Get the start of the trip for the current block. :param mask_type: Type of mask to use :type mask_type: utils.MaskEnum

Returns true if the TensorRef is non-null

Check if the object implements the DynamicExpression protocol. When partial=True, returns True if any protocol method is present.

Create an integer constant with the given type and value.

Check if this arch is equal or higher in the same family than the given arch, so that the family-specific features can be used. Exam

Returns true if the argument is a power of 2

Get the type index from the index-th field of tvm_ffi_any_type struct. Semantics as follows: .. code-block:: c int32_t g

Loads a fragment from memory with additional logical offset

Loads a fragment from memory with additional logical offset

Loads a fragment from memory with additional logical offset

Loads a fragment from memory with additional logical offset

Loads a fragment from memory with additional logical offset

Loads a fragment from memory with additional logical offset

Specialized GEMM for circular-buffered A/B from SMEM. Performs D <- A * B + C where A and B are described by circular SMEM descriptor

Element-wise minimum of 2 CuTe tensors

Logical Not

Gets the index of the problem

Acquire a producer stage and return its stage token/index. When `token` is provided, reuse the preceding `producer_try_acquire()` res

Combined producer commit + advance with automatic elect_one using internal state.

Execute a persistent batched dense blockscaled GEMM operation on Blackwell architecture with performance benchmarking. This function prepares inp

Overrides the internal iteration index

Sets the predicate mask, overriding value stored in predicate iterator

Stores a fragment to memory at the location pointed to by the iterator

Compute element-wise hyperbolic tangent of the input tensor. :param a: Input tensor :type a: Union[TensorSSA, Numeric] :param fastmath: E

Convert a raw TMA descriptor gmem pointer to the type expected by cute.copy. cute.copy requires the tma_desc_ptr to be in generic address sp

Thread-fragment scale factor A tensor for SM120 block-scaled MMA. Implements the ThrFrg partitioning for scale factor A according to the corr

Thread-fragment scale factor B tensor for SM120 block-scaled MMA. Implements the ThrFrg partitioning for scale factor B according to the corr

Returns arguments for the transposed problem

< Layout function

General atomic operation function. Atomically adds `val` to the value at memory location `ptr` and returns the old value. :param ptr: P

Add a node to dag ir

Adds a pointer offset in units of Element

Get the address of a global symbol.

Signal completion of Stage and move to the next stage (group_id) signals to (group_id+1)

Create an unconditional branch. Parameters ---------- target_block : ir.Block The target block to branch to.

Clears the predicate set efficiently

Clears the predicate set efficiently

Compiles the module by invoking the pipeline.

Compute token offset and count for a given expert from the cumsum offs tensor. :param offs: Cumulative sum tensor of token counts per expert

Create a Leaf node for a given value. Args: x: The value to create a leaf for is_numeric: Whether this is a numeric value

Returns a pointer to the shared memory buffer

Computes the determinant of a 2-by-2 matrix

Divides dividend by the cluster size

Divide the input tensor by given tiler and organize the resulting layout to 2 modes. The first mode is the tile mode and the second m

Clears the predicate set efficiently

Clears the predicate set efficiently

Clears the predicate set efficiently

Compute element-wise base-2 exponential of the input tensor. :param a: Input tensor :type a: Union[TensorSSA, Numeric] :param fastmath: E

Generate all cluster shapes for SM100 at or above the given level. Args: level: The global level to generate cluster shapes for.

Returns a pointer

Get the field name with tuple indices for shape/stride access. :param field_suffix: The field suffix (e.g., ".shape", ".strides"). :t

Get the fully qualified class name of an object. Args: x: Any object Returns: str: Fully qualified class name in format

Calculates optimal swizzle width

Gets the mask

Gets the mask

Returns a stride

Determine the number of gemm_k iterations for conv2d problem using implicit gemm algorithm