jasonliu [Fri, 21 Feb 2020 17:20:45 +0000 (17:20 +0000)]
[XCOFF][AIX] Put undefined symbol name into StringTable when neccessary
Summary:
When we have a long name for the undefined symbol, we would hit this assertion:
Assertion failed: I != StringIndexMap.end() && "String is not in table!"
This patch addresses that.
Reviewed by: DiggerLin, daltenty
Differential Revision: https://reviews.llvm.org/D74924
Nagy Mostafa [Fri, 21 Feb 2020 17:58:00 +0000 (09:58 -0800)]
[MLIR] Allow Loop dialect IfOp and ForOp to define values
This patch implements the RFCs proposed here:
https://llvm.discourse.group/t/rfc-modify-ifop-in-loop-dialect-to-yield-values/463
https://llvm.discourse.group/t/rfc-adding-operands-and-results-to-loop-for/459/19.
It introduces the following changes:
- All Loop Ops region, except for ReduceOp, terminate with a YieldOp.
- YieldOp can have variadice operands that is used to return values out of IfOp and ForOp regions.
- Change IfOp and ForOp syntax and representation to define values.
- Add unit-tests and update .td documentation.
- YieldOp is a terminator to loop.for/if/parallel
- YieldOp custom parser and printer
Lowering is not supported at the moment, and will be in a follow-up PR.
Thanks.
Reviewed By: bondhugula, nicolasvasilache, rriddle
Differential Revision: https://reviews.llvm.org/D74174
Francesco Petrogalli [Tue, 18 Feb 2020 19:13:39 +0000 (19:13 +0000)]
[llvm][CodeGen] DAG Combiner folds for vscale.
Summary:
This patch simplifies the DAGs generated when using the intrinsic `@llvm.vscale.*` as follows:
* Fold (add (vscale * C0), (vscale * C1)) to (vscale * (C0 + C1)).
* Canonicalize (sub X, (vscale * C)) to (add X, (vscale * -C)).
* Fold (mul (vscale * C0), C1) to (vscale * (C0 * C1)).
* Fold (shl (vscale * C0), C1) to (vscale * (C0 << C1)).
The test `sve-gep-ll` have been updated to reflect the folding introduced by this patch.
Reviewers: efriedma, sdesmalen, andwar, rengolin
Reviewed By: sdesmalen
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D74782
Nikita Popov [Thu, 20 Feb 2020 21:19:10 +0000 (22:19 +0100)]
[InstCombine] Improve simplify demanded bits worklist management
This fixes a small mistake from D72944: The worklist add should
happen before assigning the new operand, not after.
In case an actual replacement happens, the old operand needs to
be added for DCE. If no actual replacement happens, then old/new
are the same, so it doesn't matter.
This drops one iteration from the annotated test case.
Hiroshi Yamauchi [Wed, 12 Feb 2020 22:55:53 +0000 (14:55 -0800)]
[BFI] Fix missed BFI updates in MachineSink.
Summary:
This prevents BFI queries on new blocks (from
MachineSinking::GetAllSortedSuccessors) and fixes a bunch of assert failures
under -check-bfi-unknown-block-queries=true.
Reviewers: davidxl
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D74511
Fangrui Song [Thu, 20 Feb 2020 22:54:22 +0000 (14:54 -0800)]
[Clang interpreter] Rename Block.{h,cpp} to InterpBlock.{h,cpp}
The Blocks runtime provide a header named Block.h.
It is generally preferable to avoid name collision with system headers
(reducing reliance on -isystem order, more friendly when navigating files in
an editor, etc).
Reviewed By: gribozavr2
Differential Revision: https://reviews.llvm.org/D74934
Nikita Popov [Sun, 16 Feb 2020 09:16:17 +0000 (10:16 +0100)]
[InstCombine] Use replaceOperand() in more places
Followup to D73919 with another batch of replacements of
setOperand() -> replaceOperand(), to make sure the old
operand gets DCEd right away.
Differential Revision: https://reviews.llvm.org/D74932
Florian Hahn [Fri, 21 Feb 2020 17:34:34 +0000 (17:34 +0000)]
[DSE,MSSA] Dbg counters required assertions. Mark test accordingly.
Florian Hahn [Fri, 21 Feb 2020 17:21:44 +0000 (17:21 +0000)]
[VectorUtils] Move ToVectorTy to VectorUtils.h (NFC).
ToVectorTy is defined and used in multiple places. Hoist it to
VectorUtils.h to avoid duplication and improve re-usability.
Reviewers: rengolin, hsaito, Ayal, gilr, fpetrogalli
Reviewed By: fpetrogalli
Differential Revision: https://reviews.llvm.org/D74959
Nikita Popov [Fri, 21 Feb 2020 17:11:51 +0000 (18:11 +0100)]
[X86] Fix SDLoc initialization
Fixes -Wparentheses warning, in this case indicating a genuine
bug.
Nikita Popov [Tue, 18 Feb 2020 21:20:55 +0000 (22:20 +0100)]
[SimplifyLibCalls][IRBuilder] Accept any IRBuilder in SimplifyLibCalls
This changes the SimplifyLibCalls utility to accept an IRBuilderBase,
which allows us to pass through the IRBuilder used by InstCombine.
This will ensure that new instructions get added to the worklist.
The annotated test-case drops from 4 to 2 InstCombine iterations thanks
to this.
To achieve this, I'm adding an IRBuilderBase::OperandBundlesGuard,
which is basically the same as the existing InsertPointGuard and
FastMathFlagsGuard, but for operand bundles. Also add a
setDefaultOperandBundles() method so these can be set outside the
constructor.
Differential Revision: https://reviews.llvm.org/D74792
LLVM GN Syncbot [Fri, 21 Feb 2020 17:21:54 +0000 (17:21 +0000)]
[gn build] Port
23444edf30b
Jonas Paulsson [Thu, 20 Feb 2020 18:40:30 +0000 (10:40 -0800)]
[SystemZ] Return scalarized costs for vector instructions on older archs.
A cost query for a vector instruction should return a cost even without
target vector support, and not trigger an assert.
VectorCombine does this with an input containing source code vectors.
Review: Ulrich Weigand
Matt Arsenault [Tue, 18 Feb 2020 14:34:31 +0000 (09:34 -0500)]
AMDGPU: Use default operand for VOP3P clamp
We don't use this, and matching from the def doesn't make much sense.
There are multiple tablegen bugs with default operand
handling. undef_tied_input should work to handle the vdst_in
correctly, but this breaks the operand register class constraint which
it should be able to infer.
Danilo Carvalho Grael [Fri, 21 Feb 2020 16:59:52 +0000 (11:59 -0500)]
[AArch64][SVE] Add intrinsics for SVE2 bitwise ternary operations
Summary:
Add intrinsics for the following operations:
- eor3, bcax
- bsl, bsl1n, bsl2n, nbsl
Fix MC tests for bsl instructions.
Reviewers: kmclaughlin, c-rhodes, sdesmalen, efriedma, rengolin
Reviewed By: efriedma
Subscribers: tschuett, kristof.beyls, hiraditya, rkruppe, psnobl, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D74785
Sanjay Patel [Fri, 21 Feb 2020 16:49:11 +0000 (11:49 -0500)]
[VectorCombine] refactor matching code to reduce duplication; NFC
cmp/binop were already diverging even though they are largely
the same logic.
Yitzhak Mandelbaum [Wed, 19 Feb 2020 16:04:51 +0000 (11:04 -0500)]
[AST matchers] Add basic matchers for googletest EXPECT/ASSERT calls.
Summary:
This revision adds matchers that match calls to the gtest EXPECT and ASSERT
macros almost like function calls. The matchers are placed in separate files
(GtestMatchers...), because they are specific to the gtest library.
Reviewers: gribozavr2
Subscribers: mgorny, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D74840
Florian Hahn [Fri, 21 Feb 2020 16:55:18 +0000 (16:55 +0000)]
[DSE,MSSA] Add debug counter.
Can be used like
-debug-counter=dse-memoryssa-skip=10,dse-memoryssa-counter-count=20
Reviewers: dmgreen, rnk, efriedma, bryant, asbirlea
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D72147
Cameron McInally [Fri, 21 Feb 2020 16:55:25 +0000 (10:55 -0600)]
[AArch64][SVE] Add +fullfp16 to sve-vector-splat.ll
Add +fullfp16 to sve-vector-splat.ll so we can test folding of immediates into moves.
This attribute can go away later when SVE has a full set of fp16 patterns in place.
Differential Revision: https://reviews.llvm.org/D74965
Jay Foad [Thu, 20 Feb 2020 20:21:30 +0000 (20:21 +0000)]
GlobalISel: Fix narrowing of (G_ASHR i64:x, 32)
Reviewers: arsenm
Subscribers: jvesely, wdng, nhaehnle, rovka, hiraditya, volkan, kerbowa, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D74950
Matt Arsenault [Thu, 20 Feb 2020 14:58:03 +0000 (09:58 -0500)]
AMDGPU/GlobalISel: Commit test changes I forgot to squash
These should have been in
ac7abe0ba9ae4c6a2248cc3ef4e4fe7e6d270105
Matt Arsenault [Fri, 21 Feb 2020 02:12:04 +0000 (21:12 -0500)]
AMDGPU/GlobalISel: Fix xnor matching
We should try the generated matchers before the manual selection. This
means the patterns are now handling the common cases, but the manual
selection code is not yet dead. It's still handling the non-s32/s64
cases (like v2s16 and v2s32). Currently tablegen doesn't have a nice
way to have a single pattern that covers multiple types.
Simon Pilgrim [Fri, 21 Feb 2020 15:30:56 +0000 (15:30 +0000)]
[TargetLowering] Apply basic shift combines before recursive SimplifyDemandedBits calls.
Minor refactor/cleanup before we begin adding non-uniform support.
Fangrui Song [Thu, 20 Feb 2020 07:14:49 +0000 (23:14 -0800)]
[ELF] Shuffle .init_array/.fini_array with --shuffle-sections=
Useful for detecting static initialization order fiasco.
Reviewed By: grimar
Differential Revision: https://reviews.llvm.org/D74887
Ivan Hernandez [Fri, 21 Feb 2020 16:13:05 +0000 (08:13 -0800)]
[lldb-vscode] Use libOption with tablegen to parse command line options.
This change will bring lldb-vscode in line with how several other llvm
tools process command line arguments and make it easier to add future
options.
Differential revision: https://reviews.llvm.org/D74798
Fangrui Song [Mon, 17 Feb 2020 21:56:01 +0000 (13:56 -0800)]
[ELF] Warn changed output section address
When the output section address (addrExpr) is specified, GNU ld warns if
sh_addr is different. This patch implements the warning.
Note, LinkerScript::assignAddresses can be called more than once. We
need to record the changed section addresses, and only report the
warnings after the addresses are finalized.
Reviewed By: grimar
Differential Revision: https://reviews.llvm.org/D74741
Fangrui Song [Mon, 17 Feb 2020 18:21:31 +0000 (10:21 -0800)]
[ELF] Ignore the maximum of input section alignments for two cases
Follow-up for D74286.
Notations:
* alignExpr: the computed ALIGN value
* max_input_align: the maximum of input section alignments
This patch changes the following two cases to match GNU ld:
* When ALIGN is present, GNU ld sets output sh_addr to alignExpr, while lld use max(alignExpr, max_input_align)
* When addrExpr is specified but alignExpr is not, GNU ld sets output sh_addr to addrExpr, while lld uses `advance(0, max_input_align)`
Note, sh_addralign is still set to max(alignExpr, max_input_align).
lma-align.test is enhanced a bit to check we don't overalign sh_addr.
fixSectionAlignments() sets addrExpr but not alignExpr for the `!hasSectionsCommand` case.
This patch sets alignExpr as well so that max_input_align will be respected.
Reviewed By: grimar
Differential Revision: https://reviews.llvm.org/D74736
Matt Arsenault [Fri, 21 Feb 2020 16:09:20 +0000 (11:09 -0500)]
AMDGPU/GlobalISel: Precommit xnor matching test
David Green [Fri, 21 Feb 2020 16:07:09 +0000 (16:07 +0000)]
[ARM] Correct Formatting. NFC
Also removed an unnecessary TODO that I don't believe is relevant for
the instruction in question.
Matt Arsenault [Sat, 15 Feb 2020 22:14:17 +0000 (17:14 -0500)]
AMDGPU/GlobalISel: Manually select G_BUILD_VECTOR_TRUNC
We have patterns for s_pack* selection, but they assume the inputs are
a build_vector with 16-bit inputs, not a truncating build
vector. Since there's still outstanding work for how to handle
mismatched result and source element vector operations, and since I'm
trying a different packed vector strategy than SelectionDAG, just
manually select this for now.
Matt Arsenault [Thu, 20 Feb 2020 23:59:08 +0000 (18:59 -0500)]
AMDGPU/GlobalISel: Legalize G_FPOW
There are few differences from the DAG handling. First, the DAG
handling uses a primitive selection pattern instead of custom
legalizing it. Because of this, this makes use of source modifiers
while the DAG does not.
Also instead of promoting f16, try to use the f16 log/exp. There's no
f16 fmul_legacy, so widen just for the multiply, although I'm not sure
that's the best solution.
Matt Arsenault [Thu, 20 Feb 2020 22:40:43 +0000 (17:40 -0500)]
AMDGPU/GlobalISel: Select llvm.amdgcn.fmul.legacy
Matt Arsenault [Thu, 20 Feb 2020 20:34:51 +0000 (15:34 -0500)]
AMDGPU/GlobalISel: Fix constant bus violation with source modifiers
This looked through copies to find the source modifiers, which may
have been SGPR->VGPR copies added to avoid potential constant bus
violations. Re-insert a copy to a VGPR if this happens.
Alexandre Ganea [Fri, 21 Feb 2020 15:05:07 +0000 (10:05 -0500)]
[mlir] Silence error: call to constructor of 'llvm::APInt' is ambiguous
I was getting this error when using Clang 9.0.1 for compiling.
F:\llvm-project\mlir\lib\IR\Builders.cpp(119,27): error: call to constructor of 'llvm::APInt' is ambiguous
APInt(32, value, /*isSigned=*/false));
^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
F:\llvm-project\llvm\include\llvm/ADT/APInt.h(277,3): note: candidate constructor
APInt(unsigned numBits, uint64_t val, bool isSigned = false)
^
F:\llvm-project\llvm\include\llvm/ADT/APInt.h(304,3): note: candidate constructor
APInt(unsigned numBits, unsigned numWords, const uint64_t bigVal[]);
^
1 error generated.
Eric Astor [Thu, 20 Feb 2020 14:43:02 +0000 (09:43 -0500)]
Remove unused functions in llvm-ml
On review, these functions will likely not be needed even in the final MasmParser.
Pavel Labath [Fri, 21 Feb 2020 14:42:07 +0000 (15:42 +0100)]
[lldb/DWARF] Add support for type units in dwp files
all that was needed was to teach lldb's DWARF context about the
debug_tu_index section.
Sean Fertile [Thu, 20 Feb 2020 14:18:10 +0000 (09:18 -0500)]
[PowerPC][NFC] Add a test for vrsave usage iinline asm.
Add a lit test that that uses vrsave register in the clobber list, and
tests the extended mnemonics mtvrsave and mfvrsave.
Gabor Marton [Fri, 21 Feb 2020 14:54:58 +0000 (15:54 +0100)]
[AST][NFC] Update outdated comments in ASTStructuralEquivalence.cpp
Sean Fertile [Thu, 20 Feb 2020 20:33:18 +0000 (15:33 -0500)]
[PowerPC][NFC] Remove Darwin specific logic in frame finalization.
Remove some cumbersome Darwin specific logic for updating the frame
offsets of the condition-register spill slots. The containing function has an
early return if the subtarget is not ELF based which makes the Darwin logic
dead.
Pavel Labath [Thu, 20 Feb 2020 14:06:01 +0000 (15:06 +0100)]
[Error/unittests] Add a FailedWithMessage gtest matcher
Summary:
We already have a "Failed" matcher, which can be used to check any
property of the Error object. However, most frequently one just wants to
check the error message, and while this is possible with the "Failed"
matcher, it is also very convoluted
(Failed<ErrorInfoBase>(testing::Property(&ErrorInfoBase::message, "the
message"))).
Now, one can just write: FailedWithMessage("the message"). I expect that
most of the usages will remain this simple, but the argument of the
matcher is not limited to simple strings -- the argument of the matcher
can be any other matcher, so one can write more complicated assertions
if needed (FailedWithMessage(ContainsRegex("foo|bar"))). If one wants to
match multiple error messages, he can pass multiple arguments to the
matcher.
If one wants to match the message list as a whole (perhaps to check the
message count), I've also included a FailedWithMessageArray matcher,
which takes a single matcher receiving a vector of error message
strings.
Reviewers: sammccall, dblaikie, jhenderson
Subscribers: llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D74898
Simon Pilgrim [Fri, 21 Feb 2020 14:23:38 +0000 (14:23 +0000)]
[X86] Regenerate hi reg tests
Simon Pilgrim [Fri, 21 Feb 2020 13:29:40 +0000 (13:29 +0000)]
[TargetLowering] SimplifyDemandedBits - use getValidShiftAmountConstant helper.
Use the SelectionDAG::getValidShiftAmountConstant helper to get const/constsplat shift amounts, which allows us to drop the out of range shift amount early-out.
First step towards better non-uniform shift amount support in SimplifyDemandedBits.
Lei Zhang [Fri, 10 Jan 2020 19:48:24 +0000 (14:48 -0500)]
[mlir] Add a signedness semantics bit to IntegerType
Thus far IntegerType has been signless: a value of IntegerType does
not have a sign intrinsically and it's up to the specific operation
to decide how to interpret those bits. For example, std.addi does
two's complement arithmetic, and std.divis/std.diviu treats the first
bit as a sign.
This design choice was made some time ago when we did't have lots
of dialects and dialects were more rigid. Today we have much more
extensible infrastructure and different dialect may want different
modelling over integer signedness. So while we can say we want
signless integers in the standard dialect, we cannot dictate for
others. Requiring each dialect to model the signedness semantics
with another set of custom types is duplicating the functionality
everywhere, considering the fundamental role integer types play.
This CL extends the IntegerType with a signedness semantics bit.
This gives each dialect an option to opt in signedness semantics
if that's what they want and helps code sharing. The parser is
modified to recognize `si[1-9][0-9]*` and `ui[1-9][0-9]*` as
signed and unsigned integer types, respectively, leaving the
original `i[1-9][0-9]*` to continue to mean no indication over
signedness semantics. All existing dialects are not affected (yet)
as this is a feature to opt in.
More discussions can be found at:
https://groups.google.com/a/tensorflow.org/d/msg/mlir/XmkV8HOPWpo/7O4X0Nb_AQAJ
Differential Revision: https://reviews.llvm.org/D72533
Krzysztof Parzyszek [Thu, 20 Feb 2020 22:42:02 +0000 (16:42 -0600)]
[Hexagon] Introduce noop intrinsic to cast between vector predicate types
The (overloaded) intrinsic is llvm.hexagon.V6.pred.typecast[.128B]. The
types of the operand and the return value are HVX boolean vector types.
For each cast, there needs to be a corresponding intrinsic declared,
with different suffixes appended to the name, e.g.
; cast <128 x i1> to <32 x i1>
declare <32 x i1> @llvm.hexagon.V6.pred.typecast.128B.s1(<128 x i1>)
; cast <32 x i1> to <64 x i1>
declare <64 x i1> @llvm.hexagon.V6.pred.typecast.128B.s2(<32 x i1>)
etc.
Evgeniy Brevnov [Mon, 20 Jan 2020 12:35:39 +0000 (19:35 +0700)]
[DependenceAnalysis] Memory dependence analysis internal caching mechanism is broken in presence of TBAA (PR42733).
Summary:
There is a flaw in memory dependence analysis caching mechanism when memory accesses with TBAA are involved. Assume we first analysed and cached results for access with TBAA. Later we request dependence for the same memory but without TBAA (or different TBAA). By design these two queries should share one entry in the internal cache which corresponds to a general access (without TBAA). Thus upon second request internal cached is cleared and we continue analysis for access as if there is no TBAA.
The problem is that even though internal cache is cleared the set of visited nodes is not. That means we won't traverse visited nodes again and populate internal cache with the corresponding dependence results. So we end up with internal cache in an incomplete state. Current implementation tries to signal that situation by resetting CacheInfo->Pair at line 1104. But that doesn't actually help since later code ignores this invalidation and relies on 'Cache->empty()' property to decide on cache completeness.
Reviewers: reames, hfinkel, chandlerc, fedor.sergeev, asbirlea, fhahn, john.brawn, Prazek, sunfish
Reviewed By: john.brawn
Subscribers: DaniilSuchkov, kosarev, jfb, dantrushin, hiraditya, bmahjour, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D73032
Sanjay Patel [Fri, 21 Feb 2020 12:58:36 +0000 (07:58 -0500)]
[ConstantFold] fold fsub -0.0, undef to undef rather than NaN
A question about this behavior came up on llvm-dev:
http://lists.llvm.org/pipermail/llvm-dev/2020-February/139003.html
...and as part of backend improvements in D73978, but this is an IR
change first because we already have fairly thorough tests in place
here.
We decided not to implement a more general change that would have
folded any FP binop with nearly arbitrary constant + undef operand
to undef because that is not theoretically correct (even if it is
practically correct).
Differential Revision: https://reviews.llvm.org/D74713
serge-sans-paille [Fri, 21 Feb 2020 08:48:27 +0000 (09:48 +0100)]
Detect and disable openmp tests that require multiple hardware processor to run
Team tests seem to require at least two physical cores, and using the same trick
as in https://reviews.llvm.org/D55598 doesn't work (why?) .
Using lit configuration instead.
Differential Revision: https://reviews.llvm.org/D74921
Swiftfuchs [Fri, 21 Feb 2020 12:55:43 +0000 (13:55 +0100)]
[NFC] Corrected a minor typo in a comment
Nicolai Hähnle [Fri, 21 Feb 2020 12:10:18 +0000 (13:10 +0100)]
test/CodeGen/AMDGPU: Add a test case that shows a miscompilation
Related to https://reviews.llvm.org/D74908
Change-Id: I6ebf3b5c7a32493016994f30d6796c41e95aecde
Kern Handa [Fri, 21 Feb 2020 12:25:24 +0000 (13:25 +0100)]
[NFC][mlir] Adding more operators to EDSC TemplatedIndexedValue
This change adds some missing arithmetic and logical operators to
`TemplatedIndexedValue` for EDSC usage.
Differential Revision: https://reviews.llvm.org/D74686
Sebastian Neubauer [Fri, 24 Jan 2020 11:17:08 +0000 (12:17 +0100)]
Make unittests include path relative
This change is relevant when embedding the llvm cmake project into
another project. It should not change the build behavior of a normal
llvm build.
In the case where llvm is embedded as a cmake subproject,
CMAKE_SOURCE_DIR does not point to the expected directory and building
the tests fails.
Using CMAKE_CURRENT_SOURCE_DIR fixes this problem, as it will always
point to the same directory.
Differential Revision: https://reviews.llvm.org/D73466
Haojian Wu [Fri, 21 Feb 2020 08:57:05 +0000 (09:57 +0100)]
[clangd] Allow renaming class templates in cross-file rename.
Summary:
It was disabled because we don't handle explicit template
specialization well (due to the index limitation).
renaming templates is normal in practic, rather than disabling it, this patch
allows to rename them though it is not perfect (just a known limitation).
Context: https://github.com/clangd/clangd/issues/280
Reviewers: kbobyrev
Reviewed By: kbobyrev
Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D74709
Raphael Isemann [Thu, 20 Feb 2020 12:23:39 +0000 (13:23 +0100)]
[lldb][NFC] Split up ClangASTSource::FindExternalVisibleDecls
This function has two functions hidden inside it. Let's make
them proper functions.
Haojian Wu [Thu, 20 Feb 2020 14:22:07 +0000 (15:22 +0100)]
[clangd] Fix the incomplete template specialization in findTarget.
Summary:
FindTarget doesn't report the TemplatePattern for incomplete
specialization.
Reviewers: sammccall
Subscribers: ilya-biryukov, MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D74900
Craig Topper [Fri, 21 Feb 2020 08:29:18 +0000 (00:29 -0800)]
[X86] Don't bother avoiding illegal FCMOVs if we don't have the cmov subtarget feature.
We'll be forced to emit branches so we might as well use the most
direct condition.
Craig Topper [Fri, 21 Feb 2020 07:53:00 +0000 (23:53 -0800)]
[X86] Make combineCMov not create unsupported FCMOVs when f32/f64 are using X87.
This makes the behavior consistent with what's in LowerSELECT.
Craig Topper [Fri, 21 Feb 2020 07:49:44 +0000 (23:49 -0800)]
[X86] Autogenerate complete checks. NFC
Sam Clegg [Thu, 13 Feb 2020 16:38:18 +0000 (08:38 -0800)]
[WebAssembly] Remove unneeded getWasmKindForNamedSection function
I believe this was carried over from getELFKindForNamedSection since
the wasm backend originally used ELF object writing as a template.
Differential Revision: https://reviews.llvm.org/D74565
Craig Topper [Fri, 21 Feb 2020 05:38:49 +0000 (21:38 -0800)]
[X86] Remove unnecessary isNullConstant in LowerSelect. NFC
At this point in the code we know that Op1 or Op2 is
all ones. Y points to the other operand. In the case that
Op2 is zero, Op1 must be all ones and Y is Op2. The OR
ORs Y into Res. But if Y is 0 the OR will be folded away by
getNode so we don't need to check for it.
Jan Vesely [Tue, 4 Feb 2020 21:13:17 +0000 (16:13 -0500)]
libclc: Use acos implementation from amd_builtins
Fixes acos CTS (1 thread, scalar) on AMD Turks.
Reviewer: tstellar
Differential Revision: https://reviews.llvm.org/D74011
Craig Topper [Fri, 21 Feb 2020 03:56:44 +0000 (19:56 -0800)]
[X86] Add CMOV_VR64 pseudo instruction for MMX. Remove mmx handling from combineSelect.
The combineSelect code was casting to i64 without any check that
i64 was legal. This can break after type legalization.
It also required splitting the mmx register on 32-bit targets.
It's not clear that this makes sense. Instead switch to using
a cmov pseudo like we do for XMM/YMM/ZMM.
Martijn Vels [Wed, 19 Feb 2020 21:27:50 +0000 (16:27 -0500)]
Split _LIBCPP_STRING_EXTERN_TEMPLATE_LIST up into a V1 and UNSTABLE version.
This change splits the _LIBCPP_STRING_EXTERN_TEMPLATE_LIST up into a _LIBCPP_STRING_V1_EXTERN_TEMPLATE_LIST containing the stable ABI, and a _LIBCPP_STRING_UNSTABLE_EXTERN_TEMPLATE_LIST containing the unstable ABI.
The purpose is to explicitly define and maintain the two lists, where the unstable ABI allows for ABI breaking changes for purposes such as optimization while offering a strong guarantee that any change inside the unstable ABI does not affect the stable ABI.
As per the comment in the __string header, we do still allow etries to be added to the stable ABI list as the c++ versions and corresponding c++ std API changes.
River Riddle [Fri, 21 Feb 2020 04:08:33 +0000 (20:08 -0800)]
[mlir] Use getOperation()->setAttr when generating attribute set
methods.
This avoids the need to resolve overloads when the current operation
also defines a 'setAttr' method.
Johannes Doerfert [Thu, 20 Feb 2020 22:43:03 +0000 (16:43 -0600)]
[OpenMP][NFC] Remove leftover debug messages
Jim Lin [Fri, 21 Feb 2020 01:28:49 +0000 (09:28 +0800)]
[XCore] Add instruction pattern for bitrev
Summary:
Add support for lowering bitreverse to the bitrev instruction.
Fix https://bugs.llvm.org/show_bug.cgi?id=34628.
Reviewers: RKSimon, rtrieu, robertlytton
Reviewed By: RKSimon
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D74748
Vedant Kumar [Thu, 20 Feb 2020 22:16:25 +0000 (14:16 -0800)]
[Dominators] Use Instruction::comesBefore for block-local queries, NFC
Use the lazy instruction ordering facility for block-local dominance
queries.
Differential Revision: https://reviews.llvm.org/D74931
Bill Wendling [Fri, 21 Feb 2020 00:22:35 +0000 (16:22 -0800)]
Filter callbr insts from critical edge splitting
Similarly to how splitting predecessors with an indirectbr isn't handled
in the generic way, we also shouldn't split callbrs, for similar
reasons.
Michael Spencer [Thu, 20 Feb 2020 23:25:36 +0000 (15:25 -0800)]
[libunwind][CMake] Treat S files as C to work around CMake bug.
The OSX_ARCHITECTURES property is supposed to add the -arch flag when
targeting Apple platforms. However, due to a bug in CMake
(https://github.com/Kitware/CMake/blob/master/Source/cmLocalGenerator.cxx#L1780),
this does not apply to assembly files. This results in a linker error
when trying to build libunwind for i386 on MacOS.
rdar://
59642189
Mark de Wever [Thu, 20 Feb 2020 23:13:38 +0000 (18:13 -0500)]
[libc++][regex] Validate backreferences in the constructor.
This patch enables throwing exceptions for invalid backreferences
in the constructor when using the basic, extended, grep, or egrep grammar.
This fixes bug 34297.
Differential Revision: https://reviews.llvm.org/D62453
Craig Topper [Thu, 20 Feb 2020 22:00:17 +0000 (14:00 -0800)]
[X86] Add CMOV_VK1 pseudo so we don't crash on v1i1 ISD::SELECT
Craig Topper [Thu, 20 Feb 2020 21:46:56 +0000 (13:46 -0800)]
[X86] Expand vselect of v1i1 under avx512.
We already do this for v2i1, v4i1, etc.
Craig Topper [Thu, 20 Feb 2020 21:37:09 +0000 (13:37 -0800)]
[X86] Custom legalize v1i1 UADDSAT/USUBSAT/SADDSAT/UADDSAT to match v2i1/v4i1/v8i1 etc.
Craig Topper [Thu, 20 Feb 2020 21:28:12 +0000 (13:28 -0800)]
[X86] Fix a couple copy mistakes in v4i1 or/and/xor isel patterns.
VK1 was being used as the output of the copy to regclass, but it
should be VK2/VK4. Shouldn't matter in practice though since
VK1/VK2/VK4/VK8/VK16 are all identicaly and just have different VTs.
Craig Topper [Thu, 20 Feb 2020 21:22:38 +0000 (13:22 -0800)]
[X86] Custom legalize v1i1 add/sub/mul to xor/xor/and with avx512.
We already did this for v2i1, v4i1, v8i1, etc.
Louis Dionne [Thu, 20 Feb 2020 23:03:10 +0000 (18:03 -0500)]
[libc++] Explain XFAILs with std::uncaught_exceptions test
Jonas Devlieghere [Thu, 20 Feb 2020 22:51:57 +0000 (14:51 -0800)]
[lldb/Test] Remove stale README in test/API
Now that the test live in the same directory the README is no longer
relevant.
River Riddle [Thu, 20 Feb 2020 22:37:25 +0000 (14:37 -0800)]
[mlir][Parser] Update DenseElementsAttr to print in hex when the number of elements is over a certain threshold.
Summary: DenseElementsAttr is used to store tensor data, which in some cases can become extremely large(100s of mb). In these cases it is much more efficient to format the data as a string of hex values instead.
Differential Revision: https://reviews.llvm.org/D74922
Scott Linder [Thu, 20 Feb 2020 22:36:56 +0000 (17:36 -0500)]
Revert "[Driver] Escape the program path for -frecord-command-line"
This reverts commit
6123074d0c0de3614f1552f2f2d6b9db7b32cebe.
Quoting/escaping rules seem host specific, so the test is failing on
some bots.
Jonas Devlieghere [Thu, 20 Feb 2020 22:34:49 +0000 (14:34 -0800)]
[lldb/Plugin] Don't mark ProcessNetBSD as a plugin
ProcessNetBSD has no initializers so it's just a regular library and
shouldn't have the PLUGIN argument in add_lldb_library.
River Riddle [Thu, 20 Feb 2020 22:16:20 +0000 (14:16 -0800)]
[mlir] Use LLJIT::getMainJITDylib instead of hardcoding '<main>'
This fixes test failures caused by a change to the name of the main
dylib, now called 'main'. It also hardens the engine against potential
future changes to the name.
aartbik [Thu, 20 Feb 2020 21:57:54 +0000 (13:57 -0800)]
[mlir] [VectorOps] Multi-dim reductions for lowering vector.contract
Summary:
This implements the last step for lowering vector.contract progressively
to LLVM IR (except for masks). Multi-dimensional reductions that remain
after expanding all parallel dimensions are lowered into into simpler
vector.contract operations until a trivial 1-dim reduction remains.
Reviewers: nicolasvasilache, andydavis1
Reviewed By: andydavis1
Subscribers: mehdi_amini, rriddle, jpienaar, burmako, shauheen, antiagainst, nicolasvasilache, arpith-jacob, mgester, lucyrfox, liufengdb, Joonsoo, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D74880
Florian Hahn [Thu, 20 Feb 2020 18:46:29 +0000 (19:46 +0100)]
[SCCP] Do not mark unknown loads as overdefined.
For tracked globals that are unknown after solving, we expect all
non-store uses to be replaced.
This is a follow-up to
f8045b250d80, which removed forcedconstant.
We should not mark unknown loads as overdefined, as they either load
from an unknown pointer or an undef global. Restore the original logic
for loads.
Eli Friedman [Wed, 29 Jan 2020 03:17:30 +0000 (19:17 -0800)]
[SVE] Add support for lowering GEPs involving scalable vectors.
This includes both GEPs where the indexed type is a scalable vector, and
GEPs where the result type is a scalable vector.
Differential Revision: https://reviews.llvm.org/D73602
Scott Linder [Thu, 20 Feb 2020 21:31:17 +0000 (16:31 -0500)]
[Driver] Escape the program path for -frecord-command-line
Similar to the rest of the command line that is recorded, the program
path must also have spaces and backslashes escaped. Without this
parsing the recorded command line becomes hard on platforms like
Windows where spaces and backslashes are common.
Patch By: Ravi Ramaseshan
Differential Revision: https://reviews.llvm.org/D74811
shafik [Thu, 20 Feb 2020 20:28:46 +0000 (12:28 -0800)]
[ASTImporter] Add linkage check to ASTNodeImporter::hasSameVisibilityContext and rename to hasSameVisibilityContextAndLinkage
This fixed is based on the assert in LinkageComputer::getLVForDecl(...) which assumes that all the decls in a redecl chain have the same linkage.
Differential Revision: https://reviews.llvm.org/D74639
David Tenty [Thu, 20 Feb 2020 20:39:16 +0000 (15:39 -0500)]
[AIX] Improve 32/64-bit build configuration
Summary:
AIX supports both 32-bit and 64-bit environments (with 32-bit being the default). This patch improves support for building LLVM on AIX in both 32-bit and 64-bit mode.
- Change host detection to return correct 32/64-bit triple as config_guess does not return the correct version on 64-bit. This can confuse JIT tests and other things that care about what the host triple is.
- Remove manual setting of 64-bit flags on AIX. AIX provides OBJECT_MODE environment variable to enable the user to obtain a 64-bit development environment. CMake will properly set these flags provided the user sets the correct OBJECT_MODE before configuring and setting them manually will interfere with 32-bit builds.
- Don't present the LLVM_BUILD_32_BITS option on AIX, users should use OBJECT_MODE when running CMake instead.
Reviewers: hubert.reinterpretcast, DiggerLin, stevewan
Reviewed By: DiggerLin, stevewan
Subscribers: mgorny, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D74256
Craig Topper [Thu, 20 Feb 2020 20:28:54 +0000 (12:28 -0800)]
Recommit "[X86] Replace a bad use of MVT::getVectorVT with EVT::getVectorVT""
With the correct author this time
Craig Topper [Thu, 20 Feb 2020 20:28:01 +0000 (12:28 -0800)]
Revert
714265dabb606bfef2f85694234f152edbfa91ac "[X86] Replace a bad use of MVT::getVectorVT with EVT::getVectorVT"
I accidentally messed up the author on the previous commit somehow.
Quentin Colombet [Thu, 20 Feb 2020 19:27:36 +0000 (11:27 -0800)]
[X86] Replace a bad use of MVT::getVectorVT with EVT::getVectorVT
The type here isn't guaranteed to be a simple type.
Fixes PR44976
Nico Weber [Thu, 20 Feb 2020 20:15:54 +0000 (15:15 -0500)]
Revert "libclang: Add static build support for Windows" and
follow-up "libclang: Make shared object symbol exporting by default"
This reverts commit
7a7c753b0cca6abb61f054842fb087dfc30dc563.
This reverts commit
7ff1f55a1219719f57a6f7905c26ce41d1767e4c.
They broke building libclang.dll on Windows, see
https://reviews.llvm.org/D74564
Nico Weber [Thu, 20 Feb 2020 20:11:13 +0000 (15:11 -0500)]
Revert "[AArch64][SVE] Add intrinsics for SVE2 bitwise ternary operations"
This reverts commit
ce70e2899879e092b153a4078b993833b6696713.
It broke MC/AArch64/SVE2/bsl-diagnostics.s everywhere.
Sanjay Patel [Thu, 20 Feb 2020 18:41:56 +0000 (13:41 -0500)]
[ConstantFold] add/move tests for FP with undef operand; NFC
Sourabh Singh Tomar [Thu, 20 Feb 2020 19:42:52 +0000 (01:12 +0530)]
Revert "[NFCI][DebugInfo]: Corrected a Typo."
This reverts commit
3e1090922a0b808f424ff424b744752b0d53a3ee
as per Paul Robinson's suggestion.
Francesco Petrogalli [Thu, 20 Feb 2020 19:37:02 +0000 (19:37 +0000)]
[llvm][build] Fix shared lib builds. [NFC]
The code at https://reviews.llvm.org/D74808 has broken builds that are
configured with -DBUILD_SHARED_LIBS=On.
This patch adds the correct library dependencies.
Richard Smith [Thu, 20 Feb 2020 19:40:09 +0000 (11:40 -0800)]
[cxx_status] Update -std= instructions for C++20.
We merged support for -std=c++20 to the Clang 10 branch, so -std=c++2a
is only needed in Clang 9 and earlier.
Quentin Colombet [Thu, 20 Feb 2020 19:27:36 +0000 (11:27 -0800)]
[GISel][KnownBits] Give up on PHI analysis as soon as we don't know anything
When analyzing PHIs, we gather the known bits for every operand and
merge them together to get the known bits of the result of the PHI.
It is not unusual that merging the information leads to know nothing
on the result (e.g., phi a: i8 3, b: i8 unknown, ..., after looking at the
second argument we know we will know nothing on the result), thus, as
soon as we reach that state, stop analyzing the following operand (i.e.,
on the previous example, we won't process anything after looking at `b`).
This improves compile time in particular with PHIs with a large number
of operands.
NFC.
Jan Korous [Mon, 17 Feb 2020 22:55:49 +0000 (14:55 -0800)]
[clang][doxygen] Fix false -Wdocumentation warning for tag typedefs
For tag typedefs like this one:
/*!
@class Foo
*/
typedef class { } Foo;
clang -Wdocumentation gives:
warning: '@class' command should not be used in a comment attached to a
non-struct declaration [-Wdocumentation]
... while doxygen seems fine with it.
Differential Revision: https://reviews.llvm.org/D74746
David Tenty [Thu, 20 Feb 2020 19:22:16 +0000 (14:22 -0500)]
[AIX] Pack BasicBlockBits
Summary:
D51664 introduces a new structure BasicBlockBits which it expects to be
packed a certain way. This change is very similar to D60164, and we apply the
same fix:
"On AIX, the canonical layout of bit-fields would cause
these ... to span four bytes. Applying the pack pragma for compilers that
employ the AIX canonical layout allows these ... to fit within the expected
two bytes. In the future, the pragma would also likely need to be applied when
building with Clang on AIX."
Reviewers: jasonliu, hubert.reinterpretcast, sfertile, xingxue
Reviewed By: sfertile
Subscribers: llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D74911
Cristian Adam [Thu, 20 Feb 2020 19:24:52 +0000 (14:24 -0500)]
libclang: Make shared object symbol exporting by default
https://reviews.llvm.org/D74564 enabled static building for libclang,
and for non CMake consumers they had to set the `CMAKE_EXPORTS` define
when consuming libclang.
This commit makes the non CMake users of the static building have to define `CMAKE_NO_EXPORTS`.
Differential Revision: https://reviews.llvm.org/D74907
Jonathan Coe [Thu, 20 Feb 2020 19:23:38 +0000 (19:23 +0000)]
[clang-format] Merge name and colon into a single token for C# named arguments
Summary:
Merge 'argumentName' and ':' into a single token in foo(argumentName: bar).
Add C# named argument as a token type.
Reviewers: krasimir, MyDeveloperDay
Reviewed By: krasimir
Tags: #clang-format
Differential Revision: https://reviews.llvm.org/D74894