OSDN Git Service

android-x86/external-llvm.git
6 years ago[RISCV] Codegen support for RV32F floating point comparison operations
Alex Bradbury [Wed, 21 Mar 2018 15:11:02 +0000 (15:11 +0000)]
[RISCV] Codegen support for RV32F floating point comparison operations

This patch also includes extensive tests targeted at select and br+fcmp IR
inputs. A sequence of br+fcmp required support for FPR32 registers to be added
to RISCVInstrInfo::storeRegToStackSlot and
RISCVInstrInfo::loadRegFromStackSlot.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328104 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[RISCV] Add tests missed from r327979
Alex Bradbury [Wed, 21 Mar 2018 14:50:27 +0000 (14:50 +0000)]
[RISCV] Add tests missed from r327979

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328102 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoChange DT_* value definitions to macros in a separate file
Alexander Richardson [Wed, 21 Mar 2018 14:17:50 +0000 (14:17 +0000)]
Change DT_* value definitions to macros in a separate file

Summary:
I recently added a new dynamic tag to our fork of LLVM and when adding it
to llvm-readobj I noticed that not all DT_ values were being handled there.

Using macros in a .def file that can be included by both ELFDumper.cpp and
the ELF.h header ensures that the two don't get out of sync when new values
are added.

Reviewers: grimar, pcc, davide, espindola

Reviewed By: grimar, espindola

Subscribers: srhines, llvm-commits

Differential Revision: https://reviews.llvm.org/D44558

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328099 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[LangRef] more hyphens: always write "floating-point"
Sanjay Patel [Wed, 21 Mar 2018 14:15:33 +0000 (14:15 +0000)]
[LangRef] more hyphens: always write "floating-point"

We were inconsistent, sometimes even within a single sentence.
The consensus seems clear that the FP we're looking for is
spelled "floating-point". Without the hyphen, it's a
"surprisingly fine" jazz album.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328098 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MemCpyOpt] Update to new API for memory intrinsic alignment
Daniel Neilson [Wed, 21 Mar 2018 14:14:55 +0000 (14:14 +0000)]
[MemCpyOpt] Update to new API for memory intrinsic alignment

Summary:
This change is part of step five in the series of changes to remove alignment argument from
memcpy/memmove/memset in favour of alignment attributes. In particular, this changes the
MemCpyOpt pass to cease using:
1) The old getAlignment() API of MemoryIntrinsic in favour of getting source & dest specific
alignments through the new API.
2) The old IRBuilder CreateMemCpy/CreateMemMove single-alignment APIs in favour of the new
API that allows setting source and destination alignments independently.

We also add a few tests to fill gaps in the testing of this pass.

Steps:
Step 1) Remove alignment parameter and create alignment parameter attributes for
memcpy/memmove/memset. ( rL322965, rC322964, rL322963 )
Step 2) Expand the IRBuilder API to allow creation of memcpy/memmove with differing
source and dest alignments. ( rL323597 )
Step 3) Update Clang to use the new IRBuilder API. ( rC323617 )
Step 4) Update Polly to use the new IRBuilder API. ( rL323618 )
Step 5) Update LLVM passes that create memcpy/memmove calls to use the new IRBuilder API,
and those that use use MemIntrinsicInst::[get|set]Alignment() to use [get|set]DestAlignment()
and [get|set]SourceAlignment() instead. ( rL323886, rL323891, rL324148, rL324273, rL324278,
rL324384, rL324395, rL324402, rL324626, rL324642, rL324653, rL324654, rL324773, rL324774,
rL324781, rL324784, rL324955, rL324960, rL325816, rL327398, rL327421 )
Step 6) Remove the single-alignment IRBuilder API for memcpy/memmove, and the
MemIntrinsicInst::[get|set]Alignment() methods.

Reference
   http://lists.llvm.org/pipermail/llvm-dev/2015-August/089384.html
   http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20151109/312083.html

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328097 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRe-re-land: Teach CorrelatedValuePropagation to reduce the width of udiv/urem instruc...
Justin Lebar [Wed, 21 Mar 2018 14:08:21 +0000 (14:08 +0000)]
Re-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 causing performance regressions.  Re-landing
because we've determined that these regressions were noise.

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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328096 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "Move DataTypes.h from Support to llvm-c to fix layering."
Jonas Devlieghere [Wed, 21 Mar 2018 13:28:37 +0000 (13:28 +0000)]
Revert "Move DataTypes.h from Support to llvm-c to fix layering."

This reverts r328065.

I missed this one in r328085 and the bots were still failing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328095 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-mca] Clean up some code. NFC
Andrea Di Biagio [Wed, 21 Mar 2018 12:49:07 +0000 (12:49 +0000)]
[llvm-mca] Clean up some code. NFC

Removed a couple of methods from DispatchUnit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328094 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix build broken by r328090
Pavel Labath [Wed, 21 Mar 2018 12:18:03 +0000 (12:18 +0000)]
Fix build broken by r328090

- constexpr is needed for out-of-class definition of the Type static
  member by some compilers
- MSVC is confused by the initialization of the static constexpr char[]
  member when it happens in a template specialization. Explicitly
  specifying the length of the array seems to be enough to help it
  figure things out.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328093 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[dwarf] Unify unknown dwarf enum formatting code
Pavel Labath [Wed, 21 Mar 2018 11:46:37 +0000 (11:46 +0000)]
[dwarf] Unify unknown dwarf enum formatting code

Summary:
We have had at least three pieces of code (in DWARFAbbreviationDeclaration,
DWARFAcceleratorTable and DWARFDie) that have hand-rolled support for
dumping unknown dwarf enum values. While not terrible, they are a bit
distracting and enable small differences to creep in (Unknown_ffff vs.
Unknown_0xffff). I ended up needing to add a fourth place
(DWARFVerifier), so it seems it would be a good time to centralize.

This patch creates an alternative to the XXXString dumping functions in
the BinaryFormat library, which formats an unknown value as
DW_TYPE_unknown_1234, instead of just an empty string. It is based on
the formatv function, as that allows us to avoid materializing the
string for unknown values (and because this way I don't have to invent a
name for the new functions :P).

In this patch I add formatters for dwarf attributes, forms, tags, and
index attributes as these are the ones in use currently, but adding
other enums is straight-forward.

Reviewers: dblaikie, JDevlieghere, aprantl

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D44570

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328090 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert layering changes
Jonas Devlieghere [Wed, 21 Mar 2018 10:35:09 +0000 (10:35 +0000)]
Revert layering changes

This reverts:
  r328072 "Move Compiler.h from Support to Demangler to fix layering."
  r328073 "Fix the actual user of DataTypes.h in llvm-c to avoid the circular dependency"

Failing bots:
  http://green.lab.llvm.org/green/job/clang-stage2-coverage-R/
  http://green.lab.llvm.org/green/job/clang-stage2-configure-Rlto/

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328085 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SelectionDAG] Support multiple dangling debug info for one value
Bjorn Pettersson [Wed, 21 Mar 2018 09:44:34 +0000 (09:44 +0000)]
[SelectionDAG] Support multiple dangling debug info for one value

Summary:
When building the selection DAG we sometimes need to postpone
the handling of a dbg.value until the value it should refer to
is created. This is done by using the DanglingDebugInfoMap.
In the past this map has been limited to hold one dangling
dbg.value per value. This patch removes that restriction.

Reviewers: aprantl, rnk, probinson, vsk

Reviewed By: aprantl

Subscribers: Ka-Ka, llvm-commits, JDevlieghere

Tags: #debug-info

Differential Revision: https://reviews.llvm.org/D44610

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328084 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix build bot after r328078 "llvm-readobj] - Teach llvm-readobj to dump .note.gnu...
George Rimar [Wed, 21 Mar 2018 08:48:44 +0000 (08:48 +0000)]
Fix build bot after r328078 "llvm-readobj] - Teach llvm-readobj to dump .note.gnu.property sections."

BB was: http://lab.llvm.org:8011/builders/llvm-clang-lld-x86_64-scei-ps4-ubuntu-fast/builds/27058/steps/test/logs/stdio

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328080 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-readobj] - Teach llvm-readobj to dump .note.gnu.property sections.
George Rimar [Wed, 21 Mar 2018 08:34:55 +0000 (08:34 +0000)]
[llvm-readobj] - Teach llvm-readobj to dump .note.gnu.property sections.

NT_GNU_PROPERTY_TYPE_0 is a recently added type of .note.gnu.property
section specified in Linux Extensions to gABI.
(https://github.com/hjl-tools/linux-abi/wiki/Linux-Extensions-to-gABI)

Patch teach tool to print such notes properly.

Differential revision: https://reviews.llvm.org/D44469

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328078 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][Broadwell] Merge multiple InstrRW entries that map to the same SchedWriteRes...
Craig Topper [Wed, 21 Mar 2018 06:28:42 +0000 (06:28 +0000)]
[X86][Broadwell] Merge multiple InstrRW entries that map to the same SchedWriteRes group (NFCI) (PR35955)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328076 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[TableGen] Remove a defaulted function argument that is never called with another...
Craig Topper [Wed, 21 Mar 2018 05:13:04 +0000 (05:13 +0000)]
[TableGen] Remove a defaulted function argument that is never called with another value. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328075 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[TableGen] Move a function from llvm namespace and make it a static function. NFC
Craig Topper [Wed, 21 Mar 2018 05:13:01 +0000 (05:13 +0000)]
[TableGen] Move a function from llvm namespace and make it a static function. NFC

It's only called from one place and is defined just above that use.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328074 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix the actual user of DataTypes.h in llvm-c to avoid the circular dependency
David Blaikie [Wed, 21 Mar 2018 04:07:07 +0000 (04:07 +0000)]
Fix the actual user of DataTypes.h in llvm-c to avoid the circular dependency

(follow-up to r328065)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328073 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoMove Compiler.h from Support to Demangler to fix layering.
David Blaikie [Wed, 21 Mar 2018 04:07:05 +0000 (04:07 +0000)]
Move Compiler.h from Support to Demangler to fix layering.

Support depends on Demangle (Support/Unix/Signals.inc), so Demangle
including Support/Compiler.h created a circular dependency.

Leave a forwarding shim of Compiler.h because it makes more sense for
users (a deeper fix might involve splitting Support into lower and upper
Support - but that also sounds a bit weird/awkward) than thinking about
the dependency on the Demangler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328072 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Fix the SchedRW for XOP vpcom register form instructions to not be marked as...
Craig Topper [Wed, 21 Mar 2018 03:41:33 +0000 (03:41 +0000)]
[X86] Fix the SchedRW for XOP vpcom register form instructions to not be marked as loads.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328071 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[TableGen] Use SmallMapVector to simplify some code that was trying to keep a vector...
Craig Topper [Wed, 21 Mar 2018 02:48:34 +0000 (02:48 +0000)]
[TableGen] Use SmallMapVector to simplify some code that was trying to keep a vector unique

Summary:
This code previously had a SmallVector of std::pairs containing an unsigned and another SmallVector. The outer vector was using the unsigned effectively as a key to decide which SmallVector to add into. So each time something new needed to be added the out vector needed to be scanned. If it wasn't found a new entry needed to be added to be added. This sounds very much like a map, but the next loop iterates over the outer vector to get a deterministic order.

We can simplify this code greatly if use SmallMapVector instead. This uses more stack space since we now have a vector and a map, but the searching and creating new entries all happens behind the scenes. It should also make the search more efficient though usually there are only a few entries so that doesn't matter much.

We could probably get determinism by just using std::map which would iterate over the unsigned key, but that would generate different output from what we get with the current implementation.

Reviewers: RKSimon, dblaikie

Reviewed By: dblaikie

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D44711

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328070 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoMove DataTypes.h from Support to llvm-c to fix layering.
David Blaikie [Wed, 21 Mar 2018 00:48:05 +0000 (00:48 +0000)]
Move DataTypes.h from Support to llvm-c to fix layering.

Support depends on llvm-c (a few typedefs, macros, etc - Types.h,
Disassembler.h, and TargetMachine.h.

This could be done the other way - those macros/typedefs/etc could be
moved into Support and used from llvm-c instead. If someone feels that's
a better direction to go, happy to discuss it/try it out/etc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328065 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Change PMULLD to 10 cycles on Skylake per Agner's tables and llvm-exegesis.
Craig Topper [Tue, 20 Mar 2018 23:39:48 +0000 (23:39 +0000)]
[X86] Change PMULLD to 10 cycles on Skylake per Agner's tables and llvm-exegesis.

Also restrict to port 0 and 1 for SkylakeClient. It looks like the scheduler models don't account for client not having a full vector ALU on port 5 like server.

Fixes PR36808.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328061 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MustExecute] Shwo the effect of using full loop info variant
Philip Reames [Tue, 20 Mar 2018 23:00:54 +0000 (23:00 +0000)]
[MustExecute] Shwo the effect of using full loop info variant

Most basic possible test for the logic used by LICM.

Also contains a speculative build fix for compiles which complain about a definition of a stuct K; followed by a declaration as class K;

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328058 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Update torture compile test expectations
Derek Schuff [Tue, 20 Mar 2018 23:00:13 +0000 (23:00 +0000)]
[WebAssembly] Update torture compile test expectations

The tests compile after r328049

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328057 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MustExecute] Add simplest possible test for LoopSafetyOnfo
Philip Reames [Tue, 20 Mar 2018 22:55:20 +0000 (22:55 +0000)]
[MustExecute] Add simplest possible test for LoopSafetyOnfo

(Currently showing without, will enable and check in diff to show impact)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328056 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MustExecute] Move isGuaranteedToExecute and related rourtines to Analysis
Philip Reames [Tue, 20 Mar 2018 22:45:23 +0000 (22:45 +0000)]
[MustExecute] Move isGuaranteedToExecute and related rourtines to Analysis

Next step is to actually merge the implementations and get both implementations tested through the new printer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328055 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SchedModel] Simplify InstRegexOp::apply. NFCI.
Simon Pilgrim [Tue, 20 Mar 2018 22:20:28 +0000 (22:20 +0000)]
[SchedModel] Simplify InstRegexOp::apply. NFCI.

As discussed on D44687, there was no need for 2 separate for loops for collecting the Regex and then matching against instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328052 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Strip threadlocal attribute from globals in single thread mode
Derek Schuff [Tue, 20 Mar 2018 22:01:32 +0000 (22:01 +0000)]
[WebAssembly] Strip threadlocal attribute from globals in single thread mode

The default thread model for wasm is single, and in this mode thread-local
global variables can be lowered identically to non-thread-local variables.

Differential Revision: https://reviews.llvm.org/D44703

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328049 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Drop unnecessary InstRW overrides for WriteFMA
Simon Pilgrim [Tue, 20 Mar 2018 21:15:23 +0000 (21:15 +0000)]
[X86] Drop unnecessary InstRW overrides for WriteFMA

As noticed on D44687, these already match the WriteFMA def so can be removed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328045 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ReachingDefAnalysis] Fix what I assume to be a typo ReachingDedDefaultVal->ReachingD...
Craig Topper [Tue, 20 Mar 2018 20:53:21 +0000 (20:53 +0000)]
[ReachingDefAnalysis] Fix what I assume to be a typo ReachingDedDefaultVal->ReachingDefDefaultVal.

Unless Ded has some many I don't know about.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328043 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ObjCARC] Add funclet token to ARC marker
Shoaib Meenai [Tue, 20 Mar 2018 20:45:41 +0000 (20:45 +0000)]
[ObjCARC] Add funclet token to ARC marker

The inline assembly generated for the ARC autorelease elision marker
must have a funclet token if it's emitted inside a funclet, otherwise
the inline assembly (and all subsequent code in the funclet) will be
marked unreachable by WinEHPrepare.

Note that this only applies for the non-O0 case, since at O0, clang
emits the autorelease elision marker itself rather than deferring to the
backend. The fix for clang is handled in a separate change.

Differential Revision: https://reviews.llvm.org/D44641

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328042 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Don't use the MSVC stack protector names on mingw
Martin Storsjo [Tue, 20 Mar 2018 20:37:51 +0000 (20:37 +0000)]
[X86] Don't use the MSVC stack protector names on mingw

Mingw uses the same stack protector functions as GCC provides
on other platforms as well.

Patch by Valentin Churavy!

Differential Revision: https://reviews.llvm.org/D27296

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328039 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFor llvm-objdump and Mach-O files, fix the printing of module init and
Kevin Enderby [Tue, 20 Mar 2018 20:29:52 +0000 (20:29 +0000)]
For llvm-objdump and Mach-O files, fix the printing of module init and
term sections from .o files to look to see if the pointers have a relocation
entry and if so print the symbol name from the relocation entry.  If not fall
back to the existing code and use the pointer value to look up that value
in the symbol table.

rdar://38337506

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328037 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64] Add vmulxh_lane fp16 vector intrinsic
Abderrazek Zaafrani [Tue, 20 Mar 2018 20:25:40 +0000 (20:25 +0000)]
[AArch64] Add vmulxh_lane fp16 vector intrinsic

https://reviews.llvm.org/D44591

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328035 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[TableGen] Use range based for loop. NFC
Craig Topper [Tue, 20 Mar 2018 20:24:16 +0000 (20:24 +0000)]
[TableGen] Use range based for loop. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328034 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[TableGen] Use vector::append instead of looping and calling push_back. NFC
Craig Topper [Tue, 20 Mar 2018 20:24:14 +0000 (20:24 +0000)]
[TableGen] Use vector::append instead of looping and calling push_back. NFC

Both vectors contain unsigned so we can just use append to do the copying. Not only is this shorter, but it should be able to predict the final size and only grow the vector once if needed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328033 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[TableGen] Use llvm::transform to simplify some loops. NFCI
Craig Topper [Tue, 20 Mar 2018 20:24:12 +0000 (20:24 +0000)]
[TableGen] Use llvm::transform to simplify some loops. NFCI

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328032 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[TableGen] Pass result of std::unique to vector::erase instead of calculating a size...
Craig Topper [Tue, 20 Mar 2018 20:24:10 +0000 (20:24 +0000)]
[TableGen] Pass result of std::unique to vector::erase instead of calculating a size and calling resize.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328031 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DEBUGINFO] Add -no-dwarf-debug-ranges option.
Alexey Bataev [Tue, 20 Mar 2018 20:21:38 +0000 (20:21 +0000)]
[DEBUGINFO] Add -no-dwarf-debug-ranges option.

Summary:
Added option -no-dwarf-debug-ranges option to disable emission of
.debug_ranges section.

Reviewers: probinson, echristo

Subscribers: aprantl, JDevlieghere, llvm-commits

Differential Revision: https://reviews.llvm.org/D44384

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328030 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-mca] add keyword override to a couple of methods in BackendStatistics.
Andrea Di Biagio [Tue, 20 Mar 2018 20:18:36 +0000 (20:18 +0000)]
[llvm-mca] add keyword override to a couple of methods in BackendStatistics.

This should fix the buildbots after r328011.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328029 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[WebAssembly] Added initial AsmParser implementation.
Derek Schuff [Tue, 20 Mar 2018 20:06:35 +0000 (20:06 +0000)]
[WebAssembly] Added initial AsmParser implementation.

It uses the MC framework and the tablegen matcher to do the
heavy lifting. Can handle both explicit and implicit locals
(-disable-wasm-explicit-locals). Comes with a small regression
test.

This is a first basic implementation that can parse most llvm .s
output and round-trips most instructions succesfully, but in order
to keep the commit small, does not address all issues.

There are a fair number of mismatches between what MC / assembly
matcher think a "CPU" should look like and what WASM provides,
some already have workarounds in this commit (e.g. the way it
deals with register operands) and some that require further work.
Some of that further work may involve changing what the
Disassembler outputs (and what s2wasm parses), so are probably
best left to followups.

Some known things missing:
- Many directives are ignored and not emitted.
- Vararg calls are parsed but extra args not emitted.
- Loop signatures are likely incorrect.
- $drop= is not emitted.
- Disassembler does not output SIMD types correctly, so assembler
  can't test them.

Patch by Wouter van Oortmerssen

Differential Revision: https://reviews.llvm.org/D44329

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328028 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64] Adjust the cost model for Exynos M3
Evandro Menezes [Tue, 20 Mar 2018 20:00:29 +0000 (20:00 +0000)]
[AArch64] Adjust the cost model for Exynos M3

Fix typo in the number of integer dividers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328027 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-objcopy] Revert the tests from r328012
Alexander Shaposhnikov [Tue, 20 Mar 2018 19:50:14 +0000 (19:50 +0000)]
[llvm-objcopy] Revert the tests from r328012

Temporarily revert the tests from r328012 as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328026 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-objcopy] Revert r328012
Alexander Shaposhnikov [Tue, 20 Mar 2018 19:46:00 +0000 (19:46 +0000)]
[llvm-objcopy] Revert r328012

Temporarily revert r328012 (since it broke down the big-endian bots),
will resubmit an updated version later.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328024 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Add a few more lit tests, NFC
Krzysztof Parzyszek [Tue, 20 Mar 2018 19:35:09 +0000 (19:35 +0000)]
[Hexagon] Add a few more lit tests, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328023 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Add heuristic to exclude critical path cost for scheduling
Krzysztof Parzyszek [Tue, 20 Mar 2018 19:26:27 +0000 (19:26 +0000)]
[Hexagon] Add heuristic to exclude critical path cost for scheduling

Patch by Brendon Cahoon.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328022 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Fix fall-through warnings in HexagonMCDuplexInfo.cpp
Krzysztof Parzyszek [Tue, 20 Mar 2018 19:23:18 +0000 (19:23 +0000)]
[Hexagon] Fix fall-through warnings in HexagonMCDuplexInfo.cpp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328021 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MC,X86] Cleanup some X86 parser functions to use MCParser helpers. NFCI.
Nirav Dave [Tue, 20 Mar 2018 19:12:41 +0000 (19:12 +0000)]
[MC,X86] Cleanup some X86 parser functions to use MCParser helpers. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328019 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-mca] Remove const from a bunch of ArrayRef. NFC
Andrea Di Biagio [Tue, 20 Mar 2018 19:06:34 +0000 (19:06 +0000)]
[llvm-mca] Remove const from a bunch of ArrayRef. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328018 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[PowerPC][LegalizeFloatTypes] Move the PPC hacks for (i32 fp_to_sint/fp_to_uint ...
Craig Topper [Tue, 20 Mar 2018 18:49:28 +0000 (18:49 +0000)]
[PowerPC][LegalizeFloatTypes] Move the PPC hacks for (i32 fp_to_sint/fp_to_uint (ppcf128 X)) out of LegalizeFloatTypes and into PPC specific code

I'm not entirely sure these hacks are still needed. If you remove the hacks completely, the name of the library call that gets generated doesn't match the grep the test previously had. So the test wasn't really checking anything.

If the hack is still needed it belongs in PPC specific code. I believe the FP_TO_SINT code here is the only place in the tree where a FP_ROUND_INREG node is created today. And I don't think its even being used correctly because the legalization returned a BUILD_PAIR with the same value twice. That doesn't seem right to me. By moving the code entirely to PPC we can avoid creating the FP_ROUND_INREG at all.

I replaced the grep in the existing test with full checks generated by hacking update_llc_test_check.py to support ppc32 just long enough to generate it.

Differential Revision: https://reviews.llvm.org/D44061

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328017 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add phony registers for high halves of regs with low halves
Krzysztof Parzyszek [Tue, 20 Mar 2018 18:46:55 +0000 (18:46 +0000)]
[X86] Add phony registers for high halves of regs with low halves

Registers E[A-D]X, E[SD]I, E[BS]P, and EIP have 16-bit subregisters
that cover the low halves of these registers. This change adds artificial
subregisters for the high halves in order to differentiate (in terms of
register units) between the 32- and the low 16-bit registers.

This patch contains parts that aim to preserve the calculated register
pressure. This is in order to preserve the current codegen (minimize the
impact of this patch). The approach of having artificial subregisters
could be used to fix PR23423, but the pressure calculation would need
to be changed.

Differential Revision: https://reviews.llvm.org/D43353

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328016 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MustExecute] Use the annotation style printer
Philip Reames [Tue, 20 Mar 2018 18:43:44 +0000 (18:43 +0000)]
[MustExecute] Use the annotation style printer

As suggested in the original review (https://reviews.llvm.org/D44524), use an annotation style printer instead.

Note: The switch from -analyze to -disable-output in tests was driven by the fact that seems to be the idiomatic style used in annoation passes.  I tried to keep both working, but the old style pass API for printers really doesn't make this easy.  It invokes (runOnFunction, print(Module)) repeatedly.  I decided the extra state wasn't worth it given the old pass manager is going away soonish anyway.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328015 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "Resubmit "Support embedding natvis files in PDBs.""
Zachary Turner [Tue, 20 Mar 2018 18:37:03 +0000 (18:37 +0000)]
Revert "Resubmit "Support embedding natvis files in PDBs.""

This is still failing on a different bot this time due to some
issue related to hashing absolute paths.  Reverting until I can
figure it out.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328014 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-objcopy] Implement support for section groups
Alexander Shaposhnikov [Tue, 20 Mar 2018 18:20:42 +0000 (18:20 +0000)]
[llvm-objcopy] Implement support for section groups

This diff adds support for SHT_GROUP sections to llvm-objcopy.
Some sections are interrelated and comprise a group.
For example, a definition of an inline function might require,
in addition to the section containing its instructions,
a read-only data section containing literals referenced inside the function.
A section of the type SHT_GROUP contains the indices of the group members,
therefore, it needs to be updated whenever the indices change.
Similarly, the fields sh_link, sh_info should be recalculated as well.

Test plan: make check-all

Differential revision: https://reviews.llvm.org/D43996

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328012 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-mca] Move the logic that computes the scheduler's queue usage to the BackendSta...
Andrea Di Biagio [Tue, 20 Mar 2018 18:20:39 +0000 (18:20 +0000)]
[llvm-mca] Move the logic that computes the scheduler's queue usage to the BackendStatistics view.

This patch introduces two new callbacks in the event listener interface to
handle the "buffered resource reserved" event and the "buffered resource
released" event. Every time a buffered resource is used, an event is generated.

Before this patch, the Scheduler (with the help of the ResourceManager) was
responsible for tracking the scheduler's queue usage. However, that design
forced the Scheduler to 'publish' scheduler's queue pressure information through
the Backend interface.

The goal of this patch is to break the dependency between the BackendStatistics
view, and the Backend. Now the Scheduler knows how to notify "buffer
reserved/released" events.  The scheduler's queue usage analysis has been moved
to the BackendStatistics.

Differential Revision: https://reviews.llvm.org/D44686

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328011 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[NVPTX] Make tensor load/store intrinsics overloaded.
Artem Belevich [Tue, 20 Mar 2018 17:18:59 +0000 (17:18 +0000)]
[NVPTX] Make tensor load/store intrinsics overloaded.

This way we can support address-space specific variants without explicitly
encoding the space in the name of the intrinsic. Less intrinsics to deal with ->
less boilerplate.

Added a bit of tablegen magic to match/replace an intrinsics with a pointer
argument in particular address space with the space-specific instruction
variant.

Updated tests to use non-default address spaces.

Differential Revision: https://reviews.llvm.org/D43268

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328006 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoAdd an analysis printer for must execute reasoning
Philip Reames [Tue, 20 Mar 2018 17:09:21 +0000 (17:09 +0000)]
Add an analysis printer for must execute reasoning

Many of our loop passes make use of so called "must execute" or "guaranteed to execute" facts to prove the legality of code motion. The basic notion is that we know (by assumption) an instruction didn't fault at it's original location, so if the location we move it to is strictly post dominated by the original, then we can't have introduced a new fault.

At the moment, the testing for this logic is somewhat adhoc and done mostly through LICM. Since I'm working on that code, I want to improve the testing. This patch is the first step in that direction. It doesn't actually test the variant used by the loop passes - I need to move that to the Analysis library first - but instead exercises an alternate implementation used by SCEV. (I plan on merging both implementations.)

Note: I'll be replacing the printing logic within this with an annotation based version in the near future.  Anna suggested this in review, and it seems like a strictly better format.

Differential Revision: https://reviews.llvm.org/D44524

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328004 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoMC: fix layering violation introduced in r325139.
Rafael Espindola [Tue, 20 Mar 2018 17:07:25 +0000 (17:07 +0000)]
MC: fix layering violation introduced in r325139.

Move AsmToken from MC/MCParser/MCAsmLexer.h into MC/MCAsmMacro.h in
order to avoid MCAsmMacro definition depending on MCParser.

Patch by Roger Pau Monné <royger@freebsd.org>.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328003 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoResubmit "Support embedding natvis files in PDBs."
Zachary Turner [Tue, 20 Mar 2018 17:06:39 +0000 (17:06 +0000)]
Resubmit "Support embedding natvis files in PDBs."

The issue causing this to fail in certain configurations
should be fixed.

It was due to the fact that DIA apparently expects there to be
a null string at ID 1 in the string table.  I'm not sure why this
is important but it seems to make a difference, so set it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328002 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[LangRef] fix link formatting
Sanjay Patel [Tue, 20 Mar 2018 17:05:19 +0000 (17:05 +0000)]
[LangRef] fix link formatting

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328001 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Correct the computation of TopReadyCycle and BotReadyCycle of SU
Krzysztof Parzyszek [Tue, 20 Mar 2018 17:03:27 +0000 (17:03 +0000)]
[Hexagon] Correct the computation of TopReadyCycle and BotReadyCycle of SU

TopReadyCycle and BotReadyCycle were off by one cycle when an SU is either
the first instruction or the last instruction in a packet.

Patch by Ikhlas Ajbar.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328000 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[XRay] Lazily compute MachineLoopInfo instead of requiring it.
Michael Zolotukhin [Tue, 20 Mar 2018 17:02:29 +0000 (17:02 +0000)]
[XRay] Lazily compute MachineLoopInfo instead of requiring it.

Summary:
Currently X-Ray Instrumentation pass has a dependency on MachineLoopInfo
(and thus on MachineDominatorTree as well) and we have to compute them
even if X-Ray is not used. This patch changes it to a lazy computation
to save compile time by avoiding these redundant computations.

Reviewers: dberris, kubamracek

Subscribers: llvm-commits, hiraditya

Differential Revision: https://reviews.llvm.org/D44666

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327999 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[LangRef] describe the default FP environment
Sanjay Patel [Tue, 20 Mar 2018 16:38:22 +0000 (16:38 +0000)]
[LangRef] describe the default FP environment

Follow-up for D44216: add a section and examples to describe the FP env.
Also, add pointers from the FP instructions to this new section to reduce
bloat.

Differential Revision: https://reviews.llvm.org/D44318

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327998 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Check weak dependences when only 1 instruction is available
Krzysztof Parzyszek [Tue, 20 Mar 2018 16:22:06 +0000 (16:22 +0000)]
[Hexagon] Check weak dependences when only 1 instruction is available

Patch by Brendon Cahoon.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327997 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64] add fabs tests for PR36600; NFC
Sanjay Patel [Tue, 20 Mar 2018 16:08:47 +0000 (16:08 +0000)]
[AArch64] add fabs tests for PR36600; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327995 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DEBUGINFO] Add flag -no-dwarf-pub-sections to disable pub sections.
Alexey Bataev [Tue, 20 Mar 2018 16:04:40 +0000 (16:04 +0000)]
[DEBUGINFO] Add flag -no-dwarf-pub-sections to disable pub sections.

Summary:
Added a flag -no-dwarf-pub-sections, which allows to disable
emission of DWARF public sections.

Reviewers: probinson, echristo

Subscribers: aprantl, JDevlieghere, llvm-commits

Differential Revision: https://reviews.llvm.org/D44385

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327994 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][Haswell][Znver1] Fix typo in fldl instregexs
Simon Pilgrim [Tue, 20 Mar 2018 15:44:47 +0000 (15:44 +0000)]
[X86][Haswell][Znver1] Fix typo in fldl instregexs

Missing comma was casing 2 instregex entries to be concatenated together by mistake.

Found while investigating PR35548

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327992 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Kaleidoscope] doc fix
Sjoerd Meijer [Tue, 20 Mar 2018 15:37:15 +0000 (15:37 +0000)]
[Kaleidoscope] doc fix

This fixes the compile command of the example in Chapter 2.

Differential Revision: https://reviews.llvm.org/D43196

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327989 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Improve scheduling heuristic for large basic blocks
Krzysztof Parzyszek [Tue, 20 Mar 2018 14:54:01 +0000 (14:54 +0000)]
[Hexagon] Improve scheduling heuristic for large basic blocks

This patch changes the isLatencyBound heuristic to look at the
path length based upon the number of packets needed to schedule
a basic block. For small basic blocks, the heuristic uses a small
threshold for isLatencyBound. For large basic blocks, the
heuristic uses a large threshold.

The goal is to increase the priority of an instruction in a small
basic block that has a large height or depth relative to the code
size. For large functions, the height and depth are ignored
because it increases the live range of a register and causes more
spills. That is, for large functions, it is more important to
schedule instructions when available, and attempt to keep the defs
and uses closer together.

Patch by Brendon Cahoon.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327987 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AArch64][Falkor] Correct load/store increment scheduling details
Geoff Berry [Tue, 20 Mar 2018 13:46:35 +0000 (13:46 +0000)]
[AArch64][Falkor] Correct load/store increment scheduling details

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327982 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Fix division by zero in machine scheduler
Krzysztof Parzyszek [Tue, 20 Mar 2018 13:28:46 +0000 (13:28 +0000)]
[Hexagon] Fix division by zero in machine scheduler

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327980 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[RISCV] Add codegen for RV32F floating point load/store
Alex Bradbury [Tue, 20 Mar 2018 13:26:12 +0000 (13:26 +0000)]
[RISCV] Add codegen for RV32F floating point load/store

As part of this, add support for load/store from the constant pool. This is
used to materialise f32 constants.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327979 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-mca] Use llvm::make_unique in a few places. NFC
Andrea Di Biagio [Tue, 20 Mar 2018 12:58:34 +0000 (12:58 +0000)]
[llvm-mca] Use llvm::make_unique in a few places. NFC

Also, clang-format a couple of DEBUG functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327978 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[RISCV] Add codegen for RV32F arithmetic and conversion operations
Alex Bradbury [Tue, 20 Mar 2018 12:45:35 +0000 (12:45 +0000)]
[RISCV] Add codegen for RV32F arithmetic and conversion operations

Currently, only a soft floating point ABI is supported.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327976 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Hexagon] Improve scheduling based on register pressure
Krzysztof Parzyszek [Tue, 20 Mar 2018 12:28:43 +0000 (12:28 +0000)]
[Hexagon] Improve scheduling based on register pressure

Patch by Brendon Cahoon.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327975 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86][SandyBridge] Merge multiple InstrRW entries that map to the same SchedWriteRes...
Simon Pilgrim [Tue, 20 Mar 2018 12:26:55 +0000 (12:26 +0000)]
[X86][SandyBridge] Merge multiple InstrRW entries that map to the same SchedWriteRes group (NFCI) (PR35955)

I've also merged some VEX/non-VEX instregex strings with a (V?) prefix - there are still a lot more of these to do.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327974 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-mca] Move the routine that computes processor resource masks to its own file.
Andrea Di Biagio [Tue, 20 Mar 2018 12:25:54 +0000 (12:25 +0000)]
[llvm-mca] Move the routine that computes processor resource masks to its own file.

Function computeProcResourceMasks is used by the ResourceManager (owned by the
Scheduler) to compute resource masks for processor resources.  Before this
refactoring, there was an implicit dependency between the Scheduler and the
InstrBuilder. That is because InstrBuilder has to know about resource masks when
computing the set of processor resources consumed by a new instruction.

With this patch, the functionality that computes resource masks has been
extracted from the ResourceManager, and moved to a separate file (Support.h).
This helps removing the dependency between the Scheduler and the InstrBuilder.

No functional change intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327973 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MergeICmps] Break eargerly out of loop
Xin Tong [Tue, 20 Mar 2018 12:03:25 +0000 (12:03 +0000)]
[MergeICmps] Break eargerly out of loop

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327972 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[MergeICmp] Fix a bug in entry block shuffled to middle of the chain
Xin Tong [Tue, 20 Mar 2018 11:57:54 +0000 (11:57 +0000)]
[MergeICmp] Fix a bug in entry block shuffled to middle of the chain

Summary: Fix a bug in entry block shuffled to middle of the chain.

Reviewers: davide, courbet

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D44642

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327971 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[llvm-opt-fuzzer] Add irce to the fuzzing options
Igor Laevsky [Tue, 20 Mar 2018 11:32:13 +0000 (11:32 +0000)]
[llvm-opt-fuzzer] Add irce to the fuzzing options

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327969 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRevert "Fix layering between llvm-c and Support by factoring out some typedefs into...
Jonas Devlieghere [Tue, 20 Mar 2018 11:26:50 +0000 (11:26 +0000)]
Revert "Fix layering between llvm-c and Support by factoring out some typedefs into Support"

This reverts r327940 because it fails on several GreenDragon bots.

  http://green.lab.llvm.org/green/job/clang-stage2-configure-Rlto/
  http://green.lab.llvm.org/green/job/clang-stage2-coverage-R/

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327968 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[AVR] Add a regression test for struct return lowering
Dylan McKay [Tue, 20 Mar 2018 11:23:03 +0000 (11:23 +0000)]
[AVR] Add a regression test for struct return lowering

The test is taken from
https://github.com/avr-rust/rust/issues/57

The originally implementation of struct return lowering was made in
r325474.

Patch by Peter Nimmervoll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327967 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[Release Notes] Add release note for llvm-mca.
Andrea Di Biagio [Tue, 20 Mar 2018 10:25:36 +0000 (10:25 +0000)]
[Release Notes] Add release note for llvm-mca.

Differential Revision: https://reviews.llvm.org/D44636

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327965 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[CGP] Avoid segmentation fault when doing PHI node simplifications
Bjorn Pettersson [Tue, 20 Mar 2018 09:06:37 +0000 (09:06 +0000)]
[CGP] Avoid segmentation fault when doing PHI node simplifications

Summary:
Made PHI node simplifiations more robust in several ways:

- Minor refactoring to let the SimplificationTracker own the
sets with new PHI/Select nodes that are introduced. This is
maybe not mapping to the original intention with the
SimplificationTracker, but IMHO it encapsulates the logic behind
those sets a little bit better.

- MatchPhiNode can sometimes populate the Matched set with
several entries, where it maps one PHI node to different candidates
for replacement. The Matched set is changed into a SmallSetVector
to make sure we get a deterministic iteration when doing
the replacements.

- As described above we may get several different replacements
for a single PHI node. The loop in MatchPhiSet that is doing
the replacements could end up calling eraseFromParent several
times for the same PHI node, resulting in segmentation faults.
This problem was supposed to be fixed in rL327250, but due to
the non-determinism(?) it only appeared to be fixed (I still
got crashes sometime when turning on/off -print-after-all etc
to get different iteration order in the DenseSets).
With this patch we follow the deterministic ordering in the
Matched set when replacing the PHI nodes. If we find a new
replacement for an already replaced PHI node we replace the
new replacement by the old replacement instead. This is quite
similar to what happened in the rl327250 patch, but here we
also recursively verify that the old replacement hasn't been
replaced already.

- It was really hard to track down the fault described above
(segementation fault due to doing eraseFromParent multiple
times for the same instruction). The fault was intermittent and
small changes in the code, or simply turning on -print-after-all
etc could make the problem go away. This was basically due to
the iteration over PhiNodesToMatch in MatchPhiSet no being
deterministic. Therefore I've changed the data structure for
the SimplificationTracker::AllPhiNodes into an SmallSetVector.
This gives a deterministic behavior.

Reviewers: skatkov, john.brawn

Reviewed By: skatkov

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D44571

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327961 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[LV] Let recordVectorLoopValueForInductionCast to check if IV was created from the...
Andrei Elovikov [Tue, 20 Mar 2018 09:04:39 +0000 (09:04 +0000)]
[LV] Let recordVectorLoopValueForInductionCast to check if IV was created from the cast.

Summary:
It turned out to be error-prone to expect the callers to handle that - better to
leave the decision to this routine and make the required data to be explicitly
passed to the function.

This handles the case that was missed in the r322473 and fixes the assert
mentioned in PR36524.

Reviewers: dorit, mssimpso, Ayal, dcaballe

Reviewed By: dcaballe

Subscribers: Ka-Ka, hiraditya, dneilson, hsaito, llvm-commits

Differential Revision: https://reviews.llvm.org/D43812

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327960 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[SystemZ] Add "REQUIRES: asserts" to test case to fix build bots.
Jonas Paulsson [Tue, 20 Mar 2018 08:29:19 +0000 (08:29 +0000)]
[SystemZ]  Add "REQUIRES: asserts" to test case to fix build bots.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327958 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Properly implement the calling convention for f80 for mingw/x86_64
Martin Storsjo [Tue, 20 Mar 2018 06:19:38 +0000 (06:19 +0000)]
[X86] Properly implement the calling convention for f80 for mingw/x86_64

In these cases, both parameters and return values are passed
as a pointer to a stack allocation.

MSVC doesn't use the f80 data type at all, while it is used
for long doubles on mingw.

Normally, this part of the calling convention is handled
within clang, but for intrinsics that are lowered to libcalls,
it may need to be handled within llvm as well.

Differential Revision: https://reviews.llvm.org/D44592

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327957 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ORC] Don't fully qualify explicit destructor call -- it confuses some compilers.
Lang Hames [Tue, 20 Mar 2018 05:56:58 +0000 (05:56 +0000)]
[ORC] Don't fully qualify explicit destructor call -- it confuses some compilers.

This should fix the builder failure at
http://lab.llvm.org:8011/builders/lld-x86_64-darwin13/builds/19224

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327955 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Rename MOVSX32_NOREXrr8 to MOVSX32rr8_NOREX so that the scheduler model regular...
Craig Topper [Tue, 20 Mar 2018 05:00:20 +0000 (05:00 +0000)]
[X86] Rename MOVSX32_NOREXrr8 to MOVSX32rr8_NOREX so that the scheduler model regular expressions will pick it up with the regular version.

Do the same for MOVSX32_NOREXrm8, MOVZX32_NOREXrr8, and MOVZX32_NOREXrm8

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327948 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Fix the SchedRW for memory forms of CMP and TEST.
Craig Topper [Tue, 20 Mar 2018 03:55:17 +0000 (03:55 +0000)]
[X86] Fix the SchedRW for memory forms of CMP and TEST.

They were incorrectly marked as RMW operations. Some of the CMP instrucions worked, but the ones that use a similar encoding as RMW form of ADD ended up marked as RMW.

TEST used the same tablegen class as some of the CMPs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327947 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ORC] Rename SymbolSource to MaterializationUnit, and make the materialization
Lang Hames [Tue, 20 Mar 2018 03:49:29 +0000 (03:49 +0000)]
[ORC] Rename SymbolSource to MaterializationUnit, and make the materialization
operation all-or-nothing, rather than allowing materialization on a per-symbol
basis.

This addresses a shortcoming of per-symbol materialization: If a
MaterializationUnit (/SymbolSource) wants to materialize more symbols than
requested (which is likely: most materializers will want to materialize whole
modules) then it needs a way to notify the symbol table about the extra symbols
being materialized. This process (checking what has been requested against what
is being provided and notifying the symbol table about the difference) has to
be repeated at every level of the JIT stack. Making materialization
all-or-nothing eliminates this issue, simplifying both materializer
implementations and the symbol table (VSO class) API. The cost is that
per-symbol materialization (e.g. for individual symbols in a module) now
requires multiple MaterializationUnits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327946 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add TEST16mi/TEST32mi/TEST64mi32 to the Sandybridge/Haswell/Broadwell/Skylake...
Craig Topper [Tue, 20 Mar 2018 03:02:03 +0000 (03:02 +0000)]
[X86] Add TEST16mi/TEST32mi/TEST64mi32 to the Sandybridge/Haswell/Broadwell/Skylake scheduler models.

Move it from a load+store group on SNB to a load only group, the same group as CMP.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327944 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Add ROR/ROL/SHR/SAR by 1 instructions to the Sandy Bridge scheduler model.
Craig Topper [Tue, 20 Mar 2018 03:01:59 +0000 (03:01 +0000)]
[X86] Add ROR/ROL/SHR/SAR by 1 instructions to the Sandy Bridge scheduler model.

I assume these match the generic immediate version like they do in the other models.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327943 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[ShrinkWrap] Take into account landing pad
Quentin Colombet [Tue, 20 Mar 2018 02:44:40 +0000 (02:44 +0000)]
[ShrinkWrap] Take into account landing pad

When scanning the function for CSRs uses and defs, also check if
the basic block are landing pads.
Consider that landing pads needs the CSRs to be properly set.
That way we force the prologue/epilogue to always be pushed out
of the problematic "throw" region. The "throw" region is
problematic because the jumps are not properly modeled.

Fixes PR36513

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327942 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoFix layering between llvm-c and Support by factoring out some typedefs into Support
David Blaikie [Tue, 20 Mar 2018 02:14:50 +0000 (02:14 +0000)]
Fix layering between llvm-c and Support by factoring out some typedefs into Support

llvm-c depends on Support, but Support (TargetRegistry) uses some of
llvm-c's typedefs. Move those into a Support header to be used from both
llvm-c and Support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327940 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[RISCV] Preserve stack space for outgoing arguments when the function contain variabl...
Shiva Chen [Tue, 20 Mar 2018 01:39:17 +0000 (01:39 +0000)]
[RISCV] Preserve stack space for outgoing arguments when the function contain variable size objects

E.g.

bar (int x)
{
  char p[x];

  push outgoing variables for foo.
  call foo
}

We need to generate stack adjustment instructions for outgoing arguments by
eliminateCallFramePseudoInstr when the function contains variable size
objects to avoid outgoing variables corrupt the variable size object.

Default hasReservedCallFrame will return !hasFP().
We don't want to generate extra sp adjustment instructions when hasFP()
return true, So We override hasReservedCallFrame as !hasVarSizedObjects().

Differential Revision: https://reviews.llvm.org/D43752

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327938 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[X86] Fix the SNB scheduler for BLENDVB.
Craig Topper [Tue, 20 Mar 2018 01:30:21 +0000 (01:30 +0000)]
[X86] Fix the SNB scheduler for BLENDVB.

PBLENDVBrr0 was with the memory version of VBLENDVB and PBLENDVBrm0 was missing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327937 91177308-0d34-0410-b5e6-96231b3b80d8

6 years agoRun dos2unix on a test. NFC.
Rafael Espindola [Tue, 20 Mar 2018 01:06:29 +0000 (01:06 +0000)]
Run dos2unix on a test. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327934 91177308-0d34-0410-b5e6-96231b3b80d8

6 years ago[DebugInfoPDB] Add backward support flags to PDB_NameSearchFlags
Aaron Smith [Tue, 20 Mar 2018 01:04:21 +0000 (01:04 +0000)]
[DebugInfoPDB] Add backward support flags to PDB_NameSearchFlags

For NS_CaseInFileNameExt support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327933 91177308-0d34-0410-b5e6-96231b3b80d8