DPCT1006

Message

SYCL does not provide a standard API to differentiate between integrated and discrete GPU devices.

Detailed Help

The generated code uses a 0 (zero) as the value of this field.

Suggestions to Fix

Consider re-implementing the code that depends on this field.

For example, this original CUDA* code:

void foo() {
  int device_number = 0;
    cudaDeviceProp prop;
    cudaGetDeviceProperties(&prop, device_number);
    if (prop.integrated) {
      ...
    } else {
      ...
    }
}

results in the following migrated SYCL* code:

void foo() {
  int device_number = 0;
    dpct::device_info prop;
    dpct::dev_mgr::instance().get_device(device_number).get_device_info(prop);
    /*
    DPCT1006:0: SYCL does not provide a standard API to differentiate between
    integrated and discrete GPU devices.
    */
    if (prop.get_integrated()) {
      ...
    } else {
      ...
    }
  }

which is rewritten to:

void foo() {
  int device_number = 0;
    dpct::device_info prop;
    // 1. User can use device name or other info to select device. The `a_device_name` can be obtained by running command `   sycl-ls`
    // 2. User can use the environment variable `ONEAPI_DEVICE_SELECTOR` to filter out some devices.
    // 3. User can use the user-provided device-selector like:
    //   int prefer_my_device(const sycl::device& d) {
    //     return d.get_info<info::device::name>() == "MyDevice";
    //   }
    //   sycl::device preferred_device { prefer_my_device };
    if (dpct::dev_mgr::instance().get_device(device_numbe).get_info<sycl::info::device::name>() == "a_device_name") {
      ...
    } else {
      ...
    }
  }