Tim Keith [Wed, 15 Jul 2020 22:08:07 +0000 (15:08 -0700)]
[flang] Don't use-associate intrinsics
When an intrinsic is referenced in a module scope, a symbol for it is
added. When that module is USEd, the intrinsic should not be included.
Otherwise we can get ambiguous reference errors with the same intrinsic
coming from two difference modules.
Differential Revision: https://reviews.llvm.org/D83905
Akira Hatanaka [Wed, 15 Jul 2020 21:47:45 +0000 (14:47 -0700)]
[CodeGen] Emit a call instruction instead of an invoke if the called
llvm function is marked nounwind
This fixes cases where an invoke is emitted, despite the called llvm
function being marked nounwind, because ConstructAttributeList failed to
add the attribute to the attribute list. llvm optimization passes turn
invokes into calls and optimize away the exception handling code, but
it's better to avoid emitting the code in the front-end if the called
function is known not to raise an exception.
Differential Revision: https://reviews.llvm.org/D83906
Roman Lebedev [Wed, 15 Jul 2020 21:27:36 +0000 (00:27 +0300)]
[NFC][SimplifyCFG] HoistThenElseCodeToIf(): after hoisting terminator, do return Changed, not just true
Otherwise, if Changed was still false before that,
we would not account for that hoist in NumHoistCommonCode statistic.
Stephen Neuendorffer [Fri, 15 May 2020 17:33:13 +0000 (10:33 -0700)]
[MLIR] Add RegionKindInterface
Some dialects have semantics which is not well represented by common
SSA structures with dominance constraints. This patch allows
operations to declare the 'kind' of their contained regions.
Currently, two kinds are allowed: "SSACFG" and "Graph". The only
difference between them at the moment is that SSACFG regions are
required to have dominance, while Graph regions are not required to
have dominance. The intention is that this Interface would be
generated by ODS for existing operations, although this has not yet
been implemented. Presumably, if someone were interested in code
generation, we might also have a "CFG" dialect, which defines control
flow, but does not require SSA.
The new behavior is mostly identical to the previous behavior, since
registered operations without a RegionKindInterface are assumed to
contain SSACFG regions. However, the behavior has changed for
unregistered operations. Previously, these were checked for
dominance, however the new behavior allows dominance violations, in
order to allow the processing of unregistered dialects with Graph
regions. One implication of this is that regions in unregistered
operations with more than one op are no longer CSE'd (since it
requires dominance info).
I've also reorganized the LangRef documentation to remove assertions
about "sequential execution", "SSA Values", and "Dominance". Instead,
the core IR is simply "ordered" (i.e. totally ordered) and consists of
"Values". I've also clarified some things about how control flow
passes between blocks in an SSACFG region. Control Flow must enter a
region at the entry block and follow terminator operation successors
or be returned to the containing op. Graph regions do not define a
notion of control flow.
see discussion here:
https://llvm.discourse.group/t/rfc-allowing-dialects-to-relax-the-ssa-dominance-condition/833/53
Differential Revision: https://reviews.llvm.org/D80358
Roman Lebedev [Wed, 15 Jul 2020 21:10:20 +0000 (00:10 +0300)]
[NFC][SimplifyCFG] HoistThenElseCodeToIf(): count number of common instruction "blocks" hoisted
I.e. out of all the times HoistThenElseCodeToIf() was called,
how many times did it actually hoist something?
Roman Lebedev [Wed, 15 Jul 2020 21:05:25 +0000 (00:05 +0300)]
[NFC][SimplifyCFG] HoistThenElseCodeToIf(): count number of common instructions hoisted
Roman Lebedev [Wed, 15 Jul 2020 20:55:15 +0000 (23:55 +0300)]
[NFC][SimplifyCFG] SinkCommonCodeFromPredecessors(): count number of instruction "blocks" actually sunk
Out of all the times the function was called,
how many times did we actually sink anything?
Roman Lebedev [Wed, 15 Jul 2020 20:48:31 +0000 (23:48 +0300)]
[NFC][SimplifyCFG] SinkCommonCodeFromPredecessors(): add debug output when failing to actually sink instr
Roman Lebedev [Wed, 15 Jul 2020 20:46:35 +0000 (23:46 +0300)]
[NFC][SimplifyCFG] SinkCommonCodeFromPredecessors(): early return if nothing to sink
If we can't sink even one instruction, early return, to increase readability.
Roman Lebedev [Wed, 15 Jul 2020 20:42:40 +0000 (23:42 +0300)]
[NFC][SimplifyCFG] Rename statistic NumSinkCommons into NumSinkCommonInstrs
It really counts instructions added into common block,
not number of instruction groups sunk.
Roman Lebedev [Wed, 15 Jul 2020 20:38:43 +0000 (23:38 +0300)]
[NFC][LoopRotate] Add a statistic for how many times rotation failed due to the header size
Jared Wyles [Tue, 14 Jul 2020 06:57:59 +0000 (16:57 +1000)]
[jitlink] Adding support for PCRel32GOTLoad in ELF x86 for the jitlinker
Summary: This adds the basic support for GOT in elf x86.
Was able to just get away using the macho code by generalising the edges.
There will be a follow up patch to turn that into a generic utility for both of the x86 and Mach-O code.
This patch also lands support for relocations relative to symbol.
Reviewers: lhames
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D83748
Mehdi Amini [Wed, 15 Jul 2020 21:11:39 +0000 (21:11 +0000)]
Document the testing of Analyses in the LLVM testing guide (NFC)
This came up in a recent review, someone was wondering were was
this all documented and I couldn't find a reference to provide.
Differential Revision: https://reviews.llvm.org/D83816
Mehdi Amini [Wed, 15 Jul 2020 21:11:30 +0000 (21:11 +0000)]
Clarify a bit the guideline on omitting braces, including more examples (NFC)
Like most readability rules, it isn't absolute and there is a matter of taste
to it. I think more recent part of the project may be more consistent in the
current application of the guideline. I suspect sources like
mlir/lib/Dialect/StandardOps/IR/Ops.cpp may be examples of this at the moment.
Differential Revision: https://reviews.llvm.org/D82594
Jonas Devlieghere [Wed, 15 Jul 2020 16:59:54 +0000 (09:59 -0700)]
[lldb/Test] Remove custom tearDownHooks from GDB Remote tests
Remove custom tearDownHooks from GDB Remote tests as we now cleanup
subprocesses unconditionally. This also changes the termination order to
be the reverse of the creation order. I don't think anything is relying
on that right now, but it better fits the setup/teardown paradigm.
Abhina Sreeskantharajan [Wed, 15 Jul 2020 18:00:57 +0000 (14:00 -0400)]
[NFC] [AIX] [z/OS] Fix build failure on AIX and z/OS
Summary: This PR contains a build failure fix that occurs on both AIX and z/OS as a result of this commit https://reviews.llvm.org/rG670915094462d831e3733e5b01a76471b8cf6dd8.
Reviewers: uweigand, Kai, hubert.reinterpretcast, daltenty, lhames
Reviewed By: Kai, hubert.reinterpretcast, daltenty
Subscribers: SeanP, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D83889
Tim Keith [Wed, 15 Jul 2020 20:02:32 +0000 (13:02 -0700)]
[flang] Fix erroneous application of SAVE statement
A SAVE statement with no entity list applies the SAVE attribute only to
the entities that it is allowed on. We were applying it to automatic
data objects and reporting an error that they can't have SAVE.
The fix is to change `DeclarationVisitor::CheckSaveAttr` to check for
automatic objects. That controls both checking and setting the
attribute. This allows us to remove the check from `CheckSpecExpr`
(along with `symbolBeingChecked_`). Also, it was only called on constant
objects so the non-const overload can be eliminated.
The check in `CheckSpecExpr` is replaced by an explicit check for
automatic objects in modules. This caught an error in modfile03.f90 so
that part of the test was eliminated.
Differential Revision: https://reviews.llvm.org/D83899
Hongtao Yu [Sun, 5 Jul 2020 03:24:11 +0000 (20:24 -0700)]
[LoopUnroll] Update branch weight for remainder loop
Unrolling a loop with compile-time unknown trip count results in a remainder loop. The remainder loop executes the remaining iterations of the original loop when the original trip count is not a multiple of the unroll factor. For better profile counts maintenance throughout the optimization pipeline, I'm assigning an artificial weight to the latch branch of the remainder loop.
A remainder loop runs up to as many times as the unroll factor subtracted by 1. Therefore I'm assigning the maximum possible trip count as the back edge weight. This should be more accurate than the default non-profile weight, which assumes the back edge runs much more frequently than the exit edge.
Differential Revision: https://reviews.llvm.org/D83187
Alexey Bataev [Tue, 7 Jul 2020 18:52:45 +0000 (14:52 -0400)]
[OPENMP]Fix PR46593: Reduction initializer missing construnctor call.
Summary:
If user-defined reductions with the initializer are used with classes,
the compiler misses the constructor call when trying to create a private
copy of the reduction variable.
Reviewers: jdoerfert
Subscribers: cfe-commits, yaxunl, guansong, caomhin
Tags: #clang
Differential Revision: https://reviews.llvm.org/D83334
dfukalov [Wed, 15 Jul 2020 15:55:56 +0000 (18:55 +0300)]
[NFC] Fixed typo in tests parameters
Summary:
llc reports `fp32-denormals` is not recognized. I guess it was intended to be
`-denormal-fp-math-f32={preserve-sign|ieee} -mattr=+mad-mac-f32-insts`
Reviewers: rampitec
Reviewed By: rampitec
Subscribers: jvesely, nhaehnle, llvm-commits, kerbowa
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D83883
Uday Bondhugula [Wed, 15 Jul 2020 10:41:08 +0000 (16:11 +0530)]
[MLIR][NFC] Fix clang tidy warnings in misc utilities
Fix clang tidy warnings in misc utilities - missing const or a star in
declaration.
Differential Revision: https://reviews.llvm.org/D83861
Alexey Bataev [Mon, 13 Jul 2020 16:56:18 +0000 (12:56 -0400)]
[OPENMP]Fix PR46688: cast the type of the allocated variable to the initial one.
Summary:
If the original variable is marked for allocation in the different
address space using #pragma omp allocate, need to cast the allocated
variable to its original type with the original address space.
Otherwise, the compiler may crash trying to bitcast the type of the new
allocated variable to the original type in some cases, like passing this
variable as an argument in function calls.
Reviewers: jdoerfert
Subscribers: jholewinski, cfe-commits, yaxunl, guansong, caomhin
Tags: #clang
Differential Revision: https://reviews.llvm.org/D83696
Hiroshi Yamauchi [Tue, 7 Jul 2020 18:26:25 +0000 (11:26 -0700)]
[PGO][PGSO] Add profile guided size optimization to loop vectorization legality.
Differential Revision: https://reviews.llvm.org/D83329
Logan Smith [Wed, 15 Jul 2020 18:33:15 +0000 (11:33 -0700)]
[lldb][NFC] Add 'override' where missing in source/ and tools/
These were found by Clang's new -Wsuggest-override.
This patch doesn't touch any code in unittests/, since much of it intentionally doesn't use override to avoid massive warning spam from -Winconsistent-missing-override due to the use of MOCK_*** macros.
Differential Revision: https://reviews.llvm.org/D83847
Richard Smith [Wed, 15 Jul 2020 18:33:07 +0000 (11:33 -0700)]
Fix "unused variable" warning from recent GCC.
Rahul Joshi [Wed, 15 Jul 2020 18:12:38 +0000 (11:12 -0700)]
[MLIR] Add type checking capability to RegionBranchOpInterface
- Add function `verifyTypes` that Op's can call to do type checking verification
along the control flow edges described the Op's RegionBranchOpInterface.
- We cannot rely on the verify methods on the OpInterface because the interface
functions assume valid Ops, so they may crash if invoked on unverified Ops.
(For example, scf.for getSuccessorRegions() calls getRegionIterArgs(), which
dereferences getBody() block. If the scf.for is invalid with no body, this
can lead to a segfault). `verifyTypes` can be called post op-verification to
avoid this.
Differential Revision: https://reviews.llvm.org/D82829
Sanjay Patel [Wed, 15 Jul 2020 18:09:46 +0000 (14:09 -0400)]
[InstCombine] prevent infinite looping in or-icmp fold (PR46712)
I'm not sure if the test is truly minimal, but we need to
induce a situation where a value becomes a constant but is
not immediately folded before getting to the 'or' transform.
Sanjay Patel [Wed, 15 Jul 2020 17:32:59 +0000 (13:32 -0400)]
[InstCombine] update datalayout in test file; NFC
We need to specify legal integer widths to trigger PR46712,
so add those here. This doesn't appear to affect any existing
tests, and it's not clear why a datalayout would not include
any legal integer widths.
While here, change some variable names that include 'tmp' to
avoid warnings from the auto-generating script for CHECK lines.
LLVM GN Syncbot [Wed, 15 Jul 2020 17:53:38 +0000 (17:53 +0000)]
[gn build] Port
4a539faf74b
Hiroshi Yamauchi [Thu, 25 Jun 2020 19:19:30 +0000 (12:19 -0700)]
[PGO] Extend the value profile buckets for mem op sizes.
Extend the memop value profile buckets to be more flexible (could accommodate a
mix of individual values and ranges) and to cover more value ranges (from 11 to
22 buckets).
Disabled behind a flag (to be enabled separately) and the existing code to be
removed later.
Hiroshi Yamauchi [Tue, 7 Jul 2020 18:13:55 +0000 (11:13 -0700)]
[PGO][PGSO] Add profile guided size optimization to LegalizeDAG.
Reviewers: davidxl
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D83333
Stephan Herhut [Wed, 15 Jul 2020 07:25:39 +0000 (09:25 +0200)]
[mlir][shape] Fold shape.broadcast with one scalar operand
This folds shape.broadcast where at least one operand is a scalar to the
other operand.
Also add an assemblyFormat for shape.broadcast and shape.concat.
Differential Revision: https://reviews.llvm.org/D83854
Dmitry Preobrazhensky [Wed, 15 Jul 2020 16:44:07 +0000 (19:44 +0300)]
[AMDGPU][MC] Corrected MTBUF parsing and decoding
MTBUF implementation has many issues and this change addresses most of these:
- refactored duplicated code;
- hardcoded constants moved out of high-level code;
- fixed a decoding error when nfmt or dfmt are zero (bug 36932);
- corrected parsing of operand separators (bug 46403);
- corrected handling of missing operands (bug 46404);
- corrected handling of out-of-range modifiers (bug 46421);
- corrected default value (bug 46467).
Reviewers: arsenm, rampitec, vpykhtin, artem.tamazov, kzhuravl
Differential Revision: https://reviews.llvm.org/D83760
Arthur Eubanks [Wed, 15 Jul 2020 16:34:44 +0000 (09:34 -0700)]
[NPM][IVUsers] Rename ivusers -> iv-users
LPM passes were named iv-users, which seems nicer than ivusers.
Reviewed By: hans
Differential Revision: https://reviews.llvm.org/D83803
Stephan Herhut [Wed, 15 Jul 2020 07:31:48 +0000 (09:31 +0200)]
[mlir][shape] Mark some operations as commutative
Summary:
This makes sure that their constant arguments are sorted to the back
and hence eases the specification of rewrite patterns.
Differential Revision: https://reviews.llvm.org/D83856
Petr Hosek [Wed, 15 Jul 2020 02:24:22 +0000 (19:24 -0700)]
[CMake][runtimes] Pass the CMAKE_ARGS to runtimes build correctly
We were relying on CMAKE_ARGS argument to be passed to subbuild, but
this argument was never properly defined. This patch addresses that.
Differential Revision: https://reviews.llvm.org/D83842
Jordan Rupprecht [Wed, 15 Jul 2020 16:16:30 +0000 (09:16 -0700)]
[lldb][test] Prevent infinite loop while looking for use_lldb_suite_root.py.
Several scripts (two copies of use_lldb_suite.py, and an __init__.py) look for use_lldb_suite_root.py by checking parent directories. If for some reason it doesn't exist, it keeps checking parent directories until it finds it.
However, this only breaks when the parent directory is None, but at least on Linux, dirname('/') == '/', so this will never be None.
This changes the lookup to stop if the dirname(lldb_root) is unchanged. This was previously fixed in
67f6d842fab6d3ac8c949721be8e131cf6b17578, but only in one copy of this script.
Additionally, this makes the failure mode more visible -- if the root is not found, it complains loudly instead of silently failing, and having later modules that need lldb_root fail.
Differential Revision: https://reviews.llvm.org/D83840
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
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.
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
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.
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.
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
David Green [Wed, 15 Jul 2020 13:49:57 +0000 (14:49 +0100)]
[HardwareLoops] Add sibling loop test.
This missed being part of
9e03547cab69.
Joel E. Denny [Wed, 15 Jul 2020 15:17:00 +0000 (11:17 -0400)]
[OpenMP][Docs] Update `present` map type modifier status
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
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.
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.
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.
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.
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
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
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
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
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
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
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
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
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
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
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.
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.
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
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
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
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.
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
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.
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
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
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/
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
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
Sam McCall [Wed, 15 Jul 2020 09:03:11 +0000 (11:03 +0200)]
[clangd] Fix use-after-free in ArgStripper
Tim Northover [Wed, 15 Jul 2020 08:57:53 +0000 (09:57 +0100)]
Fix typo in identifier in assert.
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.
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
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
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
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
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
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
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
Chen Zheng [Wed, 15 Jul 2020 06:44:17 +0000 (02:44 -0400)]
[IndVarSimplify] Uniformly use emplace_back for DeadInsts, nfc
Giorgis Georgakoudis [Wed, 15 Jul 2020 06:15:59 +0000 (23:15 -0700)]
[OpenMPOpt][NFC] Update checks for parallel_deletion test
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
Giorgis Georgakoudis [Wed, 15 Jul 2020 00:01:15 +0000 (17:01 -0700)]
[OpenMPOpt] Fix preserved analyses return
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
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
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.
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
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.
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
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
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
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.
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
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.
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
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