[llvm] [NVPTX] Generalize and extend upsizing when lowering 8/16-bit-element vector loads/stores (PR #119622)

[Artem Belevich via llvm-commits]

================
@@ -4,7 +4,7 @@
 ; RUN: llc < %s -march=nvptx64 -mcpu=sm_70 -mattr=+ptx82 | FileCheck %s -check-prefixes=CHECK,SM70
 ; RUN: %if ptxas-12.2 %{ llc < %s -march=nvptx64 -mcpu=sm_70 -mattr=+ptx82 | %ptxas-verify -arch=sm_70 %}
 
-; TODO: add i1, <8 x i8>, and <6 x i8> vector tests.
+; TODO: add i1, and <6 x i8> vector tests.
----------------
Artem-B wrote:

Scalars vs vectors would be one factor.
If vectors test is still too large, it could be split by element type/size (e.g. native vectors vs the ones with small elements that get lowered as b32)

https://github.com/llvm/llvm-project/pull/119622