[TargetLowering][RISCV][AArch64][PowerPC] Enable BuildUDIV/BuildSDIV on illegal types before type legalization if we can find a larger legal type that supports MUL.

[craig.topper]

Summary

If we wait until the type is legalized, we'll lose information
about the orginal type and need to use larger magic constants.
This gets especially bad on RISCV64 where i64 is the only legal
type.

I've limited this to simple scalar types so it only works for
i8/i16/i32 which are most likely to occur. For more odd types
we might want to do a small promotion to a type where MULH is legal
instead.

Unfortunately, this does prevent some urem/srem+seteq matching since
that still require legal types.

[RKSimon]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

5101

Do we not have a legalization helper function that we can use?

5187

Pre-legalization shift amount is always nasty - is there any chance that this can be out of range?

llvm/test/CodeGen/AArch64/urem-seteq.ll

86

No idea why but its odd that we now match a multiply-subtract here but lose the multiply-add in urem-seteq-nonzero.ll above

[craig.topper]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

5101

I thought about using getTypeAction() == TypePromoteInteger and getTypeToTransformTo. And now I can't remember why I didn't.

5187

MulVT here is a legal type so the shift should be legal. So I think the target better be returning a shift amount type that should cover all possible amounts.

llvm/test/CodeGen/AArch64/urem-seteq.ll

86

The msub here is part of the remainder calculation. The case above used a divisor of 3 which is one more than a power of 2 so the multiply for the remainder was turned into "add w8, w8, w8, lsl #1"

We're using different algorithms now so its hard to say much more than that.

[RKSimon]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

5187

Would getShiftAmountConstant be safer?

[craig.topper]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

5187

It's shorter. I don't think its safer though. It just chooses between getScalarShiftAmountTy and PointerTy. It doesn't check that the constant fits.

[RKSimon]

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

5187

OK its still probably worth trying to use getShiftAmountConstant.

[craig.topper]

Use getShiftAmountConstant

[RKSimon]

Cheers Craig - any thoughts on how best to tidy up the legal type finder code?

[craig.topper]

Use getTypeAction/getTypeToTransformTo to remove the loop.

[RKSimon]

LGTM

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

 //===-- TargetLowering.cpp - Implement the TargetLowering class -----------===//
 //
 // 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 TargetLowering::BuildSDIV(SDNode *N, SelectionDAG &DAG,
                                   bool IsAfterLegalization,
                                   SmallVectorImpl<SDNode *> &Created) const {
   SDLoc dl(N);
   EVT VT = N->getValueType(0);
   EVT SVT = VT.getScalarType();
   EVT ShVT = getShiftAmountTy(VT, DAG.getDataLayout());
   EVT ShSVT = ShVT.getScalarType();
   unsigned EltBits = VT.getScalarSizeInBits();
+  MVT MulVT;
 
   // Check to see if we can do this.
   // FIXME: We should be more aggressive here.
-  if (!isTypeLegal(VT))
+  if (!isTypeLegal(VT)) {
+    // Limit this to simple scalars for now.
+    if (VT.isVector() || !VT.isSimple())
       return SDValue();
 
+    // Try to find a legal integer type with enough bits to hold a full
+    // multiply result.
+    for (unsigned t = MVT::FIRST_INTEGER_VALUETYPE;
+         t <= MVT::LAST_INTEGER_VALUETYPE; ++t) {
+      MVT NVT = (MVT::SimpleValueType)t;
+      if (NVT.getSizeInBits() >= (2 * EltBits) &&
+          isOperationLegal(ISD::MUL, NVT)) {
+        MulVT = NVT;
+        break;
+      }
+    }
+    if (MulVT == MVT::INVALID_SIMPLE_VALUE_TYPE)
+      return SDValue();
+  }
+
   // If the sdiv has an 'exact' bit we can use a simpler lowering.
   if (N->getFlags().hasExact())
     return BuildExactSDIV(*this, N, dl, DAG, Created);
 
   SmallVector<SDValue, 16> MagicFactors, Factors, Shifts, ShiftMasks;
 
   auto BuildSDIVPattern = [&](ConstantSDNode *C) {
     if (C->isNullValue())
if (VT.isFixedLengthVector()) {
     MagicFactor = MagicFactors[0];
     Factor = Factors[0];
     Shift = Shifts[0];
     ShiftMask = ShiftMasks[0];
   }
 
   // Multiply the numerator (operand 0) by the magic value.
   // FIXME: We should support doing a MUL in a wider type.
-  SDValue Q;
+  auto GetMULHS = [&](SDValue X, SDValue Y) {
+    // If the type isn't legal, use a wider mul of the the type calculated
+    // earlier.
+    if (!isTypeLegal(VT)) {
+      X = DAG.getNode(ISD::SIGN_EXTEND, dl, MulVT, X);
+      Y = DAG.getNode(ISD::SIGN_EXTEND, dl, MulVT, Y);
+      Y = DAG.getNode(ISD::MUL, dl, MulVT, X, Y);
+      Y = DAG.getNode(
+          ISD::SRL, dl, MulVT, Y,
+          DAG.getConstant(EltBits, dl,
+                          getShiftAmountTy(MulVT, DAG.getDataLayout())));
+      return DAG.getNode(ISD::TRUNCATE, dl, VT, Y);
+    }
+
     if (isOperationLegalOrCustom(ISD::MULHS, VT, IsAfterLegalization))
-    Q = DAG.getNode(ISD::MULHS, dl, VT, N0, MagicFactor);
-  else if (isOperationLegalOrCustom(ISD::SMUL_LOHI, VT, IsAfterLegalization)) {
+      return DAG.getNode(ISD::MULHS, dl, VT, X, Y);
+    if (isOperationLegalOrCustom(ISD::SMUL_LOHI, VT, IsAfterLegalization)) {
       SDValue LoHi =
-        DAG.getNode(ISD::SMUL_LOHI, dl, DAG.getVTList(VT, VT), N0, MagicFactor);
-    Q = SDValue(LoHi.getNode(), 1);
-  } else
-    return SDValue(); // No mulhs or equivalent.
+          DAG.getNode(ISD::SMUL_LOHI, dl, DAG.getVTList(VT, VT), X, Y);
+      return SDValue(LoHi.getNode(), 1);
+    }
+    return SDValue();
+  };
+
+  SDValue Q = GetMULHS(N0, MagicFactor);
+  if (!Q)
+    return SDValue();
+
   Created.push_back(Q.getNode());
 
   // (Optionally) Add/subtract the numerator using Factor.
   Factor = DAG.getNode(ISD::MUL, dl, VT, N0, Factor);
   Created.push_back(Factor.getNode());
   Q = DAG.getNode(ISD::ADD, dl, VT, Q, Factor);
   Created.push_back(Q.getNode());
 
SDValue TargetLowering::BuildUDIV(SDNode *N, SelectionDAG &DAG,
                                   bool IsAfterLegalization,
                                   SmallVectorImpl<SDNode *> &Created) const {
   SDLoc dl(N);
   EVT VT = N->getValueType(0);
   EVT SVT = VT.getScalarType();
   EVT ShVT = getShiftAmountTy(VT, DAG.getDataLayout());
   EVT ShSVT = ShVT.getScalarType();
   unsigned EltBits = VT.getScalarSizeInBits();
+  MVT MulVT;
 
   // Check to see if we can do this.
   // FIXME: We should be more aggressive here.
-  if (!isTypeLegal(VT))
+  if (!isTypeLegal(VT)) {
+    // Limit this to simple scalars for now.
+    if (VT.isVector() || !VT.isSimple())
+      return SDValue();
+
+    // Try to find a legal integer type with enough bits to hold a full
+    // multiply result.
+    for (unsigned t = MVT::FIRST_INTEGER_VALUETYPE;
+         t <= MVT::LAST_INTEGER_VALUETYPE; ++t) {
+      MVT NVT = (MVT::SimpleValueType)t;
+      if (NVT.getSizeInBits() >= (2 * EltBits) &&
+          isOperationLegal(ISD::MUL, NVT)) {
+        MulVT = NVT;
+        break;
+      }
+    }
+    if (MulVT == MVT::INVALID_SIMPLE_VALUE_TYPE)
       return SDValue();
+  }
 
   bool UseNPQ = false;
   SmallVector<SDValue, 16> PreShifts, PostShifts, MagicFactors, NPQFactors;
 
   auto BuildUDIVPattern = [&](ConstantSDNode *C) {
     if (C->isNullValue())
       return false;
     // FIXME: We should use a narrower constant when the upper
SDValue TargetLowering::BuildUDIV(SDNode *N, SelectionDAG &DAG,
   }
 
   SDValue Q = N0;
   Q = DAG.getNode(ISD::SRL, dl, VT, Q, PreShift);
   Created.push_back(Q.getNode());
 
   // FIXME: We should support doing a MUL in a wider type.
   auto GetMULHU = [&](SDValue X, SDValue Y) {
+    // If the type isn't legal, use a wider mul of the the type calculated
+    // earlier.
+    if (!isTypeLegal(VT)) {
+      X = DAG.getNode(ISD::ZERO_EXTEND, dl, MulVT, X);
+      Y = DAG.getNode(ISD::ZERO_EXTEND, dl, MulVT, Y);
+      Y = DAG.getNode(ISD::MUL, dl, MulVT, X, Y);
+      Y = DAG.getNode(
+          ISD::SRL, dl, MulVT, Y,
+          DAG.getConstant(EltBits, dl,
+                          getShiftAmountTy(MulVT, DAG.getDataLayout())));
+      return DAG.getNode(ISD::TRUNCATE, dl, VT, Y);
+    }
+
     if (isOperationLegalOrCustom(ISD::MULHU, VT, IsAfterLegalization))
       return DAG.getNode(ISD::MULHU, dl, VT, X, Y);
     if (isOperationLegalOrCustom(ISD::UMUL_LOHI, VT, IsAfterLegalization)) {
       SDValue LoHi =
           DAG.getNode(ISD::UMUL_LOHI, dl, DAG.getVTList(VT, VT), X, Y);
       return SDValue(LoHi.getNode(), 1);
     }
     return SDValue(); // No mulhu or equivalent

llvm/lib/Target/BPF/BPFISelLowering.h

 //===-- BPFISelLowering.h - BPF DAG Lowering Interface ----------*- 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
 //
 //===----------------------------------------------------------------------===//
 //
SDValue LowerReturn(SDValue Chain, CallingConv::ID CallConv, bool IsVarArg,
                       const SmallVectorImpl<SDValue> &OutVals, const SDLoc &DL,
                       SelectionDAG &DAG) const override;
 
   EVT getOptimalMemOpType(const MemOp &Op,
                           const AttributeList &FuncAttributes) const override {
     return Op.size() >= 8 ? MVT::i64 : MVT::i32;
   }
 
+  bool isIntDivCheap(EVT VT, AttributeList Attr) const override { return true; }
+
+
   bool shouldConvertConstantLoadToIntImm(const APInt &Imm,
                                          Type *Ty) const override {
     return true;
   }
 
   // Prevent reducing load width during SelectionDag phase.
   // Otherwise, we may transform the following
   //   ctx = ctx + reloc_offset

llvm/test/CodeGen/AArch64/srem-seteq.ll

 
 ;------------------------------------------------------------------------------;
 ; Even divisors
 ;------------------------------------------------------------------------------;
 
 define i16 @test_srem_even(i16 %X) nounwind {
 ; CHECK-LABEL: test_srem_even:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w9, #9363
 ; CHECK-NEXT:    sxth w8, w0
-; CHECK-NEXT:    movk w9, #37449, lsl #16
-; CHECK-NEXT:    smull x9, w8, w9
-; CHECK-NEXT:    lsr x9, x9, #32
-; CHECK-NEXT:    add w8, w9, w8
-; CHECK-NEXT:    asr w9, w8, #3
+; CHECK-NEXT:    mov w9, #18725
+; CHECK-NEXT:    mul w8, w8, w9
+; CHECK-NEXT:    asr w9, w8, #18
 ; CHECK-NEXT:    add w8, w9, w8, lsr #31
 ; CHECK-NEXT:    mov w9, #14
 ; CHECK-NEXT:    msub w8, w8, w9, w0
 ; CHECK-NEXT:    tst w8, #0xffff
 ; CHECK-NEXT:    cset w0, ne
 ; CHECK-NEXT:    ret
   %srem = srem i16 %X, 14
   %cmp = icmp ne i16 %srem, 0

llvm/test/CodeGen/AArch64/urem-seteq-nonzero.ll

 }
 
 ;-------------------------------------------------------------------------------
 ; Other widths.
 
 define i1 @t16_3_2(i16 %X) nounwind {
 ; CHECK-LABEL: t16_3_2:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w9, #43691
 ; CHECK-NEXT:    and w8, w0, #0xffff
-; CHECK-NEXT:    movk w9, #43690, lsl #16
-; CHECK-NEXT:    mov w10, #-1431655766
-; CHECK-NEXT:    madd w8, w8, w9, w10
-; CHECK-NEXT:    mov w9, #1431655765
-; CHECK-NEXT:    cmp w8, w9
-; CHECK-NEXT:    cset w0, lo
+; CHECK-NEXT:    mov w9, #43691
+; CHECK-NEXT:    mul w8, w8, w9
+; CHECK-NEXT:    lsr w8, w8, #17
+; CHECK-NEXT:    add w8, w8, w8, lsl #1
+; CHECK-NEXT:    sub w8, w0, w8
+; CHECK-NEXT:    and w8, w8, #0xffff
+; CHECK-NEXT:    cmp w8, #2 // =2
+; CHECK-NEXT:    cset w0, eq
 ; CHECK-NEXT:    ret
   %urem = urem i16 %X, 3
   %cmp = icmp eq i16 %urem, 2
   ret i1 %cmp
 }
 
 define i1 @t8_3_2(i8 %X) nounwind {
 ; CHECK-LABEL: t8_3_2:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w9, #43691
 ; CHECK-NEXT:    and w8, w0, #0xff
-; CHECK-NEXT:    movk w9, #43690, lsl #16
-; CHECK-NEXT:    mov w10, #-1431655766
-; CHECK-NEXT:    madd w8, w8, w9, w10
-; CHECK-NEXT:    mov w9, #1431655765
-; CHECK-NEXT:    cmp w8, w9
-; CHECK-NEXT:    cset w0, lo
+; CHECK-NEXT:    mov w9, #171
+; CHECK-NEXT:    mul w8, w8, w9
+; CHECK-NEXT:    lsr w8, w8, #9
+; CHECK-NEXT:    add w8, w8, w8, lsl #1
+; CHECK-NEXT:    sub w8, w0, w8
+; CHECK-NEXT:    and w8, w8, #0xff
+; CHECK-NEXT:    cmp w8, #2 // =2
+; CHECK-NEXT:    cset w0, eq
 ; CHECK-NEXT:    ret
   %urem = urem i8 %X, 3
   %cmp = icmp eq i8 %urem, 2
   ret i1 %cmp
 }
 
 define i1 @t64_3_2(i64 %X) nounwind {
 ; CHECK-LABEL: t64_3_2:

llvm/test/CodeGen/AArch64/urem-seteq.ll

 
 ;------------------------------------------------------------------------------;
 ; Even divisors
 ;------------------------------------------------------------------------------;
 
 define i16 @test_urem_even(i16 %X) nounwind {
 ; CHECK-LABEL: test_urem_even:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w9, #28087
-; CHECK-NEXT:    and w8, w0, #0xffff
-; CHECK-NEXT:    movk w9, #46811, lsl #16
+; CHECK-NEXT:    ubfx w8, w0, #1, #15
+; CHECK-NEXT:    mov w9, #18725
 ; CHECK-NEXT:    mul w8, w8, w9
-; CHECK-NEXT:    mov w9, #9362
-; CHECK-NEXT:    ror w8, w8, #1
-; CHECK-NEXT:    movk w9, #4681, lsl #16
-; CHECK-NEXT:    cmp w8, w9
-; CHECK-NEXT:    cset w0, hi
+; CHECK-NEXT:    lsr w8, w8, #17
+; CHECK-NEXT:    mov w9, #14
+; CHECK-NEXT:    msub w8, w8, w9, w0
+; CHECK-NEXT:    tst w8, #0xffff
+; CHECK-NEXT:    cset w0, ne
 ; CHECK-NEXT:    ret
   %urem = urem i16 %X, 14
   %cmp = icmp ne i16 %urem, 0
   %ret = zext i1 %cmp to i16
   ret i16 %ret
 }
 
 define i32 @test_urem_even_100(i32 %X) nounwind {

llvm/test/CodeGen/PowerPC/loop-instr-form-prepare.ll

 ;}
 
 @arr = external local_unnamed_addr global i8*, align 8
 
 define i64 @test_ds_cross_basic_blocks(i8* %0, i32 signext %1) {
 ; CHECK-LABEL: test_ds_cross_basic_blocks:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    cmplwi r4, 0
-; CHECK-NEXT:    std r26, -48(r1) # 8-byte Folded Spill
 ; CHECK-NEXT:    std r27, -40(r1) # 8-byte Folded Spill
 ; CHECK-NEXT:    std r28, -32(r1) # 8-byte Folded Spill
 ; CHECK-NEXT:    std r29, -24(r1) # 8-byte Folded Spill
 ; CHECK-NEXT:    std r30, -16(r1) # 8-byte Folded Spill
 ; CHECK-NEXT:    beq cr0, .LBB6_8
 ; CHECK-NEXT:  # %bb.1:
 ; CHECK-NEXT:    addis r5, r2, .LC0@toc@ha
 ; CHECK-NEXT:    cmpldi r4, 1
 ; CHECK-NEXT:    li r7, 1
 ; CHECK-NEXT:    addi r6, r3, 4009
 ; CHECK-NEXT:    ld r5, .LC0@toc@l(r5)
-; CHECK-NEXT:    iselgt r8, r4, r7
-; CHECK-NEXT:    lis r4, -21846
+; CHECK-NEXT:    iselgt r4, r4, r7
 ; CHECK-NEXT:    li r3, 0
-; CHECK-NEXT:    li r9, -7
-; CHECK-NEXT:    li r10, -6
+; CHECK-NEXT:    li r8, -7
+; CHECK-NEXT:    li r9, -6
+; CHECK-NEXT:    li r10, 1
 ; CHECK-NEXT:    li r11, 1
 ; CHECK-NEXT:    li r12, 1
 ; CHECK-NEXT:    li r30, 1
 ; CHECK-NEXT:    ld r5, 0(r5)
-; CHECK-NEXT:    mtctr r8
-; CHECK-NEXT:    ori r4, r4, 43691
-; CHECK-NEXT:    li r8, -9
+; CHECK-NEXT:    mtctr r4
+; CHECK-NEXT:    li r4, -9
 ; CHECK-NEXT:    li r29, 1
-; CHECK-NEXT:    li r28, 1
 ; CHECK-NEXT:    addi r5, r5, -1
 ; CHECK-NEXT:    b .LBB6_4
 ; CHECK-NEXT:    .p2align 4
 ; CHECK-NEXT:  .LBB6_2:
-; CHECK-NEXT:    ldx r0, r6, r8
-; CHECK-NEXT:    add r28, r0, r28
-; CHECK-NEXT:    ld r0, -8(r6)
+; CHECK-NEXT:    ldx r0, r6, r4
 ; CHECK-NEXT:    add r29, r0, r29
+; CHECK-NEXT:    ld r0, -8(r6)
+; CHECK-NEXT:    add r30, r0, r30
 ; CHECK-NEXT:  .LBB6_3:
-; CHECK-NEXT:    mulld r0, r29, r28
+; CHECK-NEXT:    mulld r0, r30, r29
 ; CHECK-NEXT:    addi r6, r6, 1
-; CHECK-NEXT:    mulld r0, r0, r30
 ; CHECK-NEXT:    mulld r0, r0, r12
 ; CHECK-NEXT:    mulld r0, r0, r11
+; CHECK-NEXT:    mulld r0, r0, r10
 ; CHECK-NEXT:    maddld r3, r0, r7, r3
 ; CHECK-NEXT:    bdz .LBB6_9
 ; CHECK-NEXT:  .LBB6_4:
 ; CHECK-NEXT:    lbzu r0, 1(r5)
-; CHECK-NEXT:    mulhwu r27, r0, r4
-; CHECK-NEXT:    rlwinm r26, r27, 0, 0, 30
-; CHECK-NEXT:    srwi r27, r27, 1
-; CHECK-NEXT:    add r27, r27, r26
-; CHECK-NEXT:    sub r0, r0, r27
+; CHECK-NEXT:    mulli r28, r0, 171
+; CHECK-NEXT:    rlwinm r27, r28, 24, 8, 30
+; CHECK-NEXT:    srwi r28, r28, 9
+; CHECK-NEXT:    add r28, r28, r27
+; CHECK-NEXT:    sub r0, r0, r28
+; CHECK-NEXT:    clrlwi r0, r0, 24
 ; CHECK-NEXT:    cmplwi r0, 1
 ; CHECK-NEXT:    beq cr0, .LBB6_2
 ; CHECK-NEXT:  # %bb.5:
-; CHECK-NEXT:    clrlwi r0, r0, 24
 ; CHECK-NEXT:    cmplwi r0, 2
 ; CHECK-NEXT:    bne cr0, .LBB6_7
 ; CHECK-NEXT:  # %bb.6:
-; CHECK-NEXT:    ldx r0, r6, r9
-; CHECK-NEXT:    add r30, r0, r30
-; CHECK-NEXT:    ld r0, -4(r6)
+; CHECK-NEXT:    ldx r0, r6, r8
 ; CHECK-NEXT:    add r12, r0, r12
+; CHECK-NEXT:    ld r0, -4(r6)
+; CHECK-NEXT:    add r11, r0, r11
 ; CHECK-NEXT:    b .LBB6_3
 ; CHECK-NEXT:    .p2align 4
 ; CHECK-NEXT:  .LBB6_7:
-; CHECK-NEXT:    ldx r0, r6, r10
-; CHECK-NEXT:    add r11, r0, r11
+; CHECK-NEXT:    ldx r0, r6, r9
+; CHECK-NEXT:    add r10, r0, r10
 ; CHECK-NEXT:    ld r0, 0(r6)
 ; CHECK-NEXT:    add r7, r0, r7
 ; CHECK-NEXT:    b .LBB6_3
 ; CHECK-NEXT:  .LBB6_8:
 ; CHECK-NEXT:    li r3, 0
 ; CHECK-NEXT:  .LBB6_9:
 ; CHECK-NEXT:    ld r30, -16(r1) # 8-byte Folded Reload
 ; CHECK-NEXT:    ld r29, -24(r1) # 8-byte Folded Reload
 ; CHECK-NEXT:    ld r28, -32(r1) # 8-byte Folded Reload
 ; CHECK-NEXT:    ld r27, -40(r1) # 8-byte Folded Reload
-; CHECK-NEXT:    ld r26, -48(r1) # 8-byte Folded Reload
 ; CHECK-NEXT:    blr
   %3 = sext i32 %1 to i64
   %4 = icmp eq i32 %1, 0
   br i1 %4, label %66, label %5
 
 5:                                                ; preds = %2
   %6 = load i8*, i8** @arr, align 8
   br label %7

llvm/test/CodeGen/RISCV/div.ll

 ; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    addi sp, sp, 16
 ; RV64I-NEXT:    ret
 ;
 ; RV64IM-LABEL: udiv_constant:
 ; RV64IM:       # %bb.0:
 ; RV64IM-NEXT:    slli a0, a0, 32
 ; RV64IM-NEXT:    srli a0, a0, 32
-; RV64IM-NEXT:    lui a1, 1035469
+; RV64IM-NEXT:    lui a1, 205
 ; RV64IM-NEXT:    addiw a1, a1, -819
 ; RV64IM-NEXT:    slli a1, a1, 12
 ; RV64IM-NEXT:    addi a1, a1, -819
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, -819
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, -819
-; RV64IM-NEXT:    mulhu a0, a0, a1
-; RV64IM-NEXT:    srli a0, a0, 2
+; RV64IM-NEXT:    mul a0, a0, a1
+; RV64IM-NEXT:    srli a0, a0, 34
 ; RV64IM-NEXT:    ret
   %1 = udiv i32 %a, 5
   ret i32 %1
 }
 
 define i32 @udiv_pow2(i32 %a) nounwind {
 ; RV32I-LABEL: udiv_pow2:
 ; RV32I:       # %bb.0:

 ; RV32I-NEXT:    call __udivsi3@plt
 ; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-NEXT:    addi sp, sp, 16
 ; RV32I-NEXT:    ret
 ;
 ; RV32IM-LABEL: udiv8_constant:
 ; RV32IM:       # %bb.0:
 ; RV32IM-NEXT:    andi a0, a0, 255
-; RV32IM-NEXT:    lui a1, 838861
-; RV32IM-NEXT:    addi a1, a1, -819
-; RV32IM-NEXT:    mulhu a0, a0, a1
-; RV32IM-NEXT:    srli a0, a0, 2
+; RV32IM-NEXT:    addi a1, zero, 205
+; RV32IM-NEXT:    mul a0, a0, a1
+; RV32IM-NEXT:    srli a0, a0, 10
 ; RV32IM-NEXT:    ret
 ;
 ; RV64I-LABEL: udiv8_constant:
 ; RV64I:       # %bb.0:
 ; RV64I-NEXT:    addi sp, sp, -16
 ; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64I-NEXT:    andi a0, a0, 255
 ; RV64I-NEXT:    addi a1, zero, 5
 ; RV64I-NEXT:    call __udivdi3@plt
 ; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    addi sp, sp, 16
 ; RV64I-NEXT:    ret
 ;
 ; RV64IM-LABEL: udiv8_constant:
 ; RV64IM:       # %bb.0:
 ; RV64IM-NEXT:    andi a0, a0, 255
-; RV64IM-NEXT:    lui a1, 1035469
-; RV64IM-NEXT:    addiw a1, a1, -819
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, -819
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, -819
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, -819
-; RV64IM-NEXT:    mulhu a0, a0, a1
-; RV64IM-NEXT:    srli a0, a0, 2
+; RV64IM-NEXT:    addi a1, zero, 205
+; RV64IM-NEXT:    mul a0, a0, a1
+; RV64IM-NEXT:    srli a0, a0, 10
 ; RV64IM-NEXT:    ret
   %1 = udiv i8 %a, 5
   ret i8 %1
 }
 
 define i8 @udiv8_pow2(i8 %a) nounwind {
 ; RV32I-LABEL: udiv8_pow2:
 ; RV32I:       # %bb.0:

 ; RV32I-NEXT:    addi sp, sp, 16
 ; RV32I-NEXT:    ret
 ;
 ; RV32IM-LABEL: udiv16_constant:
 ; RV32IM:       # %bb.0:
 ; RV32IM-NEXT:    lui a1, 16
 ; RV32IM-NEXT:    addi a1, a1, -1
 ; RV32IM-NEXT:    and a0, a0, a1
-; RV32IM-NEXT:    lui a1, 838861
+; RV32IM-NEXT:    lui a1, 13
 ; RV32IM-NEXT:    addi a1, a1, -819
-; RV32IM-NEXT:    mulhu a0, a0, a1
-; RV32IM-NEXT:    srli a0, a0, 2
+; RV32IM-NEXT:    mul a0, a0, a1
+; RV32IM-NEXT:    srli a0, a0, 18
 ; RV32IM-NEXT:    ret
 ;
 ; RV64I-LABEL: udiv16_constant:
 ; RV64I:       # %bb.0:
 ; RV64I-NEXT:    addi sp, sp, -16
 ; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64I-NEXT:    lui a1, 16
 ; RV64I-NEXT:    addiw a1, a1, -1
 ; RV64I-NEXT:    and a0, a0, a1
 ; RV64I-NEXT:    addi a1, zero, 5
 ; RV64I-NEXT:    call __udivdi3@plt
 ; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    addi sp, sp, 16
 ; RV64I-NEXT:    ret
 ;
 ; RV64IM-LABEL: udiv16_constant:
 ; RV64IM:       # %bb.0:
 ; RV64IM-NEXT:    lui a1, 16
 ; RV64IM-NEXT:    addiw a1, a1, -1
 ; RV64IM-NEXT:    and a0, a0, a1
-; RV64IM-NEXT:    lui a1, 1035469
+; RV64IM-NEXT:    lui a1, 13
 ; RV64IM-NEXT:    addiw a1, a1, -819
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, -819
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, -819
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, -819
-; RV64IM-NEXT:    mulhu a0, a0, a1
-; RV64IM-NEXT:    srli a0, a0, 2
+; RV64IM-NEXT:    mul a0, a0, a1
+; RV64IM-NEXT:    srli a0, a0, 18
 ; RV64IM-NEXT:    ret
   %1 = udiv i16 %a, 5
   ret i16 %1
 }
 
 define i16 @udiv16_pow2(i16 %a) nounwind {
 ; RV32I-LABEL: udiv16_pow2:
 ; RV32I:       # %bb.0:

 ; RV64I-NEXT:    call __divdi3@plt
 ; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    addi sp, sp, 16
 ; RV64I-NEXT:    ret
 ;
 ; RV64IM-LABEL: sdiv_constant:
 ; RV64IM:       # %bb.0:
 ; RV64IM-NEXT:    sext.w a0, a0
-; RV64IM-NEXT:    lui a1, 13107
-; RV64IM-NEXT:    addiw a1, a1, 819
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, 819
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, 819
-; RV64IM-NEXT:    slli a1, a1, 13
-; RV64IM-NEXT:    addi a1, a1, 1639
-; RV64IM-NEXT:    mulh a0, a0, a1
+; RV64IM-NEXT:    lui a1, 419430
+; RV64IM-NEXT:    addiw a1, a1, 1639
+; RV64IM-NEXT:    mul a0, a0, a1
 ; RV64IM-NEXT:    srli a1, a0, 63
-; RV64IM-NEXT:    srai a0, a0, 1
+; RV64IM-NEXT:    srai a0, a0, 33
 ; RV64IM-NEXT:    add a0, a0, a1
 ; RV64IM-NEXT:    ret
   %1 = sdiv i32 %a, 5
   ret i32 %1
 }
 
 define i32 @sdiv_pow2(i32 %a) nounwind {
 ; RV32I-LABEL: sdiv_pow2:

 ; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-NEXT:    addi sp, sp, 16
 ; RV32I-NEXT:    ret
 ;
 ; RV32IM-LABEL: sdiv8_constant:
 ; RV32IM:       # %bb.0:
 ; RV32IM-NEXT:    slli a0, a0, 24
 ; RV32IM-NEXT:    srai a0, a0, 24
-; RV32IM-NEXT:    lui a1, 419430
-; RV32IM-NEXT:    addi a1, a1, 1639
-; RV32IM-NEXT:    mulh a0, a0, a1
-; RV32IM-NEXT:    srli a1, a0, 31
-; RV32IM-NEXT:    srai a0, a0, 1
-; RV32IM-NEXT:    add a0, a0, a1
+; RV32IM-NEXT:    addi a1, zero, 103
+; RV32IM-NEXT:    mul a0, a0, a1
+; RV32IM-NEXT:    srai a1, a0, 9
+; RV32IM-NEXT:    srli a0, a0, 15
+; RV32IM-NEXT:    andi a0, a0, 1
+; RV32IM-NEXT:    add a0, a1, a0
 ; RV32IM-NEXT:    ret
 ;
 ; RV64I-LABEL: sdiv8_constant:
 ; RV64I:       # %bb.0:
 ; RV64I-NEXT:    addi sp, sp, -16
 ; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64I-NEXT:    slli a0, a0, 56
 ; RV64I-NEXT:    srai a0, a0, 56
 ; RV64I-NEXT:    addi a1, zero, 5
 ; RV64I-NEXT:    call __divdi3@plt
 ; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    addi sp, sp, 16
 ; RV64I-NEXT:    ret
 ;
 ; RV64IM-LABEL: sdiv8_constant:
 ; RV64IM:       # %bb.0:
 ; RV64IM-NEXT:    slli a0, a0, 56
 ; RV64IM-NEXT:    srai a0, a0, 56
-; RV64IM-NEXT:    lui a1, 13107
-; RV64IM-NEXT:    addiw a1, a1, 819
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, 819
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, 819
-; RV64IM-NEXT:    slli a1, a1, 13
-; RV64IM-NEXT:    addi a1, a1, 1639
-; RV64IM-NEXT:    mulh a0, a0, a1
-; RV64IM-NEXT:    srli a1, a0, 63
-; RV64IM-NEXT:    srai a0, a0, 1
-; RV64IM-NEXT:    add a0, a0, a1
+; RV64IM-NEXT:    addi a1, zero, 103
+; RV64IM-NEXT:    mul a0, a0, a1
+; RV64IM-NEXT:    srai a1, a0, 9
+; RV64IM-NEXT:    srli a0, a0, 15
+; RV64IM-NEXT:    andi a0, a0, 1
+; RV64IM-NEXT:    add a0, a1, a0
 ; RV64IM-NEXT:    ret
   %1 = sdiv i8 %a, 5
   ret i8 %1
 }
 
 define i8 @sdiv8_pow2(i8 %a) nounwind {
 ; RV32I-LABEL: sdiv8_pow2:
 ; RV32I:       # %bb.0:

 ; RV32I-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32I-NEXT:    addi sp, sp, 16
 ; RV32I-NEXT:    ret
 ;
 ; RV32IM-LABEL: sdiv16_constant:
 ; RV32IM:       # %bb.0:
 ; RV32IM-NEXT:    slli a0, a0, 16
 ; RV32IM-NEXT:    srai a0, a0, 16
-; RV32IM-NEXT:    lui a1, 419430
+; RV32IM-NEXT:    lui a1, 6
 ; RV32IM-NEXT:    addi a1, a1, 1639
-; RV32IM-NEXT:    mulh a0, a0, a1
+; RV32IM-NEXT:    mul a0, a0, a1
 ; RV32IM-NEXT:    srli a1, a0, 31
-; RV32IM-NEXT:    srai a0, a0, 1
+; RV32IM-NEXT:    srai a0, a0, 17
 ; RV32IM-NEXT:    add a0, a0, a1
 ; RV32IM-NEXT:    ret
 ;
 ; RV64I-LABEL: sdiv16_constant:
 ; RV64I:       # %bb.0:
 ; RV64I-NEXT:    addi sp, sp, -16
 ; RV64I-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
 ; RV64I-NEXT:    slli a0, a0, 48
 ; RV64I-NEXT:    srai a0, a0, 48
 ; RV64I-NEXT:    addi a1, zero, 5
 ; RV64I-NEXT:    call __divdi3@plt
 ; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    addi sp, sp, 16
 ; RV64I-NEXT:    ret
 ;
 ; RV64IM-LABEL: sdiv16_constant:
 ; RV64IM:       # %bb.0:
 ; RV64IM-NEXT:    slli a0, a0, 48
 ; RV64IM-NEXT:    srai a0, a0, 48
-; RV64IM-NEXT:    lui a1, 13107
-; RV64IM-NEXT:    addiw a1, a1, 819
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, 819
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, 819
-; RV64IM-NEXT:    slli a1, a1, 13
-; RV64IM-NEXT:    addi a1, a1, 1639
-; RV64IM-NEXT:    mulh a0, a0, a1
-; RV64IM-NEXT:    srli a1, a0, 63
-; RV64IM-NEXT:    srai a0, a0, 1
-; RV64IM-NEXT:    add a0, a0, a1
+; RV64IM-NEXT:    lui a1, 6
+; RV64IM-NEXT:    addiw a1, a1, 1639
+; RV64IM-NEXT:    mul a0, a0, a1
+; RV64IM-NEXT:    srai a1, a0, 17
+; RV64IM-NEXT:    srli a0, a0, 31
+; RV64IM-NEXT:    andi a0, a0, 1
+; RV64IM-NEXT:    add a0, a1, a0
 ; RV64IM-NEXT:    ret
   %1 = sdiv i16 %a, 5
   ret i16 %1
 }
 
 define i16 @sdiv16_pow2(i16 %a) nounwind {
 ; RV32I-LABEL: sdiv16_pow2:
 ; RV32I:       # %bb.0:

llvm/test/CodeGen/RISCV/srem-lkk.ll

 ; RV64I-NEXT:    addi a1, zero, 95
 ; RV64I-NEXT:    call __moddi3@plt
 ; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    addi sp, sp, 16
 ; RV64I-NEXT:    ret
 ;
 ; RV64IM-LABEL: fold_srem_positive_odd:
 ; RV64IM:       # %bb.0:
-; RV64IM-NEXT:    sext.w a0, a0
-; RV64IM-NEXT:    lui a1, 1045903
-; RV64IM-NEXT:    addiw a1, a1, -733
-; RV64IM-NEXT:    slli a1, a1, 15
-; RV64IM-NEXT:    addi a1, a1, 1035
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, -905
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, -1767
-; RV64IM-NEXT:    mulh a1, a0, a1
-; RV64IM-NEXT:    add a1, a1, a0
-; RV64IM-NEXT:    srli a2, a1, 63
-; RV64IM-NEXT:    srai a1, a1, 6
+; RV64IM-NEXT:    sext.w a1, a0
+; RV64IM-NEXT:    lui a2, 706409
+; RV64IM-NEXT:    addiw a2, a2, 389
+; RV64IM-NEXT:    mul a1, a1, a2
+; RV64IM-NEXT:    srli a1, a1, 32
+; RV64IM-NEXT:    addw a1, a1, a0
+; RV64IM-NEXT:    srliw a2, a1, 31
+; RV64IM-NEXT:    srli a1, a1, 6
 ; RV64IM-NEXT:    add a1, a1, a2
 ; RV64IM-NEXT:    addi a2, zero, 95
 ; RV64IM-NEXT:    mul a1, a1, a2
-; RV64IM-NEXT:    sub a0, a0, a1
+; RV64IM-NEXT:    subw a0, a0, a1
 ; RV64IM-NEXT:    ret
   %1 = srem i32 %x, 95
   ret i32 %1
 }
 
 
 define i32 @fold_srem_positive_even(i32 %x) nounwind {
 ; RV32I-LABEL: fold_srem_positive_even:

 ; RV64I-NEXT:    addi a1, zero, 1060
 ; RV64I-NEXT:    call __moddi3@plt
 ; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    addi sp, sp, 16
 ; RV64I-NEXT:    ret
 ;
 ; RV64IM-LABEL: fold_srem_positive_even:
 ; RV64IM:       # %bb.0:
-; RV64IM-NEXT:    sext.w a0, a0
-; RV64IM-NEXT:    lui a1, 506482
-; RV64IM-NEXT:    addiw a1, a1, -31
-; RV64IM-NEXT:    slli a1, a1, 13
-; RV64IM-NEXT:    addi a1, a1, 711
-; RV64IM-NEXT:    slli a1, a1, 19
-; RV64IM-NEXT:    addi a1, a1, 1979
-; RV64IM-NEXT:    mulh a1, a0, a1
+; RV64IM-NEXT:    sext.w a1, a0
+; RV64IM-NEXT:    lui a2, 253241
+; RV64IM-NEXT:    addiw a2, a2, -15
+; RV64IM-NEXT:    mul a1, a1, a2
 ; RV64IM-NEXT:    srli a2, a1, 63
-; RV64IM-NEXT:    srai a1, a1, 9
+; RV64IM-NEXT:    srai a1, a1, 40
 ; RV64IM-NEXT:    add a1, a1, a2
 ; RV64IM-NEXT:    addi a2, zero, 1060
 ; RV64IM-NEXT:    mul a1, a1, a2
-; RV64IM-NEXT:    sub a0, a0, a1
+; RV64IM-NEXT:    subw a0, a0, a1
 ; RV64IM-NEXT:    ret
   %1 = srem i32 %x, 1060
   ret i32 %1
 }
 
 
 define i32 @fold_srem_negative_odd(i32 %x) nounwind {
 ; RV32I-LABEL: fold_srem_negative_odd:

 ; RV64I-NEXT:    addi a1, zero, -723
 ; RV64I-NEXT:    call __moddi3@plt
 ; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    addi sp, sp, 16
 ; RV64I-NEXT:    ret
 ;
 ; RV64IM-LABEL: fold_srem_negative_odd:
 ; RV64IM:       # %bb.0:
-; RV64IM-NEXT:    sext.w a0, a0
-; RV64IM-NEXT:    lui a1, 4781
-; RV64IM-NEXT:    addiw a1, a1, 2045
-; RV64IM-NEXT:    slli a1, a1, 13
-; RV64IM-NEXT:    addi a1, a1, 1371
-; RV64IM-NEXT:    slli a1, a1, 13
-; RV64IM-NEXT:    addi a1, a1, -11
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, -1355
-; RV64IM-NEXT:    mulh a1, a0, a1
-; RV64IM-NEXT:    sub a1, a1, a0
+; RV64IM-NEXT:    sext.w a1, a0
+; RV64IM-NEXT:    lui a2, 677296
+; RV64IM-NEXT:    addiw a2, a2, -91
+; RV64IM-NEXT:    mul a1, a1, a2
 ; RV64IM-NEXT:    srli a2, a1, 63
-; RV64IM-NEXT:    srai a1, a1, 9
+; RV64IM-NEXT:    srai a1, a1, 40
 ; RV64IM-NEXT:    add a1, a1, a2
 ; RV64IM-NEXT:    addi a2, zero, -723
 ; RV64IM-NEXT:    mul a1, a1, a2
-; RV64IM-NEXT:    sub a0, a0, a1
+; RV64IM-NEXT:    subw a0, a0, a1
 ; RV64IM-NEXT:    ret
   %1 = srem i32 %x, -723
   ret i32 %1
 }
 
 
 define i32 @fold_srem_negative_even(i32 %x) nounwind {
 ; RV32I-LABEL: fold_srem_negative_even:

 ; RV64I-NEXT:    addiw a1, a1, 1595
 ; RV64I-NEXT:    call __moddi3@plt
 ; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    addi sp, sp, 16
 ; RV64I-NEXT:    ret
 ;
 ; RV64IM-LABEL: fold_srem_negative_even:
 ; RV64IM:       # %bb.0:
-; RV64IM-NEXT:    sext.w a0, a0
-; RV64IM-NEXT:    lui a1, 1036895
-; RV64IM-NEXT:    addiw a1, a1, 999
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, 11
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, -523
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, -481
-; RV64IM-NEXT:    mulh a1, a0, a1
+; RV64IM-NEXT:    sext.w a1, a0
+; RV64IM-NEXT:    lui a2, 1036895
+; RV64IM-NEXT:    addiw a2, a2, 999
+; RV64IM-NEXT:    mul a1, a1, a2
 ; RV64IM-NEXT:    srli a2, a1, 63
-; RV64IM-NEXT:    srai a1, a1, 12
+; RV64IM-NEXT:    srai a1, a1, 40
 ; RV64IM-NEXT:    add a1, a1, a2
 ; RV64IM-NEXT:    lui a2, 1048570
 ; RV64IM-NEXT:    addiw a2, a2, 1595
 ; RV64IM-NEXT:    mul a1, a1, a2
-; RV64IM-NEXT:    sub a0, a0, a1
+; RV64IM-NEXT:    subw a0, a0, a1
 ; RV64IM-NEXT:    ret
   %1 = srem i32 %x, -22981
   ret i32 %1
 }
 
 
 ; Don't fold if we can combine srem with sdiv.
 define i32 @combine_srem_sdiv(i32 %x) nounwind {

 ; RV64I-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    addi sp, sp, 32
 ; RV64I-NEXT:    ret
 ;
 ; RV64IM-LABEL: combine_srem_sdiv:
 ; RV64IM:       # %bb.0:
 ; RV64IM-NEXT:    sext.w a1, a0
-; RV64IM-NEXT:    lui a2, 1045903
-; RV64IM-NEXT:    addiw a2, a2, -733
-; RV64IM-NEXT:    slli a2, a2, 15
-; RV64IM-NEXT:    addi a2, a2, 1035
-; RV64IM-NEXT:    slli a2, a2, 12
-; RV64IM-NEXT:    addi a2, a2, -905
-; RV64IM-NEXT:    slli a2, a2, 12
-; RV64IM-NEXT:    addi a2, a2, -1767
-; RV64IM-NEXT:    mulh a2, a1, a2
-; RV64IM-NEXT:    add a1, a2, a1
-; RV64IM-NEXT:    srli a2, a1, 63
-; RV64IM-NEXT:    srai a1, a1, 6
+; RV64IM-NEXT:    lui a2, 706409
+; RV64IM-NEXT:    addiw a2, a2, 389
+; RV64IM-NEXT:    mul a1, a1, a2
+; RV64IM-NEXT:    srli a1, a1, 32
+; RV64IM-NEXT:    addw a2, a1, a0
+; RV64IM-NEXT:    srliw a2, a2, 31
+; RV64IM-NEXT:    add a1, a1, a0
+; RV64IM-NEXT:    sraiw a1, a1, 6
 ; RV64IM-NEXT:    add a1, a1, a2
 ; RV64IM-NEXT:    addi a2, zero, 95
 ; RV64IM-NEXT:    mul a2, a1, a2
 ; RV64IM-NEXT:    sub a0, a0, a2
 ; RV64IM-NEXT:    addw a0, a0, a1
 ; RV64IM-NEXT:    ret
   %1 = srem i32 %x, 95
   %2 = sdiv i32 %x, 95

llvm/test/CodeGen/RISCV/urem-lkk.ll

 ; RV64I-NEXT:    addi a1, zero, 95
 ; RV64I-NEXT:    call __umoddi3@plt
 ; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    addi sp, sp, 16
 ; RV64I-NEXT:    ret
 ;
 ; RV64IM-LABEL: fold_urem_positive_odd:
 ; RV64IM:       # %bb.0:
-; RV64IM-NEXT:    slli a0, a0, 32
-; RV64IM-NEXT:    srli a0, a0, 32
-; RV64IM-NEXT:    lui a1, 1423
-; RV64IM-NEXT:    addiw a1, a1, -733
-; RV64IM-NEXT:    slli a1, a1, 15
-; RV64IM-NEXT:    addi a1, a1, 1035
-; RV64IM-NEXT:    slli a1, a1, 13
-; RV64IM-NEXT:    addi a1, a1, -1811
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, 561
-; RV64IM-NEXT:    mulhu a1, a0, a1
+; RV64IM-NEXT:    slli a1, a0, 32
+; RV64IM-NEXT:    srli a1, a1, 32
+; RV64IM-NEXT:    lui a2, 364242
+; RV64IM-NEXT:    addiw a2, a2, 777
+; RV64IM-NEXT:    mul a1, a1, a2
+; RV64IM-NEXT:    srli a1, a1, 32
 ; RV64IM-NEXT:    sub a2, a0, a1
-; RV64IM-NEXT:    srli a2, a2, 1
+; RV64IM-NEXT:    srliw a2, a2, 1
 ; RV64IM-NEXT:    add a1, a2, a1
 ; RV64IM-NEXT:    srli a1, a1, 6
 ; RV64IM-NEXT:    addi a2, zero, 95
 ; RV64IM-NEXT:    mul a1, a1, a2
-; RV64IM-NEXT:    sub a0, a0, a1
+; RV64IM-NEXT:    subw a0, a0, a1
 ; RV64IM-NEXT:    ret
   %1 = urem i32 %x, 95
   ret i32 %1
 }
 
 
 define i32 @fold_urem_positive_even(i32 %x) nounwind {
 ; RV32I-LABEL: fold_urem_positive_even:

 ; RV64I-NEXT:    addi a1, zero, 1060
 ; RV64I-NEXT:    call __umoddi3@plt
 ; RV64I-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    addi sp, sp, 16
 ; RV64I-NEXT:    ret
 ;
 ; RV64IM-LABEL: fold_urem_positive_even:
 ; RV64IM:       # %bb.0:
-; RV64IM-NEXT:    slli a0, a0, 32
-; RV64IM-NEXT:    srli a0, a0, 32
-; RV64IM-NEXT:    lui a1, 1048020
-; RV64IM-NEXT:    addiw a1, a1, -1793
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, 139
-; RV64IM-NEXT:    slli a1, a1, 14
-; RV64IM-NEXT:    addi a1, a1, 1793
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, -139
-; RV64IM-NEXT:    mulhu a1, a0, a1
-; RV64IM-NEXT:    srli a1, a1, 10
+; RV64IM-NEXT:    slli a1, a0, 32
+; RV64IM-NEXT:    srli a1, a1, 32
+; RV64IM-NEXT:    lui a2, 62
+; RV64IM-NEXT:    addiw a2, a2, -711
+; RV64IM-NEXT:    slli a2, a2, 14
+; RV64IM-NEXT:    addi a2, a2, -61
+; RV64IM-NEXT:    mul a1, a1, a2
+; RV64IM-NEXT:    srli a1, a1, 42
 ; RV64IM-NEXT:    addi a2, zero, 1060
 ; RV64IM-NEXT:    mul a1, a1, a2
-; RV64IM-NEXT:    sub a0, a0, a1
+; RV64IM-NEXT:    subw a0, a0, a1
 ; RV64IM-NEXT:    ret
   %1 = urem i32 %x, 1060
   ret i32 %1
 }
 
 
 ; Don't fold if we can combine urem with udiv.
 define i32 @combine_urem_udiv(i32 %x) nounwind {

 ; RV64I-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    addi sp, sp, 32
 ; RV64I-NEXT:    ret
 ;
 ; RV64IM-LABEL: combine_urem_udiv:
 ; RV64IM:       # %bb.0:
-; RV64IM-NEXT:    slli a0, a0, 32
-; RV64IM-NEXT:    srli a0, a0, 32
-; RV64IM-NEXT:    lui a1, 1423
-; RV64IM-NEXT:    addiw a1, a1, -733
-; RV64IM-NEXT:    slli a1, a1, 15
-; RV64IM-NEXT:    addi a1, a1, 1035
-; RV64IM-NEXT:    slli a1, a1, 13
-; RV64IM-NEXT:    addi a1, a1, -1811
-; RV64IM-NEXT:    slli a1, a1, 12
-; RV64IM-NEXT:    addi a1, a1, 561
-; RV64IM-NEXT:    mulhu a1, a0, a1
+; RV64IM-NEXT:    slli a1, a0, 32
+; RV64IM-NEXT:    srli a1, a1, 32
+; RV64IM-NEXT:    lui a2, 364242
+; RV64IM-NEXT:    addiw a2, a2, 777
+; RV64IM-NEXT:    mul a1, a1, a2
+; RV64IM-NEXT:    srli a1, a1, 32
 ; RV64IM-NEXT:    sub a2, a0, a1
-; RV64IM-NEXT:    srli a2, a2, 1
+; RV64IM-NEXT:    srliw a2, a2, 1
 ; RV64IM-NEXT:    add a1, a2, a1
 ; RV64IM-NEXT:    srli a1, a1, 6
 ; RV64IM-NEXT:    addi a2, zero, 95
 ; RV64IM-NEXT:    mul a2, a1, a2
 ; RV64IM-NEXT:    sub a0, a0, a2
-; RV64IM-NEXT:    add a0, a0, a1
+; RV64IM-NEXT:    addw a0, a0, a1
 ; RV64IM-NEXT:    ret
   %1 = urem i32 %x, 95
   %2 = udiv i32 %x, 95
   %3 = add i32 %1, %2
   ret i32 %3
 }
 
 ; Don't fold for divisors that are a power of two.