[llvm] [offload][SYCL] Add SYCL Module splitting. (PR #131347)

[Shilei Tian via llvm-commits]

shiltian wrote:

> > Yes, a user chooses a target like:
> > ```
> > clang++ -fsycl -fsycl-targets=nvidia_gpu_sm_90  code.cpp
> > or
> > clang++ -fsycl -fsycl-targets=amd_gpu_gfx1201 code.cpp
> > ```
> > 
> > 
> >     
> >       
> >     
> > 
> >       
> >     
> > 
> >     
> >   
> > The list of supported targets in downstream can be observed here: https://intel.github.io/llvm/UsersManual.html#generic-options
> 
> This syntax unnerves me. You can kind of do this in OpenMP but the syntax is really verbose.
> 
> ```
> clang++ -fopenmp -fopenmp-targets=amdgcn-amd-amdhsa,x86_64-unknown-linux-gnu -Xopenmp-target=x86_64-unknown-linux-gnu --offload-arch=znver5 -Xopenmp-target=amdgcn-amd-amdhsa --offload-arch=gfx940
> ```
> 
> I want to let `-Xarch_` handle this but that requires changing a flag which had some detractors. I also have #125556 to make that generic but it needs to be updated.

FWIW, it might be better to adopt the offload bundler id format:

```
<target triple>:<target id>[:<target feature>]
```

You probably want to have the full target triple here because the OS part does matter.

For example,

```
# AMDGPU

-fsycl-targets=amdgcn-amd-amdhsa-gfx1201

-fsycl-targets=amdgcn-amd-amdhsa-gfx1201:xnack+

#NVIDIA GPU

-fsycl-targets=nvptx64-nvidia-unknown-sm_90
```

This also applies to `--offload-arch` @jhuber6 has mentioned.

https://github.com/llvm/llvm-project/pull/131347