[AMDGPU] should expand ROTL i16 to shifts.

[alex-t]

Summary

Instruction combining pass turns library rotl implementation to llvm.fshl.i16.
In the selection dag the intrinsic is turned to ISD::ROTL node that cannot be selected.
Need to expand it to shifts again.

[rampitec]

Needs test.

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

548

What about rotr?

[arsenm]

Needs test

[alex-t]

Tests added. ROTR case added.

[rampitec]

LGTM

[arsenm]

Should go into existing fshr and fshr tests

[alex-t]

tests moved to existing rotl/rotr tests

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

if (Subtarget->has16BitInsts()) {
     setOperationAction(ISD::SMAX, MVT::i16, Legal);
 
     setOperationAction(ISD::UMIN, MVT::i16, Legal);
     setOperationAction(ISD::UMAX, MVT::i16, Legal);
 
     setOperationAction(ISD::SIGN_EXTEND, MVT::i16, Promote);
     AddPromotedToType(ISD::SIGN_EXTEND, MVT::i16, MVT::i32);
 
     setOperationAction(ISD::ROTR, MVT::i16, Promote);
-    setOperationAction(ISD::ROTL, MVT::i16, Promote);
+    setOperationAction(ISD::ROTL, MVT::i16, Expand);
 
     setOperationAction(ISD::SDIV, MVT::i16, Promote);
     setOperationAction(ISD::UDIV, MVT::i16, Promote);
     setOperationAction(ISD::SREM, MVT::i16, Promote);
     setOperationAction(ISD::UREM, MVT::i16, Promote);
     setOperationAction(ISD::UADDSAT, MVT::i16, Legal);
     setOperationAction(ISD::USUBSAT, MVT::i16, Legal);