platform/upstream/llvm.git
4 years ago[lldb][NFC] Delete static versions of ClangASTContext::CreateFunctionType
Raphael Isemann [Sun, 29 Dec 2019 19:10:29 +0000 (20:10 +0100)]
[lldb][NFC] Delete static versions of ClangASTContext::CreateFunctionType

We can always call the member function version of this function.

4 years ago[X86] Make the AVX1 check lines in vec-strict-inttofp-256.ll test 'avx' instead of...
Craig Topper [Sun, 29 Dec 2019 19:11:26 +0000 (11:11 -0800)]
[X86] Make the AVX1 check lines in vec-strict-inttofp-256.ll test 'avx' instead of 'avx2'. Add AVX2 checks. NFC

4 years ago[mlir] Update mlir/CMakeLists.txt to install *.td files
Kern Handa [Sun, 29 Dec 2019 17:04:09 +0000 (18:04 +0100)]
[mlir] Update mlir/CMakeLists.txt to install *.td files

Currently when you build the `install` target, TableGen files don't get
installed.

TableGen files are needed when authoring new MLIR dialects, but right
now they're missing when using the pre-built binaries.

Differential Revision: https://reviews.llvm.org/D71958

4 years ago[lldb][NFC] Remove most GetASTContext calls in AST metadata code
Raphael Isemann [Sat, 28 Dec 2019 22:37:53 +0000 (23:37 +0100)]
[lldb][NFC] Remove most GetASTContext calls in AST metadata code

4 years agoFix use of named values surrounded by newlines in clang-query
Stephen Kelly [Sun, 29 Dec 2019 14:51:22 +0000 (14:51 +0000)]
Fix use of named values surrounded by newlines in clang-query

4 years agoFix newline handling in clang-query parser
Stephen Kelly [Sun, 29 Dec 2019 14:48:47 +0000 (14:48 +0000)]
Fix newline handling in clang-query parser

Don't prematurely remove characters from the end of the string

4 years agoFix handling of newlines in clang-query
Stephen Kelly [Sun, 29 Dec 2019 14:38:33 +0000 (14:38 +0000)]
Fix handling of newlines in clang-query

Replace assert with diagnostic for missing newline.

4 years ago[Attributor] AAUndefinedBehavior: Check for branches on undef value.
Hideto Ueno [Sun, 29 Dec 2019 08:34:08 +0000 (17:34 +0900)]
[Attributor] AAUndefinedBehavior: Check for branches on undef value.

A branch is considered UB if it depends on an undefined / uninitialized value.
At this point this handles simple UB branches in the form: `br i1 undef, ...`
We query `AAValueSimplify` to get a value for the branch condition, so the branch
can be more complicated than just: `br i1 undef, ...`.

Patch By: Stefanos Baziotis (@baziotis)

Reviewers: jdoerfert, sstefan1, uenoku

Reviewed By: uenoku

Differential Revision: https://reviews.llvm.org/D71799

4 years ago[X86] Stop accidentally custom type legalizing v4i32->v4f32 on SSE1 only targets.
Craig Topper [Sun, 29 Dec 2019 07:09:48 +0000 (23:09 -0800)]
[X86] Stop accidentally custom type legalizing v4i32->v4f32 on SSE1 only targets.

We had a Custom operation action for v4i32 on SSE1. But since
v4i32 isn't legal until SSE2 this was not what was intended. The
code that get executed was intended for op legalization and
creates a bunch of v4i32 nodes that all end up scalarized.

4 years ago[LV] Use getMask() when printing recipe [NFCI]
Gil Rapaport [Sat, 28 Dec 2019 17:59:31 +0000 (19:59 +0200)]
[LV] Use getMask() when printing recipe [NFCI]

Use dedicated API for getting the mask instead of duplicating it.

Differential Revision: https://reviews.llvm.org/D71964

4 years ago[X86] Remove a redundant (scalar_to_vector (extract_vector_elt X))) in LowerUINT_TO_F...
Craig Topper [Sun, 29 Dec 2019 05:49:16 +0000 (21:49 -0800)]
[X86] Remove a redundant (scalar_to_vector (extract_vector_elt X))) in LowerUINT_TO_FP_i32. NFCI

4 years ago[X86] Fix -enable-machine-outliner for x86-32 after D48683
Fangrui Song [Sun, 29 Dec 2019 01:25:12 +0000 (17:25 -0800)]
[X86] Fix -enable-machine-outliner for x86-32 after D48683

D48683 accidentally disabled -enable-machine-outliner for x86-32.

4 years ago[mlir] Fix the wrong computation of dynamic strides for lowering AllocOp to LLVM
Tung Le Duc [Tue, 24 Dec 2019 07:02:10 +0000 (16:02 +0900)]
[mlir] Fix the wrong computation of dynamic strides for lowering AllocOp to LLVM

Leftover change from before the MLIR merge, reviewed at accepted at
https://github.com/tensorflow/mlir/pull/338.

4 years agoRevert "[COFF] Make the autogenerated .weak.<name>.default symbols static"
Martin Storsjö [Sat, 28 Dec 2019 21:38:41 +0000 (23:38 +0200)]
Revert "[COFF] Make the autogenerated .weak.<name>.default symbols static"

This reverts commit 7ca86ee6494d4307333b300bae80e42df4a5140f.

Apparently this change causes MS link.exe to error out with
"LNK1235: corrupt or invalid COFF symbol table".

4 years ago[lldb][NFC] Remove GetASTContext call in ClangPersistentVariables
Raphael Isemann [Sat, 28 Dec 2019 21:00:27 +0000 (22:00 +0100)]
[lldb][NFC] Remove GetASTContext call in ClangPersistentVariables

We try to build a CompilerType from the persistent decls so we need
a ClangASTContext. With this patch the ClangPersistentVariables store
the associated ClangASTContext of the persistent decls (which is
always the scratch ClangASTContext) and no longer call GetASTContext
to map back from clang::ASTContext to ClangASTContext.

4 years agoAllow redeclaration of __declspec(uuid)
Zachary Henkel [Sat, 28 Dec 2019 21:06:13 +0000 (13:06 -0800)]
Allow redeclaration of __declspec(uuid)

msvc allows a subsequent declaration of a uuid attribute on a
struct/class.  Mirror this behavior in clang-cl.

Reviewed By: rnk

Differential Revision: https://reviews.llvm.org/D71439

4 years ago[COFF] Make the autogenerated .weak.<name>.default symbols static
Martin Storsjö [Thu, 19 Dec 2019 08:01:40 +0000 (10:01 +0200)]
[COFF] Make the autogenerated .weak.<name>.default symbols static

If we have references to the same extern_weak in multiple objects,
all of them would generate external symbols with the same name. Make
them static to avoid duplicate definitions; nothing should need to
refer to this symbol outside of the current object.

GCC/binutils seems to handle the same by not using a fixed string
for the ".default" suffix, but instead using the name of some other
defined external symbol from the same object (which is supposed to
be unique among objects unless there's other duplicate definitions).

Differential Revision: https://reviews.llvm.org/D71711

4 years ago[CMake] Fix lld detection after D69685
Fangrui Song [Sat, 28 Dec 2019 00:01:03 +0000 (16:01 -0800)]
[CMake] Fix lld detection after D69685

D69685 actually broke lld detection for my build (probably due to CMake
processing order).

Before:

```
build projects/compiler-rt/lib/sanitizer_common/tests/Sanitizer-x86_64-Test-Nolibc: ... bin/clang || ...
```

After:

```
build projects/compiler-rt/lib/sanitizer_common/tests/Sanitizer-x86_64-Test-Nolibc: ... bin/clang bin/lld || ...
```

Differential Revision: https://reviews.llvm.org/D71950

4 years ago[X86] Add test cases for v4i64->v4f32 and v8i64->v8f32 strict_sint_to_fp/strict_uint_...
Craig Topper [Sat, 28 Dec 2019 19:17:49 +0000 (11:17 -0800)]
[X86] Add test cases for v4i64->v4f32 and v8i64->v8f32 strict_sint_to_fp/strict_uint_to_fp to vec-strict-inttofp-256.ll and vec-strict-inttofp-512.ll. NFC

4 years agoFix bots after a9ad65a2b34f
Nemanja Ivanovic [Sat, 28 Dec 2019 19:07:18 +0000 (13:07 -0600)]
Fix bots after a9ad65a2b34f

In the last commit, I neglected to initialize the new subtarget feature
I added which caused failures on a few bots. This should fix that.

4 years ago[PowerPC] Change default for unaligned FP access for older subtargets
Nemanja Ivanovic [Sat, 28 Dec 2019 17:20:36 +0000 (11:20 -0600)]
[PowerPC] Change default for unaligned FP access for older subtargets

This is a fix for https://bugs.llvm.org/show_bug.cgi?id=40554

Some CPU's trap to the kernel on unaligned floating point access and there are
kernels that do not handle the interrupt. The program then fails with a SIGBUS
according to the PR. This just switches the default for unaligned access to only
allow it on recent server CPUs that are known to allow this.

Differential revision: https://reviews.llvm.org/D71954

4 years agoSimplifyDemandedBits - Remove duplicate getOperand() call. NFC.
Simon Pilgrim [Sat, 28 Dec 2019 16:41:47 +0000 (16:41 +0000)]
SimplifyDemandedBits - Remove duplicate getOperand() call. NFC.

Pulled out from D56387 - cleanup variable names, move shift amount legalization inside if() of its only user and remove duplicate getOperand() call.

4 years agoFix crash in getFullyQualifiedName for inline namespace
Alexey Bader [Sat, 28 Dec 2019 13:13:33 +0000 (16:13 +0300)]
Fix crash in getFullyQualifiedName for inline namespace

Summary: The ICE happens when the most outer namespace is an inline namespace.

Reviewers: bkramer, ilya-biryukov

Reviewed By: ilya-biryukov

Subscribers: ebevhan, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D71962

4 years ago[lldb][NFC] Remove GetASTContext call in ClangDeclVendor
Raphael Isemann [Sat, 28 Dec 2019 13:35:07 +0000 (14:35 +0100)]
[lldb][NFC] Remove GetASTContext call in ClangDeclVendor

Instead of returning NamedDecls and then calling GetASTContext
to find back the ClangASTContext we used can just implement the
FindDecl variant that returns CompilerDecls (and implement the
other function by throwing away the ClangASTContext part of the
compiler decl).

4 years ago[PowerPC] Modify the hasSideEffects of some VSX instructions from 1 to 0
Kang Zhang [Sat, 28 Dec 2019 09:04:54 +0000 (09:04 +0000)]
[PowerPC] Modify the hasSideEffects of some VSX instructions from 1 to 0

Summary:
If we didn't set the value for hasSideEffects bit in our td file,  `llvm-tblgen`
will set it as true for those instructions which has no match pattern.
Below 6 instructions don't set the hasSideEffects flag and don't have match
pattern, so their hasSideEffects flag will be set true by llvm-tblgen.

But in fact below instructions don't modify any special register and don't have
other SideEffects, they shouldn't have SideEffects.
This patch is to modify the hasSideEffects of below instructions from 1 to 0.

```
VEXTUHLX
VEXTUHRX
VEXTUWLX
VEXTUWRX
VSPLTBs
VSPLTHs
```

Reviewed By: jsji

Differential Revision: https://reviews.llvm.org/D71391

4 years ago[TargetLowering] Update comment to reference the correct compiler-rt function the...
Craig Topper [Sat, 28 Dec 2019 02:33:16 +0000 (18:33 -0800)]
[TargetLowering] Update comment to reference the correct compiler-rt function the code is based on. NFC

4 years ago[mlir] Merge the successor operand count into BlockOperand.
River Riddle [Sat, 28 Dec 2019 04:33:53 +0000 (20:33 -0800)]
[mlir] Merge the successor operand count into BlockOperand.

Summary: The successor operand counts are directly tied to block operands anyways, and this simplifies the trailing objects of Operation(i.e. one less computation to perform).

Reviewed By: mehdi_amini

Differential Revision: https://reviews.llvm.org/D71949

4 years ago[ADT] Fix FoldingSet documentation typos
Brian Gesiak [Sat, 28 Dec 2019 02:21:05 +0000 (21:21 -0500)]
[ADT] Fix FoldingSet documentation typos

* "If found then M with be non-NULL" should be "will be non-NULL".
* The documentation examples (1) and (2) declare and use a variable
  `MyNode *M`, but examples (3) and (4) switch midway to using a
  variable named `N`. Unify the examples to all use `M`.
* The examples demonstrate the use of member functions of
  `FoldingSet`, but (3) and (4) invoke these as if they were free
  functions. Modify them to call member functions on the `MyFoldingSet`
  object constructed in the code above example (1).

4 years agoDelete setjmp_undefined_for_msvc workaround after llvm.setjmp was removed
Fangrui Song [Sat, 28 Dec 2019 02:07:03 +0000 (18:07 -0800)]
Delete setjmp_undefined_for_msvc workaround after llvm.setjmp was removed

4 years ago[Intrinsic] Delete tablegen rules of llvm.{sig,}{setjmp,longjmp}
Fangrui Song [Sat, 28 Dec 2019 01:53:16 +0000 (17:53 -0800)]
[Intrinsic] Delete tablegen rules of llvm.{sig,}{setjmp,longjmp}

4 years agolld: Remove explicit copy ops from AssociatedIterator, relying on implicit operators
David Blaikie [Sat, 28 Dec 2019 01:27:20 +0000 (17:27 -0800)]
lld: Remove explicit copy ops from AssociatedIterator, relying on implicit operators

4 years agoDebugInfo: Fix rangesBaseAddress DICompileUnit bitcode serialization/deserialization
David Blaikie [Sat, 28 Dec 2019 01:18:50 +0000 (17:18 -0800)]
DebugInfo: Fix rangesBaseAddress DICompileUnit bitcode serialization/deserialization

Follow-up to r346788 review feedback from Adrian Prantl.

4 years agoAMDGPU: Adjust test so it will work with GlobalISel
Matt Arsenault [Sat, 28 Dec 2019 00:35:00 +0000 (19:35 -0500)]
AMDGPU: Adjust test so it will work with GlobalISel

This is mostly a workaround for not handling the mubuf store path yet.

4 years ago[ELF] Improve the condition to create .interp
Fangrui Song [Fri, 27 Dec 2019 21:24:41 +0000 (13:24 -0800)]
[ELF] Improve the condition to create .interp

This restores commit 1417558e4a61794347c6bfbafaff7cd96985b2c3 and its follow-up, reverted by commit c3dbd782f1e0578c7ebc342f2e92f54d9644cff7.

After this commit:

clang -fuse-ld=bfd -no-pie -nostdlib a.c => .interp not created
clang -fuse-ld=bfd -pie -fPIE -nostdlib a.c => .interp created

clang -fuse-ld=gold -no-pie -nostdlib a.c => .interp not created
clang -fuse-ld=gold -pie -fPIE -nostdlib a.c => .interp created

clang -fuse-ld=lld -no-pie -nostdlib a.c => .interp created
clang -fuse-ld=lld -pie -fPIE -nostdlib a.c => .interp created

4 years ago[sanitizer] Link Sanitizer-x86_64-Test-Nolibc with -static
Fangrui Song [Fri, 27 Dec 2019 23:08:10 +0000 (15:08 -0800)]
[sanitizer] Link Sanitizer-x86_64-Test-Nolibc with -static

Pass -static so that clang will not pass -Wl,--dynamic-linker,... to the
linker. The test is not expected to run under a ld.so. (Technically it
works under a ld.so but glibc expects to see a PT_DYNAMIC. lld
intentionally does not follow GNU ld's complex rules regarding
PT_DYNAMIC.)

This allows commit 1417558e4a61794347c6bfbafaff7cd96985b2c3 to be
relanded.

4 years agoAMDGPU/GlobalISel: Use SReg_32 for readfirstlane constraining
Matt Arsenault [Fri, 27 Dec 2019 22:41:16 +0000 (17:41 -0500)]
AMDGPU/GlobalISel: Use SReg_32 for readfirstlane constraining

This matches the DAG behavior where we don't use SReg_32_XM0
everywhere anymore, and fixes not coalescing the copies into m0.

4 years agoHexagon: Fix missing tablegen mode comment
Matt Arsenault [Sat, 17 Aug 2019 21:35:47 +0000 (17:35 -0400)]
Hexagon: Fix missing tablegen mode comment

4 years agoTII: Fix using Register for a subregister index argument
Matt Arsenault [Fri, 27 Dec 2019 20:46:10 +0000 (15:46 -0500)]
TII: Fix using Register for a subregister index argument

4 years agoAMDGPU: Use Register
Matt Arsenault [Fri, 27 Dec 2019 20:21:44 +0000 (15:21 -0500)]
AMDGPU: Use Register

4 years ago[mlir][spirv] Add basic definitions for supporting availability
Lei Zhang [Fri, 27 Dec 2019 21:24:33 +0000 (16:24 -0500)]
[mlir][spirv] Add basic definitions for supporting availability

SPIR-V has a few mechanisms to control op availability: version,
extension, and capabilities. These mechanisms are considered as
different availability classes.

This commit introduces basic definitions for modelling SPIR-V
availability classes. Specifically, an `Availability` class is
added to SPIRVBase.td, along with two subclasses: MinVersion
and MaxVersion for versioning. SPV_Op is extended to take a
list of `Availability`. Each `Availability` instance carries
information for generating op interfaces for the corresponding
availability class and also the concrete availability
requirements.

With the availability spec on ops, we can now auto-generate the
op interfaces of all SPIR-V availability classes and also
synthesize the op's implementations of these interfaces. The
interface generation is done via new TableGen backends
-gen-avail-interface-{decls|defs}. The op's implementation is
done via -gen-spirv-avail-impls.

Differential Revision: https://reviews.llvm.org/D71930

4 years agoRevert "[ELF] Improve the condition to create .interp"
Reid Kleckner [Fri, 27 Dec 2019 21:02:23 +0000 (13:02 -0800)]
Revert "[ELF] Improve the condition to create .interp"

This reverts commit 1417558e4a61794347c6bfbafaff7cd96985b2c3.
Also reverts commit 019a92bb2832447092bb5c1faf9d03bb03b8c9c8.

This causes check-sanitizer to fail. The "-Nolib" variant of the test
crashes on startup in the loader.

4 years agoSupport powerpc and sparc when building without init_array.
Sterling Augustine [Fri, 27 Dec 2019 19:52:12 +0000 (11:52 -0800)]
Support powerpc and sparc when building without init_array.

Summary: Support powerpc and sparc when building without init_array.

Reviewers: rdhindsa, gribozavr

Subscribers: jyknight, nemanjai, fedor.sergeev, jsji, shchenz, steven.zhang, #sanitizers, llvm-commits

Tags: #sanitizers, #llvm

Differential Revision: https://reviews.llvm.org/D71944

4 years ago[PowerPC][LoopVectorize]Add floating point reg usage test
Jinsong Ji [Fri, 27 Dec 2019 19:29:13 +0000 (19:29 +0000)]
[PowerPC][LoopVectorize]Add floating point reg usage test

Copied two tests from x86 to test floating point reg usage.

4 years ago[mlir][spirv] Update docs regarding how to define new ops and types
Lei Zhang [Fri, 27 Dec 2019 20:28:45 +0000 (15:28 -0500)]
[mlir][spirv] Update docs regarding how to define new ops and types

This commit expands on the steps of defining a new SPIR-V op and
also provides pointers on how to define a new SPIR-V specific type.

Differential Revision: https://reviews.llvm.org/D71928

4 years agoRevert "CWG2352: Allow qualification conversions during reference binding."
David Blaikie [Fri, 27 Dec 2019 20:17:01 +0000 (12:17 -0800)]
Revert "CWG2352: Allow qualification conversions during reference binding."

This reverts commit de21704ba96fa80d3e9402f12c6505917a3885f4.

Regressed/causes this to error due to ambiguity:

  void f(const int * const &);
  void f(int *);
  int main() {
    int * x;
    f(x);
  }

(in case it's important - the original case where this turned up was a
member function overload in a class template with, essentially:

  f(const T1&)
  f(T2*)

(where T1 == X const *, T2 == X))

It's not super clear to me if this ^ is expected behavior, in which case
I'm sorry about the revert & happy to look into ways to fix the original
code.

4 years ago[mlir] Convert std.and/std.or ops to spv.LogicalAnd/spv.LogicalOr
MaheshRavishankar [Thu, 26 Dec 2019 00:29:17 +0000 (16:29 -0800)]
[mlir] Convert std.and/std.or ops to spv.LogicalAnd/spv.LogicalOr

The conversion from std.and/std.or to spv.LogicalAnd/spv.LogicalOr is
only valid for boolean (i1) types. Modify BinaryOpPattern in
StandardToSPIRV.td to allow limiting the type of the operands for
which the pattern is applied.

Differential Revision: https://reviews.llvm.org/D71881

4 years agoReland "[msan] Intercept qsort, qsort_r."
Reid Kleckner [Fri, 27 Dec 2019 19:29:00 +0000 (11:29 -0800)]
Reland "[msan] Intercept qsort, qsort_r."

This reverts commit 8fcce5ac73d49981656d9126e6c88391c1f6bf01.

I spoke too soon, the revert does not actually cause the startup crash
to go away.

4 years agoRevert "[msan] Intercept qsort, qsort_r."
Reid Kleckner [Fri, 27 Dec 2019 19:24:07 +0000 (11:24 -0800)]
Revert "[msan] Intercept qsort, qsort_r."

This reverts commit 7a9ebe95125ea87a494d0c18f44f10bd70e12188, and
dependent commit 54c522420347e58aa7bae1892cf5c5672b57c875, which
disables qsort interception for some iOS platforms.

After this change, the -Nolibc sanitizer common test binary crashes on
startup on my regular Linux workstation, as well as on our bots:
https://ci.chromium.org/p/chromium/builders/try/linux_upload_clang/740

 ********************
  Failing Tests (1):
       SanitizerCommon-Unit ::
       ./Sanitizer-x86_64-Test/SanitizerCommon.NolibcMain

Loading it up in gdb shows that it crashes during relocation processing,
which suggests that some glibc loader versions do not support the
THREADLOCAL data added in this interceptor.

4 years agoTailDuplication: Clear NoPHIs property
Matt Arsenault [Fri, 27 Dec 2019 16:26:24 +0000 (11:26 -0500)]
TailDuplication: Clear NoPHIs property

The early tail duplicator pass introduces new ones, so a MIR test that
infers no phis since there were none on the input would fail the
verifier after running.

4 years ago[VFS] Don't run symlink test on Windows, it may pass or fail
Reid Kleckner [Fri, 27 Dec 2019 18:37:07 +0000 (10:37 -0800)]
[VFS] Don't run symlink test on Windows, it may pass or fail

This test was XFAILed because of symlinks, but some versions of ln -s
seem to work on Windows, so the test was unexpectedly passing on our
bot:
http://lab.llvm.org:8011/builders/clang-x64-windows-msvc/builds/13233
Unexpected Passing Tests (1):
    Clang :: VFS/subframework-symlink.m

I don't know how or why, but it seems dependent on system configuration,
and is not something worth debugging. Avoid the problem by marking the
test UNSUPPORTED: system-windows instead of XFAIL: system-windows.

4 years ago[compiler-rt] [netbsd] Add support for versioned statvfs interceptors
Kamil Rytarowski [Fri, 27 Dec 2019 00:12:44 +0000 (01:12 +0100)]
[compiler-rt] [netbsd] Add support for versioned statvfs interceptors

Summary:
Add support for NetBSD 9.0 and newer versions of interceptors
operating on struct statvfs: fstatvfs, fstatvfs1, getmntinfo,
getvfsstat, statvfs, statvfs1.

The default promoted interceptors are for NetBSD 9.99.26. Older
ones (currently 9.0) are kept in a new NetBSD specific file:
/sanitizer_common_interceptors_netbsd_compat.inc. This file
defines compat interceptors and mangles `INIT_*` macros,
concatenating the current interceptors and the compat ones.
This redefinition is not elegant, but it avoids preprocessor madness.

Define struct_statvfs90_sz for the compat purposes.

Reviewers: mgorny, kcc, vitalybuka, joerg

Reviewed By: mgorny

Subscribers: dberris, llvm-commits, #sanitizers

Tags: #sanitizers, #llvm

Differential Revision: https://reviews.llvm.org/D71700

4 years ago[ConstantRange] Respect destination bitwidth for cast results.
Florian Hahn [Fri, 27 Dec 2019 17:38:18 +0000 (17:38 +0000)]
[ConstantRange] Respect destination bitwidth for cast results.

We returning a full set, we should use ResultBitWidth. Otherwise we might
it assertions when the resulting constant ranges are used later on.

Reviewers: nikic, spatel, reames

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D71937

4 years ago[OpenCL] Fixed printing of __private in AMDGPU test
Anastasia Stulova [Fri, 27 Dec 2019 17:07:48 +0000 (17:07 +0000)]
[OpenCL] Fixed printing of __private in AMDGPU test

Tags: #clang

4 years ago[OpenCL] Fix inconsistency between opencl and c11 atomic fetch max/min
Yaxun (Sam) Liu [Thu, 19 Dec 2019 20:27:02 +0000 (15:27 -0500)]
[OpenCL] Fix inconsistency between opencl and c11 atomic fetch max/min

There is some inconsistency between opencl and c11 atomic fetch max/min after

https://reviews.llvm.org/D46386

https://reviews.llvm.org/D55562

It is not reasonable to have such inconsistencies. This patch fixes that.

Differential Revision: https://reviews.llvm.org/D71725

4 years ago[Matrix] Propagate and use shape info for binary operators.
Florian Hahn [Fri, 27 Dec 2019 15:44:00 +0000 (15:44 +0000)]
[Matrix] Propagate and use shape info for binary operators.

This patch extends the current shape propagation and shape aware
lowering to also support binary operators. Those operators are uniform
with respect to their shape (shape of the input operands is the same as
the shape of their result).

Reviewers: anemet, Gerolf, reames, hfinkel, andrew.w.kaylor

Reviewed By: anemet

Differential Revision: https://reviews.llvm.org/D70898

4 years agoAllow newlines in AST Matchers in clang-query files
Stephen Kelly [Wed, 18 Dec 2019 22:35:46 +0000 (22:35 +0000)]
Allow newlines in AST Matchers in clang-query files

Reviewers: aaron.ballman

Subscribers: cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D71842

4 years ago[NFC][DA] Remove duplicate code in checkSrcSubscript and checkDstSubscript
Danilo Carvalho Grael [Fri, 27 Dec 2019 15:05:28 +0000 (10:05 -0500)]
[NFC][DA] Remove duplicate code in checkSrcSubscript and checkDstSubscript

Summary:
[DA] Move common code in checkSrcSubscript and checkDstSubscript to a
new function checkSubscript. This avoids duplicate code and possible
out of sync in the future.

Reviewers: sebpop, jmolloy, reames

Reviewed By: sebpop

Subscribers: bmahjour, hiraditya, llvm-commits, amehsan

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D71087

Patch by zhongduo.

4 years ago[OpenMP] Implementation of OMPT reduction callbacks
protze@itc.rwth-aachen.de [Fri, 27 Dec 2019 13:39:50 +0000 (14:39 +0100)]
[OpenMP] Implementation of OMPT reduction callbacks

Including two tests
These callbacks were added late to the 5.0 specification, an implementation is missing.

Reviewed By: jdoerfert

Differential Review: https://reviews.llvm.org/D70395

4 years ago[MLIR][spirv] Fix links in docs after repo migration
Lei Zhang [Fri, 27 Dec 2019 13:34:26 +0000 (08:34 -0500)]
[MLIR][spirv] Fix links in docs after repo migration

Summary:
This commit updates links to SPIR-V dialect code to LLVM monorepo
on GitHub. It also points to the operation doc on mlir.llvm.org.

Reviewers: mravishankar, denis13, ftynse

Reviewed By: ftynse

Subscribers: merge_guards_bot, mehdi_amini, rriddle, jpienaar, burmako, shauheen, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D71926

4 years agoAMDGPU/GlobalISel: Fix extra result register in fdiv64 lowering
Matt Arsenault [Wed, 25 Dec 2019 12:48:37 +0000 (07:48 -0500)]
AMDGPU/GlobalISel: Fix extra result register in fdiv64 lowering

There ended up being two result registers, which would fail on
select. It was really defing a new temp register in the correct def
position, instead of the correct result register.

4 years agoAMDGPU/GlobalISel: Select some 128-bit load/stores
Matt Arsenault [Wed, 25 Dec 2019 14:43:02 +0000 (09:43 -0500)]
AMDGPU/GlobalISel: Select some 128-bit load/stores

4 years agoAMDGPU: Use correct DebugLoc
Matt Arsenault [Fri, 27 Dec 2019 13:28:30 +0000 (08:28 -0500)]
AMDGPU: Use correct DebugLoc

4 years ago[OpenCL] Pretty print __private addr space
Anastasia Stulova [Fri, 27 Dec 2019 13:38:48 +0000 (13:38 +0000)]
[OpenCL] Pretty print __private addr space

Add printing of __private address space to TypePrinter to allow
it appears in diagnostics and AST dumps as all other language
addr spaces.

Tags: #clang

Differential Revision: https://reviews.llvm.org/D71272

4 years ago[lldb] Silent random xpass on aarch64-linux buildbot
Muhammad Omair Javaid [Fri, 27 Dec 2019 12:01:52 +0000 (17:01 +0500)]
[lldb] Silent random xpass on aarch64-linux buildbot

This patch adds skipif decorator to TestWatchLocationWithWatchSet.py.
Decorator will trigger for aarch64-linux as this test passes randomly
causing buildbot failure.

4 years ago[mlir] fix typo in a comment
wyzhao [Mon, 23 Dec 2019 08:11:29 +0000 (16:11 +0800)]
[mlir] fix typo in a comment

Trivial patch, reviewed and accepted on
https://github.com/tensorflow/mlir/pull/336 before MLIR merge.

4 years ago[MLIR] [NFC] fix unused var warning
Uday Bondhugula [Fri, 27 Dec 2019 11:04:09 +0000 (12:04 +0100)]
[MLIR] [NFC] fix unused var warning

Summary:
Fix this warning:
`
[69/106] Building CXX object tools/mlir/lib/Dialect/StandardOps/CMakeFiles/MLIRStandardOps.dir/Ops.cpp.o
/home/uday/llvm-project/mlir/lib/Dialect/StandardOps/Ops.cpp: In member function ‘virtual mlir::PatternMatchResult {anonymous}::ViewOpShapeFolder::matchAndRewrite(mlir::ViewOp, mlir::PatternRewriter&) const’:
/home/uday/llvm-project/mlir/lib/Dialect/StandardOps/Ops.cpp:2575:14: warning: variable ‘dynamicOffsetOperandCount’ set but not used [-Wunused-but-set-variable]
 2575 |     unsigned dynamicOffsetOperandCount = 0;
`

Reviewers: rriddle, mehdi_amini, ftynse

Reviewed By: ftynse

Subscribers: jpienaar, burmako, shauheen, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D71922

4 years ago[mlir] Floating constants for import-llvm
Alex Zinenko [Fri, 27 Dec 2019 09:54:01 +0000 (10:54 +0100)]
[mlir] Floating constants for import-llvm

Summary:
`mlir-translate -import-llvm test.ll`  was going into segmentation fault if `test.ll` had `float` or `double` constants.
For example,
```
%3 = fadd double 3.030000e+01, %0
```
Now, it is handled in `Importer::getConstantAsAttr` (similar behaviour as normal integers)
Added tests for FP arithmetic

Reviewers: ftynse, mehdi_amini

Reviewed By: ftynse, mehdi_amini

Subscribers: shauheen, mehdi_amini, rriddle, jpienaar, burmako, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D71912

4 years ago[clangd] Reformat `HoverTests.cpp` NFC
Ilya Biryukov [Fri, 27 Dec 2019 08:26:39 +0000 (09:26 +0100)]
[clangd] Reformat `HoverTests.cpp` NFC

I accidentally broke formatting in the previous revision.

4 years ago[X86] Allow v2i32->v2f32 strict and non-strict uint_to_fp to be widened to v4i32...
Craig Topper [Fri, 27 Dec 2019 08:15:45 +0000 (00:15 -0800)]
[X86] Allow v2i32->v2f32 strict and non-strict uint_to_fp to be widened to v4i32->v4f32 under avx512.

With avx512vl we get v4i32->v4f32 uint_to_fp instructions. With
avx512f we get v16i32->v16f32 instructions which we can use to
emulate v4i32->v4f32.

4 years ago[X86] Add v2i32->v2f32 non-strict sint_to_fp/uint_to_fp tests. NFC
Craig Topper [Fri, 27 Dec 2019 08:13:30 +0000 (00:13 -0800)]
[X86] Add v2i32->v2f32 non-strict sint_to_fp/uint_to_fp tests. NFC

4 years ago[X86] Custom widen v2i32->v2f32 strict_sint_to_fp to avoid scalarization.
Craig Topper [Fri, 27 Dec 2019 07:57:33 +0000 (23:57 -0800)]
[X86] Custom widen v2i32->v2f32 strict_sint_to_fp to avoid scalarization.

4 years ago[llvm-readobj][llvm-objdump][test] - Improve dynamic section testing.
Georgii Rymar [Thu, 26 Dec 2019 11:01:05 +0000 (14:01 +0300)]
[llvm-readobj][llvm-objdump][test] - Improve dynamic section testing.

This adds --strict-whitespace --match-full-lines flags to
improve the testing and reveal formatting issues we have.

Differential revision: https://reviews.llvm.org/D71895

4 years ago[clangd] Fix crash in hover
Ilya Biryukov [Fri, 27 Dec 2019 08:11:03 +0000 (09:11 +0100)]
[clangd] Fix crash in hover

4 years agoDelete llvm.{sig,}{setjmp,longjmp} remnant after r136821
Fangrui Song [Fri, 27 Dec 2019 07:32:53 +0000 (23:32 -0800)]
Delete llvm.{sig,}{setjmp,longjmp} remnant after r136821

  Intrinsic has incorrect argument type!
  i32 (i32*)* @llvm.setjmp

*wipes tear*

4 years ago[X86] Custom widen 128/256-bit vXi32 fp_to_uint on avx512f targets without avx512vl...
Craig Topper [Fri, 27 Dec 2019 05:46:29 +0000 (21:46 -0800)]
[X86] Custom widen 128/256-bit vXi32 fp_to_uint on avx512f targets without avx512vl. Similar for vXi64 on avx512dq without avx512vl.

Summary:
Previously we did this with isel patterns that used garbage in
the widened part of the source. But that's not valid for strictfp.
So now we custom widen and use zeroes for the widened elemens for
strictfp.

This replaces D71864.

Reviewers: RKSimon, spatel, andrew.w.kaylor, pengfei, LiuChen3

Reviewed By: pengfei

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D71879

4 years ago[X86] Custom widen strict v2f32->v2i32 by padding with zeroes.
Craig Topper [Fri, 27 Dec 2019 05:25:22 +0000 (21:25 -0800)]
[X86] Custom widen strict v2f32->v2i32 by padding with zeroes.

For non-strict, generic type legalization will take care of this,
but that doesn't happen currently for strict nodes.

4 years ago[X86] Fix -Wmisleading-indentation after D71892
Fangrui Song [Fri, 27 Dec 2019 05:40:48 +0000 (21:40 -0800)]
[X86] Fix -Wmisleading-indentation after D71892

4 years ago[X86][FPEnv] Promote some float strictfp operations to double on i686-pc-windows...
Craig Topper [Fri, 27 Dec 2019 04:21:23 +0000 (20:21 -0800)]
[X86][FPEnv] Promote some float strictfp operations to double on i686-pc-windows-msvc to match what we do for non-strict.

The float libcalls are inlined in MSVC's math header where they
just cast to double and use the double libcall. Do the same when
we emit libcalls.

4 years ago[X86] Add tests for constrained float intrinsics on i686-pc-windows-msvc. NFC
Craig Topper [Fri, 27 Dec 2019 04:09:32 +0000 (20:09 -0800)]
[X86] Add tests for constrained float intrinsics on i686-pc-windows-msvc. NFC

We need to promote these to double due to missing libcalls on
Windows.

4 years ago[X86] Add custom legalization for strict_uint_to_fp v2i32->v2f32.
Craig Topper [Fri, 27 Dec 2019 01:27:44 +0000 (17:27 -0800)]
[X86] Add custom legalization for strict_uint_to_fp v2i32->v2f32.

I believe the algorithm we use for non-strict is exception safe
for strict. The fsub won't generate any exceptions. After it we
will have an exact version of the i32 integer in a double. Then
we just round it to f32. That rounding will generate a precision
exception if it can't be represented exactly.

4 years ago[X86] Add test cases for v2i32->v2f32 strict_sint_to_fp/strict_uint_to_fp. NFC
Craig Topper [Fri, 27 Dec 2019 00:41:42 +0000 (16:41 -0800)]
[X86] Add test cases for v2i32->v2f32 strict_sint_to_fp/strict_uint_to_fp. NFC

4 years agoAdd an __attribute__((unused)) to populateWithGenerated since it might
Eric Christopher [Fri, 27 Dec 2019 02:48:59 +0000 (18:48 -0800)]
Add an __attribute__((unused)) to populateWithGenerated since it might
not be used where defined and is autogenerated.

4 years agoTemporarily Revert "[compiler-rt] [netbsd] Add support for versioned statvfs intercep...
Eric Christopher [Fri, 27 Dec 2019 02:40:20 +0000 (18:40 -0800)]
Temporarily Revert "[compiler-rt] [netbsd] Add support for versioned statvfs interceptors"
as it's failing the netbsd specific linter parts of the sanitizer linter:

llvm-project/compiler-rt/lib/sanitizer_common/sanitizer_common_interceptors_netbsd_compat.inc:23:  Lines should be <= 80 characters long  [whitespace/line_length]
llvm-project/compiler-rt/lib/sanitizer_common/sanitizer_platform_limits_netbsd.cpp:2450:  Do not use variable-length arrays.  Use an appropriately named ('k' followed by CamelCase) compile-time constant for the size.

This reverts commit 78f714f824fac8aa3fdd85908c41538bccefb959.

4 years agoFix a -Wcovered-switch-default warning by moving the unreachable out of the
Eric Christopher [Fri, 27 Dec 2019 02:34:41 +0000 (18:34 -0800)]
Fix a -Wcovered-switch-default warning by moving the unreachable out of the
covered switch.

4 years agoRemove an unused static function.
Eric Christopher [Fri, 27 Dec 2019 02:34:14 +0000 (18:34 -0800)]
Remove an unused static function.

4 years agoFix a -Wcovered-switch-default warning by moving the unreachable out of the
Eric Christopher [Fri, 27 Dec 2019 02:29:54 +0000 (18:29 -0800)]
Fix a -Wcovered-switch-default warning by moving the unreachable out of the
covered switch.

4 years agoRemove unused static function.
Eric Christopher [Fri, 27 Dec 2019 02:19:48 +0000 (18:19 -0800)]
Remove unused static function.

4 years agoRevert "Allow newlines in AST Matchers in clang-query files" + 1
Evgenii Stepanov [Fri, 27 Dec 2019 02:05:25 +0000 (18:05 -0800)]
Revert "Allow newlines in AST Matchers in clang-query files" + 1

Revert "Fix -Wunused-lambda-capture warnings."
This reverts commit 2369560f4a7720b19edfbf9de14ef061307ff773.
This reverts commit 522ee29a4fb3814db604b585c8637247477ec057.

clang/lib/ASTMatchers/Dynamic/Parser.cpp:610:13: warning: implicit conversion turns string literal into bool: 'const char [35]' to 'bool' [-Wstring-conversion]
    assert(!"Newline should never be found here");

4 years agoadd custom operation for strict fpextend/fpround
Liu, Chen3 [Thu, 26 Dec 2019 09:47:37 +0000 (17:47 +0800)]
add custom operation for strict fpextend/fpround

Differential Revision: https://reviews.llvm.org/D71892

4 years ago[compiler-rt] [netbsd] Add support for versioned statvfs interceptors
Kamil Rytarowski [Fri, 27 Dec 2019 00:12:44 +0000 (01:12 +0100)]
[compiler-rt] [netbsd] Add support for versioned statvfs interceptors

Summary:
Add support for NetBSD 9.0 and newer versions of interceptors
operating on struct statvfs: fstatvfs, fstatvfs1, getmntinfo,
getvfsstat, statvfs, statvfs1.

The default promoted interceptors are for NetBSD 9.99.26. Older
ones (currently 9.0) are kept in a new NetBSD specific file:
/sanitizer_common_interceptors_netbsd_compat.inc. This file
defines compat interceptors and mangles `INIT_*` macros,
concatenating the current interceptors and the compat ones.
This redefinition is not elegant, but it avoids preprocessor madness.

Define struct_statvfs90_sz for the compat purposes.

Reviewers: mgorny, kcc, vitalybuka, joerg

Reviewed By: mgorny

Subscribers: dberris, llvm-commits, #sanitizers

Tags: #sanitizers, #llvm

Differential Revision: https://reviews.llvm.org/D71700

4 years agoRemove SrcVT only used in an assert and propagate query.
Eric Christopher [Thu, 26 Dec 2019 23:28:32 +0000 (15:28 -0800)]
Remove SrcVT only used in an assert and propagate query.

4 years agoFix -Wunused-lambda-capture warnings.
Eric Christopher [Thu, 26 Dec 2019 23:27:21 +0000 (15:27 -0800)]
Fix -Wunused-lambda-capture warnings.

4 years agoMake lazyload_metadata.ll resilient to the addition of new metadata kinds
David Herzka [Thu, 26 Dec 2019 18:56:07 +0000 (13:56 -0500)]
Make lazyload_metadata.ll resilient to the addition of new metadata kinds

Summary: The specific number of records loaded depends on the number of kinds, but the difference between the lazy and not lazy cases does not.

Reviewers: modocache

Subscribers: llvm-commits, dexonsmith, steven_wu, hiraditya, mehdi_amini

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D71882

4 years ago[X86] Custom widen 128/256-bit vXi32 uint_to_fp on avx512f targets without avx512vl...
Craig Topper [Thu, 26 Dec 2019 22:43:33 +0000 (14:43 -0800)]
[X86] Custom widen 128/256-bit vXi32 uint_to_fp on avx512f targets without avx512vl. Similar for vXi64 sint_to_fp/uint_to_fp on avx512dq without avx512vl.

Previously we widened these through isel patterns, but that
didn't work for STRICT_ nodes. Those need to be padded with
zeroes in the upper bits which is harder to do in isel patterns.

4 years agoRevert "Make lazyload_metadata.ll resilient to the addition of new metadata kinds"
Reid Kleckner [Thu, 26 Dec 2019 22:29:11 +0000 (14:29 -0800)]
Revert "Make lazyload_metadata.ll resilient to the addition of new metadata kinds"

This reverts commit be4704bd41a4dd8bb5c4dd5a614744c69fb3cf8e.

This test fails on Windows without awk.

4 years ago[ELF][test] Fix dynamic-linker.s
Fangrui Song [Thu, 26 Dec 2019 21:44:31 +0000 (13:44 -0800)]
[ELF][test] Fix dynamic-linker.s

4 years agoAdd a clang-tidy configuration file for MLIR, it is using camelBack for naming at...
Mehdi Amini [Thu, 26 Dec 2019 21:41:25 +0000 (21:41 +0000)]
Add a clang-tidy configuration file for MLIR, it is using camelBack for naming at the moment

4 years ago[X86] Add custom widening for v2i32->v2f64 strict_uint_to_fp with AVX512F, but not...
Craig Topper [Thu, 26 Dec 2019 21:18:24 +0000 (13:18 -0800)]
[X86] Add custom widening for v2i32->v2f64 strict_uint_to_fp with AVX512F, but not AVX512VL.

Previously we were widening with isel patterns, but that wasn't
exception safe for strict FP. So now we widen to v4i32->v4f64
during type legalization. And then let op legalization further
widen to v8i32->v8f64.

The vec_int_to_fp.ll changes are caused by us no longer narrowing
extracts of strict_uint_to_fp to the v4i32->v2f64 instruction
without AVX512VL only to have isel rewiden it. Now we just keep
it wide throughout. So we don't have an opportunity to narrow
the load.

4 years ago[ELF] Improve the condition to create .interp
Fangrui Song [Mon, 3 Jun 2019 05:25:03 +0000 (05:25 +0000)]
[ELF] Improve the condition to create .interp

Similar to rL362355, but with the `!config->shared` guard.

(1) {gcc,clang} -fuse-ld=bfd -pie -fPIE -nostdlib a.c => .interp created
(2) {gcc,clang} -fuse-ld=lld -pie -fPIE -nostdlib a.c => .interp not created
(3) {gcc,clang} -fuse-ld=lld -pie -fPIE -nostdlib a.c a.so => .interp created

The inconsistency of (2) is due to the condition `!Config->SharedFiles.empty()`.
To make lld behave more like ld.bfd, we could change the condition to:

    config->hasDynSymTab && !config->dynamicLinker.empty() && script->needsInterpSection();

However, that would bring another inconsistency as can be observed with:

(4) {gcc,clang} -fuse-ld=bfd -no-pie -nostdlib a.c => .interp not created

4 years ago[X86] Add custom widening for v2f64->v2i32 strict_fp_to_uint with avx512f, but not...
Craig Topper [Thu, 26 Dec 2019 20:26:39 +0000 (12:26 -0800)]
[X86] Add custom widening for v2f64->v2i32 strict_fp_to_uint with avx512f, but not avx512vl.

AVX512F added instruction for vector fp_to_uint conversions. With
AVX512VL we can use a specific instruction that does v2f64->v4i32 with
zeroes in the 2 extra elements. For non-strict nodes without AVX512VL
we relied on type legalization to turn it to v4f64->v4i32 which would
later be widened by op legalization to v8f64->v8i32. But type legalization
doesn't currently widen strict nodes since it doesn't know how to
safely and efficiently pad the extra elements. But for X86 we know
padding with zeroes is safe and efficient so do that ourselves.