OSDN Git Service

android-x86/external-llvm.git
6 years agoExpose must/may alias info in MemorySSA.
Alina Sbirlea [Thu, 8 Mar 2018 18:03:14 +0000 (18:03 +0000)]
Expose must/may alias info in MemorySSA.

Summary:
Building MemorySSA gathers alias information for Defs/Uses.
Store and expose this information when optimizing uses (when building MemorySSA),
and when optimizing defs or updating uses (getClobberingMemoryAccess).
Current patch does not propagate alias information through MemoryPhis.

Reviewers: gbiv, dberlin

Subscribers: Prazek, sanjoy, llvm-commits

Differential Revision: https://reviews.llvm.org/D38569

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327035 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AMDGPU] fix test to survive the most basic undef constant folding
Sanjay Patel [Thu, 8 Mar 2018 17:34:25 +0000 (17:34 +0000)]
[AMDGPU] fix test to survive the most basic undef constant folding

This will likely need to be changed again for anything more than:
fmul undef, undef -> undef

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327034 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert r327029
Kuba Mracek [Thu, 8 Mar 2018 17:32:00 +0000 (17:32 +0000)]
Revert r327029

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327033 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[x86] fix test to be independent of FP undef
Sanjay Patel [Thu, 8 Mar 2018 17:24:30 +0000 (17:24 +0000)]
[x86] fix test to be independent of FP undef

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327030 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[asan] Fix a false positive ODR violation due to LTO ConstantMerge pass [llvm part]
Kuba Mracek [Thu, 8 Mar 2018 17:24:06 +0000 (17:24 +0000)]
[asan] Fix a false positive ODR violation due to LTO ConstantMerge pass [llvm part]

This fixes a false positive ODR violation that is reported by ASan when using LTO. In cases, where two constant globals have the same value, LTO will merge them, which breaks ASan's ODR detection.

Differential Revision: https://reviews.llvm.org/D43959

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327029 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[StructurizeCFG] fix test to be independent of FP undef
Sanjay Patel [Thu, 8 Mar 2018 17:13:57 +0000 (17:13 +0000)]
[StructurizeCFG] fix test to be independent of FP undef

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327028 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-mca] add override keyword to method ResourcePressureView::printView().
Andrea Di Biagio [Thu, 8 Mar 2018 17:02:28 +0000 (17:02 +0000)]
[llvm-mca] add override keyword to method ResourcePressureView::printView().
NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327027 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[x86] regenerate checks; NFC
Sanjay Patel [Thu, 8 Mar 2018 16:56:49 +0000 (16:56 +0000)]
[x86] regenerate checks; NFC

This test will fail if we fix FP undef constant folding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327026 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-mca] HWEventListener is a class, not struct.
Andrea Di Biagio [Thu, 8 Mar 2018 16:34:19 +0000 (16:34 +0000)]
[llvm-mca] HWEventListener is a class, not struct.

This should appease the buildbots.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327025 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agomerge-request.sh: Update 6.0 metabug for 6.0.1
Tom Stellard [Thu, 8 Mar 2018 16:29:08 +0000 (16:29 +0000)]
merge-request.sh: Update 6.0 metabug for 6.0.1

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327024 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[StructurizeCFG] auto-generate full checks; NFC
Sanjay Patel [Thu, 8 Mar 2018 16:25:37 +0000 (16:25 +0000)]
[StructurizeCFG] auto-generate full checks; NFC

Not sure what the intent of this test is, but this will change when we fix FP undef constant folding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327022 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Power9] Add more missing instructions to the Power 9 scheduler
Stefan Pintilie [Thu, 8 Mar 2018 16:24:33 +0000 (16:24 +0000)]
[Power9] Add more missing instructions to the Power 9 scheduler

With this patch we should be able to mark the Power 9 model as complete.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327021 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAMDGPU/GlobalISel: Pass subtarget + TM to LegalizerInfo
Matt Arsenault [Thu, 8 Mar 2018 16:24:16 +0000 (16:24 +0000)]
AMDGPU/GlobalISel: Pass subtarget + TM to LegalizerInfo

These are the parameters x86 already uses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327020 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-mca] Unify the API for the various views. NFCI
Andrea Di Biagio [Thu, 8 Mar 2018 16:08:43 +0000 (16:08 +0000)]
[llvm-mca] Unify the API for the various views. NFCI

This allows the customization of the performance report.

Users can specify their own custom sequence of views.
Each view contributes a portion of the performance report generated by the
BackendPrinter.

Internally, class BackendPrinter keeps a sequence of views; views are printed
out in sequence when method 'printReport()' is called.

This patch addresses one of the two review comments from Clement in D43951.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327018 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix unused function warning in StatisticTest.cpp
Daniel Sanders [Thu, 8 Mar 2018 15:52:45 +0000 (15:52 +0000)]
Fix unused function warning in StatisticTest.cpp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327015 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[InstCombine] regenerate checks; NFC
Sanjay Patel [Thu, 8 Mar 2018 15:46:38 +0000 (15:46 +0000)]
[InstCombine] regenerate checks; NFC

We may not need any of these tests after rL327012, but leaving
them here for now until that's confirmed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327014 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[InstSimplify] add more tests for FP undef; NFC
Sanjay Patel [Thu, 8 Mar 2018 15:39:39 +0000 (15:39 +0000)]
[InstSimplify] add more tests for FP undef; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327012 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoDWARFVerifier: Basic verification of .debug_names
Pavel Labath [Thu, 8 Mar 2018 15:34:42 +0000 (15:34 +0000)]
DWARFVerifier: Basic verification of .debug_names

Summary:
This patch adds basic .debug_names verification capabilities to the
DWARF verifier. Right now, it checks that the headers and abbreviation
tables of the individual name indexes can be parsed correctly, it
verifies the buckets table and the cross-checks the CU lists for
consistency. I intend to add further checks in follow-up patches.

Reviewers: JDevlieghere, aprantl, probinson, dblaikie

Subscribers: vleschuk, echristo, clayborg, llvm-commits

Differential Revision: https://reviews.llvm.org/D44211

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327011 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-mca] Emit the 'Instruction Info' table before the resource pressure view.
Andrea Di Biagio [Thu, 8 Mar 2018 15:34:38 +0000 (15:34 +0000)]
[llvm-mca] Emit the 'Instruction Info' table before the resource pressure view.

In future, both the summary information and the 'instruction info' table should
be moved into a separate "Summary" view.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327010 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[cmake] Append -Wl,-rpath-link conditionally to GNULD
Michal Gorny [Thu, 8 Mar 2018 15:09:38 +0000 (15:09 +0000)]
[cmake] Append -Wl,-rpath-link conditionally to GNULD

Append -Wl,-rpath-link conditionally to whether GNU ld.bfd is used
rather than the Linux+!gold conditionals. Also move it out of 'else'
branch of *BSD handling. This fixes build failures with ld.bfd
on Gentoo/FreeBSD, and should cause no harm on other systems using
ld.bfd.

This patch improves the original logic by reusing results of linker
detection introduced in r307852.

Differential Revision: https://reviews.llvm.org/D43751

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327007 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[InstCombine, NewGVN] remove FP undef from tests
Sanjay Patel [Thu, 8 Mar 2018 14:57:08 +0000 (14:57 +0000)]
[InstCombine, NewGVN] remove FP undef from tests

I'm trying to preserve the intent of these tests by using
non-undef operands; if we fix FP undef folding these tests
will not pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327004 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[CMake] Add missing test dependency
David Zarzycki [Thu, 8 Mar 2018 14:43:24 +0000 (14:43 +0000)]
[CMake] Add missing test dependency

This makes 'ninja clean ; ninja check-all' work again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327002 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoUse ellipsis ... to indicate omitted commands
Ed Maste [Thu, 8 Mar 2018 13:52:04 +0000 (13:52 +0000)]
Use ellipsis ... to indicate omitted commands

In an example like "clang -fxray-instrument .." the .. could be confused
with a literal .. (parent directory), which is used in commands like
"cmake -GNinja .."

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327000 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAdd llvm-mca.rst to the table of contents in docs/CommandGuide.
Andrea Di Biagio [Thu, 8 Mar 2018 13:43:11 +0000 (13:43 +0000)]
Add llvm-mca.rst to the table of contents in docs/CommandGuide.

This should fix the documentation error reported by builder llvm-sphinx-docs
(build #16407) after r326998.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326999 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-mca] LLVM Machine Code Analyzer.
Andrea Di Biagio [Thu, 8 Mar 2018 13:05:02 +0000 (13:05 +0000)]
[llvm-mca] LLVM Machine Code Analyzer.

llvm-mca is an LLVM based performance analysis tool that can be used to
statically measure the performance of code, and to help triage potential
problems with target scheduling models.

llvm-mca uses information which is already available in LLVM (e.g. scheduling
models) to statically measure the performance of machine code in a specific cpu.
Performance is measured in terms of throughput as well as processor resource
consumption. The tool currently works for processors with an out-of-order
backend, for which there is a scheduling model available in LLVM.

The main goal of this tool is not just to predict the performance of the code
when run on the target, but also help with diagnosing potential performance
issues.

Given an assembly code sequence, llvm-mca estimates the IPC (instructions per
cycle), as well as hardware resources pressure. The analysis and reporting style
were mostly inspired by the IACA tool from Intel.

This patch is related to the RFC on llvm-dev visible at this link:
http://lists.llvm.org/pipermail/llvm-dev/2018-March/121490.html

Differential Revision: https://reviews.llvm.org/D43951

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326998 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DWARF] Don't attempt to parse line tables at invalid offsets
James Henderson [Thu, 8 Mar 2018 10:53:34 +0000 (10:53 +0000)]
[DWARF] Don't attempt to parse line tables at invalid offsets

Whilst working on improvements to the error handling of the debug line
parsing code, I noticed that if an invalid offset were to be specified
in a call to getOrParseLineTable(), an entry in the LineTableMap would
still be created, even if the offset was not within the section range.
The immediate parsing attempt afterwards would fail (it would end up
getting a version of 0), and thereafter, any subsequent calls to
getOrParseLineTable or getLineTable would return the default-
constructed, invalid line table. In reality, we shouldn't even attempt
to parse this table, and we should always return a nullptr from these
two functions for this situation.

I have tested this via a unit test, which required some new framework
for unit testing debug line. My plan is to add quite a few more unit
tests for the new error reporting mechanism that will follow shortly,
hence the reason why the supporting code for the tests are written the
way they are - I intend to extend the DwarfGenerator class to support
generating debug line. At that point, I'll make sure that there are a
few positive test cases for this and the parsing code too.

Differential Revision: https://reviews.llvm.org/D44200

Reviewers: JDevlieghere, aprantl

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326995 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[dsymutil] Embed toolchain in dSYM bundle
Jonas Devlieghere [Thu, 8 Mar 2018 10:39:12 +0000 (10:39 +0000)]
[dsymutil] Embed toolchain in dSYM bundle

Allow us to embed the (Xcode) toolchain in the dSYM bundle's property
list.

Differential revision: https://reviews.llvm.org/D44151

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326994 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MCSchedule] Always generate processor resource names.
Andrea Di Biagio [Thu, 8 Mar 2018 10:38:45 +0000 (10:38 +0000)]
[MCSchedule] Always generate processor resource names.

With this patch, the tablegen 'SubtargetEmitter' always generates processor
resource names.
The impact of this patch on the code size of other llvm tools is small.  I have
observed an average increase of 0.03% in code size when doing a release build of
LLVM (on windows, using MSVC) with all the default backends.

This change is done in preparation for the upcoming llvm-mca patch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326993 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Change X86::PMULDQ/PMULUDQ opcodes to take vXi64 type as input instead of vXi32.
Craig Topper [Thu, 8 Mar 2018 08:02:52 +0000 (08:02 +0000)]
[X86] Change X86::PMULDQ/PMULUDQ opcodes to take vXi64 type as input instead of vXi32.

This instruction can be thought of as reading either the even elements of a vXi32 input or the lower half of each element of a vXi64 input. We currently use the vXi32 interpretation, but vXi64 matches better with its broadcast behavior in EVEX.

I'm looking at moving MULDQ/MULUDQ creation to a DAG combine so we can do it when AVX512DQ is enabled without having to go through Custom lowering. But in some of the test cases we failed to use a broadcast load due to the size difference. This should help with that.

I'm also wondering if we can model these instructions in native IR and remove the intrinsics and I think using a vXi64 type will work better with that.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326991 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AMDGPU] Update AMDGOUUsage.rst descriptions
Tony Tye [Thu, 8 Mar 2018 05:46:01 +0000 (05:46 +0000)]
[AMDGPU] Update AMDGOUUsage.rst descriptions

- Improve description of XNACK ELF flag.
- Rename all uses of wave to wavefront to be consistent.

Differential Revision: https://reviews.llvm.org/D43983

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326989 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Add except_ref as a first-class type
Heejin Ahn [Thu, 8 Mar 2018 04:05:37 +0000 (04:05 +0000)]
[WebAssembly] Add except_ref as a first-class type

Summary: Add except_ref as a first-class type, according to the [[https://github.com/WebAssembly/exception-handling/blob/master/proposals/Level-1.md | Level 1 exception handling proposal ]].

Reviewers: dschuff

Subscribers: jfb, sbc100, llvm-commits

Differential Revision: https://reviews.llvm.org/D43706

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326985 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Add IntrNoReturn property to throw/rethrow intrinsics
Heejin Ahn [Thu, 8 Mar 2018 03:47:52 +0000 (03:47 +0000)]
[WebAssembly] Add IntrNoReturn property to throw/rethrow intrinsics

Reviewers: dschuff

Subscribers: jfb, sbc100, jgravelle-google, sunfish, llvm-commits

Differential Revision: https://reviews.llvm.org/D43740

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326984 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoSupport resetting STATISTIC() values using llvm::ResetStatistics()
Daniel Sanders [Thu, 8 Mar 2018 02:36:25 +0000 (02:36 +0000)]
Support resetting STATISTIC() values using llvm::ResetStatistics()

Summary:
Most of the time, compiler statistics can be obtained using a process that
performs a single compilation and terminates such as llc. However, this isn't
always the case. JITs for example, perform multiple compilations over their
lifetime and STATISTIC() will record cumulative values across all of them.

Provide tools like this with the facilities needed to measure individual
compilations by allowing them to reset the STATISTIC() values back to zero using
llvm::ResetStatistics(). It's still the tools responsibility to ensure that they
perform compilations in such a way that the results are meaningful to their
intended use.

Reviewers: qcolombet, rtereshin, bogner, aditya_nandakumar

Reviewed By: bogner

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D44181

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326981 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAdd attributes and fix some keywords in llvm-mode.el
Fangrui Song [Thu, 8 Mar 2018 01:28:45 +0000 (01:28 +0000)]
Add attributes and fix some keywords in llvm-mode.el

Reviewers: rafael, echristo

Reviewed By: echristo

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D44135

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326978 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "[LTO] Support filtering by hotness threshold"
Bob Haarman [Thu, 8 Mar 2018 01:13:10 +0000 (01:13 +0000)]
Revert "[LTO] Support filtering by hotness threshold"

This reverts commit 1f3bd185c53beb6aa68446974b7e80837abd6ef0 (r326107)
because it fails
ThinLTO/X86/diagnostic-handler-remarks-with-hotness.ll.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326975 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoReland "[DebugInfo] Support DWARF expressions in eh_frame"
Rafael Auler [Thu, 8 Mar 2018 00:46:53 +0000 (00:46 +0000)]
Reland "[DebugInfo] Support DWARF expressions in eh_frame"

Summary:
Original change was D43313 (r326932) and reverted by r326953 because it
broke an LLD test and a windows build. The LLD test was already fixed in
lld commit r326944 (thanks maskray). This is the original change with
the windows build fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326970 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64] Fix UB about shift amount exceeds data bit-width
Weiming Zhao [Thu, 8 Mar 2018 00:28:25 +0000 (00:28 +0000)]
[AArch64] Fix UB about shift amount exceeds data bit-width

Summary:
Fixes an UB caught by sanitizer. The shift amount might be larger than 32 so the operand should be 1ULL.
In this patch,  we replace the original expression with  existing API with uint64_t type.

Reviewers: eli.friedman, rengolin

Reviewed By: rengolin

Subscribers: rengolin, javed.absar, llvm-commits, kristof.beyls

Differential Revision: https://reviews.llvm.org/D44234

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326969 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Fix some isel patterns that used aligned vector load instructions with unaligne...
Craig Topper [Thu, 8 Mar 2018 00:21:17 +0000 (00:21 +0000)]
[X86] Fix some isel patterns that used aligned vector load instructions with unaligned predicates.

These patterns weren't checking the alignment of the load, but were using the aligned instructions. This will cause a GP fault if the data isn't aligned.

I believe these were introduced in r312450.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326967 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoDelete code that is probably dead since r249303.
Rafael Espindola [Thu, 8 Mar 2018 00:17:13 +0000 (00:17 +0000)]
Delete code that is probably dead since r249303.

With r249303 the expression evaluation should expand variables that
are not in sections (and so don't have an atom).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326966 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix build broken by r326959
Eugene Zemtsov [Thu, 8 Mar 2018 00:07:26 +0000 (00:07 +0000)]
Fix build broken by r326959

Adding Demangle to link time dependencies of Symbolize

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326964 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][SSE] LowerBUILD_VECTORAsVariablePermute - reorder permute types. NFCI.
Simon Pilgrim [Wed, 7 Mar 2018 23:56:42 +0000 (23:56 +0000)]
[X86][SSE] LowerBUILD_VECTORAsVariablePermute - reorder permute types. NFCI.

Reorder into 128/256/512 bit vector size groupings.

NFCI commit before some new features.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326963 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoUse itaniumDemangle in llvm-symbolizer
Eugene Zemtsov [Wed, 7 Mar 2018 23:07:34 +0000 (23:07 +0000)]
Use itaniumDemangle in llvm-symbolizer

Currently on Windows (_MSC_VER) LLVMSymbolizer supports only Microsoft mangling.
This fix just explicitly uses itaniumDemangle when mangled name starts with _Z.

Differential Revision: https://reviews.llvm.org/D44192

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326959 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[TTI] add explanatory comments for getArithmeticInstrCost; NFC
Sanjay Patel [Wed, 7 Mar 2018 22:43:08 +0000 (22:43 +0000)]
[TTI] add explanatory comments for getArithmeticInstrCost; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326956 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64] Adjust the cost of integer vector division
Evandro Menezes [Wed, 7 Mar 2018 22:35:32 +0000 (22:35 +0000)]
[AArch64] Adjust the cost of integer vector division

Since there is no instruction for integer vector division, factor in the
cost of singling out each element to be used with the scalar division
instruction.

Differential revision: https://reviews.llvm.org/D43974

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326955 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert r326932: [DebugInfo] Support DWARF expressions in eh_frame
Rui Ueyama [Wed, 7 Mar 2018 22:29:48 +0000 (22:29 +0000)]
Revert r326932: [DebugInfo] Support DWARF expressions in eh_frame

This reverts commit rr326932 because it broke lld/test/ELF/eh-frame-hdr-augmentation.s.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326953 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64] add missing pattern for insert_subvector undef
Sebastian Pop [Wed, 7 Mar 2018 22:07:13 +0000 (22:07 +0000)]
[AArch64] add missing pattern for insert_subvector undef

The attached testcase started failing after the patch to define
isExtractSubvectorCheap with the following pattern mismatch:

ISEL: Starting pattern match
  Initial Opcode index to 85068
    Match failed at index 85076
    LLVM ERROR: Cannot select: t47: v8i16 = insert_subvector undef:v8i16, t43, Constant:i64<0>

The code generated from llvm/lib/Target/AArch64/AArch64InstrInfo.td

def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i32 0)),
          (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;

is in ninja/lib/Target/AArch64/AArch64GenDAGISel.inc
At the location of the error it is:
/* 85076*/    OPC_CheckChild2Type, MVT::i32,

And it failed to match the type of operand 2.
Adding another def-pat for i64 fixes the failed def-pat error:

def : Pat<(insert_subvector undef, (v4i16 FPR64:$src), (i64 0)),
          (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)), FPR64:$src, dsub)>;

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326949 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "[llvm-objcopy] Add support for large indexes"
Jake Ehrlich [Wed, 7 Mar 2018 20:33:02 +0000 (20:33 +0000)]
Revert "[llvm-objcopy] Add support for large indexes"

Not all build bots have unzip which I used in a test.

This reverts commit 0b1f26d39ea42dd3716b525fbc8c78d8c7bb4479.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326941 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-objcopy] Add support for large indexes
Jake Ehrlich [Wed, 7 Mar 2018 19:59:15 +0000 (19:59 +0000)]
[llvm-objcopy] Add support for large indexes

Because of -ffunction-sections (and maybe other use cases I'm not aware of?) it
can occur that we need more than 0xfeff sections but ELF dosn't support that
many sections. To solve this problem SHN_XINDEX exists and with it come a whole
host of changes for section indexes everywhere. This change adds support for
those cases which should allow llvm-objcopy to copy binaries that have an
arbitrary number of sections.

Differential Revision: https://reviews.llvm.org/D42516

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326940 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Remove unused function argument. NFC
Craig Topper [Wed, 7 Mar 2018 19:45:45 +0000 (19:45 +0000)]
[X86] Remove unused function argument. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326939 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix cmake's multi-config generators after r326738
Daniel Sanders [Wed, 7 Mar 2018 19:32:36 +0000 (19:32 +0000)]
Fix cmake's multi-config generators after r326738

LLVM_ENABLE_STATS isn't known at configure-time in these generators so we must
defer it to build-time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326936 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DebugInfo] Support DWARF expressions in eh_frame
Rafael Auler [Wed, 7 Mar 2018 19:19:51 +0000 (19:19 +0000)]
[DebugInfo] Support DWARF expressions in eh_frame

This patch enhances DWARFDebugFrame with the capability of parsing and
printing DWARF expressions in CFI instructions. It also makes FDEs and
CIEs accessible to lib users, so they can process them in client tools
that rely on LLVM. To make it self-contained with a test case, it
teaches llvm-readobj to be able to dump EH frames and checks they are
correct in a unit test. The llvm-readobj code is Maksim Panchenko's work
(maksfb).

Reviewers: JDevlieghere, espindola

Reviewed By: JDevlieghere

Differential Revision: https://reviews.llvm.org/D43313

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326932 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][SSE] Regenerate float maxnum/minnum tests
Simon Pilgrim [Wed, 7 Mar 2018 19:14:05 +0000 (19:14 +0000)]
[X86][SSE] Regenerate float maxnum/minnum tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326930 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix a bug regarding a mis-identified file type in pdbutil.
Zachary Turner [Wed, 7 Mar 2018 19:12:36 +0000 (19:12 +0000)]
Fix a bug regarding a mis-identified file type in pdbutil.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326929 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoUpdate a few switch statements to handle file_magic::pdb.
Zachary Turner [Wed, 7 Mar 2018 18:58:33 +0000 (18:58 +0000)]
Update a few switch statements to handle file_magic::pdb.

This fixes a couple of warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326927 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Pipeliner] Fixed node order issue related to zero latency edges
Roorda, Jan-Willem [Wed, 7 Mar 2018 18:53:36 +0000 (18:53 +0000)]
[Pipeliner] Fixed node order issue related to zero latency edges

Summary:
A desired property of the node order in Swing Modulo Scheduling is
that for nodes outside circuits the following holds: none of them is
scheduled after both a successor and a predecessor. We call
node orders that meet this property valid.

Although invalid node orders do not lead to the generation of incorrect
code, they can cause the pipeliner not being able to find a pipelined schedule
for arbitrary II. The reason is that after scheduling the successor and the
predecessor of a node, no room may be left to schedule the node itself.

For data flow graphs with 0-latency edges, the node ordering algorithm
of Swing Modulo Scheduling can generate such undesired invalid node orders.
This patch fixes that.

In the remainder of this commit message, I will give an example
demonstrating the issue, explain the fix, and explain how the the fix is tested.

Consider, as an example, the following data flow graph with all
edge latencies 0 and all edges pointing downward.

```
   n0
  /  \
n1    n3
  \  /
   n2
    |
   n4
```

Consider the implemented node order algorithm in top-down mode. In that mode,
the algorithm orders the nodes based on greatest Height and in case of equal
Height on lowest Movability. Finally, in case of equal Height and
Movability, given two nodes with an edge between them, the algorithm prefers
the source-node.

In the graph, for every node, the Height and Movability are equal to 0.
As will be explained below, the algorithm can generate the order n0, n1, n2, n3, n4.
So, node n3 is scheduled after its predecessor n0 and after its successor n2.

The reason that the algorithm can put node n2 in the order before node n3,
even though they have an edge between them in which node n3 is the source,
is the following: Suppose the algorithm has constructed the partial node
order n0, n1. Then, the nodes left to be ordered are nodes n2, n3, and n4. Suppose
that the while-loop in the implemented algorithm considers the nodes in
the order n4, n3, n2. The algorithm will start with node n4, and look for
more preferable nodes. First, node n4 will be compared with node n3. As the nodes
have equal Height and Movability and have no edge between them, the algorithm
will stick with node n4. Then node n4 is compared with node n2. Again the
Height and Movability are equal. But, this time, there is an edge between
the two nodes, and the algorithm will prefer the source node n2.
As there are no nodes left to compare, the algorithm will add node n2 to
the node order, yielding the partial node order n0, n1, n2. In this way node n2
arrives in the node-order before node n3.

To solve this, this patch introduces the ZeroLatencyHeight (ZLH) property
for nodes. It is defined as the maximum unweighted length of a path from the
given node to an arbitrary node in which each edge has latency 0.
So, ZLH(n0)=3, ZLH(n1)=ZLH(n3)=2, ZLH(n2)=1, and ZLH(n4)=0

In this patch, the preference for a greater ZeroLatencyHeight
is added in the top-down mode of the node ordering algorithm, after the
preference for a greater Height, and before the preference for a
lower Movability.

Therefore, the two allowed node-orders are n0, n1, n3, n2, n4 and n0, n3, n1, n2, n4.
Both of them are valid node orders.

In the same way, the bottom-up mode of the node ordering algorithm is adapted
by introducing the ZeroLatencyDepth property for nodes.

The patch is tested by adding extra checks to the following existing
lit-tests:
test/CodeGen/Hexagon/SUnit-boundary-prob.ll
test/CodeGen/Hexagon/frame-offset-overflow.ll
test/CodeGen/Hexagon/vect/vect-shuffle.ll

Before this patch, the pipeliner failed to pipeline the loops in these tests
due to invalid node-orders. After the patch, the pipeliner successfully
pipelines all these loops.

Reviewers: bcahoon

Reviewed By: bcahoon

Subscribers: Ayal, mgrang, llvm-commits

Differential Revision: https://reviews.llvm.org/D43620

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326925 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoTeach identify_file_magic to identify PDB files.
Zachary Turner [Wed, 7 Mar 2018 18:40:41 +0000 (18:40 +0000)]
Teach identify_file_magic to identify PDB files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326924 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[PowerPC] Move test to correct location.
Stefan Pintilie [Wed, 7 Mar 2018 18:27:10 +0000 (18:27 +0000)]
[PowerPC] Move test to correct location.

Test was added in r326906 to an incorrect location.
Moving the test to PPC CodeGen directory as the test is PPC specific.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326923 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Make the MUL->VPMADDWD work before op legalization on AVX1 targets. Simplify...
Craig Topper [Wed, 7 Mar 2018 17:53:18 +0000 (17:53 +0000)]
[X86] Make the MUL->VPMADDWD work before op legalization on AVX1 targets. Simplify feature checks by using isTypeLegal.

The v8i32 conversion on AVX1 targets was only working after LowerMUL splits 256-bit vectors.

While I was there I've also made it so we don't have to check for AVX2 and BWI directly and instead just ask if the type is legal.

Differential Revision: https://reviews.llvm.org/D44190

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326917 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Support] Stop passing StringRefs by const reference in some of the getHostCPUname...
Craig Topper [Wed, 7 Mar 2018 17:53:16 +0000 (17:53 +0000)]
[Support] Stop passing StringRefs by const reference in some of the getHostCPUname implementations. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326916 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Rewrite non-HVX unaligned loads as pairs of aligned ones
Krzysztof Parzyszek [Wed, 7 Mar 2018 17:27:18 +0000 (17:27 +0000)]
[Hexagon] Rewrite non-HVX unaligned loads as pairs of aligned ones

This is a follow-up to r325169, this time for all types, not just HVX
vector types.

Disable this by default, since it's not always safe.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326915 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[LangRef] fix formatting in FP descriptions; NFC
Sanjay Patel [Wed, 7 Mar 2018 17:18:22 +0000 (17:18 +0000)]
[LangRef] fix formatting in FP descriptions; NFC

This is a clean-up step to reduce diffs ahead of real
changes to the FP semantics as discussed on llvm-dev:
http://lists.llvm.org/pipermail/llvm-dev/2018-February/121444.html

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326913 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AMDGPU] Increased vector length for global/constant loads.
Farhana Aleen [Wed, 7 Mar 2018 17:09:18 +0000 (17:09 +0000)]
[AMDGPU] Increased vector length for global/constant loads.

Summary: GCN ISA supports instructions that can read 16 consecutive dwords from memory through the scalar data cache;
         loadstoreVectorizer should take advantage of the wider vector length and pack 16/8 elements of dwords/quadwords.

Author: FarhanaAleen

Reviewed By: rampitec

Subscribers: llvm-commits, AMDGPU

Differential Revision: https://reviews.llvm.org/D44179

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326910 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRe-land: Teach CorrelatedValuePropagation to reduce the width of udiv/urem instructions.
Justin Lebar [Wed, 7 Mar 2018 16:56:49 +0000 (16:56 +0000)]
Re-land: Teach CorrelatedValuePropagation to reduce the width of udiv/urem instructions.

Summary:
If the operands of a udiv/urem can be proved to fit within a smaller
power-of-two-sized type, reduce the width of the udiv/urem.

Backed out for failing an assert in clang bootstrap builds.  Re-landing
with a fix for handling non-power-of-two inputs (e.g. udiv i24).

Original Differential Revision: https://reviews.llvm.org/D44102

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326908 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "[AMDGPU] Widened vector length for global/constant address space."
Farhana Aleen [Wed, 7 Mar 2018 16:55:27 +0000 (16:55 +0000)]
Revert "[AMDGPU] Widened vector length for global/constant address space."

This reverts commit ce988cc100dc65e7c6c727aff31ceb99231cab03.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326907 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[PowerPC] LSR tunings for PowerPC
Stefan Pintilie [Wed, 7 Mar 2018 16:53:09 +0000 (16:53 +0000)]
[PowerPC] LSR tunings for PowerPC

The purpose of this patch is to have LSR generate better code on Power.
This is done by overriding isLSRCostLess.

Differential Revision: https://reviews.llvm.org/D40855

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326906 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SampleFDO] Extend SampleProfReader to handle demangled names.
Wei Mi [Wed, 7 Mar 2018 16:45:33 +0000 (16:45 +0000)]
[SampleFDO] Extend SampleProfReader to handle demangled names.

SampleProfReader assumes function names in the profile are all mangled names.
However, there are cases that few demangled names are somehow contained in
the profile (usually because of debug info problems), which may trigger parsing
error in SampleProfReader and cause the whole profile to be unusable. The patch
extends SampleProfReader to handle profiles with demangled names, so that those
profiles can still be useful.

Differential revision: https://reviews.llvm.org/D44161

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326905 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AMDGPU] Widened vector length for global/constant address space.
Farhana Aleen [Wed, 7 Mar 2018 16:29:05 +0000 (16:29 +0000)]
[AMDGPU] Widened vector length for global/constant address space.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326904 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[dwarfdump] Only print CU relative offset in verbose mode
Jonas Devlieghere [Wed, 7 Mar 2018 16:28:53 +0000 (16:28 +0000)]
[dwarfdump] Only print CU relative offset in verbose mode

Instead of only printing the CU-relative offset in non-verbose mode, it
makes more sense to only printed the resolved address. In verbose mode
we still print both.

Differential revision: https://reviews.llvm.org/D44148

rdar://33525475

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326903 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "Reapply "[DWARFv5] Emit file 0 to the line table.""
Alexander Kornienko [Wed, 7 Mar 2018 16:27:44 +0000 (16:27 +0000)]
Revert "Reapply "[DWARFv5] Emit file 0 to the line table.""

This reverts commit r326839.

r326839 breaks assembly file parsing:

$ cat q.c
void g() {}
$ clang -S q.c -g
$ clang -g -c q.s
q.s:9:2: error: file number already allocated
     .file   1 "/tmp/test" "q.c"
     ^

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326902 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert rL326898: "Teach CorrelatedValuePropagation to reduce the width of udiv/urem...
Justin Lebar [Wed, 7 Mar 2018 16:05:43 +0000 (16:05 +0000)]
Revert rL326898: "Teach CorrelatedValuePropagation to reduce the width of udiv/urem instructions."

Breaks bootstrap builds: clang built with this patch asserts while
building MCDwarf.cpp: Assertion `castIsValid(op, S, Ty) && "Invalid
cast!"' failed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326900 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoTeach CorrelatedValuePropagation to reduce the width of udiv/urem instructions.
Justin Lebar [Wed, 7 Mar 2018 15:11:13 +0000 (15:11 +0000)]
Teach CorrelatedValuePropagation to reduce the width of udiv/urem instructions.

Summary:
If the operands of a udiv/urem can be proved to fit within a smaller
power-of-two-sized type, reduce the width of the udiv/urem.

Reviewers: spatel, sanjoy

Subscribers: llvm-commits, hiraditya

Differential Revision: https://reviews.llvm.org/D44102

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326898 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][X87] Add X87 fp80 conversion tests
Simon Pilgrim [Wed, 7 Mar 2018 14:13:14 +0000 (14:13 +0000)]
[X86][X87] Add X87 fp80 conversion tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326897 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[mips] Correct the definition of m(f|t)c(0|2)
Simon Dardis [Wed, 7 Mar 2018 11:39:48 +0000 (11:39 +0000)]
[mips] Correct the definition of m(f|t)c(0|2)

These instructions are defined as taking a GPR register and a
coprocessor register for ISAs up to MIPS32. MIPS32 extended the
definition to allow a selector--a value from 0 to 32--to access
another register.

These instructions are now internally defined as being MIPS-I
instructions, but are rejected for pre-MIPS32 ISA's if they have
an explicit selector which is non-zero. This deviates slightly from
GAS's behaviour which rejects assembly instructions with an
explicit selector for pre-MIPS32 ISAs.

E.g:

mfc0 $4, $5, 0
is rejected by GAS for MIPS-I to MIPS-V but will be accepted
with this patch for MIPS-I to MIPS-V.

Reviewers: atanasyan

Differential Revision: https://reviews.llvm.org/D41662

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326890 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[LoadStoreVectorizer] Differentiate between <1 x T> and T
Sven van Haastregt [Wed, 7 Mar 2018 10:29:28 +0000 (10:29 +0000)]
[LoadStoreVectorizer] Differentiate between <1 x T> and T

The LoadStoreVectorizer thought that <1 x T> and T were the same types
when merging stores, leading to a crash later.

Patch by Erik Hogeman.

Differential Revision: https://reviews.llvm.org/D44014

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326884 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ARM] Fix for PR36577
Sjoerd Meijer [Wed, 7 Mar 2018 09:10:44 +0000 (09:10 +0000)]
[ARM] Fix for PR36577

Don't PerformSHLSimplify if the given node is used by a node that also uses a
constant because we may get stuck in an infinite combine loop.

bugzilla: https://bugs.llvm.org/show_bug.cgi?id=36577

Patch by Sam Parker.

Differential Revision: https://reviews.llvm.org/D44097

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326882 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SystemZ] NFC refactoring in SystemZHazardRecognizer.
Jonas Paulsson [Wed, 7 Mar 2018 08:57:09 +0000 (08:57 +0000)]
[SystemZ]  NFC refactoring in SystemZHazardRecognizer.

Use Reset() after emitting a call.

Review: Ulrich Weigand

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326881 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SystemZ] Improve getCurrCycleIdx() in SystemZHazardRecognizer.
Jonas Paulsson [Wed, 7 Mar 2018 08:54:32 +0000 (08:54 +0000)]
[SystemZ]  Improve getCurrCycleIdx() in SystemZHazardRecognizer.

getCurrCycleIdx() returns the decoder cycle index which the next candidate SU
will be placed on.

This patch improves this method by passing the candidate SU to it so that if
SU will begin a new group, the index of that group is returned instead.

Review: Ulrich Weigand

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326880 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SystemZ] NFC refactoring in SystemZHazardRecognizer.
Jonas Paulsson [Wed, 7 Mar 2018 08:45:09 +0000 (08:45 +0000)]
[SystemZ] NFC refactoring in SystemZHazardRecognizer.

Handle the not-taken branch in emitInstruction() where the TakenBranch
argument is available. This is cleaner than relying on EmitInstruction().

Review: Ulrich Weigand

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326879 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SystemZ] Improved debug dumping during post-RA scheduling.
Jonas Paulsson [Wed, 7 Mar 2018 08:39:00 +0000 (08:39 +0000)]
[SystemZ]  Improved debug dumping during post-RA scheduling.

Review: Ulrich Weigand

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326878 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add IMUL scheduling info on sandybridge, fix it on >=haswell.
Clement Courbet [Wed, 7 Mar 2018 08:14:02 +0000 (08:14 +0000)]
[X86] Add IMUL scheduling info on sandybridge, fix it on >=haswell.

Summary:
Only IMUL16rri uses an extra P0156. IMUL32* and IMUL16rr only use
P1.
This was computed using https://github.com/google/EXEgesis/blob/master/exegesis/tools/compute_itineraries.cc

This can easily be validated by running perf on the following code:

```
int main(int argc, char**argv) {
  int a = argc;
  int b = argc;
  int c = argc;
  int d = argc;

  for (int i = 0; i < LOOP_ITERATIONS; ++i) {
    asm volatile(
      R"(
        .rept 10000
        imull $0x2, %%edx, %%eax
        imull $0x2, %%ecx, %%ebx
        imull $0x2, %%eax, %%edx
        imull $0x2, %%ebx, %%ecx
        .endr
      )"
      : "+a"(a), "+b"(b), "+c"(c), "+d"(d)
      :
      :);
  }
  return a+b+c+d;
}
```
-> test.cc

perf stat -x, -e cycles --pfm-events=uops_executed_port:port_0:u,uops_executed_port:port_1:u,uops_executed_port:port_2:u,uops_executed_port:port_3:u,uops_executed_port:port_4:u,uops_executed_port:port_5:u,uops_executed_port:port_6:u,uops_executed_port:port_7:u test

Reviewers: craig.topper, RKSimon, gadi.haber

Subscribers: llvm-commits, gchatelet, chandlerc

Differential Revision: https://reviews.llvm.org/D43460

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326877 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-pdbdump] Add guard for null pointers and remove unused code
Aaron Smith [Wed, 7 Mar 2018 02:23:08 +0000 (02:23 +0000)]
[llvm-pdbdump] Add guard for null pointers and remove unused code

Summary: This avoids crashing when a user tries to dump a pdb with the `-native` option.

Reviewers: zturner, llvm-commits, rnk

Reviewed By: zturner

Subscribers: mgrang

Differential Revision: https://reviews.llvm.org/D44117

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326863 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAdd early exit on reassociation of 0 expression.
Evgeny Stupachenko [Wed, 7 Mar 2018 02:17:08 +0000 (02:17 +0000)]
Add early exit on reassociation of 0 expression.

Summary:

Before the patch a try to reassociate ((v * 16) * 0) * 1 fall into infinite loop

Reviewers: pankajchawla

Differential Revision: http://reviews.llvm.org/D41467

From: Evgeny Stupachenko <evstupac@gmail.com>
                         <evgeny.v.stupachenko@intel.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326861 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DebugInfoPDB] Add DIA implementation for getSrcLineOnTypeDefn
Aaron Smith [Wed, 7 Mar 2018 00:33:09 +0000 (00:33 +0000)]
[DebugInfoPDB] Add DIA implementation for getSrcLineOnTypeDefn

Summary: This helps to determine the line number for a PDB type with definition

Reviewers: zturner, llvm-commits, rnk

Reviewed By: zturner

Subscribers: rengolin, JDevlieghere

Differential Revision: https://reviews.llvm.org/D44119

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326857 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Transforms] Add missing header for InstructionCombining.cpp, in order to export...
Eugene Zelenko [Tue, 6 Mar 2018 23:06:13 +0000 (23:06 +0000)]
[Transforms] Add missing header for InstructionCombining.cpp, in order to export LLVMInitializeInstCombine as extern "C". Fixes PR35947.

Patch by Brenton Bostick.

Differential revision: https://reviews.llvm.org/D44140

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326843 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Fix a typo in Host.cpp that causes us to misidentify KNL, Silvermont, Goldmont...
Craig Topper [Tue, 6 Mar 2018 22:45:31 +0000 (22:45 +0000)]
[X86] Fix a typo in Host.cpp that causes us to misidentify KNL, Silvermont, Goldmont and probably other CPUs for -march=native

I think most of the Intel Core CPUs and recent AMD CPUs are unaffected. All the CPUs that have a "subtype" should work. The ones that were broken are the ones that are a "type" with no subtypes.

Fixes PR36619.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326840 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoReapply "[DWARFv5] Emit file 0 to the line table."
Paul Robinson [Tue, 6 Mar 2018 22:37:45 +0000 (22:37 +0000)]
Reapply "[DWARFv5] Emit file 0 to the line table."

Fixes the bug found by asan. Also XFAIL the new test for Darwin,
which is stuck on DWARF v2, and fix up other tests so they stop
failing on Windows.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326839 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[TargetLowering] Add vector BITCAST support to SimplifyDemandedVectorElts
Simon Pilgrim [Tue, 6 Mar 2018 22:32:01 +0000 (22:32 +0000)]
[TargetLowering] Add vector BITCAST support to SimplifyDemandedVectorElts

Notably helps cleanup after legalization of vector types

Differential Revision: https://reviews.llvm.org/D43674

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326838 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoDA: remove uses of GEP, only ask SCEV
Sebastian Pop [Tue, 6 Mar 2018 21:55:59 +0000 (21:55 +0000)]
DA: remove uses of GEP, only ask SCEV

It's been quite some time the Dependence Analysis (DA) is broken,
as it uses the GEP representation to "identify" multi-dimensional arrays.
It even wrongly detects multi-dimensional arrays in single nested loops:

from test/Analysis/DependenceAnalysis/Coupled.ll, example @couple6
;; for (long int i = 0; i < 50; i++) {
;; A[i][3*i - 6] = i;
;; *B++ = A[i][i];

DA used to detect two subscripts, which makes no sense in the LLVM IR
or in C/C++ semantics, as there are no guarantees as in Fortran of
subscripts not overlapping into a next array dimension:

maximum nesting levels = 1
SrcPtrSCEV = %A
DstPtrSCEV = %A
using GEPs
subscript 0
    src = {0,+,1}<nuw><nsw><%for.body>
    dst = {0,+,1}<nuw><nsw><%for.body>
    class = 1
    loops = {1}
subscript 1
    src = {-6,+,3}<nsw><%for.body>
    dst = {0,+,1}<nuw><nsw><%for.body>
    class = 1
    loops = {1}
Separable = {}
Coupled = {1}

With the current patch, DA will correctly work on only one dimension:

maximum nesting levels = 1
SrcSCEV = {(-2424 + %A)<nsw>,+,1212}<%for.body>
DstSCEV = {%A,+,404}<%for.body>
subscript 0
    src = {(-2424 + %A)<nsw>,+,1212}<%for.body>
    dst = {%A,+,404}<%for.body>
    class = 1
    loops = {1}
Separable = {0}
Coupled = {}

This change removes all uses of GEP from DA, and we now only rely
on the SCEV representation.

The patch does not turn on -da-delinearize by default, and so the DA analysis
will be more conservative in the case of multi-dimensional memory accesses in
nested loops.

I disabled some interchange tests, as the DA is not able to disambiguate
the dependence anymore. To make DA stronger, we may need to
compute a bound on the number of iterations based on the access functions
and array dimensions.

The patch cleans up all the CHECKs in test/Transforms/LoopInterchange/*.ll to
avoid checking for snippets of LLVM IR: this form of checking is very hard to
maintain. Instead, we now check for output of the pass that are more meaningful
than dozens of lines of LLVM IR. Some tests now require -debug messages and thus
only enabled with asserts.

Patch written by Sebastian Pop and Aditya Kumar.

Differential Revision: https://reviews.llvm.org/D35430

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326837 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoPrintStatistics() and PrintStatisticsJSON() should take StatLock
Daniel Sanders [Tue, 6 Mar 2018 21:16:42 +0000 (21:16 +0000)]
PrintStatistics() and PrintStatisticsJSON() should take StatLock

These two functions iterate over the list of statistics but don't take the lock
that protects the iterators from being invalidated by
StatisticInfo::addStatistic().

So far, this hasn't been an issue since (in-tree at least) these functions are
called by the StatisticInfo destructor so addStatistic() shouldn't be called
anymore. However, we do expose them in the public API.

Note that this only protects against iterator invalidation and does not protect
against ordering issues caused by statistic updates that race with
PrintStatistics()/PrintStatisticsJSON().

Thanks to Roman Tereshin for spotting it

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326834 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[TargetLowering] Rename DAGCombinerInfo::isAfterLegalizeVectorOps to DAGCombiner...
Craig Topper [Tue, 6 Mar 2018 19:44:52 +0000 (19:44 +0000)]
[TargetLowering] Rename DAGCombinerInfo::isAfterLegalizeVectorOps to DAGCombiner::isAfterLegalizeDAG since that's what it checks. NFC

The code checks Level == AfterLegalizeDAG which is the fourth and last of the possible DAG combine stages that we have.

There is a Level called AfterLegalVectorOps, but that's the third DAG combine and it doesn't always run.

A function called isAfterLegalVectorOps should imply it returns true in either of the DAG combines that runs after the legalize vector ops stage, but that's not what this function does.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326832 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SymboleFilePDB] Put the test input back that my previous commit clobbered
Aaron Smith [Tue, 6 Mar 2018 19:36:17 +0000 (19:36 +0000)]
[SymboleFilePDB] Put the test input back that my previous commit clobbered

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326831 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Update more testcases
Krzysztof Parzyszek [Tue, 6 Mar 2018 19:15:58 +0000 (19:15 +0000)]
[Hexagon] Update more testcases

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326830 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Remove {{ *}} from testcases
Krzysztof Parzyszek [Tue, 6 Mar 2018 19:07:21 +0000 (19:07 +0000)]
[Hexagon] Remove {{ *}} from testcases

The spaces in the instructions are now consistent.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326829 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[InstCombine] simplify min/max canonicalization; NFCI
Sanjay Patel [Tue, 6 Mar 2018 19:01:18 +0000 (19:01 +0000)]
[InstCombine] simplify min/max canonicalization; NFCI

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326828 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Reject registers that require a REX prefix in inline asm constraints in 32...
Craig Topper [Tue, 6 Mar 2018 18:56:33 +0000 (18:56 +0000)]
[X86] Reject registers that require a REX prefix in inline asm constraints in 32-bit mode

We don't currently reject r8-r15 or xmm8-32 or bpl/spl/sil/dil in 32-bit mode.

Differential Revision: https://reviews.llvm.org/D44031

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326826 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AMDGPU] Add default ISA version targets
Stanislav Mekhanoshin [Tue, 6 Mar 2018 18:33:55 +0000 (18:33 +0000)]
[AMDGPU] Add default ISA version targets

In case if -mattr used to modify feature set bits in llvm-mc call
getIsaVersion can fail to identify specific ISA due to test mismatch.
Adding default fallback tests which will always correctly report at
least major version.

Differential Revision: https://reviews.llvm.org/D44163

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326825 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[CodeView] Emit UdtSourceLine information for enums
Aaron Smith [Tue, 6 Mar 2018 18:20:22 +0000 (18:20 +0000)]
[CodeView] Emit UdtSourceLine information for enums

Summary:
- Emit UdtSourceLine information for enums to match MSVC

- Add a method to add UDTSrcLine and call it for all Class/Struct/Union/Enum

- Update test cases to verify the changes

Reviewers: zturner, llvm-commits, rnk

Reviewed By: rnk

Differential Revision: https://reviews.llvm.org/D44116

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326824 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[PatternMatch] define m_Not using m_Xor and cst_pred_ty
Sanjay Patel [Tue, 6 Mar 2018 18:19:42 +0000 (18:19 +0000)]
[PatternMatch] define m_Not using m_Xor and cst_pred_ty

Using cst_pred_ty in the definition allows us to match vectors with undef elements.

This is a continuation of an effort to make all pattern matchers allow undef elements in vectors:
rL325437
rL325466
D43792

Differential Revision: https://reviews.llvm.org/D44076

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326823 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRefactor check for dllimport in the Verifier.
Rafael Espindola [Tue, 6 Mar 2018 17:19:23 +0000 (17:19 +0000)]
Refactor check for dllimport in the Verifier.

This avoids duplicated code and now also rejects dllimport aliases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326814 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ValueTracking] move helpers for SelectPatterns from InstCombine to ValueTracking
Sanjay Patel [Tue, 6 Mar 2018 16:57:55 +0000 (16:57 +0000)]
[ValueTracking] move helpers for SelectPatterns from InstCombine to ValueTracking

Most of the folds based on SelectPatternResult belong in InstSimplify rather than
InstCombine, so the helper code should be available to other passes/analysis.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@326812 91177308-0d34-0410-b5e6-96231b3b80d8