[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 18:10:31 PST 2026


================
@@ -108,6 +109,14 @@ generate the final executable.
 clang-sycl-linker --triple spirv64 --arch xe-bmg-g21 input.bc
 ```
 
+For an IGCA target, such as `igca_30r`, the device image is compiled ahead of
+time for the Intel GPUs given with `--device`, with a single ocloc call, or just
+in time if no `--device` is given:
----------------
YuriPlyakhin wrote:

Please provide some description first and it should be a separate clause in examples section.

Something along those lines:
```suggestion
### Specifying virtual architecture and real device
You can specify virtual architecture ....

E.g., for an IGCA target, such as `igca_30r`, the device image is compiled ahead of
time for the Intel GPUs given with `--code`, with a single ocloc call, or,
if no `--code` is given, kept as SPIR-V to be compiled just in time by the runtime:
```

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


More information about the cfe-commits mailing list