platform/upstream/llvm.git
2 years ago[mlir][sparse] Adding Action::kSparseToSparse for @newSparseTensor
wren romano [Sat, 19 Mar 2022 02:20:33 +0000 (19:20 -0700)]
[mlir][sparse] Adding Action::kSparseToSparse for @newSparseTensor

This is work towards: https://github.com/llvm/llvm-project/issues/51652

This differential doesn't yet make use of the new kSparseToSparse, just introduces it.  The differential that finally makes use of them is D122061, which is the final differential in the chain that fixes bug 51652.

Depends On D122054

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D122055

2 years ago[flang][NFC] Add forall lowering tests
Valentin Clement [Tue, 22 Mar 2022 20:32:57 +0000 (21:32 +0100)]
[flang][NFC] Add forall lowering tests

This patch adds some lowering tests for the `forall` construct.

This patch is part of the upstreaming effort from fir-dev branch.

Reviewed By: PeteSteinfeld

Differential Revision: https://reviews.llvm.org/D122253

Co-authored-by: Jean Perier <jperier@nvidia.com>
Co-authored-by: Eric Schweitz <eschweitz@nvidia.com>
2 years agoRevert "Recommit "[SLP] Fix lookahead operand reordering for splat loads.""
Arthur Eubanks [Tue, 22 Mar 2022 20:27:04 +0000 (13:27 -0700)]
Revert "Recommit "[SLP] Fix lookahead operand reordering for splat loads.""

This reverts commit 79613185d305013de743cdbd6690e4d77c8af27e.

Causes crashes, see comments in https://reviews.llvm.org/D121973.

2 years ago[InstCombine] reorder code, fix formatting; NFC
Sanjay Patel [Tue, 22 Mar 2022 20:02:09 +0000 (16:02 -0400)]
[InstCombine] reorder code, fix formatting; NFC

The affected code can be updated to solve #54364,
so make some cosmetic diffs before real changes.

2 years ago[clang][extract-api] Refactor ExtractAPI and improve docs
Zixu Wang [Mon, 21 Mar 2022 07:53:28 +0000 (00:53 -0700)]
[clang][extract-api] Refactor ExtractAPI and improve docs

- The name SymbolGraph is inappropriate and confusing for the new library
  for clang-extract-api. Refactor and rename things to make it clear that
  ExtractAPI is the core functionality and SymbolGraph is one serializer
  for the API information.
- Add documentation comments to ExtractAPI classes and methods to improve
  readability and clearness of the ExtractAPI work.

Differential Revision: https://reviews.llvm.org/D122160

2 years ago[test,instcombine] Precommit test for upcoming transform
Philip Reames [Tue, 22 Mar 2022 20:16:17 +0000 (13:16 -0700)]
[test,instcombine] Precommit test for upcoming transform

2 years ago[mlir][sparse] Introducing options for the SparseTensorConversion pass
wren romano [Sat, 19 Mar 2022 02:10:40 +0000 (19:10 -0700)]
[mlir][sparse] Introducing options for the SparseTensorConversion pass

This is work towards: https://github.com/llvm/llvm-project/issues/51652

This differential sets up the options and threads them through everywhere, but doesn't actually use them yet.  The differential that finally makes use of them is D122061, which is the final differential in the chain that fixes bug 51652.

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D122054

2 years ago[mlir][sparse] Moving lexOrder from SparseTensorCOO to Element
wren romano [Tue, 22 Mar 2022 00:39:15 +0000 (17:39 -0700)]
[mlir][sparse] Moving lexOrder from SparseTensorCOO to Element

This is the more logical place for the function to live.  If/when we factor out a separate class for just the `Coordinates` themselves, then the definition should be moved to `Coordinates::lexOrder` (and `Element::lexOrder` would become a thin wrapper delegating to that function).

This is (tangentially) work towards fixing: https://github.com/llvm/llvm-project/issues/51652

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D122057

2 years ago[flang][NFC] Add pointer dummy arguments tests
Valentin Clement [Tue, 22 Mar 2022 19:56:25 +0000 (20:56 +0100)]
[flang][NFC] Add pointer dummy arguments tests

This patch adds test for calls with POINTER dummy arguments on the caller side.

It also fixes some formatting error that was introduced when upstreaming
the other pointer tests.

This patch is part of the upstreaming effort from fir-dev branch.

Reviewed By: PeteSteinfeld

Differential Revision: https://reviews.llvm.org/D122238

Co-authored-by: Jean Perier <jperier@nvidia.com>
2 years ago[AMDGPU] Update mfma test to run gfx940 checks. NFC.
Stanislav Mekhanoshin [Tue, 22 Mar 2022 19:42:31 +0000 (12:42 -0700)]
[AMDGPU] Update mfma test to run gfx940 checks. NFC.

2 years ago[BOLT] Avoid pointless loop rotation
spupyrev [Thu, 17 Mar 2022 16:51:07 +0000 (09:51 -0700)]
[BOLT] Avoid pointless loop rotation

It seems the earlier implementation does not follow the description
in LoopRotationPass.h: It rotates loops even if they are already laid out
correctly. The diff adjusts the behaviour.

Given that the impact of LoopInversionPass is minor, this change won't
yield significant perf differences. Tested on clang-10: there seems to be a
0.1%-0.3% cpu win and a small reduction of branch misses.

**Before:**
BOLT-INFO: 120 Functions were reordered by LoopInversionPass

**After:**
BOLT-INFO: 79 Functions were reordered by LoopInversionPass

Reviewed By: yota9

Differential Revision: https://reviews.llvm.org/D121921

2 years ago[test] Autogen a test for ease of update
Philip Reames [Tue, 22 Mar 2022 19:38:20 +0000 (12:38 -0700)]
[test] Autogen a test for ease of update

2 years ago[cmake] Demote fatal error to a warning when we don't know the Apple SDK in use
Louis Dionne [Tue, 22 Mar 2022 19:35:30 +0000 (15:35 -0400)]
[cmake] Demote fatal error to a warning when we don't know the Apple SDK in use

Sometimes, we could be building for a platform where we don't link
compiler-rt, so being able to figure out the right compiler-rt suffix
isn't necessary, but we shouldn't fail the build.

2 years ago[libc++][NFC] Use struct instead of class for ranges::end
Louis Dionne [Tue, 22 Mar 2022 19:34:48 +0000 (15:34 -0400)]
[libc++][NFC] Use struct instead of class for ranges::end

This is consistent with what we do elsewhere.

2 years agollvm-dwarfdump: Including calling convention attribute in pretty printed type names
David Blaikie [Mon, 21 Mar 2022 19:37:39 +0000 (19:37 +0000)]
llvm-dwarfdump: Including calling convention attribute in pretty printed type names

2 years agoDebugInfo: Simplify const/volatile printing for function types
David Blaikie [Thu, 17 Mar 2022 22:59:18 +0000 (22:59 +0000)]
DebugInfo: Simplify const/volatile printing for function types

2 years ago[Clang][NFC] Add braces to help readability
Corentin Jabot [Tue, 22 Mar 2022 19:05:17 +0000 (20:05 +0100)]
[Clang][NFC] Add braces to help readability

In CheckConstexprFunctionStmt, as discussed in
D111400.

2 years agoRevert "Add UTF32 to/from UTF8 conversion functions"
Aaron Ballman [Tue, 22 Mar 2022 18:59:57 +0000 (14:59 -0400)]
Revert "Add UTF32 to/from UTF8 conversion functions"

This reverts commit c3460689288abc98c91d8d6bffa74be9eb16c74d.

It broke at least one of the builders:
https://lab.llvm.org/buildbot#builders/100/builds/13947

2 years ago[Clang][C++2b] P2242R3: Non-literal variables [...] in constexpr
Corentin Jabot [Tue, 5 Oct 2021 21:33:31 +0000 (23:33 +0200)]
[Clang][C++2b] P2242R3: Non-literal variables [...] in constexpr

Allow goto, labelled statements as well as `static`, `thread_local`, and
non-literal variables in `constexpr` functions.

As specified. for all of the above (except labelled statements) constant
evaluation of the construct still fails.

For `constexpr` bodies, the proposal is implemented with diagnostics as
a language extension in older language modes. For determination of
whether a lambda body satisfies the requirements for a constexpr
function, the proposal is implemented only in C++2b mode to retain the
semantics of older modes for programs conforming to them.

Reviewed By: aaron.ballman, hubert.reinterpretcast, erichkeane

Differential Revision: https://reviews.llvm.org/D111400

2 years ago[flang] Initial UTF-8 support in runtime I/O
Peter Klausler [Wed, 16 Mar 2022 19:32:03 +0000 (12:32 -0700)]
[flang] Initial UTF-8 support in runtime I/O

Implements UTF-8 encoding and decoding for external units
with OPEN(ENCODING='UTF-8').  This encoding applies to default
CHARACTER values that are not 7-bit ASCII as well as to
the wide CHARACTER kinds 2 and 4.  Basic testing is in place
via direct calls to the runtime I/O APIs, but serious checkout
awaits lowering support of the wide CHARACTER kinds.

Differential Revision: https://reviews.llvm.org/D122038

2 years ago[flang] Fix crash: ENTRY with generic interface of the same name
Peter Klausler [Wed, 16 Mar 2022 17:16:45 +0000 (10:16 -0700)]
[flang] Fix crash: ENTRY with generic interface of the same name

Name resolution was crashing while processing the ENTRY statement
due to a lack of special-case code necessary to handle the indirection
needed when the generic has the same name as the ENTRY.

Differential Revision: https://reviews.llvm.org/D122050

2 years ago[mlir][ods] Add Deprecate helper
Jacques Pienaar [Tue, 22 Mar 2022 18:08:36 +0000 (11:08 -0700)]
[mlir][ods] Add Deprecate helper

Add method to tag classes/defs as deprecated. Previously deprecations
were only verbally communicated and folks didn't have an active warning
while building about impending removal. Add mechanism to tag defs as
deprecated to allow warning users.

This doesn't change any policy, it just moves deprecation warnings from
comments to something more user visible.

Differential Revision: https://reviews.llvm.org/D122164

2 years ago[lldb] Set the TERM environment variable for the API tests
Jonas Devlieghere [Tue, 22 Mar 2022 02:18:29 +0000 (19:18 -0700)]
[lldb] Set the TERM environment variable for the API tests

Avoid "TERM environment variable not set" by either propagating the TERM
environment variable or defaulting to vt100. All of our CI is already
doing this explicitly through the --env dotest arg, but it's easy to
forget when setting up a new job. I don't see any downside in making it
the default.

2 years ago[MLIR][Presburger] Fix a comment in the unittests
Cynthia Shen [Tue, 22 Mar 2022 15:54:25 +0000 (15:54 +0000)]
[MLIR][Presburger] Fix a comment in the unittests

Reviewed By: arjunp

Differential Revision: https://reviews.llvm.org/D122197

2 years agoAdd UTF32 to/from UTF8 conversion functions
Marcus Johnson [Tue, 22 Mar 2022 17:41:43 +0000 (13:41 -0400)]
Add UTF32 to/from UTF8 conversion functions

This is anticipated to be used in new format specifier checking code.

2 years ago[ARM] select_xform.ll - re-add and fix missing CHECK prefixes
Simon Pilgrim [Tue, 22 Mar 2022 17:34:48 +0000 (17:34 +0000)]
[ARM] select_xform.ll - re-add and fix missing CHECK prefixes

We were still checking test results with the CHECK prefix but they had bit-rotted since whenever it'd been removed from the --check-prefixes list

2 years ago[RISCV] Special case sign extended scalars when type legalizing nxvXi64 .vx instrinsi...
Craig Topper [Tue, 22 Mar 2022 17:16:53 +0000 (10:16 -0700)]
[RISCV] Special case sign extended scalars when type legalizing nxvXi64 .vx instrinsics on RV32.

On RV32, we need to type legalize i64 scalar arguments to intrinsics.
We usually do this by splatting the value into a vector separately.
If the scalar happens to be sign extended, we can continue using a .vx
intrinsic.

We already special cased sign extended constants, this extends it
to any sign extended value.

I've only added tests for one case of vadd. Most intrinsics go
through the same check.

Reviewed By: khchen

Differential Revision: https://reviews.llvm.org/D122186

2 years agoAllow -Wno-gnu to silence GNU extensions related to pointer arithmetic
Aaron Ballman [Tue, 22 Mar 2022 17:14:16 +0000 (13:14 -0400)]
Allow -Wno-gnu to silence GNU extensions related to pointer arithmetic

These diagnostics were added to a diagnostic group, but that diagnostic
group was not under -Wgnu. I've now split them into their own
diagnostic group that is added both to the original group (so user's
currently opting in or out of these should not see a change) and under
the -Wgnu group so that -Wno-gnu can be used to disable all GNU
extension diagnostics. This fixes Issue 54444.

2 years ago[TableGen][RISCV] Add InstAliases with zero_reg to cover unmasked vnot.v, vncvt.x...
Craig Topper [Tue, 22 Mar 2022 16:53:39 +0000 (09:53 -0700)]
[TableGen][RISCV] Add InstAliases with zero_reg to cover unmasked vnot.v, vncvt.x.x.w, vneg.v, etc.

The mask being NoRegister prevented the existing aliases from matching
since NoRegister isn't in the VMV0 register class.

To workaround this I've added new aliases that look for zero_reg.
I had to motify tablegen to generate matching code for zero_reg.
And as a consequence, I had to change the EmitPriority for an ARM
alias that used zero_reg that started printing.

Reviewed By: frasercrmck

Differential Revision: https://reviews.llvm.org/D121496

2 years ago[RISCV][NFC] Refine and refactor RISCVVEmitter and riscv_vector.td.
Zakk Chen [Mon, 28 Feb 2022 09:07:13 +0000 (01:07 -0800)]
[RISCV][NFC] Refine and refactor RISCVVEmitter and riscv_vector.td.

1. Rename nomask as unmasked to keep with the terminology in the spec.
2. Merge UnMaskpolicy and Maskedpolicy arguments into one in RVVBuiltin class.
3. Rename HasAutoDef as HasBuiltinAlias.
4. Move header definition code into one class.

Reviewed By: rogfer01

Differential Revision: https://reviews.llvm.org/D120870

2 years ago[VP] Preserve address space of pointer for strided load/store intrinsics.
Craig Topper [Tue, 22 Mar 2022 16:41:40 +0000 (09:41 -0700)]
[VP] Preserve address space of pointer for strided load/store intrinsics.

This adds LLVMAnyPointerToElt to use instead of LLVMPointerToElt.
This allows us to preserve the address space as part of the type
overload for the intrinsic, but still require the vector element
type to match the pointer type.

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D122042

2 years ago[demangler] Add support for C++20 modules
Nathan Sidwell [Wed, 26 Jan 2022 15:22:04 +0000 (07:22 -0800)]
[demangler] Add support for C++20 modules

Add support for module name demangling.  We have two new demangler
nodes -- ModuleName and ModuleEntity. The former represents a module
name in a hierarchical fashion. The latter is the combination of a
(name) node and a module name. Because module names and entity
identities use the same substitution encoding, we have to adjust the
flow of how substitutions are handled, and examine the substituted
node to know how to deal with it.

Reviewed By: dblaikie

Differential Revision: https://reviews.llvm.org/D119933

2 years ago[clang][NFC] Refactor logic for picking standard library on Apple
Louis Dionne [Tue, 22 Mar 2022 14:58:24 +0000 (10:58 -0400)]
[clang][NFC] Refactor logic for picking standard library on Apple

Flip the logic around: always default to libc++ except on older platforms,
instead of defaulting to libstdc++ except on newer platforms. Since roughly
all supported platforms use libc++ now, it makes more sense to make that
the default, and allows the removal of some downstream diff.

Differential Revision: https://reviews.llvm.org/D122232

2 years ago[IR] Allow matching pointer to vector with opaque pointers.
Hendrik Greving [Tue, 22 Mar 2022 02:20:53 +0000 (19:20 -0700)]
[IR] Allow matching pointer to vector with opaque pointers.

Allows for skipping the pointer to vector type if opaque pointers
are enabled and the matching pointer is a vector pointer when
matching an intrinsic signature in the verifier.

No test added since lacking a target using intrinsic with pointer
to vector arguments.

Differential Revision: https://reviews.llvm.org/D122203

2 years ago[WebAssembly][NFC] Delete type checking logic for removed ref.null instruction
Alex Bradbury [Tue, 22 Mar 2022 16:24:05 +0000 (16:24 +0000)]
[WebAssembly][NFC] Delete type checking logic for removed ref.null instruction

D114979 changed the textual formal of ref.null - dropping ref.null in
favour of ref.null_extern and ref.null_func. Therefore, the type checker
no longer needs logic to handle "ref.null".

Differential Revision: https://reviews.llvm.org/D122123

2 years ago [WebAssembly] Add tests for the Asm type checker
Alex Bradbury [Tue, 22 Mar 2022 15:50:56 +0000 (15:50 +0000)]
 [WebAssembly] Add tests for the Asm type checker

While looking at bugs like PR54022, I noted that there is no real test
coverage for the asm type checker. This patch starts to address that,
adding a series of tests for the errors messages produced, as well as
some FIXMEs as an XFAIL test for some current issues.

It's not intended to be an exhaustive test, but does have test cases for
each of the instructions that the type checker has specific handling
for.

Differential Revision: https://reviews.llvm.org/D122020

2 years ago[simple] fix some the documentation
Walter Erquinigo [Tue, 22 Mar 2022 16:08:47 +0000 (09:08 -0700)]
[simple] fix some the documentation

Some links were not rendered correctly in the intel pt documentation and
some spacing was fixed in some command objects.

2 years ago[CGOpenMPRuntime] Remove some uses of deprecated Adddress ctor
Nikita Popov [Tue, 22 Mar 2022 15:19:00 +0000 (16:19 +0100)]
[CGOpenMPRuntime] Remove some uses of deprecated Adddress ctor

2 years ago[flang][NFC] Remove unused variable
Valentin Clement [Tue, 22 Mar 2022 15:11:42 +0000 (16:11 +0100)]
[flang][NFC] Remove unused variable

Fix for buildbot failure shown after fe252f8ed6369acdb13d4e290d3b9dfe2ec4eb8e

2 years ago[Flang] Lower the sqrt intrinsics
Kiran Chandramohan [Tue, 22 Mar 2022 14:13:52 +0000 (14:13 +0000)]
[Flang] Lower the sqrt intrinsics

The intrinsic computes the square root for real and complex numbers. By
default they are lowered to runtime calls to libpgmath. With the llvm
option, it can be lowered to llvm intrinsics (not all types .eg. complex
are supported for llvm lowering).

This is part of the upstreaming effort from the fir-dev branch in [1].
[1] https://github.com/flang-compiler/f18-llvm-project

Reviewed By: schweitz

Differential Revision: https://reviews.llvm.org/D122018

Co-authored-by: Eric Schweitz <eschweitz@nvidia.com>
Co-authored-by: Jean Perier <jperier@nvidia.com>
2 years ago[RISCV] Add policy operand for masked compare and vmsbf/vmsif/vmsof IR
Zakk Chen [Thu, 17 Mar 2022 03:17:06 +0000 (20:17 -0700)]
[RISCV] Add policy operand for masked compare and vmsbf/vmsif/vmsof IR
intrinsics.

Those operations are updated under a tail agnostic policy, but they
could have mask agnostic or undisturbed.

Reviewed By: rogfer01

Differential Revision: https://reviews.llvm.org/D120228

2 years ago[InstCombine] add test for abs with dominating condition; NFC
Sanjay Patel [Tue, 22 Mar 2022 14:40:05 +0000 (10:40 -0400)]
[InstCombine] add test for abs with dominating condition; NFC

There's a potential miscompile or missed optimization with
propagating 'nsw' in the transform proposed in D122013, so
we need at least one more test for coverage.

2 years ago[flang] Lower boxed procedure
Valentin Clement [Tue, 22 Mar 2022 14:40:32 +0000 (15:40 +0100)]
[flang] Lower boxed procedure

In FIR, we want to wrap function pointers in a special box known as a
boxproc value. Fortran has a limited form of dynamic scoping
[https://tinyurl.com/2p8v2hw7] between "host procedures" and "internal
procedures". There are a number of implementations possible.

Boxproc typed values abstract away the implementation details of when a
function pointer can be passed directly (as a raw address) and when a
function pointer has to account for the presence of a dynamic scope.
When lowering Fortran syntax to FIR, all function pointers are emboxed
as boxproc values.

When creating LLVM IR, we must strip away the abstraction and produce
low-level LLVM "assembly" code. This patch implements that
transformation as converting the boxproc values to either raw function
pointers or executable trampolines on the stack as needed. The
trampoline then captures the dynamic scope context within an executable
thunk that can be passed instead of the function's raw address.

Some extra handling is required for Fortran functions that return a
character value to deal with LEN values here.

Some of the code in Bridge.cpp and ConvertExpr.cpp and be re-arranged to
faciliate the upstreaming effort.

This patch is part of the upstreaming effort from fir-dev branch.

Reviewed By: jeanPerier, PeteSteinfeld

Differential Revision: https://reviews.llvm.org/D122223

Co-authored-by: mleair <leairmark@gmail.com>
Co-authored-by: Jean Perier <jperier@nvidia.com>
Co-authored-by: Eric Schweitz <eschweitz@nvidia.com>
Co-authored-by: V Donaldson <vdonaldson@nvidia.com>
Co-authored-by: Kiran Chandramohan <kiran.chandramohan@arm.com>
2 years ago[InstCombine] add tests for abs with dominating condition; NFC
chenglin.bi [Tue, 22 Mar 2022 14:34:21 +0000 (10:34 -0400)]
[InstCombine] add tests for abs with dominating condition; NFC

Baseline tests for D122013 (issue #54132).

2 years ago[CGOpenMPRuntime] Remove some uses of deprecated Address ctor
Nikita Popov [Tue, 22 Mar 2022 14:10:33 +0000 (15:10 +0100)]
[CGOpenMPRuntime] Remove some uses of deprecated Address ctor

2 years ago[clang-format] don't break up #-style comment sections
Krasimir Georgiev [Tue, 22 Mar 2022 14:28:34 +0000 (15:28 +0100)]
[clang-format] don't break up #-style comment sections

Follow-up from https://github.com/llvm/llvm-project/commit/36d13d3f8adb3d1a6bae71370afa23d11a94dc78; https://reviews.llvm.org/D121451.

Restore the old behavior in situations where we use # as comments and long strings of #'s for comment sections.

Reviewed By: MyDeveloperDay

Differential Revision: https://reviews.llvm.org/D122230

2 years ago[LV] Remove Loop argument from createInductionResumeValues (NFCI).
Florian Hahn [Tue, 22 Mar 2022 14:23:11 +0000 (14:23 +0000)]
[LV] Remove Loop argument from createInductionResumeValues (NFCI).

createInductionResumeValues only uses its loop argument only to get the
pre-header, but the pre-header is already known (we created/cached it
earlier). Remove the unneeded loop argument.

2 years ago[lldb/test] Increase pexpect termination timeouts
Pavel Labath [Tue, 22 Mar 2022 13:52:02 +0000 (14:52 +0100)]
[lldb/test] Increase pexpect termination timeouts

By default these timeouts are extremely small (0.1s). This means that
100ms after sending an EOF, pexpect will start sending the process
increasingly aggressive signals, but the small timeouts mean that (on a
loaded machine) the kernel may not have enough time to process the
signal even if the overall effect of the signal is to kill the
application.

It turns out we were already relying on this signals (instead of regular
EOF quits) in our tests. In my experiments it was sufficient to block
SIGINT and SIGHUP to cause some test to become flaky. This was most
likely the reason of a couple of flakes on the lldb-x86_64-debian bot,
and is probably the reason why the pexpect tests are flaky on several
other (e.g. asan) bots.

This patch increses the timeout to 6 seconds (60-fold increase), which
is hopefully sufficient to avoid flakes even in the most extreme
situations.

2 years ago[Flang] Lower the exp, log, log10 intrinsics
Kiran Chandramohan [Tue, 22 Mar 2022 14:04:02 +0000 (14:04 +0000)]
[Flang] Lower the exp, log, log10 intrinsics

The intrinsic computes the exponent, log real and complex numbers and
log10 for real numbers. By default they are lowered to runtime calls to
libpgmath. kind=10 and 16 are not supported. With the llvm option, it
can be lowered to llvm intrinsics (not all types .eg. complex are
supported for llvm lowering).

This is part of the upstreaming effort from the fir-dev branch in [1].
[1] https://github.com/flang-compiler/f18-llvm-project

Reviewed By: PeteSteinfeld

Differential Revision: https://reviews.llvm.org/D122132

Co-authored-by: Eric Schweitz <eschweitz@nvidia.com>
Co-authored-by: William S Moses <gh@wsmoses.com>
2 years ago[CGOpenMPRuntimeGPU] Remove uses of deprecated address constructor
Nikita Popov [Tue, 22 Mar 2022 13:38:27 +0000 (14:38 +0100)]
[CGOpenMPRuntimeGPU] Remove uses of deprecated address constructor

Worth noting that the code marked with FIXME is dead and would
produce invalid IR if hit. Someone familiar with this code should
probably look into that.

2 years agoFix _BitInt suffix width calculation
Aaron Ballman [Tue, 22 Mar 2022 14:00:05 +0000 (10:00 -0400)]
Fix _BitInt suffix width calculation

@mgehre-amd pointed out the following post-commit review feedback on
the changes in 8cba72177dcd8de5d37177dbaf2347e5c1f0f1e8:

As an example, the paper says 3wb /* Yields an _BitInt(3); two value
bits, one sign bit */.
So I would expect that 0xFwb gives _BitInt(5); four value bits, one
sign bit, but with this implementation I get _BitInt(2).
This is because ResultVal as 4 bits, and getMinSignedBits() inteprets
it as negative and thus says that 1 bit is enough to represent -1.

This corrects the behavior for calculating the bit-width and adds some
test coverage.

2 years ago[LTO] Add configuartion option to use default optimization pipeline
Joseph Huber [Tue, 22 Mar 2022 13:28:12 +0000 (09:28 -0400)]
[LTO] Add configuartion option to use default optimization pipeline

This patch adds a configuration option to simply use the default pass
pipeline in favor of the LTO-specific one. We observed some severe
performance penalties when uding device-side LTO for OpenMP offloading
applications caused by the LTO-pass pipeline. This is primarily because
OpenMP uses an LLVM bitcode library to implement a GPU runtime library.
In a standard compilation we link this bitcode library into each source
file and optimize it with the default pipeline. When performing LTO we
link it late with all the files, but the bitcode library never has the
regular optimization pipeline applied to it so we miss a few
optimizations just using the LTO pipeline to optimize it.

I'm not committed to this solution, but it's the easiest method to solve
this performance regression when using LTO without changing the
optimizatin pipeline for other users.

Reviewed By: tianshilei1992

Differential Revision: https://reviews.llvm.org/D122133

2 years ago[MLIR][Prebsurger] Add IntegerRelation::intersect supporting locals properly
Arjun P [Thu, 17 Mar 2022 22:25:40 +0000 (22:25 +0000)]
[MLIR][Prebsurger] Add IntegerRelation::intersect supporting locals properly

Reviewed By: Groverkss

Differential Revision: https://reviews.llvm.org/D122149

2 years ago[MLIR][Presburger] MultiAffineFunction::removeIdRange: fix bug where kind wasn't...
Arjun P [Mon, 21 Mar 2022 17:13:19 +0000 (17:13 +0000)]
[MLIR][Presburger] MultiAffineFunction::removeIdRange: fix bug where kind wasn't passed on to IntegerPolyhedron::removeIdRange

Reviewed By: Groverkss

Differential Revision: https://reviews.llvm.org/D122158

2 years ago[MLIR][Presburger] fix bug where Simplex::addZeroRow was not undoable
Arjun P [Fri, 18 Mar 2022 12:29:29 +0000 (12:29 +0000)]
[MLIR][Presburger] fix bug where Simplex::addZeroRow was not undoable

Previously, an UndoLogEntry was added by addRow but not by addZeroRow. So
calling directly into addZeroRow, as LexSimplex::addCut does, was not an
undoable operation. In the current usage of addCut this could never
lead to an incorrect result, and addZeroRow is protected, so it is not
currently possible to add a regression test for this. This bug needs to be
fixed for the symbolic integer lexmin algorithm.

Reviewed By: Groverkss

Differential Revision: https://reviews.llvm.org/D122162

2 years ago[InstCombine] try to canonicalize logical shift after bswap
Sanjay Patel [Tue, 22 Mar 2022 13:02:28 +0000 (09:02 -0400)]
[InstCombine] try to canonicalize logical shift after bswap

When shifting by a byte-multiple:
bswap (shl X, C) --> lshr (bswap X), C
bswap (lshr X, C) --> shl (bswap X), C

This is an IR implementation of a transform suggested in D120648.
The "swaps cancel" test models the motivating optimization from
that proposal.

Alive2 checks (as noted in the other review, we could use
knownbits to handle shift-by-variable-amount, but that can be an
enhancement patch):
https://alive2.llvm.org/ce/z/pXUaRf
https://alive2.llvm.org/ce/z/ZnaMLf

Differential Revision: https://reviews.llvm.org/D122010

2 years ago[Debugify] Use DebugifyLevel in Debugify original mode
Djordje Todorovic [Tue, 22 Mar 2022 11:16:30 +0000 (12:16 +0100)]
[Debugify] Use DebugifyLevel in Debugify original mode

Before this patch the DebugifyLevel option was used for
the synthetic mode, so after this, it will be used in
the original mode as well.

Differential Revision: https://reviews.llvm.org/D115623

2 years ago[LICM] Handle store of pointer to itself (PR54495)
Nikita Popov [Tue, 22 Mar 2022 12:57:54 +0000 (13:57 +0100)]
[LICM] Handle store of pointer to itself (PR54495)

Rather than iterating over users and comparing operands, iterate
over uses and check operand number. Otherwise, we'll end up
promoting a store twice if it has two equal operands.

This can only happen with opaque pointers, as otherwise both
operands differ by a level of indirection, so a bitcast would have
to be involved.

Fixes https://github.com/llvm/llvm-project/issues/54495.

2 years ago[NVPTX][tests] Do not run tests that require direct object generation
Igor Kudrin [Tue, 22 Mar 2022 10:14:56 +0000 (14:14 +0400)]
[NVPTX][tests] Do not run tests that require direct object generation

NVPTX does not support generating binary files, which is required for
these tests.

The majority of tests in 'DebugInfo/Generic' also require emitting
object files, so they all are disabled for NVPTX.

Differential Revision: https://reviews.llvm.org/D121996

2 years ago[tests] Make 'object-emission' imply 'default_triple'
Igor Kudrin [Tue, 22 Mar 2022 10:14:25 +0000 (14:14 +0400)]
[tests] Make 'object-emission' imply 'default_triple'

If 'config.target_triple' is empty, there is no sense to define the
'object-emission' tag.

Differential Revision: https://reviews.llvm.org/D121994

2 years ago[NVPTX] Avoid a crash when 'llc' is called with '-filetype=null'
Igor Kudrin [Tue, 22 Mar 2022 10:13:48 +0000 (14:13 +0400)]
[NVPTX] Avoid a crash when 'llc' is called with '-filetype=null'

For '-filetype=null', 'NVPTXTargetStreamer' is not created, so the
return value of 'OutStreamer->getTargetStreamer()' should be checked
before calling the methods.

Differential Revision: https://reviews.llvm.org/D122001

2 years ago[tests] Force (some) X86-specific tests to use an explicit triple
Igor Kudrin [Tue, 22 Mar 2022 10:13:37 +0000 (14:13 +0400)]
[tests] Force (some) X86-specific tests to use an explicit triple

These tests are located in 'X86' subfolders which means that they should
be compiled for that target. As they did not have the target specified
explicitly, they in fact were compiled for a default target triple. Not
all targets support all required features for these tests; for example,
if NVPTX is used as a default triple, the tests fail. The patch makes the
tests run for 'x86_64', thus they pass regardless of the default target.

Differential Revision: https://reviews.llvm.org/D121998

2 years ago[ThinLTO] Work around buggy FileCheck pattern; NFC
Bryan Chan [Tue, 22 Mar 2022 12:36:42 +0000 (08:36 -0400)]
[ThinLTO] Work around buggy FileCheck pattern; NFC

Update the FileCheck patterns in a test case to prevent a path name
containing the `@` character from causing it to fail unnecessarily,
e.g. during a Jenkins CI job.

2 years ago[analyzer] Refactor makeNull to makeNullWithWidth (NFC)
Vince Bridgers [Wed, 9 Feb 2022 00:22:32 +0000 (18:22 -0600)]
[analyzer] Refactor makeNull to makeNullWithWidth (NFC)

Usages of makeNull need to be deprecated in favor of makeNullWithWidth
for architectures where the pointer size should not be assumed. This can
occur when pointer sizes can be of different sizes, depending on address
space for example. See https://reviews.llvm.org/D118050 as an example.

This was uncovered initially in a downstream compiler project, and
tested through those systems tests.

steakhal performed systems testing across a large set of open source
projects.

Co-authored-by: steakhal
Resolves: https://github.com/llvm/llvm-project/issues/53664

Reviewed By: NoQ, steakhal

Differential Revision: https://reviews.llvm.org/D119601

2 years ago[InstCombine] try to narrow shifted bswap-of-zext
Sanjay Patel [Tue, 22 Mar 2022 11:52:39 +0000 (07:52 -0400)]
[InstCombine] try to narrow shifted bswap-of-zext

This is the IR counterpart to 370ebc9d9a573d6
which provided a bswap narrowing fix for issue #53867.

Here we can be more general (although I'm not sure yet
what would happen for illegal types in codegen - too
rare to worry about?):
https://alive2.llvm.org/ce/z/3-CPfo

This will be more effective if we have moved the shift
after the bswap as proposed in D122010, but it is
independent of that patch.

Differential Revision: https://reviews.llvm.org/D122166

2 years ago[InstCombine] add tests for shift-of-bswap; NFC
Sanjay Patel [Fri, 18 Mar 2022 20:08:37 +0000 (16:08 -0400)]
[InstCombine] add tests for shift-of-bswap; NFC

2 years ago[AArch64] Add extra insert subvector cost model tests. NFC
David Green [Tue, 22 Mar 2022 12:20:19 +0000 (12:20 +0000)]
[AArch64] Add extra insert subvector cost model tests. NFC

2 years ago[AMDGPU] use scalar shift for SALU users in frame index elimination
alex-t [Tue, 22 Mar 2022 12:13:15 +0000 (13:13 +0100)]
[AMDGPU] use scalar shift for SALU users in frame index elimination

In the frame index lowering we have to insert shift and add
instructions to adjust stack object access.  We need to take care of the stack
object user kind and use scalar shift/add for scalar users.

Reviewed By: rampitec

Differential Revision: https://reviews.llvm.org/D121524

2 years ago[Debugify] Optimize debugify original mode
Djordje Todorovic [Tue, 22 Mar 2022 09:14:36 +0000 (10:14 +0100)]
[Debugify] Optimize debugify original mode

Before we start addressing the issue with having
a lot of false positives when using debugify in
the original mode, we have made a few patches that
should speed up the execution of the testing
utility Passes.

For example, when testing a large project
(let's say LLVM project itself), we can face
a lot of potential DI issues. Usually, we use
-verify-each-debuginfo-preserve (that is very
similar to -debugify-each) -- it collects
DI metadata before each Pass, and after the Pass
it checks if the Pass preserved the DI metadata.
However, we can speed up this process, since we
don't need to collect DI metadata before each
Pass -- we could use the DI metadata that are
collected after the previous Pass from
the pipeline as an input for the next Pass.

This patch speeds up the utility for ~2x.

Differential Revision: https://reviews.llvm.org/D115622

2 years ago[flang][OpenMP] Lowering critical construct
Shraiysh Vaishay [Tue, 22 Mar 2022 09:47:52 +0000 (15:17 +0530)]
[flang][OpenMP] Lowering critical construct

This patch adds translation from PFT to FIR for critical construct.

This is part of the upstreaming effort from the fir-dev branch in [1].
[1] https://github.com/flang-compiler/f18-llvm-project

Co-authored-by: kiranchandramohan <kiranchandramohan@gmail.com>
Reviewed By: kiranchandramohan

Differential Revision: https://reviews.llvm.org/D122218

2 years ago[X86][SandyBridge] Remove superfluous mmx store from vector load schedule model group
Simon Pilgrim [Tue, 22 Mar 2022 10:48:19 +0000 (10:48 +0000)]
[X86][SandyBridge] Remove superfluous mmx store from vector load schedule model group

Noticed by D122216

2 years ago[AMDGPU] use scalar shift for SALU users in frame index elimination
alex-t [Sat, 12 Mar 2022 14:38:11 +0000 (17:38 +0300)]
[AMDGPU] use scalar shift for SALU users in frame index elimination

In the frame index lowering we have to insert shift and add
instructions to adjust stack object access.  We need to take care of the stack
object user kind and use scalar shift/add for scalar users.

Reviewed By: rampitec

Differential Revision: https://reviews.llvm.org/D121524

2 years ago[VP] Fix VPintrinsic::getStaticVectorLength for vp.merge|select
Simon Moll [Tue, 22 Mar 2022 10:41:14 +0000 (11:41 +0100)]
[VP] Fix VPintrinsic::getStaticVectorLength for vp.merge|select

VPIntrinsic::getStaticVectorLength infers the operational vector length
of a VPIntrinsic instance from a type that is used with the intrinsic.
The function used the mask operand before. Yet, vp.merge|select do not
have a mask operand (in the predicating sense that the other VP
intrinsics are using them - it is a selection mask for them). Fallback
to the return type to fix this.

Reviewed By: kaz7

Differential Revision: https://reviews.llvm.org/D121913

2 years ago[CGStmtOpenMP] Remove uses of deprecated Address constructor
Nikita Popov [Tue, 22 Mar 2022 09:25:35 +0000 (10:25 +0100)]
[CGStmtOpenMP] Remove uses of deprecated Address constructor

2 years ago[X86] Rename MMX_MOVD64from64rm to MMX_MOVD64from64mr b/c it stores sth, NFC
Shengchen Kan [Tue, 22 Mar 2022 08:49:45 +0000 (16:49 +0800)]
[X86] Rename MMX_MOVD64from64rm to MMX_MOVD64from64mr b/c it stores sth, NFC

Reviewed By: pengfei, RKSimon

Differential Revision: https://reviews.llvm.org/D122216

2 years agoFix missing include under -DEXPENSIVE_CHECK
serge-sans-paille [Tue, 22 Mar 2022 09:37:17 +0000 (10:37 +0100)]
Fix missing include under -DEXPENSIVE_CHECK

Regression introduced by f1985a3f855d3676c5aad0e5c258d2ea38598f44

2 years ago[RISCV] Add policy operand for masked vid and viota IR intrinsics.
Zakk Chen [Mon, 21 Feb 2022 08:54:46 +0000 (00:54 -0800)]
[RISCV] Add policy operand for masked vid and viota IR intrinsics.

Reviewed By: rogfer01

Differential Revision: https://reviews.llvm.org/D120227

2 years agoReland "[pseudo] Split greatergreater token."
Haojian Wu [Mon, 21 Mar 2022 14:55:46 +0000 (15:55 +0100)]
Reland "[pseudo] Split greatergreater token."

It was reverted, because the test had a lift-time issue.
Reland f66d3758bda99e9f57bfdad168212feda18792ae with a fix.

2 years ago[WebAssembly] Always emit functype directives for defined functions
Alex Bradbury [Tue, 22 Mar 2022 09:11:46 +0000 (09:11 +0000)]
[WebAssembly] Always emit functype directives for defined functions

This fixes bug <https://github.com/llvm/llvm-project/issues/54022>. For
now this means that defined functions will have two .functype directives
emitted. Given discussion in that bug has suggested interest in moving
towards using something other than .functype to mark the beginning of a
function (which would, as a side-effect, solve this issue), this patch
doesn't attempt to avoid that duplication.

Some test cases that used CHECK-LABEL: foo rather than CHECK-LABEL: foo:
are broken by this change. This patch updates those test cases to always
have a colon at the end of the CHECK-LABEL string.

Differential Revision: https://reviews.llvm.org/D122134

2 years ago[CodeGen][RISCV] Avoid deprecated address constructor
Nikita Popov [Tue, 22 Mar 2022 09:08:04 +0000 (10:08 +0100)]
[CodeGen][RISCV] Avoid deprecated address constructor

2 years ago[clang-tidy] Don't try to build CTTestTidyModule for Windows with dylibs
Martin Storsjö [Tue, 15 Mar 2022 11:30:35 +0000 (13:30 +0200)]
[clang-tidy] Don't try to build CTTestTidyModule for Windows with dylibs

In MinGW mode, it's possible to build LLVM/Clang with
LLVM_LINK_LLVM_DYLIB (which implicitly enables plugins too). Other
existing ways of building plugins on Windows is to build with
LLVM_EXPORT_SYMBOLS_FOR_PLUGINS, where each executable exports its
symbols.

With LLVM_LINK_LLVM_DYLIB, we can't generally skip building plugins
even if they are set up with PLUGIN_TOOL, as some plugins (e.g.
under clang/examples) set up that way do build properly (as
they manually call clang_target_link_libraries, which links in the
libclang-cpp.dll dylib).

For CTTestTidyModule, there's no corresponding dylib that would
provide the same exports.

Differential Revision: https://reviews.llvm.org/D121687

2 years agoCleanup includes: Transforms/IPO
serge-sans-paille [Mon, 21 Mar 2022 20:53:28 +0000 (21:53 +0100)]
Cleanup includes: Transforms/IPO

Preprocessor output diff: -238205 lines
Discourse thread: https://discourse.llvm.org/t/include-what-you-use-include-cleanup
Differential Revision: https://reviews.llvm.org/D122183

2 years ago[CodeGen] Remove some uses of deprecated Address constructor
Nikita Popov [Tue, 22 Mar 2022 08:52:12 +0000 (09:52 +0100)]
[CodeGen] Remove some uses of deprecated Address constructor

Remove two stray uses in CodeGenModule and CGCUDANV.

2 years agoCleanup includes: Linker
serge-sans-paille [Mon, 21 Mar 2022 20:52:58 +0000 (21:52 +0100)]
Cleanup includes: Linker

Preprocessor output diff: -7300 lines
Discourse thread: https://discourse.llvm.org/t/include-what-you-use-include-cleanup
Differential Revision: https://reviews.llvm.org/D122182

2 years ago[CodeGen] Avoid deprecated Address ctor in EmitLoadOfPointer()
Nikita Popov [Fri, 18 Mar 2022 12:06:42 +0000 (13:06 +0100)]
[CodeGen] Avoid deprecated Address ctor in EmitLoadOfPointer()

This requires some adjustment in caller code, because there was
a confusion regarding the meaning of the PtrTy argument: This
argument is the type of the pointer being loaded, not the addresses
being loaded from.

2 years ago[CodeGen][OpenMP] Make EmitLoadOfPointer() type consistent
Nikita Popov [Mon, 21 Mar 2022 16:52:40 +0000 (17:52 +0100)]
[CodeGen][OpenMP] Make EmitLoadOfPointer() type consistent

If necessary insert a bitcast beforehand, so the LLVM-level pointer
type and the Clang-level pointer type line up.

2 years ago[clang][dataflow] Model the behavior of optional and std swap
Stanislav Gatev [Mon, 21 Mar 2022 12:06:16 +0000 (12:06 +0000)]
[clang][dataflow] Model the behavior of optional and std swap

Differential Revision: https://reviews.llvm.org/D122129

Reviewed-by: ymandel, xazax.hun
2 years ago[RISCV] Support mask policy for RVV IR intrinsics.
Zakk Chen [Sat, 19 Feb 2022 06:19:15 +0000 (22:19 -0800)]
[RISCV] Support mask policy for RVV IR intrinsics.

Add the UsesMaskPolicy flag to indicate the operations result
would be effected by the mask policy. (ex. mask operations).

It means RISCVInsertVSETVLI should decide the mask policy according
by mask policy operand or passthru operand.
If UsesMaskPolicy is false (ex. unmasked, store, and reduction operations),
the mask policy could be either mask undisturbed or agnostic.
Currently, RISCVInsertVSETVLI sets UsesMaskPolicy operations default to
MA, otherwise to MU to keep the current mask policy would not be changed
for unmasked operations.

Add masked-tama, masked-tamu, masked-tuma and masked-tumu test cases.
I didn't add all operations because most of implementations are using
the same pseudo multiclass. Some tests maybe be duplicated in different
tests. (ex. masked vmacc with tumu shows in vmacc-rv32.ll and masked-tumu)
I think having different tests only for policy would make the testing
clear.

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D120226

2 years ago[RISCV][NFC] Add some check prefixes to remove redundant checks in some IR tests
Lian Wang [Tue, 22 Mar 2022 07:32:18 +0000 (07:32 +0000)]
[RISCV][NFC] Add some check prefixes to remove redundant checks in some IR tests

Reviewed By: frasercrmck, jacquesguan

Differential Revision: https://reviews.llvm.org/D122211

2 years ago[CMake][Fuchsia] Switch to lld on Apple platforms
Petr Hosek [Fri, 18 Mar 2022 23:07:35 +0000 (16:07 -0700)]
[CMake][Fuchsia] Switch to lld on Apple platforms

lld Mach-O backend supports all our use cases now.

Differential Revision: https://reviews.llvm.org/D122047

2 years ago[X86] Simplify attributes of the generated memory folding table (NFCI)
Shengchen Kan [Tue, 22 Mar 2022 07:58:52 +0000 (15:58 +0800)]
[X86] Simplify attributes of the generated memory folding table (NFCI)

This reduces the gaps between tables in X86GenFoldTables.inc and X86InstrFoldTables.cpp

2 years ago[mlir][affine] Add affine.min / affine.max canonicalization.
gysit [Tue, 22 Mar 2022 06:57:02 +0000 (06:57 +0000)]
[mlir][affine] Add affine.min / affine.max canonicalization.

The revision introduces a affine.min and affine.max canonicalization pattern that orders the result expressions. It flattens the result expressions to arrays of dimension and symbol coefficients plus one constant coefficient and rearranges them in lexicographic order.

Without the pattern, CSE will not eliminate two affine.min / affine.max operation if the results are ordered differently. For example, the operations
```
  %1 = affine.min affine_map<(d0) -> (8, -d0 + 27)>(%arg4)
  %2 = affine.min affine_map<(d0) -> (-d0 + 27, 8)>(%arg4)
```
doe not CSE. After applying the pattern, the two operations are equivalent
```
  %1 = affine.min affine_map<(d0) -> (8, -d0 + 27)>(%arg4)
  %2 = affine.min affine_map<(d0) -> (8, -d0 + 27)>(%arg4)
```
which enables CSE.

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D121819

2 years ago[libc][Obvious] Remove an unnecessary dep and use inline_memcpy.
Siva Chandra Reddy [Tue, 22 Mar 2022 07:04:21 +0000 (07:04 +0000)]
[libc][Obvious] Remove an unnecessary dep and use inline_memcpy.

An unnecessary dep of the getenv function is removed. From the x86_64
loader, a call to __llvm_libc::memcpy is replaced with call to
__llvm_libc::inline_memcpy.

2 years ago[NFC] [Coroutines] Remove unnecessary check and constraints on SmallVector
Chuanqi Xu [Tue, 22 Mar 2022 06:20:38 +0000 (14:20 +0800)]
[NFC] [Coroutines] Remove unnecessary check and constraints on SmallVector

The CoroSplit pass would check the existence of coroutine intrinsic
before starting work. It is not necessary and wasteful since it would
iterate over the Module.

This patch also removes the constraint on the corresponding of the
SmallVector for the possible coroutines in the Modules. The original
value is 4. Given coroutines is used actually in practice. 4 is really
relatively a low threshold.

2 years ago[RISCV] Add basic cost model for vector casting
Yeting Kuo [Wed, 16 Mar 2022 05:09:12 +0000 (13:09 +0800)]
[RISCV] Add basic cost model for vector casting

To perform the cost model of vector casting, the patch consider most vector
casts as their scalar form and consider those vector form of free scalr castings
as 1.

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D121771

2 years ago[TableGen] Fix a misuse of getValueAsBitsInit
Sheng [Tue, 22 Mar 2022 05:47:18 +0000 (05:47 +0000)]
[TableGen] Fix a misuse of getValueAsBitsInit

`getValueAsBitsInit` will assert when the "SoftFail" isn't presented.

But given the 'if' statement below, we should've allowed this situation.

This patch fix this.

2 years agoRevert "[bootstrap] Allow passing options to sub-builds for all targets"
Petr Hosek [Tue, 22 Mar 2022 01:34:59 +0000 (18:34 -0700)]
Revert "[bootstrap] Allow passing options to sub-builds for all targets"

This reverts commit 240e06dfe77feabe38a03cbb1c94875639d0f9ff.

2 years ago[mlir] Printing oilist element
Shraiysh Vaishay [Tue, 22 Mar 2022 04:38:33 +0000 (10:08 +0530)]
[mlir] Printing oilist element

This patch attempts to deduce when the oilist element must be printed
based on the optional arguments to it. This especially helps creating
an operation accurately because with the current implementation, the
inferred unit attributes must be manually added to print the clauses
appropriately.

Reviewed By: Mogball

Differential Revision: https://reviews.llvm.org/D121579

2 years ago[RISCV][NFC] Add common check prefix to reduce duplicate check lines.
jacquesguan [Mon, 21 Mar 2022 09:29:16 +0000 (17:29 +0800)]
[RISCV][NFC] Add common check prefix to reduce duplicate check lines.

Differential Revision: https://reviews.llvm.org/D122120

2 years ago[analyzer][ctu] Fix wrong 'multiple definitions' errors caused by space characters...
Ella Ma [Tue, 18 May 2021 06:22:46 +0000 (14:22 +0800)]
[analyzer][ctu] Fix wrong 'multiple definitions' errors caused by space characters in lookup names when parsing the ctu index file

This error was found when analyzing MySQL with CTU enabled.

When there are space characters in the lookup name, the current
delimiter searching strategy will make the file path wrongly parsed.
And when two lookup names have the same prefix before their first space
characters, a 'multiple definitions' error will be wrongly reported.

e.g. The lookup names for the two lambda exprs in the test case are
`c:@S@G@F@G#@Sa@F@operator int (*)(char)#1` and
`c:@S@G@F@G#@Sa@F@operator bool (*)(char)#1` respectively. And their
prefixes are both `c:@S@G@F@G#@Sa@F@operator` when using the first space
character as the delimiter.

Solving the problem by adding a length for the lookup name, making the
index items in the format of `<USR-Length>:<USR File> <Path>`.

---

In the test case of this patch, we found that it will trigger a "triple
mismatch" warning when using `clang -cc1` to analyze the source file
with CTU using the on-demand-parsing strategy in Darwin systems. And
this problem is also encountered in D75665, which is the patch
introducing the on-demand parsing strategy.
We temporarily bypass this problem by using the loading-ast-file
strategy.

Refer to the [discourse topic](https://discourse.llvm.org/t/60762) for
more details.

Differential Revision: https://reviews.llvm.org/D102669