[llvm] Handle VECREDUCE intrinsics in NVPTX backend (PR #136253)

[Princeton Ferro via llvm-commits]

================
@@ -36,17 +36,19 @@ class FeaturePTX<int version>:
 
 foreach sm = [20, 21, 30, 32, 35, 37, 50, 52, 53,
               60, 61, 62, 70, 72, 75, 80, 86, 87,
-              89, 90, 100, 101, 120] in
----------------
Prince781 wrote:

@Artem-B Should I also move the `max3`/`min3` variants to the same separate patch or can I keep them here?

https://github.com/llvm/llvm-project/pull/136253