Simon Pilgrim [Sun, 19 Sep 2021 17:41:55 +0000 (18:41 +0100)]
[X86] Fold SHUFPS(shuffle(x),shuffle(y),mask) -> SHUFPS(x,y,mask')
We can combine unary shuffles into either of SHUFPS's inputs and adjust the shuffle mask accordingly.
Unlike general shuffle combining, we can be more aggressive and handle multiuse cases as we're not going to accidentally create additional shuffles.
David Blaikie [Sun, 19 Sep 2021 16:24:25 +0000 (09:24 -0700)]
llvm-dwarfdump: Refactor type pretty printing tests
Move most type tests to a pre-generated assembly file to make it easier
to add more weird cases without having to hand craft more DWARF.
Move the novel array types that aren't reachable via clang-generated
DWARF to a separate file for easy maintenance.
Simon Pilgrim [Sun, 19 Sep 2021 15:58:40 +0000 (16:58 +0100)]
[X86][Atom] Specific uops for all IMUL/IDIV instructions
Based off a mixture of llvm-exegesis captures (PR36895) and Intel AoM / Agner / InstLatX64 reports.
Roman Lebedev [Sun, 19 Sep 2021 14:38:32 +0000 (17:38 +0300)]
[X86][TLI] SimplifyDemandedVectorEltsForTargetNode(): don't break apart broadcasts from which not just the 0'th elt is demanded
Apparently this has no test coverage before D108382,
but D108382 itself shows a few regressions that this fixes.
It doesn't seem worthwhile breaking apart broadcasts,
assuming we want the broadcasted value to be preset in several elements,
not just the 0'th one.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D108411
Roman Lebedev [Sun, 19 Sep 2021 14:32:43 +0000 (17:32 +0300)]
[X86] lowerShuffleAsDecomposedShuffleMerge(): if both inputs are broadcastable/identities, canonicalize broadcasts as such
Split off from D108253.
Broadcast is simpler than any other shuffle we might produce
to do what we want to do here, so prefer it.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D108382
Roman Lebedev [Sun, 19 Sep 2021 14:27:35 +0000 (17:27 +0300)]
[NFC] combineX86ShufflesRecursively(): actually address nits for previous patch
Roman Lebedev [Sun, 19 Sep 2021 14:21:17 +0000 (17:21 +0300)]
[X86] combineX86ShufflesRecursively(): call SimplifyMultipleUseDemandedVectorElts() on after finishing recursing
This was suggested in https://reviews.llvm.org/D108382#inline-1039018,
and it avoids regressions in that patch.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D109065
Sanjay Patel [Sun, 19 Sep 2021 13:18:07 +0000 (09:18 -0400)]
[InstCombine] add/adjust tests for min/max intrinsics; NFC
If we transform these, we have to propagate no-wrap/undef carefully.
David Green [Sun, 19 Sep 2021 13:25:21 +0000 (14:25 +0100)]
[ARM] Add VGETLANEu patterns for v4f16 and v8f16
These were apparently missing, having no pattern that could convert a
VGETLANEu of a v4f16 to an i32. Added bf16 whilst here, following the
same code.
xndcn [Sun, 19 Sep 2021 12:01:12 +0000 (20:01 +0800)]
[mlir][emitc] Fix indent in CondBranchOp and block label
1. Add missing indent in CondBranchOp
2. Remove indent in block label
Differential Revision: https://reviews.llvm.org/D109805
Simon Pilgrim [Sat, 18 Sep 2021 20:29:33 +0000 (21:29 +0100)]
[X86][Atom] Fix (U)COMISS/SD uops, latency and throughput
Both ports are required, for reg and mem variants - we can also use the WriteFComX class directly and remove the unnecessary InstRW overrides. Matches what Intel AoM / Agner / InstLatX64 report as well.
Sylvestre Ledru [Sun, 19 Sep 2021 10:17:49 +0000 (12:17 +0200)]
llvm/cmake: fix a typo
Ben Shi [Mon, 30 Aug 2021 13:14:12 +0000 (21:14 +0800)]
[RISCV] Optimize (add (shl x, c0), (shl y, c1)) with SH*ADD
Optimize (add (shl x, c0), (shl y, c1)) ->
(SLLI (SH*ADD x, y), c1), if c0-c1 == 1/2/3.
Reviewed By: craig.topper, luismarques
Differential Revision: https://reviews.llvm.org/D108916
David Blaikie [Sun, 19 Sep 2021 05:54:57 +0000 (22:54 -0700)]
DWARFDie:DWARFTypePrinter: Add common utility function for checking where parentheses are required
David Blaikie [Sun, 19 Sep 2021 05:22:25 +0000 (22:22 -0700)]
DWARFDie.cpp: Reduce indentation with early continue
Roman Lebedev [Sat, 18 Sep 2021 19:13:25 +0000 (22:13 +0300)]
[X86] Improve i8 all-ones element insertion in pre-SSE4.1
Should avoid some regressions in D109065
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D109989
Nikita Popov [Sat, 18 Sep 2021 19:14:35 +0000 (21:14 +0200)]
[Tests] Fix noalias metadata in one more test
Missed this one in
80110aafa09aff9869cdb50ed718a169888da578. This
is another test mixing up alias scopes and alias scope lists.
Nikita Popov [Thu, 16 Sep 2021 20:17:36 +0000 (22:17 +0200)]
[Tests] Fix incorrect noalias metadata
Mostly this fixes cases where !noalias or !alias.scope were passed
a scope rather than a scope list. In some cases I opted to drop
the metadata entirely instead, because it is not really relevant
to the test.
alokmishra.besu [Fri, 17 Sep 2021 21:03:01 +0000 (16:03 -0500)]
OpenMP 5.0 metadirective
This patch supports OpenMP 5.0 metadirective features.
It is implemented keeping the OpenMP 5.1 features like dynamic user condition in mind.
A new function, getBestWhenMatchForContext, is defined in llvm/Frontend/OpenMP/OMPContext.h
Currently this function return the index of the when clause with the highest score from the ones applicable in the Context.
But this function is declared with an array which can be used in OpenMP 5.1 implementation to select all the valid when clauses which can be resolved in runtime. Currently this array is set to null by default and its implementation is left for future.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D91944
Usman Nadeem [Sat, 18 Sep 2021 18:28:49 +0000 (11:28 -0700)]
Precommit tests for D109807 "[InstCombine] Narrow type of logical operation chains when possible"
Change-Id: Iae9bf18619e4926301a866c7e2bd38ced524804e
Mehdi Amini [Sat, 18 Sep 2021 02:38:17 +0000 (02:38 +0000)]
Check for invalid projects passed in LLVM_ENABLE_PROJECTS
This is catching misconfiguration. For example one of my automation
had a typo running `-DLLVM_ENABLE_PROJECTS=nlir` and it was just
silently ignored. Instead, an error will now be displayed.
Recommit
4121ac1e after fixing it to not fail on "all"
Differential Revision: https://reviews.llvm.org/D110016
Alexandre Ganea [Sat, 18 Sep 2021 16:00:17 +0000 (12:00 -0400)]
[Support] Attempt to fix deadlock in ThreadGroup
This is an attempt to fix the situation described by https://reviews.llvm.org/D104207#2826290 and PR41508.
See sequence of operations leading to the bug in https://reviews.llvm.org/D104207#3004689
We ensure that the Latch is completely "free" before decrementing the number of TaskGroupInstances.
Differential revision: https://reviews.llvm.org/D109914
David Green [Sat, 18 Sep 2021 17:47:02 +0000 (18:47 +0100)]
[ARM] Prevent large integer VQDMULH pattern crashes
Put a limit on the size of constant integers we test when looking for
VQDMULH, to prevent it from crashing from values more than 64bits.
Arjun P [Sat, 18 Sep 2021 16:10:59 +0000 (21:40 +0530)]
[MLIR] Simplex: rename num{Variables,Constraints} to getNum{Variables,Constraints}
As per the LLVM Coding Standards, function names should be verb phrases.
Kazu Hirata [Sat, 18 Sep 2021 16:29:24 +0000 (09:29 -0700)]
[CodeGen] Use make_early_inc_range (NFC)
Arjun P [Sat, 18 Sep 2021 10:34:46 +0000 (16:04 +0530)]
[MLIR] Simplex: support adding new variables dynamically
Reviewed By: Groverkss
Differential Revision: https://reviews.llvm.org/D109962
Nico Weber [Sat, 18 Sep 2021 15:15:21 +0000 (11:15 -0400)]
[lld/mac] Fix comment typo to cycle bots
Jacques Pienaar [Sat, 18 Sep 2021 13:57:51 +0000 (06:57 -0700)]
[mlir-c] Add getting fused loc
For creating a fused loc using array of locations and metadata.
Differential Revision: https://reviews.llvm.org/D110022
Nico Weber [Sat, 18 Sep 2021 13:10:37 +0000 (09:10 -0400)]
Revert "OpenMP 5.0 metadirective"
This reverts commit
c7d7b98e5263472f05b2f3cb767b5d16e1349e9a.
Breaks tests on macOS, see comment on https://reviews.llvm.org/D91944
Aaron Puchert [Sat, 18 Sep 2021 11:46:50 +0000 (13:46 +0200)]
Thread safety analysis: Warn when demoting locks on back edges
Previously in D104261 we warned about dropping locks from back edges,
this is the corresponding change for exclusive/shared joins. If we're
entering the loop with an exclusive change, which is then relaxed to a
shared lock before we loop back, we have already analyzed the loop body
with the stronger exclusive lock and thus might have false positives.
There is a minor non-observable change: we modify the exit lock set of a
function, but since that isn't used further it doesn't change anything.
Reviewed By: aaron.ballman
Differential Revision: https://reviews.llvm.org/D106713
Roman Lebedev [Sat, 18 Sep 2021 10:56:54 +0000 (13:56 +0300)]
Revert "Check for invalid projects passed in LLVM_ENABLE_PROJECTS"
```
CMake Error at CMakeLists.txt:76 (MESSAGE):
all isn't a know project:
clang;clang-tools-extra;compiler-rt;cross-project-tests;libc;libclc;libcxx;libcxxabi;libunwind;lld;lldb;mlir;openmp;parallel-libs;polly;pstl;flang;llvm
```
This reverts commit
4121ac1e7f96d826b793bd7e29c181f1eb8168bf.
Sylvestre Ledru [Sat, 18 Sep 2021 10:42:50 +0000 (12:42 +0200)]
llvm doc: remove the old reference to subversion and add a link to the GH repo instead
Sylvestre Ledru [Sat, 18 Sep 2021 10:42:09 +0000 (12:42 +0200)]
llvm doc: subversion isn't necessary in this example
Sylvestre Ledru [Sat, 18 Sep 2021 07:55:10 +0000 (09:55 +0200)]
Adjust the doc after the renaming of debuginfo-tests => cross-project-tests
Uday Bondhugula [Sun, 25 Jul 2021 13:32:38 +0000 (19:02 +0530)]
[MLIR][GPU] Add constant propagator for gpu.launch op
Add a constant propagator for gpu.launch op in cases where the
grid/thread IDs can be trivially determined to take a single constant
value of zero.
Differential Revision: https://reviews.llvm.org/D109994
Mehdi Amini [Sat, 18 Sep 2021 02:38:17 +0000 (02:38 +0000)]
Check for invalid projects passed in LLVM_ENABLE_PROJECTS
This is catching misconfiguration. For example one of my automation
had a typo running `-DLLVM_ENABLE_PROJECTS=nlir` and it was just
silently ignored. Instead, an error will now be displayed.
Differential Revision: https://reviews.llvm.org/D110016
Joseph Huber [Fri, 17 Sep 2021 18:53:58 +0000 (14:53 -0400)]
[OpenMP] Add support for dynamic shared memory in new RTL
This patch adds support for using dynamic shared memory in the new
device runtime. The new function `__kmpc_get_dynamic_shared` will return a
pointer to the buffer of dynamic shared memory. Currently the amount of memory
allocated is set by an environment variable.
In the future this amount will be added to the amount used for the smart stack
which will be configured in a similar way.
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D110006
Joseph Huber [Thu, 16 Sep 2021 18:42:14 +0000 (14:42 -0400)]
[OpenMP] Add additional fields to device environment
This patch adds fields for the device number and number of devices into
the device environment struct and debugging values.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D110004
Joseph Huber [Thu, 16 Sep 2021 12:51:12 +0000 (08:51 -0400)]
[OpenMP] Implement __assert_fail in the new device runtime
This patch implements the `__assert_fail` function in the new device
runtime. This allows users and developers to use the standars assert
function inside of the device.
Reviewed By: tianshilei1992
Differential Revision: https://reviews.llvm.org/D109886
Joseph Huber [Fri, 17 Sep 2021 19:48:54 +0000 (15:48 -0400)]
[OpenMP] Change debugging symbol to weak_odr linkage
The new device runtime uses an internal variable to set debugging. This
variable was originally privately linked because every module will have
a copy of it. This caused problems with merging the device bitcode
library because it would get renamed and there was not a way to refer to
an external, private symbol. This changes the symbol to weak_odr so it
can be defined multiply, but will not be renamed.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D109997
Amy Huang [Sat, 18 Sep 2021 00:43:28 +0000 (17:43 -0700)]
[lld] Fix small error in previous commit
6f7483b1ece4747f2aafe4baa05fc07e7dc9ed9d.
Geoffrey Martin-Noble [Sat, 18 Sep 2021 00:23:11 +0000 (17:23 -0700)]
[mlir] Fix syntax example for tensor.from_elements
Parens are not used here
Amy Huang [Thu, 16 Sep 2021 23:48:26 +0000 (16:48 -0700)]
Reland "[LLD] Remove global state in lld/COFF" after fixing asan and msan test failures
Original commit description:
[LLD] Remove global state in lld/COFF
This patch removes globals from the lldCOFF library, by moving globals
into a context class (COFFLinkingContext) and passing it around wherever
it's needed.
See https://lists.llvm.org/pipermail/llvm-dev/2021-June/151184.html for
context about removing globals from LLD.
I also haven't moved the `driver` or `config` variables yet.
Differential Revision: https://reviews.llvm.org/D109634
This reverts commit
a2fd05ada9030eab2258fff25e77a05adccae128.
Original commits were
b4fa71eed34d967195514fe9b0a5211fca2bc5bc
and
e03c7e367adb8f228332e3c2ef8f45484597b719.
Joseph Huber [Wed, 15 Sep 2021 20:16:00 +0000 (16:16 -0400)]
[Attributor] Change AAExecutionDomain to check intrinsic edges
The AAExecutionDomain instance checks if a BB is executed by the main
thread only. Currently, this only checks the `__kmpc_kernel_init` call
for generic regions to indicate the path taken by the main thread. In
the new runtime, we want to be able to detect basic blocks even in SPMD
mode. For this we enable it to check thread-ID intrinsics being compared
to zero as well.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D109849
Joseph Huber [Fri, 17 Sep 2021 19:27:33 +0000 (15:27 -0400)]
[OpenMP] Add NoSync attributes to alloc / free shared RTL calls
This patch adds the `nosync` attribute to the `__kmpc_alloc_shared` and
`__kmpc_free_shared` runtime library calls. This allows code analysis to
know that these functins dont contain any barriers. This will help
optimizations reason about the CFG of blocks containing these calls.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D109995
Jez Ng [Wed, 7 Jul 2021 17:47:26 +0000 (13:47 -0400)]
[lld-macho] Construct CFString literals by copying the ConcatInputSection
... instead of constructing a new one each time. This allows us
to take advantage of {D105305}.
I didn't see a substantial difference when linking chromium_framework,
but this paves the way for reusing similar logic for splitting compact
unwind entries into sections. There are a lot more of those, so the
performance impact is significant.
Differential Revision: https://reviews.llvm.org/D109895
Vedant Kumar [Tue, 14 Sep 2021 23:58:40 +0000 (16:58 -0700)]
[MachCore] Report arm64 thread exception state
A MachO userspace corefile may contain LC_THREAD commands which specify
thread exception state.
For arm64* only (for now), report a human-readable version of this state
as the thread stop reason, instead of 'SIGSTOP'.
As a follow-up, similar functionality can be implemented for x86 cores
by translating the trapno/err exception registers.
rdar://
82898146
Differential Revision: https://reviews.llvm.org/D109795
peter klausler [Fri, 17 Sep 2021 15:19:10 +0000 (08:19 -0700)]
[flang] More checking of NULL pointer actual arguments
Catch additional missing error cases for typed and untyped
NULL actual arguments to non-intrinsic procedures in cases
of explicit and implicit interfaces.
Differential Revision: https://reviews.llvm.org/D110003
Usman Nadeem [Sun, 12 Sep 2021 22:53:26 +0000 (15:53 -0700)]
[AArch64][SVE][InstCombine] Fold redundant zip1/2(uzp1/2) operations
zip1(uzp1(A, B), uzp2(A, B)) --> A
zip2(uzp1(A, B), uzp2(A, B)) --> B
Differential Revision: https://reviews.llvm.org/D109666
Change-Id: I4a6578db2fcef9ff71ad0e77b9fe08354e6dbfcd
Arthur Eubanks [Fri, 17 Sep 2021 21:47:28 +0000 (14:47 -0700)]
[NFC] Remove FIXMEs about calling LLVMContext::yield()
Nobody has complained about this, and the documentation for
LLVMContext::yield() states that LLVM is allowed to never call it.
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D110008
Adrian Prantl [Fri, 17 Sep 2021 17:58:46 +0000 (10:58 -0700)]
Apply proper source location to fallthrough switch cases.
This fixes a bug in clang where, when clang sees a switch with a
fallthrough to a default like this:
static void funcA(void) {}
static void funcB(void) {}
int main(int argc, char **argv) {
switch (argc) {
case 0:
funcA();
break;
case 10:
default:
funcB();
break;
}
}
It does not add a proper debug location for that switch case, such as
case 10: above.
Patch by Shubham Rastogi!
Differential Revision: https://reviews.llvm.org/D109940
cchen [Fri, 17 Sep 2021 21:44:09 +0000 (16:44 -0500)]
Revert "[OpenMP] Use irbuilder as default for masked and master construct"
This reverts commit
2908fc0d3f16f873b5019f1c62a24482c2b75e36.
alokmishra.besu [Fri, 17 Sep 2021 21:03:01 +0000 (16:03 -0500)]
OpenMP 5.0 metadirective
This patch supports OpenMP 5.0 metadirective features.
It is implemented keeping the OpenMP 5.1 features like dynamic user condition in mind.
A new function, getBestWhenMatchForContext, is defined in llvm/Frontend/OpenMP/OMPContext.h
Currently this function return the index of the when clause with the highest score from the ones applicable in the Context.
But this function is declared with an array which can be used in OpenMP 5.1 implementation to select all the valid when clauses which can be resolved in runtime. Currently this array is set to null by default and its implementation is left for future.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D91944
cchen [Fri, 17 Sep 2021 21:14:16 +0000 (16:14 -0500)]
Revert "OpenMP 5.0 metadirective"
This reverts commit
c7d7b98e5263472f05b2f3cb767b5d16e1349e9a.
cchen [Fri, 17 Sep 2021 21:03:01 +0000 (16:03 -0500)]
OpenMP 5.0 metadirective
This patch supports OpenMP 5.0 metadirective features.
It is implemented keeping the OpenMP 5.1 features like dynamic user condition in mind.
A new function, getBestWhenMatchForContext, is defined in llvm/Frontend/OpenMP/OMPContext.h
Currently this function return the index of the when clause with the highest score from the ones applicable in the Context.
But this function is declared with an array which can be used in OpenMP 5.1 implementation to select all the valid when clauses which can be resolved in runtime. Currently this array is set to null by default and its implementation is left for future.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D91944
cchen [Fri, 17 Sep 2021 20:53:49 +0000 (15:53 -0500)]
[OpenMP] Use irbuilder as default for masked and master construct
Use irbuilder as default and remove redundant Clang codegen for masked construct and master construct.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D100874
Sanjay Patel [Fri, 17 Sep 2021 18:47:49 +0000 (14:47 -0400)]
[InstCombine] add tests for min/max intrinsics with offset operand; NFC
Aart Bik [Thu, 16 Sep 2021 23:43:03 +0000 (16:43 -0700)]
[mlir][sparse] add a sparse quantized_matmul example to integration test
Note that this revision adds a very tiny bit of constant folding in the
sparse compiler lattice construction. Although I am generally trying to
avoid such canonicalizations (and rely on other passes to fix this instead),
the benefits of avoiding a very expensive disjunction lattice construction
justify having this special code (at least for now).
Reviewed By: bixia
Differential Revision: https://reviews.llvm.org/D109939
Aart Bik [Fri, 17 Sep 2021 18:55:45 +0000 (11:55 -0700)]
[mlir][sparse] add dce test for all sparse tensor ops
Reviewed By: ThomasRaoux
Differential Revision: https://reviews.llvm.org/D109992
Dávid Bolvanský [Fri, 17 Sep 2021 19:58:57 +0000 (21:58 +0200)]
[NFC] Precommit tests for D109954
Andrew Browne [Wed, 15 Sep 2021 18:50:07 +0000 (11:50 -0700)]
[DFSan] Add force_zero_label abilist option to DFSan. This can be used as a work-around for overtainting.
Reviewed By: morehouse
Differential Revision: https://reviews.llvm.org/D109847
Jan Kratochvil [Fri, 17 Sep 2021 19:51:36 +0000 (21:51 +0200)]
[nfc] [lldb] Remove unused DIEPointerSet, DeclToDIEMap and m_decl_to_die
In whole GIT history this map has been always only written but never read.
Jeremie Courreges-Anglas [Fri, 17 Sep 2021 19:15:46 +0000 (15:15 -0400)]
[compiler-rt] Implement __clear_cache() on OpenBSD/riscv64
peter klausler [Fri, 17 Sep 2021 19:39:38 +0000 (12:39 -0700)]
[flang] Fix build
An unused variable declaration from a recent change is eliciting
a fatal warning; remove it.
Samuel [Fri, 17 Sep 2021 19:18:03 +0000 (12:18 -0700)]
[llvm-reduce] Add reduce operands pass
Add reduction to set operands to default values
Reviewed By: aeubanks
Differential Revision: https://reviews.llvm.org/D108903
Hongtao Yu [Fri, 17 Sep 2021 16:21:39 +0000 (09:21 -0700)]
[CSSPGO] Tweakes to lower pseudo probe runtime overhead
A couple tweaks to
1. allow more thinlto importing by excluding probe intrinsics from IR size in module summary
2. Allow general default attributes (nofree nosync nounwind) for pseudo probe intrinsic. Without those attributes, pseudo probes will be basically treated as unknown calls which will in turn block their containing functions from annotated with those attributes.
Reviewed By: wenlei
Differential Revision: https://reviews.llvm.org/D109976
Keith Smiley [Wed, 8 Sep 2021 20:49:16 +0000 (13:49 -0700)]
[clang][darwin] Add support for --emit-static-lib
This uses darwin's default libtool since llvm-ar isn't normally
available.
Differential Revision: https://reviews.llvm.org/D109461
Krzysztof Drewniak [Thu, 16 Sep 2021 21:25:20 +0000 (21:25 +0000)]
[MLIR][Affine] Simplify nested modulo operations when able
It is the case that, for all positive a and b such that b divides a
(e mod (a * b)) mod b = e mod b. For example, ((d0 mod 35) mod 5) can
be simplified to (d0 mod 5), but ((d0 mod 35) mod 4) cannot be simplified
further (x = 36 is a counterexample).
This change enables more complex simplifications. For example,
((d0 * 72 + d1) mod 144) mod 9 can now simplify to (d0 * 72 + d1) mod 9
and thus to d1 mod 9. Expressions with chained modulus operators are
reasonably common in tensor applications, and this change _should_
improve code generation for such expressions.
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D109930
thomasraoux [Fri, 17 Sep 2021 17:56:21 +0000 (10:56 -0700)]
[mlir] Prevent crash in DropUnitDim pattern due to tensor with encoding
Differential Revision: https://reviews.llvm.org/D109984
Martin Storsjö [Thu, 16 Sep 2021 10:31:31 +0000 (13:31 +0300)]
[clang] [ARM] Don't set the strict alignment flag for armv7 on Windows
Windows on armv7 is as alignment tolerant as Linux.
The alignment considerations in the Windows on ARM ABI are documented
at https://docs.microsoft.com/en-us/cpp/build/overview-of-arm-abi-conventions?view=msvc-160#alignment.
The document doesn't explicitly say in which state the OS configures
the SCTLR.A register (and it's not accessible from user space to
inspect), but in practice, unaligned loads/stores do work and seem
to be as fast as aligned loads and stores. (Unaligned strd also does
seem to work, contrary to Linux, but significantly slower, as they're
handled by the kernel - exactly as the document describes.)
Differential Revision: https://reviews.llvm.org/D109960
peter klausler [Thu, 16 Sep 2021 00:25:15 +0000 (17:25 -0700)]
[flang] Enforce specification function rules on callee, not call
A function can't be a specification function if it has a dummy procedure
argument, even if it's optional and unused. So don't check the reference
for actual procedure arguments, but rather the characteristics of the
function.
Differential Revision: https://reviews.llvm.org/D109935
peter klausler [Wed, 15 Sep 2021 22:46:43 +0000 (15:46 -0700)]
[flang] Enforce fixed form rules about END continuation
From subclause 6.3.3.5: a program unit END statement cannot be
continued in fixed form, and other statements cannot have initial
lines that look like program unit END statements. I think this
is to avoid violating assumptions that are important to legacy
compilers' statement classification routines.
Differential Revision: https://reviews.llvm.org/D109933
Aaron Ballman [Fri, 17 Sep 2021 17:48:33 +0000 (13:48 -0400)]
Add information about C99 to the C status page.
Tozer [Fri, 10 Sep 2021 12:02:59 +0000 (13:02 +0100)]
[Dexter] Mutually exclusive argument group for --builder and --binary
Dexter currently accepts two possible arguments to determine the binary
used for testing; either --builder <builder> (and optionally
compiler/linker flags) to build the binary, or --binary <binary> to use
the provided binary directly. If both are passed, then --binary
overrides --builder; if neither are passed, then an error is raised.
This patch instead puts these arguments into a required mutually
exclusive argument group, so that an error is automatically raised by
argparse if both or neither are given.
As an additional change, the --cflags and --ldflags will now raise a
warning if they are passed without the --builder flag, as they are
meaningless if Dexter is using a pre-built binary.
Reviewed By: Orlando
Differential Revision: https://reviews.llvm.org/D109833
Alexey Bataev [Fri, 17 Sep 2021 17:30:35 +0000 (10:30 -0700)]
[SLP][NFC]Add a test for reorder of alt shuffle operands.
peter klausler [Mon, 13 Sep 2021 20:45:30 +0000 (13:45 -0700)]
[flang] Validate SIZE(x,DIM=n) dimension for assumed-size array x
Catch invalid attempts to extract the unknowable extent of the last
dimension of an assumed-size array dummy argument, and clean up
problems with assumed-rank arguments in similar circumstances
exposed by testing the fix.
Differential Revision: https://reviews.llvm.org/D109918
peter klausler [Tue, 14 Sep 2021 19:44:57 +0000 (12:44 -0700)]
[flang] Upgrade warning to error in case of PURE procedure
A procedure actual argument to a PURE procedure should be required
to have an explicit interface. Implicit-interface actual arguments
to non-PURE procedures remain a warning.
Differential Revision: https://reviews.llvm.org/D109926
thomasraoux [Fri, 17 Sep 2021 17:09:57 +0000 (10:09 -0700)]
[mlir][linalg] Extend drop unit dim pattern to all cases of reduction
Even with all parallel loops reading the output value is still allowed so we
don't have to handle reduction loops differently.
Differential Revision: https://reviews.llvm.org/D109851
thomasraoux [Fri, 17 Sep 2021 17:06:57 +0000 (10:06 -0700)]
[mlir] Linalg hoisting should ignore uses outside the loop
Differential Revision: https://reviews.llvm.org/D109859
thomasraoux [Thu, 16 Sep 2021 14:37:30 +0000 (07:37 -0700)]
[mlir] Fix potential crash in hoistRedundantVectorTransfers
Differential Revision: https://reviews.llvm.org/D107856
Vy Nguyen [Wed, 15 Sep 2021 19:49:56 +0000 (15:49 -0400)]
[lld-macho] Ignore local personality symbols if non-local with the same name exisst, to avoid "too many personalities" error.
Sometimes people intentionally re-define a dylib personlity symbol as a local defined symbol as a workaround to a ld -r bug.
As a result, we could see "too many personalities" to encode. This patch tries to handle this case by ignoring the local symbols entirely.
Differential Revision: https://reviews.llvm.org/D107533
Shilei Tian [Fri, 17 Sep 2021 16:54:40 +0000 (12:54 -0400)]
[NFC] clang-format -i /openmp/libomptarget/deviceRTLs/interface.h
peter klausler [Tue, 14 Sep 2021 18:39:48 +0000 (11:39 -0700)]
[flang] Correct overflow detection in folding of real->integer conversions
INT, NINT, FLOOR, and CEILING were failing to report overflow as an
error while folding operations with constant operands.
Differential Revision: https://reviews.llvm.org/D109922
AndreyChurbanov [Fri, 17 Sep 2021 16:49:40 +0000 (19:49 +0300)]
[OpenMP] NFC: add type casts to silence gcc warnings
AndreyChurbanov [Fri, 17 Sep 2021 16:38:34 +0000 (19:38 +0300)]
[OpenMP] libomp: Update third-party sources of ittnotify client code.
The third-party ittnotify sources updated from https://github.com/intel/ittapi.
Changes applied:
- llvm license aded to all files; initial BSD license saved in LICENSE.txt;
- clang-formatted;
- renamed *.c to *.cpp, similar to what we did with all our sources;
- added #include "kmp_config.h" with definition of INTEL_ITTNOTIFY_PREFIX macro
into ittnotify_static.cpp.
Differential Revision: https://reviews.llvm.org/D109333
peter klausler [Sat, 11 Sep 2021 00:35:47 +0000 (17:35 -0700)]
[flang] Catch errors with intrinsic array inquiry functions
The intrinsic inquiry functions SIZE and UBOUND -- but not LBOUND --
require a DIM= argument if their first argument is an assumed-size
array. The intrinsic SHAPE must not be used with an assumed-size
array.
Differential Revision: https://reviews.llvm.org/D109912
Kazu Hirata [Fri, 17 Sep 2021 16:16:39 +0000 (09:16 -0700)]
[llvm] Use drop_begin (NFC)
Roman Lebedev [Fri, 17 Sep 2021 16:07:01 +0000 (19:07 +0300)]
[X86] Improve `matchBinaryShuffle()`'s `BLEND` lowering with per-element all-zero/all-ones knowledge
We can use `OR` instead of `BLEND` if either the element we are not picking is zero (or masked away);
or the element we are picking overwhelms (e.g. it's all-ones) whatever the element we are not picking:
https://alive2.llvm.org/ce/z/RKejao
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D109726
Nico Weber [Fri, 17 Sep 2021 16:01:55 +0000 (12:01 -0400)]
[gn build] (semi-manually) port
750d5fc65c92
peter klausler [Mon, 13 Sep 2021 17:41:27 +0000 (10:41 -0700)]
[flang] Accept SIZE(assumed-rank, DIM=n)
Silence a bogus error message about an out-of-range DIM= argument
when the argument is assumed-rank. (More generally, don't pretend
to be able to discern a shape of known rank for an assumed-rank
object.)
Differential Revision: https://reviews.llvm.org/D109915
Sanjay Patel [Fri, 17 Sep 2021 14:54:59 +0000 (10:54 -0400)]
[InstCombine] allow splat vectors for narrowing masked fold
Mostly cosmetic diffs, but the use of m_APInt matches splat constants.
Sanjay Patel [Fri, 17 Sep 2021 13:45:56 +0000 (09:45 -0400)]
[InstCombine] add vector tests for 'and' folds; NFC
Tobias Gysi [Fri, 17 Sep 2021 14:54:04 +0000 (14:54 +0000)]
[mlir][linalg] Add helper to update IndexOps after tiling (NFC).
Add the addTileLoopIvsToIndexOpResults method to shift the IndexOp results after tiling.
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D109761
peter klausler [Thu, 9 Sep 2021 22:23:48 +0000 (15:23 -0700)]
[flang] Revamp C1502 checking of END INTERFACE [generic-spec]
Validation of the optional generic-spec on an END INTERFACE statement
was missing many possible error cases; reimplement it.
Differential Revision: https://reviews.llvm.org/D109910
Thomas Preud'homme [Fri, 17 Sep 2021 09:23:40 +0000 (10:23 +0100)]
Fix CodeGen/pgo-sample-thinlto-summary.c with old PM
Re-add -fexperimental-new-pass-manager to
Clang::CodeGen/pgo-sample-thinlto-summary.c for the test to work on
builds that still default to the old pass manager.
Reviewed By: tejohnson
Differential Revision: https://reviews.llvm.org/D109956
Matt Morehouse [Fri, 17 Sep 2021 14:08:57 +0000 (07:08 -0700)]
[HWASan] Intercept setjmp/longjmp on x86_64.
Reviewed By: xiangzhangllvm
Differential Revision: https://reviews.llvm.org/D109790
PeixinQiao [Fri, 17 Sep 2021 13:53:07 +0000 (21:53 +0800)]
[flang][OpenMP] Add semantic checks for ordered construct
This patch implements the following semantic checks according to
OpenMP Version 5.1 Ordered construct restriction:
```
At most one threads clause can appear on an ordered construct; At most
one simd clause can appear on an ordered construct; At most one
depend(source) clause can appear on an ordered construct; Either
depend(sink:vec) clauses or depend(source) clauses may appear on an
ordered construct, but not both.
```
This patch also implements the following semantic checks according to
the syntax and descriptions in OpenMP Version 5.1 Ordered construct:
```
The dependence types of sink or source are only allowed on an ordered
construct. The depend(*) clauses are not allowed when ordered construct
is a block construct with an ordered region. The threads or simd clauses
are not allowed when the ordered construct is a standalone construct
with no ordered region.
```
Co-authored-by: Sameeran Joshi <sameeranjayant.joshi@amd.com>
Reviewed By: kiranchandramohan
Differential Revision: https://reviews.llvm.org/D108512
Dávid Bolvanský [Fri, 17 Sep 2021 13:48:36 +0000 (15:48 +0200)]
[NFC] Added testcase for PR25725
Erich Keane [Fri, 17 Sep 2021 13:20:55 +0000 (06:20 -0700)]
Fix test failure from
e3b10525b489b604d6a1e540be78bda80afb5868
Seemingly, names in anonymous namespaces are ALWAYS given the unique
internal linkage name on windows, and I was not aware of this when I put
the names in my test! Replaced them with a wildcard.
Justas Janickas [Wed, 1 Sep 2021 14:20:01 +0000 (15:20 +0100)]
[OpenCL] Supports optional same image reads and writes in C++ for OpenCL 2021
Adds support for a feature macro `__opencl_c_read_write_images` in
C++ for OpenCL 2021 enabling a respective optional core feature
from OpenCL 3.0.
This change aims to achieve compatibility between C++ for OpenCL
2021 and OpenCL 3.0.
Differential Revision: https://reviews.llvm.org/D109307