platform/upstream/llvm.git
4 years ago[Support] Simplify and optimize ThreadPool
Fangrui Song [Sat, 25 Apr 2020 03:50:06 +0000 (20:50 -0700)]
[Support] Simplify and optimize ThreadPool

* Merge QueueLock and CompletionLock.
* Avoid spurious CompletionCondition.notify_all() when ActiveThreads is greater than 0.
* Use default member initializers.

Reviewed By: mehdi_amini

Differential Revision: https://reviews.llvm.org/D78856

4 years ago[ELF][PPC64] Don't perform toc-indirect to toc-relative relaxation for R_PPC64_TOC16_...
Fangrui Song [Sat, 18 Apr 2020 06:08:11 +0000 (23:08 -0700)]
[ELF][PPC64] Don't perform toc-indirect to toc-relative relaxation for R_PPC64_TOC16_HA not followed by R_PPC64_TOC16_LO_DS

The current implementation assumes that R_PPC64_TOC16_HA is always followed
by R_PPC64_TOC16_LO_DS. This can break with:

// Load the address of the TOC entry, instead of the value stored at that address
  addis 3, 2, .LC0@tloc@ha  # R_PPC64_TOC16_HA
  addi  3, 3, .LC0@tloc@l   # R_PPC64_TOC16_LO
  blr

which is used by boringssl's util/fipstools/delocate/delocate.go
https://github.com/google/boringssl/blob/master/crypto/fipsmodule/FIPS.md has some documentation.
In short, this tool converts an assembly file to avoid any potential relocations.
The distance to an input .toc is not a constant after linking, so the assembly cannot use an `addis;ld` pair.
Instead, delocate changes the code to jump to a stub (`addis;addi`) which loads the TOC entry address.

Reviewed By: sfertile

Differential Revision: https://reviews.llvm.org/D78431

4 years ago[PowerPC][AIX] Pass ByVal formal args that span registers and stack.
Sean Fertile [Wed, 15 Apr 2020 13:58:22 +0000 (09:58 -0400)]
[PowerPC][AIX] Pass ByVal formal args that span registers and stack.

Implement passing of ByVal formal arguments when the argument is passed
partly in the argument registers, with the remainder of the argument
passed on the stack.

Differential Revision: https://reviews.llvm.org/D78515

4 years ago[Windows] Autolink with basenames and add libdir to libpath
Reid Kleckner [Wed, 31 Jul 2019 21:52:00 +0000 (14:52 -0700)]
[Windows] Autolink with basenames and add libdir to libpath

Prior to this change, for a few compiler-rt libraries such as ubsan and
the profile library, Clang would embed "-defaultlib:path/to/rt-arch.lib"
into the .drective section of every object compiled with
-finstr-profile-generate or -fsanitize=ubsan as appropriate.

These paths assume that the link step will run from the same working
directory as the compile step. There is also evidence that sometimes the
paths become absolute, such as when clang is run from a different drive
letter from the current working directory. This is fragile, and I'd like
to get away from having paths embedded in the object if possible. Long
ago it was suggested that we use this for ASan, and apparently I felt
the same way back then:
https://reviews.llvm.org/D4428#56536

This is also consistent with how all other autolinking usage works for
PS4, Mac, and Windows: they all use basenames, not paths.

To keep things working for people using the standard GCC driver
workflow, the driver now adds the resource directory to the linker
library search path when it calls the linker. This is enough to make
check-ubsan pass, and seems like a generally good thing.

Users that invoke the linker directly (most clang-cl users) will have to
add clang's resource library directory to their linker search path in
their build system. I'm not sure where I can document this. Ideally I'd
also do it in the MSBuild files, but I can't figure out where they go.
I'd like to start with this for now.

Reviewed By: hans

Differential Revision: https://reviews.llvm.org/D65543

4 years ago[X86] Update costs for truncates from less than 128-bit vectors to vXi1 on pre-avx512...
Craig Topper [Tue, 28 Apr 2020 17:50:59 +0000 (10:50 -0700)]
[X86] Update costs for truncates from less than 128-bit vectors to vXi1 on pre-avx512 targets

vXi1 types are legalized by promoting, but the narrow vectors
are legalized by widening. This results in some truncates turning
into any_extends.

4 years ago[AArch64][GlobalISel] Select immediate forms of compares by wiggling constants
Jessica Paquette [Wed, 22 Apr 2020 23:43:31 +0000 (16:43 -0700)]
[AArch64][GlobalISel] Select immediate forms of compares by wiggling constants

Similar to code in `getAArch64Cmp` in AArch64ISelLowering.

When we get a compare against a constant, sometimes, that constant isn't valid
for selecting an immediate form.

However, sometimes, you can get a valid constant by adding 1 or subtracting 1,
and updating the condition code.

This implements the following transformations when valid:

- x slt c => x sle c - 1
- x sge c => x sgt c - 1
- x ult c => x ule c - 1
- x uge c => x ugt c - 1

- x sle c => x slt c + 1
- x sgt c => s sge c + 1
- x ule c => x ult c + 1
- x ugt c => s uge c + 1

Valid meaning the constant doesn't wrap around when we fudge it, and the result
gives us a compare which can be selected into an immediate form.

This also moves `getImmedFromMO` higher up in the file so we can use it.

Differential Revision: https://reviews.llvm.org/D78769

4 years agoRevert "[lld-macho] Add basic symbol table output"
Shoaib Meenai [Tue, 28 Apr 2020 18:31:39 +0000 (11:31 -0700)]
Revert "[lld-macho] Add basic symbol table output"

This reverts commit a3d95a50ee3397c89327fce6983624446cd8e9d7.

Reverting due to UBSan failures:
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fast/builds/40817/steps/check-lld%20ubsan/logs/stdio

4 years agoRevert "[lld-macho] Add support for emitting dylibs with a single symbol"
Shoaib Meenai [Tue, 28 Apr 2020 18:30:57 +0000 (11:30 -0700)]
Revert "[lld-macho] Add support for emitting dylibs with a single symbol"

This reverts commit 9598778bd1910e77ccd399f2c9e979c8ecf98e55.

Reverting due to UBSan failures:
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fast/builds/40817/steps/check-lld%20ubsan/logs/stdio

4 years ago[MachO] Fix UB in memcpy
Shoaib Meenai [Tue, 28 Apr 2020 18:29:30 +0000 (11:29 -0700)]
[MachO] Fix UB in memcpy

UBSan complains about a memcpy with a null pointer, so just skip the
memcpy call if the data is empty.

4 years agoMustBeExecutedContextPrinter::runOnModule: Use unique_ptr to simplify/clarify ownership
David Blaikie [Tue, 28 Apr 2020 18:30:22 +0000 (11:30 -0700)]
MustBeExecutedContextPrinter::runOnModule: Use unique_ptr to simplify/clarify ownership

4 years agoRevert "Recommit [libc++] Move abs and div into stdlib.h to fix header cycle."
Raphael Isemann [Tue, 28 Apr 2020 18:23:01 +0000 (20:23 +0200)]
Revert "Recommit [libc++] Move abs and div into stdlib.h to fix header cycle."

It seems that D74892 still hasn't fixed the issue on the bot. Currently
investigating the bot breakage and meanwhile (again) reverting this...

4 years ago[SVE] Update EmitSVEPredicateCast to take a ScalableVectorType
Christopher Tetreault [Tue, 28 Apr 2020 17:32:55 +0000 (10:32 -0700)]
[SVE] Update EmitSVEPredicateCast to take a ScalableVectorType

Summary:
Removes usage of VectorType::getNumElements identified by test located
at CodeGen/aarch64-sve-intrinsics/acle_sve_abs.c. Since the type is an
SVE predicate vector, it makes sense to specialize the code for scalable
vectors only.

Reviewers: rengolin, efriedma

Reviewed By: efriedma

Subscribers: tschuett, kristof.beyls, rkruppe, psnobl, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D78958

4 years ago[FileCollector] Lock Mutex in copyFiles
Jan Korous [Tue, 28 Apr 2020 17:43:17 +0000 (10:43 -0700)]
[FileCollector] Lock Mutex in copyFiles

We should synchronize reading of VFSWriter's data with the rest of the methods.

Differential revision: https://reviews.llvm.org/D78956

4 years ago[X86] Handle more cases in combineAddOrSubToADCOrSBB.
Craig Topper [Tue, 28 Apr 2020 17:24:44 +0000 (10:24 -0700)]
[X86] Handle more cases in combineAddOrSubToADCOrSBB.

This adds support for

X + SETAE --> sbb X, -1
X - SETAE --> adc X, -1

Fixes PR45700

Differential Revision: https://reviews.llvm.org/D78984

4 years ago[X86] Pre-commit tests for D78984. NFC
Craig Topper [Tue, 28 Apr 2020 17:08:49 +0000 (10:08 -0700)]
[X86] Pre-commit tests for D78984. NFC

These tests show some missed opportunities to use sbb/adc.

4 years ago[X86][CostModel] Correct the costs for truncate to a mask register with avx512
Craig Topper [Tue, 28 Apr 2020 17:06:29 +0000 (10:06 -0700)]
[X86][CostModel] Correct the costs for truncate to a mask register with avx512

I've modified isTruncateFree to get an accurate cost for types that need to be split. I'm planning to look into fixing it for all vectors, but need more cost cleanups first.

Differential Revision: https://reviews.llvm.org/D78973

4 years ago[COFF] Add a fastpath for /INCLUDE: in .drective sections
Reid Kleckner [Sat, 25 Apr 2020 00:26:17 +0000 (17:26 -0700)]
[COFF] Add a fastpath for /INCLUDE: in .drective sections

This speeds up linking chrome.dll with PGO instrumentation by 13%
(154271ms -> 134033ms).

LLVM's Option library is very slow. In particular, it allocates at least
one large-ish heap object (Arg) for every argument. When PGO
instrumentation is enabled, all the __profd_* symbols are added to the
@llvm.used list, which compiles down to these /INCLUDE: directives. This
means we have O(#symbols) directives to parse in the section, so we end
up allocating an Arg for every function symbol in the object file. This
is unnecessary.

To address the issue and speed up the link, extend the fast path that we
already have for /EXPORT:, which has similar scaling issues.

I promise that I took a hard look at optimizing the Option library, but
its data structures are very general and would need a lot of cleanup. We
have accumulated lots of optional features (option groups, aliases,
multiple values) over the years, and these are now properties of every
parsed argument, when the vast majority of arguments do not use these
features.

Reviewed By: thakis

Differential Revision: https://reviews.llvm.org/D78845

4 years ago[libc++] Avoid using distutils.util in the DSL
Louis Dionne [Tue, 28 Apr 2020 17:32:05 +0000 (13:32 -0400)]
[libc++] Avoid using distutils.util in the DSL

Some bots apparently don't have that package.

4 years agobuild: use Python3 for compiler-rt
Saleem Abdulrasool [Tue, 28 Apr 2020 17:25:23 +0000 (17:25 +0000)]
build: use Python3 for compiler-rt

compiler-rt is built in a unified configuration on some of the builders
which requires that this is updated to follow the same pattern as LLVM.

4 years ago[llvm/DebugInfo] Fix invalid verifier error for DWARF5
Jonas Devlieghere [Tue, 28 Apr 2020 17:24:40 +0000 (10:24 -0700)]
[llvm/DebugInfo] Fix invalid verifier error for DWARF5

We unconditionally compared the DW_AT_ranges offset to the length of the
.debug_ranges section. For DWARF5 we should look at the debug_rnglists
section instead.

Differential revision: https://reviews.llvm.org/D78971

4 years ago[gn build] (manually) merge c4c3883
Reid Kleckner [Tue, 28 Apr 2020 17:20:30 +0000 (10:20 -0700)]
[gn build] (manually) merge c4c3883

PYTHON_EXECUTABLE changed to Python3_EXECUTABLE in the lit test suite.

Committing without review, since the gn build is currently broken. Going
forward, more gn build changes may be necessary to get gn to find Python
3 instead of Python 2.

4 years ago[clangd] Fix remote index build without shared libs mode
Kirill Bobyrev [Tue, 28 Apr 2020 17:15:49 +0000 (19:15 +0200)]
[clangd] Fix remote index build without shared libs mode

Summary:
Generated Protobuf library has to be in CLANG_EXPORTS and should also be
installed appropriately. The easiest way to do that is via CMake's
add_clang_library. That unfortunately applies "one directory - one
clang_(library|tool)" policy so .proto files should be in a separate directory
and complicates the layout.

This setup works both in shared and static libs mode.

Resolves: https://github.com/clangd/clangd/issues/351

Reviewers: sammccall

Reviewed By: sammccall

Subscribers: mgorny, ilya-biryukov, MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits, llvm-commits

Tags: #clang, #llvm

Differential Revision: https://reviews.llvm.org/D78885

4 years ago[libc++] Temporarily disable test that fails on Windows
Louis Dionne [Tue, 28 Apr 2020 17:09:21 +0000 (13:09 -0400)]
[libc++] Temporarily disable test that fails on Windows

4 years agoFix .arclint on Windows
Reid Kleckner [Sat, 25 Apr 2020 00:51:15 +0000 (17:51 -0700)]
Fix .arclint on Windows

Run "bash myscript.sh". This will work if bash is on PATH, which it
probably is, since developers have to use git.

Reviewed By: scott.linder

Differential Revision: https://reviews.llvm.org/D78846

4 years ago[AST] Fix a crash on a dependent vector_size attribute
Erik Pilkington [Tue, 28 Apr 2020 16:24:54 +0000 (12:24 -0400)]
[AST] Fix a crash on a dependent vector_size attribute

Looks like this was just a copy & paste mistake from
getDependentSizedExtVectorType. rdar://60092165

Differential revision: https://reviews.llvm.org/D79012

4 years ago[ELF] Clear lazyObjFiles in lld::elf::link after D46034
Fangrui Song [Tue, 28 Apr 2020 16:49:00 +0000 (09:49 -0700)]
[ELF] Clear lazyObjFiles in lld::elf::link after D46034

4 years ago[SystemZ] Avoid __INTPTR_TYPE__ conversions in vecintrin.h
Ulrich Weigand [Tue, 28 Apr 2020 16:46:37 +0000 (18:46 +0200)]
[SystemZ] Avoid __INTPTR_TYPE__ conversions in vecintrin.h

Some intrinsics in vecintrin.h are currently implemented by
performing address arithmetic in __INTPTR_TYPE__ and converting
the result to some pointer type.  While this works correctly,
it leads to suboptimal code generation since many optimizers
cannot trace the provenance of the resulting pointers.

Fixed by using "char *" pointer arithmetic instead.

4 years ago[SystemZ] Use reserved keywords in vecintrin.h
Ulrich Weigand [Tue, 28 Apr 2020 16:24:19 +0000 (18:24 +0200)]
[SystemZ] Use reserved keywords in vecintrin.h

System headers should avoid using the "vector" and "bool" keywords
since those might be redefined by user code.  For example, using
<stdbool.h> before <vecintrin.h> will currently lead to compiler
errors.

Fixed by using the reserved "__vector" and "__bool" keywords
instead.  NFC otherwise.

4 years agoRevert "[clang-format] C# property formatting can be controlled by config options"
Jonathan Coe [Tue, 28 Apr 2020 16:41:17 +0000 (17:41 +0100)]
Revert "[clang-format] C# property formatting can be controlled by config options"

Committed in error without approval https://reviews.llvm.org/D79000

This reverts commit 015bca3e67cbb88f74f01fb5ae4e46392bec6416.

4 years ago[clang-format] C# property formatting can be controlled by config options
Jonathan Coe [Tue, 28 Apr 2020 16:35:33 +0000 (17:35 +0100)]
[clang-format] C# property formatting can be controlled by config options

Summary:
Allow brace wrapping in C# property accessors to be controlled by configuration options.

Add new tests and revert old test results for MS style to their old state (as intended).

`FormatStyle.BraceWrapping.AfterFunction = true;` will change automatic property formatting from

```
Type MyType { get; set }
```

to

```
Type MyType
{ get; set }
```

Reviewers: krasimir, MyDeveloperDay

Subscribers: cfe-commits

Tags: #clang-format, #clang

Differential Revision: https://reviews.llvm.org/D79000

4 years ago[clang-format] insert space after C# keyword var in var (key, value)
Jonathan Coe [Tue, 28 Apr 2020 16:26:51 +0000 (17:26 +0100)]
[clang-format] insert space after C# keyword var in var (key, value)

Reviewers: krasimir, MyDeveloperDay

Reviewed By: MyDeveloperDay

Subscribers: cfe-commits

Tags: #clang-format, #clang

Differential Revision: https://reviews.llvm.org/D79008

4 years ago[mlir][assemblyFormat] Fix bug when using AttrSizedOperandSegments trait with only...
Martin Erhart [Tue, 28 Apr 2020 16:25:04 +0000 (18:25 +0200)]
[mlir][assemblyFormat] Fix bug when using AttrSizedOperandSegments trait with only non-buildable operand types

Summary:
When creating an operation with
* `AttrSizedOperandSegments` trait
* Variadic operands of only non-buildable types
* assemblyFormat to automatically generate the parser
the `builder` local variable is used, but never declared.
This adds a fix as well as a test for this case as existing ones use buildable types only.

Reviewers: rriddle, Kayjukh, grosser

Reviewed By: Kayjukh

Subscribers: mehdi_amini, jpienaar, shauheen, antiagainst, nicolasvasilache, arpith-jacob, mgester, lucyrfox, liufengdb, Joonsoo, grosul1, frgossen, llvm-commits

Tags: #mlir, #llvm

Differential Revision: https://reviews.llvm.org/D79004

4 years agobuild: use `find_package(Python3)` if available
Saleem Abdulrasool [Thu, 23 Apr 2020 20:46:47 +0000 (20:46 +0000)]
build: use `find_package(Python3)` if available

This is primarily motivated by the desire to move from Python2 to
Python3.  `PYTHON_EXECUTABLE` is ambiguous.  This explicitly identifies
the python interpreter in use.  Since the LLVM build seems to be able to
completed successfully with python3, use that across the build.  The old
path aliases `PYTHON_EXECUTABLE` to be treated as Python3.

4 years ago[InstCombine] Negator: 'or' with no common bits set is just 'add'
Roman Lebedev [Tue, 28 Apr 2020 10:20:04 +0000 (13:20 +0300)]
[InstCombine] Negator: 'or' with no common bits set is just 'add'

In `InstCombiner::visitAdd()`, we have
```
  // A+B --> A|B iff A and B have no bits set in common.
  if (haveNoCommonBitsSet(LHS, RHS, DL, &AC, &I, &DT))
    return BinaryOperator::CreateOr(LHS, RHS);
```
so we should handle such `or`'s here, too.

4 years ago[NFC][InstCombine] Tests for negation of 'or' with no common bits set
Roman Lebedev [Tue, 28 Apr 2020 10:07:39 +0000 (13:07 +0300)]
[NFC][InstCombine] Tests for negation of 'or' with no common bits set

4 years ago[CMSE] Clear padding bits of struct/unions/fp16 passed by value
Momchil Velikov [Tue, 28 Apr 2020 15:27:52 +0000 (16:27 +0100)]
[CMSE] Clear padding bits of struct/unions/fp16 passed by value

When passing a value of a struct/union type from secure to non-secure
state (that is returning from a CMSE entry function or passing an
argument to CMSE-non-secure call), there is a potential sensitive
information leak via the padding bits in the structure. It is not
possible in the general case to ensure those bits are cleared by using
Standard C/C++.

This patch makes the compiler emit code to clear such padding
bits. Since type information is lost in LLVM IR, the code generation
is done by Clang.

For each interesting record type, we build a bitmask, in which all the
bits, corresponding to user declared members, are set. Values of
record types are returned by coercing them to an integer. After the
coercion, the coerced value is masked (with bitwise AND) and then
returned by the function. In a similar manner, values of record types
are passed as arguments by coercing them to an array of integers, and
the coerced values themselves are masked.

For union types, we effectively clear only bits, which aren't part of
any member, since we don't know which is the currently active one.
The compiler will issue a warning, whenever a union is passed to
non-secure state.

Values of half-precision floating-point types are passed in the least
significant bits of a 32-bit register (GPR or FPR) with the most
significant bits unspecified. Since this is also a potential leak of
sensitive information, this patch also clears those unspecified bits.

Differential Revision: https://reviews.llvm.org/D76369

4 years ago[AArch64] Add support for -ffixed-x30
Francis Visoiu Mistrih [Sat, 22 Feb 2020 01:35:19 +0000 (17:35 -0800)]
[AArch64] Add support for -ffixed-x30

Add support for reserving LR in:

* the driver through `-ffixed-x30`
* cc1 through `-target-feature +reserve-x30`
* the backend through `-mattr=+reserve-x30`
* a subtarget feature `reserve-x30`

the same way we're doing for the other registers.

4 years ago[TII] remove overrides of isUnpredicatedTerminator
Nick Desaulniers [Tue, 28 Apr 2020 15:34:01 +0000 (08:34 -0700)]
[TII] remove overrides of isUnpredicatedTerminator

Summary:
They all match the base implementation in
TargetInstrInfo::isUnpredicatedTerminator.

Follow up to D62749.

Reviewers: echristo, MaskRay, hfinkel

Reviewed By: echristo

Subscribers: wuzish, nemanjai, hiraditya, kbarton, llvm-commits, srhines

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D78976

4 years ago[SveEmitter] Add builtins for logical and predicate operations.
Sander de Smalen [Tue, 28 Apr 2020 15:06:26 +0000 (16:06 +0100)]
[SveEmitter] Add builtins for logical and predicate operations.

This patch adds builtins for logical ops:
- svand, svbic, sveor, svorr, svcnot, svnot

and builtins for predicate operations:
- svand_b_z, svbic_b_z, sveor_b_z, svnand_b_z, svnor_b_z, svorn_b_z, svorr_b_z
- svbrka_b_z, svbrkb_b_z, svbrkpa_b_z, svbrkpb_b_z, svbrkn_b_z
- svpfirst_b
- svpnext
- svptest_any
- svptest_first
- svptest_last

4 years ago[ARM] Always replace FP16 bitcasts with VMOVhr or VMOVrh
David Green [Tue, 28 Apr 2020 09:10:04 +0000 (10:10 +0100)]
[ARM] Always replace FP16 bitcasts with VMOVhr or VMOVrh

This changes the logic with lowering fp16 bitcasts to always produce
either a VMOVhr or a VMOVrh, instead of only trying to do it with
certain surrounding nodes. To perform the same optimisations demand bits
and known bits information has been added for them.

Differential Revision: https://reviews.llvm.org/D78587

4 years agoHandle part-word LL/SC in atomic expansion pass
Krzysztof Parzyszek [Mon, 23 Mar 2020 17:47:32 +0000 (12:47 -0500)]
Handle part-word LL/SC in atomic expansion pass

Differential Revision: https://reviews.llvm.org/D77213

4 years ago[libc++][Take 2] Create a small DSL for defining Lit features and parameters
Louis Dionne [Tue, 28 Apr 2020 13:30:24 +0000 (09:30 -0400)]
[libc++][Take 2] Create a small DSL for defining Lit features and parameters

This allows defining Lit features that can be enabled or disabled based
on compiler support, and parameters that are passed on the command line.

The main benefits are:
- Feature detection is entirely based on the substitutions provided in
  the TestingConfig object, which is simpler and decouples it from the
  complicated compiler emulation infrastructure.
- The syntax is declarative, which makes it easy to see what features
  and parameters are accepted by the test suite. This is significantly
  less entangled than the current config.py logic.
- Since feature detection is based on substitutions, it works really
  well on top of the new format, and custom Lit configurations can be
  created easily without being based on `config.py`.

This commit is a reapplication of 6d58030c8c7d, which was reverted in
8f24c4b72f3e because it broke Python 3 support. This re-application
supports Python 3.

Differential Revision: https://reviews.llvm.org/D78381

4 years ago[MLIR] Give AffineStoreOp and AffineLoadOp Memory SideEffects.
Tres Popp [Mon, 27 Apr 2020 14:40:00 +0000 (16:40 +0200)]
[MLIR] Give AffineStoreOp and AffineLoadOp Memory SideEffects.

Summary:
This change results in tests also being changed to prevent dead
affine.load operations from being folded away during rewrites.

Also move AffineStoreOp and AffineLoadOp to an ODS file.

Differential Revision: https://reviews.llvm.org/D78930

4 years agoRecommit [libc++] Move abs and div into stdlib.h to fix header cycle.
Eric Fiselier [Sat, 15 Feb 2020 23:55:07 +0000 (18:55 -0500)]
Recommit [libc++] Move abs and div into stdlib.h to fix header cycle.

This relands this commit as it broke the LLDB bot the first time it landed.
See also the discussion on https://reviews.llvm.org/rG82b47b2978405f802a33b00d046e6f18ef6a47be

Since D74892 this code should now also work on macOS.

Original description:

libc++ is careful to not fracture overload sets. When one overload
is visible to a user, all of them should be. Anything less causes
subtle bugs and ODR violations.

Previously, in order to support ::abs and ::div being supplied by
both <cmath> and <cstdlib> we had to do awful things that make
<math.h> and <stdlib.h> have header cycles and be non-modular.
This really breaks with modules.

Specifically the problem was that in C++ ::abs introduces overloads
for floating point numbers, these overloads forward to ::fabs,
which are defined in math.h. Therefore ::abs needed to be in math.h
too. But this required stdlib.h to include math.h and math.h to
include stdlib.h.

To avoid these problems the definitions have been moved to stddef.h
(which math includes), and the floating point overloads of ::abs
have been changed to call __builtin_fabs, which both Clang and GCC
support.

4 years ago[DebugInfo] Fix crash caused by unhandled error.
Xing GUO [Tue, 28 Apr 2020 13:38:50 +0000 (21:38 +0800)]
[DebugInfo] Fix crash caused by unhandled error.

Summary: This patch helps fix LLVM crash caused by unhandled error.

Reviewers: clayborg, aprantl

Reviewed By: clayborg

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D78924

4 years agorecommit c77a4078e01033aa2206c31a579d217c8a07569b with fix
Yaxun (Sam) Liu [Fri, 24 Apr 2020 20:41:24 +0000 (16:41 -0400)]
recommit c77a4078e01033aa2206c31a579d217c8a07569b with fix

https://reviews.llvm.org/D77954 caused a regression about ambiguity of new operator
in file scope.

This patch recovered the previous behavior for comparison without a caller.

This is a workaround. For real fix we need D71227

https://reviews.llvm.org/D78970

4 years ago[clang-format] Improved parser for C# properties
Jonathan Coe [Tue, 28 Apr 2020 13:11:09 +0000 (14:11 +0100)]
[clang-format] Improved parser for C# properties

Summary:
Added some examples of properties from Microsoft documentation as test cases.

https://docs.microsoft.com/en-us/dotnet/csharp/programming-guide/classes-and-structs/properties

Configuration support will be added in a follow up patch to address whether automatic properties are formatted as

```
Type MyType { get; set }
```

or

```
Type MyType
{ get; set }
```

Reviewers: krasimir, MyDeveloperDay

Reviewed By: krasimir

Subscribers: cfe-commits

Tags: #clang-format, #clang

Differential Revision: https://reviews.llvm.org/D78915

4 years ago[SveEmitter] Add builtins for zero/sign extension and bit/byte reversal.
Sander de Smalen [Tue, 28 Apr 2020 12:56:44 +0000 (13:56 +0100)]
[SveEmitter] Add builtins for zero/sign extension and bit/byte reversal.

This patch adds builtins for predicated unary builtins
svext[bhw] and svrev[bhw] and svrbit.

4 years ago[lldb][cmake] Also use local submodule visibility on Darwin
Raphael Isemann [Mon, 27 Apr 2020 12:58:50 +0000 (14:58 +0200)]
[lldb][cmake] Also use local submodule visibility on Darwin

Summary:
Currently building LLVM on macOS and on other platforms with LLVM_ENABLE_MODULES is using different module flags,
which means that a passing modules build on macOS might fail on Linux and vice versa. -fmodules-local-submodule-visibility
is the mode that has clearer semantics and is closer to the actual C++ module standard, so let's make this the default everywhere.

We can still test building without local submodule visibility on an additional bot by just changing the respective CMake flag. However,
if building without local-submodule-visibility breaks we won't revert other commits and we won't loose LLDB's/Clang's test run
information.

Reviewers: aprantl, bruno, Bigcheese

Reviewed By: Bigcheese

Subscribers: abidh, dexonsmith, JDevlieghere, lldb-commits, mgorny, llvm-commits

Tags: #llvm, #lldb

Differential Revision: https://reviews.llvm.org/D74892

4 years agoFix Wparentheses gcc warning. NFC.
Simon Pilgrim [Tue, 28 Apr 2020 12:57:47 +0000 (13:57 +0100)]
Fix Wparentheses gcc warning. NFC.

Wrap the 'anyof' hasAttribute checks so that we don't get precedence warnings with the assertion message.

4 years ago[SveEmitter] Add builtins for bitcount operations
Sander de Smalen [Tue, 28 Apr 2020 12:48:16 +0000 (13:48 +0100)]
[SveEmitter] Add builtins for bitcount operations

This patch adds builtins for svcls, svclz and svcnt.

For merging (_m), zeroing (_z) and don't-care (_x) predication.

4 years agoRevert "[MLIR] Introduce op trait PolyhedralScope"
Dmitri Gribenko [Tue, 28 Apr 2020 12:39:46 +0000 (14:39 +0200)]
Revert "[MLIR] Introduce op trait PolyhedralScope"

This reverts commit dd2c639c3cd397dfef941186fb85c82e4e918425. It broke a
few things -- the explanation will be posted to the review thread.

4 years ago[SveEmitter] Add builtins for permutations and selection
Sander de Smalen [Tue, 28 Apr 2020 12:38:41 +0000 (13:38 +0100)]
[SveEmitter] Add builtins for permutations and selection

This patch adds builtins for:
- svlasta and svlastb
- svclasta and svclastb
- svunpkhi and svunpklo
- svuzp1 and svuzp2
- svzip1 and svzip2
- svrev
- svsel
- svcompact
- svsplice
- svtbl

4 years agoRevert: [libc++] Create a small DSL for defining Lit features and parameters
David Zarzycki [Tue, 28 Apr 2020 12:09:45 +0000 (08:09 -0400)]
Revert: [libc++] Create a small DSL for defining Lit features and parameters

This reverts commit 6d58030c8c7d481d08a549246fd3103aceb61c7b due to lack
of Python 3 support. As a reminder, the Python community ended their
support for 2.x at the start of 2020.

4 years ago[LLD][ELF] Eliminate symbols of merged .ARM.exidx sections.
Igor Kudrin [Fri, 24 Apr 2020 03:32:49 +0000 (10:32 +0700)]
[LLD][ELF] Eliminate symbols of merged .ARM.exidx sections.

GNU tools generate mapping symbols "$d" for .ARM.exidx sections. The
symbols are added to the symbol table much earlier than the merging
takes place, and after that, they become dangling. Before the patch,
LLD output those symbols as SHN_ABS with the value of 0. The patch
removes such symbols from the symbol table.

Differential Revision: https://reviews.llvm.org/D78820

4 years ago[OpenCL] Fixed test for the cast operators.
Anastasia Stulova [Tue, 28 Apr 2020 10:49:31 +0000 (11:49 +0100)]
[OpenCL] Fixed test for the cast operators.

The test had unused variable because it missed to cover
case with __constant address space. This change now
completes the testing fully.

4 years ago[lldb/unittest] Adjust CheckIPSupport function to avoid double-consume of llvm::Error
Pavel Labath [Tue, 28 Apr 2020 11:33:52 +0000 (13:33 +0200)]
[lldb/unittest] Adjust CheckIPSupport function to avoid double-consume of llvm::Error

The problem caught by clang-tidy and reported by Tobias Bosch.

4 years ago[lldb-vscode] A couple of small style fixes
Pavel Labath [Tue, 28 Apr 2020 11:13:45 +0000 (13:13 +0200)]
[lldb-vscode] A couple of small style fixes

to make the code conform to llvm style better:
- avoid use of auto where the type is not obivous
- avoid StringRef::data where it is not needed

No functional change intended.

4 years ago[Flang][CMake] Add explicit libFortranCommon dependency for f18 etc.
Rainer Orth [Tue, 28 Apr 2020 11:28:41 +0000 (13:28 +0200)]
[Flang][CMake] Add explicit libFortranCommon dependency for f18 etc.

When I tried Solaris builds with `-DBUILD_SHARED_LIBS=ON`, some commands failed
to link:

  [ 94%] Linking CXX executable ../../../../bin/f18
  Undefined                       first referenced
   symbol                             in file
  Fortran::common::IntrinsicTypeDefaultKinds::set_sizeIntegerKind(int) CMakeFiles/f18.dir/f18.cpp.o  (symbol belongs to implicit dependency /var/llvm/local-amd64-release-shared-gcc8-make/lib/libFortranCommon.so.11git)
  Fortran::common::IntrinsicTypeDefaultKinds::set_subscriptIntegerKind(int) CMakeFiles/f18.dir/f18.cpp.o  (symbol belongs to implicit dependency /var/llvm/local-amd64-release-shared-gcc8-make/lib/libFortranCommon.so.11git)
  Fortran::common::EnumIndexToString[abi:cxx11](int, char const*) CMakeFiles/f18.dir/f18.cpp.o  (symbol belongs to implicit dependency /var/llvm/local-amd64-release-shared-gcc8-make/lib/libFortranCommon.so.11git)
  Fortran::common::IntrinsicTypeDefaultKinds::set_defaultIntegerKind(int) CMakeFiles/f18.dir/f18.cpp.o  (symbol belongs to implicit dependency /var/llvm/local-amd64-release-shared-gcc8-make/lib/libFortranCommon.so.11git)
  Fortran::common::IntrinsicTypeDefaultKinds::IntrinsicTypeDefaultKinds() CMakeFiles/f18.dir/f18.cpp.o  (symbol belongs to implicit dependency /var/llvm/local-amd64-release-shared-gcc8-make/lib/libFortranCommon.so.11git)
  Fortran::common::IntrinsicTypeDefaultKinds::set_defaultRealKind(int) CMakeFiles/f18.dir/f18.cpp.o  (symbol belongs to implicit dependency /var/llvm/local-amd64-release-shared-gcc8-make/lib/libFortranCommon.so.11git)
  ld: fatal: symbol referencing errors

This patch fixes this by adding explicit dependencies on `libFortranCommon`
to the affected commands.

Tested on `amd64-pc-solaris2.11`, `sparcv9-sun-solaris2.11`, and
`x86-64-pc-linux-gnu`.

Differential Revision: https://reviews.llvm.org/D78761

4 years ago[gcov][test] Work around PR45673 - NFC
KAWASHIMA Takahiro [Tue, 28 Apr 2020 11:16:24 +0000 (20:16 +0900)]
[gcov][test] Work around PR45673 - NFC

Work around PR45673 until the test code is fixed.

4 years ago[SveEmitter] Add builtins for ternary ops (fmla, fmad, etc)
Sander de Smalen [Tue, 28 Apr 2020 09:59:38 +0000 (10:59 +0100)]
[SveEmitter] Add builtins for ternary ops (fmla, fmad, etc)

This patch adds builtins for:
- svmad, svmla, svmls, svmsb
  svnmad, svnmla, svnmls, svnmsb
  svmla_lane, svmls_lane

These builtins come in several flavours:
- Merge into first source vector (`_m`)
- False lanes are undef (`_x`)
- False lanes are zeroed (`_z`)

And can also have `_n` to indicate the last operand is a scalar.

For example:

  svint32_t svmla[_n_s32]_z(svbool_t pg, svint32_t op1, svint32_t op2, int32_t op3)

Reviewed By: efriedma

Differential Revision: https://reviews.llvm.org/D78960

4 years ago[Powerpc] add triple for new added qpx test case - NFC
Chen Zheng [Tue, 28 Apr 2020 09:27:06 +0000 (05:27 -0400)]
[Powerpc] add triple for new added qpx test case - NFC

4 years ago[PowerPC] Fix downcast from nullptr for target streamer
Ng Zhi An [Tue, 28 Apr 2020 09:20:10 +0000 (09:20 +0000)]
[PowerPC] Fix downcast from nullptr for target streamer

getTargetStreamer() might return null (e.g. when running inlined-strings.ll test),
downcasting to a reference will be wrong. This is detectable with -fsanitize=null.

Reviewed By: steven.zhang

Differential Revision: https://reviews.llvm.org/D78686

4 years ago[PowerPC] add test case for reorder operands of qpx fma instr - nfc.
Chen Zheng [Tue, 28 Apr 2020 08:28:23 +0000 (04:28 -0400)]
[PowerPC] add test case for reorder operands of qpx fma instr - nfc.

4 years ago[mlir][flang] use OpBuilder& instead of Builder* in <Op>::build methods
Alex Zinenko [Thu, 23 Apr 2020 14:02:46 +0000 (16:02 +0200)]
[mlir][flang] use OpBuilder& instead of Builder* in <Op>::build methods

As we start defining more complex Ops, we increasingly see the need for
Ops-with-regions to be able to construct Ops within their regions in
their ::build methods. However, these methods only have access to
Builder, and not OpBuilder. Creating a local instance of OpBuilder
inside ::build and using it fails to trigger the operation creation
hooks in derived builders (e.g., ConversionPatternRewriter). In this
case, we risk breaking the logic of the derived builder. At the same
time, OpBuilder::create, which is by far the largest user of ::build
already passes "this" as the first argument, so an OpBuilder instance is
already available.

Update all ::build methods in all Ops in MLIR and Flang to take
"OpBuilder &" instead of "Builder *". Note the change from pointer and
to reference to comply with the common style in MLIR, this also ensures
all other users must change their ::build methods.

Differential Revision: https://reviews.llvm.org/D78713

4 years agoProviding buffer assignment for MLIR
Ehsan Toosi [Mon, 20 Apr 2020 10:41:14 +0000 (12:41 +0200)]
Providing buffer assignment for MLIR

We have provided a generic buffer assignment transformation ported from
TensorFlow. This generic transformation pass automatically analyzes the values
and their aliases (also in other blocks) and returns the valid positions for
Alloc and Dealloc operations. To find these positions, the algorithm uses the
block Dominator and Post-Dominator analyses. In our proposed algorithm, we have
considered aliasing, liveness, nested regions, branches, conditional branches,
critical edges, and independency to custom block terminators. This
implementation doesn't support block loops. However, we have considered this in
our design. For this purpose, it is only required to have a loop analysis to
insert Alloc and Dealloc operations outside of these loops in some special
cases.

Differential Revision: https://reviews.llvm.org/D78484

4 years agoRevert f8990feb125a "[libclang] Install both libclang.a and libclang.so when LIBCLANG...
Hans Wennborg [Tue, 28 Apr 2020 08:10:33 +0000 (10:10 +0200)]
Revert f8990feb125a "[libclang] Install both libclang.a and libclang.so when LIBCLANG_BUILD_STATIC=ON"

This broke builds configured with

$ cmake -GNinja -DCMAKE_BUILD_TYPE=Release '-DLLVM_ENABLE_PROJECTS=clang' '-DLLVM_TARGETS_TO_BUILD=X86' -DLLVM_ENABLE_PIC=OFF ../llvm

CMake Error at
/b/s/w/ir/cache/builder/src/third_party/llvm/clang/tools/libclang/CMakeLists.txt:123
(target_compile_definitions):
    target_compile_definitions called with non-compilable target type

This reverts commit f8990feb125a0f8d3f2892a589bc6fad3c430858.

4 years ago[analyzer] StdLibraryFunctionsChecker: Associate summaries to FunctionDecls
Gabor Marton [Tue, 7 Apr 2020 09:44:33 +0000 (11:44 +0200)]
[analyzer] StdLibraryFunctionsChecker: Associate summaries to FunctionDecls

Summary:
Currently we map function summaries to names (i.e. strings). We can
associate more summaries with different signatures to one name, this way
we support overloading. During a call event we check whether the
signature of the summary matches the signature of the callee and we
apply the summary only in that case.

In this patch we change this mapping to associate a summary to a
FunctionDecl. We do lookup operations when the summary map is
initialized. We lookup the given name and we match the signature of the
given summary against the lookup results. If the summary matches the
FunctionDecl (got from the lookup result) then we add that to the
summary map. During a call event we compare FunctionDecl pointers.
Advantages of this new refactor:
- Cleaner mapping and structure for the checker.
- Possibly way more efficient handling of call events.
- A summary is added only if that is relevant for the given TU.
- We can get the concrete FunctionDecl by the time when we create the
  summary, this opens up possibilities of further sanity checks
  regarding the summary cases and argument constraints.
- Opens up to future work when we'd like to store summaries from IR to a
  FunctionDecl (or from the Attributor results of the given
  FunctionDecl).

Note, we cannot support old C functions without prototypes.

Reviewers: NoQ, Szelethus, balazske, jdoerfert, sstefan1, uenoku

Subscribers: whisperity, xazax.hun, baloghadamsoftware, szepet, rnkovacs, a.sidorin, mikhail.ramalho, donat.nagy, dkrupp, gamesh411, Charusso, steakhal, uenoku, ASDenysPetrov, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D77641

4 years ago[TTI] Add TargetCostKind argument to getUserCost
Sam Parker [Mon, 27 Apr 2020 08:02:14 +0000 (09:02 +0100)]
[TTI] Add TargetCostKind argument to getUserCost

There are several different types of cost that TTI tries to provide
explicit information for: throughput, latency, code size along with
a vague 'intersection of code-size cost and execution cost'.

The vectorizer is a keen user of RecipThroughput and there's at least
'getInstructionThroughput' and 'getArithmeticInstrCost' designed to
help with this cost. The latency cost has a single use and a single
implementation. The intersection cost appears to cover most of the
rest of the API.

getUserCost is explicitly called from within TTI when the user has
been explicit in wanting the code size (also only one use) as well
as a few passes which are concerned with a mixture of size and/or
a relative cost. In many cases these costs are closely related, such
as when multiple instructions are required, but one evident diverging
cost in this function is for div/rem.

This patch adds an argument so that the cost required is explicit,
so that we can make the important distinction when necessary.

Differential Revision: https://reviews.llvm.org/D78635

4 years agoUse components instead of libraries in Polly linkage step
serge-sans-paille [Sat, 18 Apr 2020 06:14:52 +0000 (08:14 +0200)]
Use components instead of libraries in Polly linkage step

As a side effect, this tests (and fix a bug) in the compiler extension handling
of components.

Differential Revision: https://reviews.llvm.org/D78358

4 years ago[SystemZ] Fix test case.
Jonas Paulsson [Tue, 28 Apr 2020 07:37:43 +0000 (09:37 +0200)]
[SystemZ] Fix test case.

Remove bad kill flags fom load-and-test.mir as discovered by
https://reviews.llvm.org/D78586: "[MachineVerifier] Add more checks for
registers in live-in lists".

Review: Ulrich Weigand

4 years ago[VE] Update branch instructions
Kazushi (Jam) Marukawa [Tue, 28 Apr 2020 07:41:01 +0000 (09:41 +0200)]
[VE] Update branch instructions

Summary:
Changing all mnemonic to match assembly instructions to simplify mnemonic
naming rules. This time update all branch instructions.  This also change
to use %s10 register consistently.

Differential Revision: https://reviews.llvm.org/D78889

4 years ago[VE] Support floating point immediate values
Kazushi (Jam) Marukawa [Tue, 28 Apr 2020 07:35:58 +0000 (09:35 +0200)]
[VE] Support floating point immediate values

Summary:
Add simm7fp/mimmfp to represent floating point immediate values.
Also clean multiclasses to define floating point arithmetic instructions
to handle simm7fp/mimmfp operands.  Also add several regression tests
for new operands.

Differential Revision: https://reviews.llvm.org/D78887

4 years ago[PowerPC] use inst-level fast-math-flags to drive MachineCombiner
Chen Zheng [Tue, 28 Apr 2020 07:31:12 +0000 (03:31 -0400)]
[PowerPC] use inst-level fast-math-flags to drive MachineCombiner

Currently, on PowerPC target, it uses function scope UnsafeFPMath
option to drive Machine Combiner pass.

This is not accurate in two ways:
1: the scope is not accurate. Machine Combiner pass only requires
   instruction-level flags instead of the function scope.
2: the float point flag is not accurate. Machine Combiner pass
   only requires float point flags reassoc and nsz.

Reviewed By: steven.zhang

Differential Revision: https://reviews.llvm.org/D78183

4 years agoFix the -Wunused-variable warning.
Haojian Wu [Tue, 28 Apr 2020 06:44:15 +0000 (08:44 +0200)]
Fix the -Wunused-variable warning.

4 years ago[SelectionDAGBuilder] Use CallBase::isInlineAsm in a couple places. NFC
Craig Topper [Tue, 28 Apr 2020 05:36:25 +0000 (22:36 -0700)]
[SelectionDAGBuilder] Use CallBase::isInlineAsm in a couple places. NFC

These lines were just changed from using CallBase::getCalledValue
to getCallledOperand. Go aheand change them to isInlineAsm.

4 years ago[IR] Replace all uses of CallBase::getCalledValue() with getCalledOperand().
Craig Topper [Tue, 28 Apr 2020 03:15:59 +0000 (20:15 -0700)]
[IR] Replace all uses of CallBase::getCalledValue() with getCalledOperand().

This method has been commented as deprecated for a while. Remove
it and replace all uses with the equivalent getCalledOperand().

I also made a few cleanups in here. For example, to removes use
of getElementType on a pointer when we could just use getFunctionType
from the call.

Differential Revision: https://reviews.llvm.org/D78882

4 years ago[AMDGPU] DWARF proposal review feedback
Tony [Mon, 20 Apr 2020 20:55:34 +0000 (16:55 -0400)]
[AMDGPU] DWARF proposal review feedback

- Rename DW_OP_LLVM_offset_constu to DW_OP_LLVM_offset_uconst to
  matches DW_OP_plus_uconst.
- Correct DW_OP_LLVM_call_ref to be DW_OP_call_ref.
- Move proposed changes to a separate section to clarify that the
  introduction section is not part of the changes.
- Fix formatting typos and add missing reference.
- Clarify why DW_OP_LLVM_offset et al do not wrap on overflow.
- Correct syntax of augmentation string.

Differential Revision: https://reviews.llvm.org/D70523

4 years ago[MLIR] Introduce op trait PolyhedralScope
Uday Bondhugula [Sat, 25 Apr 2020 04:24:18 +0000 (09:54 +0530)]
[MLIR] Introduce op trait PolyhedralScope

Introduce op trait `PolyhedralScope` for ops to define a new scope for
polyhedral optimization / affine dialect purposes, thus generalizing
such scopes beyond FuncOp. Ops to which this trait is attached will
define a new scope for the consideration of SSA values as valid symbols
for the purposes of polyhedral analysis and optimization. Update methods
that check for dim/symbol validity to work based on this trait.

Differential Revision: https://reviews.llvm.org/D78863

4 years agoFix missing namespace in API implementation.
Mircea Trofin [Tue, 28 Apr 2020 04:05:33 +0000 (21:05 -0700)]
Fix missing namespace in API implementation.

4 years ago[llvm][NFC] Use CallBase instead of Instruction in ProfileSummaryInfo
Mircea Trofin [Mon, 27 Apr 2020 16:07:18 +0000 (09:07 -0700)]
[llvm][NFC] Use CallBase instead of Instruction in ProfileSummaryInfo

Summary:
getProfileCount requires the parameter be a valid CallBase, and its uses
reflect that.

Reviewers: dblaikie, craig.topper, wmi

Subscribers: eraman, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D78940

4 years agoRevert "[gn build] (manually) merge cd84bfb8142bc7ff3a0"
Nico Weber [Tue, 28 Apr 2020 03:31:07 +0000 (23:31 -0400)]
Revert "[gn build] (manually) merge cd84bfb8142bc7ff3a0"

This reverts commit 825f583c86ca856e930b434c6e3462398461eb9a.
This reverts commit 45417ecbc1cddb667d044c709ee980c9426d5eb2.

cd84bfb8142bc7ff3a0 was reverted in be884b79352de89

4 years ago[PowerPC] Fix the liveins for ppc-expand-isel pass
Kang Zhang [Tue, 28 Apr 2020 03:22:48 +0000 (03:22 +0000)]
[PowerPC] Fix the liveins for ppc-expand-isel pass

Summary:
In the ppc-expand-isel pass, we use stepForward() to update the
liveins, this function is not recommended, because it needs the
accurate kill info.

This patch uses the function computeAndAddLiveIns() to update the
liveins, it's the recommended method and can fix the liveins bug for
ppc-expand-isel pass..

Reviewed By: efriedma, lkail

Differential Revision: https://reviews.llvm.org/D78657

4 years agoRevert "build: use `find_package(Python3)` if available"
Saleem Abdulrasool [Tue, 28 Apr 2020 03:03:32 +0000 (20:03 -0700)]
Revert "build: use `find_package(Python3)` if available"

This reverts commit cd84bfb8142bc7ff3a07a188ffb809f1d86d1fd7.  Although
this passed the CI in phabricator, some of the bots are missing python3
packages, revert it temporarily.

4 years ago[gn build] (manually) merge cd84bfb8142bc7ff3a0 more precisely
Nico Weber [Tue, 28 Apr 2020 01:55:02 +0000 (21:55 -0400)]
[gn build] (manually) merge cd84bfb8142bc7ff3a0 more precisely

4 years ago[gn build] (manually) merge cd84bfb8142bc7ff3a0
Nico Weber [Tue, 28 Apr 2020 01:45:49 +0000 (21:45 -0400)]
[gn build] (manually) merge cd84bfb8142bc7ff3a0

4 years agoMake MLIR Pass Timing output configurable through injection
Mehdi Amini [Mon, 27 Apr 2020 23:38:17 +0000 (23:38 +0000)]
Make MLIR Pass Timing output configurable through injection

This makes it possible for the client to control where the pass timings will
be printed.

Differential Revision: https://reviews.llvm.org/D78891

4 years agobuild: use `find_package(Python3)` if available
Saleem Abdulrasool [Thu, 23 Apr 2020 20:46:47 +0000 (20:46 +0000)]
build: use `find_package(Python3)` if available

This is primarily motivated by the desire to move from Python2 to
Python3.  `PYTHON_EXECUTABLE` is ambiguous.  This explicitly identifies
the python interpreter in use.  Since the LLVM build seems to be able to
completed successfully with python3, use that across the build.  The old
path aliases `PYTHON_EXECUTABLE` to be treated as Python3.

4 years ago[llvm-objcopy][MachO] Fix build
Alexander Shaposhnikov [Tue, 28 Apr 2020 01:20:01 +0000 (18:20 -0700)]
[llvm-objcopy][MachO] Fix build

Some compilers are confused when the same name is used in different contexts.
Rename the field Section to unbreak the build.
(Caught by the buildbot http://lab.llvm.org:8011/builders/clang-with-thin-lto-ubuntu/builds/22374)

4 years ago[llvm-objcopy][MachO] Handle relocation entries where r_extern is zero
Alexander Shaposhnikov [Mon, 27 Apr 2020 05:39:50 +0000 (22:39 -0700)]
[llvm-objcopy][MachO] Handle relocation entries where r_extern is zero

Fix handling of relocations with r_extern == 0.
If r_extern == 0 then r_symbolnum is an index of a section rather than a symbol index.

Patch by Seiya Nuta and Alexander Shaposhnikov.

Test plan: make check-all

Differential revision: https://reviews.llvm.org/D78946

4 years ago[flang] Upstream recent work on FIR to llvm-project.
Eric Schweitz [Fri, 24 Apr 2020 22:24:20 +0000 (15:24 -0700)]
[flang] Upstream recent work on FIR to llvm-project.

Summary:

Reviewers: DavidTruby, sscalpone, jeanPerier

Subscribers: mgorny, aartbik, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D78835

4 years ago[X86] remove derived method w/ same impl as base
Nick Desaulniers [Tue, 28 Apr 2020 00:40:01 +0000 (17:40 -0700)]
[X86] remove derived method w/ same impl as base

Summary:
While looking into issues with IfConverter, I noticed that
X86InstrInfo::isUnpredicatedTerminator matched its overriden
implementation in TargetInstrInfo::isUnpredicatedTerminator.

Reviewers: craig.topper, hfinkel, MaskRay, echristo

Reviewed By: MaskRay, echristo

Subscribers: hiraditya, llvm-commits, srhines

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D62749

4 years ago[mlir][LLVMDebugTranslation] Only insert the location mapping after translation
River Riddle [Mon, 27 Apr 2020 23:42:58 +0000 (16:42 -0700)]
[mlir][LLVMDebugTranslation] Only insert the location mapping after translation

This fixes an iteration invalidation bug when the map grows beyond capacity and the iterator for the location to translate becomes invalid.

4 years ago[llvm/DebugInfo] Fix typo in DWARFTypeUnit ctor call
Jonas Devlieghere [Mon, 27 Apr 2020 22:51:37 +0000 (15:51 -0700)]
[llvm/DebugInfo] Fix typo in DWARFTypeUnit ctor call

We were passing the AppleObjCSection instead of the AddrSection. Maybe
the API changed and this remained unnoticed because the types are the
same, or maybe it's just a typo.

4 years ago[llvm/DebugInfo] Print DW_AT_ranges offset as part of verifier error.
Jonas Devlieghere [Mon, 27 Apr 2020 22:41:56 +0000 (15:41 -0700)]
[llvm/DebugInfo] Print DW_AT_ranges offset as part of verifier error.

Print the DW_AT_ranges offset as part of the verifier error, like we do
for the DW_AT_stmt_list offset.

4 years ago[SVE][NFC] Use ScalableVectorType in CGBuiltin
Christopher Tetreault [Mon, 27 Apr 2020 23:11:40 +0000 (16:11 -0700)]
[SVE][NFC] Use ScalableVectorType in CGBuiltin

Summary: * Upgrade some usages of VectorType to use ScalableVectorType

Reviewers: efriedma, david-arm, fpetrogalli, kmclaughlin

Reviewed By: efriedma

Subscribers: tschuett, rkruppe, psnobl, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D78842

4 years agoAdd IR constructs for preallocated (inalloca replacement)
Arthur Eubanks [Fri, 14 Feb 2020 22:16:53 +0000 (14:16 -0800)]
Add IR constructs for preallocated (inalloca replacement)

Add llvm.call.preallocated.{setup,arg} instrinsics.
Add "preallocated" operand bundle which takes a token produced by llvm.call.preallocated.setup.
Add "preallocated" parameter attribute, which is like byval but without the copy.

Verifier changes for these IR constructs.

See https://github.com/rnk/llvm-project/blob/call-setup-docs/llvm/docs/CallSetup.md

Subscribers: hiraditya, jdoerfert, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D74651

4 years ago[X86][CostModel] Add vXiY->vXi1 truncate tests to min-legal-vector-width.ll. NFC
Craig Topper [Mon, 27 Apr 2020 22:47:53 +0000 (15:47 -0700)]
[X86][CostModel] Add vXiY->vXi1 truncate tests to min-legal-vector-width.ll. NFC

4 years ago[SVE] Add specialized overloads of VectorType::get
Christopher Tetreault [Mon, 27 Apr 2020 22:18:14 +0000 (15:18 -0700)]
[SVE] Add specialized overloads of VectorType::get

Summary:
Add overloads of get to FixedVectorType and ScalableVectorType that
mirror VectorType::get(Type *Ty, VectorType *VTy) that take and return
derived vector types

Reviewers: efriedma, c-rhodes, david-arm, fpetrogalli

Reviewed By: david-arm

Subscribers: tschuett, rkruppe, psnobl, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D78841

4 years agoAdd a folder for division by one.
Phoenix Meadowlark [Mon, 27 Apr 2020 19:59:16 +0000 (19:59 +0000)]
Add a folder for division by one.

- Adds a folder for integer division by one with the `divi_signed` and `divi_unsigned` ops.
- Creates tests for scalar and tensor versions of these ops.
- Modifies the test in `parallel-loop-collapsing.mlir` so that it doesn't assume division by one will be in the output.

Differential Revision: https://reviews.llvm.org/D78518