Method generate_build_options

src/core/CL/CLCompileContext.cpp:245–295

}

std::string CLCompileContext::generate_build_options(const StringSet   &build_options_set,
                                                     const std::string &kernel_path) const
{
    std::string concat_str;
    bool        ext_supported = false;
    std::string ext_buildopts;

#if defined(ARM_COMPUTE_DEBUG_ENABLED)
    // Enable debug properties in CL kernels
    concat_str += " -DARM_COMPUTE_DEBUG_ENABLED";
#endif // defined(ARM_COMPUTE_DEBUG_ENABLED)

    GPUTarget gpu_arch = get_arch_from_target(_device.target());
    concat_str +=
        " -DGPU_ARCH=" + support::cpp11::to_string(static_cast<std::underlying_type<GPUTarget>::type>(gpu_arch));

    if (_device.supported("cl_khr_fp16"))
    {
        concat_str += " -DARM_COMPUTE_OPENCL_FP16_ENABLED=1 ";
    }

    if (_device.supported("cl_arm_integer_dot_product_int8") || _device.supported("cl_khr_integer_dot_product"))
    {
        concat_str += " -DARM_COMPUTE_OPENCL_DOT8_ENABLED=1 ";
    }

    if (_device.supported("cl_arm_integer_dot_product_accumulate_int8"))
    {
        concat_str += " -DARM_COMPUTE_OPENCL_DOT8_ACC_ENABLED=1 ";
    }

    std::tie(ext_supported, ext_buildopts) = _device.is_non_uniform_workgroup_supported();

    if (ext_supported)
    {
        concat_str += ext_buildopts;
    }
    else
    {
        ARM_COMPUTE_ERROR("Non uniform workgroup size is not supported!!");
    }

    if (gpu_arch != GPUTarget::UNKNOWN && gpu_arch != GPUTarget::MIDGARD && get_ddk_version() >= 11)
    {
        concat_str += " -DUNROLL_WITH_PRAGMA ";
    }

    std::string build_options = stringify_set(build_options_set, kernel_path) + concat_str;

    return build_options;
}

bool CLCompileContext::fp16_supported() const
{