value profile annotated after inlining.

In https://reviews.llvm.org/D96806 and https://reviews.llvm.org/D97350, we
use the magic number -1 in the value profile to avoid repeated indirect call
promotion to the same target for an indirect call. Function updateIDTMetaData
is used to mark an target as being promoted in the value profile with the
magic number. updateIDTMetaData is also used to update the value profile
when an indirect call is inlined and new inline instance profile should be
applied. For the second case, currently updateIDTMetaData mixes up the
existing value profile of the indirect call with the new profile, leading
to the problematic senario that a target count is larger than the total count
in the value profile.

The patch fixes the problem. When updateIDTMetaData is used to update the
value profile after inlining, all the values in the existing value profile
will be dropped except the values with the magic numbe...

This reverts commit 11b70b9e.

The bot failure was due to ArgumentPromotion deleting functions
without deleting their analyses. This was separately fixed in 4b1c8070.

This is a test commit to verify my access.

Not doing it here can lead to subtle bugs - the analysis results are
associated by the Function object's address. Nothing stops the memory
allocator from allocating new functions at the same address.

Added basic parsing/sema/serialization support to extend the
existing 'destroy' clause for use with the 'interop' directive.

Differential Revision: https://reviews.llvm.org/D98834

Valgrind is reporting this bogus warning because it doesn't model
pthread_sigmask fully accurately.  This is a valgrind bug, but
silencing it has effectively no cost, so just do it.

==73662== Syscall param __pthread_sigmask(set) points to uninitialised byte(s)
==73662==    at 0x101E9D4C2: __pthread_sigmask (in /usr/lib/system/libsystem_kernel.dylib)
==73662==    by 0x101EFB5EA: pthread_sigmask (in /usr/lib/system/libsystem_pthread.dylib)
==73662==    by 0x1000D9F6D: llvm::sys::Process::SafelyCloseFileDescriptor(int) (in /Users/chrisl/Projects/circt/build/bin/firtool)
==73662==    by 0x100072795: llvm::ErrorOr<std::__1::unique_ptr<llvm::MemoryBuffer, std::__1::default_delete<llvm::MemoryBuffer> > > getFileAux<llvm::MemoryBuffer>(llvm::Twine const&, long long, unsigned long long, unsigned long long, bool, bool) (in /Users/chrisl/Projects/circt/build/bin/firtool)
==73662==    by 0x100072573: llvm::MemoryBuffer::getFileOrSTDIN(llvm::Twine const&, long long, bool) (in /Users/chrisl/Projects/circt/build/bin/firtool)
==73662==    by 0x100282C25: mlir::openInputFile(llvm::StringRef, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >*) (in /Users/chrisl/Projects/circt/build/bin

Differential Revision: https://reviews.llvm.org/D98830

Differential Revision: https://reviews.llvm.org/D98804

[amdgpu] Update med3 combine to skip i64

Fixes an assumption that a type which is not i32 will be i16. This asserts
when trying to sign/zero extend an i64 to i32.

Test case was cut down from an openmp application. Variations on it are hit by
other combines before reaching the problematic one, e.g. replacing the
immediate values with other function arguments changes the codegen path and
misses this combine.

Reviewed By: arsenm

Differential Revision: https://reviews.llvm.org/D98872

Create test fixture for runtime tests which enables verification
of failure cases. Test some runtime IO APIs for failure cases.
Support testing efforts in D98303. Expand on effort discussed
in D98601.

Reviewed By: awarzynski

Differential Revision: https://reviews.llvm.org/D98652

The pass can be seen on staging buildbot clang-xcore-ubuntu-20-x64.

Differential Revision: https://reviews.llvm.org/D98352

Extend this to support ComputeNumSignBits of the (used) source vector elements so that we can handle more than just the case where we're sext_in_reg from the source element signbit.

Noticed while investigating the poor codegen in D98587.

Also use `ArrayAttr` to pass iterator pass to the TiledLoopOp builder.

Differential Revision: https://reviews.llvm.org/D98871

Allow inline assembly code to referece cs0 and cs1.

Implement command-line options -map

Reviewed By: int3, #lld-macho

Differential Revision: https://reviews.llvm.org/D98323

Fix the test to account for recent test infrastructure changes, and make
it run locally to increase the chances of it continuing to work in the
future.

Adds support for `-fget-symbols-sources` in the new Flang driver. All
relevant tests are updated to use the new driver when
`FLANG_BUILD_NEW_DRIVER` is set.

`RUN` lines in tests are updated so `-fsyntax-only`
comes before `-fget-symbols-sources`. That's because:
  * both `-fsyntax-only` and `-fget-symbols-sources` are
    action flags, and
  * the new driver, flang-new, will only consider the right-most
    action flag.
In other words, this change is needed so that the tests work with both
`f18` (requires both flags) and `flang-new` (only considers the last
action flag).

Differential Revision: https://reviews.llvm.org/D98191

Delay the issue of a new instruction if that leads to out-of-order
commits of writes.

This patch fixes the problem described in:
https://bugs.llvm.org/show_bug.cgi?id=41796#c3

Differential Revision: https://reviews.llvm.org/D98604

Co-authored-by: Kiran Chandramohan <kiran.chandramohan@arm.com>

Reviewed By: ftynse

Differential Revision: https://reviews.llvm.org/D92327

Pass through the original argument IR value in order to preserve the
aliasing information in the memcpy memory operands.

byval requires an implicit copy between the caller and callee such
that the callee may write into the stack area without it modifying the
value in the parent. Previously, this was passing through the raw
pointer value which would break if the callee wrote into it.

Most of the time, this copy can be optimized out (however we don't
have the optimization SelectionDAG does yet).

This will trigger more fallbacks for AMDGPU now, since we don't have
legalization for memcpy yet (although we should stop using byval
anyway).

If SLP vectorizer tries to extend the scheduling region and runs out of
the budget too early, but still extends the region to the new ending
instructions (i.e., it was able to extend the region for the first
instruction in the bundle, but not for the second), the compiler need to
recalculate dependecies in full, just like if the extending was
successfull. Without it, the schedule data chunks may end up with the
wrong number of (unscheduled) dependecies and it may end up with the
incorrect function, where the vectorized instruction does not dominate
on the extractelement instruction.

Differential Revision: https://reviews.llvm.org/D98531

This patch is follow-up for D91028. It implements direct writing into the
output stream for wasm.

Depends on D91028

Differential Revision: https://reviews.llvm.org/D95478

If producing libraries with an arch suffix (i.e. if
LLVM_ENABLE_PER_TARGET_RUNTIME_DIR isn't set), we append the
architecture name. However, for arm, clang doesn't look for libraries
with the full architecture name, but only looks for "arm" and "armhf".

Try to deduce what the full target triple might have been, and use
that for deciding between "arm" and "armhf".

This tries to reapply this bit from D98173, that had to be reverted
in 7b153b43 due to affecting how
the builtins themselves are compiled, not only affecting the output
file name.

Differential Revision: https://reviews.llvm.org/D98452

Just makes code simpler.

Provides API that allows to check predicate for being true or
false with one call. Current implementation is naive and just
calls isKnownPredicate twice, but further we can rework this
logic trying to use one check to prove both facts.

LLVM test CodeGen/AArch64/machine-outliner-retaddr-sign-thunk.ll uses
a string substitution block that contains a regex matching block. This
seems like as a copy/paste from other similar test where the match also
defines a variable, hence the [[]] syntax. In this case however this is
a CHECK-NOT variable so nothing should match. No variable definition is
thus expected and the square brackets can be dropped.

Reviewed By: chill

Differential Revision: https://reviews.llvm.org/D98853

The `int` and `long` versions of these builtins already provide the
necessary overloads for `intptr_t` and `uintptr_t` arguments, as
`ASTContext` defines `atomic_(u)intptr_t` in terms of the `int` or
`long` types.

Prior to this patch, calls to those builtins with particular argument
types resulted in call-is-ambiguous errors.

Differential Revision: https://reviews.llvm.org/D98520

Clang test acle_sve_ld1.sh is missing the colon in one of the string
variable definition separating the variable name from the regex. This
leads the substitution block to be parsed as a numeric variable use.

Reviewed By: sdesmalen

Differential Revision: https://reviews.llvm.org/D98852

This makes the induction part of the loop vectorizer match the reduction part.
We do not need all of the fast-math-flags. For example, there are some that
clearly are not in play like arcp or afn.

If we want to make FMF constraints consistent across the IR optimizer, we
might want to add nsz too, but that's up for debate (users can't expect
associative FP math and preservation of sign-of-zero at the same time?).

The calling code was fixed to avoid miscompiles with:
1bee5497

Differential Revision: https://reviews.llvm.org/D98708

After the import, we did not copy the `TSCSpec`.
This commit resolves that.

Reviewed By: balazske

Differential Revision: https://reviews.llvm.org/D98707

This patch is a second attempt at fixing a link error for MSVC
entry points when calling conventions are specified using a flag.

Calling conventions specified using flags should not be applied to MSVC
entry points. The default calling convention is set in this case. The
default calling convention for MSVC entry points main and wmain is cdecl.
For WinMain, wWinMain and DllMain, the default calling convention is
stdcall on 32 bit Windows.

Explicitly specified calling conventions are applied to MSVC entry points.

For MinGW, the default calling convention for all MSVC entry points is
cdecl.

First attempt: 4cff1b40
Revert of first attempt: bebfc3b9

Differential Revision: https://reviews.llvm.org/D97941

This patch removes creation of the resulting file from the
executeObjcopyOnBinary() function. For the most use cases, the
executeObjcopyOnBinary receives output file as a parameter
- raw_ostream &Out. The splitting .dwo file is implemented differently:
file containg .dwo tables is created inside executeObjcopyOnBinary().
When objcopy functionality would be moved into separate library,
current implementation will become inconvenient. The goal of that
refactoring is to separate concerns: It might be convenient to
to do dwo tables splitting but to create resulting file differently.

Differential Revision: https://reviews.llvm.org/D98582

Add ISD::ABS to the existing unary instructions handling for splat detection

This is similar to D83605, but doesn't appear to need to touch any of the wasm refactoring.

Differential Revision: https://reviews.llvm.org/D98778