[SelectionDAG] Optimize expansion for rotates/funnel shifts

[RKSimon]

Summary

(Branched from D108058 as getting this completed should help unlock some other WIP patches).

If the type of a funnel shift needs to be expanded, expand it to two funnel shifts instead of regular shifts. For constant shifts, this doesn't make much difference, but for variable shifts it allows a more optimal lowering.

Also use the optimized funnel shift lowering for rotates.

Original Patch: @efriedma (Eli Friedman)

[craig.topper]

llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

4411

Does passing "false" to this really solve the original problem I raised? That makes getShiftAmountTy use pointer type, but pointer type on some targets is i16 which wouldn't be enough to represent a shift amount for an i65356. Should this instead use the getShiftAmountTyForConstant helper? The name of that function is outdated.

[RKSimon]

llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

4411

Sorry I missed that comment on the other phab

[craig.topper]

llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

4411

I'm working on a patch to fold getShiftAmountTyForConstant into TLI.getShiftAmountTy and then we'll never have to thing about this again.

[RKSimon]

Use getShiftAmountTyForConstant

[craig.topper]

Code change LGTM. X86 and RISCV tests LGTM.

llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

4402

Add a full stop at the end of this comment to match the other comments in this function.

[RKSimon]

@nemanjai The PPC codegen is no longer branchless, how much of an issue is that?

[dmgreen]

Arm parts obviously look great

[RKSimon]

rebase on D112469

[RKSimon]

rebase (for real this time....)

[nemanjai]

FWIW, the PPC changes look fine.

[nemanjai]

llvm/test/CodeGen/PowerPC/funnel-shift-rot.ll

95–96

These OR's with zero seem odd, but they are essentially just register copies so there isn't an issue with performance there.

[RKSimon]

Thanks everyone - anyone willing to officially accept this?

[craig.topper]

llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

4411

Drop the false. I don't think it is needed.

[RKSimon]

Use default getShiftAmountTy arguments

[RKSimon]

using default legal type args caused some x86 instruction reordering

[RKSimon]

I finally remembered to add the missing '.' to the end of the comment

[RKSimon]

ping?

[RKSimon]

ping - any more comments?

[lebedev.ri]

LGTM, thanks.
https://alive2.llvm.org/ce/z/TvHDB- / https://alive2.llvm.org/ce/z/yzPept

[RKSimon]

In D112443#3102370, @lebedev.ri wrote:

LGTM, thanks.
https://alive2.llvm.org/ce/z/TvHDB- / https://alive2.llvm.org/ce/z/yzPept

And thanks for those links!

llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

void DAGTypeLegalizer::ExpandIntRes_VECREDUCE(SDNode *N,
   // TODO For VECREDUCE_(AND|OR|XOR) we could split the vector and calculate
   // both halves independently.
   SDValue Res = TLI.expandVecReduce(N, DAG);
   SplitInteger(Res, Lo, Hi);
 }
 
 void DAGTypeLegalizer::ExpandIntRes_Rotate(SDNode *N,
                                            SDValue &Lo, SDValue &Hi) {
-  // Lower the rotate to shifts and ORs which can be expanded.
-  SDValue Res;
-  TLI.expandROT(N, true /*AllowVectorOps*/, Res, DAG);
+  // Delegate to funnel-shift expansion.
+  SDLoc DL(N);
+  unsigned Opcode = N->getOpcode() == ISD::ROTL ? ISD::FSHL : ISD::FSHR;
+  SDValue Res = DAG.getNode(Opcode, DL, N->getValueType(0), N->getOperand(0),
+                            N->getOperand(0), N->getOperand(1));
   SplitInteger(Res, Lo, Hi);
 }
 
-void DAGTypeLegalizer::ExpandIntRes_FunnelShift(SDNode *N,
-                                                SDValue &Lo, SDValue &Hi) {
-  // Lower the funnel shift to shifts and ORs which can be expanded.
-  SDValue Res;
-  TLI.expandFunnelShift(N, Res, DAG);
-  SplitInteger(Res, Lo, Hi);
+void DAGTypeLegalizer::ExpandIntRes_FunnelShift(SDNode *N, SDValue &Lo,
+                                                SDValue &Hi) {
+  // Values numbered from least significant to most significant.
+  SDValue In1, In2, In3, In4;
+  GetExpandedInteger(N->getOperand(0), In3, In4);
+  GetExpandedInteger(N->getOperand(1), In1, In2);
+  EVT HalfVT = In1.getValueType();
+
+  SDLoc DL(N);
+  unsigned Opc = N->getOpcode();
+  SDValue ShAmt = N->getOperand(2);
+  EVT ShAmtVT = ShAmt.getValueType();
+  EVT ShAmtCCVT = getSetCCResultType(ShAmtVT);
+
+  // If the shift amount is at least half the bitwidth, swap the inputs.
+  unsigned HalfVTBits = HalfVT.getScalarSizeInBits();
+  SDValue AndNode = DAG.getNode(ISD::AND, DL, ShAmtVT, ShAmt,
+                                DAG.getConstant(HalfVTBits, DL, ShAmtVT));
+  SDValue Cond =
+      DAG.getSetCC(DL, ShAmtCCVT, AndNode, DAG.getConstant(0, DL, ShAmtVT),
+                   Opc == ISD::FSHL ? ISD::SETNE : ISD::SETEQ);
+
+  // Expand to a pair of funnel shifts.
+  EVT NewShAmtVT = TLI.getShiftAmountTy(HalfVT, DAG.getDataLayout());
+  SDValue NewShAmt = DAG.getAnyExtOrTrunc(ShAmt, DL, NewShAmtVT);
+
+  SDValue Select1 = DAG.getNode(ISD::SELECT, DL, HalfVT, Cond, In1, In2);
+  SDValue Select2 = DAG.getNode(ISD::SELECT, DL, HalfVT, Cond, In2, In3);
+  SDValue Select3 = DAG.getNode(ISD::SELECT, DL, HalfVT, Cond, In3, In4);
+  Lo = DAG.getNode(Opc, DL, HalfVT, Select2, Select1, NewShAmt);
+  Hi = DAG.getNode(Opc, DL, HalfVT, Select3, Select2, NewShAmt);
 }
 
 void DAGTypeLegalizer::ExpandIntRes_VSCALE(SDNode *N, SDValue &Lo,
                                            SDValue &Hi) {
   EVT VT = N->getValueType(0);
   EVT HalfVT =
       EVT::getIntegerVT(*DAG.getContext(), N->getValueSizeInBits(0) / 2);
   SDLoc dl(N);

llvm/test/CodeGen/AArch64/funnel-shift.ll

 ; CHECK-NEXT:    ret
   %f = call i64 @llvm.fshl.i64(i64 %x, i64 %y, i64 %z)
   ret i64 %f
 }
 
 define i128 @fshl_i128(i128 %x, i128 %y, i128 %z) nounwind {
 ; CHECK-LABEL: fshl_i128:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    tst x4, #0x40
 ; CHECK-NEXT:    mvn w8, w4
-; CHECK-NEXT:    extr x9, x3, x2, #1
-; CHECK-NEXT:    lsr x10, x3, #1
-; CHECK-NEXT:    and x12, x8, #0x7f
-; CHECK-NEXT:    lsl x11, x10, #1
-; CHECK-NEXT:    tst x12, #0x40
-; CHECK-NEXT:    lsl x11, x11, x4
+; CHECK-NEXT:    csel x9, x2, x3, ne
+; CHECK-NEXT:    csel x10, x3, x0, ne
+; CHECK-NEXT:    lsr x9, x9, #1
+; CHECK-NEXT:    lsl x11, x10, x4
+; CHECK-NEXT:    csel x12, x0, x1, ne
+; CHECK-NEXT:    lsr x10, x10, #1
 ; CHECK-NEXT:    lsr x9, x9, x8
-; CHECK-NEXT:    orr x9, x11, x9
-; CHECK-NEXT:    lsr x11, x0, #1
-; CHECK-NEXT:    lsr x10, x10, x8
-; CHECK-NEXT:    lsl x12, x1, x4
-; CHECK-NEXT:    lsr x8, x11, x8
-; CHECK-NEXT:    and x11, x4, #0x7f
-; CHECK-NEXT:    csel x9, x10, x9, ne
-; CHECK-NEXT:    csel x10, xzr, x10, ne
-; CHECK-NEXT:    orr x8, x12, x8
-; CHECK-NEXT:    lsl x12, x0, x4
-; CHECK-NEXT:    tst x11, #0x40
-; CHECK-NEXT:    csel x8, x12, x8, ne
-; CHECK-NEXT:    csel x11, xzr, x12, ne
-; CHECK-NEXT:    orr x1, x8, x10
+; CHECK-NEXT:    lsl x12, x12, x4
+; CHECK-NEXT:    lsr x8, x10, x8
 ; CHECK-NEXT:    orr x0, x11, x9
+; CHECK-NEXT:    orr x1, x12, x8
 ; CHECK-NEXT:    ret
   %f = call i128 @llvm.fshl.i128(i128 %x, i128 %y, i128 %z)
   ret i128 %f
 }
 
 ; Verify that weird types are minimally supported.
 declare i37 @llvm.fshl.i37(i37, i37, i37)
 define i37 @fshl_i37(i37 %x, i37 %y, i37 %z) {

llvm/test/CodeGen/ARM/funnel-shift-rot.ll

 ; CHECK-NEXT:    rsb r1, r1, #0
 ; CHECK-NEXT:    ror r0, r0, r1
 ; CHECK-NEXT:    bx lr
   %f = call i32 @llvm.fshl.i32(i32 %x, i32 %x, i32 %z)
   ret i32 %f
 }
 
 define i64 @rotl_i64(i64 %x, i64 %z) {
-; SCALAR-LABEL: rotl_i64:
-; SCALAR:       @ %bb.0:
-; SCALAR-NEXT:    .save {r4, r5, r11, lr}
-; SCALAR-NEXT:    push {r4, r5, r11, lr}
-; SCALAR-NEXT:    rsb r3, r2, #0
-; SCALAR-NEXT:    and r4, r2, #63
-; SCALAR-NEXT:    and lr, r3, #63
-; SCALAR-NEXT:    rsb r3, lr, #32
-; SCALAR-NEXT:    lsl r2, r0, r4
-; SCALAR-NEXT:    lsr r12, r0, lr
-; SCALAR-NEXT:    orr r3, r12, r1, lsl r3
-; SCALAR-NEXT:    subs r12, lr, #32
-; SCALAR-NEXT:    lsrpl r3, r1, r12
-; SCALAR-NEXT:    subs r5, r4, #32
-; SCALAR-NEXT:    movwpl r2, #0
-; SCALAR-NEXT:    cmp r5, #0
-; SCALAR-NEXT:    orr r2, r2, r3
-; SCALAR-NEXT:    rsb r3, r4, #32
-; SCALAR-NEXT:    lsr r3, r0, r3
-; SCALAR-NEXT:    orr r3, r3, r1, lsl r4
-; SCALAR-NEXT:    lslpl r3, r0, r5
-; SCALAR-NEXT:    lsr r0, r1, lr
-; SCALAR-NEXT:    cmp r12, #0
-; SCALAR-NEXT:    movwpl r0, #0
-; SCALAR-NEXT:    orr r1, r3, r0
-; SCALAR-NEXT:    mov r0, r2
-; SCALAR-NEXT:    pop {r4, r5, r11, pc}
-;
-; NEON-LABEL: rotl_i64:
-; NEON:       @ %bb.0:
-; NEON-NEXT:    .save {r4, r5, r11, lr}
-; NEON-NEXT:    push {r4, r5, r11, lr}
-; NEON-NEXT:    and r12, r2, #63
-; NEON-NEXT:    rsb r2, r2, #0
-; NEON-NEXT:    rsb r3, r12, #32
-; NEON-NEXT:    and r4, r2, #63
-; NEON-NEXT:    subs lr, r12, #32
-; NEON-NEXT:    lsr r3, r0, r3
-; NEON-NEXT:    lsr r2, r1, r4
-; NEON-NEXT:    orr r3, r3, r1, lsl r12
-; NEON-NEXT:    lslpl r3, r0, lr
-; NEON-NEXT:    subs r5, r4, #32
-; NEON-NEXT:    movwpl r2, #0
-; NEON-NEXT:    cmp r5, #0
-; NEON-NEXT:    orr r2, r3, r2
-; NEON-NEXT:    lsr r3, r0, r4
-; NEON-NEXT:    rsb r4, r4, #32
-; NEON-NEXT:    lsl r0, r0, r12
-; NEON-NEXT:    orr r3, r3, r1, lsl r4
-; NEON-NEXT:    lsrpl r3, r1, r5
-; NEON-NEXT:    cmp lr, #0
-; NEON-NEXT:    movwpl r0, #0
-; NEON-NEXT:    mov r1, r2
-; NEON-NEXT:    orr r0, r0, r3
-; NEON-NEXT:    pop {r4, r5, r11, pc}
+; CHECK-LABEL: rotl_i64:
+; CHECK:       @ %bb.0:
+; CHECK-NEXT:    .save {r4, lr}
+; CHECK-NEXT:    push {r4, lr}
+; CHECK-NEXT:    ands r3, r2, #32
+; CHECK-NEXT:    and r12, r2, #31
+; CHECK-NEXT:    mov r3, r0
+; CHECK-NEXT:    mov r4, #31
+; CHECK-NEXT:    movne r3, r1
+; CHECK-NEXT:    movne r1, r0
+; CHECK-NEXT:    bic r2, r4, r2
+; CHECK-NEXT:    lsl lr, r3, r12
+; CHECK-NEXT:    lsr r0, r1, #1
+; CHECK-NEXT:    lsl r1, r1, r12
+; CHECK-NEXT:    lsr r3, r3, #1
+; CHECK-NEXT:    orr r0, lr, r0, lsr r2
+; CHECK-NEXT:    orr r1, r1, r3, lsr r2
+; CHECK-NEXT:    pop {r4, pc}
   %f = call i64 @llvm.fshl.i64(i64 %x, i64 %x, i64 %z)
   ret i64 %f
 }
 
 ; Vector rotate.
 
 define <4 x i32> @rotl_v4i32(<4 x i32> %x, <4 x i32> %z) {
 ; SCALAR-LABEL: rotl_v4i32:

 ; CHECK-NEXT:    bx lr
   %f = call i32 @llvm.fshr.i32(i32 %x, i32 %x, i32 %z)
   ret i32 %f
 }
 
 define i64 @rotr_i64(i64 %x, i64 %z) {
 ; CHECK-LABEL: rotr_i64:
 ; CHECK:       @ %bb.0:
-; CHECK-NEXT:    .save {r4, r5, r11, lr}
-; CHECK-NEXT:    push {r4, r5, r11, lr}
-; CHECK-NEXT:    and lr, r2, #63
-; CHECK-NEXT:    rsb r2, r2, #0
-; CHECK-NEXT:    rsb r3, lr, #32
-; CHECK-NEXT:    and r4, r2, #63
-; CHECK-NEXT:    lsr r12, r0, lr
-; CHECK-NEXT:    orr r3, r12, r1, lsl r3
-; CHECK-NEXT:    subs r12, lr, #32
-; CHECK-NEXT:    lsl r2, r0, r4
-; CHECK-NEXT:    lsrpl r3, r1, r12
-; CHECK-NEXT:    subs r5, r4, #32
-; CHECK-NEXT:    movwpl r2, #0
-; CHECK-NEXT:    cmp r5, #0
-; CHECK-NEXT:    orr r2, r3, r2
-; CHECK-NEXT:    rsb r3, r4, #32
-; CHECK-NEXT:    lsr r3, r0, r3
-; CHECK-NEXT:    orr r3, r3, r1, lsl r4
-; CHECK-NEXT:    lslpl r3, r0, r5
-; CHECK-NEXT:    lsr r0, r1, lr
-; CHECK-NEXT:    cmp r12, #0
-; CHECK-NEXT:    movwpl r0, #0
-; CHECK-NEXT:    orr r1, r0, r3
-; CHECK-NEXT:    mov r0, r2
-; CHECK-NEXT:    pop {r4, r5, r11, pc}
+; CHECK-NEXT:    ands r3, r2, #32
+; CHECK-NEXT:    mov r3, r1
+; CHECK-NEXT:    moveq r3, r0
+; CHECK-NEXT:    moveq r0, r1
+; CHECK-NEXT:    mov r1, #31
+; CHECK-NEXT:    lsl r12, r0, #1
+; CHECK-NEXT:    bic r1, r1, r2
+; CHECK-NEXT:    and r2, r2, #31
+; CHECK-NEXT:    lsl r12, r12, r1
+; CHECK-NEXT:    orr r12, r12, r3, lsr r2
+; CHECK-NEXT:    lsl r3, r3, #1
+; CHECK-NEXT:    lsl r1, r3, r1
+; CHECK-NEXT:    orr r1, r1, r0, lsr r2
+; CHECK-NEXT:    mov r0, r12
+; CHECK-NEXT:    bx lr
   %f = call i64 @llvm.fshr.i64(i64 %x, i64 %x, i64 %z)
   ret i64 %f
 }
 
 ; Vector rotate.
 
 define <4 x i32> @rotr_v4i32(<4 x i32> %x, <4 x i32> %z) {
 ; SCALAR-LABEL: rotr_v4i32:

llvm/test/CodeGen/ARM/funnel-shift.ll

 ; CHECK-NEXT:    bx lr
   %f = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 %z)
   ret i32 %f
 }
 
 ; Verify that weird types are minimally supported.
 declare i37 @llvm.fshl.i37(i37, i37, i37)
 define i37 @fshl_i37(i37 %x, i37 %y, i37 %z) {
-; CHECK-LABEL: fshl_i37:
-; CHECK:       @ %bb.0:
-; CHECK-NEXT:    .save {r4, r5, r6, r7, r8, lr}
-; CHECK-NEXT:    push {r4, r5, r6, r7, r8, lr}
-; CHECK-NEXT:    mov r8, r1
-; CHECK-NEXT:    mov r4, r0
-; CHECK-NEXT:    ldr r0, [sp, #24]
-; CHECK-NEXT:    mov r6, r3
-; CHECK-NEXT:    ldr r1, [sp, #28]
-; CHECK-NEXT:    mov r7, r2
-; CHECK-NEXT:    mov r2, #37
-; CHECK-NEXT:    mov r3, #0
-; CHECK-NEXT:    bl __aeabi_uldivmod
-; CHECK-NEXT:    mov r0, #63
-; CHECK-NEXT:    bic r1, r0, r2
-; CHECK-NEXT:    lsl r0, r6, #27
-; CHECK-NEXT:    lsl r3, r7, #27
-; CHECK-NEXT:    orr r0, r0, r7, lsr #5
-; CHECK-NEXT:    and r2, r2, #63
-; CHECK-NEXT:    lsrs r7, r0, #1
-; CHECK-NEXT:    rrx r0, r3
-; CHECK-NEXT:    rsb r3, r1, #32
-; CHECK-NEXT:    lsr r0, r0, r1
-; CHECK-NEXT:    lsl r6, r4, r2
-; CHECK-NEXT:    orr r0, r0, r7, lsl r3
-; CHECK-NEXT:    subs r3, r1, #32
-; CHECK-NEXT:    lsr r1, r7, r1
-; CHECK-NEXT:    lsrpl r0, r7, r3
-; CHECK-NEXT:    subs r5, r2, #32
-; CHECK-NEXT:    movwpl r6, #0
-; CHECK-NEXT:    orr r0, r6, r0
-; CHECK-NEXT:    rsb r6, r2, #32
-; CHECK-NEXT:    cmp r5, #0
-; CHECK-NEXT:    lsr r6, r4, r6
-; CHECK-NEXT:    orr r2, r6, r8, lsl r2
-; CHECK-NEXT:    lslpl r2, r4, r5
-; CHECK-NEXT:    cmp r3, #0
-; CHECK-NEXT:    movwpl r1, #0
-; CHECK-NEXT:    orr r1, r2, r1
-; CHECK-NEXT:    pop {r4, r5, r6, r7, r8, pc}
+; SCALAR-LABEL: fshl_i37:
+; SCALAR:       @ %bb.0:
+; SCALAR-NEXT:    .save {r4, r5, r6, r7, r8, lr}
+; SCALAR-NEXT:    push {r4, r5, r6, r7, r8, lr}
+; SCALAR-NEXT:    mov r4, r1
+; SCALAR-NEXT:    mov r8, r0
+; SCALAR-NEXT:    ldr r0, [sp, #24]
+; SCALAR-NEXT:    mov r5, r3
+; SCALAR-NEXT:    ldr r1, [sp, #28]
+; SCALAR-NEXT:    mov r6, r2
+; SCALAR-NEXT:    mov r2, #37
+; SCALAR-NEXT:    mov r3, #0
+; SCALAR-NEXT:    bl __aeabi_uldivmod
+; SCALAR-NEXT:    lsl r1, r5, #27
+; SCALAR-NEXT:    ands r12, r2, #32
+; SCALAR-NEXT:    orr r1, r1, r6, lsr #5
+; SCALAR-NEXT:    mov r3, r8
+; SCALAR-NEXT:    and r5, r2, #31
+; SCALAR-NEXT:    mov r0, #31
+; SCALAR-NEXT:    movne r3, r1
+; SCALAR-NEXT:    cmp r12, #0
+; SCALAR-NEXT:    bic r2, r0, r2
+; SCALAR-NEXT:    lslne r1, r6, #27
+; SCALAR-NEXT:    movne r4, r8
+; SCALAR-NEXT:    lsl r7, r3, r5
+; SCALAR-NEXT:    lsr r0, r1, #1
+; SCALAR-NEXT:    lsl r1, r4, r5
+; SCALAR-NEXT:    lsr r3, r3, #1
+; SCALAR-NEXT:    orr r0, r7, r0, lsr r2
+; SCALAR-NEXT:    orr r1, r1, r3, lsr r2
+; SCALAR-NEXT:    pop {r4, r5, r6, r7, r8, pc}
+;
+; NEON-LABEL: fshl_i37:
+; NEON:       @ %bb.0:
+; NEON-NEXT:    .save {r4, r5, r6, r7, r11, lr}
+; NEON-NEXT:    push {r4, r5, r6, r7, r11, lr}
+; NEON-NEXT:    mov r4, r1
+; NEON-NEXT:    mov r5, r0
+; NEON-NEXT:    ldr r0, [sp, #24]
+; NEON-NEXT:    mov r7, r3
+; NEON-NEXT:    ldr r1, [sp, #28]
+; NEON-NEXT:    mov r6, r2
+; NEON-NEXT:    mov r2, #37
+; NEON-NEXT:    mov r3, #0
+; NEON-NEXT:    bl __aeabi_uldivmod
+; NEON-NEXT:    mov r0, #31
+; NEON-NEXT:    bic r1, r0, r2
+; NEON-NEXT:    lsl r0, r7, #27
+; NEON-NEXT:    ands r12, r2, #32
+; NEON-NEXT:    orr r0, r0, r6, lsr #5
+; NEON-NEXT:    mov r7, r5
+; NEON-NEXT:    and r2, r2, #31
+; NEON-NEXT:    movne r7, r0
+; NEON-NEXT:    lslne r0, r6, #27
+; NEON-NEXT:    cmp r12, #0
+; NEON-NEXT:    lsl r3, r7, r2
+; NEON-NEXT:    lsr r0, r0, #1
+; NEON-NEXT:    movne r4, r5
+; NEON-NEXT:    orr r0, r3, r0, lsr r1
+; NEON-NEXT:    lsr r3, r7, #1
+; NEON-NEXT:    lsl r2, r4, r2
+; NEON-NEXT:    orr r1, r2, r3, lsr r1
+; NEON-NEXT:    pop {r4, r5, r6, r7, r11, pc}
   %f = call i37 @llvm.fshl.i37(i37 %x, i37 %y, i37 %z)
   ret i37 %f
 }
 
 ; extract(concat(0b1110000, 0b1111111) << 2) = 0b1000011
 
 declare i7 @llvm.fshl.i7(i7, i7, i7)
 define i7 @fshl_i7_const_fold() {
; CHECK-NEXT:    bx lr
   ret i32 %f
 }
 
 ; 64-bit should also work.
 
 define i64 @fshl_i64_const_overshift(i64 %x, i64 %y) {
 ; CHECK-LABEL: fshl_i64_const_overshift:
 ; CHECK:       @ %bb.0:
-; CHECK-NEXT:    lsr r1, r2, #23
-; CHECK-NEXT:    orr r2, r1, r3, lsl #9
+; CHECK-NEXT:    lsl r1, r3, #9
+; CHECK-NEXT:    orr r2, r1, r2, lsr #23
 ; CHECK-NEXT:    lsl r0, r0, #9
 ; CHECK-NEXT:    orr r1, r0, r3, lsr #23
 ; CHECK-NEXT:    mov r0, r2
 ; CHECK-NEXT:    bx lr
   %f = call i64 @llvm.fshl.i64(i64 %x, i64 %y, i64 105)
   ret i64 %f
 }
 
; CHECK-NEXT:    bx lr
   ret i32 %f
 }
 
 ; Verify that weird types are minimally supported.
 declare i37 @llvm.fshr.i37(i37, i37, i37)
 define i37 @fshr_i37(i37 %x, i37 %y, i37 %z) {
 ; CHECK-LABEL: fshr_i37:
 ; CHECK:       @ %bb.0:
-; CHECK-NEXT:    .save {r4, r5, r6, r7, r8, r9, r11, lr}
-; CHECK-NEXT:    push {r4, r5, r6, r7, r8, r9, r11, lr}
-; CHECK-NEXT:    mov r8, r1
-; CHECK-NEXT:    mov r9, r0
-; CHECK-NEXT:    ldr r0, [sp, #32]
-; CHECK-NEXT:    mov r6, r3
-; CHECK-NEXT:    ldr r1, [sp, #36]
+; CHECK-NEXT:    .save {r4, r5, r6, r7, r11, lr}
+; CHECK-NEXT:    push {r4, r5, r6, r7, r11, lr}
+; CHECK-NEXT:    mov r4, r1
+; CHECK-NEXT:    mov r6, r0
+; CHECK-NEXT:    ldr r0, [sp, #24]
+; CHECK-NEXT:    mov r5, r3
+; CHECK-NEXT:    ldr r1, [sp, #28]
 ; CHECK-NEXT:    mov r7, r2
 ; CHECK-NEXT:    mov r2, #37
 ; CHECK-NEXT:    mov r3, #0
 ; CHECK-NEXT:    bl __aeabi_uldivmod
+; CHECK-NEXT:    lsl r3, r5, #27
 ; CHECK-NEXT:    add r0, r2, #27
-; CHECK-NEXT:    lsl r6, r6, #27
-; CHECK-NEXT:    and r1, r0, #63
-; CHECK-NEXT:    lsl r2, r7, #27
-; CHECK-NEXT:    orr r7, r6, r7, lsr #5
-; CHECK-NEXT:    mov r6, #63
-; CHECK-NEXT:    rsb r3, r1, #32
-; CHECK-NEXT:    lsr r2, r2, r1
-; CHECK-NEXT:    subs r12, r1, #32
-; CHECK-NEXT:    bic r6, r6, r0
-; CHECK-NEXT:    orr r2, r2, r7, lsl r3
-; CHECK-NEXT:    lsl r5, r9, #1
-; CHECK-NEXT:    lsrpl r2, r7, r12
-; CHECK-NEXT:    lsl r0, r5, r6
-; CHECK-NEXT:    subs r4, r6, #32
-; CHECK-NEXT:    lsl r3, r8, #1
-; CHECK-NEXT:    movwpl r0, #0
-; CHECK-NEXT:    orr r3, r3, r9, lsr #31
-; CHECK-NEXT:    orr r0, r0, r2
-; CHECK-NEXT:    rsb r2, r6, #32
-; CHECK-NEXT:    cmp r4, #0
-; CHECK-NEXT:    lsr r1, r7, r1
-; CHECK-NEXT:    lsr r2, r5, r2
-; CHECK-NEXT:    orr r2, r2, r3, lsl r6
-; CHECK-NEXT:    lslpl r2, r5, r4
+; CHECK-NEXT:    orr r3, r3, r7, lsr #5
+; CHECK-NEXT:    mov r1, #31
+; CHECK-NEXT:    ands r12, r0, #32
+; CHECK-NEXT:    mov r5, r6
+; CHECK-NEXT:    moveq r5, r3
+; CHECK-NEXT:    bic r1, r1, r0
+; CHECK-NEXT:    lsl r2, r5, #1
+; CHECK-NEXT:    lsleq r3, r7, #27
 ; CHECK-NEXT:    cmp r12, #0
-; CHECK-NEXT:    movwpl r1, #0
-; CHECK-NEXT:    orr r1, r2, r1
-; CHECK-NEXT:    pop {r4, r5, r6, r7, r8, r9, r11, pc}
+; CHECK-NEXT:    and r7, r0, #31
+; CHECK-NEXT:    lsl r2, r2, r1
+; CHECK-NEXT:    moveq r4, r6
+; CHECK-NEXT:    orr r0, r2, r3, lsr r7
+; CHECK-NEXT:    lsl r2, r4, #1
+; CHECK-NEXT:    lsl r1, r2, r1
+; CHECK-NEXT:    orr r1, r1, r5, lsr r7
+; CHECK-NEXT:    pop {r4, r5, r6, r7, r11, pc}
   %f = call i37 @llvm.fshr.i37(i37 %x, i37 %y, i37 %z)
   ret i37 %f
 }
 
 ; extract(concat(0b1110000, 0b1111111) >> 2) = 0b0011111
 
 declare i7 @llvm.fshr.i7(i7, i7, i7)
 define i7 @fshr_i7_const_fold() {

llvm/test/CodeGen/Mips/funnel-shift-rot.ll

 ; CHECK-NEXT:    or $2, $1, $2
   %f = call i32 @llvm.fshl.i32(i32 %x, i32 %x, i32 %z)
   ret i32 %f
 }
 
 define i64 @rotl_i64(i64 %x, i64 %z) {
 ; CHECK-BE-LABEL: rotl_i64:
 ; CHECK-BE:       # %bb.0:
-; CHECK-BE-NEXT:    negu $1, $7
-; CHECK-BE-NEXT:    andi $3, $1, 63
-; CHECK-BE-NEXT:    srlv $6, $4, $3
-; CHECK-BE-NEXT:    andi $1, $1, 32
-; CHECK-BE-NEXT:    andi $2, $7, 63
-; CHECK-BE-NEXT:    move $8, $6
-; CHECK-BE-NEXT:    movn $8, $zero, $1
-; CHECK-BE-NEXT:    sllv $9, $4, $2
-; CHECK-BE-NEXT:    srl $10, $5, 1
-; CHECK-BE-NEXT:    not $11, $2
-; CHECK-BE-NEXT:    srlv $10, $10, $11
-; CHECK-BE-NEXT:    or $9, $9, $10
-; CHECK-BE-NEXT:    sllv $10, $5, $2
-; CHECK-BE-NEXT:    andi $7, $7, 32
-; CHECK-BE-NEXT:    movn $9, $10, $7
-; CHECK-BE-NEXT:    or $2, $9, $8
-; CHECK-BE-NEXT:    srlv $5, $5, $3
-; CHECK-BE-NEXT:    not $3, $3
-; CHECK-BE-NEXT:    sll $4, $4, 1
-; CHECK-BE-NEXT:    sllv $3, $4, $3
-; CHECK-BE-NEXT:    or $3, $3, $5
-; CHECK-BE-NEXT:    movn $3, $6, $1
-; CHECK-BE-NEXT:    movn $10, $zero, $7
+; CHECK-BE-NEXT:    srl $1, $7, 5
+; CHECK-BE-NEXT:    andi $1, $1, 1
+; CHECK-BE-NEXT:    move $3, $4
+; CHECK-BE-NEXT:    movn $3, $5, $1
+; CHECK-BE-NEXT:    andi $6, $7, 31
+; CHECK-BE-NEXT:    sllv $2, $3, $6
+; CHECK-BE-NEXT:    movn $5, $4, $1
+; CHECK-BE-NEXT:    srl $1, $5, 1
+; CHECK-BE-NEXT:    not $4, $7
+; CHECK-BE-NEXT:    andi $4, $4, 31
+; CHECK-BE-NEXT:    srlv $1, $1, $4
+; CHECK-BE-NEXT:    or $2, $2, $1
+; CHECK-BE-NEXT:    sllv $1, $5, $6
+; CHECK-BE-NEXT:    srl $3, $3, 1
+; CHECK-BE-NEXT:    srlv $3, $3, $4
 ; CHECK-BE-NEXT:    jr $ra
-; CHECK-BE-NEXT:    or $3, $10, $3
+; CHECK-BE-NEXT:    or $3, $1, $3
 ;
 ; CHECK-LE-LABEL: rotl_i64:
 ; CHECK-LE:       # %bb.0:
-; CHECK-LE-NEXT:    negu $1, $6
-; CHECK-LE-NEXT:    andi $2, $1, 63
-; CHECK-LE-NEXT:    srlv $7, $5, $2
-; CHECK-LE-NEXT:    andi $1, $1, 32
-; CHECK-LE-NEXT:    andi $3, $6, 63
-; CHECK-LE-NEXT:    move $8, $7
-; CHECK-LE-NEXT:    movn $8, $zero, $1
-; CHECK-LE-NEXT:    sllv $9, $5, $3
-; CHECK-LE-NEXT:    srl $10, $4, 1
-; CHECK-LE-NEXT:    not $11, $3
-; CHECK-LE-NEXT:    srlv $10, $10, $11
-; CHECK-LE-NEXT:    or $9, $9, $10
-; CHECK-LE-NEXT:    sllv $10, $4, $3
-; CHECK-LE-NEXT:    andi $6, $6, 32
-; CHECK-LE-NEXT:    movn $9, $10, $6
-; CHECK-LE-NEXT:    or $3, $9, $8
-; CHECK-LE-NEXT:    srlv $4, $4, $2
-; CHECK-LE-NEXT:    not $2, $2
-; CHECK-LE-NEXT:    sll $5, $5, 1
-; CHECK-LE-NEXT:    sllv $2, $5, $2
-; CHECK-LE-NEXT:    or $2, $2, $4
-; CHECK-LE-NEXT:    movn $2, $7, $1
-; CHECK-LE-NEXT:    movn $10, $zero, $6
+; CHECK-LE-NEXT:    srl $1, $6, 5
+; CHECK-LE-NEXT:    andi $1, $1, 1
+; CHECK-LE-NEXT:    move $3, $4
+; CHECK-LE-NEXT:    movn $3, $5, $1
+; CHECK-LE-NEXT:    andi $7, $6, 31
+; CHECK-LE-NEXT:    sllv $2, $3, $7
+; CHECK-LE-NEXT:    movn $5, $4, $1
+; CHECK-LE-NEXT:    srl $1, $5, 1
+; CHECK-LE-NEXT:    not $4, $6
+; CHECK-LE-NEXT:    andi $4, $4, 31
+; CHECK-LE-NEXT:    srlv $1, $1, $4
+; CHECK-LE-NEXT:    or $2, $2, $1
+; CHECK-LE-NEXT:    sllv $1, $5, $7
+; CHECK-LE-NEXT:    srl $3, $3, 1
+; CHECK-LE-NEXT:    srlv $3, $3, $4
 ; CHECK-LE-NEXT:    jr $ra
-; CHECK-LE-NEXT:    or $2, $10, $2
+; CHECK-LE-NEXT:    or $3, $1, $3
   %f = call i64 @llvm.fshl.i64(i64 %x, i64 %x, i64 %z)
   ret i64 %f
 }
 
 ; Vector rotate.
 
 define <4 x i32> @rotl_v4i32(<4 x i32> %x, <4 x i32> %z) {
 ; CHECK-LABEL: rotl_v4i32:

 ; CHECK-NEXT:    or $2, $1, $2
   %f = call i32 @llvm.fshr.i32(i32 %x, i32 %x, i32 %z)
   ret i32 %f
 }
 
 define i64 @rotr_i64(i64 %x, i64 %z) {
 ; CHECK-BE-LABEL: rotr_i64:
 ; CHECK-BE:       # %bb.0:
-; CHECK-BE-NEXT:    negu $1, $7
-; CHECK-BE-NEXT:    andi $2, $1, 63
-; CHECK-BE-NEXT:    sllv $6, $5, $2
-; CHECK-BE-NEXT:    andi $1, $1, 32
-; CHECK-BE-NEXT:    andi $3, $7, 63
-; CHECK-BE-NEXT:    move $8, $6
-; CHECK-BE-NEXT:    movn $8, $zero, $1
-; CHECK-BE-NEXT:    srlv $9, $5, $3
-; CHECK-BE-NEXT:    sll $10, $4, 1
-; CHECK-BE-NEXT:    not $11, $3
-; CHECK-BE-NEXT:    sllv $10, $10, $11
-; CHECK-BE-NEXT:    or $9, $10, $9
-; CHECK-BE-NEXT:    srlv $10, $4, $3
-; CHECK-BE-NEXT:    andi $7, $7, 32
-; CHECK-BE-NEXT:    movn $9, $10, $7
-; CHECK-BE-NEXT:    or $3, $9, $8
-; CHECK-BE-NEXT:    sllv $4, $4, $2
-; CHECK-BE-NEXT:    not $2, $2
-; CHECK-BE-NEXT:    srl $5, $5, 1
-; CHECK-BE-NEXT:    srlv $2, $5, $2
-; CHECK-BE-NEXT:    or $2, $4, $2
-; CHECK-BE-NEXT:    movn $2, $6, $1
-; CHECK-BE-NEXT:    movn $10, $zero, $7
+; CHECK-BE-NEXT:    andi $1, $7, 32
+; CHECK-BE-NEXT:    move $3, $5
+; CHECK-BE-NEXT:    movz $3, $4, $1
+; CHECK-BE-NEXT:    andi $6, $7, 31
+; CHECK-BE-NEXT:    srlv $2, $3, $6
+; CHECK-BE-NEXT:    movz $4, $5, $1
+; CHECK-BE-NEXT:    sll $1, $4, 1
+; CHECK-BE-NEXT:    not $5, $7
+; CHECK-BE-NEXT:    andi $5, $5, 31
+; CHECK-BE-NEXT:    sllv $1, $1, $5
+; CHECK-BE-NEXT:    or $2, $1, $2
+; CHECK-BE-NEXT:    srlv $1, $4, $6
+; CHECK-BE-NEXT:    sll $3, $3, 1
+; CHECK-BE-NEXT:    sllv $3, $3, $5
 ; CHECK-BE-NEXT:    jr $ra
-; CHECK-BE-NEXT:    or $2, $10, $2
+; CHECK-BE-NEXT:    or $3, $3, $1
 ;
 ; CHECK-LE-LABEL: rotr_i64:
 ; CHECK-LE:       # %bb.0:
-; CHECK-LE-NEXT:    negu $1, $6
-; CHECK-LE-NEXT:    andi $3, $1, 63
-; CHECK-LE-NEXT:    sllv $7, $4, $3
-; CHECK-LE-NEXT:    andi $1, $1, 32
-; CHECK-LE-NEXT:    andi $2, $6, 63
-; CHECK-LE-NEXT:    move $8, $7
-; CHECK-LE-NEXT:    movn $8, $zero, $1
-; CHECK-LE-NEXT:    srlv $9, $4, $2
-; CHECK-LE-NEXT:    sll $10, $5, 1
-; CHECK-LE-NEXT:    not $11, $2
-; CHECK-LE-NEXT:    sllv $10, $10, $11
-; CHECK-LE-NEXT:    or $9, $10, $9
-; CHECK-LE-NEXT:    srlv $10, $5, $2
-; CHECK-LE-NEXT:    andi $6, $6, 32
-; CHECK-LE-NEXT:    movn $9, $10, $6
-; CHECK-LE-NEXT:    or $2, $9, $8
-; CHECK-LE-NEXT:    sllv $5, $5, $3
-; CHECK-LE-NEXT:    not $3, $3
-; CHECK-LE-NEXT:    srl $4, $4, 1
-; CHECK-LE-NEXT:    srlv $3, $4, $3
-; CHECK-LE-NEXT:    or $3, $5, $3
-; CHECK-LE-NEXT:    movn $3, $7, $1
-; CHECK-LE-NEXT:    movn $10, $zero, $6
+; CHECK-LE-NEXT:    andi $1, $6, 32
+; CHECK-LE-NEXT:    move $3, $5
+; CHECK-LE-NEXT:    movz $3, $4, $1
+; CHECK-LE-NEXT:    andi $7, $6, 31
+; CHECK-LE-NEXT:    srlv $2, $3, $7
+; CHECK-LE-NEXT:    movz $4, $5, $1
+; CHECK-LE-NEXT:    sll $1, $4, 1
+; CHECK-LE-NEXT:    not $5, $6
+; CHECK-LE-NEXT:    andi $5, $5, 31
+; CHECK-LE-NEXT:    sllv $1, $1, $5
+; CHECK-LE-NEXT:    or $2, $1, $2
+; CHECK-LE-NEXT:    srlv $1, $4, $7
+; CHECK-LE-NEXT:    sll $3, $3, 1
+; CHECK-LE-NEXT:    sllv $3, $3, $5
 ; CHECK-LE-NEXT:    jr $ra
-; CHECK-LE-NEXT:    or $3, $10, $3
+; CHECK-LE-NEXT:    or $3, $3, $1
   %f = call i64 @llvm.fshr.i64(i64 %x, i64 %x, i64 %z)
   ret i64 %f
 }
 
 ; Vector rotate.
 
 define <4 x i32> @rotr_v4i32(<4 x i32> %x, <4 x i32> %z) {
 ; CHECK-LABEL: rotr_v4i32:

llvm/test/CodeGen/Mips/funnel-shift.ll

 ; CHECK-BE-NEXT:    move $18, $5
 ; CHECK-BE-NEXT:    move $19, $4
 ; CHECK-BE-NEXT:    lw $4, 56($sp)
 ; CHECK-BE-NEXT:    lw $5, 60($sp)
 ; CHECK-BE-NEXT:    addiu $6, $zero, 0
 ; CHECK-BE-NEXT:    jal __umoddi3
 ; CHECK-BE-NEXT:    addiu $7, $zero, 37
 ; CHECK-BE-NEXT:    not $1, $3
-; CHECK-BE-NEXT:    andi $2, $3, 63
-; CHECK-BE-NEXT:    not $4, $2
-; CHECK-BE-NEXT:    srl $5, $18, 1
-; CHECK-BE-NEXT:    sllv $6, $19, $2
-; CHECK-BE-NEXT:    srlv $4, $5, $4
-; CHECK-BE-NEXT:    andi $5, $1, 63
-; CHECK-BE-NEXT:    srl $7, $16, 5
-; CHECK-BE-NEXT:    sll $8, $17, 27
-; CHECK-BE-NEXT:    or $7, $8, $7
-; CHECK-BE-NEXT:    srl $8, $7, 1
-; CHECK-BE-NEXT:    srlv $9, $8, $5
-; CHECK-BE-NEXT:    andi $1, $1, 32
-; CHECK-BE-NEXT:    move $10, $9
-; CHECK-BE-NEXT:    movn $10, $zero, $1
-; CHECK-BE-NEXT:    or $4, $6, $4
-; CHECK-BE-NEXT:    sllv $6, $18, $2
-; CHECK-BE-NEXT:    andi $3, $3, 32
-; CHECK-BE-NEXT:    movn $4, $6, $3
-; CHECK-BE-NEXT:    sll $7, $7, 31
-; CHECK-BE-NEXT:    sll $2, $16, 27
-; CHECK-BE-NEXT:    srl $11, $2, 1
-; CHECK-BE-NEXT:    or $2, $4, $10
-; CHECK-BE-NEXT:    movn $6, $zero, $3
-; CHECK-BE-NEXT:    or $3, $11, $7
-; CHECK-BE-NEXT:    srlv $3, $3, $5
-; CHECK-BE-NEXT:    not $4, $5
-; CHECK-BE-NEXT:    sll $5, $8, 1
-; CHECK-BE-NEXT:    sllv $4, $5, $4
-; CHECK-BE-NEXT:    or $3, $4, $3
-; CHECK-BE-NEXT:    movn $3, $9, $1
-; CHECK-BE-NEXT:    or $3, $6, $3
+; CHECK-BE-NEXT:    srl $2, $3, 5
+; CHECK-BE-NEXT:    andi $4, $2, 1
+; CHECK-BE-NEXT:    movn $19, $18, $4
+; CHECK-BE-NEXT:    andi $3, $3, 31
+; CHECK-BE-NEXT:    sllv $2, $19, $3
+; CHECK-BE-NEXT:    andi $1, $1, 31
+; CHECK-BE-NEXT:    srl $5, $16, 5
+; CHECK-BE-NEXT:    sll $6, $17, 27
+; CHECK-BE-NEXT:    or $5, $6, $5
+; CHECK-BE-NEXT:    movn $18, $5, $4
+; CHECK-BE-NEXT:    srl $6, $18, 1
+; CHECK-BE-NEXT:    srlv $6, $6, $1
+; CHECK-BE-NEXT:    or $2, $2, $6
+; CHECK-BE-NEXT:    sllv $3, $18, $3
+; CHECK-BE-NEXT:    sll $6, $16, 27
+; CHECK-BE-NEXT:    movn $5, $6, $4
+; CHECK-BE-NEXT:    srl $4, $5, 1
+; CHECK-BE-NEXT:    srlv $1, $4, $1
+; CHECK-BE-NEXT:    or $3, $3, $1
 ; CHECK-BE-NEXT:    lw $16, 20($sp) # 4-byte Folded Reload
 ; CHECK-BE-NEXT:    lw $17, 24($sp) # 4-byte Folded Reload
 ; CHECK-BE-NEXT:    lw $18, 28($sp) # 4-byte Folded Reload
 ; CHECK-BE-NEXT:    lw $19, 32($sp) # 4-byte Folded Reload
 ; CHECK-BE-NEXT:    lw $ra, 36($sp) # 4-byte Folded Reload
 ; CHECK-BE-NEXT:    jr $ra
 ; CHECK-BE-NEXT:    addiu $sp, $sp, 40
 ;

 ; CHECK-LE-NEXT:    move $17, $6
 ; CHECK-LE-NEXT:    move $18, $5
 ; CHECK-LE-NEXT:    move $19, $4
 ; CHECK-LE-NEXT:    lw $4, 56($sp)
 ; CHECK-LE-NEXT:    lw $5, 60($sp)
 ; CHECK-LE-NEXT:    addiu $6, $zero, 37
 ; CHECK-LE-NEXT:    jal __umoddi3
 ; CHECK-LE-NEXT:    addiu $7, $zero, 0
-; CHECK-LE-NEXT:    not $1, $2
-; CHECK-LE-NEXT:    andi $3, $2, 63
-; CHECK-LE-NEXT:    not $4, $3
-; CHECK-LE-NEXT:    srl $5, $19, 1
-; CHECK-LE-NEXT:    sllv $6, $18, $3
-; CHECK-LE-NEXT:    srlv $4, $5, $4
-; CHECK-LE-NEXT:    andi $5, $1, 63
-; CHECK-LE-NEXT:    srl $7, $17, 5
-; CHECK-LE-NEXT:    sll $8, $16, 27
-; CHECK-LE-NEXT:    or $7, $8, $7
-; CHECK-LE-NEXT:    srl $8, $7, 1
-; CHECK-LE-NEXT:    srlv $9, $8, $5
-; CHECK-LE-NEXT:    andi $1, $1, 32
-; CHECK-LE-NEXT:    move $10, $9
-; CHECK-LE-NEXT:    movn $10, $zero, $1
-; CHECK-LE-NEXT:    or $4, $6, $4
-; CHECK-LE-NEXT:    sllv $6, $19, $3
-; CHECK-LE-NEXT:    andi $2, $2, 32
-; CHECK-LE-NEXT:    movn $4, $6, $2
-; CHECK-LE-NEXT:    sll $7, $7, 31
-; CHECK-LE-NEXT:    sll $3, $17, 27
-; CHECK-LE-NEXT:    srl $11, $3, 1
-; CHECK-LE-NEXT:    or $3, $4, $10
-; CHECK-LE-NEXT:    movn $6, $zero, $2
-; CHECK-LE-NEXT:    or $2, $11, $7
-; CHECK-LE-NEXT:    srlv $2, $2, $5
-; CHECK-LE-NEXT:    not $4, $5
-; CHECK-LE-NEXT:    sll $5, $8, 1
-; CHECK-LE-NEXT:    sllv $4, $5, $4
-; CHECK-LE-NEXT:    or $2, $4, $2
-; CHECK-LE-NEXT:    movn $2, $9, $1
+; CHECK-LE-NEXT:    srl $1, $2, 5
+; CHECK-LE-NEXT:    andi $1, $1, 1
+; CHECK-LE-NEXT:    srl $3, $17, 5
+; CHECK-LE-NEXT:    sll $4, $16, 27
+; CHECK-LE-NEXT:    or $3, $4, $3
+; CHECK-LE-NEXT:    move $4, $19
+; CHECK-LE-NEXT:    movn $4, $3, $1
+; CHECK-LE-NEXT:    andi $5, $2, 31
+; CHECK-LE-NEXT:    sllv $6, $4, $5
+; CHECK-LE-NEXT:    not $2, $2
+; CHECK-LE-NEXT:    andi $7, $2, 31
+; CHECK-LE-NEXT:    sll $2, $17, 27
+; CHECK-LE-NEXT:    movn $3, $2, $1
+; CHECK-LE-NEXT:    srl $2, $3, 1
+; CHECK-LE-NEXT:    srlv $2, $2, $7
 ; CHECK-LE-NEXT:    or $2, $6, $2
+; CHECK-LE-NEXT:    movn $18, $19, $1
+; CHECK-LE-NEXT:    sllv $1, $18, $5
+; CHECK-LE-NEXT:    srl $3, $4, 1
+; CHECK-LE-NEXT:    srlv $3, $3, $7
+; CHECK-LE-NEXT:    or $3, $1, $3
 ; CHECK-LE-NEXT:    lw $16, 20($sp) # 4-byte Folded Reload
 ; CHECK-LE-NEXT:    lw $17, 24($sp) # 4-byte Folded Reload
 ; CHECK-LE-NEXT:    lw $18, 28($sp) # 4-byte Folded Reload
 ; CHECK-LE-NEXT:    lw $19, 32($sp) # 4-byte Folded Reload
 ; CHECK-LE-NEXT:    lw $ra, 36($sp) # 4-byte Folded Reload
 ; CHECK-LE-NEXT:    jr $ra
 ; CHECK-LE-NEXT:    addiu $sp, $sp, 40
   %f = call i37 @llvm.fshl.i37(i37 %x, i37 %y, i37 %z)

 ; 64-bit should also work.
 
 define i64 @fshl_i64_const_overshift(i64 %x, i64 %y) {
 ; CHECK-BE-LABEL: fshl_i64_const_overshift:
 ; CHECK-BE:       # %bb.0:
 ; CHECK-BE-NEXT:    srl $1, $6, 23
 ; CHECK-BE-NEXT:    sll $2, $5, 9
 ; CHECK-BE-NEXT:    or $2, $2, $1
-; CHECK-BE-NEXT:    sll $1, $6, 9
-; CHECK-BE-NEXT:    srl $3, $7, 23
+; CHECK-BE-NEXT:    srl $1, $7, 23
+; CHECK-BE-NEXT:    sll $3, $6, 9
 ; CHECK-BE-NEXT:    jr $ra
 ; CHECK-BE-NEXT:    or $3, $3, $1
 ;
 ; CHECK-LE-LABEL: fshl_i64_const_overshift:
 ; CHECK-LE:       # %bb.0:
-; CHECK-LE-NEXT:    sll $1, $7, 9
-; CHECK-LE-NEXT:    srl $2, $6, 23
+; CHECK-LE-NEXT:    srl $1, $6, 23
+; CHECK-LE-NEXT:    sll $2, $7, 9
 ; CHECK-LE-NEXT:    or $2, $2, $1
 ; CHECK-LE-NEXT:    srl $1, $7, 23
 ; CHECK-LE-NEXT:    sll $3, $4, 9
 ; CHECK-LE-NEXT:    jr $ra
 ; CHECK-LE-NEXT:    or $3, $3, $1
   %f = call i64 @llvm.fshl.i64(i64 %x, i64 %y, i64 105)
   ret i64 %f
 }

 ; CHECK-BE-NEXT:    move $18, $5
 ; CHECK-BE-NEXT:    move $19, $4
 ; CHECK-BE-NEXT:    lw $4, 56($sp)
 ; CHECK-BE-NEXT:    lw $5, 60($sp)
 ; CHECK-BE-NEXT:    addiu $6, $zero, 0
 ; CHECK-BE-NEXT:    jal __umoddi3
 ; CHECK-BE-NEXT:    addiu $7, $zero, 37
 ; CHECK-BE-NEXT:    addiu $1, $3, 27
-; CHECK-BE-NEXT:    andi $2, $1, 63
-; CHECK-BE-NEXT:    not $3, $2
-; CHECK-BE-NEXT:    srl $4, $16, 5
-; CHECK-BE-NEXT:    sll $5, $17, 27
-; CHECK-BE-NEXT:    or $4, $5, $4
-; CHECK-BE-NEXT:    sll $5, $4, 1
-; CHECK-BE-NEXT:    sll $6, $16, 27
-; CHECK-BE-NEXT:    srlv $6, $6, $2
-; CHECK-BE-NEXT:    sllv $3, $5, $3
-; CHECK-BE-NEXT:    not $5, $1
-; CHECK-BE-NEXT:    andi $7, $5, 63
-; CHECK-BE-NEXT:    sll $8, $18, 1
-; CHECK-BE-NEXT:    sllv $8, $8, $7
-; CHECK-BE-NEXT:    andi $5, $5, 32
-; CHECK-BE-NEXT:    move $9, $8
-; CHECK-BE-NEXT:    movn $9, $zero, $5
-; CHECK-BE-NEXT:    or $3, $3, $6
-; CHECK-BE-NEXT:    srlv $2, $4, $2
-; CHECK-BE-NEXT:    andi $1, $1, 32
-; CHECK-BE-NEXT:    movn $3, $2, $1
-; CHECK-BE-NEXT:    srl $4, $18, 31
+; CHECK-BE-NEXT:    andi $3, $1, 32
+; CHECK-BE-NEXT:    srl $2, $16, 5
+; CHECK-BE-NEXT:    sll $4, $17, 27
+; CHECK-BE-NEXT:    or $4, $4, $2
+; CHECK-BE-NEXT:    movz $19, $18, $3
+; CHECK-BE-NEXT:    movz $18, $4, $3
+; CHECK-BE-NEXT:    andi $5, $1, 31
+; CHECK-BE-NEXT:    srlv $2, $18, $5
+; CHECK-BE-NEXT:    not $1, $1
+; CHECK-BE-NEXT:    andi $1, $1, 31
 ; CHECK-BE-NEXT:    sll $6, $19, 1
-; CHECK-BE-NEXT:    or $4, $6, $4
-; CHECK-BE-NEXT:    or $3, $9, $3
-; CHECK-BE-NEXT:    movn $2, $zero, $1
-; CHECK-BE-NEXT:    sllv $1, $4, $7
-; CHECK-BE-NEXT:    not $4, $7
-; CHECK-BE-NEXT:    lui $6, 32767
-; CHECK-BE-NEXT:    ori $6, $6, 65535
-; CHECK-BE-NEXT:    and $6, $18, $6
-; CHECK-BE-NEXT:    srlv $4, $6, $4
-; CHECK-BE-NEXT:    or $1, $1, $4
-; CHECK-BE-NEXT:    movn $1, $8, $5
-; CHECK-BE-NEXT:    or $2, $1, $2
+; CHECK-BE-NEXT:    sllv $6, $6, $1
+; CHECK-BE-NEXT:    or $2, $6, $2
+; CHECK-BE-NEXT:    sll $6, $16, 27
+; CHECK-BE-NEXT:    movz $4, $6, $3
+; CHECK-BE-NEXT:    srlv $3, $4, $5
+; CHECK-BE-NEXT:    sll $4, $18, 1
+; CHECK-BE-NEXT:    sllv $1, $4, $1
+; CHECK-BE-NEXT:    or $3, $1, $3
 ; CHECK-BE-NEXT:    lw $16, 20($sp) # 4-byte Folded Reload
 ; CHECK-BE-NEXT:    lw $17, 24($sp) # 4-byte Folded Reload
 ; CHECK-BE-NEXT:    lw $18, 28($sp) # 4-byte Folded Reload
 ; CHECK-BE-NEXT:    lw $19, 32($sp) # 4-byte Folded Reload
 ; CHECK-BE-NEXT:    lw $ra, 36($sp) # 4-byte Folded Reload
 ; CHECK-BE-NEXT:    jr $ra
 ; CHECK-BE-NEXT:    addiu $sp, $sp, 40
 ;

 ; CHECK-LE-NEXT:    move $18, $5
 ; CHECK-LE-NEXT:    move $19, $4
 ; CHECK-LE-NEXT:    lw $4, 56($sp)
 ; CHECK-LE-NEXT:    lw $5, 60($sp)
 ; CHECK-LE-NEXT:    addiu $6, $zero, 37
 ; CHECK-LE-NEXT:    jal __umoddi3
 ; CHECK-LE-NEXT:    addiu $7, $zero, 0
 ; CHECK-LE-NEXT:    addiu $1, $2, 27
-; CHECK-LE-NEXT:    andi $2, $1, 63
-; CHECK-LE-NEXT:    not $3, $2
-; CHECK-LE-NEXT:    srl $4, $17, 5
-; CHECK-LE-NEXT:    sll $5, $16, 27
-; CHECK-LE-NEXT:    or $4, $5, $4
-; CHECK-LE-NEXT:    sll $5, $4, 1
-; CHECK-LE-NEXT:    sll $6, $17, 27
-; CHECK-LE-NEXT:    srlv $6, $6, $2
-; CHECK-LE-NEXT:    sllv $3, $5, $3
-; CHECK-LE-NEXT:    not $5, $1
-; CHECK-LE-NEXT:    andi $7, $5, 63
-; CHECK-LE-NEXT:    sll $8, $19, 1
-; CHECK-LE-NEXT:    sllv $8, $8, $7
-; CHECK-LE-NEXT:    andi $5, $5, 32
-; CHECK-LE-NEXT:    move $9, $8
-; CHECK-LE-NEXT:    movn $9, $zero, $5
-; CHECK-LE-NEXT:    or $3, $3, $6
-; CHECK-LE-NEXT:    srlv $4, $4, $2
-; CHECK-LE-NEXT:    andi $1, $1, 32
-; CHECK-LE-NEXT:    movn $3, $4, $1
-; CHECK-LE-NEXT:    srl $2, $19, 31
-; CHECK-LE-NEXT:    sll $6, $18, 1
-; CHECK-LE-NEXT:    or $6, $6, $2
-; CHECK-LE-NEXT:    or $2, $9, $3
-; CHECK-LE-NEXT:    movn $4, $zero, $1
-; CHECK-LE-NEXT:    sllv $1, $6, $7
-; CHECK-LE-NEXT:    not $3, $7
-; CHECK-LE-NEXT:    lui $6, 32767
-; CHECK-LE-NEXT:    ori $6, $6, 65535
-; CHECK-LE-NEXT:    and $6, $19, $6
-; CHECK-LE-NEXT:    srlv $3, $6, $3
-; CHECK-LE-NEXT:    or $1, $1, $3
-; CHECK-LE-NEXT:    movn $1, $8, $5
+; CHECK-LE-NEXT:    andi $3, $1, 32
+; CHECK-LE-NEXT:    srl $2, $17, 5
+; CHECK-LE-NEXT:    sll $4, $16, 27
+; CHECK-LE-NEXT:    or $2, $4, $2
+; CHECK-LE-NEXT:    sll $4, $17, 27
+; CHECK-LE-NEXT:    move $5, $19
+; CHECK-LE-NEXT:    movz $5, $2, $3
+; CHECK-LE-NEXT:    movz $2, $4, $3
+; CHECK-LE-NEXT:    andi $4, $1, 31
+; CHECK-LE-NEXT:    srlv $2, $2, $4
+; CHECK-LE-NEXT:    not $1, $1
+; CHECK-LE-NEXT:    andi $1, $1, 31
+; CHECK-LE-NEXT:    sll $6, $5, 1
+; CHECK-LE-NEXT:    sllv $6, $6, $1
+; CHECK-LE-NEXT:    or $2, $6, $2
+; CHECK-LE-NEXT:    srlv $4, $5, $4
+; CHECK-LE-NEXT:    movz $18, $19, $3
+; CHECK-LE-NEXT:    sll $3, $18, 1
+; CHECK-LE-NEXT:    sllv $1, $3, $1
 ; CHECK-LE-NEXT:    or $3, $1, $4
 ; CHECK-LE-NEXT:    lw $16, 20($sp) # 4-byte Folded Reload
 ; CHECK-LE-NEXT:    lw $17, 24($sp) # 4-byte Folded Reload
 ; CHECK-LE-NEXT:    lw $18, 28($sp) # 4-byte Folded Reload
 ; CHECK-LE-NEXT:    lw $19, 32($sp) # 4-byte Folded Reload
 ; CHECK-LE-NEXT:    lw $ra, 36($sp) # 4-byte Folded Reload
 ; CHECK-LE-NEXT:    jr $ra
 ; CHECK-LE-NEXT:    addiu $sp, $sp, 40

llvm/test/CodeGen/PowerPC/funnel-shift-rot.ll

 ; CHECK-NEXT:    blr
   %f = call i32 @llvm.fshl.i32(i32 %x, i32 %x, i32 %z)
   ret i32 %f
 }
 
 define i64 @rotl_i64(i64 %x, i64 %z) {
 ; CHECK32_32-LABEL: rotl_i64:
 ; CHECK32_32:       # %bb.0:
-; CHECK32_32-NEXT:    clrlwi 5, 6, 26
-; CHECK32_32-NEXT:    subfic 8, 5, 32
-; CHECK32_32-NEXT:    neg 6, 6
-; CHECK32_32-NEXT:    slw 7, 3, 5
-; CHECK32_32-NEXT:    addi 9, 5, -32
-; CHECK32_32-NEXT:    srw 8, 4, 8
-; CHECK32_32-NEXT:    clrlwi 6, 6, 26
-; CHECK32_32-NEXT:    slw 9, 4, 9
-; CHECK32_32-NEXT:    or 7, 7, 8
-; CHECK32_32-NEXT:    subfic 8, 6, 32
-; CHECK32_32-NEXT:    or 7, 7, 9
-; CHECK32_32-NEXT:    addi 9, 6, -32
-; CHECK32_32-NEXT:    slw 8, 3, 8
-; CHECK32_32-NEXT:    srw 9, 3, 9
-; CHECK32_32-NEXT:    srw 3, 3, 6
-; CHECK32_32-NEXT:    srw 6, 4, 6
-; CHECK32_32-NEXT:    or 6, 6, 8
-; CHECK32_32-NEXT:    or 6, 6, 9
-; CHECK32_32-NEXT:    slw 4, 4, 5
-; CHECK32_32-NEXT:    or 3, 7, 3
-; CHECK32_32-NEXT:    or 4, 4, 6
+; CHECK32_32-NEXT:    andi. 5, 6, 32
+; CHECK32_32-NEXT:    clrlwi 5, 6, 27
+; CHECK32_32-NEXT:    subfic 6, 5, 32
+; CHECK32_32-NEXT:    bc 12, 2, .LBB4_2
+; CHECK32_32-NEXT:  # %bb.1:
+; CHECK32_32-NEXT:    ori 7, 3, 0
+; CHECK32_32-NEXT:    ori 3, 4, 0
+; CHECK32_32-NEXT:    b .LBB4_3
+; CHECK32_32-NEXT:  .LBB4_2:
+; CHECK32_32-NEXT:    addi 7, 4, 0
+; CHECK32_32-NEXT:  .LBB4_3:
+; CHECK32_32-NEXT:    srw 4, 7, 6
+; CHECK32_32-NEXT:    slw 8, 3, 5
+; CHECK32_32-NEXT:    srw 6, 3, 6
+; CHECK32_32-NEXT:    slw 5, 7, 5
+; CHECK32_32-NEXT:    or 3, 8, 4
+; CHECK32_32-NEXT:    or 4, 5, 6
 ; CHECK32_32-NEXT:    blr
 ;
 ; CHECK32_64-LABEL: rotl_i64:
 ; CHECK32_64:       # %bb.0:
-; CHECK32_64-NEXT:    clrlwi 5, 6, 26
-; CHECK32_64-NEXT:    neg 6, 6
-; CHECK32_64-NEXT:    subfic 8, 5, 32
-; CHECK32_64-NEXT:    slw 7, 3, 5
-; CHECK32_64-NEXT:    clrlwi 6, 6, 26
-; CHECK32_64-NEXT:    srw 8, 4, 8
-; CHECK32_64-NEXT:    addi 9, 5, -32
-; CHECK32_64-NEXT:    or 7, 7, 8
-; CHECK32_64-NEXT:    subfic 8, 6, 32
-; CHECK32_64-NEXT:    slw 5, 4, 5
-; CHECK32_64-NEXT:    slw 9, 4, 9
-; CHECK32_64-NEXT:    srw 10, 3, 6
-; CHECK32_64-NEXT:    srw 4, 4, 6
-; CHECK32_64-NEXT:    addi 6, 6, -32
-; CHECK32_64-NEXT:    slw 8, 3, 8
-; CHECK32_64-NEXT:    srw 3, 3, 6
-; CHECK32_64-NEXT:    or 4, 4, 8
-; CHECK32_64-NEXT:    or 6, 7, 9
-; CHECK32_64-NEXT:    or 4, 4, 3
-; CHECK32_64-NEXT:    or 3, 6, 10
-; CHECK32_64-NEXT:    or 4, 5, 4
+; CHECK32_64-NEXT:    andi. 5, 6, 32
+; CHECK32_64-NEXT:    clrlwi 5, 6, 27
+; CHECK32_64-NEXT:    bc 12, 2, .LBB4_2
+; CHECK32_64-NEXT:  # %bb.1:
+; CHECK32_64-NEXT:    ori 7, 3, 0
+; CHECK32_64-NEXT:    ori 3, 4, 0
+; CHECK32_64-NEXT:    b .LBB4_3
+; CHECK32_64-NEXT:  .LBB4_2:
+; CHECK32_64-NEXT:    addi 7, 4, 0
+; CHECK32_64-NEXT:  .LBB4_3:
+; CHECK32_64-NEXT:    subfic 6, 5, 32
+; CHECK32_64-NEXT:    srw 4, 7, 6
+; CHECK32_64-NEXT:    slw 8, 3, 5
+; CHECK32_64-NEXT:    srw 6, 3, 6
+; CHECK32_64-NEXT:    slw 5, 7, 5
+; CHECK32_64-NEXT:    or 3, 8, 4
+; CHECK32_64-NEXT:    or 4, 5, 6
 ; CHECK32_64-NEXT:    blr
 ;
 ; CHECK64-LABEL: rotl_i64:
 ; CHECK64:       # %bb.0:
 ; CHECK64-NEXT:    rotld 3, 3, 4
 ; CHECK64-NEXT:    blr
   %f = call i64 @llvm.fshl.i64(i64 %x, i64 %x, i64 %z)
   ret i64 %f

 ; CHECK-NEXT:    blr
   %f = call i32 @llvm.fshr.i32(i32 %x, i32 %x, i32 %z)
   ret i32 %f
 }
 
 define i64 @rotr_i64(i64 %x, i64 %z) {
 ; CHECK32_32-LABEL: rotr_i64:
 ; CHECK32_32:       # %bb.0:
-; CHECK32_32-NEXT:    clrlwi 5, 6, 26
-; CHECK32_32-NEXT:    subfic 8, 5, 32
-; CHECK32_32-NEXT:    neg 6, 6
-; CHECK32_32-NEXT:    srw 7, 4, 5
-; CHECK32_32-NEXT:    addi 9, 5, -32
-; CHECK32_32-NEXT:    slw 8, 3, 8
-; CHECK32_32-NEXT:    clrlwi 6, 6, 26
-; CHECK32_32-NEXT:    srw 9, 3, 9
-; CHECK32_32-NEXT:    or 7, 7, 8
-; CHECK32_32-NEXT:    subfic 8, 6, 32
-; CHECK32_32-NEXT:    or 7, 7, 9
-; CHECK32_32-NEXT:    addi 9, 6, -32
-; CHECK32_32-NEXT:    srw 8, 4, 8
-; CHECK32_32-NEXT:    slw 9, 4, 9
-; CHECK32_32-NEXT:    slw 4, 4, 6
-; CHECK32_32-NEXT:    slw 6, 3, 6
-; CHECK32_32-NEXT:    or 6, 6, 8
-; CHECK32_32-NEXT:    or 6, 6, 9
-; CHECK32_32-NEXT:    srw 3, 3, 5
-; CHECK32_32-NEXT:    or 4, 7, 4
-; CHECK32_32-NEXT:    or 3, 3, 6
+; CHECK32_32-NEXT:    andi. 5, 6, 32
+; CHECK32_32-NEXT:    clrlwi 5, 6, 27
+; CHECK32_32-NEXT:    subfic 6, 5, 32
+; CHECK32_32-NEXT:    bc 12, 2, .LBB11_2
+; CHECK32_32-NEXT:  # %bb.1:
+; CHECK32_32-NEXT:    ori 7, 4, 0
+; CHECK32_32-NEXT:    b .LBB11_3
+; CHECK32_32-NEXT:  .LBB11_2:
+; CHECK32_32-NEXT:    addi 7, 3, 0
+; CHECK32_32-NEXT:    addi 3, 4, 0
+; CHECK32_32-NEXT:  .LBB11_3:
+; CHECK32_32-NEXT:    srw 4, 7, 5
+; CHECK32_32-NEXT:    slw 8, 3, 6
+; CHECK32_32-NEXT:    srw 5, 3, 5
+; CHECK32_32-NEXT:    slw 6, 7, 6
+; CHECK32_32-NEXT:    or 3, 8, 4
+; CHECK32_32-NEXT:    or 4, 6, 5
 ; CHECK32_32-NEXT:    blr
 ;
 ; CHECK32_64-LABEL: rotr_i64:
 ; CHECK32_64:       # %bb.0:
-; CHECK32_64-NEXT:    clrlwi 5, 6, 26
-; CHECK32_64-NEXT:    neg 6, 6
-; CHECK32_64-NEXT:    subfic 8, 5, 32
-; CHECK32_64-NEXT:    srw 7, 4, 5
-; CHECK32_64-NEXT:    clrlwi 6, 6, 26
-; CHECK32_64-NEXT:    slw 8, 3, 8
-; CHECK32_64-NEXT:    addi 9, 5, -32
-; CHECK32_64-NEXT:    or 7, 7, 8
-; CHECK32_64-NEXT:    subfic 8, 6, 32
+; CHECK32_64-NEXT:    andi. 5, 6, 32
+; CHECK32_64-NEXT:    clrlwi 5, 6, 27
+; CHECK32_64-NEXT:    bc 12, 2, .LBB11_2
+; CHECK32_64-NEXT:  # %bb.1:
+; CHECK32_64-NEXT:    ori 7, 4, 0
+; CHECK32_64-NEXT:    b .LBB11_3
+; CHECK32_64-NEXT:  .LBB11_2:
+; CHECK32_64-NEXT:    addi 7, 3, 0
+; CHECK32_64-NEXT:    addi 3, 4, 0
+; CHECK32_64-NEXT:  .LBB11_3:
+; CHECK32_64-NEXT:    subfic 6, 5, 32
+; CHECK32_64-NEXT:    srw 4, 7, 5
+; CHECK32_64-NEXT:    slw 8, 3, 6
 ; CHECK32_64-NEXT:    srw 5, 3, 5
-; CHECK32_64-NEXT:    srw 9, 3, 9
-; CHECK32_64-NEXT:    slw 10, 4, 6
-; CHECK32_64-NEXT:    slw 3, 3, 6
-; CHECK32_64-NEXT:    addi 6, 6, -32
-; CHECK32_64-NEXT:    srw 8, 4, 8
-; CHECK32_64-NEXT:    slw 4, 4, 6
-; CHECK32_64-NEXT:    or 3, 3, 8
-; CHECK32_64-NEXT:    or 6, 7, 9
-; CHECK32_64-NEXT:    or 3, 3, 4
-; CHECK32_64-NEXT:    or 4, 6, 10
-; CHECK32_64-NEXT:    or 3, 5, 3
+; CHECK32_64-NEXT:    slw 6, 7, 6
+; CHECK32_64-NEXT:    or 3, 8, 4
+; CHECK32_64-NEXT:    or 4, 6, 5
 ; CHECK32_64-NEXT:    blr
 ;
 ; CHECK64-LABEL: rotr_i64:
 ; CHECK64:       # %bb.0:
 ; CHECK64-NEXT:    neg 4, 4
 ; CHECK64-NEXT:    rotld 3, 3, 4
 ; CHECK64-NEXT:    blr
   %f = call i64 @llvm.fshr.i64(i64 %x, i64 %x, i64 %z)

llvm/test/CodeGen/PowerPC/funnel-shift.ll

 ; CHECK64-NEXT:    blr
   %f = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 %z)
   ret i32 %f
 }
 
 define i64 @fshl_i64(i64 %x, i64 %y, i64 %z) {
 ; CHECK32_32-LABEL: fshl_i64:
 ; CHECK32_32:       # %bb.0:
-; CHECK32_32-NEXT:    clrlwi 7, 8, 26
-; CHECK32_32-NEXT:    not 8, 8
-; CHECK32_32-NEXT:    rotlwi 6, 6, 31
-; CHECK32_32-NEXT:    subfic 10, 7, 32
-; CHECK32_32-NEXT:    srwi 9, 5, 1
+; CHECK32_32-NEXT:    andi. 7, 8, 32
+; CHECK32_32-NEXT:    clrlwi 7, 8, 27
+; CHECK32_32-NEXT:    subfic 8, 7, 32
+; CHECK32_32-NEXT:    bc 12, 2, .LBB1_2
+; CHECK32_32-NEXT:  # %bb.1:
+; CHECK32_32-NEXT:    ori 9, 5, 0
+; CHECK32_32-NEXT:    ori 3, 4, 0
+; CHECK32_32-NEXT:    ori 4, 6, 0
+; CHECK32_32-NEXT:    b .LBB1_3
+; CHECK32_32-NEXT:  .LBB1_2:
+; CHECK32_32-NEXT:    addi 9, 4, 0
+; CHECK32_32-NEXT:    addi 4, 5, 0
+; CHECK32_32-NEXT:  .LBB1_3:
+; CHECK32_32-NEXT:    srw 5, 9, 8
 ; CHECK32_32-NEXT:    slw 3, 3, 7
-; CHECK32_32-NEXT:    clrlwi 8, 8, 26
-; CHECK32_32-NEXT:    rlwimi 6, 5, 31, 0, 0
-; CHECK32_32-NEXT:    srw 5, 4, 10
-; CHECK32_32-NEXT:    srw 10, 9, 8
-; CHECK32_32-NEXT:    srw 6, 6, 8
+; CHECK32_32-NEXT:    srw 4, 4, 8
+; CHECK32_32-NEXT:    slw 6, 9, 7
 ; CHECK32_32-NEXT:    or 3, 3, 5
-; CHECK32_32-NEXT:    subfic 5, 8, 32
-; CHECK32_32-NEXT:    addi 8, 8, -32
-; CHECK32_32-NEXT:    slw 5, 9, 5
-; CHECK32_32-NEXT:    srw 8, 9, 8
-; CHECK32_32-NEXT:    addi 9, 7, -32
-; CHECK32_32-NEXT:    slw 9, 4, 9
-; CHECK32_32-NEXT:    or 5, 6, 5
-; CHECK32_32-NEXT:    or 3, 3, 9
-; CHECK32_32-NEXT:    or 5, 5, 8
-; CHECK32_32-NEXT:    slw 4, 4, 7
-; CHECK32_32-NEXT:    or 3, 3, 10
-; CHECK32_32-NEXT:    or 4, 4, 5
+; CHECK32_32-NEXT:    or 4, 6, 4
 ; CHECK32_32-NEXT:    blr
 ;
 ; CHECK32_64-LABEL: fshl_i64:
 ; CHECK32_64:       # %bb.0:
-; CHECK32_64-NEXT:    clrlwi 7, 8, 26
-; CHECK32_64-NEXT:    not 8, 8
-; CHECK32_64-NEXT:    subfic 9, 7, 32
-; CHECK32_64-NEXT:    rotlwi 6, 6, 31
+; CHECK32_64-NEXT:    andi. 7, 8, 32
+; CHECK32_64-NEXT:    clrlwi 7, 8, 27
+; CHECK32_64-NEXT:    bc 12, 2, .LBB1_2
+; CHECK32_64-NEXT:  # %bb.1:
+; CHECK32_64-NEXT:    ori 9, 5, 0
+; CHECK32_64-NEXT:    ori 3, 4, 0
+; CHECK32_64-NEXT:    ori 5, 6, 0
+; CHECK32_64-NEXT:    b .LBB1_3
+; CHECK32_64-NEXT:  .LBB1_2:
+; CHECK32_64-NEXT:    addi 9, 4, 0
+; CHECK32_64-NEXT:  .LBB1_3:
+; CHECK32_64-NEXT:    subfic 8, 7, 32
+; CHECK32_64-NEXT:    srw 4, 9, 8
 ; CHECK32_64-NEXT:    slw 3, 3, 7
-; CHECK32_64-NEXT:    clrlwi 8, 8, 26
-; CHECK32_64-NEXT:    srw 9, 4, 9
-; CHECK32_64-NEXT:    rlwimi 6, 5, 31, 0, 0
-; CHECK32_64-NEXT:    srwi 5, 5, 1
-; CHECK32_64-NEXT:    addi 10, 7, -32
-; CHECK32_64-NEXT:    or 3, 3, 9
-; CHECK32_64-NEXT:    subfic 9, 8, 32
-; CHECK32_64-NEXT:    slw 7, 4, 7
-; CHECK32_64-NEXT:    slw 4, 4, 10
-; CHECK32_64-NEXT:    srw 10, 5, 8
-; CHECK32_64-NEXT:    srw 6, 6, 8
-; CHECK32_64-NEXT:    addi 8, 8, -32
-; CHECK32_64-NEXT:    slw 9, 5, 9
 ; CHECK32_64-NEXT:    srw 5, 5, 8
-; CHECK32_64-NEXT:    or 6, 6, 9
+; CHECK32_64-NEXT:    slw 6, 9, 7
 ; CHECK32_64-NEXT:    or 3, 3, 4
 ; CHECK32_64-NEXT:    or 4, 6, 5
-; CHECK32_64-NEXT:    or 3, 3, 10
-; CHECK32_64-NEXT:    or 4, 7, 4
 ; CHECK32_64-NEXT:    blr
 ;
 ; CHECK64-LABEL: fshl_i64:
 ; CHECK64:       # %bb.0:
 ; CHECK64-NEXT:    clrlwi 5, 5, 26
 ; CHECK64-NEXT:    subfic 6, 5, 64
 ; CHECK64-NEXT:    sld 3, 3, 5
 ; CHECK64-NEXT:    srd 4, 4, 6
 ; CHECK64-NEXT:    or 3, 3, 4
 ; CHECK64-NEXT:    blr
   %f = call i64 @llvm.fshl.i64(i64 %x, i64 %y, i64 %z)
   ret i64 %f
 }
 
 define i128 @fshl_i128(i128 %x, i128 %y, i128 %z) nounwind {
 ; CHECK32_32-LABEL: fshl_i128:
 ; CHECK32_32:       # %bb.0:
-; CHECK32_32-NEXT:    stwu 1, -64(1)
-; CHECK32_32-NEXT:    lwz 0, 84(1)
-; CHECK32_32-NEXT:    rotlwi 12, 8, 31
-; CHECK32_32-NEXT:    srwi 11, 7, 1
-; CHECK32_32-NEXT:    rlwimi 12, 7, 31, 0, 0
-; CHECK32_32-NEXT:    andi. 7, 0, 127
-; CHECK32_32-NEXT:    stw 27, 44(1) # 4-byte Folded Spill
-; CHECK32_32-NEXT:    rotlwi 10, 10, 31
-; CHECK32_32-NEXT:    stw 30, 56(1) # 4-byte Folded Spill
-; CHECK32_32-NEXT:    rotlwi 30, 9, 31
-; CHECK32_32-NEXT:    subfic 27, 7, 32
-; CHECK32_32-NEXT:    stw 22, 24(1) # 4-byte Folded Spill
-; CHECK32_32-NEXT:    rlwimi 10, 9, 31, 0, 0
-; CHECK32_32-NEXT:    stw 25, 36(1) # 4-byte Folded Spill
-; CHECK32_32-NEXT:    rlwimi 30, 8, 31, 0, 0
-; CHECK32_32-NEXT:    stw 28, 48(1) # 4-byte Folded Spill
-; CHECK32_32-NEXT:    not 8, 0
-; CHECK32_32-NEXT:    subfic 9, 7, 96
-; CHECK32_32-NEXT:    addi 0, 7, -64
-; CHECK32_32-NEXT:    slw 28, 3, 7
-; CHECK32_32-NEXT:    subfic 25, 7, 64
-; CHECK32_32-NEXT:    srw 22, 4, 27
-; CHECK32_32-NEXT:    stw 20, 16(1) # 4-byte Folded Spill
-; CHECK32_32-NEXT:    srw 9, 6, 9
-; CHECK32_32-NEXT:    stw 23, 28(1) # 4-byte Folded Spill
-; CHECK32_32-NEXT:    slw 23, 5, 0
-; CHECK32_32-NEXT:    stw 29, 52(1) # 4-byte Folded Spill
-; CHECK32_32-NEXT:    addi 29, 7, -96
-; CHECK32_32-NEXT:    srw 20, 5, 25
-; CHECK32_32-NEXT:    or 28, 28, 22
-; CHECK32_32-NEXT:    srw 22, 6, 25
-; CHECK32_32-NEXT:    subfic 25, 25, 32
-; CHECK32_32-NEXT:    stw 24, 32(1) # 4-byte Folded Spill
+; CHECK32_32-NEXT:    lwz 11, 20(1)
+; CHECK32_32-NEXT:    andi. 12, 11, 64
 ; CHECK32_32-NEXT:    mcrf 1, 0
-; CHECK32_32-NEXT:    stw 26, 40(1) # 4-byte Folded Spill
-; CHECK32_32-NEXT:    addi 26, 7, -32
-; CHECK32_32-NEXT:    andi. 8, 8, 127
-; CHECK32_32-NEXT:    slw 24, 5, 7
-; CHECK32_32-NEXT:    slw 29, 6, 29
-; CHECK32_32-NEXT:    or 9, 23, 9
-; CHECK32_32-NEXT:    slw 25, 5, 25
-; CHECK32_32-NEXT:    srw 5, 5, 27
-; CHECK32_32-NEXT:    srw 27, 6, 27
-; CHECK32_32-NEXT:    stw 21, 20(1) # 4-byte Folded Spill
-; CHECK32_32-NEXT:    slw 21, 4, 26
-; CHECK32_32-NEXT:    subfic 23, 8, 32
-; CHECK32_32-NEXT:    or 27, 24, 27
-; CHECK32_32-NEXT:    subfic 24, 8, 96
-; CHECK32_32-NEXT:    or 9, 9, 29
-; CHECK32_32-NEXT:    addi 29, 8, -64
-; CHECK32_32-NEXT:    or 25, 22, 25
-; CHECK32_32-NEXT:    stw 19, 12(1) # 4-byte Folded Spill
-; CHECK32_32-NEXT:    srw 19, 12, 8
-; CHECK32_32-NEXT:    or 28, 28, 21
-; CHECK32_32-NEXT:    slw 21, 11, 23
-; CHECK32_32-NEXT:    slw 24, 11, 24
-; CHECK32_32-NEXT:    srw 22, 12, 29
-; CHECK32_32-NEXT:    slw 26, 6, 26
-; CHECK32_32-NEXT:    or 5, 25, 5
-; CHECK32_32-NEXT:    addi 25, 8, -96
-; CHECK32_32-NEXT:    or 21, 19, 21
-; CHECK32_32-NEXT:    srw 19, 10, 8
-; CHECK32_32-NEXT:    or 24, 22, 24
-; CHECK32_32-NEXT:    slw 22, 30, 23
-; CHECK32_32-NEXT:    or 27, 27, 26
-; CHECK32_32-NEXT:    addi 26, 8, -32
-; CHECK32_32-NEXT:    srw 25, 11, 25
-; CHECK32_32-NEXT:    or 22, 19, 22
-; CHECK32_32-NEXT:    or 28, 28, 20
-; CHECK32_32-NEXT:    srw 20, 11, 26
-; CHECK32_32-NEXT:    or 25, 24, 25
-; CHECK32_32-NEXT:    subfic 24, 8, 64
-; CHECK32_32-NEXT:    srw 26, 30, 26
-; CHECK32_32-NEXT:    or 26, 22, 26
-; CHECK32_32-NEXT:    subfic 22, 24, 32
-; CHECK32_32-NEXT:    slw 23, 12, 23
-; CHECK32_32-NEXT:    srw 22, 12, 22
-; CHECK32_32-NEXT:    slw 12, 12, 24
-; CHECK32_32-NEXT:    slw 24, 11, 24
-; CHECK32_32-NEXT:    cmplwi 5, 7, 64
-; CHECK32_32-NEXT:    or 24, 24, 22
-; CHECK32_32-NEXT:    slw 22, 6, 0
-; CHECK32_32-NEXT:    slw 6, 6, 7
-; CHECK32_32-NEXT:    slw 7, 4, 7
-; CHECK32_32-NEXT:    srw 29, 11, 29
-; CHECK32_32-NEXT:    srw 11, 11, 8
-; CHECK32_32-NEXT:    cmplwi 6, 8, 64
-; CHECK32_32-NEXT:    srw 8, 30, 8
-; CHECK32_32-NEXT:    or 5, 7, 5
-; CHECK32_32-NEXT:    or 7, 26, 12
-; CHECK32_32-NEXT:    or 12, 24, 23
-; CHECK32_32-NEXT:    bc 12, 20, .LBB2_1
-; CHECK32_32-NEXT:    b .LBB2_2
-; CHECK32_32-NEXT:  .LBB2_1:
-; CHECK32_32-NEXT:    addi 9, 28, 0
+; CHECK32_32-NEXT:    andi. 12, 11, 32
+; CHECK32_32-NEXT:    clrlwi 11, 11, 27
+; CHECK32_32-NEXT:    bc 12, 6, .LBB2_2
+; CHECK32_32-NEXT:  # %bb.1:
+; CHECK32_32-NEXT:    ori 4, 6, 0
+; CHECK32_32-NEXT:    ori 12, 7, 0
+; CHECK32_32-NEXT:    ori 3, 5, 0
+; CHECK32_32-NEXT:    ori 5, 8, 0
+; CHECK32_32-NEXT:    ori 6, 9, 0
+; CHECK32_32-NEXT:    ori 7, 10, 0
+; CHECK32_32-NEXT:    b .LBB2_3
 ; CHECK32_32-NEXT:  .LBB2_2:
-; CHECK32_32-NEXT:    li 28, 0
-; CHECK32_32-NEXT:    bc 12, 20, .LBB2_4
-; CHECK32_32-NEXT:  # %bb.3:
-; CHECK32_32-NEXT:    ori 5, 22, 0
-; CHECK32_32-NEXT:    b .LBB2_4
-; CHECK32_32-NEXT:  .LBB2_4:
-; CHECK32_32-NEXT:    bc 12, 24, .LBB2_6
-; CHECK32_32-NEXT:  # %bb.5:
-; CHECK32_32-NEXT:    ori 7, 25, 0
+; CHECK32_32-NEXT:    addi 12, 5, 0
+; CHECK32_32-NEXT:    addi 5, 6, 0
+; CHECK32_32-NEXT:    addi 6, 7, 0
+; CHECK32_32-NEXT:    addi 7, 8, 0
+; CHECK32_32-NEXT:  .LBB2_3:
+; CHECK32_32-NEXT:    subfic 8, 11, 32
+; CHECK32_32-NEXT:    bc 12, 2, .LBB2_5
+; CHECK32_32-NEXT:  # %bb.4:
+; CHECK32_32-NEXT:    ori 9, 12, 0
+; CHECK32_32-NEXT:    ori 3, 4, 0
+; CHECK32_32-NEXT:    ori 4, 5, 0
+; CHECK32_32-NEXT:    ori 5, 6, 0
+; CHECK32_32-NEXT:    ori 6, 7, 0
 ; CHECK32_32-NEXT:    b .LBB2_6
+; CHECK32_32-NEXT:  .LBB2_5:
+; CHECK32_32-NEXT:    addi 9, 4, 0
+; CHECK32_32-NEXT:    addi 4, 12, 0
 ; CHECK32_32-NEXT:  .LBB2_6:
-; CHECK32_32-NEXT:    or 8, 8, 12
-; CHECK32_32-NEXT:    or 21, 21, 20
-; CHECK32_32-NEXT:    bc 12, 20, .LBB2_8
-; CHECK32_32-NEXT:  # %bb.7:
-; CHECK32_32-NEXT:    ori 6, 28, 0
-; CHECK32_32-NEXT:    b .LBB2_8
-; CHECK32_32-NEXT:  .LBB2_8:
-; CHECK32_32-NEXT:    bc 12, 6, .LBB2_10
-; CHECK32_32-NEXT:  # %bb.9:
-; CHECK32_32-NEXT:    ori 4, 5, 0
-; CHECK32_32-NEXT:    b .LBB2_10
-; CHECK32_32-NEXT:  .LBB2_10:
-; CHECK32_32-NEXT:    bc 12, 2, .LBB2_12
-; CHECK32_32-NEXT:  # %bb.11:
-; CHECK32_32-NEXT:    ori 5, 7, 0
-; CHECK32_32-NEXT:    b .LBB2_13
-; CHECK32_32-NEXT:  .LBB2_12:
-; CHECK32_32-NEXT:    addi 5, 10, 0
-; CHECK32_32-NEXT:  .LBB2_13:
-; CHECK32_32-NEXT:    bc 12, 24, .LBB2_15
-; CHECK32_32-NEXT:  # %bb.14:
-; CHECK32_32-NEXT:    ori 7, 29, 0
-; CHECK32_32-NEXT:    ori 11, 28, 0
-; CHECK32_32-NEXT:    ori 0, 28, 0
-; CHECK32_32-NEXT:    b .LBB2_16
-; CHECK32_32-NEXT:  .LBB2_15:
-; CHECK32_32-NEXT:    addi 7, 8, 0
-; CHECK32_32-NEXT:    addi 0, 21, 0
-; CHECK32_32-NEXT:  .LBB2_16:
-; CHECK32_32-NEXT:    bc 12, 6, .LBB2_18
-; CHECK32_32-NEXT:  # %bb.17:
-; CHECK32_32-NEXT:    ori 3, 9, 0
-; CHECK32_32-NEXT:    b .LBB2_18
-; CHECK32_32-NEXT:  .LBB2_18:
-; CHECK32_32-NEXT:    lwz 29, 52(1) # 4-byte Folded Reload
-; CHECK32_32-NEXT:    or 6, 6, 5
-; CHECK32_32-NEXT:    bc 12, 20, .LBB2_20
-; CHECK32_32-NEXT:  # %bb.19:
-; CHECK32_32-NEXT:    ori 5, 28, 0
-; CHECK32_32-NEXT:    b .LBB2_21
-; CHECK32_32-NEXT:  .LBB2_20:
-; CHECK32_32-NEXT:    addi 5, 27, 0
-; CHECK32_32-NEXT:  .LBB2_21:
-; CHECK32_32-NEXT:    bc 12, 2, .LBB2_22
-; CHECK32_32-NEXT:    b .LBB2_23
-; CHECK32_32-NEXT:  .LBB2_22:
-; CHECK32_32-NEXT:    addi 7, 30, 0
-; CHECK32_32-NEXT:  .LBB2_23:
-; CHECK32_32-NEXT:    or 3, 3, 11
-; CHECK32_32-NEXT:    or 4, 4, 0
-; CHECK32_32-NEXT:    or 5, 5, 7
-; CHECK32_32-NEXT:    lwz 30, 56(1) # 4-byte Folded Reload
-; CHECK32_32-NEXT:    lwz 28, 48(1) # 4-byte Folded Reload
-; CHECK32_32-NEXT:    lwz 27, 44(1) # 4-byte Folded Reload
-; CHECK32_32-NEXT:    lwz 26, 40(1) # 4-byte Folded Reload
-; CHECK32_32-NEXT:    lwz 25, 36(1) # 4-byte Folded Reload
-; CHECK32_32-NEXT:    lwz 24, 32(1) # 4-byte Folded Reload
-; CHECK32_32-NEXT:    lwz 23, 28(1) # 4-byte Folded Reload
-; CHECK32_32-NEXT:    lwz 22, 24(1) # 4-byte Folded Reload
-; CHECK32_32-NEXT:    lwz 21, 20(1) # 4-byte Folded Reload
-; CHECK32_32-NEXT:    lwz 20, 16(1) # 4-byte Folded Reload
-; CHECK32_32-NEXT:    lwz 19, 12(1) # 4-byte Folded Reload
-; CHECK32_32-NEXT:    addi 1, 1, 64
+; CHECK32_32-NEXT:    srw 7, 9, 8
+; CHECK32_32-NEXT:    slw 3, 3, 11
+; CHECK32_32-NEXT:    srw 10, 4, 8
+; CHECK32_32-NEXT:    slw 9, 9, 11
+; CHECK32_32-NEXT:    srw 12, 5, 8
+; CHECK32_32-NEXT:    slw 0, 4, 11
+; CHECK32_32-NEXT:    srw 6, 6, 8
+; CHECK32_32-NEXT:    slw 8, 5, 11
+; CHECK32_32-NEXT:    or 3, 3, 7
+; CHECK32_32-NEXT:    or 4, 9, 10
+; CHECK32_32-NEXT:    or 5, 0, 12
+; CHECK32_32-NEXT:    or 6, 8, 6
 ; CHECK32_32-NEXT:    blr
 ;
 ; CHECK32_64-LABEL: fshl_i128:
 ; CHECK32_64:       # %bb.0:
-; CHECK32_64-NEXT:    stwu 1, -64(1)
-; CHECK32_64-NEXT:    lwz 12, 84(1)
-; CHECK32_64-NEXT:    rotlwi 11, 8, 31
-; CHECK32_64-NEXT:    rotlwi 10, 10, 31
-; CHECK32_64-NEXT:    rlwimi 10, 9, 31, 0, 0
-; CHECK32_64-NEXT:    rlwimi 11, 7, 31, 0, 0
-; CHECK32_64-NEXT:    stw 30, 56(1) # 4-byte Folded Spill
-; CHECK32_64-NEXT:    rotlwi 30, 9, 31
-; CHECK32_64-NEXT:    stw 27, 44(1) # 4-byte Folded Spill
-; CHECK32_64-NEXT:    not 9, 12
-; CHECK32_64-NEXT:    rlwimi 30, 8, 31, 0, 0
-; CHECK32_64-NEXT:    andi. 8, 12, 127
-; CHECK32_64-NEXT:    stw 22, 24(1) # 4-byte Folded Spill
+; CHECK32_64-NEXT:    stwu 1, -16(1)
+; CHECK32_64-NEXT:    lwz 11, 36(1)
+; CHECK32_64-NEXT:    andi. 12, 11, 64
+; CHECK32_64-NEXT:    stw 30, 8(1) # 4-byte Folded Spill
 ; CHECK32_64-NEXT:    mcrf 1, 0
-; CHECK32_64-NEXT:    subfic 12, 8, 96
-; CHECK32_64-NEXT:    addi 0, 8, -64
-; CHECK32_64-NEXT:    subfic 27, 8, 32
-; CHECK32_64-NEXT:    stw 23, 28(1) # 4-byte Folded Spill
-; CHECK32_64-NEXT:    andi. 9, 9, 127
-; CHECK32_64-NEXT:    srw 12, 6, 12
-; CHECK32_64-NEXT:    stw 25, 36(1) # 4-byte Folded Spill
-; CHECK32_64-NEXT:    subfic 25, 8, 64
-; CHECK32_64-NEXT:    slw 23, 5, 0
-; CHECK32_64-NEXT:    stw 26, 40(1) # 4-byte Folded Spill
-; CHECK32_64-NEXT:    addi 26, 8, -32
-; CHECK32_64-NEXT:    srw 22, 4, 27
-; CHECK32_64-NEXT:    srwi 7, 7, 1
-; CHECK32_64-NEXT:    or 12, 23, 12
-; CHECK32_64-NEXT:    stw 28, 48(1) # 4-byte Folded Spill
-; CHECK32_64-NEXT:    slw 28, 3, 8
-; CHECK32_64-NEXT:    srw 23, 6, 25
-; CHECK32_64-NEXT:    stw 18, 8(1) # 4-byte Folded Spill
-; CHECK32_64-NEXT:    subfic 18, 9, 32
-; CHECK32_64-NEXT:    or 28, 28, 22
-; CHECK32_64-NEXT:    srw 22, 5, 27
-; CHECK32_64-NEXT:    srw 27, 6, 27
-; CHECK32_64-NEXT:    stw 20, 16(1) # 4-byte Folded Spill
-; CHECK32_64-NEXT:    srw 20, 5, 25
-; CHECK32_64-NEXT:    subfic 25, 25, 32
-; CHECK32_64-NEXT:    stw 21, 20(1) # 4-byte Folded Spill
-; CHECK32_64-NEXT:    slw 21, 4, 26
-; CHECK32_64-NEXT:    slw 26, 6, 26
-; CHECK32_64-NEXT:    or 28, 28, 21
-; CHECK32_64-NEXT:    slw 21, 7, 18
-; CHECK32_64-NEXT:    stw 24, 32(1) # 4-byte Folded Spill
-; CHECK32_64-NEXT:    slw 24, 5, 8
-; CHECK32_64-NEXT:    slw 5, 5, 25
-; CHECK32_64-NEXT:    stw 29, 52(1) # 4-byte Folded Spill
-; CHECK32_64-NEXT:    addi 29, 8, -96
-; CHECK32_64-NEXT:    subfic 25, 9, 96
-; CHECK32_64-NEXT:    slw 29, 6, 29
-; CHECK32_64-NEXT:    or 27, 24, 27
-; CHECK32_64-NEXT:    stw 19, 12(1) # 4-byte Folded Spill
-; CHECK32_64-NEXT:    srw 19, 11, 9
-; CHECK32_64-NEXT:    addi 24, 9, -64
-; CHECK32_64-NEXT:    or 12, 12, 29
-; CHECK32_64-NEXT:    srw 29, 10, 9
-; CHECK32_64-NEXT:    slw 25, 7, 25
-; CHECK32_64-NEXT:    or 21, 19, 21
-; CHECK32_64-NEXT:    srw 19, 11, 24
-; CHECK32_64-NEXT:    or 5, 23, 5
-; CHECK32_64-NEXT:    slw 23, 30, 18
-; CHECK32_64-NEXT:    or 27, 27, 26
-; CHECK32_64-NEXT:    addi 26, 9, -96
-; CHECK32_64-NEXT:    or 25, 19, 25
-; CHECK32_64-NEXT:    lwz 19, 12(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    or 29, 29, 23
-; CHECK32_64-NEXT:    addi 23, 9, -32
-; CHECK32_64-NEXT:    srw 26, 7, 26
-; CHECK32_64-NEXT:    or 28, 28, 20
-; CHECK32_64-NEXT:    srw 20, 7, 23
-; CHECK32_64-NEXT:    or 26, 25, 26
-; CHECK32_64-NEXT:    subfic 25, 9, 64
-; CHECK32_64-NEXT:    srw 23, 30, 23
-; CHECK32_64-NEXT:    or 29, 29, 23
-; CHECK32_64-NEXT:    subfic 23, 25, 32
-; CHECK32_64-NEXT:    or 5, 5, 22
-; CHECK32_64-NEXT:    slw 22, 11, 18
-; CHECK32_64-NEXT:    lwz 18, 8(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    srw 23, 11, 23
-; CHECK32_64-NEXT:    slw 11, 11, 25
-; CHECK32_64-NEXT:    slw 25, 7, 25
-; CHECK32_64-NEXT:    cmplwi 5, 8, 64
-; CHECK32_64-NEXT:    bc 12, 20, .LBB2_1
-; CHECK32_64-NEXT:    b .LBB2_2
-; CHECK32_64-NEXT:  .LBB2_1:
-; CHECK32_64-NEXT:    addi 12, 28, 0
+; CHECK32_64-NEXT:    clrlwi 12, 11, 27
+; CHECK32_64-NEXT:    andi. 11, 11, 32
+; CHECK32_64-NEXT:    bc 12, 6, .LBB2_2
+; CHECK32_64-NEXT:  # %bb.1:
+; CHECK32_64-NEXT:    ori 4, 6, 0
+; CHECK32_64-NEXT:    ori 30, 7, 0
+; CHECK32_64-NEXT:    ori 3, 5, 0
+; CHECK32_64-NEXT:    ori 7, 9, 0
+; CHECK32_64-NEXT:    b .LBB2_3
 ; CHECK32_64-NEXT:  .LBB2_2:
-; CHECK32_64-NEXT:    lwz 28, 48(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    or 25, 25, 23
-; CHECK32_64-NEXT:    bc 12, 6, .LBB2_4
-; CHECK32_64-NEXT:  # %bb.3:
-; CHECK32_64-NEXT:    ori 3, 12, 0
-; CHECK32_64-NEXT:    b .LBB2_4
-; CHECK32_64-NEXT:  .LBB2_4:
-; CHECK32_64-NEXT:    slw 23, 6, 0
-; CHECK32_64-NEXT:    slw 6, 6, 8
-; CHECK32_64-NEXT:    slw 8, 4, 8
-; CHECK32_64-NEXT:    cmplwi 6, 9, 64
-; CHECK32_64-NEXT:    or 5, 8, 5
-; CHECK32_64-NEXT:    bc 12, 20, .LBB2_6
-; CHECK32_64-NEXT:  # %bb.5:
-; CHECK32_64-NEXT:    ori 5, 23, 0
+; CHECK32_64-NEXT:    addi 30, 5, 0
+; CHECK32_64-NEXT:  .LBB2_3:
+; CHECK32_64-NEXT:    bc 12, 2, .LBB2_5
+; CHECK32_64-NEXT:  # %bb.4:
+; CHECK32_64-NEXT:    ori 5, 30, 0
+; CHECK32_64-NEXT:    ori 3, 4, 0
 ; CHECK32_64-NEXT:    b .LBB2_6
+; CHECK32_64-NEXT:  .LBB2_5:
+; CHECK32_64-NEXT:    addi 5, 4, 0
 ; CHECK32_64-NEXT:  .LBB2_6:
-; CHECK32_64-NEXT:    lwz 23, 28(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    li 8, 0
-; CHECK32_64-NEXT:    srw 24, 7, 24
 ; CHECK32_64-NEXT:    bc 12, 6, .LBB2_8
 ; CHECK32_64-NEXT:  # %bb.7:
-; CHECK32_64-NEXT:    ori 4, 5, 0
-; CHECK32_64-NEXT:    b .LBB2_8
+; CHECK32_64-NEXT:    ori 4, 8, 0
+; CHECK32_64-NEXT:    ori 8, 10, 0
+; CHECK32_64-NEXT:    b .LBB2_9
 ; CHECK32_64-NEXT:  .LBB2_8:
-; CHECK32_64-NEXT:    bc 12, 20, .LBB2_10
-; CHECK32_64-NEXT:  # %bb.9:
-; CHECK32_64-NEXT:    ori 6, 8, 0
-; CHECK32_64-NEXT:    b .LBB2_10
-; CHECK32_64-NEXT:  .LBB2_10:
-; CHECK32_64-NEXT:    srw 7, 7, 9
-; CHECK32_64-NEXT:    srw 9, 30, 9
-; CHECK32_64-NEXT:    bc 12, 24, .LBB2_12
-; CHECK32_64-NEXT:  # %bb.11:
+; CHECK32_64-NEXT:    addi 4, 6, 0
+; CHECK32_64-NEXT:  .LBB2_9:
+; CHECK32_64-NEXT:    subfic 11, 12, 32
+; CHECK32_64-NEXT:    bc 12, 2, .LBB2_11
+; CHECK32_64-NEXT:  # %bb.10:
+; CHECK32_64-NEXT:    ori 0, 4, 0
+; CHECK32_64-NEXT:    ori 4, 7, 0
 ; CHECK32_64-NEXT:    ori 7, 8, 0
 ; CHECK32_64-NEXT:    b .LBB2_12
+; CHECK32_64-NEXT:  .LBB2_11:
+; CHECK32_64-NEXT:    addi 0, 30, 0
 ; CHECK32_64-NEXT:  .LBB2_12:
-; CHECK32_64-NEXT:    or 0, 25, 22
-; CHECK32_64-NEXT:    or 11, 29, 11
-; CHECK32_64-NEXT:    lwz 29, 52(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    bc 12, 24, .LBB2_14
-; CHECK32_64-NEXT:  # %bb.13:
-; CHECK32_64-NEXT:    ori 5, 26, 0
-; CHECK32_64-NEXT:    b .LBB2_15
-; CHECK32_64-NEXT:  .LBB2_14:
-; CHECK32_64-NEXT:    addi 5, 11, 0
-; CHECK32_64-NEXT:  .LBB2_15:
-; CHECK32_64-NEXT:    or 9, 9, 0
-; CHECK32_64-NEXT:    or 21, 21, 20
-; CHECK32_64-NEXT:    bc 12, 2, .LBB2_16
-; CHECK32_64-NEXT:    b .LBB2_17
-; CHECK32_64-NEXT:  .LBB2_16:
-; CHECK32_64-NEXT:    addi 5, 10, 0
-; CHECK32_64-NEXT:  .LBB2_17:
-; CHECK32_64-NEXT:    bc 12, 24, .LBB2_19
-; CHECK32_64-NEXT:  # %bb.18:
-; CHECK32_64-NEXT:    ori 0, 8, 0
-; CHECK32_64-NEXT:    b .LBB2_20
-; CHECK32_64-NEXT:  .LBB2_19:
-; CHECK32_64-NEXT:    addi 0, 21, 0
-; CHECK32_64-NEXT:  .LBB2_20:
-; CHECK32_64-NEXT:    bc 12, 20, .LBB2_21
-; CHECK32_64-NEXT:    b .LBB2_22
-; CHECK32_64-NEXT:  .LBB2_21:
-; CHECK32_64-NEXT:    addi 8, 27, 0
-; CHECK32_64-NEXT:  .LBB2_22:
-; CHECK32_64-NEXT:    lwz 27, 44(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    or 3, 3, 7
-; CHECK32_64-NEXT:    bc 12, 24, .LBB2_24
-; CHECK32_64-NEXT:  # %bb.23:
-; CHECK32_64-NEXT:    ori 7, 24, 0
-; CHECK32_64-NEXT:    b .LBB2_25
-; CHECK32_64-NEXT:  .LBB2_24:
-; CHECK32_64-NEXT:    addi 7, 9, 0
-; CHECK32_64-NEXT:  .LBB2_25:
-; CHECK32_64-NEXT:    or 4, 4, 0
-; CHECK32_64-NEXT:    bc 12, 2, .LBB2_26
-; CHECK32_64-NEXT:    b .LBB2_27
-; CHECK32_64-NEXT:  .LBB2_26:
-; CHECK32_64-NEXT:    addi 7, 30, 0
-; CHECK32_64-NEXT:  .LBB2_27:
-; CHECK32_64-NEXT:    or 6, 6, 5
-; CHECK32_64-NEXT:    or 5, 8, 7
-; CHECK32_64-NEXT:    lwz 30, 56(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    lwz 26, 40(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    lwz 25, 36(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    lwz 24, 32(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    lwz 22, 24(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    lwz 21, 20(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    lwz 20, 16(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    addi 1, 1, 64
+; CHECK32_64-NEXT:    srw 6, 5, 11
+; CHECK32_64-NEXT:    lwz 30, 8(1) # 4-byte Folded Reload
+; CHECK32_64-NEXT:    slw 3, 3, 12
+; CHECK32_64-NEXT:    srw 9, 0, 11
+; CHECK32_64-NEXT:    slw 5, 5, 12
+; CHECK32_64-NEXT:    srw 10, 4, 11
+; CHECK32_64-NEXT:    slw 0, 0, 12
+; CHECK32_64-NEXT:    srw 7, 7, 11
+; CHECK32_64-NEXT:    slw 8, 4, 12
+; CHECK32_64-NEXT:    or 3, 3, 6
+; CHECK32_64-NEXT:    or 4, 5, 9
+; CHECK32_64-NEXT:    or 5, 0, 10
+; CHECK32_64-NEXT:    or 6, 8, 7
+; CHECK32_64-NEXT:    addi 1, 1, 16
 ; CHECK32_64-NEXT:    blr
 ;
 ; CHECK64-LABEL: fshl_i128:
 ; CHECK64:       # %bb.0:
-; CHECK64-NEXT:    clrlwi 8, 7, 25
-; CHECK64-NEXT:    rotldi 5, 5, 63
-; CHECK64-NEXT:    not 7, 7
-; CHECK64-NEXT:    rldicl 9, 6, 63, 1
-; CHECK64-NEXT:    subfic 10, 8, 64
-; CHECK64-NEXT:    addi 11, 8, -64
-; CHECK64-NEXT:    rldimi 5, 6, 63, 0
-; CHECK64-NEXT:    clrlwi 6, 7, 25
-; CHECK64-NEXT:    srd 7, 3, 10
-; CHECK64-NEXT:    sld 10, 3, 11
-; CHECK64-NEXT:    subfic 11, 6, 64
-; CHECK64-NEXT:    addi 12, 6, -64
-; CHECK64-NEXT:    sld 4, 4, 8
-; CHECK64-NEXT:    srd 5, 5, 6
-; CHECK64-NEXT:    sld 11, 9, 11
-; CHECK64-NEXT:    or 4, 4, 7
-; CHECK64-NEXT:    or 5, 5, 11
-; CHECK64-NEXT:    srd 7, 9, 12
-; CHECK64-NEXT:    or 4, 4, 10
-; CHECK64-NEXT:    srd 6, 9, 6
-; CHECK64-NEXT:    or 5, 5, 7
-; CHECK64-NEXT:    sld 3, 3, 8
-; CHECK64-NEXT:    or 4, 4, 6
-; CHECK64-NEXT:    or 3, 3, 5
+; CHECK64-NEXT:    andi. 8, 7, 64
+; CHECK64-NEXT:    clrlwi 7, 7, 26
+; CHECK64-NEXT:    iseleq 5, 6, 5
+; CHECK64-NEXT:    subfic 8, 7, 64
+; CHECK64-NEXT:    iseleq 6, 3, 6
+; CHECK64-NEXT:    iseleq 3, 4, 3
+; CHECK64-NEXT:    srd 4, 5, 8
+; CHECK64-NEXT:    sld 5, 6, 7
+; CHECK64-NEXT:    srd 6, 6, 8
+; CHECK64-NEXT:    sld 7, 3, 7
+; CHECK64-NEXT:    or 3, 5, 4
+; CHECK64-NEXT:    or 4, 7, 6
 ; CHECK64-NEXT:    blr
   %f = call i128 @llvm.fshl.i128(i128 %x, i128 %y, i128 %z)
   ret i128 %f
 }
 
 ; Verify that weird types are minimally supported.
 declare i37 @llvm.fshl.i37(i37, i37, i37)
 define i37 @fshl_i37(i37 %x, i37 %y, i37 %z) {

 ; CHECK32_32-NEXT:    mr 29, 5
 ; CHECK32_32-NEXT:    stw 30, 24(1) # 4-byte Folded Spill
 ; CHECK32_32-NEXT:    mr 30, 6
 ; CHECK32_32-NEXT:    mr 3, 7
 ; CHECK32_32-NEXT:    mr 4, 8
 ; CHECK32_32-NEXT:    li 5, 0
 ; CHECK32_32-NEXT:    li 6, 37
 ; CHECK32_32-NEXT:    bl __umoddi3
-; CHECK32_32-NEXT:    clrlwi 6, 4, 26
-; CHECK32_32-NEXT:    srwi 3, 30, 6
-; CHECK32_32-NEXT:    not 4, 4
-; CHECK32_32-NEXT:    subfic 8, 6, 32
-; CHECK32_32-NEXT:    slwi 5, 30, 26
-; CHECK32_32-NEXT:    rlwimi 3, 29, 26, 1, 5
-; CHECK32_32-NEXT:    slw 7, 27, 6
-; CHECK32_32-NEXT:    clrlwi 4, 4, 26
-; CHECK32_32-NEXT:    srw 8, 28, 8
-; CHECK32_32-NEXT:    srw 9, 3, 4
-; CHECK32_32-NEXT:    srw 5, 5, 4
-; CHECK32_32-NEXT:    or 7, 7, 8
-; CHECK32_32-NEXT:    subfic 8, 4, 32
-; CHECK32_32-NEXT:    addi 4, 4, -32
-; CHECK32_32-NEXT:    slw 8, 3, 8
-; CHECK32_32-NEXT:    srw 4, 3, 4
-; CHECK32_32-NEXT:    addi 3, 6, -32
-; CHECK32_32-NEXT:    slw 3, 28, 3
-; CHECK32_32-NEXT:    or 5, 5, 8
-; CHECK32_32-NEXT:    or 3, 7, 3
-; CHECK32_32-NEXT:    or 4, 5, 4
-; CHECK32_32-NEXT:    slw 5, 28, 6
-; CHECK32_32-NEXT:    or 3, 3, 9
-; CHECK32_32-NEXT:    or 4, 5, 4
+; CHECK32_32-NEXT:    rotlwi 3, 30, 27
+; CHECK32_32-NEXT:    slwi 5, 30, 27
+; CHECK32_32-NEXT:    andi. 6, 4, 32
+; CHECK32_32-NEXT:    rlwimi 3, 29, 27, 0, 4
+; CHECK32_32-NEXT:    clrlwi 4, 4, 27
+; CHECK32_32-NEXT:    subfic 6, 4, 32
+; CHECK32_32-NEXT:    bc 12, 2, .LBB3_2
+; CHECK32_32-NEXT:  # %bb.1:
+; CHECK32_32-NEXT:    ori 7, 3, 0
+; CHECK32_32-NEXT:    ori 8, 28, 0
+; CHECK32_32-NEXT:    ori 3, 5, 0
+; CHECK32_32-NEXT:    b .LBB3_3
+; CHECK32_32-NEXT:  .LBB3_2:
+; CHECK32_32-NEXT:    addi 7, 28, 0
+; CHECK32_32-NEXT:    addi 8, 27, 0
+; CHECK32_32-NEXT:  .LBB3_3:
 ; CHECK32_32-NEXT:    lwz 30, 24(1) # 4-byte Folded Reload
+; CHECK32_32-NEXT:    srw 5, 7, 6
+; CHECK32_32-NEXT:    slw 8, 8, 4
+; CHECK32_32-NEXT:    srw 6, 3, 6
+; CHECK32_32-NEXT:    slw 4, 7, 4
+; CHECK32_32-NEXT:    or 3, 8, 5
+; CHECK32_32-NEXT:    or 4, 4, 6
 ; CHECK32_32-NEXT:    lwz 29, 20(1) # 4-byte Folded Reload
 ; CHECK32_32-NEXT:    lwz 28, 16(1) # 4-byte Folded Reload
 ; CHECK32_32-NEXT:    lwz 27, 12(1) # 4-byte Folded Reload
 ; CHECK32_32-NEXT:    lwz 0, 36(1)
 ; CHECK32_32-NEXT:    addi 1, 1, 32
 ; CHECK32_32-NEXT:    mtlr 0
 ; CHECK32_32-NEXT:    blr
 ;

 ; CHECK32_64-NEXT:    mr 4, 8
 ; CHECK32_64-NEXT:    stw 29, 20(1) # 4-byte Folded Spill
 ; CHECK32_64-NEXT:    mr 29, 5
 ; CHECK32_64-NEXT:    li 5, 0
 ; CHECK32_64-NEXT:    stw 30, 24(1) # 4-byte Folded Spill
 ; CHECK32_64-NEXT:    mr 30, 6
 ; CHECK32_64-NEXT:    li 6, 37
 ; CHECK32_64-NEXT:    bl __umoddi3
-; CHECK32_64-NEXT:    clrlwi 6, 4, 26
-; CHECK32_64-NEXT:    not 4, 4
-; CHECK32_64-NEXT:    subfic 8, 6, 32
-; CHECK32_64-NEXT:    srwi 3, 30, 6
-; CHECK32_64-NEXT:    slw 7, 27, 6
-; CHECK32_64-NEXT:    clrlwi 4, 4, 26
+; CHECK32_64-NEXT:    rotlwi 3, 30, 27
+; CHECK32_64-NEXT:    andi. 5, 4, 32
+; CHECK32_64-NEXT:    bc 12, 2, .LBB3_2
+; CHECK32_64-NEXT:  # %bb.1:
+; CHECK32_64-NEXT:    ori 8, 28, 0
+; CHECK32_64-NEXT:    b .LBB3_3
+; CHECK32_64-NEXT:  .LBB3_2:
+; CHECK32_64-NEXT:    addi 8, 27, 0
+; CHECK32_64-NEXT:  .LBB3_3:
 ; CHECK32_64-NEXT:    lwz 27, 12(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    srw 8, 28, 8
-; CHECK32_64-NEXT:    rlwimi 3, 29, 26, 1, 5
-; CHECK32_64-NEXT:    lwz 29, 20(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    slwi 5, 30, 26
-; CHECK32_64-NEXT:    or 7, 7, 8
-; CHECK32_64-NEXT:    subfic 8, 4, 32
+; CHECK32_64-NEXT:    rlwimi 3, 29, 27, 0, 4
+; CHECK32_64-NEXT:    clrlwi 4, 4, 27
+; CHECK32_64-NEXT:    bc 12, 2, .LBB3_5
+; CHECK32_64-NEXT:  # %bb.4:
+; CHECK32_64-NEXT:    ori 7, 3, 0
+; CHECK32_64-NEXT:    b .LBB3_6
+; CHECK32_64-NEXT:  .LBB3_5:
+; CHECK32_64-NEXT:    addi 7, 28, 0
+; CHECK32_64-NEXT:  .LBB3_6:
+; CHECK32_64-NEXT:    slwi 5, 30, 27
 ; CHECK32_64-NEXT:    lwz 30, 24(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    addi 9, 6, -32
-; CHECK32_64-NEXT:    srw 10, 3, 4
-; CHECK32_64-NEXT:    srw 5, 5, 4
-; CHECK32_64-NEXT:    addi 4, 4, -32
-; CHECK32_64-NEXT:    slw 8, 3, 8
-; CHECK32_64-NEXT:    slw 9, 28, 9
-; CHECK32_64-NEXT:    srw 3, 3, 4
-; CHECK32_64-NEXT:    or 4, 5, 8
-; CHECK32_64-NEXT:    slw 6, 28, 6
-; CHECK32_64-NEXT:    or 5, 7, 9
+; CHECK32_64-NEXT:    bc 12, 2, .LBB3_8
+; CHECK32_64-NEXT:  # %bb.7:
+; CHECK32_64-NEXT:    ori 3, 5, 0
+; CHECK32_64-NEXT:    b .LBB3_8
+; CHECK32_64-NEXT:  .LBB3_8:
+; CHECK32_64-NEXT:    subfic 6, 4, 32
+; CHECK32_64-NEXT:    slw 8, 8, 4
+; CHECK32_64-NEXT:    lwz 29, 20(1) # 4-byte Folded Reload
+; CHECK32_64-NEXT:    srw 9, 7, 6
+; CHECK32_64-NEXT:    srw 5, 3, 6
+; CHECK32_64-NEXT:    slw 4, 7, 4
+; CHECK32_64-NEXT:    or 3, 8, 9
 ; CHECK32_64-NEXT:    lwz 28, 16(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    or 4, 4, 3
-; CHECK32_64-NEXT:    or 3, 5, 10
+; CHECK32_64-NEXT:    or 4, 4, 5
 ; CHECK32_64-NEXT:    lwz 0, 36(1)
-; CHECK32_64-NEXT:    or 4, 6, 4
 ; CHECK32_64-NEXT:    addi 1, 1, 32
 ; CHECK32_64-NEXT:    mtlr 0
 ; CHECK32_64-NEXT:    blr
 ;
 ; CHECK64-LABEL: fshl_i37:
 ; CHECK64:       # %bb.0:
 ; CHECK64-NEXT:    lis 6, 28339
 ; CHECK64-NEXT:    sldi 4, 4, 27

 ; CHECK64-NEXT:    blr
   %f = call i32 @llvm.fshr.i32(i32 %x, i32 %y, i32 %z)
   ret i32 %f
 }
 
 define i64 @fshr_i64(i64 %x, i64 %y, i64 %z) {
 ; CHECK32_32-LABEL: fshr_i64:
 ; CHECK32_32:       # %bb.0:
-; CHECK32_32-NEXT:    clrlwi 7, 8, 26
-; CHECK32_32-NEXT:    slwi 9, 4, 1
-; CHECK32_32-NEXT:    not 8, 8
-; CHECK32_32-NEXT:    rotlwi 4, 4, 1
-; CHECK32_32-NEXT:    subfic 10, 7, 32
-; CHECK32_32-NEXT:    srw 6, 6, 7
-; CHECK32_32-NEXT:    clrlwi 8, 8, 26
-; CHECK32_32-NEXT:    rlwimi 4, 3, 1, 0, 30
-; CHECK32_32-NEXT:    slw 3, 5, 10
-; CHECK32_32-NEXT:    slw 10, 9, 8
-; CHECK32_32-NEXT:    slw 4, 4, 8
-; CHECK32_32-NEXT:    or 3, 6, 3
-; CHECK32_32-NEXT:    subfic 6, 8, 32
-; CHECK32_32-NEXT:    addi 8, 8, -32
-; CHECK32_32-NEXT:    srw 6, 9, 6
-; CHECK32_32-NEXT:    slw 8, 9, 8
-; CHECK32_32-NEXT:    addi 9, 7, -32
-; CHECK32_32-NEXT:    srw 9, 5, 9
-; CHECK32_32-NEXT:    or 3, 3, 9
-; CHECK32_32-NEXT:    or 6, 4, 6
-; CHECK32_32-NEXT:    or 4, 10, 3
-; CHECK32_32-NEXT:    or 3, 6, 8
-; CHECK32_32-NEXT:    srw 5, 5, 7
+; CHECK32_32-NEXT:    andi. 7, 8, 32
+; CHECK32_32-NEXT:    clrlwi 7, 8, 27
+; CHECK32_32-NEXT:    subfic 8, 7, 32
+; CHECK32_32-NEXT:    bc 12, 2, .LBB10_2
+; CHECK32_32-NEXT:  # %bb.1:
+; CHECK32_32-NEXT:    ori 9, 4, 0
+; CHECK32_32-NEXT:    ori 4, 5, 0
+; CHECK32_32-NEXT:    b .LBB10_3
+; CHECK32_32-NEXT:  .LBB10_2:
+; CHECK32_32-NEXT:    addi 9, 5, 0
+; CHECK32_32-NEXT:    addi 3, 4, 0
+; CHECK32_32-NEXT:    addi 4, 6, 0
+; CHECK32_32-NEXT:  .LBB10_3:
+; CHECK32_32-NEXT:    srw 5, 9, 7
+; CHECK32_32-NEXT:    slw 3, 3, 8
+; CHECK32_32-NEXT:    srw 4, 4, 7
+; CHECK32_32-NEXT:    slw 6, 9, 8
 ; CHECK32_32-NEXT:    or 3, 3, 5
+; CHECK32_32-NEXT:    or 4, 6, 4
 ; CHECK32_32-NEXT:    blr
 ;
 ; CHECK32_64-LABEL: fshr_i64:
 ; CHECK32_64:       # %bb.0:
-; CHECK32_64-NEXT:    rotlwi 7, 4, 1
-; CHECK32_64-NEXT:    slwi 4, 4, 1
-; CHECK32_64-NEXT:    rlwimi 7, 3, 1, 0, 30
-; CHECK32_64-NEXT:    clrlwi 3, 8, 26
-; CHECK32_64-NEXT:    not 8, 8
-; CHECK32_64-NEXT:    subfic 9, 3, 32
-; CHECK32_64-NEXT:    srw 6, 6, 3
-; CHECK32_64-NEXT:    clrlwi 8, 8, 26
-; CHECK32_64-NEXT:    slw 9, 5, 9
-; CHECK32_64-NEXT:    addi 10, 3, -32
-; CHECK32_64-NEXT:    or 6, 6, 9
-; CHECK32_64-NEXT:    subfic 9, 8, 32
-; CHECK32_64-NEXT:    srw 3, 5, 3
-; CHECK32_64-NEXT:    srw 5, 5, 10
-; CHECK32_64-NEXT:    slw 10, 4, 8
-; CHECK32_64-NEXT:    slw 7, 7, 8
-; CHECK32_64-NEXT:    addi 8, 8, -32
-; CHECK32_64-NEXT:    srw 9, 4, 9
-; CHECK32_64-NEXT:    slw 4, 4, 8
-; CHECK32_64-NEXT:    or 7, 7, 9
-; CHECK32_64-NEXT:    or 5, 6, 5
-; CHECK32_64-NEXT:    or 6, 7, 4
-; CHECK32_64-NEXT:    or 4, 10, 5
-; CHECK32_64-NEXT:    or 3, 6, 3
+; CHECK32_64-NEXT:    andi. 7, 8, 32
+; CHECK32_64-NEXT:    clrlwi 7, 8, 27
+; CHECK32_64-NEXT:    bc 12, 2, .LBB10_2
+; CHECK32_64-NEXT:  # %bb.1:
+; CHECK32_64-NEXT:    ori 9, 4, 0
+; CHECK32_64-NEXT:    b .LBB10_3
+; CHECK32_64-NEXT:  .LBB10_2:
+; CHECK32_64-NEXT:    addi 9, 5, 0
+; CHECK32_64-NEXT:    addi 3, 4, 0
+; CHECK32_64-NEXT:    addi 5, 6, 0
+; CHECK32_64-NEXT:  .LBB10_3:
+; CHECK32_64-NEXT:    subfic 8, 7, 32
+; CHECK32_64-NEXT:    srw 4, 9, 7
+; CHECK32_64-NEXT:    slw 3, 3, 8
+; CHECK32_64-NEXT:    srw 5, 5, 7
+; CHECK32_64-NEXT:    slw 6, 9, 8
+; CHECK32_64-NEXT:    or 3, 3, 4
+; CHECK32_64-NEXT:    or 4, 6, 5
 ; CHECK32_64-NEXT:    blr
 ;
 ; CHECK64-LABEL: fshr_i64:
 ; CHECK64:       # %bb.0:
 ; CHECK64-NEXT:    clrlwi 5, 5, 26
 ; CHECK64-NEXT:    subfic 6, 5, 64
 ; CHECK64-NEXT:    srd 4, 4, 5
 ; CHECK64-NEXT:    sld 3, 3, 6

 ; CHECK32_32-NEXT:    mr 29, 5
 ; CHECK32_32-NEXT:    stw 30, 24(1) # 4-byte Folded Spill
 ; CHECK32_32-NEXT:    mr 30, 6
 ; CHECK32_32-NEXT:    mr 3, 7
 ; CHECK32_32-NEXT:    mr 4, 8
 ; CHECK32_32-NEXT:    li 5, 0
 ; CHECK32_32-NEXT:    li 6, 37
 ; CHECK32_32-NEXT:    bl __umoddi3
+; CHECK32_32-NEXT:    rotlwi 3, 30, 27
 ; CHECK32_32-NEXT:    addi 4, 4, 27
-; CHECK32_32-NEXT:    rotlwi 5, 30, 27
-; CHECK32_32-NEXT:    clrlwi 8, 4, 26
-; CHECK32_32-NEXT:    slwi 3, 30, 27
-; CHECK32_32-NEXT:    rotlwi 7, 28, 1
-; CHECK32_32-NEXT:    rlwimi 5, 29, 27, 0, 4
-; CHECK32_32-NEXT:    not 4, 4
-; CHECK32_32-NEXT:    subfic 9, 8, 32
-; CHECK32_32-NEXT:    slwi 6, 28, 1
-; CHECK32_32-NEXT:    rlwimi 7, 27, 1, 0, 30
-; CHECK32_32-NEXT:    srw 3, 3, 8
-; CHECK32_32-NEXT:    clrlwi 4, 4, 26
-; CHECK32_32-NEXT:    slw 9, 5, 9
-; CHECK32_32-NEXT:    slw 10, 6, 4
-; CHECK32_32-NEXT:    slw 7, 7, 4
-; CHECK32_32-NEXT:    or 3, 3, 9
-; CHECK32_32-NEXT:    subfic 9, 4, 32
-; CHECK32_32-NEXT:    addi 4, 4, -32
-; CHECK32_32-NEXT:    srw 9, 6, 9
-; CHECK32_32-NEXT:    slw 6, 6, 4
-; CHECK32_32-NEXT:    addi 4, 8, -32
-; CHECK32_32-NEXT:    srw 4, 5, 4
-; CHECK32_32-NEXT:    or 3, 3, 4
-; CHECK32_32-NEXT:    or 7, 7, 9
-; CHECK32_32-NEXT:    or 4, 10, 3
-; CHECK32_32-NEXT:    or 3, 7, 6
-; CHECK32_32-NEXT:    srw 5, 5, 8
-; CHECK32_32-NEXT:    or 3, 3, 5
+; CHECK32_32-NEXT:    slwi 5, 30, 27
+; CHECK32_32-NEXT:    rlwimi 3, 29, 27, 0, 4
+; CHECK32_32-NEXT:    andi. 6, 4, 32
+; CHECK32_32-NEXT:    clrlwi 4, 4, 27
+; CHECK32_32-NEXT:    subfic 6, 4, 32
+; CHECK32_32-NEXT:    bc 12, 2, .LBB11_2
+; CHECK32_32-NEXT:  # %bb.1:
+; CHECK32_32-NEXT:    ori 7, 28, 0
+; CHECK32_32-NEXT:    ori 8, 27, 0
+; CHECK32_32-NEXT:    b .LBB11_3
+; CHECK32_32-NEXT:  .LBB11_2:
+; CHECK32_32-NEXT:    addi 7, 3, 0
+; CHECK32_32-NEXT:    addi 8, 28, 0
+; CHECK32_32-NEXT:    addi 3, 5, 0
+; CHECK32_32-NEXT:  .LBB11_3:
 ; CHECK32_32-NEXT:    lwz 30, 24(1) # 4-byte Folded Reload
+; CHECK32_32-NEXT:    srw 5, 7, 4
+; CHECK32_32-NEXT:    slw 8, 8, 6
+; CHECK32_32-NEXT:    srw 4, 3, 4
+; CHECK32_32-NEXT:    slw 6, 7, 6
+; CHECK32_32-NEXT:    or 3, 8, 5
+; CHECK32_32-NEXT:    or 4, 6, 4
 ; CHECK32_32-NEXT:    lwz 29, 20(1) # 4-byte Folded Reload
 ; CHECK32_32-NEXT:    lwz 28, 16(1) # 4-byte Folded Reload
 ; CHECK32_32-NEXT:    lwz 27, 12(1) # 4-byte Folded Reload
 ; CHECK32_32-NEXT:    lwz 0, 36(1)
 ; CHECK32_32-NEXT:    addi 1, 1, 32
 ; CHECK32_32-NEXT:    mtlr 0
 ; CHECK32_32-NEXT:    blr
 ;

 ; CHECK32_64-NEXT:    mr 29, 5
 ; CHECK32_64-NEXT:    li 5, 0
 ; CHECK32_64-NEXT:    stw 30, 24(1) # 4-byte Folded Spill
 ; CHECK32_64-NEXT:    mr 30, 6
 ; CHECK32_64-NEXT:    li 6, 37
 ; CHECK32_64-NEXT:    bl __umoddi3
 ; CHECK32_64-NEXT:    addi 4, 4, 27
 ; CHECK32_64-NEXT:    rotlwi 3, 30, 27
-; CHECK32_64-NEXT:    clrlwi 8, 4, 26
+; CHECK32_64-NEXT:    andi. 5, 4, 32
 ; CHECK32_64-NEXT:    rlwimi 3, 29, 27, 0, 4
 ; CHECK32_64-NEXT:    lwz 29, 20(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    slwi 6, 30, 27
-; CHECK32_64-NEXT:    lwz 30, 24(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    not 4, 4
-; CHECK32_64-NEXT:    subfic 9, 8, 32
-; CHECK32_64-NEXT:    rotlwi 5, 28, 1
-; CHECK32_64-NEXT:    srw 6, 6, 8
-; CHECK32_64-NEXT:    clrlwi 4, 4, 26
-; CHECK32_64-NEXT:    slw 9, 3, 9
-; CHECK32_64-NEXT:    rlwimi 5, 27, 1, 0, 30
-; CHECK32_64-NEXT:    slwi 7, 28, 1
+; CHECK32_64-NEXT:    bc 12, 2, .LBB11_2
+; CHECK32_64-NEXT:  # %bb.1:
+; CHECK32_64-NEXT:    ori 7, 28, 0
+; CHECK32_64-NEXT:    ori 8, 27, 0
+; CHECK32_64-NEXT:    b .LBB11_3
+; CHECK32_64-NEXT:  .LBB11_2:
+; CHECK32_64-NEXT:    addi 7, 3, 0
+; CHECK32_64-NEXT:    addi 8, 28, 0
+; CHECK32_64-NEXT:  .LBB11_3:
+; CHECK32_64-NEXT:    clrlwi 4, 4, 27
 ; CHECK32_64-NEXT:    lwz 28, 16(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    addi 10, 8, -32
+; CHECK32_64-NEXT:    slwi 5, 30, 27
+; CHECK32_64-NEXT:    subfic 6, 4, 32
+; CHECK32_64-NEXT:    bc 12, 2, .LBB11_4
+; CHECK32_64-NEXT:    b .LBB11_5
+; CHECK32_64-NEXT:  .LBB11_4:
+; CHECK32_64-NEXT:    addi 3, 5, 0
+; CHECK32_64-NEXT:  .LBB11_5:
+; CHECK32_64-NEXT:    srw 9, 7, 4
+; CHECK32_64-NEXT:    slw 8, 8, 6
+; CHECK32_64-NEXT:    lwz 30, 24(1) # 4-byte Folded Reload
+; CHECK32_64-NEXT:    srw 4, 3, 4
+; CHECK32_64-NEXT:    slw 5, 7, 6
 ; CHECK32_64-NEXT:    lwz 27, 12(1) # 4-byte Folded Reload
-; CHECK32_64-NEXT:    or 6, 6, 9
-; CHECK32_64-NEXT:    subfic 9, 4, 32
-; CHECK32_64-NEXT:    srw 8, 3, 8
-; CHECK32_64-NEXT:    srw 3, 3, 10
+; CHECK32_64-NEXT:    or 3, 8, 9
+; CHECK32_64-NEXT:    or 4, 5, 4
 ; CHECK32_64-NEXT:    lwz 0, 36(1)
-; CHECK32_64-NEXT:    slw 10, 7, 4
-; CHECK32_64-NEXT:    slw 5, 5, 4
-; CHECK32_64-NEXT:    addi 4, 4, -32
-; CHECK32_64-NEXT:    srw 9, 7, 9
-; CHECK32_64-NEXT:    slw 4, 7, 4
-; CHECK32_64-NEXT:    or 5, 5, 9
-; CHECK32_64-NEXT:    or 3, 6, 3
-; CHECK32_64-NEXT:    or 5, 5, 4
-; CHECK32_64-NEXT:    or 4, 10, 3
-; CHECK32_64-NEXT:    or 3, 5, 8
 ; CHECK32_64-NEXT:    addi 1, 1, 32
 ; CHECK32_64-NEXT:    mtlr 0
 ; CHECK32_64-NEXT:    blr
 ;
 ; CHECK64-LABEL: fshr_i37:
 ; CHECK64:       # %bb.0:
 ; CHECK64-NEXT:    lis 6, 28339
 ; CHECK64-NEXT:    sldi 4, 4, 27

llvm/test/CodeGen/RISCV/rv32zbb-zbp.ll

 ; This test is presented here in case future expansions of the experimental-b
 ; extension introduce instructions suitable for this pattern.
 
 declare i64 @llvm.fshl.i64(i64, i64, i64)
 
 define i64 @rol_i64(i64 %a, i64 %b) nounwind {
 ; RV32I-LABEL: rol_i64:
 ; RV32I:       # %bb.0:
-; RV32I-NEXT:    mv a7, a1
-; RV32I-NEXT:    andi a1, a2, 63
-; RV32I-NEXT:    addi t0, a1, -32
-; RV32I-NEXT:    addi a6, zero, 31
-; RV32I-NEXT:    bltz t0, .LBB7_2
+; RV32I-NEXT:    srli a3, a2, 5
+; RV32I-NEXT:    andi a3, a3, 1
+; RV32I-NEXT:    mv a4, a1
+; RV32I-NEXT:    bnez a3, .LBB7_2
 ; RV32I-NEXT:  # %bb.1:
-; RV32I-NEXT:    sll a1, a0, t0
-; RV32I-NEXT:    j .LBB7_3
+; RV32I-NEXT:    mv a4, a0
 ; RV32I-NEXT:  .LBB7_2:
-; RV32I-NEXT:    sll a3, a7, a2
-; RV32I-NEXT:    sub a1, a6, a1
-; RV32I-NEXT:    srli a4, a0, 1
-; RV32I-NEXT:    srl a1, a4, a1
-; RV32I-NEXT:    or a1, a3, a1
-; RV32I-NEXT:  .LBB7_3:
-; RV32I-NEXT:    neg a5, a2
-; RV32I-NEXT:    andi a3, a5, 63
-; RV32I-NEXT:    addi a4, a3, -32
-; RV32I-NEXT:    bltz a4, .LBB7_5
-; RV32I-NEXT:  # %bb.4:
-; RV32I-NEXT:    srl a3, a7, a4
-; RV32I-NEXT:    bltz t0, .LBB7_6
-; RV32I-NEXT:    j .LBB7_7
-; RV32I-NEXT:  .LBB7_5:
-; RV32I-NEXT:    srl a4, a7, a5
-; RV32I-NEXT:    or a1, a1, a4
-; RV32I-NEXT:    srl a4, a0, a5
-; RV32I-NEXT:    sub a3, a6, a3
-; RV32I-NEXT:    slli a5, a7, 1
-; RV32I-NEXT:    sll a3, a5, a3
-; RV32I-NEXT:    or a3, a4, a3
-; RV32I-NEXT:    bgez t0, .LBB7_7
-; RV32I-NEXT:  .LBB7_6:
+; RV32I-NEXT:    sll a6, a4, a2
+; RV32I-NEXT:    bnez a3, .LBB7_4
+; RV32I-NEXT:  # %bb.3:
+; RV32I-NEXT:    mv a0, a1
+; RV32I-NEXT:  .LBB7_4:
+; RV32I-NEXT:    srli a1, a0, 1
+; RV32I-NEXT:    not a5, a2
+; RV32I-NEXT:    srl a1, a1, a5
+; RV32I-NEXT:    or a3, a6, a1
 ; RV32I-NEXT:    sll a0, a0, a2
-; RV32I-NEXT:    or a3, a3, a0
-; RV32I-NEXT:  .LBB7_7:
+; RV32I-NEXT:    srli a1, a4, 1
+; RV32I-NEXT:    srl a1, a1, a5
+; RV32I-NEXT:    or a1, a0, a1
 ; RV32I-NEXT:    mv a0, a3
 ; RV32I-NEXT:    ret
 ;
 ; RV32ZBB-LABEL: rol_i64:
 ; RV32ZBB:       # %bb.0:
-; RV32ZBB-NEXT:    mv a7, a1
-; RV32ZBB-NEXT:    andi a1, a2, 63
-; RV32ZBB-NEXT:    addi t0, a1, -32
-; RV32ZBB-NEXT:    addi a6, zero, 31
-; RV32ZBB-NEXT:    bltz t0, .LBB7_2
+; RV32ZBB-NEXT:    srli a3, a2, 5
+; RV32ZBB-NEXT:    andi a3, a3, 1
+; RV32ZBB-NEXT:    mv a4, a1
+; RV32ZBB-NEXT:    bnez a3, .LBB7_2
 ; RV32ZBB-NEXT:  # %bb.1:
-; RV32ZBB-NEXT:    sll a1, a0, t0
-; RV32ZBB-NEXT:    j .LBB7_3
+; RV32ZBB-NEXT:    mv a4, a0
 ; RV32ZBB-NEXT:  .LBB7_2:
-; RV32ZBB-NEXT:    sll a3, a7, a2
-; RV32ZBB-NEXT:    sub a1, a6, a1
-; RV32ZBB-NEXT:    srli a4, a0, 1
-; RV32ZBB-NEXT:    srl a1, a4, a1
-; RV32ZBB-NEXT:    or a1, a3, a1
-; RV32ZBB-NEXT:  .LBB7_3:
-; RV32ZBB-NEXT:    neg a5, a2
-; RV32ZBB-NEXT:    andi a3, a5, 63
-; RV32ZBB-NEXT:    addi a4, a3, -32
-; RV32ZBB-NEXT:    bltz a4, .LBB7_5
-; RV32ZBB-NEXT:  # %bb.4:
-; RV32ZBB-NEXT:    srl a3, a7, a4
-; RV32ZBB-NEXT:    bltz t0, .LBB7_6
-; RV32ZBB-NEXT:    j .LBB7_7
-; RV32ZBB-NEXT:  .LBB7_5:
-; RV32ZBB-NEXT:    srl a4, a7, a5
-; RV32ZBB-NEXT:    or a1, a1, a4
-; RV32ZBB-NEXT:    srl a4, a0, a5
-; RV32ZBB-NEXT:    sub a3, a6, a3
-; RV32ZBB-NEXT:    slli a5, a7, 1
-; RV32ZBB-NEXT:    sll a3, a5, a3
-; RV32ZBB-NEXT:    or a3, a4, a3
-; RV32ZBB-NEXT:    bgez t0, .LBB7_7
-; RV32ZBB-NEXT:  .LBB7_6:
+; RV32ZBB-NEXT:    sll a6, a4, a2
+; RV32ZBB-NEXT:    bnez a3, .LBB7_4
+; RV32ZBB-NEXT:  # %bb.3:
+; RV32ZBB-NEXT:    mv a0, a1
+; RV32ZBB-NEXT:  .LBB7_4:
+; RV32ZBB-NEXT:    srli a1, a0, 1
+; RV32ZBB-NEXT:    not a5, a2
+; RV32ZBB-NEXT:    srl a1, a1, a5
+; RV32ZBB-NEXT:    or a3, a6, a1
 ; RV32ZBB-NEXT:    sll a0, a0, a2
-; RV32ZBB-NEXT:    or a3, a3, a0
-; RV32ZBB-NEXT:  .LBB7_7:
+; RV32ZBB-NEXT:    srli a1, a4, 1
+; RV32ZBB-NEXT:    srl a1, a1, a5
+; RV32ZBB-NEXT:    or a1, a0, a1
 ; RV32ZBB-NEXT:    mv a0, a3
 ; RV32ZBB-NEXT:    ret
 ;
 ; RV32ZBP-LABEL: rol_i64:
 ; RV32ZBP:       # %bb.0:
-; RV32ZBP-NEXT:    mv a7, a1
-; RV32ZBP-NEXT:    andi a1, a2, 63
-; RV32ZBP-NEXT:    addi t0, a1, -32
-; RV32ZBP-NEXT:    addi a6, zero, 31
-; RV32ZBP-NEXT:    bltz t0, .LBB7_2
+; RV32ZBP-NEXT:    srli a3, a2, 5
+; RV32ZBP-NEXT:    andi a3, a3, 1
+; RV32ZBP-NEXT:    mv a4, a1
+; RV32ZBP-NEXT:    bnez a3, .LBB7_2
 ; RV32ZBP-NEXT:  # %bb.1:
-; RV32ZBP-NEXT:    sll a1, a0, t0
-; RV32ZBP-NEXT:    j .LBB7_3
+; RV32ZBP-NEXT:    mv a4, a0
 ; RV32ZBP-NEXT:  .LBB7_2:
-; RV32ZBP-NEXT:    sll a3, a7, a2
-; RV32ZBP-NEXT:    sub a1, a6, a1
-; RV32ZBP-NEXT:    srli a4, a0, 1
-; RV32ZBP-NEXT:    srl a1, a4, a1
-; RV32ZBP-NEXT:    or a1, a3, a1
-; RV32ZBP-NEXT:  .LBB7_3:
-; RV32ZBP-NEXT:    neg a5, a2
-; RV32ZBP-NEXT:    andi a3, a5, 63
-; RV32ZBP-NEXT:    addi a4, a3, -32
-; RV32ZBP-NEXT:    bltz a4, .LBB7_5
-; RV32ZBP-NEXT:  # %bb.4:
-; RV32ZBP-NEXT:    srl a3, a7, a4
-; RV32ZBP-NEXT:    bltz t0, .LBB7_6
-; RV32ZBP-NEXT:    j .LBB7_7
-; RV32ZBP-NEXT:  .LBB7_5:
-; RV32ZBP-NEXT:    srl a4, a7, a5
-; RV32ZBP-NEXT:    or a1, a1, a4
-; RV32ZBP-NEXT:    srl a4, a0, a5
-; RV32ZBP-NEXT:    sub a3, a6, a3
-; RV32ZBP-NEXT:    slli a5, a7, 1
-; RV32ZBP-NEXT:    sll a3, a5, a3
-; RV32ZBP-NEXT:    or a3, a4, a3
-; RV32ZBP-NEXT:    bgez t0, .LBB7_7
-; RV32ZBP-NEXT:  .LBB7_6:
+; RV32ZBP-NEXT:    sll a6, a4, a2
+; RV32ZBP-NEXT:    bnez a3, .LBB7_4
+; RV32ZBP-NEXT:  # %bb.3:
+; RV32ZBP-NEXT:    mv a0, a1
+; RV32ZBP-NEXT:  .LBB7_4:
+; RV32ZBP-NEXT:    srli a1, a0, 1
+; RV32ZBP-NEXT:    not a5, a2
+; RV32ZBP-NEXT:    srl a1, a1, a5
+; RV32ZBP-NEXT:    or a3, a6, a1
 ; RV32ZBP-NEXT:    sll a0, a0, a2
-; RV32ZBP-NEXT:    or a3, a3, a0
-; RV32ZBP-NEXT:  .LBB7_7:
+; RV32ZBP-NEXT:    srli a1, a4, 1
+; RV32ZBP-NEXT:    srl a1, a1, a5
+; RV32ZBP-NEXT:    or a1, a0, a1
 ; RV32ZBP-NEXT:    mv a0, a3
 ; RV32ZBP-NEXT:    ret
   %or = tail call i64 @llvm.fshl.i64(i64 %a, i64 %a, i64 %b)
   ret i64 %or
 }
 
 declare i32 @llvm.fshr.i32(i32, i32, i32)
 

 ; This test is presented here in case future expansions of the experimental-b
 ; extension introduce instructions suitable for this pattern.
 
 declare i64 @llvm.fshr.i64(i64, i64, i64)
 
 define i64 @ror_i64(i64 %a, i64 %b) nounwind {
 ; RV32I-LABEL: ror_i64:
 ; RV32I:       # %bb.0:
-; RV32I-NEXT:    mv t0, a0
-; RV32I-NEXT:    andi a0, a2, 63
-; RV32I-NEXT:    addi a7, a0, -32
-; RV32I-NEXT:    addi a6, zero, 31
-; RV32I-NEXT:    bltz a7, .LBB9_2
+; RV32I-NEXT:    andi a4, a2, 32
+; RV32I-NEXT:    mv a3, a0
+; RV32I-NEXT:    beqz a4, .LBB9_2
 ; RV32I-NEXT:  # %bb.1:
-; RV32I-NEXT:    srl a0, a1, a7
-; RV32I-NEXT:    j .LBB9_3
+; RV32I-NEXT:    mv a3, a1
 ; RV32I-NEXT:  .LBB9_2:
-; RV32I-NEXT:    srl a3, t0, a2
-; RV32I-NEXT:    sub a0, a6, a0
-; RV32I-NEXT:    slli a4, a1, 1
-; RV32I-NEXT:    sll a0, a4, a0
-; RV32I-NEXT:    or a0, a3, a0
-; RV32I-NEXT:  .LBB9_3:
-; RV32I-NEXT:    neg a5, a2
-; RV32I-NEXT:    andi a4, a5, 63
-; RV32I-NEXT:    addi a3, a4, -32
-; RV32I-NEXT:    bltz a3, .LBB9_5
-; RV32I-NEXT:  # %bb.4:
-; RV32I-NEXT:    sll a3, t0, a3
-; RV32I-NEXT:    bltz a7, .LBB9_6
-; RV32I-NEXT:    j .LBB9_7
-; RV32I-NEXT:  .LBB9_5:
-; RV32I-NEXT:    sll a3, t0, a5
-; RV32I-NEXT:    or a0, a0, a3
-; RV32I-NEXT:    sll a3, a1, a5
-; RV32I-NEXT:    sub a4, a6, a4
-; RV32I-NEXT:    srli a5, t0, 1
-; RV32I-NEXT:    srl a4, a5, a4
-; RV32I-NEXT:    or a3, a3, a4
-; RV32I-NEXT:    bgez a7, .LBB9_7
-; RV32I-NEXT:  .LBB9_6:
+; RV32I-NEXT:    srl a5, a3, a2
+; RV32I-NEXT:    beqz a4, .LBB9_4
+; RV32I-NEXT:  # %bb.3:
+; RV32I-NEXT:    mv a1, a0
+; RV32I-NEXT:  .LBB9_4:
+; RV32I-NEXT:    slli a0, a1, 1
+; RV32I-NEXT:    not a4, a2
+; RV32I-NEXT:    sll a0, a0, a4
+; RV32I-NEXT:    or a0, a0, a5
 ; RV32I-NEXT:    srl a1, a1, a2
-; RV32I-NEXT:    or a3, a3, a1
-; RV32I-NEXT:  .LBB9_7:
-; RV32I-NEXT:    mv a1, a3
+; RV32I-NEXT:    slli a2, a3, 1
+; RV32I-NEXT:    sll a2, a2, a4
+; RV32I-NEXT:    or a1, a2, a1
 ; RV32I-NEXT:    ret
 ;
 ; RV32ZBB-LABEL: ror_i64:
 ; RV32ZBB:       # %bb.0:
-; RV32ZBB-NEXT:    mv t0, a0
-; RV32ZBB-NEXT:    andi a0, a2, 63
-; RV32ZBB-NEXT:    addi a7, a0, -32
-; RV32ZBB-NEXT:    addi a6, zero, 31
-; RV32ZBB-NEXT:    bltz a7, .LBB9_2
+; RV32ZBB-NEXT:    andi a4, a2, 32
+; RV32ZBB-NEXT:    mv a3, a0
+; RV32ZBB-NEXT:    beqz a4, .LBB9_2
 ; RV32ZBB-NEXT:  # %bb.1:
-; RV32ZBB-NEXT:    srl a0, a1, a7
-; RV32ZBB-NEXT:    j .LBB9_3
+; RV32ZBB-NEXT:    mv a3, a1
 ; RV32ZBB-NEXT:  .LBB9_2:
-; RV32ZBB-NEXT:    srl a3, t0, a2
-; RV32ZBB-NEXT:    sub a0, a6, a0
-; RV32ZBB-NEXT:    slli a4, a1, 1
-; RV32ZBB-NEXT:    sll a0, a4, a0
-; RV32ZBB-NEXT:    or a0, a3, a0
-; RV32ZBB-NEXT:  .LBB9_3:
-; RV32ZBB-NEXT:    neg a5, a2
-; RV32ZBB-NEXT:    andi a4, a5, 63
-; RV32ZBB-NEXT:    addi a3, a4, -32
-; RV32ZBB-NEXT:    bltz a3, .LBB9_5
-; RV32ZBB-NEXT:  # %bb.4:
-; RV32ZBB-NEXT:    sll a3, t0, a3
-; RV32ZBB-NEXT:    bltz a7, .LBB9_6
-; RV32ZBB-NEXT:    j .LBB9_7
-; RV32ZBB-NEXT:  .LBB9_5:
-; RV32ZBB-NEXT:    sll a3, t0, a5
-; RV32ZBB-NEXT:    or a0, a0, a3
-; RV32ZBB-NEXT:    sll a3, a1, a5
-; RV32ZBB-NEXT:    sub a4, a6, a4
-; RV32ZBB-NEXT:    srli a5, t0, 1
-; RV32ZBB-NEXT:    srl a4, a5, a4
-; RV32ZBB-NEXT:    or a3, a3, a4
-; RV32ZBB-NEXT:    bgez a7, .LBB9_7
-; RV32ZBB-NEXT:  .LBB9_6:
+; RV32ZBB-NEXT:    srl a5, a3, a2
+; RV32ZBB-NEXT:    beqz a4, .LBB9_4
+; RV32ZBB-NEXT:  # %bb.3:
+; RV32ZBB-NEXT:    mv a1, a0
+; RV32ZBB-NEXT:  .LBB9_4:
+; RV32ZBB-NEXT:    slli a0, a1, 1
+; RV32ZBB-NEXT:    not a4, a2
+; RV32ZBB-NEXT:    sll a0, a0, a4
+; RV32ZBB-NEXT:    or a0, a0, a5
 ; RV32ZBB-NEXT:    srl a1, a1, a2
-; RV32ZBB-NEXT:    or a3, a3, a1
-; RV32ZBB-NEXT:  .LBB9_7:
-; RV32ZBB-NEXT:    mv a1, a3
+; RV32ZBB-NEXT:    slli a2, a3, 1
+; RV32ZBB-NEXT:    sll a2, a2, a4
+; RV32ZBB-NEXT:    or a1, a2, a1
 ; RV32ZBB-NEXT:    ret
 ;
 ; RV32ZBP-LABEL: ror_i64:
 ; RV32ZBP:       # %bb.0:
-; RV32ZBP-NEXT:    mv t0, a0
-; RV32ZBP-NEXT:    andi a0, a2, 63
-; RV32ZBP-NEXT:    addi a7, a0, -32
-; RV32ZBP-NEXT:    addi a6, zero, 31
-; RV32ZBP-NEXT:    bltz a7, .LBB9_2
+; RV32ZBP-NEXT:    andi a4, a2, 32
+; RV32ZBP-NEXT:    mv a3, a0
+; RV32ZBP-NEXT:    beqz a4, .LBB9_2
 ; RV32ZBP-NEXT:  # %bb.1:
-; RV32ZBP-NEXT:    srl a0, a1, a7
-; RV32ZBP-NEXT:    j .LBB9_3
+; RV32ZBP-NEXT:    mv a3, a1
 ; RV32ZBP-NEXT:  .LBB9_2:
-; RV32ZBP-NEXT:    srl a3, t0, a2
-; RV32ZBP-NEXT:    sub a0, a6, a0
-; RV32ZBP-NEXT:    slli a4, a1, 1
-; RV32ZBP-NEXT:    sll a0, a4, a0
-; RV32ZBP-NEXT:    or a0, a3, a0
-; RV32ZBP-NEXT:  .LBB9_3:
-; RV32ZBP-NEXT:    neg a5, a2
-; RV32ZBP-NEXT:    andi a4, a5, 63
-; RV32ZBP-NEXT:    addi a3, a4, -32
-; RV32ZBP-NEXT:    bltz a3, .LBB9_5
-; RV32ZBP-NEXT:  # %bb.4:
-; RV32ZBP-NEXT:    sll a3, t0, a3
-; RV32ZBP-NEXT:    bltz a7, .LBB9_6
-; RV32ZBP-NEXT:    j .LBB9_7
-; RV32ZBP-NEXT:  .LBB9_5:
-; RV32ZBP-NEXT:    sll a3, t0, a5
-; RV32ZBP-NEXT:    or a0, a0, a3
-; RV32ZBP-NEXT:    sll a3, a1, a5
-; RV32ZBP-NEXT:    sub a4, a6, a4
-; RV32ZBP-NEXT:    srli a5, t0, 1
-; RV32ZBP-NEXT:    srl a4, a5, a4
-; RV32ZBP-NEXT:    or a3, a3, a4
-; RV32ZBP-NEXT:    bgez a7, .LBB9_7
-; RV32ZBP-NEXT:  .LBB9_6:
+; RV32ZBP-NEXT:    srl a5, a3, a2
+; RV32ZBP-NEXT:    beqz a4, .LBB9_4
+; RV32ZBP-NEXT:  # %bb.3:
+; RV32ZBP-NEXT:    mv a1, a0
+; RV32ZBP-NEXT:  .LBB9_4:
+; RV32ZBP-NEXT:    slli a0, a1, 1
+; RV32ZBP-NEXT:    not a4, a2
+; RV32ZBP-NEXT:    sll a0, a0, a4
+; RV32ZBP-NEXT:    or a0, a0, a5
 ; RV32ZBP-NEXT:    srl a1, a1, a2
-; RV32ZBP-NEXT:    or a3, a3, a1
-; RV32ZBP-NEXT:  .LBB9_7:
-; RV32ZBP-NEXT:    mv a1, a3
+; RV32ZBP-NEXT:    slli a2, a3, 1
+; RV32ZBP-NEXT:    sll a2, a2, a4
+; RV32ZBP-NEXT:    or a1, a2, a1
 ; RV32ZBP-NEXT:    ret
   %or = tail call i64 @llvm.fshr.i64(i64 %a, i64 %a, i64 %b)
   ret i64 %or
 }
 
 define i32 @rori_i32_fshl(i32 %a) nounwind {
 ; RV32I-LABEL: rori_i32_fshl:
 ; RV32I:       # %bb.0:

 ; RV32ZBP-NEXT:    ret
   %1 = tail call i32 @llvm.fshr.i32(i32 %a, i32 %a, i32 31)
   ret i32 %1
 }
 
 define i64 @rori_i64(i64 %a) nounwind {
 ; RV32I-LABEL: rori_i64:
 ; RV32I:       # %bb.0:
-; RV32I-NEXT:    slli a2, a1, 31
-; RV32I-NEXT:    srli a3, a0, 1
+; RV32I-NEXT:    srli a2, a0, 1
+; RV32I-NEXT:    slli a3, a1, 31
 ; RV32I-NEXT:    or a2, a3, a2
 ; RV32I-NEXT:    srli a1, a1, 1
 ; RV32I-NEXT:    slli a0, a0, 31
 ; RV32I-NEXT:    or a1, a0, a1
 ; RV32I-NEXT:    mv a0, a2
 ; RV32I-NEXT:    ret
 ;
 ; RV32ZBB-LABEL: rori_i64:
 ; RV32ZBB:       # %bb.0:
-; RV32ZBB-NEXT:    slli a2, a1, 31
-; RV32ZBB-NEXT:    srli a3, a0, 1
+; RV32ZBB-NEXT:    srli a2, a0, 1
+; RV32ZBB-NEXT:    slli a3, a1, 31
 ; RV32ZBB-NEXT:    or a2, a3, a2
 ; RV32ZBB-NEXT:    srli a1, a1, 1
 ; RV32ZBB-NEXT:    slli a0, a0, 31
 ; RV32ZBB-NEXT:    or a1, a0, a1
 ; RV32ZBB-NEXT:    mv a0, a2
 ; RV32ZBB-NEXT:    ret
 ;
 ; RV32ZBP-LABEL: rori_i64:
 ; RV32ZBP:       # %bb.0:
-; RV32ZBP-NEXT:    slli a2, a1, 31
-; RV32ZBP-NEXT:    srli a3, a0, 1
+; RV32ZBP-NEXT:    srli a2, a0, 1
+; RV32ZBP-NEXT:    slli a3, a1, 31
 ; RV32ZBP-NEXT:    or a2, a3, a2
 ; RV32ZBP-NEXT:    srli a1, a1, 1
 ; RV32ZBP-NEXT:    slli a0, a0, 31
 ; RV32ZBP-NEXT:    or a1, a0, a1
 ; RV32ZBP-NEXT:    mv a0, a2
 ; RV32ZBP-NEXT:    ret
   %1 = tail call i64 @llvm.fshl.i64(i64 %a, i64 %a, i64 63)
   ret i64 %1
 }
 
 define i64 @rori_i64_fshr(i64 %a) nounwind {
 ; RV32I-LABEL: rori_i64_fshr:
 ; RV32I:       # %bb.0:
-; RV32I-NEXT:    slli a2, a0, 1
-; RV32I-NEXT:    srli a3, a1, 31
+; RV32I-NEXT:    srli a2, a1, 31
+; RV32I-NEXT:    slli a3, a0, 1
 ; RV32I-NEXT:    or a2, a3, a2
 ; RV32I-NEXT:    srli a0, a0, 31
 ; RV32I-NEXT:    slli a1, a1, 1
 ; RV32I-NEXT:    or a1, a1, a0
 ; RV32I-NEXT:    mv a0, a2
 ; RV32I-NEXT:    ret
 ;
 ; RV32ZBB-LABEL: rori_i64_fshr:
 ; RV32ZBB:       # %bb.0:
-; RV32ZBB-NEXT:    slli a2, a0, 1
-; RV32ZBB-NEXT:    srli a3, a1, 31
+; RV32ZBB-NEXT:    srli a2, a1, 31
+; RV32ZBB-NEXT:    slli a3, a0, 1
 ; RV32ZBB-NEXT:    or a2, a3, a2
 ; RV32ZBB-NEXT:    srli a0, a0, 31
 ; RV32ZBB-NEXT:    slli a1, a1, 1
 ; RV32ZBB-NEXT:    or a1, a1, a0
 ; RV32ZBB-NEXT:    mv a0, a2
 ; RV32ZBB-NEXT:    ret
 ;
 ; RV32ZBP-LABEL: rori_i64_fshr:
 ; RV32ZBP:       # %bb.0:
-; RV32ZBP-NEXT:    slli a2, a0, 1
-; RV32ZBP-NEXT:    srli a3, a1, 31
+; RV32ZBP-NEXT:    srli a2, a1, 31
+; RV32ZBP-NEXT:    slli a3, a0, 1
 ; RV32ZBP-NEXT:    or a2, a3, a2
 ; RV32ZBP-NEXT:    srli a0, a0, 31
 ; RV32ZBP-NEXT:    slli a1, a1, 1
 ; RV32ZBP-NEXT:    or a1, a1, a0
 ; RV32ZBP-NEXT:    mv a0, a2
 ; RV32ZBP-NEXT:    ret
   %1 = tail call i64 @llvm.fshr.i64(i64 %a, i64 %a, i64 63)
   ret i64 %1

llvm/test/CodeGen/RISCV/rv32zbt.ll

 ; This test is presented here in case future expansions of the experimental-b
 ; extension introduce instructions that can match more efficiently this pattern.
 
 declare i64 @llvm.fshl.i64(i64, i64, i64)
 
 define i64 @fshl_i64(i64 %a, i64 %b, i64 %c) nounwind {
 ; RV32I-LABEL: fshl_i64:
 ; RV32I:       # %bb.0:
-; RV32I-NEXT:    andi a5, a4, 63
-; RV32I-NEXT:    addi a7, a5, -32
-; RV32I-NEXT:    addi a6, zero, 31
-; RV32I-NEXT:    bltz a7, .LBB13_2
+; RV32I-NEXT:    srli a5, a4, 5
+; RV32I-NEXT:    andi a5, a5, 1
+; RV32I-NEXT:    mv a6, a3
+; RV32I-NEXT:    bnez a5, .LBB13_2
 ; RV32I-NEXT:  # %bb.1:
-; RV32I-NEXT:    sll a1, a0, a7
-; RV32I-NEXT:    j .LBB13_3
+; RV32I-NEXT:    mv a6, a0
 ; RV32I-NEXT:  .LBB13_2:
-; RV32I-NEXT:    sll t0, a1, a4
-; RV32I-NEXT:    sub a5, a6, a5
-; RV32I-NEXT:    srli a1, a0, 1
-; RV32I-NEXT:    srl a1, a1, a5
-; RV32I-NEXT:    or a1, t0, a1
-; RV32I-NEXT:  .LBB13_3:
-; RV32I-NEXT:    not t2, a4
-; RV32I-NEXT:    andi t1, t2, 63
-; RV32I-NEXT:    addi a5, t1, -32
-; RV32I-NEXT:    srli t0, a3, 1
-; RV32I-NEXT:    bltz a5, .LBB13_5
-; RV32I-NEXT:  # %bb.4:
-; RV32I-NEXT:    srl a2, t0, a5
-; RV32I-NEXT:    bltz a7, .LBB13_6
-; RV32I-NEXT:    j .LBB13_7
-; RV32I-NEXT:  .LBB13_5:
-; RV32I-NEXT:    srl a5, t0, t2
-; RV32I-NEXT:    or a1, a1, a5
-; RV32I-NEXT:    slli a3, a3, 31
+; RV32I-NEXT:    sll a7, a6, a4
+; RV32I-NEXT:    bnez a5, .LBB13_4
+; RV32I-NEXT:  # %bb.3:
+; RV32I-NEXT:    mv a2, a3
+; RV32I-NEXT:  .LBB13_4:
 ; RV32I-NEXT:    srli a2, a2, 1
-; RV32I-NEXT:    or a2, a2, a3
-; RV32I-NEXT:    srl a2, a2, t2
-; RV32I-NEXT:    sub a3, a6, t1
-; RV32I-NEXT:    slli a5, t0, 1
-; RV32I-NEXT:    sll a3, a5, a3
-; RV32I-NEXT:    or a2, a2, a3
-; RV32I-NEXT:    bgez a7, .LBB13_7
+; RV32I-NEXT:    not a3, a4
+; RV32I-NEXT:    srl a2, a2, a3
+; RV32I-NEXT:    or a2, a7, a2
+; RV32I-NEXT:    bnez a5, .LBB13_6
+; RV32I-NEXT:  # %bb.5:
+; RV32I-NEXT:    mv a0, a1
 ; RV32I-NEXT:  .LBB13_6:
 ; RV32I-NEXT:    sll a0, a0, a4
-; RV32I-NEXT:    or a2, a2, a0
-; RV32I-NEXT:  .LBB13_7:
+; RV32I-NEXT:    srli a1, a6, 1
+; RV32I-NEXT:    srl a1, a1, a3
+; RV32I-NEXT:    or a1, a0, a1
 ; RV32I-NEXT:    mv a0, a2
 ; RV32I-NEXT:    ret
 ;
 ; RV32ZBT-LABEL: fshl_i64:
 ; RV32ZBT:       # %bb.0:
-; RV32ZBT-NEXT:    sll a7, a1, a4
-; RV32ZBT-NEXT:    andi a5, a4, 63
-; RV32ZBT-NEXT:    addi a6, zero, 31
-; RV32ZBT-NEXT:    sub t0, a6, a5
-; RV32ZBT-NEXT:    srli a1, a0, 1
-; RV32ZBT-NEXT:    srl a1, a1, t0
-; RV32ZBT-NEXT:    or a7, a7, a1
-; RV32ZBT-NEXT:    addi t1, a5, -32
-; RV32ZBT-NEXT:    sll t0, a0, t1
-; RV32ZBT-NEXT:    slti a1, t1, 0
-; RV32ZBT-NEXT:    cmov t0, a1, a7, t0
-; RV32ZBT-NEXT:    not a5, a4
-; RV32ZBT-NEXT:    srli a7, a3, 1
-; RV32ZBT-NEXT:    srl t4, a7, a5
-; RV32ZBT-NEXT:    andi t2, a5, 63
-; RV32ZBT-NEXT:    addi t3, t2, -32
-; RV32ZBT-NEXT:    srai a1, t3, 31
-; RV32ZBT-NEXT:    and a1, a1, t4
-; RV32ZBT-NEXT:    or a1, t0, a1
-; RV32ZBT-NEXT:    fsri a2, a2, a3, 1
-; RV32ZBT-NEXT:    srl a2, a2, a5
-; RV32ZBT-NEXT:    sub a3, a6, t2
-; RV32ZBT-NEXT:    slli a5, a7, 1
-; RV32ZBT-NEXT:    sll a3, a5, a3
-; RV32ZBT-NEXT:    or a2, a2, a3
-; RV32ZBT-NEXT:    srl a3, a7, t3
-; RV32ZBT-NEXT:    slti a5, t3, 0
+; RV32ZBT-NEXT:    srli a5, a4, 5
+; RV32ZBT-NEXT:    andi a5, a5, 1
 ; RV32ZBT-NEXT:    cmov a2, a5, a2, a3
-; RV32ZBT-NEXT:    sll a0, a0, a4
-; RV32ZBT-NEXT:    srai a3, t1, 31
-; RV32ZBT-NEXT:    and a0, a3, a0
-; RV32ZBT-NEXT:    or a0, a0, a2
+; RV32ZBT-NEXT:    cmov a3, a5, a3, a0
+; RV32ZBT-NEXT:    andi a4, a4, 31
+; RV32ZBT-NEXT:    fsl a2, a3, a2, a4
+; RV32ZBT-NEXT:    cmov a0, a5, a0, a1
+; RV32ZBT-NEXT:    fsl a1, a0, a3, a4
+; RV32ZBT-NEXT:    mv a0, a2
 ; RV32ZBT-NEXT:    ret
   %1 = tail call i64 @llvm.fshl.i64(i64 %a, i64 %b, i64 %c)
   ret i64 %1
 }
 
 declare i32 @llvm.fshr.i32(i32, i32, i32)
 
 define i32 @fshr_i32(i32 %a, i32 %b, i32 %c) nounwind {

 ; This test is presented here in case future expansions of the experimental-b
 ; extension introduce instructions that can match more efficiently this pattern.
 
 declare i64 @llvm.fshr.i64(i64, i64, i64)
 
 define i64 @fshr_i64(i64 %a, i64 %b, i64 %c) nounwind {
 ; RV32I-LABEL: fshr_i64:
 ; RV32I:       # %bb.0:
-; RV32I-NEXT:    mv t0, a0
-; RV32I-NEXT:    andi a0, a4, 63
-; RV32I-NEXT:    addi a6, a0, -32
-; RV32I-NEXT:    addi a7, zero, 31
-; RV32I-NEXT:    bltz a6, .LBB15_2
+; RV32I-NEXT:    andi a5, a4, 32
+; RV32I-NEXT:    beqz a5, .LBB15_2
 ; RV32I-NEXT:  # %bb.1:
-; RV32I-NEXT:    srl a0, a3, a6
-; RV32I-NEXT:    j .LBB15_3
+; RV32I-NEXT:    mv a2, a3
 ; RV32I-NEXT:  .LBB15_2:
-; RV32I-NEXT:    srl a2, a2, a4
-; RV32I-NEXT:    sub a0, a7, a0
-; RV32I-NEXT:    slli a5, a3, 1
-; RV32I-NEXT:    sll a0, a5, a0
-; RV32I-NEXT:    or a0, a2, a0
-; RV32I-NEXT:  .LBB15_3:
-; RV32I-NEXT:    not t2, a4
-; RV32I-NEXT:    andi a5, t2, 63
-; RV32I-NEXT:    addi a2, a5, -32
-; RV32I-NEXT:    slli t1, t0, 1
-; RV32I-NEXT:    bltz a2, .LBB15_5
-; RV32I-NEXT:  # %bb.4:
-; RV32I-NEXT:    sll a1, t1, a2
-; RV32I-NEXT:    bltz a6, .LBB15_6
-; RV32I-NEXT:    j .LBB15_7
-; RV32I-NEXT:  .LBB15_5:
-; RV32I-NEXT:    sll a2, t1, t2
-; RV32I-NEXT:    or a0, a0, a2
-; RV32I-NEXT:    lui a2, 524288
-; RV32I-NEXT:    addi a2, a2, -1
-; RV32I-NEXT:    and a2, t0, a2
-; RV32I-NEXT:    sub a5, a7, a5
-; RV32I-NEXT:    srl a2, a2, a5
-; RV32I-NEXT:    srli a5, t0, 31
-; RV32I-NEXT:    slli a1, a1, 1
-; RV32I-NEXT:    or a1, a1, a5
-; RV32I-NEXT:    sll a1, a1, t2
-; RV32I-NEXT:    or a1, a1, a2
-; RV32I-NEXT:    bgez a6, .LBB15_7
+; RV32I-NEXT:    srl a6, a2, a4
+; RV32I-NEXT:    beqz a5, .LBB15_4
+; RV32I-NEXT:  # %bb.3:
+; RV32I-NEXT:    mv a3, a0
+; RV32I-NEXT:  .LBB15_4:
+; RV32I-NEXT:    slli a7, a3, 1
+; RV32I-NEXT:    not t0, a4
+; RV32I-NEXT:    sll a2, a7, t0
+; RV32I-NEXT:    or a6, a2, a6
+; RV32I-NEXT:    srl a3, a3, a4
+; RV32I-NEXT:    beqz a5, .LBB15_6
+; RV32I-NEXT:  # %bb.5:
+; RV32I-NEXT:    mv a0, a1
 ; RV32I-NEXT:  .LBB15_6:
-; RV32I-NEXT:    srl a2, a3, a4
-; RV32I-NEXT:    or a1, a1, a2
-; RV32I-NEXT:  .LBB15_7:
+; RV32I-NEXT:    slli a0, a0, 1
+; RV32I-NEXT:    sll a0, a0, t0
+; RV32I-NEXT:    or a1, a0, a3
+; RV32I-NEXT:    mv a0, a6
 ; RV32I-NEXT:    ret
 ;
 ; RV32ZBT-LABEL: fshr_i64:
 ; RV32ZBT:       # %bb.0:
-; RV32ZBT-NEXT:    srl a7, a2, a4
-; RV32ZBT-NEXT:    andi a5, a4, 63
-; RV32ZBT-NEXT:    addi a6, zero, 31
-; RV32ZBT-NEXT:    sub t0, a6, a5
-; RV32ZBT-NEXT:    slli a2, a3, 1
-; RV32ZBT-NEXT:    sll a2, a2, t0
-; RV32ZBT-NEXT:    or a7, a7, a2
-; RV32ZBT-NEXT:    addi t2, a5, -32
-; RV32ZBT-NEXT:    srl t0, a3, t2
-; RV32ZBT-NEXT:    slti a2, t2, 0
-; RV32ZBT-NEXT:    cmov a7, a2, a7, t0
-; RV32ZBT-NEXT:    not t4, a4
-; RV32ZBT-NEXT:    slli t0, a0, 1
-; RV32ZBT-NEXT:    sll t1, t0, t4
-; RV32ZBT-NEXT:    andi t3, t4, 63
-; RV32ZBT-NEXT:    addi a5, t3, -32
-; RV32ZBT-NEXT:    srai a2, a5, 31
-; RV32ZBT-NEXT:    and a2, a2, t1
-; RV32ZBT-NEXT:    or a7, a2, a7
-; RV32ZBT-NEXT:    lui a2, 524288
-; RV32ZBT-NEXT:    addi a2, a2, -1
-; RV32ZBT-NEXT:    and t1, a0, a2
-; RV32ZBT-NEXT:    sub a2, a6, t3
-; RV32ZBT-NEXT:    srl a2, t1, a2
-; RV32ZBT-NEXT:    fsri a0, a0, a1, 31
-; RV32ZBT-NEXT:    sll a0, a0, t4
-; RV32ZBT-NEXT:    or a0, a0, a2
-; RV32ZBT-NEXT:    sll a1, t0, a5
-; RV32ZBT-NEXT:    slti a2, a5, 0
-; RV32ZBT-NEXT:    cmov a0, a2, a0, a1
-; RV32ZBT-NEXT:    srl a1, a3, a4
-; RV32ZBT-NEXT:    srai a2, t2, 31
-; RV32ZBT-NEXT:    and a1, a2, a1
-; RV32ZBT-NEXT:    or a1, a0, a1
-; RV32ZBT-NEXT:    mv a0, a7
+; RV32ZBT-NEXT:    andi a5, a4, 32
+; RV32ZBT-NEXT:    cmov a6, a5, a0, a3
+; RV32ZBT-NEXT:    cmov a2, a5, a3, a2
+; RV32ZBT-NEXT:    andi a3, a4, 31
+; RV32ZBT-NEXT:    fsr a2, a2, a6, a3
+; RV32ZBT-NEXT:    cmov a0, a5, a1, a0
+; RV32ZBT-NEXT:    fsr a1, a6, a0, a3
+; RV32ZBT-NEXT:    mv a0, a2
 ; RV32ZBT-NEXT:    ret
   %1 = tail call i64 @llvm.fshr.i64(i64 %a, i64 %b, i64 %c)
   ret i64 %1
 }
 
 define i32 @fshri_i32(i32 %a, i32 %b) nounwind {
 ; RV32I-LABEL: fshri_i32:
 ; RV32I:       # %bb.0:
 ; RV32I-NEXT:    srli a1, a1, 5
 ; RV32I-NEXT:    slli a0, a0, 27
 ; RV32I-NEXT:    or a0, a0, a1
 ; RV32I-NEXT:    ret
 ;
 ; RV32ZBT-LABEL: fshri_i32:
 ; RV32ZBT:       # %bb.0:
 ; RV32ZBT-NEXT:    fsri a0, a1, a0, 5
 ; RV32ZBT-NEXT:    ret
   %1 = tail call i32 @llvm.fshr.i32(i32 %a, i32 %b, i32 5)
   ret i32 %1
 }
 
 define i64 @fshri_i64(i64 %a, i64 %b) nounwind {
 ; RV32I-LABEL: fshri_i64:
 ; RV32I:       # %bb.0:
-; RV32I-NEXT:    slli a1, a3, 27
-; RV32I-NEXT:    srli a2, a2, 5
+; RV32I-NEXT:    srli a1, a2, 5
+; RV32I-NEXT:    slli a2, a3, 27
 ; RV32I-NEXT:    or a2, a2, a1
 ; RV32I-NEXT:    srli a1, a3, 5
 ; RV32I-NEXT:    slli a0, a0, 27
 ; RV32I-NEXT:    or a1, a0, a1
 ; RV32I-NEXT:    mv a0, a2
 ; RV32I-NEXT:    ret
 ;
 ; RV32ZBT-LABEL: fshri_i64:

llvm/test/CodeGen/RISCV/shifts.ll

 ; RV64I-NEXT:    ret
   %1 = shl i128 %a, %b
   ret i128 %1
 }
 
 define i64 @fshr64_minsize(i64 %a, i64 %b) minsize nounwind {
 ; RV32I-LABEL: fshr64_minsize:
 ; RV32I:       # %bb.0:
-; RV32I-NEXT:    addi sp, sp, -32
-; RV32I-NEXT:    sw ra, 28(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s0, 24(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s1, 20(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s2, 16(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s3, 12(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s4, 8(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    mv s0, a2
-; RV32I-NEXT:    mv s2, a1
-; RV32I-NEXT:    mv s1, a0
-; RV32I-NEXT:    andi a2, a2, 63
-; RV32I-NEXT:    call __lshrdi3@plt
-; RV32I-NEXT:    mv s3, a0
-; RV32I-NEXT:    mv s4, a1
-; RV32I-NEXT:    neg a0, s0
-; RV32I-NEXT:    andi a2, a0, 63
-; RV32I-NEXT:    mv a0, s1
-; RV32I-NEXT:    mv a1, s2
-; RV32I-NEXT:    call __ashldi3@plt
-; RV32I-NEXT:    or a0, s3, a0
-; RV32I-NEXT:    or a1, s4, a1
-; RV32I-NEXT:    lw s4, 8(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw s3, 12(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw s2, 16(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw s1, 20(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw s0, 24(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw ra, 28(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    addi sp, sp, 32
+; RV32I-NEXT:    andi a4, a2, 32
+; RV32I-NEXT:    mv a3, a0
+; RV32I-NEXT:    beqz a4, .LBB9_2
+; RV32I-NEXT:  # %bb.1:
+; RV32I-NEXT:    mv a3, a1
+; RV32I-NEXT:  .LBB9_2:
+; RV32I-NEXT:    srl a5, a3, a2
+; RV32I-NEXT:    beqz a4, .LBB9_4
+; RV32I-NEXT:  # %bb.3:
+; RV32I-NEXT:    mv a1, a0
+; RV32I-NEXT:  .LBB9_4:
+; RV32I-NEXT:    slli a0, a1, 1
+; RV32I-NEXT:    not a4, a2
+; RV32I-NEXT:    sll a0, a0, a4
+; RV32I-NEXT:    or a0, a0, a5
+; RV32I-NEXT:    srl a1, a1, a2
+; RV32I-NEXT:    slli a2, a3, 1
+; RV32I-NEXT:    sll a2, a2, a4
+; RV32I-NEXT:    or a1, a2, a1
 ; RV32I-NEXT:    ret
 ;
 ; RV64I-LABEL: fshr64_minsize:
 ; RV64I:       # %bb.0:
 ; RV64I-NEXT:    srl a2, a0, a1
 ; RV64I-NEXT:    neg a1, a1
 ; RV64I-NEXT:    sll a0, a0, a1
 ; RV64I-NEXT:    or a0, a2, a0
 ; RV64I-NEXT:    ret
   %res = tail call i64 @llvm.fshr.i64(i64 %a, i64 %a, i64 %b)
   ret i64 %res
 }
 
 define i128 @fshr128_minsize(i128 %a, i128 %b) minsize nounwind {
 ; RV32I-LABEL: fshr128_minsize:
 ; RV32I:       # %bb.0:
-; RV32I-NEXT:    addi sp, sp, -64
-; RV32I-NEXT:    sw ra, 60(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s0, 56(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s1, 52(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s2, 48(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s3, 44(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s4, 40(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s5, 36(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s6, 32(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s7, 28(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s8, 24(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s9, 20(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s10, 16(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    sw s11, 12(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    lw s5, 0(a1)
-; RV32I-NEXT:    lw s6, 4(a1)
-; RV32I-NEXT:    lw s4, 8(a1)
-; RV32I-NEXT:    lw s3, 12(a1)
-; RV32I-NEXT:    lw s11, 0(a2)
-; RV32I-NEXT:    mv s2, a0
-; RV32I-NEXT:    andi s0, s11, 127
-; RV32I-NEXT:    addi a2, s0, -64
-; RV32I-NEXT:    mv a0, s4
-; RV32I-NEXT:    mv a1, s3
-; RV32I-NEXT:    call __lshrdi3@plt
-; RV32I-NEXT:    mv s8, a0
-; RV32I-NEXT:    sw a1, 8(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    mv a0, s5
-; RV32I-NEXT:    mv a1, s6
-; RV32I-NEXT:    mv a2, s0
-; RV32I-NEXT:    call __lshrdi3@plt
-; RV32I-NEXT:    mv s1, a0
-; RV32I-NEXT:    sw a1, 0(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    addi s9, zero, 64
-; RV32I-NEXT:    sub a2, s9, s0
-; RV32I-NEXT:    mv a0, s4
-; RV32I-NEXT:    mv a1, s3
-; RV32I-NEXT:    call __ashldi3@plt
-; RV32I-NEXT:    mv s10, a1
-; RV32I-NEXT:    bgeu s0, s9, .LBB10_2
+; RV32I-NEXT:    lw t2, 8(a1)
+; RV32I-NEXT:    lw a3, 0(a1)
+; RV32I-NEXT:    lw a2, 0(a2)
+; RV32I-NEXT:    lw a7, 4(a1)
+; RV32I-NEXT:    lw t1, 12(a1)
+; RV32I-NEXT:    andi a1, a2, 64
+; RV32I-NEXT:    mv a5, a7
+; RV32I-NEXT:    mv a6, a3
+; RV32I-NEXT:    beqz a1, .LBB10_2
 ; RV32I-NEXT:  # %bb.1:
-; RV32I-NEXT:    or s8, s1, a0
+; RV32I-NEXT:    mv a5, t1
+; RV32I-NEXT:    mv a6, t2
 ; RV32I-NEXT:  .LBB10_2:
-; RV32I-NEXT:    mv s7, s5
-; RV32I-NEXT:    beqz s0, .LBB10_4
+; RV32I-NEXT:    andi a4, a2, 32
+; RV32I-NEXT:    mv t0, a6
+; RV32I-NEXT:    bnez a4, .LBB10_13
 ; RV32I-NEXT:  # %bb.3:
-; RV32I-NEXT:    mv s7, s8
+; RV32I-NEXT:    bnez a1, .LBB10_14
 ; RV32I-NEXT:  .LBB10_4:
-; RV32I-NEXT:    neg a0, s11
-; RV32I-NEXT:    andi s1, a0, 127
-; RV32I-NEXT:    mv a0, s5
-; RV32I-NEXT:    mv a1, s6
-; RV32I-NEXT:    mv a2, s1
-; RV32I-NEXT:    call __ashldi3@plt
-; RV32I-NEXT:    sw a1, 4(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    bgeu s1, s9, .LBB10_6
-; RV32I-NEXT:  # %bb.5:
-; RV32I-NEXT:    or s7, s7, a0
+; RV32I-NEXT:    beqz a4, .LBB10_6
+; RV32I-NEXT:  .LBB10_5:
+; RV32I-NEXT:    mv a5, t2
 ; RV32I-NEXT:  .LBB10_6:
-; RV32I-NEXT:    bltu s0, s9, .LBB10_8
+; RV32I-NEXT:    slli t3, a5, 1
+; RV32I-NEXT:    not a3, a2
+; RV32I-NEXT:    beqz a1, .LBB10_8
 ; RV32I-NEXT:  # %bb.7:
-; RV32I-NEXT:    lw a0, 8(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    j .LBB10_9
+; RV32I-NEXT:    mv t1, a7
 ; RV32I-NEXT:  .LBB10_8:
-; RV32I-NEXT:    lw a0, 0(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    or a0, a0, s10
-; RV32I-NEXT:  .LBB10_9:
-; RV32I-NEXT:    mv s8, s6
-; RV32I-NEXT:    beqz s0, .LBB10_11
-; RV32I-NEXT:  # %bb.10:
-; RV32I-NEXT:    mv s8, a0
-; RV32I-NEXT:  .LBB10_11:
-; RV32I-NEXT:    sub a2, s9, s1
-; RV32I-NEXT:    mv a0, s5
-; RV32I-NEXT:    mv a1, s6
-; RV32I-NEXT:    call __lshrdi3@plt
-; RV32I-NEXT:    mv s10, a0
-; RV32I-NEXT:    sw a1, 8(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    mv a0, s4
-; RV32I-NEXT:    mv a1, s3
-; RV32I-NEXT:    mv a2, s1
-; RV32I-NEXT:    call __ashldi3@plt
-; RV32I-NEXT:    mv s11, a0
-; RV32I-NEXT:    sw a1, 0(sp) # 4-byte Folded Spill
-; RV32I-NEXT:    addi a2, s1, -64
-; RV32I-NEXT:    mv a0, s5
-; RV32I-NEXT:    mv a1, s6
-; RV32I-NEXT:    call __ashldi3@plt
-; RV32I-NEXT:    mv s5, a1
-; RV32I-NEXT:    bgeu s1, s9, .LBB10_13
-; RV32I-NEXT:  # %bb.12:
-; RV32I-NEXT:    lw a0, 4(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    or s8, s8, a0
-; RV32I-NEXT:    or a0, s11, s10
+; RV32I-NEXT:    srl a7, t0, a2
+; RV32I-NEXT:    sll a1, t3, a3
+; RV32I-NEXT:    srl a5, a5, a2
+; RV32I-NEXT:    beqz a4, .LBB10_10
+; RV32I-NEXT:  # %bb.9:
+; RV32I-NEXT:    mv t2, t1
+; RV32I-NEXT:  .LBB10_10:
+; RV32I-NEXT:    or a7, a1, a7
+; RV32I-NEXT:    slli a1, t2, 1
+; RV32I-NEXT:    sll a1, a1, a3
+; RV32I-NEXT:    or a5, a1, a5
+; RV32I-NEXT:    srl a1, t2, a2
+; RV32I-NEXT:    beqz a4, .LBB10_12
+; RV32I-NEXT:  # %bb.11:
+; RV32I-NEXT:    mv t1, a6
+; RV32I-NEXT:  .LBB10_12:
+; RV32I-NEXT:    slli a4, t1, 1
+; RV32I-NEXT:    sll a4, a4, a3
+; RV32I-NEXT:    or a1, a4, a1
+; RV32I-NEXT:    srl a2, t1, a2
+; RV32I-NEXT:    slli a4, t0, 1
+; RV32I-NEXT:    sll a3, a4, a3
+; RV32I-NEXT:    or a2, a3, a2
+; RV32I-NEXT:    sw a2, 12(a0)
+; RV32I-NEXT:    sw a1, 8(a0)
+; RV32I-NEXT:    sw a5, 4(a0)
+; RV32I-NEXT:    sw a7, 0(a0)
+; RV32I-NEXT:    ret
 ; RV32I-NEXT:  .LBB10_13:
-; RV32I-NEXT:    mv s6, s4
-; RV32I-NEXT:    beqz s1, .LBB10_15
-; RV32I-NEXT:  # %bb.14:
-; RV32I-NEXT:    mv s6, a0
-; RV32I-NEXT:  .LBB10_15:
-; RV32I-NEXT:    mv a0, s4
-; RV32I-NEXT:    mv a1, s3
-; RV32I-NEXT:    mv a2, s0
-; RV32I-NEXT:    call __lshrdi3@plt
-; RV32I-NEXT:    bltu s0, s9, .LBB10_21
-; RV32I-NEXT:  # %bb.16:
-; RV32I-NEXT:    bltu s1, s9, .LBB10_22
-; RV32I-NEXT:  .LBB10_17:
-; RV32I-NEXT:    bnez s1, .LBB10_23
-; RV32I-NEXT:  .LBB10_18:
-; RV32I-NEXT:    bgeu s0, s9, .LBB10_20
-; RV32I-NEXT:  .LBB10_19:
-; RV32I-NEXT:    or s3, s3, a1
-; RV32I-NEXT:  .LBB10_20:
-; RV32I-NEXT:    sw s8, 4(s2)
-; RV32I-NEXT:    sw s7, 0(s2)
-; RV32I-NEXT:    sw s3, 12(s2)
-; RV32I-NEXT:    sw s6, 8(s2)
-; RV32I-NEXT:    lw s11, 12(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw s10, 16(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw s9, 20(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw s8, 24(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw s7, 28(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw s6, 32(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw s5, 36(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw s4, 40(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw s3, 44(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw s2, 48(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw s1, 52(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw s0, 56(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw ra, 60(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    addi sp, sp, 64
-; RV32I-NEXT:    ret
-; RV32I-NEXT:  .LBB10_21:
-; RV32I-NEXT:    or s6, s6, a0
-; RV32I-NEXT:    bgeu s1, s9, .LBB10_17
-; RV32I-NEXT:  .LBB10_22:
-; RV32I-NEXT:    lw a0, 8(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    lw a2, 0(sp) # 4-byte Folded Reload
-; RV32I-NEXT:    or s5, a2, a0
-; RV32I-NEXT:    beqz s1, .LBB10_18
-; RV32I-NEXT:  .LBB10_23:
-; RV32I-NEXT:    mv s3, s5
-; RV32I-NEXT:    bltu s0, s9, .LBB10_19
-; RV32I-NEXT:    j .LBB10_20
+; RV32I-NEXT:    mv t0, a5
+; RV32I-NEXT:    beqz a1, .LBB10_4
+; RV32I-NEXT:  .LBB10_14:
+; RV32I-NEXT:    mv t2, a3
+; RV32I-NEXT:    bnez a4, .LBB10_5
+; RV32I-NEXT:    j .LBB10_6
 ;
 ; RV64I-LABEL: fshr128_minsize:
 ; RV64I:       # %bb.0:
-; RV64I-NEXT:    addi sp, sp, -48
-; RV64I-NEXT:    sd ra, 40(sp) # 8-byte Folded Spill
-; RV64I-NEXT:    sd s0, 32(sp) # 8-byte Folded Spill
-; RV64I-NEXT:    sd s1, 24(sp) # 8-byte Folded Spill
-; RV64I-NEXT:    sd s2, 16(sp) # 8-byte Folded Spill
-; RV64I-NEXT:    sd s3, 8(sp) # 8-byte Folded Spill
-; RV64I-NEXT:    sd s4, 0(sp) # 8-byte Folded Spill
-; RV64I-NEXT:    mv s0, a2
-; RV64I-NEXT:    mv s2, a1
-; RV64I-NEXT:    mv s1, a0
-; RV64I-NEXT:    andi a2, a2, 127
-; RV64I-NEXT:    call __lshrti3@plt
-; RV64I-NEXT:    mv s3, a0
-; RV64I-NEXT:    mv s4, a1
-; RV64I-NEXT:    neg a0, s0
-; RV64I-NEXT:    andi a2, a0, 127
-; RV64I-NEXT:    mv a0, s1
-; RV64I-NEXT:    mv a1, s2
-; RV64I-NEXT:    call __ashlti3@plt
-; RV64I-NEXT:    or a0, s3, a0
-; RV64I-NEXT:    or a1, s4, a1
-; RV64I-NEXT:    ld s4, 0(sp) # 8-byte Folded Reload
-; RV64I-NEXT:    ld s3, 8(sp) # 8-byte Folded Reload
-; RV64I-NEXT:    ld s2, 16(sp) # 8-byte Folded Reload
-; RV64I-NEXT:    ld s1, 24(sp) # 8-byte Folded Reload
-; RV64I-NEXT:    ld s0, 32(sp) # 8-byte Folded Reload
-; RV64I-NEXT:    ld ra, 40(sp) # 8-byte Folded Reload
-; RV64I-NEXT:    addi sp, sp, 48
+; RV64I-NEXT:    andi a4, a2, 64
+; RV64I-NEXT:    mv a3, a0
+; RV64I-NEXT:    beqz a4, .LBB10_2
+; RV64I-NEXT:  # %bb.1:
+; RV64I-NEXT:    mv a3, a1
+; RV64I-NEXT:  .LBB10_2:
+; RV64I-NEXT:    srl a5, a3, a2
+; RV64I-NEXT:    beqz a4, .LBB10_4
+; RV64I-NEXT:  # %bb.3:
+; RV64I-NEXT:    mv a1, a0
+; RV64I-NEXT:  .LBB10_4:
+; RV64I-NEXT:    slli a0, a1, 1
+; RV64I-NEXT:    not a4, a2
+; RV64I-NEXT:    sll a0, a0, a4
+; RV64I-NEXT:    or a0, a0, a5
+; RV64I-NEXT:    srl a1, a1, a2
+; RV64I-NEXT:    slli a2, a3, 1
+; RV64I-NEXT:    sll a2, a2, a4
+; RV64I-NEXT:    or a1, a2, a1
 ; RV64I-NEXT:    ret
   %res = tail call i128 @llvm.fshr.i128(i128 %a, i128 %a, i128 %b)
   ret i128 %res
 }

llvm/test/CodeGen/X86/fshl.ll

 ; X64-NEXT:    retq
   %tmp = tail call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 %z)
   ret i32 %tmp
 }
 
 define i64 @var_shift_i64(i64 %x, i64 %y, i64 %z) nounwind {
 ; X86-FAST-LABEL: var_shift_i64:
 ; X86-FAST:       # %bb.0:
-; X86-FAST-NEXT:    pushl %ebx
 ; X86-FAST-NEXT:    pushl %edi
 ; X86-FAST-NEXT:    pushl %esi
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X86-FAST-NEXT:    movb {{[0-9]+}}(%esp), %ch
-; X86-FAST-NEXT:    movb %ch, %cl
-; X86-FAST-NEXT:    notb %cl
-; X86-FAST-NEXT:    shrdl $1, %edi, %esi
-; X86-FAST-NEXT:    shrl %edi
-; X86-FAST-NEXT:    shrdl %cl, %edi, %esi
-; X86-FAST-NEXT:    shrl %cl, %edi
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-FAST-NEXT:    testb $32, %cl
-; X86-FAST-NEXT:    je .LBB5_2
-; X86-FAST-NEXT:  # %bb.1:
-; X86-FAST-NEXT:    movl %edi, %esi
-; X86-FAST-NEXT:    xorl %edi, %edi
-; X86-FAST-NEXT:  .LBB5_2:
-; X86-FAST-NEXT:    movl %ebx, %eax
-; X86-FAST-NEXT:    movb %ch, %cl
-; X86-FAST-NEXT:    shll %cl, %eax
-; X86-FAST-NEXT:    shldl %cl, %ebx, %edx
-; X86-FAST-NEXT:    testb $32, %ch
-; X86-FAST-NEXT:    je .LBB5_4
-; X86-FAST-NEXT:  # %bb.3:
-; X86-FAST-NEXT:    movl %eax, %edx
-; X86-FAST-NEXT:    xorl %eax, %eax
-; X86-FAST-NEXT:  .LBB5_4:
-; X86-FAST-NEXT:    orl %edi, %edx
-; X86-FAST-NEXT:    orl %esi, %eax
+; X86-FAST-NEXT:    jne .LBB5_1
+; X86-FAST-NEXT:  # %bb.2:
+; X86-FAST-NEXT:    movl %edx, %edi
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-FAST-NEXT:    jmp .LBB5_3
+; X86-FAST-NEXT:  .LBB5_1:
+; X86-FAST-NEXT:    movl %esi, %edi
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-FAST-NEXT:  .LBB5_3:
+; X86-FAST-NEXT:    movl %edi, %eax
+; X86-FAST-NEXT:    shldl %cl, %esi, %eax
+; X86-FAST-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-FAST-NEXT:    shldl %cl, %edi, %edx
 ; X86-FAST-NEXT:    popl %esi
 ; X86-FAST-NEXT:    popl %edi
-; X86-FAST-NEXT:    popl %ebx
 ; X86-FAST-NEXT:    retl
 ;
 ; X86-SLOW-LABEL: var_shift_i64:
 ; X86-SLOW:       # %bb.0:
-; X86-SLOW-NEXT:    pushl %ebp
 ; X86-SLOW-NEXT:    pushl %ebx
 ; X86-SLOW-NEXT:    pushl %edi
 ; X86-SLOW-NEXT:    pushl %esi
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X86-SLOW-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-SLOW-NEXT:    shrl %eax
-; X86-SLOW-NEXT:    movl %esi, %edi
-; X86-SLOW-NEXT:    shll $31, %edi
-; X86-SLOW-NEXT:    orl %eax, %edi
-; X86-SLOW-NEXT:    movl %ecx, %eax
-; X86-SLOW-NEXT:    movb %cl, %ch
-; X86-SLOW-NEXT:    notb %ch
-; X86-SLOW-NEXT:    movb %ch, %cl
-; X86-SLOW-NEXT:    shrl %cl, %edi
-; X86-SLOW-NEXT:    shrl %esi
-; X86-SLOW-NEXT:    leal (%esi,%esi), %ebp
-; X86-SLOW-NEXT:    movb %al, %cl
-; X86-SLOW-NEXT:    shll %cl, %ebp
-; X86-SLOW-NEXT:    shll %cl, %ebx
-; X86-SLOW-NEXT:    movl %edx, %eax
-; X86-SLOW-NEXT:    shrl %eax
-; X86-SLOW-NEXT:    movb %ch, %cl
-; X86-SLOW-NEXT:    shrl %cl, %eax
-; X86-SLOW-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-SLOW-NEXT:    shll %cl, %edx
-; X86-SLOW-NEXT:    testb $32, {{[0-9]+}}(%esp)
+; X86-SLOW-NEXT:    testb $32, %bl
 ; X86-SLOW-NEXT:    jne .LBB5_1
 ; X86-SLOW-NEXT:  # %bb.2:
-; X86-SLOW-NEXT:    orl %eax, %ebx
+; X86-SLOW-NEXT:    movl %edx, %esi
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-SLOW-NEXT:    jmp .LBB5_3
 ; X86-SLOW-NEXT:  .LBB5_1:
-; X86-SLOW-NEXT:    movl %edx, %ebx
-; X86-SLOW-NEXT:    xorl %edx, %edx
+; X86-SLOW-NEXT:    movl %eax, %esi
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-SLOW-NEXT:  .LBB5_3:
-; X86-SLOW-NEXT:    movb %ch, %cl
+; X86-SLOW-NEXT:    movl %esi, %edi
+; X86-SLOW-NEXT:    movl %ebx, %ecx
+; X86-SLOW-NEXT:    shll %cl, %edi
+; X86-SLOW-NEXT:    shrl %eax
+; X86-SLOW-NEXT:    notb %cl
+; X86-SLOW-NEXT:    shrl %cl, %eax
+; X86-SLOW-NEXT:    orl %edi, %eax
+; X86-SLOW-NEXT:    shrl %esi
 ; X86-SLOW-NEXT:    shrl %cl, %esi
-; X86-SLOW-NEXT:    testb $32, %ch
-; X86-SLOW-NEXT:    jne .LBB5_4
-; X86-SLOW-NEXT:  # %bb.5:
-; X86-SLOW-NEXT:    orl %edi, %ebp
-; X86-SLOW-NEXT:    jmp .LBB5_6
-; X86-SLOW-NEXT:  .LBB5_4:
-; X86-SLOW-NEXT:    movl %esi, %ebp
-; X86-SLOW-NEXT:    xorl %esi, %esi
-; X86-SLOW-NEXT:  .LBB5_6:
-; X86-SLOW-NEXT:    orl %ebp, %edx
-; X86-SLOW-NEXT:    orl %esi, %ebx
-; X86-SLOW-NEXT:    movl %edx, %eax
-; X86-SLOW-NEXT:    movl %ebx, %edx
+; X86-SLOW-NEXT:    movl %ebx, %ecx
+; X86-SLOW-NEXT:    shll %cl, %edx
+; X86-SLOW-NEXT:    orl %esi, %edx
 ; X86-SLOW-NEXT:    popl %esi
 ; X86-SLOW-NEXT:    popl %edi
 ; X86-SLOW-NEXT:    popl %ebx
-; X86-SLOW-NEXT:    popl %ebp
 ; X86-SLOW-NEXT:    retl
 ;
 ; X64-FAST-LABEL: var_shift_i64:
 ; X64-FAST:       # %bb.0:
 ; X64-FAST-NEXT:    movq %rdx, %rcx
 ; X64-FAST-NEXT:    movq %rdi, %rax
 ; X64-FAST-NEXT:    # kill: def $cl killed $cl killed $rcx
 ; X64-FAST-NEXT:    shldq %cl, %rsi, %rax

 
 define i128 @var_shift_i128(i128 %x, i128 %y, i128 %z) nounwind {
 ; X86-FAST-LABEL: var_shift_i128:
 ; X86-FAST:       # %bb.0:
 ; X86-FAST-NEXT:    pushl %ebp
 ; X86-FAST-NEXT:    pushl %ebx
 ; X86-FAST-NEXT:    pushl %edi
 ; X86-FAST-NEXT:    pushl %esi
-; X86-FAST-NEXT:    subl $72, %esp
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-FAST-NEXT:    movl %edx, %edi
-; X86-FAST-NEXT:    shldl $31, %eax, %edi
-; X86-FAST-NEXT:    movl %ebx, %eax
-; X86-FAST-NEXT:    notl %ebx
-; X86-FAST-NEXT:    andl $127, %ebx
-; X86-FAST-NEXT:    movb $64, %cl
-; X86-FAST-NEXT:    subb %bl, %cl
-; X86-FAST-NEXT:    shrl %edx
-; X86-FAST-NEXT:    movl %edx, %ebp
-; X86-FAST-NEXT:    shldl %cl, %edi, %edx
-; X86-FAST-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl %edi, %edx
-; X86-FAST-NEXT:    shll %cl, %edx
-; X86-FAST-NEXT:    testb $32, %cl
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edi
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-FAST-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X86-FAST-NEXT:    testb $64, %cl
 ; X86-FAST-NEXT:    jne .LBB6_1
 ; X86-FAST-NEXT:  # %bb.2:
-; X86-FAST-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    jmp .LBB6_3
-; X86-FAST-NEXT:  .LBB6_1:
-; X86-FAST-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-FAST-NEXT:  .LBB6_3:
-; X86-FAST-NEXT:    andl $127, %eax
-; X86-FAST-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movb %al, %ch
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    movb %ch, %cl
-; X86-FAST-NEXT:    shldl %cl, %esi, %eax
-; X86-FAST-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movb %bl, %cl
-; X86-FAST-NEXT:    addb $-64, %cl
 ; X86-FAST-NEXT:    movl %edi, %eax
-; X86-FAST-NEXT:    movl %ebp, %edx
-; X86-FAST-NEXT:    shrdl %cl, %ebp, %eax
-; X86-FAST-NEXT:    shrl %cl, %ebp
+; X86-FAST-NEXT:    movl %esi, %edi
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-FAST-NEXT:    movl %ebx, %ebp
+; X86-FAST-NEXT:    movl %edx, %ebx
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-FAST-NEXT:    testb $32, %cl
-; X86-FAST-NEXT:    jne .LBB6_4
-; X86-FAST-NEXT:  # %bb.5:
-; X86-FAST-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    jmp .LBB6_6
+; X86-FAST-NEXT:    je .LBB6_5
 ; X86-FAST-NEXT:  .LBB6_4:
-; X86-FAST-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-FAST-NEXT:  .LBB6_6:
+; X86-FAST-NEXT:    movl %edx, %esi
+; X86-FAST-NEXT:    movl %edi, %edx
+; X86-FAST-NEXT:    movl %ebx, %edi
+; X86-FAST-NEXT:    movl %eax, %ebx
+; X86-FAST-NEXT:    jmp .LBB6_6
+; X86-FAST-NEXT:  .LBB6_1:
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X86-FAST-NEXT:    movb %ch, %cl
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    shldl %cl, %eax, %ebp
-; X86-FAST-NEXT:    shll %cl, %eax
-; X86-FAST-NEXT:    shll %cl, %esi
-; X86-FAST-NEXT:    testb $32, %ch
-; X86-FAST-NEXT:    jne .LBB6_7
-; X86-FAST-NEXT:  # %bb.8:
-; X86-FAST-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    jmp .LBB6_9
-; X86-FAST-NEXT:  .LBB6_7:
-; X86-FAST-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl %eax, %ebp
-; X86-FAST-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-FAST-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-FAST-NEXT:  .LBB6_9:
-; X86-FAST-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-FAST-NEXT:    jb .LBB6_11
-; X86-FAST-NEXT:  # %bb.10:
-; X86-FAST-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-FAST-NEXT:  .LBB6_11:
-; X86-FAST-NEXT:    movb %bl, %cl
-; X86-FAST-NEXT:    shrdl %cl, %edx, %edi
-; X86-FAST-NEXT:    shrl %cl, %edx
-; X86-FAST-NEXT:    shldl $31, %eax, %esi
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X86-FAST-NEXT:    shrdl $1, %ebp, %eax
-; X86-FAST-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    shrdl %cl, %esi, %eax
-; X86-FAST-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl %esi, %eax
-; X86-FAST-NEXT:    shrl %cl, %eax
-; X86-FAST-NEXT:    testb $32, %bl
-; X86-FAST-NEXT:    je .LBB6_13
-; X86-FAST-NEXT:  # %bb.12:
-; X86-FAST-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl %edx, %edi
-; X86-FAST-NEXT:    xorl %eax, %eax
-; X86-FAST-NEXT:    xorl %edx, %edx
-; X86-FAST-NEXT:  .LBB6_13:
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X86-FAST-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-FAST-NEXT:    jb .LBB6_15
-; X86-FAST-NEXT:  # %bb.14:
-; X86-FAST-NEXT:    xorl %ebp, %ebp
-; X86-FAST-NEXT:  .LBB6_15:
-; X86-FAST-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movb $64, %cl
-; X86-FAST-NEXT:    subb %ch, %cl
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X86-FAST-NEXT:    shrl %cl, %ebp
-; X86-FAST-NEXT:    movb %cl, {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Spill
-; X86-FAST-NEXT:    testb $32, %cl
-; X86-FAST-NEXT:    movl $0, %edx
-; X86-FAST-NEXT:    jne .LBB6_17
-; X86-FAST-NEXT:  # %bb.16:
-; X86-FAST-NEXT:    movl %ebp, %edx
-; X86-FAST-NEXT:  .LBB6_17:
-; X86-FAST-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    addb $-64, %ch
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X86-FAST-NEXT:    movl %edi, %esi
-; X86-FAST-NEXT:    movb %ch, %cl
-; X86-FAST-NEXT:    shll %cl, %esi
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-FAST-NEXT:    shldl %cl, %edi, %edx
-; X86-FAST-NEXT:    testb $32, %ch
-; X86-FAST-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    jne .LBB6_19
-; X86-FAST-NEXT:  # %bb.18:
-; X86-FAST-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:  .LBB6_19:
-; X86-FAST-NEXT:    cmpl $64, %ebx
-; X86-FAST-NEXT:    jb .LBB6_21
-; X86-FAST-NEXT:  # %bb.20:
-; X86-FAST-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-FAST-NEXT:  .LBB6_21:
-; X86-FAST-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X86-FAST-NEXT:    jae .LBB6_23
-; X86-FAST-NEXT:  # %bb.22:
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X86-FAST-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X86-FAST-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X86-FAST-NEXT:  .LBB6_23:
-; X86-FAST-NEXT:    testb $32, %ch
-; X86-FAST-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-FAST-NEXT:    jne .LBB6_25
-; X86-FAST-NEXT:  # %bb.24:
-; X86-FAST-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:  .LBB6_25:
-; X86-FAST-NEXT:    cmpl $64, %ebx
-; X86-FAST-NEXT:    jb .LBB6_27
-; X86-FAST-NEXT:  # %bb.26:
-; X86-FAST-NEXT:    xorl %edx, %edx
-; X86-FAST-NEXT:  .LBB6_27:
-; X86-FAST-NEXT:    movb {{[-0-9]+}}(%e{{[sb]}}p), %cl # 1-byte Reload
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-FAST-NEXT:    shrdl %cl, %esi, %edi
 ; X86-FAST-NEXT:    testb $32, %cl
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X86-FAST-NEXT:    jne .LBB6_29
-; X86-FAST-NEXT:  # %bb.28:
-; X86-FAST-NEXT:    movl %edi, %ebp
-; X86-FAST-NEXT:  .LBB6_29:
-; X86-FAST-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X86-FAST-NEXT:    jae .LBB6_31
-; X86-FAST-NEXT:  # %bb.30:
-; X86-FAST-NEXT:    orl %ebp, %esi
-; X86-FAST-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:  .LBB6_31:
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X86-FAST-NEXT:    cmpl $64, %ebx
-; X86-FAST-NEXT:    jae .LBB6_33
-; X86-FAST-NEXT:  # %bb.32:
-; X86-FAST-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X86-FAST-NEXT:    jne .LBB6_4
+; X86-FAST-NEXT:  .LBB6_5:
 ; X86-FAST-NEXT:    movl %eax, %ebp
-; X86-FAST-NEXT:  .LBB6_33:
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X86-FAST-NEXT:    cmpl $64, %ebx
-; X86-FAST-NEXT:    jae .LBB6_35
-; X86-FAST-NEXT:  # %bb.34:
-; X86-FAST-NEXT:    movl %edx, %ecx
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X86-FAST-NEXT:    orl %eax, %edx
-; X86-FAST-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl %ecx, %edx
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X86-FAST-NEXT:  .LBB6_35:
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    testl %ebx, %ebx
-; X86-FAST-NEXT:    je .LBB6_37
-; X86-FAST-NEXT:  # %bb.36:
-; X86-FAST-NEXT:    movl %ebp, %ecx
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X86-FAST-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:  .LBB6_37:
-; X86-FAST-NEXT:    orl %ecx, %edi
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X86-FAST-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X86-FAST-NEXT:    cmpl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-FAST-NEXT:    je .LBB6_39
-; X86-FAST-NEXT:  # %bb.38:
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X86-FAST-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X86-FAST-NEXT:  .LBB6_39:
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X86-FAST-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X86-FAST-NEXT:    orl %edx, %esi
-; X86-FAST-NEXT:    movl %ecx, 12(%eax)
-; X86-FAST-NEXT:    movl %esi, 8(%eax)
-; X86-FAST-NEXT:    movl %edi, 4(%eax)
-; X86-FAST-NEXT:    movl %ebx, (%eax)
-; X86-FAST-NEXT:    addl $72, %esp
+; X86-FAST-NEXT:  .LBB6_6:
+; X86-FAST-NEXT:    movl %ebx, %eax
+; X86-FAST-NEXT:    shldl %cl, %ebp, %eax
+; X86-FAST-NEXT:    movl %edi, %ebp
+; X86-FAST-NEXT:    shldl %cl, %ebx, %ebp
+; X86-FAST-NEXT:    movl %edx, %ebx
+; X86-FAST-NEXT:    shldl %cl, %edi, %ebx
+; X86-FAST-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-FAST-NEXT:    shldl %cl, %edx, %esi
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-FAST-NEXT:    movl %esi, 12(%ecx)
+; X86-FAST-NEXT:    movl %ebx, 8(%ecx)
+; X86-FAST-NEXT:    movl %ebp, 4(%ecx)
+; X86-FAST-NEXT:    movl %eax, (%ecx)
+; X86-FAST-NEXT:    movl %ecx, %eax
 ; X86-FAST-NEXT:    popl %esi
 ; X86-FAST-NEXT:    popl %edi
 ; X86-FAST-NEXT:    popl %ebx
 ; X86-FAST-NEXT:    popl %ebp
 ; X86-FAST-NEXT:    retl $4
 ;
 ; X86-SLOW-LABEL: var_shift_i128:
 ; X86-SLOW:       # %bb.0:
 ; X86-SLOW-NEXT:    pushl %ebp
 ; X86-SLOW-NEXT:    pushl %ebx
 ; X86-SLOW-NEXT:    pushl %edi
 ; X86-SLOW-NEXT:    pushl %esi
-; X86-SLOW-NEXT:    subl $76, %esp
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X86-SLOW-NEXT:    pushl %eax
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edi
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SLOW-NEXT:    andl $127, %eax
-; X86-SLOW-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    # kill: def $al killed $al killed $eax
-; X86-SLOW-NEXT:    movl %eax, %ecx
-; X86-SLOW-NEXT:    shll %cl, %edx
-; X86-SLOW-NEXT:    movl %ebx, %esi
-; X86-SLOW-NEXT:    shrl %esi
-; X86-SLOW-NEXT:    movb %al, %ah
-; X86-SLOW-NEXT:    notb %ah
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movb %ah, %cl
-; X86-SLOW-NEXT:    shrl %cl, %esi
-; X86-SLOW-NEXT:    movl %esi, (%esp) # 4-byte Spill
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X86-SLOW-NEXT:    movl %eax, %ecx
-; X86-SLOW-NEXT:    shll %cl, %ebp
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X86-SLOW-NEXT:    shrl %edi
-; X86-SLOW-NEXT:    movb %ah, %cl
-; X86-SLOW-NEXT:    shrl %cl, %edi
-; X86-SLOW-NEXT:    movl %ebx, %esi
-; X86-SLOW-NEXT:    movl %eax, %ecx
-; X86-SLOW-NEXT:    shll %cl, %esi
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X86-SLOW-NEXT:    shll %cl, %ebx
-; X86-SLOW-NEXT:    movb %al, {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Spill
-; X86-SLOW-NEXT:    testb $32, %al
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SLOW-NEXT:    testb $64, %al
 ; X86-SLOW-NEXT:    jne .LBB6_1
 ; X86-SLOW-NEXT:  # %bb.2:
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    orl (%esp), %edx # 4-byte Folded Reload
-; X86-SLOW-NEXT:    movl %edx, (%esp) # 4-byte Spill
-; X86-SLOW-NEXT:    orl %edi, %ebp
+; X86-SLOW-NEXT:    movl %ebp, %ecx
+; X86-SLOW-NEXT:    movl %edi, %ebp
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-SLOW-NEXT:    movl %edx, %ebx
+; X86-SLOW-NEXT:    movl %esi, %edx
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-SLOW-NEXT:    jmp .LBB6_3
 ; X86-SLOW-NEXT:  .LBB6_1:
-; X86-SLOW-NEXT:    movl %ebx, %ebp
-; X86-SLOW-NEXT:    movl %esi, (%esp) # 4-byte Spill
-; X86-SLOW-NEXT:    xorl %ebx, %ebx
-; X86-SLOW-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-SLOW-NEXT:  .LBB6_3:
-; X86-SLOW-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X86-SLOW-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X86-SLOW-NEXT:    movb {{[-0-9]+}}(%e{{[sb]}}p), %dl # 1-byte Reload
-; X86-SLOW-NEXT:    jb .LBB6_5
-; X86-SLOW-NEXT:  # %bb.4:
-; X86-SLOW-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-SLOW-NEXT:  .LBB6_5:
-; X86-SLOW-NEXT:    shrl %edi
-; X86-SLOW-NEXT:    notl %ebx
-; X86-SLOW-NEXT:    andl $127, %ebx
-; X86-SLOW-NEXT:    movl %edi, %ebp
-; X86-SLOW-NEXT:    movl %ebx, %ecx
-; X86-SLOW-NEXT:    shrl %cl, %ebp
-; X86-SLOW-NEXT:    movl %esi, %ecx
-; X86-SLOW-NEXT:    shrl %ecx
-; X86-SLOW-NEXT:    movl %eax, %esi
-; X86-SLOW-NEXT:    shll $31, %esi
-; X86-SLOW-NEXT:    orl %ecx, %esi
-; X86-SLOW-NEXT:    movl %esi, %ecx
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl %ebx, %ecx
-; X86-SLOW-NEXT:    shrl %cl, %esi
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    testb $32, %bl
-; X86-SLOW-NEXT:    movl $0, %esi
-; X86-SLOW-NEXT:    movl $0, %ecx
-; X86-SLOW-NEXT:    jne .LBB6_7
-; X86-SLOW-NEXT:  # %bb.6:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X86-SLOW-NEXT:    movl %ebp, %ecx
-; X86-SLOW-NEXT:  .LBB6_7:
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    shrl %eax
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-SLOW-NEXT:    shll $31, %esi
-; X86-SLOW-NEXT:    orl %eax, %esi
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl %ebx, %ecx
-; X86-SLOW-NEXT:    shrl %cl, %esi
-; X86-SLOW-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    addl %edi, %edi
-; X86-SLOW-NEXT:    notb %cl
-; X86-SLOW-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movb %cl, {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Spill
-; X86-SLOW-NEXT:    shll %cl, %edi
-; X86-SLOW-NEXT:    testb $32, %bl
-; X86-SLOW-NEXT:    jne .LBB6_9
-; X86-SLOW-NEXT:  # %bb.8:
-; X86-SLOW-NEXT:    orl %esi, %edi
-; X86-SLOW-NEXT:    movl %edi, %ebp
-; X86-SLOW-NEXT:  .LBB6_9:
-; X86-SLOW-NEXT:    movb %bl, %dh
-; X86-SLOW-NEXT:    addb $-64, %dh
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X86-SLOW-NEXT:    movb %dh, %cl
-; X86-SLOW-NEXT:    shrl %cl, %esi
-; X86-SLOW-NEXT:    testb $32, %dh
-; X86-SLOW-NEXT:    movl $0, %ecx
-; X86-SLOW-NEXT:    movl (%esp), %eax # 4-byte Reload
-; X86-SLOW-NEXT:    jne .LBB6_11
-; X86-SLOW-NEXT:  # %bb.10:
-; X86-SLOW-NEXT:    movl %esi, %ecx
-; X86-SLOW-NEXT:  .LBB6_11:
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X86-SLOW-NEXT:    jb .LBB6_13
-; X86-SLOW-NEXT:  # %bb.12:
-; X86-SLOW-NEXT:    xorl %eax, %eax
-; X86-SLOW-NEXT:  .LBB6_13:
-; X86-SLOW-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X86-SLOW-NEXT:    movb $64, %ch
-; X86-SLOW-NEXT:    movb $64, %ah
-; X86-SLOW-NEXT:    subb %dl, %ah
-; X86-SLOW-NEXT:    movb %ah, %cl
-; X86-SLOW-NEXT:    shrl %cl, %esi
-; X86-SLOW-NEXT:    notb %cl
-; X86-SLOW-NEXT:    leal (%ebp,%ebp), %edi
+; X86-SLOW-NEXT:    testb $32, %al
+; X86-SLOW-NEXT:    jne .LBB6_4
+; X86-SLOW-NEXT:  # %bb.5:
+; X86-SLOW-NEXT:    movl %ecx, %ebx
+; X86-SLOW-NEXT:    movl %edi, (%esp) # 4-byte Spill
+; X86-SLOW-NEXT:    jmp .LBB6_6
+; X86-SLOW-NEXT:  .LBB6_4:
+; X86-SLOW-NEXT:    movl %esi, (%esp) # 4-byte Spill
+; X86-SLOW-NEXT:    movl %ebp, %esi
+; X86-SLOW-NEXT:    movl %edx, %ebp
+; X86-SLOW-NEXT:    movl %ecx, %edx
+; X86-SLOW-NEXT:  .LBB6_6:
+; X86-SLOW-NEXT:    movl %edx, %edi
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SLOW-NEXT:    movl %eax, %ecx
 ; X86-SLOW-NEXT:    shll %cl, %edi
-; X86-SLOW-NEXT:    movb %ah, %cl
-; X86-SLOW-NEXT:    shrl %cl, %ebp
-; X86-SLOW-NEXT:    testb $32, %ah
-; X86-SLOW-NEXT:    jne .LBB6_14
-; X86-SLOW-NEXT:  # %bb.15:
-; X86-SLOW-NEXT:    orl %esi, %edi
-; X86-SLOW-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl %edi, %ebp
-; X86-SLOW-NEXT:    jmp .LBB6_16
-; X86-SLOW-NEXT:  .LBB6_14:
-; X86-SLOW-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-SLOW-NEXT:  .LBB6_16:
-; X86-SLOW-NEXT:    addb $-64, %dl
-; X86-SLOW-NEXT:    movb %dl, %cl
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-SLOW-NEXT:    shrl %ebx
+; X86-SLOW-NEXT:    movb %al, %ch
+; X86-SLOW-NEXT:    notb %ch
+; X86-SLOW-NEXT:    movb %ch, %cl
+; X86-SLOW-NEXT:    shrl %cl, %ebx
+; X86-SLOW-NEXT:    orl %edi, %ebx
+; X86-SLOW-NEXT:    movl %ebp, %edi
+; X86-SLOW-NEXT:    movb %al, %cl
 ; X86-SLOW-NEXT:    shll %cl, %edi
-; X86-SLOW-NEXT:    notb %cl
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X86-SLOW-NEXT:    shrl %cl, %eax
-; X86-SLOW-NEXT:    movb %dl, %cl
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-SLOW-NEXT:    shll %cl, %esi
-; X86-SLOW-NEXT:    testb $32, %dl
-; X86-SLOW-NEXT:    jne .LBB6_17
-; X86-SLOW-NEXT:  # %bb.18:
-; X86-SLOW-NEXT:    orl %eax, %edi
-; X86-SLOW-NEXT:    cmpl $64, %ebx
-; X86-SLOW-NEXT:    jae .LBB6_20
-; X86-SLOW-NEXT:    jmp .LBB6_21
-; X86-SLOW-NEXT:  .LBB6_17:
+; X86-SLOW-NEXT:    shrl %edx
+; X86-SLOW-NEXT:    movb %ch, %cl
+; X86-SLOW-NEXT:    shrl %cl, %edx
+; X86-SLOW-NEXT:    orl %edi, %edx
 ; X86-SLOW-NEXT:    movl %esi, %edi
-; X86-SLOW-NEXT:    xorl %esi, %esi
-; X86-SLOW-NEXT:    cmpl $64, %ebx
-; X86-SLOW-NEXT:    jb .LBB6_21
-; X86-SLOW-NEXT:  .LBB6_20:
-; X86-SLOW-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-SLOW-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-SLOW-NEXT:  .LBB6_21:
-; X86-SLOW-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-SLOW-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    jae .LBB6_23
-; X86-SLOW-NEXT:  # %bb.22:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X86-SLOW-NEXT:    orl %ebp, %esi
-; X86-SLOW-NEXT:  .LBB6_23:
-; X86-SLOW-NEXT:    movl (%esp), %ebp # 4-byte Reload
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X86-SLOW-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    jae .LBB6_25
-; X86-SLOW-NEXT:  # %bb.24:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X86-SLOW-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:  .LBB6_25:
-; X86-SLOW-NEXT:    shrl %edi
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-SLOW-NEXT:    shll $31, %esi
-; X86-SLOW-NEXT:    orl %edi, %esi
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movb %bl, %cl
-; X86-SLOW-NEXT:    shrl %cl, %esi
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X86-SLOW-NEXT:    addl %edi, %edi
-; X86-SLOW-NEXT:    movb {{[-0-9]+}}(%e{{[sb]}}p), %cl # 1-byte Reload
+; X86-SLOW-NEXT:    movb %al, %cl
 ; X86-SLOW-NEXT:    shll %cl, %edi
-; X86-SLOW-NEXT:    testb $32, %bl
-; X86-SLOW-NEXT:    jne .LBB6_27
-; X86-SLOW-NEXT:  # %bb.26:
-; X86-SLOW-NEXT:    orl %esi, %edi
-; X86-SLOW-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:  .LBB6_27:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X86-SLOW-NEXT:    movl %edi, %eax
-; X86-SLOW-NEXT:    movb %dh, %cl
-; X86-SLOW-NEXT:    shrl %cl, %eax
-; X86-SLOW-NEXT:    notb %cl
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X86-SLOW-NEXT:    shll %cl, %esi
-; X86-SLOW-NEXT:    testb $32, %dh
-; X86-SLOW-NEXT:    jne .LBB6_29
-; X86-SLOW-NEXT:  # %bb.28:
-; X86-SLOW-NEXT:    orl %eax, %esi
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:  .LBB6_29:
-; X86-SLOW-NEXT:    subb %bl, %ch
-; X86-SLOW-NEXT:    movl %edi, %eax
+; X86-SLOW-NEXT:    shrl %ebp
 ; X86-SLOW-NEXT:    movb %ch, %cl
+; X86-SLOW-NEXT:    shrl %cl, %ebp
+; X86-SLOW-NEXT:    orl %edi, %ebp
+; X86-SLOW-NEXT:    movb %al, %cl
+; X86-SLOW-NEXT:    movl (%esp), %eax # 4-byte Reload
 ; X86-SLOW-NEXT:    shll %cl, %eax
-; X86-SLOW-NEXT:    shrl %edi
-; X86-SLOW-NEXT:    notb %cl
-; X86-SLOW-NEXT:    shrl %cl, %edi
+; X86-SLOW-NEXT:    shrl %esi
 ; X86-SLOW-NEXT:    movb %ch, %cl
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X86-SLOW-NEXT:    shll %cl, %edx
-; X86-SLOW-NEXT:    testb $32, %ch
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-SLOW-NEXT:    movl %edi, %ecx
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X86-SLOW-NEXT:    jne .LBB6_30
-; X86-SLOW-NEXT:  # %bb.31:
-; X86-SLOW-NEXT:    orl %ecx, %edx
-; X86-SLOW-NEXT:    movl %edx, %ecx
-; X86-SLOW-NEXT:    cmpl $64, %ebx
-; X86-SLOW-NEXT:    jb .LBB6_33
-; X86-SLOW-NEXT:    jmp .LBB6_34
-; X86-SLOW-NEXT:  .LBB6_30:
-; X86-SLOW-NEXT:    movl %eax, %ecx
-; X86-SLOW-NEXT:    xorl %eax, %eax
-; X86-SLOW-NEXT:    cmpl $64, %ebx
-; X86-SLOW-NEXT:    jae .LBB6_34
-; X86-SLOW-NEXT:  .LBB6_33:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X86-SLOW-NEXT:    orl %eax, %edx
-; X86-SLOW-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:  .LBB6_34:
-; X86-SLOW-NEXT:    cmpl $64, %ebx
-; X86-SLOW-NEXT:    jb .LBB6_35
-; X86-SLOW-NEXT:  # %bb.36:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X86-SLOW-NEXT:    jmp .LBB6_37
-; X86-SLOW-NEXT:  .LBB6_35:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X86-SLOW-NEXT:    orl %ecx, %eax
-; X86-SLOW-NEXT:    movl %eax, %ecx
-; X86-SLOW-NEXT:  .LBB6_37:
+; X86-SLOW-NEXT:    shrl %cl, %esi
+; X86-SLOW-NEXT:    orl %eax, %esi
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SLOW-NEXT:    testl %ebx, %ebx
-; X86-SLOW-NEXT:    je .LBB6_39
-; X86-SLOW-NEXT:  # %bb.38:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X86-SLOW-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl %ecx, %ebx
-; X86-SLOW-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:  .LBB6_39:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X86-SLOW-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X86-SLOW-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X86-SLOW-NEXT:    cmpl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-SLOW-NEXT:    je .LBB6_41
-; X86-SLOW-NEXT:  # %bb.40:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X86-SLOW-NEXT:  .LBB6_41:
-; X86-SLOW-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X86-SLOW-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X86-SLOW-NEXT:    movl %esi, 12(%eax)
-; X86-SLOW-NEXT:    movl %edi, 8(%eax)
+; X86-SLOW-NEXT:    movl %ebp, 8(%eax)
+; X86-SLOW-NEXT:    movl %edx, 4(%eax)
 ; X86-SLOW-NEXT:    movl %ebx, (%eax)
-; X86-SLOW-NEXT:    movl %ebp, 4(%eax)
-; X86-SLOW-NEXT:    addl $76, %esp
+; X86-SLOW-NEXT:    addl $4, %esp
 ; X86-SLOW-NEXT:    popl %esi
 ; X86-SLOW-NEXT:    popl %edi
 ; X86-SLOW-NEXT:    popl %ebx
 ; X86-SLOW-NEXT:    popl %ebp
 ; X86-SLOW-NEXT:    retl $4
 ;
 ; X64-FAST-LABEL: var_shift_i128:
 ; X64-FAST:       # %bb.0:
-; X64-FAST-NEXT:    movq %r8, %r9
-; X64-FAST-NEXT:    movq %rcx, %r8
-; X64-FAST-NEXT:    movl %r9d, %ecx
-; X64-FAST-NEXT:    shldq %cl, %rdi, %rsi
-; X64-FAST-NEXT:    shrdq $1, %r8, %rdx
-; X64-FAST-NEXT:    shrq %r8
-; X64-FAST-NEXT:    notb %cl
-; X64-FAST-NEXT:    shrdq %cl, %r8, %rdx
-; X64-FAST-NEXT:    shrq %cl, %r8
-; X64-FAST-NEXT:    xorl %eax, %eax
-; X64-FAST-NEXT:    testb $64, %cl
-; X64-FAST-NEXT:    cmovneq %r8, %rdx
-; X64-FAST-NEXT:    cmovneq %rax, %r8
-; X64-FAST-NEXT:    movl %r9d, %ecx
-; X64-FAST-NEXT:    shlq %cl, %rdi
-; X64-FAST-NEXT:    testb $64, %r9b
+; X64-FAST-NEXT:    testb $64, %r8b
 ; X64-FAST-NEXT:    cmovneq %rdi, %rsi
-; X64-FAST-NEXT:    cmoveq %rdi, %rax
-; X64-FAST-NEXT:    orq %rdx, %rax
-; X64-FAST-NEXT:    orq %rsi, %r8
-; X64-FAST-NEXT:    movq %r8, %rdx
+; X64-FAST-NEXT:    cmoveq %rcx, %rdx
+; X64-FAST-NEXT:    cmovneq %rcx, %rdi
+; X64-FAST-NEXT:    movq %rdi, %rax
+; X64-FAST-NEXT:    movl %r8d, %ecx
+; X64-FAST-NEXT:    shldq %cl, %rdx, %rax
+; X64-FAST-NEXT:    shldq %cl, %rdi, %rsi
+; X64-FAST-NEXT:    movq %rsi, %rdx
 ; X64-FAST-NEXT:    retq
 ;
 ; X64-SLOW-LABEL: var_shift_i128:
 ; X64-SLOW:       # %bb.0:
-; X64-SLOW-NEXT:    movq %rcx, %r11
-; X64-SLOW-NEXT:    movq %rdx, %r9
+; X64-SLOW-NEXT:    testb $64, %r8b
+; X64-SLOW-NEXT:    cmovneq %rdi, %rsi
+; X64-SLOW-NEXT:    cmoveq %rcx, %rdx
+; X64-SLOW-NEXT:    cmovneq %rcx, %rdi
+; X64-SLOW-NEXT:    movq %rdi, %rax
 ; X64-SLOW-NEXT:    movl %r8d, %ecx
-; X64-SLOW-NEXT:    shlq %cl, %rsi
-; X64-SLOW-NEXT:    movq %rdi, %rdx
+; X64-SLOW-NEXT:    shlq %cl, %rax
 ; X64-SLOW-NEXT:    shrq %rdx
-; X64-SLOW-NEXT:    movl %r8d, %r10d
-; X64-SLOW-NEXT:    notb %r10b
-; X64-SLOW-NEXT:    movl %r10d, %ecx
+; X64-SLOW-NEXT:    movl %r8d, %r9d
+; X64-SLOW-NEXT:    notb %r9b
+; X64-SLOW-NEXT:    movl %r9d, %ecx
 ; X64-SLOW-NEXT:    shrq %cl, %rdx
-; X64-SLOW-NEXT:    orq %rsi, %rdx
-; X64-SLOW-NEXT:    shrq %r9
-; X64-SLOW-NEXT:    movq %r11, %rax
-; X64-SLOW-NEXT:    shlq $63, %rax
-; X64-SLOW-NEXT:    orq %r9, %rax
-; X64-SLOW-NEXT:    shrq %cl, %rax
-; X64-SLOW-NEXT:    shrq %r11
-; X64-SLOW-NEXT:    leaq (%r11,%r11), %rsi
+; X64-SLOW-NEXT:    orq %rdx, %rax
 ; X64-SLOW-NEXT:    movl %r8d, %ecx
 ; X64-SLOW-NEXT:    shlq %cl, %rsi
-; X64-SLOW-NEXT:    orq %rax, %rsi
-; X64-SLOW-NEXT:    movl %r10d, %ecx
-; X64-SLOW-NEXT:    shrq %cl, %r11
-; X64-SLOW-NEXT:    xorl %eax, %eax
-; X64-SLOW-NEXT:    testb $64, %r10b
-; X64-SLOW-NEXT:    cmovneq %r11, %rsi
-; X64-SLOW-NEXT:    cmovneq %rax, %r11
-; X64-SLOW-NEXT:    movl %r8d, %ecx
-; X64-SLOW-NEXT:    shlq %cl, %rdi
-; X64-SLOW-NEXT:    testb $64, %r8b
-; X64-SLOW-NEXT:    cmovneq %rdi, %rdx
-; X64-SLOW-NEXT:    cmoveq %rdi, %rax
-; X64-SLOW-NEXT:    orq %rsi, %rax
-; X64-SLOW-NEXT:    orq %r11, %rdx
+; X64-SLOW-NEXT:    shrq %rdi
+; X64-SLOW-NEXT:    movl %r9d, %ecx
+; X64-SLOW-NEXT:    shrq %cl, %rdi
+; X64-SLOW-NEXT:    orq %rsi, %rdi
+; X64-SLOW-NEXT:    movq %rdi, %rdx
 ; X64-SLOW-NEXT:    retq
   %tmp = tail call i128 @llvm.fshl.i128(i128 %x, i128 %y, i128 %z)
   ret i128 %tmp
 }
 
 ;
 ; Const Funnel Shift
 ;

llvm/test/CodeGen/X86/fshr.ll

 ; X64-NEXT:    retq
   %tmp = tail call i32 @llvm.fshr.i32(i32 %x, i32 %y, i32 %z)
   ret i32 %tmp
 }
 
 define i64 @var_shift_i64(i64 %x, i64 %y, i64 %z) nounwind {
 ; X86-FAST-LABEL: var_shift_i64:
 ; X86-FAST:       # %bb.0:
-; X86-FAST-NEXT:    pushl %ebp
-; X86-FAST-NEXT:    pushl %ebx
-; X86-FAST-NEXT:    pushl %edi
 ; X86-FAST-NEXT:    pushl %esi
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebp
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-FAST-NEXT:    movb {{[0-9]+}}(%esp), %bl
-; X86-FAST-NEXT:    movb %bl, %ch
-; X86-FAST-NEXT:    notb %ch
-; X86-FAST-NEXT:    shldl $1, %eax, %edx
-; X86-FAST-NEXT:    addl %eax, %eax
-; X86-FAST-NEXT:    movb %ch, %cl
-; X86-FAST-NEXT:    shldl %cl, %eax, %edx
-; X86-FAST-NEXT:    movl %ebp, %edi
-; X86-FAST-NEXT:    movb %bl, %cl
-; X86-FAST-NEXT:    shrl %cl, %edi
-; X86-FAST-NEXT:    shrdl %cl, %ebp, %esi
-; X86-FAST-NEXT:    testb $32, %bl
-; X86-FAST-NEXT:    je .LBB5_2
-; X86-FAST-NEXT:  # %bb.1:
-; X86-FAST-NEXT:    movl %edi, %esi
-; X86-FAST-NEXT:    xorl %edi, %edi
-; X86-FAST-NEXT:  .LBB5_2:
-; X86-FAST-NEXT:    movb %ch, %cl
-; X86-FAST-NEXT:    shll %cl, %eax
-; X86-FAST-NEXT:    testb $32, %ch
-; X86-FAST-NEXT:    je .LBB5_4
-; X86-FAST-NEXT:  # %bb.3:
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-FAST-NEXT:    testb $32, %cl
+; X86-FAST-NEXT:    je .LBB5_1
+; X86-FAST-NEXT:  # %bb.2:
+; X86-FAST-NEXT:    movl %esi, %edx
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-FAST-NEXT:    jmp .LBB5_3
+; X86-FAST-NEXT:  .LBB5_1:
 ; X86-FAST-NEXT:    movl %eax, %edx
-; X86-FAST-NEXT:    xorl %eax, %eax
-; X86-FAST-NEXT:  .LBB5_4:
-; X86-FAST-NEXT:    orl %edi, %edx
-; X86-FAST-NEXT:    orl %esi, %eax
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-FAST-NEXT:  .LBB5_3:
+; X86-FAST-NEXT:    shrdl %cl, %edx, %eax
+; X86-FAST-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-FAST-NEXT:    shrdl %cl, %esi, %edx
 ; X86-FAST-NEXT:    popl %esi
-; X86-FAST-NEXT:    popl %edi
-; X86-FAST-NEXT:    popl %ebx
-; X86-FAST-NEXT:    popl %ebp
 ; X86-FAST-NEXT:    retl
 ;
 ; X86-SLOW-LABEL: var_shift_i64:
 ; X86-SLOW:       # %bb.0:
-; X86-SLOW-NEXT:    pushl %ebp
 ; X86-SLOW-NEXT:    pushl %ebx
 ; X86-SLOW-NEXT:    pushl %edi
 ; X86-SLOW-NEXT:    pushl %esi
-; X86-SLOW-NEXT:    pushl %eax
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-SLOW-NEXT:    movb {{[0-9]+}}(%esp), %bl
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-SLOW-NEXT:    movl %eax, %edi
-; X86-SLOW-NEXT:    andl $2147483647, %edi # imm = 0x7FFFFFFF
-; X86-SLOW-NEXT:    movl %ebx, %ecx
-; X86-SLOW-NEXT:    shrl %cl, %edi
-; X86-SLOW-NEXT:    movl %eax, %ecx
-; X86-SLOW-NEXT:    shrl $31, %ecx
-; X86-SLOW-NEXT:    leal (%ecx,%edx,2), %edx
-; X86-SLOW-NEXT:    movb %bl, %ch
-; X86-SLOW-NEXT:    notb %ch
-; X86-SLOW-NEXT:    movb %ch, %cl
-; X86-SLOW-NEXT:    shll %cl, %edx
-; X86-SLOW-NEXT:    movb %bl, %cl
-; X86-SLOW-NEXT:    shrl %cl, %ebp
-; X86-SLOW-NEXT:    movl %ebp, (%esp) # 4-byte Spill
-; X86-SLOW-NEXT:    leal (%esi,%esi), %ebp
-; X86-SLOW-NEXT:    movb %ch, %cl
-; X86-SLOW-NEXT:    shll %cl, %ebp
-; X86-SLOW-NEXT:    movb %bl, %cl
-; X86-SLOW-NEXT:    shrl %cl, %esi
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebx
 ; X86-SLOW-NEXT:    testb $32, %bl
-; X86-SLOW-NEXT:    jne .LBB5_1
+; X86-SLOW-NEXT:    je .LBB5_1
 ; X86-SLOW-NEXT:  # %bb.2:
-; X86-SLOW-NEXT:    orl (%esp), %ebp # 4-byte Folded Reload
+; X86-SLOW-NEXT:    movl %edx, %esi
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-SLOW-NEXT:    jmp .LBB5_3
 ; X86-SLOW-NEXT:  .LBB5_1:
-; X86-SLOW-NEXT:    movl %esi, %ebp
-; X86-SLOW-NEXT:    xorl %esi, %esi
+; X86-SLOW-NEXT:    movl %eax, %esi
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-SLOW-NEXT:  .LBB5_3:
-; X86-SLOW-NEXT:    addl %eax, %eax
+; X86-SLOW-NEXT:    leal (%esi,%esi), %edi
+; X86-SLOW-NEXT:    movb %bl, %ch
+; X86-SLOW-NEXT:    notb %ch
 ; X86-SLOW-NEXT:    movb %ch, %cl
-; X86-SLOW-NEXT:    shll %cl, %eax
-; X86-SLOW-NEXT:    testb $32, %ch
-; X86-SLOW-NEXT:    jne .LBB5_4
-; X86-SLOW-NEXT:  # %bb.5:
-; X86-SLOW-NEXT:    orl %edi, %edx
-; X86-SLOW-NEXT:    jmp .LBB5_6
-; X86-SLOW-NEXT:  .LBB5_4:
-; X86-SLOW-NEXT:    movl %eax, %edx
-; X86-SLOW-NEXT:    xorl %eax, %eax
-; X86-SLOW-NEXT:  .LBB5_6:
+; X86-SLOW-NEXT:    shll %cl, %edi
+; X86-SLOW-NEXT:    movb %bl, %cl
+; X86-SLOW-NEXT:    shrl %cl, %eax
+; X86-SLOW-NEXT:    orl %edi, %eax
+; X86-SLOW-NEXT:    shrl %cl, %esi
+; X86-SLOW-NEXT:    addl %edx, %edx
+; X86-SLOW-NEXT:    movb %ch, %cl
+; X86-SLOW-NEXT:    shll %cl, %edx
 ; X86-SLOW-NEXT:    orl %esi, %edx
-; X86-SLOW-NEXT:    orl %ebp, %eax
-; X86-SLOW-NEXT:    addl $4, %esp
 ; X86-SLOW-NEXT:    popl %esi
 ; X86-SLOW-NEXT:    popl %edi
 ; X86-SLOW-NEXT:    popl %ebx
-; X86-SLOW-NEXT:    popl %ebp
 ; X86-SLOW-NEXT:    retl
 ;
 ; X64-FAST-LABEL: var_shift_i64:
 ; X64-FAST:       # %bb.0:
 ; X64-FAST-NEXT:    movq %rdx, %rcx
 ; X64-FAST-NEXT:    movq %rsi, %rax
 ; X64-FAST-NEXT:    # kill: def $cl killed $cl killed $rcx
 ; X64-FAST-NEXT:    shrdq %cl, %rdi, %rax

 
 define i128 @var_shift_i128(i128 %x, i128 %y, i128 %z) nounwind {
 ; X86-FAST-LABEL: var_shift_i128:
 ; X86-FAST:       # %bb.0:
 ; X86-FAST-NEXT:    pushl %ebp
 ; X86-FAST-NEXT:    pushl %ebx
 ; X86-FAST-NEXT:    pushl %edi
 ; X86-FAST-NEXT:    pushl %esi
-; X86-FAST-NEXT:    subl $76, %esp
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebp
+; X86-FAST-NEXT:    pushl %eax
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X86-FAST-NEXT:    movl %ebx, %ecx
-; X86-FAST-NEXT:    andl $127, %ecx
-; X86-FAST-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movb %cl, %ch
-; X86-FAST-NEXT:    movb $64, %cl
-; X86-FAST-NEXT:    subb %ch, %cl
-; X86-FAST-NEXT:    shll %cl, %edi
-; X86-FAST-NEXT:    movb %cl, (%esp) # 1-byte Spill
-; X86-FAST-NEXT:    testb $32, %cl
-; X86-FAST-NEXT:    movl $0, %esi
-; X86-FAST-NEXT:    jne .LBB6_2
-; X86-FAST-NEXT:  # %bb.1:
-; X86-FAST-NEXT:    movl %edi, %esi
-; X86-FAST-NEXT:  .LBB6_2:
-; X86-FAST-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X86-FAST-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-FAST-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl %eax, %edi
-; X86-FAST-NEXT:    movl %ebp, %eax
-; X86-FAST-NEXT:    shldl $1, %ebp, %edi
-; X86-FAST-NEXT:    addl %ebp, %eax
-; X86-FAST-NEXT:    notl %ebx
-; X86-FAST-NEXT:    andl $127, %ebx
-; X86-FAST-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movb %bl, %cl
-; X86-FAST-NEXT:    shldl %cl, %eax, %edi
-; X86-FAST-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    shll %cl, %eax
-; X86-FAST-NEXT:    testb $32, %bl
-; X86-FAST-NEXT:    movl %eax, %esi
-; X86-FAST-NEXT:    jne .LBB6_4
-; X86-FAST-NEXT:  # %bb.3:
-; X86-FAST-NEXT:    movl %edi, %esi
-; X86-FAST-NEXT:  .LBB6_4:
-; X86-FAST-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-FAST-NEXT:    movb %ch, %cl
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X86-FAST-NEXT:    shrdl %cl, %edi, %esi
-; X86-FAST-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl %edi, %esi
-; X86-FAST-NEXT:    shrl %cl, %esi
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X86-FAST-NEXT:    shrl %cl, %edi
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebp
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-FAST-NEXT:    testb $64, %cl
+; X86-FAST-NEXT:    je .LBB6_1
+; X86-FAST-NEXT:  # %bb.2:
+; X86-FAST-NEXT:    movl %edi, %ebp
+; X86-FAST-NEXT:    movl %ebx, %edi
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X86-FAST-NEXT:    movl %esi, (%esp) # 4-byte Spill
+; X86-FAST-NEXT:    movl %edx, %esi
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-FAST-NEXT:    shrdl %cl, %edx, %ebp
-; X86-FAST-NEXT:    testb $32, %ch
-; X86-FAST-NEXT:    jne .LBB6_5
-; X86-FAST-NEXT:  # %bb.6:
-; X86-FAST-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    jmp .LBB6_7
-; X86-FAST-NEXT:  .LBB6_5:
-; X86-FAST-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    xorl %edi, %edi
-; X86-FAST-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-FAST-NEXT:  .LBB6_7:
-; X86-FAST-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    testb $32, %bl
-; X86-FAST-NEXT:    movl $0, %esi
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X86-FAST-NEXT:    jne .LBB6_9
-; X86-FAST-NEXT:  # %bb.8:
-; X86-FAST-NEXT:    movl %eax, %esi
-; X86-FAST-NEXT:  .LBB6_9:
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X86-FAST-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-FAST-NEXT:    jb .LBB6_11
-; X86-FAST-NEXT:  # %bb.10:
-; X86-FAST-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-FAST-NEXT:  .LBB6_11:
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    shrdl $31, %edi, %eax
-; X86-FAST-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movb %bl, %cl
-; X86-FAST-NEXT:    shll %cl, %eax
-; X86-FAST-NEXT:    testb $32, %bl
-; X86-FAST-NEXT:    movl $0, %edi
-; X86-FAST-NEXT:    jne .LBB6_13
-; X86-FAST-NEXT:  # %bb.12:
-; X86-FAST-NEXT:    movl %eax, %edi
-; X86-FAST-NEXT:  .LBB6_13:
-; X86-FAST-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    movb (%esp), %cl # 1-byte Reload
-; X86-FAST-NEXT:    shldl %cl, %ebp, %eax
 ; X86-FAST-NEXT:    testb $32, %cl
-; X86-FAST-NEXT:    jne .LBB6_15
-; X86-FAST-NEXT:  # %bb.14:
-; X86-FAST-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:  .LBB6_15:
-; X86-FAST-NEXT:    movb %bl, %dh
-; X86-FAST-NEXT:    addb $-64, %dh
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X86-FAST-NEXT:    movb %dh, %cl
-; X86-FAST-NEXT:    shll %cl, %eax
-; X86-FAST-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X86-FAST-NEXT:    testb $32, %dh
-; X86-FAST-NEXT:    movl $0, %eax
-; X86-FAST-NEXT:    jne .LBB6_17
-; X86-FAST-NEXT:  # %bb.16:
-; X86-FAST-NEXT:    movl (%esp), %eax # 4-byte Reload
-; X86-FAST-NEXT:  .LBB6_17:
-; X86-FAST-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-FAST-NEXT:    jb .LBB6_19
-; X86-FAST-NEXT:  # %bb.18:
-; X86-FAST-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-FAST-NEXT:  .LBB6_19:
-; X86-FAST-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    cmpl $64, %ebx
-; X86-FAST-NEXT:    jb .LBB6_21
-; X86-FAST-NEXT:  # %bb.20:
-; X86-FAST-NEXT:    xorl %esi, %esi
-; X86-FAST-NEXT:  .LBB6_21:
-; X86-FAST-NEXT:    addb $-64, %ch
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    movb %ch, %cl
-; X86-FAST-NEXT:    shrl %cl, %eax
-; X86-FAST-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    testb $32, %ch
-; X86-FAST-NEXT:    movl $0, %eax
-; X86-FAST-NEXT:    jne .LBB6_23
-; X86-FAST-NEXT:  # %bb.22:
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X86-FAST-NEXT:  .LBB6_23:
-; X86-FAST-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-FAST-NEXT:    jae .LBB6_25
-; X86-FAST-NEXT:  # %bb.24:
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X86-FAST-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X86-FAST-NEXT:  .LBB6_25:
-; X86-FAST-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X86-FAST-NEXT:    movb %ch, %cl
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    shrdl %cl, %eax, %ebp
-; X86-FAST-NEXT:    testb $32, %ch
-; X86-FAST-NEXT:    jne .LBB6_27
-; X86-FAST-NEXT:  # %bb.26:
-; X86-FAST-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:  .LBB6_27:
-; X86-FAST-NEXT:    cmpl $64, %ebx
-; X86-FAST-NEXT:    jb .LBB6_29
-; X86-FAST-NEXT:  # %bb.28:
-; X86-FAST-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-FAST-NEXT:  .LBB6_29:
-; X86-FAST-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
+; X86-FAST-NEXT:    je .LBB6_4
+; X86-FAST-NEXT:    jmp .LBB6_5
+; X86-FAST-NEXT:  .LBB6_1:
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebp
+; X86-FAST-NEXT:    movl %ebp, (%esp) # 4-byte Spill
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X86-FAST-NEXT:    jae .LBB6_31
-; X86-FAST-NEXT:  # %bb.30:
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X86-FAST-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X86-FAST-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:  .LBB6_31:
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    shldl $1, %eax, %ebp
-; X86-FAST-NEXT:    movl %ebp, %eax
-; X86-FAST-NEXT:    movl %ebx, %ecx
-; X86-FAST-NEXT:    shldl %cl, %edi, %eax
-; X86-FAST-NEXT:    testb $32, %bl
-; X86-FAST-NEXT:    jne .LBB6_33
-; X86-FAST-NEXT:  # %bb.32:
-; X86-FAST-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:  .LBB6_33:
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X86-FAST-NEXT:    movb %dh, %cl
-; X86-FAST-NEXT:    shldl %cl, %esi, %eax
-; X86-FAST-NEXT:    testb $32, %dh
-; X86-FAST-NEXT:    jne .LBB6_35
-; X86-FAST-NEXT:  # %bb.34:
-; X86-FAST-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X86-FAST-NEXT:  .LBB6_35:
-; X86-FAST-NEXT:    movb $64, %cl
-; X86-FAST-NEXT:    subb %bl, %cl
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X86-FAST-NEXT:    shrdl %cl, %eax, %esi
-; X86-FAST-NEXT:    shrl %cl, %eax
 ; X86-FAST-NEXT:    testb $32, %cl
-; X86-FAST-NEXT:    je .LBB6_37
-; X86-FAST-NEXT:  # %bb.36:
-; X86-FAST-NEXT:    movl %eax, %esi
-; X86-FAST-NEXT:    xorl %eax, %eax
-; X86-FAST-NEXT:  .LBB6_37:
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X86-FAST-NEXT:    cmpl $64, %ebx
-; X86-FAST-NEXT:    jae .LBB6_39
-; X86-FAST-NEXT:  # %bb.38:
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X86-FAST-NEXT:    orl %eax, %ecx
-; X86-FAST-NEXT:    movl %ecx, (%esp) # 4-byte Spill
-; X86-FAST-NEXT:  .LBB6_39:
-; X86-FAST-NEXT:    cmpl $64, %ebx
-; X86-FAST-NEXT:    jae .LBB6_41
-; X86-FAST-NEXT:  # %bb.40:
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X86-FAST-NEXT:    orl %esi, %eax
-; X86-FAST-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:  .LBB6_41:
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-FAST-NEXT:    testl %ebx, %ebx
-; X86-FAST-NEXT:    je .LBB6_43
-; X86-FAST-NEXT:  # %bb.42:
+; X86-FAST-NEXT:    jne .LBB6_5
+; X86-FAST-NEXT:  .LBB6_4:
+; X86-FAST-NEXT:    movl %edx, %ebx
+; X86-FAST-NEXT:    movl %edi, %edx
+; X86-FAST-NEXT:    movl %esi, %edi
+; X86-FAST-NEXT:    movl %ebp, %esi
 ; X86-FAST-NEXT:    movl (%esp), %ebp # 4-byte Reload
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X86-FAST-NEXT:  .LBB6_43:
-; X86-FAST-NEXT:    orl %edx, %ebp
-; X86-FAST-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X86-FAST-NEXT:    cmpl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-FAST-NEXT:    je .LBB6_45
-; X86-FAST-NEXT:  # %bb.44:
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X86-FAST-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X86-FAST-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-FAST-NEXT:  .LBB6_45:
-; X86-FAST-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X86-FAST-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X86-FAST-NEXT:    movl %ecx, 4(%eax)
-; X86-FAST-NEXT:    movl %esi, (%eax)
-; X86-FAST-NEXT:    movl %ebp, 12(%eax)
+; X86-FAST-NEXT:  .LBB6_5:
+; X86-FAST-NEXT:    shrdl %cl, %esi, %ebp
+; X86-FAST-NEXT:    shrdl %cl, %edi, %esi
+; X86-FAST-NEXT:    shrdl %cl, %edx, %edi
+; X86-FAST-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-FAST-NEXT:    shrdl %cl, %ebx, %edx
+; X86-FAST-NEXT:    movl %edx, 12(%eax)
 ; X86-FAST-NEXT:    movl %edi, 8(%eax)
-; X86-FAST-NEXT:    addl $76, %esp
+; X86-FAST-NEXT:    movl %esi, 4(%eax)
+; X86-FAST-NEXT:    movl %ebp, (%eax)
+; X86-FAST-NEXT:    addl $4, %esp
 ; X86-FAST-NEXT:    popl %esi
 ; X86-FAST-NEXT:    popl %edi
 ; X86-FAST-NEXT:    popl %ebx
 ; X86-FAST-NEXT:    popl %ebp
 ; X86-FAST-NEXT:    retl $4
 ;
 ; X86-SLOW-LABEL: var_shift_i128:
 ; X86-SLOW:       # %bb.0:
 ; X86-SLOW-NEXT:    pushl %ebp
 ; X86-SLOW-NEXT:    pushl %ebx
 ; X86-SLOW-NEXT:    pushl %edi
 ; X86-SLOW-NEXT:    pushl %esi
-; X86-SLOW-NEXT:    subl $72, %esp
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-SLOW-NEXT:    subl $8, %esp
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SLOW-NEXT:    andl $127, %eax
-; X86-SLOW-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl %eax, %edx
-; X86-SLOW-NEXT:    movl %eax, %ecx
-; X86-SLOW-NEXT:    shrl %cl, %esi
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    leal (%edi,%edi), %ebp
-; X86-SLOW-NEXT:    notb %al
-; X86-SLOW-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl %eax, %ecx
-; X86-SLOW-NEXT:    shll %cl, %ebp
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X86-SLOW-NEXT:    movl %edx, %ecx
-; X86-SLOW-NEXT:    shrl %cl, %edi
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-SLOW-NEXT:    leal (%esi,%esi), %ebx
-; X86-SLOW-NEXT:    movl %eax, %ecx
-; X86-SLOW-NEXT:    shll %cl, %ebx
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SLOW-NEXT:    movl %edx, %ecx
-; X86-SLOW-NEXT:    shrl %cl, %eax
-; X86-SLOW-NEXT:    shrl %cl, %esi
-; X86-SLOW-NEXT:    testb $32, %dl
-; X86-SLOW-NEXT:    jne .LBB6_1
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebp
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-SLOW-NEXT:    testb $64, %cl
+; X86-SLOW-NEXT:    je .LBB6_1
 ; X86-SLOW-NEXT:  # %bb.2:
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X86-SLOW-NEXT:    orl %edi, %ebx
-; X86-SLOW-NEXT:    movl %ebx, %esi
-; X86-SLOW-NEXT:    jmp .LBB6_3
-; X86-SLOW-NEXT:  .LBB6_1:
-; X86-SLOW-NEXT:    movl %eax, %ebp
-; X86-SLOW-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-SLOW-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-SLOW-NEXT:  .LBB6_3:
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X86-SLOW-NEXT:    movl %ebx, %edx
+; X86-SLOW-NEXT:    movl %edi, %ebx
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-SLOW-NEXT:    movl %ebp, %eax
+; X86-SLOW-NEXT:    movl %esi, %ebp
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-SLOW-NEXT:    testb $32, %cl
+; X86-SLOW-NEXT:    jne .LBB6_5
+; X86-SLOW-NEXT:  .LBB6_4:
+; X86-SLOW-NEXT:    movl %esi, %edi
+; X86-SLOW-NEXT:    movl %ebx, (%esp) # 4-byte Spill
+; X86-SLOW-NEXT:    movl %ebp, %esi
+; X86-SLOW-NEXT:    movl %edx, %ebp
+; X86-SLOW-NEXT:    movl %eax, %edx
+; X86-SLOW-NEXT:    jmp .LBB6_6
+; X86-SLOW-NEXT:  .LBB6_1:
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SLOW-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-SLOW-NEXT:    jb .LBB6_5
-; X86-SLOW-NEXT:  # %bb.4:
-; X86-SLOW-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-SLOW-NEXT:    testb $32, %cl
+; X86-SLOW-NEXT:    je .LBB6_4
 ; X86-SLOW-NEXT:  .LBB6_5:
-; X86-SLOW-NEXT:    leal (%ecx,%ecx), %esi
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X86-SLOW-NEXT:    notl %ebx
-; X86-SLOW-NEXT:    andl $127, %ebx
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X86-SLOW-NEXT:    movl %esi, (%esp) # 4-byte Spill
+; X86-SLOW-NEXT:    movl %ebx, %esi
+; X86-SLOW-NEXT:  .LBB6_6:
+; X86-SLOW-NEXT:    shrl %cl, %edx
+; X86-SLOW-NEXT:    movl %ecx, %ebx
+; X86-SLOW-NEXT:    notb %bl
+; X86-SLOW-NEXT:    leal (%ebp,%ebp), %eax
 ; X86-SLOW-NEXT:    movl %ebx, %ecx
-; X86-SLOW-NEXT:    shll %cl, %esi
-; X86-SLOW-NEXT:    movl %eax, %ecx
-; X86-SLOW-NEXT:    shrl $31, %ecx
-; X86-SLOW-NEXT:    leal (%ecx,%edi,2), %ecx
-; X86-SLOW-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl %ecx, %edi
+; X86-SLOW-NEXT:    shll %cl, %eax
+; X86-SLOW-NEXT:    orl %edx, %eax
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-SLOW-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-SLOW-NEXT:    shrl %cl, %ebp
+; X86-SLOW-NEXT:    leal (%esi,%esi), %edx
 ; X86-SLOW-NEXT:    movl %ebx, %ecx
-; X86-SLOW-NEXT:    shll %cl, %edi
-; X86-SLOW-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    testb $32, %bl
-; X86-SLOW-NEXT:    movl $0, %edi
-; X86-SLOW-NEXT:    movl $0, %ecx
-; X86-SLOW-NEXT:    jne .LBB6_7
-; X86-SLOW-NEXT:  # %bb.6:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X86-SLOW-NEXT:    movl %esi, %ecx
-; X86-SLOW-NEXT:  .LBB6_7:
-; X86-SLOW-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X86-SLOW-NEXT:    movl %edi, %ecx
-; X86-SLOW-NEXT:    shrl $31, %ecx
-; X86-SLOW-NEXT:    leal (%ecx,%eax,2), %esi
+; X86-SLOW-NEXT:    shll %cl, %edx
+; X86-SLOW-NEXT:    orl %ebp, %edx
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-SLOW-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-SLOW-NEXT:    shrl %cl, %esi
 ; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X86-SLOW-NEXT:    movl (%esp), %esi # 4-byte Reload
+; X86-SLOW-NEXT:    leal (%esi,%esi), %ebp
 ; X86-SLOW-NEXT:    movl %ebx, %ecx
-; X86-SLOW-NEXT:    shll %cl, %esi
-; X86-SLOW-NEXT:    andl $2147483647, %edi # imm = 0x7FFFFFFF
-; X86-SLOW-NEXT:    movl %ebx, %ecx
-; X86-SLOW-NEXT:    notb %cl
-; X86-SLOW-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movb %cl, {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Spill
-; X86-SLOW-NEXT:    shrl %cl, %edi
-; X86-SLOW-NEXT:    testb $32, %bl
-; X86-SLOW-NEXT:    jne .LBB6_9
-; X86-SLOW-NEXT:  # %bb.8:
-; X86-SLOW-NEXT:    orl %edi, %esi
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:  .LBB6_9:
-; X86-SLOW-NEXT:    movb %bl, %dh
-; X86-SLOW-NEXT:    addb $-64, %dh
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X86-SLOW-NEXT:    movb %dh, %cl
-; X86-SLOW-NEXT:    shll %cl, %esi
-; X86-SLOW-NEXT:    testb $32, %dh
-; X86-SLOW-NEXT:    movl $0, %ecx
-; X86-SLOW-NEXT:    jne .LBB6_11
-; X86-SLOW-NEXT:  # %bb.10:
-; X86-SLOW-NEXT:    movl %esi, %ecx
-; X86-SLOW-NEXT:  .LBB6_11:
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-SLOW-NEXT:    jb .LBB6_13
-; X86-SLOW-NEXT:  # %bb.12:
-; X86-SLOW-NEXT:    xorl %ebp, %ebp
-; X86-SLOW-NEXT:  .LBB6_13:
-; X86-SLOW-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movb $64, %ch
-; X86-SLOW-NEXT:    movb $64, %ah
-; X86-SLOW-NEXT:    subb %dl, %ah
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X86-SLOW-NEXT:    movb %ah, %cl
-; X86-SLOW-NEXT:    shll %cl, %ebp
-; X86-SLOW-NEXT:    notb %cl
-; X86-SLOW-NEXT:    movl %esi, %edi
-; X86-SLOW-NEXT:    shrl %edi
-; X86-SLOW-NEXT:    shrl %cl, %edi
-; X86-SLOW-NEXT:    movb %ah, %cl
-; X86-SLOW-NEXT:    shll %cl, %esi
-; X86-SLOW-NEXT:    testb $32, %ah
-; X86-SLOW-NEXT:    jne .LBB6_14
-; X86-SLOW-NEXT:  # %bb.15:
-; X86-SLOW-NEXT:    orl %edi, %ebp
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl %ebp, %esi
-; X86-SLOW-NEXT:    jmp .LBB6_16
-; X86-SLOW-NEXT:  .LBB6_14:
-; X86-SLOW-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-SLOW-NEXT:  .LBB6_16:
-; X86-SLOW-NEXT:    addb $-64, %dl
-; X86-SLOW-NEXT:    movb %dl, %cl
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SLOW-NEXT:    shrl %cl, %eax
-; X86-SLOW-NEXT:    notb %cl
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X86-SLOW-NEXT:    shll %cl, %ebp
-; X86-SLOW-NEXT:    movb %dl, %cl
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X86-SLOW-NEXT:    shrl %cl, %edi
-; X86-SLOW-NEXT:    testb $32, %dl
-; X86-SLOW-NEXT:    jne .LBB6_17
-; X86-SLOW-NEXT:  # %bb.18:
-; X86-SLOW-NEXT:    orl %eax, %ebp
-; X86-SLOW-NEXT:    cmpl $64, %ebx
-; X86-SLOW-NEXT:    jae .LBB6_20
-; X86-SLOW-NEXT:    jmp .LBB6_21
-; X86-SLOW-NEXT:  .LBB6_17:
-; X86-SLOW-NEXT:    movl %edi, %ebp
-; X86-SLOW-NEXT:    xorl %edi, %edi
-; X86-SLOW-NEXT:    cmpl $64, %ebx
-; X86-SLOW-NEXT:    jb .LBB6_21
-; X86-SLOW-NEXT:  .LBB6_20:
-; X86-SLOW-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-SLOW-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X86-SLOW-NEXT:  .LBB6_21:
-; X86-SLOW-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-SLOW-NEXT:    jae .LBB6_23
-; X86-SLOW-NEXT:  # %bb.22:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X86-SLOW-NEXT:    orl %esi, %edi
-; X86-SLOW-NEXT:  .LBB6_23:
-; X86-SLOW-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-SLOW-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    jb .LBB6_24
-; X86-SLOW-NEXT:  # %bb.25:
-; X86-SLOW-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    jmp .LBB6_26
-; X86-SLOW-NEXT:  .LBB6_24:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X86-SLOW-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X86-SLOW-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:  .LBB6_26:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SLOW-NEXT:    shrl $31, %eax
-; X86-SLOW-NEXT:    leal (%eax,%esi,2), %esi
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movb %bl, %cl
-; X86-SLOW-NEXT:    shll %cl, %esi
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X86-SLOW-NEXT:    shrl %edi
-; X86-SLOW-NEXT:    movb {{[-0-9]+}}(%e{{[sb]}}p), %cl # 1-byte Reload
-; X86-SLOW-NEXT:    shrl %cl, %edi
-; X86-SLOW-NEXT:    testb $32, %bl
-; X86-SLOW-NEXT:    jne .LBB6_28
-; X86-SLOW-NEXT:  # %bb.27:
-; X86-SLOW-NEXT:    orl %edi, %esi
-; X86-SLOW-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:  .LBB6_28:
-; X86-SLOW-NEXT:    movl %ebp, %eax
-; X86-SLOW-NEXT:    movb %dh, %cl
-; X86-SLOW-NEXT:    shll %cl, %eax
-; X86-SLOW-NEXT:    notb %cl
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X86-SLOW-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-SLOW-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X86-SLOW-NEXT:    shrl %cl, %esi
-; X86-SLOW-NEXT:    testb $32, %dh
-; X86-SLOW-NEXT:    jne .LBB6_30
-; X86-SLOW-NEXT:  # %bb.29:
-; X86-SLOW-NEXT:    orl %esi, %eax
-; X86-SLOW-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:  .LBB6_30:
-; X86-SLOW-NEXT:    subb %bl, %ch
-; X86-SLOW-NEXT:    movl %ebp, %eax
-; X86-SLOW-NEXT:    movb %ch, %cl
-; X86-SLOW-NEXT:    shrl %cl, %eax
-; X86-SLOW-NEXT:    addl %ebp, %ebp
-; X86-SLOW-NEXT:    notb %cl
-; X86-SLOW-NEXT:    shll %cl, %ebp
-; X86-SLOW-NEXT:    movb %ch, %cl
-; X86-SLOW-NEXT:    movl %ebp, %esi
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X86-SLOW-NEXT:    shrl %cl, %ebp
-; X86-SLOW-NEXT:    testb $32, %ch
+; X86-SLOW-NEXT:    addl %edi, %edi
+; X86-SLOW-NEXT:    movl %ebx, %ecx
+; X86-SLOW-NEXT:    shll %cl, %edi
+; X86-SLOW-NEXT:    orl %esi, %edi
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X86-SLOW-NEXT:    jne .LBB6_31
-; X86-SLOW-NEXT:  # %bb.32:
-; X86-SLOW-NEXT:    orl %ebp, %esi
-; X86-SLOW-NEXT:    movl %esi, %ebp
-; X86-SLOW-NEXT:    cmpl $64, %ebx
-; X86-SLOW-NEXT:    jb .LBB6_34
-; X86-SLOW-NEXT:    jmp .LBB6_35
-; X86-SLOW-NEXT:  .LBB6_31:
-; X86-SLOW-NEXT:    movl %eax, %ebp
-; X86-SLOW-NEXT:    xorl %eax, %eax
-; X86-SLOW-NEXT:    cmpl $64, %ebx
-; X86-SLOW-NEXT:    jae .LBB6_35
-; X86-SLOW-NEXT:  .LBB6_34:
-; X86-SLOW-NEXT:    movl %ebp, %esi
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X86-SLOW-NEXT:    orl %eax, %ebp
-; X86-SLOW-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:    movl %esi, %ebp
-; X86-SLOW-NEXT:  .LBB6_35:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X86-SLOW-NEXT:    cmpl $64, %ebx
-; X86-SLOW-NEXT:    jae .LBB6_37
-; X86-SLOW-NEXT:  # %bb.36:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X86-SLOW-NEXT:    orl %ebp, %eax
-; X86-SLOW-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-SLOW-NEXT:  .LBB6_37:
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-SLOW-NEXT:    testl %ebx, %ebx
-; X86-SLOW-NEXT:    je .LBB6_39
-; X86-SLOW-NEXT:  # %bb.38:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X86-SLOW-NEXT:  .LBB6_39:
-; X86-SLOW-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X86-SLOW-NEXT:    cmpl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X86-SLOW-NEXT:    je .LBB6_41
-; X86-SLOW-NEXT:  # %bb.40:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X86-SLOW-NEXT:  .LBB6_41:
-; X86-SLOW-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X86-SLOW-NEXT:    orl %ecx, %ebx
-; X86-SLOW-NEXT:    orl %ebp, %edx
-; X86-SLOW-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X86-SLOW-NEXT:    movl %ebx, (%eax)
-; X86-SLOW-NEXT:    movl %esi, 12(%eax)
-; X86-SLOW-NEXT:    movl %edx, 4(%eax)
-; X86-SLOW-NEXT:    movl %edi, 8(%eax)
-; X86-SLOW-NEXT:    addl $72, %esp
+; X86-SLOW-NEXT:    movl %edi, 12(%ecx)
+; X86-SLOW-NEXT:    movl %ebp, 8(%ecx)
+; X86-SLOW-NEXT:    movl %edx, 4(%ecx)
+; X86-SLOW-NEXT:    movl %eax, (%ecx)
+; X86-SLOW-NEXT:    movl %ecx, %eax
+; X86-SLOW-NEXT:    addl $8, %esp
 ; X86-SLOW-NEXT:    popl %esi
 ; X86-SLOW-NEXT:    popl %edi
 ; X86-SLOW-NEXT:    popl %ebx
 ; X86-SLOW-NEXT:    popl %ebp
 ; X86-SLOW-NEXT:    retl $4
 ;
 ; X64-FAST-LABEL: var_shift_i128:
 ; X64-FAST:       # %bb.0:
-; X64-FAST-NEXT:    movq %r8, %r9
-; X64-FAST-NEXT:    movq %rcx, %r8
-; X64-FAST-NEXT:    movl %r9d, %ecx
-; X64-FAST-NEXT:    shrdq %cl, %r8, %rdx
-; X64-FAST-NEXT:    shrq %cl, %r8
-; X64-FAST-NEXT:    xorl %eax, %eax
-; X64-FAST-NEXT:    testb $64, %r9b
-; X64-FAST-NEXT:    cmovneq %r8, %rdx
-; X64-FAST-NEXT:    cmovneq %rax, %r8
-; X64-FAST-NEXT:    shldq $1, %rdi, %rsi
-; X64-FAST-NEXT:    addq %rdi, %rdi
-; X64-FAST-NEXT:    notb %r9b
-; X64-FAST-NEXT:    movl %r9d, %ecx
-; X64-FAST-NEXT:    shldq %cl, %rdi, %rsi
-; X64-FAST-NEXT:    shlq %cl, %rdi
-; X64-FAST-NEXT:    testb $64, %r9b
-; X64-FAST-NEXT:    cmovneq %rdi, %rsi
-; X64-FAST-NEXT:    cmoveq %rdi, %rax
-; X64-FAST-NEXT:    orq %rdx, %rax
-; X64-FAST-NEXT:    orq %rsi, %r8
-; X64-FAST-NEXT:    movq %r8, %rdx
+; X64-FAST-NEXT:    movq %rdx, %rax
+; X64-FAST-NEXT:    testb $64, %r8b
+; X64-FAST-NEXT:    cmoveq %rdi, %rsi
+; X64-FAST-NEXT:    cmoveq %rcx, %rdi
+; X64-FAST-NEXT:    cmovneq %rcx, %rax
+; X64-FAST-NEXT:    movl %r8d, %ecx
+; X64-FAST-NEXT:    shrdq %cl, %rdi, %rax
+; X64-FAST-NEXT:    shrdq %cl, %rsi, %rdi
+; X64-FAST-NEXT:    movq %rdi, %rdx
 ; X64-FAST-NEXT:    retq
 ;
 ; X64-SLOW-LABEL: var_shift_i128:
 ; X64-SLOW:       # %bb.0:
-; X64-SLOW-NEXT:    movq %rcx, %r9
-; X64-SLOW-NEXT:    movq %rdx, %r10
-; X64-SLOW-NEXT:    movabsq $9223372036854775807, %rax # imm = 0x7FFFFFFFFFFFFFFF
-; X64-SLOW-NEXT:    andq %rdi, %rax
-; X64-SLOW-NEXT:    movl %r8d, %ecx
-; X64-SLOW-NEXT:    shrq %cl, %rax
-; X64-SLOW-NEXT:    movq %rdi, %rcx
-; X64-SLOW-NEXT:    shrq $63, %rcx
-; X64-SLOW-NEXT:    leaq (%rcx,%rsi,2), %rdx
-; X64-SLOW-NEXT:    movl %r8d, %r11d
-; X64-SLOW-NEXT:    notb %r11b
-; X64-SLOW-NEXT:    movl %r11d, %ecx
-; X64-SLOW-NEXT:    shlq %cl, %rdx
-; X64-SLOW-NEXT:    orq %rax, %rdx
+; X64-SLOW-NEXT:    testb $64, %r8b
+; X64-SLOW-NEXT:    cmoveq %rdi, %rsi
+; X64-SLOW-NEXT:    cmoveq %rcx, %rdi
+; X64-SLOW-NEXT:    cmovneq %rcx, %rdx
 ; X64-SLOW-NEXT:    movl %r8d, %ecx
-; X64-SLOW-NEXT:    shrq %cl, %r10
-; X64-SLOW-NEXT:    leaq (%r9,%r9), %rsi
-; X64-SLOW-NEXT:    movl %r11d, %ecx
-; X64-SLOW-NEXT:    shlq %cl, %rsi
-; X64-SLOW-NEXT:    orq %r10, %rsi
+; X64-SLOW-NEXT:    shrq %cl, %rdx
+; X64-SLOW-NEXT:    leaq (%rdi,%rdi), %rax
+; X64-SLOW-NEXT:    movl %r8d, %r9d
+; X64-SLOW-NEXT:    notb %r9b
+; X64-SLOW-NEXT:    movl %r9d, %ecx
+; X64-SLOW-NEXT:    shlq %cl, %rax
+; X64-SLOW-NEXT:    orq %rdx, %rax
 ; X64-SLOW-NEXT:    movl %r8d, %ecx
-; X64-SLOW-NEXT:    shrq %cl, %r9
-; X64-SLOW-NEXT:    xorl %eax, %eax
-; X64-SLOW-NEXT:    testb $64, %r8b
-; X64-SLOW-NEXT:    cmovneq %r9, %rsi
-; X64-SLOW-NEXT:    cmovneq %rax, %r9
-; X64-SLOW-NEXT:    addq %rdi, %rdi
-; X64-SLOW-NEXT:    movl %r11d, %ecx
-; X64-SLOW-NEXT:    shlq %cl, %rdi
-; X64-SLOW-NEXT:    testb $64, %r11b
-; X64-SLOW-NEXT:    cmovneq %rdi, %rdx
-; X64-SLOW-NEXT:    cmoveq %rdi, %rax
-; X64-SLOW-NEXT:    orq %rsi, %rax
-; X64-SLOW-NEXT:    orq %r9, %rdx
+; X64-SLOW-NEXT:    shrq %cl, %rdi
+; X64-SLOW-NEXT:    leaq (%rsi,%rsi), %rdx
+; X64-SLOW-NEXT:    movl %r9d, %ecx
+; X64-SLOW-NEXT:    shlq %cl, %rdx
+; X64-SLOW-NEXT:    orq %rdi, %rdx
 ; X64-SLOW-NEXT:    retq
   %tmp = tail call i128 @llvm.fshr.i128(i128 %x, i128 %y, i128 %z)
   ret i128 %tmp
 }
 
 ;
 ; Const Funnel Shift
 ;

 ; X64-SLOW-NEXT:    retq
   %tmp = tail call i32 @llvm.fshr.i32(i32 %x, i32 %y, i32 7)
   ret i32 %tmp
 }
 
 define i64 @const_shift_i64(i64 %x, i64 %y) nounwind {
 ; X86-FAST-LABEL: const_shift_i64:
 ; X86-FAST:       # %bb.0:
+; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-FAST-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-FAST-NEXT:    shldl $25, %ecx, %edx
 ; X86-FAST-NEXT:    shrdl $7, %ecx, %eax
 ; X86-FAST-NEXT:    retl
 ;
 ; X86-SLOW-LABEL: const_shift_i64:
 ; X86-SLOW:       # %bb.0:
 ; X86-SLOW-NEXT:    pushl %esi
 ; X86-SLOW-NEXT:    movl {{[0-9]+}}(%esp), %edx

llvm/test/CodeGen/X86/funnel-shift-rot.ll

 ; X64-AVX2-NEXT:    retq
   %f = call i16 @llvm.fshr.i16(i16 %x, i16 %x, i16 %z)
   ret i16 %f
 }
 
 define i64 @rotr_i64(i64 %x, i64 %z) nounwind {
 ; X32-SSE2-LABEL: rotr_i64:
 ; X32-SSE2:       # %bb.0:
-; X32-SSE2-NEXT:    pushl %ebp
-; X32-SSE2-NEXT:    pushl %ebx
-; X32-SSE2-NEXT:    pushl %edi
 ; X32-SSE2-NEXT:    pushl %esi
-; X32-SSE2-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-SSE2-NEXT:    movl %edx, %esi
-; X32-SSE2-NEXT:    shrl %cl, %esi
-; X32-SSE2-NEXT:    movl %ebx, %edi
-; X32-SSE2-NEXT:    shrdl %cl, %edx, %edi
-; X32-SSE2-NEXT:    xorl %ebp, %ebp
-; X32-SSE2-NEXT:    testb $32, %cl
-; X32-SSE2-NEXT:    cmovnel %esi, %edi
-; X32-SSE2-NEXT:    cmovnel %ebp, %esi
-; X32-SSE2-NEXT:    negb %cl
-; X32-SSE2-NEXT:    movl %ebx, %eax
-; X32-SSE2-NEXT:    shll %cl, %eax
-; X32-SSE2-NEXT:    shldl %cl, %ebx, %edx
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-SSE2-NEXT:    testb $32, %cl
-; X32-SSE2-NEXT:    cmovnel %eax, %edx
-; X32-SSE2-NEXT:    cmovnel %ebp, %eax
-; X32-SSE2-NEXT:    orl %edi, %eax
-; X32-SSE2-NEXT:    orl %esi, %edx
+; X32-SSE2-NEXT:    movl %eax, %edx
+; X32-SSE2-NEXT:    cmovel %esi, %edx
+; X32-SSE2-NEXT:    cmovel %eax, %esi
+; X32-SSE2-NEXT:    movl %esi, %eax
+; X32-SSE2-NEXT:    shrdl %cl, %edx, %eax
+; X32-SSE2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X32-SSE2-NEXT:    shrdl %cl, %esi, %edx
 ; X32-SSE2-NEXT:    popl %esi
-; X32-SSE2-NEXT:    popl %edi
-; X32-SSE2-NEXT:    popl %ebx
-; X32-SSE2-NEXT:    popl %ebp
 ; X32-SSE2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: rotr_i64:
 ; X64-AVX2:       # %bb.0:
 ; X64-AVX2-NEXT:    movq %rsi, %rcx
 ; X64-AVX2-NEXT:    movq %rdi, %rax
 ; X64-AVX2-NEXT:    # kill: def $cl killed $cl killed $rcx
 ; X64-AVX2-NEXT:    rorq %cl, %rax

llvm/test/CodeGen/X86/funnel-shift.ll

 ; X64-AVX2-NEXT:    retq
   %f = call i32 @llvm.fshl.i32(i32 %x, i32 %y, i32 %z)
   ret i32 %f
 }
 
 define i64 @fshl_i64(i64 %x, i64 %y, i64 %z) nounwind {
 ; X32-SSE2-LABEL: fshl_i64:
 ; X32-SSE2:       # %bb.0:
-; X32-SSE2-NEXT:    pushl %ebp
-; X32-SSE2-NEXT:    pushl %ebx
 ; X32-SSE2-NEXT:    pushl %edi
 ; X32-SSE2-NEXT:    pushl %esi
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X32-SSE2-NEXT:    movb {{[0-9]+}}(%esp), %ch
-; X32-SSE2-NEXT:    movb %ch, %cl
-; X32-SSE2-NEXT:    notb %cl
-; X32-SSE2-NEXT:    shrdl $1, %ebx, %esi
-; X32-SSE2-NEXT:    shrl %ebx
-; X32-SSE2-NEXT:    shrdl %cl, %ebx, %esi
-; X32-SSE2-NEXT:    shrl %cl, %ebx
-; X32-SSE2-NEXT:    xorl %ebp, %ebp
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-SSE2-NEXT:    testb $32, %cl
-; X32-SSE2-NEXT:    cmovnel %ebx, %esi
-; X32-SSE2-NEXT:    cmovnel %ebp, %ebx
+; X32-SSE2-NEXT:    movl %edx, %edi
+; X32-SSE2-NEXT:    cmovnel %esi, %edi
+; X32-SSE2-NEXT:    cmovel {{[0-9]+}}(%esp), %edx
+; X32-SSE2-NEXT:    cmovnel {{[0-9]+}}(%esp), %esi
 ; X32-SSE2-NEXT:    movl %edi, %eax
-; X32-SSE2-NEXT:    movb %ch, %cl
-; X32-SSE2-NEXT:    shll %cl, %eax
+; X32-SSE2-NEXT:    shldl %cl, %esi, %eax
+; X32-SSE2-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X32-SSE2-NEXT:    shldl %cl, %edi, %edx
-; X32-SSE2-NEXT:    testb $32, %ch
-; X32-SSE2-NEXT:    cmovnel %eax, %edx
-; X32-SSE2-NEXT:    cmovnel %ebp, %eax
-; X32-SSE2-NEXT:    orl %esi, %eax
-; X32-SSE2-NEXT:    orl %ebx, %edx
 ; X32-SSE2-NEXT:    popl %esi
 ; X32-SSE2-NEXT:    popl %edi
-; X32-SSE2-NEXT:    popl %ebx
-; X32-SSE2-NEXT:    popl %ebp
 ; X32-SSE2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: fshl_i64:
 ; X64-AVX2:       # %bb.0:
 ; X64-AVX2-NEXT:    movq %rdx, %rcx
 ; X64-AVX2-NEXT:    movq %rdi, %rax
 ; X64-AVX2-NEXT:    # kill: def $cl killed $cl killed $rcx
 ; X64-AVX2-NEXT:    shldq %cl, %rsi, %rax
 ; X64-AVX2-NEXT:    retq
   %f = call i64 @llvm.fshl.i64(i64 %x, i64 %y, i64 %z)
   ret i64 %f
 }
 
 define i128 @fshl_i128(i128 %x, i128 %y, i128 %z) nounwind {
 ; X32-SSE2-LABEL: fshl_i128:
 ; X32-SSE2:       # %bb.0:
 ; X32-SSE2-NEXT:    pushl %ebp
 ; X32-SSE2-NEXT:    pushl %ebx
 ; X32-SSE2-NEXT:    pushl %edi
 ; X32-SSE2-NEXT:    pushl %esi
-; X32-SSE2-NEXT:    subl $64, %esp
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-SSE2-NEXT:    movl %esi, %edi
-; X32-SSE2-NEXT:    shldl $31, %ecx, %edi
-; X32-SSE2-NEXT:    notl %ebx
-; X32-SSE2-NEXT:    andl $127, %ebx
-; X32-SSE2-NEXT:    movb $64, %cl
-; X32-SSE2-NEXT:    subb %bl, %cl
-; X32-SSE2-NEXT:    shrl %esi
-; X32-SSE2-NEXT:    movl %esi, (%esp) # 4-byte Spill
-; X32-SSE2-NEXT:    shldl %cl, %edi, %esi
-; X32-SSE2-NEXT:    movl %edi, %ebp
-; X32-SSE2-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    shll %cl, %ebp
-; X32-SSE2-NEXT:    xorl %eax, %eax
-; X32-SSE2-NEXT:    testb $32, %cl
-; X32-SSE2-NEXT:    cmovnel %ebp, %esi
-; X32-SSE2-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    cmovnel %eax, %ebp
-; X32-SSE2-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-SSE2-NEXT:    andl $127, %eax
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-SSE2-NEXT:    movl %eax, %ecx
-; X32-SSE2-NEXT:    shldl %cl, %ebp, %edx
-; X32-SSE2-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    movl %ebx, %ecx
-; X32-SSE2-NEXT:    addb $-64, %cl
-; X32-SSE2-NEXT:    movl (%esp), %esi # 4-byte Reload
-; X32-SSE2-NEXT:    shrdl %cl, %esi, %edi
-; X32-SSE2-NEXT:    shrl %cl, %esi
-; X32-SSE2-NEXT:    testb $32, %cl
-; X32-SSE2-NEXT:    cmovnel %esi, %edi
-; X32-SSE2-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    movl $0, %ecx
-; X32-SSE2-NEXT:    cmovnel %ecx, %esi
-; X32-SSE2-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edi
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-SSE2-NEXT:    movl %eax, %ecx
-; X32-SSE2-NEXT:    shldl %cl, %edi, %esi
-; X32-SSE2-NEXT:    movl %edi, %edx
-; X32-SSE2-NEXT:    shll %cl, %edx
-; X32-SSE2-NEXT:    shll %cl, %ebp
-; X32-SSE2-NEXT:    testb $32, %al
-; X32-SSE2-NEXT:    movl %eax, %ecx
-; X32-SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-SSE2-NEXT:    cmovnel %ebp, %eax
-; X32-SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    cmovnel %edx, %esi
-; X32-SSE2-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    movl $0, %eax
-; X32-SSE2-NEXT:    cmovnel %eax, %ebp
-; X32-SSE2-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    cmovnel %eax, %edx
-; X32-SSE2-NEXT:    xorl %eax, %eax
-; X32-SSE2-NEXT:    cmpl $64, %ecx
-; X32-SSE2-NEXT:    cmovael %eax, %edx
-; X32-SSE2-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X32-SSE2-NEXT:    shldl $31, %eax, %ebp
-; X32-SSE2-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-SSE2-NEXT:    shrdl $1, %eax, %esi
-; X32-SSE2-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    movl %ebx, %ecx
-; X32-SSE2-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-SSE2-NEXT:    movl (%esp), %edx # 4-byte Reload
-; X32-SSE2-NEXT:    shrdl %cl, %edx, %eax
-; X32-SSE2-NEXT:    shrl %cl, %edx
-; X32-SSE2-NEXT:    movl %esi, %ebx
-; X32-SSE2-NEXT:    shrdl %cl, %ebp, %ebx
-; X32-SSE2-NEXT:    movl %ebp, %esi
-; X32-SSE2-NEXT:    shrl %cl, %esi
-; X32-SSE2-NEXT:    testb $32, %cl
-; X32-SSE2-NEXT:    cmovnel %esi, %ebx
-; X32-SSE2-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    movl %edx, %ecx
-; X32-SSE2-NEXT:    cmovnel %edx, %eax
-; X32-SSE2-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    movl $0, %eax
-; X32-SSE2-NEXT:    cmovnel %eax, %esi
-; X32-SSE2-NEXT:    cmovnel %eax, %ecx
-; X32-SSE2-NEXT:    movl %ecx, (%esp) # 4-byte Spill
-; X32-SSE2-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-SSE2-NEXT:    cmpl $64, %ebx
-; X32-SSE2-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-SSE2-NEXT:    cmovael %eax, %ecx
-; X32-SSE2-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    xorl %ebp, %ebp
-; X32-SSE2-NEXT:    movb $64, %ch
-; X32-SSE2-NEXT:    subb %bl, %ch
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-SSE2-NEXT:    movb %ch, %cl
-; X32-SSE2-NEXT:    shrl %cl, %edx
-; X32-SSE2-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    testb $32, %ch
-; X32-SSE2-NEXT:    cmovnel %ebp, %edx
-; X32-SSE2-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-SSE2-NEXT:    movb %bl, %cl
-; X32-SSE2-NEXT:    addb $-64, %cl
-; X32-SSE2-NEXT:    movl %edi, %ebp
-; X32-SSE2-NEXT:    shll %cl, %ebp
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-SSE2-NEXT:    shldl %cl, %edi, %eax
-; X32-SSE2-NEXT:    testb $32, %cl
-; X32-SSE2-NEXT:    cmovnel %ebp, %eax
-; X32-SSE2-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X32-SSE2-NEXT:    movl (%esp), %ebx # 4-byte Reload
-; X32-SSE2-NEXT:    movl $0, %edi
-; X32-SSE2-NEXT:    cmovael %edi, %ebx
-; X32-SSE2-NEXT:    movl %ebx, (%esp) # 4-byte Spill
-; X32-SSE2-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-SSE2-NEXT:    cmpl $64, %ebx
-; X32-SSE2-NEXT:    cmovbl %edx, %eax
-; X32-SSE2-NEXT:    testb $32, %cl
-; X32-SSE2-NEXT:    movl $0, %edi
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-SSE2-NEXT:    testb $64, %cl
+; X32-SSE2-NEXT:    movl %esi, %eax
+; X32-SSE2-NEXT:    cmovnel %ebx, %eax
+; X32-SSE2-NEXT:    movl %edx, %ebp
 ; X32-SSE2-NEXT:    cmovnel %edi, %ebp
-; X32-SSE2-NEXT:    cmpl $64, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X32-SSE2-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-SSE2-NEXT:    cmovael %edi, %edx
-; X32-SSE2-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-SSE2-NEXT:    movb %ch, %cl
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-SSE2-NEXT:    shrdl %cl, %edx, %edi
-; X32-SSE2-NEXT:    testb $32, %ch
-; X32-SSE2-NEXT:    cmovnel {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-SSE2-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-SSE2-NEXT:    cmpl $64, %ebx
-; X32-SSE2-NEXT:    cmovael %ebp, %edi
-; X32-SSE2-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-SSE2-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-SSE2-NEXT:    cmpl $64, %edx
-; X32-SSE2-NEXT:    cmovael {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-SSE2-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-SSE2-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-SSE2-NEXT:    cmpl $64, %edx
-; X32-SSE2-NEXT:    cmovael {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-SSE2-NEXT:    testl %edx, %edx
-; X32-SSE2-NEXT:    cmovel {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-SSE2-NEXT:    cmovel {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-SSE2-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-SSE2-NEXT:    movl %ecx, %edx
-; X32-SSE2-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-SSE2-NEXT:    testl %ebx, %ebx
-; X32-SSE2-NEXT:    cmovel {{[0-9]+}}(%esp), %eax
-; X32-SSE2-NEXT:    cmovel {{[0-9]+}}(%esp), %edi
-; X32-SSE2-NEXT:    orl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-SSE2-NEXT:    orl (%esp), %eax # 4-byte Folded Reload
+; X32-SSE2-NEXT:    cmovnel {{[0-9]+}}(%esp), %edi
+; X32-SSE2-NEXT:    cmovnel {{[0-9]+}}(%esp), %ebx
+; X32-SSE2-NEXT:    cmovel {{[0-9]+}}(%esp), %edx
+; X32-SSE2-NEXT:    cmovel {{[0-9]+}}(%esp), %esi
+; X32-SSE2-NEXT:    testb $32, %cl
+; X32-SSE2-NEXT:    cmovnel %esi, %edx
+; X32-SSE2-NEXT:    cmovnel %ebp, %esi
+; X32-SSE2-NEXT:    cmovnel %eax, %ebp
+; X32-SSE2-NEXT:    cmovel %edi, %ebx
+; X32-SSE2-NEXT:    cmovel %eax, %edi
+; X32-SSE2-NEXT:    movl %edi, %eax
+; X32-SSE2-NEXT:    shldl %cl, %ebx, %eax
+; X32-SSE2-NEXT:    movl %ebp, %ebx
+; X32-SSE2-NEXT:    shldl %cl, %edi, %ebx
+; X32-SSE2-NEXT:    movl %esi, %edi
+; X32-SSE2-NEXT:    shldl %cl, %ebp, %edi
+; X32-SSE2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X32-SSE2-NEXT:    shldl %cl, %esi, %edx
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-SSE2-NEXT:    movl %eax, 12(%ecx)
+; X32-SSE2-NEXT:    movl %edx, 12(%ecx)
 ; X32-SSE2-NEXT:    movl %edi, 8(%ecx)
-; X32-SSE2-NEXT:    movl %esi, 4(%ecx)
-; X32-SSE2-NEXT:    movl %edx, (%ecx)
+; X32-SSE2-NEXT:    movl %ebx, 4(%ecx)
+; X32-SSE2-NEXT:    movl %eax, (%ecx)
 ; X32-SSE2-NEXT:    movl %ecx, %eax
-; X32-SSE2-NEXT:    addl $64, %esp
 ; X32-SSE2-NEXT:    popl %esi
 ; X32-SSE2-NEXT:    popl %edi
 ; X32-SSE2-NEXT:    popl %ebx
 ; X32-SSE2-NEXT:    popl %ebp
 ; X32-SSE2-NEXT:    retl $4
 ;
 ; X64-AVX2-LABEL: fshl_i128:
 ; X64-AVX2:       # %bb.0:
-; X64-AVX2-NEXT:    movq %r8, %r9
-; X64-AVX2-NEXT:    movq %rcx, %r8
-; X64-AVX2-NEXT:    movl %r9d, %ecx
-; X64-AVX2-NEXT:    shldq %cl, %rdi, %rsi
-; X64-AVX2-NEXT:    shrdq $1, %r8, %rdx
-; X64-AVX2-NEXT:    shrq %r8
-; X64-AVX2-NEXT:    notb %cl
-; X64-AVX2-NEXT:    shrdq %cl, %r8, %rdx
-; X64-AVX2-NEXT:    shrq %cl, %r8
-; X64-AVX2-NEXT:    xorl %eax, %eax
-; X64-AVX2-NEXT:    testb $64, %cl
-; X64-AVX2-NEXT:    cmovneq %r8, %rdx
-; X64-AVX2-NEXT:    cmovneq %rax, %r8
-; X64-AVX2-NEXT:    movl %r9d, %ecx
-; X64-AVX2-NEXT:    shlq %cl, %rdi
-; X64-AVX2-NEXT:    testb $64, %r9b
+; X64-AVX2-NEXT:    testb $64, %r8b
 ; X64-AVX2-NEXT:    cmovneq %rdi, %rsi
-; X64-AVX2-NEXT:    cmoveq %rdi, %rax
-; X64-AVX2-NEXT:    orq %rdx, %rax
-; X64-AVX2-NEXT:    orq %rsi, %r8
-; X64-AVX2-NEXT:    movq %r8, %rdx
+; X64-AVX2-NEXT:    cmoveq %rcx, %rdx
+; X64-AVX2-NEXT:    cmovneq %rcx, %rdi
+; X64-AVX2-NEXT:    movq %rdi, %rax
+; X64-AVX2-NEXT:    movl %r8d, %ecx
+; X64-AVX2-NEXT:    shldq %cl, %rdx, %rax
+; X64-AVX2-NEXT:    shldq %cl, %rdi, %rsi
+; X64-AVX2-NEXT:    movq %rsi, %rdx
 ; X64-AVX2-NEXT:    retq
   %f = call i128 @llvm.fshl.i128(i128 %x, i128 %y, i128 %z)
   ret i128 %f
 }
 
 ; Verify that weird types are minimally supported.
 declare i37 @llvm.fshl.i37(i37, i37, i37)
 define i37 @fshl_i37(i37 %x, i37 %y, i37 %z) nounwind {
 ; X32-SSE2-LABEL: fshl_i37:
 ; X32-SSE2:       # %bb.0:
-; X32-SSE2-NEXT:    pushl %ebp
 ; X32-SSE2-NEXT:    pushl %ebx
 ; X32-SSE2-NEXT:    pushl %edi
 ; X32-SSE2-NEXT:    pushl %esi
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebx
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edi
 ; X32-SSE2-NEXT:    shldl $27, %ebx, %edi
-; X32-SSE2-NEXT:    shll $27, %ebx
-; X32-SSE2-NEXT:    shrdl $1, %edi, %ebx
-; X32-SSE2-NEXT:    shrl %edi
 ; X32-SSE2-NEXT:    pushl $0
 ; X32-SSE2-NEXT:    pushl $37
 ; X32-SSE2-NEXT:    pushl {{[0-9]+}}(%esp)
 ; X32-SSE2-NEXT:    pushl {{[0-9]+}}(%esp)
 ; X32-SSE2-NEXT:    calll __umoddi3
 ; X32-SSE2-NEXT:    addl $16, %esp
-; X32-SSE2-NEXT:    movl %eax, %edx
-; X32-SSE2-NEXT:    movl %edx, %ecx
-; X32-SSE2-NEXT:    notb %cl
-; X32-SSE2-NEXT:    shrdl %cl, %edi, %ebx
-; X32-SSE2-NEXT:    shrl %cl, %edi
-; X32-SSE2-NEXT:    xorl %eax, %eax
+; X32-SSE2-NEXT:    movl %eax, %ecx
 ; X32-SSE2-NEXT:    testb $32, %cl
-; X32-SSE2-NEXT:    cmovnel %edi, %ebx
-; X32-SSE2-NEXT:    cmovnel %eax, %edi
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-SSE2-NEXT:    movl %edx, %ecx
-; X32-SSE2-NEXT:    shll %cl, %eax
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X32-SSE2-NEXT:    shldl %cl, %ebp, %esi
-; X32-SSE2-NEXT:    testb $32, %dl
-; X32-SSE2-NEXT:    cmovnel %eax, %esi
-; X32-SSE2-NEXT:    movl $0, %ecx
-; X32-SSE2-NEXT:    cmovnel %ecx, %eax
-; X32-SSE2-NEXT:    orl %ebx, %eax
-; X32-SSE2-NEXT:    orl %edi, %esi
+; X32-SSE2-NEXT:    jne .LBB3_1
+; X32-SSE2-NEXT:  # %bb.2:
+; X32-SSE2-NEXT:    movl %edi, %ebx
+; X32-SSE2-NEXT:    movl %esi, %edi
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-SSE2-NEXT:    jmp .LBB3_3
+; X32-SSE2-NEXT:  .LBB3_1:
+; X32-SSE2-NEXT:    shll $27, %ebx
+; X32-SSE2-NEXT:  .LBB3_3:
+; X32-SSE2-NEXT:    movl %edi, %eax
+; X32-SSE2-NEXT:    shldl %cl, %ebx, %eax
+; X32-SSE2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X32-SSE2-NEXT:    shldl %cl, %edi, %esi
 ; X32-SSE2-NEXT:    movl %esi, %edx
 ; X32-SSE2-NEXT:    popl %esi
 ; X32-SSE2-NEXT:    popl %edi
 ; X32-SSE2-NEXT:    popl %ebx
-; X32-SSE2-NEXT:    popl %ebp
 ; X32-SSE2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: fshl_i37:
 ; X64-AVX2:       # %bb.0:
 ; X64-AVX2-NEXT:    movq %rdx, %rcx
 ; X64-AVX2-NEXT:    movabsq $-2492803253203993461, %rdx # imm = 0xDD67C8A60DD67C8B
 ; X64-AVX2-NEXT:    movq %rcx, %rax
 ; X64-AVX2-NEXT:    mulq %rdx
; X64-AVX2-NEXT:    retq
   ret i32 %f
 }
 
 ; Verify that weird types are minimally supported.
 declare i37 @llvm.fshr.i37(i37, i37, i37)
 define i37 @fshr_i37(i37 %x, i37 %y, i37 %z) nounwind {
 ; X32-SSE2-LABEL: fshr_i37:
 ; X32-SSE2:       # %bb.0:
-; X32-SSE2-NEXT:    pushl %ebp
 ; X32-SSE2-NEXT:    pushl %ebx
 ; X32-SSE2-NEXT:    pushl %edi
 ; X32-SSE2-NEXT:    pushl %esi
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebx
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ebx
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-SSE2-NEXT:    shldl $1, %edi, %esi
-; X32-SSE2-NEXT:    addl %edi, %edi
+; X32-SSE2-NEXT:    shldl $27, %ebx, %esi
 ; X32-SSE2-NEXT:    pushl $0
 ; X32-SSE2-NEXT:    pushl $37
 ; X32-SSE2-NEXT:    pushl {{[0-9]+}}(%esp)
 ; X32-SSE2-NEXT:    pushl {{[0-9]+}}(%esp)
 ; X32-SSE2-NEXT:    calll __umoddi3
 ; X32-SSE2-NEXT:    addl $16, %esp
-; X32-SSE2-NEXT:    movl %eax, %edx
-; X32-SSE2-NEXT:    addb $27, %dl
-; X32-SSE2-NEXT:    movl %edx, %eax
-; X32-SSE2-NEXT:    notb %al
-; X32-SSE2-NEXT:    movl %eax, %ecx
-; X32-SSE2-NEXT:    shldl %cl, %edi, %esi
-; X32-SSE2-NEXT:    shldl $27, %ebp, %ebx
-; X32-SSE2-NEXT:    shll $27, %ebp
-; X32-SSE2-NEXT:    movl %edx, %ecx
-; X32-SSE2-NEXT:    shrdl %cl, %ebx, %ebp
-; X32-SSE2-NEXT:    shrl %cl, %ebx
-; X32-SSE2-NEXT:    xorl %ecx, %ecx
-; X32-SSE2-NEXT:    testb $32, %dl
-; X32-SSE2-NEXT:    cmovnel %ebx, %ebp
-; X32-SSE2-NEXT:    cmovnel %ecx, %ebx
-; X32-SSE2-NEXT:    xorl %edx, %edx
 ; X32-SSE2-NEXT:    movl %eax, %ecx
-; X32-SSE2-NEXT:    shll %cl, %edi
-; X32-SSE2-NEXT:    testb $32, %al
-; X32-SSE2-NEXT:    cmovnel %edi, %esi
-; X32-SSE2-NEXT:    cmovnel %edx, %edi
-; X32-SSE2-NEXT:    orl %ebp, %edi
-; X32-SSE2-NEXT:    orl %ebx, %esi
-; X32-SSE2-NEXT:    movl %edi, %eax
+; X32-SSE2-NEXT:    addl $27, %ecx
+; X32-SSE2-NEXT:    testb $32, %cl
+; X32-SSE2-NEXT:    je .LBB10_1
+; X32-SSE2-NEXT:  # %bb.2:
+; X32-SSE2-NEXT:    movl %edi, %edx
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-SSE2-NEXT:    jmp .LBB10_3
+; X32-SSE2-NEXT:  .LBB10_1:
+; X32-SSE2-NEXT:    shll $27, %ebx
 ; X32-SSE2-NEXT:    movl %esi, %edx
+; X32-SSE2-NEXT:    movl %ebx, %esi
+; X32-SSE2-NEXT:  .LBB10_3:
+; X32-SSE2-NEXT:    shrdl %cl, %edx, %esi
+; X32-SSE2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X32-SSE2-NEXT:    shrdl %cl, %edi, %edx
+; X32-SSE2-NEXT:    movl %esi, %eax
 ; X32-SSE2-NEXT:    popl %esi
 ; X32-SSE2-NEXT:    popl %edi
 ; X32-SSE2-NEXT:    popl %ebx
-; X32-SSE2-NEXT:    popl %ebp
 ; X32-SSE2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: fshr_i37:
 ; X64-AVX2:       # %bb.0:
 ; X64-AVX2-NEXT:    movq %rdx, %rcx
 ; X64-AVX2-NEXT:    movabsq $-2492803253203993461, %rdx # imm = 0xDD67C8A60DD67C8B
 ; X64-AVX2-NEXT:    movq %rcx, %rax
 ; X64-AVX2-NEXT:    mulq %rdx
; X64-AVX2-NEXT:    retq
   ret i32 %f
 }
 
 ; 64-bit should also work. 105-64 = 41, but right-shift became left, so 64-41=23.
 
 define i64 @fshr_i64_const_overshift(i64 %x, i64 %y) nounwind {
 ; X32-SSE2-LABEL: fshr_i64_const_overshift:
 ; X32-SSE2:       # %bb.0:
-; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-SSE2-NEXT:    shrdl $9, %ecx, %eax
 ; X32-SSE2-NEXT:    shldl $23, %ecx, %edx
 ; X32-SSE2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: fshr_i64_const_overshift:
 ; X64-AVX2:       # %bb.0:
 ; X64-AVX2-NEXT:    movq %rdi, %rax
 ; X64-AVX2-NEXT:    shldq $23, %rsi, %rax