This reverts commit 7dbba39e.

Revert as there are reports this triggers during ThinLTO in some
configurations.

Combines G_SHUFFLE_VECTOR whose sources comes from G_CONCAT_VECTORS into
a single G_CONCAT_VECTORS instruction.

a = G_CONCAT_VECTORS x, y, undef, undef
b = G_CONCAT_VECTORS z, undef, undef, undef
c = G_SHUFFLE_VECTORS a, b, <0, 1, 4, undef>
===>
c = G_CONCAT_VECTORS x, y, z, undef`

Use sd_match to match commutative inner AND/OR/XOR node arguments instead of some messy manual matching of each commutation.

This is the first proof of concept of the modification of FIR codegen to
fully support a variety of top level operations (beyond just func.func)
proposed in
https://discourse.llvm.org/t/rfc-add-an-interface-for-top-level-container-operations

Commit #87622 broke the build. Ninja was happy with creating the output
directories as necessary, but Unix Makefiles isn't. Ensure they are
always created.

Fixes #88626.

Instead of doing a isa<> + getAs<>

This reverts commit eef57988.

This breaks two tests on an arm builder:
https://lab.llvm.org/buildbot/#/builders/245/builds/23496

Previously these were declared as having the 2008 behavior, with
underspecified signed zero handling. Currently, AMDGPU, PPC and
LoongArch mark these as legal. The AMDGPU and PPC instructions respect
the signed zero behavior. The LoongArch documentation doesn't state, but
I'm assuming it also does.

See https://mlir.llvm.org/deprecation and
https://discourse.llvm.org/t/preferred-casting-style-going-forward/68443/4

This relands the c8e65e19, which was
reverted in b48ea2d3 due to the breakage
of windows builtbot.

The reland contains some adjustments in the lit test deduction-gudie.cpp, to
make the checking text less strict.

[VectorCombine] foldShuffleOfBinops - don't fold shuffle(divrem(x,y),divrem(z,w)) if mask contains poison

Fixes #89390

It turns out that if any of the operations can be zero, and neither of
the operands can be proven to be positive, it is possible for smax to be
zero, and KnownBits cannot prove otherwise even with KnownBits::smax. In
fact, proving it based on the KnownBits itself at that point without
increasing the depth is actually, provably impossible.

Same with smin.

This covers all the possible cases and is proven to be complete.

This patch folds fabs over select if it is beneficial. I also tried
other interger/fp intrinsics. Only handling fabs shows benefit to some
real-world applications.

Reverts llvm/llvm-project#89213

This is causing buildbot failures.

Allow targets to implement custom module splitting logic for
--lto-partitions, see #89245

https://discourse.llvm.org/t/rfc-lto-target-specific-module-splittting/77252

This patch adds support for consumer fusion to the tiling interface, and
implements fuse consumers on FuseIntoContainingOp.

- Add interface method 'getIterDomainTilePositionFromOperandPosition' to
tiling interface which get iteration domain position from operand
position.
- Add interface method 'getTiledImplementationFromOperandPosition' to
tiling interface which generate tiled implementation according to
operand position.
- Implemented the above two methods and supported consumer fusion for
FuseIntoContainingOp.

Signed-off-by: Donald Chen

This resolves issue #88501.

We don't need to create these instances of ArrayRef because
ConstantDataVector::get takes ArrayRef, and ArrayRef can be implicitly
constructed from C arrays.

Fixes #89516

Similar to #87934, this adds costs to the shuffles in a canonical LD3/LD4
pattern, which are represented in LLVM as deinterleaving-shuffle(load). This
likely has less effect at the moment than the ST3/ST4 costs as instcombine will
perform certain transforms without considering the cost.

Reverts llvm/llvm-project#69267

this broke some bots.

This patch generalizes tensor.expand_shape and memref.expand_shape to
consume the output shape as a list of SSA values. This enables us to
implement generic reshape operations with dynamic shapes using
collapse_shape/expand_shape pairs.

The output_shape input to expand_shape follows the static/dynamic
representation that's also used in `tensor.extract_slice`.

Differential Revision: https://reviews.llvm.org/D140821



Co-authored-by: Ramiro Leal-Cavazos <ramiroleal050@gmail.com>

In a pack expression, if the id-expression is not valid, do no try to
detect whether it is a pack as that would lead to a crash trying to
print a recovery expression.

Fixes #88929

Revert "[X86] X86LowerTileCopy: Find dead register to use to prevent save-reload of tile register (#83628)"

This reverts commit 34acbb38.

This change causes major compile-time regressions.

To accomodate for recent changes in array index calculations.

foldFDivPowDivisor can address A / powi(x, y) to A * powi(x, -y),
while for small const value y, for example y=2, the instcombine will
transform powi(x, 2) to fmul x, x, so it is not optimal for A / powi(x, 2).

Fix https://github.com/llvm/llvm-project/issues/77171

Since PR86428, foldPowiReassoc is called by both FMul and FDiv,
as the optimization of FDiv is placed after the FMul, so now
it is correct we don't add the checking of FDiv for powi(X, Y) / X.
But, we may add more matching scenarios later, so add the checking opcode
explicitly is easier to understand.