Haojian Wu [Mon, 19 Oct 2020 07:26:32 +0000 (09:26 +0200)]
[clang-rename] Fix rename on function template specializations.
previously, we missed to rename occurrences to explicit function template
specilizations.
Differential Revision: https://reviews.llvm.org/D89221
David Sherwood [Tue, 29 Sep 2020 11:08:40 +0000 (12:08 +0100)]
[SVE][CodeGen][NFC] Replace TypeSize comparison operators with their scalar equivalents
In certain places in llvm/lib/CodeGen we were relying upon the TypeSize
comparison operators when in fact the code was only ever expecting
either scalar values or fixed width vectors. I've changed some of these
places to use the equivalent scalar operator.
Differential Revision: https://reviews.llvm.org/D88482
Lang Hames [Mon, 19 Oct 2020 07:25:15 +0000 (00:25 -0700)]
[examples] Fix Kaleidoscope-Ch3, which was broken by
ad92f16ccc5.
David Sherwood [Wed, 7 Oct 2020 07:26:17 +0000 (08:26 +0100)]
[SVE] Replace TypeSize comparison operators in llvm/utils/TableGen
In CodeGenDAGPatterns.cpp we were relying upon TypeSize comparison
operators for ordering types, when we can actually just use the known
minimum size since the scalable property is already being taken into
account. Also, in TypeInfer::EnforceSameSize I fixed some implicit
TypeSize->uint64_t casts by changing the code to test the equality
of TypeSize objects instead.
In other places I have replaced calls to getSizeInBits() with
getFixedSizeInBits() because we are only ever expecting integer values.
Differential Revision: https://reviews.llvm.org/D88947
Lang Hames [Mon, 19 Oct 2020 07:15:58 +0000 (00:15 -0700)]
[ORC] Move SimpleResourceManager function object inits out of constructor.
MSVC doesn't seem to like capturing references to variables in lambdas passed to
the variable's constructor. This should fix the windows bots that have been
unable to build the new ResourceTracker unit test.
Lang Hames [Mon, 19 Oct 2020 06:24:39 +0000 (23:24 -0700)]
[ORC] Add a missing '&'.
David Sherwood [Tue, 29 Sep 2020 07:03:13 +0000 (08:03 +0100)]
[SVE][CodeGen] Replace uses of TypeSize comparison operators
In certain places in the code we can never end up in a situation where
we're mixing fixed width and scalable vector types. For example,
we can't have truncations and extends that change the lane count. Also,
in other places such as GenWidenVectorStores and GenWidenVectorLoads we
know from the behaviour of FindMemType that we can never choose a vector
type with a different scalable property.
In various places I have used EVT::bitsXY functions instead of
TypeSize::isKnownXY, where it probably makes sense to keep an assert
that scalable properties match.
Differential Revision: https://reviews.llvm.org/D88654
David Sherwood [Fri, 9 Oct 2020 11:03:20 +0000 (12:03 +0100)]
[SVE][AArch64] Replace TypeSize comparisons with their integer equivalents
In many places in the AArch64 backend we are comparing TypeSize objects,
but in fact we are only ever expecting fixed width types. I've changed
all such comparisons to use their integer equivalents by replacing
calls to getSizeInBits() with getFixedSizeInBits(), etc.
Differential Revision: https://reviews.llvm.org/D89116
Max Kazantsev [Mon, 19 Oct 2020 06:38:20 +0000 (13:38 +0700)]
[Test] Add one more SCEV range test
Max Kazantsev [Mon, 19 Oct 2020 05:55:35 +0000 (12:55 +0700)]
[NFC][SCEV] Use getMinusOne where possible
Christian Sigg [Wed, 14 Oct 2020 08:31:08 +0000 (10:31 +0200)]
[mlir][gpu] NFC: Make room for more than one GPU rewrite pattern.
AllReduceLowering is currently the only GPU rewrite pattern, but more are coming. This is a preparation change.
Reviewed By: herhut
Differential Revision: https://reviews.llvm.org/D89370
Fangrui Song [Mon, 19 Oct 2020 05:36:58 +0000 (22:36 -0700)]
[PrologEpilogInserter] Fix prolog-params.mir
Christian Sigg [Sun, 18 Oct 2020 05:40:42 +0000 (07:40 +0200)]
[mlir] Allow space literals (` `) in assemblyFormat.
Spaces are only printed, not parsed.
Reviewed By: rriddle
Differential Revision: https://reviews.llvm.org/D89585
Kai Luo [Mon, 19 Oct 2020 03:41:05 +0000 (03:41 +0000)]
[PowerPC] Skip combining (uint_to_fp x) if x is not simple type
Current powerpc64le backend hits
```
Combining: t7: f64 = uint_to_fp t6
llc: llvm-project/llvm/include/llvm/CodeGen/ValueTypes.h:291: llvm::MVT llvm::EVT::getSimpleVT() const: Assertion `isSimple() && "Expected a SimpleValueType!"' failed.
```
This patch fixes it by skipping combination if `t6` is not simple type.
Fixed https://bugs.llvm.org/show_bug.cgi?id=47660.
Reviewed By: #powerpc, steven.zhang
Differential Revision: https://reviews.llvm.org/D88388
Fangrui Song [Mon, 19 Oct 2020 05:17:22 +0000 (22:17 -0700)]
[PrologEpilogInserter] Reduce PR16393 test and fix a prologue parameter in a debuginfo test
LLVM GN Syncbot [Mon, 19 Oct 2020 04:23:08 +0000 (04:23 +0000)]
[gn build] Port
6154c4115cd
LLVM GN Syncbot [Mon, 19 Oct 2020 04:23:07 +0000 (04:23 +0000)]
[gn build] Port
0aec49c8531
Lang Hames [Wed, 23 Sep 2020 05:37:25 +0000 (22:37 -0700)]
[ORC][examples] Update Kaleidoscope and BuildingAJIT tutorial series to OrcV2.
This patch updates the Kaleidoscope and BuildingAJIT tutorial series (chapter
1-4) to OrcV2. Chapter 5 of the BuildingAJIT series is removed -- it will be
re-instated once we have in-tree support for out-of-process JITing.
This patch only updates the tutorial code, not the text. Patches welcome for
that, otherwise I will try to update it in a few weeks.
Lang Hames [Fri, 11 Sep 2020 16:50:41 +0000 (09:50 -0700)]
[ORC] Add support for resource tracking/removal (removable code).
This patch introduces new APIs to support resource tracking and removal in Orc.
It is intended as a thread-safe generalization of the removeModule concept from
OrcV1.
Clients can now create ResourceTracker objects (using
JITDylib::createResourceTracker) to track resources for each MaterializationUnit
(code, data, aliases, absolute symbols, etc.) added to the JIT. Every
MaterializationUnit will be associated with a ResourceTracker, and
ResourceTrackers can be re-used for multiple MaterializationUnits. Each JITDylib
has a default ResourceTracker that will be used for MaterializationUnits added
to that JITDylib if no ResourceTracker is explicitly specified.
Two operations can be performed on ResourceTrackers: transferTo and remove. The
transferTo operation transfers tracking of the resources to a different
ResourceTracker object, allowing ResourceTrackers to be merged to reduce
administrative overhead (the source tracker is invalidated in the process). The
remove operation removes all resources associated with a ResourceTracker,
including any symbols defined by MaterializationUnits associated with the
tracker, and also invalidates the tracker. These operations are thread safe, and
should work regardless of the the state of the MaterializationUnits. In the case
of resource transfer any existing resources associated with the source tracker
will be transferred to the destination tracker, and all future resources for
those units will be automatically associated with the destination tracker. In
the case of resource removal all already-allocated resources will be
deallocated, any if any program representations associated with the tracker have
not been compiled yet they will be destroyed. If any program representations are
currently being compiled then they will be prevented from completing: their
MaterializationResponsibility will return errors on any attempt to update the
JIT state.
Clients (usually Layer writers) wishing to track resources can implement the
ResourceManager API to receive notifications when ResourceTrackers are
transferred or removed. The MaterializationResponsibility::withResourceKeyDo
method can be used to create associations between the key for a ResourceTracker
and an allocated resource in a thread-safe way.
RTDyldObjectLinkingLayer and ObjectLinkingLayer are updated to use the
ResourceManager API to enable tracking and removal of memory allocated by the
JIT linker.
The new JITDylib::clear method can be used to trigger removal of every
ResourceTracker associated with the JITDylib (note that this will only
remove resources for the JITDylib, it does not run static destructors).
This patch includes unit tests showing basic usage. A follow-up patch will
update the Kaleidoscope and BuildingAJIT tutorial series to OrcV2 and will
use this API to release code associated with anonymous expressions.
Lang Hames [Tue, 8 Sep 2020 04:21:28 +0000 (21:21 -0700)]
[ORC] Remove OrcV1 APIs.
This removes all legacy layers, legacy utilities, the old Orc C bindings,
OrcMCJITReplacement, and OrcMCJITReplacement regression tests.
ExecutionEngine and MCJIT are not affected by this change.
John Demme [Mon, 19 Oct 2020 01:07:44 +0000 (01:07 +0000)]
[MLIR] ODS TypeDefs: getChecked() and internal enhancements
Have the ODS TypeDef generator write the getChecked() definition.
Also add to TypeParamCommaFormatter a `JustParams` format and
refactor around that.
Reviewed By: rriddle
Differential Revision: https://reviews.llvm.org/D89438
Craig Topper [Mon, 19 Oct 2020 00:16:14 +0000 (17:16 -0700)]
[X86] Sort the tables before printing in X86FoldTablesEmitter.
This makes diffing with the manual tables easier. And if we ever
directly use the autogenerated tables instead of the manual tables
we'll need them to be in sorted order for the binary search.
John Demme [Mon, 19 Oct 2020 00:20:53 +0000 (00:20 +0000)]
[MLIR] ODS TypeDef documentation
Adds documentation for the new ODS TypeDef support.
Reviewed By: rriddle
Differential Revision: https://reviews.llvm.org/D89440
Richard Smith [Sun, 18 Oct 2020 21:15:12 +0000 (14:15 -0700)]
PR47893: Synthesis of a comparison operator from an 'operator<=>'
inherits the SFINAEness of its enclosing context.
Richard Smith [Sun, 18 Oct 2020 20:52:31 +0000 (13:52 -0700)]
PR47870: Properly mangle placeholders for deduced class template
specializations that have no deduced type.
River Riddle [Sun, 18 Oct 2020 20:43:09 +0000 (13:43 -0700)]
[mlir] Add canonicalization for cond_br that feed into a cond_br on the same condition
```
...
cond_br %cond, ^bb1(...), ^bb2(...)
...
^bb1: // has single predecessor
...
cond_br %cond, ^bb3(...), ^bb4(...)
```
->
```
...
cond_br %cond, ^bb1(...), ^bb2(...)
...
^bb1: // has single predecessor
...
br ^bb3(...)
```
Differential Revision: https://reviews.llvm.org/D89604
Craig Topper [Sun, 18 Oct 2020 19:45:01 +0000 (12:45 -0700)]
[X86] Add test cases for PR47874. NFC
Alex Richardson [Sun, 18 Oct 2020 15:06:00 +0000 (16:06 +0100)]
[libc++][dsl] Run checks for locale names aliases using a single %exec
This changes the checking for available locales to use one program that
iterates over argv to test multiple locale names instead of checking each
name with a separate executable.
This massively speeds up running individual tests using an SSH executor
(it can take up to 10 seconds to compile and run a single test in some
emulated environments) in case no locales are installed since then all
fallback names are tested idividually. But even on a native machine
this reduces the libc++ lit startup time by ~1-2 second for me on a machine
that does not have locale data installed.
Reviewed By: #libc, ldionne
Differential Revision: https://reviews.llvm.org/D88884
Alex Richardson [Sun, 18 Oct 2020 15:05:36 +0000 (16:05 +0100)]
[clang-format] Add a SpaceAroundPointerQualifiers style option
Some projects (e.g. FreeBSD) align pointers to the right but expect a
space between the '*' and any pointer qualifiers such as const. To handle
these cases this patch adds a new config option SpaceAroundPointerQualifiers
that can be used to configure whether spaces need to be added before/after
pointer qualifiers.
PointerAlignment = Right
SpaceAroundPointerQualifiers = Default/After:
void *const *x = NULL;
SpaceAroundPointerQualifiers = Before/Both
void * const *x = NULL;
PointerAlignment = Left
SpaceAroundPointerQualifiers = Default/Before:
void* const* x = NULL;
SpaceAroundPointerQualifiers = After/Both
void* const * x = NULL;
PointerAlignment = Middle
SpaceAroundPointerQualifiers = Default/Before/After/Both:
void * const * x = NULL;
Reviewed By: MyDeveloperDay
Differential Revision: https://reviews.llvm.org/D88227
Alex Richardson [Sun, 18 Oct 2020 15:05:01 +0000 (16:05 +0100)]
[libc++] Fix aligned_alloc tests FreeBSD
On FreeBSD we get the following error when passing zero as the requested
alignment: error: requested alignment is not a power of 2
Reviewed By: #libc, ldionne
Differential Revision: https://reviews.llvm.org/D88820
Mark de Wever [Sun, 18 Oct 2020 12:11:41 +0000 (14:11 +0200)]
[NFC] Make non-modifying members const.
Implementing the likelihood attributes for the iteration statements adds
a new helper function. This function can't be const qualified since
these non-modifying members aren't const qualified.
Hubert Tong [Sun, 18 Oct 2020 14:30:14 +0000 (10:30 -0400)]
[PowerPC][AIX] Make `__vector [un]signed long` an error
The semantics associated with `__vector [un]signed long` are neither
consistently specified nor consistently implemented.
The IBM XL compilers on AIX traditionally treated these as deprecated
aliases for the corresponding `__vector int` type in both 32-bit and
64-bit modes. The newer, Clang-based, IBM XL compilers on AIX make usage
of the previously deprecated types an error. This is also consistent
with IBM XL C/C++ for Linux on Power (on little endian distributions).
In line with the above, this patch upgrades (on AIX) the deprecation of
`__vector long` to become removal.
Reviewed By: ZarkoCA
Differential Revision: https://reviews.llvm.org/D89443
Hubert Tong [Sun, 18 Oct 2020 15:41:52 +0000 (11:41 -0400)]
Fix various format specifier mismatches
Format specifiers of incorrect length are replaced with format specifier
macros from `<cinttypes>` matching the typedefs used to declare the type
of the value being printed.
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D89637
Hubert Tong [Sun, 18 Oct 2020 02:22:43 +0000 (22:22 -0400)]
[test][NFC] Do ptrdiff_t comparison with signed instead of unsigned constants
... because using unsigned constants for comparing against signed values
is liable to mutate the signed value via conversion to an unsigned type
due to the usual arithmetic conversions.
Nikita Popov [Sun, 18 Oct 2020 15:34:22 +0000 (17:34 +0200)]
[BasicAA] Accept AATags by const reference (NFC)
Rather than swapping the value, the sizes, the AA tags and the
underlying objects multiple times, invoke the helper methods
with swapped arguments.
Nico Weber [Sun, 18 Oct 2020 16:17:55 +0000 (12:17 -0400)]
[gn build] (manually) port
d09b08919ca
Dávid Bolvanský [Sun, 18 Oct 2020 15:28:05 +0000 (17:28 +0200)]
[ASAN] Restore and adjust tests
There are optimized out with -fno-builtin
Nathan James [Sun, 18 Oct 2020 15:02:11 +0000 (16:02 +0100)]
[NFC][clang-tidy] Use isInStdNamespace matcher instead of check defined alternatives
Jan Kratochvil [Sun, 18 Oct 2020 14:48:47 +0000 (16:48 +0200)]
[nfc] [lldb] [testsuite] Fix DW_FORM_ref* in DW_AT_declaration-with-children.s .
There were invalid DIE references which nobody used. If LLDB starts to
report invalid DIE references it would lock up (mutex lock).
These invalid DIE references are there since initial check-in by:
https://reviews.llvm.org/D83302
Nathan James [Sun, 18 Oct 2020 14:56:39 +0000 (15:56 +0100)]
[clang tidy] Fix SIMDIntrinsicsCheck not storing options
Florian Hahn [Sun, 18 Oct 2020 14:19:05 +0000 (15:19 +0100)]
[DSE] Do not consider 'noop' intrinsics as read-clobbers.
isNoopIntrinsic returns true for some intrinsics that are modeled in
MemorySSA but do not actually read or write any memory and do not block
DSE. Such intrinsics should not be considered as read-clobbers.
Paul C. Anagnostopoulos [Sat, 17 Oct 2020 14:42:26 +0000 (10:42 -0400)]
[TableGen] Change Programmer's Reference to use "DAG argument" rather than "operand".
Differential Revision: https://reviews.llvm.org/D89624
Nikita Popov [Sun, 18 Oct 2020 14:41:44 +0000 (16:41 +0200)]
[AA] Add helper to update result (NFC)
This pattern was repeated a few times, and for some reason always
using insert or try_emplace, even though we know in advance that
we're looking for an existing entry and not trying to create a
new one.
Florian Hahn [Sun, 18 Oct 2020 12:55:26 +0000 (13:55 +0100)]
[DSE] Add tests for elimination at end of function with lifetime.
David Carlier [Sun, 18 Oct 2020 05:01:29 +0000 (06:01 +0100)]
[Sanitizers] Remove OpenBSD support
- Removing unused and unusable code.
Reviewers: krytarowski
Reviewed By: krytarowski
Differential Revision: https://reviews.llvm.org/D89640
Dávid Bolvanský [Sun, 18 Oct 2020 12:06:34 +0000 (14:06 +0200)]
[LoopIdiom] Regenerate test checks; NFC
Mark de Wever [Sun, 18 Oct 2020 11:47:04 +0000 (13:47 +0200)]
[NFC] Fixes a documentation typo.
Mark de Wever [Sun, 18 Oct 2020 11:34:41 +0000 (13:34 +0200)]
[Sema, CodeGen] Implement [[likely]] and [[unlikely]] in SwitchStmt
This implements the likelihood attribute for the switch statement. Based on the
discussion in D85091 and D86559 it only handles the attribute when placed on
the case labels or the default labels.
It also marks the likelihood attribute as feature complete. There are more QoI
patches in the pipeline.
Differential Revision: https://reviews.llvm.org/D89210
Florian Hahn [Sat, 17 Oct 2020 15:50:31 +0000 (16:50 +0100)]
[DSE] Add tests with noalias store between noop load/store.
This adds 2 new tests from PR47887 and regenerates the check lines for
the file.
Jan Kratochvil [Sun, 18 Oct 2020 08:16:00 +0000 (10:16 +0200)]
[nfc] [lldb] Delete an empty trailing line.
A patchset being prepared shows needless diffs.
Fangrui Song [Sun, 18 Oct 2020 04:34:43 +0000 (21:34 -0700)]
Delete unneeded X86RegisterInfo::hasReservedSpillSlot. NFC
Only PowerPC and RISCV need to override it.
Craig Topper [Sun, 18 Oct 2020 05:20:07 +0000 (22:20 -0700)]
[X86] Remove unnecessary defaulted argument from function in X86FoldTablesEmitter.cpp. NFC
Nothing ever calls it with anything other than the default value.
So just delete it and the code that handled the non-default value.
Fangrui Song [Sun, 18 Oct 2020 03:36:21 +0000 (20:36 -0700)]
[PrologEpilogInserter][test] Improve SpilledToReg test
D39386 made CalleeSavedInfo possible to spill a register to another register
(vector register for POWER9) but did not actually test live-in.
Craig Topper [Sat, 17 Oct 2020 23:35:37 +0000 (16:35 -0700)]
[X86] Mark the Key Locker instructions as NotMemoryFoldable to make the X86FoldTablesEmitter not crash.
loadiwkey and aesenc128kl share the same opcode but one is memory
and one is register. But they're behavior is quite different. We
were crashing because one has an output register and one doesn't
and the backend couldn't account for that. But since they aren't
foldable we can just add NotMemoryFoldable so they won't be looked at.
Dávid Bolvanský [Sun, 18 Oct 2020 00:06:39 +0000 (02:06 +0200)]
[Tests] Added tests for D88328
Dávid Bolvanský [Sat, 17 Oct 2020 20:23:39 +0000 (22:23 +0200)]
[InferAttrs] Add argmemonly attribute to string libcalls
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D89602
Nikita Popov [Sat, 17 Oct 2020 21:18:22 +0000 (23:18 +0200)]
[BasicAA] Avoid alias query if result cannot be used (NFCI)
Rather then querying first and then checking additional conditions,
check the conditions first. They are much cheaper than the alias
query.
Nikita Popov [Sat, 17 Oct 2020 20:09:32 +0000 (22:09 +0200)]
[BasicAA] Fix stale comment (NFC)
DataLayout is always around...
Dávid Bolvanský [Sat, 17 Oct 2020 21:29:02 +0000 (23:29 +0200)]
Revert "[InferAttrs] Add argmemonly attribute to string libcalls"
This reverts commit
b77dd32a6fcc53908aaffc065d4d5b05026ddda7. Sanitizer tests are broken.
Dávid Bolvanský [Sat, 17 Oct 2020 20:23:39 +0000 (22:23 +0200)]
[InferAttrs] Add argmemonly attribute to string libcalls
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D89602
Roman Lebedev [Sat, 17 Oct 2020 18:04:26 +0000 (21:04 +0300)]
[SCEV] Model `ashr exact x, C` as `(abs(x) EXACT/u (1<<C)) * signum(x)`
It's not pretty, but probably better than modelling it
as an opaque SCEVUnknown, i guess.
It is relevant e.g. for the loop that was brought up in
https://bugs.llvm.org/show_bug.cgi?id=46786#c26
as an example of what we'd be able to better analyze
once SCEV handles `ptrtoint` (D89456).
But as it is evident, even if we deal with `ptrtoint` there,
we also fail to model such an `ashr`.
Also, modeling of mul-of-exact-shr/div could use improvement.
As per alive2:
https://alive2.llvm.org/ce/z/tnfZKd
```
define i8 @src(i8 %0) {
%2 = ashr exact i8 %0, 4
ret i8 %2
}
declare i8 @llvm.abs(i8, i1)
declare i8 @llvm.smin(i8, i8)
declare i8 @llvm.smax(i8, i8)
define i8 @tgt(i8 %x) {
%abs_x = call i8 @llvm.abs(i8 %x, i1 false)
%div = udiv exact i8 %abs_x, 16
%t0 = call i8 @llvm.smax(i8 %x, i8 -1)
%t1 = call i8 @llvm.smin(i8 %t0, i8 1)
%r = mul nsw i8 %div, %t1
ret i8 %r
}
```
Transformation seems to be correct!
Roman Lebedev [Sat, 17 Oct 2020 17:53:11 +0000 (20:53 +0300)]
[NFC][SCEV] Refactor getAbsExpr() out of createSCEV()
Roman Lebedev [Sat, 17 Oct 2020 13:24:49 +0000 (16:24 +0300)]
[NFC][SCEV] Add 'getMinusOne()' method
Roman Lebedev [Sat, 17 Oct 2020 12:43:32 +0000 (15:43 +0300)]
[NFC][SCEV] Add some more ptrtoint/PR46786 -related tests
Sylvestre Ledru [Sat, 17 Oct 2020 17:51:05 +0000 (19:51 +0200)]
Revert "[clang-format] Fix AlignConsecutive on PP blocks"
This reverts commit
b2eb439317576ce718193763c12bff9fccdfc166.
Caused the regression:
https://bugs.llvm.org/show_bug.cgi?id=47589
Reviewed By: MyDeveloperDay
Differential Revision: https://reviews.llvm.org/D89464
Sanjay Patel [Sat, 17 Oct 2020 16:19:04 +0000 (12:19 -0400)]
[InstCombine] (~A & B) ^ A -> A | B
Differential Revision: https://reviews.llvm.org/D86395
Mircea Trofin [Sat, 17 Oct 2020 16:06:24 +0000 (09:06 -0700)]
[NFC][ML] Avoid source of some signed/unsigned warnings in TFUtilsTest
Luís Marques [Sat, 17 Oct 2020 15:52:29 +0000 (16:52 +0100)]
[RISCV][ASAN] Fix passing XFAIL tests
These tests pass for RV64 Linux, but they are marked as XFAIL. This patch
fixes that.
Differential Revision: https://reviews.llvm.org/D89299
Nikita Popov [Sat, 17 Oct 2020 13:29:19 +0000 (15:29 +0200)]
[MemCpyOpt] Extract common function for unwinding check
These two cases should be using the same logic. Not NFC, as this
resolves the TODO regarding use of the underlying object.
Pedro Tammela [Sat, 17 Oct 2020 13:20:55 +0000 (14:20 +0100)]
[NFC] fix some typos in LoopUnrollPass
This patch fixes a couple of typos in the LoopUnrollPass.cpp comments
Differential Revision: https://reviews.llvm.org/D89603
David Green [Sat, 17 Oct 2020 09:29:00 +0000 (10:29 +0100)]
[ARM] Basic getArithmeticReductionCost reduction costs
This adds some basic costs for MVE reductions - currently just costing
the simple legal add vectors as a single MVE instruction. More complex
costing can be added in the future when the framework more readily
allows it.
Differential Revision: https://reviews.llvm.org/D88980
David Green [Sat, 17 Oct 2020 09:09:42 +0000 (10:09 +0100)]
[ARM] Add a very basic active_lane_mask cost
This adds a very basic cost for active_lane_mask under MVE - making the
assumption that they will be free and then apologizing for that in a
comment.
In reality they may either be free (by being nicely folded into a tail
predicated loop), cost the same as a VCTP or be expanded into vdup's,
adds and cmp's. It is difficult to detect the difference from a single
getIntrinsicInstrCost call, so makes the assumption that the vectorizer
is adding them, and only added them where it makes sense.
We may need to change this in the future to better model predicate costs
in the vectorizer, especially at -Os or non-tail predicated loops. The
vectorizer currently does not query the cost of these instructions but
that will change in the future and a zero cost there probably makes the
most sense at the moment.
Differential Revision: https://reviews.llvm.org/D88989
Dave Lee [Sat, 17 Oct 2020 06:02:50 +0000 (23:02 -0700)]
[lldb] Enable cmake policy CMP0077 for option()
In lldb, explicitly set the "option() honors normal variables" CMake policy. This applies for
standalone lldb builds and matches what llvm, clang, etc do. This prevents potentially unwanted
clearing of variables like `LLVM_ENABLE_WARNINGS`, and also prevents unnecessary build warnings.
See: https://cmake.org/cmake/help/latest/policy/CMP0077.html
Differential Revision: https://reviews.llvm.org/D89614
Dave Lee [Sat, 17 Oct 2020 07:13:48 +0000 (00:13 -0700)]
Revert "Register TargetCXXABI.def as a textual header"
Unbreak module builds.
TargetCXXABI.def has been removed in a revert:
79829a47040512fe54001db839ac59146ca55aec.
This reverts commit
0ff9116b36781d6fa61c25841edd53dc8f366bec.
Jason Molenda [Sat, 17 Oct 2020 06:23:25 +0000 (23:23 -0700)]
Un-indent the section on 'Writing Target Stop-Hooks in Python'
it was ending up in a code block from the previous section,
instead of being its own section.
Juneyoung Lee [Sat, 17 Oct 2020 04:44:00 +0000 (13:44 +0900)]
Add support for !noundef metatdata on loads
This patch adds metadata !noundef and makes load instructions can optionally have it.
A load with !noundef always return a well-defined value (has no undef bit or isn't poison).
If the loaded value isn't well defined, the behavior is undefined.
This metadata can be used to encode the assumption from C/C++ that certain reads of variables should have well-defined values.
It is helpful for optimizing freeze instructions away, because freeze can be removed when its operand has well-defined value, and showing that a load from arbitrary location is well-defined is usually hard otherwise.
The same information can be encoded with llvm.assume with operand bundle; using metadata is chosen because I wasn't sure whether code motion can be freely done when llvm.assume is inserted from clang instead.
The existing codebase already is stripping unknown metadata when doing code motion, so using metadata is UB-safe as well.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D89050
Sriraman Tallam [Sat, 17 Oct 2020 04:28:18 +0000 (21:28 -0700)]
This test includes a source that will produce basic blocks and hence sections with -fbasic-block-sections=all.
The test reorders the basic blocks to be dis-contiguous in the address space and checks if the back trace contains the right symbol.
Reviewed By: labath
Differential Revision: https://reviews.llvm.org/D89179
Juneyoung Lee [Sat, 17 Oct 2020 04:29:58 +0000 (13:29 +0900)]
[LangRef] Rename the names of metadata in load/store's syntax (NFC)
Discussed in D89050
Alok Kumar Sharma [Sat, 17 Oct 2020 02:54:08 +0000 (08:24 +0530)]
[DebugInfo] Support for DWARF operator DW_OP_over
LLVM rejects DWARF operator DW_OP_over. This DWARF operator is needed
for Flang to support assumed rank array.
Summary:
Currently LLVM rejects DWARF operator DW_OP_over. Below error is
produced when llvm finds this operator.
[..]
invalid expression
!DIExpression(151, 20, 16, 48, 30, 35, 80, 34, 6)
warning: ignoring invalid debug info in over.ll
[..]
There were some parts missing in support of this operator, which are
now completed.
Testing
-added a unit testcase
-check-debuginfo
-check-llvm
Reviewed By: aprantl
Differential Revision: https://reviews.llvm.org/D89208
Craig Topper [Fri, 16 Oct 2020 22:42:49 +0000 (15:42 -0700)]
[TargetLowering] Extract simplifySetCCs ctpop into a separate function. NFCI
As requested in D89346. This allows us to add some early outs.
I reordered some checks a little bit to make the more common bail outs happen earlier. Like checking opcode before checking hasOneUse. And I moved the bit width check to make sure it was safe to look through a truncate to the spot where we look through truncates instead of after.
Reviewed By: spatel
Differential Revision: https://reviews.llvm.org/D89494
Alina Sbirlea [Sat, 17 Oct 2020 00:25:30 +0000 (17:25 -0700)]
[MemorySSA] Verify clobbering within reachable blocks.
Resolves PR45976.
Arthur Eubanks [Fri, 16 Oct 2020 21:58:32 +0000 (14:58 -0700)]
[NPM] Fix some PhaseOrdering tests under NPM
These either already have corresponding NPM RUN lines,
or need to be fixed to not use -analyze.
Amara Emerson [Thu, 15 Oct 2020 22:44:27 +0000 (15:44 -0700)]
[AArch64][GlobalISel] Select csinc if a select has a 1 on RHS.
Differential Revision: https://reviews.llvm.org/D89513
Richard Smith [Fri, 16 Oct 2020 23:46:49 +0000 (16:46 -0700)]
Make the check for whether we should memset(0) an aggregate
initialization a little smarter.
Look through casts that preserve zero-ness when determining if an
initializer is zero, so that we can handle cases like an {0} initializer
whose corresponding field is a type other than 'int'.
George Rokos [Fri, 16 Oct 2020 21:39:30 +0000 (14:39 -0700)]
[libomptarget] Fix copy-to motion for PTR_AND_OBJ entries where PTR is a struct member.
This patch fixes a problem whereby the pointee object of a PTR_AND_OBJ entry with a `map(to)` motion clause can be overwritten on the device even if its reference counter is >=1.
Currently, we check the reference counter of the parent struct in order to determine whether the motion clause should be respected, but since the pointee object is not part of the struct, it's got its own reference counter which should be used to enqueue the copy or discard it.
The same behavior has already been implemented in targetDataEnd (omptarget.cpp:539-540), but we somehow missed doing the same in targetDataBegin.
Differential Revision: https://reviews.llvm.org/D89597
Richard Smith [Fri, 16 Oct 2020 22:13:23 +0000 (15:13 -0700)]
Fix accidental use of VLAs that causes these tests to fail after Clang
commit
552c6c2328723a248c2b4d2765f75d49129dff20.
Albion Fung [Fri, 16 Oct 2020 22:03:09 +0000 (18:03 -0400)]
[PowerPC] Implementation of 128-bit Binary Vector Rotate builtins
This patch implements 128-bit Binary Vector Rotate builtins for PowerPC10.
Differential Revision: https://reviews.llvm.org/D86819
Rob Suderman [Fri, 16 Oct 2020 20:51:07 +0000 (13:51 -0700)]
Fixed a failure when const matcher fails, added a test to catch
Differential Revision: https://reviews.llvm.org/D89593
Richard Smith [Fri, 16 Oct 2020 02:32:15 +0000 (19:32 -0700)]
PR44406: Follow behavior of array bound constant folding in more recent versions of GCC.
Old GCC used to aggressively fold VLAs to constant-bound arrays at block
scope in GNU mode. That's non-conforming, and more modern versions of
GCC only do this at file scope. Update Clang to do the same.
Also promote the warning for this from off-by-default to on-by-default
in all cases; more recent versions of GCC likewise warn on this by
default.
This is still slightly more permissive than GCC, as pointed out in
PR44406, as we still fold VLAs to constant arrays in structs, but that
seems justifiable given that we don't support VLA-in-struct (and don't
intend to ever support it), but GCC does.
Differential Revision: https://reviews.llvm.org/D89523
Peng Guo [Fri, 16 Oct 2020 20:47:46 +0000 (13:47 -0700)]
[objdump][macho] Check arch before formating reloc name as arm64 addend
Before formating ARM64_RELOC_ADDEND relocation target name as a hex
number, the architecture need to be checked since other architectures
can define a different relocation type with the same integer as
ARM64_RELOC_ADDEND.
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D89094
Dave Lee [Fri, 16 Oct 2020 18:46:35 +0000 (11:46 -0700)]
[lldb] Implement ObjCExceptionThrowFrameRecognizer::GetName()
Implement `GetName` for `ObjCExceptionThrowFrameRecognizer`. Otherwise, `frame recognizer list` shows "(internal)" for the name.
Differential Revision: https://reviews.llvm.org/D89589
Jameson Nash [Fri, 16 Oct 2020 21:22:07 +0000 (17:22 -0400)]
Revert "make the AsmPrinterHandler array public"
I messed up one of the tests.
Ben Hamilton [Thu, 15 Oct 2020 21:00:00 +0000 (15:00 -0600)]
[Format/ObjC] Correctly handle base class with lightweight generics and protocol
ClangFormat does not correctly handle an Objective-C interface declaration
with both lightweight generics and a protocol conformance.
This simple example:
```
@interface Foo : Bar <Baz> <Blech>
@end
```
means `Foo` extends `Bar` (a lightweight generic class whose type
parameter is `Baz`) and also conforms to the protocol `Blech`.
ClangFormat should not apply any changes to the above example, but
instead it currently formats it quite poorly:
```
@interface Foo : Bar <Baz>
<Blech>
@end
```
The bug is that `UnwrappedLineParser` assumes an open-angle bracket
after a base class name is a protocol list, but it can also be a
lightweight generic specification.
This diff fixes the bug by factoring out the logic to parse
lightweight generics so it can apply both to the declared class
as well as the base class.
Test Plan: New tests added. Ran tests with:
% ninja FormatTests && ./tools/clang/unittests/Format/FormatTests
Confirmed tests failed before diff and passed after diff.
Reviewed By: sammccall, MyDeveloperDay
Differential Revision: https://reviews.llvm.org/D89496
Renato Golin [Fri, 16 Oct 2020 20:49:47 +0000 (21:49 +0100)]
Revert "Fix Windows llvm-objdump tests"
It had already been fixed in a different way.
Stanislav Mekhanoshin [Fri, 16 Oct 2020 17:18:47 +0000 (10:18 -0700)]
[AMDGPU] Fix gfx1032 description in AMDGPUUsage.rst. NFC.
Differential Revision: https://reviews.llvm.org/D89565
Renato Golin [Fri, 16 Oct 2020 20:26:46 +0000 (21:26 +0100)]
Fix Windows llvm-objdump tests
Broken in
e95f9a23fad with path issues.
Jameson Nash [Thu, 15 Oct 2020 21:19:10 +0000 (17:19 -0400)]
make the AsmPrinterHandler array public
This lets external consumers customize the output, similar to how
AssemblyAnnotationWriter lets the caller define callbacks when printing
IR. The array of handlers already existed, this just cleans up the code
so that it can be exposed publically.
Differential Revision: https://reviews.llvm.org/D74158
Richard Smith [Wed, 14 Oct 2020 02:14:13 +0000 (19:14 -0700)]
Treat constant contexts as being in the default rounding mode.
This addresses a regression where pretty much all C++ compilations using
-frounding-math now fail, due to rounding being performed in constexpr
function definitions in the standard library.
This follows the "manifestly constant evaluated" approach described in
https://reviews.llvm.org/D87528#2270676 -- evaluations that are required
to succeed at compile time are permitted even in regions with dynamic
rounding modes, as are (unfortunately) the evaluation of the
initializers of local variables of const integral types.
Differential Revision: https://reviews.llvm.org/D89360
Richard Smith [Fri, 16 Oct 2020 20:18:52 +0000 (13:18 -0700)]
Extend memset-to-zero optimization to C++11 aggregate functional casts
Aggr{...}.
We previously missed these cases due to not stepping over the additional
AST nodes representing their syntactic form.
Artem Belevich [Fri, 16 Oct 2020 18:19:27 +0000 (11:19 -0700)]
[VectorCombine] Avoid crossing address space boundaries.
We can not bitcast pointers across different address spaces, and VectorCombine
should be careful when it attempts to find the original source of the loaded
data.
Differential Revision: https://reviews.llvm.org/D89577