[Mlir-commits] [mlir] [mlir] Decouple NVPTX target from CUDA toolkit presence (PR #93008)

[Oleksandr Alex Zinenko]

================
@@ -167,6 +167,15 @@ class NVPTXSerializer : public SerializeGPUModuleBase {
 
   gpu::GPUModuleOp getOperation();
 
+  std::optional<SmallVector<char, 0>>
----------------
ftynse wrote:

I see that you are only moving this, but consider adding some documentation if you understand what this does.

https://github.com/llvm/llvm-project/pull/93008