[llvm] [offload] add support for aligned allocations (PR #203353)

[via llvm-commits]

================
@@ -600,20 +615,44 @@ struct CUDADeviceTy : public GenericDeviceTy {
     switch (Kind) {
     case TARGET_ALLOC_DEFAULT:
     case TARGET_ALLOC_DEVICE:
+      if (Alignment > 0 && Alignment > Granularity) {
+        return Plugin::error(
+            ErrorCode::UNSUPPORTED,
+            "too large alignment size for the device allocation");
+      }
+
----------------
EuphoricThinking wrote:

We can, so it's done. Outside the switch the error messages are less accurate, but I suppose that the user would know which type of the allocation they have requested.

https://github.com/llvm/llvm-project/pull/203353