[SYCL] Does sycl support CUDA Multi-Device System?

[malixian]

For example, I have two same GPUs . How can I specify different operators to execute on different GPUs in a sycl program? Like cudaSetDevice(int device_id) function.

[keryell]

The SYCL standard handles any number of accelerators through the concept of sycl::device and you can execute kernels on them through different sycl::queue you have to create.
You can use the concept of device selector to select the device you want or directly look at all the device list from sycl::device::get_devices()
Look for example at https://www.khronos.org/registry/SYCL/specs/sycl-2020-provisional.pdf for these classes or at some SYCL tutorials or book on https://sycl.tech/
The SYCL committee did not follow the path of old APIs likes cudaSetDevice() because this is not thread-safe to have so much hidden state handled by the current C++ std::thread.
Instead, we have a clearer and safer explicit API.

[malixian]

thx for @keryell, this question has been resolved.