platform/upstream/llvm.git
3 years ago[NFC][sanitizer] Add a few consts
Vitaly Buka [Mon, 11 Oct 2021 05:47:02 +0000 (22:47 -0700)]
[NFC][sanitizer] Add a few consts

3 years ago[NFC][sanitizer] Clang-format sanitizer_flat_map.h
Vitaly Buka [Mon, 11 Oct 2021 04:57:12 +0000 (21:57 -0700)]
[NFC][sanitizer] Clang-format sanitizer_flat_map.h

3 years ago[NFC][sanitizer] Add constexpr to FlatMap::size
Vitaly Buka [Sun, 10 Oct 2021 19:45:47 +0000 (12:45 -0700)]
[NFC][sanitizer] Add constexpr to FlatMap::size

3 years ago[NFC][sanitizer] Rename ByteMap to Map
Vitaly Buka [Sun, 10 Oct 2021 19:37:31 +0000 (12:37 -0700)]
[NFC][sanitizer] Rename ByteMap to Map

3 years ago[NFC] Allow to include sanitizer_allocator_bytemap.h
Vitaly Buka [Sun, 10 Oct 2021 19:33:49 +0000 (12:33 -0700)]
[NFC] Allow to include sanitizer_allocator_bytemap.h

3 years ago[MLIR] Fix affine loop unroll corner case for full unroll
Uday Bondhugula [Sat, 2 Oct 2021 06:51:31 +0000 (12:21 +0530)]
[MLIR] Fix affine loop unroll corner case for full unroll

Fix affine loop unroll for zero trip count loops. Add missing check.

Differential Revision: https://reviews.llvm.org/D111375

3 years ago[ORC] Add TaskDispatcher::shutdown calls to TaskDispatchTest.cpp unit tests.
Lang Hames [Mon, 11 Oct 2021 04:09:29 +0000 (21:09 -0700)]
[ORC] Add TaskDispatcher::shutdown calls to TaskDispatchTest.cpp unit tests.

These calls were left out of 4d7cea3d2e8. In the InPlaceDispatcher test case
the operation is a no-op, but it's good form to include it. In the
DynamicThreadPoolTaskDispatcher test the shutdown call is required to ensure
that we don't exit the test (and tear down the dispatcher) before the thread
running the dispatch has completed.

3 years ago[ORC] Add optional RunPolicy to ExecutorProcessControl::callWrapperAsync.
Lang Hames [Sun, 10 Oct 2021 22:49:08 +0000 (15:49 -0700)]
[ORC] Add optional RunPolicy to ExecutorProcessControl::callWrapperAsync.

The callWrapperAsync and callSPSWrapperAsync methods take a handler object
that is run on the return value of the call when it is ready. The new RunPolicy
parameters allow clients to control how these handlers are run. If no policy is
specified then the handler will be packaged as a GenericNamedTask and dispatched
using the ExecutorProcessControl's TaskDispatch member. Callers can use the
ExecutorProcessControl::RunInPlace policy to cause the handler to be run
directly instead, which may be preferrable for simple handlers, or they can
write their own policy object (e.g. to dispatch as some other kind of Task,
rather than GenericNamedTask).

3 years ago[examples] Fix LLJITWithRemoteDebugging example after f3411616896.
Lang Hames [Mon, 11 Oct 2021 03:25:44 +0000 (20:25 -0700)]
[examples] Fix LLJITWithRemoteDebugging example after f3411616896.

3 years ago[XCOFF] Improve error message context.
Esme-Yi [Mon, 11 Oct 2021 02:52:20 +0000 (02:52 +0000)]
[XCOFF] Improve error message context.

Summary: This patch improves the error message context of the
XCOFF interfaces by providing more details.

Reviewed By: jhenderson

Differential Revision: https://reviews.llvm.org/D110320

3 years ago[Clang] [PowerPC] Fix header include typo in smmintrin.h
Qiu Chaofan [Mon, 11 Oct 2021 02:44:08 +0000 (10:44 +0800)]
[Clang] [PowerPC] Fix header include typo in smmintrin.h

The SSE4 header (smmintrin.h) should include SSSE3 (tmmintrin.h) instead
of SSE2 (emmintrin.h).

Reviewed By: jsji

Differential Revision: https://reviews.llvm.org/D111482

3 years ago[ORC] Add dependence on pthreads library to ORC.
Lang Hames [Mon, 11 Oct 2021 02:32:54 +0000 (19:32 -0700)]
[ORC] Add dependence on pthreads library to ORC.

f3411616896 introduced a dependence (for builds with LLVM_ENABLE_THREADS) on
pthreads. This commit updates the CMakeLists.txt file to include a LINK_LIBS
entry for pthreads.

3 years ago[gn build] Port f34116168964
LLVM GN Syncbot [Mon, 11 Oct 2021 02:15:38 +0000 (02:15 +0000)]
[gn build] Port f34116168964

3 years ago[gn build] Port 3df094d31eac
LLVM GN Syncbot [Mon, 11 Oct 2021 02:15:37 +0000 (02:15 +0000)]
[gn build] Port 3df094d31eac

3 years ago[ORC] Add missing headers.
Lang Hames [Mon, 11 Oct 2021 02:11:46 +0000 (19:11 -0700)]
[ORC] Add missing headers.

These were accidentally left out of f3411616896.

3 years ago[libc++] [P1614] Implement std::compare_three_way.
Arthur O'Dwyer [Thu, 29 Jul 2021 03:40:29 +0000 (23:40 -0400)]
[libc++] [P1614] Implement std::compare_three_way.

Differential Revision: https://reviews.llvm.org/D110735

3 years ago[ORC] Add TaskDispatch API and thread it through ExecutorProcessControl.
Lang Hames [Sat, 9 Oct 2021 00:12:06 +0000 (17:12 -0700)]
[ORC] Add TaskDispatch API and thread it through ExecutorProcessControl.

ExecutorProcessControl objects will now have a TaskDispatcher member which
should be used to dispatch work (in particular, handling incoming packets in
the implementation of remote EPC implementations like SimpleRemoteEPC).

The GenericNamedTask template can be used to wrap function objects that are
callable as 'void()' (along with an optional name to describe the task).
The makeGenericNamedTask functions can be used to create GenericNamedTask
instances without having to name the function object type.

In a future patch ExecutionSession will be updated to use the
ExecutorProcessControl's dispatcher, instead of its DispatchTaskFunction.

3 years ago[NFC][llvm-reduce] Cleanup types
Arthur Eubanks [Tue, 5 Oct 2021 07:05:37 +0000 (00:05 -0700)]
[NFC][llvm-reduce] Cleanup types

Use Module& wherever possible.
Since every reduction immediately turns Chunks into an Oracle, directly pass Oracle instead.

Reviewed By: hans

Differential Revision: https://reviews.llvm.org/D111122

3 years ago[AArch64][GlobalISel] Legalize G_VECREDUCE_XOR. Treated same as other bitwise reductions.
Amara Emerson [Sun, 10 Oct 2021 22:32:58 +0000 (15:32 -0700)]
[AArch64][GlobalISel] Legalize G_VECREDUCE_XOR. Treated same as other bitwise reductions.

3 years ago[llvm-profdata] Allow overlap/similarity comparison to use custom hot threshold cutoff
Wenlei He [Fri, 8 Oct 2021 04:53:04 +0000 (21:53 -0700)]
[llvm-profdata] Allow overlap/similarity comparison to use custom hot threshold cutoff

Allow overlap/similarity comparison to use custom hot threshold cutoff, instead of using hard coded 990000 as hot cutoff.

Differential Revision: https://reviews.llvm.org/D111385

3 years ago[llvm-profgen] Deduplicate PID when processing perf input
Wenlei He [Fri, 8 Oct 2021 04:27:17 +0000 (21:27 -0700)]
[llvm-profgen] Deduplicate PID when processing perf input

When parsing mmap to retrieve PID, deduplicate them before passing PID list to perf script. Perf script would error out when there's duplicated PID in the input, however raw perf data may main duplicated PID for large binary where more than one mmap is needed to load executable segment.

Differential Revision: https://reviews.llvm.org/D111384

3 years agoclang release notes: improve the wording
Sylvestre Ledru [Sun, 10 Oct 2021 20:25:33 +0000 (22:25 +0200)]
clang release notes: improve the wording

3 years ago[ORC] Reorder callWrapperAsync and callSPSWrapperAsync parameters.
Lang Hames [Sun, 10 Oct 2021 19:56:37 +0000 (12:56 -0700)]
[ORC] Reorder callWrapperAsync and callSPSWrapperAsync parameters.

The callee address is now the first parameter and the 'SendResult' function
the second. This change improves consistentency with the non-async functions
where the callee is the first address and the return value the second.

3 years agoRevert "Add missing include after dfd74db9"
Lang Hames [Fri, 8 Oct 2021 20:59:48 +0000 (13:59 -0700)]
Revert "Add missing include after dfd74db9"

This reverts commit dd384d2814094bf5d3ab44f917f759fa24a41158.

dfd74db9 was reverted in 8fe3d9df0ed, so this is no longer needed.

3 years ago[DSE] Re-enable calloc transformation with extra care (PR25892)
Dawid Jurczak [Sun, 10 Oct 2021 17:52:33 +0000 (19:52 +0200)]
[DSE] Re-enable calloc transformation with extra care (PR25892)

Transformation from malloc+memset to calloc is always correct and in many situations
it brings significant observable benefits in terms of execution speed and memory consumption [1][2].
Unfortunately there are cases when producing calloc cause performance drops [3].
As discussed here: https://reviews.llvm.org/D103009 it's possible to differentiate between those 2 scenarios.
If optimizer is able to prove that after malloc call it's _very_ likely to reach memset branch then after
calloc emission we shouldn't observe any performance hits. Therefore finding "null pointer check" pattern
before memset basic block sounds like good justification for performing transformation.
Also that method was already suggested by GCC folks [4]. Main reason for change is that for now
to be safe we check for post dominance relation which is way too conservative approach making transformation
"almost" disabled in practice. This patch tends to enable transformation again but with extra care.

[1] https://stackoverflow.com/questions/2688466/why-mallocmemset-is-slower-than-calloc
[2] https://vorpus.org/blog/why-does-calloc-exist/
[3] http://smalldatum.blogspot.com/2017/11/a-new-optimization-in-gcc-5x-and-mysql.html
[4] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83022

Differential Revision: https://reviews.llvm.org/D110021

3 years agoclang release notes: document the -Wbool-operation improvement
Sylvestre Ledru [Sun, 10 Oct 2021 19:28:24 +0000 (21:28 +0200)]
clang release notes: document the -Wbool-operation improvement

Reviewed By: xbolva00

Differential Revision: https://reviews.llvm.org/D111215

3 years agoclang: Add range-based CFG::try_blocks()
Nico Weber [Sun, 10 Oct 2021 19:14:46 +0000 (15:14 -0400)]
clang: Add range-based CFG::try_blocks()

..and use it. No behavior change.

3 years agoclang: Convert two loops to for-each
Nico Weber [Sun, 10 Oct 2021 18:32:51 +0000 (14:32 -0400)]
clang: Convert two loops to for-each

And rewrap a line at 80 columns while here. No behavior change.

3 years ago[libc++][test] Replace a TEST_NOEXCEPT_FALSE with noexcept(false). NFC.
Joe Loser [Sun, 10 Oct 2021 18:46:35 +0000 (14:46 -0400)]
[libc++][test] Replace a TEST_NOEXCEPT_FALSE with noexcept(false). NFC.

Replace `TEST_NOEXCEPT_FALSE` directly with `noexcept(false)` in
optional hash test which is only run in C++17 or later.
`TEST_NOEXCEPT_FALSE` is only useful in C++03 context where `noexcept`
isn't supported by clang. `TEST_NOEXCEPT_FALSE` now only has one remaining use
in `hash_unique_ptr.pass.cpp`.

3 years ago[libc++] Remove empty namespace std in type_traits. NFCI.
Joe Loser [Sun, 10 Oct 2021 18:35:00 +0000 (14:35 -0400)]
[libc++] Remove empty namespace std in type_traits. NFCI.

There is an empty `namespace std` in `type_traits` which was originally
used when `std::byte` was added in
c97d8aa86650ed795bf75a7dd735ecfaef3b8f55. At some point, the bitwise operators
on `std::byte` got relocated but this empty namespace was left around.
Remove it.

Reviewed By: Quuxplusone, Mordante, #libc

Differential Revision: https://reviews.llvm.org/D111512

3 years ago[fir] Add character conversion pass
Jean Perier [Sun, 10 Oct 2021 18:18:45 +0000 (20:18 +0200)]
[fir] Add character conversion pass

Upstream the character conversion pass.
Translates entities of one CHARACTER KIND to another.
By default the translation is to naively zero-extend or truncate a code
point to fit the destination size.

This patch is part of the upstreaming effort from fir-dev branch.

Co-authored-by: Eric Schweitz <eschweitz@nvidia.com>
Co-authored-by: Valentin Clement <clementval@gmail.com>
Reviewed By: schweitz

Differential Revision: https://reviews.llvm.org/D111405

3 years ago[libc++][NFC] Replace tab with whitespace in comment
Joe Loser [Sun, 10 Oct 2021 16:53:35 +0000 (12:53 -0400)]
[libc++][NFC] Replace tab with whitespace in comment

There is a stray tab character in a comment block. Replace the tab
character with a space for consistency with other comments.

3 years ago[Basic] Use llvm::is_contained (NFC)
Kazu Hirata [Sun, 10 Oct 2021 15:52:14 +0000 (08:52 -0700)]
[Basic] Use llvm::is_contained (NFC)

3 years ago[InstCombine] move fold for "(X-Y) == 0"; NFC
Sanjay Patel [Sun, 10 Oct 2021 15:26:03 +0000 (11:26 -0400)]
[InstCombine] move fold for "(X-Y) == 0"; NFC

This consolidates related folds that all have a
similar use restriction that may not be necessary.

3 years ago[InstCombine] add tests for (X - Y) == 0; NFC
Sanjay Patel [Sun, 10 Oct 2021 15:13:46 +0000 (11:13 -0400)]
[InstCombine] add tests for (X - Y) == 0; NFC

3 years ago[InstCombine] canonicalize "(C2 - Y) > C" as (Y + ~C2) < ~C
Sanjay Patel [Sun, 10 Oct 2021 14:58:58 +0000 (10:58 -0400)]
[InstCombine] canonicalize "(C2 - Y) > C" as (Y + ~C2) < ~C

The test diffs show that we have better analysis/folds for 'add'
(although we should at least have the simplifications
independently, so we don't have the one-use restriction).

This is related to solving regressions that would appear in
transforms related to D111410, and that is part of a series
of enhancements that may eventually helpi solve PR34047.

https://alive2.llvm.org/ce/z/3tB9KG

  define i1 @src(i8 %x, i8 %C, i8 %C2) {
    %sub = sub nuw i8 %C2, %x
    %r = icmp slt i8 %sub, %C
    ret i1 %r
  }

  define i1 @tgt(i8 %x, i8 %C, i8 %C2) {
    %Cnot = xor i8 %C, -1
    %C2not = xor i8 %C2, -1
    %add = add nuw i8 %x, %C2not
    %r = icmp sgt i8 %add, %Cnot
    ret i1 %r
  }

3 years ago[InstCombine] add test for or-of-icmps; NFC
Sanjay Patel [Sun, 10 Oct 2021 14:41:28 +0000 (10:41 -0400)]
[InstCombine] add test for or-of-icmps; NFC

3 years ago[PowerPC] update test case using the scripts; nfc
Chen Zheng [Sun, 10 Oct 2021 14:39:20 +0000 (14:39 +0000)]
[PowerPC] update test case using the scripts; nfc

3 years ago[libc++][nfc] Remove a duplicated include.
Mark de Wever [Sun, 10 Oct 2021 12:21:01 +0000 (14:21 +0200)]
[libc++][nfc] Remove a duplicated include.

3 years ago[NFC] Added tests for PR52056
Dávid Bolvanský [Sun, 10 Oct 2021 09:34:03 +0000 (11:34 +0200)]
[NFC] Added tests for PR52056

3 years ago[BitcodeAnalyzer] allow a motivated user to dump BLOCKINFO
william woodruff [Sun, 10 Oct 2021 04:10:22 +0000 (09:40 +0530)]
[BitcodeAnalyzer] allow a motivated user to dump BLOCKINFO

This adds the `--dump-blockinfo` flag to `llvm-bcanalyzer`, allowing a sufficiently motivated user to dump (parts of) the `BLOCKINFO_BLOCK` block. The default behavior is unchanged, and `--dump-blockinfo` only takes effect in the same context as other flags that control dump behavior (i.e., requires that `--dump` is also passed).

Reviewed By: tejohnson

Differential Revision: https://reviews.llvm.org/D107536

3 years ago[GlobalISel] Fix the stores of truncates -> wide store combine for non-evenly dividin...
Amara Emerson [Sun, 10 Oct 2021 03:43:21 +0000 (20:43 -0700)]
[GlobalISel] Fix the stores of truncates -> wide store combine for non-evenly dividing type sizes.

If the wide store we'd generate is not a multiple of the memory type of the
narrow stores (e.g. s48 and s32), we'd assert. Fix that.

3 years ago[clang] Fix JSON AST output when a filter is used
william woodruff [Sun, 10 Oct 2021 00:44:08 +0000 (06:14 +0530)]
[clang] Fix JSON AST output when a filter is used

Without this, the combination of `-ast-dump=json` and `-ast-dump-filter FILTER` produces invalid JSON: the first line is a string that says `Dumping $SOME_DECL_NAME: `.

Reviewed By: aaron.ballman

Differential Revision: https://reviews.llvm.org/D108441

3 years ago[lldb/test] Disable 'TestScriptedProcess.py' on macOS
Med Ismail Bennani [Sun, 10 Oct 2021 01:28:36 +0000 (03:28 +0200)]
[lldb/test] Disable 'TestScriptedProcess.py' on macOS

This is disabling 'TestScriptedProcess.py' on macOS since it fails on
Green Dragon: https://green.lab.llvm.org/green/view/LLDB/job/lldb-cmake/35974

Signed-off-by: Med Ismail Bennani <medismail.bennani@gmail.com>
3 years ago[libc++][test] Remove empty {ind.move.subsumption.compile.pass.cpp}
Joe Loser [Sat, 9 Oct 2021 21:20:19 +0000 (17:20 -0400)]
[libc++][test] Remove empty {ind.move.subsumption.compile.pass.cpp}

`{ind.move.subsumption.compile.pass.cpp}` was accidentally commited in
https://reviews.llvm.org/D102639. Per the conversation on Discord in

3 years ago[mlir] Vectorize induction variables
Amy Zhuang [Sat, 9 Oct 2021 19:40:13 +0000 (12:40 -0700)]
[mlir] Vectorize induction variables

1. Add support to vectorize induction variables of loops that are
   not mapped to any vector dimension in SuperVectorize pass.
2. Fix a bug in getForInductionVarOwner.

Reviewed By: dcaballe

Differential Revision: https://reviews.llvm.org/D111370

3 years ago[clang-format][NFC] improve the visual of the "clang-formatted %"
mydeveloperday [Sat, 9 Oct 2021 18:34:30 +0000 (19:34 +0100)]
[clang-format][NFC] improve the visual of the "clang-formatted %"

NOTE: some files are being removed from those files that are clang-formatted
which means some lack of formatting is slipping through the net on reviews

3 years agoFix a comment at call-site to match the declared parameter (NFC)
Mehdi Amini [Sat, 9 Oct 2021 17:56:23 +0000 (17:56 +0000)]
Fix a comment at call-site to match the declared parameter (NFC)

(clang-tidy warning)

3 years ago[libomptarget][amdgpu][NFC] tweak a comment
Ron Lieberman [Sat, 9 Oct 2021 16:51:53 +0000 (12:51 -0400)]
[libomptarget][amdgpu][NFC] tweak a comment

3 years ago[IR] Remove arg_operands and getNumArgOperands (NFC)
Kazu Hirata [Sat, 9 Oct 2021 16:38:15 +0000 (09:38 -0700)]
[IR] Remove arg_operands and getNumArgOperands (NFC)

The last uses were removed on Oct 8, 2021 in commit
46ef2e0bf995d8db4cbdf69f3d1bbc2487030ba0.

This is a relanding of b2ee408dde374d6a27a34746fd7c7b5bab97ea89.

3 years ago[InstCombine] enhance icmp with sub folds
Sanjay Patel [Sat, 9 Oct 2021 15:34:48 +0000 (11:34 -0400)]
[InstCombine] enhance icmp with sub folds

There were 2 related but over-specified folds for:
C1 - X == C

One allowed multi-use but was limited to equal constants.
The other allowed different constants but disallowed multi-use.

This combines the 2 folds into a more general match.
The test diffs show the multi-use cases that were falling
through the cracks.

https://alive2.llvm.org/ce/z/4_hEt2

  define i1 @src(i8 %x, i8 %subC, i8 %C) {
    %s = sub i8 %subC, %x
    %r = icmp eq i8 %s, %C
    ret i1 %r
  }

  define i1 @tgt(i8 %x, i8 %subC, i8 %C) {
    %newC = sub i8 %subC, %C
    %isneg = icmp eq i8 %x, %newC
    ret i1 %isneg
  }

3 years ago[InstCombine] add tests for icmp of negated op; NFC
Sanjay Patel [Fri, 8 Oct 2021 20:30:55 +0000 (16:30 -0400)]
[InstCombine] add tests for icmp of negated op; NFC

3 years ago[InstCombine] add tests for (iN X s>> N-1) | Y; NFC
Sanjay Patel [Fri, 8 Oct 2021 17:03:19 +0000 (13:03 -0400)]
[InstCombine] add tests for (iN X s>> N-1) | Y; NFC

These are for a sibling fold suggested in D111410.
The tests correspond to the 'and' tests added with:
a35673f4cfc4

3 years agoFixed some errors detected by PVS Studio
Dávid Bolvanský [Sat, 9 Oct 2021 15:27:41 +0000 (17:27 +0200)]
Fixed some errors detected by PVS Studio

3 years agoFixed some errors detected by PVS Studio
Dávid Bolvanský [Sat, 9 Oct 2021 15:19:53 +0000 (17:19 +0200)]
Fixed some errors detected by PVS Studio

3 years ago[CanonicalizeFreeze] Drop IVUsers.h include (NFC)
Nikita Popov [Sat, 9 Oct 2021 14:53:07 +0000 (16:53 +0200)]
[CanonicalizeFreeze] Drop IVUsers.h include (NFC)

Looking for users of IVUsers, this was a false positive. Only LSR
uses IVUsers.

3 years ago[AArch64] Make -mcpu=generic schedule for an in-order core
David Green [Sat, 9 Oct 2021 14:58:31 +0000 (15:58 +0100)]
[AArch64] Make -mcpu=generic schedule for an in-order core

We would like to start pushing -mcpu=generic towards enabling the set of
features that improves performance for some CPUs, without hurting any
others. A blend of the performance options hopefully beneficial to all
CPUs. The largest part of that is enabling in-order scheduling using the
Cortex-A55 schedule model. This is similar to the Arm backend change
from eecb353d0e25ba which made -mcpu=generic perform in-order scheduling
using the cortex-a8 schedule model.

The idea is that in-order cpu's require the most help in instruction
scheduling, whereas out-of-order cpus can for the most part out-of-order
schedule around different codegen. Our benchmarking suggests that
hypothesis holds. When running on an in-order core this improved
performance by 3.8% geomean on a set of DSP workloads, 2% geomean on
some other embedded benchmark and between 1% and 1.8% on a set of
singlecore and multicore workloads, all running on a Cortex-A55 cluster.

On an out-of-order cpu the results are a lot more noisy but show flat
performance or an improvement. On the set of DSP and embedded
benchmarks, run on a Cortex-A78 there was a very noisy 1% speed
improvement. Using the most detailed results I could find, SPEC2006 runs
on a Neoverse N1 show a small increase in instruction count (+0.127%),
but a decrease in cycle counts (-0.155%, on average). The instruction
count is very low noise, the cycle count is more noisy with a 0.15%
decrease not being significant. SPEC2k17 shows a small decrease (-0.2%)
in instruction count leading to a -0.296% decrease in cycle count. These
results are within noise margins but tend to show a small improvement in
general.

When specifying an Apple target, clang will set "-target-cpu apple-a7"
on the command line, so should not be affected by this change when
running from clang. This also doesn't enable more runtime unrolling like
-mcpu=cortex-a55 does, only changing the schedule used.

A lot of existing tests have updated. This is a summary of the important
differences:
 - Most changes are the same instructions in a different order.
 - Sometimes this leads to very minor inefficiencies, such as requiring
   an extra mov to move variables into r0/v0 for the return value of a test
   function.
 - misched-fusion.ll was no longer fusing the pairs of instructions it
   should, as per D110561. I've changed the schedule used in the test
   for now.
 - neon-mla-mls.ll now uses "mul; sub" as opposed to "neg; mla" due to
   the different latencies. This seems fine to me.
 - Some SVE tests do not always remove movprfx where they did before due
   to different register allocation giving different destructive forms.
 - The tests argument-blocks-array-of-struct.ll and arm64-windows-calls.ll
   produce two LDR where they previously produced an LDP due to
   store-pair-suppress kicking in.
 - arm64-ldp.ll and arm64-neon-copy.ll are missing pre/postinc on LPD.
 - Some tests such as arm64-neon-mul-div.ll and
   ragreedy-local-interval-cost.ll have more, less or just different
   spilling.
 - In aarch64_generated_funcs.ll.generated.expected one part of the
   function is no longer outlined. Interestingly if I switch this to use
   any other scheduled even less is outlined.

Some of these are expected to happen, such as differences in outlining
or register spilling. There will be places where these result in worse
codegen, places where they are better, with the SPEC instruction counts
suggesting it is not a decrease overall, on average.

Differential Revision: https://reviews.llvm.org/D110830

3 years agoRevert "Reland "[gn build] (manually) port 6fe2beba7d2a (ExceptionTests)""
Nico Weber [Sat, 9 Oct 2021 14:18:52 +0000 (10:18 -0400)]
Revert "Reland "[gn build] (manually) port 6fe2beba7d2a (ExceptionTests)""

This reverts commit 842035d8bdf470af05848114ce1808802c5d4aef.
1dba6b3 was reverted yet again in 04aff395047a.

3 years ago[lldb] [DynamicRegisterInfo] Remove obsolete dwarf typedefs (NFC)
Michał Górny [Sat, 9 Oct 2021 13:42:34 +0000 (15:42 +0200)]
[lldb] [DynamicRegisterInfo] Remove obsolete dwarf typedefs (NFC)

3 years ago[lldb][NFC] Early-exit in DWARFASTParserClang::ParseSingleMember
Raphael Isemann [Sat, 9 Oct 2021 12:15:56 +0000 (14:15 +0200)]
[lldb][NFC] Early-exit in DWARFASTParserClang::ParseSingleMember

ParseSingleMember has two large ifs around the back of it's body:
`if (!is_artificial)` and `if (member_type)`. This patch just converts those
to early-exits. The patch is NFC. It even retains the curious fact that
Objective-C properties that fail to parse are silently ignored, but now there
is at least a FIXME that points this out.

3 years agoFix a diagnoses-valid in C++20 with variadic macros
Aaron Ballman [Sat, 9 Oct 2021 12:20:20 +0000 (08:20 -0400)]
Fix a diagnoses-valid in C++20 with variadic macros

C++20 and later allow you to pass no argument for the ... parameter in
a variadic macro, whereas earlier language modes and C disallow it.

We no longer diagnose in C++20 and later modes. This fixes PR51609.

3 years ago[NFC][libc++] Update back_insert_iterator style.
Mark de Wever [Sat, 9 Oct 2021 11:31:20 +0000 (13:31 +0200)]
[NFC][libc++] Update back_insert_iterator style.

As suggested in D110573 land the rename part separately.

3 years ago[libc++][doc] Update format status.
Mark de Wever [Sat, 9 Oct 2021 11:28:38 +0000 (13:28 +0200)]
[libc++][doc] Update format status.

Updated based on recent commits.

3 years ago[clang-format][NFC] Fix spelling mistakes
mydeveloperday [Sat, 9 Oct 2021 11:26:07 +0000 (12:26 +0100)]
[clang-format][NFC] Fix spelling mistakes

3 years ago[Driver][OpenBSD] Use ToolChain reference instead of getToolChain().
Frederic Cambus [Sat, 9 Oct 2021 11:21:39 +0000 (13:21 +0200)]
[Driver][OpenBSD] Use ToolChain reference instead of getToolChain().

Differential Revision: https://reviews.llvm.org/D111462

3 years ago[clang-format][NFC] Fix spelling mistake
mydeveloperday [Sat, 9 Oct 2021 11:18:25 +0000 (12:18 +0100)]
[clang-format][NFC] Fix spelling mistake

3 years ago[clang-format][docs][NFC] correct the "first supported versions" of some of the clang...
mydeveloperday [Sat, 9 Oct 2021 10:02:49 +0000 (11:02 +0100)]
[clang-format][docs][NFC] correct the "first supported versions" of some of the clang-format options

Some of the first supported version field were incorrectly attributed to a later branch.

It wasn't possible to correctly determine the "introduced version" with my naive implementation
using git blame alone, (especially if the type had been changed from a bool -> enum)

I saw more things attributed to clang-format 13 than I remembered and reviewed
those options to determine their introduced version.

Reviewed By: HazardyKnusperkeks

Differential Revision: https://reviews.llvm.org/D110803

3 years ago[Type] Avoid APFloat.h include (NFC)
Nikita Popov [Sat, 9 Oct 2021 09:28:11 +0000 (11:28 +0200)]
[Type] Avoid APFloat.h include (NFC)

This is only used by a handful of methods working on fltSemantics,
and having these defined inline in the header does not look
particularly important.

3 years ago[MCPseudoProbe] Clean up includes (NFC)
Nikita Popov [Sat, 9 Oct 2021 08:30:39 +0000 (10:30 +0200)]
[MCPseudoProbe] Clean up includes (NFC)

This was including various things that don't appear to be used in
the header at all.

3 years ago[Orc] Fix global variable destructor function support when --jit-kind=orc-lazy
luxufan [Sat, 9 Oct 2021 07:40:03 +0000 (15:40 +0800)]
[Orc] Fix global variable destructor function support when --jit-kind=orc-lazy

The bug was reported here https://bugs.llvm.org/show_bug.cgi?id=52030

This patch follows the idea that @lhames commented in the above webpage.

Reviewed By: lhames

Differential Revision: https://reviews.llvm.org/D110990

3 years ago[LoopDeletion] Support selects when symbolically evaluating 1st iteration
Max Kazantsev [Sat, 9 Oct 2021 07:47:44 +0000 (14:47 +0700)]
[LoopDeletion] Support selects when symbolically evaluating 1st iteration

Adds support for selects for which we know value on the 1st iteration.

Differential Revision: https://reviews.llvm.org/D104111
Reviewed By: nikic

3 years ago[Test] Add commit justifying revert of D110922
Max Kazantsev [Sat, 9 Oct 2021 07:32:46 +0000 (14:32 +0700)]
[Test] Add commit justifying revert of D110922

Test by Arthur Eubanks!

3 years ago[Orc] Support atexit in Orc(JITLink)
luxufan [Sat, 9 Oct 2021 00:36:41 +0000 (08:36 +0800)]
[Orc] Support atexit in Orc(JITLink)

There is a bug reported at https://bugs.llvm.org/show_bug.cgi?id=48938

After looking through the glibc, I found the `atexit(f)` is the same as `__cxa_atexit(f, NULL, NULL)`. In orc runtime, we identify different JITDylib by their dso_handle value, so that a NULL dso_handle is invalid. So in this patch, I added a `PlatformJDDSOHandle` to ELFNixRuntimeState, and functions which are registered by atexit will be registered at PlatformJD.

Reviewed By: lhames

Differential Revision: https://reviews.llvm.org/D111413

3 years ago[BitcodeReader] fix a logic error in vector type element validation
william woodruff [Sat, 9 Oct 2021 03:29:45 +0000 (08:59 +0530)]
[BitcodeReader] fix a logic error in vector type element validation

The current code checks whether the vector's element type is a valid structure element type, rather than a valid vector element type. The two have separate implementations and but only accept very slightly different sets of types, which is probably why this wasn't caught before.

Differential Revision: https://reviews.llvm.org/D109655

3 years ago[OpenBSD] Use cortex-a8 as default CPU for ARMv7
Brad Smith [Sat, 9 Oct 2021 03:56:52 +0000 (23:56 -0400)]
[OpenBSD] Use cortex-a8 as default CPU for ARMv7

3 years ago[CFE][Codegen][In-progress] Remove CodeGenFunction::InitTempAlloca()
hsmahesha [Sat, 9 Oct 2021 03:52:59 +0000 (09:22 +0530)]
[CFE][Codegen][In-progress] Remove CodeGenFunction::InitTempAlloca()

CodeGenFunction::InitTempAlloca() inits the static alloca within the
entry block which may *not* necessarily be correct always.

For example, the current instruction insertion point (pointed by the
instruction builder) could be a program point which is hit multiple
times during the program execution, and it is expected that the static
alloca is initialized every time the program point is hit.

Hence remove CodeGenFunction::InitTempAlloca(), and initialize the
static alloca where the instruction insertion point is at the moment.

This patch, as a starting attempt, removes the calls to
CodeGenFunction::InitTempAlloca() which do not have any side effect on
the lit tests.

Reviewed By: rjmccall

Differential Revision: https://reviews.llvm.org/D111293

3 years ago[Polly] Fix test case fixing the colon.
Michael Kruse [Sat, 9 Oct 2021 03:42:27 +0000 (22:42 -0500)]
[Polly] Fix test case fixing the colon.

Commit 573531fb1f529b1413b789fa9eee11c7b41ac83d fixed the colon at the
end of a CHECK line (was a semicolon by mistake). With the check
enabled, it turned out that it was failing. Check for the correct
content.

Also add the missing colon to the next CHECK line.

3 years agoRevert a LIT typo fix in a RUN line
Qiu Chaofan [Sat, 9 Oct 2021 03:26:01 +0000 (11:26 +0800)]
Revert a LIT typo fix in a RUN line

Commit 573531f changes the behavior of the test, revert it back.

3 years agoDisable mlir/test/mlir-cpu-runner/async-group.mlir with ASAN
Mehdi Amini [Sat, 9 Oct 2021 03:01:42 +0000 (03:01 +0000)]
Disable mlir/test/mlir-cpu-runner/async-group.mlir with ASAN

This test is crashing 9 out of 10 runs in CI, but I can't reproduce
locally right now. Disabling to get the CI back to green and avoid
backsliding with more ASAN issues that would go unnoticed.

3 years agoDon't update the vptr at the start of the destructor of a final class.
Richard Smith [Sat, 9 Oct 2021 02:06:22 +0000 (19:06 -0700)]
Don't update the vptr at the start of the destructor of a final class.

In this case, we know statically that we're destroying the most-derived
class, so the vptr must already point to the current class and never
needs to be updated.

3 years ago[Clang] Enable _Complex __ibm128 type
Qiu Chaofan [Sat, 9 Oct 2021 02:48:44 +0000 (10:48 +0800)]
[Clang] Enable _Complex __ibm128 type

fae0dfa implemented the new __ibm128 type, this patch enables its
complex form.

Reviewed By: rjmccall

Differential Revision: https://reviews.llvm.org/D109948

3 years ago[NFC] [Clang] Use global enum for explicit float mode
Qiu Chaofan [Sat, 9 Oct 2021 02:39:10 +0000 (10:39 +0800)]
[NFC] [Clang] Use global enum for explicit float mode

Currently, there're multiple float types that can be represented by
__attribute__((mode(xx))). It's parsed, and then a corresponding type is
created if available.

This refactor moves the enum for mode into a global enum class visible
to ASTContext.

Reviewed By: aaron.ballman, erichkeane

Differential Revision: https://reviews.llvm.org/D111391

3 years ago[OpenMP] Add RTL function for getting number of threads in block.
Joseph Huber [Sat, 9 Oct 2021 00:08:28 +0000 (20:08 -0400)]
[OpenMP] Add RTL function for getting number of threads in block.

This patch adds support for the
`__kmpc_get_hardware_num_threads_in_block` function that returns the
number of threads. This was missing in the new runtime and was used by
the AMDGPU plugin which prevented it from using the new runtime. This
patchs also unified the interface for getting the thread numbers in the
frontend.

Originally authored by jdoerfert.

Reviewed By: JonChesterfield

Differential Revision: https://reviews.llvm.org/D111475

3 years ago[APFloat] Set size of PPCDoubleDouble to 128
Qiu Chaofan [Sat, 9 Oct 2021 02:12:10 +0000 (10:12 +0800)]
[APFloat] Set size of PPCDoubleDouble to 128

566690b0 uses size information in float semantics, but PPCDoubleDouble
left them empty.

As follow-up, we can consider remove PPCDoubleDoubleLegacy and fill
other fields in the future.

Reviewed By: foad

Differential Revision: https://reviews.llvm.org/D111398

3 years agoFix typo of colon to semicolon in lit tests
Qiu Chaofan [Sat, 9 Oct 2021 02:01:27 +0000 (10:01 +0800)]
Fix typo of colon to semicolon in lit tests

3 years ago[OpenMP] Avoid calling `isSPMDMode` during RT initialization
Joseph Huber [Sat, 9 Oct 2021 01:52:54 +0000 (21:52 -0400)]
[OpenMP] Avoid calling `isSPMDMode` during RT initialization

Until we hit the first barrier we should not call `mapping::isSPMDMode`
with all threads. Instead, we now have (and use during initialization) a
`mapping::isMainThreadInGenericMode` overload that takes the known
SPMD-mode state and one that queries it.

Reviewed By: tianshilei1992

Differential Revision: https://reviews.llvm.org/D111381

3 years agoPR51079: Treat thread_local variables with an incomplete class type as
Richard Smith [Sat, 9 Oct 2021 01:38:30 +0000 (18:38 -0700)]
PR51079: Treat thread_local variables with an incomplete class type as
being not trivially destructible when determining if we can skip calling
their thread wrapper function.

3 years ago[Polly] Add greedy fusion algorithm.
Michael Kruse [Sat, 9 Oct 2021 00:49:40 +0000 (19:49 -0500)]
[Polly] Add greedy fusion algorithm.

When the option -polly-loopfusion-greedy is set, the ScheduleOptimizer
tries to aggressively fuse any band it can and does not violate any
dependences.

As part if the implementation, the functionalty for copying a band
into an new schedule was extracted out of the ScheduleTreeRewriter.

3 years ago[LICM] Use Align instead of int
Arthur Eubanks [Sat, 9 Oct 2021 01:26:15 +0000 (18:26 -0700)]
[LICM] Use Align instead of int

3 years agoRemove unnecessary StringRef convesion in llvm-config
John Ericson [Thu, 7 Oct 2021 15:39:01 +0000 (11:39 -0400)]
Remove unnecessary StringRef convesion in llvm-config

We have a string litteral (via CPP) used to construct `StringRef`, which
is used to construct a `SmallString`. Just construct the latter
directly.

Differential Revision: https://reviews.llvm.org/D111322

3 years agoAdd no_instrument_function attribute to Objective C methods as well
Aditya Kumar [Fri, 8 Oct 2021 23:28:15 +0000 (16:28 -0700)]
Add no_instrument_function attribute to Objective C methods as well

There are functions where we do not want function instrumentation which is why we have `__attribute__((no_instrument_function))`. Extending this functionality to disable instrumentation for Objective-C methods as well. Objective C methods like `+load` run premain and having instrumentation on them causes runtime errors depending on the implementation of `__cyg_profile_func_enter` etc. functions

Reviewed By: rjmccall, aaron.ballman

Differential Revision: https://reviews.llvm.org/D111286

3 years agoRevert "Reland "[clang-repl] Re-implement clang-interpreter as a test case.""
Leonard Chan [Sat, 9 Oct 2021 00:43:23 +0000 (17:43 -0700)]
Revert "Reland "[clang-repl] Re-implement clang-interpreter as a test case.""

This reverts commit 1dba6b37bdc70210f75a480eff3715ebe1f1d8be.

Reverting because the ClangReplInterpreterExceptionTests test fails on
our builders with this patch.

3 years ago[LangRef] Fix a typo in DISubrange section
Yuanfang Chen [Fri, 8 Oct 2021 23:46:03 +0000 (16:46 -0700)]
[LangRef] Fix a typo in DISubrange section

3 years agoMake more places that use alignment use uint64_t
Arthur Eubanks [Fri, 8 Oct 2021 23:34:22 +0000 (16:34 -0700)]
Make more places that use alignment use uint64_t

Followup to D110451.

3 years ago[libc++][spaceship] Implement std::tuple::operator<=>
Kent Ross [Fri, 8 Oct 2021 21:54:28 +0000 (14:54 -0700)]
[libc++][spaceship] Implement std::tuple::operator<=>

Implement parts of P1614, including three-way comparison for tuples, and expand testing.

Reviewed By: ldionne, Mordante, #libc

Differential Revision: https://reviews.llvm.org/D108250

3 years agoFix TargetRegistry shlib build, clang edition
Reid Kleckner [Fri, 8 Oct 2021 22:43:43 +0000 (15:43 -0700)]
Fix TargetRegistry shlib build, clang edition

3 years ago[InlineCost] model calls to llvm.is.constant* more carefully
Nick Desaulniers [Fri, 8 Oct 2021 22:17:54 +0000 (15:17 -0700)]
[InlineCost] model calls to llvm.is.constant* more carefully

llvm.is.constant* intrinsics are evaluated to 0 or 1 integral values.

A common use case for llvm.is.constant comes from the higher level
__builtin_constant_p. A common usage pattern of __builtin_constant_p in
the Linux kernel is:

    void foo (int bar) {
      if (__builtin_constant_p(bar)) {
        // lots of code that will fold away to a constant.
      } else {
        // a little bit of code, usually a libcall.
      }
    }

A minor issue in InlineCost calculations is when `bar` is _not_ Constant
and still will not be after inlining, we don't discount the true branch
and the inline cost of `foo` ends up being the cost of both branches
together, rather than just the false branch.

This leads to code like the above where inlining will not help prove bar
Constant, but it still would be beneficial to inline foo, because the
"true" branch is irrelevant from a cost perspective.

For example, IPSCCP can sink a passed constant argument to foo:

    const int x = 42;
    void bar (void) { foo(x); }

This improves our inlining decisions, and fixes a few head scratching
cases were the disassembly shows a relatively small `foo` not inlined
into a lone caller.

We could further improve this modeling by tracking whether the argument
to llvm.is.constant* is a parameter of the function, and if inlining
would allow that parameter to become Constant. This idea is noted in a
FIXME comment.

Link: https://github.com/ClangBuiltLinux/linux/issues/1302
Reviewed By: kazu

Differential Revision: https://reviews.llvm.org/D111272

3 years ago[ADT] Mark IntervalMap::overlaps const
Vedant Kumar [Fri, 8 Oct 2021 22:00:02 +0000 (15:00 -0700)]
[ADT] Mark IntervalMap::overlaps const

This allows the overlaps() predicate to be used on a const IntervalMap.
Tested by building ADTTests, llc, and lldb-test.

3 years agoFix shlib builds for all lib/Target/*/TargetInfo libs
Reid Kleckner [Fri, 8 Oct 2021 22:18:58 +0000 (15:18 -0700)]
Fix shlib builds for all lib/Target/*/TargetInfo libs

They all must depend on MC now that the target registry is in MC.
Also fix llvm-cxxdump

3 years agoFix shared library build after TargetRegistry move
Reid Kleckner [Fri, 8 Oct 2021 22:06:03 +0000 (15:06 -0700)]
Fix shared library build after TargetRegistry move