platform/upstream/llvm.git
3 years ago[WebAssembly] Fix remapping branch dests in fixCatchUnwindMismatches
Heejin Ahn [Thu, 25 Feb 2021 20:50:04 +0000 (12:50 -0800)]
[WebAssembly] Fix remapping branch dests in fixCatchUnwindMismatches

This is a case D97178 tried to solve but missed. D97178 could not handle
the case when
multiple consecutive delegates are generated:
- Before:
```
block
  br (a)
  try
  catch
  end_try
end_block
          <- (a)
```

- After
```
block
  br (a)
  try
    ...
    try
      try
      catch
      end_try
            <- (a)
    delegate
  delegate
end_block
          <- (b)
```
(The `br` should point to (b) now)

D97178 assumed `end_block` exists two BBs later than `end_try`, because
it assumed the order as `end_try` BB -> `delegate` BB -> `end_block` BB.
But it turned out there can be multiple `delegate`s in between. This
patch changes the logic so we just search from `end_try` BB until we
find `end_block`.

Fixes https://github.com/emscripten-core/emscripten/issues/13515.
(More precisely, fixes
https://github.com/emscripten-core/emscripten/issues/13515#issuecomment-784711318.)

Reviewed By: dschuff, tlively

Differential Revision: https://reviews.llvm.org/D97569

3 years ago[tests] Precommit for upcoming patch
Philip Reames [Fri, 26 Feb 2021 21:11:13 +0000 (13:11 -0800)]
[tests] Precommit for upcoming patch

3 years ago[MLIR][TOSA] Lower tosa.reshape to linalg.reshape
Rob Suderman [Wed, 24 Feb 2021 22:12:03 +0000 (14:12 -0800)]
[MLIR][TOSA] Lower tosa.reshape to linalg.reshape

Lowering from the tosa.reshape op to linalg.reshape. For same-rank or
non-collapsed/expanded cases two linalg.reshapes are inserted.

Differential Revision: https://reviews.llvm.org/D97439

3 years ago[AMDGPU] Avoid second rescheduling for some regions
Stanislav Mekhanoshin [Thu, 25 Feb 2021 23:03:34 +0000 (15:03 -0800)]
[AMDGPU] Avoid second rescheduling for some regions

If a region was not constrained by a high register pressure
and was not rescheduled without clustering we can skip
rescheduling it ClusteredLowOccupancyReschedule stage.

This improves scheduling speed by 25% on some kernels.

Differential Revision: https://reviews.llvm.org/D97506

3 years ago[AMDGPU] Skip unclusterd rescheduling w/o ld/st
Stanislav Mekhanoshin [Tue, 23 Feb 2021 23:26:12 +0000 (15:26 -0800)]
[AMDGPU] Skip unclusterd rescheduling w/o ld/st

We are attempting rescheduling without load store clustering
if occupancy limits were not met with clustering. Skip this
for regions which do not have any loads or stores at all.

In a set of kernels I am experimenting with this improves
scheduling time by ~30%.

Differential Revision: https://reviews.llvm.org/D97342

3 years ago[SystemZ] Introducing assembler dialects for the Z backend
Anirudh Prasad [Fri, 26 Feb 2021 19:19:40 +0000 (14:19 -0500)]
[SystemZ] Introducing assembler dialects for the Z backend

- This patch introduces a different assembler dialect ("hlasm") for z/OS.
  The default dialect has now been given the "att" dialect name. For this
  appropriate changes have been added to SystemZ.td.
- This patch also makes a few changes to SystemZInstrFormats.td which
  restrict a few condition code mnemonics to just the "att" dialect
  variant (he, le, lh, nhe, nle, nlh). These extended condition code
  mnemonics are not available in HLASM.
- A new private function has been introduced in SystemZAsmParser.cpp to
  return the assembler dialect set in SystemZMCAsmInfo.cpp. The reason we
  couldn't/haven't explicitly queried the overriden getAssemblerDialect
  function from AsmParser is outlined in this thread here. This returned
  dialect is directly passed onto the relevant matcher functions which taken
  in a variantID, so that the matcher functions can appropriately choose an
  instruction based on the variant.

Reviewed By: uweigand

Differential Revision: https://reviews.llvm.org/D94250

3 years agoUse getAlign() on atomicrmw/cmpxchg instructions, now that it's available.
James Y Knight [Mon, 22 Feb 2021 20:36:20 +0000 (15:36 -0500)]
Use getAlign() on atomicrmw/cmpxchg instructions, now that it's available.

These locations were missed as part of adding alignment to the
instructions, and were still making their own alignment assumptions.

3 years ago[cgp] Minor code improvement - reuse an existing named helper [NFC]
Philip Reames [Fri, 26 Feb 2021 19:50:40 +0000 (11:50 -0800)]
[cgp] Minor code improvement - reuse an existing named helper [NFC]

3 years ago[dfsan] Do not test origin-tracking in atomic.cpp
Jianzhou Zhao [Fri, 26 Feb 2021 19:44:18 +0000 (19:44 +0000)]
[dfsan] Do not test origin-tracking in atomic.cpp

This would cause linking errors after https://reviews.llvm.org/D97483
that introduced new prefixes for ABI wrappers with origin tracking mode.
We will renable this after the full origin tracking is checked in.

3 years ago[RISCV] Call SelectBaseAddr on the base pointer in the custom isel for vector loads...
Craig Topper [Fri, 26 Feb 2021 18:53:55 +0000 (10:53 -0800)]
[RISCV] Call SelectBaseAddr on the base pointer in the custom isel for vector loads and stores.

This will allow FrameIndex as the base address instead of
emitting a separate ADDI from isel. eliminateFrameIndex will likely turn
it back into an ADDI, but this makes things consistent with the
SDPatterns and VLPatterns.

I only tested one case for simplicity. I can test more if reviewers
want.

Reviewed By: frasercrmck

Differential Revision: https://reviews.llvm.org/D97221

3 years agoBe more mathematicly precise about definition of recurrence [NFC]
Philip Reames [Fri, 26 Feb 2021 19:17:47 +0000 (11:17 -0800)]
Be more mathematicly precise about definition of recurrence [NFC]

This clarifies the interface of the matchSimpleRecurrence helper introduced in 8020be0b8 for non-commutative operators.  After ebd3aeba, I realized the original way I framed the routine was inconsistent.  For shifts, we only matched the the LHS form, but for sub we matched both and the caller wanted that information.  So, instead, we now consistently match both forms for non-commutative operators and the caller becomes responsible for filtering if needed.  I tried to put a clear warning in the header because I suspect the RHS form of e.g. a sub recurrence is non-obvious for most folks.  (It was for me.)

3 years ago[scudo][test] Disable -Wfree-nonheap-object
Leonard Chan [Fri, 26 Feb 2021 19:13:42 +0000 (11:13 -0800)]
[scudo][test] Disable -Wfree-nonheap-object

As of 4f395db86b5cc11bb56853323d3cb1d4b6db5a0b which contains updates to
-Wfree-nonheap-object, a line in this test will trigger the warning. This
particular line is ok though since it's meant to test a free on a bad pointer.

Differential Revision: https://reviews.llvm.org/D97516

3 years ago[dfsan] Propagate origins for callsites
Jianzhou Zhao [Thu, 25 Feb 2021 05:35:07 +0000 (05:35 +0000)]
[dfsan] Propagate origins for callsites

This is a part of https://reviews.llvm.org/D95835.

Each customized function has two wrappers. The
first one dfsw is for the normal shadow propagation. The second one dfso is used
when origin tracking is on. It calls the first one, and does additional
origin propagation. Which one to use can be decided at instrumentation
time. This is to ensure minimal additional overhead when origin tracking
is off.

Reviewed-by: morehouse
Differential Revision: https://reviews.llvm.org/D97483

3 years ago[clang-tidy][NFC] Tweak some generation of diag messages
Nathan James [Fri, 26 Feb 2021 19:10:25 +0000 (19:10 +0000)]
[clang-tidy][NFC] Tweak some generation of diag messages

Fix up cases where diag is called by piecing together a string in favour of placeholders.
Fix up cases where select could be used instead of duplicating the message for sake of 1 word difference.

Reviewed By: aaron.ballman

Differential Revision: https://reviews.llvm.org/D97488

3 years ago[SanitizerCoverage] Clarify llvm.used/llvm.compiler.used and partially fix unmatched...
Fangrui Song [Fri, 26 Feb 2021 19:10:02 +0000 (11:10 -0800)]
[SanitizerCoverage] Clarify llvm.used/llvm.compiler.used and partially fix unmatched metadata sections on Windows

`__sancov_pcs` parallels the other metadata section(s). While some optimizers
(e.g. GlobalDCE) respect linker semantics for comdat and retain or discard the
sections as a unit, some (e.g.  GlobalOpt/ConstantMerge) do not. So we have to
conservatively retain all unconditionally in the compiler.

When a comdat is used, the COFF/ELF linkers' GC semantics ensure the
associated parallel array elements are retained or discarded together,
so `llvm.compiler.used` is sufficient.

Otherwise (MachO (see rL311955/rL311959), COFF special case where comdat is not
used), we have to use `llvm.used` to conservatively make all sections retain by
the linker. This will fix the Windows problem once internal linkage
GlobalObject's in `llvm.used` are retained via `/INCLUDE:`.

Reviewed By: morehouse, vitalybuka

Differential Revision: https://reviews.llvm.org/D97432

3 years ago[clang-tidy] Remove some test c++ mode restrictions.
Nathan James [Fri, 26 Feb 2021 19:09:27 +0000 (19:09 +0000)]
[clang-tidy] Remove some test c++ mode restrictions.

Some comments are redundant, others just simple fixes.

Reviewed By: aaron.ballman

Differential Revision: https://reviews.llvm.org/D97544

3 years agoUse helper introduced in 8020be0b8 to simplify ValueTracking [NFC]
Philip Reames [Fri, 26 Feb 2021 18:46:28 +0000 (10:46 -0800)]
Use helper introduced in 8020be0b8 to simplify ValueTracking [NFC]

Direct rewrite of the code the helper was extracted from.

3 years ago[libcxx] cleans up __cpp_concepts mess
Christopher Di Bella [Fri, 19 Feb 2021 01:54:52 +0000 (01:54 +0000)]
[libcxx] cleans up __cpp_concepts mess

libc++ was previously a bit confused by what the value of __cpp_concepts
should be. Also replaces `__floating_point` with `floating_point` now
that it exists.

Differential Revision: https://reviews.llvm.org/D97015

3 years agoChange some addUsedGlobal to addUsedOrCompilerUsedGlobal
Fangrui Song [Fri, 26 Feb 2021 18:42:07 +0000 (10:42 -0800)]
Change some addUsedGlobal to addUsedOrCompilerUsedGlobal

An global value in the `llvm.used` list does not have GC root semantics on ELF targets.
This will be changed in a subsequent backend patch.

Change some `llvm.used` in the ELF code path to use `llvm.compiler.used` to
prevent undesired GC root semantics.

Change one extern "C" alias (due to `__attribute__((used))` in extern "C") to use `llvm.compiler.used` on all targets.

GNU ld has a rule "`__start_/__stop_` references from a live input section retain the associated C identifier name sections",
which LLD may drop entirely (currently refined to exclude SHF_LINK_ORDER/SHF_GROUP) in a future release (the rule makes it clumsy to GC metadata sections; D96914 added a way to try the potential future behavior).
For `llvm.used` global values defined in a C identifier name section, keep using `llvm.used` so that
the future LLD change will not affect them.

rnk kindly categorized the changes:
```
ObjC/blocks: this wants GC root semantics, since ObjC mainly runs on Mac.
MS C++ ABI stuff: wants GC root semantics, no change
OpenMP: unsure, but GC root semantics probably don't hurt
CodeGenModule: affected in this patch to *not* use GC root semantics so that __attribute__((used)) behavior remains the same on ELF, plus two other minor use cases that don't want GC semantics
Coverage: Probably want GC root semantics
CGExpr.cpp: refers to LTO, wants GC root
CGDeclCXX.cpp: one is MS ABI specific, so yes GC root, one is some other C++ init functionality, which should form GC roots (C++ initializers can have side effects and must run)
CGDecl.cpp: Changed in this patch for __attribute__((used))
```

Differential Revision: https://reviews.llvm.org/D97446

3 years agoAdd a helper for matching simple recurrence cycles
Philip Reames [Fri, 26 Feb 2021 18:18:56 +0000 (10:18 -0800)]
Add a helper for matching simple recurrence cycles

This helper came up in another review, and I've got about 4 different patches with copies of this copied into it.  Time to precommit the routine.  :)

3 years ago[libc++] Make the from-scratch config work on macOS out-of-the-box
Louis Dionne [Fri, 26 Feb 2021 18:11:00 +0000 (13:11 -0500)]
[libc++] Make the from-scratch config work on macOS out-of-the-box

3 years ago[InstCombine][NFC]Add a test for logical reductions.
Alexey Bataev [Fri, 26 Feb 2021 17:52:03 +0000 (09:52 -0800)]
[InstCombine][NFC]Add a test for logical reductions.

3 years ago[NFC][regalloc] const-ed APIs, using MCRegister instead of unsigned
Mircea Trofin [Fri, 26 Feb 2021 17:54:20 +0000 (09:54 -0800)]
[NFC][regalloc] const-ed APIs, using MCRegister instead of unsigned

3 years ago[NFC] Const-ed 2 APIs in VirtRegMap
Mircea Trofin [Fri, 26 Feb 2021 17:32:09 +0000 (09:32 -0800)]
[NFC] Const-ed 2 APIs in VirtRegMap

3 years ago[libc++] Allow running CI on macOS when Ninja isn't installed outside of Xcode
Louis Dionne [Fri, 26 Feb 2021 17:13:51 +0000 (12:13 -0500)]
[libc++] Allow running CI on macOS when Ninja isn't installed outside of Xcode

Xcode does bundle Ninja, so we can use that Ninja if there's no system-wide
Ninja installed. This is useful on some CI bots we have that don't come
with Ninja pre-installed.

3 years ago[NFC] MCRegister fixes in RegisterClassInfo, and const-ed APIs
Mircea Trofin [Fri, 26 Feb 2021 16:52:34 +0000 (08:52 -0800)]
[NFC] MCRegister fixes in RegisterClassInfo, and const-ed APIs

3 years ago[AMDGPU] Add selection pattern for v_xnor_b32
Jay Foad [Fri, 26 Feb 2021 16:31:11 +0000 (16:31 +0000)]
[AMDGPU] Add selection pattern for v_xnor_b32

This allows GlobalISel to use this instruction where available. I assume
SelectionDAG always selects s_xnor_b32 so it isn't affected by this
change.

Differential Revision: https://reviews.llvm.org/D97560

3 years ago[MLIR][linalg] Make integer matmul ops cast before multiplying
Geoffrey Martin-Noble [Fri, 26 Feb 2021 01:20:25 +0000 (17:20 -0800)]
[MLIR][linalg] Make integer matmul ops cast before multiplying

Right now they multiply before casting which means they would frequently
overflow. There are various reasonable ways to do this, but until we
have robust op description infra, this is a simple and safe default. More
careful treatments are likely to be hardware specific, as well (e.g.
using an i8*i8->i16 mul instruction).

Reviewed By: nicolasvasilache, mravishankar

Differential Revision: https://reviews.llvm.org/D97505

3 years ago[PDB] Fix unsigned integer overflow
Nico Weber [Fri, 26 Feb 2021 16:14:26 +0000 (11:14 -0500)]
[PDB] Fix unsigned integer overflow

When building with -fsanitize=unsigned-integer-overflow, this code
causes a diagnostic like:

    ../../llvm/lib/DebugInfo/PDB/Native/GSIStreamBuilder.cpp:159:15:
    runtime error: unsigned integer overflow:
    90 - 229 cannot be represented in type 'unsigned long'

unsigned integer overflow is well defined and it isn't an issue in
practice, but in obscure scenarios (S1.size() small, S2.size over 2GB
on 32-bit systems) it could even be a bug.

So use the usual idiom for implementing cmp functions instead of the
gernally considered buggy idiom :)
See e.g. https://www.gnu.org/software/libc/manual/html_node/Comparison-Functions.html
or https://stackoverflow.com/questions/10996418/efficient-integer-compare-function/10997428#10997428

Differential Revision: https://reviews.llvm.org/D97557

3 years ago[X86][AVX] SimplifyDemandedBitsForTargetNode - add basic X86ISD::VBROADCAST handling.
Simon Pilgrim [Fri, 26 Feb 2021 16:12:59 +0000 (16:12 +0000)]
[X86][AVX] SimplifyDemandedBitsForTargetNode - add basic X86ISD::VBROADCAST handling.

Simplify through to the scalar/vector source operand.

3 years ago[llvm][utils] Fix innocuous off by one in lldb formatters
Dave Lee [Fri, 26 Feb 2021 16:04:53 +0000 (08:04 -0800)]
[llvm][utils] Fix innocuous off by one in lldb formatters

num_children is "last_index" + 1, thus
    num_children + 1 = "last_index" + 2

this worked anyway because the index of `$$dereference$$` would work as long as
it was past the last index.

3 years agoFix assert to use getTypeStoreSize instead of getPrimitiveSizeInBits,
James Y Knight [Fri, 26 Feb 2021 15:33:24 +0000 (10:33 -0500)]
Fix assert to use getTypeStoreSize instead of getPrimitiveSizeInBits,
per comment on D97223.

3 years ago[AMDGPU] Better codegen for i64 bitreverse
Jay Foad [Fri, 26 Feb 2021 13:40:03 +0000 (13:40 +0000)]
[AMDGPU] Better codegen for i64 bitreverse

Differential Revision: https://reviews.llvm.org/D97547

3 years ago[llvm][utils] Rename lldb dict variables to internal_dict
Dave Lee [Fri, 26 Feb 2021 15:45:04 +0000 (07:45 -0800)]
[llvm][utils] Rename lldb dict variables to internal_dict

Most lldb scripts use `internal_dict`. Also, `dict` is a builtin constructor,
it's good habit to avoid using it as a variable name.

3 years ago[llvm][utils] Support dereferencing llvm::Optional lldb formatter
Dave Lee [Mon, 22 Feb 2021 05:45:04 +0000 (21:45 -0800)]
[llvm][utils] Support dereferencing llvm::Optional lldb formatter

Add deref support to `llvm::Optional` in `lldbDataFormatters.py`.

This creates a synthetic provider that adds dereference support, but otherwise proxies all access to the underlying value.

With this change, an optional value can be displayed by running `v *someOptional`, and its contents can be accessed with the arrow `operator->`, for example `v someOpt->HasThing`. This matches expressions usable from expression evaluation.

See also D97165 and D97524.

Differential Revision: https://reviews.llvm.org/D97525

3 years ago[lldb] Add deref support to libc++ unique_ptr synthetic
Dave Lee [Thu, 25 Feb 2021 21:27:22 +0000 (13:27 -0800)]
[lldb] Add deref support to libc++ unique_ptr synthetic

Add frame variable dereference suppport to libc++ `std::unique_ptr`.

This change allows for commands like `v *thing_up` and `v thing_up->m_id`. These commands now work the same way they would with raw pointers, and as they would with expression. This is done by adding an unaccounted for child member named `$$dereference$$`.

Without this change, the command would have to be written as `v *thing_up.__value_` or v thing_up.__value_->m_id` which exposes internal structure and is more clumsy to type.

Additionally, the existing tests were updated. See also https://reviews.llvm.org/D97165 which added deref support for `std::shared_ptr`.

Differential Revision: https://reviews.llvm.org/D97524

3 years ago[ADT][NFC] Add extra typedefs to `ArrayRef` and `MutableArrayRef`
Vladislav Vinogradov [Thu, 28 Jan 2021 12:51:58 +0000 (15:51 +0300)]
[ADT][NFC] Add extra typedefs to `ArrayRef` and `MutableArrayRef`

* `value_type`
* `pointer`
* `const_pointer`
* `reference`
* `const_reference`
* `const_reverse_iterator`
* `size_type`
* `difference_type`

It makes `ArrayRef` and `MutableArrayRef` types fully compliant with STL Container concept.

Reviewed By: lattner, courbet

Differential Revision: https://reviews.llvm.org/D95611

3 years ago[libc++] Remove the now unused macos-trunk and macos-backdeployment CI scripts
Louis Dionne [Fri, 26 Feb 2021 15:26:33 +0000 (10:26 -0500)]
[libc++] Remove the now unused macos-trunk and macos-backdeployment CI scripts

We use the run-buildbot script everywhere now.

3 years ago[mlir][linalg] Add pooling ops to Linalg TC ops.
Hanhan Wang [Fri, 26 Feb 2021 15:17:02 +0000 (07:17 -0800)]
[mlir][linalg] Add pooling ops to Linalg TC ops.

- Add EDSC builders for std_cmpf_ogt and std_cmpf_olt.
- Add pooling_nhwc_min/max/sum ops

Depends On D97384

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D97385

3 years ago[Utils] collectBitParts - bail for integers > 128-bits
Simon Pilgrim [Fri, 26 Feb 2021 14:55:02 +0000 (14:55 +0000)]
[Utils] collectBitParts - bail for integers > 128-bits

collectBitParts uses int8_t for the bit indices, leaving a 128-bit limit.

We already test for this before calling collectBitParts, but rGb94c215592bd added truncate handling which meant we could end up processing wider integers.

Thanks to @manojgupta for the repro.

3 years ago[mlir][linalg] Allow TC ops taking an unused shaped operand.
Hanhan Wang [Fri, 26 Feb 2021 14:45:08 +0000 (06:45 -0800)]
[mlir][linalg] Allow TC ops taking an unused shaped operand.

If one operand is not used in the formula, it will be considered a
shaped operand. And the result of indexing map of the operand will be the first
reduction dims.

Depends On D97383

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D97384

3 years ago[mlir] Silence some deprecation warnings after dffc487b07d9a50f8c65eccb9241f72af3486c79
Benjamin Kramer [Fri, 26 Feb 2021 14:14:16 +0000 (15:14 +0100)]
[mlir] Silence some deprecation warnings after dffc487b07d9a50f8c65eccb9241f72af3486c79

3 years ago[libcxxabi] Fewer assumptions about path from libcxx to libcxxabi
Nico Weber [Fri, 26 Feb 2021 14:00:32 +0000 (09:00 -0500)]
[libcxxabi] Fewer assumptions about path from libcxx to libcxxabi

This is useful for projects that pull in libcxx and libcxxabi and build
them using out-of-tree build files, but don't make them sibling
directories (or don't call the sibling directories libcxx and libcxxabi
for some reason).

Fixes PR49313.

Differential Revision: https://reviews.llvm.org/D97379

3 years ago[clangd][remote] Add flag to set idletimeout
Kadir Cetinkaya [Mon, 22 Feb 2021 08:49:08 +0000 (09:49 +0100)]
[clangd][remote] Add flag to set idletimeout

By default gRPC has no idletimeout and some firewalls might drop idle
connections after a certain period. This results in idle clients
shouting into void until server resets the connection.

Differential Revision: https://reviews.llvm.org/D97536

3 years ago[X86] Allow PTILEZEROV and PTILELOADDV to be rematerializable
Wang, Pengfei [Fri, 26 Feb 2021 13:20:03 +0000 (21:20 +0800)]
[X86] Allow PTILEZEROV and PTILELOADDV to be rematerializable

Spilling and reloading AMX registers are expensive. We allow PTILEZEROV
and PTILELOADDV to be rematerializable to avoid the register spilling.

Reviewed By: LuoYuanke

Differential Revision: https://reviews.llvm.org/D97453

3 years ago[mlir] Replace deprecated 'getAttrs'
Marius Brehler [Fri, 26 Feb 2021 13:28:32 +0000 (13:28 +0000)]
[mlir] Replace deprecated 'getAttrs'

'getAttrs' has been explicitly marked deprecated. This patch refactors
to use Operation::getAttrs().

Reviewed By: csigg

Differential Revision: https://reviews.llvm.org/D97546

3 years ago[InstCombine] Avoid redundant or out-of-order debug value sinking
Stephen Tozer [Fri, 26 Feb 2021 13:04:33 +0000 (13:04 +0000)]
[InstCombine] Avoid redundant or out-of-order debug value sinking

This patch modifies TryToSinkInstruction in the InstCombine pass, to prevent
redundant debug intrinsics from being produced, and also prevent the intrinsics
from being emitted in an incorrect order. It does this by ensuring that when
this pass sinks an instruction and creates clones of the debug intrinsics that
use that instruction, it inserts those debug intrinsics in their original order,
and only inserts the last debug intrinsic for each variable in the Instruction's
block.

Differential revision: https://reviews.llvm.org/D95463

3 years agoRevert "[NARY-REASSOCIATE] Support reassociation of min/max"
Evgeniy Brevnov [Fri, 26 Feb 2021 12:23:32 +0000 (19:23 +0700)]
Revert "[NARY-REASSOCIATE] Support reassociation of min/max"

This reverts commit 83d134c3c4222e8b8d3d90c099f749a3b3abc8e0.

3 years ago[clang][cli] Implement '-cuid=' marshalling
Jan Svoboda [Thu, 25 Feb 2021 12:20:59 +0000 (13:20 +0100)]
[clang][cli] Implement '-cuid=' marshalling

This patch adds marshalling to the `-cuid=` option introduced in D95007.

Reviewed By: Bigcheese

Differential Revision: https://reviews.llvm.org/D97461

3 years ago[clang][cli] Fix build failure in CompilerInvocation
Jan Svoboda [Fri, 26 Feb 2021 11:54:42 +0000 (12:54 +0100)]
[clang][cli] Fix build failure in CompilerInvocation

There is a report that https://reviews.llvm.org/D96280 causes a build failure with error: too few template arguments for class template 'SmallVector'. This patch attempts to fix that by explicitly specifying N for SmallVector<T, N>.

3 years ago[Orc] Use extensible RTTI for the orc::ObjectLayer class hierarchy
Stefan Gränitz [Fri, 26 Feb 2021 12:11:57 +0000 (13:11 +0100)]
[Orc] Use extensible RTTI for the orc::ObjectLayer class hierarchy

So far we had no way to distinguish between JITLink and RuntimeDyld in lli. Instead, we used implicit knowledge that RuntimeDyld would be used for linking ELF. In order to get D97337 to work with lli though, we have to move on and allow JITLink for ELF. This patch uses extensible RTTI to allow external clients to add their own layers without touching the LLVM sources.

Reviewed By: lhames

Differential Revision: https://reviews.llvm.org/D97338

3 years ago[X86] Remove unnecessary custom lowering of vXi1 SADDSAT/SSUBSAT/UADDSAT/USUBSAT
Simon Pilgrim [Fri, 26 Feb 2021 12:10:10 +0000 (12:10 +0000)]
[X86] Remove unnecessary custom lowering of vXi1 SADDSAT/SSUBSAT/UADDSAT/USUBSAT

As discussed on D97478. The removal of the custom tag causes some changes in the add/sub-overflow expansion as it no longer expands to sat-arith codegen.

3 years ago[Driver] Flush file in locked area
Serge Pavlov [Fri, 26 Feb 2021 10:44:29 +0000 (17:44 +0700)]
[Driver] Flush file in locked area

When writing report file by option -proc-stat-report some part of output
can be written to unlocked file because destructor of raw_fd_ostream
calls `flush()`. In high thread contention environment it can result in
file operation failure. With this change `flush` is called explicitly when
file is locked, so call of `flush()` in the destructor does not cause
write to file.

3 years ago[docs][JITLink] Few typo fixes in JITLink design/API doc
Stefan Gränitz [Fri, 26 Feb 2021 11:56:31 +0000 (12:56 +0100)]
[docs][JITLink] Few typo fixes in JITLink design/API doc

3 years ago[DAG] Fold vXi1 multiplies -> and
Simon Pilgrim [Fri, 26 Feb 2021 11:45:59 +0000 (11:45 +0000)]
[DAG] Fold vXi1 multiplies -> and

This allows us to remove X86 custom lowering of vXi1 MUL, which helps simplify a load of mask math.

Mentioned in D97478 post review.

3 years ago[X86] Remove unnecessary custom lowering of v16i1/v32i1 ADD/SUB
Simon Pilgrim [Fri, 26 Feb 2021 11:10:54 +0000 (11:10 +0000)]
[X86] Remove unnecessary custom lowering of v16i1/v32i1 ADD/SUB

These were missed in D97478

3 years ago[mlir] Mark OpState::removeAttr() deprecated.
Christian Sigg [Fri, 26 Feb 2021 09:08:40 +0000 (10:08 +0100)]
[mlir] Mark OpState::removeAttr() deprecated.

Fix call sites.

The method will be removed 2 weeks later.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D97530

3 years ago[DAG] expandAddSubSat - break if-else chain. NFCI.
Simon Pilgrim [Fri, 26 Feb 2021 10:59:54 +0000 (10:59 +0000)]
[DAG] expandAddSubSat - break if-else chain. NFCI.

Fix styleguide issue - each if() block always returns so we don't need to make them a if-else chain.

3 years ago[Test] Two more interesting test cases & their codegen counterparts
Max Kazantsev [Fri, 26 Feb 2021 10:19:50 +0000 (17:19 +0700)]
[Test] Two more interesting test cases & their codegen counterparts

3 years ago[MLIR][affine] Certain Call Ops to prevent fusion
Vinayaka Bandishti [Fri, 26 Feb 2021 09:56:12 +0000 (15:26 +0530)]
[MLIR][affine] Certain Call Ops to prevent fusion

Fixes a bug in affine fusion pipeline where an incorrect fusion is performed
despite a Call Op that potentially modifies memrefs under consideration
exists between source and target.

Fixes part of https://bugs.llvm.org/show_bug.cgi?id=49220

Reviewed By: bondhugula, dcaballe

Differential Revision: https://reviews.llvm.org/D97252

3 years ago[RISCV] Use existing method for the LMUL1 type. NFCI.
Fraser Cormack [Thu, 25 Feb 2021 14:23:59 +0000 (14:23 +0000)]
[RISCV] Use existing method for the LMUL1 type. NFCI.

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D97467

3 years ago[llvm-exegesis][X86] Ignore a few unmeasurable opcodes.
Clement Courbet [Mon, 2 Nov 2020 14:01:21 +0000 (15:01 +0100)]
[llvm-exegesis][X86] Ignore a few unmeasurable opcodes.

Differential Revision: https://reviews.llvm.org/D90744

3 years ago[Sema] Fix MSVC "attribute is not recognized" warning (NFC)
Yang Fan [Fri, 26 Feb 2021 09:27:44 +0000 (17:27 +0800)]
[Sema] Fix MSVC "attribute is not recognized" warning (NFC)

MSVC warning:
```
\llvm-project\clang\lib\Sema\SemaChecking.cpp(10323): warning C5030: attribute 'clang::fallthrough' is not recognized
```

3 years ago[Test] Add one more test with corner cases for non-wrapping IVs
Max Kazantsev [Fri, 26 Feb 2021 08:44:06 +0000 (15:44 +0700)]
[Test] Add one more test with corner cases for non-wrapping IVs

3 years ago[Basic] Fix Wreturn-type gcc warning (NFC)
Yang Fan [Fri, 26 Feb 2021 09:14:13 +0000 (17:14 +0800)]
[Basic] Fix Wreturn-type gcc warning (NFC)

3 years ago[clang][NFC] Extract Target and AuxTarget creation in CompilerInstance to new function
Yu-Hsun Chiang [Fri, 26 Feb 2021 08:12:32 +0000 (09:12 +0100)]
[clang][NFC] Extract Target and AuxTarget creation in CompilerInstance to new function

As @sammccall mentioned in [[ https://reviews.llvm.org/D97109 | D97109 ]], I've extract the logic of creating Target and AuxTarget into a new function called `createTargetAndAuxTarget`.

Since there are many similar code in clang or other related tools, consolidating them into a single function may help others to maintain the logic handling target related things.

Reviewed By: sammccall

Differential Revision: https://reviews.llvm.org/D97493

3 years ago[clang-format] [PR19056] Add support for access modifiers indentation
Jakub Budiský [Fri, 26 Feb 2021 08:05:35 +0000 (09:05 +0100)]
[clang-format] [PR19056] Add support for access modifiers indentation

Adds support for coding styles that make a separate indentation level for access modifiers, such as Code::Blocks or QtCreator.

The new option, `IndentAccessModifiers`, if enabled, forces the content inside classes, structs and unions (“records”) to be indented twice while removing a level for access modifiers. The value of `AccessModifierOffset` is disregarded in this case, aiming towards an ease of use.

======
The PR (https://bugs.llvm.org/show_bug.cgi?id=19056) had an implementation attempt by @MyDeveloperDay already (https://reviews.llvm.org/D60225) but I've decided to start from scratch. They differ in functionality, chosen approaches, and even the option name. The code tries to re-use the existing functionality to achieve this behavior, limiting possibility of breaking something else.

Reviewed By: MyDeveloperDay, curdeius, HazardyKnusperkeks

Differential Revision: https://reviews.llvm.org/D94661

3 years ago[mlir] Remove some rarely used OpState members and use Operation members instead.
Christian Sigg [Thu, 25 Feb 2021 21:20:45 +0000 (22:20 +0100)]
[mlir] Remove some rarely used OpState members and use Operation members instead.

Skipping the deprecation dance here.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D97494

3 years ago[Driver] Don't pass -ffile-compilation-dir through to cc1
Petr Hosek [Fri, 26 Feb 2021 06:27:18 +0000 (22:27 -0800)]
[Driver] Don't pass -ffile-compilation-dir through to cc1

This is a driver only flag so it has to be expanded when invoking cc1.

Differential Revision: https://reviews.llvm.org/D97528

3 years ago[RISCV] Add test cases for fixed vector extract element with non-constant index. NFC
Craig Topper [Fri, 26 Feb 2021 05:16:47 +0000 (21:16 -0800)]
[RISCV] Add test cases for fixed vector extract element with non-constant index. NFC

3 years ago[MC] Allow .cfi_sections with empty section list
Fangrui Song [Fri, 26 Feb 2021 06:29:49 +0000 (22:29 -0800)]
[MC] Allow .cfi_sections with empty section list

GNU as supports this. This mode silently ignores
.cfi_startproc/.cfi_endproc and .cfi_* in between.

Also drop a diagnostic `in '.cfi_sections' directive`: the diagnostic
already includes the line and it is clear the line is a `.cfi_sections` directive.

3 years ago[Driver] Rename -fprofile-{prefix-map,compilation-dir} to -fcoverage-{prefix-map...
Petr Hosek [Thu, 25 Feb 2021 03:04:36 +0000 (19:04 -0800)]
[Driver] Rename -fprofile-{prefix-map,compilation-dir} to -fcoverage-{prefix-map,compilation-dir}

These flags affect coverage mapping (-fcoverage-mapping), not
-fprofile-[instr-]generate so it makes more sense to use the
-fcoverage-* prefix.

Differential Revision: https://reviews.llvm.org/D97434

3 years ago[Driver] Create -ffile-compilation-dir alias
Petr Hosek [Thu, 25 Feb 2021 01:48:27 +0000 (17:48 -0800)]
[Driver] Create -ffile-compilation-dir alias

We introduce -ffile-compilation-dir shorthand to avoid having to set
-fdebug-compilation-dir and -fprofile-compilation-dir separately. This
is similar to -ffile-prefix-map.

Differential Revision: https://reviews.llvm.org/D97433

3 years ago[X86] Use correct padding when in 16-bit mode
Bill Wendling [Tue, 23 Feb 2021 20:39:39 +0000 (12:39 -0800)]
[X86] Use correct padding when in 16-bit mode

In 16-bit mode, some of the nop patterns used in 32-bit mode can end up
mangling other instructions. For instance, an aligned "movz" instruction
may have the 0x66 and 0x67 prefixes omitted, because the nop that's used
messes things up.

       xorl    %ebx, %ebx
       .p2align 4, 0x90
       movzbl  (%esi,%ebx), %ecx

Use instead nop patterns we know 16-bit mode can handle.

Differential Revision: https://reviews.llvm.org/D97268

3 years ago[llvm-dwarfdump] StringMap -> MapVector to make iteration order stable
Fangrui Song [Fri, 26 Feb 2021 04:05:05 +0000 (20:05 -0800)]
[llvm-dwarfdump] StringMap -> MapVector to make iteration order stable

Exposed by D97396

3 years ago[Scalar] Use range-based for loops (NFC)
Kazu Hirata [Fri, 26 Feb 2021 03:54:38 +0000 (19:54 -0800)]
[Scalar] Use range-based for loops (NFC)

3 years ago[TableGen] Use ListSeparator (NFC)
Kazu Hirata [Fri, 26 Feb 2021 03:54:36 +0000 (19:54 -0800)]
[TableGen] Use ListSeparator (NFC)

3 years ago[flang][fir] Add array value operations.
Eric Schweitz [Wed, 24 Feb 2021 16:02:58 +0000 (08:02 -0800)]
[flang][fir] Add array value operations.

We lower expressions with rank > 0 to a set of high-level array operations.
These operations are then analyzed and refined to more primitve
operations in subsequent pass(es).

This patch upstreams these array operations and some other helper ops.

Authors: Eric Schweitz, Rajan Walia, Kiran Chandramohan, et.al.

https://github.com/flang-compiler/f18-llvm-project/pull/565

Differential Revision: https://reviews.llvm.org/D97421

3 years agoFix signed-compare warning.
Justin Lebar [Fri, 26 Feb 2021 02:14:40 +0000 (18:14 -0800)]
Fix signed-compare warning.

Introduced in my c90dac27e94ec354a3e8919556ac5bc89b62c731.

3 years agoRevert [MLIR][TOSA] Added Tosa to Standard/SCF Lowerings (const, if, while)
Rob Suderman [Fri, 26 Feb 2021 02:04:34 +0000 (18:04 -0800)]
Revert [MLIR][TOSA] Added Tosa to Standard/SCF Lowerings (const, if, while)

This reverts commit a813e9be5bc91203508bde239c1a15c5b8f8c0cc.

Results in an ASAN failure due to bypassing rewriter.

Differential Revision: https://reviews.llvm.org/D97518

3 years ago[debug-info] refactor emitDwarfUnitLength
Chen Zheng [Fri, 26 Feb 2021 01:39:45 +0000 (20:39 -0500)]
[debug-info] refactor emitDwarfUnitLength

remove `Hi` `Lo` argument from `emitDwarfUnitLength`, so we
can make caller of emitDwarfUnitLength easier.

Reviewed By: MaskRay, dblaikie, ikudrin

Differential Revision: https://reviews.llvm.org/D96409

3 years ago[clang] Print 32 candidates on the first failure, with -fshow-overloads=best.
Justin Lebar [Sun, 31 Jan 2021 03:00:24 +0000 (19:00 -0800)]
[clang] Print 32 candidates on the first failure, with -fshow-overloads=best.

Previously, -fshow-overloads=best always showed 4 candidates.  The
problem is, when this isn't enough, you're kind of up a creek; the only
option available is to recompile with different flags.  This can be
quite expensive!

With this change, we try to strike a compromise.  The *first* error with
more than 4 candidates will show up to 32 candidates.  All further
errors continue to show only 4 candidates.

The hope is that this way, users will have *some chance* of making
forward progress, without facing unbounded amounts of error spam.

Differential Revision: https://reviews.llvm.org/D95754

3 years agoRevert 7a0da8894348, "scudo: Support memory tagging in the secondary allocator."
Peter Collingbourne [Fri, 26 Feb 2021 00:50:02 +0000 (16:50 -0800)]
Revert 7a0da8894348, "scudo: Support memory tagging in the secondary allocator."

We measured a 2.5 seconds (17.5%) regression in Android boot time
performance with this change.

3 years ago[RISCV] Support fixed vector extract_element for FP types.
Craig Topper [Fri, 26 Feb 2021 00:30:21 +0000 (16:30 -0800)]
[RISCV] Support fixed vector extract_element for FP types.

3 years ago[ELF] Add -z start-stop-gc to let __start_/__stop_ not retain C identifier name sections
Fangrui Song [Thu, 25 Feb 2021 23:46:37 +0000 (15:46 -0800)]
[ELF] Add -z start-stop-gc to let __start_/__stop_ not retain C identifier name sections

For one metadata section usage, each text section references a metadata section.
The metadata sections have a C identifier name to allow the runtime to collect them via `__start_/__stop_` symbols.

Since `__start_`/`__stop_` references are always present from live sections, the
C identifier name sections appear like GC roots, which means they cannot be
discarded by `ld --gc-sections`.

To make such sections GCable, either SHF_LINK_ORDER or a section group is needed.

SHF_LINK_ORDER is not suitable for the references can be inlined into other functions
(See D97430:
Function A (in the section .text.A) references its `__sancov_guard` section.
Function B inlines A (so now .text.B references `__sancov_guard` - this is invalid with the semantics of SHF_LINK_ORDER).

In the linking stage,
if `.text.A` gets discarded, and `__sancov_guard` is retained via the reference from `.text.B`,
the output will be invalid because `__sancov_guard` references the discarded `.text.A`.
LLD errors "sh_link points to discarded section".
)

A section group have size overhead, and is cumbersome when there is just one metadata section.

Add `-z start-stop-gc` to drop the "__start_/__stop_ references retain
non-SHF_LINK_ORDER non-SHF_GROUP C identifier name sections" rule.
We reserve the rights to switch the default in the future.

Reviewed By: phosek, jrtc27

Differential Revision: https://reviews.llvm.org/D96914

3 years agoBPF: Add LLVMTransformUtils in CMakefile LINK_COMPONENTS
Yonghong Song [Thu, 25 Feb 2021 23:34:48 +0000 (15:34 -0800)]
BPF: Add LLVMTransformUtils in CMakefile LINK_COMPONENTS

Commit 1959ead525b8 ("BPF: Implement TTI.getCmpSelInstrCost()
properly") introduced a dependency on LLVMTransformUtils
library. Let us encode this dependency explicitly in
CMakefile to avoid build error.

3 years ago[Android] Use -l:libunwind.a with --rtlib=compiler-rt
Ryan Prichard [Wed, 10 Feb 2021 09:25:58 +0000 (01:25 -0800)]
[Android] Use -l:libunwind.a with --rtlib=compiler-rt

On Android, the unwinder isn't part of the C++ STL and isn't (in older
versions) exported from libc.so. Instead, the driver links the static
unwinder archive implicitly. Currently, the Android NDK implicitly
links libgcc.a to provide both builtins and the unwinder.

To support switching to compiler-rt builtins and libunwind, make
--rtlib=compiler-rt behave the same way on Android, and implicitly pass
-l:libunwind.a to the linker.

Adjust the -ldl logic. For the Android NDK, the unwinder (whether
libgcc.a or libunwind.a) is linked statically and calls a function in
the dynamic loader for finding unwind tables (e.g. dl_iterate_phdr).
On Android, this function is in libc.a for static executables and
libdl.so otherwise, so -ldl is needed. (glibc doesn't need -ldl because
its libc.so exports dl_iterate_phdr.)

Differential Revision: https://reviews.llvm.org/D96403

3 years ago[dfsan] Conservative solution to atomic load/store
Jianzhou Zhao [Tue, 23 Feb 2021 16:48:26 +0000 (16:48 +0000)]
[dfsan] Conservative solution to atomic load/store

DFSan at store does store shadow data; store app data; and at load does
load shadow data; load app data.

When an application data is atomic, one overtainting case is

thread A: load shadow
thread B: store shadow
thread B: store app
thread A: load app

If the application address had been used by other flows, thread A reads
previous shadow, causing overtainting.

The change is similar to MSan's solution.
1) enforce ordering of app load/store
2) load shadow after load app; store shadow before shadow app
3) do not track atomic store by reseting its shadow to be 0.
The last one is to address a case like this.

Thread A: load app
Thread B: store shadow
Thread A: load shadow
Thread B: store app

This approach eliminates overtainting as a trade-off between undertainting
flows via shadow data race.

Note that this change addresses only native atomic instructions, but
does not support builtin libcalls yet.
   https://llvm.org/docs/Atomics.html#libcalls-atomic

Reviewed-by: morehouse
Differential Revision: https://reviews.llvm.org/D97310

3 years ago[Clang][ASan] Correct AsanDtorKindToString to return non-void in default case
Kostya Kortchinsky [Thu, 25 Feb 2021 21:30:40 +0000 (13:30 -0800)]
[Clang][ASan] Correct AsanDtorKindToString to return non-void in default case

Post D96572, a warning started showing up for me:
`clang/lib/Basic/Sanitizers.cpp:73:1: warning: control reaches end of non-void function [-Wreturn-type]`

So this adds a default to the case to return invalid, which seems appropriate,
and appears to correct the issue.

Differential Revision: https://reviews.llvm.org/D97496

3 years agoAdd Alignment argument to IRBuilder CreateAtomicRMW and CreateAtomicCmpXchg.
James Y Knight [Tue, 9 Feb 2021 04:07:12 +0000 (23:07 -0500)]
Add Alignment argument to IRBuilder CreateAtomicRMW and CreateAtomicCmpXchg.

And then push those change throughout LLVM.

Keep the old signature in Clang's CGBuilder for now -- that will be
updated in a follow-on patch (D97224).

The MLIR LLVM-IR dialect is not updated to support the new alignment
attribute, but preserves its existing behavior.

Differential Revision: https://reviews.llvm.org/D97223

3 years ago[gn build] Port 4753a69a316b
LLVM GN Syncbot [Thu, 25 Feb 2021 23:16:39 +0000 (23:16 +0000)]
[gn build] Port 4753a69a316b

3 years ago[Remarks] Provide more information about auto-init calls
Francis Visoiu Mistrih [Thu, 25 Feb 2021 17:49:53 +0000 (09:49 -0800)]
[Remarks] Provide more information about auto-init calls

This now analyzes calls to both intrinsics and functions.

For intrinsics, grab the ones we know and care about (mem* family) and
analyze the arguments.

For calls, use TLI to get more information about the libcalls, then
analyze the arguments if known.

```
auto-init.c:4:7: remark: Call to memset inserted by -ftrivial-auto-var-init. Memory operation size: 4096 bytes. [-Rpass-missed=annotation-remarks]
  int var[1024];
      ^
```

Differential Revision: https://reviews.llvm.org/D97489

3 years ago[Remarks] Provide more information about auto-init stores
Francis Visoiu Mistrih [Wed, 24 Feb 2021 21:19:39 +0000 (13:19 -0800)]
[Remarks] Provide more information about auto-init stores

This adds support for analyzing the instruction with the !annotation
"auto-init" in order to generate a more user-friendly remark.

For now, support the store size, and whether it's atomic/volatile.

Example:

```
auto-init.c:4:7: remark: Store inserted by -ftrivial-auto-var-init.Store size: 4 bytes. [-Rpass-missed=annotation-remarks]
  int var;
      ^
```

Differential Revision: https://reviews.llvm.org/D97412

3 years ago[Remarks] Emit remarks for "auto-init" !annotations
Francis Visoiu Mistrih [Wed, 24 Feb 2021 19:17:19 +0000 (11:17 -0800)]
[Remarks] Emit remarks for "auto-init" !annotations

Using the !annotation metadata, emit remarks pointing to code added by
`-ftrivial-auto-var-init` that survived the optimizer.

Example:

```
auto-init.c:4:7: remark: Initialization inserted by -ftrivial-auto-var-init. [-Rpass-missed=annotation-remarks]
  int buf[1024];
      ^
```

The tests are testing various situations like calls/stores/other
instructions, with debug locations, and extra debug information on
purpose: more patches will come to improve the reporting to make it more
user-friendly, and these tests will show how the reporting evolves.

Differential Revision: https://reviews.llvm.org/D97405

3 years agoAdd more historic DWARF vendor extensions
Adrian Prantl [Wed, 24 Feb 2021 01:52:21 +0000 (17:52 -0800)]
Add more historic DWARF vendor extensions

The maintainer of libdwarf kindly provided this patch with a bunch of
historic DWARF extensions that are missing from Dwarf.def. This list
is helpful to avoid potential conflicts in the user-defined vendor
extension space in the future.

Patch by David Anderson!

[Relanded with an updated test.]

Differential Revision: https://reviews.llvm.org/D97242

3 years ago[Clang][Attributes] Allow not_tail_called attribute to be applied to virtual function.
Zequan Wu [Wed, 17 Feb 2021 02:45:56 +0000 (18:45 -0800)]
[Clang][Attributes] Allow not_tail_called attribute to be applied to virtual function.

It would be beneficial to allow not_tail_called attribute to be applied to
virtual functions. I don't see any drawback of allowing this.

Differential Revision: https://reviews.llvm.org/D96832

3 years agoBPF: Implement TTI.getCmpSelInstrCost() properly
Yonghong Song [Wed, 24 Feb 2021 18:02:48 +0000 (10:02 -0800)]
BPF: Implement TTI.getCmpSelInstrCost() properly

The Select insn in BPF is expensive as BPF backend
needs to resolve with conditionals.  This patch set
the getCmpSelInstrCost() to SCEVCheapExpansionBudget
for Select insn to prevent some Select insn related
optimizations.

This change is motivated during bcc code review for
   https://github.com/iovisor/bcc/pull/3270
where IndVarSimplifyPass eventually caused generating
the following asm code:
  ;       for (i = 0; (i < VIRTIO_MAX_SGS) && (i < num); i++) {
      14:       16 05 40 00 00 00 00 00 if w5 == 0 goto +64 <LBB0_6>
      15:       bc 51 00 00 00 00 00 00 w1 = w5
      16:       04 01 00 00 ff ff ff ff w1 += -1
      17:       67 05 00 00 20 00 00 00 r5 <<= 32
      18:       77 05 00 00 20 00 00 00 r5 >>= 32
      19:       a6 01 01 00 05 00 00 00 if w1 < 5 goto +1 <LBB0_4>
      20:       b7 05 00 00 06 00 00 00 r5 = 6
  00000000000000a8 <LBB0_4>:
      21:       b7 02 00 00 00 00 00 00 r2 = 0
      22:       b7 01 00 00 00 00 00 00 r1 = 0
  ;       for (i = 0; (i < VIRTIO_MAX_SGS) && (i < num); i++) {
      23:       7b 1a e0 ff 00 00 00 00 *(u64 *)(r10 - 32) = r1
      24:       7b 5a c0 ff 00 00 00 00 *(u64 *)(r10 - 64) = r5
Note that insn #15 has w1 = w5 and w1 is refined later but r5(w5) is
eventually saved on stack at insn #24 for later use. This cause
later verifier failures.

With this change, IndVarSimplifyPass won't do the above
transformation any more.

Differential Revision: https://reviews.llvm.org/D97479

3 years ago[www] Add cxx_status tracking for C++23.
Richard Smith [Thu, 25 Feb 2021 22:47:02 +0000 (14:47 -0800)]
[www] Add cxx_status tracking for C++23.

Convert the list of standards to a table; it's starting to get unwieldy.

3 years ago[MLIR][TOSA] Added Tosa to Standard/SCF Lowerings (const, if, while)
Rob Suderman [Thu, 25 Feb 2021 22:34:49 +0000 (14:34 -0800)]
[MLIR][TOSA] Added Tosa to Standard/SCF Lowerings (const, if, while)

Includes a lowering for tosa.const, tosa.if, and tosa.while to Standard/SCF dialects. TosaToStandard is
used for constant lowerings and TosaToSCF handles the if/while ops.

Reviewed By: silvas

Differential Revision: https://reviews.llvm.org/D97352

3 years agoFix constructor declarations that are invalid in C++20 onwards.
Richard Smith [Tue, 23 Feb 2021 22:07:13 +0000 (14:07 -0800)]
Fix constructor declarations that are invalid in C++20 onwards.

Under C++ CWG DR 2237, the constructor for a class template C must be
written as 'C(...)' not as 'C<T>(...)'. This fixes a build failure with
GCC in C++20 mode.

In passing, remove some other redundant '<T>' qualification from the
affected classes.