[clang] [CUDA] Add device-side kernel launch support (PR #165519)

[Joseph Huber via cfe-commits]

================
@@ -525,6 +564,11 @@ Expected<SmallVector<StringRef>> getInput(const ArgList &Args) {
           object::Archive::create(Buffer);
       if (!LibFile)
         return LibFile.takeError();
+      // Skip extracting archives with fat binaries. Forward them to nvlink.
+      if (hasFatBinary(**LibFile)) {
+        ForwardArchives.emplace_back(Args.MakeArgString(*Filename));
----------------
jhuber6 wrote:

Right, I guess this is kind of complicated by the fact that I need to do my own static library resolution. What are the chances we can just extract the object file from the fatbinary ourselves and handle it that way? Presumably nvlink doesn't actually do anything with the host sections, but I've never used this much before.

https://github.com/llvm/llvm-project/pull/165519