platform/upstream/llvm.git
21 months ago[InstCombine] fold sdiv with common shl amount in operands
Sanjay Patel [Wed, 12 Oct 2022 15:18:55 +0000 (11:18 -0400)]
[InstCombine] fold sdiv with common shl amount in operands

(X << Z) / (Y << Z) --> X / Y

https://alive2.llvm.org/ce/z/CLKzqT

This requires a surprising "nuw" constraint because we have
to guard against immediate UB via signed-div overflow with
-1 divisor.

This extends 008a89037a49ca0d9 and is another transform
derived from issue #58137.

21 months ago[libc] New version of the mem* framework
Guillaume Chatelet [Tue, 4 Oct 2022 08:49:02 +0000 (08:49 +0000)]
[libc] New version of the mem* framework

This version is more composable and also simpler at the expense of being more explicit and more verbose.
This patch is not meant to be submitted but gives an idea of the change.
Codegen can be checked in https://godbolt.org/z/6z1dEoWbs by removing the "static inline" before individual functions.

Unittests are coming.

Suggested review order:
 - utils
 - op_base
 - op_builtin
 - op_generic
 - op_x86 / op_aarch64
 - *_implementations.h

Differential Revision: https://reviews.llvm.org/D135134

21 months ago[SLP]Fix insertpoint of the extractellements instructions to avoid reshuffle crash.
Alexey Bataev [Mon, 10 Oct 2022 14:40:35 +0000 (07:40 -0700)]
[SLP]Fix insertpoint of the extractellements instructions to avoid reshuffle crash.

Need to set the insertpoint for extractelement to point to the first
instruction in the node to avoid possible crash during external uses
combine  process. Without it we may endup with the incorrect
transformation.

Differential Revision: https://reviews.llvm.org/D135591

21 months ago[libc] Update benchmarks/README.md
Andre Vieira [Wed, 12 Oct 2022 15:12:23 +0000 (16:12 +0100)]
[libc] Update benchmarks/README.md

This patch updates the instructions on how to run the mem* benchmarks in
README.md.

Differential Revision: https://reviews.llvm.org/D135582

21 months ago[InstCombine] fold udiv with common shl amount in operands
Sanjay Patel [Wed, 12 Oct 2022 14:58:03 +0000 (10:58 -0400)]
[InstCombine] fold udiv with common shl amount in operands

(X << Z) / (Y << Z) --> X / Y

https://alive2.llvm.org/ce/z/E5eaxU

This fixes the motivating example from issue #58137,
but it is not the most general transform. We should
probably also convert left-shift in the divisor to
right-shift in the dividend for that, but that exposes
another missed canonicalization for shifts and adds.

21 months ago[FunctionAttrs] Regenerate test checks (NFC)
Nikita Popov [Wed, 12 Oct 2022 14:59:42 +0000 (16:59 +0200)]
[FunctionAttrs] Regenerate test checks (NFC)

21 months ago[NFC] Ignore unused var in no-asserts builds
Jordan Rupprecht [Wed, 12 Oct 2022 15:11:10 +0000 (08:11 -0700)]
[NFC] Ignore unused var in no-asserts builds

21 months ago[UpdateTestChecks] Add basic BPF triple handling
Simon Pilgrim [Wed, 12 Oct 2022 14:57:45 +0000 (15:57 +0100)]
[UpdateTestChecks] Add basic BPF triple handling

Working on Issue #57872 - its really useful to be able to autogenerate checks

21 months ago[libc++] Remove _LIBCPP_HAS_NO_CXX20_COROUTINES
Louis Dionne [Wed, 5 Oct 2022 15:35:53 +0000 (11:35 -0400)]
[libc++] Remove _LIBCPP_HAS_NO_CXX20_COROUTINES

As far as I can tell, all the compilers we support have support for
C++20 coroutines now.

Differential Revision: https://reviews.llvm.org/D135274

21 months ago[libc++] Disable the GDB pretty printer tests again, since they break the CI
Louis Dionne [Wed, 12 Oct 2022 14:40:06 +0000 (10:40 -0400)]
[libc++] Disable the GDB pretty printer tests again, since they break the CI

21 months ago[CGStmt] Use helper functions to set memory attributes (NFC)
Nikita Popov [Wed, 12 Oct 2022 12:34:03 +0000 (14:34 +0200)]
[CGStmt] Use helper functions to set memory attributes (NFC)

21 months ago[SLP]Fix PR58177: Improve isUndefVector function to avoid extra freeze.
Alexey Bataev [Thu, 6 Oct 2022 18:01:54 +0000 (11:01 -0700)]
[SLP]Fix PR58177: Improve isUndefVector function to avoid extra freeze.

Freeze instruction in some cases makes codegen worse, so need to be very
careful when emitting it. Instead improve analysis in isUndefVector
function to generate mask of unused elements and use it in the analysis.

Differential Revision: https://reviews.llvm.org/D135382

21 months ago[mlir][Linalg] Drop filter-based splitReduction
Nicolas Vasilache [Wed, 12 Oct 2022 12:21:52 +0000 (05:21 -0700)]
[mlir][Linalg] Drop filter-based splitReduction

This transformation is available and tested via the transform dialect.

Differential Revision: https://reviews.llvm.org/D135767

21 months ago[mlir][Linalg] NFC - Drop LinalgLoweringPattern
Nicolas Vasilache [Wed, 12 Oct 2022 12:06:18 +0000 (05:06 -0700)]
[mlir][Linalg] NFC - Drop LinalgLoweringPattern

Linalg ops are generally lowered to loops using a pass and the individual application
of patterns has no concrete uses.
If/when necessary, this can later be reintroduced via a proper transform dialect op.

Differential Revision: https://reviews.llvm.org/D135766

21 months ago[SystemZ] Revert migration of SystemZ/Large/branch-01.ll
Kai Nacke [Wed, 12 Oct 2022 14:10:05 +0000 (14:10 +0000)]
[SystemZ] Revert migration of SystemZ/Large/branch-01.ll

Root cause for the regression is that value %6598 is
defined as a label, and later as an i32 value.

Reviewed By: uweigand

Differential Revision: https://reviews.llvm.org/D135778

21 months ago[LV] Add extra tests for epilogue vectorization with widened inductions.
Florian Hahn [Wed, 12 Oct 2022 14:21:38 +0000 (15:21 +0100)]
[LV] Add extra tests for epilogue vectorization with widened inductions.

Extend test coverage to also include inductions with step > 1 and also
with runtime trip counts.

21 months ago[flang] Use fir.rebox for tbp fir.dispatch call with allocatable or pointer
Valentin Clement [Wed, 12 Oct 2022 14:18:18 +0000 (16:18 +0200)]
[flang] Use fir.rebox for tbp fir.dispatch call with allocatable or pointer

Polymorphic entities with allocatable or pointer attribute cannot be passed
directly as passed-object when the type-bound procedure is expecting a
simply dummy polymorphic entity. Use fir.rebox to adapt the fir.class box
to the tbp type.

Depends on D135649

Reviewed By: jeanPerier

Differential Revision: https://reviews.llvm.org/D135654

21 months ago[mlir] Fix bitwidth of memref-to-llvm constant.
Tobias Gysi [Wed, 12 Oct 2022 14:12:21 +0000 (17:12 +0300)]
[mlir] Fix bitwidth of memref-to-llvm constant.

One constant generated in MemRefToLLVM had a hardcoded bitwidth of
64 bits. The fix uses the typeConverter to create a constant that
matches the bitwidth of the provided by the data layout. The issue was
detected in an attempt to add a verifier to the LLVM ICmp operation that
checks that the types of the compared arguments match.

Reviewed By: ftynse

Differential Revision: https://reviews.llvm.org/D135775

21 months ago[runtimes] Always define cxx_shared, cxx_static & other targets
Louis Dionne [Mon, 19 Sep 2022 20:17:02 +0000 (16:17 -0400)]
[runtimes] Always define cxx_shared, cxx_static & other targets

However, mark them as EXCLUDE_FROM_ALL when we don't want to build them.
Simply declaring the targets should be of no harm, and it allows other
projects to mention these targets regardless of whether they end up
being built or not.

While the diff may not make that obvious, this patch basically
moves the definition of e.g. `cxx_shared` out of the `if (LIBCXX_ENABLE_SHARED)`
and instead marks it as `EXCLUDE_FROM_ALL` conditionally on whether
LIBCXX_ENABLE_SHARED is passed. It then does the same for libunwind
and libc++abi targets.

Differential Revision: https://reviews.llvm.org/D134221

21 months ago[LoongArch] Handle missing CondCodes
gonglingqin [Wed, 12 Oct 2022 13:11:29 +0000 (21:11 +0800)]
[LoongArch] Handle missing CondCodes

Support SETLE/SETEQ and expand SETGE/SETNE/SETGT

Differential Revision: https://reviews.llvm.org/D135511

21 months ago[InstCombine] propagate "exact" through folds of div
Sanjay Patel [Wed, 12 Oct 2022 13:05:34 +0000 (09:05 -0400)]
[InstCombine] propagate "exact" through folds of div

These folds were added recently with:
6b869be8100d
8da2fa856f1b
...but they didn't account for the "exact" attribute,
and that can be safely propagated:
https://alive2.llvm.org/ce/z/F_WhnR
https://alive2.llvm.org/ce/z/ft9Cgr

21 months ago[InstCombine] add "exact" to div tests to show missing propagation; NFC
Sanjay Patel [Wed, 12 Oct 2022 12:53:39 +0000 (08:53 -0400)]
[InstCombine] add "exact" to div tests to show missing propagation; NFC

21 months ago[InstCombine] add helper function for div+shl folds; NFC
Sanjay Patel [Wed, 12 Oct 2022 12:50:13 +0000 (08:50 -0400)]
[InstCombine] add helper function for div+shl folds; NFC

There are at least 2 similar patterns that could be added here,
and the existing fold can be improved because it fails to
propagate "exact".

21 months ago[InstCombine] add tests for div with shl operands; NFC
Sanjay Patel [Tue, 11 Oct 2022 21:06:01 +0000 (17:06 -0400)]
[InstCombine] add tests for div with shl operands; NFC

21 months ago[flang] Lower type-bound procedure call needing dynamic dispatch to fir.dispatch
Valentin Clement [Wed, 12 Oct 2022 13:24:21 +0000 (15:24 +0200)]
[flang] Lower type-bound procedure call needing dynamic dispatch to fir.dispatch

Lower call with polymorphic entities to fir.dispatch operation. This patch only
focus one lowering with simple scalar polymorphic entities. A follow-up patch
will deal with allocatble, pointer and array of polymorphic entities as they
require box manipulation for the passed-object.

Reviewed By: jeanPerier

Differential Revision: https://reviews.llvm.org/D135649

21 months ago[LoongArch] Add earlyclobber of destination register to atomic instructions
gonglingqin [Wed, 12 Oct 2022 12:43:22 +0000 (20:43 +0800)]
[LoongArch] Add earlyclobber of destination register to atomic instructions

If the AM* atomic memory access instruction has the same register number as
rd and rj, the execution will trigger an Instruction Non-defined Exception.
If the AM* atomic memory access instruction has the same register number as
rd and rk, the execution result is uncertain.

Reference: https://github.com/loongson/LoongArch-Documentation

Differential Revision: https://reviews.llvm.org/D135641

21 months agoFix false positive related to handling of [[noreturn]] function pointers
Arseniy Zaostrovnykh [Wed, 12 Oct 2022 12:46:32 +0000 (14:46 +0200)]
Fix false positive related to handling of [[noreturn]] function pointers

Before this change, the `NoReturnFunctionChecker` was missing function pointers
with a `[[noreturn]]` attribute, while `CFG` was constructed taking that into
account, which leads CSA to take impossible paths. The reason was that the
`NoReturnFunctionChecker` was looking for the attribute in the type of the
entire call expression rather than the type of the function being called.

This change makes the `[[noreturn]]` attribute of a function pointer visible
to `NoReturnFunctionChecker`. This leads to a more coherent behavior of the
CSA on the AST involving.

Reviewed By: xazax.hun

Differential Revision: https://reviews.llvm.org/D135682

21 months ago[flang][NFC] Fix FIRTypes.td
Peixin Qiao [Wed, 12 Oct 2022 12:25:34 +0000 (20:25 +0800)]
[flang][NFC] Fix FIRTypes.td

Fix two typos and remove the incorrectly added `extraClassDeclaration` in
`FIRTypes.td`.

Reviewed By: clementval

Differential Revision: https://reviews.llvm.org/D135665

21 months ago[PowerPC][NFC] refactor some test cases.
Chen Zheng [Wed, 12 Oct 2022 12:15:15 +0000 (12:15 +0000)]
[PowerPC][NFC] refactor some test cases.

21 months ago[lldb] Fix member access in GetExpressionPath
Tonko Sabolčec [Wed, 12 Oct 2022 11:46:23 +0000 (11:46 +0000)]
[lldb] Fix member access in GetExpressionPath

This change fixes two issues in ValueObject::GetExpressionPath method:

 1. Accessing members of struct references used to produce expression
    paths such as "str.&str.member" (instead of the expected
    "str.member"). This is fixed by assigning the flag tha the child
    value is a dereference when calling Dereference() on references
    and adjusting logic in expression path creation.

 2. If the parent of member access is dereference, the produced
    expression path was "*(ptr).member". This is incorrect, since it
    dereferences the member instead of the pointer. This is fixed by
    wrapping dereference expression into parenthesis, resulting with
    "(*ptr).member".

Reviewed By: werat, clayborg

Differential Revision: https://reviews.llvm.org/D132734

21 months ago[mlir][Bazel] Port 8446f24ef03928f511ab59dae17a8aa581b0ab62
Adrian Kuegel [Wed, 12 Oct 2022 12:07:31 +0000 (14:07 +0200)]
[mlir][Bazel] Port 8446f24ef03928f511ab59dae17a8aa581b0ab62

21 months ago[mlir][Linalg] Add MapOp to Linalg structured ops.
Adrian Kuegel [Wed, 12 Oct 2022 08:18:38 +0000 (10:18 +0200)]
[mlir][Linalg] Add MapOp to Linalg structured ops.

This will allow to model elementwise ops with this special op instead of using
GenericOp.
Also allow MapOp and ReduceOp to have no result if the output type is not a tensor.
This is needed for buffer semantics.

Differential Revision: https://reviews.llvm.org/D135754

21 months ago[VPlan] Remove dead recipes before sinking.
Florian Hahn [Wed, 12 Oct 2022 11:47:18 +0000 (12:47 +0100)]
[VPlan] Remove dead recipes before sinking.

optimizeInductions may leave dead recipes which can prevent sinking.
Sinking on the other hand should not introduce new dead recipes, so
clean up dead recipes before sinking.

Reviewed By: Ayal

Differential Revision: https://reviews.llvm.org/D133762

21 months ago[mlir][Linalg] NFC - Drop filter from LinalgGeneralizationPattern
Nicolas Vasilache [Wed, 12 Oct 2022 11:14:33 +0000 (04:14 -0700)]
[mlir][Linalg] NFC - Drop filter from LinalgGeneralizationPattern

Differential Revision: https://reviews.llvm.org/D135761

21 months ago[mlir][Linalg] NFC - Drop CodegenStrategy and remaining strategy passes
Nicolas Vasilache [Wed, 12 Oct 2022 10:26:55 +0000 (03:26 -0700)]
[mlir][Linalg] NFC - Drop CodegenStrategy and remaining strategy passes

Differential Revision: https://reviews.llvm.org/D135759

21 months ago[mlir][llvm] Tablegen based operation import from LLVM IR.
Tobias Gysi [Wed, 12 Oct 2022 11:23:10 +0000 (14:23 +0300)]
[mlir][llvm] Tablegen based operation import from LLVM IR.

The revision uses tablegen generated builders to convert the most common
LLVM IR instructions to MLIR LLVM dialect operations. All instructions
with special handlers, except for alloca and fence, still use manual
handlers. The revision also introduces an additional "instructions.ll"
test file to test the import of instructions that have tablegen builders
(except for the resume instruction whose test remains untouched). A part
of the test cases are new, for example the integer instruction test,
while others are migrated from the "basic.ll" test file.

Reviewed By: ftynse

Differential Revision: https://reviews.llvm.org/D135709

21 months ago[NFC] Perform all legality checks for non-trivial unswitch in one function
Max Kazantsev [Wed, 12 Oct 2022 11:14:06 +0000 (18:14 +0700)]
[NFC] Perform all legality checks for non-trivial unswitch in one function

They have been scattered over the code. For better structuring, perform
them in one place. Potential CT drop is possible because we collect exit
blocks twice, but it's small price to pay for much better code structure.

21 months agoDon't widen shuffle element with AVX512
Luo, Yuanke [Wed, 12 Oct 2022 10:42:32 +0000 (18:42 +0800)]
Don't widen shuffle element with AVX512

Fix crash issue of D129537 and reopen it.

Currently the X86 shuffle lowering would widen the element type for
shuffle if the mask element value is adjacent. For below example

  %t2 = add nsw <16 x i32> %t0, %t1
  %t3 = sub nsw <16 x i32> %t0, %t1
  %t4 = shufflevector <16 x i32> %t2, <16 x i32> %t3,
                      <16 x i32> <i32 16, i32 17, i32 2, i32 3, i32 4,
                       i32 5, i32 6, i32 7, i32 8, i32 9, i32 10,
                       i32 11, i32 12, i32 13, i32 14, i32 15>

  ret <16 x i32> %t4

Compiler would transform the shuffle to
  %t4 = shufflevector <8 x i64> %t2, <8 x i64> %t3,
                      <8 x i64> <i32 8, i32 1, i32 2, i32 3, i32 4,
                                 i32 5, i32 6, i32 7>
This may lose the oppotunity to let ISel select mask instruction when
avx512 is enabled.

This patch is to prevent the tranform when avx512 feature is enabled.
Thank Simon for the idea.

Reviewed By: RKSimon

Differential Revision: https://reviews.llvm.org/D130830

21 months ago[Dexter] Ignore step information in __libc_start_call_main
OCHyams [Wed, 12 Oct 2022 11:11:50 +0000 (12:11 +0100)]
[Dexter] Ignore step information in __libc_start_call_main

The test dexter/feature_tests/commands/perfect/expect_step_kind/direction.cpp
was failing on a machine because __libc_start_call_main was not identified as a
"frame below main" (a frame we don't want to gather information from), causing
dexter to count one more step than expected in the test.

Add __libc_start_call_main to the list of "frames below main". There may
be a more robust way of doing this but this is a pragmatic solution we can
use for now.

21 months ago[VPlan] Add test for sinking pointer induction increments.
Florian Hahn [Wed, 12 Oct 2022 11:06:40 +0000 (12:06 +0100)]
[VPlan] Add test for sinking pointer induction increments.

Extra test for D133762.

21 months ago[mlir][doc] move transform dialect docs to .md, NFC
Alex Zinenko [Wed, 12 Oct 2022 11:02:01 +0000 (11:02 +0000)]
[mlir][doc] move transform dialect docs to .md, NFC

The description of the Transform dialect has become quite lengthy to be
kept as a Tablegen string literal. Move it to a proper Markdown file.

21 months ago[mlir] document transform dialect types
Alex Zinenko [Wed, 12 Oct 2022 10:55:54 +0000 (10:55 +0000)]
[mlir] document transform dialect types

21 months ago[test][DebugInfo] Convert some test cases to opaque pointers. NFC
Bjorn Pettersson [Wed, 12 Oct 2022 10:00:25 +0000 (12:00 +0200)]
[test][DebugInfo] Convert some test cases to opaque pointers. NFC

Conversion performed using the script at:
https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34

21 months ago[mlir] Apply ClangTidyLegacy findings.
Adrian Kuegel [Wed, 12 Oct 2022 10:45:35 +0000 (12:45 +0200)]
[mlir] Apply ClangTidyLegacy findings.

Use '= default' to define a trivial default constructor.

21 months ago[llvm-objdump] Support nonzero section addresses in addSymbolizer
Pierre van Houtryve [Fri, 7 Oct 2022 08:12:36 +0000 (08:12 +0000)]
[llvm-objdump] Support nonzero section addresses in addSymbolizer

The previous calculations seem to have assumed that the section address would be zero.
This is true for relocatable object files, but certainly not for linked files like shared libraries.

Fixed the calculations to make them identical to the "real" `getInstruction` call below & added a regression test.

Reviewed By: scott.linder, simon_tatham

Differential Revision: https://reviews.llvm.org/D135430

21 months agoRevert "[AArch64] Add support for 128-bit non temporal loads."
David Green [Wed, 12 Oct 2022 10:11:32 +0000 (11:11 +0100)]
Revert "[AArch64] Add support for 128-bit non temporal loads."

This reverts commit 661403b85c219a83baa37335a870d4d93dc4b1c3 as the
custom lowering of loads prevents expanding unaligned loads with
strict-align.

21 months ago[AArch64] Exclude instructions after setting the FP from SEH prologues
Martin Storsjö [Tue, 11 Oct 2022 12:11:39 +0000 (15:11 +0300)]
[AArch64] Exclude instructions after setting the FP from SEH prologues

After setting up the FP, the rest of the prologue doesn't need to
be replayed for unwinding the stack frame.

This allows reverting the functional parts of
2f7fbf837625267193351cc334e506a3a9161958 (but fixing inconsistent
duplicate setting of HasWinCFI).

Differential Revision: https://reviews.llvm.org/D135686

21 months ago[LLDB] Fix x86_64 build
David Spickett [Wed, 12 Oct 2022 09:19:29 +0000 (09:19 +0000)]
[LLDB] Fix x86_64 build

28e65a6a63ab39be97d1a88fe7b4d0fa2f532643 changed the parameter
type of SetType but I forgot to build on x86 as well as arm64.

21 months ago[AArch64][SVE] Add instcombine for PTEST_ANY(X=OP(PG,...), X) -> PTEST_ANY(PG, X))
Cullen Rhodes [Wed, 12 Oct 2022 08:36:03 +0000 (08:36 +0000)]
[AArch64][SVE] Add instcombine for PTEST_ANY(X=OP(PG,...), X) -> PTEST_ANY(PG, X))

Given this is an OR reduction the two are equivalent and later
optimizations (AArch64InstrInfo::optimizePTestInstr) may rewrite the
sequence to use the flag-setting variant of instruction X, to remove the
PTEST altogether.

Reviewed By: paulwalker-arm, bsmith

Differential Revision: https://reviews.llvm.org/D134946

21 months ago[AArch64][SVE] NFC: auto-generate llvm/test/Transforms/InstCombine/AArch64/sve-intrin...
Cullen Rhodes [Fri, 7 Oct 2022 14:29:17 +0000 (14:29 +0000)]
[AArch64][SVE] NFC: auto-generate llvm/test/Transforms/InstCombine/AArch64/sve-intrinsics-ptest.ll

21 months ago[CMake] Add option LLVM_FORCE_CREATE_SYMLINKS
Tobias Hieta [Wed, 12 Oct 2022 08:04:21 +0000 (10:04 +0200)]
[CMake] Add option LLVM_FORCE_CREATE_SYMLINKS

On Windows we don't create symlinks for the binaries (clang++, clang-cl)
since the support requires special setup (group policy settings and
you need to know exactly our distribution story). But if you know
about these things and have a controlled environment there is a lot
of storage to be saved, so let's add a manual opt-in for using symlinks
on Windows with LLVM_FORCE_CREATE_SYMLINKS=ON.

Reviewed By: phosek

Differential Revision: https://reviews.llvm.org/D135578

21 months ago[clang-tidy] Dump effective diagnostics level in YAML output
Dmitry Polukhin [Thu, 6 Oct 2022 14:56:19 +0000 (07:56 -0700)]
[clang-tidy] Dump effective diagnostics level in YAML output

Before this patch YAML output had default diagnostic level instead of effective level reported to the user on stdout. Wrapper scripts for clang-tidy usually use YAML output and they pick wrong diagnostics level without this patch.

Test Plan: check-clang-tools

Differential Revision: https://reviews.llvm.org/D135367

21 months ago[Attributes] Make intrinsic attribute generation more flexible (NFC)
Nikita Popov [Tue, 11 Oct 2022 11:56:06 +0000 (13:56 +0200)]
[Attributes] Make intrinsic attribute generation more flexible (NFC)

Currently attributes for intrinsics are emitted using the
ArrayRef<AttrKind> based constructor for AttributeLists. This works
out fine for simple enum attributes, but doesn't really generalize
to attributes that accept values. We're already doing something
awkward for alignment attributes, and I'd like to have a cleaner
solution to this with
https://discourse.llvm.org/t/rfc-unify-memory-effect-attributes/65579 in mind.

The new generation approach is to instead directly construct
Attributes, giving us access to the full generality of that
interface. To avoid significantly increasing the size of the
generated code, we now also deduplicate the attribute sets. The
code generated per unique AttributeList looks like this:

  case 204: {
    AS[0] = {1, getIntrinsicArgAttributeSet(C, 5)};
    AS[1] = {AttributeList::FunctionIndex, getIntrinsicFnAttributeSet(C, 10)};
    NumAttrs = 2;
    break;
  }

and then the helper functions contain something like

  case 5:
    return AttributeSet::get(C, {
      Attribute::get(C, Attribute::NoCapture),
    });

and

  case 10:
    return AttributeSet::get(C, {
      Attribute::get(C, Attribute::NoUnwind),
      Attribute::get(C, Attribute::ArgMemOnly),
    });

A casualty of this change is the intrin-properties.td test, as I
don't think that FileCheck allows matching this kind of output.

Differential Revision: https://reviews.llvm.org/D135679

21 months ago[LLDB] Change EmulateInstriction::ReadRegister to return Optional
David Spickett [Wed, 28 Sep 2022 15:32:33 +0000 (15:32 +0000)]
[LLDB] Change EmulateInstriction::ReadRegister to return Optional

Making it easier to understand and harder to misuse.
This only applies to the ReadRegister(const RegisterInfo &reg_info) variant.

Depends on D135671

Reviewed By: clayborg

Differential Revision: https://reviews.llvm.org/D135672

21 months ago[AArch64][SVE] Fix BRKNS bug in optimizePTestInstr
Cullen Rhodes [Wed, 12 Oct 2022 08:17:45 +0000 (08:17 +0000)]
[AArch64][SVE] Fix BRKNS bug in optimizePTestInstr

The BRKNS instruction is unlike the other instructions that set flags
since it has an all active implicit predicate, so the existing

  PTEST(PG, BRKN(PG, A, B)) -> BRKNS(PG, A, B)

in AArch64InstrInfo::optimizePTestInstr is incorrect, however

  PTEST(PTRUE_B(31), BRKN(PG, A, B)) -> BRKNS(PG, A, B)

is correct.

Spotted by @paulwalker-arm in D134946.

Reviewed By: paulwalker-arm

Differential Revision: https://reviews.llvm.org/D135655

21 months ago[AArch64] NFC: Auto-generate llvm/test/CodeGen/AArch64/sve-ptest-removal-brk.ll
Cullen Rhodes [Wed, 28 Sep 2022 09:25:21 +0000 (09:25 +0000)]
[AArch64] NFC: Auto-generate llvm/test/CodeGen/AArch64/sve-ptest-removal-brk.ll

21 months ago[LLDB] Change RegisterValue::SetType param to const RegisterInfo&
David Spickett [Wed, 28 Sep 2022 14:52:12 +0000 (14:52 +0000)]
[LLDB] Change RegisterValue::SetType param to const RegisterInfo&

No one was pasing nullptr here.

Depends on D135670

Reviewed By: clayborg

Differential Revision: https://reviews.llvm.org/D135671

21 months ago[LLDB] Pass const RegisterInfo& to RegisterValue::SetValueFromData
David Spickett [Wed, 28 Sep 2022 14:44:17 +0000 (14:44 +0000)]
[LLDB] Pass const RegisterInfo& to RegisterValue::SetValueFromData

Familiar story, callers are either checking upfront that the pointer
wasn't null or not checking at all. SetValueFromData itself didn't
check either.

So make the parameter a ref and fixup the few places where a nullptr
check seems needed.

Depends on D135668

Reviewed By: clayborg

Differential Revision: https://reviews.llvm.org/D135670

21 months ago[mlir] add transform dialect entry point
Alex Zinenko [Tue, 11 Oct 2022 15:23:48 +0000 (15:23 +0000)]
[mlir] add transform dialect entry point

Introduce `transform::applyTransforms` as a top-level entry point to the
Transform dialect-driven transformation infrastructure, by analogy with
`applyFull/PartialConversion`. Clients are expected to use this function
and no longer need to maintain the transformation state. Make the
constructor of the TransformState private for that purpose.

Reviewed By: springerm

Differential Revision: https://reviews.llvm.org/D135681

21 months ago[LLDB] Change RegisterValue::SetFromMemoryData to const RegisterInfo&
David Spickett [Wed, 28 Sep 2022 14:24:57 +0000 (14:24 +0000)]
[LLDB] Change RegisterValue::SetFromMemoryData to const RegisterInfo&

All callers were either assuming their pointer was not null before calling
this, or checking beforehand.

Reviewed By: clayborg

Differential Revision: https://reviews.llvm.org/D135668

21 months ago[LLD] [MinGW] Look for libs named "<libname>.lib" even with -static
Martin Storsjö [Tue, 11 Oct 2022 08:11:31 +0000 (11:11 +0300)]
[LLD] [MinGW] Look for libs named "<libname>.lib" even with -static

This matches how ld.bfd works in practice; this fixes
https://github.com/mstorsjo/llvm-mingw/issues/305.

Adding a test for the new lib name combination that this allows, but
also adding a few negative tests for combinations that aren't
matched when -static is specified (because this change in itself
didn't break any of the existing tests either).

The logic in how ld.bfd looks for various libraries based on
an -l<libname> argument is rather complex; the
ldemul_open_dynamic_archive function looks for various combinations:
https://sourceware.org/git/?p=binutils-gdb.git;a=blob;f=ld/emultempl/pep.em;h=e68d1e69f17ad73af065b6bed19ae89ded913172;hb=b51c2fec1da205ea3e7354cbb3e253018d64873c#l2066
This function is only called if looking for dynamic libraries
(i.e. if -static wasn't specified):
https://sourceware.org/git/?p=binutils-gdb.git;a=blob;f=ld/ldfile.c;h=731ae5f7aedcf921bd36a1b32a3e0f5bfa189071;hb=b51c2fec1da205ea3e7354cbb3e253018d64873c#l365

However even this function is skipped, it still looks for libraries
in the form of "lib<libname>.a" (this is what lld did before):
https://sourceware.org/git/?p=binutils-gdb.git;a=blob;f=ld/ldfile.c;h=731ae5f7aedcf921bd36a1b32a3e0f5bfa189071;hb=b51c2fec1da205ea3e7354cbb3e253018d64873c#l440
But it also calls a format specific function called
ldemul_find_potential_libraries, which for PE targets looks for
files named "<libname>.lib":
https://sourceware.org/git/?p=binutils-gdb.git;a=blob;f=ld/emultempl/pep.em;h=e68d1e69f17ad73af065b6bed19ae89ded913172;hb=b51c2fec1da205ea3e7354cbb3e253018d64873c#l2175

Differential Revision: https://reviews.llvm.org/D135651

21 months ago[MC] [Win64EH] Generate ARM64 packed unwind info with signed return addresses
Martin Storsjö [Tue, 11 Oct 2022 11:29:36 +0000 (14:29 +0300)]
[MC] [Win64EH] Generate ARM64 packed unwind info with signed return addresses

Differential Revision: https://reviews.llvm.org/D135660

21 months ago[AArch64] Add support for the SEH opcode for return address signing
Martin Storsjö [Wed, 5 Oct 2022 10:42:46 +0000 (13:42 +0300)]
[AArch64] Add support for the SEH opcode for return address signing

This was documented upstream in
https://github.com/MicrosoftDocs/cpp-docs/pull/4202.

Differential Revision: https://reviews.llvm.org/D135276

21 months ago[runtimes] Use a response file for runtimes test suites
Petr Hosek [Thu, 18 Aug 2022 08:25:13 +0000 (08:25 +0000)]
[runtimes] Use a response file for runtimes test suites

We don't know which test suites are going to be included by runtimes
builds so we cannot include those before running the sub-build, but
that's not possible during the LLVM build configuration. We instead use
a response file that's populated by the runtimes build as a level of
indirection.

This addresses the issue described in:
https://discourse.llvm.org/t/cmake-regeneration-is-broken/62788

Differential Revision: https://reviews.llvm.org/D132438

21 months ago[mlir][Math] Change regex to match fp value on different target.
jacquesguan [Thu, 29 Sep 2022 03:19:25 +0000 (11:19 +0800)]
[mlir][Math] Change regex to match fp value on different target.

Link: https://github.com/llvm/llvm-project/issues/58048
Reviewed By: ftynse, Mogball

Differential Revision: https://reviews.llvm.org/D134850

21 months agoRevert "Fix locating of COMPILER_RT_LIBCXX_PATH"
Vitaly Buka [Wed, 12 Oct 2022 06:24:13 +0000 (23:24 -0700)]
Revert "Fix locating of COMPILER_RT_LIBCXX_PATH"

clang-aarch64-sve-vla crashes in msan
clang-s390x-linux-lnt fails to compile libcxx

This reverts commit 14b2b67cd7d5e439c4bc5c8051f68a5fbb683480.

21 months ago[SimpleLoopUnswitch][NFC] Separate legality checks from cost computation
Max Kazantsev [Wed, 12 Oct 2022 06:17:28 +0000 (13:17 +0700)]
[SimpleLoopUnswitch][NFC] Separate legality checks from cost computation

These are semantically two different stages, but were entwined in the
old implementation. Now cost computation does not do legality checks,
and they all are done beforehead.

21 months ago[NFC] Factor out computation of best unswitch cost candidate
Max Kazantsev [Wed, 12 Oct 2022 05:21:29 +0000 (12:21 +0700)]
[NFC] Factor out computation of best unswitch cost candidate

Split out a major peice of this method to make code more readable.

21 months ago[llvm] [cmake] Support finding both static and shared zstd via FindZstd
Michał Górny [Fri, 7 Oct 2022 16:42:27 +0000 (18:42 +0200)]
[llvm] [cmake] Support finding both static and shared zstd via FindZstd

Improve the logic in FindZstd to support finding both shared and static
variants of the zstd library simultaneously.  Otherwise, if the shared
library is installed, zstd::libzstd_static is not declared at all
and CMake fails if LLVM_USE_STATIC_ZSTD is used.

Differential Revision: https://reviews.llvm.org/D135457

21 months ago[clang] Do not override libclang.so's SOVERSION if CLANG_FORCE_MATCHING_LIBCLANG_SOVE...
Michał Górny [Tue, 11 Oct 2022 18:01:11 +0000 (20:01 +0200)]
[clang] Do not override libclang.so's SOVERSION if CLANG_FORCE_MATCHING_LIBCLANG_SOVERSION

Instead of setting libclang.so's SOVERSION to CLANG_MAJOR_VERSION
when CLANG_FORCE_MATCHING_LIBCLANG_SOVERSION is enabled, do not override
it and let LLVM set the correct SOVERSION.  This fixes libclang.so
missing LLVM_VERSION_SUFFIX, and harmonizes the SOVERSION with other
shared libraries.

Differential Revision: https://reviews.llvm.org/D135701

21 months ago[mlir][async] Fix missing dialect dependency in async-to-async-runtime
rkayaith [Tue, 11 Oct 2022 20:46:26 +0000 (16:46 -0400)]
[mlir][async] Fix missing dialect dependency in async-to-async-runtime

The specific case I hit this was when lowering an `async.execute` that's
not inside a `func.func` op, e.g.:
```
llvm.func @foo() {
  %token = async.execute {
    async.yield
  }
  llvm.return
}
```

Reviewed By: bondhugula

Differential Revision: https://reviews.llvm.org/D135742

21 months agoApply clang-tidy fixes for performance-unnecessary-value-param in TileUsingInterface...
Mehdi Amini [Thu, 6 Oct 2022 19:19:41 +0000 (19:19 +0000)]
Apply clang-tidy fixes for performance-unnecessary-value-param in TileUsingInterface.cpp (NFC)

21 months agoApply clang-tidy fixes for performance-for-range-copy in TileUsingInterface.cpp ...
Mehdi Amini [Thu, 6 Oct 2022 19:19:08 +0000 (19:19 +0000)]
Apply clang-tidy fixes for performance-for-range-copy in TileUsingInterface.cpp (NFC)

21 months agoFix typo in tablegen option: '-gen-typedef-docs' -> '-gen-typedef-doc'
Mehdi Amini [Wed, 12 Oct 2022 05:03:05 +0000 (05:03 +0000)]
Fix typo in tablegen option: '-gen-typedef-docs' -> '-gen-typedef-doc'

21 months agoFix locating of COMPILER_RT_LIBCXX_PATH
Vitaly Buka [Thu, 29 Sep 2022 01:53:39 +0000 (18:53 -0700)]
Fix locating of COMPILER_RT_LIBCXX_PATH

COMPILER_RT_LIBCXX_PATH is required for msan unittests.

21 months agoFix assert in generated `direct` property getter/setters due to removal of `_cmd...
Michael Wyman [Tue, 11 Oct 2022 21:32:21 +0000 (14:32 -0700)]
Fix assert in generated `direct` property getter/setters due to removal of `_cmd` parameter.

This fixes a bug from https://reviews.llvm.org/D131424 that removed the implicit `_cmd` parameter as an argument to `objc_direct` method implementations. In many cases the generated getter/setter will call `objc_getProperty` or `objc_setProperty`, both of which require the selector of the getter/setter; since `_cmd` didn't automatically have backing storage, attempting to load the address asserted.

For direct property generated getters/setters, this now passes an undefined/uninitialized/poison value as the `_cmd` argument to `objc_getProperty`/`objc_setProperty`. Prior to removing the `_cmd` argument from the ABI of direct methods, it was left uninitialized/undefined; although references within hand-implemented methods would load the selector in the method prologue, generated getters/setters never did and just forwarded the undefined value that was passed as the argument.

This change keeps the generated code mostly similar to before, passing an uninitialized/undefined/poison value; for setters, the value argument may be moved to another register.

Added a test that triggers the assert prior to the implementation code.

Differential Revision: https://reviews.llvm.org/D135091

21 months ago[lld-macho] Canonicalize personality pointers in EH frames
Jez Ng [Wed, 12 Oct 2022 03:50:46 +0000 (23:50 -0400)]
[lld-macho] Canonicalize personality pointers in EH frames

We already do this for personality pointers referenced from compact
unwind entries; this patch extends that behavior to personalities
referenced via EH frames as well.

This reduces the number of distinct personalities we need in the final
binary, and helps us avoid hitting the "too many personalities" error.

I renamed `UnwindInfoSection::prepareRelocations()` to simply `prepare`
since we now do some non-reloc-specific stuff within.

Fixes #58277.

Reviewed By: #lld-macho, oontvoo

Differential Revision: https://reviews.llvm.org/D135728

21 months ago[AArch64][Windows] Check sret attribute also for inreg attribute
chenglin.bi [Wed, 12 Oct 2022 01:57:35 +0000 (09:57 +0800)]
[AArch64][Windows] Check sret attribute also for inreg attribute

Fix the issue: #57684

Reviewed By: efriedma

Differential Revision: https://reviews.llvm.org/D135512

21 months ago[RISCV] Add isel patterns for vmacc, vnmsac.
Yeting Kuo [Wed, 5 Oct 2022 15:34:37 +0000 (23:34 +0800)]
[RISCV] Add isel patterns for vmacc, vnmsac.

The patch selects VSELECT/VP_MERGE_VL which uses fmadd/fnmsub as true operand
and the adden of the fmadd/fnmsub as false operand.

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D135330

21 months agoApply clang-tidy fixes for readability-simplify-boolean-expr in BufferizableOpInterfa...
Mehdi Amini [Thu, 6 Oct 2022 19:16:56 +0000 (19:16 +0000)]
Apply clang-tidy fixes for readability-simplify-boolean-expr in BufferizableOpInterfaceImpl.cpp (NFC)

21 months agoApply clang-tidy fixes for llvm-else-after-return in MultiBuffer.cpp (NFC)
Mehdi Amini [Thu, 6 Oct 2022 19:13:00 +0000 (19:13 +0000)]
Apply clang-tidy fixes for llvm-else-after-return in MultiBuffer.cpp (NFC)

21 months ago[RISCV] Use mask/tail agnostic if tied source is IMPLICIT_DEF regardless of the polic...
Craig Topper [Tue, 11 Oct 2022 23:36:21 +0000 (16:36 -0700)]
[RISCV] Use mask/tail agnostic if tied source is IMPLICIT_DEF regardless of the policy operand.

If the source is implicit_def, the register allocator won't have
any constraint on what register it picks for the destination. This
doesn't give the user much control of what register is being used.

So in my mind that means the only reason to honor the policy operand
is to control what policy is used in vsetvli to maybe avoid a vtype
change. Given the other optimizations we do on the policy field, I
don't think allowing the user this control is reliable.

Therefore, I think we should use agnostic policies if the source is
undef.

This should give better performance on some CPUs for VP intrinsics where
there is no merge operand and the backend adds IMPLICIT_DEF to the instruction.

Differential Revision: https://reviews.llvm.org/D135396

21 months agoRevert "[DAGCombiner] Fold (mul (sra X, BW-1), Y) -> (neg (and (sra X, BW-1), Y))"
Craig Topper [Tue, 11 Oct 2022 23:30:40 +0000 (16:30 -0700)]
Revert "[DAGCombiner] Fold (mul (sra X, BW-1), Y) -> (neg (and (sra X, BW-1), Y))"

This reverts commit 0148df8157f05ecf3b1064508e6f012aefb87dad.

Getting a lit test failures on AMDGPU but I can't reproduce it so far.
Reverting to investigate.

21 months ago[RISCV] Remove unused SchedClass WriteVFCvtFToFV. NFC
Craig Topper [Tue, 11 Oct 2022 23:25:28 +0000 (16:25 -0700)]
[RISCV] Remove unused SchedClass WriteVFCvtFToFV. NFC

This isn't bound to any instruction. From the section comment it
was for single-width F-to-F conversions, but those don't exist.

21 months ago[llvm-ocaml] Replace all typed pointer functions with opaque pointer functions
Alan Hu [Tue, 11 Oct 2022 23:25:29 +0000 (16:25 -0700)]
[llvm-ocaml] Replace all typed pointer functions with opaque pointer functions

Reviewed By: aeubanks

Differential Revision: https://reviews.llvm.org/D135524

21 months ago[mlir][sparse] Add MLIR interface for sparse tensor file input and output.
bixia1 [Tue, 11 Oct 2022 22:28:20 +0000 (15:28 -0700)]
[mlir][sparse] Add MLIR interface for sparse tensor file input and output.

Reviewed By: aartbik, wrengr

Differential Revision: https://reviews.llvm.org/D135480

21 months ago[DAGCombiner] Fold (mul (sra X, BW-1), Y) -> (neg (and (sra X, BW-1), Y))
Craig Topper [Tue, 11 Oct 2022 23:05:04 +0000 (16:05 -0700)]
[DAGCombiner] Fold (mul (sra X, BW-1), Y) -> (neg (and (sra X, BW-1), Y))

(sra X, BW-1) is either 0 or -1. So the multiply is a conditional
negate of Y.

This pattern shows up when type legalizing wide multiplies involving
a sign extended value.

Fixes PR57549.

Reviewed By: RKSimon

Differential Revision: https://reviews.llvm.org/D133399

21 months ago[lld-macho] Add --ignore-auto-link-option
Keith Smiley [Sun, 9 Oct 2022 04:28:02 +0000 (21:28 -0700)]
[lld-macho] Add --ignore-auto-link-option

This provides a workaround for a small difference in ld64 behavior where
ld64 ignores invalid LC_LINKER_OPTIONS unless the link fails. Instead of
fully adopting that behavior, this provides an escape hatch by allowing
users to specify `--ignore-auto-link-option` passing the invalid library
or framework name

Fixes https://github.com/llvm/llvm-project/issues/56939

Differential Revision: https://reviews.llvm.org/D135530

21 months ago[clang][deps] Prevent emitting diagnostics outside of source file
Jan Svoboda [Tue, 11 Oct 2022 22:38:48 +0000 (15:38 -0700)]
[clang][deps] Prevent emitting diagnostics outside of source file

The dependency scanner needs to report the module map file describing the module whose implementation is being compiled (see D134222). However, calling `Preprocessor::getCurrentModuleImplementation()` in the scanner might cause a diagnostic during module map parsing and emitting a diagnostic without being "in" a source file is illegal (e.g. in `TextDiagnosticPrinter`). This patch ensures the module map parse is triggered while the compiler is still "in" a source file, avoiding the failure case.

Reviewed By: Bigcheese

Differential Revision: https://reviews.llvm.org/D135637

21 months ago[clang][deps] Remove more codegen options
Jan Svoboda [Tue, 11 Oct 2022 22:38:41 +0000 (15:38 -0700)]
[clang][deps] Remove more codegen options

Codegen options are typically unused by modules. Reset some of them to increase sharing between TUs with different flags.

Reviewed By: Bigcheese

Differential Revision: https://reviews.llvm.org/D135720

21 months ago[Clang] Do not build the OffloadActionBuilder when using the new driver
Joseph Huber [Tue, 11 Oct 2022 21:11:21 +0000 (16:11 -0500)]
[Clang] Do not build the OffloadActionBuilder when using the new driver

The Offloading toolchain currently has two methods for construction the
requires actions. The "new" driver and the old `OffloadActionBuilder`.
Using either one is mutually exclusive, so we should not initialize this
when using the new driver. This was causing some error messages to be
printed multiple times because we were checking them in both the old and
the new driver.

Reviewed By: tra

Differential Revision: https://reviews.llvm.org/D135715

21 months agoPrevent lldb-vscode tests from source lldbinit file
Jeffrey Tan [Fri, 30 Sep 2022 19:11:58 +0000 (12:11 -0700)]
Prevent lldb-vscode tests from source lldbinit file

lldb-vscode is hard-coded to source .lldbinit file which causes some tests to
fail on my machine.
This patch adds a new option to control this:
1. vscode.py and lldb-vscode tests will not source .lldbinit by default
2. lldb-vscode will source .lldbinit in production if not specified otherwise

Differential Revision: https://reviews.llvm.org/D135620

21 months ago[GlobalISel] Allow vectors in redundant or + add combines
Jessica Paquette [Tue, 11 Oct 2022 21:49:41 +0000 (14:49 -0700)]
[GlobalISel] Allow vectors in redundant or + add combines

We support KnownBits for vectors, so we can enable these.

https://godbolt.org/z/r9a9W4Gj1

Differential Revision: https://reviews.llvm.org/D135719

21 months ago[llvm] Update WPD tests to use opaque pointers
Leonard Chan [Tue, 11 Oct 2022 21:10:07 +0000 (21:10 +0000)]
[llvm] Update WPD tests to use opaque pointers

Differential Revision: https://reviews.llvm.org/D135611

21 months ago[LTO] Make local linkage GlobalValue in non-prevailing COMDAT available_externally
Fangrui Song [Tue, 11 Oct 2022 22:30:07 +0000 (15:30 -0700)]
[LTO] Make local linkage GlobalValue in non-prevailing COMDAT available_externally

For a local linkage GlobalObject in a non-prevailing COMDAT, it remains defined while its
leader has been made available_externally. This violates the COMDAT rule that
its members must be retained or discarded as a unit.

To fix this, update the regular LTO change D34803 to track local linkage
GlobalValues, and port the code to ThinLTO (GlobalAliases are not handled.)

This fixes two problems.

(a) `__cxx_global_var_init` in a non-prevailing COMDAT group used to
linger around (unreferenced, hence benign), and is now correctly discarded.
```
int foo();
inline int v = foo();
```

(b) Fix https://github.com/llvm/llvm-project/issues/58215:
as a size optimization, we place private `__profd_` in a COMDAT with a
`__profc_` key. When FuncImport.cpp makes `__profc_` available_externally due to
a non-prevailing COMDAT, `__profd_` incorrectly remains private. This change
makes the `__profd_` available_externally.

```
cat > c.h <<'eof'
extern void bar();
inline __attribute__((noinline)) void foo() {}
eof
cat > m1.cc <<'eof'
int main() {
  bar();
  foo();
}
eof
cat > m2.cc <<'eof'
__attribute__((noinline)) void bar() {
  foo();
}
eof

clang -O2 -fprofile-generate=./t m1.cc m2.cc -flto -fuse-ld=lld -o t_gen
rm -fr t && ./t_gen && llvm-profdata show -function=foo t/default_*.profraw

clang -O2 -fprofile-generate=./t m1.cc m2.cc -flto=thin -fuse-ld=lld -o t_gen
rm -fr t && ./t_gen && llvm-profdata show -function=foo t/default_*.profraw
```

Reviewed By: tejohnson

Differential Revision: https://reviews.llvm.org/D135427

21 months ago[mlir][sparse] Add rewriting rules for the sparse_tensor.new operator.
bixia1 [Tue, 11 Oct 2022 20:47:45 +0000 (13:47 -0700)]
[mlir][sparse] Add rewriting rules for the sparse_tensor.new operator.

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D135692

21 months ago[MC] .addrsig_sym: ignore unregistered symbols
Fangrui Song [Tue, 11 Oct 2022 22:07:14 +0000 (15:07 -0700)]
[MC] .addrsig_sym: ignore unregistered symbols

.addrsig_sym forces registering the symbol regardless whether it is otherwise
registered. This creates an undefined symbol which is inconvenient/undesired:

* `extern int x; void f() { (void)x; }` has inconsistent behavior whether `x` is emitted as an undefined symbol.
  `-O0 -faddrsig` makes `x` undefined while other -O levels and -fno-addrsig eliminate the symbol.
* In ThinLTO, after a non-prevailing linkonce_odr definition is converted to available_externally, and then a declaration,
  the addrsig code emits a symbol while the symbol is otherwise unseen.

D135427 fixed a bug that a non-prevailing `__cxx_global_var_init` was
incorrectly retained. However, the IR declaration causes an undesired
`.addrsig_sym __cxx_global_var_init`. This can be addressed in a way similar
to D101512 (`isTransitiveUsedByMetadataOnly`) but the increased
`OutStreamer->emitAddrsigSym(getSymbol(&GV));` complexity makes me nervous.
Just ignoring unregistered symbols circumvents the problem.

Reviewed By: rnk

Differential Revision: https://reviews.llvm.org/D135642

21 months ago[mlir][sparse] Renaming mlir_sparse_tensor_utils library to SparseTensorRuntime
wren romano [Mon, 10 Oct 2022 20:41:58 +0000 (13:41 -0700)]
[mlir][sparse] Renaming mlir_sparse_tensor_utils library to SparseTensorRuntime

The "mlir_xxx_utils" naming scheme is reserved/intended for shared libraries, whereas this library must be static due to issues of linking DLLs on Windows.  So we rename the library to avoid any potential confusion.   In addition we also rename the ExecutionEngine/SparseTensorUtils.{h,cpp} files to match the new library name.

Reviewed By: aartbik, stella.stamenova

Differential Revision: https://reviews.llvm.org/D135613

21 months ago[CallGraph] Port -print-callgraph-sccs to new pass manager
Arthur Eubanks [Fri, 7 Oct 2022 21:08:43 +0000 (14:08 -0700)]
[CallGraph] Port -print-callgraph-sccs to new pass manager

And remove the legacy opt-specific pass.

Reviewed By: asbirlea

Differential Revision: https://reviews.llvm.org/D135487

21 months ago[mlir][sparse] Move a few routines to CodegenUtils.
bixia1 [Tue, 11 Oct 2022 20:22:14 +0000 (13:22 -0700)]
[mlir][sparse] Move a few routines to CodegenUtils.

Move a few supporting routines for generating function calls to CodegenUtils so
that they can be used by the codegen path for sparse tensor file input and
output.

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D135691