Method configure

src/gpu/cl/kernels/ClGemmReshapeLhsMatrixKernel.cpp:127–191

}

void ClGemmReshapeLhsMatrixKernel::configure(const CLCompileContext  &compile_context,
                                             ITensorInfo             *src,
                                             ITensorInfo             *dst,
                                             const GEMMLHSMatrixInfo &lhs_info,
                                             bool                     reinterpret_input_as_3d)
{
    ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);

    // Perform validate step
    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst, lhs_info, reinterpret_input_as_3d));

    auto padding_info = get_padding_info({src});

    const unsigned int src_w      = src->dimension(0);
    const unsigned int m          = reinterpret_input_as_3d ? src->dimension(1) * src->dimension(2) : src->dimension(1);
    const unsigned int partial_m0 = m % lhs_info.m0;
    const unsigned int partial_k0 = src_w % lhs_info.k0;

    // Create build options
    CLBuildOptions build_opts;
    build_opts.add_option("-DM0=" + support::cpp11::to_string(lhs_info.m0));
    build_opts.add_option("-DK0=" + support::cpp11::to_string(lhs_info.k0));
    build_opts.add_option_if(lhs_info.interleave, "-DINTERLEAVE");
    build_opts.add_option_if_else(lhs_info.transpose, "-DRESHAPE_LHS_T", "-DRESHAPE_LHS_NT");
    build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(src->element_size()));
    build_opts.add_option("-DPARTIAL_M0=" + support::cpp11::to_string(partial_m0));
    build_opts.add_option("-DPARTIAL_K0=" + support::cpp11::to_string(partial_k0));

    std::string kernel_name("gemm_reshape_lhs_matrix_");
    kernel_name += lhs_info.transpose ? "t" : "nt";

    // Create kernel
    _kernel = create_kernel(compile_context, kernel_name, build_opts.options());

    // Configure kernel window
    auto win_config = configure_window(src, dst, lhs_info, reinterpret_input_as_3d);
    ICLKernel::configure_internal(win_config);

    unsigned int idx = 2 * num_arguments_per_3d_tensor_nhw();
    _kernel.setArg<cl_int>(idx++, m);
    _kernel.setArg<cl_int>(idx++, lhs_info.v0);

    // Set config_id for enabling LWS tuning
    _config_id = "gemm_reshape_lhs_matrix_";
    _config_id += (reinterpret_input_as_3d ? "3d_" : "");
    _config_id += lower_string(string_from_data_type(src->data_type()));
    _config_id += "_";
    _config_id += support::cpp11::to_string(dst->dimension(0));
    _config_id += "_";
    _config_id += support::cpp11::to_string(dst->dimension(1));
    _config_id += "_";
    _config_id += support::cpp11::to_string(dst->dimension(2));
    _config_id += "_";
    _config_id += support::cpp11::to_string(lhs_info.m0);
    _config_id += "_";
    _config_id += support::cpp11::to_string(lhs_info.k0);
    _config_id += "_";
    _config_id += support::cpp11::to_string(lhs_info.v0);