Index: lib/Target/RISCV/RISCVISelLowering.h
===================================================================
--- lib/Target/RISCV/RISCVISelLowering.h
+++ lib/Target/RISCV/RISCVISelLowering.h
@@ -31,7 +31,10 @@
   SELECT_CC,
   BuildPairF64,
   SplitF64,
-  TAIL
+  TAIL,
+  SHLW,
+  SRAW,
+  SRLW
 };
 }
 
@@ -57,6 +60,8 @@
 
   // Provide custom lowering hooks for some operations.
   SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const override;
+  void ReplaceNodeResults(SDNode *N, SmallVectorImpl<SDValue> &Results,
+                          SelectionDAG &DAG) const override;
 
   SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const override;
 
Index: lib/Target/RISCV/RISCVISelLowering.cpp
===================================================================
--- lib/Target/RISCV/RISCVISelLowering.cpp
+++ lib/Target/RISCV/RISCVISelLowering.cpp
@@ -80,10 +80,10 @@
     setOperationAction(ISD::SIGN_EXTEND_INREG, VT, Expand);
 
   if (Subtarget.is64Bit()) {
-    setTargetDAGCombine(ISD::SHL);
-    setTargetDAGCombine(ISD::SRL);
-    setTargetDAGCombine(ISD::SRA);
     setTargetDAGCombine(ISD::ANY_EXTEND);
+    setOperationAction(ISD::SHL, MVT::i32, Custom);
+    setOperationAction(ISD::SRA, MVT::i32, Custom);
+    setOperationAction(ISD::SRL, MVT::i32, Custom);
   }
 
   if (!Subtarget.hasStdExtM()) {
@@ -512,15 +512,53 @@
   return DAG.getCopyFromReg(DAG.getEntryNode(), DL, Reg, XLenVT);
 }
 
-// Return true if the given node is a shift with a non-constant shift amount.
-static bool isVariableShift(SDValue Val) {
-  switch (Val.getOpcode()) {
+// Returns the opcode of the target-specific SDNode that implements the 32-bit
+// form of the given shift Opcode.
+static RISCVISD::NodeType getRISCVShiftWOpcode(unsigned Opcode) {
+  switch (Opcode) {
   default:
-    return false;
+    llvm_unreachable("Unexpected opcode");
   case ISD::SHL:
+    return RISCVISD::SHLW;
   case ISD::SRA:
+    return RISCVISD::SRAW;
   case ISD::SRL:
-    return Val.getOperand(1).getOpcode() != ISD::Constant;
+    return RISCVISD::SRLW;
+  }
+}
+
+void RISCVTargetLowering::ReplaceNodeResults(SDNode *N,
+                                             SmallVectorImpl<SDValue> &Results,
+                                             SelectionDAG &DAG) const {
+  SDLoc DL(N);
+  switch (N->getOpcode()) {
+  default:
+    llvm_unreachable("Do not know how to custom type legalize this operation!");
+  case ISD::SHL:
+  case ISD::SRA:
+  case ISD::SRL: {
+    // Custom legalize 32-bit variable shifts to target-specific nodes on RV64.
+    // Because i32 isn't a legal type for RV64 they would otherwise be
+    // expanded to i64 operations, making it very difficult to select
+    // SLLW/SRLW/SRAW as the fact the shifts were originally 32-bit is lost.
+    assert(N->getValueType(0) == MVT::i32 && Subtarget.is64Bit() &&
+           "Unexpected custom legalisation");
+    if (N->getOperand(1).getOpcode() == ISD::Constant)
+      return;
+    SDValue NewOp0 =
+        DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(0));
+    SDValue NewOp1 =
+        DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(1));
+    SDValue NewShift = DAG.getNode(getRISCVShiftWOpcode(N->getOpcode()), DL,
+                                   MVT::i64, NewOp0, NewOp1);
+    // We need to replace the i32 shift with another i32 node. Replacing with
+    // an i64 node doesn't trigger an assert, but does lead to problems in the
+    // case that the result was extended. This is because the DAG combiner
+    // assumes that ty->ty no-op extends don't exist due to the folding logic
+    // in DAG.getNode.
+    Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, NewShift));
+    break;
+  }
   }
 }
 
@@ -545,34 +583,14 @@
   switch (N->getOpcode()) {
   default:
     break;
-  case ISD::SHL:
-  case ISD::SRL:
-  case ISD::SRA: {
-    assert(Subtarget.getXLen() == 64 && "Combine should be 64-bit only");
-    if (!DCI.isBeforeLegalize())
-      break;
-    SDValue RHS = N->getOperand(1);
-    if (N->getValueType(0) != MVT::i32 || RHS->getOpcode() == ISD::Constant ||
-        (RHS->getOpcode() == ISD::AssertZext &&
-         cast<VTSDNode>(RHS->getOperand(1))->getVT().getSizeInBits() <= 5))
-      break;
-    SDValue LHS = N->getOperand(0);
-    SDLoc DL(N);
-    SDValue NewRHS =
-        DAG.getNode(ISD::AssertZext, DL, RHS.getValueType(), RHS,
-                    DAG.getValueType(EVT::getIntegerVT(*DAG.getContext(), 5)));
-    return DCI.CombineTo(
-        N, DAG.getNode(N->getOpcode(), DL, LHS.getValueType(), LHS, NewRHS));
-  }
   case ISD::ANY_EXTEND: {
-    // If any-extending an i32 variable-length shift or sdiv/udiv/urem to i64,
-    // then instead sign-extend in order to increase the chance of being able
-    // to select the sllw/srlw/sraw/divw/divuw/remuw instructions.
+    // If any-extending an i32 sdiv/udiv/urem to i64, then instead sign-extend
+    // in order to increase the chance of being able to select the
+    // divw/divuw/remuw instructions.
     SDValue Src = N->getOperand(0);
     if (N->getValueType(0) != MVT::i64 || Src.getValueType() != MVT::i32)
       break;
-    if (!isVariableShift(Src) &&
-        !(Subtarget.hasStdExtM() && isVariableSDivUDivURem(Src)))
+    if (!(Subtarget.hasStdExtM() && isVariableSDivUDivURem(Src)))
       break;
     SDLoc DL(N);
     // Don't add the new node to the DAGCombiner worklist, in order to avoid
@@ -1682,6 +1700,12 @@
     return "RISCVISD::SplitF64";
   case RISCVISD::TAIL:
     return "RISCVISD::TAIL";
+  case RISCVISD::SHLW:
+    return "RISCVISD::SHLW";
+  case RISCVISD::SRAW:
+    return "RISCVISD::SRAW";
+  case RISCVISD::SRLW:
+    return "RISCVISD::SRLW";
   }
   return nullptr;
 }
Index: lib/Target/RISCV/RISCVInstrInfo.td
===================================================================
--- lib/Target/RISCV/RISCVInstrInfo.td
+++ lib/Target/RISCV/RISCVInstrInfo.td
@@ -51,6 +51,9 @@
 def riscv_tail      : SDNode<"RISCVISD::TAIL", SDT_RISCVCall,
                              [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
                               SDNPVariadic]>;
+def riscv_shlw      : SDNode<"RISCVISD::SHLW", SDTIntShiftOp>;
+def riscv_sraw      : SDNode<"RISCVISD::SRAW", SDTIntShiftOp>;
+def riscv_srlw      : SDNode<"RISCVISD::SRLW", SDTIntShiftOp>;
 
 //===----------------------------------------------------------------------===//
 // Operand and SDNode transformation definitions.
@@ -667,21 +670,14 @@
 def assertzexti32 : PatFrag<(ops node:$src), (assertzext node:$src), [{
   return cast<VTSDNode>(N->getOperand(1))->getVT() == MVT::i32;
 }]>;
-def assertzexti5 : PatFrag<(ops node:$src), (assertzext node:$src), [{
-  return cast<VTSDNode>(N->getOperand(1))->getVT().getSizeInBits() <= 5;
-}]>;
 def zexti32 : PatFrags<(ops node:$src),
                        [(and node:$src, 0xffffffff),
                         (assertzexti32 node:$src)]>;
-// Defines a legal mask for (assertzexti5 (and src, mask)) to be combinable
-// with a shiftw operation. The mask mustn't modify the lower 5 bits or the
-// upper 32 bits.
+// Defines a legal mask for (and src, mask) to be combinable with a shiftw
+// operation. The mask mustn't modify the lower 5 bits or the upper 32 bits.
 def shiftwamt_mask : ImmLeaf<XLenVT, [{
   return countTrailingOnes<uint64_t>(Imm) >= 5 && isUInt<32>(Imm);
 }]>;
-def shiftwamt : PatFrags<(ops node:$src),
-                         [(assertzexti5 (and node:$src, shiftwamt_mask)),
-                          (assertzexti5 node:$src)]>;
 
 /// Immediates
 
@@ -941,28 +937,20 @@
 def : Pat<(sra (sext_inreg GPR:$rs1, i32), uimm5:$shamt),
           (SRAIW GPR:$rs1, uimm5:$shamt)>;
 
-// For variable-length shifts, we rely on assertzexti5 being inserted during
-// lowering (see RISCVTargetLowering::PerformDAGCombine). This enables us to
-// guarantee that selecting a 32-bit variable shift is legal (as the variable
-// shift is known to be <= 32). We must also be careful not to create
-// semantically incorrect patterns. For instance, selecting SRLW for
-// (srl (zexti32 GPR:$rs1), (shiftwamt GPR:$rs2)),
-// is not guaranteed to be safe, as we don't know whether the upper 32-bits of
-// the result are used or not (in the case where rs2=0, this is a
-// sign-extension operation).
-
-def : Pat<(sext_inreg (shl GPR:$rs1, (shiftwamt GPR:$rs2)), i32),
-          (SLLW GPR:$rs1, GPR:$rs2)>;
-def : Pat<(zexti32 (shl GPR:$rs1, (shiftwamt GPR:$rs2))),
-          (SRLI (SLLI (SLLW GPR:$rs1, GPR:$rs2), 32), 32)>;
-
-def : Pat<(sext_inreg (srl (zexti32 GPR:$rs1), (shiftwamt GPR:$rs2)), i32),
-          (SRLW GPR:$rs1, GPR:$rs2)>;
-def : Pat<(zexti32 (srl (zexti32 GPR:$rs1), (shiftwamt GPR:$rs2))),
-          (SRLI (SLLI (SRLW GPR:$rs1, GPR:$rs2), 32), 32)>;
-
-def : Pat<(sra (sexti32 GPR:$rs1), (shiftwamt GPR:$rs2)),
-          (SRAW GPR:$rs1, GPR:$rs2)>;
+multiclass ShiftWPat<PatFrag ShiftWOp, RVInst Inst> {
+  def : Pat<(ShiftWOp GPR:$rs1, (and GPR:$rs2, shiftwamt_mask)),
+            (Inst GPR:$rs1, GPR:$rs2)>;
+  def : Pat<(ShiftWOp GPR:$rs1, GPR:$rs2),
+            (Inst GPR:$rs1, GPR:$rs2)>;
+  def : Pat<(sext_inreg (ShiftWOp GPR:$rs1, (and GPR:$rs2, shiftwamt_mask)), i32),
+            (Inst GPR:$rs1, GPR:$rs2)>;
+  def : Pat<(sext_inreg (ShiftWOp GPR:$rs1, GPR:$rs2), i32),
+            (Inst GPR:$rs1, GPR:$rs2)>;
+}
+
+defm : ShiftWPat<riscv_shlw, SLLW>;
+defm : ShiftWPat<riscv_srlw, SRLW>;
+defm : ShiftWPat<riscv_sraw, SRAW>;
 
 /// Loads
 
Index: test/CodeGen/RISCV/atomic-cmpxchg.ll
===================================================================
--- test/CodeGen/RISCV/atomic-cmpxchg.ll
+++ test/CodeGen/RISCV/atomic-cmpxchg.ll
@@ -61,8 +61,8 @@
 ;
 ; RV64IA-LABEL: cmpxchg_i8_monotonic_monotonic:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    addi a4, zero, 255
 ; RV64IA-NEXT:    sllw a4, a4, a3
 ; RV64IA-NEXT:    andi a2, a2, 255
@@ -139,8 +139,8 @@
 ;
 ; RV64IA-LABEL: cmpxchg_i8_acquire_monotonic:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    addi a4, zero, 255
 ; RV64IA-NEXT:    sllw a4, a4, a3
 ; RV64IA-NEXT:    andi a2, a2, 255
@@ -217,8 +217,8 @@
 ;
 ; RV64IA-LABEL: cmpxchg_i8_acquire_acquire:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    addi a4, zero, 255
 ; RV64IA-NEXT:    sllw a4, a4, a3
 ; RV64IA-NEXT:    andi a2, a2, 255
@@ -295,8 +295,8 @@
 ;
 ; RV64IA-LABEL: cmpxchg_i8_release_monotonic:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    addi a4, zero, 255
 ; RV64IA-NEXT:    sllw a4, a4, a3
 ; RV64IA-NEXT:    andi a2, a2, 255
@@ -373,8 +373,8 @@
 ;
 ; RV64IA-LABEL: cmpxchg_i8_release_acquire:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    addi a4, zero, 255
 ; RV64IA-NEXT:    sllw a4, a4, a3
 ; RV64IA-NEXT:    andi a2, a2, 255
@@ -451,8 +451,8 @@
 ;
 ; RV64IA-LABEL: cmpxchg_i8_acq_rel_monotonic:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    addi a4, zero, 255
 ; RV64IA-NEXT:    sllw a4, a4, a3
 ; RV64IA-NEXT:    andi a2, a2, 255
@@ -529,8 +529,8 @@
 ;
 ; RV64IA-LABEL: cmpxchg_i8_acq_rel_acquire:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    addi a4, zero, 255
 ; RV64IA-NEXT:    sllw a4, a4, a3
 ; RV64IA-NEXT:    andi a2, a2, 255
@@ -607,8 +607,8 @@
 ;
 ; RV64IA-LABEL: cmpxchg_i8_seq_cst_monotonic:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    addi a4, zero, 255
 ; RV64IA-NEXT:    sllw a4, a4, a3
 ; RV64IA-NEXT:    andi a2, a2, 255
@@ -685,8 +685,8 @@
 ;
 ; RV64IA-LABEL: cmpxchg_i8_seq_cst_acquire:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    addi a4, zero, 255
 ; RV64IA-NEXT:    sllw a4, a4, a3
 ; RV64IA-NEXT:    andi a2, a2, 255
@@ -763,8 +763,8 @@
 ;
 ; RV64IA-LABEL: cmpxchg_i8_seq_cst_seq_cst:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    addi a4, zero, 255
 ; RV64IA-NEXT:    sllw a4, a4, a3
 ; RV64IA-NEXT:    andi a2, a2, 255
@@ -846,8 +846,8 @@
 ; RV64IA-NEXT:    addiw a3, a3, -1
 ; RV64IA-NEXT:    and a1, a1, a3
 ; RV64IA-NEXT:    and a2, a2, a3
-; RV64IA-NEXT:    andi a4, a0, 3
-; RV64IA-NEXT:    slli a4, a4, 3
+; RV64IA-NEXT:    slli a4, a0, 3
+; RV64IA-NEXT:    andi a4, a4, 24
 ; RV64IA-NEXT:    sllw a3, a3, a4
 ; RV64IA-NEXT:    sllw a2, a2, a4
 ; RV64IA-NEXT:    sllw a1, a1, a4
@@ -926,8 +926,8 @@
 ; RV64IA-NEXT:    addiw a3, a3, -1
 ; RV64IA-NEXT:    and a1, a1, a3
 ; RV64IA-NEXT:    and a2, a2, a3
-; RV64IA-NEXT:    andi a4, a0, 3
-; RV64IA-NEXT:    slli a4, a4, 3
+; RV64IA-NEXT:    slli a4, a0, 3
+; RV64IA-NEXT:    andi a4, a4, 24
 ; RV64IA-NEXT:    sllw a3, a3, a4
 ; RV64IA-NEXT:    sllw a2, a2, a4
 ; RV64IA-NEXT:    sllw a1, a1, a4
@@ -1006,8 +1006,8 @@
 ; RV64IA-NEXT:    addiw a3, a3, -1
 ; RV64IA-NEXT:    and a1, a1, a3
 ; RV64IA-NEXT:    and a2, a2, a3
-; RV64IA-NEXT:    andi a4, a0, 3
-; RV64IA-NEXT:    slli a4, a4, 3
+; RV64IA-NEXT:    slli a4, a0, 3
+; RV64IA-NEXT:    andi a4, a4, 24
 ; RV64IA-NEXT:    sllw a3, a3, a4
 ; RV64IA-NEXT:    sllw a2, a2, a4
 ; RV64IA-NEXT:    sllw a1, a1, a4
@@ -1086,8 +1086,8 @@
 ; RV64IA-NEXT:    addiw a3, a3, -1
 ; RV64IA-NEXT:    and a1, a1, a3
 ; RV64IA-NEXT:    and a2, a2, a3
-; RV64IA-NEXT:    andi a4, a0, 3
-; RV64IA-NEXT:    slli a4, a4, 3
+; RV64IA-NEXT:    slli a4, a0, 3
+; RV64IA-NEXT:    andi a4, a4, 24
 ; RV64IA-NEXT:    sllw a3, a3, a4
 ; RV64IA-NEXT:    sllw a2, a2, a4
 ; RV64IA-NEXT:    sllw a1, a1, a4
@@ -1166,8 +1166,8 @@
 ; RV64IA-NEXT:    addiw a3, a3, -1
 ; RV64IA-NEXT:    and a1, a1, a3
 ; RV64IA-NEXT:    and a2, a2, a3
-; RV64IA-NEXT:    andi a4, a0, 3
-; RV64IA-NEXT:    slli a4, a4, 3
+; RV64IA-NEXT:    slli a4, a0, 3
+; RV64IA-NEXT:    andi a4, a4, 24
 ; RV64IA-NEXT:    sllw a3, a3, a4
 ; RV64IA-NEXT:    sllw a2, a2, a4
 ; RV64IA-NEXT:    sllw a1, a1, a4
@@ -1246,8 +1246,8 @@
 ; RV64IA-NEXT:    addiw a3, a3, -1
 ; RV64IA-NEXT:    and a1, a1, a3
 ; RV64IA-NEXT:    and a2, a2, a3
-; RV64IA-NEXT:    andi a4, a0, 3
-; RV64IA-NEXT:    slli a4, a4, 3
+; RV64IA-NEXT:    slli a4, a0, 3
+; RV64IA-NEXT:    andi a4, a4, 24
 ; RV64IA-NEXT:    sllw a3, a3, a4
 ; RV64IA-NEXT:    sllw a2, a2, a4
 ; RV64IA-NEXT:    sllw a1, a1, a4
@@ -1326,8 +1326,8 @@
 ; RV64IA-NEXT:    addiw a3, a3, -1
 ; RV64IA-NEXT:    and a1, a1, a3
 ; RV64IA-NEXT:    and a2, a2, a3
-; RV64IA-NEXT:    andi a4, a0, 3
-; RV64IA-NEXT:    slli a4, a4, 3
+; RV64IA-NEXT:    slli a4, a0, 3
+; RV64IA-NEXT:    andi a4, a4, 24
 ; RV64IA-NEXT:    sllw a3, a3, a4
 ; RV64IA-NEXT:    sllw a2, a2, a4
 ; RV64IA-NEXT:    sllw a1, a1, a4
@@ -1406,8 +1406,8 @@
 ; RV64IA-NEXT:    addiw a3, a3, -1
 ; RV64IA-NEXT:    and a1, a1, a3
 ; RV64IA-NEXT:    and a2, a2, a3
-; RV64IA-NEXT:    andi a4, a0, 3
-; RV64IA-NEXT:    slli a4, a4, 3
+; RV64IA-NEXT:    slli a4, a0, 3
+; RV64IA-NEXT:    andi a4, a4, 24
 ; RV64IA-NEXT:    sllw a3, a3, a4
 ; RV64IA-NEXT:    sllw a2, a2, a4
 ; RV64IA-NEXT:    sllw a1, a1, a4
@@ -1486,8 +1486,8 @@
 ; RV64IA-NEXT:    addiw a3, a3, -1
 ; RV64IA-NEXT:    and a1, a1, a3
 ; RV64IA-NEXT:    and a2, a2, a3
-; RV64IA-NEXT:    andi a4, a0, 3
-; RV64IA-NEXT:    slli a4, a4, 3
+; RV64IA-NEXT:    slli a4, a0, 3
+; RV64IA-NEXT:    andi a4, a4, 24
 ; RV64IA-NEXT:    sllw a3, a3, a4
 ; RV64IA-NEXT:    sllw a2, a2, a4
 ; RV64IA-NEXT:    sllw a1, a1, a4
@@ -1566,8 +1566,8 @@
 ; RV64IA-NEXT:    addiw a3, a3, -1
 ; RV64IA-NEXT:    and a1, a1, a3
 ; RV64IA-NEXT:    and a2, a2, a3
-; RV64IA-NEXT:    andi a4, a0, 3
-; RV64IA-NEXT:    slli a4, a4, 3
+; RV64IA-NEXT:    slli a4, a0, 3
+; RV64IA-NEXT:    andi a4, a4, 24
 ; RV64IA-NEXT:    sllw a3, a3, a4
 ; RV64IA-NEXT:    sllw a2, a2, a4
 ; RV64IA-NEXT:    sllw a1, a1, a4
Index: test/CodeGen/RISCV/atomic-rmw.ll
===================================================================
--- test/CodeGen/RISCV/atomic-rmw.ll
+++ test/CodeGen/RISCV/atomic-rmw.ll
@@ -52,8 +52,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_xchg_i8_monotonic:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -118,8 +118,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_xchg_i8_acquire:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -184,8 +184,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_xchg_i8_release:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -250,8 +250,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_xchg_i8_acq_rel:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -316,8 +316,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_xchg_i8_seq_cst:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -382,8 +382,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_add_i8_monotonic:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -448,8 +448,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_add_i8_acquire:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -514,8 +514,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_add_i8_release:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -580,8 +580,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_add_i8_acq_rel:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -646,8 +646,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_add_i8_seq_cst:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -712,8 +712,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_sub_i8_monotonic:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -778,8 +778,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_sub_i8_acquire:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -844,8 +844,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_sub_i8_release:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -910,8 +910,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_sub_i8_acq_rel:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -976,8 +976,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_sub_i8_seq_cst:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -1037,11 +1037,11 @@
 ; RV64IA-LABEL: atomicrmw_and_i8_monotonic:
 ; RV64IA:       # %bb.0:
 ; RV64IA-NEXT:    andi a1, a1, 255
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    addi a3, zero, 255
-; RV64IA-NEXT:    sll a3, a3, a2
+; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    not a3, a3
 ; RV64IA-NEXT:    or a1, a3, a1
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -1091,11 +1091,11 @@
 ; RV64IA-LABEL: atomicrmw_and_i8_acquire:
 ; RV64IA:       # %bb.0:
 ; RV64IA-NEXT:    andi a1, a1, 255
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    addi a3, zero, 255
-; RV64IA-NEXT:    sll a3, a3, a2
+; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    not a3, a3
 ; RV64IA-NEXT:    or a1, a3, a1
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -1145,11 +1145,11 @@
 ; RV64IA-LABEL: atomicrmw_and_i8_release:
 ; RV64IA:       # %bb.0:
 ; RV64IA-NEXT:    andi a1, a1, 255
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    addi a3, zero, 255
-; RV64IA-NEXT:    sll a3, a3, a2
+; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    not a3, a3
 ; RV64IA-NEXT:    or a1, a3, a1
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -1199,11 +1199,11 @@
 ; RV64IA-LABEL: atomicrmw_and_i8_acq_rel:
 ; RV64IA:       # %bb.0:
 ; RV64IA-NEXT:    andi a1, a1, 255
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    addi a3, zero, 255
-; RV64IA-NEXT:    sll a3, a3, a2
+; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    not a3, a3
 ; RV64IA-NEXT:    or a1, a3, a1
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -1253,11 +1253,11 @@
 ; RV64IA-LABEL: atomicrmw_and_i8_seq_cst:
 ; RV64IA:       # %bb.0:
 ; RV64IA-NEXT:    andi a1, a1, 255
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    addi a3, zero, 255
-; RV64IA-NEXT:    sll a3, a3, a2
+; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    not a3, a3
 ; RV64IA-NEXT:    or a1, a3, a1
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -1313,8 +1313,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_nand_i8_monotonic:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -1381,8 +1381,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_nand_i8_acquire:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -1449,8 +1449,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_nand_i8_release:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -1517,8 +1517,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_nand_i8_acq_rel:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -1585,8 +1585,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_nand_i8_seq_cst:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a3, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -1643,9 +1643,9 @@
 ; RV64IA-LABEL: atomicrmw_or_i8_monotonic:
 ; RV64IA:       # %bb.0:
 ; RV64IA-NEXT:    andi a1, a1, 255
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoor.w a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -1689,9 +1689,9 @@
 ; RV64IA-LABEL: atomicrmw_or_i8_acquire:
 ; RV64IA:       # %bb.0:
 ; RV64IA-NEXT:    andi a1, a1, 255
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoor.w.aq a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -1735,9 +1735,9 @@
 ; RV64IA-LABEL: atomicrmw_or_i8_release:
 ; RV64IA:       # %bb.0:
 ; RV64IA-NEXT:    andi a1, a1, 255
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoor.w.rl a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -1781,9 +1781,9 @@
 ; RV64IA-LABEL: atomicrmw_or_i8_acq_rel:
 ; RV64IA:       # %bb.0:
 ; RV64IA-NEXT:    andi a1, a1, 255
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoor.w.aqrl a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -1827,9 +1827,9 @@
 ; RV64IA-LABEL: atomicrmw_or_i8_seq_cst:
 ; RV64IA:       # %bb.0:
 ; RV64IA-NEXT:    andi a1, a1, 255
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoor.w.aqrl a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -1873,9 +1873,9 @@
 ; RV64IA-LABEL: atomicrmw_xor_i8_monotonic:
 ; RV64IA:       # %bb.0:
 ; RV64IA-NEXT:    andi a1, a1, 255
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoxor.w a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -1919,9 +1919,9 @@
 ; RV64IA-LABEL: atomicrmw_xor_i8_acquire:
 ; RV64IA:       # %bb.0:
 ; RV64IA-NEXT:    andi a1, a1, 255
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoxor.w.aq a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -1965,9 +1965,9 @@
 ; RV64IA-LABEL: atomicrmw_xor_i8_release:
 ; RV64IA:       # %bb.0:
 ; RV64IA-NEXT:    andi a1, a1, 255
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoxor.w.rl a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -2011,9 +2011,9 @@
 ; RV64IA-LABEL: atomicrmw_xor_i8_acq_rel:
 ; RV64IA:       # %bb.0:
 ; RV64IA-NEXT:    andi a1, a1, 255
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoxor.w.aqrl a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -2057,9 +2057,9 @@
 ; RV64IA-LABEL: atomicrmw_xor_i8_seq_cst:
 ; RV64IA:       # %bb.0:
 ; RV64IA-NEXT:    andi a1, a1, 255
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoxor.w.aqrl a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -3738,8 +3738,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_umax_i8_monotonic:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a6, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -3880,8 +3880,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_umax_i8_acquire:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a6, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -4022,8 +4022,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_umax_i8_release:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a6, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -4170,8 +4170,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_umax_i8_acq_rel:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a6, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -4312,8 +4312,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_umax_i8_seq_cst:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a6, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -4448,8 +4448,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_umin_i8_monotonic:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a6, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -4590,8 +4590,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_umin_i8_acquire:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a6, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -4732,8 +4732,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_umin_i8_release:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a6, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -4880,8 +4880,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_umin_i8_acq_rel:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a6, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -5022,8 +5022,8 @@
 ;
 ; RV64IA-LABEL: atomicrmw_umin_i8_seq_cst:
 ; RV64IA:       # %bb.0:
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
 ; RV64IA-NEXT:    addi a3, zero, 255
 ; RV64IA-NEXT:    sllw a6, a3, a2
 ; RV64IA-NEXT:    andi a1, a1, 255
@@ -5096,8 +5096,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -5164,8 +5164,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -5232,8 +5232,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -5300,8 +5300,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -5368,8 +5368,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -5436,8 +5436,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -5504,8 +5504,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -5572,8 +5572,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -5640,8 +5640,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -5708,8 +5708,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -5776,8 +5776,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -5844,8 +5844,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -5912,8 +5912,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -5980,8 +5980,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -6048,8 +6048,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -6110,10 +6110,10 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
-; RV64IA-NEXT:    sll a1, a1, a3
-; RV64IA-NEXT:    sll a2, a2, a3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
+; RV64IA-NEXT:    sllw a1, a1, a3
+; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    not a2, a2
 ; RV64IA-NEXT:    or a1, a2, a1
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -6166,10 +6166,10 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
-; RV64IA-NEXT:    sll a1, a1, a3
-; RV64IA-NEXT:    sll a2, a2, a3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
+; RV64IA-NEXT:    sllw a1, a1, a3
+; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    not a2, a2
 ; RV64IA-NEXT:    or a1, a2, a1
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -6222,10 +6222,10 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
-; RV64IA-NEXT:    sll a1, a1, a3
-; RV64IA-NEXT:    sll a2, a2, a3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
+; RV64IA-NEXT:    sllw a1, a1, a3
+; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    not a2, a2
 ; RV64IA-NEXT:    or a1, a2, a1
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -6278,10 +6278,10 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
-; RV64IA-NEXT:    sll a1, a1, a3
-; RV64IA-NEXT:    sll a2, a2, a3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
+; RV64IA-NEXT:    sllw a1, a1, a3
+; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    not a2, a2
 ; RV64IA-NEXT:    or a1, a2, a1
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -6334,10 +6334,10 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
-; RV64IA-NEXT:    sll a1, a1, a3
-; RV64IA-NEXT:    sll a2, a2, a3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
+; RV64IA-NEXT:    sllw a1, a1, a3
+; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    not a2, a2
 ; RV64IA-NEXT:    or a1, a2, a1
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -6397,8 +6397,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -6467,8 +6467,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -6537,8 +6537,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -6607,8 +6607,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -6677,8 +6677,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a2, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -6737,9 +6737,9 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoor.w a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -6787,9 +6787,9 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoor.w.aq a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -6837,9 +6837,9 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoor.w.rl a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -6887,9 +6887,9 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoor.w.aqrl a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -6937,9 +6937,9 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoor.w.aqrl a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -6987,9 +6987,9 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoxor.w a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -7037,9 +7037,9 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoxor.w.aq a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -7087,9 +7087,9 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoxor.w.rl a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -7137,9 +7137,9 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoxor.w.aqrl a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -7187,9 +7187,9 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a2, a0, 3
-; RV64IA-NEXT:    slli a2, a2, 3
-; RV64IA-NEXT:    sll a1, a1, a2
+; RV64IA-NEXT:    slli a2, a0, 3
+; RV64IA-NEXT:    andi a2, a2, 24
+; RV64IA-NEXT:    sllw a1, a1, a2
 ; RV64IA-NEXT:    andi a0, a0, -4
 ; RV64IA-NEXT:    amoxor.w.aqrl a0, a1, (a0)
 ; RV64IA-NEXT:    srlw a0, a0, a2
@@ -8900,8 +8900,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a6, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -9052,8 +9052,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a6, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -9204,8 +9204,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a6, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -9362,8 +9362,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a6, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -9514,8 +9514,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a6, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -9660,8 +9660,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a6, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -9812,8 +9812,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a6, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -9964,8 +9964,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a6, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -10122,8 +10122,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a6, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
@@ -10274,8 +10274,8 @@
 ; RV64IA-NEXT:    lui a2, 16
 ; RV64IA-NEXT:    addiw a2, a2, -1
 ; RV64IA-NEXT:    and a1, a1, a2
-; RV64IA-NEXT:    andi a3, a0, 3
-; RV64IA-NEXT:    slli a3, a3, 3
+; RV64IA-NEXT:    slli a3, a0, 3
+; RV64IA-NEXT:    andi a3, a3, 24
 ; RV64IA-NEXT:    sllw a6, a2, a3
 ; RV64IA-NEXT:    sllw a1, a1, a3
 ; RV64IA-NEXT:    andi a0, a0, -4
Index: test/CodeGen/RISCV/pr40333.ll
===================================================================
--- test/CodeGen/RISCV/pr40333.ll
+++ test/CodeGen/RISCV/pr40333.ll
@@ -13,11 +13,7 @@
 define signext i8 @foo(i32 %a, i32 %b) nounwind {
 ; RV64I-LABEL: foo:
 ; RV64I:       # %bb.0:
-; RV64I-NEXT:    slli a1, a1, 32
-; RV64I-NEXT:    srli a1, a1, 32
-; RV64I-NEXT:    slli a0, a0, 32
-; RV64I-NEXT:    srli a0, a0, 32
-; RV64I-NEXT:    srl a0, a0, a1
+; RV64I-NEXT:    srlw a0, a0, a1
 ; RV64I-NEXT:    slli a0, a0, 56
 ; RV64I-NEXT:    srai a0, a0, 56
 ; RV64I-NEXT:    ret