OSDN Git Service

android-x86/external-llvm.git
7 years agoObject: Add IRObjectFile::getTargetTriple().
Peter Collingbourne [Thu, 24 Nov 2016 01:13:09 +0000 (01:13 +0000)]
Object: Add IRObjectFile::getTargetTriple().

This lets us remove a use of IRObjectFile::getModule() in llvm-nm.

Differential Revision: https://reviews.llvm.org/D27074

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287846 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoObject: Simplify the IRObjectFile symbol iterator implementation.
Peter Collingbourne [Thu, 24 Nov 2016 00:41:05 +0000 (00:41 +0000)]
Object: Simplify the IRObjectFile symbol iterator implementation.

Change the IRObjectFile symbol iterator to be a pointer into a vector of
PointerUnions representing either IR symbols or asm symbols.

This change is in preparation for a future change for supporting multiple
modules in an IRObjectFile. Although it causes an increase in memory
consumption, we can deal with that issue separately by introducing a bitcode
symbol table.

Differential Revision: https://reviews.llvm.org/D26928

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287845 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU: Preserve m0 value when spilling
Matt Arsenault [Thu, 24 Nov 2016 00:26:50 +0000 (00:26 +0000)]
AMDGPU: Preserve m0 value when spilling

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287844 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoTRI: Add hook to pass scavenger during frame elimination
Matt Arsenault [Thu, 24 Nov 2016 00:26:47 +0000 (00:26 +0000)]
TRI: Add hook to pass scavenger during frame elimination

The scavenger was not passed if requiresFrameIndexScavenging was
enabled. I need to be able to test for the availability of an
unallocatable register here, so I can't create a virtual register for
it.

It might be better to just always use the scavenger and stop
creating virtual registers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287843 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU: Remove m0 spilling code
Matt Arsenault [Thu, 24 Nov 2016 00:26:44 +0000 (00:26 +0000)]
AMDGPU: Remove m0 spilling code

Since m0 isn't allocatable it should never be spilled anymore.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287842 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU: Make m0 unallocatable
Matt Arsenault [Thu, 24 Nov 2016 00:26:40 +0000 (00:26 +0000)]
AMDGPU: Make m0 unallocatable

m0 may need to be written for spill code, so
we don't want general code uses relying on the
value stored in it.

This introduces a few code quality regressions where copies
from m0 are not coalesced into copies of a copy of m0.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287841 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[lib/LTO] Rename few instances of Lto to LTO.
Davide Italiano [Thu, 24 Nov 2016 00:23:09 +0000 (00:23 +0000)]
[lib/LTO] Rename few instances of Lto to LTO.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287840 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRely on a single DWARF version instead of having two copies
Greg Clayton [Wed, 23 Nov 2016 23:30:37 +0000 (23:30 +0000)]
Rely on a single DWARF version instead of having two copies

This patch makes AsmPrinter less reliant on DwarfDebug by relying on the DWARF version in the AsmPrinter's MCStreamer's MCContext. This allows us to remove the redundant DWARF version from DwarfDebug. It also lets us change code that used to access the AsmPrinter's DwarfDebug just to get to the DWARF version by changing the DWARF version accessor on AsmPrinter so that it grabs the version from its MCStreamer's MCContext.

Differential Revision: https://reviews.llvm.org/D27032

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287839 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DebugInfo] Fix some Clang-tidy modernize-use-default and Include What You Use warnin...
Eugene Zelenko [Wed, 23 Nov 2016 23:16:32 +0000 (23:16 +0000)]
[DebugInfo] Fix some Clang-tidy modernize-use-default and Include What You Use warnings; other minor fixes (NFC).

Per Zachary Turner and Mehdi Amini suggestion to make only post-commit reviews.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287838 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][SSE] Add awareness of (v)cvtpd2dq and vcvtpd2udq implicit zeroing of upper...
Simon Pilgrim [Wed, 23 Nov 2016 22:35:06 +0000 (22:35 +0000)]
[X86][SSE] Add awareness of (v)cvtpd2dq and vcvtpd2udq implicit zeroing of upper 64-bits of xmm result

We've already added the equivalent for (v)cvttpd2dq (rL284459) and vcvttpd2udq

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287835 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[IR] Fix some Clang-tidy modernize-use-default, modernize-use-equal-delete and Includ...
Eugene Zelenko [Wed, 23 Nov 2016 22:25:16 +0000 (22:25 +0000)]
[IR] Fix some Clang-tidy modernize-use-default, modernize-use-equal-delete and Include What You Use warnings; other minor fixes (NFC).

Per Zachary Turner and Mehdi Amini suggestion to make only post-commit reviews.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287834 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SelectionDAG] Early-out in TargetLowering::expandMUL (NFC)
Nicolai Haehnle [Wed, 23 Nov 2016 22:14:20 +0000 (22:14 +0000)]
[SelectionDAG] Early-out in TargetLowering::expandMUL (NFC)

Summary: Reduce indentation level; preparation for D24956.

Reviewers: efriedma

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D27063

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287831 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][AVX512VL] Add v2f64 -> v2i32/v2f32 + zero codegen tests
Simon Pilgrim [Wed, 23 Nov 2016 22:01:50 +0000 (22:01 +0000)]
[X86][AVX512VL] Add v2f64 -> v2i32/v2f32 + zero codegen tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287821 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU: Cleanup immediate folding code
Matt Arsenault [Wed, 23 Nov 2016 21:51:07 +0000 (21:51 +0000)]
AMDGPU: Cleanup immediate folding code

Move code down to use, reorder to avoid hard to follow
immediate folding logic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287818 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU: Fix debug printing
Matt Arsenault [Wed, 23 Nov 2016 21:51:05 +0000 (21:51 +0000)]
AMDGPU: Fix debug printing

The uint8_t was printed as a char which didn't really work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287817 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][SSE] Add v2i64 -> v2i32 + zero codegen test
Simon Pilgrim [Wed, 23 Nov 2016 21:19:57 +0000 (21:19 +0000)]
[X86][SSE] Add v2i64 -> v2i32 + zero codegen test

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287813 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU: Fix not setting kill flag on temp reg when spilling
Matt Arsenault [Wed, 23 Nov 2016 21:00:12 +0000 (21:00 +0000)]
AMDGPU: Fix not setting kill flag on temp reg when spilling

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287808 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU: Fix adding extra implicit def of register
Matt Arsenault [Wed, 23 Nov 2016 21:00:10 +0000 (21:00 +0000)]
AMDGPU: Fix adding extra implicit def of register

In the scalar case, there's no reason to add an additional
def of the same register.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287807 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU: Fix MMO when splitting spill
Matt Arsenault [Wed, 23 Nov 2016 20:52:53 +0000 (20:52 +0000)]
AMDGPU: Fix MMO when splitting spill

The size and offset were wrong. The size of the object was
being used for the size of the access, when here it is really
being split into 4-byte accesses. The underlying object size
is set in the MachinePointerInfo, which also didn't have the
offset set.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287806 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert "[lit] When setting SDKROOT on Darwin, use '--sdk macosx' to find the right...
Vedant Kumar [Wed, 23 Nov 2016 20:51:09 +0000 (20:51 +0000)]
Revert "[lit] When setting SDKROOT on Darwin, use '--sdk macosx' to find the right SDK path."

This reverts commit r287403. It breaks an internal asan bot. According
to Kuba, a fix is up for review here: https://reviews.llvm.org/D26929

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287804 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agollvm-nm: Print correct symbol types for init and fini sections
Meador Inge [Wed, 23 Nov 2016 20:17:20 +0000 (20:17 +0000)]
llvm-nm: Print correct symbol types for init and fini sections

This patch fixes a small bug where symbols defined in the INIT
and FINI sections were incorrectly getting a type of 'n'.

Differential Revision: https://reviews.llvm.org/D26937

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287803 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agollvm-nm: Don't print value or size for undefined or weak symbols
Meador Inge [Wed, 23 Nov 2016 20:17:15 +0000 (20:17 +0000)]
llvm-nm: Don't print value or size for undefined or weak symbols

Undefined and weak symbols don't have a meaningful size or value.
As such, nothing should be printed for those attributes (this is
already done for the address with 'U') with the BSD format.  This
matches what GNU nm does.

Note that for the POSIX.2 format [1] zero values are still
printed for the size and value.  This seems in spirit with
the format strings in that specification, but is debatable.

[1] http://pubs.opengroup.org/onlinepubs/9699919799/

Differential Revision: https://reviews.llvm.org/D26936

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287802 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SLP] Add more tests for SLP Vectorizer.
Alexey Bataev [Wed, 23 Nov 2016 20:10:32 +0000 (20:10 +0000)]
[SLP] Add more tests for SLP Vectorizer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287801 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LoopUnroll] Move code to exit early. NFC.
Haicheng Wu [Wed, 23 Nov 2016 19:39:26 +0000 (19:39 +0000)]
[LoopUnroll] Move code to exit early. NFC.

Just to save some compilation time.

Differential Revision: https://reviews.llvm.org/D26784

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287800 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert "[Triple] Add Facebook vendor"
Daniel Berlin [Wed, 23 Nov 2016 19:03:54 +0000 (19:03 +0000)]
Revert "[Triple] Add Facebook vendor"

This reverts commit r287684

Objections on the review thread had not been addressed to
prior to commit.  I asked the committer to revert, but i expect they
are gone for the US holiday or something.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287798 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Allow folding of stack reloads when loading a subreg of the spilled reg
Michael Kuperstein [Wed, 23 Nov 2016 18:33:49 +0000 (18:33 +0000)]
[X86] Allow folding of stack reloads when loading a subreg of the spilled reg

We did not support subregs in InlineSpiller:foldMemoryOperand() because targets
may not deal with them correctly.

This adds a target hook to let the spiller know that a target can handle
subregs, and actually enables it for x86 for the case of stack slot reloads.
This fixes PR30832.

Differential Revision: https://reviews.llvm.org/D26521

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287792 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agollvm-readobj: Use hash tables to print dynamic symbols.
Hemant Kulkarni [Wed, 23 Nov 2016 18:04:23 +0000 (18:04 +0000)]
llvm-readobj: Use hash tables to print dynamic symbols.

-symbols prints both .symtab and .dynsym symbols for GNU style in ELF.
-dyn-symbols prints symbols looking up through hash tables. This helps validate hash tables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287786 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PM] Change the static object whose address is used to uniquely identify
Chandler Carruth [Wed, 23 Nov 2016 17:53:26 +0000 (17:53 +0000)]
[PM] Change the static object whose address is used to uniquely identify
analyses to have a common type which is enforced rather than using
a char object and a `void *` type when used as an identifier.

This has a number of advantages. First, it at least helps some of the
confusion raised in Justin Lebar's code review of why `void *` was being
used everywhere by having a stronger type that connects to documentation
about this.

However, perhaps more importantly, it addresses a serious issue where
the alignment of these pointer-like identifiers was unknown. This made
it hard to use them in pointer-like data structures. We were already
dodging this in dangerous ways to create the "all analyses" entry. In
a subsequent patch I attempted to use these with TinyPtrVector and
things fell apart in a very bad way.

And it isn't just a compile time or type system issue. Worse than that,
the actual alignment of these pointer-like opaque identifiers wasn't
guaranteed to be a useful alignment as they were just characters.

This change introduces a type to use as the "key" object whose address
forms the opaque identifier. This both forces the objects to have proper
alignment, and provides type checking that we get it right everywhere.
It also makes the types somewhat less mysterious than `void *`.

We could go one step further and introduce a truly opaque pointer-like
type to return from the `ID()` static function rather than returning
`AnalysisKey *`, but that didn't seem to be a clear win so this is just
the initial change to get to a reliably typed and aligned object serving
is a key for all the analyses.

Thanks to Richard Smith and Justin Lebar for helping pick plausible
names and avoid making this refactoring many times. =] And thanks to
Sean for the super fast review!

While here, I've tried to move away from the "PassID" nomenclature
entirely as it wasn't really helping and is overloaded with old pass
manager constructs. Now we have IDs for analyses, and key objects whose
address can be used as IDs. Where possible and clear I've shortened this
to just "ID". In a few places I kept "AnalysisID" to make it clear what
was being identified.

Differential Revision: https://reviews.llvm.org/D27031

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287783 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LoadStoreVectorizer] Enable vectorization of stores in the presence of an aliasing...
Alina Sbirlea [Wed, 23 Nov 2016 17:43:15 +0000 (17:43 +0000)]
[LoadStoreVectorizer] Enable vectorization of stores in the presence of an aliasing load

Summary:
The "getVectorizablePrefix" method would give up if it found an aliasing load for a store chain.
In practice, the aliasing load can be treated as a memory barrier and all stores that precede it
are a valid vectorizable prefix.
Issue found by volkan in D26962. Testcase is a pruned version of the one in the original patch.

Reviewers: jlebar, arsenm, tstellarAMD

Subscribers: mzolotukhin, wdng, nhaehnle, anna, volkan, llvm-commits

Differential Revision: https://reviews.llvm.org/D27008

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287781 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DAG] Improve loads-from-store forwarding to handle TokenFactor
Nirav Dave [Wed, 23 Nov 2016 16:48:35 +0000 (16:48 +0000)]
[DAG] Improve loads-from-store forwarding to handle TokenFactor

Forward store values to matching loads down through token
factors. Factored from D14834.

Reviewers: jyknight, hfinkel

Subscribers: hfinkel, nemanjai, llvm-commits

Differential Revision: https://reviews.llvm.org/D26080

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287773 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix doc of `llvm.bitreverse.iN`
Yichao Yu [Wed, 23 Nov 2016 16:25:31 +0000 (16:25 +0000)]
Fix doc of `llvm.bitreverse.iN`

Summary:
The return type is `iN` rather than always `i16`

Seems to be a typo in https://reviews.llvm.org/rL252878 .

Reviewers: jmolloy

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D27047

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287769 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DAGCombiner] Fix infinite loop in vector mul/shl combining
John Brawn [Wed, 23 Nov 2016 16:05:51 +0000 (16:05 +0000)]
[DAGCombiner] Fix infinite loop in vector mul/shl combining

We have the following DAGCombiner transformations:
 (mul (shl X, c1), c2) -> (mul X, c2 << c1)
 (mul (shl X, C), Y) -> (shl (mul X, Y), C)
 (shl (mul x, c1), c2) -> (mul x, c1 << c2)
Usually the constant shift is optimised by SelectionDAG::getNode when it is
constructed, by SelectionDAG::FoldConstantArithmetic, but when we're dealing
with vectors and one of those vector constants contains an undef element
FoldConstantArithmetic does not fold and we enter an infinite loop.

Fix this by making FoldConstantArithmetic use getNode to decide how to fold each
vector element, the same as FoldConstantVectorArithmetic does, and rather than
adding the constant shift to the work list instead only apply the transformation
if it's already been folded into a constant, as if it's not we're going to loop
endlessly. Additionally add missing NoOpaques to one of those transformations,
which I noticed when writing the tests for this.

Differential Revision: https://reviews.llvm.org/D26605

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287766 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PowerPC] Remove InstAlias definitions that cause incorrect assembly
Nemanja Ivanovic [Wed, 23 Nov 2016 15:51:52 +0000 (15:51 +0000)]
[PowerPC] Remove InstAlias definitions that cause incorrect assembly

In rL283190, I added some InstAlias definitions to generate extended mnemonics
for some uses of the XXPERMDI instruction. However, when the assembler matches
these extended mnemonics, it matches the new instruction in situations where it
should match the old one.
This patch removes these definitions and accomplishes that by defining these
mnemonics with additional instructions that are isCodeGenOnly.

Fixes PR31127.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287765 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][AVX512] Add support for v4i64 fptosi/fptoui/sitofp/uitofp on AVX512DQ-only...
Simon Pilgrim [Wed, 23 Nov 2016 14:01:18 +0000 (14:01 +0000)]
[X86][AVX512] Add support for v4i64 fptosi/fptoui/sitofp/uitofp on AVX512DQ-only targets

Use 512-bit instructions with subvector insertion/extraction like we do in a number of similar circumstances

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287762 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoType legalization for compressstore and expandload intrinsics.
Elena Demikhovsky [Wed, 23 Nov 2016 13:58:24 +0000 (13:58 +0000)]
Type legalization for compressstore and expandload intrinsics.

Implemented widening (v2f32) and splitting (v16f64).
On splitting, I use "popcnt" to calculate memory increment.
More type legalization work will come in the next patches.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287761 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CostModel][X86] Add missing AVX512DQ v8i64 fptosi/sitofp costs
Simon Pilgrim [Wed, 23 Nov 2016 13:42:09 +0000 (13:42 +0000)]
[CostModel][X86] Add missing AVX512DQ v8i64 fptosi/sitofp costs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287760 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[MD5] Use write32le instead of spelling it out with shifts.
Benjamin Kramer [Wed, 23 Nov 2016 11:49:28 +0000 (11:49 +0000)]
[MD5] Use write32le instead of spelling it out with shifts.

No functionality change intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287757 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CostModel][X86] Add v2f32 -> v2i64 fptosi/fptoui cost tests
Simon Pilgrim [Wed, 23 Nov 2016 11:43:00 +0000 (11:43 +0000)]
[CostModel][X86] Add v2f32 -> v2i64 fptosi/fptoui cost tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287756 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AVX-512] Remove intrinsics for valignd/q and autoupgrade them to native shuffles.
Craig Topper [Wed, 23 Nov 2016 06:54:55 +0000 (06:54 +0000)]
[AVX-512] Remove intrinsics for valignd/q and autoupgrade them to native shuffles.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287744 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Simplify lowerVectorShuffleAsBitMask to handle only integer VT's
Zvi Rackover [Wed, 23 Nov 2016 06:45:25 +0000 (06:45 +0000)]
[X86] Simplify lowerVectorShuffleAsBitMask to handle only integer VT's

Summary: This function is only called with integer VT arguments, so remove code that handles FP vectors.

Reviewers: RKSimon, craig.topper, delena, andreadb

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D26985

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287743 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix builbots.
Rui Ueyama [Wed, 23 Nov 2016 03:58:12 +0000 (03:58 +0000)]
Fix builbots.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287735 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[xray] Add XRay support for Mach-O in CodeGen
Kuba Mracek [Wed, 23 Nov 2016 02:07:04 +0000 (02:07 +0000)]
[xray] Add XRay support for Mach-O in CodeGen

Currently, XRay only supports emitting the XRay table (xray_instr_map) on ELF binaries. Let's add Mach-O support.

Differential Revision: https://reviews.llvm.org/D26983

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287734 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SCCP] Add a test for switches on undef.
Davide Italiano [Wed, 23 Nov 2016 01:42:39 +0000 (01:42 +0000)]
[SCCP] Add a test for switches on undef.

Without this test, you can just remove the code fixing the
switch to the first constant in ResolvedUndefs in and everything
pass. This test, instead, fails with an assertion if the code
is removed. Found while refactoring SCCP to integrate undef in
the solver.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287731 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdd convenient functions to compute hashes of byte vectors.
Rui Ueyama [Wed, 23 Nov 2016 00:46:09 +0000 (00:46 +0000)]
Add convenient functions to compute hashes of byte vectors.

In many sitautions, you just want to compute a hash for one chunk
of data. This patch adds convenient functions for that purpose.

Differential Revision: https://reviews.llvm.org/D26988

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287726 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ADT] Fix some Clang-tidy modernize-use-default and Include What You Use warnings...
Eugene Zelenko [Wed, 23 Nov 2016 00:30:24 +0000 (00:30 +0000)]
[ADT] Fix some Clang-tidy modernize-use-default and Include What You Use warnings; other minor fixes.

Differential revision: https://reviews.llvm.org/D27001

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287725 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoMake STL range adapter naming consistent.
Zachary Turner [Wed, 23 Nov 2016 00:27:23 +0000 (00:27 +0000)]
Make STL range adapter naming consistent.

Differential Revision: https://reviews.llvm.org/D27009

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287724 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdd some searching functions for ArrayRef<T>.
Zachary Turner [Tue, 22 Nov 2016 23:22:19 +0000 (23:22 +0000)]
Add some searching functions for ArrayRef<T>.

Differential Revision: https://reviews.llvm.org/D26999

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287722 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[StructurizeCFG] Refactor OrderNodes.
Justin Lebar [Tue, 22 Nov 2016 23:14:11 +0000 (23:14 +0000)]
[StructurizeCFG] Refactor OrderNodes.

Summary:
No need to copy the RPOT vector before using it.  Switch from std::map
to SmallDenseMap.  Get rid of an unused variable (TempVisited).  Get rid
of a typedef, RNVector, which is now used only once.

Differential Revision: https://reviews.llvm.org/D26997

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287721 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[StructurizeCFG] Add whitespace in getAnalysisUsage.
Justin Lebar [Tue, 22 Nov 2016 23:14:07 +0000 (23:14 +0000)]
[StructurizeCFG] Add whitespace in getAnalysisUsage.

Summary:
"addRequired" and "addPreserved" look very similar when squished up next
to each other -- without the newline this code looked to me like it was
addRequired'ing DominatorTreeWrapperPass twice.

Reviewers: arsenm

Subscribers: wdng, llvm-commits

Differential Revision: https://reviews.llvm.org/D26996

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287720 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[StructurizeCFG] Remove unnecessary "using" in class.
Justin Lebar [Tue, 22 Nov 2016 23:13:49 +0000 (23:13 +0000)]
[StructurizeCFG] Remove unnecessary "using" in class.

Reviewers: arsenm

Subscribers: wdng, llvm-commits

Differential Revision: https://reviews.llvm.org/D26995

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287719 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[StructurizeCFG] Merge the two constructors into one.
Justin Lebar [Tue, 22 Nov 2016 23:13:44 +0000 (23:13 +0000)]
[StructurizeCFG] Merge the two constructors into one.

Reviewers: arsenm

Subscribers: wdng, llvm-commits

Differential Revision: https://reviews.llvm.org/D26994

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287718 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[StructurizeCFG] Use a for-each loop instead of iterators in runOnRegion.
Justin Lebar [Tue, 22 Nov 2016 23:13:37 +0000 (23:13 +0000)]
[StructurizeCFG] Use a for-each loop instead of iterators in runOnRegion.

Summary:

Reviewers: arsenm

Subscribers: wdng, llvm-commits

Differential Revision: https://reviews.llvm.org/D26993

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287717 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[StructurizeCFG] Make hasOnlyUniformBranches a non-member function.
Justin Lebar [Tue, 22 Nov 2016 23:13:33 +0000 (23:13 +0000)]
[StructurizeCFG] Make hasOnlyUniformBranches a non-member function.

Summary: Lets us get rid of one member variable too.

Reviewers: arsenm

Subscribers: wdng, llvm-commits

Differential Revision: https://reviews.llvm.org/D26992

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287716 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CUDA] Note in docs that you need to build with -lcudart on MacOS
Justin Lebar [Tue, 22 Nov 2016 23:13:29 +0000 (23:13 +0000)]
[CUDA] Note in docs that you need to build with -lcudart on MacOS

-lcudart_static doesn't work.  We don't know why.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287715 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoadd and use isBitwiseLogicOp() helper function; NFCI
Sanjay Patel [Tue, 22 Nov 2016 22:54:36 +0000 (22:54 +0000)]
add and use isBitwiseLogicOp() helper function; NFCI

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287712 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoBefore sample pgo annotation, do not inline a function that has no debug info. (NFC)
Dehao Chen [Tue, 22 Nov 2016 22:50:01 +0000 (22:50 +0000)]
Before sample pgo annotation, do not inline a function that has no debug info. (NFC)

If there is no debug info in the callee, inlining it will not help annotator. This avoids infinite loop as reported in PR/31119.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287710 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SCCP] Remove code in visitBinaryOperator (and add tests).
Davide Italiano [Tue, 22 Nov 2016 22:11:25 +0000 (22:11 +0000)]
[SCCP] Remove code in visitBinaryOperator (and add tests).

We visit and/or, we try to derive a lattice value for the
instruction even if one of the operands is overdefined.
If the non-overdefined value is still 'unknown' just return and wait
for ResolvedUndefsIn to "plug in" the correct value. This simplifies
the logic a bit. While I'm here add tests for missing cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287709 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoTargetSubtargetInfo: Move implementation to lib/CodeGen; NFC
Matthias Braun [Tue, 22 Nov 2016 22:09:03 +0000 (22:09 +0000)]
TargetSubtargetInfo: Move implementation to lib/CodeGen; NFC

TargetSubtargetInfo is filled with CodeGen specific interfaces nowadays
(getInstrInfo(), getFrameLowering(), getSelectionDAGInfo()) most of the
tuning flags like enablePostRAScheduler(), getAntiDepBreakMode(),
enableRALocalReassignment(), ... also do not seem to be universal enough
to make sense outside of CodeGen.

Differential Revision: https://reviews.llvm.org/D26948

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287708 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstCombine] change bitwise logic type to eliminate bitcasts
Sanjay Patel [Tue, 22 Nov 2016 22:05:48 +0000 (22:05 +0000)]
[InstCombine] change bitwise logic type to eliminate bitcasts

In PR27925:
https://llvm.org/bugs/show_bug.cgi?id=27925

...we proposed adding this fold to eliminate a bitcast. In D20774, there was
some concern about changing the type of a bitwise op as well as creating
bitcasts that might not be free for a target. However, if we're strictly
eliminating an instruction (by limiting this to one-use ops), then we should
be able to do this in InstCombine.

But we're cautiously restricting the transform for now to vector types to
avoid possible backend problems. A transform to make sure the logic op is
legal for the target should be added to reverse this transform and improve
codegen.

Differential Revision: https://reviews.llvm.org/D26641

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287707 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][AVX512DQ] Add fp <-> int tests for AVX512DQ/AVX512DQ+VL
Simon Pilgrim [Tue, 22 Nov 2016 22:04:50 +0000 (22:04 +0000)]
[X86][AVX512DQ] Add fp <-> int tests for AVX512DQ/AVX512DQ+VL

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287706 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LCG] Add a previously missing assert about the relationship of RefSCCs.
Chandler Carruth [Tue, 22 Nov 2016 21:40:10 +0000 (21:40 +0000)]
[LCG] Add a previously missing assert about the relationship of RefSCCs.

No intended change, everything seems to be in working order already.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287705 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoLTO: Remove a now-unused InputFile accessor.
Peter Collingbourne [Tue, 22 Nov 2016 21:25:30 +0000 (21:25 +0000)]
LTO: Remove a now-unused InputFile accessor.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287702 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFixed the lost FastMathFlags in GVN(Global Value Numbering).
Vyacheslav Klochkov [Tue, 22 Nov 2016 20:52:53 +0000 (20:52 +0000)]
Fixed the lost FastMathFlags in GVN(Global Value Numbering).
Reviewer: Hal Finkel.
Differential Revision: https://reviews.llvm.org/D26952

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287700 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LCG] Start using SCC relationship predicates in the unittest.
Chandler Carruth [Tue, 22 Nov 2016 20:35:32 +0000 (20:35 +0000)]
[LCG] Start using SCC relationship predicates in the unittest.

This mostly gives us nice unittesting of the predicates themselves. I'll
start using them further in subsequent commits to help test the actual
operations performed on the graph.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287698 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRemove PDBFileBuilder::build() and related functions.
Rui Ueyama [Tue, 22 Nov 2016 20:32:22 +0000 (20:32 +0000)]
Remove PDBFileBuilder::build() and related functions.

PDBFileBuilder supports two different ways to create files.
One is PDBFileBuilder::commit. That function takes a filename
and write a result to the file. The other is PDBFileBuilder::build.
That returns a new PDBFile object.

This patch removes the latter because no one is using it and
in a real life situation we are very unlikely to need it.
Even if you need it, it'd be easy to write a new PDB to a memory
buffer and read it back.

Removing PDBFileBuilder::build enables us to remove other classes
build transitively.

Differential Revision: https://reviews.llvm.org/D26987

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287697 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFixed the lost FastMathFlags in Reassociate optimization.
Vyacheslav Klochkov [Tue, 22 Nov 2016 20:23:04 +0000 (20:23 +0000)]
Fixed the lost FastMathFlags in Reassociate optimization.
Reviewer: Hal Finkel.
Differential Revision: https://reviews.llvm.org/D26957

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287695 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRestructure DwarfDebug::beginInstruction(). [NFC]
Paul Robinson [Tue, 22 Nov 2016 19:46:51 +0000 (19:46 +0000)]
Restructure DwarfDebug::beginInstruction(). [NFC]
Will help a pending patch.

Differential Revision: http://reviews.llvm.org/D26982

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287686 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Triple] Add Facebook vendor
Shoaib Meenai [Tue, 22 Nov 2016 19:36:26 +0000 (19:36 +0000)]
[Triple] Add Facebook vendor

Add a compiler vendor for Facebook, to enable future vendor-specific
behavior.

Differential Revision: https://reviews.llvm.org/D25136

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287684 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LCG] Add utilities to compute parent and ascestor relationships between
Chandler Carruth [Tue, 22 Nov 2016 19:23:31 +0000 (19:23 +0000)]
[LCG] Add utilities to compute parent and ascestor relationships between
SCCs.

These will be fairly expensive routines to call and might be abused in
real code, but are quite useful when debugging or in asserts and are
reasonable and well formed properties to query.

I've used one of them in an assert that was requested in a code review
here. In subsequent commits I'll start using these routines more
heavily, for example in unittests etc. But this at least gets the
groundwork in place.

Differential Revision: https://reviews.llvm.org/D25506

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287682 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[mips] seb, seh instruction aliases
Simon Dardis [Tue, 22 Nov 2016 19:17:23 +0000 (19:17 +0000)]
[mips] seb, seh instruction aliases

Add the single operand form.

Reviewers: vkalintiris

Differential Revision: https://reviews.llvm.org/D26961

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287681 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdd IntrInaccessibleMemOnly property for intrinsics
Andrew Kaylor [Tue, 22 Nov 2016 19:16:04 +0000 (19:16 +0000)]
Add IntrInaccessibleMemOnly property for intrinsics

Differential Revision: https://reviews.llvm.org/D26485

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287680 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PowerPC] Emit VMX loads/stores for aligned ops to avoid adding swaps on LE
Nemanja Ivanovic [Tue, 22 Nov 2016 19:02:07 +0000 (19:02 +0000)]
[PowerPC] Emit VMX loads/stores for aligned ops to avoid adding swaps on LE

This patch corresponds to review:
https://reviews.llvm.org/D26861

It also fixes PR30730.

Committing on behalf of Lei Huang.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287679 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CostModel][X86] Updated sitofp/uitofp scalar/vector cost tests
Simon Pilgrim [Tue, 22 Nov 2016 18:55:49 +0000 (18:55 +0000)]
[CostModel][X86] Updated sitofp/uitofp scalar/vector cost tests

Better coverage of all legal types + special cases.

Removed old fptoui tests which are all handled in fptoui.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287678 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][SSE] Combine UNPCKL(FHADD,FHADD) -> FHADD for v2f64 shuffles.
Simon Pilgrim [Tue, 22 Nov 2016 17:50:06 +0000 (17:50 +0000)]
[X86][SSE] Combine UNPCKL(FHADD,FHADD) -> FHADD for v2f64 shuffles.

This occurs during UINT_TO_FP v2f64 lowering.

We can easily generalize this to other horizontal ops (FHSUB, PACKSS, PACKUS) as required - we are doing something similar with PACKUS in lowerV2I64VectorShuffle

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287676 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[mips] Add support for unaligned load/store macros.
Vasileios Kalintiris [Tue, 22 Nov 2016 16:43:49 +0000 (16:43 +0000)]
[mips] Add support for unaligned load/store macros.

Add missing unaligned store macros (ush/usw) and fix the exisiting
implementation of the unaligned load macros in order to generate
identical expansions with the GNU assembler.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287646 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoCodeGen: simplify TargetMachine::getSymbol interface. NFC.
Tim Northover [Tue, 22 Nov 2016 16:17:20 +0000 (16:17 +0000)]
CodeGen: simplify TargetMachine::getSymbol interface. NFC.

No-one actually had a mangler handy when calling this function, and
getSymbol itself went most of the way towards getting its own mangler
(with a local TLOF variable) so forcing all callers to supply one was
just extra complication.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287645 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Change lowerBuildVectorToBitOp() to take a BuildVectorSDNode. NFC.
Zvi Rackover [Tue, 22 Nov 2016 15:33:28 +0000 (15:33 +0000)]
[X86] Change lowerBuildVectorToBitOp() to take a BuildVectorSDNode. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287644 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Remove dead code from LowerVectorBroadcast
Zvi Rackover [Tue, 22 Nov 2016 15:17:52 +0000 (15:17 +0000)]
[X86] Remove dead code from LowerVectorBroadcast

Summary: Splat vectors are canonicalized to BUILD_VECTOR's so the code can be simplified. NFC-ish.

Reviewers: craig.topper, delena, RKSimon, andreadb

Subscribers: RKSimon, llvm-commits

Differential Revision: https://reviews.llvm.org/D26678

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287643 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AArch64] Set the max interleave factor for Falkor.
Chad Rosier [Tue, 22 Nov 2016 14:25:02 +0000 (14:25 +0000)]
[AArch64] Set the max interleave factor for Falkor.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287642 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AArch64] Maximize 80-column. NFC.
Chad Rosier [Tue, 22 Nov 2016 14:12:09 +0000 (14:12 +0000)]
[AArch64] Maximize 80-column. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287640 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix line endings
Simon Pilgrim [Tue, 22 Nov 2016 13:27:29 +0000 (13:27 +0000)]
Fix line endings

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287638 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[wasm] hack around test failure after r287553.
Benjamin Kramer [Tue, 22 Nov 2016 13:13:33 +0000 (13:13 +0000)]
[wasm] hack around test failure after r287553.

This test is very brittle as small changes to block layout break the
check patterns. Hack around a change one more time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287637 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SelectionDAG] ComputeNumSignBits of TRUNCATE operations
Simon Pilgrim [Tue, 22 Nov 2016 11:29:19 +0000 (11:29 +0000)]
[SelectionDAG] ComputeNumSignBits of TRUNCATE operations

Add basic ComputeNumSignBits support for TRUNCATE ops for cases where the source's number of sign bits overlaps with the truncated size.

Improves X86 SIGN_EXTEND_IN_REG vector cases which were needlessly sign extending boolean vector results.

Differential Revision: https://reviews.llvm.org/D26851

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287635 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AVX512][inline-asm] Fix AVX512 inline assembly instruction resolution when the size...
Coby Tayree [Tue, 22 Nov 2016 09:30:29 +0000 (09:30 +0000)]
[AVX512][inline-asm] Fix AVX512 inline assembly instruction resolution when the size qualifier of a memory operand is not specified explicitly.

This commit handles cases where the size qualifier of an indirect memory reference operand in Intel syntax is missing (e.g. "vaddps xmm1, xmm2, [a]").

GCC will deduce the size qualifier for AVX512 vector and broadcast memory operands based on the possible matches:
"vaddps xmm1, xmm2, [a]" matches only “XMMWORD PTR” qualifier.
"vaddps xmm1, xmm2, [a]{1to4}" matches only “DWORD PTR” qualifier.

This is different from the current behavior of LLVM, which deduces the size qualifier based on the size of the memory operand.
For "vaddps xmm1, xmm2, [a]"
"char a;" will imply "BYTE PTR" qualifier
"short a;" will imply "WORD PTR" qualifier.

This commit aligns LLVM to GCC’s behavior.

This is the LLVM part of the review.
The Clang part of the review: https://reviews.llvm.org/D26587

Differential Revision: https://reviews.llvm.org/D26586

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287630 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRename option to -lto-pass-remarks-output
Adam Nemet [Tue, 22 Nov 2016 07:35:14 +0000 (07:35 +0000)]
Rename option to -lto-pass-remarks-output

The new option -pass-remarks-output broke LLVM_LINK_LLVM_DYLIB because
of the duplicate option name with opt.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287627 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[TableGen][ISel] When factoring ScopeMatcher, if the child of the ScopeMatcher we...
Craig Topper [Tue, 22 Nov 2016 07:00:06 +0000 (07:00 +0000)]
[TableGen][ISel] When factoring ScopeMatcher, if the child of the ScopeMatcher we're working on is also a ScopeMatcher, merge all its children into the one we're working on.

There were several cases in X86 where we were unable to fully factor a ScopeMatcher but created nested ScopeMatchers for some portions of it. Then we created a SwitchType that split it up and further factored it so that we ended up with something like this:

SwitchType
  Scope
    Scope
      Sequence of matchers
      Some other sequence of matchers
    EndScope
    Another sequence of matchers
  EndScope
...Next type

This change turns it into this:

SwitchType
  Scope
    Sequence of matchers
    Some other sequence of matchers
    Another sequence of matchers
  EndScope
...Next type

Several other in-tree targets had similar nested scopes like this. Overall this doesn't save many bytes, but makes the isel output a little more regular.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287624 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Remove alternate CodeGenOnly version of (v)movq that declared the load size...
Craig Topper [Tue, 22 Nov 2016 05:31:43 +0000 (05:31 +0000)]
[X86] Remove alternate CodeGenOnly version of (v)movq that declared the load size as i128mem. Change all uses to the use the i64mem version.

I'm sure this caused the load size to misprint in Intel syntax output. We were also inconsistent about which patterns used which instruction between VEX and EVEX.

There are two different reg/reg versions of movq, one from a GPR and one from the lower 64-bits of an XMM register. This changes the loading folding table to use the single i64mem memory form for folding both cases. But we need to use TB_NO_REVERSE to prevent a duplicate entry in the unfolding table.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287622 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D...
Craig Topper [Tue, 22 Nov 2016 04:57:34 +0000 (04:57 +0000)]
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD).

Summary:
The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities.

We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too.

Reviewers: igorb, delena, Ayal, Farhana, RKSimon

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D25652

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287621 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoMC: ensure that we have a section before accessing it
Saleem Abdulrasool [Tue, 22 Nov 2016 04:32:54 +0000 (04:32 +0000)]
MC: ensure that we have a section before accessing it

We would attempt to access the symbol section without ensuring that the symbol
was not absolute.  When the assembler referenced relocation is not evaluated to
the absolute, but when we record the relocation, we would query the section.
Because the symbol is absolute, it does not have a section associated with it,
triggering an assertion.  Just be more careful about the access of the section.

Addresses PR31064!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287619 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AVX-512] Add support for changing the element size of PALIGNR/VALIGND/VALIGNQ shuffl...
Craig Topper [Tue, 22 Nov 2016 03:51:53 +0000 (03:51 +0000)]
[AVX-512] Add support for changing the element size of PALIGNR/VALIGND/VALIGNQ shuffles if they feed a vselect with a different type

Summary:
Shuffle lowering widens the element size of a shuffle if elements are contiguous. This is sometimes help because wider element types have more shuffle options. If the shuffle is one of the arguments to a vselect this shuffle widening can introduce a bitcast between the vselect and the shuffle. This will prevent isel from selecting a masked operation. If the shuffle can be written equally efficiently with a different element size to match the vselect type we should change the shuffle type to allow masking.

This patch does this conversion for all VALIGND/VALIGNQ sizes. It also supports turning 128-bit PALIGNR into VALIGND/VALIGNQ. This fixes the case shown in PR31018.

I plan to add support for more operations in future patches.

Reviewers: RKSimon, zvi, delena

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D26902

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287612 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoObject: Make SymbolicFile::symbol_{begin,end}() virtual and remove unnecessary wrappers.
Peter Collingbourne [Tue, 22 Nov 2016 03:38:40 +0000 (03:38 +0000)]
Object: Make SymbolicFile::symbol_{begin,end}() virtual and remove unnecessary wrappers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287611 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ADT] Add initializer list support to SmallPtrSet so that sets can be
Chandler Carruth [Tue, 22 Nov 2016 03:27:43 +0000 (03:27 +0000)]
[ADT] Add initializer list support to SmallPtrSet so that sets can be
easily initialized with some initial values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287610 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AMDGPU] Fix multiple vreg definitions in si-lower-control-flow
Stanislav Mekhanoshin [Tue, 22 Nov 2016 01:42:34 +0000 (01:42 +0000)]
[AMDGPU] Fix multiple vreg definitions in si-lower-control-flow

Differential Revision: https://reviews.llvm.org/D26939

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287608 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAnalysis: gep inbounds (gep inbounds (...)) is inbounds.
Peter Collingbourne [Tue, 22 Nov 2016 01:03:40 +0000 (01:03 +0000)]
Analysis: gep inbounds (gep inbounds (...)) is inbounds.

Differential Revision: https://reviews.llvm.org/D26441

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287604 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRemove LLVM_NODISCARD in one more place.
Zachary Turner [Mon, 21 Nov 2016 23:17:15 +0000 (23:17 +0000)]
Remove LLVM_NODISCARD in one more place.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287596 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRemove LLVM_NODISCARD from two more StringRef members.
Zachary Turner [Mon, 21 Nov 2016 23:02:28 +0000 (23:02 +0000)]
Remove LLVM_NODISCARD from two more StringRef members.

This should be everything.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287594 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoDAG: Ignore call site attributes when emitting target intrinsic
Matt Arsenault [Mon, 21 Nov 2016 22:56:42 +0000 (22:56 +0000)]
DAG: Ignore call site attributes when emitting target intrinsic

A target intrinsic may be defined as possibly reading memory,
but the call site may have additional knowledge that it doesn't read
memory. The intrinsic lowering will expect the pessimistic
assumption of the intrinsic definition, so the chain should
still be used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287593 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AArch64LoadStoreOptimizer] Don't treat write to XZR/WZR as a clobber.
Geoff Berry [Mon, 21 Nov 2016 22:51:10 +0000 (22:51 +0000)]
[AArch64LoadStoreOptimizer] Don't treat write to XZR/WZR as a clobber.

Summary:
When searching for load/store instructions to pair/merge don't treat
writes to WZR/XZR as clobbers since they don't change the value read
from WZR/XZR (which is always 0).

Reviewers: mcrosier, junbuml, jmolloy, t.p.northover

Subscribers: aemerson, llvm-commits, rengolin

Differential Revision: https://reviews.llvm.org/D26921

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287592 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CodeGenPrepare] Don't sink non-cheap addrspacecasts.
Justin Lebar [Mon, 21 Nov 2016 22:49:15 +0000 (22:49 +0000)]
[CodeGenPrepare] Don't sink non-cheap addrspacecasts.

Summary:
Previously, CGP would unconditionally sink addrspacecast instructions,
even going so far as to sink them into a loop.

Now we check that the cast is "cheap", as defined by TLI.

We introduce a new "is-cheap" function to TLI rather than using
isNopAddrSpaceCast because some GPU platforms want the ability to ask
for non-nop casts to be sunk.

Reviewers: arsenm, tra

Subscribers: jholewinski, wdng, llvm-commits

Differential Revision: https://reviews.llvm.org/D26923

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287591 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CodeGenPrepare] Rewrite a loop in terms of llvm::none_of. NFC.
Justin Lebar [Mon, 21 Nov 2016 22:49:11 +0000 (22:49 +0000)]
[CodeGenPrepare] Rewrite a loop in terms of llvm::none_of.  NFC.

Reviewers: arsenm

Subscribers: wdng, llvm-commits

Differential Revision: https://reviews.llvm.org/D26924

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287590 91177308-0d34-0410-b5e6-96231b3b80d8