arch Class — pytorch Architecture

Source Code

aten/src/ATen/native/cuda/cutlass_extensions/gemm/kernel/default_fpA_intB_traits.h lines 21–35

template<typename arch>
struct MixedGemmArchTraits<float, float, arch> {
    static constexpr int Stages = 2;
    using OperatorClass         = cutlass::arch::OpClassSimt;
    using AccType               = float;
    using LayoutB               = cutlass::layout::RowMajor;

    static constexpr int ElementsPerAccessA = 1;
    static constexpr int ElementsPerAccessB = 1;
    static constexpr int ElementsPerAccessC = 1;
    static constexpr int ThreadblockK       = 8;
    using InstructionShape                  = cutlass::gemm::GemmShape<1, 1, 1>;

    using Operator = cutlass::arch::OpMultiplyAdd;
};