This target will depend on each individual extension and represent "all"
Python bindings in the repo. User projects can get a finer grain control by
depending directly on some individual targets as needed.

Differential Revision: https://reviews.llvm.org/D90816

The Python bindings now require -DLLVM_BUILD_LLVM_DYLIB=ON to build.
This change is needed to be able to build multiple Python native
extension without having each of them embedding a copy of MLIR, which
would make them incompatible with each other. Instead they should all
link to the same copy of MLIR.

Differential Revision: https://reviews.llvm.org/D90813

This is the cmp/sel sibling to D90692.
Again, the reasoning is: the throughput cost is number of instructions/uops,
so size/blended costs are identical except in special cases (for example,
fdiv or other known-expensive machine instructions or things like MVE that
may require cracking into >1 uops).

We need to check for a valid (non-null) condition type parameter because
SimplifyCFG may pass nullptr for that (and so we will crash multiple
regression tests without that check). I'm not sure if passing nullptr makes
sense, but other code in the cost model does appear to check if that param
is set or not.

Differential Revision: https://reviews.llvm.org/D90781

Add IgnoreMainLikeFunctions to the per file config. This can be extended for new options added to the check easily.

Reviewed By: aaron.ballman

Differential Revision: https://reviews.llvm.org/D90832

Remove the thread name caching code.  It does not handle the possibility
of thread name changing between requests, therefore breaking
TestGdbRemoteThreadName.  While technically we could cache the results
and reset the cache on resuming process, the gain from doing that
does not seem worth the effort.

Differential Revision: https://reviews.llvm.org/D90863

Fix TestGdbRemoteThreadName to call ::pthread_setname_np instead
of ::pthread_set_name_np on FreeBSD.  While technically both names
are correct, the former is preferable because of compatibility
with Linux.  Furthermore, the latter requires `#include <pthread_np.h>`
that was missing causing the test to fail to compile.

Differential Revision: https://reviews.llvm.org/D90862

This is more or less a port of rL329598 (D45275) to the COFF linker.
Since there were already LTO-related settings under -opt:, I added
them there instead of new flags.

Differential Revision: https://reviews.llvm.org/D90624

The debug location is removed from any outlined instruction. This
causes the MachineVerifier to crash on outlined DBG_VALUE
instructions.

Then, debug instructions are "invisible" to the outliner, that is, two
ranges of instructions from different functions are considered
identical if the only difference is debug instructions. Since a debug
instruction from one function is unlikely to provide sensible debug
information about all functions, sharing an outlined sequence, this
patch just removes debug instructions from the outlined functions.

Differential Revision: https://reviews.llvm.org/D89485

This functionality is superceded by BufferResultsToOutParams pass (see
https://reviews.llvm.org/D90071) for users the require buffers to be
out-params. That pass should be run immediately after all tensors are gone from
the program (before buffer optimizations and deallocation insertion), such as
immediately after a "finalizing" bufferize pass.

The -test-finalizing-bufferize pass now defaults to what used to be the
`allowMemrefFunctionResults=true` flag. and the
finalizing-bufferize-allowed-memref-results.mlir file is moved
to test/Transforms/finalizing-bufferize.mlir.

Differential Revision: https://reviews.llvm.org/D90778

These were previously handled by pattern matching shuffles in the selector, but
adding a new opcode and making it equivalent to the AArch64duplane SDAG node
allows us to select more patterns, like lane indexed FMLAs (patch adding a test
for that will be committed later).

The pattern matching code has been simply moved to postlegalize lowering.

Differential Revision: https://reviews.llvm.org/D90820

This corrects WrapperGen generating incorrect wrappers for functions
that take no arguments. Previously it would generate a wrapper with a
single argument of type `void`.

Reviewed By: sivachandra

Differential Revision: https://reviews.llvm.org/D90800

[LegalizeTypes] Remove unnecessary if around switch in ScalarizeVectorOperand and SplitVectorOperand. NFC

The if was checking !Res.getNode() but that's always true since
Res was initialized to SDValue() and not touched before the if.

This appears to be a leftover from a previous implementation of
Custom legalization where Res was updated instead of returning
immediately.

This adds the skeleton of the YAML Compiler for APINotes.  This change
only adds the YAML IO model for the API Notes along with a new testing
tool `apinotes-test` which can be used to verify that can round trip the
YAML content properly.  It provides the basis for the future work which
will add a binary serialization and deserialization format to the data
model.

This is based on the code contributed by Apple at
https://github.com/llvm/llvm-project-staging/tree/staging/swift/apinotes.

Differential Revision: https://reviews.llvm.org/D88859
Reviewed By: Gabor Marton

Looks accidentally omitted, it's present on `readsRegister`,
`definesRegister` and few others.

Differential Revision: https://reviews.llvm.org/D89625

D80526 added custom lowering to pick the si lib call on RV64, but this custom handling is only enabled when the F and D extension are both disabled. This prevents the si library call from being used for double when F is enabled but D is not.

This patch changes the behavior so we always enable the Custom hook on RV64 and decide in ReplaceNodeResults if we should emit a libcall based on whether the FP type should be softened or not.

Differential Revision: https://reviews.llvm.org/D90817

This change adds a real glc operand to the return atomic
instead of just string " glc" in the middle of the asm
string.

Improves asm parser diagnostics.

Differential Revision: https://reviews.llvm.org/D90730

https://reviews.llvm.org/D90811 is breaking our CI builders because
InitializePlatformCommonFlags is not defined. This just adds an empty definition.

This would've been caught on our upstream buildbot, but it's red at the moment
and most likely won't be sending out alert emails for recent failures.

Differential Revision: https://reviews.llvm.org/D90864

These options are in practice passes to the external tool (defined with
F18_FC), i.e. fall into "unrecognised" category. No need to keep them
among other options that are actually parsed.

This is the only -Werror- form warning option GCC supports (gcc/c-family/c.opt).
Fortunately no other form is used anywhere.

Differential Revision: https://reviews.llvm.org/D90828

[RISCV] Remove shadow register list passed to AllocateReg when allocating FP registers for calling convention

The _F and _D registers are already sub/super registers. When one gets allocated all its aliases are already marked as allocated. We don't need to explicitly shadow it too.

I believe shadow is for calling conventions like 64-bit Windows on X86 where have rules like this

CCIfType<[i32], CCAssignToRegWithShadow<[ECX , EDX , R8D , R9D ],
                                         [XMM0, XMM1, XMM2, XMM3]>>

For that calling convention the argument number determines which register is used regardless of how many scalars or vectors came before it.

Removing this removes a question I had in D90738.

Differential Revision: https://reviews.llvm.org/D90801

There is no FSLI instruction, but we can emulate it using FSRI by swapping operands and subtracting the immediate from the bitwidth.

Differential Revision: https://reviews.llvm.org/D90826

Currently, vendor-specific availability markup is enabled by default.
This means that even when building against trunk libc++, the headers
will by default prevent you from using some features that were not
released in the dylib on your target platform. This is a source of
frustration since people building libc++ from sources are usually not
trying to use some vendor's released dylib.

For that reason, I've been thinking for a long time that availability
annotations should be off by default, which is the primary change that
this commit enables.

In addition, it reworks the implementation to make it easier for new
vendors to add availability annotations for their platform, and it
refreshes the documentation to reflect the current state of the codebase.

Finally, a CMake configuration option is added to control whether
availability annotations should be turned on for the flavor of libc++
being created. The intent is for vendors like Apple to turn it on, and
for the upstream libc++ to leave it off (the default).

Differential Revision: https://reviews.llvm.org/D90843

This reverts commit 15694fd6.
Need to investigate and fix a failing clang test: synchronized.m.
Might need a test update.

There is no need to memset released pages because they are already
zero. On db845c, before:

BM_stdlib_malloc_free_default/131072      34562 ns        34547 ns        20258 bytes_per_second=3.53345G/s

after:

BM_stdlib_malloc_free_default/131072      29618 ns        29589 ns        23485 bytes_per_second=4.12548G/s

Differential Revision: https://reviews.llvm.org/D90814

This patch adds the mask and ident_t function to get the
openmp version. It also adds logic to force monotonic:dynamic
behavior when OpenMP version less than 5.0.

The OpenMP version is stored in the format:
major*10+minor e.g., OpenMP 5.0 = 50

Differential Revision: https://reviews.llvm.org/D90632

Copy+Paste typo - we were storing getNumOperands() opcounts in a Register type instead of just an unsigned.

Differential Revision: https://reviews.llvm.org/D90858

This moves WidenIV from IndVarSimplify to Utils/SimplifyIndVar so that we have
createWideIV available as a generic helper utility. I.e., this is not only
useful in IndVarSimplify, but could be useful for loop transformations. For
example, motivation for this refactoring is the loop flatten transformation: if
induction variables in a loop nest can be widened, we can avoid having to
perform certain overflow checks, enabling this transformation.

Differential Revision: https://reviews.llvm.org/D90421

The new FreeBSDRemote plugin has reached feature parity on i386
and amd64 targets.  Use it by default on these architectures, while
allowing the use of the legacy plugin via FREEBSD_LEGACY_PLUGIN envvar.

Revisit the method of switching plugins.  Apparently, the return value
of PlatformFreeBSD::CanDebugProcess() is what really decides whether
the legacy or the new plugin is used.

Update the test status.  Reenable the tests that were previously
disabled on FreeBSD and do not cause hangs or are irrelevant to FreeBSD.
Mark all tests that fail reliably as expectedFailure.  For now, tests
that are flaky (i.e. produce unstable results) are left enabled
and cause unpredictable test failures.

Differential Revision: https://reviews.llvm.org/D90757

Pass -pthread on all systems except for Darwin and Windows.
Suggested by Pavel Labath.

CapturesBefore tracker has an overly restrictive dominates check when
the `BeforeHere` and the capture point are in different basic blocks.
All we need to check is that there is no path from the capture point
to `BeforeHere` (which is less stricter than the dominates check).
See added testcase in one of the users of CapturesBefore.

Reviewed-By: jdoerfert
Differential Revision: https://reviews.llvm.org/D90688

5d645743 removes this enum value and now
all the switch statements that previously relied on handling this in the
'default' branch are causes compiler warnings due to redundant default cases.

This just removes the now unreachable code in there.

In D89056 the default value for architecture was moved to `build` so that
all called functions see the same architecture value. It seems there are a
few functions that call buildDefault directly (and not via build), so
on some test configurations that set a custom arch value the architecture
value is no longer available.

This just adds the architecture code from build to buildDefault to get
the bots green again while I'm looking for a better solution.

As described here:
https://devblogs.microsoft.com/oldnewthing/20150220-00/?p=44623

In order to allow Lambdas to be used with traditional Win32 APIs, they
emit a conversion function for (what Raymond Chen claims is all) a
number of the calling conventions.  Through experimentation, we
discovered that the list isn't quite 'all'.

This patch implements this by taking the list of conversions that MSVC
emits (across 'all' architectures, I don't see any CCs on ARM), then
emits them if they are supported by the current target.

However, we also add 3 other options (which may be duplicates):
free-function, member-function, and operator() calling conventions.  We
do this because we have an extension where we generate both free and
member for these cases so th at people specifying a calling convention
on the lambda will have the expected behavior when specifying one of
those two.

MSVC doesn't seem to permit specifying calling-convention on lambdas,
but we do, so we need to make sure those are emitted as well. We do this
so that clang-only conventions are supported if the user specifies them.

Differential Revision: https://reviews.llvm.org/D90634