Adrian Kuegel [Mon, 17 May 2021 11:21:24 +0000 (13:21 +0200)]
[mlir] Add folder for complex.ReOp and complex.ImOp.
Now that complex constants are supported, we can also fold.
Differential Revision: https://reviews.llvm.org/D102609
Jan Svoboda [Mon, 17 May 2021 11:23:05 +0000 (13:23 +0200)]
Revert "[clang][deps] Support inferred modules"
This reverts commit
1d9e8e13
Something is off with handling Windows paths: http://45.33.8.238/win/38908/step_7.txt
Most likely introduced in https://reviews.llvm.org/D102491
Max Kazantsev [Mon, 17 May 2021 11:25:59 +0000 (18:25 +0700)]
[Test] Auto-generate checks in a test (prepring to update)
Anastasia Stulova [Mon, 17 May 2021 10:58:32 +0000 (11:58 +0100)]
[OpenCL] Drop pragma handling for extension types/decls.
Drop non-conformant extension pragma implementation as
it does not properly disable anything and therefore
enabling non-disabled logic has no meaning.
This simplifies clang code and user interface to the extension
functionality. With this patch extension pragma 'begin'/'end'
and 'enable'/'disable' are only accepted for backward
compatibility and no longer have any default behavior.
Differential Revision: https://reviews.llvm.org/D101043
Nemanja Ivanovic [Mon, 17 May 2021 11:05:36 +0000 (06:05 -0500)]
[PowerPC] Do not emit dssall on AIX
This instruction is a nop on all server cores (certainly on all
cores that AIX supports) so it is fine to emit a nop instead of it.
In fact, that is exactly what XL emits. So we emit a nop on AIX
and we leave the codegen as is on other platforms since there may
indeed be cores out there for which this actually does some prefetching.
Jan Svoboda [Mon, 17 May 2021 11:02:25 +0000 (13:02 +0200)]
[clang][deps] Specify Python 3 in shebang
Some build bots don't like plain `python` in the shebang: https://lab.llvm.org/buildbot/#/builders/139/builds/4147.
Nico Weber [Mon, 17 May 2021 10:59:43 +0000 (06:59 -0400)]
[gn build] reformat all gn files
$ git ls-files '*.gn' '*.gni' | xargs llvm/utils/gn/gn.py format
Nico Weber [Sat, 15 May 2021 03:07:09 +0000 (23:07 -0400)]
[gn build] Add build file for msan runtime
Works for the examples on
https://clang.llvm.org/docs/MemorySanitizer.html
Differential Revision: https://reviews.llvm.org/D102554
Tim Northover [Tue, 12 Jan 2021 13:12:40 +0000 (13:12 +0000)]
X86: support Swift Async context
This adds support to the X86 backend for the newly committed swiftasync
function parameter. If such a (pointer) parameter is present it gets stored
into an augmented frame record (populated in IR, but generally containing
enhanced backtrace for coroutines using lots of tail calls back and forth).
The context frame is identical to AArch64 (primarily so that unwinders etc
don't get extra complexity). Specfically, the new frame record is [AsyncCtx,
%rbp, ReturnAddr], and its presence is signalled by bit 60 of the stored %rbp
being set to 1. %rbp still points to the frame pointer in memory for backwards
compatibility (only partial on x86, but OTOH the weird AsyncCtx before the rest
of the record is because of x86).
Tim Northover [Mon, 17 May 2021 10:34:16 +0000 (11:34 +0100)]
AArch64: mark x22 livein if it's an async context that gets stored.
This fixes a crash with expensive checks enabled (the verifier was not happy).
Kadir Cetinkaya [Thu, 13 May 2021 18:56:04 +0000 (20:56 +0200)]
[clangd][QueryDriver] Dont check for existence of driver
Execute implementations already checks for permissions and existence
and returns relevant errors as necessary, so instead of printing our own errors,
we just print theirs.
This also fixes a case in windows where the driver might be missing the `.exe`
suffix. Previously, clangd would reject such a driver because sys::fs::exists is
strict, whereas the underlying Execute implementation would check with `.exe`
suffix too.
Fixes https://github.com/clangd/clangd/issues/93
Differential Revision: https://reviews.llvm.org/D102431
Max Kazantsev [Mon, 17 May 2021 10:13:49 +0000 (17:13 +0700)]
[Test] Fix test to make the transform for which is was added legal
%limit in these tests is supposed to be positive.
Adam Paszke [Mon, 17 May 2021 10:14:02 +0000 (10:14 +0000)]
Add `mlirModuleFromOperation` to C API
At the moment `MlirModule`s can be converted to `MlirOperation`s, but not
the other way around (at least not without going around the C API). This
makes it impossible to e.g. run passes over a `ModuleOp` created through
`mlirOperationCreate`.
Reviewed By: nicolasvasilache, mehdi_amini
Differential Revision: https://reviews.llvm.org/D102497
Simon Pilgrim [Mon, 17 May 2021 10:03:12 +0000 (11:03 +0100)]
[TargetLowering] prepareUREMEqFold/prepareSREMEqFold - account for non legal shift types
Ensure we tell getShiftAmountTy that we're working with pre-legalized types to prevent cases where the (legalized) shift type can no longer handle the (non-legalized) type width.
Fixes https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=34366
Florian Hahn [Mon, 17 May 2021 09:57:59 +0000 (10:57 +0100)]
Revert "tsan: mark sigwait as blocking"
This reverts commit
5dad3d1ba9ad01152be21e94cfbbfb31659ea3e1.
The added test (signal_block2.cpp) does not terminate on some Darwin
configurations and is causing Green Dragon bots to fail. First
failure of the test started in
http://green.lab.llvm.org/green/job/clang-stage1-RA/20767/
Tim Northover [Thu, 19 Nov 2020 12:32:50 +0000 (12:32 +0000)]
IR/AArch64/X86: add "swifttailcc" calling convention.
Swift's new concurrency features are going to require guaranteed tail calls so
that they don't consume excessive amounts of stack space. This would normally
mean "tailcc", but there are also Swift-specific ABI desires that don't
naturally go along with "tailcc" so this adds another calling convention that's
the combination of "swiftcc" and "tailcc".
Support is added for AArch64 and X86 for now.
Michael Spencer [Mon, 17 May 2021 08:41:44 +0000 (10:41 +0200)]
[clang][deps] Support inferred modules
This patch adds support for inferred modules to the dependency scanner.
Effectively a cherry-pick of https://github.com/apple/llvm-project/pull/699 authored by @Bigcheese with libclang and other changes omitted.
Contains following changes:
1. [Clang][ScanDeps] Ignore __inferred_module.map dependency.
* This shows up with inferred modules, but it doesn't exist on disk, so don't report it as a dependency.
2. [Clang][ScanDeps] Use the module map a module was inferred from for inferred modules.
Also includes a smoke test that uses clang-scan-deps output to perform an explicit build. There's no intention to duplicate whatever `test/Modules` contains, just to verify the produced command-line does "work" (with very loose definition of work).
Split from D100934.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D102495
Julian Gross [Fri, 7 May 2021 10:38:12 +0000 (12:38 +0200)]
[WIP][mlir] Resolve memref dependency in canonicalize pass.
Splitting the memref dialect lead to an introduction of several dependencies
to avoid compilation issues. The canonicalize pass also depends on the
memref dialect, but it shouldn't. This patch resolves the dependencies
and the unintuitive includes are removed. However, the dependency moves
to the constructor of the std dialect.
Differential Revision: https://reviews.llvm.org/D102060
Tobias Gysi [Mon, 17 May 2021 08:50:15 +0000 (08:50 +0000)]
[mlir][linalg] Remove template parameter from loop lowering.
Replace the templated linalgLowerOpToLoops method by three specialized methods linalgOpToLoops, LinalgOpToParallelLoops, and linalgOpToAffineLoops.
Differential Revision: https://reviews.llvm.org/D102324
Jacob Bramley [Tue, 27 Apr 2021 09:12:11 +0000 (10:12 +0100)]
[AArch64] Lower fpto*i.sat intrinsics.
AArch64's fctv* instructions implement the saturating behaviour that the
fpto*i.sat intrinsics require, in cases where the destination width
matches the saturation width. Lowering them removes a lot of unnecessary
generated code.
Only scalar lowerings are supported for now.
Differential Revision: https://reviews.llvm.org/D102353
Raphael Isemann [Mon, 17 May 2021 09:02:36 +0000 (11:02 +0200)]
[lldb] Let dotest check for the `cassert` header instead of `algorithm` for libc++ check
This is just a dotest check to see if we can compile a simple program that uses
libc++. Right now we are parsing the rather big `algorithm` header in the test
program, but the test really just checks whether we can find *any* libc++
headers and link against some libc++ SO. Using the much smaller `cassert` header
for checking whether we can find libc++ headers speeds up this check by a bit.
After some incredibly unscientific performance testing this saves a few seconds
when running the test suite on Linux (on macOS we hardcoded that libc++ is
always there, so this check won't be used there and we don't save any time).
Reviewed By: jankratochvil
Differential Revision: https://reviews.llvm.org/D101056
Utkarsh Saxena [Fri, 14 May 2021 18:22:09 +0000 (20:22 +0200)]
[clangd] Set FileSystem for tweaks in Check tool.
Tweaks like DefineOutline depend on FS to be set at `apply()` time.
After https://reviews.llvm.org/D93978, tweaks run from Check tool lost
access to FS. This makes the available to apply() once again.
Differential Revision: https://reviews.llvm.org/D102519
Raphael Isemann [Mon, 17 May 2021 08:32:45 +0000 (10:32 +0200)]
[lldb][NFC] Specify guidelines for API tests
This patch specifies a few guidelines that our API tests should follow.
The motivations for this are twofold:
1. API tests have unexpected pitfalls that especially new contributors run into
when writing tests. To prevent the frustration of letting people figure those
pitfalls out by trial-and-error, let's just document them briefly in one place.
2. It prevents some arguing about what is the right way to write tests. I really
like to have fast and reliable API test suite, but I also don't want to be the
bogeyman that has to insist in every review that the test should be rewritten to
not launch a process for no good reason. It's much easier to just point to a
policy document.
I omitted some guidelines that I think could be controversial (e.g., the whole
"should assert message describe failure or success").
Reviewed By: shafik
Differential Revision: https://reviews.llvm.org/D101153
Michael Spencer [Mon, 17 May 2021 07:40:29 +0000 (09:40 +0200)]
[clang][modules] Build inferred modules
This patch enables explicitly building inferred modules.
Effectively a cherry-pick of https://github.com/apple/llvm-project/pull/699 authored by @Bigcheese with libclang and dependency scanner changes omitted.
Contains the following changes:
1. [Clang] Fix the header paths in clang::Module for inferred modules.
* The UmbrellaAsWritten and NameAsWritten fields in clang::Module are a lie for framework modules. For those they actually are the path to the header or umbrella relative to the clang::Module::Directory.
* The exception to this case is for inferred modules. Here it actually is the name as written, because we print out the module and read it back in when implicitly building modules. This causes a problem when explicitly building an inferred module, as we skip the printing out step.
* In order to fix this issue this patch adds a new field for the path we want to use in getInputBufferForModule. It also makes NameAsWritten actually be the name written in the module map file (or that would be, in the case of an inferred module).
2. [Clang] Allow explicitly building an inferred module.
* Building the actual module still fails, but make sure it fails for the right reason.
Split from D100934.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D102491
Florian Hahn [Mon, 17 May 2021 08:29:49 +0000 (09:29 +0100)]
Revert "[NFC][LSAN] Limit the number of concurrent threads is the test"
This reverts commit
2a73b7bd8cf7620fc0e478ac838b07ee6649dd8a.
This appears to be causing the following failures on GreenDragon:
LeakSanitizer-AddressSanitizer-x86_64 :: TestCases/many_threads_detach.cpp
LeakSanitizer-Standalone-x86_64 :: TestCases/many_threads_detach.cpp
First failure:
http://green.lab.llvm.org/green/job/clang-stage1-RA/20754/
Still failing in latest build:
http://green.lab.llvm.org/green/job/clang-stage1-RA/20928/
Fraser Cormack [Thu, 13 May 2021 11:18:36 +0000 (12:18 +0100)]
[DAGCombiner] Relax an assertion to an early return
The select-of-constants transform was asserting that its constant vector
inputs did not implicitly truncate their input without that as an
explicit precondition to the function. This patch relaxes that assertion
into an early return to skip the optimization.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D102393
Kristina Bessonova [Sat, 15 May 2021 11:15:56 +0000 (13:15 +0200)]
[libcxx] NFC. Add missed articles to _LIPCPP_ASSERT messages in <list>
OCHyams [Mon, 17 May 2021 08:07:46 +0000 (09:07 +0100)]
[dexter] Remove requirement for a condition in DexLimitSteps
Currently the DexLimitSteps command requires at least one condition. This patch
lets users elide the condition to specify that the breakpoint range should
always be activated when the leading line is stepped on. This patch also
updates the terminology used in the `ConditionalController` class from the
terms 'conditional' and 'unconditional' to 'leading' and 'trailing' when
referring to the breakpoints in the DexLimitSteps range because the leading
breakpoint can now be unconditional.
Reviewed By: chrisjackson
Differential Revision: https://reviews.llvm.org/D101438
OCHyams [Mon, 17 May 2021 07:57:43 +0000 (08:57 +0100)]
[dexter] Remove unnecessary double check on conditional breakpoints
Remove the `ConditionalController._conditional_met` method. This was missed in
the recent ConditionalController refactor (D98699). We don't need to check that
the conditions for a conditional breakpoint have been met because
`DebuggerBase.get_triggered_breakpoint_ids` returns the set of ids for
breakpoints which have been triggered.
To get the "triggered breakpoints" from lldb we use `GetStopReasonDataCount`
and `GetStopReasonDataAtIndex`. It seems that these functions count all
breakpoints associated with the location which lldb has stopped at, regardless
of their condition. i.e. Even if we have two breakpoints at the same source
location that have mutually exclusive conditions, both will be found this way
when either condition is true. To get around this, we store a map of breakpoint
{id: condition} `_breakpoint_conditions` and evaluate the conditions of the
triggered breakpoints to filter the set down to those which are unconditional
or have a condition which evaluates to true.
Essentially we are just moving the condition double check from a general
debugger controller into the lldb specific wrapper. This tidy up will help make
upcoming patches simpler.
Reviewed By: chrisjackson
Differential Revision: https://reviews.llvm.org/D101431
Mateusz Mikuła [Mon, 17 May 2021 07:39:59 +0000 (10:39 +0300)]
[LLD][MinGW] Add --fatal-warnings and --no-fatal-warnings flags
Differential Revision: https://reviews.llvm.org/D102514
Jan Svoboda [Mon, 17 May 2021 07:29:42 +0000 (09:29 +0200)]
[clang] NFC: Remove CompilerInstance::getInvocationPtr
This member function was introduced in
0a92e09c ([clang][deps] Generate the full command-line for modules) in order to keep the CompilerInvocation object alive after CompilerInstance goes out of scope. However,
d3fb4b90 ([clang][deps] NFC: Report modules' context hash) removes that use-case, making this function dead.
Chen Zheng [Mon, 17 May 2021 07:12:16 +0000 (03:12 -0400)]
[PowerPC] add a testcase for reverse memory op; nfc
Jan Svoboda [Mon, 17 May 2021 07:16:25 +0000 (09:16 +0200)]
[clang][deps] NFC: Report modules' context hash
This patch eagerly constructs and modifies CompilerInvocation of modular dependencies in order to report the correct context hash instead of the hash of the original translation unit.
No functionality change here, since we currently don't modify CompilerInvocation in a way that affects the context hash.
Depends on D102473.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D102482
Jan Svoboda [Mon, 17 May 2021 06:50:49 +0000 (08:50 +0200)]
[clang][deps] NFC: Stop assuming the TU's context hash
The context hash of modular dependencies can be different from the context hash of the original translation unit if we modify their `CompilerInvocation`s.
Stop assuming the TU's context hash everywhere.
No functionality change here, since we're still currently using the unmodified TU CompilerInvocation to compute the context hash.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D102473
Adrian Kuegel [Fri, 14 May 2021 11:00:38 +0000 (13:00 +0200)]
Add support for complex constants to MLIR core.
BEGIN_PUBLIC
Add support for complex constants to MLIR core.
END_PUBLIC
Differential Revision: https://reviews.llvm.org/D101908
Matthias Springer [Mon, 17 May 2021 06:30:07 +0000 (15:30 +0900)]
[mlir] Lower permutation maps on TransferWriteOps
Add TransferWritePermutationLowering, which replaces permutation maps of TransferWriteOps with vector.transpose.
Differential Revision: https://reviews.llvm.org/D102548
Matthias Springer [Mon, 17 May 2021 06:26:26 +0000 (15:26 +0900)]
[mlir] Fix in_bounds attr handling in TransferReadPermutationLowering
The in_bounds attribute should also be transposed.
Differential Revision: https://reviews.llvm.org/D102572
Hongtao Yu [Thu, 13 May 2021 18:06:44 +0000 (11:06 -0700)]
[CSSPGO] Update pseudo probe distribution factor based on inline context.
With prelink inlining, pseudo probes with same ID can come from different inline contexts. Such probes should not share samples and their factors should be fixed up separately.
I'm seeing 0.3% speedup for SPEC2017 overall. Benchmark 631.deepsjeng_s benefits the most, about 4%.
Reviewed By: wenlei, wmi
Differential Revision: https://reviews.llvm.org/D102429
Arthur Eubanks [Mon, 17 May 2021 05:02:10 +0000 (22:02 -0700)]
Revert "[TargetLowering] Only inspect attributes in the arguments for ArgListEntry"
This reverts commit
16748bd2fb1fe10d7d097961f1988327338f3f9f.
Causes https://crbug.com/1209013
Arthur Eubanks [Mon, 17 May 2021 05:00:54 +0000 (22:00 -0700)]
Revert "[NFC] Use ArgListEntry indirect types more in ISel lowering"
This reverts commit
85af8a8c1b574faa0d5d57d189ae051debdfada8.
Pan, Tao [Mon, 17 May 2021 03:22:47 +0000 (11:22 +0800)]
[SelectionDAG] Make fast and linearize visible by clang -pre-RA-sched
ScheduleDAGFast.cpp is compiled to object file, but the ScheduleDAGFast
object file isn't linked into clang executable file as no symbol is
referred by outside. Add calling to createXxx of ScheduleDAGFast.cpp,
then the ScheduleDAGFast object file will be linked into clang
executable file. The static RegisterScheduler will register scheduler
fast and linearize at clang boot time.
Reviewed By: pengfei
Differential Revision: https://reviews.llvm.org/D101601
Ben Shi [Mon, 17 May 2021 02:59:52 +0000 (10:59 +0800)]
[RISCV] Optimize or/xor with immediate in the zbs extension
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D102398
Vitaly Buka [Mon, 17 May 2021 01:59:19 +0000 (18:59 -0700)]
[NFC][sanitizer] Fix 'macro redefined' warning in test
Ben Shi [Mon, 17 May 2021 01:47:23 +0000 (09:47 +0800)]
[RISCV][test] Add new tests of or/xor in the zbs extension
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D102396
Harald van Dijk [Sun, 16 May 2021 23:13:00 +0000 (00:13 +0100)]
[lld][X86] Restore gotEntrySize.
D62727 removed GotEntrySize and GotPltEntrySize with a comment that they
are always equal to wordsize(), but that is not entirely true: X32 has a
word size of 4, but needs 8-byte GOT entries. This restores gotEntrySize
for both, adjusted for current naming conventions, but defaults it to
config->wordsize to keep things simple for architectures other than
x86_64.
This partially reverts D62727.
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D102509
David Blaikie [Sun, 16 May 2021 22:58:40 +0000 (15:58 -0700)]
Fix some deprecated copy ops in google mock
Craig Topper [Sun, 16 May 2021 19:17:47 +0000 (12:17 -0700)]
[RISCV] Replace AddiPair ComplexPattern with a PatLeaf. NFC
The ComplexPattern is looking for an immediate in a certain range
that has a single use. This can be handled with a PatLeaf since
we aren't matching multiple patterns or checking any complicated
relationships between nodes.
This shrinks the isel table a little bit since tablegen no longer
has to generate patterns with commuted operands. With the PatLeaf,
tablegen can see we're matching an immediate which should always
be on the right hand side of add.
Reviewed By: benshi001
Differential Revision: https://reviews.llvm.org/D102510
Fangrui Song [Sun, 16 May 2021 18:17:09 +0000 (11:17 -0700)]
[test] Improve CodeGen/*/semantic-interposition-asm.ll
Alessandro Decina [Sun, 16 May 2021 16:32:36 +0000 (09:32 -0700)]
[BPF] add support for 32 bit registers in inline asm
Add "w" constraint type which allows selecting 32 bit registers.
32 bit registers were added in https://reviews.llvm.org/rGca31c3bb3ff149850b664838fbbc7d40ce571879.
Differential Revision: https://reviews.llvm.org/D102118
Florian Hahn [Sun, 16 May 2021 11:03:36 +0000 (12:03 +0100)]
Recommit "[Clang,Driver] Add -fveclib=Darwin_libsystem_m support."
Recommit D102489, with the test case requiring the AArch64 backend.
This reverts the revert
59b419adc6e608db8d7c31efcc37f34c0b57b7d0.
Danila Kutenin [Fri, 14 May 2021 23:56:48 +0000 (16:56 -0700)]
[clang] Fix ternary operator in the second for loop statement
Fix ternary operator in for loop argument, it was by mistake not set as CanBeForRangeDecl and led to incorrect codegen. It fixes https://bugs.llvm.org/show_bug.cgi?id=50038. I don't have commit rights. Danila Kutenin. kutdanila@yandex.ru
Reviewed By: rsmith
Differential Revision: https://reviews.llvm.org/D102502
Lang Hames [Sun, 16 May 2021 17:09:35 +0000 (10:09 -0700)]
[JITLink] Fix symbol comparator in LinkGraph::dump.
The existing implementation did not provide a strict weak ordering.
David Green [Sun, 16 May 2021 16:54:06 +0000 (17:54 +0100)]
[CPG][ARM] Optimize towards branch on zero in codegenprepare
This adds a simple fold into codegenprepare that converts comparison of
branches towards comparison with zero if possible. For example:
%c = icmp ult %x, 8
br %c, bla, blb
%tc = lshr %x, 3
becomes
%tc = lshr %x, 3
%c = icmp eq %tc, 0
br %c, bla, blb
As a first order approximation, this can reduce the number of
instructions needed to perform the branch as the shift is (often) needed
anyway. At the moment this does not effect very much, as llvm tends to
prefer the opposite form. But it can protect against regressions from
commits like rG9423f78240a2.
Simple cases of Add and Sub are added along with Shift, equally as the
comparison to zero can often be folded with cpsr flags.
Differential Revision: https://reviews.llvm.org/D101778
Philip Reames [Sun, 16 May 2021 16:29:25 +0000 (09:29 -0700)]
Revert "Do actual DCE in LoopUnroll (try 2)"
This reverts commit
653fa0b46ae34c06495b542414b704b30381cd02.
Reported to trigger pr50354. Reverting until investigated.
David Green [Sun, 16 May 2021 16:22:52 +0000 (17:22 +0100)]
[ARM] Extra branch on zero tests. NFC
Kai Luo [Sun, 16 May 2021 14:34:09 +0000 (22:34 +0800)]
[Utils] Fix indentation error in utils/wciia.py
Running this script gives
```
"llvm-project/llvm/./utils/wciia.py", line 56
if word == "N:":
TabError: inconsistent use of tabs and spaces in indentation
```
Under emacs' whitespace-mode, it shows
```
for·line·in·code_owners_file:$
····for·word·in·line.split():$
» if·word·==·"N:":$
» » name·=·line[2:].strip()$
» » if·code_owner:$
» » » process_code_owner(code_owner)$
» » » code_owner·=·{}$
```
I use `yapf` to format this script directly and it's running correctly.
Nikita Popov [Sun, 16 May 2021 14:04:10 +0000 (16:04 +0200)]
[CaptureTracking] Simplify reachability check (NFCI)
This code was re-implementing the same-BB case of
isPotentiallyReachable(). Historically, this was done because
CaptureTracking used additional caching for local dominance
queries. Now that it is no longer needed, the code is effectively
the same as isPotentiallyReachable().
The only difference are extra checks for invoke/phis. These are
misleading checks related to dominance in the value availability
sense that are not relevant for control reachability. The invoke
check was correct but redundant in that invokes are always
terminators, so `I` could never come before the invoke. The phi
check is a matter of interpretation (should an earlier phi node be
considered reachable from a later phi node in the same block?)
but ultimately doesn't matter because phis don't capture anyway.
Nikita Popov [Sun, 16 May 2021 08:40:02 +0000 (10:40 +0200)]
Reapply [CaptureTracking] Do not check domination
Reapply after adjusting the synchronized.m test case, where the
TODO is now resolved. The pointer is only captured on the exception
handling path.
-----
For the CapturesBefore tracker, it is sufficient to check that
I can not reach BeforeHere. This does not necessarily require
that BeforeHere dominates I, it can also occur if the capture
happens on an entirely disjoint path.
This change was previously accepted in D90688, but had to be
reverted due to large compile-time impact in some cases: It
increases the number of reachability queries that are performed.
After recent changes, the compile-time impact is largely mitigated,
so I'm reapplying this patch. The remaining compile-time impact
is largely proportional to changes in code-size.
Florian Hahn [Sun, 16 May 2021 13:11:09 +0000 (14:11 +0100)]
[Matrix] Fix some newpm check lines, which fail on some bots. (2)
Simon Pilgrim [Sun, 16 May 2021 09:53:04 +0000 (10:53 +0100)]
[X86][SSE] Pull out combineToHorizontalAddSub helper from inside (F)ADD/SUB combines (REAPPLIED). NFCI.
The intention is to be able to run this from additional locations (such as shuffle combining) in the future.
Reapplies rGb95a103808ac (after reversion at rGc012a388a15b), with SSE3/SSSE3 typo fix, test added at rG0afb10de1449.
Florian Hahn [Sun, 16 May 2021 12:47:40 +0000 (13:47 +0100)]
[Matrix] Fix some newpm check lines, which fail on some bots.
Florian Hahn [Sun, 16 May 2021 12:08:51 +0000 (13:08 +0100)]
[Matrix] Extend test to also check new PM pipeline.
Florian Hahn [Sun, 16 May 2021 10:45:30 +0000 (11:45 +0100)]
[PhaseOrdering] Add test for insert/extract code for matrixes.
Florian Hahn [Sun, 16 May 2021 10:12:55 +0000 (11:12 +0100)]
[VectorCombine] Add tests with and & urem guaranteeing idx is valid.
Roman Lebedev [Sun, 16 May 2021 10:58:07 +0000 (13:58 +0300)]
[NFC][X86][Costmodel] Add tests for load/store with i1 element type
Nikita Popov [Sun, 16 May 2021 09:04:45 +0000 (11:04 +0200)]
Revert "[CaptureTracking] Do not check domination"
This reverts commit
6b8b43e7af3074124e3c9e429e1fb08165799be4.
This causes clang test to fail (CodeGenObjC/synchronized.m).
Revert until I can figure out whether that's an expected change.
Nikita Popov [Sun, 16 May 2021 08:40:02 +0000 (10:40 +0200)]
[CaptureTracking] Do not check domination
For the CapturesBefore tracker, it is sufficient to check that
I can not reach BeforeHere. This does not necessarily require
that BeforeHere dominates I, it can also occur if the capture
happens on an entirely disjoint path.
This change was previously accepted in D90688, but had to be
reverted due to large compile-time impact in some cases: It
increases the number of reachability queries that are performed.
After recent changes, the compile-time impact is largely mitigated,
so I'm reapplying this patch. The remaining compile-time impact
is largely proportional to changes in code-size.
Nikita Popov [Sun, 16 May 2021 08:35:30 +0000 (10:35 +0200)]
[MemCpyOpt] Add test for unreachable capture (NFC)
This is based on the test from D90688, without the argmemonly
attribute. The argmemonly attribute would guaranteed no modref
by itself and the question of captures would not arise in the
first place.
Uday Bondhugula [Wed, 5 May 2021 02:44:31 +0000 (08:14 +0530)]
[MLIR][PYTHON] Provide opt level for ExecutionEngine Python binding
Provide an option to specify optimization level when creating an
ExecutionEngine via the MLIR JIT Python binding. Not only is the
specified optimization level used for code generation, but all LLVM
optimization passes at the optimization level are also run prior to
machine code generation (akin to the mlir-cpu-runner tool).
Default opt level continues to remain at level two (-O2).
Contributions in part from Prashant Kumar <prashantk@polymagelabs.com>
as well.
Differential Revision: https://reviews.llvm.org/D102551
Saurabh Jha [Sun, 9 May 2021 07:17:10 +0000 (08:17 +0100)]
[Matrix] Implement static cast for matrix types
This patch implements static casts for matrix types. This patch finishes all the work needed for https://bugs.llvm.org/show_bug.cgi?id=47141
Reviewed By: fhahn
Differential Revision: https://reviews.llvm.org/D102125
Uday Bondhugula [Sun, 16 May 2021 06:55:25 +0000 (12:25 +0530)]
[MLIR][NFC] Fix clang-tidy warnings in IntegerSet.h
Fix clang-tidy warnings and some comments in IntegerSet.h. NFC.
Differential Revision: https://reviews.llvm.org/D102387
Kai Luo [Sun, 16 May 2021 05:03:10 +0000 (05:03 +0000)]
[AIX][compiler-rt] Build and install standalone libatomic
On AIX, we have to ship `libatomic.a` for compatibility. First, a new `clang_rt.atomic` is added. Second, use added cmake modules for AIX, we are able to build a compatible libatomic.a for AIX. The second step can't be perfectly implemented with cmake now since AIX's archive approach is kinda unique, i.e., archiving shared libraries into a static archive file.
Reviewed By: jsji
Differential Revision: https://reviews.llvm.org/D102155
Jinsong Ji [Sun, 16 May 2021 03:04:44 +0000 (03:04 +0000)]
[AIX][AsmPrinter] Print Global Variable in comments
The default AsmPrinter print GV in comments,
AIX should do so too.
This also fix LLVM :: CodeGen/Generic/inline-asm-mem-clobber.ll.
Reviewed By: hubert.reinterpretcast
Differential Revision: https://reviews.llvm.org/D102534
Hubert Tong [Sun, 16 May 2021 02:38:42 +0000 (22:38 -0400)]
[AIX][clang-repl][test] Mark unsupported pending XCOFF64 integrated-as
This patch replaces the `powerpc64` token with the `system-aix` one in
the UNSUPPORTED line of a test. The `powerpc64` token was originally
added temporarily in
71a0609a2b53.
If AIX uses integrated-as by default and it works both for 32-bit and
64-bit objects, then the issues encountered so far (see comments in
D96033) would be mostly solved.
As it is, marking the test as expected-to-fail (as opposed to
unsupported) on AIX might cause more trouble in the form of 32-bit
versus 64-bit differences. I am not aware of other situations where LIT
tests are dependent on whether the LLVM build is 64-bit or 32-bit.
Reviewed By: jsji
Differential Revision: https://reviews.llvm.org/D102560
Tomasz Miąsko [Sat, 15 May 2021 21:33:24 +0000 (23:33 +0200)]
[Demangle][Rust] Parse trait definitions
Part of https://reviews.llvm.org/D102549
Tomasz Miąsko [Sat, 15 May 2021 21:25:24 +0000 (23:25 +0200)]
[Demangle][Rust] Parse trait implementations
Part of https://reviews.llvm.org/D102549
Tomasz Miąsko [Sat, 15 May 2021 21:24:16 +0000 (23:24 +0200)]
[Demangle][Rust] Parse inherent implementations
Part of https://reviews.llvm.org/D102549
Kuter Dinel [Sat, 15 May 2021 16:28:50 +0000 (16:28 +0000)]
[Attributor] Call site specific AAValueSimplification and AAIsDead.
This patch makes it possible to do call site specific deductions
for AAValueSimplification and AAIsDead.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D84722
Simon Pilgrim [Sat, 15 May 2021 21:04:49 +0000 (22:04 +0100)]
[X86] Remove unused check-prefixes
Nikita Popov [Sat, 15 May 2021 17:05:08 +0000 (19:05 +0200)]
[CaptureTracking] Only check reachability for capture candidates
Reachability queries are very expensive, and currently performed
for each instruction we look at, even though most of them will
not lead to a capture and are thus ultimately irrelevant. It is
more efficient to walk a few unnecessary instructions than to
perform unnecessary reachability queries.
Theoretically, this may produce worse results, because the additional
instructions considered may cause us to hit the use count limit
earlier. In practice, this does not appear to be a problem, e.g.
on test-suite O3 we report only one more captured-before with this
change, with no resulting codegen differences.
This makes PointerMayBeCapturedBefore() significantly cheaper in
practice, hopefully allowing it to be used in more places.
Simon Pilgrim [Sat, 15 May 2021 20:25:34 +0000 (21:25 +0100)]
[X86][SSE] Add SSE3 coverage to PHADD/SUB generation tests
This would have caught the regression identified in rGc012a388a15b.
Alex Orlov [Sat, 15 May 2021 19:01:12 +0000 (23:01 +0400)]
NFC. Refactored DIPrinter for support embedded source.
This patch introduces source loading and pruning functions.
It will allow to use the DWARF embedded source and use the same code for JSON printout.
No functional changes.
Reviewed By: dblaikie
Differential Revision: https://reviews.llvm.org/D102539
Dan Liew [Sat, 15 May 2021 18:09:34 +0000 (11:09 -0700)]
[Compiler-rt] Downgrade another fatal error to warning
https://reviews.llvm.org/D101681 landed a change to check the testing
configuration which relies on using the `-print-runtime-dir` flag of
clang to determine where the runtime testing library is.
The patch treated not being able to find the path reported by clang
as an error. Unfortunately this seems to break the
`llvm-clang-win-x-aarch64` bot. Either the bot is misconfigured or
clang is reporting a bogus path.
To temporarily unbreak the bot downgrade the fatal error to a warning.
While we're here also print information about the command used to
determine the path to aid debugging.
Dan Liew [Sat, 15 May 2021 17:52:10 +0000 (10:52 -0700)]
[Compiler-rt] Downgrade fatal error about unsupported test configuration
to a warning.
https://reviews.llvm.org/D101681 introduced a check to make sure the
compiler and compiler-rt were using the same library path when
`COMPILER_RT_TEST_STANDALONE_BUILD_LIBS=ON`, i.e. the developer's
intention is to test the just built libs rather that shipped with the
compiler used for testing.
It seems this broken some bots that are likely misconfigured.
So to unbreak them, for now let's make this a warning so the bot
owners can investigate without breaking their builds.
Pengxuan Zheng [Wed, 14 Apr 2021 20:54:47 +0000 (13:54 -0700)]
Support GCC's -fstack-usage flag
This patch adds support for GCC's -fstack-usage flag. With this flag, a stack
usage file (i.e., .su file) is generated for each input source file. The format
of the stack usage file is also similar to what is used by GCC. For each
function defined in the source file, a line with the following information is
produced in the .su file.
<source_file>:<line_number>:<function_name> <size_in_byte> <static/dynamic>
"Static" means that the function's frame size is static and the size info is an
accurate reflection of the frame size. While "dynamic" means the function's
frame size can only be determined at run-time because the function manipulates
the stack dynamically (e.g., due to variable size objects). The size info only
reflects the size of the fixed size frame objects in this case and therefore is
not a reliable measure of the total frame size.
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D100509
Nico Weber [Sat, 15 May 2021 16:18:35 +0000 (12:18 -0400)]
Revert "[X86][SSE] Pull out combineToHorizontalAddSub helper from inside (F)ADD/SUB combines. NFCI."
This reverts commit
b95a103808acfd8f33290d3e80e28af434454b28.
Makes clang assert very early in a Chromium build. See
https://bugs.chromium.org/p/chromium/issues/detail?id=1209490#c1
for a standalone repro.
David Green [Sat, 15 May 2021 15:48:58 +0000 (16:48 +0100)]
[ARM] Add an extra memset test showing reverted WLSTP loops. NFC
Simon Pilgrim [Sat, 15 May 2021 15:28:20 +0000 (16:28 +0100)]
[X86] X86OptimizeLEAPass::replaceDebugValue - take a copy of the DebugLoc not a reference as it may be deleted.
Fixes msan warning due to rG9ca2c50b3601
Simon Pilgrim [Sat, 15 May 2021 15:13:34 +0000 (16:13 +0100)]
[X86] X86CmovConverterPass::convertCmovInstsToBranches - take a copy of the DebugLoc not a reference as it may be deleted.
Fixes msan warning due to rG9ca2c50b3601
Simon Pilgrim [Sat, 15 May 2021 14:49:14 +0000 (15:49 +0100)]
Revert rG632668c1c0e7dcf97154d2e377491cdc8cb6963c : "[TableGen] TreePatternNode::isIsomorphicTo - early out for matching leafs. NFCI."
Revert premature (and very broken....) experimental commit.
Simon Pilgrim [Sat, 15 May 2021 14:34:43 +0000 (15:34 +0100)]
[TableGen] TreePatternNode::isIsomorphicTo - early out for matching leafs. NFCI.
If the leafs are the same then no need to perform DefInit matching.
Simon Pilgrim [Sat, 15 May 2021 14:13:28 +0000 (15:13 +0100)]
X86SpeculativeLoadHardeningPass::hardenValueInRegister - assert that we have a i8/i16/i32/i64 sized register. NFCI.
Silence static analyzer warning for out-of-range access to the SubRegImms[] array.
Simon Pilgrim [Sat, 15 May 2021 14:02:52 +0000 (15:02 +0100)]
SampleProfileLoader::inlineHotFunctionsWithPriority - Fix uninitialized variable warning. NFCI.
findIndirectCallFunctionSamples will leave Sum uninitialized if it returns an empty vector, we don't really use Sum in this case (but we do make a copy that isn't used either) - so ensure we initialize the value to zero to at least silence the static analysis warning.
Simon Pilgrim [Sat, 15 May 2021 13:51:37 +0000 (14:51 +0100)]
IfConverter::MeetIfcvtSizeLimit - Fix uninitialized variable warnings. NFCI.
Ensure the duplication instruction counts are initialized to zero (even though they aren't used) to silence static analysis warnings.
Nikita Popov [Sat, 15 May 2021 11:05:18 +0000 (13:05 +0200)]
[CFG] Move reachable from entry checks into basic block variant
These checks are not specific to the instruction based variant of
isPotentiallyReachable(), they are equally valid for the basic
block based variant. Move them there, to make sure that switching
between the instruction and basic block variants cannot introduce
regressions.
Simon Pilgrim [Sat, 15 May 2021 13:25:48 +0000 (14:25 +0100)]
[X86][Atom] Fix vector integer multiplication resource/throughputs
Match whats documented in the Intel AOM (and Agner/instlatx64 agree) - vector integer multiplies are pipelined - all Port0, throughput = 2 @ 128bits, 1 @ 64bits.
Noticed while checking reduction costs - now that we can use in-order models in llvm-mca, the atom model is the "worst case scenario" we have in x86.
Simon Pilgrim [Sat, 15 May 2021 12:58:42 +0000 (13:58 +0100)]
[Local] collectBitParts - bail out if we find more than one root input value.
All the uses that we have for collectBitParts revolve around us matching down to an operation with a single root value - I don't think we're intending to change that (and a lot of collectBitParts assumes it).
The binops cases (OR/FSHL/FSHR) already check if the providers are the same, but that would still mean we waste time collecting through unaryops before getting to them.
Simon Pilgrim [Sat, 15 May 2021 12:39:09 +0000 (13:39 +0100)]
[InstCombine] InstCombinerImpl::visitOr - enable bitreverse matching
Currently we only match bswap intrinsics from or(shl(),lshr()) style patterns when we could often match bitreverse intrinsics almost as cheaply.
Differential Revision: https://reviews.llvm.org/D90170
Simon Pilgrim [Sat, 15 May 2021 12:23:28 +0000 (13:23 +0100)]
[X86] Try to pass DebugLoc by const-ref to avoid costly TrackingMDNodeRef copies (REAPPLIED). NFCI.
Reapply rG5ed56a821c06 (after reverted by rG7aa89c4a22fd) - don't take reference from struct that will be erased in X86FrameLowering::eliminateCallFramePseudoInstr
Simon Pilgrim [Fri, 14 May 2021 18:09:33 +0000 (19:09 +0100)]
[Local] collectBitParts - early-out from binops. NFCI.
Minor speedup by not bothering to attempt to collect the second operand's bit parts if we already know its failed in the first operand.