platform/upstream/llvm.git
3 years ago[AArch64][SVE] Add missing target require to test
Bradley Smith [Mon, 21 Jun 2021 14:18:25 +0000 (15:18 +0100)]
[AArch64][SVE] Add missing target require to test

Differential revision: https://reviews.llvm.org/D104643

3 years ago[FuncSpec] Add minsize test. NFC.
Sjoerd Meijer [Mon, 21 Jun 2021 14:13:40 +0000 (15:13 +0100)]
[FuncSpec] Add minsize test. NFC.

3 years ago[NFC][libomptarget] Remove redundant libelf dependency for elf_common.
Vyacheslav Zakharin [Fri, 18 Jun 2021 16:46:10 +0000 (09:46 -0700)]
[NFC][libomptarget] Remove redundant libelf dependency for elf_common.

Differential Revision: https://reviews.llvm.org/D104549

3 years ago[clang-cl] Implement /external:I, /external:env, and EXTERNAL_INCLUDE support (PR36003)
Hans Wennborg [Wed, 16 Jun 2021 13:22:17 +0000 (15:22 +0200)]
[clang-cl] Implement /external:I, /external:env, and EXTERNAL_INCLUDE support (PR36003)

This patch does three things:

- Map the /external:I flag to -isystem

- Add support for the /external:env:<var> flag which reads system
  include paths from the <var> environment variable

- Pick up system include dirs EXTERNAL_INCLUDE in addition to the old
  INCLUDE environment variable.

Differential revision: https://reviews.llvm.org/D104387

3 years ago[ARM] Transform a fixed-point to floating-point conversion into a VCVT_fix
Sam Tebbs [Mon, 7 Jun 2021 12:29:38 +0000 (13:29 +0100)]
[ARM] Transform a fixed-point to floating-point conversion into a VCVT_fix

Conversion from a fixed-point number to a floating-point number is done by
multiplying the fixed-point number by 2^(-n) where n is the number of
fractional bits. Currently this is lowered to a vcvt
(integer to floating-point) then a vmul, but it can instead be lowered
directly to a vcvt (fixed-point to floating-point). This patch enables
such transformations as long as the multiplication factor is a power of 2.

Differential Revision: https://reviews.llvm.org/D103903

3 years ago[NFC] Fix typo
Sebastian Neubauer [Mon, 21 Jun 2021 12:59:30 +0000 (14:59 +0200)]
[NFC] Fix typo

3 years ago[AArch64][SVE] Wire up vscale_range attribute to SVE min/max vector queries
Bradley Smith [Fri, 4 Jun 2021 14:24:35 +0000 (15:24 +0100)]
[AArch64][SVE] Wire up vscale_range attribute to SVE min/max vector queries

Differential Revision: https://reviews.llvm.org/D103702

3 years ago[Sema][SVE] Properly match builtin ID when using aux target
Bradley Smith [Fri, 18 Jun 2021 12:48:13 +0000 (13:48 +0100)]
[Sema][SVE] Properly match builtin ID when using aux target

Differential Revision: https://reviews.llvm.org/D104539

3 years ago[lldb][NFC] Remove redundant deleted constructors in HostInfoBase subclasses
Raphael Isemann [Mon, 21 Jun 2021 11:19:15 +0000 (13:19 +0200)]
[lldb][NFC] Remove redundant deleted constructors in HostInfoBase subclasses

HostInfoBase has a deleted dtor/ctor so there is no need to do the same for
all the classes inheriting from it.

Reviewed By: DavidSpickett, JDevlieghere

Differential Revision: https://reviews.llvm.org/D104221

3 years ago[LoopIdiom] Add test case that involves adds with flags and zero exts.
Florian Hahn [Mon, 21 Jun 2021 09:49:26 +0000 (10:49 +0100)]
[LoopIdiom] Add test case that involves adds with flags and zero exts.

Test coverage to ensure D104319 does not introduce a regression here.

3 years ago[mlir][ODS] Fix copy ctor for generate Pass classes
Vladislav Vinogradov [Tue, 15 Jun 2021 15:09:31 +0000 (18:09 +0300)]
[mlir][ODS] Fix copy ctor for generate Pass classes

Redirect the copy ctor to the actual class instead of
overwriting it with `TypeID` based ctor.

This allows the final Pass classes to have extra fields and logic for their copy.

Reviewed By: lattner

Differential Revision: https://reviews.llvm.org/D104302

3 years ago[NFC] Wrap entire assert-only block in LLVM_DEBUG
Jordan Rupprecht [Mon, 21 Jun 2021 11:01:09 +0000 (04:01 -0700)]
[NFC] Wrap entire assert-only block in LLVM_DEBUG

3 years ago[mlir][Linalg] Silence warnings in Release builds. NFC.
Benjamin Kramer [Mon, 21 Jun 2021 10:55:25 +0000 (12:55 +0200)]
[mlir][Linalg] Silence warnings in Release builds. NFC.

mlir/lib/Dialect/Linalg/Transforms/ComprehensiveBufferize.cpp:940:8: warning: unused variable 'opProducesRootRead' [-Wunused-variable]
  bool opProducesRootRead =
       ^
mlir/lib/Dialect/Linalg/Transforms/ComprehensiveBufferize.cpp:942:8: warning: unused variable 'opProducesRootWrite' [-Wunused-variable]
  bool opProducesRootWrite =
       ^
mlir/lib/Dialect/Linalg/Transforms/ComprehensiveBufferize.cpp:1498:11: warning: unused variable 'resultNumber' [-Wunused-variable]
  int64_t resultNumber = result.getResultNumber();
          ^
mlir/lib/Dialect/Linalg/Transforms/ComprehensiveBufferize.cpp:1497:11: warning: unused variable 'operandNumber' [-Wunused-variable]
  int64_t operandNumber = operand.getOperandNumber();
          ^
mlir/lib/Dialect/Linalg/Transforms/ComprehensiveBufferize.cpp:267:20: warning: unused function 'getInPlace' [-Wunused-function]
static InPlaceSpec getInPlace(Value v) {
                   ^

3 years ago[VP][NFCI] Address various clang-tidy warnings
Fraser Cormack [Mon, 21 Jun 2021 08:50:27 +0000 (09:50 +0100)]
[VP][NFCI] Address various clang-tidy warnings

Reviewed By: simoll

Differential Revision: https://reviews.llvm.org/D104288

3 years ago[lldb] Assert that CommandResultObject error messages are not empty
David Spickett [Fri, 18 Jun 2021 12:17:53 +0000 (12:17 +0000)]
[lldb] Assert that CommandResultObject error messages are not empty

The intention is now that AppendError/SetError/AppendRawError only
be called with some message to show. This enforces that.

For SetError with a Status and a fallback string first assert
that the Status is a failure Status. Then it calls SetError(StringRef)
which checks the message is valid. (which could be the fallback
or the Status')

Reviewed By: dblaikie

Differential Revision: https://reviews.llvm.org/D104525

3 years ago[AMDGPU] Fix linking with shared libraries
Sebastian Neubauer [Mon, 21 Jun 2021 09:11:09 +0000 (11:11 +0200)]
[AMDGPU] Fix linking with shared libraries

AMDGPULDSUtils depends on llvm::CallGraph.

3 years ago[Mem2Reg] Regenerate test checks (NFC)
Nikita Popov [Mon, 21 Jun 2021 09:05:31 +0000 (11:05 +0200)]
[Mem2Reg] Regenerate test checks (NFC)

3 years ago[mlir][NFC] Remove Standard dialect dependency on MemRef dialect
Matthias Springer [Mon, 21 Jun 2021 08:45:16 +0000 (17:45 +0900)]
[mlir][NFC] Remove Standard dialect dependency on MemRef dialect

* Remove dependency: Standard --> MemRef
* Add dependencies: GPUToNVVMTransforms --> MemRef, Linalg --> MemRef, MemRef --> Tensor
* Note: The `subtensor_insert_propagate_dest_cast` test case in MemRef/canonicalize.mlir will be moved to Tensor/canonicalize.mlir in a subsequent commit, which moves over the remaining Tensor ops from the Standard dialect to the Tensor dialect.

Differential Revision: https://reviews.llvm.org/D104506

3 years ago[Mem2Reg] Use poison for unreachable cases
Nikita Popov [Mon, 21 Jun 2021 08:47:14 +0000 (10:47 +0200)]
[Mem2Reg] Use poison for unreachable cases

Use poison instead of undef for cases dealing with unreachable
code. This still leaves the more interesting case of "load from
uninitialized memory" as undef.

3 years ago[Mem2Reg] Regenerate test checks (NFC)
Nikita Popov [Mon, 21 Jun 2021 08:47:59 +0000 (10:47 +0200)]
[Mem2Reg] Regenerate test checks (NFC)

3 years ago[InstCombine] Fold icmp (select c,const,arg), null if icmp arg, null can be simplified
Juneyoung Lee [Mon, 21 Jun 2021 07:50:54 +0000 (16:50 +0900)]
[InstCombine] Fold icmp (select c,const,arg), null if icmp arg, null can be simplified

This patch folds icmp (select c,const,arg), null if icmp arg, null can be simplified.

Resolves llvm.org/pr48975.

Reviewed By: nikic, xbolva00

Differential Revision: https://reviews.llvm.org/D96663

3 years ago[FuncSpec] Don't specialise functions with NoDuplicate instructions.
Sjoerd Meijer [Thu, 17 Jun 2021 13:43:02 +0000 (14:43 +0100)]
[FuncSpec] Don't specialise functions with NoDuplicate instructions.

getSpecializationCost was returning INT_MAX for a case when specialisation
shouldn't happen, but this wasn't properly checked if specialisation was
forced.

Differential Revision: https://reviews.llvm.org/D104461

3 years ago[mlir][linalg] Support low padding in subtensor(pad_tensor) lowering
Matthias Springer [Mon, 21 Jun 2021 07:29:42 +0000 (16:29 +0900)]
[mlir][linalg] Support low padding in subtensor(pad_tensor) lowering

Differential Revision: https://reviews.llvm.org/D104591

3 years ago[gn build] Port 208332de8abf
LLVM GN Syncbot [Mon, 21 Jun 2021 07:27:34 +0000 (07:27 +0000)]
[gn build] Port 208332de8abf

3 years ago[AMDGPU] Add Optimize VGPR LiveRange Pass.
Ruiling Song [Mon, 19 Apr 2021 02:45:41 +0000 (10:45 +0800)]
[AMDGPU] Add Optimize VGPR LiveRange Pass.

This pass aims to optimize VGPR live-range in a typical divergent if-else
control flow. For example:

def(a)
if(cond)
  use(a)
  ... // A
else
  use(a)

As AMDGPU access vgpr with respect to active-mask, we can mark `a` as
dead in region A. For details, please refer to the comments in
implementation file.

The pass is enabled by default, the frontend can disable it through
"-amdgpu-opt-vgpr-liverange=false".

Differential Revision: https://reviews.llvm.org/D102212

3 years ago[mlir][Linalg] NFC - Drop unused variable definition.
Nicolas Vasilache [Mon, 21 Jun 2021 07:08:02 +0000 (07:08 +0000)]
[mlir][Linalg] NFC - Drop unused variable definition.

3 years ago[mlir][Linalg] Introduce a BufferizationAliasInfo (6/n)
Nicolas Vasilache [Wed, 16 Jun 2021 11:39:40 +0000 (11:39 +0000)]
[mlir][Linalg] Introduce a BufferizationAliasInfo (6/n)

This revision adds a BufferizationAliasInfo which maintains and updates information about which tensors will alias once bufferized, which bufferized tensors are equivalent to others and how to handle clobbers.

Bufferization greedily tries to bufferize inplace by:

1. first trying to bufferize SubTensorInsertOp inplace, in reverse order (these are deemed the most expensives).
2. then trying to bufferize all non SubTensorOp / SubTensorInsertOp, in reverse order.
3. lastly trying to bufferize all SubTensorOp in reverse order.

Reverse order is a heuristic that seems to work nicely because structured tensor codegen very often proceeds by:

1. take a subset of a tensor
2. compute on that subset
3. insert the result subset into the full tensor and yield a new tensor.

BufferizationAliasInfo + equivalence sets + clobber analysis allows bufferizing nested
subtensor/compute/subtensor_insert sequences inplace to a certain extent.
To fully realize inplace bufferization, additional container-containee analysis will be necessary and is left for a subsequent commit.

Differential revision: https://reviews.llvm.org/D104110

3 years ago[gn build] Port 80fd5fa5269c
LLVM GN Syncbot [Mon, 21 Jun 2021 06:23:08 +0000 (06:23 +0000)]
[gn build] Port 80fd5fa5269c

3 years ago[AMDGPU] Replace non-kernel function uses of LDS globals by pointers.
hsmahesha [Mon, 21 Jun 2021 05:25:23 +0000 (10:55 +0530)]
[AMDGPU] Replace non-kernel function uses of LDS globals by pointers.

The main motivation behind pointer replacement of LDS use within non-kernel
functions is - to *avoid* subsequent LDS lowering pass from directly packing
LDS (assume large LDS) into a struct type which would otherwise cause allocating
huge memory for struct instance within every kernel.

Reviewed By: rampitec

Differential Revision: https://reviews.llvm.org/D103225

3 years ago[AMDGPU][Libomptarget] Remove redundant functions
Pushpinder Singh [Fri, 18 Jun 2021 09:34:08 +0000 (09:34 +0000)]
[AMDGPU][Libomptarget] Remove redundant functions

There does not seem to be any use of these functions. They just
put the value to a local which is never used again.

Reviewed By: JonChesterfield

Differential Revision: https://reviews.llvm.org/D104512

3 years ago[Test] Add some tests showing room for optimization exploiting undef and UB
Max Kazantsev [Mon, 21 Jun 2021 06:11:15 +0000 (13:11 +0700)]
[Test] Add some tests showing room for optimization exploiting undef and UB

3 years ago[clangd] Type hints for C++14 return type deduction
Nathan Ridge [Mon, 7 Jun 2021 06:53:37 +0000 (02:53 -0400)]
[clangd] Type hints for C++14 return type deduction

Differential Revision: https://reviews.llvm.org/D103789

3 years ago[yaml2obj] Add support for writing the long symbol name.
Esme-Yi [Mon, 21 Jun 2021 05:09:56 +0000 (05:09 +0000)]
[yaml2obj] Add support for writing the long symbol name.

Summary: This patch, as a follow-up of D95505, adds
support for writing the long symbol name by implementing
the StringTable. Only XCOFF32 is suppoted now.

Reviewed By: jhenderson, shchenz

Differential Revision: https://reviews.llvm.org/D103455

3 years ago[LoopDeletion] Handle Phis with similar inputs from different blocks
Max Kazantsev [Mon, 21 Jun 2021 04:37:06 +0000 (11:37 +0700)]
[LoopDeletion] Handle Phis with similar inputs from different blocks

This patch lifts the requirement to have the only incoming live block
for Phis. There can be multiple live blocks if the same value comes to
phi from all of them.

Differential Revision: https://reviews.llvm.org/D103959
Reviewed By: nikic, lebedev.ri

3 years ago[clang] NFC: adjust indentation of statements with more than one lines
Zhouyi Zhou [Mon, 21 Jun 2021 00:54:27 +0000 (08:54 +0800)]
[clang] NFC: adjust indentation of statements with more than one lines

Hi,

I think it will be more beautiful to adjust indentation of statements with more than one lines.

In function TreeTransform<Derived>::TransformDependentScopeDeclRefExpr
the second line of statement
NestedNameSpecifierLoc QualifierLoc \newline = getDerived().TransformNestedNameSpecifierLoc(E->getQualifierLoc());
is no more indent than the first line

There is a similar case in function TreeTransform<Derived>::TransformUnresolvedMemberExpr

Also I use clang-format to fix above functions

Thanks alot

Reviewed By: pengfei

Differential Revision: https://reviews.llvm.org/D104145

3 years ago[lld/mac] Make a variable more local; no behavior change
Nico Weber [Mon, 21 Jun 2021 01:59:11 +0000 (21:59 -0400)]
[lld/mac] Make a variable more local; no behavior change

The variable used to need the wider scope, but doesn't after the
reland. See LC_LINKER_OPTIONS-related discussion on
https://reviews.llvm.org/D104353 for background.

3 years ago[InstCombine] Use poison constant to represent the result of unreachable instrs
Juneyoung Lee [Sun, 20 Jun 2021 06:00:15 +0000 (15:00 +0900)]
[InstCombine] Use poison constant to represent the result of unreachable instrs

This patch updates InstCombine to use poison constant to represent the resulting value of (either semantically or syntactically) unreachable instrs, or a don't-care value of an unreachable store instruction.

This allows more aggressive folding of unused results, as shown in llvm/test/Transforms/InstCombine/getelementptr.ll .

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D104602

3 years ago[lld/mac] Test zerofill sections after __thread_bss
Nico Weber [Sun, 20 Jun 2021 23:39:09 +0000 (19:39 -0400)]
[lld/mac] Test zerofill sections after __thread_bss

Real zerofill sections go after __thread_bss, since zerofill sections
must all be at the end of their segment and __thread_bss must be right
after __thread_data.

Works fine already, but wasn't tested as far as I can tell.

Also tweak comment about zerofill sections a bit.

No behavior change.

Differential Revision: https://reviews.llvm.org/D104609

3 years ago[NFC][ScalarEvolution] Clean up ExitLimit constructors.
Eli Friedman [Mon, 21 Jun 2021 00:16:28 +0000 (17:16 -0700)]
[NFC][ScalarEvolution] Clean up ExitLimit constructors.

Make all the constructors forward to one constructor.  Remove redundant
assertions.

3 years ago[IVDescriptors] Fix comment that getUnsafeAlgebraInst has been renamed to getExactFPM...
Jim Lin [Sun, 20 Jun 2021 23:56:22 +0000 (07:56 +0800)]
[IVDescriptors] Fix comment that getUnsafeAlgebraInst has been renamed to getExactFPMathInst

https://reviews.llvm.org/rG36a489d194750dc888f214240e9dec9122ca1f0e renamed the function call
in the test from getUnsafeAlgebraInst to getExactFPMathInst.

Reviewed By: spatel

Differential Revision: https://reviews.llvm.org/D104441

3 years ago[lld-macho] Have inputOrder default to less than INT_MAX
Jez Ng [Sun, 20 Jun 2021 23:49:13 +0000 (19:49 -0400)]
[lld-macho] Have inputOrder default to less than INT_MAX

We make it less than INT_MAX in order not to conflict with the ordering
of zerofill sections, which must always be placed at the end of their
segment.

This is the more structural fix for the issue addressed in {D104596}.

Reviewed By: #lld-macho, thakis

Differential Revision: https://reviews.llvm.org/D104607

3 years ago[ELF] Delete --no-cref which does not exist in GNU ld
Fangrui Song [Sun, 20 Jun 2021 21:28:56 +0000 (14:28 -0700)]
[ELF] Delete --no-cref which does not exist in GNU ld

Also delete the single dash form which does not appear to be used.

3 years ago[ELF][test] Add missing tests for --no-export-dynamic & --no-warn-backrefs
Fangrui Song [Sun, 20 Jun 2021 21:20:14 +0000 (14:20 -0700)]
[ELF][test] Add missing tests for --no-export-dynamic & --no-warn-backrefs

3 years ago[GCOVProfiling][test] Ensure that 'opt' drops any files in a temp directory
Dmitri Gribenko [Sun, 20 Jun 2021 20:48:35 +0000 (22:48 +0200)]
[GCOVProfiling][test] Ensure that 'opt' drops any files in a temp directory

3 years agoTry to unbreak the windows CI
Jason Molenda [Sun, 20 Jun 2021 20:12:38 +0000 (13:12 -0700)]
Try to unbreak the windows CI

MSVC and clang seem to disagree with whether I can do this.

3 years ago[TypePromotion] Prune Intrinsic includes. NFC
Craig Topper [Sun, 20 Jun 2021 02:52:52 +0000 (19:52 -0700)]
[TypePromotion] Prune Intrinsic includes. NFC

TypePromotion is meant to be a generic pass and doesn't reference
any ARM intrinsics so it shouldn't include IntrinsicsARM.h.
The other Intrinsic related headers appear to be unneeded as well.

3 years agoAdd a corefile style option to process save-core; skinny corefiles
Jason Molenda [Sun, 20 Jun 2021 19:19:50 +0000 (12:19 -0700)]
Add a corefile style option to process save-core; skinny corefiles

Add a new feature to process save-core on Darwin systems -- for
lldb to create a user process corefile with only the dirty (modified
memory) pages included.  All of the binaries that were used in the
corefile are assumed to still exist on the system for the duration
of the use of the corefile.  A new --style option to process save-core
is added, so a full corefile can be requested if portability across
systems, or across time, is needed for this corefile.

debugserver can now identify the dirty pages in a memory region
when queried with qMemoryRegionInfo, and the size of vm pages is
given in qHostInfo.

Create a new "all image infos" LC_NOTE for Mach-O which allows us
to describe all of the binaries that were loaded in the process --
load address, UUID, file path, segment load addresses, and optionally
whether code from the binary was executing on any thread.  The old
"read dyld_all_image_infos and then the in-memory Mach-O load
commands to get segment load addresses" no longer works when we
only have dirty memory.

rdar://69670807
Differential Revision: https://reviews.llvm.org/D88387

3 years ago[LoopUnroll] Use smallest exact trip count from any exit
Nikita Popov [Sat, 19 Jun 2021 07:44:28 +0000 (09:44 +0200)]
[LoopUnroll] Use smallest exact trip count from any exit

This is a more general alternative/extension to D102635. Rather than
handling the special case of "header exit with non-exiting latch",
this unrolls against the smallest exact trip count from any exit.
The latch exit is no longer treated as priviledged when it comes to
full unrolling.

The motivating case is in full-unroll-one-unpredictable-exit.ll.
Here the header exit is an IV-based exit, while the latch exit is
a data comparison. This kind of loop does not get rotated, because
the latch is already exiting, and loop rotation doesn't try to
distinguish IV-based/analyzable latches.

Differential Revision: https://reviews.llvm.org/D102982

3 years ago[mlir] Fix -Wunused-but-set-variable in -DLLVM_ENABLE_ASSERTIONS=off build. NFC
Fangrui Song [Sun, 20 Jun 2021 18:55:00 +0000 (11:55 -0700)]
[mlir] Fix -Wunused-but-set-variable in -DLLVM_ENABLE_ASSERTIONS=off build. NFC

3 years ago[lld-link] Fix -Wunused-but-set-variable in -DLLVM_ENABLE_ASSERTIONS=off build. NFC
Fangrui Song [Sun, 20 Jun 2021 18:35:01 +0000 (11:35 -0700)]
[lld-link] Fix -Wunused-but-set-variable in -DLLVM_ENABLE_ASSERTIONS=off build. NFC

3 years agoFix -Wunused-variable and -Wunused-but-set-variable in -DLLVM_ENABLE_ASSERTIONS=off...
Fangrui Song [Sun, 20 Jun 2021 18:09:07 +0000 (11:09 -0700)]
Fix -Wunused-variable and -Wunused-but-set-variable in -DLLVM_ENABLE_ASSERTIONS=off build. NFC

3 years ago[lldb] [Process/elf-core] Fix reading NetBSD/i386 core dumps
Michał Górny [Thu, 22 Apr 2021 17:42:52 +0000 (19:42 +0200)]
[lldb] [Process/elf-core] Fix reading NetBSD/i386 core dumps

Add support for extracting basic data from NetBSD/i386 core dumps.
FPU registers are not supported at the moment.

Differential Revision: https://reviews.llvm.org/D101091

3 years ago[DSE] Remove stores in the same loop iteration
David Green [Sun, 20 Jun 2021 16:03:30 +0000 (17:03 +0100)]
[DSE] Remove stores in the same loop iteration

DSE will currently only remove stores in the same block unless they can
be guaranteed to be loop invariant. This expands that to any stores that
are in the same Loop, at the same loop level. This should still account
for where AA/MSSA will not handle aliasing between loops, but allow the
dead stores to be removed where they overlap in the same loop iteration.
It requires adding loop info to DSE, but that looks fairly harmless.

The test case this helps is from code like this, which can come up in
certain matrix operations:
  for(i=..)
    dst[i] = 0;
    for(j=..)
      dst[i] += src[i*n+j];

After LICM, this becomes:
for(i=..)
  dst[i] = 0;
  sum = 0;
  for(j=..)
    sum += src[i*n+j];
  dst[i] = sum;

The first store is dead, and with this patch is now removed.

Differntial Revision: https://reviews.llvm.org/D100464

3 years ago[InstCombine] fold ctpop-of-select with 1 or more constant arms
Sanjay Patel [Sun, 20 Jun 2021 15:26:11 +0000 (11:26 -0400)]
[InstCombine] fold ctpop-of-select with 1 or more constant arms

The general pattern is mentioned in:
https://llvm.org/PR50140
...but we need to do a bit more to handle intrinsics with extra operands
like ctlz/cttz.

3 years ago[libcxx] Implement P0883R2 ("Fixing Atomic Initialization")
Raul Tambre [Sun, 6 Jun 2021 16:02:20 +0000 (19:02 +0300)]
[libcxx] Implement P0883R2 ("Fixing Atomic Initialization")

Reviewed By: Quuxplusone

Differential Revision: https://reviews.llvm.org/D103769

3 years ago[flang] Implement constant folding for the NOT intrinsic
Peter Steinfeld [Sat, 19 Jun 2021 02:18:27 +0000 (19:18 -0700)]
[flang] Implement constant folding for the NOT intrinsic

I implemented constant folding for the NOT intrinsic and added some tests.

Differential Revision: https://reviews.llvm.org/D104587

3 years ago[InstCombine] avoid infinite loops with select folds of constant expressions
Sanjay Patel [Sun, 20 Jun 2021 13:41:59 +0000 (09:41 -0400)]
[InstCombine] avoid infinite loops with select folds of constant expressions

This pair of transforms was added recently with:
8591640379ac9175a

And could lead to conflicting folds:
https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=35399

3 years ago[NFC][AArch64][ARM][Thumb][Hexagon] Autogenerate some tests
Roman Lebedev [Sun, 20 Jun 2021 10:24:31 +0000 (13:24 +0300)]
[NFC][AArch64][ARM][Thumb][Hexagon] Autogenerate some tests

These all (and some others) are being affected by D104597,
but they are manually-written, which rather complicates
checking the effect that change has on them.

3 years ago[UpdateTestUtils] Print test filename when complaining about conflicting prefix
Roman Lebedev [Sun, 20 Jun 2021 10:14:33 +0000 (13:14 +0300)]
[UpdateTestUtils] Print test filename when complaining about conflicting prefix

Now that FileCheck eagerly complains when prefixes are unused,
the update script does the same, and  is becoming very common
to need to drop some prefixes, yet figuring out the file
it complains about isn't obvious unless it actually tells us.

3 years ago[SimplifyCFG] FoldTwoEntryPHINode(): don't fold if either block has it's address...
Roman Lebedev [Sun, 20 Jun 2021 09:33:57 +0000 (12:33 +0300)]
[SimplifyCFG] FoldTwoEntryPHINode(): don't fold if either block has it's address taken

Same as with HoistThenElseCodeToIf() (ad87761925c2790aab272138b5bbbde4a93e0383).

3 years ago[SimplifyCFG] HoistThenElseCodeToIf(): don't hoist if either block has it's address...
Roman Lebedev [Sun, 20 Jun 2021 09:18:15 +0000 (12:18 +0300)]
[SimplifyCFG] HoistThenElseCodeToIf(): don't hoist if either block has it's address taken

This problem is exposed by D104598, after it tail-merges `ret` in
`@test_inline_constraint_S_label`, the verifier would start complaining
`invalid operand for inline asm constraint 'S'`.

Essentially, taking address of a block is mismodelled in IR.
It should probably be an explicit instruction, a first one in block,
that isn't identical to any other instruction of the same type,
so that it can't be hoisted.

3 years ago[InstSimplify] icmp poison, X -> poison
Juneyoung Lee [Sun, 20 Jun 2021 06:29:05 +0000 (15:29 +0900)]
[InstSimplify] icmp poison, X -> poison

This adds a simple transformation from icmp with poison constant to poison.
Comparing poison with something else is poison, so this is okay.

https://alive2.llvm.org/ce/z/e8iReb
https://alive2.llvm.org/ce/z/q4MurY

3 years ago[llvm-cov gcov] Support GCC 12 format
Fangrui Song [Sun, 20 Jun 2021 05:51:20 +0000 (22:51 -0700)]
[llvm-cov gcov] Support GCC 12 format

GCC 12 will change the length field to represent the number of bytes instead of
32-bit words. This avoids padding for strings.

3 years ago[llvm-cov gcov] Change case to match the prevailing style && replace getString with...
Fangrui Song [Sun, 20 Jun 2021 05:02:14 +0000 (22:02 -0700)]
[llvm-cov gcov] Change case to match the prevailing style && replace getString with readString

3 years ago[Flang][test] Fix Windows buildbot.
Michael Kruse [Sun, 20 Jun 2021 03:23:02 +0000 (22:23 -0500)]
[Flang][test] Fix Windows buildbot.

Add

    REQUIRES: shell

to tests that execute a UNIX shell script to not run on Windows.

3 years ago[test] Fix nocompress.test
Fangrui Song [Sat, 19 Jun 2021 23:27:52 +0000 (16:27 -0700)]
[test] Fix nocompress.test

3 years ago[profile] Fix variable name
Petr Hosek [Sat, 19 Jun 2021 21:55:32 +0000 (14:55 -0700)]
[profile] Fix variable name

This fixes a bug introduced in d85c258fd1e7459cc8085b5f364e356f50b490a4.

3 years ago[llvm-profdata] Make diagnostics consistent with the (no capitalization, no period...
Fangrui Song [Sat, 19 Jun 2021 21:54:24 +0000 (14:54 -0700)]
[llvm-profdata] Make diagnostics consistent with the (no capitalization, no period) style

The format is currently inconsistent. Use the https://llvm.org/docs/CodingStandards.html#error-and-warning-messages style.

And add `error:` or `warning:` to CHECK lines wherever appropriate.

3 years ago[profile] Don't publish VMO if there are no counters
Petr Hosek [Wed, 19 May 2021 17:14:00 +0000 (10:14 -0700)]
[profile] Don't publish VMO if there are no counters

If there are no counters, there's no need to publish the VMO.

Differential Revision: https://reviews.llvm.org/D102786

3 years ago[LLD] [COFF] Avoid doing repeated fuzzy symbol lookup for each iteration. NFC.
Martin Storsjö [Thu, 17 Jun 2021 13:28:47 +0000 (16:28 +0300)]
[LLD] [COFF] Avoid doing repeated fuzzy symbol lookup for each iteration. NFC.

This is run every time around in the main linker loop. Once a match
has been found, stop trying to rematch such a symbol.

Not sure if this has any actual measurable performance impact though
(SymbolTable::findMangle() iterates over the whole symbol table for
each call and does fuzzy matching on top of that) but this makes the
code more reassuring to read at least. (This is in practice run for def
files listing undecorated stdcall functions to be exported.)

Differential Revision: https://reviews.llvm.org/D104529

3 years ago[LLD] [MinGW] Print errors/warnings in lld-link with a "ld.lld" prefix
Martin Storsjö [Fri, 18 Jun 2021 11:29:55 +0000 (14:29 +0300)]
[LLD] [MinGW] Print errors/warnings in lld-link with a "ld.lld" prefix

Pass the original argv[0] to the coff linker, as the coff linker uses
the basename of argv[0] as the log prefix.

This makes error messages to be printed with a "ld.lld:" prefix
instead of "lld-link:". The current "lld-link:" prefix can be confusing
to users, as they're invoking the MinGW linker (and might not even have
a lld-link executable).

Keep the first argument as lld-link when printing the command line, to
make it an actually reproducible standalone command.

Differential Revision: https://reviews.llvm.org/D104526

3 years ago[llvm-profdata] Delete unneeded empty output filename check
Fangrui Song [Sat, 19 Jun 2021 19:20:45 +0000 (12:20 -0700)]
[llvm-profdata] Delete unneeded empty output filename check

3 years ago[RISCV] Prevent formation of shXadd(.uw) and add.uw if it prevents the use of addi.
Craig Topper [Sat, 19 Jun 2021 18:54:59 +0000 (11:54 -0700)]
[RISCV] Prevent formation of shXadd(.uw) and add.uw if it prevents the use of addi.

If the outer add has an simm12 immediate operand we should prefer
it instead of materializing it in a register. This would guarantee
and extra instruction and temporary register. Since we don't check
one use on the shl or zext we might generate more instructions if
there is an additional user.

3 years ago[NFC] AMD Zen 3: fix typo in a comment
Roman Lebedev [Sat, 19 Jun 2021 19:04:42 +0000 (22:04 +0300)]
[NFC] AMD Zen 3: fix typo in a comment

3 years agoSimplify some typedef struct
Fangrui Song [Sat, 19 Jun 2021 18:36:44 +0000 (11:36 -0700)]
Simplify some typedef struct

3 years ago[gn build] (manually) port b9c05aff205b (MIRTests)
Nico Weber [Sat, 19 Jun 2021 17:03:51 +0000 (13:03 -0400)]
[gn build] (manually) port b9c05aff205b (MIRTests)

3 years ago[lld/mac] Make sure __thread_ptrs is in front of __thread_bss
Nico Weber [Sat, 19 Jun 2021 14:55:48 +0000 (10:55 -0400)]
[lld/mac] Make sure __thread_ptrs is in front of __thread_bss

The exact location doesn't matter, but it should be in front
of __thread_bss. We put it right in front of __thread_data
which is where ld64 seems to put it as well.

Fixes PR50769.

(As mentioned on the bug, there is probably a more structural
fix too, see comment 5. If we don't address this, it's likely
we'll run into this again with other synthetic sections. But
for now, let's fix the immediate breakage.)

Differential Revision: https://reviews.llvm.org/D104596

3 years ago[lld/mac] Give __DATA,__thread_ptrs type S_THREAD_LOCAL_VARIABLE_POINTERS
Nico Weber [Sat, 19 Jun 2021 13:54:11 +0000 (09:54 -0400)]
[lld/mac] Give __DATA,__thread_ptrs type S_THREAD_LOCAL_VARIABLE_POINTERS

...instead of S_NON_LAZY_SYMBOL_POINTERS. This matches ld64.

Part of PR50769.

While here, also remove an old TODO that was done in D87178.

Differential Revision: https://reviews.llvm.org/D104594

3 years ago[MIRPrinter] Add machine metadata support.
Michael Liao [Wed, 26 May 2021 00:20:52 +0000 (20:20 -0400)]
[MIRPrinter] Add machine metadata support.

- Distinct metadata needs generating in the codegen to attach correct
  AAInfo on the loads/stores after lowering, merging, and other relevant
  transformations.
- This patch adds 'MachhineModuleSlotTracker' to help assign slot
  numbers to these newly generated unnamed metadata nodes.
- To help 'MachhineModuleSlotTracker' track machine metadata, the
  original 'SlotTracker' is rebased from 'AbstractSlotTrackerStorage',
  which provides basic interfaces to create/retrive metadata slots. In
  addition, once LLVM IR is processsed, additional hooks are also
  introduced to help collect machine metadata and assign them slot
  numbers.
- Finally, if there is any such machine metadata, 'MIRPrinter' outputs
  an additional 'machineMetadataNodes' field containing all the
  definition of those nodes.

Reviewed By: arsenm

Differential Revision: https://reviews.llvm.org/D103205

3 years ago[amdgpu] Improve the from f32 to i64.
Michael Liao [Fri, 11 Jun 2021 21:21:09 +0000 (17:21 -0400)]
[amdgpu] Improve the from f32 to i64.

- Take the same principle as the conversion from f64 to i64 with extra
  necessary pre- and post-processing. It helps to reduce that conversion
  sequence by half compared to legacy one.

Reviewed By: foad

Differential Revision: https://reviews.llvm.org/D104427

3 years ago[InstCombine][test] add tests for select-of-bit-manip; NFC
Sanjay Patel [Sat, 19 Jun 2021 16:20:45 +0000 (12:20 -0400)]
[InstCombine][test] add tests for select-of-bit-manip; NFC

3 years agoRevert "Re-Revert "DirectoryWatcher: add an implementation for Windows""
Saleem Abdulrasool [Fri, 18 Jun 2021 23:00:01 +0000 (16:00 -0700)]
Revert "Re-Revert "DirectoryWatcher: add an implementation for Windows""

This reverts commit fb32de9e97af0921242a021e30020ffacf7aa6e2.

Remove the secondary synchronization point as noted by Adrian.  This is
technically only to make the builders happier about tests and should not
be needed.  This also pushes the condition variable setting to after the
watch is actually established (which was the source of the original race
condition, but would normally succeed as the thread shouldn't get put to
sleep immediately on the trigger of the condition variable).

This also was pretested on the chromium builders:
https://ci.chromium.org/ui/p/chromium/builders/try/win_upload_clang/1612/overview.

3 years agoAllow building for release with EXPENSIVE_CHECKS
Tomas Matheson [Fri, 18 Jun 2021 09:57:59 +0000 (10:57 +0100)]
Allow building for release with EXPENSIVE_CHECKS

D97225 moved LazyCallGraph verify() calls behind EXPENSIVE_CHECKS,
but verity() is defined for debug builds only so this had the unintended
effect of breaking release builds with EXPENSIVE_CHECKS.

Fix by enabling verify() for both debug and EXPENSIVE_CHECKS.

Differential Revision: https://reviews.llvm.org/D104514

3 years ago[ARM][NFC] Tidy up subtarget frame pointer routines
Tomas Matheson [Thu, 17 Jun 2021 18:11:25 +0000 (19:11 +0100)]
[ARM][NFC] Tidy up subtarget frame pointer routines

getFramePointerReg only depends on information in ARMSubtarget,
so move it in there so it can be accessed from more places.

Make use of ARMSubtarget::getFramePointerReg to remove duplicated code.

The main use of useR7AsFramePointer is getFramePointerReg, so inline it.

Differential Revision: https://reviews.llvm.org/D104476

3 years agoRevert "[clang][FPEnv] Clang floatng point model ffp-model=precise enables ffp-contra...
Melanie Blower [Sat, 19 Jun 2021 11:59:21 +0000 (07:59 -0400)]
Revert "[clang][FPEnv] Clang floatng point model ffp-model=precise enables ffp-contract=on"

This reverts commit a1449a10dbcfcf353f4f761281c4e572b4ce9308.
Seems like my changes to LNT had no effect -- puzzled.
The 21 tests pass on my sandbox with the clang patch but are
failing in exec time in the bot

3 years ago[gn build] Port 134723edd5bf
LLVM GN Syncbot [Sat, 19 Jun 2021 11:49:56 +0000 (11:49 +0000)]
[gn build] Port 134723edd5bf

3 years ago[libcxx] Move all algorithms into their own headers
Louis Dionne [Thu, 17 Jun 2021 15:30:11 +0000 (11:30 -0400)]
[libcxx] Move all algorithms into their own headers

This is a fairly mechanical change, it just moves each algorithm into
its own header. This is intended to be a NFC.

This commit re-applies 7ed7d4ccb899, which was reverted in 692d7166f771
because the Modules build got broken. The modules build has now been
fixed, so we're re-committing this.

Differential Revision: https://reviews.llvm.org/D103583

Attribution note
----------------
I'm only committing this. This commit is a mix of D103583, D103330 and
D104171 authored by:

Co-authored-by: Christopher Di Bella <cjdb@google.com>
Co-authored-by: zoecarver <z.zoelec2@gmail.com>
3 years ago[clang-cl] Don't expand /permissive- to /ZC:strictStrings yet
Markus Böck [Sat, 19 Jun 2021 11:28:12 +0000 (13:28 +0200)]
[clang-cl] Don't expand /permissive- to /ZC:strictStrings yet

Follow up on rGc70b0e808da8

/Zc:strictStrings is an alias to an option part of the -W group. When the driver tries to render the option back to a string for the cc1 invocation, it sadly gets rendered with the original spelling instead of the alias, causing issues reported here: https://reviews.llvm.org/D103773#inline-989447

I am thinking it's the best to revert this part of the patch until I figured out how to correctly add the arg and until /Zc:strictStrings- exists/is needed.

3 years ago[clang][FPEnv] Clang floatng point model ffp-model=precise enables ffp-contract=on
Melanie Blower [Sat, 19 Jun 2021 10:49:27 +0000 (06:49 -0400)]
[clang][FPEnv] Clang floatng point model ffp-model=precise enables ffp-contract=on

This patch changes the ffp-model=precise to enables -ffp-contract=on
(previously -ffp-model=precise enabled -ffp-contract=fast). This is a
follow-up to Andy Kaylor's comments in the llvm-dev discussion
"Floating Point semantic modes". From the same email thread, I put
Andy's distillation of floating point options and floating point modes
into UsersManual.rst

Differential Revision: https://reviews.llvm.org/D74436

3 years ago[mlir] Add EmitC dialect
Marius Brehler [Wed, 9 Jun 2021 13:38:10 +0000 (13:38 +0000)]
[mlir] Add EmitC dialect

This upstreams the EmitC dialect and the corresponding Cpp target, both
initially presented with [1], from [2] to MLIR core. For the related
discussion, see [3].

[1] https://reviews.llvm.org/D76571
[2] https://github.com/iml130/mlir-emitc
[3] https://llvm.discourse.group/t/emitc-generating-c-c-from-mlir/3388

Co-authored-by: Jacques Pienaar <jpienaar@google.com>
Co-authored-by: Simon Camphausen <simon.camphausen@iml.fraunhofer.de>
Co-authored-by: Oliver Scherf <oliver.scherf@iml.fraunhofer.de>
Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D103969

3 years ago[LoopUnroll] Push runtime unrolling decision up into tryToUnrollLoop()
Nikita Popov [Thu, 17 Jun 2021 19:49:29 +0000 (21:49 +0200)]
[LoopUnroll] Push runtime unrolling decision up into tryToUnrollLoop()

Currently, UnrollLoop() is passed an AllowRuntime flag and decides
itself whether runtime unrolling should be used or not. This patch
pushes the decision into the caller and allows us to eliminate the
ULO.TripCount and ULO.TripMultiple parameters.

Differential Revision: https://reviews.llvm.org/D104487

3 years ago[RISCV] Optimize add-mul in the zba extension with SH*ADD
Ben Shi [Sat, 19 Jun 2021 03:01:43 +0000 (11:01 +0800)]
[RISCV] Optimize add-mul in the zba extension with SH*ADD

This patch does the following optimization.

Rx + Ry * 18 => (SH1ADD (SH3ADD Rx, Rx), Ry)
Rx + Ry * 20 => (SH2ADD (SH2ADD Rx, Rx), Ry)
Rx + Ry * 24 => (SH3ADD (SH1ADD Rx, Rx), Ry)
Rx + Ry * 36 => (SH2ADD (SH3ADD Rx, Rx), Ry)
Rx + Ry * 40 => (SH3ADD (SH2ADD Rx, Rx), Ry)
Rx + Ry * 72 => (SH3ADD (SH3ADD Rx, Rx), Ry)

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D104588

3 years ago[RISCV][test] Add new tests for add-mul optimization in the zba extension with SH*ADD
Ben Shi [Sat, 19 Jun 2021 02:09:08 +0000 (10:09 +0800)]
[RISCV][test] Add new tests for add-mul optimization in the zba extension with SH*ADD

These tests will show the following optimization by future patches.

Rx + Ry * 18  => (SH1ADD (SH3ADD Rx, Rx), Ry)
Rx + Ry * 20  => (SH2ADD (SH2ADD Rx, Rx), Ry)
Rx + Ry * 24  => (SH3ADD (SH1ADD Rx, Rx), Ry)
Rx + Ry * 36  => (SH2ADD (SH3ADD Rx, Rx), Ry)
Rx + Ry * 40  => (SH3ADD (SH2ADD Rx, Rx), Ry)
Rx + Ry * 72  => (SH3ADD (SH3ADD Rx, Rx), Ry)
Rx * (3 << C) => (SLLI (SH1ADD Rx, Rx), C)
Rx * (5 << C) => (SLLI (SH2ADD Rx, Rx), C)
Rx * (9 << C) => (SLLI (SH3ADD Rx, Rx), C)

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D104507

3 years ago[ORC][examples] Add missing library dependence
Lang Hames [Sat, 19 Jun 2021 04:44:38 +0000 (14:44 +1000)]
[ORC][examples] Add missing library dependence

3 years ago[mlir][linalg] Lower subtensor(pad_tensor) to pad_tensor(subtensor)
Matthias Springer [Sat, 19 Jun 2021 04:25:47 +0000 (13:25 +0900)]
[mlir][linalg] Lower subtensor(pad_tensor) to pad_tensor(subtensor)

Only high padding is supported at the moment. Low padding will be added in a separate commit.

Differential Revision: https://reviews.llvm.org/D104357

3 years ago[re-land][lld-macho] Avoid force-loading the same archive twice
Jez Ng [Sat, 19 Jun 2021 02:30:57 +0000 (22:30 -0400)]
[re-land][lld-macho] Avoid force-loading the same archive twice

This reverts commit c9b241efd68c5a0f1f67e9250960ade454f3bc11, which was
a backout diff to fix the buildbots.

The real culprit of the crash is
https://github.com/llvm/llvm-project/commit/1d31fb8d122b1117cf20a9edc09812db8472e930,
which is being reverted.

Differential Revision: https://reviews.llvm.org/D104353

3 years agoRevert "[lld-macho] Have path-related functions return std::string, not StringRef"
Jez Ng [Sat, 19 Jun 2021 02:19:09 +0000 (22:19 -0400)]
Revert "[lld-macho] Have path-related functions return std::string, not StringRef"

This reverts commit 1d31fb8d122b1117cf20a9edc09812db8472e930.

Making `rerootPath` return a temporary std::string caused a
use-after-free:

https://ci.chromium.org/ui/p/chromium/builders/try/win_upload_clang/1608/overview

3 years ago[llvm][Inliner] Add an optional PriorityInlineOrder
Liqiang Tao [Sat, 19 Jun 2021 02:17:19 +0000 (10:17 +0800)]
[llvm][Inliner] Add an optional PriorityInlineOrder

This patch adds an optional PriorityInlineOrder, which uses the heap to order inlining.
The callsite which size is smaller would have a higher priority.

Reviewed By: mtrofin

Differential Revision: https://reviews.llvm.org/D104028

3 years ago[ORC][C-bindings] Add access to LLJIT IRTransformLayer, ThreadSafeModule utils.
Lang Hames [Sat, 19 Jun 2021 01:41:42 +0000 (11:41 +1000)]
[ORC][C-bindings] Add access to LLJIT IRTransformLayer, ThreadSafeModule utils.

This patch was derived from Valentin Churavy's work in
https://reviews.llvm.org/D104480. It adds support for setting the transform on
an IRTransformLayer, and for accessing the IRTransformLayer in LLJIT. It also
adds access to the ThreadSafeModule::withModuleDo method for thread-safe
access to modules.

A new example has been added to show how to use these APIs to optimize a module
during materialization.

Thanks Valentin!

Reviewed By: lhames

Differential Revision: https://reviews.llvm.org/D103855

3 years ago[ORC][examples] Fix file name in comment.
Lang Hames [Fri, 18 Jun 2021 23:22:45 +0000 (09:22 +1000)]
[ORC][examples] Fix file name in comment.