platform/upstream/llvm.git
4 years ago[MLIR][Standard] Update `assert` documentation post commit
Frederik Gossen [Wed, 15 Jul 2020 16:13:10 +0000 (16:13 +0000)]
[MLIR][Standard] Update `assert` documentation post commit

Update line wrapping.

Differential Revision: https://reviews.llvm.org/D83769

4 years ago[NFC] Add float aggregate ABI test for PowerPC
Qiu Chaofan [Wed, 15 Jul 2020 16:05:25 +0000 (00:05 +0800)]
[NFC] Add float aggregate ABI test for PowerPC

4c5a93bd landed adjustment to handle C++20 no_unique_address attribute
correctly, clang treats empty members in aggregate type differently if
having this attribute. This commit adds necessary test for PowerPC
target to reflect this change.

4 years ago[Legalize] Hoist invariant condition in ExpandVectorBuildThroughStack(...)
Cameron McInally [Wed, 15 Jul 2020 16:04:22 +0000 (11:04 -0500)]
[Legalize] Hoist invariant condition in ExpandVectorBuildThroughStack(...)

The operands of a BUILD_VECTOR must all have the same type, so we can hoist this invariant condition out of the loop.

Differential Revision: https://reviews.llvm.org/D83882

4 years ago[lldb/Test] Remove support for forking a subprocess from the test suite.
Jonas Devlieghere [Wed, 15 Jul 2020 15:51:24 +0000 (08:51 -0700)]
[lldb/Test] Remove support for forking a subprocess from the test suite.

Remove the forkSubprocess method and its bookkeeping.
TestCreateAfterAttach is the only test using the fork method and I'm not
convinced it adds enough to warrant the maintenance. Pavel suggested the
same thing in D83815.

4 years agoRevert "[TSan] Optimize handling of racy address"
Joachim Protze [Wed, 15 Jul 2020 15:39:30 +0000 (17:39 +0200)]
Revert "[TSan] Optimize handling of racy address"

This reverts commit 00e3a1ddec95c0b48ce216220d7e3481dab3bc78.
The commit broke most build bots, investigating.

4 years ago[MLIR][Shape] Fix `shape_of` lowering to `scf`
Frederik Gossen [Wed, 15 Jul 2020 15:37:13 +0000 (15:37 +0000)]
[MLIR][Shape] Fix `shape_of` lowering to `scf`

The use of the `scf.for` callback builder does not allow for a rollback of the
emitted conversions. Instead, we populate the loop body through the conversion
rewriter directly.

Differential Revision: https://reviews.llvm.org/D83873

4 years ago[HardwareLoops] Add sibling loop test.
David Green [Wed, 15 Jul 2020 13:49:57 +0000 (14:49 +0100)]
[HardwareLoops] Add sibling loop test.

This missed being part of 9e03547cab69.

4 years ago[OpenMP][Docs] Update `present` map type modifier status
Joel E. Denny [Wed, 15 Jul 2020 15:17:00 +0000 (11:17 -0400)]
[OpenMP][Docs] Update `present` map type modifier status

4 years ago[TSan] Optimize handling of racy address
Joachim Protze [Wed, 15 Jul 2020 14:45:27 +0000 (16:45 +0200)]
[TSan] Optimize handling of racy address

This patch splits the handling of racy address and racy stack into separate
functions. If a race was already reported for the address, we can avoid the
cost for collecting the involved stacks.

This patch also removes the race condition in storing the racy address / racy
stack. This race condition allowed all threads to report the race.

This patch changes the transitive suppression of reports. Previously
suppression could transitively chain memory location and racy stacks.
Now racy memory and racy stack are separate suppressions.

Reviewed by: dvyukov

Differential Revision: https://reviews.llvm.org/D83625

4 years agoRevert "[lldb] Use the basename of the Python test for the log name instead of the...
Raphael Isemann [Wed, 15 Jul 2020 14:26:37 +0000 (16:26 +0200)]
Revert "[lldb] Use the basename of the Python test for the log name instead of the class name"

This reverts commit 29aab9b5c748b28b231e2ca0f9b95453638ade1a.

It seems on Windows the file name is just always "lldbsuite.test.lldbtest" for
all tests and that breaks pretty much everything. Reverting until we have
a better solution.

4 years agoCodeGenPrep: remove AssertingVH references before deleting dead instructions.
Tim Northover [Wed, 15 Jul 2020 08:49:49 +0000 (09:49 +0100)]
CodeGenPrep: remove AssertingVH references before deleting dead instructions.

CodeGenPrepare keeps fairly close track of various instructions it's
seen, particularly GEPs, in maps and vectors. However, sometimes those
instructions become dead and get removed while it's still executing.
This triggers AssertingVH references to them in an asserts build and
could lead to miscompiles in a release build (I've only seen a later
segfault though).

So this patch adds a callback to
RecursivelyDeleteTriviallyDeadInstructions which can make sure the
instruction about to be deleted is removed from CodeGenPrepare's data
structures.

4 years ago[lldb/test] Remove JOIN_CMD from Makefile.rules
Pavel Labath [Wed, 15 Jul 2020 13:42:00 +0000 (15:42 +0200)]
[lldb/test] Remove JOIN_CMD from Makefile.rules

It's possible to achieve the same effect by providing multi-step recipe
instead of a single-step recipe where the step happens to contain
multiple commands.

4 years ago[lldb] Delete useless assertion
Pavel Labath [Wed, 15 Jul 2020 13:42:53 +0000 (15:42 +0200)]
[lldb] Delete useless assertion

It served a puprose while we were using the test name to provide a name
for the created file. Now that the files are created in memory, we don't
need that.

4 years ago[clangd] Use llvm::errs() instead of outs() for errors
Kirill Bobyrev [Wed, 15 Jul 2020 12:34:23 +0000 (14:34 +0200)]
[clangd] Use llvm::errs() instead of outs() for errors

Summary: errs() is more appropriate for error messages in dexp and clangd-index-server.

Reviewers: sammccall

Reviewed By: sammccall

Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D83827

4 years ago[lldb] Make expect_expr fall back to the dummy target if no target is selected
Raphael Isemann [Wed, 15 Jul 2020 11:55:32 +0000 (13:55 +0200)]
[lldb] Make expect_expr fall back to the dummy target if no target is selected

Summary:

Currently expect_expr will not run the expression if no target is selected. This
patch changes this behavior so that expect_expr will instead fall back to the
dummy target similar to what the `expression` command is doing. This way we
don't have to compile an empty executable to be able to use `expect_expr` (which
is a waste of resources for tests that just test generic type system features).

As a test I modernized the TestTypeOfDeclTypeExpr into a Python test +
expect_expr (as it relied on the dummy target fallback of the expression
command).

Reviewers: labath, JDevlieghere

Reviewed By: labath

Subscribers: abidh

Differential Revision: https://reviews.llvm.org/D83388

4 years ago[lldb] Use the basename of the Python test for the log name instead of the class...
Raphael Isemann [Wed, 15 Jul 2020 11:53:44 +0000 (13:53 +0200)]
[lldb] Use the basename of the Python test for the log name instead of the class name

Summary:

From what I know we already have the restriction that every test in the test
suite needs to have a unique file name as that's used for generating the unique
build directory for a test. It seems there is also a restriction that every test
case class in the test suite needs to have a unique name as that's used to
generate the unique log file name for the test run.

This changes the log file format to use the basename of the test file instead so
that we only have to keep worrying about the 'unique file name' restriction from
now on.

This came up because I started naming the test classes "TestCase" (as repeating
the file name in the test class seems like redudant information that just makes
renaming tests a pain).

Reviewers: labath, JDevlieghere

Reviewed By: labath

Subscribers: mgorny, abidh

Differential Revision: https://reviews.llvm.org/D83767

4 years ago[mips] Rename FeatureMadd4 to FeatureNoMadd4. NFC
YunQiang Su [Wed, 15 Jul 2020 09:11:36 +0000 (12:11 +0300)]
[mips] Rename FeatureMadd4 to FeatureNoMadd4. NFC

`FeatureMadd4` is used to disable `madd4`, and the corresponding feature
option is `(+-)nomadd4`. Renaming to the `FeatureNoMadd4` makes its
purpose clear.

Patch by YunQiang Su.

Differential Revision: https://reviews.llvm.org/D83780

4 years ago[RISCV] Add matching of codegen patterns to RISCV Bit Manipulation Zbt asm instructions
lewis-revill [Wed, 15 Jul 2020 10:59:47 +0000 (11:59 +0100)]
[RISCV] Add matching of codegen patterns to RISCV Bit Manipulation Zbt asm instructions

This patch provides optimization of bit manipulation operations by
enabling the +experimental-b target feature.
It adds matching of single block patterns of instructions to specific
bit-manip instructions from the ternary subset (zbt subextension) of the
experimental B extension of RISC-V.
It adds also the correspondent codegen tests.

This patch is based on Claire Wolf's proposal for the bit manipulation
extension of RISCV:
https://github.com/riscv/riscv-bitmanip/blob/master/bitmanip-0.92.pdf

Differential Revision: https://reviews.llvm.org/D79875

4 years ago[RISCV] Add matching of codegen patterns to RISCV Bit Manipulation Zbs asm instructions
lewis-revill [Wed, 15 Jul 2020 10:57:29 +0000 (11:57 +0100)]
[RISCV] Add matching of codegen patterns to RISCV Bit Manipulation Zbs asm instructions

This patch provides optimization of bit manipulation operations by
enabling the +experimental-b target feature.
It adds matching of single block patterns of instructions to specific
bit-manip instructions from the single-bit subset (zbs subextension) of
the experimental B extension of RISC-V.
It adds also the correspondent codegen tests.

This patch is based on Claire Wolf's proposal for the bit manipulation
extension of RISCV:
https://github.com/riscv/riscv-bitmanip/blob/master/bitmanip-0.92.pdf

Differential Revision: https://reviews.llvm.org/D79874

4 years ago[RISCV] Add matching of codegen patterns to RISCV Bit Manipulation Zbbp asm instructions
lewis-revill [Wed, 15 Jul 2020 10:55:44 +0000 (11:55 +0100)]
[RISCV] Add matching of codegen patterns to RISCV Bit Manipulation Zbbp asm instructions

This patch provides optimization of bit manipulation operations by
enabling the +experimental-b target feature.
It adds matching of single block patterns of instructions to specific
bit-manip instructions belonging to both the permutation and the base
subsets of the experimental B extension of RISC-V.
It adds also the correspondent codegen tests.

This patch is based on Claire Wolf's proposal for the bit manipulation
extension of RISCV:
https://github.com/riscv/riscv-bitmanip/blob/master/bitmanip-0.92.pdf

Differential Revision: https://reviews.llvm.org/D79873

4 years ago[RISCV] Add matching of codegen patterns to RISCV Bit Manipulation Zbp asm instructions
lewis-revill [Wed, 15 Jul 2020 10:53:06 +0000 (11:53 +0100)]
[RISCV] Add matching of codegen patterns to RISCV Bit Manipulation Zbp asm instructions

This patch provides optimization of bit manipulation operations by
enabling the +experimental-b target feature.
It adds matching of single block patterns of instructions to specific
bit-manip instructions from the permutation subset (zbp subextension) of
the experimental B extension of RISC-V.
It adds also the correspondent codegen tests.

This patch is based on Claire Wolf's proposal for the bit manipulation
extension of RISCV:
https://github.com/riscv/riscv-bitmanip/blob/master/bitmanip-0.92.pdf

Differential Revision: https://reviews.llvm.org/D79871

4 years ago[RISCV] Add matching of codegen patterns to RISCV Bit Manipulation Zbb asm instructions
lewis-revill [Wed, 15 Jul 2020 10:50:03 +0000 (11:50 +0100)]
[RISCV] Add matching of codegen patterns to RISCV Bit Manipulation Zbb asm instructions

This patch provides optimization of bit manipulation operations by
enabling the +experimental-b target feature.
It adds matching of single block patterns of instructions to specific
bit-manip instructions from the base subset (zbb subextension) of the
experimental B extension of RISC-V.
It adds also the correspondent codegen tests.

This patch is based on Claire Wolf's proposal for the bit manipulation
extension of RISCV:
https://github.com/riscv/riscv-bitmanip/blob/master/bitmanip-0.92.pdf

Differential Revision: https://reviews.llvm.org/D79870

4 years ago[DSE,MSSA] Recognise init_trampoline in getLocForWriteEx
John Brawn [Wed, 15 Jul 2020 11:17:31 +0000 (12:17 +0100)]
[DSE,MSSA] Recognise init_trampoline in getLocForWriteEx

This fixes an instance where MemorySSA-using Dead Store Elimination is failing
to do a transformation that the non-MemorySSA-using version does.

Differential Revision: https://reviews.llvm.org/D83783

4 years ago[DWARFYAML] Remove useless conditional statement. NFC.
Xing GUO [Wed, 15 Jul 2020 10:53:55 +0000 (18:53 +0800)]
[DWARFYAML] Remove useless conditional statement. NFC.

The conditional statement is useless after
3a48a632d00ef1c98042140f402337fe13cdff52.

4 years ago[LoopRotate] Remove unnecessary verifyMemorySSA calls.
Florian Hahn [Wed, 15 Jul 2020 10:49:24 +0000 (11:49 +0100)]
[LoopRotate] Remove unnecessary verifyMemorySSA calls.

The actual rotation happens in processLoop, so the second removed
call to verifyMemorySSA was unnecessary.

In fact, processLoop/rotateLoop already verify MemorySSA before
and after transforming each loop. Hence, both calls can be removed.

Pointed out by @lebedev.ri post-commit D51718.

4 years ago[MLIR][Shape] Add `shape.shape_eq` operation
Frederik Gossen [Wed, 15 Jul 2020 10:28:29 +0000 (10:28 +0000)]
[MLIR][Shape] Add `shape.shape_eq` operation

Add `shape.shape_eq` operation to the shape dialect.
The operation allows to test shapes and extent tensors for equality.

Differential Revision: https://reviews.llvm.org/D82528

4 years ago[clangd] Config: also propagate in sync (testing) mode
Sam McCall [Tue, 14 Jul 2020 18:24:32 +0000 (20:24 +0200)]
[clangd] Config: also propagate in sync (testing) mode

Summary:
I hit this while trying to add a config-over-LSP lit test, which I think
is an appropriate way to test this feature.

That needs a few more changes though...

Reviewers: kadircet

Subscribers: ilya-biryukov, javed.absar, MaskRay, jkorous, arphaman, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D83802

4 years ago[Analyzer] Report every bug if only uniqueing location differs.
Balázs Kéri [Wed, 15 Jul 2020 09:05:22 +0000 (11:05 +0200)]
[Analyzer] Report every bug if only uniqueing location differs.

Summary:
Two CSA bug reports where only the uniqueing location is different
should be treated as different problems. The role of uniqueing location
is to differentiate bug reports.

Reviewers: Szelethus, baloghadamsoftware, NoQ, vsavchenko, xazax.hun, martong

Reviewed By: NoQ

Subscribers: NoQ, rnkovacs, xazax.hun, baloghadamsoftware, szepet, a.sidorin, mikhail.ramalho, Szelethus, donat.nagy, dkrupp, gamesh411, Charusso, martong, ASDenysPetrov, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D83115

4 years agoBump the trunk major version to 12 llvmorg-12-init
Hans Wennborg [Wed, 15 Jul 2020 09:40:53 +0000 (11:40 +0200)]
Bump the trunk major version to 12

and clear the release notes.

4 years ago[LLD][ELF] - Update test after yaml2obj change. NFC.
Georgii Rymar [Wed, 15 Jul 2020 10:00:02 +0000 (13:00 +0300)]
[LLD][ELF] - Update test after yaml2obj change. NFC.

Names of YAML keys changed in rG7a587ca93200c49e47fe205ce037895c81c5a542

4 years ago[RISCV] Duplicate pseudo expansion comment to RISCVMCCodeEmitter
Jessica Clarke [Wed, 15 Jul 2020 09:52:42 +0000 (10:52 +0100)]
[RISCV] Duplicate pseudo expansion comment to RISCVMCCodeEmitter

Follow-on from D77443. Although we're not fixing any of these
pseudo-instructions, the potential for them to be out of sync still
exists.

4 years ago[RISCV] Fix RISCVInstrInfo::getInstSizeInBytes for atomics pseudos
Jessica Clarke [Wed, 15 Jul 2020 09:48:41 +0000 (10:48 +0100)]
[RISCV] Fix RISCVInstrInfo::getInstSizeInBytes for atomics pseudos

Summary:
Without these, the generic branch relaxation pass will underestimate the
range required for branches spanning these and we can end up with
"fixup value out of range" errors rather than relaxing the branches.
Some of the instructions in the expansion may end up being compressed
but exactly determining that is awkward, and these conservative values
should be safe, if slightly suboptimal in rare cases.

Reviewers: asb, lenary, luismarques, lewis-revill

Reviewed By: asb, luismarques

Subscribers: hiraditya, rbar, johnrusso, simoncook, sabuasal, niosHD, kito-cheng, shiva0217, MaskRay, zzheng, edward-jones, rogfer01, MartinMosbeck, brucehoult, the_o, rkruppe, jfb, PkmX, jocewei, psnobl, benna, Jim, s.egerton, pzheng, sameer.abuasal, apazos, evandro, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D77443

4 years ago[yaml2obj] - Rename FileHeader::SH* fields.
Georgii Rymar [Tue, 14 Jul 2020 12:48:03 +0000 (15:48 +0300)]
[yaml2obj] - Rename FileHeader::SH* fields.

In D83482 we agreed to name e_* fields that are used for overriding
values (like e_phoff) as EPh* (e.g. EPhOff).

Currently we have a set of e_sh* fields that are named inconsistently
with this rule. This patch renames all of them.

Differential revision: https://reviews.llvm.org/D83766

4 years ago[Matrix] Only pass vector arg as overloaded type in MatrixBuilder.
Florian Hahn [Wed, 15 Jul 2020 09:30:48 +0000 (10:30 +0100)]
[Matrix] Only pass vector arg as overloaded type in MatrixBuilder.

In 2b3c505, the pointer arguments for the matrix load and store
intrinsics was changed to always be the element type of the vector
argument.

This patch updates the MatrixBuilder to not add the pointer type to the
overloaded types and adjusts the clang/mlir tests.

This should fix a few build failures on GreenDragon, including
 http://green.lab.llvm.org/green/job/test-suite-verify-machineinstrs-x86_64-O0-g/7891/

4 years ago[llvm-readobj] - Verify the location of program headers better.
Georgii Rymar [Tue, 14 Jul 2020 14:14:06 +0000 (17:14 +0300)]
[llvm-readobj] - Verify the location of program headers better.

This improves condition in the ELFFile::program_headers().
Previously if was possible to read the headers from the wrong place when
the value of e_phoff was so large that computation overflowed.

Differential revision: https://reviews.llvm.org/D83774

4 years ago[Support] Fix Windows directory_iterator_construct out of bounds
Andrew Ng [Mon, 13 Jul 2020 12:36:33 +0000 (13:36 +0100)]
[Support] Fix Windows directory_iterator_construct out of bounds

Fix incorrect use of the size of Path when accessing PathUTF16, as the
UTF-16 path can be shorter. Added unit test for coverage of this test
case.

Thanks to Ding Fei (danix800) for the code fix, see
https://reviews.llvm.org/D83321.

Differential Revision: https://reviews.llvm.org/D83689

4 years ago[clangd] Fix use-after-free in ArgStripper
Sam McCall [Wed, 15 Jul 2020 09:03:11 +0000 (11:03 +0200)]
[clangd] Fix use-after-free in ArgStripper

4 years agoFix typo in identifier in assert.
Tim Northover [Wed, 15 Jul 2020 08:57:53 +0000 (09:57 +0100)]
Fix typo in identifier in assert.

4 years agoAArch64+ARM: make LLVM consider system registers volatile.
Tim Northover [Wed, 15 Jul 2020 08:11:36 +0000 (09:11 +0100)]
AArch64+ARM: make LLVM consider system registers volatile.

Some of the system registers readable on AArch64 and ARM platforms
return different values with each read (for example a timer counter),
these shouldn't be hoisted outside loops or otherwise interfered with,
but the normal @llvm.read_register intrinsic is only considered to read
memory.

This introduces a separate @llvm.read_volatile_register intrinsic and
maps all system-registers on ARM platforms to use it for the
__builtin_arm_rsr calls. Registers declared with asm("r9") or similar
are unaffected.

4 years ago[clang][Driver] Fix tool path priority test failures
David Spickett [Thu, 2 Jul 2020 15:04:06 +0000 (16:04 +0100)]
[clang][Driver] Fix tool path priority test failures

Summary:
Failure type 1:
This test can fail when the path of the build includes the strings
we're checking for. E.g "/gcc" is found in ".../gcc_7.3.0/..."

To correct this look for '"' on the end of all matches. So that we
only match the end of paths printed by clang -###.
(which would be ".../gcc_7.3.0/.../gcc" for the example)

Also look for other gcc names like gcc-x.y.z in the first check.
This confirms that the copy of clang we made is isolated as expected.

Failure type 2:
If you use a triple like "powerpc64le-linux-gnu" clang actually reports
"powerpc64le-unknown-linux-gnu". Then it searches for the
former.

That combined with Mac OS adding a version number to cmake's triple
means we can't trust cmake or clang to give us the one default triple.
To fix the test, write to both names. As they don't overlap with our
fake triple, we're still showing that the lookup works.

Reviewers: MaskRay, stevewan

Reviewed By: stevewan

Subscribers: miyuki, JDevlieghere, steven.zhang, stevewan, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D83055

4 years ago[lldb][formatters] Add support for printing NSConstantDate and fix distantPast value
Raphael Isemann [Tue, 14 Jul 2020 12:26:59 +0000 (14:26 +0200)]
[lldb][formatters] Add support for printing NSConstantDate and fix distantPast value

Summary:

Certain `NSDate` constructors return a special `NSConstantDate` class which
currently ends up being unformatted as it's not in the list of supported classes
for the NSDate formatter. This patch adds that class to the supported class list
so LLDB produces a summary for it.

One of these special constructors is `[NSDate distantPast]` which returns the
date for `0001-01-01 00:00:00 UTC`. LLDB has a special case for formatting this
date but for some reason we did hardcode the wrong summary string in that
special case. Maybe the summary string was correct back when the code was
written but it isn't correct anymore (`distantPast` isn't actually defined to be
a special date but just some 'a guaranteed temporal boundary.' so maybe someone
changed the value in the last 10 years).

If someone else is wondering why we even have this special case for
`distantPast` but not for the future. The reason seems to be that our date
formatting for really old dates is off by 24 hours. So for example, adding one
second to `distantPast` will cause LLDB to print `0000-12-30 00:00:01 UTC`
(which is 24 hours behind the expected result). So to make our code appear to be
correct it seems we just hardcoded the most common NSDate result from that time
span. I'll replace that logic with a generic solution in a probably more
invasive follow up patch.

I also took the freedom to replace the magic value `-63114076800` with some
constant + documentation. I heard there are some people that don't know from the
top of their head that there are 63114076800 seconds between 1. Jan 0001 and 1.
January 2001 in whatever calendar system NSDate is using.

Reviewers: mib, davide

Reviewed By: mib

Subscribers: JDevlieghere

Differential Revision: https://reviews.llvm.org/D83217

4 years ago[RISCV] Add error checking for extensions missing separating underscores
Simon Cook [Wed, 15 Jul 2020 08:23:35 +0000 (09:23 +0100)]
[RISCV] Add error checking for extensions missing separating underscores

Currently if two multi-letter extensions are provided in a -march=
string, the verification code checks the version of the first and
consumes the second, resulting in that part of the architecture
string being ignored. This adds a test that when a version number has
been parsed for an extension, there are no subsequent characters.

Differential Revision: https://reviews.llvm.org/D83819

4 years ago[DAGCombiner] Rebuild (setcc x, y, ==) from (xor (xor x, y), 1)
Roger Ferrer Ibanez [Wed, 7 Aug 2019 15:31:29 +0000 (15:31 +0000)]
[DAGCombiner] Rebuild (setcc x, y, ==) from (xor (xor x, y), 1)

The existing code already considered this case. Unfortunately a typo in
the condition prevents it from triggering. Also the existing code, had
it run, forgot to do the folding.

This fixes PR42876.

Differential Revision: https://reviews.llvm.org/D65802

4 years ago[NFC] Add tests for boolean comparisons
Roger Ferrer Ibanez [Wed, 7 Aug 2019 12:43:07 +0000 (12:43 +0000)]
[NFC] Add tests for boolean comparisons

They currently show that the not equal case may be improved.

See PR42876

Differential Revision: https://reviews.llvm.org/D65801

4 years ago[MLIR][SPIRVToLLVM] SPIRV function fix and nits
George Mitenkov [Wed, 15 Jul 2020 07:02:01 +0000 (10:02 +0300)]
[MLIR][SPIRVToLLVM] SPIRV function fix and nits

This patch addresses the comments from https://reviews.llvm.org/D83030 and
https://reviews.llvm.org/D82639. `this->` is removed when not inside the
template. Also, type conversion for `spv.func` takes `convertRegionTypes()`
in order to apply type conversion on all blocks within the function.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D83786

4 years agoMake linalg::ReshapeOp implement ViewLikeOpInterface
Stephan Herhut [Tue, 14 Jul 2020 14:23:19 +0000 (16:23 +0200)]
Make linalg::ReshapeOp implement ViewLikeOpInterface

Summary: A reshape aliases its input memref, so it acts like a view.

Differential Revision: https://reviews.llvm.org/D83773

4 years ago[IndVarSimplify] Uniformly use emplace_back for DeadInsts, nfc
Chen Zheng [Wed, 15 Jul 2020 06:44:17 +0000 (02:44 -0400)]
[IndVarSimplify] Uniformly use emplace_back for DeadInsts, nfc

4 years ago[OpenMPOpt][NFC] Update checks for parallel_deletion test
Giorgis Georgakoudis [Wed, 15 Jul 2020 06:15:59 +0000 (23:15 -0700)]
[OpenMPOpt][NFC] Update checks for parallel_deletion test

4 years ago[mlir][Vector] Degrade masking information when forwarding linalg.copy to vector...
Nicolas Vasilache [Tue, 14 Jul 2020 15:45:59 +0000 (11:45 -0400)]
[mlir][Vector] Degrade masking information when forwarding linalg.copy to vector.transfer

Summary:
linalg.copy + linalg.fill can be used to create a padded local buffer.
The `masked` attribute is only valid on this padded buffer.
When forwarding to vector.transfer ops, the attribute must be reset
conservatively.

Differential Revision: https://reviews.llvm.org/D83782

4 years ago[OpenMPOpt] Fix preserved analyses return
Giorgis Georgakoudis [Wed, 15 Jul 2020 00:01:15 +0000 (17:01 -0700)]
[OpenMPOpt] Fix preserved analyses return

4 years ago[LLD] [MinGW] Implement the --file-alignment and --section-alignment options
Martin Storsjö [Mon, 13 Jul 2020 21:12:13 +0000 (00:12 +0300)]
[LLD] [MinGW] Implement the --file-alignment and --section-alignment options

Differential Revision: https://reviews.llvm.org/D83720

4 years ago[llvm][NFC] Hide the tensorflow dependency from headers.
Mircea Trofin [Wed, 15 Jul 2020 02:32:37 +0000 (19:32 -0700)]
[llvm][NFC] Hide the tensorflow dependency from headers.

Summary:
This change avoids exposing tensorflow types when including TFUtils.h.
They are just an implementation detail, and don't need to be used
directly when implementing an analysis requiring ML model evaluation.

The TFUtils APIs, while generically typed, are still not exposed unless
the tensorflow C library is present, as they currently have no use
otherwise.

Reviewers: mehdi_amini, davidxl

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D83843

4 years ago[Attributor] Fix build of unittest with DBUILD_SHARED_LIBS=True
Alex Bradbury [Wed, 15 Jul 2020 04:04:09 +0000 (05:04 +0100)]
[Attributor] Fix build of unittest with DBUILD_SHARED_LIBS=True

The dependencies in llvm/unittests/Transforms/IPO/CMakeLists.txt
introduced in revision 0750757e were incomplete, leading to link errors
for a DBUILD_SHARED_LIBS=True build.

4 years ago[OpenMP][FIX] Check only for deterministic part of a generated function name
Johannes Doerfert [Wed, 15 Jul 2020 03:46:38 +0000 (22:46 -0500)]
[OpenMP][FIX] Check only for deterministic part of a generated function name

4 years agoRevert "[Attributor] Track AA dependency using dependency graph"
Luofan Chen [Wed, 15 Jul 2020 03:48:08 +0000 (11:48 +0800)]
Revert "[Attributor] Track AA dependency using dependency graph"

This reverts commit 8df7af560aebce3f3de3541d039e17331c479831.

4 years ago[OpenMP][IRBuilder] Support nested parallel regions
Johannes Doerfert [Sun, 28 Jun 2020 18:41:33 +0000 (13:41 -0500)]
[OpenMP][IRBuilder] Support nested parallel regions

During code generation we might change/add basic blocks so keeping a
list of them is fairly easy to break. Nested parallel regions were
enough. The new scheme does recompute the list of blocks to be outlined
once it is needed.

Reviewed By: anchu-rajendran

Differential Revision: https://reviews.llvm.org/D82722

4 years ago[CallGraph] Update callback call sites in RefreshCallGraph
Johannes Doerfert [Tue, 14 Jul 2020 02:35:03 +0000 (21:35 -0500)]
[CallGraph] Update callback call sites in RefreshCallGraph

Since D82572, we keep "reference" edges for callback call sites. While
not strictly necessary they can improve the traversal order. However, we
did not update them properly in case a pass removed the callback call
site which caused a verification error (PR46687). With this patch we
update these reference edges properly during the invocation of
`CallGraphSCCPass::RefreshCallGraph` in non-checking mode.

Reviewed By: sdmitriev

Differential Revision: https://reviews.llvm.org/D83718

4 years ago[OpenMP] Emit remarks during GPU state machine optimization
Johannes Doerfert [Wed, 15 Jul 2020 00:11:30 +0000 (19:11 -0500)]
[OpenMP] Emit remarks during GPU state machine optimization

Since D83271 we can optimize the GPU state machine to avoid spurious
call edges that increase the register usage of kernels. With this patch
we inform the user why and if this optimization is happening and when it
is not.

Reviewed By: ye-luo

Differential Revision: https://reviews.llvm.org/D83707

4 years agoRevert "[CMake] Simplify CMake handling for zlib"
Petr Hosek [Wed, 15 Jul 2020 02:56:10 +0000 (19:56 -0700)]
Revert "[CMake] Simplify CMake handling for zlib"

This reverts commit 8c1a79dc12f3cc600e16153961cd8cc50ba2c33b because
it fails when zlib isn't installed.

4 years ago[Attributor] Track AA dependency using dependency graph
Luofan Chen [Wed, 15 Jul 2020 02:38:21 +0000 (10:38 +0800)]
[Attributor] Track AA dependency using dependency graph

Summary: This patch added dependency graph to the attributor so that we can dump the dependencies between AAs more easily. We can also apply general graph algorithms to the graph, making it easier for us to create deep wrappers.

Reviewers: jdoerfert, sstefan1, uenoku, homerdin, baziotis

Reviewed By: jdoerfert

Subscribers: jfb, okura, mgrang, kuter, lebedev.ri, hiraditya, uenoku, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D78861

4 years agoRevert "[Attributor] [WIP] Track AA dependency using dependency graph"
Luofan Chen [Wed, 15 Jul 2020 02:33:55 +0000 (10:33 +0800)]
Revert "[Attributor] [WIP] Track AA dependency using dependency graph"

This reverts commit 6b78ed60708b56d85c6d028e9a06ce24ec3c1ae5.

4 years ago[CMake] Simplify CMake handling for zlib
Petr Hosek [Thu, 30 Apr 2020 20:07:13 +0000 (13:07 -0700)]
[CMake] Simplify CMake handling for zlib

Rather than handling zlib handling manually, use find_package from CMake
to find zlib properly. Use this to normalize the LLVM_ENABLE_ZLIB,
HAVE_ZLIB, HAVE_ZLIB_H. Furthermore, require zlib if LLVM_ENABLE_ZLIB is
set to YES, which requires the distributor to explicitly select whether
zlib is enabled or not. This simplifies the CMake handling and usage in
the rest of the tooling.

This is a reland of abb0075 with all followup changes and fixes that
should address issues that were reported in PR44780.

Differential Revision: https://reviews.llvm.org/D79219

4 years ago[Attributor] [WIP] Track AA dependency using dependency graph
Luofan Chen [Wed, 15 Jul 2020 02:19:37 +0000 (10:19 +0800)]
[Attributor] [WIP] Track AA dependency using dependency graph

Summary: This patch added dependency graph to the attributor so that we can dump the dependencies between AAs more easily. We can also apply general graph algorithms to the graph, making it easier for us to create deep wrappers.

Reviewers: jdoerfert, sstefan1, uenoku, homerdin, baziotis

Reviewed By: jdoerfert

Subscribers: jfb, okura, mgrang, kuter, lebedev.ri, hiraditya, uenoku, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D78861

4 years ago[gn build] Port 0750757e80c
LLVM GN Syncbot [Wed, 15 Jul 2020 02:13:43 +0000 (02:13 +0000)]
[gn build] Port 0750757e80c

4 years ago[Attributor] Unittest for Attributor
Luofan Chen [Wed, 15 Jul 2020 02:09:11 +0000 (10:09 +0800)]
[Attributor] Unittest for Attributor

Summary: This patch introduces basic unittest interface for the Attributor and a simple test case for casting.

Reviewers: jdoerfert, sstefan1, uenoku, homerdin, baziotis

Reviewed By: jdoerfert

Subscribers: mgorny, uenoku, kuter, okura, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D83754

4 years ago[AMDGPU] Apply pre-emit s_cbranch_vcc optimation to more patterns
Carl Ritson [Wed, 15 Jul 2020 01:22:42 +0000 (10:22 +0900)]
[AMDGPU] Apply pre-emit s_cbranch_vcc optimation to more patterns

Add handling of s_andn2 and mask of 0.
This eliminates redundant instructions from uniform control flow.

Reviewed By: rampitec

Differential Revision: https://reviews.llvm.org/D83641

4 years ago[Attributor] Create getter function for the ID of the abstract attribute
Luofan Chen [Wed, 15 Jul 2020 01:31:00 +0000 (09:31 +0800)]
[Attributor] Create getter function for the ID of the abstract attribute

Summary: The `getIdAddr()` function returns the address of the ID of the abstract attribute

Reviewers: jdoerfert, sstefan1, uenoku, homerdin, baziotis

Reviewed By: jdoerfert

Subscribers: okura, hiraditya, uenoku, kuter, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D83172

4 years ago[ObjectFilePECOFF] Try to avoid unaligned access.
Davide Italiano [Wed, 15 Jul 2020 01:51:56 +0000 (18:51 -0700)]
[ObjectFilePECOFF] Try to avoid unaligned access.

Fixes an UBSAN error.

4 years ago[IR] Add a few asserts to provide a better failure signature if you try to create...
Craig Topper [Wed, 15 Jul 2020 01:47:14 +0000 (18:47 -0700)]
[IR] Add a few asserts to provide a better failure signature if you try to create a load/store/alloca with no alignment or insertion position

If no alignment is specified we try to find the datalayout by using the insert position to get the module so we can get the datalayout. But if those are null, then we deference a null pointer.

This patch adds asserts to make the failure a little more obvious than just seg faulting.

Differential Revision: https://reviews.llvm.org/D83829

4 years agoMake LazyCallThroughManager destructor virtual rather than arranging for
Richard Smith [Wed, 15 Jul 2020 01:19:46 +0000 (18:19 -0700)]
Make LazyCallThroughManager destructor virtual rather than arranging for
all clients to delete it via the correct dynamic type.

4 years agoFix the skipIfRosetta decorator
Adrian Prantl [Wed, 15 Jul 2020 01:12:19 +0000 (18:12 -0700)]
Fix the skipIfRosetta decorator

the form that takes func as an argument isn't compatible with the
optional bugnumber argument. This means that only correct for to use it is now
@skipIfRosetta(bugnumber='url')

4 years ago[flang] Fix out-of-tree build with missing acc_gen target
Valentin Clement [Wed, 15 Jul 2020 00:50:10 +0000 (20:50 -0400)]
[flang] Fix out-of-tree build with missing acc_gen target

This pacth fix out-of-tree build of Flang after the introduction of acc_gen.

Reviewed By: sscalpone

Differential Revision: https://reviews.llvm.org/D83835

4 years ago[flang][openacc] Semantic checks for OpenACC 3.0 clauses validity
Valentin Clement [Wed, 15 Jul 2020 00:43:40 +0000 (20:43 -0400)]
[flang][openacc] Semantic checks for OpenACC 3.0 clauses validity

Summary: This patch adds semantic checking for the OpenACC 3.0 clauses validity.

Reviewers: sscalpone, tskeith, klausler, ichoyjx, DavidTruby, jdoerfert

Reviewed By: tskeith, klausler

Subscribers: mgorny, llvm-commits

Tags: #llvm, #flang

Differential Revision: https://reviews.llvm.org/D83807

4 years ago[gn build] (manually) merge c1e2f73c
Nico Weber [Wed, 15 Jul 2020 00:43:59 +0000 (20:43 -0400)]
[gn build] (manually) merge c1e2f73c

4 years agoFix undefined behavior due to deleting an object with a non-virtual
Richard Smith [Wed, 15 Jul 2020 00:12:46 +0000 (17:12 -0700)]
Fix undefined behavior due to deleting an object with a non-virtual
destructor via a pointer of the wrong static type.

This caused crashes during deallocation in C++14 builds when using a
deallocator whose sized delete requires the size argument to be correct.

Also make the LazyCallThroughManager destructor protected to catch this
sort of bug in the future.

4 years agoFix test that was accidentally adding the llvm-as binary into an
Richard Smith [Wed, 15 Jul 2020 00:06:27 +0000 (17:06 -0700)]
Fix test that was accidentally adding the llvm-as binary into an
IR archive.

4 years ago[flang] Fix list-directed input (repeated nulls and LOGICAL)
peter klausler [Tue, 14 Jul 2020 19:31:16 +0000 (12:31 -0700)]
[flang] Fix list-directed input (repeated nulls and LOGICAL)

Allow repeated nulls in list-directed input (e.g., "4*,") and
ignore excess characters in list-directed LOGICAL input after the
T or F.

Fixes FCVS test fm923.f.

Reviewed By: sscalpone

Differential Revision: https://reviews.llvm.org/D83810

4 years ago[lldb/Test] Create reproducer dir if necessary
Jonas Devlieghere [Tue, 14 Jul 2020 23:17:10 +0000 (16:17 -0700)]
[lldb/Test] Create reproducer dir if necessary

Create the reproducer directory under the build root if it doesn't
exists. The reproducer will only create the top level directory.

4 years ago[flang] Implement anonymous units in the runtime
peter klausler [Tue, 14 Jul 2020 18:28:03 +0000 (11:28 -0700)]
[flang] Implement anonymous units in the runtime

I/O from/to an unopened unit number needs to open &/or create
a "fort.$UNIT" file.

Fixes FCVS test fm401.f.

Reviewed By: tskeith

Differential Revision: https://reviews.llvm.org/D83809

4 years agoAMDGPU: Add @LINE to assembler error test checks
Matt Arsenault [Tue, 14 Jul 2020 21:06:32 +0000 (17:06 -0400)]
AMDGPU: Add @LINE to assembler error test checks

It was basically impossible to figure out where the failure point was

4 years ago[libFuzzer] Fix compilation error by including missing platform macro definitions.
Dokyung Song [Tue, 14 Jul 2020 22:25:51 +0000 (22:25 +0000)]
[libFuzzer] Fix compilation error by including missing platform macro definitions.

Summary: This patch fixes sanitizer-windows build errors.

Reviewers: morehouse, hctim

Reviewed By: morehouse, hctim

Subscribers: #sanitizers

Tags: #sanitizers

Differential Revision: https://reviews.llvm.org/D83823

4 years ago[flang] Fix a crash when an array constructor contains an unlimited polymorphic value
Pete Steinfeld [Tue, 14 Jul 2020 17:08:34 +0000 (10:08 -0700)]
[flang] Fix a crash when an array constructor contains an unlimited polymorphic value

Summary:
C7113 States that "An ac-value shall not be unlimited polymorphic."  We failed
to detect this situation which resulted in a crash when trying to get the
underlying derived type specification of the unlimited polymorphic value.

I added code to avoid the crash, code to emit an error message, and a test.

Reviewers: klausler, tskeith, DavidTruby

Subscribers: llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D83793

4 years agoRevert "[compiler-rt][Android] Stop using detect_target_arch"
Arthur Eubanks [Tue, 14 Jul 2020 22:04:30 +0000 (15:04 -0700)]
Revert "[compiler-rt][Android] Stop using detect_target_arch"

This reverts commit 9b7e24c2a5b32e25b773bc8b4ca84dbda995d959.

See comments in https://reviews.llvm.org/D82148.

4 years ago[clang] Teach -fembed-bitcode option not to embed W_value Group
Steven Wu [Tue, 14 Jul 2020 21:39:51 +0000 (14:39 -0700)]
[clang] Teach -fembed-bitcode option not to embed W_value Group

Summary:
-fembed-bitcode options doesn't embed warning options since they are
useless to code generation. Make sure it handles the W_value group and
not embed those options in the output.

Reviewers: zixuw, arphaman

Reviewed By: zixuw

Subscribers: jkorous, dexonsmith, ributzka, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D83813

4 years ago[libFuzzer] Separate platform related macros out from FuzzerDefs.h into FuzzerPlatfor...
Dokyung Song [Tue, 14 Jul 2020 21:02:32 +0000 (21:02 +0000)]
[libFuzzer] Separate platform related macros out from FuzzerDefs.h into FuzzerPlatform.h, and adjust includes in other files.

Summary: This patch separates platform related macros in lib/fuzzer/FuzzerDefs.h into lib/fuzzer/FuzzerPlatform.h, and use FuzzerPlatform.h where necessary. This separation helps when compiling libFuzzer's interceptor module (under review); an unnecessary include of standard headers (such as string.h) may produce conflicts/ambiguation with the interceptor's declarations/definitions of library functions, which complicates interceptor implementation.

Reviewers: morehouse, hctim

Reviewed By: morehouse

Subscribers: krytarowski, #sanitizers

Tags: #sanitizers

Differential Revision: https://reviews.llvm.org/D83805

4 years ago[SVE] Make cstfp_pred_ty and cst_pred_ty work with scalable splats
Christopher Tetreault [Tue, 14 Jul 2020 20:36:33 +0000 (13:36 -0700)]
[SVE] Make cstfp_pred_ty and cst_pred_ty work with scalable splats

Reviewers: efriedma, lebedev.ri, fhahn, c-rhodes, david-arm

Reviewed By: efriedma, david-arm

Subscribers: tschuett, rkruppe, psnobl, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D83001

4 years ago[clang-tidy] OptionsView::store specialized on bool
Nathan James [Tue, 14 Jul 2020 21:19:36 +0000 (22:19 +0100)]
[clang-tidy] OptionsView::store specialized on bool

Following on fcf7cc268fe and 672207c319a which granted checks the ability to read boolean configuration arguments as `true` or `false`.
This enables storing the options back to the configuration file using `true` and `false`.
This is in line with how clang-format dumps boolean options in its style config.

Reviewed By: aaron.ballman

Differential Revision: https://reviews.llvm.org/D83053

4 years agodocument -fpch-instantiate-templates in release notes
Luboš Luňák [Sat, 11 Jul 2020 13:07:27 +0000 (15:07 +0200)]
document -fpch-instantiate-templates in release notes

This should have been included in D69585.

Differential Revision: https://reviews.llvm.org/D83622

4 years ago[flang] Refine CR handling
peter klausler [Tue, 14 Jul 2020 17:16:30 +0000 (10:16 -0700)]
[flang] Refine CR handling

We need to retain carriage return characters in source files
that are not parts of multi-byte line endings; they are
significant in CHARACTER literal constants.

Reviewed By: tskeith

Differential Revision: https://reviews.llvm.org/D83808

4 years ago[llvm][NFC] expose LLVM_HAVE_TF_API through llvm-config.h
Mircea Trofin [Tue, 14 Jul 2020 05:30:25 +0000 (22:30 -0700)]
[llvm][NFC] expose LLVM_HAVE_TF_API through llvm-config.h

Summary:
This allows users of the llvm library discover whether llvm was built
with the tensorflow c API dependency, which helps if using the TFUtils
wrapper, for example.

We don't do the same for the LLVM_HAVE_TF_AOT flag, because that does
not expose any API.

Reviewers: mehdi_amini, davidxl

Subscribers: mgorny, aaron.ballman, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D83746

4 years ago[lldb/Test] Skip TestProcessConnect.py on Windows
Jonas Devlieghere [Tue, 14 Jul 2020 21:07:06 +0000 (14:07 -0700)]
[lldb/Test] Skip TestProcessConnect.py on Windows

Remote connections are not supported on Windows.

4 years ago[lldb/Test] Always set the cleanupSubprocesses tear down hook
Jonas Devlieghere [Tue, 14 Jul 2020 19:12:05 +0000 (12:12 -0700)]
[lldb/Test] Always set the cleanupSubprocesses tear down hook

Always clean up subprocesses on tear down instead of relying on the
caller to do so. This is not only less error prone but also means the
tests can be more concise.

Differential revision: https://reviews.llvm.org/D83787

4 years agoFix tests after 16f777f4217cfcdcf6ddce8eb1e3525a65563c43
Tyker [Tue, 14 Jul 2020 20:52:12 +0000 (22:52 +0200)]
Fix tests after 16f777f4217cfcdcf6ddce8eb1e3525a65563c43

4 years ago[libc++] Automatically detect whether RTTI is enabled
Louis Dionne [Tue, 14 Jul 2020 20:28:41 +0000 (16:28 -0400)]
[libc++] Automatically detect whether RTTI is enabled

Instead of detecting it automatically but also allowing for the setting
to be specified explicitly, always detect whether exceptions are enabled
based on whether -fno-rtti (or equivalent) is used. It's less confusing
to have a single way of tweaking that knob.

This change follows the lead of 71d88cebfb42.

4 years ago[MC] Support .reloc sym+constant, *, *
Fangrui Song [Tue, 14 Jul 2020 20:44:00 +0000 (13:44 -0700)]
[MC] Support .reloc sym+constant, *, *

For `.reloc offset, *, *`, currently offset can be a constant or symbol.
This patch makes it support any expression which can be folded to sym+constant.

Reviewed By: stefanp

Differential Revision: https://reviews.llvm.org/D83751

4 years ago[flang] Tuning up binary->decimal conversion
peter klausler [Tue, 14 Jul 2020 19:38:03 +0000 (12:38 -0700)]
[flang] Tuning up binary->decimal conversion

Use short division of big-radix values by powers of two when
converting values with negative unbiased exponents rather than
multiplication by smaller powers of five; this reduces the overall
outer iteration count. This change is a win across the entire range
of inputs.

Reviewed By: tskeith

Differential Revision: https://reviews.llvm.org/D83806

4 years ago[mlir] Add support for parsing optional Attribute values.
River Riddle [Tue, 14 Jul 2020 20:14:14 +0000 (13:14 -0700)]
[mlir] Add support for parsing optional Attribute values.

This adds a `parseOptionalAttribute` method to the OpAsmParser that allows for parsing optional attributes, in a similar fashion to how optional types are parsed. This also enables the use of attribute values as the first element of an assembly format optional group.

Differential Revision: https://reviews.llvm.org/D83712

4 years ago[CallGraph] Ignore callback uses
Giorgis Georgakoudis [Wed, 8 Jul 2020 05:43:24 +0000 (22:43 -0700)]
[CallGraph] Ignore callback uses

Summary:
Ignore callback uses when adding a callback function
in the CallGraph. Callback functions are typically
created when outlining, e.g. for OpenMP, so they have
internal scope and linkage. They should not be added
to the ExternalCallingNode since they are only callable
by the specified caller function at creation time.

A CGSCC pass, such as OpenMPOpt, may need to update
the CallGraph by adding a new outlined callback function.
Without ignoring callback uses, adding breaks CGSCC
pass restrictions and results to a broken CallGraph.

Reviewers: jdoerfert

Subscribers: hiraditya, sstefan1, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D83370

4 years ago[mlir][DialectConversion] Emit an error if an operation marked as erased has live...
River Riddle [Tue, 14 Jul 2020 19:55:58 +0000 (12:55 -0700)]
[mlir][DialectConversion] Emit an error if an operation marked as erased has live users after conversion

Up until now, there has been an implicit agreement that when an operation is marked as
"erased" all uses of that operation's results are guaranteed to be removed during conversion. How this works in practice is that there is either an assert/crash/asan failure/etc. This revision adds support for properly detecting when an erased operation has dangling users, emits and error and fails the conversion.

Differential Revision: https://reviews.llvm.org/D82830

4 years agoCall Frame Information (CFI) Handling for Basic Block Sections
Krzysztof Pszeniczny [Tue, 14 Jul 2020 18:55:41 +0000 (11:55 -0700)]
Call Frame Information (CFI) Handling for Basic Block Sections

This patch handles CFI with basic block sections, which unlike DebugInfo does
not support ranges. The DWARF standard explicitly requires emitting separate
CFI Frame Descriptor Entries for each contiguous fragment of a function. Thus,
the CFI information for all callee-saved registers (possibly including the
frame pointer, if necessary) have to be emitted along with redefining the
Call Frame Address (CFA), viz. where the current frame starts.

CFI directives are emitted in FDE’s in the object file with a low_pc, high_pc
specification. So, a single FDE must point to a contiguous code region unlike
debug info which has the support for ranges. This is what complicates CFI for
basic block sections.

Now, what happens when we start placing individual basic blocks in unique
sections:

* Basic block sections allow the linker to randomly reorder basic blocks in the
address space such that a given basic block can become non-contiguous with the
original function.
* The different basic block sections can no longer share the cfi_startproc and
cfi_endproc directives. So, each basic block section should emit this
independently.
* Each (cfi_startproc, cfi_endproc) directive will result in a new FDE that
caters to that basic block section.
* Now, this basic block section needs to duplicate the information from the
entry block to compute the CFA as it is an independent entity. It cannot refer
to the FDE of the original function and hence must duplicate all the stuff that
is needed to compute the CFA on its own.
* We are working on a de-duplication patch that can share common information in
FDEs in a CIE (Common Information Entry) and we will present this as a follow up
patch. This can significantly reduce the duplication overhead and is
particularly useful when several basic block sections are created.
* The CFI directives are emitted similarly for registers that are pushed onto
the stack, like callee saved registers in the prologue. There are cfi
directives that emit how to retrieve the value of the register at that point
when the push happened. This has to be duplicated too in a basic block that is
floated as a separate section.

Differential Revision: https://reviews.llvm.org/D79978