Operations#

Load/Store#

`bid`	Gets the index of current block.
`num_blocks`	Gets the number of blocks along the axis.
`num_tiles`	Gets the number of tiles in the tile space of the array along the axis.
`load`	Loads a tile from the array which is partitioned into a tile space.
`store`	Stores a tile value into the array at the index of its tile space.
`load_advanced_indexing`	Loads a tile from non-contiguous slices of array.
`store_advanced_indexing`	Stores a tile into non-contiguous slices of array.
`gather`	Loads a tile from the array elements specified by indices.
`scatter`	Stores a tile value into the array elements specified by indices.

`arange`	Creates a tile with value starting from 0 to size - 1.
`astile`	Creates a tile from a value.
`full`	Creates a tile filled with given value.
`ones`	Creates a tile filled with ones.
`zeros`	Creates a tile filled with zeros.

`cat`	Concatenates two tiles along the axis.
`broadcast_to`	Broadcasts a tile to the specified shape following Numpy broadcasting rule.
`expand_dims`	Reshapes the tile by inserting a new axis of size 1 at given position.
`reshape`	Reshapes a tile to the specified shape.
`permute`	Permutes the axes of the input tile.
`transpose`	Transposes two axes of the input tile with at least 2 dimensions.
`astype`	Converts a tile to the specified data type.
`bitcast`	Reinterpets tile as being of specified data type.
`pack_to_bytes`	Flattens a tile and reinterprets its raw bytes as uint8 elements.
`unpack_from_bytes`	Reinterprets a 1D uint8 byte tile as a 1D tile of the target data type.

`sum`	Performs sum reduction on tile along the axis.
`max`	Performs max reduction on tile along the axis.
`min`	Performs min reduction on tile along the axis.
`prod`	Performs prod reduction on tile along the axis.
`argmax`	Performs argmax reduction on tile along the axis.
`argmin`	Performs argmin reduction on tile along the axis.
`reduce`	Apply custom reduction function along axis.

`cumsum`	Performs cumsum on tile along the axis.
`cumprod`	Performs cumprod on tile along the axis.
`scan`	Apply custom scan (inclusive prefix) function along axis.

`mma`	Matrix multiply-accumulate.
`mma_scaled`	Block-scaled matrix multiply-accumulate.
`matmul`	Performs matrix multiply on the given tiles.

`where`	Returns elements chosen from x or y depending on condition.
`extract`	Extracts a smaller tile from input tile.

`add`	Elementwise add on two tiles.
`sub`	Elementwise sub on two tiles.
`mul`	Elementwise mul on two tiles.
`truediv`	Elementwise truediv on two tiles.
`floordiv`	Elementwise floordiv on two tiles.
`cdiv`	Computes ceil(x / y).
`pow`	Elementwise pow on two tiles.
`mod`	Elementwise mod on two tiles.
`minimum`	Elementwise minimum on two tiles.
`maximum`	Elementwise maximum on two tiles.
`negative`	Same as -x.
`abs`	Perform abs on a tile.
`isnan`	Perform isnan on a tile.
`exp`	Perform exp on a tile.
`exp2`	Perform exp2 on a tile.
`log`	Perform log on a tile.
`log2`	Perform log2 on a tile.
`sqrt`	Perform sqrt on a tile.
`rsqrt`	Perform rsqrt on a tile.
`sin`	Perform sin on a tile.
`cos`	Perform cos on a tile.
`tan`	Perform tan on a tile.
`sinh`	Perform sinh on a tile.
`cosh`	Perform cosh on a tile.
`tanh`	Perform tanh on a tile.
`floor`	Perform floor on a tile.
`ceil`	Perform ceil on a tile.

`bitwise_and`	Elementwise bitwise_and on two tiles.
`bitwise_or`	Elementwise bitwise_or on two tiles.
`bitwise_xor`	Elementwise bitwise_xor on two tiles.
`bitwise_lshift`	Elementwise bitwise_lshift on two tiles.
`bitwise_rshift`	Elementwise bitwise_rshift on two tiles.
`bitwise_not`	Elementwise bitwise not on a tile.

`greater`	Compare two tiles elementwise with >.
`greater_equal`	Compare two tiles elementwise with >=.
`less`	Compare two tiles elementwise with <.
`less_equal`	Compare two tiles elementwise with <=.
`equal`	Compare two tiles elementwise with ==.
`not_equal`	Compare two tiles elementwise with !=.

`atomic_cas`	Bulk atomic compare-and-swap on array elements with given indices.
`atomic_xchg`	Bulk atomic exchange of array elements at given indices.
`atomic_add`	Bulk atomic post-increment of array elements at given indices.
`atomic_max`	Bulk atomic maximum value assignment on array elements at given indices.
`atomic_min`	Bulk atomic minimum value assignment on array elements at given indices.
`atomic_and`	Bulk atomic AND operation on array elements at given indices.
`atomic_or`	Bulk atomic OR operation on array elements at given indices.
`atomic_xor`	Bulk atomic XOR operation on array elements at given indices.

`printf`	Print the values at runtime from the device
`print`	Print values at runtime from the device using Python-style syntax.
`assert_`	Assert that all elements of the given tile are True.

`static_assert`	Asserts that a condition is true at compile time.
`static_eval`	Evaluates the given Python expression at compile time.
`static_iter`	Iterates at compile time.

`Array`	Class for global array objects.
`TiledView`	Class for tiled view objects.
`Slice`

`RoundingMode`	Rounding mode for floating-point operations.
`PaddingMode`	Padding mode for load operation.

Searches the entire search space and return the best configuration.

`tune.TuningResult`	Holds the measurement result for each config.
`tune.Measurement`	Holds a configuration and its timing result.

`kernel.replace_hints`	Return a new kernel with updated compiler hints.
`compiler_timeout`	Context manager that temporarily sets the compiler timeout.

Launch a cuTile kernel from a JAX-traced graph.

`jax.OutputPlaceholder`	Represents an output buffer passed to cutile_call.
`jax.InputOutput`	Wraps an input buffer to alias an output buffer to be returned by cutile_call.