Method compile

python/cutlass_cppgen/op/gemm.py:478–507

Emits and compiles the kernel currently specified. If ``tile_description`` and any of the ``alignment`` parameters are set, the kernel will be chosen using this tile description and alignments. Otherwise, a default tile description and alignment will be used.

(self, tile_description: TileDescription = None,
                alignment_A: int = None, alignment_B: int = None, alignment_C: int = None,
                print_module: bool = False)

        return operation

    def compile(self, tile_description: TileDescription = None,
                alignment_A: int = None, alignment_B: int = None, alignment_C: int = None,
                print_module: bool = False) -> cutlass_cppgen.backend.GemmOperationUniversal:
        """
        Emits and compiles the kernel currently specified. If ``tile_description`` and any
        of the ``alignment`` parameters are set, the kernel will be chosen using this
        tile description and alignments. Otherwise, a default tile description and alignment
        will be used.

        :param tile_description: tile description specifying shapes and operand types to use in the kernel
        :type tile_description: cutlass_cppgen.backend.TileDescription
        :param alignment_A: alignment of operand A
        :type alignment_A: int
        :param alignment_B: alignment of operand B
        :type alignment_B: int
        :param alignment_C: alignment of operand C
        :type alignment_C: int
        :param print_module: whether to print the emitted C++ code
        :type print_module: bool

        :return: operation that was compiled
        :rtype: cutlass_cppgen.backend.GemmOperationUniversal
        """
        self.operation = self.construct(tile_description, alignment_A, alignment_B, alignment_C)

        if print_module:
            print(self.operation.rt_module.emit())

        compiler.add_module([self.operation,])
        return self.operation

    def _verify_rank(self, tensor):
        """