[clang] [llvm] [Clang][SYCL] Compile device code once per IGCA target (PR #239010)

Yury Plyakhin via cfe-commits cfe-commits at lists.llvm.org
Thu Nov 12 18:10:30 PST 2026


================
@@ -37,6 +37,7 @@ USAGE: clang-sycl-linker [options] <input bitcode files>
 
 OPTIONS:
   --arch <value>                Specify the name of the target architecture.
+  --device=<value>              An Intel GPU to compile ahead of time for. Can be repeated.
----------------
YuriPlyakhin wrote:

1. I'd prefer `--code` over `--device`. It matches the code image metadata it is derived from, the `--offload-code` option in the IGCA design, and the established arch/code (virtual/real) terminology. "device" is already overloaded in these tools.
2. Can we avoid limiting description to Intel GPUs here?
3. Please, make sure the option documentation here matches `--help`.

I'd suggest this description: 
```
Specify a real device to compile ahead of time for. Can be repeated. Without it, a real --arch is compiled ahead of time and a virtual one is kept for just-in-time compilation.
```

https://github.com/llvm/llvm-project/pull/239010


More information about the cfe-commits mailing list