OSDN Git Service

android-x86/external-llvm.git
7 years agofix formatting; NFC
Hiroshi Inoue [Tue, 11 Jul 2017 15:41:31 +0000 (15:41 +0000)]
fix formatting; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307662 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[globalisel][tablegen] Change method of squashing unused variable warnings following...
Daniel Sanders [Tue, 11 Jul 2017 14:23:14 +0000 (14:23 +0000)]
[globalisel][tablegen] Change method of squashing unused variable warnings following post-commit comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307659 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[SystemZ] Minor fixing in SystemZScheduleZ13.td
Jonas Paulsson [Tue, 11 Jul 2017 14:07:55 +0000 (14:07 +0000)]
[SystemZ]  Minor fixing in SystemZScheduleZ13.td

Some minor corrections for the recently added instructions.

Review: Ulrich Weigand

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307658 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ARM] GlobalISel: Tighten G_FCMP selection test. NFC
Diana Picus [Tue, 11 Jul 2017 12:34:33 +0000 (12:34 +0000)]
[ARM] GlobalISel: Tighten G_FCMP selection test. NFC

Use CHECK-NEXT for the comparison sequence, to make sure we don't get
any unexpected instructions in the middle of our flag manipulation
efforts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307656 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DWARF] - Add testcase for checking message about broken relocations.
George Rimar [Tue, 11 Jul 2017 12:29:07 +0000 (12:29 +0000)]
[DWARF] - Add testcase for checking message about broken relocations.

Addresses comments for r306677, which fixed error message itself.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307655 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][AVX512] regenerate avx512-insert-extract.ll
Guy Blank [Tue, 11 Jul 2017 11:51:49 +0000 (11:51 +0000)]
[X86][AVX512] regenerate avx512-insert-extract.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307654 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ARM] GlobalISel: Add reg mapping for s64 G_FCMP
Diana Picus [Tue, 11 Jul 2017 11:47:45 +0000 (11:47 +0000)]
[ARM] GlobalISel: Add reg mapping for s64 G_FCMP

Map the result into GPR and the operands into FPR.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307653 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PM] Another post-commit fix in NewPMDriver
Philip Pfaffe [Tue, 11 Jul 2017 11:17:44 +0000 (11:17 +0000)]
[PM] Another post-commit fix in NewPMDriver

There were two errors in the parsing of opt's command line options for
extension point pipelines. The EP callbacks are not supposed to return a
value. To check the pipeline text for correctness, I now try to parse it
into a temporary PM object, and print a message on failure. This solves
the compile time error for the lambda return type, as well as correctly
handles unparsable pipelines now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307649 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ARM] GlobalISel: Tighten legalizer tests. NFC
Diana Picus [Tue, 11 Jul 2017 10:52:08 +0000 (10:52 +0000)]
[ARM] GlobalISel: Tighten legalizer tests. NFC

Make sure that all the legalizer tests where the original instruction
needs to be removed check for the removal. We do this by adding
CHECK-NOT lines before and after the replacement sequence. This won't
catch pathological cases where the instruction remains somewhere in the
middle of the instruction sequence that's supposed to replace it, but
hopefully that won't occur in practice (since ideally we'd be setting
the insert point for the new instruction sequence either before or after
the original instruction and not fiddle with it while building the
sequence).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307647 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[globalisel][tablegen] Fix an multi-insn match bug where ComplexPattern is used on...
Daniel Sanders [Tue, 11 Jul 2017 10:40:18 +0000 (10:40 +0000)]
[globalisel][tablegen] Fix an multi-insn match bug where ComplexPattern is used on multiple insns.

In each rule, each use of ComplexPattern is assigned an element in the Renderers
array. The matcher then collects renderer functions in this array and they are
used to render instructions. This works well for a single instruction but a
bug in the allocation mechanism causes the elements to be assigned on a
per-instruction basis rather than a per-rule basis.

So in the case of:
  (set GPR32:$dst, (Op complex:$src1, complex:$src2))
tablegen currently assigns elements 0 and 1 to $src1 and $src2 respectively,
but for:
  (set GPR32:$dst, (Op complex:$src1, (Op complex:$src2)))
it currently assigned both $src1 and $src2 the same element (0). This results in
one complex operand being rendered twice and the other being forgotten.
This patch corrects the allocation such that $src1 and $src2 are still allocated
different elements in this case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307646 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ARM] ldr pc,=expression should be allowed in Thumb2
Peter Smith [Tue, 11 Jul 2017 09:47:12 +0000 (09:47 +0000)]
[ARM] ldr pc,=expression should be allowed in Thumb2

This change allows the pc to be used as a destination register for the
pseudo instruction LDR pc,=expression . The pseudo instruction must not be
transformed into a MOV, but it can use the Thumb2 LDR (literal) instruction
to a constant pool entry. See (A7.7.43 from ARMv7M ARM ARM).

Differential Revision: https://reviews.llvm.org/D34751

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307640 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ARM] GlobalISel: Fix oversight in G_FCMP legalization
Diana Picus [Tue, 11 Jul 2017 09:43:51 +0000 (09:43 +0000)]
[ARM] GlobalISel: Fix oversight in G_FCMP legalization

We used to forget to erase the original instruction when replacing a
G_FCMP true/false. Fix this bug and make sure the tests check for it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307639 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[globalisel][tablegen] Correct matching of intrinsic ID's.
Daniel Sanders [Tue, 11 Jul 2017 08:57:29 +0000 (08:57 +0000)]
[globalisel][tablegen] Correct matching of intrinsic ID's.

TreePatternNode considers them to be plain integers but MachineInstr considers
them to be a distinct kind of operand.

The tweak to AArch64InstrInfo.td to produce a simple test case is a NFC for
everything except GlobalISelEmitter (confirmed by diffing the tablegenerated
files). GlobalISelEmitter is currently unable to infer the type of operands in
the Dst pattern from the operands in the Src pattern.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307634 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ARM] GlobalISel: Legalize s64 G_FCMP
Diana Picus [Tue, 11 Jul 2017 08:50:01 +0000 (08:50 +0000)]
[ARM] GlobalISel: Legalize s64 G_FCMP

Same as the s32 version, for both hard and soft float.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307633 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert Revert [MBP] do not rotate loop if it creates extra branch
Serguei Katkov [Tue, 11 Jul 2017 08:34:58 +0000 (08:34 +0000)]
Revert Revert [MBP] do not rotate loop if it creates extra branch

This is a second attempt to land this patch.

The first one resulted in a crash of clang sanitizer buildbot.
The fix is here and regression test is added.

This is a last fix for the corner case of PR32214. Actually this is not really corner case in general.

We should not do a loop rotation if we create an additional branch due to it.
Consider the case where we have a loop chain H, M, B, C , where
H is header with viable fallthrough from pre-header and exit from the loop
M - some middle block
B - backedge to Header but with exit from the loop also.
C - some cold block of the loop.

Let's H is determined as a best exit. If we do a loop rotation M, B, C, H we can introduce the extra branch.
Let's compute the change in number of branches:
+1 branch from pre-header to header
-1 branch from header to exit
+1 branch from header to middle block if there is such
-1 branch from cold bock to header if there is one

So if C is not a predecessor of H then we introduce extra branch.

This change actually prohibits rotation of the loop if both true
  Best Exit has next element in chain as successor.
  Last element in chain is not a predecessor of first element of chain.

Reviewers: iteratee, xur, sammccall, chandlerc
Reviewed By: iteratee
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D34745

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307631 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[GlobalISel][X86] Use correct AND instructions.
Igor Breger [Tue, 11 Jul 2017 08:04:51 +0000 (08:04 +0000)]
[GlobalISel][X86] Use correct AND instructions.

AND8ri8 not supported in 64bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307630 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CGP] Relax a bit restriction for optimizeMemoryInst to extend scope
Serguei Katkov [Tue, 11 Jul 2017 06:24:44 +0000 (06:24 +0000)]
[CGP] Relax a bit restriction for optimizeMemoryInst to extend scope

CodeGenPrepare::optimizeMemoryInst contains a check that we do nothing
if all instructions combining the address for memory instruction is in the same
block as memory instruction itself.

However if any of these instruction are placed after memory instruction then
address calculation will not be folded to memory instruction.

The added test case shows an example.

Reviewers: loladiro, spatel, efriedma
Reviewed By: efriedma
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D34862

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307628 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agofix typos in comments; NFC
Hiroshi Inoue [Tue, 11 Jul 2017 06:04:59 +0000 (06:04 +0000)]
fix typos in comments; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307626 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PM/ThinLTO] Fix PR33536, a bug where the ThinLTO bitcode writer was
Chandler Carruth [Tue, 11 Jul 2017 05:39:20 +0000 (05:39 +0000)]
[PM/ThinLTO] Fix PR33536, a bug where the ThinLTO bitcode writer was
querying for analysis results on a function declaration rather than
a definition.

The only reason this worked previously is by chance -- because the way
we got alias analysis results with the legacy PM, we happened to not
compute a dominator tree and so we happened to not hit an assert even
though it didn't make any real sense. Now we bail out before trying to
compute alias analysis so that we don't hit these asserts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307625 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PowerPC] fix latency for simple integer instructions in POWER9 scheduler
Hiroshi Inoue [Tue, 11 Jul 2017 05:37:16 +0000 (05:37 +0000)]
[PowerPC] fix latency for simple integer instructions in POWER9 scheduler

In the POWER9 instruction scheduler, SchedWriteRes for the simple integer instructions are misconfigured to use that of (costly) DFU instructions.
This results in surprisingly long instruction latency estimation and causes misbehavior in some optimizers such as if-conversion.

Differential Revision: https://reviews.llvm.org/D34869

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307624 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PowerPC] avoid redundant analysis while lowering an immediate; NFC
Hiroshi Inoue [Tue, 11 Jul 2017 05:28:26 +0000 (05:28 +0000)]
[PowerPC] avoid redundant analysis while lowering an immediate; NFC

This patch reduces compilation time by avoiding redundant analysis while selecting instructions to create an immediate.
If the instruction count required to create the input number without rotate is 2, we do not need further analysis to find a shorter instruction sequence with rotate; rotate + load constant cannot be done by 1 instruction (i.e. getInt64CountDirectnever return 0).
This patch should not change functionality.

Differential Revision: https://reviews.llvm.org/D34986

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307623 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AVR] Remove a few very old TODOs that don't have enough context to understand
Dylan McKay [Tue, 11 Jul 2017 05:14:40 +0000 (05:14 +0000)]
[AVR] Remove a few very old TODOs that don't have enough context to understand

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307622 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[InstCombine] Add test case for PR33721.
Craig Topper [Tue, 11 Jul 2017 05:12:52 +0000 (05:12 +0000)]
[InstCombine] Add test case for PR33721.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307621 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AVR] Rename 'ZREGS' to 'ZREG'
Dylan McKay [Tue, 11 Jul 2017 04:53:43 +0000 (04:53 +0000)]
[AVR] Rename 'ZREGS' to 'ZREG'

It will only ever contain one register.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307620 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AVR] Rename 'AVRTiny' to 'Tiny'
Dylan McKay [Tue, 11 Jul 2017 04:45:15 +0000 (04:45 +0000)]
[AVR] Rename 'AVRTiny' to 'Tiny'

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307619 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AVR] Use the generic branch relaxer
Dylan McKay [Tue, 11 Jul 2017 04:17:13 +0000 (04:17 +0000)]
[AVR] Use the generic branch relaxer

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307617 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoWhitespace.
NAKAMURA Takumi [Tue, 11 Jul 2017 02:31:54 +0000 (02:31 +0000)]
Whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307614 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert r307581, "Avoid doing conservative phi checks in aliasSameBasePointerGEPs...
NAKAMURA Takumi [Tue, 11 Jul 2017 02:31:51 +0000 (02:31 +0000)]
Revert r307581, "Avoid doing conservative phi checks in aliasSameBasePointerGEPs() if no phis have been visited yet."

It broke stage2 tests in selfhosting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307613 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[WebAssembly] Fix use of cast vs dyn_cast
Sam Clegg [Tue, 11 Jul 2017 02:21:57 +0000 (02:21 +0000)]
[WebAssembly] Fix use of cast vs dyn_cast

Differential Revision: https://reviews.llvm.org/D35233

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307612 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[llvm-cov] Disable threading in a test. NFC.
Vedant Kumar [Tue, 11 Jul 2017 01:42:12 +0000 (01:42 +0000)]
[llvm-cov] Disable threading in a test. NFC.

PR30735 reports an issue where llvm-cov hangs with a worker thread
waiting on a condition, and the main thread waiting to join() the
workers. While this doesn't appear to be a bug in llvm-cov or the
ThreadPool implementation, it would be helpful to disable the use of
threading in the llvm-cov tests where no test coverage is added.

More context: https://bugs.llvm.org/show_bug.cgi?id=30735

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307610 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[llvm-cov] Add a cl::opt to control the number of threads
Vedant Kumar [Tue, 11 Jul 2017 01:23:29 +0000 (01:23 +0000)]
[llvm-cov] Add a cl::opt to control the number of threads

When an output directory is specified, llvm-cov spawns some threads to
speed up the process of writing out file reports. Add an option which
allows users to control how many threads llvm-cov uses.

A CommandGuide.rst update + test is included.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307609 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agollvm-profdata: Improve memory usage by tuning SmallDenseMap size
David Blaikie [Tue, 11 Jul 2017 01:18:28 +0000 (01:18 +0000)]
llvm-profdata: Improve memory usage by tuning SmallDenseMap size

This takes memory usage from 5.1GB to 970MB - it could go further by
using a small size of 2 instead of the default of 4, but given the
rather high cost of going over this limit by much, I figured a little
slosh would be worth the ~130MB of memory usage.

& this'll might not be such a big deal if we use a custom slab allocator
for the DenseMaps here anyway

While the vast majority (99.9%) of records use only 1 entry, the tuning
parameter to SmallDenseMap is the the number of buckets, not the number
of entries - so a small size of 1 wasn't useful, even for 1 element, it
would tip over into allocating (much, 64 slots worth) more space - none
of them ended up small.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307608 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[cmake] Check for Haiku when setting LIB_NAMES for GNU ld
Andrew Wilkins [Tue, 11 Jul 2017 01:17:44 +0000 (01:17 +0000)]
[cmake] Check for Haiku when setting LIB_NAMES for GNU ld

Haiku uses GNU ld for linking, but is not captured in the
conditional when setting LIB_NAMES. This causes a shared
library with no symbols on Haiku. This patch simply adds
a check for whether the CMake system name is Haiku in
addition to the existing checks.

Patch by Jérôme Duval.

Differential Revision: https://reviews.llvm.org/D34998

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307607 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRemove circular dependency from runtimes/CMakeLists
George Karpenkov [Tue, 11 Jul 2017 00:48:00 +0000 (00:48 +0000)]
Remove circular dependency from runtimes/CMakeLists

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307605 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[lld/pdb] Create an empty public symbol record stream.
Zachary Turner [Mon, 10 Jul 2017 22:40:20 +0000 (22:40 +0000)]
[lld/pdb] Create an empty public symbol record stream.

This is part of the continuing effort to increase parity between
LLD and MSVC PDBs.  link still doesn't like our PDBs, so the most
obvious thing to check was whether adding an empty publics stream
would get it to do something else.  It still fails in the same way
but at least this removes one more variable from the equation.
The next logical step would be to try creating an empty globals
stream.

Differential Revision: https://reviews.llvm.org/D35224

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307598 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoDoxygen formatting. NFCI
Joel Jones [Mon, 10 Jul 2017 22:11:50 +0000 (22:11 +0000)]
Doxygen formatting. NFCI

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307597 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoInstrProf: Fix unit test which accidentally used a duplicate name
Vedant Kumar [Mon, 10 Jul 2017 21:44:43 +0000 (21:44 +0000)]
InstrProf: Fix unit test which accidentally used a duplicate name

This unit test constructed some profile records incorrectly. One of the
records had a duplicate name: adding that record into the writer caused
an error unrelated to what needed to be tested.

Reported by David Blaikie!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307596 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[lld/pdb] Add some basic linker module symbols.
Zachary Turner [Mon, 10 Jul 2017 21:01:37 +0000 (21:01 +0000)]
[lld/pdb] Add some basic linker module symbols.

Differential Revision: https://reviews.llvm.org/D35152

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307590 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert "[DAG] Improve Aliasing of operations to static alloca"
Matthias Braun [Mon, 10 Jul 2017 20:51:30 +0000 (20:51 +0000)]
Revert "[DAG] Improve Aliasing of operations to static alloca"

Reverting as it breaks tramp3d-v4 in the llvm test-suite. I added some
comments to https://reviews.llvm.org/D33345 about it.

This reverts commit r307546.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307589 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[WebAssembly] Add more details to llvm-readobj for wasm files
Sam Clegg [Mon, 10 Jul 2017 20:47:12 +0000 (20:47 +0000)]
[WebAssembly] Add more details to llvm-readobj for wasm files

Differential Revision: https://reviews.llvm.org/D35208

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307588 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ConstantHoisting] Remove dupliate logic in constant hoisting
Leo Li [Mon, 10 Jul 2017 20:45:34 +0000 (20:45 +0000)]
[ConstantHoisting] Remove dupliate logic in constant hoisting

Summary:
As metioned in https://reviews.llvm.org/D34576, checkings in
`collectConstantCandidates` can be replaced by using
`llvm::canReplaceOperandWithVariable`.

The only special case is that `collectConstantCandidates` return false for
all `IntrinsicInst` but it is safe for us to collect constant candidates from
`IntrinsicInst`.

Reviewers: pirama, efriedma, srhines

Reviewed By: efriedma

Subscribers: llvm-commits, javed.absar

Differential Revision: https://reviews.llvm.org/D34921

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307587 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[NewGVN] Simplify a lambda a little bit. NFCI.
Davide Italiano [Mon, 10 Jul 2017 20:45:00 +0000 (20:45 +0000)]
[NewGVN] Simplify a lambda a little bit. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307586 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[WebAssembly] Be consistent in generating trivial test input files
Sam Clegg [Mon, 10 Jul 2017 20:43:26 +0000 (20:43 +0000)]
[WebAssembly] Be consistent in generating trivial test input files

For each checked-in wasm file, make sure the there is
corresponding .ll file that can be used to regenerate it
if needed.

Add test/Object/Inputs/trivial-object-test.wasm to match other
formats and add some new wasm tests in test/Object.

Differential Revision: https://reviews.llvm.org/D35213

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307585 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAdd DAG argument to canMergeStoresTo NFC.
Nirav Dave [Mon, 10 Jul 2017 20:25:54 +0000 (20:25 +0000)]
Add DAG argument to canMergeStoresTo NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307583 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Hexagon] Convert typed ISD opcodes to generic ones, NFC
Krzysztof Parzyszek [Mon, 10 Jul 2017 20:16:44 +0000 (20:16 +0000)]
[Hexagon] Convert typed ISD opcodes to generic ones, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307582 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAvoid doing conservative phi checks in aliasSameBasePointerGEPs() if no phis have...
Farhana Aleen [Mon, 10 Jul 2017 20:15:40 +0000 (20:15 +0000)]
Avoid doing conservative phi checks in aliasSameBasePointerGEPs() if no phis have been visited yet.

Reviewers: Daniel Berlin

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D34478

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307581 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Hexagon] Remove unused ISD opcodes, NFC
Krzysztof Parzyszek [Mon, 10 Jul 2017 20:13:44 +0000 (20:13 +0000)]
[Hexagon] Remove unused ISD opcodes, NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307580 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoUse DenseMap instead std::map for GVSummaryMapTy.
Dehao Chen [Mon, 10 Jul 2017 20:12:54 +0000 (20:12 +0000)]
Use DenseMap instead std::map for GVSummaryMapTy.

Summary: This speeds-up thin-link by ~47% for large programs.

Reviewers: tejohnson

Reviewed By: tejohnson

Subscribers: sanjoy, llvm-commits, mehdi_amini

Differential Revision: https://reviews.llvm.org/D35148

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307578 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU: Allow SIShrinkInstructions to fold FrameIndexes
Matt Arsenault [Mon, 10 Jul 2017 20:04:35 +0000 (20:04 +0000)]
AMDGPU: Allow SIShrinkInstructions to fold FrameIndexes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307576 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU: Allow SIShrinkInstructions to work in non-SSA
Matt Arsenault [Mon, 10 Jul 2017 19:53:57 +0000 (19:53 +0000)]
AMDGPU: Allow SIShrinkInstructions to work in non-SSA

Immediates can be folded as long as the immediate is a vreg.

Also undo commuting instructions if it didn't fold an immediate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307575 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU: Remove unnecessary check for constant operands
Matt Arsenault [Mon, 10 Jul 2017 19:33:38 +0000 (19:33 +0000)]
AMDGPU: Remove unnecessary check for constant operands

An instruction that has an immediate operand can't reach
this point. This is only called for a freshly shrunk instruction,
which prevously couldn't have had a literal constant operand.
This was also not conservative enough since it woudl also have
had to filter other constant-like inputs like frame indexes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307574 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoAMDGPU: Do not test for SI in getIsaVersion
Konstantin Zhuravlyov [Mon, 10 Jul 2017 19:24:05 +0000 (19:24 +0000)]
AMDGPU: Do not test for SI in getIsaVersion

SI is being tested by isa version in the first two if statements of the function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307573 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[CMake] Dependencies for tests in "runtimes"
George Karpenkov [Mon, 10 Jul 2017 19:23:28 +0000 (19:23 +0000)]
[CMake] Dependencies for tests in "runtimes"

Many of the test cases in the runtimes require LLVM's testing tools, to
facilitate this working as expected we need to have all the test targets
in the runtimes depend on all LLVM testing tools used in the runtimes.

Differential Revision: https://reviews.llvm.org/D33048

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307572 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoResubmit "Add pdb-diff test."
Zachary Turner [Mon, 10 Jul 2017 19:16:49 +0000 (19:16 +0000)]
Resubmit "Add pdb-diff test."

This was originally reverted because of two issues.
  1) Printing ANSI color escape codes even when outputting to
     a file
  2) Module name comparisons were failing when comparing a PDB
     generated on one machine to a PDB generated on another
     machine.

I attempted to fix #2 by adding command line options which let
you specify prefixes to strip from the beginning of embedded
paths, which effectively lets us specify a path to "base" each
PDB from and only compare the parts under the base.  But this is
tricky because PDB paths always use Windows path syntax, even
when they are created on non-Windows hosts.  A problem still
existed when constructing the prefix to strip, where we were
accidentally using a host-specific path separator instead of
a Windows path separator.

This resubmission fixes the issue on Linux (and I have verified
that the test now passes on Linux).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307571 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[cmake] Remove obsolete unset in CMake.
George Karpenkov [Mon, 10 Jul 2017 18:58:05 +0000 (18:58 +0000)]
[cmake] Remove obsolete unset in CMake.

Differential Revision: https://reviews.llvm.org/D35024

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307569 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[docs] Remove obsolete section in CMake documentation.
George Karpenkov [Mon, 10 Jul 2017 18:56:51 +0000 (18:56 +0000)]
[docs] Remove obsolete section in CMake documentation.

Differential Revision: https://reviews.llvm.org/D35023

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307568 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Hexagon] Fix check for HMOTF_ConstExtend operand flag
Krzysztof Parzyszek [Mon, 10 Jul 2017 18:38:52 +0000 (18:38 +0000)]
[Hexagon] Fix check for HMOTF_ConstExtend operand flag

This fixes https://llvm.org/PR33718.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307566 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[WebAssembly] Use the correct size for MCFillFragment
Sam Clegg [Mon, 10 Jul 2017 18:36:34 +0000 (18:36 +0000)]
[WebAssembly] Use the correct size for MCFillFragment

Summary: When implementing MCFillFragment, use the size of the fragment,
rather than the size of the section.

Patch by Dan Gohman

Differential Revision: https://reviews.llvm.org/D35090

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307565 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Hexagon] Handle Hexagon-specific machine operand target flags in MIR
Krzysztof Parzyszek [Mon, 10 Jul 2017 18:31:02 +0000 (18:31 +0000)]
[Hexagon] Handle Hexagon-specific machine operand target flags in MIR

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307564 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PPC CodeGen] Expand the bitreverse.i64 intrinsic.
Tony Jiang [Mon, 10 Jul 2017 18:11:23 +0000 (18:11 +0000)]
[PPC CodeGen] Expand the bitreverse.i64 intrinsic.

Differential Revision: https://reviews.llvm.org/D34908
Fix PR: https://bugs.llvm.org/show_bug.cgi?id=33093

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307563 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[Docs] Updating CMake docs to include LLVM_REVERSE_ITERATION
Mandeep Singh Grang [Mon, 10 Jul 2017 17:37:36 +0000 (17:37 +0000)]
[Docs] Updating CMake docs to include LLVM_REVERSE_ITERATION

Reviewers: beanz, mehdi_amini

Reviewed By: beanz

Subscribers: mgorny, llvm-commits

Differential Revision: https://reviews.llvm.org/D35080

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307560 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoRevert "Build fixes for pdb-diff test."
Zachary Turner [Mon, 10 Jul 2017 17:32:47 +0000 (17:32 +0000)]
Revert "Build fixes for pdb-diff test."

This reverts commit 180af3fdbdb17ec35b45ec1f925fd743b28d37e1.

This is still breaking due to linux-specific path differences.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307559 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoBuild fixes for pdb-diff test.
Zachary Turner [Mon, 10 Jul 2017 17:01:28 +0000 (17:01 +0000)]
Build fixes for pdb-diff test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307556 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix pdb-diff test.
Zachary Turner [Mon, 10 Jul 2017 16:52:15 +0000 (16:52 +0000)]
Fix pdb-diff test.

A test was checked in on Friday that worked by checking in an
object file and PDB generated locally by MSVC, and then having
the test run lld-link on the object file and diffing LLD's PDB
against the checked in PDB.

This failed because part of the diffing algorithm involves
determining if two modules are the same, and if so drilling into
the module and diffing individual fields of the module.  The
only thing we can use to make this determination though is the
"name" of the module, which is a path to where the module (obj
file) was read from on the machine where it was linked.  This
fails for obvious reasons when comparing a PDB generated on one
machine to a PDB on another machine.

The fix employed here is to add two command line options to the
diff subcommand, which allow the user to specify a "binary root
path".  The bin root path, if specified, is stripped from the
beginning of any embedded PDB paths.  The test is updated to
specify the user's local test output directory for the left
PDB, and is hardcoded to the location where the original PDB
was created for the right PDB.  This way all the equivalence
comparisons should succeed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307555 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoFix invalid cast in instcombine UMul/ZExt idiom
Serge Guelton [Mon, 10 Jul 2017 16:51:40 +0000 (16:51 +0000)]
Fix invalid cast in instcombine UMul/ZExt idiom

Fixes https://bugs.llvm.org/show_bug.cgi?id=25454

Do not assume IRBuilder creates Instruction where it can create Value.
Do not assume idiom operands are constant, leave generalisation ot the IRBuilder.

Differential Revision: https://reviews.llvm.org/D35114

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307554 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PowerPC] Reduce register pressure by not materializing a constant just for use as...
Lei Huang [Mon, 10 Jul 2017 16:44:45 +0000 (16:44 +0000)]
[PowerPC] Reduce register pressure by not materializing a constant just for use as an index register for X-Form loads/stores.

For this example:
float test (int *arr) {
    return arr[2];
}

We currently generate the following code:
  li r4, 8
  lxsiwax f0, r3, r4
  xscvsxdsp f1, f0

With this patch, we will now generate:
  addi r3, r3, 8
  lxsiwax f0, 0, r3
  xscvsxdsp f1, f0

Originally reported in: https://bugs.llvm.org/show_bug.cgi?id=27204
Differential Revision: https://reviews.llvm.org/D35027

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307553 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Model 256-bit AVX instructions in the AMD Jaguar scheduler Part-1 (PR28573).
Andrew V. Tischenko [Mon, 10 Jul 2017 16:36:03 +0000 (16:36 +0000)]
[X86] Model 256-bit AVX instructions in the AMD Jaguar scheduler Part-1 (PR28573).
The new version of the model is definitely faster.

Differential Revision:
https://reviews.llvm.org/D35198

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307552 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LLVM] Get rid of white spaces in file names in a DebugInfo test. NFC
Eric Liu [Mon, 10 Jul 2017 16:05:18 +0000 (16:05 +0000)]
[LLVM] Get rid of white spaces in file names in a DebugInfo test. NFC

Summary: White spaces in file names are causing Phabricator/SVN to crash.

Reviewers: bkramer

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D35206

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307550 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DAG] Improve Aliasing of operations to static alloca
Nirav Dave [Mon, 10 Jul 2017 15:39:41 +0000 (15:39 +0000)]
[DAG] Improve Aliasing of operations to static alloca

Memory accesses offset from frame indices may alias, e.g., we
may merge write from function arguments passed on the stack when they
are contiguous. As a result, when checking aliasing, we consider the
underlying frame index's offset from the stack pointer.

Static allocs are realized as stack objects in SelectionDAG, but its
offset is not set until post-DAG causing DAGCombiner's alias check to
consider access to static allocas to frequently alias. Modify isAlias
to consider access between static allocas and access from other frame
objects to be considered aliasing.

Many test changes are included here. Most are fixes for tests which
indirectly relied on our aliasing ability and needed to be modified to
preserve their original intent.

The remaining tests have minor improvements due to relaxed
ordering. The exception is CodeGen/X86/2011-10-19-widen_vselect.ll
which has a minor degradation dispite though the pre-legalized DAG is
improved.

Reviewers: rnk, mkuper, jonpa, hfinkel, uweigand

Reviewed By: rnk

Subscribers: sdardis, nemanjai, javed.absar, llvm-commits

Differential Revision: https://reviews.llvm.org/D33345

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307546 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoUse emplace_back to replace size() and resize().
Dehao Chen [Mon, 10 Jul 2017 15:31:53 +0000 (15:31 +0000)]
Use emplace_back to replace size() and resize().

Summary: This speeds-up thin-link for ~29% for large programs.

Reviewers: tejohnson

Reviewed By: tejohnson

Subscribers: grandinj, sanjoy, llvm-commits

Differential Revision: https://reviews.llvm.org/D35145

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307543 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[LoopUnrollRuntime] Remove strict assert about VMap requirement
Anna Thomas [Mon, 10 Jul 2017 15:29:38 +0000 (15:29 +0000)]
[LoopUnrollRuntime] Remove strict assert about VMap requirement

When unrolling under multiple exits which is under off-by-default option,
the assert that checks for VMap entry in loop exit values is too strong.
(assert if VMap entry did not exist, the value should be a
constant). However, values derived from
constants or from values outside loop, does not have a VMap entry too.

Removed the assert and added a testcase showcasing the property for
non-constant values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307542 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PM] Fix a warning.
Philip Pfaffe [Mon, 10 Jul 2017 13:54:23 +0000 (13:54 +0000)]
[PM] Fix a warning.

The DebugLogging argument was unused in the EP callbacks registration.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307536 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PM] Fix r307532: Get rid of a dangling reference.
Philip Pfaffe [Mon, 10 Jul 2017 12:48:51 +0000 (12:48 +0000)]
[PM] Fix r307532: Get rid of a dangling reference.

Escaping lambda by-reference capture of local variable caused a dangling
reference.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307534 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agofix typos in comments and error messages; NFC
Hiroshi Inoue [Mon, 10 Jul 2017 12:44:25 +0000 (12:44 +0000)]
fix typos in comments and error messages; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307533 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PM] Enable registration of out-of-tree passes with PassBuilder
Philip Pfaffe [Mon, 10 Jul 2017 10:57:55 +0000 (10:57 +0000)]
[PM] Enable registration of out-of-tree passes with PassBuilder

Summary:
This patch adds a callback registration API to the PassBuilder,
enabling registering out-of-tree passes with it.

Through the Callback API, callers may register callbacks with the
various stages at which passes are added into pass managers, including
parsing of a pass pipeline as well as at extension points within the
default -O pipelines.

Registering utilities like `require<>` and `invalidate<>` needs to be
handled manually by the caller, but a helper is provided.

Additionally, adding passes at pipeline extension points is exposed
through the opt tool. This patch adds a `-passes-ep-X` commandline
option for every extension point X, which opt parses into pipelines
inserted into that extension point.

Reviewers: chandlerc

Reviewed By: chandlerc

Subscribers: lksbhm, grosser, davide, mehdi_amini, llvm-commits, mgorny

Differential Revision: https://reviews.llvm.org/D33464

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307532 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ARM] Tidy up ARMBaseRegisterInfo implementation. NFC
Javed Absar [Mon, 10 Jul 2017 10:42:55 +0000 (10:42 +0000)]
[ARM] Tidy up ARMBaseRegisterInfo implementation. NFC

Clean up ARMBaseRegisterInfo implementation a bit.
Differential Revision: https://reviews.llvm.org/D35116

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307531 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DWARF] - Rename variable. NFC.
George Rimar [Mon, 10 Jul 2017 10:04:51 +0000 (10:04 +0000)]
[DWARF] - Rename variable. NFC.

Variable was called 'Name' and contained text
name of relocation type. Problem was that
outside of this error handling scope we already
have different 'Name' variable that contains
section name.

Change helps to avoid confusion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307530 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoThis patch completely replaces the scheduling information for the SandyBridge archite...
Gadi Haber [Mon, 10 Jul 2017 09:53:16 +0000 (09:53 +0000)]
This patch completely replaces the scheduling information for the SandyBridge architecture target by modifying the file X86SchedSandyBridge.td located under the X86 Target.
The SandyBridge architects have provided us with a more accurate information about each instruction latency, number of uOPs and used ports and I used it to replace the existing estimated SNB instructions scheduling and to add missing scheduling information.

Please note that the patch extensively affects the X86 MC instr scheduling for SNB.

Also note that this patch will be followed by additional patches for the remaining target architectures HSW, IVB, BDW, SKL and SKX.

The updated and extended information about each instruction includes the following details:
•static latency of the instruction
•number of uOps from which the instruction consists of
•all ports used by the instruction's' uOPs

For example, the following code dictates that instructions, ADC64mr, ADC8mr, SBB64mr, SBB8mr have a static latency of 9 cycles. Each of these instructions is decoded into 6 micro operations which use ports 4, ports 2 or 3 and port 0 and ports 0 or 1 or 5:

def SBWriteResGroup94 : SchedWriteRes<[SBPort4,SBPort23,SBPort0,SBPort015]> {
let Latency = 9;
let NumMicroOps = 6;
let ResourceCycles = [1,2,2,1];

}
def: InstRW<[SBWriteResGroup94], (instregex "ADC64mr")>;
def: InstRW<[SBWriteResGroup94], (instregex "ADC8mr")>;
def: InstRW<[SBWriteResGroup94], (instregex "SBB64mr")>;
def: InstRW<[SBWriteResGroup94], (instregex "SBB8mr")>;

Note that apart for the header, most of the X86SchedSandyBridge.td file was generated by a script.

Reviewers: zvi, chandlerc, RKSimon, m_zuckerman, craig.topper, igorb

Differential Revision:  https://reviews.llvm.org/D35019#inline-304691

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307529 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[DWARF] - Remove unused variables. NFC.
George Rimar [Mon, 10 Jul 2017 09:36:44 +0000 (09:36 +0000)]
[DWARF] - Remove unused variables. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307528 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[GlobalISel][X86] Support G_LOAD/G_STORE i1.
Igor Breger [Mon, 10 Jul 2017 09:26:09 +0000 (09:26 +0000)]
[GlobalISel][X86] Support G_LOAD/G_STORE i1.

Summary: Support G_LOAD/G_STORE i1.

Reviewers: zvi, guyblank

Reviewed By: guyblank

Subscribers: rovka, kristof.beyls, llvm-commits

Differential Revision: https://reviews.llvm.org/D35178

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307527 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[GlobalISel][X86] extend G_ZEXT support.
Igor Breger [Mon, 10 Jul 2017 09:07:34 +0000 (09:07 +0000)]
[GlobalISel][X86] extend G_ZEXT support.

Summary:
Mark G_ZEXT/G_SEXT i1 to i8/i16,  i8 to i16 as legal.
Support G_ZEXT i1 to i8/i16 instruction selection ( C++ code).
This patch requred to support G_LOAD/G_STORE i1.

Reviewers: zvi, guyblank

Reviewed By: guyblank

Subscribers: rovka, llvm-commits, kristof.beyls

Differential Revision: https://reviews.llvm.org/D35177

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307526 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[docs] NFC: Fix links in the tutorial
Kirill Bobyrev [Mon, 10 Jul 2017 09:07:23 +0000 (09:07 +0000)]
[docs] NFC: Fix links in the tutorial

r274441 introduced Chapter 10 of "Implementing a Language with LLVM" tutorial,
which caused all files in the tutorial to start using two digit numbering. But
many links were not changed and therefore appear to be broken. This patch
addresses described issue.

As a result, following command does not produce any output anymore:

$ grep -nR '<LangImpl[0-9].html>' ./docs/tutorial/

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307525 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agofix formatting; NFC
Hiroshi Inoue [Mon, 10 Jul 2017 06:32:52 +0000 (06:32 +0000)]
fix formatting; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307523 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Fix typo in comment. NFC
Craig Topper [Mon, 10 Jul 2017 06:09:22 +0000 (06:09 +0000)]
[X86] Fix typo in comment. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307522 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ArgumentPromotion] Change use of removed argument in llvm.dbg.value to undef
Mikael Holmen [Mon, 10 Jul 2017 06:07:24 +0000 (06:07 +0000)]
[ArgumentPromotion] Change use of removed argument in llvm.dbg.value to undef

Summary:
This solves PR33641.

When removing a dead argument we must also handle possibly existing calls
to llvm.dbg.value that use the removed argument. Now we change the use
of the otherwise dead argument to an undef for some other pass to cleanup
later.

If the calls are left untouched, they will later on cause errors:
 "function-local metadata used in wrong function"
since the ArgumentPromotion rewrites the code by creating a new function
with the wanted signature, but the metadata is not recreated so the new
function may then erroneously use metadata from the old function.

Reviewers: mstorsjo, rnk, arsenm

Reviewed By: rnk

Subscribers: wdng, llvm-commits

Differential Revision: https://reviews.llvm.org/D34874

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307521 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Remove asserts from getX86CpuIDAndInfo/getX86CpuIDAndInfoEx. Restore past behav...
Craig Topper [Mon, 10 Jul 2017 06:04:11 +0000 (06:04 +0000)]
[X86] Remove asserts from getX86CpuIDAndInfo/getX86CpuIDAndInfoEx. Restore past behavior of returning an unsupported indication to the caller instead.

These asserts could only occur if we fail to properly detect the compiler, but an assert is not a good way to do that because it doesn't work in release builds.

I wonder if we could use #error?

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307520 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ADT] Fix another "oops" spotted by eddyb and reported in IRC.
Chandler Carruth [Mon, 10 Jul 2017 05:41:14 +0000 (05:41 +0000)]
[ADT] Fix another "oops" spotted by eddyb and reported in IRC.

This test pretty clearly should be calling 'maxnum' here. =]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307519 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agollvm-profdata: Reduce memory usage by using Error callback rather than member
David Blaikie [Mon, 10 Jul 2017 03:04:59 +0000 (03:04 +0000)]
llvm-profdata: Reduce memory usage by using Error callback rather than member

Reduces llvm-profdata memory usage on a large profile from 7.8GB to 5.1GB.

The ProfData API now supports reporting all the errors/warnings rather
than only the first, though llvm-profdata ignores everything after the
first for now to preserve existing behavior. (if there's a desire for
other behavior, happy to implement that - but might be as well left for
a separate patch)

Reviewers: davidxl

Differential Revision: https://reviews.llvm.org/D35149

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307516 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoCGSCCPassManagerTest.cpp: Fix warnings. [-Wunused-variable]
NAKAMURA Takumi [Sun, 9 Jul 2017 23:06:05 +0000 (23:06 +0000)]
CGSCCPassManagerTest.cpp: Fix warnings. [-Wunused-variable]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307511 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Relax an assertion when legalizing vector types.
Davide Italiano [Sun, 9 Jul 2017 19:22:48 +0000 (19:22 +0000)]
[X86] Relax an assertion when legalizing vector types.

WidenVSELECTAndMask can fold (and it folds in this case) so we
get a BUILD_VECTOR of constants as mask. convertMask() seems to
work fine when the input is a vector of constants, and we still
need to call it to extend/add elements at the end. but the current
code just asserts on anything but a SETCC or AND/OR/XOR of 2xSETCC.
This change was discussed briefly with Simon Pilgrim, who also
suggests we might consider dropping this assertion in the future.

Fixes PR33715.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307508 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Allow GHC calling convention to use YMM and ZMM registers
Simon Pilgrim [Sun, 9 Jul 2017 16:57:10 +0000 (16:57 +0000)]
[X86] Allow GHC calling convention to use YMM and ZMM registers

GHC 8.4 will know how to use YMM and ZMM registers for calls.

Submitted on behalf of @bgamari (Ben Gamari)

Differential Revision: https://reviews.llvm.org/D34854

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307504 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[AVR] Fix test errors due to tied operands not matching
Dylan McKay [Sun, 9 Jul 2017 16:36:35 +0000 (16:36 +0000)]
[AVR] Fix test errors due to tied operands not matching

Broken due to r307259.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307503 91177308-0d34-0410-b5e6-96231b3b80d8

7 years agoHandle ConstantExpr correctly in SelectionDAGBuilder
Simon Pilgrim [Sun, 9 Jul 2017 16:01:04 +0000 (16:01 +0000)]
Handle ConstantExpr correctly in SelectionDAGBuilder

This change fixes a bug in SelectionDAGBuilder::visitInsertValue and SelectionDAGBuilder::visitExtractValue where constant expressions (InsertValueConstantExpr and ExtractValueConstantExpr) would be treated as non-constant instructions (InsertValueInst and ExtractValueInst). This bug resulted in an incorrect memory access, which manifested as an assertion failure in SDValue::SDValue.

Fixes PR#33094.

Submitted on behalf of @Praetonus (Benoit Vey)

Differential Revision: https://reviews.llvm.org/D34538

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307502 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86][AVX512] Regenerate AVX512VL comparison tests.
Simon Pilgrim [Sun, 9 Jul 2017 15:47:43 +0000 (15:47 +0000)]
[X86][AVX512] Regenerate AVX512VL comparison tests.

Show poor codegen on KNL targets as mentioned on D35179

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307500 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PM] Fix a nasty bug in the new PM where we failed to properly
Chandler Carruth [Sun, 9 Jul 2017 13:45:11 +0000 (13:45 +0000)]
[PM] Fix a nasty bug in the new PM where we failed to properly
invalidation of analyses when merging SCCs.

While I've added a bunch of testing of this, it takes something much
more like the inliner to really trigger this as you need to have
partially-analyzed SCCs with updates at just the right time. So I've
added a direct test for this using the inliner and verifying the
domtree. Without the changes here, this test ends up finding a stale
dominator tree.

However, to handle this properly, we need to invalidate analyses
*before* merging the SCCs. After talking to Philip and Sanjoy about this
they convinced me this was the right approach. To do this, we need
a callback mechanism when merging SCCs so we can observe the cycle that
will be merged before the merge happens. This API update ended up being
surprisingly easy.

With this commit, the new PM passes the test-suite again. It hadn't
since MemorySSA was enabled for EarlyCSE as that also will find this bug
very quickly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307498 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PM] Add unittesting of the call graph update logic with complex
Chandler Carruth [Sun, 9 Jul 2017 13:16:55 +0000 (13:16 +0000)]
[PM] Add unittesting of the call graph update logic with complex
dependencies between analyses.

This uncovers even more issues with the proxies and the splitting apart
of SCCs which are fixed in this patch. I discovered this while trying to
add more rigorous testing for a change I'm making to the call graph
update invalidation logic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307497 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[ADT] Fix a test case to use a correct escape for a null byte followed
Chandler Carruth [Sun, 9 Jul 2017 07:37:47 +0000 (07:37 +0000)]
[ADT] Fix a test case to use a correct escape for a null byte followed
by a valid octal digit.

The length argument shows that this was in fact the intent.

This was pointed out in IRC, thanks to eddyb!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307496 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[X86] Remove check for AVX512 support from skylake-avx512 detection in getHostCPUName.
Craig Topper [Sun, 9 Jul 2017 07:26:14 +0000 (07:26 +0000)]
[X86] Remove check for AVX512 support from skylake-avx512 detection in getHostCPUName.

Users of getHostCPUName should also use getHostCPUFeatures which will take care of making sure avx512 is disabled if the CPU doesn't support it. This is consistent with what we do for other CPUs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307495 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[GlobalISel][X86] Add legalizer tests for G_LOAD/G_STORE operations. NFC.
Igor Breger [Sun, 9 Jul 2017 07:25:57 +0000 (07:25 +0000)]
[GlobalISel][X86] Add legalizer tests for G_LOAD/G_STORE operations. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307494 91177308-0d34-0410-b5e6-96231b3b80d8

7 years ago[PM] Teach PreservedAnalyses to have an `allInSet` static factory
Chandler Carruth [Sun, 9 Jul 2017 07:23:27 +0000 (07:23 +0000)]
[PM] Teach PreservedAnalyses to have an `allInSet` static factory
function template to simplify building a quick object with a set marked
as preserved.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307493 91177308-0d34-0410-b5e6-96231b3b80d8