lang Hames [Tue, 12 Jun 2018 20:43:17 +0000 (20:43 +0000)]
[RuntimeDyld] Add an assert to catch misbehaving symbol resolvers.

Resolvers are required to find results for all requested symbols or return an
error, but if a resolver fails to adhere to this contract (by returning results
for only a subset of the requested symbols) then this code will infinite loop.
This assertion catches resolvers that fail to adhere to the contract.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334536 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MCJIT] Call materializeAll on modules before compiling them in MCJIT.
Lang Hames [Tue, 12 Jun 2018 20:43:15 +0000 (20:43 +0000)]
[MCJIT] Call materializeAll on modules before compiling them in MCJIT.

This only affects modules with lazy GVMaterializers attached (usually modules
read off disk using the lazy bitcode reader). For such modules, materializing
before compiling prevents crashes due to missing function bodies /
initializers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334535 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64] Support reserving x20 register
Petr Hosek [Tue, 12 Jun 2018 20:00:50 +0000 (20:00 +0000)]
[AArch64] Support reserving x20 register

Register x20 is a callee-saved register which may be used for other
purposes in certain contexts, for example to hold special variables
within the kernel. This change adds support for reserving this register
both to frontend and backend to make this register usable for these
purposes.

Differential Revision: https://reviews.llvm.org/D46552

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334531 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Remove unnecessary include from one of the tblgen emitters.
Craig Topper [Tue, 12 Jun 2018 19:59:10 +0000 (19:59 +0000)]
[X86] Remove unnecessary include from one of the tblgen emitters.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334530 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Remove mayLoad flag from AVX512 truncating store instructions.
Craig Topper [Tue, 12 Jun 2018 19:59:08 +0000 (19:59 +0000)]
[X86] Remove mayLoad flag from AVX512 truncating store instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334529 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MS][ARM64] Hoist __ImageBase handling into TargetLoweringObjectFileCOFF
Reid Kleckner [Tue, 12 Jun 2018 18:56:05 +0000 (18:56 +0000)]
[MS][ARM64] Hoist __ImageBase handling into TargetLoweringObjectFileCOFF

All COFF targets should use @IMGREL32 relocations for symbol differences
against __ImageBase. Do the same for getSectionForConstant, so that
immediates lowered to globals get merged across TUs.

Patch by Chris January

Differential Revision: https://reviews.llvm.org/D47783

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334523 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDHSA/NFC: Code object v3 updates (additional):
Konstantin Zhuravlyov [Tue, 12 Jun 2018 18:33:51 +0000 (18:33 +0000)]
AMDHSA/NFC: Code object v3 updates (additional):

- Move section selection and alignment to AMDGPUAsmPrinter

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334521 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MIR][MachineCSE] Implementing proper MachineInstr::getNumExplicitDefs()
Roman Tereshin [Tue, 12 Jun 2018 18:30:37 +0000 (18:30 +0000)]
[MIR][MachineCSE] Implementing proper MachineInstr::getNumExplicitDefs()

Apparently, MachineInstr class definition as well as pretty much all of
the machine passes assume that the only kind of MachineInstr's operands
that is variadic for variadic opcodes is explicit non-definitions.

In particular, this assumption is made by MachineInstr::defs(), uses(),
and explicit_uses() methods, as well as by MachineCSE pass.

The assumption is incorrect judging from at least TableGen backend
implementation, that recognizes variable_ops in OutOperandList, and the
very existence of G_UNMERGE_VALUES generic opcode, or ARM load multiple
instructions, all of which have variadic defs.

In particular, MachineCSE pass breaks MIR with CSE'able G_UNMERGE_VALUES
instructions in it.

This commit implements MachineInstr::getNumExplicitDefs() similar to
pre-existing MachineInstr::getNumExplicitOperands(), fixes
MachineInstr::defs(), uses(), and explicit_uses(), and fixes MachineCSE
pass.

As the issue addressed seems to affect only machine passes that could be
ran mid-GlobalISel pipeline at the moment, the other passes aren't fixed
by this commit, like MachineLICM: that could be done on per-pass basis
when (if ever) they get adopted for GlobalISel.

Reviewed By: arsenm

Differential Revision: https://reviews.llvm.org/D45640

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334520 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDHSA: Code object v3 updates
Konstantin Zhuravlyov [Tue, 12 Jun 2018 18:02:46 +0000 (18:02 +0000)]
AMDHSA: Code object v3 updates

- Do not emit following assembler directives:
  - .hsa_code_object_version
  - .hsa_code_object_isa
  - .amd_amdgpu_isa
  - .amd_amdgpu_hsa_metadata
  - .amd_amdgpu_pal_metadata
- Do not emit .note entries
- Cleanup and bring in sync kernel descriptor header file
- Emit kernel descriptor into .rodata with appropriate relocations and
  alignments

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334519 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRefactor ExecuteAndWait to take StringRefs.
Zachary Turner [Tue, 12 Jun 2018 17:43:52 +0000 (17:43 +0000)]
Refactor ExecuteAndWait to take StringRefs.

This simplifies some code which had StringRefs to begin with, and
makes other code more complicated which had const char* to begin
with.

In the end, I think this makes for a more idiomatic and platform
agnostic API.  Not all platforms launch process with null terminated
c-string arrays for the environment pointer and argv, but the api
was designed that way because it allowed easy pass-through for
posix-based platforms.  There's a little additional overhead now
since on posix based platforms we'll be takign StringRefs which
were constructed from null terminated strings and then copying
them to null terminate them again, but from a readability and
usability standpoint of the API user, I think this API signature
is strictly better.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334518 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MC] [X86] Teach leaq _GLOBAL_OFFSET_TABLE(%rip), %r15 to use R_X86_64_GOTPC32 instea...
Fangrui Song [Tue, 12 Jun 2018 16:20:44 +0000 (16:20 +0000)]
[MC] [X86] Teach leaq _GLOBAL_OFFSET_TABLE(%rip), %r15 to use R_X86_64_GOTPC32 instead of R_X86_64_PC32

Summary:
This is similar to D46319 (ARM). x86-64 psABI p40 gives an example:

  leaq _GLOBAL_OFFSET_TABLE(%rip), %r15 # GOTPC32 reloc

GNU as creates R_X86_64_GOTPC32. However, MC currently emits R_X86_64_PC32.

Reviewers: javed.absar, echristo

Subscribers: kristof.beyls, llvm-commits, peter.smith, grimar

Differential Revision: https://reviews.llvm.org/D47507

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334515 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoUtilize new SDNode flag functionality to expand current support for fmul
Michael Berg [Tue, 12 Jun 2018 16:13:11 +0000 (16:13 +0000)]
Utilize new SDNode flag functionality to expand current support for fmul

Summary: This patch originated from D46562 and is a proper subset, with some issues addressed for fmul.

Reviewers: spatel, hfinkel, wristow, arsenm

Reviewed By: spatel

Subscribers: nhaehnle, wdng

Differential Revision: https://reviews.llvm.org/D47911

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334514 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[CostModel] Replace ShuffleKind::SK_Alternate with ShuffleKind::SK_Select (PR33744)
Simon Pilgrim [Tue, 12 Jun 2018 16:12:29 +0000 (16:12 +0000)]
[CostModel] Replace ShuffleKind::SK_Alternate with ShuffleKind::SK_Select (PR33744)

As discussed on PR33744, this patch relaxes ShuffleKind::SK_Alternate which requires shuffle masks to only match an alternating pattern from its 2 sources:

e.g. v4f32: <0,5,2,7> or <4,1,6,3>

This seems far too restrictive as most SIMD hardware which will implement it using a general blend/bit-select instruction, so replaces it with SK_Select, permitting elements from either source as long as they are inline:

e.g. v4f32: <0,5,2,7>, <4,1,6,3>, <0,1,6,7>, <4,1,2,3> etc.

This initial patch just updates the name and cost model shuffle mask analysis, later patch reviews will update SLP to better utilise this - it still limits itself to SK_Alternate style patterns.

Differential Revision: https://reviews.llvm.org/D47985

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334513 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DWARFv5] llvm-mc -dwarf-version does not imply -g.
Paul Robinson [Tue, 12 Jun 2018 16:09:03 +0000 (16:09 +0000)]
[DWARFv5] llvm-mc -dwarf-version does not imply -g.

Don't provide the assembler source as the "root file" unless the user
asked to have debug info for the assembler source (with -g).

If the source doesn't provide an explicit ".file 0" then (a) use the
compilation directory as directory #0, and (b) use the file #1 info
for file #0 also.

Differential Revision: https://reviews.llvm.org/D48055

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334512 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Remove TB_ALIGN_16 from VEXTRACTF128/VEXTRACTI128 in the memory folding table.
Craig Topper [Tue, 12 Jun 2018 15:48:03 +0000 (15:48 +0000)]
[X86] Remove TB_ALIGN_16 from VEXTRACTF128/VEXTRACTI128 in the memory folding table.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334511 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix signed/unsigned warning. NFCI.
Simon Pilgrim [Tue, 12 Jun 2018 15:14:34 +0000 (15:14 +0000)]
Fix signed/unsigned warning. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334509 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Make floating point operations expensive for vectorization
Krzysztof Parzyszek [Tue, 12 Jun 2018 15:12:50 +0000 (15:12 +0000)]
[Hexagon] Make floating point operations expensive for vectorization

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334508 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[CostModel] Treat Identity shuffle masks as zero cost
Simon Pilgrim [Tue, 12 Jun 2018 14:47:13 +0000 (14:47 +0000)]
[CostModel] Treat Identity shuffle masks as zero cost

As discussed on D47985, identity shuffle masks should probably be free.

I've limited this to the case where the input and output types all match - but we could probably accept all cases.

Differential Revision: https://reviews.llvm.org/D47986

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334506 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[x86] move shrunkblend transform to helper function; NFCI
Sanjay Patel [Tue, 12 Jun 2018 14:21:51 +0000 (14:21 +0000)]
[x86] move shrunkblend transform to helper function; NFCI

We should be able to obsolete D48043 by easing the constraints
on this existing code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334504 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-exegesis] Sum counter values when several counters are specified for a ProcRes.
Clement Courbet [Tue, 12 Jun 2018 13:28:37 +0000 (13:28 +0000)]
[llvm-exegesis] Sum counter values when several counters are specified for a ProcRes.

Summary: This allows handling memory ports on SNB.

Reviewers: gchatelet

Subscribers: tschuett, llvm-commits

Differential Revision: https://reviews.llvm.org/D48076

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334502 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-exegesis] Move libpfm linking to LLVMExegesis.
Guillaume Chatelet [Tue, 12 Jun 2018 13:07:16 +0000 (13:07 +0000)]
[llvm-exegesis] Move libpfm linking to LLVMExegesis.

Summary: This patch moves linking of libpfm from different places to a single one.

Reviewers: courbet

Subscribers: mgorny, tschuett, llvm-commits

Differential Revision: https://reviews.llvm.org/D48075

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334499 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SelectionDAG] Provide default expansion for rotates
Krzysztof Parzyszek [Tue, 12 Jun 2018 12:49:36 +0000 (12:49 +0000)]
[SelectionDAG] Provide default expansion for rotates

Implement default legalization of rotates: either in terms of the rotation
in the opposite direction (if legal), or in terms of shifts and ors.

Implement generating of rotate instructions for Hexagon. Hexagon only
supports rotates by an immediate value, so implement custom lowering of
ROTL/ROTR on Hexagon. If a rotate is not legal, use the default expansion.

Differential Revision: https://reviews.llvm.org/D47725

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334497 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoUse SmallPtrSet explicitly for SmallSets with pointer types (NFC).
Florian Hahn [Tue, 12 Jun 2018 11:16:56 +0000 (11:16 +0000)]
Use SmallPtrSet explicitly for SmallSets with pointer types (NFC).

Currently SmallSet<PointerTy> inherits from SmallPtrSet<PointerTy>. This
patch replaces such types with SmallPtrSet, because IMO it is slightly
clearer and allows us to get rid of unnecessarily including SmallSet.h

Reviewers: dblaikie, craig.topper

Reviewed By: dblaikie

Differential Revision: https://reviews.llvm.org/D47836

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334492 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips] Guard some floating point instructions correctly
Simon Dardis [Tue, 12 Jun 2018 10:28:06 +0000 (10:28 +0000)]
[mips] Guard some floating point instructions correctly

Reviewers: smaksimovic, atanasyan, abeserminji

Differential Revision: https://reviews.llvm.org/D47636

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334491 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips] Extend LONG_BRANCH_LUi/ADDiu with extra parameter
Aleksandar Beserminji [Tue, 12 Jun 2018 10:23:49 +0000 (10:23 +0000)]
[mips] Extend LONG_BRANCH_LUi/ADDiu with extra parameter

Extend LONG_BRANCH_LUi and LONG_BRANCH_ADDiu pseudo instructions with
additional flag, so instead of always lowering to lui %hi(...),
addiu %lo(...) or addiu %hi(...), now they can lower to either %lo, %hi,
%higher or %highest depending on the added flag.

Differential Revision: https://reviews.llvm.org/D47941

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334490 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64] Audit on rL333879 to fix FP16 64bit bitpatterns
Luke Geeson [Tue, 12 Jun 2018 09:35:20 +0000 (09:35 +0000)]
[AArch64] Audit on rL333879 to fix FP16 64bit bitpatterns

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334488 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[CostModel][X86] Add extra Identity shuffle mask cost tests (D47986)
Simon Pilgrim [Tue, 12 Jun 2018 09:18:13 +0000 (09:18 +0000)]
[CostModel][X86] Add extra Identity shuffle mask cost tests (D47986)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334486 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add NotMemoryFoldable to the VPCOMPRESS instructions.
Craig Topper [Tue, 12 Jun 2018 07:32:19 +0000 (07:32 +0000)]
[X86] Add NotMemoryFoldable to the VPCOMPRESS instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334481 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Don't add stores to the autogenerated load folding tables if the register and...
Craig Topper [Tue, 12 Jun 2018 07:32:18 +0000 (07:32 +0000)]
[X86] Don't add stores to the autogenerated load folding tables if the register and memory operands have different widths.

This can cause the amount of the memory written to be changed which would be bad.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334480 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add NotMemoryFoldable to more instructions.
Craig Topper [Tue, 12 Jun 2018 07:32:17 +0000 (07:32 +0000)]
[X86] Add NotMemoryFoldable to more instructions.

These include PUSH/POP instructions that don't match the manual table. This also includes CMPXCHG which we never emit in non-locked form.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334479 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Update folding table generator to properly detect RMW arithmetic instructions.
Craig Topper [Tue, 12 Jun 2018 07:32:15 +0000 (07:32 +0000)]
[X86] Update folding table generator to properly detect RMW arithmetic instructions.

The RMW instructions are detected by reading the SchedRW data, but the RMW instructions have had their SchedRW changed in recent months. This broke the expectation.

We probably should fix this to use the mayLoad/mayStore flags if possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334478 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[NFC] Change sample profile format enum name SPF_Raw_Binary to SPF_Binary.
Wei Mi [Tue, 12 Jun 2018 05:53:49 +0000 (05:53 +0000)]
[NFC] Change sample profile format enum name SPF_Raw_Binary to SPF_Binary.

Some out-of-tree targets depend on the enum name SPF_Binary. Keep the name
can avoid unnecessary churn to those targets.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334476 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix a typo in rL334447.
Wei Mi [Tue, 12 Jun 2018 04:43:09 +0000 (04:43 +0000)]
Fix a typo in rL334447.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334475 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add NotMemoryFoldable to a bunch of instructions to suppress them from the...
Craig Topper [Tue, 12 Jun 2018 04:34:59 +0000 (04:34 +0000)]
[X86] Add NotMemoryFoldable to a bunch of instructions to suppress them from the autogenerated load folding table.

Most of these are system instructions or other instructions we don't use in CodeGen. No point wasting space for them in the table. Removing them from the autogenerated table makes it easier to review the manual table.

A few are real opcode collisions where the memory and register forms are completely different instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334474 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Enforce agreement of AdSize field when autogenerating load folding table entries.
Craig Topper [Tue, 12 Jun 2018 04:34:58 +0000 (04:34 +0000)]
[X86] Enforce agreement of AdSize field when autogenerating load folding table entries.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334473 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoNFC, some additional tests added and some renaming for planned fma support changes
Michael Berg [Tue, 12 Jun 2018 00:52:43 +0000 (00:52 +0000)]
NFC, some additional tests added and some renaming for planned fma support changes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334461 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add isel patterns for folding loads when creating ROUND instructions from ffloo...
Craig Topper [Tue, 12 Jun 2018 00:48:57 +0000 (00:48 +0000)]
[X86] Add isel patterns for folding loads when creating ROUND instructions from ffloor/fnearbyint/fceil/frint/ftrunc.

We were missing packed isel folding patterns for all of sse41, avx, and avx512.

For some reason avx512 had scalar load folding patterns under optsize(due to partial/undef reg update), but we didn't have the equivalent sse41 and avx patterns.

Sometimes we would get load folding due to peephole pass anyway, but we're also missing avx512 instructions from the load folding table. I'll try to fix that in another patch.

Some of this was spotted in the review for D47993.

This patch adds all the folds to isel, adds a few spot tests, and disables the peephole pass on a few tests to ensure we're testing some of these patterns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334460 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AMDGPU] prevent hitting Assertion `isReg() && "Wrong MachineOperand accessor"'
Mark Searles [Tue, 12 Jun 2018 00:41:26 +0000 (00:41 +0000)]
[AMDGPU] prevent hitting Assertion `isReg() && "Wrong MachineOperand accessor"'

The use iterator, used within findMaskOperands(), can return anything which is
not a def. isUse() requires a register, so check isReg() before calling isUse().

Differential Revision: https://reviews.llvm.org/D48047

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334459 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix a buildbot error reported by sanitizer-x86_64-linux-fast:
Wei Mi [Mon, 11 Jun 2018 23:39:02 +0000 (23:39 +0000)]
Fix a buildbot error reported by sanitizer-x86_64-linux-fast:
Function::getGUID may complain if its input is an empty StringRef.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334455 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix a warning issued by clang.
Wei Mi [Mon, 11 Jun 2018 23:09:04 +0000 (23:09 +0000)]
Fix a warning issued by clang.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334453 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[CMake] Fix dropped dependency in install-llvm-headers
Justin Bogner [Mon, 11 Jun 2018 23:05:28 +0000 (23:05 +0000)]
[CMake] Fix dropped dependency in install-llvm-headers

This dependency was accidentally dropped in r319480, causing
install-distribution and install-llvm-headers to install an incomplete
set of headers (the generated Intrinsics and Attributes would be
missing).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334452 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoSimplify; NFC
George Burgess IV [Mon, 11 Jun 2018 22:58:32 +0000 (22:58 +0000)]
Simplify; NFC

Not shown in the diff: AQ is a `vector<SUnit *>`, and SU is a `SUnit *`

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334451 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix a warning reported by clang but not by gcc.
Wei Mi [Mon, 11 Jun 2018 22:51:28 +0000 (22:51 +0000)]
Fix a warning reported by clang but not by gcc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334449 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SampleFDO] Add a new compact binary format for sample profile.
Wei Mi [Mon, 11 Jun 2018 22:40:43 +0000 (22:40 +0000)]
[SampleFDO] Add a new compact binary format for sample profile.

Name table occupies a big chunk of size in current binary format sample profile.
In order to reduce its size, the patch changes the sample writer/reader to
save/restore MD5Hash of names in the name table. Sample annotation phase will
also use MD5Hash of name to query samples accordingly.

Experiment shows compact binary format can reduce the size of sample profile by
2/3 compared with binary format generally.

Differential Revision: https://reviews.llvm.org/D47955

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334447 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoTableGen: Change some pointer parameters to references since they're never null anyway
David Blaikie [Mon, 11 Jun 2018 22:14:43 +0000 (22:14 +0000)]
TableGen: Change some pointer parameters to references since they're never null anyway

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334446 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU: Add 64-bit relative variant kind
Konstantin Zhuravlyov [Mon, 11 Jun 2018 21:37:57 +0000 (21:37 +0000)]
AMDGPU: Add 64-bit relative variant kind

Differential Revision: https://reviews.llvm.org/D47601

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334443 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoDAG: Fix extract_subvector combine for a single element
Matt Arsenault [Mon, 11 Jun 2018 21:27:41 +0000 (21:27 +0000)]
DAG: Fix extract_subvector combine for a single element

This would fail before because 1x vectors aren't legal,
so instead just use the scalar type.

Avoids regressions in a future AMDGPU commit to add
v4i16/v4f16 as legal types.

Test update is just the one test that this triggers
on in tree now. It wasn't checking anything before.
The result is completely  changed since the selects
are eliminated. Not sure if it's considered better
or not.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334440 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Push some variable declarations down into the individual switch cases that...
Craig Topper [Mon, 11 Jun 2018 20:50:58 +0000 (20:50 +0000)]
[X86] Push some variable declarations down into the individual switch cases that need them. NFC

All of the cases are already wrapped in curly braces so declaring a variable there isn't an issue. And the variables aren't assigned or used in the larger scope.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334436 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SLP] Add testcases of min/max reduction pattern for AMDGPU.
Farhana Aleen [Mon, 11 Jun 2018 20:29:31 +0000 (20:29 +0000)]
[SLP] Add testcases of min/max reduction pattern for AMDGPU.

Author: FarhanaAleen

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334435 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix incorrect CHECK-LABEL
Tim Shen [Mon, 11 Jun 2018 19:56:12 +0000 (19:56 +0000)]
Fix incorrect CHECK-LABEL

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334434 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Reorder some type constraints to force things to be vectors and integer/fp...
Craig Topper [Mon, 11 Jun 2018 19:20:15 +0000 (19:20 +0000)]
[X86] Reorder some type constraints to force things to be vectors and integer/fp before forcing them to be the same size.

This may be needed by another patch that I'm working on. It should have no effect on any of the generated outputs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334430 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SCEV] Add transform zext((A * B * ...)<nuw>) --> (zext(A) * zext(B) * ...)<nuw>.
Justin Lebar [Mon, 11 Jun 2018 18:57:58 +0000 (18:57 +0000)]
[SCEV] Add transform zext((A * B * ...)<nuw>) --> (zext(A) * zext(B) * ...)<nuw>.

Reviewers: sanjoy

Subscribers: hiraditya, llvm-commits

Differential Revision: https://reviews.llvm.org/D48041

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334429 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SCEV] Add nuw/nsw to mul ops in StrengthenNoWrapFlags where safe.
Justin Lebar [Mon, 11 Jun 2018 18:57:42 +0000 (18:57 +0000)]
[SCEV] Add nuw/nsw to mul ops in StrengthenNoWrapFlags where safe.

Summary:
Previously we would add them for adds, but not multiplies.

Reviewers: sanjoy

Subscribers: llvm-commits, hiraditya

Differential Revision: https://reviews.llvm.org/D48038

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334428 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix indentation in ScalarEvolution.cpp.
Justin Lebar [Mon, 11 Jun 2018 18:57:27 +0000 (18:57 +0000)]
Fix indentation in ScalarEvolution.cpp.

Whitespace-only change.  (clang-formatted the whole block.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334427 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Late predicate producers cannot be used as dot-new sources
Krzysztof Parzyszek [Mon, 11 Jun 2018 18:45:52 +0000 (18:45 +0000)]
[Hexagon] Late predicate producers cannot be used as dot-new sources

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@334426 91177308-0d34-0410-b5e6-96231b3b80d8