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

Srividya Sundaram via cfe-commits cfe-commits at lists.llvm.org
Mon Nov 9 18:06:11 PST 2026


================
@@ -1480,16 +1595,20 @@ getDeviceInput(const ArgList &Args) {
     if (!Binary.getBinary())
       continue;
 
-    SmallVector<OffloadFile::TargetID> CompatibleTargets = {Binary};
+    // A generic member can still be linked into any compatible image; other
+    // members only into images with the same "code".
+    ImageID Target = GetImageID(Binary);
+    bool IsGeneric = Target.first.second == "generic";
+    SmallVector<ImageID> CompatibleTargets = {Target};
     for (const auto &[ID, Input] : InputFiles)
-      if (OffloadFile::TargetID(Binary) != ID &&
-          object::areTargetsCompatible(Binary, ID))
+      if (Target != ID && (IsGeneric || Target.second == ID.second) &&
+          object::areTargetsCompatible(Target.first, ID.first))
----------------
srividya-sundaram wrote:

Yes, the archive member is extracted, since both GPUs map to igca_30r. 
But device images are grouped by the same arch and code, and the codes differ here, so the library's device code is linked separately from the application's:


```
clang-linker-wrapper app.o libl.a
  |-- arch=igca_30r, code=xe-mtl-h,xe-mtl-u
  |     app.o --> clang --sycl-link --> clang-sycl-linker --> ocloc -device xe-mtl-h,xe-mtl-u --> offload binary 1
  |-- arch=igca_30r, code=xe-mtl-u
  |     lib.o --> clang --sycl-link --> clang-sycl-linker --> ocloc -device xe-mtl-u  --> offload binary 2
  |
  --> offload binary 1 + offload binary 2 merged into one offload binary --> one wrapper object
```

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


More information about the cfe-commits mailing list