[SYCL][Doc] Update SYCL --offload-arch option.

[srividya-sundaram]

Adopt LLVM Clang's --offload-arch option for SYCL offloading to Intel GPUs, Intel CPUs, NVidia and AMD GPUs.

[maarquitos14]

I see we only have lists with the accepted value for --offload-arch for Intel devices, should we do the same for third party devices?

[srividya-sundaram]

I see we only have lists with the accepted value for --offload-arch for Intel devices, should we do the same for third party devices?

For SYCL offloading to Intel CPUs and GPUs, there is a mapping between the accepted values for --offload-arch and what is passed to the target-specific LLVM compiler (OCLOC or opencl-aot).
For NVidia GPUs and AMD GPUs, there is no such mapping. I can maybe add a list of accepted values for --offload-arch for NVidia and AMD GPUs.

[maarquitos14]

For NVidia GPUs and AMD GPUs, there is no such mapping. I can maybe add a list of accepted values for --offload-arch for NVidia and AMD GPUs.

I think that would be helpful for users.

[mdtoguchi]

@gmlueck, please take a look as you were part of the original conversation for these values.