[llvm] [clang] [LinkerWrapper] Handle AMDGPU Target-IDs correctly when linking (PR #78359)

[Joseph Huber via cfe-commits]

================
@@ -162,6 +162,19 @@ class OffloadFile : public OwningBinary<OffloadBinary> {
               std::unique_ptr<MemoryBuffer> Buffer)
       : OwningBinary<OffloadBinary>(std::move(Binary), std::move(Buffer)) {}
 
+  /// Make a deep copy of this offloading file.
+  OffloadFile copy() const {
+    std::unique_ptr<MemoryBuffer> Buffer = MemoryBuffer::getMemBufferCopy(
+        getBinary()->getMemoryBufferRef().getBuffer());
+
+    // This parsing should never fail because it has already been parsed.
+    auto NewBinaryOrErr = OffloadBinary::create(*Buffer);
+    assert(NewBinaryOrErr && "Failed to parse a copy of the binary?");
+    if (!NewBinaryOrErr)
----------------
jhuber6 wrote:

Errors always need to be checked, even if they were successful. If the user did not compile with assertions on this would abort the program otherwise.

https://github.com/llvm/llvm-project/pull/78359