Simon Pilgrim [Fri, 3 Jul 2020 14:51:06 +0000 (15:51 +0100)]
[InstCombine] canEvaluateTruncated - use KnownBits to check for inrange shift amounts
Currently canEvaluateTruncated can only attempt to truncate shifts if they are scalar/uniform constant amounts that are in range.
This patch replaces the constant extraction code with KnownBits handling, using the KnownBits::getMaxValue to check that the amounts are inrange.
This enables support for nonuniform constant cases, and also variable shift amounts that have been masked somehow. Annoyingly, this still won't work for vectors with (demanded) undefs as KnownBits returns nothing in those cases, but its a definite improvement on what we currently have.
Differential Revision: https://reviews.llvm.org/D83127
Dmitry Preobrazhensky [Fri, 3 Jul 2020 11:36:57 +0000 (14:36 +0300)]
[AMDGPU] Added support of new inline assembler constraints
Added support for constraints 'I', 'J', 'L', 'B', 'C', 'Kf', 'DA', 'DB'.
See https://gcc.gnu.org/onlinedocs/gcc/Machine-Constraints.html#Machine-Constraints.
Reviewers: arsenm, rampitec
Differential Revision: https://reviews.llvm.org/D81657
Petre-Ionut Tudor [Wed, 3 Jun 2020 14:28:43 +0000 (15:28 +0100)]
[ARM] Generate [SU]RHADD from (b - (~a)) >> 1
Summary:
Teach LLVM to recognize the above pattern, which is usually a
transformation of (a + b + 1) >> 1, where the operands are either
signed or unsigned types.
Subscribers: kristof.beyls, hiraditya, danielkiss, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D82669
Pavel Labath [Wed, 1 Jul 2020 14:46:24 +0000 (16:46 +0200)]
[lldb/DWARF] Look for complete member definitions in other modules
With -flimit-debug-info, we can have a definition of a class, but no
definition for some of its members. This extends the same logic we were
using for incomplete base classes to cover incomplete members too.
Test forward-declarations.s is removed as it is no longer applicable --
we don't warn anymore when encountering incomplete members as they could
be completed elsewhere. New checks added to TestLimitDebugInfo cover the
handling of incomplete members more thoroughly.
Julian Gross [Thu, 2 Jul 2020 13:10:24 +0000 (15:10 +0200)]
[mlir] Add check for ViewLikeOpInterface that creates additional aliases.
ViewLikeOpInterfaces introduce new aliases that need to be added to the alias
list. This is necessary to place deallocs in the right positions.
Differential Revision: https://reviews.llvm.org/D83044
Xing GUO [Fri, 3 Jul 2020 10:13:49 +0000 (18:13 +0800)]
[ObjectYAML][ELF] Add support for emitting the .debug_gnu_pubnames/pubtypes sections.
This patch helps add support for emitting the .debug_gnu_pubnames and .debug_gnu_pubtypes sections.
The .debug_gnu_pub* sections is verified by llvm-dwarfdump.
Known issues:
- Doesn't support emitting multiple pub-tables.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D82367
Pavel Labath [Fri, 3 Jul 2020 14:31:50 +0000 (16:31 +0200)]
[lldb/Utility] Simplify more Scalar methods
A lot of the methods handle all integral and all floating point types
the same way. They can be changed to switch on the category of the type,
instead of the actual type, saving a lot of boilerplate.
This patch does that for the methods where I could be reasonably certain
of their expected semantics.
Xing GUO [Fri, 3 Jul 2020 14:34:27 +0000 (22:34 +0800)]
[DWARFYAML][unittest] Use parseDWARFYAML() in unit test. NFC.
Ehsan Toosi [Thu, 25 Jun 2020 15:02:11 +0000 (17:02 +0200)]
[mlir] Add redundant copy removal transform
This pass removes redundant dialect-independent Copy operations in different
situations like the following:
%from = ...
%to = ...
... (no user/alias for %to)
copy(%from, %to)
... (no user/alias for %from)
dealloc %from
use(%to)
Differential Revision: https://reviews.llvm.org/D82757
Sam Parker [Fri, 3 Jul 2020 13:20:57 +0000 (14:20 +0100)]
[NFC][SimplifyCFG] Move X86 tests into subdir
Georgii Rymar [Fri, 3 Jul 2020 11:17:08 +0000 (14:17 +0300)]
[llvm-readobj] - Use cantFail() for all `Obj->sections()` calls. NFCI.
`ELFDumper<ELFT>::ELFDumper` calls `Obj->sections()` in its constructor:
https://github.com/llvm/llvm-project/blob/master/llvm/tools/llvm-readobj/ELFDumper.cpp#L2046
this means that all subsequent calls can't fail and can be
wrapped into `cantFail` in instead of `unwrapOrError` for simplicity.
Actually we already do it in a few places. In this patch I've fixed all
other places I've found.
Differential revision: https://reviews.llvm.org/D83126
Danila Malyutin [Tue, 23 Jun 2020 11:43:02 +0000 (14:43 +0300)]
[IR] Short-circuit comparison with itself for Attributes
Differential Revision: https://reviews.llvm.org/D82295
Bruno Ricci [Fri, 3 Jul 2020 12:58:59 +0000 (13:58 +0100)]
[clang][NFC] Add a missing /dev/null in test/AST/ast-dump-lambda.cpp
Bruno Ricci [Fri, 3 Jul 2020 12:58:19 +0000 (13:58 +0100)]
[clang][NFC] Also test for serialization in test/AST/ast-dump-comment.cpp
Bruno Ricci [Fri, 3 Jul 2020 12:54:10 +0000 (13:54 +0100)]
[clang][NFC] Store a pointer to the ASTContext in ASTDumper and TextNodeDumper
In general there is no way to get to the ASTContext from most AST nodes
(Decls are one of the exception). This will be a problem when implementing
the rest of APValue::dump since we need the ASTContext to dump some kinds of
APValues.
The ASTContext* in ASTDumper and TextNodeDumper is not always non-null.
This is because we still want to be able to use the various dump() functions
in a debugger.
No functional changes intended.
Reverted in
fcf4d5e4499a391dff42ea1a096f146db44147b6 since a few dump()
functions in lldb where missed.
Simon Pilgrim [Fri, 3 Jul 2020 12:39:16 +0000 (13:39 +0100)]
Add tests for trunc(shl/lshr/ashr(*ext(x),zext(and(y,c)))) patterns with variable shifts with clamped shift amounts
Simon Pilgrim [Fri, 3 Jul 2020 10:51:59 +0000 (11:51 +0100)]
Add vector trunc(or(shl(zext(x),c1),zext(x))) tests
Ben Dunbobbin [Fri, 3 Jul 2020 11:54:24 +0000 (12:54 +0100)]
[LLD][ELF][Windows] Allow LLD to overwrite existing output files that are in use
On Windows co-operative programs can be expected to open LLD's
output in FILE_SHARE_DELETE mode. This allows us to delete the
file (by moving it to a temporary filename and then deleting
it) so that we can link another output file that overwrites
the existing file, even if the current file is in use.
A similar strategy is documented here:
https://boostgsoc13.github.io/boost.afio/doc/html/afio/FAQ/deleting_open_files.html
Differential Revision: https://reviews.llvm.org/D82567
vpykhtin [Thu, 25 Jun 2020 14:24:58 +0000 (17:24 +0300)]
[AMDGPU] Don't combine DPP if DPP register is used more than once per instruction
Reviewers: arsenm, rampitec, foad
Reviewed By: rampitec, foad
Subscribers: wuzish, kzhuravl, nemanjai, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, hiraditya, kbarton, kerbowa, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D82551
Luke Geeson [Tue, 30 Jun 2020 15:45:36 +0000 (16:45 +0100)]
[ARM] Add Cortex-A77 Support for Clang and LLVM
This patch upstreams support for the Arm-v8 Cortex-A77
processor for AArch64 and ARM.
In detail:
- Adding cortex-a77 as a cpu option for aarch64 and arm targets in clang
- Cortex-A77 CPU name and ProcessorModel in llvm
details of the CPU can be found here:
https://www.arm.com/products/silicon-ip-cpu/cortex-a/cortex-a77
and a similar submission to GCC can be found here:
https://github.com/gcc-mirror/gcc/commit/
e0664b7a63ed8305e9f8539309df7fb3eb13babe
The following people contributed to this patch:
- Luke Geeson
- Mikhail Maltsev
Reviewers: t.p.northover, dmgreen, ostannard, SjoerdMeijer
Reviewed By: dmgreen
Subscribers: dmgreen, kristof.beyls, hiraditya, danielkiss, cfe-commits,
llvm-commits, miyuki
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D82887
Dmitri Gribenko [Fri, 3 Jul 2020 11:46:59 +0000 (13:46 +0200)]
Revert RecursiveASTVisitor fixes.
This reverts commit
8bf4c40af813e73de77739b33b8808f6bd13497b.
This reverts commit
7b0be962d681c408c8ecf7180c6ad8f9fbcdaf2d.
This reverts commit
94454442c3c15a67ae70ef3a73616632968973fc.
Some compilers on some buildbots didn't accept the specialization of
is_same_method_impl in a non-namespace scope.
Dmitri Gribenko [Fri, 3 Jul 2020 10:39:14 +0000 (12:39 +0200)]
Make RecursiveASTVisitor call WalkUpFrom for operators when the data recursion queue is absent
Reviewers: eduucaldas, ymandel, rsmith
Reviewed By: eduucaldas
Subscribers: gribozavr2, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D82889
Dmitri Gribenko [Fri, 3 Jul 2020 10:39:03 +0000 (12:39 +0200)]
Make RecursiveASTVisitor call WalkUpFrom for unary and binary operators in post-order traversal mode
Reviewers: ymandel, eduucaldas, rsmith
Reviewed By: eduucaldas, rsmith
Subscribers: gribozavr2, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D82787
Dmitri Gribenko [Fri, 3 Jul 2020 10:38:45 +0000 (12:38 +0200)]
RecursiveASTVisitor: don't call WalkUp unnecessarily in post-order traversal
Summary:
How does RecursiveASTVisitor call the WalkUp callback for expressions?
* In pre-order traversal mode, RecursiveASTVisitor calls the WalkUp
callback from the default implementation of Traverse callbacks.
* In post-order traversal mode when we don't have a DataRecursionQueue,
RecursiveASTVisitor also calls the WalkUp callback from the default
implementation of Traverse callbacks.
* However, in post-order traversal mode when we have a DataRecursionQueue,
RecursiveASTVisitor calls the WalkUp callback from PostVisitStmt.
As a result, when the user overrides the Traverse callback, in pre-order
traversal mode they never get the corresponding WalkUp callback. However
in the post-order traversal mode the WalkUp callback is invoked or not
depending on whether the data recursion optimization could be applied.
I had to adjust the implementation of TraverseCXXForRangeStmt in the
syntax tree builder to call the WalkUp method directly, as it was
relying on this behavior. There is an existing test for this
functionality and it prompted me to make this extra fix.
In addition, I had to fix the default implementation implementation of
RecursiveASTVisitor::TraverseSynOrSemInitListExpr to call WalkUpFrom in
the same manner as the implementation generated by the DEF_TRAVERSE_STMT
macro. Without this fix, the InitListExprIsPostOrderNoQueueVisitedTwice
test was failing because WalkUpFromInitListExpr was never called.
Reviewers: eduucaldas, ymandel
Reviewed By: eduucaldas, ymandel
Subscribers: gribozavr2, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D82486
Dmitri Gribenko [Fri, 3 Jul 2020 10:38:35 +0000 (12:38 +0200)]
Added tests for RecursiveASTVisitor for AST nodes that are special cased
Summary:
RecursiveASTVisitor has special code for handling operator AST nodes,
specifically, unary, binary, and compound assignment operators. In this
change I'm adding tests for operator AST nodes that follow the existing
pattern of tests for the CallExpr node (an AST node that triggers the
common code path).
Reviewers: ymandel, eduucaldas
Reviewed By: ymandel, eduucaldas
Subscribers: gribozavr2, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D82875
David Zarzycki [Fri, 3 Jul 2020 10:53:44 +0000 (06:53 -0400)]
[libcxx testing] Remove ALLOW_RETRIES from another test
James Henderson [Thu, 2 Jul 2020 13:13:43 +0000 (14:13 +0100)]
[DebugInfo] Use Cursor to detect errors in debug line prologue parser
Previously, the debug line parser would keep attempting to read data
even if it had run out of data to read. This meant errors in parsing
would often end up being reported as something else, such as an unknown
version or malformed directory/filename table. This patch fixes the
issues by using the Cursor API to capture errors.
Reviewed by: labath
Differential Revision: https://reviews.llvm.org/D83043
Simon Pilgrim [Fri, 3 Jul 2020 10:41:21 +0000 (11:41 +0100)]
Regenerate apint-cast tests and replace %tmp variable names to silence update_test_checks warnings
Simon Pilgrim [Fri, 3 Jul 2020 10:40:17 +0000 (11:40 +0100)]
Add nonuniform vector trunc(or(shl(zext(x),c1),srl(zext(x),c2))) tests
Simon Pilgrim [Fri, 3 Jul 2020 10:37:26 +0000 (11:37 +0100)]
Regenerate mul-trunc tests, add vector variants and replace %tmp variable names to silence update_test_checks warnings
Martin Svensson [Fri, 3 Jul 2020 09:36:34 +0000 (11:36 +0200)]
[lldb] Fix missing characters when autocompleting LLDB commands in REPL
Summary:
When tabbing to complete LLDB commands in REPL, characters would at best be
missing but at worst cause the REPL to crash due to out of range string access.
This patch appends the command character to the completion results to fulfill
the assumption that all matches are prefixed by the request's cursor argument
prefix.
Bug report for the Swift REPL
https://bugs.swift.org/browse/SR-12867
Reviewers: teemperor
Reviewed By: teemperor
Subscribers: lldb-commits
Tags: #lldb
Differential Revision: https://reviews.llvm.org/D82835
Xing GUO [Fri, 3 Jul 2020 09:56:02 +0000 (17:56 +0800)]
[DWARFYAML][debug_gnu_*] Add the missing context `IsGNUStyle`. NFC.
This patch helps add the missing context `IsGNUStyle`. Before this patch, yaml2obj cannot parse the YAML description of 'debug_gnu_pubnames' and 'debug_gnu_pubtypes' correctly due to the missing context.
In other words, if we have
```
DWARF:
debug_gnu_pubtypes:
Length:
TotalLength: 0x1234
Version: 2
UnitOffset: 0x1234
UnitSize: 0x4321
Entries:
- DieOffset: 0x12345678
Name: abc
Descriptor: 0x00 ## Descriptor can never be mapped into Entry.Descriptor
```
yaml2obj will complain that "error: unknown key 'Descriptor'".
This patch helps resolve this problem.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D82435
Simon Pilgrim [Fri, 3 Jul 2020 09:42:54 +0000 (10:42 +0100)]
Fix MSVC "result of 32-bit shift implicitly converted to 64 bits" warning.
Adam Czachorowski [Fri, 3 Jul 2020 09:20:22 +0000 (11:20 +0200)]
[clangd] Improve hover on arguments to function call
Summary:
In cases like:
foo(a, ^b);
We now additionally show the name and type of the parameter to foo that
corresponds that "b" is passed as.
The name should help with understanding what it's used for and type can
be useful to find out if call to foo() can mutate variable "b" or not
(i.e. if it is pass by value, reference, const reference, etc).
Patch By: adamcz@ !
Reviewers: kadircet
Reviewed By: kadircet
Subscribers: nridge, ilya-biryukov, MaskRay, jkorous, arphaman, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D81169
Simon Pilgrim [Thu, 2 Jul 2020 18:24:38 +0000 (19:24 +0100)]
[InstCombine] Add sext(ashr(shl(trunc(x),c),c)) folding support for vectors
Replacing m_ConstantInt with m_Constant permits folding of vectors as well as scalars.
Differential Revision: https://reviews.llvm.org/D83058
Simon Pilgrim [Thu, 2 Jul 2020 17:30:34 +0000 (18:30 +0100)]
Regenerate PR19420 tests
Georgii Rymar [Thu, 2 Jul 2020 13:06:26 +0000 (16:06 +0300)]
[llvm-readelf] - Do not report a misleading warning when there is no string table.
This is a follow-up for D82955, which allows to continue dumping when a symbol table is broken.
When we are unable to get the string table and trying to print symbols,
the existent tool logic together with D82955 reports an error:
"st_name (0x??) is past the end of the string table of size 0x??"
Though, when there is no string table, this message becomes misleading and excessive.
It is easy to fix it though and that is what this patch does.
Differential revision: https://reviews.llvm.org/D83042
Georgii Rymar [Wed, 1 Jul 2020 12:25:33 +0000 (15:25 +0300)]
[llvm-readelf] - Do not error out when dumping symbols.
When the --symbols option/--dyn-symbols is given we might report an
error and exit when something goes not right. E.g. when the SHT_SYMTAB
section is broken. Though we could report a warning and try to continue
dumping instead in many cases.
This patch removes `unwrapOrErr` calls from the code involved in the
flow described.
Differential revision: https://reviews.llvm.org/D82955
Guillaume Chatelet [Fri, 3 Jul 2020 08:54:27 +0000 (08:54 +0000)]
[Alignment][NFC] Use 5 bits to store Instructions Alignment
As per [MaxAlignmentExponent]{https://github.com/llvm/llvm-project/blob/
b7338fb1a6a464472850211165391983d2c8fdf3/llvm/include/llvm/IR/Value.h#L688} alignment is not allowed to be more than 2^29.
Encoded as Log2, this means that storing alignment uses 5 bits.
This patch makes sure all instructions store their alignment in a consistent way, encoded as Log2 and using 5 bits.
Differential Revision: https://reviews.llvm.org/D83119
Jean Perier [Fri, 3 Jul 2020 08:01:35 +0000 (10:01 +0200)]
[flang][NFC] Move and rework pgmath description used in folding
This change prepares usage of lipgmath description in lowering.
- Removes the static variable templates that were used to abstract
libpgmath description
- Move the description to pgmath.h.inc header and rework the macros
so that they can both be used to declare pgmath functions and use
them.
The way they are to be used is left to pgmath.h.inc user that
must define PGMATH_USE_XX macros that will be called for all pgmath
functions in pgmath.h.inc.
- In intrinsic-library.cpp define PGMATH_USE_XX macro callbacks in
order to capture function pointers to pgmath functions as well as
a description of their type. This will be used for constant folding
using pgmath.
- Change atan/atan2 handling to use atan2 instead of atan when there are two
arguments because it is easier to handle in the runtime description.
Also fixes lipgmath linking regression after D78215 cmake changes.
This change is motivated by the need to use a similar pgmath
description in lowering. The difference is that no function pointers will
be taken there, and instead only the function name and type are needed.
Reviewed By: schweitz, sscalpone
Differential Revision: https://reviews.llvm.org/D83051
Guillaume Chatelet [Fri, 3 Jul 2020 08:06:43 +0000 (08:06 +0000)]
[Alignment][NFC] Use proper getter to retrieve alignment from ConstantInt and ConstantSDNode
This patch is part of a series to introduce an Alignment type.
See this thread for context: http://lists.llvm.org/pipermail/llvm-dev/2019-July/133851.html
See this patch for the introduction of the type: https://reviews.llvm.org/D64790
Differential Revision: https://reviews.llvm.org/D83082
Joachim Protze [Mon, 15 Jun 2020 22:31:14 +0000 (00:31 +0200)]
[OpenMP][OMPT]Add event callbacks for taskwait with depend
This adds the missing event callbacks to express dependencies on included tasks
and taskwait with depend clause.
The test fails for GCC, see bug report:
https://bugs.llvm.org/show_bug.cgi?id=46573
Reviewed by: hbae
Differential Revision: https://reviews.llvm.org/D81891
Luofan Chen [Fri, 3 Jul 2020 07:22:18 +0000 (15:22 +0800)]
[Attributor] Create getName() method for abstract attribute
Summary: The `getName()` method returns the name of the abstract attribute
Reviewers: jdoerfert, sstefan1, uenoku, homerdin, baziotis
Reviewed By: sstefan1
Subscribers: uenoku, kuter, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D83109
serge-sans-paille [Tue, 30 Jun 2020 12:04:00 +0000 (14:04 +0200)]
Fix stack-clash probing for large static alloca
Differential Revision: https://reviews.llvm.org/D82867
Guillaume Chatelet [Fri, 3 Jul 2020 07:20:22 +0000 (07:20 +0000)]
[NFC] Use ADT/Bitfields in Instructions
This is an example patch for D81580.
Differential Revision: https://reviews.llvm.org/D81662
Craig Topper [Fri, 3 Jul 2020 07:13:59 +0000 (00:13 -0700)]
[X86] Remove MODRM_SPLITREGM from the disassembler tables.
This offers a very minor table size reduction due to only being
used for one AMX opcode.
Haojian Wu [Fri, 3 Jul 2020 06:54:36 +0000 (08:54 +0200)]
[clang] Check ValueDependent instead of InstantiationDependent before executing the align expr for builtin align functions.
in general, value dependent is a subset of instnatiation dependent. This
would allows us to produce diagnostics for the align expression (which
is instantiation dependent but not value dependent).
Differential Revision: https://reviews.llvm.org/D83074
Sam Parker [Thu, 2 Jul 2020 11:13:23 +0000 (12:13 +0100)]
[CostModel] Fix cast crash
Don't presume instruction operands while matching reductions.
Bugzilla: https://bugs.llvm.org/show_bug.cgi?id=46430
Differential Revision: https://reviews.llvm.org/D82453
Kai Luo [Fri, 3 Jul 2020 05:27:25 +0000 (05:27 +0000)]
[PowerPC] Implement probing for dynamic stack allocation
This patch is part of supporting `-fstack-clash-protection`. Mainly do
such things compared to existing `lowerDynamicAlloc`
- Added a new pseudo instruction PPC::PREPARE_PROBED_ALLOC to get
actual frame pointer and final stack pointer.
- Synthesize a loop to probe by blocks.
- Use DYNAREAOFFSET to get MaxCallFrameSize which is calculated in
prologepilog.
Differential Revision: https://reviews.llvm.org/D81358
Craig Topper [Fri, 3 Jul 2020 05:11:52 +0000 (22:11 -0700)]
[X86] Add back support for matching VPTERNLOG from back to back logic ops.
I think this mostly looks ok. The only weird thing I noticed was
a couple rotate vXi8 tests picked up an extra logic op where we have
(and (or (and), (andn)), X). Previously we matched the (or (and), (andn))
to vpternlog, but now we match the (and (or), X) and leave the and/andn
unmatched.
Carl Ritson [Fri, 3 Jul 2020 03:25:33 +0000 (12:25 +0900)]
[AMDGPU] Insert PS early exit at end of control flow
Exit early if the exec mask is zero at the end of control flow.
Mark the ends of control flow during control flow lowering and
convert these to exits during the insert skips pass.
Reviewed By: nhaehnle
Differential Revision: https://reviews.llvm.org/D82737
Kai Luo [Fri, 3 Jul 2020 04:17:01 +0000 (04:17 +0000)]
[PowerPC][NFC] Prevent unused error when assertion is disabled.
Jez Ng [Fri, 3 Jul 2020 04:19:55 +0000 (21:19 -0700)]
[lld-macho] Support binding dysyms to any section
Previously, we only supported binding dysyms to the GOT. This
diff adds support for binding them to any arbitrary section. C++
programs appear to use this, I believe for vtables and type_info.
This diff also makes our bind opcode encoding a bit smarter -- we now
encode just the differences between bindings, which will make things
more compact.
I was initially concerned about the performance overhead of iterating
over these relocations, but it turns out that the number of such
relocations is small. A quick analysis of my llvm-project build
directory showed that < 1.3% out of ~7M relocations are RELOC_UNSIGNED
bindings to symbols (including both dynamic and static symbols).
Reviewed By: #lld-macho, smeenai
Differential Revision: https://reviews.llvm.org/D83103
Carl Ritson [Fri, 3 Jul 2020 04:03:33 +0000 (13:03 +0900)]
Revert "[AMDGPU] Insert PS early exit at end of control flow"
This reverts commit
2bfcacf0ad362956277a1c2c9ba00ddc453a42ce.
There appears to be an issue to analysis preservation.
Kai Luo [Fri, 3 Jul 2020 03:30:38 +0000 (03:30 +0000)]
[PowerPC][NFC] Refactor lowerDynamicAlloc
When performing dynamic stack allocation, calculation of frame pointer
and actual negsize can be separated. This patch refactors
`lowerDynamicAlloc` in preparation of supporting
`-fstack-clash-protection` which also has to calculate actual frame
pointer and negsize.
Differential Revision: https://reviews.llvm.org/D81354
Carl Ritson [Fri, 3 Jul 2020 03:25:33 +0000 (12:25 +0900)]
[AMDGPU] Insert PS early exit at end of control flow
Exit early if the exec mask is zero at the end of control flow.
Mark the ends of control flow during control flow lowering and
convert these to exits during the insert skips pass.
Reviewed By: nhaehnle
Differential Revision: https://reviews.llvm.org/D82737
Carl Ritson [Fri, 3 Jul 2020 00:57:42 +0000 (09:57 +0900)]
[AMDGPU] Unify early PS termination blocks
Generate a single early exit block out-of-line and branch to this
if all lanes are killed. This avoids branching if lanes are active.
Reviewed By: nhaehnle
Differential Revision: https://reviews.llvm.org/D82641
peter klausler [Thu, 2 Jul 2020 22:51:07 +0000 (15:51 -0700)]
[flang] External I/O runtime work, repackaged (part 2)
Clean up the input editing path so external input works better
when combined with further changes. List-directed input needed
to allow for advancement to following records.
Reviewed By: tskeith, sscalpone
Differential Revision: https://reviews.llvm.org/D83104
Roman Lebedev [Thu, 2 Jul 2020 23:37:29 +0000 (02:37 +0300)]
[NFC][Scalarizer] Also scalarize loads in newly-added tests
Should help better showcase improvements
Roman Lebedev [Thu, 2 Jul 2020 22:53:13 +0000 (01:53 +0300)]
[NFC][Scalarizer] Add some insertelement/extractelement tests
See D82961/D82970/D83101/D83102.
Nico Weber [Thu, 2 Jul 2020 22:52:05 +0000 (18:52 -0400)]
[gn build] get everything to build when llvm_targets_to_build is just AArch64
Craig Topper [Thu, 2 Jul 2020 22:35:47 +0000 (15:35 -0700)]
[X86] Teach lower512BitShuffle to try bitmask and bitblend before splitting v32i16/v64i8 on av512f only targets.
We consider v32i16/v64i8 to be legal types on avx512f, but we
don't have most operations until avx512bw. But we can use
and/or/xor operations. So try those before splitting.
This is especially helpful since we turn some ands with constant
masks into shuffles in early DAG combines. So we should make sure
we recover those back to AND.
peter klausler [Thu, 2 Jul 2020 21:11:14 +0000 (14:11 -0700)]
[flang] External I/O runtime work, repackaged (part 1)
Add a isFixedRecordLength flag member to Connection to
disambiguate the state of "record has known variable length"
from "record has fixed length". Code that sets and tests this
flag will appear in later patches. Rearrange data members to
reduce storage requirements, since Connection might indirectly
end up on a program stack frame. Add a utility member function
BeginRecord(); use it in internal I/O processing.
Reviewed By: tskeith, sscalpone
Differential Revision: https://reviews.llvm.org/D83098
Biplob Mishra [Thu, 2 Jul 2020 21:14:27 +0000 (16:14 -0500)]
[PowerPC] Implement Vector Blend Builtins in LLVM/Clang
Implements vec_blendv()
Differential Revision: https://reviews.llvm.org/D82774
Sameer Arora [Thu, 2 Jul 2020 21:42:01 +0000 (14:42 -0700)]
Fix typo and check commit access.
Sanjay Patel [Thu, 2 Jul 2020 21:45:57 +0000 (17:45 -0400)]
[x86] remove redundant tests with no check lines; NFC
These were accidentally included with:
rGb93e6650c8ac
Sanjay Patel [Thu, 2 Jul 2020 20:48:09 +0000 (16:48 -0400)]
[SelectionDAG] don't split branch on logic-of-vector-compares
SelectionDAGBuilder converts logic-of-compares into multiple branches based
on a boolean TLI setting in isJumpExpensive(). But that probably never
considered the pattern of extracted bools from a vector compare - it seems
unlikely that we would want to turn vector logic into control-flow.
The motivating x86 reduction case is shown in PR44565:
https://bugs.llvm.org/show_bug.cgi?id=44565
...and that test shows the expected improvement from using pmovmsk codegen.
For AArch64, I modified the test to include an extra op because the simpler
test gets transformed by a codegen invocation of SimplifyCFG.
Differential Revision: https://reviews.llvm.org/D82602
Amy Kwan [Thu, 2 Jul 2020 20:15:14 +0000 (15:15 -0500)]
[PowerPC]Add Vector Insert Instruction Definitions and MC Test
Adds td definitions and asm/disasm tests for the following instructions:
VINSBVLX
VINSBVRX
VINSHVLX
VINSHVRX
VINSWVLX
VINSWVRX
VINSBLX
VINSBRX
VINSHLX
VINSHRX
VINSWLX
VINSWRX
VINSDLX
VINSDRX
VINSW
VINSD
Differential Revision: https://reviews.llvm.org/D83052
Craig Topper [Thu, 2 Jul 2020 20:43:27 +0000 (13:43 -0700)]
[X86] Add vpternlog to the broadcast unfolding table.
Craig Topper [Thu, 2 Jul 2020 20:29:30 +0000 (13:29 -0700)]
[X86] Add test case for unfolding broadcast load from vpternlog.
Vedant Kumar [Thu, 2 Jul 2020 20:28:48 +0000 (13:28 -0700)]
[test] Deflake test/profile/ContinuousSyncMode/online-merging.c
This test spawns 32 child processes which race to update counters on
shared memory pages. On some Apple-internal machines, two processes race
to perform an update in approximately 0.5% of the test runs, leading to
dropped counter updates. Deflake the test by using atomic increments.
Tested with:
```
$ for I in $(seq 1 1000); do echo ":: Test run $I..."; ./bin/llvm-lit projects/compiler-rt/test/profile/Profile-x86_64h/ContinuousSyncMode/online-merging.c -av || break; done
```
rdar://
64956774
Nikita Popov [Thu, 2 Jul 2020 20:21:23 +0000 (22:21 +0200)]
[InstSimplify] Add test for sext/zext comparisons (NFC)
aartbik [Thu, 2 Jul 2020 20:21:14 +0000 (13:21 -0700)]
[mlir] [VectorOps] Add choice between dot and axpy lowering of vector.contract
Default vector.contract lowering essentially yields a series of sdot/ddot
operations. However, for some layouts a series of saxpy/daxpy operations,
chained through fma are more efficient. This CL introduces a choice between
the two lowering paths. A default heuristic is to follow.
Some preliminary avx2 performance numbers for matrix-times-vector.
Here, dot performs best for 64x64 A x b and saxpy for 64x64 A^T x b.
```
------------------------------------------------------------
A x b A^T x b
------------------------------------------------------------
GFLOPS sdot (reassoc) saxpy sdot (reassoc) saxpy
------------------------------------------------------------
1x1 0.6 0.9 0.6 0.9
2x2 2.5 3.2 2.4 3.5
4x4 6.4 8.4 4.9 11.8
8x8 11.7 6.1 5.0 29.6
16x16 20.7 10.8 7.3 43.3
32x32 29.3 7.9 6.4 51.8
64x64 38.9 79.3
128x128 32.4 40.7
------------------------------------------------------------
```
Reviewed By: nicolasvasilache, ftynse
Differential Revision: https://reviews.llvm.org/D83012
Sanjay Patel [Wed, 1 Jul 2020 22:17:16 +0000 (18:17 -0400)]
[x86] add tests for vector select with bit-test condition; NFC
Craig Topper [Thu, 2 Jul 2020 19:23:02 +0000 (12:23 -0700)]
[X86] Modify the conditions for when we stop making v16i8/v32i8 rotate Custom based on having avx512 features.
The comments here indicate that we prefer to promote the shifts
instead of allowing rotate to be pattern matched. But we weren't
taking into account whether 512-bit registers are enabled or
whethever we have vpsllvw/vpsrlvw instructions.
splatvar_rotate_v32i8 is a slight regrssion, but the other cases
are neutral or improved.
Craig Topper [Thu, 2 Jul 2020 18:26:32 +0000 (11:26 -0700)]
[X86] Add test cases for v32i8 rotate with min-legal-vector-width=256
We currently don't mark ROTL as custom when avx512bw is enabled
under the assumption we'll be able to promote the shifts in the
rotate idiom. But if we don't have 512-bit registers enabled we
can't promote.
Lei Zhang [Thu, 2 Jul 2020 19:42:10 +0000 (15:42 -0400)]
Revert "[MLIR][SPIRV] Support two memory access attributes in OpCopyMemory."
This reverts commit
ef2f46e1f6a63040734c48ed53893298df14b6fa, which
likely triggers a compiler internal error for MSVC.
Differential Revision: https://reviews.llvm.org/D83075
Biplob Mishra [Thu, 2 Jul 2020 19:49:47 +0000 (14:49 -0500)]
[PowerPC]Implement Vector Permute Extended Builtin
Implements vector permute builtin: vec_permx()
Differential Revision: https://reviews.llvm.org/D82869
Nico Weber [Thu, 2 Jul 2020 19:30:39 +0000 (15:30 -0400)]
ld64.lld: Make janky support for tbd files actually work sometimes
Also fix a bug in the test input that made the test miss this issue.
peter klausler [Thu, 2 Jul 2020 16:52:48 +0000 (09:52 -0700)]
[flang] Clean up binary dependences of runtime libraries
There were dependences upon LLVM libraries in the Fortran
runtime support library binaries due to some indirect #includes
of llvm/Support/raw_ostream.h, which caused some kind of internal
ABI version consistency checking to get pulled in. Fixed by
cleaning up some includes.
Reviewed By: tskeith, PeteSteinfeld, sscalpone
Differential Revision: https://reviews.llvm.org/D83060
Tim Keith [Thu, 2 Jul 2020 19:24:34 +0000 (12:24 -0700)]
[flang] Fix bug determining alternate return
The arguments have been moved out of the analyzer so we can't get the
expected number there. Instead use the argument count from the newly
built callee.
Differential Revision: https://reviews.llvm.org/D83063
Bruno Ricci [Thu, 2 Jul 2020 18:38:46 +0000 (19:38 +0100)]
Revert "[clang][NFC] Store a pointer to the ASTContext in ASTDumper and TextNodeDumper"
This reverts commit
aa7fd905e4e1bc510448431da9310e8cf5197523.
I missed some dump() functions.
David Tenty [Thu, 2 Jul 2020 18:35:30 +0000 (14:35 -0400)]
[DebugInfo] Fix LineTest byteswap for cross-targeting builds
Summary:
The byte swap fix for big endian hosts in
9782c922cb21 (for D81570)
swaps based on the host endianess, but for cross-targeting builds (i.e.
big endian host targeting little endian) the host-endianess won't
necessarily match the generated DWARF. This change updates the test
to use symmetrical constants so the results aren't endian dependent.
Reviewers: jhenderson, hubert.reinterpretcast, stevewan, ikudrin
Reviewed By: ikudrin
Subscribers: ikudrin, aprantl, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D82827
Bruno Ricci [Thu, 2 Jul 2020 17:55:07 +0000 (18:55 +0100)]
[clang][NFC] Store a pointer to the ASTContext in ASTDumper and TextNodeDumper
In general there is no way to get to the ASTContext from most AST nodes
(Decls are one of the exception). This will be a problem when implementing
the rest of APValue::dump since we need the ASTContext to dump some kinds of
APValues.
The ASTContext* in ASTDumper and TextNodeDumper is not always
non-null. This is because we still want to be able to use the various
dump() functions in a debugger.
No functional changes intended.
Arthur Eubanks [Thu, 2 Jul 2020 18:27:38 +0000 (11:27 -0700)]
[NewPM] Add -basic-aa to pr33196.ll
The legacy pass manager implicitly adds BasicAA, but the new PM does
not. This causes pr33196.ll to fail under NPM.
There are almost certainly lots of other failures like this, wanted to
get some input on if adding -basic-aa to tests makes sense at scale.
Reviewed By: fhahn
Differential Revision: https://reviews.llvm.org/D82915
George Mitenkov [Thu, 2 Jul 2020 18:21:35 +0000 (14:21 -0400)]
[MLIR][SPIRVToLLVM] Convert spv.constant scalars and vectors
This patch introduces conversion pattern for `spv.constant` with scalar
and vector types. There is a special case when the constant value is a
signed/unsigned integer (vector of integers). Since LLVM dialect does not
have signedness semantics, the types had to be converted to signless ints.
Reviewed By: antiagainst
Differential Revision: https://reviews.llvm.org/D82936
Nico Weber [Thu, 2 Jul 2020 18:22:31 +0000 (14:22 -0400)]
[gn build] make building on an arm mac work
Currently requires `llvm_targets_to_build = [ "X86", "AArch64" ]`:
building just the host arch (i.e. aarch64) causes some linker errors.
Arthur Eubanks [Thu, 2 Jul 2020 18:15:29 +0000 (11:15 -0700)]
[NewPM][LSR] Rename strength-reduce -> loop-reduce
The legacy pass was called "loop-reduce".
This lowers the number of check-llvm failures under NPM by 83.
Reviewed By: ychen
Differential Revision: https://reviews.llvm.org/D82925
sstefan1 [Thu, 2 Jul 2020 17:50:39 +0000 (19:50 +0200)]
[OpenMPOpt][Fix] Remove double initialization of omp::types.
Jonas Hahnfeld [Thu, 2 Jul 2020 17:34:03 +0000 (19:34 +0200)]
[OpenMP][CMake] Fix version detection of testing compiler
When configuring in-tree, the correct names are LLVM_VERSION_MAJOR
and LLVM_VERSION_MINOR. This has been wrong since the code was added
in commits
fc473dee98 and
821649229e.
Nemanja Ivanovic [Thu, 2 Jul 2020 17:28:17 +0000 (12:28 -0500)]
[LLD] Add required dependency after shared libs break due to
ba5087f13025
The dependency on TextAPI was not added and is required for shared
libs builds.
Nemanja Ivanovic [Thu, 2 Jul 2020 15:14:54 +0000 (10:14 -0500)]
[PowerPC] Remove undefs from splat input when changing shuffle mask
As of
1fed131660b2c5d3ea7007e273a7a5da80699445, we have code that
changes shuffle masks so that we can put the shuffle in a canonical
form that can be matched to a single instruction. However, it
does not properly account for undef elements in the BUILD_VECTOR
that is the RHS splat so we can end up with undefs where they
shouldn't be. This patch converts the splat input with undefs to
one without.
ergawy [Thu, 2 Jul 2020 17:14:29 +0000 (13:14 -0400)]
[MLIR][SPIRV] Support two memory access attributes in OpCopyMemory.
This commit augments spv.CopyMemory's implementation to support 2 memory
access operands. Hence, more closely following the spec. The following
changes are introduces:
- Customize logic for spv.CopyMemory serialization and deserialization.
- Add 2 additional attributes for source memory access operand.
Reviewed By: antiagainst
Differential Revision: https://reviews.llvm.org/D82710
George Mitenkov [Thu, 2 Jul 2020 16:34:27 +0000 (12:34 -0400)]
[MLIR][SPIRVToLLVM] SPIR-V function call conversion pattern
Added conversion pattern for SPIR-V `FunctionCallOp`. Based on
specification, it returns no results or a single result, so
can be mapped directly to LLVM dialect's `llvm.call`.
Reviewed By: antiagainst, ftynse
Differential Revision: https://reviews.llvm.org/D83030
George Mitenkov [Thu, 2 Jul 2020 16:19:05 +0000 (12:19 -0400)]
[MLIR][SPIRVToLLVM] Implementation of spv.BitFieldInsert pattern
This patch introduces conversion pattern for `spv.BitFiledInsert` op,
as well as some utility functions to facilitate code reading.
Since `spv.BitFiledInsert` may take both vector and integer operands,
this case was specifically handled by broadcasting values (`count`
and `offset` here) to vectors. Moreover, the types had to be converted
to same bitwidth in order to conform with LLVM dialect rules.
This was done with `zext` when extending (Note that `count` and
`offset` are treated as unsigned) and `trunc` in the opposite case.
For the latter one, truncation is safe since the op is defined only when
`count`/`offset`/their sum is less than the bitwidth of the result.
This introduces a natural bound of the value of 64, which can be
expressed as `i8`.
Reviewed By: antiagainst, ftynse
Differential Revision: https://reviews.llvm.org/D82639
Simon Pilgrim [Thu, 2 Jul 2020 16:11:38 +0000 (17:11 +0100)]
[InstCombine] Add some sext/trunc tests to show missing support for non-uniform vectors
Pavel Labath [Mon, 29 Jun 2020 14:17:29 +0000 (16:17 +0200)]
[lldb] Fix type conversion in the Scalar getters
Summary:
The Scalar class claims to follow the C type conversion rules. This is
true for the Promote function, but it is not true for the implicit
conversions done in the getter methods.
These functions had a subtle bug: when extending the type, they used the
signedness of the *target* type in order to determine whether to do
sign-extension or zero-extension. This is not how things work in C,
which uses the signedness of the *source* type. I.e., C does
(sign-)extension before it does signed->unsigned conversion, and not the
other way around.
This means that: (unsigned long)(int)-1
is equal to (unsigned long)0xffffffffffffffff
and not (unsigned long)0x00000000ffffffff
Unsurprisingly, we have accumulated code which depended on this
inconsistent behavior. It mainly manifested itself as code calling
"ULongLong/SLongLong" as a way to get the value of the Scalar object in
a primitive type that is "large enough". Previously, the ULongLong
conversion did not do sign-extension, but now it does.
This patch makes the Scalar getters consistent with the declared
semantics, and fixes the couple of call sites that were using it
incorrectly.
Reviewers: teemperor, JDevlieghere
Subscribers: lldb-commits
Tags: #lldb
Differential Revision: https://reviews.llvm.org/D82772
Sander de Smalen [Thu, 2 Jul 2020 13:40:08 +0000 (14:40 +0100)]
[AArch64][SVE] NFC: Rename isOrig -> isReverseInstr
This is a non-functional to clarify some of the terminology in the
AArch64SVEInstrInfo/SVEInstrFormats.td files around the tables
for mapping an instruction to it's reverse instruction counter part,
and vice versa. e.g. DIV -> DIVR and DIVR -> DIV.
Reviewers: paulwalker-arm, cameron.mcinally, rengolin, efriedma
Reviewed By: paulwalker-arm, efriedma
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D82979
Simon Pilgrim [Thu, 2 Jul 2020 15:56:33 +0000 (16:56 +0100)]
[InstCombine] Add (vXi1 trunc(lshr(x,c))) -> icmp_eq(and(x,c')) support for non-uniform vectors
As noted on PR46531, we were only performing this transform on uniform vectors as we were using the m_APInt pattern matcher to extract the shift amount.
Differential Revision: https://reviews.llvm.org/D83035
Simon Pilgrim [Thu, 2 Jul 2020 14:09:52 +0000 (15:09 +0100)]
[InstCombine] Add some (vXi1 trunc(lshr(x,c))) -> icmp_eq(and(x,c')) tests for vectors with undef elements
Suggested on D83035