Canonicalize operations

Composite fixed point pass

Sink operations into conditional blocks

Convert ElementwiseMappable ops to linalg

Lower the operations from the linalg dialect into affine loops

Lower the operations from the linalg dialect into loops

Lower the operations from the linalg dialect into parallel loops

Eliminate common sub-expressions

Generate additional runtime op verification checks

Make GPU ops async

Decomposes memref index computation into explicit ops.

Erase unnecessary barriers

Outline gpu.launch bodies to kernel functions

Sink index computations into gpu.launch body

Greedily maps loops to GPU hardware dimensions.

Transforms a GPU module into a GPU binary.

Inline function calls

Convert linalg matmul ops to block layout and back

Detensorize linalg ops

Remove unit-extent dimension in Linalg ops on tensors

Fuse elementwise operations on tensors

Convert named ops into generic ops

Inline scalar operands into linalg generic ops

Convert from one named linalg op to another.

Convert generic ops back to named ops

Hoist loop invariant instructions outside of the loop

Hoist loop invariant subset ops outside of the loop

Promotes memory slots into values.

Attaches an NVVM target attribute to a GPU Module.

Print IR on the debug stream

Print statistics of operations

Remove dead values

Attaches a ROCDL target attribute to a GPU Module.

Sparse Conditional Constant Propagation

Generate new locations from the current IR

Attaches an SPIR-V target attribute to a GPU Module.

Scalar Replacement of Aggregates

Strip debug info from all operations

Eliminate dead symbols

Mark symbols private

Sort regions without SSA dominance in topological order

Print Graphviz visualization of an operation