platform/upstream/llvm.git
5 years ago[NFC][InstSimplify] rewrite test added in r371537 to use non-null pointer instead
Roman Lebedev [Tue, 10 Sep 2019 18:40:00 +0000 (18:40 +0000)]
[NFC][InstSimplify] rewrite test added in r371537 to use non-null pointer instead

I only want to ensure that %offset is non-zero there,
it doesn't matter how that info is conveyed.
As filed in PR43267, the assumption way does not work.

llvm-svn: 371546

5 years ago[lldbtest] Add an "expected_cmd_failure" option to the filecheck helper
Vedant Kumar [Tue, 10 Sep 2019 18:36:53 +0000 (18:36 +0000)]
[lldbtest] Add an "expected_cmd_failure" option to the filecheck helper

llvm-svn: 371544

5 years ago[Function] Factor out GetCallEdgeForReturnAddress, NFC
Vedant Kumar [Tue, 10 Sep 2019 18:36:50 +0000 (18:36 +0000)]
[Function] Factor out GetCallEdgeForReturnAddress, NFC

Finding the call edge in a function which corresponds to a particular
return address is a generic/useful operation.

llvm-svn: 371543

5 years agoGlobalISel/TableGen: Handle REG_SEQUENCE patterns
Matt Arsenault [Tue, 10 Sep 2019 17:57:33 +0000 (17:57 +0000)]
GlobalISel/TableGen: Handle REG_SEQUENCE patterns

The scalar f64 patterns don't work yet because they fail on multiple
results from the unused implicit def of scc in the result bit
operation.

llvm-svn: 371542

5 years ago[BPI] Adjust the probability for floating point unordered comparison
Guozhi Wei [Tue, 10 Sep 2019 17:25:11 +0000 (17:25 +0000)]
[BPI] Adjust the probability for floating point unordered comparison

Since NaN is very rare in normal programs, so the probability for floating point unordered comparison should be extremely small. Current probability is 3/8, it is too large, this patch changes it to a tiny number.

Differential Revision: https://reviews.llvm.org/D65303

llvm-svn: 371541

5 years agoAMDGPU/GlobalISel: Select G_FABS/G_FNEG
Matt Arsenault [Tue, 10 Sep 2019 17:19:46 +0000 (17:19 +0000)]
AMDGPU/GlobalISel: Select G_FABS/G_FNEG

f64 doesn't work yet because tablegen currently doesn't handlde
REG_SEQUENCE.

This does regress some multi use VALU fneg cases since now the
immediate remains in an SGPR, and more moves are used for legalizing
the xor. This is a SIFixSGPRCopies deficiency.

llvm-svn: 371540

5 years agoAMDGPU/GlobalISel: Select cvt pk intrinsics
Matt Arsenault [Tue, 10 Sep 2019 17:17:05 +0000 (17:17 +0000)]
AMDGPU/GlobalISel: Select cvt pk intrinsics

llvm-svn: 371539

5 years agoAMDGPU/GlobalISel: Select llvm.amdgcn.sffbh
Matt Arsenault [Tue, 10 Sep 2019 17:16:59 +0000 (17:16 +0000)]
AMDGPU/GlobalISel: Select llvm.amdgcn.sffbh

llvm-svn: 371538

5 years ago[NFC][InstCombine][InstSimplify] PR43251 - and some patterns with offset != 0
Roman Lebedev [Tue, 10 Sep 2019 17:13:59 +0000 (17:13 +0000)]
[NFC][InstCombine][InstSimplify] PR43251 - and some patterns with offset != 0

https://rise4fun.com/Alive/21b

llvm-svn: 371537

5 years agoAMDGPU/GlobalISel: RegBankSelect for G_ZEXTLOAD/G_SEXTLOAD
Matt Arsenault [Tue, 10 Sep 2019 16:42:37 +0000 (16:42 +0000)]
AMDGPU/GlobalISel: RegBankSelect for G_ZEXTLOAD/G_SEXTLOAD

llvm-svn: 371536

5 years agoAMDGPU/GlobalISel: Legalize constant 32-bit loads
Matt Arsenault [Tue, 10 Sep 2019 16:42:31 +0000 (16:42 +0000)]
AMDGPU/GlobalISel: Legalize constant 32-bit loads

Legalize by casting to a 64-bit constant address. This isn't how the
DAG implements it, but it should.

llvm-svn: 371535

5 years ago[RISCV] Support llvm-objdump -M no-aliases and -M numeric
Sam Elliott [Tue, 10 Sep 2019 16:24:03 +0000 (16:24 +0000)]
[RISCV] Support llvm-objdump -M no-aliases and -M numeric

Summary:
Now that llvm-objdump allows target-specific options, we match the
`no-aliases` and `numeric` options for RISC-V, as supported by GNU objdump.

This is done by overriding the variables used for the command-line options, so
that the command-line options are still supported.

This patch updates all tests using `llvm-objdump -riscv-no-aliases` to use
`llvm-objdump -M no-aliases`.

Reviewers: luismarques, asb

Reviewed By: luismarques, asb

Subscribers: pzheng, hiraditya, rbar, johnrusso, simoncook, apazos, sabuasal, niosHD, kito-cheng, shiva0217, jrtc27, MaskRay, zzheng, edward-jones, rogfer01, MartinMosbeck, brucehoult, the_o, rkruppe, PkmX, jocewei, psnobl, benna, Jim, s.egerton, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66139

llvm-svn: 371534

5 years agoAMDGPU/GlobalISel: First pass at attempting to legalize load/stores
Matt Arsenault [Tue, 10 Sep 2019 16:20:14 +0000 (16:20 +0000)]
AMDGPU/GlobalISel: First pass at attempting to legalize load/stores

There's still a lot more to do, but this handles decomposing due to
alignment. I've gotten it to the point where nothing crashes or
infinite loops the legalizer.

llvm-svn: 371533

5 years agoImplement DW_OP_convert
Adrian Prantl [Tue, 10 Sep 2019 16:17:38 +0000 (16:17 +0000)]
Implement DW_OP_convert

This patch adds basic support for DW_OP_convert[1] for integer
types. Recent versions of LLVM's optimizer may insert this opcode into
DWARF expressions. DW_OP_convert is effectively a type cast operation
that takes a reference to a base type DIE (or zero) and then casts the
value at the top of the DWARF stack to that type. Internally this
works by changing the bit size of the APInt that is used as backing
storage for LLDB's DWARF stack.

I managed to write a unit test for this by implementing a mock YAML
object file / module that takes debug info sections in yaml2obj
format.

[1] Typed DWARF stack. http://www.dwarfstd.org/ShowIssue.php?issue=140425.1

<rdar://problem/48167864>

Differential Revision: https://reviews.llvm.org/D67369

llvm-svn: 371532

5 years ago[RISCV] Add Option for Printing Architectural Register Names
Sam Elliott [Tue, 10 Sep 2019 15:55:55 +0000 (15:55 +0000)]
[RISCV] Add Option for Printing Architectural Register Names

Summary:
This is an option primarily to use during testing. Instead of always
printing registers using their ABI names, this allows a user to request they
are printed with their architectural name.

This is then used in the register constraint tests to ensure the mapping
between architectural and abi names is correct.

Reviewers: asb, luismarques

Reviewed By: asb

Subscribers: pzheng, hiraditya, rbar, johnrusso, simoncook, apazos, sabuasal, niosHD, kito-cheng, shiva0217, jrtc27, MaskRay, zzheng, edward-jones, rogfer01, MartinMosbeck, brucehoult, the_o, rkruppe, PkmX, jocewei, psnobl, benna, Jim, s.egerton, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D65950

llvm-svn: 371531

5 years agoDon't emit .gnu_pubnames when tuning for LLDB.
Adrian Prantl [Tue, 10 Sep 2019 15:53:18 +0000 (15:53 +0000)]
Don't emit .gnu_pubnames when tuning for LLDB.

LLDB reads the various .apple* accelerator tables (and in the near
future: the DWARF 5 accelerator tables) which should make
.gnu_pubnames redundant. This changes the Clang driver to no longer
pass -ggnu-pubnames when tuning for LLDB.

Thanks to David Blaikie for pointing this out!
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20190422/thread.html#646062

rdar://problem/50142073

Differential Revision: https://reviews.llvm.org/D67373

llvm-svn: 371530

5 years ago[x86] add a test for BreakFalseDeps; NFC
Sanjay Patel [Tue, 10 Sep 2019 15:42:22 +0000 (15:42 +0000)]
[x86] add a test for BreakFalseDeps; NFC

As discussed in D67363

llvm-svn: 371528

5 years agoRevert "[utils] Implement the llvm-locstats tool"
Djordje Todorovic [Tue, 10 Sep 2019 14:48:52 +0000 (14:48 +0000)]
Revert "[utils] Implement the llvm-locstats tool"

This reverts commit rL371520.

llvm-svn: 371527

5 years ago[ARM] add test for BreakFalseDeps with minsize attribute; NFC
Sanjay Patel [Tue, 10 Sep 2019 14:29:02 +0000 (14:29 +0000)]
[ARM] add test for BreakFalseDeps with minsize attribute; NFC

llvm-svn: 371526

5 years ago[X86] Add AVX partial dependency tests as noted on D67363
Simon Pilgrim [Tue, 10 Sep 2019 14:28:29 +0000 (14:28 +0000)]
[X86] Add AVX partial dependency tests as noted on D67363

llvm-svn: 371525

5 years ago[ARM] auto-generate complete test checks; NFC
Sanjay Patel [Tue, 10 Sep 2019 14:26:25 +0000 (14:26 +0000)]
[ARM] auto-generate complete test checks; NFC

llvm-svn: 371524

5 years ago[VectorUtils] Remove unused include. [NFC]
Florian Hahn [Tue, 10 Sep 2019 14:20:20 +0000 (14:20 +0000)]
[VectorUtils] Remove unused include. [NFC]

Patch by Francesco Petrogalli <francesco.petrogalli@arm.com>

Reviewers: fhahn

Reviewed By: fhahn

Differential Revision: https://reviews.llvm.org/D67365

llvm-svn: 371523

5 years ago[clang][codegen][NFC] Make test patterns more permissive.
Clement Courbet [Tue, 10 Sep 2019 14:20:08 +0000 (14:20 +0000)]
[clang][codegen][NFC] Make test patterns more permissive.

See the discussion in:
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20190909/692736.html

llvm-svn: 371522

5 years ago[CMake] Don't pass all LLVM_COMPILE_FLAGS to the C compiler
David Zarzycki [Tue, 10 Sep 2019 14:19:52 +0000 (14:19 +0000)]
[CMake] Don't pass all LLVM_COMPILE_FLAGS to the C compiler

GCC (unlike clang!) warns about C++ flags when compiling C.

https://reviews.llvm.org/D67171

llvm-svn: 371521

5 years ago[utils] Implement the llvm-locstats tool
Djordje Todorovic [Tue, 10 Sep 2019 13:47:03 +0000 (13:47 +0000)]
[utils] Implement the llvm-locstats tool

The tool reports verbose output for the DWARF debug location coverage.
The llvm-locstats for each variable or formal parameter DIE computes what
percentage from the code section bytes, where it is in scope, it has
location description. The line 0 shows the number (and the percentage) of
DIEs with no location information, but the line 100 shows the number (and
the percentage) of DIEs where there is location information in all code
section bytes (where the variable or parameter is in the scope). The line
50..59 shows the number (and the percentage) of DIEs where the location
information is in between 50 and 59 percentage of its scope covered.

The tool will be very useful for tracking improvements regarding the
"debugging optimized code" support with LLVM ecosystem.

Differential Revision: https://reviews.llvm.org/D66526

llvm-svn: 371520

5 years ago[NFC][InstCombine] PR43251 - valid for other predicates too
Roman Lebedev [Tue, 10 Sep 2019 13:29:40 +0000 (13:29 +0000)]
[NFC][InstCombine] PR43251 - valid for other predicates too

llvm-svn: 371519

5 years ago[InstCombine] Use SimplifyFMulInst to simplify multiply in fma.
Florian Hahn [Tue, 10 Sep 2019 13:10:28 +0000 (13:10 +0000)]
[InstCombine] Use SimplifyFMulInst to simplify multiply in fma.

This allows us to fold fma's that multiply with 0.0. Also, the
multiply by 1.0 case is handled there as well. The fneg/fabs cases
are not handled by SimplifyFMulInst, so we need to keep them.

Reviewers: spatel, anemet, lebedev.ri

Reviewed By: spatel

Differential Revision: https://reviews.llvm.org/D67351

llvm-svn: 371518

5 years ago[InstCombine] Precommit tests for D67351.
Florian Hahn [Tue, 10 Sep 2019 13:05:34 +0000 (13:05 +0000)]
[InstCombine] Precommit tests for D67351.

llvm-svn: 371517

5 years ago[BreakFalseDeps] fix typos/grammar in documentation comment; NFC
Sanjay Patel [Tue, 10 Sep 2019 13:00:31 +0000 (13:00 +0000)]
[BreakFalseDeps] fix typos/grammar in documentation comment; NFC

llvm-svn: 371516

5 years ago[Object] Implement relocation resolver for COFF ARM/ARM64
Martin Storsjo [Tue, 10 Sep 2019 12:31:40 +0000 (12:31 +0000)]
[Object] Implement relocation resolver for COFF ARM/ARM64

Adding testscases for this via llvm-dwarfdump.

Also add testcases for the existing resolver support for X86.

Differential Revision: https://reviews.llvm.org/D67340

llvm-svn: 371515

5 years ago[ELF][test] Make tests more tolerant to exact symbol addresses
Fangrui Song [Tue, 10 Sep 2019 12:28:07 +0000 (12:28 +0000)]
[ELF][test] Make tests more tolerant to exact symbol addresses

Delete relocation-local.s and relocation-shared.s - covered by various tests

llvm-svn: 371514

5 years ago Fix the "git modified" issue on the preserve-comments-crlf.s.
Haojian Wu [Tue, 10 Sep 2019 12:17:49 +0000 (12:17 +0000)]
 Fix the "git modified" issue on the preserve-comments-crlf.s.

Summary:
 rL371473 corrected the line ending (to crlf) in preserve-comments-crlf.s, but it causes a wired issue on git
 repository (I ran git pull today, it showed a local change of that file even I did not change anything).

Reviewers: gribozavr

Reviewed By: gribozavr

Subscribers: llvm-commits, cfe-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D67391

llvm-svn: 371513

5 years ago[lldb] Readd missing functionalities/breakpoint tests
Raphael Isemann [Tue, 10 Sep 2019 12:04:04 +0000 (12:04 +0000)]
[lldb] Readd missing functionalities/breakpoint tests

It seems when I restructured the test folders the functionalities/breakpoint
was deleted. This just reverts this change and re-adds the tests.

llvm-svn: 371512

5 years ago[Alignment][NFC] Use llvm::Align for TargetLowering::getPrefLoopAlignment
Guillaume Chatelet [Tue, 10 Sep 2019 12:00:43 +0000 (12:00 +0000)]
[Alignment][NFC] Use llvm::Align for TargetLowering::getPrefLoopAlignment

Summary:
This is patch is part of a series to introduce an Alignment type.
See this thread for context: http://lists.llvm.org/pipermail/llvm-dev/2019-July/133851.html
See this patch for the introduction of the type: https://reviews.llvm.org/D64790

Reviewers: courbet

Reviewed By: courbet

Subscribers: wuzish, arsenm, nemanjai, jvesely, nhaehnle, hiraditya, kbarton, MaskRay, jsji, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D67386

llvm-svn: 371511

5 years agoReland [DWARF] Add a unit test for DWARFUnit::getLength().
Igor Kudrin [Tue, 10 Sep 2019 11:54:32 +0000 (11:54 +0000)]
Reland [DWARF] Add a unit test for DWARFUnit::getLength().

This is a follow-up of rL369529, where the return value of
DWARFUnit::getLength() was changed from uint32_t to uint64_t.
The test checks that a unit header with Length > 4G can be successfully
parsed and the value of the Length field is not truncated.

Differential Revision: https://reviews.llvm.org/D67276

llvm-svn: 371510

5 years ago[LLD][COFF] Add index to disambiguate archive members when using -wholearchive
Rui Ueyama [Tue, 10 Sep 2019 11:50:26 +0000 (11:50 +0000)]
[LLD][COFF] Add index to disambiguate archive members when using -wholearchive

Patch by Markus Böck.

PR42951: When linking an archive with members that have the same name linking
fails when using the -wholearchive option. This patch passes the index
of the member in the archive to the offset parameter to disambiguate the
member.

Differential Revision: https://reviews.llvm.org/D66239

llvm-svn: 371509

5 years ago[AMDGPU]: PHI Elimination hooks added for custom COPY insertion.
Alexander Timofeev [Tue, 10 Sep 2019 10:58:57 +0000 (10:58 +0000)]
[AMDGPU]: PHI Elimination hooks added for custom COPY insertion.

  Reviewers: rampitec, vpykhtin

  Differential Revision: https://reviews.llvm.org/D67101

llvm-svn: 371508

5 years agoRevert "Reland "r364412 [ExpandMemCmp][MergeICmps] Move passes out of CodeGen into...
Dmitri Gribenko [Tue, 10 Sep 2019 10:39:09 +0000 (10:39 +0000)]
Revert "Reland "r364412 [ExpandMemCmp][MergeICmps] Move passes out of CodeGen into opt pipeline.""

This reverts commit r371502, it broke tests
(clang/test/CodeGenCXX/auto-var-init.cpp).

llvm-svn: 371507

5 years ago[llvm-dwarfdump] Add additional stats fields
Djordje Todorovic [Tue, 10 Sep 2019 10:37:28 +0000 (10:37 +0000)]
[llvm-dwarfdump] Add additional stats fields

The additional fields will be parsed by the llvm-locstats tool in order to
produce more human readable output of the DWARF debug location quality
generated.

Differential Revision: https://reviews.llvm.org/D66525

llvm-svn: 371506

5 years agoRevert "[DWARF] Add a unit test for DWARFUnit::getLength()" because it broke ASAN...
Igor Kudrin [Tue, 10 Sep 2019 10:17:48 +0000 (10:17 +0000)]
Revert "[DWARF] Add a unit test for DWARFUnit::getLength()" because it broke ASAN bot.

llvm-svn: 371505

5 years ago[clangd] Collect location of macro definition in the ParsedAST
Haojian Wu [Tue, 10 Sep 2019 10:10:36 +0000 (10:10 +0000)]
[clangd] Collect location of macro definition in the ParsedAST

allows semantic hightlighting macro definition

Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, kadircet, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D67264

llvm-svn: 371504

5 years ago[ExpandMemCmp] Add lit.local.cfg
Clement Courbet [Tue, 10 Sep 2019 10:00:15 +0000 (10:00 +0000)]
[ExpandMemCmp] Add lit.local.cfg

To prevent AArch64 tests from running when the target is not compiled.

Fixes r371502:

/home/buildslave/ps4-buildslave4/llvm-clang-lld-x86_64-scei-ps4-ubuntu-fast/llvm.src/test/Transforms/ExpandMemCmp/AArch64/memcmp.ll:11:15: error: CHECK-NEXT: expected string not found in input
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i8* [[S1:%.*]] to i64*

llvm-svn: 371503

5 years agoReland "r364412 [ExpandMemCmp][MergeICmps] Move passes out of CodeGen into opt pipeline."
Clement Courbet [Tue, 10 Sep 2019 09:18:00 +0000 (09:18 +0000)]
Reland "r364412 [ExpandMemCmp][MergeICmps] Move passes out of CodeGen into opt pipeline."

With a fix for sanitizer breakage (see explanation in D60318).

llvm-svn: 371502

5 years ago[yaml2obj] Set p_align to the maximum sh_addralign of contained sections
Fangrui Song [Tue, 10 Sep 2019 09:16:34 +0000 (09:16 +0000)]
[yaml2obj] Set p_align to the maximum sh_addralign of contained sections

The address difference between two sections in a PT_LOAD is a constant.
Consider a hypothetical case (pagesize can be very small, say, 4).

```
.text     sh_addralign=4
.text.hot sh_addralign=16
```

If we set p_align to 4, the PT_LOAD will be loaded at an address which
is a multiple of 4. The address of .text.hot is guaranteed to be a
multiple of 4, but not necessarily a multiple of 16.

This patch deletes the constraint

  if (SHeader->sh_offset == PHeader.p_offset)

Reviewed By: grimar, jhenderson

Differential Revision: https://reviews.llvm.org/D67260

llvm-svn: 371501

5 years ago[DWARF] Add a unit test for DWARFUnit::getLength().
Igor Kudrin [Tue, 10 Sep 2019 09:03:24 +0000 (09:03 +0000)]
[DWARF] Add a unit test for DWARFUnit::getLength().

This is a follow-up of rL369529, where the return value of
DWARFUnit::getLength() was changed from uint32_t to uint64_t.
The test checks that a unit header with Length > 4G can be successfully
parsed and the value of the Length field is not truncated.

Differential Revision: https://reviews.llvm.org/D67276

llvm-svn: 371499

5 years ago[Alignment] Use Align for TargetLowering::MinStackArgumentAlignment
Guillaume Chatelet [Tue, 10 Sep 2019 09:01:18 +0000 (09:01 +0000)]
[Alignment] Use Align for TargetLowering::MinStackArgumentAlignment

Summary:
This is patch is part of a series to introduce an Alignment type.
See this thread for context: http://lists.llvm.org/pipermail/llvm-dev/2019-July/133851.html
See this patch for the introduction of the type: https://reviews.llvm.org/D64790

Reviewers: courbet

Subscribers: sdardis, nemanjai, hiraditya, kbarton, jrtc27, MaskRay, atanasyan, jsji, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D67288

llvm-svn: 371498

5 years agoRevert Remove REQUIRES:shell from tests that pass for me on Windows
James Henderson [Tue, 10 Sep 2019 08:48:33 +0000 (08:48 +0000)]
Revert Remove REQUIRES:shell from tests that pass for me on Windows

This reverts r371478 (git commit a9980f60ce083fa6d5fd03c12c58ca0b293e3d60)

llvm-svn: 371497

5 years ago[RISCV] Make -march=rv{32,64}gc the default in RISC-V Linux
Roger Ferrer Ibanez [Tue, 10 Sep 2019 08:16:24 +0000 (08:16 +0000)]
[RISCV] Make -march=rv{32,64}gc the default in RISC-V Linux

This is the logical follow-up of D65634.

Differential Revision: https://reviews.llvm.org/D66003

llvm-svn: 371496

5 years ago[clangd] some tweaks on the vscode readme, NFC
Haojian Wu [Tue, 10 Sep 2019 08:07:12 +0000 (08:07 +0000)]
[clangd] some tweaks on the vscode readme, NFC

llvm-svn: 371495

5 years ago[RISCV] Default to ilp32d/lp64d in RISC-V Linux
Roger Ferrer Ibanez [Tue, 10 Sep 2019 07:57:36 +0000 (07:57 +0000)]
[RISCV] Default to ilp32d/lp64d in RISC-V Linux

When running clang as a native compiler in RISC-V Linux the flag
-mabi=ilp32d / -mabi=lp64d is always mandatory. This change makes it the
default there.

Differential Revision: https://reviews.llvm.org/D65634

llvm-svn: 371494

5 years ago[LegalizeTypes] Teach SoftenFloatOp_SELECT_CC to handle operand 2 or 3 being softened.
Craig Topper [Tue, 10 Sep 2019 07:56:02 +0000 (07:56 +0000)]
[LegalizeTypes] Teach SoftenFloatOp_SELECT_CC to handle operand 2 or 3 being softened.

This can only happen on X86 when fp128 is a legal type, but we
go through softening to generate libcalls. This causes fp128 to
be softened to fp128 instead of an integer type. This can be
removed if D67128 lands.

llvm-svn: 371493

5 years ago[RISCV] Move architecture parsing code into its own function
Roger Ferrer Ibanez [Tue, 10 Sep 2019 07:47:34 +0000 (07:47 +0000)]
[RISCV] Move architecture parsing code into its own function

I plan to reuse it in a later patch.

This is almost NFC except a small change in control flow when diagnosing
+d without +f.

Differential Revision: https://reviews.llvm.org/D66002

llvm-svn: 371492

5 years ago[LLDB] FreeBSD fix new SetFile call.
David Carlier [Tue, 10 Sep 2019 07:33:39 +0000 (07:33 +0000)]
[LLDB] FreeBSD fix new SetFile call.

llvm-svn: 371491

5 years agogn build: Merge r371488
Nico Weber [Tue, 10 Sep 2019 06:31:59 +0000 (06:31 +0000)]
gn build: Merge r371488

llvm-svn: 371489

5 years agoRevert "clang-misexpect: Profile Guided Validation of Performance Annotations in...
Petr Hosek [Tue, 10 Sep 2019 06:25:13 +0000 (06:25 +0000)]
Revert "clang-misexpect: Profile Guided Validation of Performance Annotations in LLVM"

This reverts commit r371484: this broke sanitizer-x86_64-linux-fast bot.

llvm-svn: 371488

5 years ago[X86] Add broadcast load unfolding support for VCMPPS/PD.
Craig Topper [Tue, 10 Sep 2019 05:49:53 +0000 (05:49 +0000)]
[X86] Add broadcast load unfolding support for VCMPPS/PD.

llvm-svn: 371487

5 years ago[X86] Add broadcast load unfold tests for VCMPPS/PD.
Craig Topper [Tue, 10 Sep 2019 05:49:48 +0000 (05:49 +0000)]
[X86] Add broadcast load unfold tests for VCMPPS/PD.

llvm-svn: 371486

5 years agogn build: Merge r371484
Nico Weber [Tue, 10 Sep 2019 03:18:25 +0000 (03:18 +0000)]
gn build: Merge r371484

llvm-svn: 371485

5 years agoclang-misexpect: Profile Guided Validation of Performance Annotations in LLVM
Petr Hosek [Tue, 10 Sep 2019 03:11:39 +0000 (03:11 +0000)]
clang-misexpect: Profile Guided Validation of Performance Annotations in LLVM

This patch contains the basic functionality for reporting potentially
incorrect usage of __builtin_expect() by comparing the developer's
annotation against a collected PGO profile. A more detailed proposal and
discussion appears on the CFE-dev mailing list
(http://lists.llvm.org/pipermail/cfe-dev/2019-July/062971.html) and a
prototype of the initial frontend changes appear here in D65300

We revised the work in D65300 by moving the misexpect check into the
LLVM backend, and adding support for IR and sampling based profiles, in
addition to frontend instrumentation.

We add new misexpect metadata tags to those instructions directly
influenced by the llvm.expect intrinsic (branch, switch, and select)
when lowering the intrinsics. The misexpect metadata contains
information about the expected target of the intrinsic so that we can
check against the correct PGO counter when emitting diagnostics, and the
compiler's values for the LikelyBranchWeight and UnlikelyBranchWeight.
We use these branch weight values to determine when to emit the
diagnostic to the user.

A future patch should address the comment at the top of
LowerExpectIntrisic.cpp to hoist the LikelyBranchWeight and
UnlikelyBranchWeight values into a shared space that can be accessed
outside of the LowerExpectIntrinsic pass. Once that is done, the
misexpect metadata can be updated to be smaller.

In the long term, it is possible to reconstruct portions of the
misexpect metadata from the existing profile data. However, we have
avoided this to keep the code simple, and because some kind of metadata
tag will be required to identify which branch/switch/select instructions
are influenced by the use of llvm.expect

Patch By: paulkirth
Differential Revision: https://reviews.llvm.org/D66324

llvm-svn: 371484

5 years ago[PowerPC][NFC] Update test assertions using update_llc_test_checks.py
Kai Luo [Tue, 10 Sep 2019 02:28:24 +0000 (02:28 +0000)]
[PowerPC][NFC] Update test assertions using update_llc_test_checks.py

Summary:
This patch is made due to https://reviews.llvm.org/rL371289 where typo
fixes failed.

Differential Revision: https://reviews.llvm.org/D67317

llvm-svn: 371483

5 years agoRevert [git-llvm] Do not reinvent `@{upstream}`
Mehdi Amini [Tue, 10 Sep 2019 01:26:36 +0000 (01:26 +0000)]
Revert [git-llvm] Do not reinvent `@{upstream}`

This reverts r371290 (git commit 7faffd544b16f851a632d6b8f93e3c8485ff34bb)

The change wasnt NFC and broke some users' workflow. Reverting while figuring
out the best alternative to move forward.

llvm-svn: 371480

5 years agogn build: Merge r371466
Nico Weber [Tue, 10 Sep 2019 01:11:30 +0000 (01:11 +0000)]
gn build: Merge r371466

llvm-svn: 371479

5 years agoRemove REQUIRES:shell from tests that pass for me on Windows
Reid Kleckner [Tue, 10 Sep 2019 00:50:32 +0000 (00:50 +0000)]
Remove REQUIRES:shell from tests that pass for me on Windows

I see in the history for some of these tests REQUIRES:shell was used as
a way to disable tests on Windows because they are flaky there. I tried
not to re-enable such tests, but it's possible that I missed some and
this will re-enable flaky tests on Windows. If so, we should disable
them with UNSUPPORTED:system-windows and add a comment that they are
flaky there. So far as I can tell, the lit internal shell is capable of
running all of these tests, and we shouldn't use REQUIRES:shell as a
proxy for Windows.

llvm-svn: 371478

5 years agogn build: (manually) merge r371429
Nico Weber [Tue, 10 Sep 2019 00:48:20 +0000 (00:48 +0000)]
gn build: (manually) merge r371429

llvm-svn: 371477

5 years agoFix crash mangling an explicit lambda non-type template parameter pack
Richard Smith [Tue, 10 Sep 2019 00:39:53 +0000 (00:39 +0000)]
Fix crash mangling an explicit lambda non-type template parameter pack
that is not a pack expansion.

llvm-svn: 371476

5 years ago[llvm][ADT][NFC] Add test for makeArrayRef(std::array)
Jan Korous [Tue, 10 Sep 2019 00:29:35 +0000 (00:29 +0000)]
[llvm][ADT][NFC] Add test for makeArrayRef(std::array)

llvm-svn: 371475

5 years ago[Utility] Replace `lldb_private::CleanUp` by `llvm::scope_exit`
Jonas Devlieghere [Tue, 10 Sep 2019 00:20:50 +0000 (00:20 +0000)]
[Utility] Replace `lldb_private::CleanUp` by `llvm::scope_exit`

This removes the CleanUp class and replaces its usages with llvm's
ScopeExit, which has similar semantics.

Differential revision: https://reviews.llvm.org/D67378

llvm-svn: 371474

5 years agoRemove some unnecessary REQUIRES: shell lines
Reid Kleckner [Tue, 10 Sep 2019 00:06:52 +0000 (00:06 +0000)]
Remove some unnecessary REQUIRES: shell lines

This means these tests will run on Windows. Replace one with
UNSUPPORTED: system-windows.

llvm-svn: 371473

5 years ago[Expression] Remove unused header from LLVMUserExpression
Alex Langford [Mon, 9 Sep 2019 23:59:54 +0000 (23:59 +0000)]
[Expression] Remove unused header from LLVMUserExpression

llvm-svn: 371472

5 years agoAMDGPU/GlobalISel: Fix insert point when lowering fminnum/fmaxnum
Matt Arsenault [Mon, 9 Sep 2019 23:30:11 +0000 (23:30 +0000)]
AMDGPU/GlobalISel: Fix insert point when lowering fminnum/fmaxnum

llvm-svn: 371471

5 years ago[Symbol] Give ClangASTContext a PersistentExpressionState instead of a ClangPersisten...
Alex Langford [Mon, 9 Sep 2019 23:11:43 +0000 (23:11 +0000)]
[Symbol] Give ClangASTContext a PersistentExpressionState instead of a ClangPersistentVariables

ClangASTContext doesn't use m_persistent_variables in a way specific to
ClangPersistentVariables. Therefore, it should hold a unique pointer to
PersistentExpressionState instead of a ClangPersistentVariablesUP.
This also prevents you from pulling in a plugin header when including
ClangASTContext.h

Doing this exposed an implicit dependency in ObjCLanguage that was
corrected by including ClangModulesDeclVendor.h

llvm-svn: 371470

5 years agoFix incorrect demangling of call operator of lambda with explicit
Richard Smith [Mon, 9 Sep 2019 23:07:25 +0000 (23:07 +0000)]
Fix incorrect demangling of call operator of lambda with explicit
template parameters due to registering template parameters twice.

llvm-svn: 371469

5 years agoPR43242: Fix crash when typo-correcting to an operator() that should not
Richard Smith [Mon, 9 Sep 2019 23:07:22 +0000 (23:07 +0000)]
PR43242: Fix crash when typo-correcting to an operator() that should not
have been visible.

llvm-svn: 371468

5 years agoAMDGPU/GlobalISel: Rename MIRBuilder to B. NFC
Austin Kerbow [Mon, 9 Sep 2019 23:06:13 +0000 (23:06 +0000)]
AMDGPU/GlobalISel: Rename MIRBuilder to B. NFC

Reviewers: arsenm

Reviewed By: arsenm

Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, rovka, dstuttard, tpr, t-tye, hiraditya, Petar.Avramovic, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D67374

llvm-svn: 371467

5 years ago[Windows] Replace TrapUnreachable with an int3 insertion pass
Reid Kleckner [Mon, 9 Sep 2019 23:04:25 +0000 (23:04 +0000)]
[Windows] Replace TrapUnreachable with an int3 insertion pass

This is an alternative to D66980, which was reverted. Instead of
inserting a pseudo instruction that optionally expands to nothing, add a
pass that inserts int3 when appropriate after basic block layout.

Reviewers: hans

Differential Revision: https://reviews.llvm.org/D67201

llvm-svn: 371466

5 years ago[GlobalISel]: Fix a bug where we could dereference None
Aditya Nandakumar [Mon, 9 Sep 2019 22:51:41 +0000 (22:51 +0000)]
[GlobalISel]: Fix a bug where we could dereference None

getConstantVRegVal returns None when dealing with constants > 64 bits.
Don't assume we always have a value in GISelKnownBits.

llvm-svn: 371465

5 years agoSimplify demangler rule for lambda-expressions to match discussion on
Richard Smith [Mon, 9 Sep 2019 22:26:04 +0000 (22:26 +0000)]
Simplify demangler rule for lambda-expressions to match discussion on
cxx-abi list.

llvm-svn: 371462

5 years agoLangRef: mention MSan's problem with speculative conditional branches.
Evgeniy Stepanov [Mon, 9 Sep 2019 22:24:57 +0000 (22:24 +0000)]
LangRef: mention MSan's problem with speculative conditional branches.

Summary:
This short blurb aims to disallow optimizations like we had to revert
(under MSan) in
  https://reviews.llvm.org/D21165
  https://bugs.llvm.org/show_bug.cgi?id=28054
  https://reviews.llvm.org/D67205

Reviewers: vitalybuka, efriedma

Subscribers: llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D67244

llvm-svn: 371461

5 years agoRevert "[Reproducer] Add a `cont` to ModuleCXX.test"
Jonas Devlieghere [Mon, 9 Sep 2019 22:07:45 +0000 (22:07 +0000)]
Revert "[Reproducer] Add a `cont` to ModuleCXX.test"

This should no longer be necessary after r371459.

llvm-svn: 371460

5 years ago[Reproducer] Disconnect when the replay server is out of packets.
Jonas Devlieghere [Mon, 9 Sep 2019 22:05:48 +0000 (22:05 +0000)]
[Reproducer] Disconnect when the replay server is out of packets.

This is a fix for the issue described in r371144.

> On more than one occasion I've found this test got stuck during replay
> while waiting for a packet from debugserver when the debugger was in
> the process of being destroyed.

When the replay server is out of packets we should just disconnect so
the debugger doesn't have to do any cleanup that it wouldn't do during
capture.

llvm-svn: 371459

5 years ago[mips] Make another set of test cases more tolerant to exact symbol addresses. NFC
Simon Atanasyan [Mon, 9 Sep 2019 22:04:20 +0000 (22:04 +0000)]
[mips] Make another set of test cases more tolerant to exact symbol addresses. NFC

llvm-svn: 371458

5 years agoFix ELF core file memory reading for PT_LOAD program headers with no p_filesz
Greg Clayton [Mon, 9 Sep 2019 21:45:49 +0000 (21:45 +0000)]
Fix ELF core file memory reading for PT_LOAD program headers with no p_filesz

Prior to this fix, ELF files might contain PT_LOAD program headers that had a valid p_vaddr, and a valid file p_offset, but the p_filesz would be zero. For example in llvm-project/lldb/test/testcases/functionalities/postmortem/elf-core/thread_crash/linux-i386.core we see:

Program Headers:
Index   p_type           p_flags    p_offset           p_vaddr            p_paddr            p_filesz           p_memsz            p_align
======= ---------------- ---------- ------------------ ------------------ ------------------ ------------------ ------------------ ------------------
[    0] PT_NOTE          0x00000000 0x0000000000000474 0x0000000000000000 0x0000000000000000 0x0000000000001940 0x0000000000000000 0x0000000000000000
[    1] PT_LOAD          0x00000005 0x0000000000002000 0x0000000008048000 0x0000000000000000 0x0000000000000000 0x0000000000003000 0x0000000000001000
[    2] PT_LOAD          0x00000004 0x0000000000002000 0x000000000804b000 0x0000000000000000 0x0000000000000000 0x0000000000001000 0x0000000000001000
[    3] PT_LOAD          0x00000006 0x0000000000002000 0x000000000804c000 0x0000000000000000 0x0000000000000000 0x0000000000001000 0x0000000000001000
[    4] PT_LOAD          0x00000006 0x0000000000002000 0x0000000009036000 0x0000000000000000 0x0000000000000000 0x0000000000025000 0x0000000000001000
[    5] PT_LOAD          0x00000000 0x0000000000002000 0x00000000f63a1000 0x0000000000000000 0x0000000000000000 0x0000000000001000 0x0000000000001000
[    6] PT_LOAD          0x00000006 0x0000000000002000 0x00000000f63a2000 0x0000000000000000 0x0000000000000000 0x0000000000800000 0x0000000000001000
[    7] PT_LOAD          0x00000000 0x0000000000002000 0x00000000f6ba2000 0x0000000000000000 0x0000000000000000 0x0000000000001000 0x0000000000001000
[    8] PT_LOAD          0x00000006 0x0000000000002000 0x00000000f6ba3000 0x0000000000000000 0x0000000000000000 0x0000000000804000 0x0000000000001000
[    9] PT_LOAD          0x00000005 0x0000000000002000 0x00000000f73a7000 0x0000000000000000 0x0000000000000000 0x00000000001b1000 0x0000000000001000
[   10] PT_LOAD          0x00000004 0x0000000000002000 0x00000000f7558000 0x0000000000000000 0x0000000000000000 0x0000000000002000 0x0000000000001000
[   11] PT_LOAD          0x00000006 0x0000000000002000 0x00000000f755a000 0x0000000000000000 0x0000000000000000 0x0000000000001000 0x0000000000001000
[   12] PT_LOAD          0x00000006 0x0000000000002000 0x00000000f755b000 0x0000000000000000 0x0000000000000000 0x0000000000003000 0x0000000000001000
[   13] PT_LOAD          0x00000005 0x0000000000002000 0x00000000f755e000 0x0000000000000000 0x0000000000000000 0x0000000000019000 0x0000000000001000
[   14] PT_LOAD          0x00000004 0x0000000000002000 0x00000000f7577000 0x0000000000000000 0x0000000000000000 0x0000000000001000 0x0000000000001000
[   15] PT_LOAD          0x00000006 0x0000000000002000 0x00000000f7578000 0x0000000000000000 0x0000000000000000 0x0000000000001000 0x0000000000001000
[   16] PT_LOAD          0x00000006 0x0000000000002000 0x00000000f7579000 0x0000000000000000 0x0000000000000000 0x0000000000002000 0x0000000000001000
[   17] PT_LOAD          0x00000005 0x0000000000002000 0x00000000f757b000 0x0000000000000000 0x0000000000000000 0x000000000001c000 0x0000000000001000
[   18] PT_LOAD          0x00000004 0x0000000000002000 0x00000000f7597000 0x0000000000000000 0x0000000000000000 0x0000000000001000 0x0000000000001000
[   19] PT_LOAD          0x00000006 0x0000000000002000 0x00000000f7598000 0x0000000000000000 0x0000000000000000 0x0000000000001000 0x0000000000001000
[   20] PT_LOAD          0x00000005 0x0000000000002000 0x00000000f7599000 0x0000000000000000 0x0000000000000000 0x0000000000053000 0x0000000000001000
[   21] PT_LOAD          0x00000004 0x0000000000002000 0x00000000f75ec000 0x0000000000000000 0x0000000000000000 0x0000000000001000 0x0000000000001000
[   22] PT_LOAD          0x00000006 0x0000000000002000 0x00000000f75ed000 0x0000000000000000 0x0000000000000000 0x0000000000001000 0x0000000000001000
[   23] PT_LOAD          0x00000005 0x0000000000002000 0x00000000f75ee000 0x0000000000000000 0x0000000000000000 0x0000000000176000 0x0000000000001000
[   24] PT_LOAD          0x00000004 0x0000000000002000 0x00000000f7764000 0x0000000000000000 0x0000000000000000 0x0000000000006000 0x0000000000001000
[   25] PT_LOAD          0x00000006 0x0000000000002000 0x00000000f776a000 0x0000000000000000 0x0000000000000000 0x0000000000001000 0x0000000000001000
[   26] PT_LOAD          0x00000006 0x0000000000002000 0x00000000f776b000 0x0000000000000000 0x0000000000000000 0x0000000000003000 0x0000000000001000
[   27] PT_LOAD          0x00000006 0x0000000000002000 0x00000000f778a000 0x0000000000000000 0x0000000000000000 0x0000000000002000 0x0000000000001000
[   28] PT_LOAD          0x00000004 0x0000000000002000 0x00000000f778c000 0x0000000000000000 0x0000000000002000 0x0000000000002000 0x0000000000001000
[   29] PT_LOAD          0x00000005 0x0000000000004000 0x00000000f778e000 0x0000000000000000 0x0000000000002000 0x0000000000002000 0x0000000000001000
[   30] PT_LOAD          0x00000005 0x0000000000006000 0x00000000f7790000 0x0000000000000000 0x0000000000000000 0x0000000000022000 0x0000000000001000
[   31] PT_LOAD          0x00000004 0x0000000000006000 0x00000000f77b3000 0x0000000000000000 0x0000000000000000 0x0000000000001000 0x0000000000001000
[   32] PT_LOAD          0x00000006 0x0000000000006000 0x00000000f77b4000 0x0000000000000000 0x0000000000000000 0x0000000000001000 0x0000000000001000
[   33] PT_LOAD          0x00000006 0x0000000000006000 0x00000000ffa25000 0x0000000000000000 0x0000000000000000 0x0000000000022000 0x0000000000001000
Prior to this fix if users tried to read memory from one of these addresses like 0x8048000, they would end up incorrectly reading from the next memory region that actually had a p_filesz which would be 0x00000000f778c000 in this case. This fix correctly doesn't include program headers with zero p_filesz in the ProcessELFCore::m_core_aranges that is used to read memory. I found two cores files that have this same issue and added tests.

Differential Revision: https://reviews.llvm.org/D67370

llvm-svn: 371457

5 years ago[Tests] Fix a typo in a test
Philip Reames [Mon, 9 Sep 2019 21:33:59 +0000 (21:33 +0000)]
[Tests] Fix a typo in a test

llvm-svn: 371456

5 years ago[Tests] Precommit test case for D67372
Philip Reames [Mon, 9 Sep 2019 21:32:16 +0000 (21:32 +0000)]
[Tests] Precommit test case for D67372

llvm-svn: 371455

5 years agoFix MSVC "not all control paths return a value" warning. NFCI.
Simon Pilgrim [Mon, 9 Sep 2019 21:30:11 +0000 (21:30 +0000)]
Fix MSVC "not all control paths return a value" warning. NFCI.

llvm-svn: 371454

5 years ago[UBSan] Follow up fix for r371442.
Max Moroz [Mon, 9 Sep 2019 21:00:25 +0000 (21:00 +0000)]
[UBSan] Follow up fix for r371442.

Reviewers: vitalybuka, hctim, Dor1s

Reviewed By: Dor1s

Subscribers: delcypher, #sanitizers, llvm-commits

Tags: #llvm, #sanitizers

Differential Revision: https://reviews.llvm.org/D67371

llvm-svn: 371453

5 years ago[LoopVectorize] Leverage speculation safety to avoid masked.loads
Philip Reames [Mon, 9 Sep 2019 20:54:13 +0000 (20:54 +0000)]
[LoopVectorize] Leverage speculation safety to avoid masked.loads

If we're vectorizing a load in a predicated block, check to see if the load can be speculated rather than predicated.  This allows us to generate a normal vector load instead of a masked.load.

To do so, we must prove that all bytes accessed on any iteration of the original loop are dereferenceable, and that all loads (across all iterations) are properly aligned.  This is equivelent to proving that hoisting the load into the loop header in the original scalar loop is safe.

Note: There are a couple of code motion todos in the code.  My intention is to wait about a day - to be sure this sticks - and then perform the NFC motion without furthe review.

Differential Revision: https://reviews.llvm.org/D66688

llvm-svn: 371452

5 years ago[analyzer] NFC: Simplify bug report equivalence classes to not be ilists.
Artem Dergachev [Mon, 9 Sep 2019 20:34:44 +0000 (20:34 +0000)]
[analyzer] NFC: Simplify bug report equivalence classes to not be ilists.

Use a vector of unique pointers instead.

Differential Revision: https://reviews.llvm.org/D67024

llvm-svn: 371451

5 years ago[analyzer] NFC: Introduce sub-classes for path-sensitive and basic reports.
Artem Dergachev [Mon, 9 Sep 2019 20:34:40 +0000 (20:34 +0000)]
[analyzer] NFC: Introduce sub-classes for path-sensitive and basic reports.

Checkers are now required to specify whether they're creating a
path-sensitive report or a path-insensitive report by constructing an
object of the respective type.

This makes BugReporter more independent from the rest of the Static Analyzer
because all Analyzer-specific code is now in sub-classes.

Differential Revision: https://reviews.llvm.org/D66572

llvm-svn: 371450

5 years ago[Tests] Add anyextend tests for unordered atomics
Philip Reames [Mon, 9 Sep 2019 20:26:52 +0000 (20:26 +0000)]
[Tests] Add anyextend tests for unordered atomics

Motivated by work on changing our representation of unordered atomics in SelectionDAG, but as an aside, all our lowerings for O3 are terrible.  Even the ones which ignore the atomicity.

llvm-svn: 371449

5 years agoRelax opcode checks in test to check for only a number instead of a specific number.
Douglas Yung [Mon, 9 Sep 2019 20:12:29 +0000 (20:12 +0000)]
Relax opcode checks in test to check for only a number instead of a specific number.

llvm-svn: 371447

5 years ago[TSan] Add AnnotateIgnoreReadsBegin declaration to tsan/test.h
Julian Lettner [Mon, 9 Sep 2019 20:07:03 +0000 (20:07 +0000)]
[TSan] Add AnnotateIgnoreReadsBegin declaration to tsan/test.h

Declare the family of AnnotateIgnore[Read,Write][Begin,End] TSan
annotations in compiler-rt/test/tsan/test.h so that we don't have to
declare them separately in every test that needs them.  Replace usages.

Leave usages that explicitly test the annotation mechanism:
  thread_end_with_ignore.cpp
  thread_end_with_ignore3.cpp

llvm-svn: 371446

5 years ago[SDAG] Add a isSimple cover functon to MemSDNode, just as we have in IR/MI [NFC]
Philip Reames [Mon, 9 Sep 2019 20:06:19 +0000 (20:06 +0000)]
[SDAG] Add a isSimple cover functon to MemSDNode, just as we have in IR/MI [NFC]

Uses are in reviews D66322 and D66318.  Submitted separately to control rebuild times.

llvm-svn: 371445

5 years ago[Driver] Handle default case in refactored addOpenMPRuntime
Pirama Arumuga Nainar [Mon, 9 Sep 2019 19:52:39 +0000 (19:52 +0000)]
[Driver] Handle default case in refactored addOpenMPRuntime

Summary:
Appease failed builds (due to -Werror and -Wswitch) where OMPRT_Unknown
is not handled in the switch statement (even though it's handled by the
early exit).

This fixes -Wswitch triggered by r371442.

Reviewers: srhines, danalbert, jdoerfert

Subscribers: guansong, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D67364

llvm-svn: 371444

5 years ago[Remarks] Fix warning for uint8_t < 0 comparison
Francis Visoiu Mistrih [Mon, 9 Sep 2019 19:47:25 +0000 (19:47 +0000)]
[Remarks] Fix warning for uint8_t < 0 comparison

http://lab.llvm.org:8011/builders/clang-with-thin-lto-ubuntu/builds/19109/steps/build-stage1-compiler/logs/stdio

llvm-svn: 371443

5 years ago[UBSan] Do not overwrite the default print_summary sanitizer option.
Max Moroz [Mon, 9 Sep 2019 19:30:48 +0000 (19:30 +0000)]
[UBSan] Do not overwrite the default print_summary sanitizer option.

Summary:
This option is true by default in sanitizer common. The default
false value was added a while ago without any reasoning in
https://github.com/llvm-mirror/compiler-rt/commit/524e934112a593ac081bf2b05aa0d60a67987f05

so, presumably it's safe to remove for consistency.

Reviewers: hctim, samsonov, morehouse, kcc, vitalybuka

Reviewed By: hctim, samsonov, vitalybuka

Subscribers: delcypher, #sanitizers, llvm-commits, kcc

Tags: #llvm, #sanitizers

Differential Revision: https://reviews.llvm.org/D67193

llvm-svn: 371442

5 years agoIntroduce infrastructure for an incremental port of SelectionDAG atomic load/store...
Philip Reames [Mon, 9 Sep 2019 19:23:22 +0000 (19:23 +0000)]
Introduce infrastructure for an incremental port of SelectionDAG atomic load/store handling

This is the first patch in a large sequence. The eventual goal is to have unordered atomic loads and stores - and possibly ordered atomics as well - handled through the normal ISEL codepaths for loads and stores. Today, there handled w/instances of AtomicSDNodes. The result of which is that all transforms need to be duplicated to work for unordered atomics. The benefit of the current design is that it's harder to introduce a silent miscompile by adding an transform which forgets about atomicity.  See the thread on llvm-dev titled "FYI: proposed changes to atomic load/store in SelectionDAG" for further context.

Note that this patch is NFC unless the experimental flag is set.

The basic strategy I plan on taking is:

    introduce infrastructure and a flag for testing (this patch)
    Audit uses of isVolatile, and apply isAtomic conservatively*
    piecemeal conservative* update generic code and x86 backedge code in individual reviews w/tests for cases which didn't check volatile, but can be found with inspection
    flip the flag at the end (with minimal diffs)
    Work through todo list identified in (2) and (3) exposing performance ops

(*) The "conservative" bit here is aimed at minimizing the number of diffs involved in (4). Ideally, there'd be none. In practice, getting it down to something reviewable by a human is the actual goal. Note that there are (currently) no paths which produce LoadSDNode or StoreSDNode with atomic MMOs, so we don't need to worry about preserving any behaviour there.

We've taken a very similar strategy twice before with success - once at IR level, and once at the MI level (post ISEL).

Differential Revision: https://reviews.llvm.org/D66309

llvm-svn: 371441

5 years agoAMDGPU/GlobalISel: Legalize G_BUILD_VECTOR v2s16
Matt Arsenault [Mon, 9 Sep 2019 18:57:51 +0000 (18:57 +0000)]
AMDGPU/GlobalISel: Legalize G_BUILD_VECTOR v2s16

Handle it the same way as G_BUILD_VECTOR_TRUNC. Arguably only
G_BUILD_VECTOR_TRUNC should be legal for this, but G_BUILD_VECTOR will
probably be more convenient in most cases.

llvm-svn: 371440

5 years ago[TSan] Add interceptors for mach_vm_[de]allocate
Julian Lettner [Mon, 9 Sep 2019 18:57:32 +0000 (18:57 +0000)]
[TSan] Add interceptors for mach_vm_[de]allocate

I verified that the test is red without the interceptors.

rdar://40334350

Reviewed By: kubamracek, vitalybuka

Differential Revision: https://reviews.llvm.org/D66616

llvm-svn: 371439

5 years agoAMDGPU: Make VReg_1 size be 1
Matt Arsenault [Mon, 9 Sep 2019 18:43:29 +0000 (18:43 +0000)]
AMDGPU: Make VReg_1 size be 1

This was getting chosen as the preferred 32-bit register class based
on how TableGen selects subregister classes.

llvm-svn: 371438