[LegalizeTypes] Don't assume fshl/fshr shift amount type matches the other operands.

[craig.topper]

Summary

Like other shifts, the type isn't required to match. We shouldn't
assume we can call ZExtPromotedInteger.

I tested the PromoteIntOp_FunnelShift locally by removing the promotion
of the shift amount from PromoteIntRes_FunnelShift. But with the final
version of this patch it is never executed on any tests.

[spatel]

LGTM

llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

 //===----- LegalizeIntegerTypes.cpp - Legalization of integer types -------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
SDValue DAGTypeLegalizer::PromoteIntRes_Rotate(SDNode *N) {
   SDValue Res = TLI.expandROT(N, true /*AllowVectorOps*/, DAG);
   ReplaceValueWith(SDValue(N, 0), Res);
   return SDValue();
 }
 
 SDValue DAGTypeLegalizer::PromoteIntRes_FunnelShift(SDNode *N) {
   SDValue Hi = GetPromotedInteger(N->getOperand(0));
   SDValue Lo = GetPromotedInteger(N->getOperand(1));
-  SDValue Amt = ZExtPromotedInteger(N->getOperand(2));
+  SDValue Amt = N->getOperand(2);
+  if (getTypeAction(Amt.getValueType()) == TargetLowering::TypePromoteInteger)
+    Amt = ZExtPromotedInteger(Amt);
+  EVT AmtVT = Amt.getValueType();
 
   SDLoc DL(N);
   EVT OldVT = N->getOperand(0).getValueType();
   EVT VT = Lo.getValueType();
   unsigned Opcode = N->getOpcode();
   bool IsFSHR = Opcode == ISD::FSHR;
   unsigned OldBits = OldVT.getScalarSizeInBits();
   unsigned NewBits = VT.getScalarSizeInBits();
 
   // Amount has to be interpreted modulo the old bit width.
-  Amt = DAG.getNode(ISD::UREM, DL, VT, Amt, DAG.getConstant(OldBits, DL, VT));
+  Amt = DAG.getNode(ISD::UREM, DL, AmtVT, Amt,
+                    DAG.getConstant(OldBits, DL, AmtVT));
 
   // If the promoted type is twice the size (or more), then we use the
   // traditional funnel 'double' shift codegen. This isn't necessary if the
   // shift amount is constant.
   // fshl(x,y,z) -> (((aext(x) << bw) | zext(y)) << (z % bw)) >> bw.
   // fshr(x,y,z) -> (((aext(x) << bw) | zext(y)) >> (z % bw)).
   if (NewBits >= (2 * OldBits) && !isa<ConstantSDNode>(Amt) &&
       !TLI.isOperationLegalOrCustom(Opcode, VT)) {
     SDValue HiShift = DAG.getConstant(OldBits, DL, VT);
     Hi = DAG.getNode(ISD::SHL, DL, VT, Hi, HiShift);
     Lo = DAG.getZeroExtendInReg(Lo, DL, OldVT);
     SDValue Res = DAG.getNode(ISD::OR, DL, VT, Hi, Lo);
     Res = DAG.getNode(IsFSHR ? ISD::SRL : ISD::SHL, DL, VT, Res, Amt);
     if (!IsFSHR)
       Res = DAG.getNode(ISD::SRL, DL, VT, Res, HiShift);
     return Res;
   }
 
   // Shift Lo up to occupy the upper bits of the promoted type.
-  SDValue ShiftOffset = DAG.getConstant(NewBits - OldBits, DL, VT);
+  SDValue ShiftOffset = DAG.getConstant(NewBits - OldBits, DL, AmtVT);
   Lo = DAG.getNode(ISD::SHL, DL, VT, Lo, ShiftOffset);
 
   // Increase Amount to shift the result into the lower bits of the promoted
   // type.
   if (IsFSHR)
-    Amt = DAG.getNode(ISD::ADD, DL, VT, Amt, ShiftOffset);
+    Amt = DAG.getNode(ISD::ADD, DL, AmtVT, Amt, ShiftOffset);
 
   return DAG.getNode(Opcode, DL, VT, Hi, Lo, Amt);
 }
 
 SDValue DAGTypeLegalizer::PromoteIntRes_TRUNCATE(SDNode *N) {
   EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
   SDValue Res;
   SDValue InOp = N->getOperand(0);
bool DAGTypeLegalizer::PromoteIntegerOperand(SDNode *N, unsigned OpNo) {
   case ISD::INSERT_SUBVECTOR: Res = PromoteIntOp_INSERT_SUBVECTOR(N); break;
 
   case ISD::SHL:
   case ISD::SRA:
   case ISD::SRL:
   case ISD::ROTL:
   case ISD::ROTR: Res = PromoteIntOp_Shift(N); break;
 
+  case ISD::FSHL:
+  case ISD::FSHR: Res = PromoteIntOp_FunnelShift(N); break;
+
   case ISD::SADDO_CARRY:
   case ISD::SSUBO_CARRY:
   case ISD::ADDCARRY:
   case ISD::SUBCARRY: Res = PromoteIntOp_ADDSUBCARRY(N, OpNo); break;
 
   case ISD::FRAMEADDR:
   case ISD::RETURNADDR: Res = PromoteIntOp_FRAMERETURNADDR(N); break;
 
SDValue DAGTypeLegalizer::PromoteIntOp_SETCC(SDNode *N, unsigned OpNo) {
   return SDValue(DAG.UpdateNodeOperands(N, LHS, RHS, N->getOperand(2)), 0);
 }
 
 SDValue DAGTypeLegalizer::PromoteIntOp_Shift(SDNode *N) {
   return SDValue(DAG.UpdateNodeOperands(N, N->getOperand(0),
                                 ZExtPromotedInteger(N->getOperand(1))), 0);
 }
 
+SDValue DAGTypeLegalizer::PromoteIntOp_FunnelShift(SDNode *N) {
+  return SDValue(DAG.UpdateNodeOperands(N, N->getOperand(0), N->getOperand(1),
+                                ZExtPromotedInteger(N->getOperand(2))), 0);
+}
+
 SDValue DAGTypeLegalizer::PromoteIntOp_SIGN_EXTEND(SDNode *N) {
   SDValue Op = GetPromotedInteger(N->getOperand(0));
   SDLoc dl(N);
   Op = DAG.getNode(ISD::ANY_EXTEND, dl, N->getValueType(0), Op);
   return DAG.getNode(ISD::SIGN_EXTEND_INREG, dl, Op.getValueType(),
                      Op, DAG.getValueType(N->getOperand(0).getValueType()));
 }
 

llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h

 //===-- LegalizeTypes.h - DAG Type Legalizer class definition ---*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
private:
   SDValue PromoteIntOp_INSERT_SUBVECTOR(SDNode *N);
   SDValue PromoteIntOp_CONCAT_VECTORS(SDNode *N);
   SDValue PromoteIntOp_SCALAR_TO_VECTOR(SDNode *N);
   SDValue PromoteIntOp_SPLAT_VECTOR(SDNode *N);
   SDValue PromoteIntOp_SELECT(SDNode *N, unsigned OpNo);
   SDValue PromoteIntOp_SELECT_CC(SDNode *N, unsigned OpNo);
   SDValue PromoteIntOp_SETCC(SDNode *N, unsigned OpNo);
   SDValue PromoteIntOp_Shift(SDNode *N);
+  SDValue PromoteIntOp_FunnelShift(SDNode *N);
   SDValue PromoteIntOp_SIGN_EXTEND(SDNode *N);
   SDValue PromoteIntOp_SINT_TO_FP(SDNode *N);
   SDValue PromoteIntOp_STRICT_SINT_TO_FP(SDNode *N);
   SDValue PromoteIntOp_STORE(StoreSDNode *N, unsigned OpNo);
   SDValue PromoteIntOp_TRUNCATE(SDNode *N);
   SDValue PromoteIntOp_UINT_TO_FP(SDNode *N);
   SDValue PromoteIntOp_STRICT_UINT_TO_FP(SDNode *N);
   SDValue PromoteIntOp_ZERO_EXTEND(SDNode *N);