platform/upstream/llvm.git
4 years ago[x86] Propagate memory operands during ISel DAG postprocessing
Jean-Michel Gorius [Mon, 18 May 2020 19:30:43 +0000 (21:30 +0200)]
[x86] Propagate memory operands during ISel DAG postprocessing

Summary:
Propagate memory operands when folding test instructions.

This was split from D80062.

Reviewers: craig.topper, rnk, lebedev.ri

Reviewed By: craig.topper

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D80140

4 years agoAMDGPU: Fix illegally constant folding from V_MOV_B32_sdwa
Matt Arsenault [Sat, 16 May 2020 17:48:55 +0000 (13:48 -0400)]
AMDGPU: Fix illegally constant folding from V_MOV_B32_sdwa

This was assumed to be a simple move, and interpreting the immediate
modifier operand as a materialized immediate. Apparently the SDWA pass
never produces these, but GlobalISel does emit these for some vector
shuffles.

4 years agoRevert "[CUDA][HIP] Workaround for resolving host device function against wrong-sided...
Artem Belevich [Mon, 18 May 2020 19:17:39 +0000 (12:17 -0700)]
Revert "[CUDA][HIP] Workaround for resolving host device function against wrong-sided function"

Still breaks CUDA compilation.

This reverts commit e03394c6a6ff5832aa43259d4b8345f40ca6a22c.

4 years ago[mlir][vulkan-runner] Add back accidentially removed header include
Lei Zhang [Sun, 17 May 2020 14:14:36 +0000 (10:14 -0400)]
[mlir][vulkan-runner] Add back accidentially removed header include

Differential Revision: https://reviews.llvm.org/D80149

4 years agoAMDGPU/GlobalISel: Fix f64 G_FDIV lowering
Matt Arsenault [Mon, 18 May 2020 18:53:26 +0000 (14:53 -0400)]
AMDGPU/GlobalISel: Fix f64 G_FDIV lowering

This was using an integer multiply instead of FP.

4 years agoLoadStoreVectorizer: Match nested adds to prove vectorization is safe
Volkan Keles [Mon, 18 May 2020 19:11:46 +0000 (12:11 -0700)]
LoadStoreVectorizer: Match nested adds to prove vectorization is safe

If both OpA and OpB is an add with NSW/NUW and with the same LHS operand,
we can guarantee that the transformation is safe if we can prove that OpA
won't overflow when IdxDiff added to the RHS of OpA.

Review: https://reviews.llvm.org/D79817

4 years ago[Loads] Require Align in isSafeToLoadUnconditionally() (NFC)
Nikita Popov [Mon, 18 May 2020 17:09:03 +0000 (19:09 +0200)]
[Loads] Require Align in isSafeToLoadUnconditionally() (NFC)

Now that load/store have required alignment, accept Align here.
This also avoids uses of getPointerElementType(), which is
incompatible with opaque pointers.

4 years agoFixed a typo in the comment for allocateBuffer()
George [Mon, 18 May 2020 18:40:32 +0000 (14:40 -0400)]
Fixed a typo in the comment for allocateBuffer()

Differential Revision: https://reviews.llvm.org/D80087

4 years ago[dsymutil] Make sure the --help output and man page are consistent
Jonas Devlieghere [Mon, 18 May 2020 18:34:09 +0000 (11:34 -0700)]
[dsymutil] Make sure the --help output and man page are consistent

As suggested by Adrian in D79398.

4 years agoAdd verifier check that musttail and preallocated are not used together
Arthur Eubanks [Mon, 18 May 2020 16:02:12 +0000 (09:02 -0700)]
Add verifier check that musttail and preallocated are not used together

Summary:
Currently they are not supported together. Supporting them will require
a LangRef change. See discussion in https://reviews.llvm.org/D77689.

Reviewers: rnk, efriedma

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D80132

4 years ago[IR] Revert r2694 in BasicBlock::removePredecessor
Jay Foad [Mon, 18 May 2020 15:06:42 +0000 (16:06 +0100)]
[IR] Revert r2694 in BasicBlock::removePredecessor

r2694 fixed a bug where removePredecessor could create IR with a use not
dominated by its def in a self loop. But this could only happen in an
unreachable loop, and since that time the rules have been relaxed so
that defs don't have to dominate uses in unreachable code, so the fix is
unnecessary. The regression test added in r2691 still stands.

Differential Revision: https://reviews.llvm.org/D80128

4 years ago[SystemZ] Don't create PERMUTE nodes with an undef operand.
Jonas Paulsson [Mon, 18 May 2020 16:20:40 +0000 (18:20 +0200)]
[SystemZ] Don't create PERMUTE nodes with an undef operand.

It's better to reuse the first source value than to use an undef second
operand, because that will make more resulting VPERMs have identical operands
and therefore MachineCSE more successful.

Review: Ulrich Weigand

4 years ago[llvm][NFC] Fixed non-compliant style in InlineAdvisor.h
Mircea Trofin [Mon, 18 May 2020 17:26:45 +0000 (10:26 -0700)]
[llvm][NFC] Fixed non-compliant style in InlineAdvisor.h

Changed OnPass{Entry|Exit} -> onPass{Entry|Exit}

Also fixed a small typo in a comment.

4 years ago[lldb/Reproducers] Add skipIfReproducer to more tests
Jonas Devlieghere [Mon, 18 May 2020 16:55:33 +0000 (09:55 -0700)]
[lldb/Reproducers] Add skipIfReproducer to more tests

Mark more tests as unsupported with reproducers.

4 years ago[Local] Do not ignore zexts in salvageDebugInfo, PR45923
Vedant Kumar [Fri, 15 May 2020 20:58:40 +0000 (13:58 -0700)]
[Local] Do not ignore zexts in salvageDebugInfo, PR45923

Summary:
When salvaging a dead zext instruction, append a convert operation to
the DIExpressions of the debug uses of the instruction, to prevent the
salvaged value from being sign-extended.

I confirmed that lldb prints out the correct unsigned result for "f" in
the example from PR45923 with this changed applied.

rdar://63246143

Reviewers: aprantl, jmorse, chrisjackson, davide

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D80034

4 years ago[MLIR] Use `MLIR_INCLUDE_TESTS` to conditionally compile tests.
Alexandre Rames [Mon, 18 May 2020 16:44:26 +0000 (18:44 +0200)]
[MLIR] Use `MLIR_INCLUDE_TESTS` to conditionally compile tests.

This is equivalent to what is done for other projects (e.g. clang).

Differential Revision: https://reviews.llvm.org/D80022

4 years ago[mlir] Support optional attributes in assembly formats
Tobias Gysi [Mon, 18 May 2020 16:30:39 +0000 (18:30 +0200)]
[mlir] Support optional attributes in assembly formats

Summary: This revision adds support for assembly formats with optional attributes. It elides optional attributes that are part of the syntax from the attribute dictionary.

Reviewers: ftynse, Kayjukh

Reviewed By: ftynse, Kayjukh

Subscribers: mehdi_amini, rriddle, jpienaar, shauheen, antiagainst, nicolasvasilache, arpith-jacob, mgester, lucyrfox, liufengdb, stephenneuendorffer, Joonsoo, grosul1, frgossen, jurahul, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D80113

4 years ago[MLIR] [Linalg] Add option to use the partial view after promotion.
Pierre Oechsel [Mon, 18 May 2020 16:25:23 +0000 (18:25 +0200)]
[MLIR] [Linalg] Add option to use the partial view after promotion.

For now the promoted buffer is indexed using the `full view`. The full view might be
slightly bigger than the partial view (which is accounting for boundaries).
Unfortunately this does not compose easily with other transformations when multiple buffers
with shapes related to each other are involved.
Take `linalg.matmul A B C` (with A of size MxK, B of size KxN and C of size MxN) and suppose we are:
- Tiling over M by 100
- Promoting A only

This is producing a `linalg.matmul promoted_A B subview_C` where `promoted_A` is a promoted buffer
of `A` of size (100xK) and `subview_C` is a subview of size mxK where m could be smaller than 100 due
to boundaries thus leading to a possible incorrect behavior.

We propose to:
- Add a new parameter to the tiling promotion allowing to enable the use of the full tile buffer.
- By default all promoted buffers will be indexed by the partial view.

Note that this could be considered as a breaking change in comparison to the way the tiling promotion
was working.

Differential Revision: https://reviews.llvm.org/D79927

4 years agoAMDGPU/GlobalISel: Fix splitting wide VALU, non-vector loads
Matt Arsenault [Sun, 17 May 2020 23:06:10 +0000 (19:06 -0400)]
AMDGPU/GlobalISel: Fix splitting wide VALU, non-vector loads

4 years agoAMDGPU: Remove outdated comment
Matt Arsenault [Sun, 17 May 2020 16:37:27 +0000 (12:37 -0400)]
AMDGPU: Remove outdated comment

4 years ago[mlir][Vector] Add an optional "masked" boolean array attribute to vector transfer...
Nicolas Vasilache [Mon, 18 May 2020 15:51:56 +0000 (11:51 -0400)]
[mlir][Vector] Add an optional "masked" boolean array attribute to vector transfer operations

Summary:
Vector transfer ops semantic is extended to allow specifying a per-dimension `masked`
attribute. When the attribute is false on a particular dimension, lowering to LLVM emits
unmasked load and store operations.

Differential Revision: https://reviews.llvm.org/D80098

4 years ago[mlir][Vector] Make minor identity permutation map optional in transfer op printing...
Nicolas Vasilache [Mon, 18 May 2020 15:40:59 +0000 (11:40 -0400)]
[mlir][Vector] Make minor identity permutation map optional in transfer op printing and parsing

Summary:
This revision makes the use of vector transfer operatons more idiomatic by
allowing to omit and inferring the permutation_map.

Differential Revision: https://reviews.llvm.org/D80092

4 years ago[SVE] Ignore scalable vectors in InterleavedLoadCombinePass
David Sherwood [Wed, 6 May 2020 10:23:04 +0000 (11:23 +0100)]
[SVE] Ignore scalable vectors in InterleavedLoadCombinePass

I have changed the pass so that we ignore shuffle vectors with
scalable vector types, and replaced VectorType with FixedVectorType
in the rest of the pass. I couldn't think of an easy way to test
this change, since for scalable vectors we shouldn't be using
shufflevectors for interleaving. This change fixes up some
type size assert warnings I found in the following test:

  CodeGen/AArch64/sve-intrinsics-int-arith-imm.ll

Differential Revision: https://reviews.llvm.org/D79700

4 years ago[WebAssembly] iterate stack in DebugFixup from the top.
Wouter van Oortmerssen [Fri, 15 May 2020 23:09:20 +0000 (16:09 -0700)]
[WebAssembly] iterate stack in DebugFixup from the top.

Differential Revision: https://reviews.llvm.org/D80045

4 years ago[Analyzer][StreamChecker] Fixed compile error - NFC.
Balázs Kéri [Mon, 18 May 2020 15:14:39 +0000 (17:14 +0200)]
[Analyzer][StreamChecker] Fixed compile error - NFC.

4 years ago[Analyzer][StreamChecker] Changed representation of stream error state - NFC.
Balázs Kéri [Mon, 18 May 2020 13:07:32 +0000 (15:07 +0200)]
[Analyzer][StreamChecker] Changed representation of stream error state - NFC.

Summary:
State of error flags for a stream is handled by having separate flags
that allow combination of multiple error states to be described with one
error state object.
After a failed function the error state is set in the stream state
and must not be determined later based on the last failed function
like before this change. The error state can not always be determined
from the last failed function and it was not the best design.

Reviewers: Szelethus

Reviewed By: Szelethus

Subscribers: 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/D80009

4 years ago[pstl] A fix for move placement-new (and destroy) allocated objects from raw memory.
Dvorskiy, Mikhail [Tue, 10 Mar 2020 16:09:57 +0000 (19:09 +0300)]
[pstl] A fix for move placement-new (and destroy) allocated objects from raw memory.

https://reviews.llvm.org/D74123

The fix affects follow algorithms:
remove_if, unique, rotate, inplace_merge, partial_sort_copy, set_union, set_intersection, set_difference, set_symmetric_difference.

For "is_trivial" types there are no problems with "creating objects/clean-up"
For non-trivial types the algo results are also correct, but possible incorrect copying/moving "operator=" calls "by raw memory" within one of mentioned algo or incorrect destructor calls in the end of algo.

4 years ago[mlir] Add BoolArrayAttr in Tablegen + Builder support
Nicolas Vasilache [Mon, 18 May 2020 13:41:59 +0000 (09:41 -0400)]
[mlir] Add BoolArrayAttr in Tablegen + Builder support

Differential Revision: https://reviews.llvm.org/D80090

4 years ago[mlir] NFC - Fix OperationSupport.cpp::findNamedAttr
Nicolas Vasilache [Mon, 18 May 2020 13:33:35 +0000 (09:33 -0400)]
[mlir] NFC - Fix OperationSupport.cpp::findNamedAttr

Summary:
When NamedAttrList::get is called against a StringRef and the
entry is not present, the Identifier::operator== crashes.
Interestingly there seems to be no use of the NamedAttrList::get(StringRef) in
the codebase so far. A subsequent commit will introduce such a use.

Differential Revision: https://reviews.llvm.org/D80089

4 years ago[openmp] Fixed hang if detached task was serialized.
AndreyChurbanov [Mon, 18 May 2020 12:32:13 +0000 (15:32 +0300)]
[openmp] Fixed hang if detached task was serialized.

The patch fixes https://bugs.llvm.org/show_bug.cgi?id=45904.

Differential Revision: https://reviews.llvm.org/D79944

4 years agoReorder Type fields to make various isa< > check more concise
serge-sans-paille [Tue, 12 May 2020 16:46:32 +0000 (18:46 +0200)]
Reorder Type fields to make various isa< > check more concise

Depending on the order of fields, some isa < > checks can be faster because of
tests that check a range of type, leading to assembly simplification.

To find a relevant ordering, I... brute-forced the permutation among the derived
types and pick the combination that resulted in the smallest libLLVM-11.so.

On my laptop (x86_64), this reduces the size of libLLVM-11.so from 127344064 bytes to 127335336,
that's 8728 bytes shaved without much effort.

Also removed obsolete comments in the process.

Differential Revision: https://reviews.llvm.org/D79996

4 years ago[libcxx][type_traits] Add C++20 changes to common_type
Michael Schellenberger Costa [Mon, 18 May 2020 12:01:55 +0000 (14:01 +0200)]
[libcxx][type_traits] Add C++20 changes to common_type

Summary: This already implements the expected changes for LWG-3205

Reviewers: mclow.lists, EricWF, CaseyCarter, cjdb, #libc, ldionne

Reviewed By: #libc, ldionne

Subscribers: dexonsmith, broadwaylamb, christof, ldionne, libcxx-commits

Tags: #libc

Differential Revision: https://reviews.llvm.org/D74350

4 years ago[libcxx testing] Remove ALLOW_RETRIES from lock_guard tests
David Zarzycki [Mon, 18 May 2020 11:39:12 +0000 (07:39 -0400)]
[libcxx testing] Remove ALLOW_RETRIES from lock_guard tests

These two tests were clumsily using time measurements to determine
whether std::lock_guard was working correctly. In practice, this
approach merely verified that the underlying lock properly waits.

Now these two tests verify that lock is acquired, not dropped
prematurely, and finally, actually dropped at the end of the scope.

4 years agoTurn -Wmax-tokens off by default
Hans Wennborg [Fri, 15 May 2020 15:20:27 +0000 (17:20 +0200)]
Turn -Wmax-tokens off by default

On the one hand, one might interpret the use of the max-token pragmas or
-fmax-tokens flag as an opt-in to the warning. However, in Chromium
we've found it useful to only opt in selected build configurations, even
though we have the pragmas in the code. For that reason, we think it
makes sense to turn it off by default.

Differential revision: https://reviews.llvm.org/D80014

4 years ago[InstCombine][NFC] Simplify check in sinking
Max Kazantsev [Mon, 18 May 2020 11:09:24 +0000 (18:09 +0700)]
[InstCombine][NFC] Simplify check in sinking

We just need to check that the only predecessor of user parent is
BB, we don't need to iterate through BB's successors for it.

4 years ago[OpenCL] Added addrspace_cast operator in C++ mode.
Anastasia Stulova [Mon, 18 May 2020 10:02:01 +0000 (11:02 +0100)]
[OpenCL] Added addrspace_cast operator in C++ mode.

This operator is intended for casting between
pointers to objects in different address spaces
and follows similar logic as const_cast in C++.

Tags: #clang

Differential Revision: https://reviews.llvm.org/D60193

4 years ago[AMDGPU][MC] Corrected branch relocation handling to detect undefined labels
Dmitry Preobrazhensky [Mon, 18 May 2020 11:04:58 +0000 (14:04 +0300)]
[AMDGPU][MC] Corrected branch relocation handling to detect undefined labels

Fixed ELF object writer to die gracefully when an undefined label is encountered in a branch instruction.
See https://bugs.llvm.org/show_bug.cgi?id=41914.

Reviewers: arsenm, rampitec

Differential Revision: https://reviews.llvm.org/D79943

4 years ago[clang-tidy] NFC: Cleanup Python scripts
Kirill Bobyrev [Mon, 18 May 2020 10:20:11 +0000 (12:20 +0200)]
[clang-tidy] NFC: Cleanup Python scripts

Summary: Silence few PEP8 warnings.

Reviewers: hokein

Reviewed By: hokein

Subscribers: Eugene.Zelenko, xazax.hun, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D78985

4 years ago[Test] Opportunity for sinking to unreachable in InstCombine
Max Kazantsev [Mon, 18 May 2020 09:25:16 +0000 (16:25 +0700)]
[Test] Opportunity for sinking to unreachable in InstCombine

4 years agoRevert 76c5f277f2 "Re-land [Debug][CodeView] Emit fully qualified names for globals"
Hans Wennborg [Mon, 18 May 2020 09:20:42 +0000 (11:20 +0200)]
Revert 76c5f277f2 "Re-land [Debug][CodeView] Emit fully qualified names for globals"

> Before this patch, S_[L|G][THREAD32|DATA32] records were emitted with a simple name, not the fully qualified name (namespace + class scope).
>
> Differential Revision: https://reviews.llvm.org/D79447

This causes asserts in Chromium builds:

CodeViewDebug.cpp:2997: void llvm::CodeViewDebug::emitDebugInfoForUDTs(const std::vector<std::pair<std::string, const DIType *>> &):
Assertion `OriginalSize == UDTs.size()' failed.

I will follow up on the Phabricator issue.

4 years ago[NFC][clang-tidy] use hasOperands in place of hasEitherOperand
Nathan James [Mon, 18 May 2020 09:11:22 +0000 (10:11 +0100)]
[NFC][clang-tidy] use hasOperands in place of hasEitherOperand

4 years ago[DebugInfo][DWARF] Emit a single location instead of a location list
OCHyams [Mon, 18 May 2020 08:25:29 +0000 (09:25 +0100)]
[DebugInfo][DWARF] Emit a single location instead of a location list

for variables in nested scopes (including inlined functions) if there is a
single location which covers the entire scope and the scope is contained in a
single block.

Based on work by @jmorse.

Reviewed By: vsk, aprantl

Differential Revision: https://reviews.llvm.org/D79571

4 years ago[compiler-rt] Don't error out on missing builtins when builting builtins themselves
Martin Storsjö [Mon, 18 May 2020 08:04:42 +0000 (11:04 +0300)]
[compiler-rt] Don't error out on missing builtins when builting builtins themselves

This fixes bootstrapping the builtins when no previous version of
them exists after 2fe66bdb2e5fb556e15d4a07158e89ec007c587b.

Also fix a whitespace issue in that commit.

4 years ago[TargetLibraryInfo] memcpy and memset have optimized codegen.
Clement Courbet [Mon, 27 Apr 2020 08:45:41 +0000 (10:45 +0200)]
[TargetLibraryInfo] memcpy and memset have optimized codegen.

Summary:
Add memcpy and memset to the list of function with optimized codegen.
hasOptimizedCodeGen() was returning false for them.
This was not an issue when working with c++ as clang typically emits an
intrinsic rather than a libfunc call, but when working with IR,
this would cause some issues with sanitizers which would not correctly
mark the function as nobuiltin in `maybeMarkSanitizerLibraryCallNoBuiltin()`.

Subscribers: llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D78991

4 years ago[RISCV] Fix passing two floating-point values in complex separately by two GPRs on...
Jim Lin [Mon, 18 May 2020 03:10:46 +0000 (11:10 +0800)]
[RISCV] Fix passing two floating-point values in complex separately by two GPRs on RV64

Summary:
This patch fixed the error of counting the remaining FPRs. Complex floating-point
values should be passed by two FPRs for the hard-float ABI. If no two FPRs are
available, it should be passed via a 64-bit GPR (fp+fp). `ArgFPRsLeft` is only
decreased one while the type is complex floating-point. It causes two floating-point
values in the complex are passed separately by two GPRs.

Reviewers: asb, luismarques, lenary

Reviewed By: asb

Subscribers: rbar, johnrusso, simoncook, sabuasal, niosHD, kito-cheng, shiva0217, jrtc27, zzheng, edward-jones, rogfer01, MartinMosbeck, brucehoult, the_o, rkruppe, PkmX, jocewei, psnobl, benna, s.egerton, pzheng, sameer.abuasal, apazos, evandro, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D79770

4 years ago[llvm-objcopy][MachO] Fix code formatting
Alexander Shaposhnikov [Mon, 18 May 2020 03:46:17 +0000 (20:46 -0700)]
[llvm-objcopy][MachO] Fix code formatting

Apply clang-format to MachOObjcopy.cpp. NFC.

Test plan: make check-all

4 years ago[PowerPC] add more high latency opcodes for machinecombiner - NFC
Chen Zheng [Mon, 18 May 2020 01:00:09 +0000 (21:00 -0400)]
[PowerPC] add more high latency opcodes for machinecombiner - NFC

4 years ago[MLIR][cmake] Fix unittest/Tablegen by special casing LLVMMLIRTableGen
Stephen Neuendorffer [Sun, 17 May 2020 22:17:04 +0000 (15:17 -0700)]
[MLIR][cmake] Fix unittest/Tablegen by special casing LLVMMLIRTableGen

unittest/Tablegen generates an executable that depends on MLIRIR and
LLVMMLIRTableGen.  Avoid specifying linkage dependence on LLVM
libraries here because then everyone has to depend on those libraries.

Differential Revision: https://reviews.llvm.org/D80093

4 years agoFix warning "defined but not used" for debug function (NFC)
Mehdi Amini [Sun, 17 May 2020 23:16:21 +0000 (23:16 +0000)]
Fix warning "defined but not used" for debug function (NFC)

4 years agoReplace dyn_cast with isa when the result isn't used (NFC)
Mehdi Amini [Sun, 17 May 2020 19:25:35 +0000 (19:25 +0000)]
Replace dyn_cast with isa when the result isn't used (NFC)

Fix build warning: unused variable 'BB'

4 years agoRemove spurious semicolon after function definition (NFC)
Mehdi Amini [Sun, 17 May 2020 19:24:33 +0000 (19:24 +0000)]
Remove spurious semicolon after function definition (NFC)

This fixes some GCC pedantic warnings.

4 years agoValueMapper does not preserve inline assembly dialect when remapping the type
Craig Topper [Sun, 17 May 2020 21:57:04 +0000 (14:57 -0700)]
ValueMapper does not preserve inline assembly dialect when remapping the type

Bug report: https://bugs.llvm.org/show_bug.cgi?id=45291

Patch by Tomasz Miąsko

Differential Revision: https://reviews.llvm.org/D80066

4 years ago[MLIR] Fix linkage for libMLIR.so
Stephen Neuendorffer [Sun, 17 May 2020 06:01:52 +0000 (23:01 -0700)]
[MLIR] Fix linkage for libMLIR.so

Generally:
1) don't use target_link_libraries() and add_mlir_library() on the same target, use LINK_LIBS PUBLIC instead.
2) don't use LINK_LIBS to specify LLVM libraries.  Use LINK_COMPONENTS instead
3) no need to link against LLVMSupport.  We pull it in by default.

Differential Revision: https://reviews.llvm.org/D80076

4 years ago[MLIR][cmake][NFC] Update linkage checker for mlir-opt
Stephen Neuendorffer [Sun, 17 May 2020 05:46:57 +0000 (22:46 -0700)]
[MLIR][cmake][NFC] Update linkage checker for mlir-opt

New CMakeLists.txt for MLIRStandardOpsTransforms was incorrect, but wasn't
caught by the check.

Differential Revision: https://reviews.llvm.org/D80075

4 years ago[Alignment] Remove unnecessary getValueOrABITypeAlignment calls (NFC)
Nikita Popov [Sun, 17 May 2020 20:14:42 +0000 (22:14 +0200)]
[Alignment] Remove unnecessary getValueOrABITypeAlignment calls (NFC)

Now that load/store alignment is required, we no longer need most
of them. Also switch the getLoadStoreAlignment() helper to return
Align instead of MaybeAlign.

4 years ago[InstCombine] visitMaskedMerge(): when unfolding, sanitize undef constants (PR45955)
Roman Lebedev [Sun, 17 May 2020 18:58:55 +0000 (21:58 +0300)]
[InstCombine] visitMaskedMerge(): when unfolding, sanitize undef constants (PR45955)

We can't leave undef vector element constants as-is,
it is a miscompile, so we need to sanitize them.

We have two vectors (C and ~C):
* We can't replace undef with 0 in both of them
* We can't replace undef with 0 in only one of them
* We could replace undef with -1 in both of them
* We could replace undef with -1 in only one(!) of them
* We could replace undef with -1 in one and 0 in another one of them.

Therefore, it seems best to go with the last option, since otherwise
we'd loose knowledge that C and ~C have no common bits set,
which seems more important than preserving partial undef knowledge.

Fixes https://bugs.llvm.org/show_bug.cgi?id=45955

4 years agoDebugInfo: Reduce long-distance dependence on what will/won't emit a debug_addr section
David Blaikie [Sun, 17 May 2020 19:17:31 +0000 (12:17 -0700)]
DebugInfo: Reduce long-distance dependence on what will/won't emit a debug_addr section

This is a no-op/NFC at the moment & generally makes the code /somewhat/
cleaner/less reliant on assumptions about what will produce a debug_addr
section.

It's still a bit "spooky action at a distance" - the add ranges code
pre-emptively inserts addresses into the address pool it knows will
eventually be used by the range emission code (or low/high pc).

The 'ideal' would be either to actually compute the addresses needed for
range (& loc) emission earlier - which would mean decanonicalizing the
range/loc representation earlier to account for whether it was going to
use addrx encodings or not (which would be unfortunate, but could be
refactored to be relatively unobtrusive).

Alternatively, emitting the range/loc sections earlier would cause them
to request the needed addresses sooner - but then you endup having to
split finalizeModuleInfo because some things need to be handled there
before the ranges/locs are emitted, I think...

4 years ago[LVI] Don't use dominator tree in isValidAssumeForContext()
Nikita Popov [Wed, 25 Mar 2020 19:51:01 +0000 (20:51 +0100)]
[LVI] Don't use dominator tree in isValidAssumeForContext()

LVI and its consumers currently have quite a bit of complexity
related to dominator tree management. However, it doesn't look
like it is actually needed...

The only use of the dominator tree is inside isValidAssumeForContext().
However, due to the way LVI queries work, it is not needed:
If we query a value for some block, we will first get the edge values
from all predecessor blocks, which also includes an intersection with
assumptions that apply to the terminator of the predecessor. As such,
we will already have processed all assumptions from predecessor blocks
(this is actually stronger than what isValidAssumeForContext() does
with a DT, because this is capable of combining non-dominating
assumptions). The only additional assumptions we need to take into
account are those in the block being queried. And we don't need a
dominator tree for that.

This patch only removes the use of DT, I will drop the machinery
around it in a followup.

Differential Revision: https://reviews.llvm.org/D76797

4 years agoAdd missing include Host.h in llvm-mc-* fuzzers. NFC.
Fedor Sergeev [Sun, 17 May 2020 19:17:56 +0000 (02:17 +0700)]
Add missing include Host.h in llvm-mc-* fuzzers. NFC.

Fixes build failure in these fuzzers.

4 years ago[ASTMatchers] Added BinaryOperator hasOperands matcher
Nathan James [Sat, 16 May 2020 01:00:33 +0000 (02:00 +0100)]
[ASTMatchers] Added BinaryOperator hasOperands matcher

Summary: Adds a matcher called `hasOperands` for `BinaryOperator`'s when you need to match both sides but the order isn't important, usually on commutative operators.

Reviewers: klimek, aaron.ballman, gribozavr2, alexfh

Reviewed By: aaron.ballman

Subscribers: cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D80054

4 years agotest commit
JinGu Kang [Sun, 17 May 2020 18:49:37 +0000 (19:49 +0100)]
test commit

4 years agoRevert rGca18ce1a00cd8b7cb7ce0e130440f5ae1ffe86ee "GlobPattern.h - remove unnecessary...
Simon Pilgrim [Sun, 17 May 2020 17:51:21 +0000 (18:51 +0100)]
Revert rGca18ce1a00cd8b7cb7ce0e130440f5ae1ffe86ee "GlobPattern.h - remove unnecessary BitVector.h/StringRef.h includes. NFC"

Causes lld build errors

4 years agoGlobPattern.h - remove unnecessary BitVector.h/StringRef.h includes. NFC
Simon Pilgrim [Sun, 17 May 2020 17:29:27 +0000 (18:29 +0100)]
GlobPattern.h - remove unnecessary BitVector.h/StringRef.h includes. NFC

Use forward declarations (BitVector already had one) and an headers to source file that were implicitly using them.

4 years agoImmutableGraph.h - remove unused raw_ostream.h include. NFC
Simon Pilgrim [Sun, 17 May 2020 17:03:15 +0000 (18:03 +0100)]
ImmutableGraph.h - remove unused raw_ostream.h include. NFC

4 years ago[XRay] Migrate xray_naive_log=true tests to xray_mode=xray-basic
Fangrui Song [Sun, 17 May 2020 16:31:09 +0000 (09:31 -0700)]
[XRay] Migrate xray_naive_log=true tests to xray_mode=xray-basic

4 years ago[x86] favor vector constant load to avoid GPR to XMM transfer
Sanjay Patel [Sun, 17 May 2020 15:56:26 +0000 (11:56 -0400)]
[x86] favor vector constant load to avoid GPR to XMM transfer

This build vector lowering pattern came up in D79886.
I've tried to limit the improvement to cases where it looks
clearly better to load, but we could remove the 'TODO'
predicates already if we are willing to overlook some
corner cases.

Differential Revision: https://reviews.llvm.org/D80013

4 years ago[InstCombine] add tests for FP cast of cast; NFC
Sanjay Patel [Sun, 17 May 2020 15:40:27 +0000 (11:40 -0400)]
[InstCombine] add tests for FP cast of cast; NFC

A fold of casts is proposed as a backend transform in D79187,
but we can also do that in IR (and that may obsolete the need
for a backend transform).

4 years ago[ObjectYAML][DWARF] Take into account other debug sections in DWARFYAML::Data::isEmpty().
Xing GUO [Sun, 17 May 2020 14:48:40 +0000 (22:48 +0800)]
[ObjectYAML][DWARF] Take into account other debug sections in DWARFYAML::Data::isEmpty().

4 years ago[AVR] Explicitly set the address of the data section when invoking the linker
Dylan McKay [Sun, 17 May 2020 14:18:54 +0000 (02:18 +1200)]
[AVR] Explicitly set the address of the data section when invoking the linker

This is required to get avr-gdb correctly showing values at the right
addresses. This problem was discovered by using debug symbols in an
external program to lookup values in an AVR simulator.

4 years ago[mlir] NFC - VectorTransforms use OpBuilder where relevant
Nicolas Vasilache [Sun, 17 May 2020 14:15:58 +0000 (10:15 -0400)]
[mlir] NFC - VectorTransforms use OpBuilder where relevant

Summary: This will allow using unrolling outside of only rewrite patterns.

Differential Revision: https://reviews.llvm.org/D80083

4 years ago[X86] Add getTargetConstantFromBasePtr helper. NFC.
Simon Pilgrim [Sun, 17 May 2020 13:58:17 +0000 (14:58 +0100)]
[X86] Add getTargetConstantFromBasePtr helper. NFC.

Allows us to share code from LoadSDNode and MemIntrinsicSDNode constant pool loads.

4 years ago[X86] getTargetConstantBitsFromNode - remove unnecessary X86ISD::VBROADCAST handling.
Simon Pilgrim [Sun, 17 May 2020 13:39:51 +0000 (14:39 +0100)]
[X86] getTargetConstantBitsFromNode - remove unnecessary X86ISD::VBROADCAST handling.

We create X86ISD::VBROADCAST_LOAD for constant pool folds now.

4 years ago[InstCombine] improve analysis of FP->int->FP to eliminate fpextend
Sanjay Patel [Sun, 17 May 2020 12:43:40 +0000 (08:43 -0400)]
[InstCombine] improve analysis of FP->int->FP to eliminate fpextend

This was originally in D79116.
Converting from a narrow-enough FP source value to integer and
back to FP guarantees that the conversion to FP is exact because
of UB/poison-on-overflow.

This was suggested in PR36617:
https://bugs.llvm.org/show_bug.cgi?id=36617#c19

4 years ago[LoopUnroll] Extend test case with additional loop with larger TC.
Florian Hahn [Sun, 17 May 2020 12:55:11 +0000 (13:55 +0100)]
[LoopUnroll] Extend test case with additional loop with larger TC.

4 years ago[LoopUnroll] Precommit test for PR459393.
Florian Hahn [Sun, 17 May 2020 12:26:51 +0000 (13:26 +0100)]
[LoopUnroll] Precommit test for PR459393.

4 years ago[AMDGPU] Enable base pointer.
Christudasan Devadasan [Tue, 21 Apr 2020 09:34:33 +0000 (15:04 +0530)]
[AMDGPU] Enable base pointer.

When the callee requires a dynamic stack realignment,
it is not possible to correcty access the incoming
stack arguments using the stack pointer. We reserve a
base pointer in such cases to access the function arguments
inside the callee. The base pointer will hold the incoming
stack pointer value before any kind of delta added to it.

Reviewed By: arsenm, scott.linder

Differential Revision: https://reviews.llvm.org/D78811

4 years ago[OpenMP] Fix race condition in the completion/freeing of detached tasks
Joachim Protze [Sun, 17 May 2020 10:25:02 +0000 (12:25 +0200)]
[OpenMP] Fix race condition in the completion/freeing of detached tasks

Spurious assertion failures are symptoms of a race condition for the handling
of detached tasks:
Assertion failure at kmp_tasking.cpp(3744): taskdata->td_flags.complete == 1.
Assertion failure at kmp_tasking.cpp(710): taskdata->td_flags.executing == 0.

in the case of detach=true, all accesses to taskdata in __kmp_task_finish need
to happen before (~line 873):

taskdata->td_flags.proxy = TASK_PROXY;

This assignment signals to __kmp_fulfill_event, that the task will need to be
freed there. So, conceptionally the ownership of taskdata is moved.

Reviewed By: AndreyChurbanov

Differential Revision: https://reviews.llvm.org/D79702

4 years ago[Inliner][NFC] silence gcc 'overloaded-virtual' warning on hiding of Pass::doInitiali...
Fedor Sergeev [Sun, 17 May 2020 09:31:33 +0000 (16:31 +0700)]
[Inliner][NFC] silence gcc 'overloaded-virtual' warning on hiding of Pass::doInitialization

When compiling with -Werror=overloaded-virtual, gcc emits this:
====
llvm/include/llvm/Pass.h:102:16: error: ‘virtual bool llvm::Pass::doInitialization(llvm::Module&)’ was hidden [-Werror=overloaded-virtual]
   virtual bool doInitialization(Module &)  { return false; }
                ^~~~~~~~~~~~~~~~
In file included from llvm/lib/Transforms/IPO/Inliner.cpp:20:0:
llvm/include/llvm/Transforms/IPO/Inliner.h:38:8: error:   by ‘virtual bool llvm::LegacyInlinerBase::doInitialization(llvm::CallGraph&)’ [-Werror=overloaded-virtual]
   bool doInitialization(CallGraph &CG) override;
        ^~~~~~~~~~~~~~~~
====

This is an old issue which has just started biting our downstream after
a slight rearrangement of includes around Inliner.
Fixing it similar to how doFinalization was done years ago.

4 years ago[LLVM][AVR] Support for R_AVR_6 fixup
Dylan McKay [Sun, 17 May 2020 04:46:32 +0000 (16:46 +1200)]
[LLVM][AVR] Support for R_AVR_6 fixup

Summary: Handle the emission of `R_AVR_6` ELF relocation type.

Reviewers: dylanmckay

Reviewed By: dylanmckay

Subscribers: hiraditya, Jim, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D78721

Patch by @LemonBoy https://reviews.llvm.org/p/LemonBoy/

4 years ago[AVR] Fix I/O instructions on XMEGA
Dylan McKay [Sun, 17 May 2020 04:25:12 +0000 (16:25 +1200)]
[AVR]  Fix I/O instructions on XMEGA

Summary:
On XMEGA, I/O address space is same as data address space - there is no 0x20 offset,
because CPU General Purpose Registers are not mapped in data address space.

From https://en.wikipedia.org/wiki/AVR_microcontrollers
> In the XMEGA variant, the working register file is not mapped into the data address space; as such, it is not possible to treat any of the XMEGA's working registers as though they were SRAM. Instead, the I/O registers are mapped into the data address space starting at the very beginning of the address space.

Reviewers: dylanmckay

Reviewed By: dylanmckay

Subscribers: hiraditya, Jim, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D77207

Patch by Vlastimil Labsky.

4 years ago[Driver] Render -T for Gnu.cpp
Fangrui Song [Sun, 17 May 2020 06:32:28 +0000 (23:32 -0700)]
[Driver] Render -T for Gnu.cpp

clang -T a.lds a.c currently does not render -T.

4 years ago[MLIR][cmake] use LINK_LIBS PUBLIC for MLIRStandardOpsTransforms
Stephen Neuendorffer [Sun, 17 May 2020 05:42:32 +0000 (22:42 -0700)]
[MLIR][cmake] use LINK_LIBS PUBLIC for MLIRStandardOpsTransforms

Without this LLVM_LINK_LLVM_DYLIB is broken

Differential Revision: https://reviews.llvm.org/D80074

4 years ago[llvm-xray] consumeError when trying big-endian
Fangrui Song [Sun, 17 May 2020 05:42:52 +0000 (22:42 -0700)]
[llvm-xray] consumeError when trying big-endian

Follow-up of rL341226.

Fixes "Expected<T> must be checked before access or destruction"

4 years ago[NFC] Run clang-format on ISDOpcodes.h
Arthur Eubanks [Fri, 15 May 2020 23:31:17 +0000 (16:31 -0700)]
[NFC] Run clang-format on ISDOpcodes.h

Subscribers: jfb, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D80050

4 years ago[Compiler-rt] Emit error if builtins library cannot be found
Yi Kong [Sat, 16 May 2020 03:59:21 +0000 (11:59 +0800)]
[Compiler-rt] Emit error if builtins library cannot be found

Since setting COMPILER_RT_USE_BUILTINS_LIBRARY would remove -z,defs
flag, missing builtins library would continue to build unnoticed.
Explicitly emit an error in such case.

Differential Revision: https://reviews.llvm.org/D79470

4 years agoFix a few doc typos to cycle bots.
Nico Weber [Sun, 17 May 2020 00:38:28 +0000 (20:38 -0400)]
Fix a few doc typos to cycle bots.

4 years agoTry to heal bots after https://reviews.llvm.org/D79655
Nico Weber [Sun, 17 May 2020 00:32:58 +0000 (20:32 -0400)]
Try to heal bots after https://reviews.llvm.org/D79655

4 years ago[LegalizeDAG] Use MachinePointerInfo::getUnknownStack in place of MachinePointerInfo...
Craig Topper [Sat, 16 May 2020 22:48:08 +0000 (15:48 -0700)]
[LegalizeDAG] Use MachinePointerInfo::getUnknownStack in place of MachinePointerInfo() in a couple places. NFC

We know the pointer somewhere on the stack, we just don't know
exactly where since the index may be variable.

Differential Revision: https://reviews.llvm.org/D80060

4 years agoAllocaInst should store Align instead of MaybeAlign.
Eli Friedman [Fri, 15 May 2020 20:23:14 +0000 (13:23 -0700)]
AllocaInst should store Align instead of MaybeAlign.

Along the lines of D77454 and D79968.  Unlike loads and stores, the
default alignment is getPrefTypeAlign, to match the existing handling in
various places, including SelectionDAG and InstCombine.

Differential Revision: https://reviews.llvm.org/D80044

4 years ago[X86] Replace selectScalarSSELoad ComplexPattern with PatFrags to handle the 3 types...
Craig Topper [Sat, 16 May 2020 19:38:34 +0000 (12:38 -0700)]
[X86] Replace selectScalarSSELoad ComplexPattern with PatFrags to handle the 3 types of loads we currently match.

This ensures we create mem operands for these instructions fixing PR45949.

Unfortunately, it increases the size of X86GenDAGISel.inc, but some dag
combine canonicalization could reduce the types of load we need to match.

4 years agoHarden IR and bitcode parsers against infinite size types.
Eli Friedman [Sat, 16 May 2020 21:01:54 +0000 (14:01 -0700)]
Harden IR and bitcode parsers against infinite size types.

If isSized is passed a SmallPtrSet, it uses that set to catch infinitely
recursive types (for example, a struct that has itself as a member).
Otherwise, it just crashes on such types.

4 years agoRevert "[nfc] test commit"
faisal vali [Sat, 16 May 2020 20:12:04 +0000 (15:12 -0500)]
Revert "[nfc] test commit"

This reverts commit 0ee46e857d81ea815e5b11d266c0c118a2c2e714.

4 years ago[nfc] test commit
faisal vali [Sat, 16 May 2020 20:08:30 +0000 (15:08 -0500)]
[nfc] test commit

4 years agoExpose IRGen API to add the default IR attributes to a function definition.
John McCall [Sat, 16 May 2020 18:44:54 +0000 (14:44 -0400)]
Expose IRGen API to add the default IR attributes to a function definition.

I've also made a stab at imposing some more order on where and how we add
attributes; this part should be NFC.  I wasn't sure whether the CUDA use
case for libdevice should propagate CPU/features attributes, so there's a
bit of unnecessary duplication.

4 years agoThe release notes for ObjCBreakBeforeNestedBlockParam was placed between the release...
mydeveloperday [Sat, 16 May 2020 17:48:13 +0000 (18:48 +0100)]
The release notes for ObjCBreakBeforeNestedBlockParam was placed between the release note for IndentCaseBlocks and its example code

Remove other whitespace and line limit issues and double blank line issues

4 years ago[VectorCombine] forward walk through instructions to improve chaining of transforms
Sanjay Patel [Sat, 16 May 2020 17:08:01 +0000 (13:08 -0400)]
[VectorCombine] forward walk through instructions to improve chaining of transforms

This is split off from D79799 - where I was proposing to fully iterate
over a function until there are no more transforms. I suspect we are
still going to want to do something like that eventually.

But we can achieve the same gains much more efficiently on the current
set of regression tests just by reversing the order that we visit the
instructions.

This may also reduce the motivation for D79078, but we are still not
getting the optimal pattern for a reduction.

4 years ago[PhaseOrdering] add vector reduction tests; NFC
Sanjay Patel [Sat, 16 May 2020 16:50:16 +0000 (12:50 -0400)]
[PhaseOrdering] add vector reduction tests; NFC

These are based on tests originally included in:
D79078

4 years ago[InstCombine] Clean up alignment handling (NFC)
Nikita Popov [Sat, 16 May 2020 16:46:03 +0000 (18:46 +0200)]
[InstCombine] Clean up alignment handling (NFC)

Now that load/store alignment is required, we can simplify code
in some places.

4 years ago[ARM] Patterns for VQSHRN
David Green [Sat, 16 May 2020 14:27:20 +0000 (15:27 +0100)]
[ARM] Patterns for VQSHRN

Given a VQMOVN(VSHR), we can fold that into a VQSHRN simply enough using
a few tablegen patterns.

Differential Revision: https://reviews.llvm.org/D77720

4 years ago[VectorCombine] add reduction-like patterns; NFC
Sanjay Patel [Sat, 16 May 2020 16:44:00 +0000 (12:44 -0400)]
[VectorCombine] add reduction-like patterns; NFC

These are based on tests originally included in:
D79078