Operand Class — pytorch Architecture

Source Code

aten/src/ATen/native/cuda/cutlass_extensions/gemm/warp/mma_tensorop_dequantizer.h lines 68–83

template<
    /// Matrix multiply operator
    typename MmaOperator_,
    /// Size of the matrix to load (concept: MatrixShape)
    typename Shape_,
    /// Operand identity
    Operand Operand,
    /// Data type of Scale elements
    typename Element_,
    /// Layout of operand
    typename Layout_,
    /// Number of threads participating in one matrix operation
    int Threads,
    ///
    typename Enable = void>
class MmaTensorOpDequantizer;