Computes the grid shape

Get all ir.Values from this leaf (may be multiple for DynamicExpression).

Check if an ``ir.Value`` matches this parameterized vector type.

Get the v_ptr from the index-th field of tvm_ffi_any_type struct. Semantics as follows: .. code-block:: c void* get_v_p

Loads a fragment from memory

Loads a fragment

Creates a tensor in register memory with the specified layout/shape and data type. This function allocates a tensor in register memory (rmem) usu

Returns a TensorRef to the B operand

Construct A Single Stage's Accumulator Shape

After reduce call, before smem async fence. Smem stores usually performed here. Upon exit, all smem stores for TMA must have been issued

Issue Tma Descriptor Prefetch -- ideally from a single thread for best performance

Before visit callback. Smem broadcasts usually performed here. Upon entry, all producer loads for this subtile are completed and visible.

The full procedural name indicates architecture, extended name, tile size, and layout.

Accesses raw internal state

Perform reduce on selected modes with given predefined reduction op. :param op: The reduction operator to use (operator.add or opera

Executes one test

Stores a fragment

A helper function to get dynamic SFA/SFB layout by filling dynamic A/B shape to the scale factor atom layout. :param Shape: The shape of the

Indicate arrival of warp issuing TMA_STORE

Pack work tile info fields into an rmem tensor of shape (4,) for vectorized smem copy.

Get the scale value. :return: The scale value

Wraps the input into a tuple if not a tuple.

Create zero or more CUTLASS 3 two-dimensional convolution operators. Create a CUTLASS 3 two-dimensional convolution operator for all feasible

Default ctor

Load gemm problem (m,n,k,l) for the specified group from global memory to register. :param problem_shape_mnkl: Tensor in global memo

Helper function to normalize pointer types to MLIR ir.Value. Supports: - ir.Value (LLVM pointer): returned as-is - cute.ptr (_Pointe

Reset the alignment of the current epilogue functor based on alignment C

Resets the kernel options based on cc :param cc: compute capability to reset to :type cc: int

Adds a pointer offset in units of Element

Adds a pointer offset in units of Element

Adds a pointer offset in units of Element

Adds a tile offset

Advances an iterator along logical dimensions of matrix in units of whole tiles

Return the round-up offset up to the next multiple of align.

Create a logical AND operation between two values.

Logical OR operation for any element in an iterable. Returns True if any element in the iterable is truthy, otherwise False. This is the DSL

Determines whether the GEMM can execute the given problem.

Check if the current work index is valid for the given query sequence length. This method verifies that the current work tile index

Clear the cache.

< Efficiently disables all accesses guarded by mask

Constructs a ``cutlass_cppgen.backend.Conv2dOperation`` based on the input parameters and current kernel specification of the ``Conv2

Returns whether the current work_tile_info passed in should continue to be used. This occurs only in the stream-K decomposition with stream-K work uni

Context manager for entering a new dynamic control-flow scope. This scope rule diverge from Python's local scope, variables defined