This adds additional support for XL compatibility. There are a number
of functions in altivec.h that produce a single instruction (or a
very short sequence) for Power8 but can be done on Power7 without
scalarization. XL provides these implementations.
This patch adds the following overloads for doubleword vectors:
vec_add
vec_cmpeq
vec_cmpgt
vec_cmpge
vec_cmplt
vec_cmple
vec_sl
vec_sr
vec_sra

Differential Revision: https://reviews.llvm.org/D101766

This patch simplifies the parser and makes the language semantics
consistent. There is no extension pragma requirement in the spec
for the subgroup functions in enqueue kernel or pipes and all other
builtin functions are available without the pragama.

Differential Revision: https://reviews.llvm.org/D100984

[amdgpu-arch] Fix rpath to run from build dir

Prior to this, amdgpu-arch has RUNPATH set to $ORIGIN/../lib which works
for some installs, but not from the build directory where clang executes
the tool from when running tests.

This cmake option adds the location of the rocr runtime to the RUNPATH
(note, it amends RUNPATH here, despite the cmake option referring to RPATH)
to create a binary that runs from build or install location.

Before:
RUNPATH [$ORIGIN/../lib]
After:
RUNPATH [$ORIGIN/../lib:$HOME/llvm-install/lib]

Credit to Greg for knowing this trick and pointing to examples of it in use
for the aomp build scripts.

Reviewed By: pdhaliwal

Differential Revision: https://reviews.llvm.org/D101926

Instruction test for inactive kill/demote needs to be based on
actual opcode not whether instruction would be lowered to demote.

Reviewed By: piotr

Differential Revision: https://reviews.llvm.org/D101966

Reverting commit since it causes failure (10462).
This reverts commit b856f4a2.

The loop vectorizer will currently assume a large trip count when
calculating which of several vectorization factors are more profitable.
That is often not a terrible assumption to make as small trip count
loops will usually have been fully unrolled. There are cases however
where we will try to vectorize them, and especially when folding the
tail by masking can incorrectly choose to vectorize loops that are not
beneficial, due to the folded tail rounding the iteration count up for
the vectorized loop.

The motivating example here has a trip count of 5, so either performs 5
scalar iterations or 2 vector iterations (with VF=4). At a high enough
trip count the vectorization becomes profitable, but the rounding up to
2 vector iterations vs only 5 scalar makes it unprofitable.

This adds an alternative cost calculation when we know the max trip
count and are folding tail by masking, rounding the iteration count up
to the correct number for the vector width. We still do not account for
anything like setup cost or the mixture of vector and scalar loops, but
this is at least an improvement in a few cases that we have had
reported.

Differential Revision: https://reviews.llvm.org/D101726

This is a slight improvement to the help text, as I was slightly
surprised when strip-all did more than remove the symbol table.

Currently, we match gold's help text for strip-all and strip-debug.
I think that the GNU documentation for these options is not particularly
clear. However, I have opted to make only a minor change here and keep
the help text similar to gold's as these are mature options that are
well understood.

ld.bfd (https://sourceware.org/binutils/docs/ld/Options.html) has a
similar implication although it defines strip-debug as a subset of
strip-all. However, felt that noting that strip-all implies strip-debug
is better; because, with the ld.bfd approach you have to read both the
--strip-debug and the --strip-all help text to understand the behaviour
of --strip-all (and the --strip-all help text doesn't indicate that he
--strip-debug help text is related).

Differential Revision: https://reviews.llvm.org/D101890

Reviewed By: herhut

Differential Revision: https://reviews.llvm.org/D101757

In order to use __builtin_frame_address(0) with packed stack and no
backchain, the address of where the backchain would have been written is
returned (like GCC).

This address may either contain a saved register or be unused.

Review: Ulrich Weigand

Differential Revision: https://reviews.llvm.org/D101897

Adds support for scalable vectorization of loops containing first-order recurrences, e.g:
```
for(int i = 0; i < n; i++)
  b[i] =  a[i] + a[i - 1]
```
This patch changes fixFirstOrderRecurrence for scalable vectors to take vscale into
account when inserting into and extracting from the last lane of a vector.
CreateVectorSplice has been added to construct a vector for the recurrence, which
returns a splice intrinsic for scalable types. For fixed-width the behaviour
remains unchanged as CreateVectorSplice will return a shufflevector instead.

The tests included here are the same as test/Transform/LoopVectorize/first-order-recurrence.ll

Reviewed By: david-arm, fhahn

Differential Revision: https://reviews.llvm.org/D101076

Reviewed By: curdeius

Differential Revision: https://reviews.llvm.org/D101862

This fixes two errors:

Previously, clang-format was splitting up type identifiers from the
nullable ?. This changes this behavior so that the type name sticks with
the operator.

Additionally, nullable operators attached to return types in interface
functions were not parsed correctly. Digging deeper, it looks like
interface bodies were being parsed differently than classes and structs,
causing MustBeDeclaration to be incorrect for interface members. They
now share the same logic.

One other change is reintroducing the CSharpNullable type independent of
JsTypeOptionalQuestion. Despite having a similar semantic purpose, their
actual syntax differs quite a bit.

Reviewed By: MyDeveloperDay, curdeius

Differential Revision: https://reviews.llvm.org/D101860

This adds support for the null-coalescing assignment and null-forgiving
operators.

https://docs.microsoft.com/en-us/dotnet/csharp/language-reference/operators/null-coalescing-operator

https://docs.microsoft.com/en-us/dotnet/csharp/language-reference/operators/null-forgiving

Reviewed By: krasimir, curdeius

Differential Revision: https://reviews.llvm.org/D101702

First clean up the strange API of tryConstantFoldOp where it took an
immediate operand value, but no indication of which operand it was the
value for.

Second clean up the loop that calls tryConstantFoldOp so that it does
not have to restart from the beginning every time it folds an
instruction.

This is NFCI but there are some minor changes caused by the order in
which things are folded.

Differential Revision: https://reviews.llvm.org/D100031

`%f18` was originally introduced to represent the old Flang driver,
`f18`. With the introduction of the new driver, `flang-new`, we have
been switching to `%flang` (compiler driver) and `%flang_fc1` (frontend
driver) as more generic alternatives.

As most tests have been portend to use the new LIT variables instead of
`%f18`, this is good time to remove it from lit.cfg.py. There's only one
test left that requires the old driver to run. It's updated with:
```
! REQUIRES: old-flang-driver
```
This way we preserve its semantics while reducing the number of
variables in LIT configuration.

Differential Revision: https://reviews.llvm.org/D101281

This patch converts llvm.memcpy intrinsic into Tail Predicated
Hardware loops for a target that supports the Arm M-profile
Vector Extension (MVE).

From an implementation point of view, the patch

- adds an ARM specific SDAG Node (to which the llvm.memcpy intrinsic is lowered to, during first phase of ISel)
- adds a corresponding TableGen entry to generate a pseudo instruction, with a custom inserter,
  on matching the above node.
- Adds a custom inserter function that expands the pseudo instruction into MIR suitable
   to be (by later passes) into a WLSTP loop.

Note: A cli option is used to control the conversion of memcpy to TP
loop and this option is currently disabled by default. It may be enabled
in the future after further downstream testing.

Reviewed By: dmgreen

Differential Revision: https://reviews.llvm.org/D99723

Previously, if the search_env argument was specified, and the tool was
found at that location, the path was not reported, unlike other
situations when this function was called. Adding the reporting makes the
function consistent.

Reviewed by: thopre

Differential Revision: https://reviews.llvm.org/D101896

Fix up my recent commit rG1128311a to
use std::make_unique instead of std::unique_ptr(new), as requested by
David Blaikie.

Differential Revision: https://reviews.llvm.org/D101822

Differential Revision: https://reviews.llvm.org/D101910

Differential Revision: https://reviews.llvm.org/D101909

Differential Revision: https://reviews.llvm.org/D101907

Differential Revision: https://reviews.llvm.org/D101906

Differential Revision: https://reviews.llvm.org/D101905

This patch fixes various issues with our prior `declare target` handling
and extends it to support `omp begin declare target` as well.

This started with PR49649 in mind, trying to provide a way for users to
avoid the "ref" global use introduced for globals with internal linkage.
From there it went down the rabbit hole, e.g., all variables, even
`nohost` ones, were emitted into the device code so it was impossible to
determine if "ref" was needed late in the game (based on the name only).
To make it really useful, `begin declare target` was needed as it can
carry the `device_type`. Not emitting variables eagerly had a ripple
effect. Finally, the precedence of the (explicit) declare target list
items needed to be taken into account, that meant we cannot just look
for any declare target attribute to make a decision. This caused the
handling of functions to require fixup as well.

I tried to clean up things while I was at it, e.g., we should not "parse
declarations and defintions" as part of OpenMP parsing, this will always
break at some point. Instead, we keep track what region we are in and
act on definitions and declarations instead, this is what we do for
declare variant and other begin/end directives already.

Highlights:
  - new diagnosis for restrictions specificed in the standard,
  - delayed emission of globals not mentioned in an explicit
    list of a declare target,
  - omission of `nohost` globals on the host and `host` globals on the
    device,
  - no explicit parsing of declarations in-between `omp [begin] declare
    variant` and the corresponding end anymore, regular parsing instead,
  - precedence for explicit mentions in `declare target` lists over
    implicit mentions in the declaration-definition-seq, and
  - `omp allocate` declarations will now replace an earlier emitted
    global, if necessary.

---

Notes:

The patch is larger than I hoped but it turns out that most changes do
on their own lead to "inconsistent states", which seem less desirable
overall.

After working through this I feel the standard should remove the
explicit declare target forms as the delayed emission is horrible.
That said, while we delay things anyway, it seems to me we check too
often for the current status even though that is often not sufficient to
act upon. There seems to be a lot of duplication that can probably be
trimmed down. Eagerly emitting some things seems pretty weak as an
argument to keep so much logic around.

---

Reviewed By: ABataev

Differential Revision: https://reviews.llvm.org/D101030

A user reported an assertion (below) but without a reproducer. I failed to
create a test myself but from the assertion one can derive the problem.
I set the DefaultMapperId location now to make sure this doesn't cause
trouble.

```
clang-13: .../DeclTemplate.h:1940:
void clang::ClassTemplateSpecializationDecl::setPointOfInstantiation(clang::SourceLocation):
Assertion `Loc.isValid() && "point of instantiation must be valid!"' failed.
```

Reviewed By: JonChesterfield

Differential Revision: https://reviews.llvm.org/D100621

We do provide `operator delete(void*)` in `<new>` but it should be
available by default. This is mostly boilerplate to test it and the
unconditional include of `<new>` in the header we always in include
on the device.

Reviewed By: JonChesterfield

Differential Revision: https://reviews.llvm.org/D100620

Add warp synchronous matrix-multiply accumulate ops in GPU and NVVM
dialect. Add following three ops to GPU dialect :-
  1.) subgroup_mma_load_matrix
  2.) subgroup_mma_store_matrix
  3.) subgroup_mma_compute
Add following three ops to NVVM dialect :-
  1.) wmma.m16n16k16.load.[a,b,c].[f16,f32].row.stride
  2.) wmma.m16n16k16.store.d.[f16,f32].row.stride
  3.) wmma.m16n16k16.mma.row.row.[f16,f32].[f16,f32]

Reviewed By: bondhugula, ftynse, ThomasRaoux

Differential Revision: https://reviews.llvm.org/D95330

Having nested macros in the C code could cause clangd to fail an assert in clang::Preprocessor::setLoadedMacroDirective() and crash.

 #1 0x00000000007ace30 PrintStackTraceSignalHandler(void*) /qdelacru/llvm-project/llvm/lib/Support/Unix/Signals.inc:632:1
 #2 0x00000000007aaded llvm::sys::RunSignalHandlers() /qdelacru/llvm-project/llvm/lib/Support/Signals.cpp:76:20
 #3 0x00000000007ac7c1 SignalHandler(int) /qdelacru/llvm-project/llvm/lib/Support/Unix/Signals.inc:407:1
 #4 0x00007f096604db20 __restore_rt (/lib64/libpthread.so.0+0x12b20)
 #5 0x00007f0964b307ff raise (/lib64/libc.so.6+0x377ff)
 #6 0x00007f0964b1ac35 abort (/lib64/libc.so.6+0x21c35)
 #7 0x00007f0964b1ab09 _nl_load_domain.cold.0 (/lib64/libc.so.6+0x21b09)
 #8 0x00007f0964b28de6 (/lib64/libc.so.6+0x2fde6)
 #9 0x0000000001004d1a clang::Preprocessor::setLoadedMacroDirective(clang::IdentifierInfo*, clang::MacroDirective*, clang::MacroDirective*) /qdelacru/llvm-project/clang/lib/Lex/PPMacroExpansion.cpp:116:5

An example of the code that causes the assert failure:
```
...
```

During code completion in clangd, the macros will be loaded in loadMainFilePreambleMacros() by iterating over the macro names and calling PreambleIdentifiers->get(). Since these macro names are store in a StringSet (has StringMap underlying container), the order of the iterator is not guaranteed to be same as the order seen in the source code.

When clangd is trying to resolve nested macros it sometimes attempts to load them out of order which causes a macro to be stored twice. In the example above, ECHO2 macro gets resolved first, but since it uses another macro that has not been resolved it will try to resolve/store that as well. Now there are two MacroDirectives stored in the Preprocessor, ECHO and ECHO2. When clangd tries to load the next macro, ECHO, the preprocessor fails an assert in clang::Preprocessor::setLoadedMacroDirective() because there is already a MacroDirective stored for that macro name.

In this diff, I check if the macro is already inside the IdentifierTable and if it is skip it so that it is not resolved twice.

Reviewed By: kadircet

Differential Revision: https://reviews.llvm.org/D101870

This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D101849

[SelectionDAG][Mips][PowerPC][RISCV][WebAssembly] Teach computeKnownBits/ComputeNumSignBits about atomics

Unlike normal loads these don't have an extension field, but we know
from TargetLowering whether these are sign-extending or zero-extending,
and so can optimise away unnecessary extensions.

This was noticed on RISC-V, where sign extensions in the calling
convention would result in unnecessary explicit extension instructions,
but this also fixes some Mips inefficiencies. PowerPC sees churn in the
tests as all the zero extensions are only for promoting 32-bit to
64-bit, but these zero extensions are still not optimised away as they
should be, likely due to i32 being a legal type.

This also simplifies the WebAssembly code somewhat, which currently
works around the lack of target-independent combines with some ugly
patterns that break once they're optimised away.

Re-landed with correct handling in ComputeNumSignBits for Tmp == VTBits,
where zero-extending atomics were incorrectly returning 0 rather than
the (slightly confusing) required return value of 1.

Reviewed By: RKSimon, atanasyan

Differential Revision: https://reviews.llvm.org/D101342

https://reviews.llvm.org/D101194 changed the default getMultiarchTriple in toolchain.
So -march=bpf on AIX will get triple of bpf-ibm-aix now,
this is unexpected and causing test failures.

BPF on AIX is not supported (yet), disable the codegen test on AIX in lit cfg.

Reviewed By: yonghong-song

Differential Revision: https://reviews.llvm.org/D101866

Codegen for OpeMP copyin has non-deterministic IR output due to the unspecified evaluation order in a codegen conditional branch, which makes automatic test generation unreliable. This patch refactors codegen code to avoid this non-determinism.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D101952

This can be useful for clients constructing custom JIT stacks: If the C API
for your custom stack exposes API to obtain a reference to an object layer
(e.g. LLVMOrcLLJITGetObjLinkingLayer) then the newly added
LLVMOrcObjectLayerAddObjectFile and LLVMOrcObjectLayerAddObjectFileWithRT
functions can be used to add objects directly to that layer.

Fixes compilation on Android which has a TSDSharedRegistry object in the config.

Reviewed By: cryptoad, vitalybuka

Differential Revision: https://reviews.llvm.org/D101951