Method configure

src/cpu/kernels/CpuGemmMatrixMultiplyKernel.cpp:143–192

} // namespace

void CpuGemmMatrixMultiplyKernel::configure(const ITensorInfo     *lhs,
                                            const ITensorInfo     *rhs,
                                            ITensorInfo           *dst,
                                            float                  alpha,
                                            bool                   is_interleaved,
                                            const GEMMReshapeInfo &reshape_info)
{
    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU,
                            "CpuGemmMatrixMultiplyKernel::configure");
    ARM_COMPUTE_ERROR_ON_NULLPTR(lhs, rhs, dst);

    // dst tensor auto inizialitation if not yet initialized
    TensorShape tensor_shape{lhs->tensor_shape()};
    tensor_shape.set(0, is_interleaved ? reshape_info.n() : rhs->dimension(0));
    tensor_shape.set(1, is_interleaved ? reshape_info.m() : lhs->dimension(1));

    auto_init_if_empty(*dst, lhs->clone()->set_tensor_shape(tensor_shape));

    // Perform validate step
    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(lhs, rhs, dst, alpha, is_interleaved, reshape_info));

    _alpha = alpha;

    // Configure kernel window
    Window win{};

    // Check if the dst tensor is a vector. If so,the kernel runs the vector-matrix multiplication
    const bool is_dst_vector = (dst->dimension(1) == 1);
    if (is_dst_vector)
    {
        const unsigned int num_elems_processed_per_iteration_x = (lhs->data_type() == DataType::F32) ? 16 : 32;

        win = calculate_max_window(*dst, Steps(num_elems_processed_per_iteration_x));
    }
    else
    {
        constexpr unsigned int num_elems_processed_per_iteration_x = 8;
        constexpr unsigned int num_elems_processed_per_iteration_y = 4;

        win =
            calculate_max_window(*dst, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
    }

    const auto uk = CpuGemmMatrixMultiplyKernel::get_implementation(
        DataTypeISASelectorData{lhs->data_type(), CPUInfo::get().get_isa()});
    ARM_COMPUTE_ERROR_ON_NULLPTR(uk);
    _func = uk->ukernel;

    ICPPKernel::configure(win);
}

Status CpuGemmMatrixMultiplyKernel::validate(const ITensorInfo     *lhs,
                                             const ITensorInfo     *rhs,