platform/upstream/llvm.git
18 months agoInstSimplify: Perform cheaper check first
Matt Arsenault [Tue, 11 Apr 2023 20:39:39 +0000 (16:39 -0400)]
InstSimplify: Perform cheaper check first

18 months agoValueTracking: Handle extractelement and extractvalue in computeKnownFPClass
Matt Arsenault [Wed, 12 Apr 2023 23:09:17 +0000 (19:09 -0400)]
ValueTracking: Handle extractelement and extractvalue in computeKnownFPClass

18 months agoValueTracking: Implement computeKnownFPClass for sin/cos
Matt Arsenault [Sat, 8 Apr 2023 23:15:26 +0000 (19:15 -0400)]
ValueTracking: Implement computeKnownFPClass for sin/cos

18 months ago[Docs] Update ClangFormatStyleOptions.rst's versionbadge's to have min-width
Shivam Gupta [Fri, 14 Apr 2023 18:14:39 +0000 (23:44 +0530)]
[Docs] Update ClangFormatStyleOptions.rst's versionbadge's to have min-width

This fix https://github.com/llvm/llvm-project/issues/61934

By default, the width property is used to set the width of a table column.
However, if the content of a cell in that column is wider than the width
specified by the width property, the cell's content will overflow the
column and the table will become wider than the specified width.
This causes the version numbers to be displayed outside of their table cells.

Using the min-width property instead of width ensures that the column is
wide enough to accommodate the content of its cells. If a cell's content is
wider than the specified min-width, the column will expand to fit the content.

18 months ago[mlir][linalg] Fix a bug in lower_pack when there are no padding values.
Hanhan Wang [Fri, 14 Apr 2023 18:23:16 +0000 (11:23 -0700)]
[mlir][linalg] Fix a bug in lower_pack when there are no padding values.

Reviewed By: chelini

Differential Revision: https://reviews.llvm.org/D148061

18 months ago[StringView] remove ctor incompatible with std::string_view
Nick Desaulniers [Fri, 14 Apr 2023 18:11:25 +0000 (11:11 -0700)]
[StringView] remove ctor incompatible with std::string_view

Towards replacing llvm::StringView with std::string_view, remove ctor
that std::string_view doesn't have an analog for.

Reviewed By: erichkeane, MaskRay

Differential Revision: https://reviews.llvm.org/D148353

18 months ago[InstCombine] Make `FoldOpIntoSelect` handle non-constants and use condition to deduc...
Noah Goldstein [Fri, 14 Apr 2023 17:35:14 +0000 (12:35 -0500)]
[InstCombine] Make `FoldOpIntoSelect` handle non-constants and use condition to deduce constants.

Make the fold use the information present in the condition for deducing constants i.e:
```
%c = icmp eq i8 %x, 10
%s = select i1 %c, i8 3, i8 2
%r = mul i8 %x, %s
```

If we fold the `mul` into the select, on the true side we insert `10` for `%x` in the `mul`.

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D146349

18 months ago[InstCombine] Add tests for deducing constants based on condition in select; NFC
Noah Goldstein [Fri, 14 Apr 2023 17:34:53 +0000 (12:34 -0500)]
[InstCombine] Add tests for deducing constants based on condition in select; NFC

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D148180

18 months ago[LIBC] Actually assert no errors in `pthread_call_once` in its test; NFC
Noah Goldstein [Fri, 14 Apr 2023 17:34:27 +0000 (12:34 -0500)]
[LIBC] Actually assert no errors in `pthread_call_once` in its test; NFC

Just seemed to be missing an assertion.

Reviewed By: sivachandra

Differential Revision: https://reviews.llvm.org/D148289

18 months ago[LIBC] Clarify namespace of `thread_exit` in `pthread_exit`
Noah Goldstein [Fri, 14 Apr 2023 17:33:59 +0000 (12:33 -0500)]
[LIBC] Clarify namespace of `thread_exit` in `pthread_exit`

Just add `__llvm_libc::` to makes things clearer.

Reviewed By: sivachandra

Differential Revision: https://reviews.llvm.org/D148292

18 months ago[TableGen] Replace std::shared_ptr with InstrusiveRefCntPtr for TreePatternNode. NFC
Craig Topper [Fri, 14 Apr 2023 17:37:14 +0000 (10:37 -0700)]
[TableGen] Replace std::shared_ptr with InstrusiveRefCntPtr for TreePatternNode. NFC

An intrusive reference counter uses less memory than the control
block of std::shared_ptr.

This should allow some additional code simplifications if we
don't need to pass around shared_ptr in order to create new
shared_ptrs.

18 months ago[gn build] Port 5abe338f2a7f
LLVM GN Syncbot [Fri, 14 Apr 2023 17:45:35 +0000 (17:45 +0000)]
[gn build] Port 5abe338f2a7f

18 months ago[clang-tidy][NFC] Remove reference to Visual studio native plugin
Jorge Pinto Sousa [Fri, 14 Apr 2023 17:22:47 +0000 (17:22 +0000)]
[clang-tidy][NFC] Remove reference to Visual studio native plugin

The clang-tidy documentation contained:

> MS Visual Studio has a native clang-tidy-vs plugin

This plugin was removed via 559ae14.

Fixes: https://github.com/llvm/llvm-project/issues/62142

Reviewed By: PiotrZSL

Differential Revision: https://reviews.llvm.org/D148354

18 months ago[clang-tidy][NFC] Improved hungarian notation regression test at post-commit review
Douglas Chen [Fri, 14 Apr 2023 17:22:36 +0000 (17:22 +0000)]
[clang-tidy][NFC] Improved hungarian notation regression test at post-commit review

Improve the (`D144510`)[https://reviews.llvm.org/D144510] patch with prefix string. Using "my" instead of "cust" would increase
readability.

Take some examples:
- const char* `custszNamePtr` = "Name"; --> `myszNamePtr`
- uint8_t `custu8ValueU8` = 0; --> `myu8ValueU8`
- DWORD `custdwMsDword` = 0; --> `mydwMsDword`

Reviewed By: PiotrZSL

Differential Revision: https://reviews.llvm.org/D148314

18 months ago[clang-tidy] Fix false positve for defaulted move constructor in performance-noexcept...
AMS21 [Fri, 14 Apr 2023 17:06:30 +0000 (17:06 +0000)]
[clang-tidy] Fix false positve for defaulted move constructor in performance-noexcept-move-constructor

Previously a struct like this:

template <typename>
struct A { A(A&&) = default; };

Would trigger a false positive, since even though it is not marked as
noexcept it still is due to the `= default`.
Now we only give a warning if the defaulted move constructor is
actually declared as throwing and correctly resolve it if they are
defaulted.

This fixes llvm#56026, llvm#41414, llvm#38081

Reviewed By: PiotrZSL

Differential Revision: https://reviews.llvm.org/D146922

18 months ago[StringView] remove dropFront
Nick Desaulniers [Fri, 14 Apr 2023 17:24:07 +0000 (10:24 -0700)]
[StringView] remove dropFront

Towards converting our use of llvm::StringView to std::string_view,
remove a method that std::string_view doesn't have.
llvm::StringView::dropFront is semantically similar to
std::string_view::substr but with the input clamped to the size. No code
was relying on clamping other than the rust demangler, which I fixed in
https://reviews.llvm.org/D148272.  Removing this method makes it easier
to switch over code later.

Reviewed By: MaskRay

Differential Revision: https://reviews.llvm.org/D148348

18 months ago[lsan] Move allocator base to avoid conflict with high-entropy ASLR for x86-64 Linux
Fangrui Song [Fri, 14 Apr 2023 17:18:54 +0000 (10:18 -0700)]
[lsan] Move allocator base to avoid conflict with high-entropy ASLR for x86-64 Linux

This ports D148280 for ASan.

On x86-64 Linux, when mmap_rnd_bits is set to 32 (the maximum `ARCH_MMAP_RND_BITS_MAX`),
the allocator space `[kAllocatorSpace,kAllocatorSpace+kAllocatorSize)`
collides with the PIE load base range (0x555555554000 upto `2**mmap_rnd_bits * pagesize` away),
which can cause the allocation to fail.
Using 0x500000000000ULL as the base address avoids this problem and works with
AArch64 Linux and FreeBSD as well.

While here, change s390x to use 0x500000000000ULL as well. See D78644
discussions that the address works.

Reviewed By: thurston, vitalybuka

Differential Revision: https://reviews.llvm.org/D148193

18 months ago[Pipeline] Don't limit ArgumentPromotion to -O3
Arthur Eubanks [Thu, 13 Apr 2023 20:25:08 +0000 (13:25 -0700)]
[Pipeline] Don't limit ArgumentPromotion to -O3

This is a cheap pass so there's no need to limit to -O3.
This removes some differences between various pipelines.

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D148269

18 months ago[passes] Remove the legacy PM version of IRCE
Bjorn Pettersson [Fri, 14 Apr 2023 13:04:10 +0000 (15:04 +0200)]
[passes] Remove the legacy PM version of IRCE

Differential Revision: https://reviews.llvm.org/D148338

18 months ago[passes] Remove the legacy PM version of AlignmentFromAssumptions
Bjorn Pettersson [Fri, 14 Apr 2023 13:00:36 +0000 (15:00 +0200)]
[passes] Remove the legacy PM version of AlignmentFromAssumptions

Differential Revision: https://reviews.llvm.org/D148337

18 months ago[Passes] Remove the legacy DemandedBitsWrapperPass
Bjorn Pettersson [Thu, 13 Apr 2023 11:25:43 +0000 (13:25 +0200)]
[Passes] Remove the legacy DemandedBitsWrapperPass

Last user of DemandedBitsWrapperPass was the BDCE pass. Since
the legacy PM version of BDCE was removed in an earlier commit, this
patch removes the now unused DemandedBitsWrapperPass.

Differential Revision: https://reviews.llvm.org/D148336

18 months ago[Passes] Remove legacy PM version of BDCE (aka BitTrackingDCEPass)
Bjorn Pettersson [Thu, 13 Apr 2023 08:42:16 +0000 (10:42 +0200)]
[Passes] Remove legacy PM version of BDCE (aka BitTrackingDCEPass)

BDCE is not used by the codegen pipeline so we should not need the
legacy PM version of the pass any longer.

Differential Revision: https://reviews.llvm.org/D148335

18 months ago[clangd] Handle destructors in DefineOutline tweak
Nathan James [Fri, 14 Apr 2023 16:51:44 +0000 (17:51 +0100)]
[clangd] Handle destructors in DefineOutline tweak

Fix destructors being incorrectly defined in the DefineOutline tweak
Currently it doesn't prepend the class name to the destructor
```lang=c++
class A { ~A() {} };
// Destructor definition after outline
~A() {}
// After this fix
A::~A() {}
```

Reviewed By: kadircet

Differential Revision: https://reviews.llvm.org/D147802

18 months ago[AArch64] Combine SELECT_CC patterns that match smin(a,0) and smax(a,0)
Cameron McInally [Fri, 14 Apr 2023 16:26:16 +0000 (09:26 -0700)]
[AArch64] Combine SELECT_CC patterns that match smin(a,0) and smax(a,0)

With a previous patch to canonicalize SPF to min/max intrinsics (a266af721153),
we saw a performance regression on the AArch64 backend.

This patch recovers from the SPF canonicalization by combining smin(a,0) and
smax(a,0) SELECT_CC patterns during AArch64ISelLowering.

GitHub Issue: #61767

Reviewed By: dmgreen

Differential Revision: https://reviews.llvm.org/D148249

18 months ago[mlir][Linalg] Make HoistRedundantTensorSubsetsOp not a FunctionalStyleTransformOpTrait
Nicolas Vasilache [Fri, 31 Mar 2023 15:00:39 +0000 (08:00 -0700)]
[mlir][Linalg] Make HoistRedundantTensorSubsetsOp not a FunctionalStyleTransformOpTrait

Differential Revision: https://reviews.llvm.org/D148350

18 months ago[flang][openacc] Add kernels construct lowering
Valentin Clement [Fri, 14 Apr 2023 16:43:44 +0000 (09:43 -0700)]
[flang][openacc] Add kernels construct lowering

Lower the parse tree to acc dialects operations. Make use
of the compute construct lowering.

Depends on D148277

Reviewed By: PeteSteinfeld

Differential Revision: https://reviews.llvm.org/D148278

18 months ago[mlir][Tensor] Add support for insert_slice in FoldTensorSubsetOps
Nicolas Vasilache [Fri, 14 Apr 2023 12:44:41 +0000 (05:44 -0700)]
[mlir][Tensor] Add support for insert_slice in FoldTensorSubsetOps

Differential Revision: https://reviews.llvm.org/D148334

18 months ago[RustDemangle] remove StringView::dropFront
Nick Desaulniers [Fri, 14 Apr 2023 16:18:04 +0000 (09:18 -0700)]
[RustDemangle] remove StringView::dropFront

Toward the goal of replacing llvm::StringView with std::string_view,
first replacing users of llvm::StringView::dropFront, this case in the
Rust demangling scheme seemed worth its own commit+review.

Reviewed By: erichkeane, MaskRay

Differential Revision: https://reviews.llvm.org/D148272

18 months ago[lldb][ClangExpression] Filter out non-root namespaces in FindNamespace
Michael Buch [Mon, 3 Apr 2023 10:09:23 +0000 (11:09 +0100)]
[lldb][ClangExpression] Filter out non-root namespaces in FindNamespace

**Summary**

In a program such as:
```
namespace A {
namespace B {
struct Bar {};
}
}

namespace B {
struct Foo {};
}
```
...LLDB would run into issues such as:
```
(lldb) expr ::B::Foo f
error: expression failed to parse:
error: <user expression 0>:1:6: no type named 'Foo' in namespace 'A::B'
::B::Foo f
~~~~~^
```

This is because the `SymbolFileDWARF::FindNamespace` implementation
will return *any* namespace it finds if the `parent_decl_ctx` provided
is empty. In `FindExternalVisibleDecls` we use this API to find the
namespace that symbol `B` refers to. If `A::B` happened to be the one
that `SymbolFileDWARF::FindNamespace` looked at first, we would try
to find `struct Foo` in `A::B`. Hence the error.

This patch proposes a new `SymbolFileDWARF::FindNamespace` API that
will only find a match for top-level namespaces, which is what
`FindExternalVisibleDecls` is attempting anyway; it just never
accounted for multiple namespaces of the same name.

**Testing**

* Added API test-case

Differential Revision: https://reviews.llvm.org/D147436

18 months ago[OpenMP][Docs] Replace broken design document link with the git repo
Joseph Huber [Fri, 14 Apr 2023 16:09:33 +0000 (11:09 -0500)]
[OpenMP][Docs] Replace broken design document link with the git repo

Summary:
At some point we stopped copying this file to the server, but
realistically this is just a static `.pdf` hosted in the LLVM repository
so we can link it directly.

18 months ago[AArch64] Teach the AArch64 backend patterns to generate the RAX1 instruction.
Shreyansh Chouhan [Fri, 14 Apr 2023 16:11:05 +0000 (17:11 +0100)]
[AArch64] Teach the AArch64 backend patterns to generate the RAX1 instruction.

Differential Revision: https://reviews.llvm.org/D147887

18 months ago[lldb] Allow evaluating expressions in C++20 mode
Michael Buch [Fri, 14 Apr 2023 16:01:49 +0000 (17:01 +0100)]
[lldb] Allow evaluating expressions in C++20 mode

This patch allows users to evaluate expressions using
`expr -l c++20`. Currently DWARF keeps the CU's at
`DW_AT_language` at `DW_LANG_C_plus_plus_14` even
when compiling with `-std=c++20`. So even in "C++20
programs" expression evaluation will by default be
performed in `C++11` mode for now.

Enabling `C++14` has been previously attempted at
https://reviews.llvm.org/D80308

There are some remaining issues around evaluating C++20
expressions. Mainly, lack of support for C++20 AST nodes in
`clang::ASTImporter`. But these can be addressed in follow-up
patches.

18 months ago[lldb][Language] Add more language types
Michael Buch [Wed, 1 Feb 2023 03:58:05 +0000 (03:58 +0000)]
[lldb][Language] Add more language types

Adds more languages to the `language_names` list in
preparation for adding support for C++20 expression
evaluation.

The language constants were taken from the DWARFv5
constants defined in LLVM's `Dwarf.def`. Two vendor
constants overlap with the DWARFv5 constants so bump
their values. Their actual value is not important,
whereas keeping the enum values consecutive is (since
they are used for array lookups).

Differential Revision: https://reviews.llvm.org/D143061

18 months ago[fuzzer][test] Disable noasan-strncmp test for AArch64
Antoine Moynault [Fri, 14 Apr 2023 15:32:41 +0000 (15:32 +0000)]
[fuzzer][test] Disable noasan-strncmp test for AArch64

This test fails on several aarch64 bots
  clang-aarch64-lld-2stage ( https://lab.llvm.org/buildbot/#/builders/185/builds/3525 )
  clang-aarch64-full-2stage ( https://lab.llvm.org/buildbot/#/builders/179/builds/5904 )
  clang-aarch64-sve-vla ( https://lab.llvm.org/buildbot/#/builders/197/builds/4519 )
Disable it while this is analyzed.

Reviewed By: DavidSpickett

Differential Revision: https://reviews.llvm.org/D148332

18 months ago[mlir][openacc] Add acc.kernels operation
Valentin Clement [Fri, 14 Apr 2023 15:29:39 +0000 (08:29 -0700)]
[mlir][openacc] Add acc.kernels operation

The acc.kernels operation models the OpenACC kernels construct.
The kernels construct defines a region of a program that is
compiled into a sequence of kernels to be executed on the current device.
The operation is modelled on the acc.parallel operation and will
receive similar updates when the data operands operations will
be implemented.

Reviewed By: PeteSteinfeld

Differential Revision: https://reviews.llvm.org/D148277

18 months ago[libc++] Remove redundant assertion in std::span::subspan
Louis Dionne [Tue, 11 Apr 2023 16:53:39 +0000 (17:53 +0100)]
[libc++] Remove redundant assertion in std::span::subspan

That is already checked later in the function as `__count <= size() - __offset`.

rdar://107884996

Differential Revision: https://reviews.llvm.org/D148030

18 months ago[FPEnv][LangRef] Update doc for strictfp attribute
Kevin P. Neal [Wed, 12 Apr 2023 15:42:44 +0000 (11:42 -0400)]
[FPEnv][LangRef] Update doc for strictfp attribute

Based on the direction of IR Verifier changes in D146845, this documentation
needs to be updated.

Differential Revision: https://reviews.llvm.org/D148138

18 months ago[lldb] Add `operator StringRef` to ConstString
Dave Lee [Wed, 12 Apr 2023 21:44:19 +0000 (15:44 -0600)]
[lldb] Add `operator StringRef` to ConstString

Add a `StringRef` conversion function to `ConstString`.

This will make using llvm, and other non-ConstString, APIs more convenient.

For demonstration, this updates Module.cpp.

Differential Revision: https://reviews.llvm.org/D148175

18 months ago[mlir] Fix test after 2d4e856
Christian Sigg [Fri, 14 Apr 2023 15:04:14 +0000 (17:04 +0200)]
[mlir] Fix test after 2d4e856

The test CHECKs for `-nan`, but printing the sign is optional.

This change removes the sign because it has no meaning.

18 months ago[ConstraintElimination] Add tests to check for transfering facts from sgt to ugt.
Zain Jaffal [Fri, 14 Apr 2023 14:37:06 +0000 (15:37 +0100)]
[ConstraintElimination] Add tests to check for transfering facts from sgt to ugt.

If we have a >s b and we know that b >s 0 then we also can infer that a >u b.

Reviewed By: fhahn

Differential Revision: https://reviews.llvm.org/D148321

18 months ago[OpenMP] Replace HeapToShared's initial value with `poison`
Joseph Huber [Thu, 6 Apr 2023 16:18:47 +0000 (11:18 -0500)]
[OpenMP] Replace HeapToShared's initial value with `poison`

There's a desire to move away from `undef` in LLVM. Currently we want to
have the `addressspace(3)` variables use `poison` instead.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D147719

18 months ago[OpenMP][Docs] Add documentation for using configuration files
Joseph Huber [Mon, 10 Apr 2023 14:50:40 +0000 (09:50 -0500)]
[OpenMP][Docs] Add documentation for using configuration files

We recently reverted a patch that automatically set the rpath on OpenMP
executables. This was used because the `libomptarget.so` library is only
expected to work with the same version of compiler that will be using
it. This patch adds some documentation for how to get similar behaviour
as before using a clang configuration file.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D147943

18 months ago[Flang] Add code owner for the Driver
Kiran Chandramohan [Fri, 14 Apr 2023 14:23:30 +0000 (14:23 +0000)]
[Flang] Add code owner for the Driver

Andrzej proposed the design and led the work for the Flang Driver
over the past couple of years. He is continuing to function as
the primary reviewer and gate-keeper of the Driver.

Reviewed By: sscalpone

Differential Revision: https://reviews.llvm.org/D147573

18 months ago[X86] combineSetCCMOVMSK - fold anyof/noneof movmskps/movmskpd -> testps/testpd
Simon Pilgrim [Fri, 14 Apr 2023 14:37:40 +0000 (15:37 +0100)]
[X86] combineSetCCMOVMSK - fold anyof/noneof movmskps/movmskpd -> testps/testpd

Another part of Issue #60007

18 months ago[X86] Add additional add of mul test (NFC)
Nikita Popov [Fri, 14 Apr 2023 14:28:03 +0000 (16:28 +0200)]
[X86] Add additional add of mul test (NFC)

18 months ago[mlir][math] Expand math.powf to exp, log and multiply
Balaji V. Iyer [Fri, 14 Apr 2023 13:52:17 +0000 (13:52 +0000)]
[mlir][math] Expand math.powf to exp, log and multiply

Powf functions are pushed directly to libm. This is problematic for
situations where libm is not available. This patch will decompose the
powf function into log of exponent multiplied by log of base and raise
it to the exp.

Reviewed By: rsuderman

Differential Revision: https://reviews.llvm.org/D148164

18 months ago[X86] Add test for foldable add of mul (NFC)
Nikita Popov [Fri, 14 Apr 2023 13:56:42 +0000 (15:56 +0200)]
[X86] Add test for foldable add of mul (NFC)

18 months agoRevert " [sanitizer] adapt for 75f1f158812dabc03e70697b6b9c272230bce63d"
Krasimir Georgiev [Fri, 14 Apr 2023 13:55:20 +0000 (13:55 +0000)]
Revert " [sanitizer] adapt for 75f1f158812dabc03e70697b6b9c272230bce63d"

This reverts commit de4c038c7ba2c6a8d529cb094f1a7c3deaae9b75.

The change that triggered this fix got reverted.

18 months ago[AMDGPU] Less aggressively break large PHIs
pvanhout [Fri, 7 Apr 2023 12:54:50 +0000 (14:54 +0200)]
[AMDGPU] Less aggressively break large PHIs

In some cases, breaking large PHIs can very negatively affect
performance (3x more instructions observed in a particular test case).

This patch adds some basic profitability heuristics to help with some of these issues without affecting the "good" cases.
e.g. avoid breaking PHIs if it causes back-and-forth between vector/scalar form for no good reason.

Fixes SWDEV-392803
Fixes SWDEV-393781
Fixes SWDEV-394228

Reviewed By: arsenm

Differential Revision: https://reviews.llvm.org/D147786

18 months ago[mlir][Linalg] Support tensor.parallel_insert_slice in transform.insert_slice_to_copy
Nicolas Vasilache [Fri, 14 Apr 2023 12:43:07 +0000 (05:43 -0700)]
[mlir][Linalg] Support tensor.parallel_insert_slice in transform.insert_slice_to_copy

Differential Revision: https://reviews.llvm.org/D148333

18 months ago[VP] Mark llvm.vp.copysign as a binary op
Luke Lau [Thu, 13 Apr 2023 15:43:36 +0000 (16:43 +0100)]
[VP] Mark llvm.vp.copysign as a binary op

I was working on legalising vector binary ops to their VP counterparts when I noticed that `vp_copysign` wasn't marked as a binary op. It looks like this might have been an oversight, so this marks it as one.

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D148241

18 months ago[lldb] Add a sleep to TestObjectFileJSON
Pavel Labath [Fri, 14 Apr 2023 12:36:27 +0000 (14:36 +0200)]
[lldb] Add a sleep to TestObjectFileJSON

The test fails when the two generated files have the same timestamp
(lldb uses second granularity).

18 months ago[SCEV] Preserve NSW for AddRec multiplied by -1 if it cannot be signed minimum
Dmitry Makogon [Wed, 12 Apr 2023 08:15:47 +0000 (15:15 +0700)]
[SCEV] Preserve NSW for AddRec multiplied by -1 if it cannot be signed minimum

This preserves NSW flag for AddRecs multiplied by -1 if we can prove
via constant ranges that the AddRec cannot be signed minimum.

An explanation:
Let M be signed minimum. If AddRec's range contains M, then M * (-1) will
stay M and (M + 1) * (-1) will be signed maximum, so we get a signed overflow.
In all other cases if an AddRec didn't signed overflow,
then AddRec * (-1) wouldn't too.

Differential Revision: https://reviews.llvm.org/D148084

18 months ago[mlir][scf] WhileOp patterns cleanup
Ivan Butygin [Thu, 13 Apr 2023 16:28:57 +0000 (18:28 +0200)]
[mlir][scf] WhileOp patterns cleanup

Fix review comments from https://reviews.llvm.org/D146252
Merge `WhileRemoveUnusedArgs` pattern with (unused) `WhileUnusedArg`,
use `getConditionOp`, use `SmallPtrSet` and early check, move tests

Differential Revision: https://reviews.llvm.org/D148256

18 months ago[bazel] Fix build after e5f50bd and 62ef97e
Christian Sigg [Fri, 14 Apr 2023 11:24:15 +0000 (13:24 +0200)]
[bazel] Fix build after e5f50bd and 62ef97e

18 months ago[mlir][Analysis] CFGLoopInfo instantiation in C++
Christian Ulmann [Fri, 14 Apr 2023 11:09:54 +0000 (11:09 +0000)]
[mlir][Analysis] CFGLoopInfo instantiation in C++

This commit moves the CFGLoopInfo instantiation into the C++ file to
ensure that it is only compiled once. Instantiating the template
explicitly revealed two missing functions that this commit also adds.

Reviewed By: ftynse

Differential Revision: https://reviews.llvm.org/D148219

18 months agoModel list initialization more directly; fixes an assert with coverage mapping
Aaron Ballman [Fri, 14 Apr 2023 11:18:46 +0000 (07:18 -0400)]
Model list initialization more directly; fixes an assert with coverage mapping

Instead of using the validity of a brace's source location as a flag
for list initialization, this now uses a PointerIntPair to model it so
we do not increase the size of the AST node to track this information.
This allows us to retain the valid source location information, which
fixes the coverage assertion.

Fixes https://github.com/llvm/llvm-project/issues/62105
Differential Revision: https://reviews.llvm.org/D148245

18 months ago[Test] Fix tests showing flags improvement for SCEV sub expressions (NFC)
Dmitry Makogon [Fri, 14 Apr 2023 11:12:25 +0000 (18:12 +0700)]
[Test] Fix tests showing flags improvement for SCEV sub expressions (NFC)

This fixes a failing check in tests added by 05a142cc6fd1.

18 months ago[X86] SimplifyDemandedBitsForTargetNode - improve TESTPS/TESTPD handling for duplicat...
Simon Pilgrim [Fri, 14 Apr 2023 11:03:38 +0000 (12:03 +0100)]
[X86] SimplifyDemandedBitsForTargetNode - improve TESTPS/TESTPD handling for duplicated operands

We often repeat the vector operand in TESTPS(X,X)/TESTPD(X,X) for anyof comparisons - ensure we still only demand the sign bits when the TESTP is the only user of that operand

18 months ago[RISCV][NFC] Add test case for accept hexadecimal floating-point for fli instructions
Kito Cheng [Fri, 14 Apr 2023 10:46:52 +0000 (18:46 +0800)]
[RISCV][NFC] Add test case for accept hexadecimal floating-point for fli instructions

Zfa provide fli instruction to load a floating point immediate value,
and some of those are not easy to write and read by human, so
hexadecimal floating-point format should be a good alternative way to write:

Consider this case: 1^-16 = 1.52587890625e-05 (decimal) vs 0x1p-16 (hexadecimal)

The hexadecimal format is easier to write for human.

Fortunately hexadecimal floating-point constants already supported in
C99, so actually we don't need to add any extra code to support that.

This patch added test case for demonstrate we support that and also make
sure this will be supported in future.

I also gonna to talk with ISA folks to adding hexadecimal floating-point
to the ISA spec, so that user will know fli accept value in this format.

Reviewed By: asb

Differential Revision: https://reviews.llvm.org/D148077

18 months agoRevert "[symbolizer] Change error message if module not found"
Serge Pavlov [Fri, 14 Apr 2023 10:37:25 +0000 (17:37 +0700)]
Revert "[symbolizer] Change error message if module not found"

This reverts commit 75f1f158812dabc03e70697b6b9c272230bce63d.
It caused fail on https://lab.llvm.org/buildbot#builders/37/builds/21461

18 months ago[Matrix] Refine cost estimate for dot-product.
Florian Hahn [Fri, 14 Apr 2023 10:35:01 +0000 (11:35 +0100)]
[Matrix] Refine cost estimate for dot-product.

Adjust lowerDotProduct cost estimate to include the cost benefits of:
 * emitting a wide load
 * emitting a wide multiply.

Reviewed By: thegameg

Differential Revision: https://reviews.llvm.org/D147330

18 months ago[X86] Add AVX2 test coverage to TESTPS/TESTPD combines
Simon Pilgrim [Fri, 14 Apr 2023 10:07:37 +0000 (11:07 +0100)]
[X86] Add AVX2 test coverage to TESTPS/TESTPD combines

18 months ago[X86] Add TESTPS/TESTPD test coverage showing failure to simplify demanded sign bit...
Simon Pilgrim [Fri, 14 Apr 2023 10:05:51 +0000 (11:05 +0100)]
[X86] Add TESTPS/TESTPD test coverage showing failure to simplify demanded sign bit when the operands has multiple uses

18 months ago[Test] Add tests showing flags improvement for SCEV sub expressions (NFC)
Dmitry Makogon [Fri, 14 Apr 2023 10:12:17 +0000 (17:12 +0700)]
[Test] Add tests showing flags improvement for SCEV sub expressions (NFC)

18 months ago[AIX] enable the cases that are excluded by XCOFF 64 integrated-as support
Chen Zheng [Fri, 14 Apr 2023 10:16:01 +0000 (06:16 -0400)]
[AIX] enable the cases that are excluded by XCOFF 64 integrated-as support

These case are excluded in https://reviews.llvm.org/D113049.
Now AIX XCOFF 64 integrated-as support improves a lot and all these
cases pass now, so enable them.

18 months ago[llvm-c] Remove PassRegistry and initialization APIs
Nikita Popov [Wed, 1 Mar 2023 08:41:59 +0000 (09:41 +0100)]
[llvm-c] Remove PassRegistry and initialization APIs

Remove C APIs for interacting with PassRegistry and pass
initialization. These are legacy PM concepts, and are no longer
relevant for the new pass manager.

Calls to these initialization functions can simply be dropped.

Differential Revision: https://reviews.llvm.org/D145043

18 months ago[gn] Port fcc5f9ef50e1 (GenVT)
Nico Weber [Fri, 14 Apr 2023 10:08:51 +0000 (06:08 -0400)]
[gn] Port fcc5f9ef50e1 (GenVT)

18 months ago[FunctionAttrs] Fix nounwind inference for landingpads
Nikita Popov [Thu, 6 Apr 2023 08:49:42 +0000 (10:49 +0200)]
[FunctionAttrs] Fix nounwind inference for landingpads

Currently, FunctionAttrs treats landingpads as non-throwing, and
will infer nounwind for functions with landingpads (assuming they
can't unwind in some other way, e.g. via resum). There are two
problems with this:

* Non-cleanup landingpads with catch/filter clauses do not
  necessarily catch all exceptions. Unless there are catch ptr null
  or filter [0 x ptr] zeroinitializer clauses, we should assume
  that we may unwind past this landingpad. This seems like an
  outright bug.
* Cleanup landingpads are skipped during phase one unwinding, so
  we effectively need to support unwinding past them. Marking these
  nounwind is technically correct, but not compatible with how
  unwinding works in reality.

Fixes https://github.com/llvm/llvm-project/issues/61945.

Differential Revision: https://reviews.llvm.org/D147694

18 months ago[Matrix] Add dot product tests with builtin loads with variable strides
Florian Hahn [Fri, 14 Apr 2023 09:40:47 +0000 (10:40 +0100)]
[Matrix] Add dot product tests with builtin loads with variable strides

Extra tests for D147330.

18 months ago[PowerPC] Update `incr` after resetting the register in MI
Kai Luo [Fri, 14 Apr 2023 09:24:30 +0000 (17:24 +0800)]
[PowerPC] Update `incr` after resetting the register in MI

After performing signed extension, we update the register in MI. We should also update `incr` register which is tracking the register in `MI`.

Fixes https://github.com/llvm/llvm-project/issues/61882.

Reviewed By: #powerpc, shchenz

Differential Revision: https://reviews.llvm.org/D147594

18 months ago[InstCombine] Support multiple comparisons in foldAllocaCmp()
Nikita Popov [Mon, 20 Feb 2023 14:55:17 +0000 (15:55 +0100)]
[InstCombine] Support multiple comparisons in foldAllocaCmp()

foldAllocaCmp() needs to fold all comparisons of an alloca at the
same time, to ensure that there is a consistent view of the alloca
address. Currently, it folds "all" comparisons by limiting to the
case where there is only one. This patch switches the algorithm to
instead actually collect and fold all comparisons.

Something we need to be careful about here is that there may be
comparisons where both sides of the icmp are based on the alloca.
Such comparisons are comparing offsets of the alloca, and as such
can be ignored here, but shouldn't be folded to false.

Differential Revision: https://reviews.llvm.org/D144492

18 months ago[InstSimplify] Add MaxRecurse argument to simplifyInstructionWithOperands (NFC)
Nikita Popov [Fri, 14 Apr 2023 09:18:18 +0000 (11:18 +0200)]
[InstSimplify] Add MaxRecurse argument to simplifyInstructionWithOperands (NFC)

18 months agoFix bazel build after e5f50bd26807
Christian Sigg [Fri, 14 Apr 2023 09:11:39 +0000 (11:11 +0200)]
Fix bazel build after e5f50bd26807

18 months ago[LLD][ARM] Handle .ARM.exidx sections at non-zero output sec offset
Peter Smith [Tue, 11 Apr 2023 17:14:40 +0000 (18:14 +0100)]
[LLD][ARM] Handle .ARM.exidx sections at non-zero output sec offset

Embedded systems that do not use an ELF loader locate the
.ARM.exidx exception table via linker defined __exidx_start and
__exidx_end rather than use the PT_ARM_EXIDX program header. This
means that some linker scripts such as the picolibc C library's
linker script, do not have the .ARM.exidx sections at offset 0 in
the OutputSection. For example:

.except_unordered : {
    . = ALIGN(8);
    PROVIDE(__exidx_start = .);
    *(.ARM.exidx*)
    PROVIDE(__exidx_end = .);
} >flash AT>flash :text

This is within the specification of Arm exception tables, and is
handled correctly by ld.bfd.

This patch has 2 parts. The first updates the writing of the data
of the .ARM.exidx SyntheticSection to account for a non-zero
OutputSection offset. The second part makes the PT_ARM_EXIDX program
header generation a special case so that it covers only the
SyntheticSection and not the parent OutputSection. While not strictly
necessary for programs locating the exception tables via the symbols
it may cause ELF utilities that locate the exception tables via
the PT_ARM_EXIDX program header to fail. This does not seem to be the
case for GNU and LLVM readelf which seems to look for the
SHT_ARM_EXIDX section.

Differential Revision: https://reviews.llvm.org/D148033

18 months ago[docs] Clarify that CoC docs are under a CC-BY license.
Kristof Beyls [Wed, 12 Apr 2023 13:01:00 +0000 (15:01 +0200)]
[docs] Clarify that CoC docs are under a CC-BY license.

Differential Revision: https://reviews.llvm.org/D148121

18 months ago[AMDGPU] Add backend support for new PAL ELF Metadata 3.0
David Stuttard [Mon, 20 Feb 2023 17:11:22 +0000 (17:11 +0000)]
[AMDGPU] Add backend support for new PAL ELF Metadata 3.0

PAL Metadata 3.0 introduces an explicit structure in metadata for the
programmable registers written out by the compiler backend.
Rather than using opaque registers which can change between different
architectures and requires encoding the bitfield information in the backend,
which may change between versions.

This is the initial minimal implementation that enables the use of PAL Metadata
3.0.

The change itself should be NFC for non-PAL, although the way RSRC2 register is
handled has been changed slightly.

The test is fairly minimal, but checks that the metadata format looks as
expected and verifies a couple of special cases such as tgid_[xyz]_en handling
and PsInputAddr/Ena which also change to explicit fields.

Differential Revision: https://reviews.llvm.org/D147143

18 months ago[LangRef][Local] dereferenceable metadata violation is UB
Nikita Popov [Thu, 13 Apr 2023 08:38:56 +0000 (10:38 +0200)]
[LangRef][Local] dereferenceable metadata violation is UB

I believe !dereferencable violation is immediate undefined behavior,
but this was not explicitly spelled out in LangRef. We already
assume that !dereferenceable is implicitly !noundef and cannot
return poison in isGuaranteedNotToBeUndefOrPoison().

The reason why we made dereferenceable implicitly noundef is that
the purpose of this metadata is to allow speculation, and that
would not be legal on a potential poison pointer.

Differential Revision: https://reviews.llvm.org/D148202

18 months ago[InstSimplify] Remove unused ORE argument (NFC)
Nikita Popov [Fri, 14 Apr 2023 08:35:17 +0000 (10:35 +0200)]
[InstSimplify] Remove unused ORE argument (NFC)

18 months ago[AMDGPU] Don't S_MOV_B32 into $scc
Diana Picus [Wed, 12 Apr 2023 09:52:44 +0000 (11:52 +0200)]
[AMDGPU] Don't S_MOV_B32 into $scc

The peephole optimizer tries to replace
```
%n:sgpr_32 = S_MOV_B32 x
$scc = COPY %n
```
with a `S_MOV_B32` directly into `$scc`.

This crashes because `S_MOV_B32` cannot take `$scc` as input.

We currently generate code like this from GlobalISel when lowering a
G_BRCOND with a constant condition. We should probably look into
removing this kind of branch altogether, but until then we should at
least not crash.

This patch fixes the issue by making sure we don't apply the peephole
optimization when trying to move into a physical register that
doesn't belong to the correct register class.

Differential Revision: https://reviews.llvm.org/D148117

18 months ago[Coroutines] Directly remove unnecessary lifetime intrinsics
Nikita Popov [Thu, 13 Apr 2023 15:34:30 +0000 (17:34 +0200)]
[Coroutines] Directly remove unnecessary lifetime intrinsics

The insertSpills() code will currently skip lifetime intrinsic users
when replacing the alloca with a frame reference. Rather than
leaving behind the dead lifetime intrinsics working on the old
alloca, directly remove them. This makes sure the alloca can be
dropped as well.

I noticed this as a regression when converting tests to opaque
pointers. Without opaque pointers, this code didn't really do
anything, because there would usually be a bitcast in between.
The lifetimes would get rewritten to the frame pointer. With
opaque pointers, this code now triggers and leaves behind users
of the old allocas.

Differential Revision: https://reviews.llvm.org/D148240

18 months ago [sanitizer] adapt for 75f1f158812dabc03e70697b6b9c272230bce63d
Krasimir Georgiev [Fri, 14 Apr 2023 08:18:43 +0000 (08:18 +0000)]
 [sanitizer] adapt for 75f1f158812dabc03e70697b6b9c272230bce63d

 No functional changes intended.

18 months ago[mlir][llvm] Move the LLVM dialect definition (NFC).
Tobias Gysi [Fri, 14 Apr 2023 07:10:15 +0000 (07:10 +0000)]
[mlir][llvm] Move the LLVM dialect definition (NFC).

The revision separates out the LLVM dialect definition in a separate
tablegen file and ensures the LLVMOpBase.td can include the attributes
defined by LLVMAttrDefs.td. The change allows us to use LLVM dialect
attributes in the definition of the intrinsic and memory operation
base classes, e.g. to represent alias analysis metadata using
attributes.

Reviewed By: Dinistro

Differential Revision: https://reviews.llvm.org/D148007

18 months agoRevert "[clang-format] Handle object instansiation in if-statements"
Tobias Hieta [Fri, 14 Apr 2023 07:44:35 +0000 (09:44 +0200)]
Revert "[clang-format] Handle object instansiation in if-statements"

This reverts commit 70de684d44135b4025d92b2b36ad387cf5ab8b5a.

This causes a regression as described in #61785

18 months ago[llvm-exegesis] Fix -Wc++98-compat-extra-semi in BenchmarkRunner.cpp (NFC)
Jie Fu [Fri, 14 Apr 2023 07:46:09 +0000 (15:46 +0800)]
[llvm-exegesis] Fix -Wc++98-compat-extra-semi in BenchmarkRunner.cpp (NFC)

/data/llvm-project/llvm/tools/llvm-exegesis/lib/BenchmarkRunner.cpp:66:2: error: extra ';' outside of a function is incompatible with C++98 [-Werror,-W
c++98-compat-extra-semi]
};
 ^
1 error generated.

18 months ago[llvm-exegesis] Refactor common parts out of FunctionExecutorImpl
Aiden Grossman [Fri, 14 Apr 2023 07:19:10 +0000 (07:19 +0000)]
[llvm-exegesis] Refactor common parts out of FunctionExecutorImpl

This patch refactors some code out of FunctionExecutorImpl into the base
class that should be common across all implementations of
FunctionExecutor. Particularly, this patch factors out
accumulateCounterValues, and also factors out runAndSample, moving
implementation specific code into a new runWithCounter function. This
makes adding new implementations of FunctinExecutor easier.

Reviewed By: gchatelet

Differential Revision: https://reviews.llvm.org/D148079

18 months ago[clangd] Fix test failure in initialize-params.test
Nathan Ridge [Fri, 14 Apr 2023 07:25:40 +0000 (03:25 -0400)]
[clangd] Fix test failure in initialize-params.test

18 months ago[llvm-exegesis][NFC] remove runAndMeasure
Aiden Grossman [Fri, 14 Apr 2023 07:05:37 +0000 (07:05 +0000)]
[llvm-exegesis][NFC] remove runAndMeasure

This completes the FIXME listed in FunctionExecutor in regards to
deprecating this function. It simply makes the appropriate call into
runAndSample and grabs the first counter value. This patch completely
removes the function, moving that logic into the callers (currently only
uopsBenchmarkRunner). This makes creating new FunctionExecutors easier
as an implementation no longer needs to worry about this detail.

Reviewed By: gchatelet

Differential Revision: https://reviews.llvm.org/D147878

18 months ago[clang] Add test for CWG1894 and CWG2199
Vlad Serebrennikov [Fri, 14 Apr 2023 07:14:36 +0000 (10:14 +0300)]
[clang] Add test for CWG1894 and CWG2199

[[https://wg21.link/p1787 | P1787]]: CWG1894 and its duplicate CWG2199 are resolved per Richard’s proposal for [[ https://listarchives.isocpp.org/cgi-bin/wg21/message?wg=core&msg=28415 | “dr407 still leaves open questions about typedef / tag hiding” ]], using generic conflicting-declaration rules even for typedef, and discarding a redundant typedef-name when looking up an elaborated-type-specifier.
Wording: See changes to [dcl.typedef], [basic.lookup.elab], and [basic.lookup]/4.

Generic conflicting-declaration rules are specified in changes to [basic.scope.scope]. [[ https://cplusplus.github.io/CWG/issues/407.html | CWG407]], [[ https://cplusplus.github.io/CWG/issues/1894.html | CWG1894 ]], and [[ https://cplusplus.github.io/CWG/issues/2199.html | CWG2199 ]] discuss how elaborated type specifiers interact with typedefs, using directives, and using declarations. Since existing test for CWG407 covers examples provided in CWG1894 and CWG2199, and does it in accordance with P1787, I reused parts of it.

Reviewed By: #clang-language-wg, cor3ntin

Differential Revision: https://reviews.llvm.org/D148136

18 months ago[clangd] Inactive regions support via dedicated protocol
Nathan Ridge [Tue, 5 Apr 2022 07:19:15 +0000 (03:19 -0400)]
[clangd] Inactive regions support via dedicated protocol

This implements the server side of the approach discussed at
https://github.com/clangd/vscode-clangd/pull/193#issuecomment-1044315732

Differential Revision: https://reviews.llvm.org/D143974

18 months ago[JITLink][RISCV] Handle R_RISCV_CALL_PLT fixups
Job Noorman [Fri, 14 Apr 2023 07:11:08 +0000 (09:11 +0200)]
[JITLink][RISCV] Handle R_RISCV_CALL_PLT fixups

In the default link configuration, PLT stubs are created automatically
for R_RISCV_CALL_PLT relocations and the relocation itself is
transformed to R_RISCV_CALL (PerGraphGOTAndPLTStubsBuilder_ELF_riscv).
Only the latter is later handled when applying fixups and the former is
simply ignored.

This patch proposes to handle R_RISCV_CALL_PLT anyway when applying
fixups to support custom configurations that do not need automatic PLT
creation. An example of this is BOLT where PLT entries from the input
binary are reused (D147544).

Reviewed By: StephenFan

Differential Revision: https://reviews.llvm.org/D148238

18 months ago[MLGO] Change MBB Profile Dump from using MBB numbers to MBB IDs
Aiden Grossman [Fri, 14 Apr 2023 06:55:35 +0000 (06:55 +0000)]
[MLGO] Change MBB Profile Dump from using MBB numbers to MBB IDs

Currenty, setting the -mbb-profile-dump dumps a CSV file with blocks
inside an individual function identified by their MBB numbers. This
patch changes the MBBs to be identified by their ID which is set at MBB
creation and not changed afterwards, making it inherently stable
throughout the backend. This alleviates concerns with the MBB IDs
changing between the profile dump and what ends up in the final object
file. The MBBs inside the SHT_LLVM_BB_ADDR_MAP sections are also
identified using their MBB ID rather than number, so if we want to match
them up we need to identify the MBBs here by number.

Reviewed By: mtrofin, rahmanl

Differential Revision: https://reviews.llvm.org/D147366

18 months ago[bazel] Remove remnant HAVE_LSEEK64 after 5cd554303ead0f8891eee3cd6d25cb07f5a7bf67
Fangrui Song [Fri, 14 Apr 2023 06:54:08 +0000 (23:54 -0700)]
[bazel] Remove remnant HAVE_LSEEK64 after 5cd554303ead0f8891eee3cd6d25cb07f5a7bf67

18 months ago[flang] fix fir.array_coor of fir.box with component references
Jean Perier [Fri, 14 Apr 2023 06:47:25 +0000 (08:47 +0200)]
[flang] fix fir.array_coor of fir.box with component references

When dealing with "derived_array(j)%component" where derived_array
is not a contiguous array, but for which we know the extent, lowering
generates a fir.array_coor op on a !fir.box<!fir.array<cst x T>> with
a fir.slice containing "j" in the component path.

Codegen first computes "derived_array(j)" address using the byte
strides inside the descriptor, and then computes the offset of "j"
from that address with a second GEP.
The type of the address in that second GEP matters since "j" is passed
in the GEP via an index indicating its component position in the type.

The code was using the LLVM type of "derived_array" instead of
"derived_array(j)".
In general, with fir.box, the extent ("cst" above) is unknown and those
types match. But if the extent of "derived_array" is a compile time
constant, its LLVM type will be [cst x T] instead of T*, and the produced
GEP will compute the address of the nth T instead of the nth component
inside T leading to undefined behaviors.

Fix this by computing the element type for the second GEP.

Differential Revision: https://reviews.llvm.org/D148226

18 months ago[flang] Change TYPE(*) arrays passing convention
Jean Perier [Fri, 14 Apr 2023 06:43:33 +0000 (08:43 +0200)]
[flang] Change TYPE(*) arrays passing convention

- Fix the BIND(C) assumed-shape case: TYPE(*) assumed shape are passed
  via CFI_cdesc_t according to Fortran 2018 standard 18.3.6 point 2 (5).
- Align the none BIND(C) case with the BIND(C) case. There is little
  point passing TYPE(*) assumed size via descriptor, use a simple
  address. C710 ensures there is no way the knowledge of the actual
  type will be required when manipulating the dummy.

Differential Revision: https://reviews.llvm.org/D148130

18 months ago[flang] Fold IS_CONTIGUOUS for TYPE(*) when possible
Jean Perier [Fri, 14 Apr 2023 06:42:15 +0000 (08:42 +0200)]
[flang] Fold IS_CONTIGUOUS for TYPE(*) when possible

TYPE(*) arguments fell through in IS_CONTIGUOUS folding
because they are not Expr<SomeType>. Expose entry point for
symbols in IsContiguous and use that.

The added test revealed that IS_CONTIGUOUS was folded to
false for assumed rank arguments. Fix this: the contiguity of
assumed rank without the CONTIGUOUS argument can only be
verified at runtime.

Differential Revision: https://reviews.llvm.org/D148128

18 months ago[symbolizer] Change error message if module not found
Serge Pavlov [Fri, 14 Apr 2023 06:03:28 +0000 (13:03 +0700)]
[symbolizer] Change error message if module not found

If llvm-symbolize did not find module, the error looked like:

    LLVMSymbolizer: error reading file: No such file or directory

This message does not follow common practice: LLVMSymbolizer is not an
utility name. Also the message did not not contain the name of missed file.

With this change the error message looks differently:

    llvm-symbolizer: error: 'abc': No such file or directory

This format is closer to messages produced by other utilities and allow
proper coloring.

Differential Revision: https://reviews.llvm.org/D148032

18 months ago[IRCE][NFC] Refactor parseRangeCheckICmp to compute SCEVs instead of Values
Max Kazantsev [Fri, 14 Apr 2023 05:24:32 +0000 (12:24 +0700)]
[IRCE][NFC] Refactor parseRangeCheckICmp to compute SCEVs instead of Values

The motivation is to make an opportunity to compute and return
expressions after parsing ICmp into a range check (e.g. Length + 1).

Patch by Aleksandr Popov!

Differential Revision: https://reviews.llvm.org/D148205

18 months ago[compiler-rt] Fix signed shift overflows in absvdi2.c, absvsi2.c, negvdi2.c and negvsi2.c
Karl-Johan Karlsson [Fri, 14 Apr 2023 05:44:44 +0000 (07:44 +0200)]
[compiler-rt] Fix signed shift overflows in absvdi2.c, absvsi2.c, negvdi2.c and negvsi2.c

When compiling compiler-rt with -fsanitize=undefined and running testcases you
end up with the following warnings:

UBSan: absvdi2.c:21:23: left shift of 1 by 63 places cannot be represented in type 'di_int' (aka 'long long')
UBSan: absvsi2.c:21:23: left shift of 1 by 31 places cannot be represented in type 'si_int' (aka 'long')
UBSan: negvdi2.c:20:32: left shift of 1 by 63 places cannot be represented in type 'di_int' (aka 'long long')
UBSan: negvsi2.c:20:32: left shift of 1 by 31 places cannot be represented in type 'si_int' (aka 'long')

This can be avoided by doing the shift in a matching unsigned variant of the
type.

This was found in an out of tree target.

Reviewed By: MaskRay

Differential Revision: https://reviews.llvm.org/D146932

18 months agoRe-land 'ASan: move allocator base to avoid conflict with high-entropy ASLR for x86...
Thurston Dang [Thu, 13 Apr 2023 23:55:01 +0000 (23:55 +0000)]
Re-land 'ASan: move allocator base to avoid conflict with high-entropy ASLR for x86-64 Linux'

D147984 was reverted because it broke lit tests on Mac. This revision is based on D147984
but maintains the old behavior for Apple.

Note that, per the follow-up discussion with MaskRay in D147984, this patch excludes Apple
but includes other platforms (e.g., aarch64, MIPS64) and OSes (e.g., FreeBSD, S390X), not just
x86-64 Linux.

Original commit message from D147984:

Users have discovered [*] that when CONFIG_ARCH_MMAP_RND_BITS == 32,
it will frequently conflict with ASan's allocator on x86-64 Linux, because the
PIE program segment base address of 0x555555555554 plus an ASLR shift of up to
((2**32) * 4K == 0x100000000000) will sometimes exceed ASan's hardcoded
base address of 0x600000000000. We fix this by simply moving the allocator base
to 0x500000000000, which is below the PIE program segment base address. This is
cleaner than trying to move it to another location that is sandwiched between
the PIE program and library segments, because if either of those grow too large,
it will collide with the allocator region.

Note that we will never need to change this base address again (unless we want to increase
the size of the allocator), because ASLR cannot be set above 32-bits for x86-64 Linux (the
PIE program segment and library segments would collide with each other; see also
ARCH_MMAP_RND_BITS_MAX in https://github.com/torvalds/linux/blob/master/arch/x86/Kconfig).

[*] see https://b.corp.google.com/issues/276925478
and https://groups.google.com/a/google.com/g/chrome-os-gardeners/c/BbfzCP3dEeo/m/h3C_vVUxCQAJ

Differential Revision: https://reviews.llvm.org/D148280