[llvm] [AMDGPU] selecting v_sat_pk instruction, version 2 (PR #123297)

[Matt Arsenault via llvm-commits]

================
@@ -1975,6 +1982,10 @@ bool SITargetLowering::isTypeDesirableForOp(unsigned Op, EVT VT) const {
   if (VT == MVT::i1 && Op == ISD::SETCC)
     return false;
 
+  // v2i8 is illegal and only allowed in specific cases
+  if (VT == MVT::v2i8 && Op == ISD::TRUNCATE_SSAT_U)
----------------
arsenm wrote:

The combiner checks isOperationLegalOrCustom though


Plus this wouldn't be a check specific to this vector size. We would want wider vectors split into pieces 

https://github.com/llvm/llvm-project/pull/123297