Replaces the given op with the contents of the given single-block region, using the operands of the block terminator to replace operation results.

Run a command and return the result

If the given value is an attribute, return it. Otherwise, turn it into a FloatAttr or IntegerAttr, depending on the given type.

Get element bit width matching C++ getTileSizeInBits logic.

Infer MLIR type from a Python value (int, float, bool).

Initialize MMA configurations using automatic subclass discovery

Initialize MMA Scaled configurations using automatic subclass discovery

Helper function for reduce and scan operations. Prepares common components such as element type handling and attribute creation. Args:

This method adds a debug info attribute to an operation.

Float comparison operation.

Integer comparison operation.

Create a copy of the loop with new bounds & partial copy of if-blocks

Returns a new file attribute based on the given file location.

Helper function to insert SelectOp for given cond & values

Helper function to insert XOrIOp with tile of ones

Extracts the attribute tag name from a TableGen definition name.

Finalize parsing by resolving operands and setting attributes.

Signed integer floor division operation.

Helper function to generate common attribute serialization logic.

Returns the bytecode file version.

This method gets or creates an index for an operation.

Reads an OpcodeRange definition from TableGen records. Returns {minValue, maxValue} or nullopt if not defined.

Helper function to extract names from a trait's "values" field

Gets or creates an index for a type in the type table.

Convert an enum value to the corresponding AtomicRMWModeAttr. Args: mode: AtomicRMWMode enum value context: Optional MLIR co

Convert an enum value to the corresponding ComparisonPredicateAttr.

Convert an enum value to the corresponding PaddingValueAttr.

Check if parameter is a Type parameter (Generic or Specific).

Check if version string is >= 13.3 (unified bitfield format). Types introduced at 13.3+ always use unified format.

Check if the given MLIR types match this configuration

Performs element-wise multiplication of two tiles.

----------------------------------------------------------------------------- 2) optimize CUDA Tile IR module - shared optimization pass with CAPI ---

Performs element-wise, bit-wise "or" of two tiles.

Parses function argument list with variadic support.

Parses and validates the bytecode header, including the magic number and version.

Parses memory ordering semantics and scope attributes for token-ordered operations

Parses an optional comma followed by a dimMap, returning the identity dimMap if absent.

Parses the section header from the bytecode.

Parse optimization pipeline from text

Parse common elements for LoadViewTkoOp and StoreViewTkoOp: - Optional token - Optional optimization_hints - Optional attribute dict Returns the parse

Raises lhs to the power of rhs element-wise.

Prints shape and element type in "8x16xf32" syntax.

Performs element-wise remainder of two tiles.

Replace TableGen placeholders with C++ code.

Performs element-wise subtraction of two tiles.

Validate allow_tma parameter and return the value if valid. Returns nullopt with error message if invalid (when key is provided). Pass empty key from

Validate latency parameter and return the value if valid. Returns nullopt with error message if invalid (when key is provided). Pass empty key from ge

Validate num_cta_in_cga parameter and return the value if valid. Returns nullopt with error message if invalid (when key is provided). Pass empty key

Validate occupancy parameter and return the value if valid. Returns nullopt with error message if invalid (when key is provided). Pass empty key from

Common verification logic for operations with aggregation semantics (Reduce, Scan, etc.)

Common verification logic for operations with aggregation semantics (Reduce, Scan, etc.)

Verifies requirements that are shared between all grid-like views.

Helper function to serialize the APFloat representation of a FloatAttr.