For the conversion to nvgpu `mma.sync` and `ldmatrix` pathways, the code
was missing support for the `i4` data type. While fixing this, another
bug was discoverd that caused the number of ldmatrix tiles calculated for
certain operand types and configurations to be incorrect. This change
fixes both issues and adds additional tests.

Differential Revision: https://reviews.llvm.org/D128074

TestObjCXXBridgedPO is broken on macOS Ventura (but not on macOS
Monterey). I took a look but it doesn't seem trivial. I'm XFAILing the
test until Adrian, who wrote the test, can take a look.

rdar://96224141

Differential Revision: https://reviews.llvm.org/D128681

The check to see if the arguments for the MIN/MAX intrinsics were of CHARACTER
type was not handling assumed length characters.  In this case, the FIR type is
"!fir.ref<!fir.char<1,?>>".

This patch is part of the upstreaming effort from fir-dev branch.

Reviewed By: PeteSteinfeld

Differential Revision: https://reviews.llvm.org/D128922



Co-authored-by: Peter Steinfeld <psteinfeld@nvidia.com>

Many of our tests are currently written using `TEST_F` where the test fixture class doesn't have any `SetUp` or `TearDown` methods, and just one helper method. In those cases, this patch deletes the class and pulls its method out into a standalone function, using `TEST` instead of `TEST_F`.

There are still a few test files leftover in `clang/unittests/Analysis/FlowSensitive/` that use `TEST_F`:

- `DataflowAnalysisContextTest.cpp` because the class contains a `Context` field which is used
- `DataflowEnvironmentTest.cpp` because the class contains an `Environment` field which is used
- `SolverTest.cpp` because the class contains a `Vals` field which is used
- `TypeErasedDataflowAnalysisTest.cpp` because there are several different classes which all share the same method name

Reviewed By: ymandel, sgatev

Differential Revision: https://reviews.llvm.org/D128924

Similar for a subtract with a constant left hand side.

(sra (add (shl X, 32), C1<<32), 32) is the canonical IR from InstCombine
for (sext (add (trunc X to i32), 32) to i32).

For RISCV, we should lower this as addiw which means turning it into
(sext_inreg (add X, C1)).

There is an existing DAG combine to convert back to (sext (add (trunc X
to i32), 32) to i32), but it requires isTruncateFree to return true
and for i32 to be a legal type as it used sign_extend and truncate
nodes. So that doesn't work for RISCV.

If the outer sra happens be used by a shl by constant, it will be
folded and the shift amount of the sra will be changed before we
can do our own DAG combine. This requires us to match the more
general pattern and restore the shl.

I had wanted to do this as a separate (add (shl X, 32), C1<<32) ->
(shl (add X, C1), 32) combine, but that hit an infinite loop for some
values of C1.

Reviewed By: asb

Differential Revision: https://reviews.llvm.org/D128869

The sext_inreg can often be folded into an earlier instruction by
using a W instruction. The sext_inreg also works better with our ABI.

This is one of the steps to improving the generated code for this https://godbolt.org/z/hssn6sPco

Reviewed By: asb

Differential Revision: https://reviews.llvm.org/D128843

Fixes error: missing '#include "llvm/IR/FMF.h"'; 'FastMathFlags' must be
defined before it is used in llvm/include/llvm/IR/NoFolder.h.

Adds support for reading and writing LTO bitcode files.

  - Emit a summary if the original bitcode file had a summary
  - Use split LTO units if the original bitcode file used them.

Reviewed By: arsenm

Differential Revision: https://reviews.llvm.org/D127168

As Fortran 2018 16.9.163, the reshape is the only intrinsic which
requires the shape argument to be rank-one integer array and the SIZE
of it to be one constant expression. The current expression lowering
converts the shape expression with slice in intrinsic into one box value
with the box element type of unknown extent. However, the genReshape
requires the box element type to be constant size. So, convert the box
value into one with box element type of sequence of 1 x constant. This
corner case is found in cam4 in SPEC 2017
https://github.com/llvm/llvm-project/issues/56140.

Reviewed By: Jean Perier

Differential Revision: https://reviews.llvm.org/D128597

gfx11 introduces new WMMA (Wave Matrix Multiply-accumulate)
instructions.

Reviewed By: arsenm, #amdgpu

Differential Revision: https://reviews.llvm.org/D128756

Reviewed By: ldionne, #libc

Spies: libcxx-commits

Differential Revision: https://reviews.llvm.org/D128906

Reviewed By: ldionne, #libc

Spies: cfe-commits, llvm-commits, libcxx-commits

Differential Revision: https://reviews.llvm.org/D127313

D128285 only changed the stable (v1) layout, so the matching change in
D128694 broke the formatting of the unstable strings. This fixes that,
and ensures compatibility with all older layouts as well.

Migrate all binops to use FoldXYZ rather than CreateXYZ APIs,
which are compatible with InstSimplifyFolder and fallible constant
folding.

Rather than continuing to add one method for every single operator,
add a generic FoldBinOp (plus variants for nowrap, exact and fmf
operators), which we would need anyway for CreateBinaryOp.

This change is not NFC because IRBuilder with InstSimplifyFolder
may perform more folding. However, this patch changes SCEVExpander
to not use the folder in InsertBinOp to minimize practical impact
and keep this change as close to NFC as possible.

PDB/func-symbols.test was orignally written for 32bit x86, keeping in
mind cdecl and stdcall calling conventions which does name mangling for
example like adding "_" underscore before function name.
This is only x86 specific but purpose of pointers.test is NOT to test
calling convention.
I have made a minor change to make this test pass on Windows/Arm.

TestCommandScript.py fails on Arm/Windows due following issues:
https://llvm.org/pr56288
https://llvm.org/pr56292

LLDB fails to skip prologue and also step over library function or
nodebug functions fails due to PDB/DWARF mismatch.

This patch replace function breakpoint with line breakpoint so that we
can expect LLDB to stop on desired line. Also replace dwarf with PDB
debug info for this test only.

This is a continuation of D119544.  Based on @rsmith 's feed back
showing me https://eel.is/c++draft/temp#friend-9, We should properly
handle friend functions now.

Differential Revision: https://reviews.llvm.org/D126907

The original assertion is not necessarily correct since the shape
argument may involve a slice of an array (an expression) and not a whole
vector with constant length. In the presence of a slice operation, the
size must be computed (left as a TODO for now).

This patch is part of the upstreaming effort from fir-dev branch.

Reviewed By: jeanPerier

Differential Revision: https://reviews.llvm.org/D128894



Co-authored-by: Eric Schweitz <eschweitz@nvidia.com>

This means we no longer need to have the same API between IRBuilder
and IRBuilderFolder.

The constant case is substantially simpler, so implementing it
separately isn't an undue burden.

Change WaveActiveCountBits from builtin into wrapper of __builtin_hlsl_wave_active_count_bits.
For comment at
https://reviews.llvm.org/D126857#inline-1235949

Reviewed By: beanz

Differential Revision: https://reviews.llvm.org/D128855

Fixing issue "incorrect -Winfinite-recursion warning on potentially-
unevaluated operand".

We add a dedicated visit function (VisitCXXTypeidExpr) for typeid,
instead of using the default (VisitStmt). In this new function we skip
over building the CFG for unevaluated operands of typeid.

Fixes #21668

Differential Revision: https://reviews.llvm.org/D128747

Nowdays we have a generic constant folding API to load a type from
an offset. It should be able to do anything that VNCoercion can do.

This avoids the weird templating between IRBuilder and ConstantFolder
in one function, which is will stop working as the IRBuilderFolder
moves from CreateXYZ to FoldXYZ APIs.

Unfortunately, this doesn't eliminate this pattern from VNCoercion
entirely yet.

It took me multiple hours of debugging plus asking an expert for help to
figure out why this function didn't do what it promised to do. It turns
out there is a flag that needs to be set. Document this, in an attempt
to save the next person the surprise.

Reviewed By: ymandel

Differential Revision: https://reviews.llvm.org/D128774

Reviewed By: ldionne, #libc, EricWF

Spies: EricWF, libcxx-commits

Differential Revision: https://reviews.llvm.org/D127919

At the moment LoopVersioning is only created for inner-loop
vectorization. This patch moves it to LVP::execute, which means it will
also be added for epilogue vectorization. As a consequence, the proper
noalias metadata is now also added to epilogue vector loops.

LVer will be moved to VPTransformState as follow-up.

Reviewed By: Ayal

Differential Revision: https://reviews.llvm.org/D127966

This allows including the headers without risk of conflict with
user-defined macros e.g. max

Differential Revision: https://reviews.llvm.org/D128728

[mlir][Linalg] Uniformize SplitReduction transforms and add option to use Bufferization::AllocTensor

This revision merges the 2 split_reduction transforms and adds extra control by using attributes.

SplitReduction is known to require a concrete additional buffer to store tempoaray information.
Add an option to introduce a `bufferization.alloc_tensor` instead of `linalg.init_tensor`.
This behaves better with subset-based tiling and bufferization.

Differential Revision: https://reviews.llvm.org/D128722

The assert was added with 0399473d and is correct for that
pattern, but it is off-by-1 with the enhancement in d4f39d83.

The transforms are still correct with the new pre-condition:
https://alive2.llvm.org/ce/z/6_6ghm
https://alive2.llvm.org/ce/z/_GTBUt

And as shown in the new test, the transform is expected with
'ult' - in that case, the icmp reduces to test if the shift
amount is 0.

This allows all constant folding to happen through a single
function, without requiring special handling for loads at each
call-site.

This may not be NFC because some callers currently don't do that
special handling.