platform/upstream/llvm.git
3 years ago[profile] Add -fprofile-instr-generate tests for weak definition and various linkages
Fangrui Song [Fri, 4 Jun 2021 17:26:55 +0000 (10:26 -0700)]
[profile] Add -fprofile-instr-generate tests for weak definition and various linkages

3 years ago[scudo] Always exclude Tag 0
Vitaly Buka [Fri, 28 May 2021 09:08:44 +0000 (02:08 -0700)]
[scudo] Always exclude Tag 0

prepareTaggedChunk uses Tag 0 for header.

Android already PR_MTE_TAG_MASK to 0xfffe,
but with the patch we will not need to deppend
on the system configuration.

Reviewed By: pcc

Differential Revision: https://reviews.llvm.org/D103134

3 years ago[OpenCL] Add DWARF address spaces mapping for SPIR
Jason Zheng [Tue, 1 Jun 2021 09:22:00 +0000 (10:22 +0100)]
[OpenCL] Add DWARF address spaces mapping for SPIR

Extend debug info handling by adding DWARF address space mapping for
SPIR, with corresponding test case.

Reviewed By: Anastasia

Differential Revision: https://reviews.llvm.org/D103097

3 years agoCUDA/HIP: Change device-use-host-var.cu's NOT "external" check to include variable...
Konstantin Zhuravlyov [Fri, 4 Jun 2021 12:55:26 +0000 (08:55 -0400)]
CUDA/HIP: Change device-use-host-var.cu's NOT "external" check to include variable name

Otherwise it is causing one of our build jobs to fail,
it is using "external" as directory, and NOT is
failing because "external" is found in ModuleID.

Differential Revision: https://reviews.llvm.org/D103658

3 years ago[AArch64][GlobalISel] Handle multiple phis in fixupPHIOpBanks
Jessica Paquette [Thu, 3 Jun 2021 00:45:58 +0000 (17:45 -0700)]
[AArch64][GlobalISel] Handle multiple phis in fixupPHIOpBanks

If we ended up with two phi instructions in a block, and we needed to fix up
the banks for the first one, we'd end up inserting our COPY before the second
phi.

E.g.

```
%x = G_PHI ...
%fixup = COPY ...
%y = G_PHI ...
```

This is invalid MIR, and breaks assumptions made by the register allocator later
down the line. With the verifier enabled, it also emits a verification error.

This teaches fixupPHIOpBanks to walk past any phi instructions in the block
when emitting the fixup copies.

Here's an example of the crashing code (same as added testcase):
https://godbolt.org/z/h5j1x3o6e

Differential Revision: https://reviews.llvm.org/D103582

3 years ago[libcxx][ranges] Add concepts in range.utility.helpers.
zoecarver [Tue, 1 Jun 2021 21:54:29 +0000 (14:54 -0700)]
[libcxx][ranges] Add concepts in range.utility.helpers.

There are no changes to public APIs.

Differential Revision: https://reviews.llvm.org/D103493

3 years agoAdd commutable attribute to opcodes for ARC
Mark Schimmel [Fri, 4 Jun 2021 16:47:50 +0000 (19:47 +0300)]
Add commutable attribute to opcodes for ARC

This patch sets the isCommutable attribute for several opcodes that have
the "reg = OPCODE reg, reg" format.

Differential Revision: https://reviews.llvm.org/D103653

3 years ago[gn build] Port d31a2e7554ea
LLVM GN Syncbot [Fri, 4 Jun 2021 16:41:04 +0000 (16:41 +0000)]
[gn build] Port d31a2e7554ea

3 years ago[gn build] Port 7ed7d4ccb899
LLVM GN Syncbot [Fri, 4 Jun 2021 16:41:03 +0000 (16:41 +0000)]
[gn build] Port 7ed7d4ccb899

3 years ago[libcxx][ranges] Add `ranges::empty_view`.
zoecarver [Wed, 26 May 2021 21:25:02 +0000 (14:25 -0700)]
[libcxx][ranges] Add `ranges::empty_view`.

Differential Revision: https://reviews.llvm.org/D103208

3 years ago[Attributor] Check HeapToStack's state for isKnownHeapToStack
Joseph Huber [Thu, 3 Jun 2021 00:20:17 +0000 (20:20 -0400)]
[Attributor] Check HeapToStack's state for isKnownHeapToStack

This patch changes the `isKnownHeapToStack` and `isAssumedHeapToStack`
member functions to return if a function call is going to be altered by
HeapToStack.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D103574

3 years ago[libcxx][gardening] Move all algorithms into their own headers.
zoecarver [Thu, 3 Jun 2021 18:26:03 +0000 (11:26 -0700)]
[libcxx][gardening] Move all algorithms into their own headers.

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

Note: during this change, I burned down all the includes, so this follows "include only and exactly what you use."

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

3 years ago[Attributor] Allow lookupAAFor to return null on invalid state
Joseph Huber [Wed, 2 Jun 2021 21:32:42 +0000 (17:32 -0400)]
[Attributor] Allow lookupAAFor to return null on invalid state

This patch adds an option to `lookupAAFor` that allows it to return a
nullptr if the state of the looked up attribute is invalid. This is so
future passes can use this to query other attributes with the guarantee
that they are valid.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D103556

3 years ago[OpenMP] fix spelling error in message-converter.pl
Peyton, Jonathan L [Fri, 4 Jun 2021 16:19:27 +0000 (11:19 -0500)]
[OpenMP] fix spelling error in message-converter.pl

3 years ago[RISCV] Teach vsetvli insertion pass that operations on masks don't care about SEW...
Craig Topper [Fri, 4 Jun 2021 15:57:07 +0000 (08:57 -0700)]
[RISCV] Teach vsetvli insertion pass that operations on masks don't care about SEW/LMUL.

All that really matters is that the VLMAX of the preceding
instructions is the same as the VLMAX required by the mask
operation.

Also update the vmsge(u) handling to use the SEW/LMUL we use for
other mask register operations. We were matching it to the compare
before. Some cases will be improve if we fix masked compares to
use tail agnostic policy. I think they ignore the tail policy
anyway.

Reviewed By: frasercrmck

Differential Revision: https://reviews.llvm.org/D103299

3 years ago[clang] use a different name for generated test cdb
Mikhail Goncharov [Fri, 4 Jun 2021 16:10:17 +0000 (18:10 +0200)]
[clang] use a different name for generated test cdb

if build system copied source files as readonly, then override of db_tu.json
will fail

3 years ago[clangd] Run code completion on each token coverd by --check-lines
Adam Czachorowski [Wed, 2 Jun 2021 17:45:11 +0000 (19:45 +0200)]
[clangd] Run code completion on each token coverd by --check-lines

In --check mode we do not run code completion because it is too slow,
especially on larger files. With the introducation of --check-lines we
can narrow down the scope and thus we can afford to do code completion.

We vlog() the top completion result, but that's not really the point.
The most value will come from being able to reproduce crashes that occur
during code completion and require preamble build or index (and thus are
more difficult to reproduce with -code-complete-at).

Differential Revision: https://reviews.llvm.org/D103538

3 years ago[ConstantFolding] add copysign tests for more FP types; NFC
Sanjay Patel [Fri, 4 Jun 2021 15:40:39 +0000 (11:40 -0400)]
[ConstantFolding] add copysign tests for more FP types; NFC

D102673 proposes to ease the current type check, but
there doesn't appear to be any test coverage for that.

3 years ago[OPENMP]Fix PR50129: omp cancel parallel not working as expected.
Alexey Bataev [Thu, 3 Jun 2021 21:13:35 +0000 (14:13 -0700)]
[OPENMP]Fix PR50129: omp cancel parallel not working as expected.

Need to emit a call for __kmpc_cancel_barrier in the exit block for
__kmpc_cancel function call if cancellation of the parallel block is
requested.

Differential Revision: https://reviews.llvm.org/D103646

3 years ago[AArch64] Remove SETCC of CSEL when the latter's condition can be inverted
Bradley Smith [Thu, 27 May 2021 12:59:06 +0000 (13:59 +0100)]
[AArch64] Remove SETCC of CSEL when the latter's condition can be inverted

  setcc (csel 0, 1, cond, X), 1, ne ==> csel 0, 1, !cond, X

Where X is a condition code setting instruction.

Co-authored-by: Paul Walker <paul.walker@arm.com>
Differential Revision: https://reviews.llvm.org/D103256

3 years ago[mlir][linalg] Refactor PadTensorOpVectorizationPattern (NFC)
Matthias Springer [Fri, 4 Jun 2021 14:44:18 +0000 (23:44 +0900)]
[mlir][linalg] Refactor PadTensorOpVectorizationPattern (NFC)

* Rename PadTensorOpVectorizationPattern to GenericPadTensorOpVectorizationPattern.
* Make GenericPadTensorOpVectorizationPattern a private pattern, to be instantiated via populatePadTensorOpVectorizationPatterns.
* Factor out parts of PadTensorOpVectorizationPattern into helper functions.

This commit prepares PadTensorOpVectorizationPattern for a series of subsequent commits that add more specialized PadTensorOp vectorization patterns.

Differential Revision: https://reviews.llvm.org/D103681

3 years agoTemporarily remove another test added in one of the tests added in effb87d
Nico Weber [Fri, 4 Jun 2021 14:42:37 +0000 (10:42 -0400)]
Temporarily remove another test added in one of the tests added in effb87d

This test reads the test file removed in db3e4faa4d2ca.

3 years agoRevert test fixups after e9a9c850989e (which reverted a14fc74).
Nico Weber [Fri, 4 Jun 2021 14:39:49 +0000 (10:39 -0400)]
Revert test fixups after e9a9c850989e (which reverted a14fc74).

This reverts commit da3ed58b97c1cc1356b7732d5dcbb6e4de3057da.
This reverts commit ba1fb0ff8c9f9ef7f9b7d1fe43cb95c8d1363f78.

3 years ago[mlir] Mark VectorToSCF patterns as recursive
Matthias Springer [Fri, 4 Jun 2021 14:30:28 +0000 (23:30 +0900)]
[mlir] Mark VectorToSCF patterns as recursive

Differential Revision: https://reviews.llvm.org/D103599

3 years ago[gn build] (manually) port de07b1e84d8de9
Nico Weber [Fri, 4 Jun 2021 14:37:53 +0000 (10:37 -0400)]
[gn build] (manually) port de07b1e84d8de9

3 years ago[OPENMP]Fix PR49790: Constexpr values not handled in `omp declare mapper` clause.
Alexey Bataev [Thu, 3 Jun 2021 19:13:10 +0000 (12:13 -0700)]
[OPENMP]Fix PR49790: Constexpr values not handled in `omp declare mapper` clause.

Patch allows using of constexpr vars evaluatable to constant calue to be
used in declare mapper construct.

Differential Revision: https://reviews.llvm.org/D103642

3 years ago[mlir][openacc] Conversion of data operands in acc.data to LLVM IR dialect
Valentin Clement [Fri, 4 Jun 2021 14:25:50 +0000 (10:25 -0400)]
[mlir][openacc] Conversion of data operands in acc.data to LLVM IR dialect

Convert data operands from the acc.data operation using the same conversion pattern than D102170.

Reviewed By: ftynse

Differential Revision: https://reviews.llvm.org/D103332

3 years ago[flang][driver] Add checks for missing option arguments
Andrzej Warzynski [Wed, 2 Jun 2021 20:53:02 +0000 (21:53 +0100)]
[flang][driver] Add checks for missing option arguments

With this patch, the following invocation of the frontend driver will
return an error:
```
flang-new -fc1 input-file.f90 -o
```
Similar logic applies to other options that require arguments.

Similar checks are already available in the compiler driver, flang-new
(that's implemented in clangDriver).

Differential Revision: https://reviews.llvm.org/D103554

3 years ago[InstCombine] add tests for pow() reassociation; NFC
Sanjay Patel [Fri, 4 Jun 2021 14:15:22 +0000 (10:15 -0400)]
[InstCombine] add tests for pow() reassociation; NFC

Baseline tests for D102574

3 years agoTemporarily remove one of the tests added in effb87dfa810a
Nico Weber [Fri, 4 Jun 2021 14:02:22 +0000 (10:02 -0400)]
Temporarily remove one of the tests added in effb87dfa810a

It fails on some Linux systems. Remove the test until we've figured
out what's going on. See https://crbug.com/1216005 for details.

3 years ago[clang-format] NFC, 2% improvement in overall clang-formatted status (now 50%)
mydeveloperday [Fri, 4 Jun 2021 13:59:12 +0000 (14:59 +0100)]
[clang-format] NFC, 2% improvement in overall clang-formatted status (now 50%)

3 years agoRevert "[InstrProfiling] If no value profiling, make data variable private and (for...
Nico Weber [Fri, 4 Jun 2021 13:51:24 +0000 (09:51 -0400)]
Revert "[InstrProfiling] If no value profiling, make data variable private and (for Windows) use one comdat"

This reverts commit a14fc749aab2c8e1a45d19d512255ebfc69357c3.
Breaks check-profile on macOS. See https://reviews.llvm.org/D103372 for details.

3 years ago[libc++] Use the using_if_exists attribute when provided
Louis Dionne [Wed, 2 Jun 2021 14:41:37 +0000 (10:41 -0400)]
[libc++] Use the using_if_exists attribute when provided

As discussed on cfe-dev [1], use the using_if_exists Clang attribute when
the compiler supports it. This makes it easier to port libc++ on top of
new platforms that don't fully support the C Standard library.

Previously, libc++ would fail to build when trying to import a missing
declaration in a <cXXXX> header. With the attribute, the declaration will
simply not be imported into namespace std, and hence it won't be available
for libc++ to use. In many cases, the declarations were *not* actually
required for libc++ to work (they were only surfaced for users to use
them as std::XXXX), so not importing them into namespace std is acceptable.

The same thing could be achieved by conscious usage of `#ifdef` along
with platform detection, however that quickly creates a maintenance
problem as libc++ is ported to new platforms. Furthermore, this problem
is exacerbated when mixed with vendor internal-only platforms, which can
lead to difficulties maintaining a downstream fork of the library.

For the time being, we only use the using_if_exists attribute when it
is supported. At some point in the future, we will start removing #ifdef
paths that are unnecessary when the attribute is supported, and folks
who need those #ifdef paths will be required to use a compiler that
supports the attribute.

[1]: http://lists.llvm.org/pipermail/cfe-dev/2020-June/066038.html

Differential Revision: https://reviews.llvm.org/D90257

3 years ago[OpenCL][NFC] Test commit: tidy up whitespace in comment
Stuart Brady [Fri, 4 Jun 2021 13:44:12 +0000 (14:44 +0100)]
[OpenCL][NFC] Test commit: tidy up whitespace in comment

3 years ago[AArch64] Further enable UnrollAndJam
Nicholas Guy [Wed, 26 May 2021 13:49:58 +0000 (14:49 +0100)]
[AArch64] Further enable UnrollAndJam

Due to the dependency on runtime unrolling, UnJ is only
enabled by default on in-order scheduling models,
and if a cpu is specified through -mcpu.

Differential Revision: https://reviews.llvm.org/D103604

3 years ago[HIP] Fix spack HIP device lib detection
Yaxun (Sam) Liu [Thu, 27 May 2021 19:51:02 +0000 (15:51 -0400)]
[HIP] Fix spack HIP device lib detection

spack HIP device library is installed at amdgcn directory under llvm/clang
directory.

This patch fixes detection of HIP device library for spack.

Reviewed by: Artem Belevich, Harmen Stoppels

Differential Revision: https://reviews.llvm.org/D103281

3 years ago[InstCombine] add/adjust test comments; NFC
Sanjay Patel [Fri, 4 Jun 2021 13:03:32 +0000 (09:03 -0400)]
[InstCombine] add/adjust test comments; NFC

Follow-up to post-commit comment:
https://reviews.llvm.org/rG23a116c8c446

3 years ago[clang][deps] Support object files
Jan Svoboda [Wed, 2 Jun 2021 12:49:14 +0000 (14:49 +0200)]
[clang][deps] Support object files

When a project uses PCH with explicit modules, the build will look like this:

1. scan PCH dependencies
2. explicitly build PCH
3. scan TU dependencies
4. explicitly build TU

Step 2 produces an object file for the PCH, which the dependency scanner needs to read in step 3. This patch adds support for this.

The `clang-scan-deps` invocation in the attached test would fail without this change.

Depends on D103516.

Reviewed By: Bigcheese

Differential Revision: https://reviews.llvm.org/D103519

3 years ago[flang][driver] Add support for `-module-suffix`
Andrzej Warzynski [Fri, 4 Jun 2021 12:58:03 +0000 (13:58 +0100)]
[flang][driver] Add support for `-module-suffix`

This option is supported in `f18`, but not yet available in `flang-new`.
It is required in order to call `flang-new` from the `flang` bash
script.

Differential Revision: https://reviews.llvm.org/D103613

3 years ago[AMDGPU][GlobalISel] Legalize G_ABS
Mirko Brkusanin [Thu, 3 Jun 2021 16:09:45 +0000 (18:09 +0200)]
[AMDGPU][GlobalISel] Legalize G_ABS

Legalize and select G_ABS so that we can use llvm.abs intrinsic

Differential Revision: https://reviews.llvm.org/D102391

3 years ago[clang][deps] Add argument for customizing PCM paths
Jan Svoboda [Wed, 2 Jun 2021 12:07:14 +0000 (14:07 +0200)]
[clang][deps] Add argument for customizing PCM paths

Dependency scanning currently performs an implicit build. When testing that Clang can build modules with the command-lines generated by `clang-scan-deps`, the actual compilation would overwrite artifacts created during the scan, which makes debugging harder than it should be and can lead to errors in multi-step builds.

To prevent this, this patch adds new flag to `clang-scan-deps` that allows developers to customize the directory to use when generating module map paths, instead of always using the module cache. Moreover, the explicit context hash in now part of the PCM path, which will be useful in D102488, where the context hash can change due to command-line pruning.

Reviewed By: Bigcheese

Differential Revision: https://reviews.llvm.org/D103516

3 years ago[AMDGPU][MC][NFC] Fixed typos in parser
Dmitry Preobrazhensky [Fri, 4 Jun 2021 12:38:37 +0000 (15:38 +0300)]
[AMDGPU][MC][NFC] Fixed typos in parser

Differential Revision: https://reviews.llvm.org/D103680

3 years ago[flang] Generate type info symbols outside of derived type scopes
Jean Perier [Fri, 4 Jun 2021 12:27:23 +0000 (14:27 +0200)]
[flang] Generate type info symbols outside of derived type scopes

A recent change was made in https://reviews.llvm.org/D101482 to cope
with kind parameters. It had the side effect of generating some type
info symbols inside derived type scopes. Derived type scope symbols
are meant for components, and other/later compilation phases might
choke when finding compiler generated symbols there that are not
components.

This patch preserves the fix from D101482 while still generating the
symbols outside of derived type scopes.

Differential Revision: https://reviews.llvm.org/D103621

3 years ago[clang][AST] Set correct DeclContext in ASTImporter lookup table for ParmVarDecl.
Balázs Kéri [Fri, 4 Jun 2021 10:21:20 +0000 (12:21 +0200)]
[clang][AST] Set correct DeclContext in ASTImporter lookup table for ParmVarDecl.

ParmVarDecl is created with translation unit as the parent DeclContext
and later moved to the correct DeclContext. ASTImporterLookupTable
should be updated at this move.

Reviewed By: martong

Differential Revision: https://reviews.llvm.org/D103231

3 years ago[AArch64][SVE] Add support for using reverse forms of SVE2 shifts
Bradley Smith [Wed, 26 May 2021 11:07:17 +0000 (12:07 +0100)]
[AArch64][SVE] Add support for using reverse forms of SVE2 shifts

When using and ACLE intrinsic for an SVE2 shift, if the predicate passed
has all relevant lanes active, then use a reversed version of the
instruction if beneficial.

3 years ago[mlir] Catch nonconvertible types in async conversion
Christian Sigg [Thu, 3 Jun 2021 07:25:22 +0000 (09:25 +0200)]
[mlir] Catch nonconvertible types in async conversion

Reviewed By: ezhulenev, ftynse

Differential Revision: https://reviews.llvm.org/D103592

3 years ago[InstCombine] convert lshr to ashr to eliminate cast op
Sanjay Patel [Thu, 3 Jun 2021 20:13:16 +0000 (16:13 -0400)]
[InstCombine] convert lshr to ashr to eliminate cast op

This is similar to b865eead7657 ( D103617 ) and fixes:
https://llvm.org/PR50575

41b71f718b94c6f12b did this and more (noted with TODO
comments in the tests), but it didn't handle the case
where the destination is narrower than the source, so
it got reverted.

This is a simple match-and-replace. If there's evidence
that the TODO cases are useful, we can revisit/extend.

3 years ago[InstCombine] add tests for sext-of-trunc-of-lshr; NFC
Sanjay Patel [Thu, 3 Jun 2021 19:53:11 +0000 (15:53 -0400)]
[InstCombine] add tests for sext-of-trunc-of-lshr; NFC

3 years agoRevert "[gn build] port d1d36f7ad (llvm-tapi-diff)"
Nico Weber [Fri, 4 Jun 2021 10:46:19 +0000 (06:46 -0400)]
Revert "[gn build] port d1d36f7ad (llvm-tapi-diff)"

This reverts commit 13155138c1ce1e91032d467e20e557f9cdbf08f5.
d1d36f7ad was reverted in 5337c7550d.

3 years agoRe-land ae4303b42c, "Track PHI values through register coalescing"
Jeremy Morse [Fri, 4 Jun 2021 10:18:40 +0000 (11:18 +0100)]
Re-land ae4303b42c, "Track PHI values through register coalescing"

Was reverted in 0507fc2ffc9, in phi-coalesce-subreg.mir I'd explicitly named
some passes to run instead of specifying a range. As a result some
two-address-instrs weren't correctly rewritten and the verifier got upset.
Original commit message:

[DebugInstrRef][2/3] Track PHI values through register coalescing

In the instruction referencing variable location model, we store variable
locations that point at PHIs in MachineFunction during register allocation.
Unfortunately, register coalescing can substantially change the locations
of registers, and so that PHI-variable-location side table needs
maintenence during the pass.

This patch builds an index from the side table, and whenever a vreg gets
coalesced into another vreg, update the index to record the new vreg that
the PHI happens in. It also accepts a limited range of subregister
coalescing, for example merging a subregister into a larger class.

Differential Revision: https://reviews.llvm.org/D86813

3 years ago[test] Fix accidental match in parent_recurse_depth.s
Thomas Preud'homme [Fri, 4 Jun 2021 08:55:33 +0000 (09:55 +0100)]
[test] Fix accidental match in parent_recurse_depth.s

The CHECK-NOT directives in
tools/llvm-dwarfdump/X86/parent_recurse_depth.s can accidentally match
something in the path of the object file created by yaml2obj, for
example:

llvm-project/llvm/test/tools/llvm-dwarfdump/X86/parent_recurse_depth.s:13:12:
error: ONE-NOT: excluded string found in input
           ^
<stdin>:1:22: note: found here
builds/llvm-projects/mainline/release/test/tools/llvm-dwarfdump/X86/Output/parent_recurse_depth.s.tmp.o: file format elf64-x86-64
                     ^~~~

This commit alleviate this issue by consuming the file name from the
output, forcing all the CHECK-NOT to match what comes after.

Reviewed By: Higuoxing

Differential Revision: https://reviews.llvm.org/D103676

3 years ago[lldb][NFC] Remove a redundant call to weak_ptr::expired
Raphael Isemann [Fri, 4 Jun 2021 08:59:31 +0000 (10:59 +0200)]
[lldb][NFC] Remove a redundant call to weak_ptr::expired

The `lock` call directly will check for us if the `weak_ptr` is expired and
returns an invalid `shared_ptr` (which we correctly handle), so this check is
redundant.

Reviewed By: JDevlieghere

Differential Revision: https://reviews.llvm.org/D103442

3 years ago[libcxx] Don't use an undefined '+' in unsigned/octal/hexal print formats
Martin Storsjö [Tue, 1 Jun 2021 09:04:58 +0000 (09:04 +0000)]
[libcxx] Don't use an undefined '+' in unsigned/octal/hexal print formats

If building code like this:

    unsigned long val = 1000;
    snprintf(buf, sizeof(buf), "%+lu", val);

with clang, clang warns

    warning: flag '+' results in undefined behavior with 'u' conversion specifier [-Wformat]

Therefore, don't construct such undefined format strings. (There's
no compiler warnings here, as the compiler can't inspect dynamically
assembled format strings.)

This fixes number formatting in mingw-w64 if built with
`__USE_MINGW_ANSI_STDIO` defined (there, the '+' flag causes a
leading plus to be printed when formatting unsigned numbers too,
while the '+' flag doesn't cause any extra leading plus in other
stdio implementations).

Differential Revision: https://reviews.llvm.org/D103444

3 years ago[SelectionDAG] Extend FoldConstantVectorArithmetic to SPLAT_VECTOR
Fraser Cormack [Thu, 27 May 2021 12:45:23 +0000 (13:45 +0100)]
[SelectionDAG] Extend FoldConstantVectorArithmetic to SPLAT_VECTOR

This patch extends the SelectionDAG's ability to constant-fold vector
arithmetic to include support for SPLAT_VECTOR. This is not only for
scalable-vector types but also for fixed-length vector types, which
helps Hexagon in a couple of cases.

The original RISC-V test case was in fact an infinite DAGCombine loop.
The pattern `and (truncate v1), (truncate v2)` can be combined to
`truncate (and v1, v2)` but the truncate can similarly be combined back
to `truncate (and v1, v2)` (but, crucially, only when one of `v1` or
`v2` is a constant vector).

It wasn't exposed in on fixed-length types because a TRUNCATE of a
constant BUILD_VECTOR was folded into the BUILD_VECTOR itself, whereas
this did not happen for the equivalent (scalable-vector) SPLAT_VECTOR.

Reviewed By: RKSimon, craig.topper

Differential Revision: https://reviews.llvm.org/D103246

3 years ago[mlir][linalg] Cleanup left over uses of deprecated LinalgOp methods.
Tobias Gysi [Fri, 4 Jun 2021 08:18:19 +0000 (08:18 +0000)]
[mlir][linalg] Cleanup left over uses of deprecated LinalgOp methods.

Replace all remaining uses of deprecated Structured Op Interface methods. This patch is based on https://reviews.llvm.org/D103394.

Differential Revision: https://reviews.llvm.org/D103673

3 years agoAArch64: support atomic zext/sextloads
Tim Northover [Wed, 5 May 2021 12:12:55 +0000 (13:12 +0100)]
AArch64: support atomic zext/sextloads

3 years ago[flang] Change capitalization for Adjustl/r
Diana Picus [Mon, 31 May 2021 08:51:11 +0000 (08:51 +0000)]
[flang] Change capitalization for Adjustl/r

Rename the definitions of the character runtime functions Adjustl and
Adjustr (used to be AdjustL and AdjustR respectively).

Also add unit tests (and move some of the helpers to the top of the
file, since they're now used in more than one place).

Differential Revision: https://reviews.llvm.org/D103392

3 years ago[Debug-Info] handle DW_CC_pass_by_value/DW_CC_pass_by_reference under strict DWARF.
Esme-Yi [Fri, 4 Jun 2021 08:14:47 +0000 (08:14 +0000)]
[Debug-Info] handle DW_CC_pass_by_value/DW_CC_pass_by_reference under strict DWARF.

Summary: When -strict-dwarf=true is specified, the calling convention info
    DW_CC_pass_by_value or DW_CC_pass_by_reference can only be generated at DWARF5.

Reviewed By: shchenz, dblaikie

Differential Revision: https://reviews.llvm.org/D103300

3 years ago[Format] Fix incorrect pointer detection
Yilong Guo [Fri, 4 Jun 2021 07:37:21 +0000 (09:37 +0200)]
[Format] Fix incorrect pointer detection

https://llvm.org/PR50429

Before:
    void f() { f(float(1), a *a); }

After:
    void f() { f(float(1), a * a); }

Signed-off-by: Yilong Guo <yilong.guo@intel.com>
Reviewed By: HazardyKnusperkeks, curdeius

Differential Revision: https://reviews.llvm.org/D103589

3 years ago[Sema][RISCV][SVE] Allow ?: to select Typedef BuiltinType in C
ShihPo Hung [Thu, 3 Jun 2021 08:10:21 +0000 (16:10 +0800)]
[Sema][RISCV][SVE] Allow ?: to select Typedef BuiltinType in C

This patch solves an error such as:
  incompatible operand types ('vbool4_t' (aka '__rvv_bool4_t') and '__rvv_bool4_t')
when one of the value is a TypedefType of the other value in ?:.

Reviewed By: rjmccall

Differential Revision: https://reviews.llvm.org/D103603

3 years agoAdd LLDB in release binaries by default
Muhammad Omair Javaid [Fri, 4 Jun 2021 06:56:28 +0000 (11:56 +0500)]
Add LLDB in release binaries by default

LLDB is currently not selected in LLVM release testing and thus it
doesnt make its way into prebuilt binaries which build with default
configuration. This patch enables LLDB by default in test-release
script.

Assuming LLDB build by default was disabled back in 2016 LLDB support
for various architectures has a long way since then. It has buildbots
for most architectures and supports a case to be included by default.

Also lldb build can easily be disabled in case some release managers
choose to do so.

Reviewed By: tstellar

Differential Revision: https://reviews.llvm.org/D101864

3 years ago[AMDGPU] [IndirectCalls] Don't propagate attributes to address taken functions and...
madhur13490 [Wed, 26 May 2021 05:47:03 +0000 (11:17 +0530)]
[AMDGPU] [IndirectCalls] Don't propagate attributes to address taken functions and their callees

Don't propagate launch bound related attributes to
address taken functions and their callees. The idea
is to do a traversal over the call graph starting at
address taken functions and erase the attributes
set by previous logic i.e. process().

This two phase approach makes sure that we don't
miss out on deep nested callees from address taken
functions as a function might be called directly as
well as indirectly.

This patch is also reattempt to D94585 as latent issues
are fixed in hasAddressTaken function in the recent
past.

Reviewed By: arsenm

Differential Revision: https://reviews.llvm.org/D103138

3 years agoRevert "[AMDGPU] Increase alignment of LDS globals if necessary before LDS lowering."
hsmahesha [Fri, 4 Jun 2021 05:46:46 +0000 (11:16 +0530)]
Revert "[AMDGPU] Increase alignment of LDS globals if necessary before LDS lowering."

This reverts commit d71ff907ef23eaef86ad66ba2d711e4986cd6cb2.

3 years agoRevert "[llvm] llvm-tapi-diff"
Cyndy Ishida [Fri, 4 Jun 2021 03:53:06 +0000 (20:53 -0700)]
Revert "[llvm] llvm-tapi-diff"

This reverts commit d1d36f7ad2ae82bea8a6fcc40d6c42a72e21f096.
Reverting this patch to investigate linux bot failures
 + fix with author offline

3 years ago[AMDGPU] Increase alignment of LDS globals if necessary before LDS lowering.
hsmahesha [Fri, 4 Jun 2021 04:04:37 +0000 (09:34 +0530)]
[AMDGPU] Increase alignment of LDS globals if necessary before LDS lowering.

Before packing LDS globals into a sorted structure, make sure that
their alignment is properly updated based on their size. This will make
sure that the members of sorted structure are properly aligned, and
hence it will further reduce the probability of unaligned LDS access.

Reviewed By: rampitec

Differential Revision: https://reviews.llvm.org/D103261

3 years ago[NFC] Remove checking pointee type for byval/preallocated type
Arthur Eubanks [Fri, 4 Jun 2021 01:57:36 +0000 (18:57 -0700)]
[NFC] Remove checking pointee type for byval/preallocated type

These currently always require a type parameter. The bitcode reader
already upgrades old bitcode without the type parameter to use the
pointee type.

3 years ago[scudo] Rework Vector/String
Kostya Kortchinsky [Thu, 3 Jun 2021 19:11:05 +0000 (12:11 -0700)]
[scudo] Rework Vector/String

Some platforms (eg: Trusty) are extremelly memory constrained, which
doesn't necessarily work well with some of Scudo's current assumptions.

`Vector` by default (and as such `String` and `ScopedString`) maps a
page, which is a bit of a waste. This CL changes `Vector` to use a
buffer local to the class first, then potentially map more memory if
needed (`ScopedString` currently are all stack based so it would be
stack data). We also want to allow a platform to prevent any dynamic
resizing, so I added a `CanGrow` templated parameter that for now is
always `true` but would be set to `false` on Trusty.

Differential Revision: https://reviews.llvm.org/D103641

3 years agoRevert "Update and improve compiler-rt tests for -mllvm -asan_use_after_return=(never...
Nico Weber [Fri, 4 Jun 2021 01:01:11 +0000 (21:01 -0400)]
Revert "Update and improve compiler-rt tests for -mllvm -asan_use_after_return=(never|[runtime]|always)."

This reverts commit 41b3088c3f33d712e3d2f64b66ae4eb701fa4bfb.
Doesn't build on macOS, see comments on https://reviews.llvm.org/D103304

3 years ago[lld/mac] Add test coverage for --reproduce + -flat_namespace
Nico Weber [Thu, 3 Jun 2021 19:29:22 +0000 (15:29 -0400)]
[lld/mac] Add test coverage for --reproduce + -flat_namespace

Works fine already, now it has a test too.

Differential Revision: https://reviews.llvm.org/D103643

3 years ago[CSSPGO][llvm-profgen] Make extended binary the default output format
Wenlei He [Thu, 3 Jun 2021 20:42:24 +0000 (13:42 -0700)]
[CSSPGO][llvm-profgen] Make extended binary the default output format

Make extended binary the default output format for CSSPGO. This avoids having to pass flag every time when generating profile. It also matches llvm-profdata where binary profile is the default (should we switch to extbinary as default for llvm-profdata?).

We plan to compress name table for context profile, which depends on the built-in compression of extbinary.

Differential Revision: https://reviews.llvm.org/D103650

3 years ago[RISCV] Simplify some code in RISCVInsertVSETVLI by calling an existing function...
Craig Topper [Fri, 4 Jun 2021 00:31:54 +0000 (17:31 -0700)]
[RISCV] Simplify some code in RISCVInsertVSETVLI by calling an existing function that does the same thing. NFCI

3 years ago[NFC] Add ArrayRef includes to two files.
Stella Laurenzo [Thu, 3 Jun 2021 23:38:12 +0000 (16:38 -0700)]
[NFC] Add ArrayRef includes to two files.

These started failing on one of our buildbots. I didn't completely root cause the situation and just added the explicit includes that correct the issue.

Reviewed By: rriddle, jpienaar

Differential Revision: https://reviews.llvm.org/D103657

3 years ago[mlir][SPIRV] Add lowering for math.log1p operation to SPIR-V dialect.
MaheshRavishankar [Thu, 3 Jun 2021 23:25:56 +0000 (16:25 -0700)]
[mlir][SPIRV] Add lowering for math.log1p operation to SPIR-V dialect.

Differential Revision: https://reviews.llvm.org/D103635

3 years ago[gn build] port d1d36f7ad (llvm-tapi-diff)
Nico Weber [Thu, 3 Jun 2021 23:22:21 +0000 (19:22 -0400)]
[gn build] port d1d36f7ad (llvm-tapi-diff)

3 years ago[mlir-lsp-server] Add support for tracking the use/def chains of symbols
River Riddle [Thu, 3 Jun 2021 23:03:57 +0000 (16:03 -0700)]
[mlir-lsp-server] Add support for tracking the use/def chains of symbols

This revision adds assembly state tracking for uses of symbols, allowing for go-to-definition and references support for SymbolRefAttrs.

Differential Revision: https://reviews.llvm.org/D103585

3 years ago[HIP] Fix amdgcn builtin for long type
Yaxun (Sam) Liu [Wed, 2 Jun 2021 22:24:12 +0000 (18:24 -0400)]
[HIP] Fix amdgcn builtin for long type

Currently some amdgcn builtins are defined with long int type,
which causes invalid IR on Windows since long int is 32 bit
on Windows whereas these builtins have 64 bit arguments.

long long int type cannot be used since it is 128 bit in OpenCL.

This patch uses 64 bit int type instead of long int to define 64 bit int
arguments or return for amdgcn builtins.

Reviewed by: Artem Belevich

Differential Revision: https://reviews.llvm.org/D103563

3 years ago[mlir] Remove redundant loads
Amy Zhuang [Thu, 3 Jun 2021 22:27:25 +0000 (15:27 -0700)]
[mlir] Remove redundant loads

Reviewed By: vinayaka-polymage, bondhugula

Differential Revision: https://reviews.llvm.org/D103294

3 years ago[TargetLowering] Only inspect attributes in the arguments for ArgListEntry
Arthur Eubanks [Mon, 31 May 2021 16:37:11 +0000 (09:37 -0700)]
[TargetLowering] Only inspect attributes in the arguments for ArgListEntry

Parameter attributes are considered part of the function [1], and like
mismatched calling conventions [2], we can't have the verifier check for
mismatched parameter attributes.

Issues can be diagnosed with D103412.

[1] https://llvm.org/docs/LangRef.html#parameter-attributes
[2] https://llvm.org/docs/FAQ.html#why-does-instcombine-simplifycfg-turn-a-call-to-a-function-with-a-mismatched-calling-convention-into-unreachable-why-not-make-the-verifier-reject-it

Reviewed By: rnk

Differential Revision: https://reviews.llvm.org/D101806

3 years ago[BuildLibCalls] Properly set ABI attributes on arguments
Arthur Eubanks [Mon, 31 May 2021 16:15:25 +0000 (09:15 -0700)]
[BuildLibCalls] Properly set ABI attributes on arguments

Some floating point lib calls have ABI attributes that need to be set on
the caller. Found via D103412.

Reviewed By: rnk

Differential Revision: https://reviews.llvm.org/D103415

3 years ago[flang] Fix spurious "already declared" errors for interfaces
peter klausler [Thu, 3 Jun 2021 00:33:34 +0000 (17:33 -0700)]
[flang] Fix spurious "already declared" errors for interfaces

When a subroutine or function symbol is defined in an INTERFACE
block, it's okay if a symbol of the same name appears in a
scope between the global scope and the scope of the INTERFACE.

Differential Revision: https://reviews.llvm.org/D103580

3 years ago[libc++] Simplify apple-install-libcxx since we always use the same CMake cache
Louis Dionne [Thu, 3 Jun 2021 22:26:31 +0000 (18:26 -0400)]
[libc++] Simplify apple-install-libcxx since we always use the same CMake cache

3 years ago[CMake][ELF] Add -fno-semantic-interposition for GCC and Clang>=13
Fangrui Song [Thu, 3 Jun 2021 22:26:34 +0000 (15:26 -0700)]
[CMake][ELF] Add -fno-semantic-interposition for GCC and Clang>=13

In a `-DCMAKE_BUILD_TYPE=Release -DLLVM_TARGETS_TO_BUILD=X86 -DLLVM_LINK_LLVM_DYLIB=on -DCLANG_LINK_CLANG_DYLIB=on`
build, libLLVM-13git.so is 2% smaller and libclang-cpp.so is 1% smaller (on top of -Wl,-Bsymbolic-functions).
There may be some small performance improvement as well because GCC
-fPIC suppresses interprocedural optimizations for non-inline
definitions by default.

Note: we cannot add -fno-semantic-interposition for Clang<13.  Clang<13's
implementation additionally optimizes global variables, which is incompatible
with unfortunate ELF -fno-pic default: direct access relocations for external
data. If the executable has a -fno-pic object file referencing a global variable
declared in a public header, the direct access relocation will cause a copy
relocation. The executable and libLLVM.so/libclang-cpp.so will disagree on the
address.

Reviewed By: phosek

Differential Revision: https://reviews.llvm.org/D102453

3 years ago[flang] Catch errors in function calls instead of crashing
peter klausler [Thu, 3 Jun 2021 00:28:50 +0000 (17:28 -0700)]
[flang] Catch errors in function calls instead of crashing

Add some missing error messages, and permit the appearance
of EntityDetails symbols in dummy argument type characterization.

Differential Revision: https://reviews.llvm.org/D103576

3 years ago[flang] Emit error about missing interface when needed
peter klausler [Thu, 3 Jun 2021 00:25:29 +0000 (17:25 -0700)]
[flang] Emit error about missing interface when needed

When a procedure pointer with no interface is called by a
function reference, complain about the lack.

Differential Revision: https://reviews.llvm.org/D103573

3 years ago[libc++] Define _LIBCPP_NO_NATIVE_SEMAPHORES even outside of pthread
Louis Dionne [Thu, 3 Jun 2021 22:18:41 +0000 (18:18 -0400)]
[libc++] Define _LIBCPP_NO_NATIVE_SEMAPHORES even outside of pthread

<semaphore> needs to know about whether native semaphores are supported
or not, even if we're not using the pthread API.

3 years ago[libc++] Also build the static library in the Apple cache
Louis Dionne [Thu, 3 Jun 2021 22:15:13 +0000 (18:15 -0400)]
[libc++] Also build the static library in the Apple cache

3 years ago[flang] Fix crash on structure constructor as selector
peter klausler [Thu, 3 Jun 2021 00:21:41 +0000 (17:21 -0700)]
[flang] Fix crash on structure constructor as selector

In something like "ASSOCIATE(X=>T(1))", the "T(1)" is parsed
as a Variable because it looks like a function reference or
array reference; if it turns out to be a structure constructor,
which is something we can't know until we're able to attempt
generic interface resolution in semantics, the parse tree needs
to be fixed up by replacing the Variable with an Expr.

The compiler could already do this for putative function references
encapsulated as Exprs, so this patch moves some code around and
adds parser::Selector to the overloads of expression analysis.

Differential Revision: https://reviews.llvm.org/D103572

3 years agoKill a variable which is unused after cddcc4cf [nfc]
Philip Reames [Thu, 3 Jun 2021 21:35:43 +0000 (14:35 -0700)]
Kill a variable which is unused after cddcc4cf [nfc]

3 years ago[NFC] Add missing includes for LLVM_ENABLE_MODULES builds
zero9178 [Thu, 3 Jun 2021 21:28:47 +0000 (23:28 +0200)]
[NFC] Add missing includes for LLVM_ENABLE_MODULES builds

Building LLVM with the LLVM_ENABLE_MODULES cmake option fails when the modules are being compiled due to missing includes. This is a side effect of some transitive includes that changed recently.

Differential Revision: https://reviews.llvm.org/D103645

3 years ago[NFC][compiler-rt][hwasan] Refactor hwasan functions
Leonard Chan [Thu, 3 Jun 2021 21:22:37 +0000 (14:22 -0700)]
[NFC][compiler-rt][hwasan] Refactor hwasan functions

This moves the implementations for HandleTagMismatch, __hwasan_tag_mismatch4,
and HwasanAtExit from hwasan_linux.cpp to hwasan.cpp and declares them in hwasan.h.
This way, calls to those functions can be shared with the fuchsia implementation
without duplicating code.

Differential Revision: https://reviews.llvm.org/D103562

3 years ago[flang] Support known constant lengths in DynamicType
peter klausler [Thu, 3 Jun 2021 00:13:55 +0000 (17:13 -0700)]
[flang] Support known constant lengths in DynamicType

The constexpr-capable class evaluate::DynamicType represented
CHARACTER length only with a nullable pointer into the declared
parameters of types in the symbol table, which works fine for
anything with a declaration but turns out to not suffice to
describe the results of the ACHAR() and CHAR() intrinsic
functions.  So extend DynamicType to also accommodate known
constant CHARACTER lengths, too; use them for ACHAR & CHAR;
clean up several use sites and fix regressions found in test.

Differential Revision: https://reviews.llvm.org/D103571

3 years ago[LTO] Fix -fwhole-program-vtables handling after HIP ThinLTO patch
Teresa Johnson [Wed, 2 Jun 2021 23:37:23 +0000 (16:37 -0700)]
[LTO] Fix -fwhole-program-vtables handling after HIP ThinLTO patch

A recent change (D99683) to support ThinLTO for HIP caused a regression
when compiling cuda code with -flto=thin -fwhole-program-vtables.
Specifically, we now get an error:
error: invalid argument '-fwhole-program-vtables' only allowed with '-flto'

This error is coming from the device offload cc1 action being set up for
the cuda compile, for which -flto=thin doesn't apply and gets dropped.
This is a regression, but points to a potential issue that was silently
occurring before the patch, details below.

Before D99683, the check for fwhole-program-vtables in the driver looked
like:

  if (WholeProgramVTables) {
    if (!D.isUsingLTO())
      D.Diag(diag::err_drv_argument_only_allowed_with)
          << "-fwhole-program-vtables"
          << "-flto";
    CmdArgs.push_back("-fwhole-program-vtables");
  }

And D.isUsingLTO() returned true since we have -flto=thin. However,
because the cuda cc1 compile is doing device offloading, which didn't
support any LTO, there was other code that suppressed -flto* options
from being passed to the cc1 invocation. So the cc1 invocation silently
had -fwhole-program-vtables without any -flto*. This seems potentially
problematic, since if we had any virtual calls we would get type test
assume sequences without the corresponding LTO pass that handles them.

However, with the patch, which adds support for device offloading LTO
option -foffload-lto=thin, the code has changed so that we set a bool
IsUsingLTO based on either -flto* or -foffload-lto*, depending on
whether this is the device offloading action. For the device offload
action in our compile, since we don't have -foffload-lto, IsUsingLTO is
false, and the check for LTO with -fwhole-program-vtables now fails.

What we should do is only pass through -fwhole-program-vtables to the
cc1 invocation that has LTO enabled (either the device offload action
with -foffload-lto, or the non-device offload action with -flto), and
otherwise drop the -fwhole-program-vtables for the non-LTO action.
Then we should error only if we have -fwhole-program-vtables without any
-f*lto* options.

Differential Revision: https://reviews.llvm.org/D103579

3 years agoA couple style tweaks on top of 5c0d1b2f9 [nfc]
Philip Reames [Thu, 3 Jun 2021 21:14:59 +0000 (14:14 -0700)]
A couple style tweaks on top of 5c0d1b2f9 [nfc]

3 years ago[LoopUnroll] Eliminate PreserveCondBr parameter and fix a bug in the process
Philip Reames [Thu, 3 Jun 2021 21:03:21 +0000 (14:03 -0700)]
[LoopUnroll] Eliminate PreserveCondBr parameter and fix a bug in the process

This builds on D103584. The change eliminates the coupling between unroll heuristic and implementation w.r.t. knowing when the passed in trip count is an exact trip count or a max trip count. In theory the new code is slightly less powerful (since it relies on exact computable trip counts), but in practice, it appears to cover all the same cases. It can also be extended if needed.

The test change shows what appears to be a bug in the existing code around the interaction of peeling and unrolling. The original loop only ran 8 iterations. The previous output had the loop peeled by 2, and then an exact unroll of 8. This meant the loop ran a total of 10 iterations which appears to have been a miscompile.

Differential Revision: https://reviews.llvm.org/D103620

3 years ago[flang] Catch procedure pointer interface error
peter klausler [Thu, 3 Jun 2021 00:09:42 +0000 (17:09 -0700)]
[flang] Catch procedure pointer interface error

A procedure pointer is allowed to name a specific intrinsic function
from F'2018 table 16.2 as its interface, but not other intrinsic
procedures.  Catch this error, and thereby also fix a crash resulting
from a failure later in compilation from failed characteristics;
while here, also catch the similar error with initializers.

Differential Revision: https://reviews.llvm.org/D103570

3 years ago[test] Update more clang/test/Profile tests
Fangrui Song [Thu, 3 Jun 2021 20:43:19 +0000 (13:43 -0700)]
[test] Update more clang/test/Profile tests

3 years ago[AMDGPU] Fix a crash when selecting a particular case of buffer_load_format_d16
Julien Pagès [Thu, 3 Jun 2021 20:32:15 +0000 (16:32 -0400)]
[AMDGPU] Fix a crash when selecting a particular case of buffer_load_format_d16

In this particular example, we had a crash when compiling it
for several architectures. This patch extends the legalization
of extract_subvector to avoid this problem.

Differential Revision: https://reviews.llvm.org/D103344

3 years ago[test] Update c-linkage-available_externally.c
Fangrui Song [Thu, 3 Jun 2021 20:40:01 +0000 (13:40 -0700)]
[test] Update c-linkage-available_externally.c

3 years ago[flang] Accept BOZ in array constructors w/o types
peter klausler [Thu, 3 Jun 2021 00:02:43 +0000 (17:02 -0700)]
[flang] Accept BOZ in array constructors w/o types

As a benign extension common to other Fortran compilers,
accept BOZ literals in array constructors w/o explicit
types, treating them as integers.

Differential Revision: https://reviews.llvm.org/D103569

3 years ago[Constants][PowerPC] Check exactlyValue for ppc_fp128 in isNullValue
Jinsong Ji [Thu, 3 Jun 2021 20:08:35 +0000 (20:08 +0000)]
[Constants][PowerPC] Check exactlyValue for ppc_fp128 in isNullValue

 PPC_FP128 determines isZero/isNan/isInf using high-order double value
 only. Checking isZero/isNegative might return the isNullValue unexpectedly.
 eg:
   0xM0000000000000000FFFFFFFFFFFFFFFFF

isZero, but it is not NullValue.

Reviewed By: efriedma

Differential Revision: https://reviews.llvm.org/D103634