diff --git a/lib/Target/RISCV/RISCVISelLowering.cpp b/lib/Target/RISCV/RISCVISelLowering.cpp index 571eae6cb75..75ea3aa5d73 100644 --- a/lib/Target/RISCV/RISCVISelLowering.cpp +++ b/lib/Target/RISCV/RISCVISelLowering.cpp @@ -772,9 +772,9 @@ RISCVTargetLowering::RISCVTargetLowering(const TargetMachine &TM, // We can use any register for comparisons setHasMultipleConditionRegisters(); - if (Subtarget.hasStdExtZbp()) { - setTargetDAGCombine(ISD::OR); - } + setTargetDAGCombine(ISD::AND); + setTargetDAGCombine(ISD::OR); + setTargetDAGCombine(ISD::XOR); if (Subtarget.hasStdExtV()) { setTargetDAGCombine(ISD::FCOPYSIGN); setTargetDAGCombine(ISD::MGATHER); @@ -5213,6 +5213,99 @@ static SDValue combineGREVI_GORCI(SDNode *N, SelectionDAG &DAG) { DAG.getConstant(CombinedShAmt, DL, N->getOperand(1).getValueType())); } +// Combine a constant select operand into its use: +// +// (and (select_cc lhs, rhs, cc, -1, c), x) +// -> (select_cc lhs, rhs, cc, x, (and, x, c)) [AllOnes=1] +// (or (select_cc lhs, rhs, cc, 0, c), x) +// -> (select_cc lhs, rhs, cc, x, (or, x, c)) [AllOnes=0] +// (xor (select_cc lhs, rhs, cc, 0, c), x) +// -> (select_cc lhs, rhs, cc, x, (xor, x, c)) [AllOnes=0] +static SDValue combineSelectCCAndUse(SDNode *N, SDValue Slct, SDValue OtherOp, + SelectionDAG &DAG, bool AllOnes) { + EVT VT = N->getValueType(0); + + if (Slct.getOpcode() != RISCVISD::SELECT_CC || !Slct.hasOneUse()) + return SDValue(); + + auto isZeroOrAllOnes = [](SDValue N, bool AllOnes) { + return AllOnes ? isAllOnesConstant(N) : isNullConstant(N); + }; + + bool SwapSelectOps; + SDValue TrueVal = Slct.getOperand(3); + SDValue FalseVal = Slct.getOperand(4); + SDValue NonConstantVal; + if (isZeroOrAllOnes(TrueVal, AllOnes)) { + SwapSelectOps = false; + NonConstantVal = FalseVal; + } else if (isZeroOrAllOnes(FalseVal, AllOnes)) { + SwapSelectOps = true; + NonConstantVal = TrueVal; + } else + return SDValue(); + + // Slct is now know to be the desired identity constant when CC is true. + TrueVal = OtherOp; + FalseVal = DAG.getNode(N->getOpcode(), SDLoc(N), VT, OtherOp, NonConstantVal); + // Unless SwapSelectOps says CC should be false. + if (SwapSelectOps) + std::swap(TrueVal, FalseVal); + + return DAG.getNode(RISCVISD::SELECT_CC, SDLoc(N), VT, + {Slct.getOperand(0), Slct.getOperand(1), + Slct.getOperand(2), TrueVal, FalseVal}); +} + +// Attempt combineSelectAndUse on each operand of a commutative operator N. +static SDValue combineSelectCCAndUseCommutative(SDNode *N, SelectionDAG &DAG, + bool AllOnes) { + SDValue N0 = N->getOperand(0); + SDValue N1 = N->getOperand(1); + if (SDValue Result = combineSelectCCAndUse(N, N0, N1, DAG, AllOnes)) + return Result; + if (SDValue Result = combineSelectCCAndUse(N, N1, N0, DAG, AllOnes)) + return Result; + return SDValue(); +} + +static SDValue performANDCombine(SDNode *N, + TargetLowering::DAGCombinerInfo &DCI, + const RISCVSubtarget &Subtarget) { + SelectionDAG &DAG = DCI.DAG; + + // fold (and (select_cc lhs, rhs, cc, -1, y), x) -> + // (select lhs, rhs, cc, x, (and x, y)) + return combineSelectCCAndUseCommutative(N, DAG, true); +} + +static SDValue performORCombine(SDNode *N, TargetLowering::DAGCombinerInfo &DCI, + const RISCVSubtarget &Subtarget) { + SelectionDAG &DAG = DCI.DAG; + if (Subtarget.hasStdExtZbp()) { + if (auto GREV = combineORToGREV(SDValue(N, 0), DAG, Subtarget)) + return GREV; + if (auto GORC = combineORToGORC(SDValue(N, 0), DAG, Subtarget)) + return GORC; + if (auto SHFL = combineORToSHFL(SDValue(N, 0), DAG, Subtarget)) + return SHFL; + } + + // fold (or (select_cc lhs, rhs, cc, 0, y), x) -> + // (select lhs, rhs, cc, x, (or x, y)) + return combineSelectCCAndUseCommutative(N, DAG, false); +} + +static SDValue performXORCombine(SDNode *N, + TargetLowering::DAGCombinerInfo &DCI, + const RISCVSubtarget &Subtarget) { + SelectionDAG &DAG = DCI.DAG; + + // fold (xor (select_cc lhs, rhs, cc, 0, y), x) -> + // (select lhs, rhs, cc, x, (xor x, y)) + return combineSelectCCAndUseCommutative(N, DAG, false); +} + SDValue RISCVTargetLowering::PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const { SelectionDAG &DAG = DCI.DAG; @@ -5431,14 +5524,12 @@ SDValue RISCVTargetLowering::PerformDAGCombine(SDNode *N, return DAG.getNode(ISD::AND, DL, MVT::i64, NewFMV, DAG.getConstant(~SignBit, DL, MVT::i64)); } + case ISD::AND: + return performANDCombine(N, DCI, Subtarget); case ISD::OR: - if (auto GREV = combineORToGREV(SDValue(N, 0), DCI.DAG, Subtarget)) - return GREV; - if (auto GORC = combineORToGORC(SDValue(N, 0), DCI.DAG, Subtarget)) - return GORC; - if (auto SHFL = combineORToSHFL(SDValue(N, 0), DCI.DAG, Subtarget)) - return SHFL; - break; + return performORCombine(N, DCI, Subtarget); + case ISD::XOR: + return performXORCombine(N, DCI, Subtarget); case RISCVISD::SELECT_CC: { // Transform SDValue LHS = N->getOperand(0); diff --git a/test/CodeGen/RISCV/rv32zbb-zbp.ll b/test/CodeGen/RISCV/rv32zbb-zbp.ll index 18cca70b11b..b8856124bb9 100644 --- a/test/CodeGen/RISCV/rv32zbb-zbp.ll +++ b/test/CodeGen/RISCV/rv32zbb-zbp.ll @@ -218,48 +218,43 @@ 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: andi a3, a2, 63 -; RV32I-NEXT: addi t1, a3, -32 +; 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 t1, .LBB7_2 +; RV32I-NEXT: bltz t0, .LBB7_2 ; RV32I-NEXT: # %bb.1: -; RV32I-NEXT: sll a7, a0, t1 +; RV32I-NEXT: sll a1, a0, t0 ; RV32I-NEXT: j .LBB7_3 ; RV32I-NEXT: .LBB7_2: -; RV32I-NEXT: sll a4, a1, a2 -; RV32I-NEXT: sub a3, a6, a3 -; RV32I-NEXT: srli a5, a0, 1 -; RV32I-NEXT: srl a3, a5, a3 -; RV32I-NEXT: or a7, a4, a3 +; 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 a4, a2 -; RV32I-NEXT: andi a5, a4, 63 -; RV32I-NEXT: addi a3, a5, -32 -; RV32I-NEXT: bltz a3, .LBB7_7 +; 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: mv t0, zero -; RV32I-NEXT: bgez a3, .LBB7_8 +; RV32I-NEXT: srl a3, a7, a4 +; RV32I-NEXT: bltz t0, .LBB7_6 +; RV32I-NEXT: j .LBB7_7 ; RV32I-NEXT: .LBB7_5: -; RV32I-NEXT: srl a3, a0, a4 -; RV32I-NEXT: sub a4, a6, a5 -; RV32I-NEXT: slli a1, a1, 1 -; RV32I-NEXT: sll a1, a1, a4 -; RV32I-NEXT: or a4, a3, a1 -; RV32I-NEXT: or a1, a7, t0 -; RV32I-NEXT: bgez t1, .LBB7_9 +; 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 a0, a0, a2 -; RV32I-NEXT: or a0, a0, a4 -; RV32I-NEXT: ret +; RV32I-NEXT: or a3, a3, a0 ; RV32I-NEXT: .LBB7_7: -; RV32I-NEXT: srl t0, a1, a4 -; RV32I-NEXT: bltz a3, .LBB7_5 -; RV32I-NEXT: .LBB7_8: -; RV32I-NEXT: srl a4, a1, a3 -; RV32I-NEXT: or a1, a7, t0 -; RV32I-NEXT: bltz t1, .LBB7_6 -; RV32I-NEXT: .LBB7_9: -; RV32I-NEXT: or a0, zero, a4 +; RV32I-NEXT: mv a0, a3 ; RV32I-NEXT: ret ; ; RV32IB-LABEL: rol_i64: @@ -299,94 +294,84 @@ define i64 @rol_i64(i64 %a, i64 %b) nounwind { ; ; RV32IBB-LABEL: rol_i64: ; RV32IBB: # %bb.0: -; RV32IBB-NEXT: andi a3, a2, 63 -; RV32IBB-NEXT: addi t1, a3, -32 +; RV32IBB-NEXT: mv a7, a1 +; RV32IBB-NEXT: andi a1, a2, 63 +; RV32IBB-NEXT: addi t0, a1, -32 ; RV32IBB-NEXT: addi a6, zero, 31 -; RV32IBB-NEXT: bltz t1, .LBB7_2 +; RV32IBB-NEXT: bltz t0, .LBB7_2 ; RV32IBB-NEXT: # %bb.1: -; RV32IBB-NEXT: sll a7, a0, t1 +; RV32IBB-NEXT: sll a1, a0, t0 ; RV32IBB-NEXT: j .LBB7_3 ; RV32IBB-NEXT: .LBB7_2: -; RV32IBB-NEXT: sll a4, a1, a2 -; RV32IBB-NEXT: sub a3, a6, a3 -; RV32IBB-NEXT: srli a5, a0, 1 -; RV32IBB-NEXT: srl a3, a5, a3 -; RV32IBB-NEXT: or a7, a4, a3 +; RV32IBB-NEXT: sll a3, a7, a2 +; RV32IBB-NEXT: sub a1, a6, a1 +; RV32IBB-NEXT: srli a4, a0, 1 +; RV32IBB-NEXT: srl a1, a4, a1 +; RV32IBB-NEXT: or a1, a3, a1 ; RV32IBB-NEXT: .LBB7_3: -; RV32IBB-NEXT: neg a4, a2 -; RV32IBB-NEXT: andi a5, a4, 63 -; RV32IBB-NEXT: addi a3, a5, -32 -; RV32IBB-NEXT: bltz a3, .LBB7_7 +; RV32IBB-NEXT: neg a5, a2 +; RV32IBB-NEXT: andi a3, a5, 63 +; RV32IBB-NEXT: addi a4, a3, -32 +; RV32IBB-NEXT: bltz a4, .LBB7_5 ; RV32IBB-NEXT: # %bb.4: -; RV32IBB-NEXT: mv t0, zero -; RV32IBB-NEXT: bgez a3, .LBB7_8 +; RV32IBB-NEXT: srl a3, a7, a4 +; RV32IBB-NEXT: bltz t0, .LBB7_6 +; RV32IBB-NEXT: j .LBB7_7 ; RV32IBB-NEXT: .LBB7_5: -; RV32IBB-NEXT: srl a3, a0, a4 -; RV32IBB-NEXT: sub a4, a6, a5 -; RV32IBB-NEXT: slli a1, a1, 1 -; RV32IBB-NEXT: sll a1, a1, a4 -; RV32IBB-NEXT: or a4, a3, a1 -; RV32IBB-NEXT: or a1, a7, t0 -; RV32IBB-NEXT: bgez t1, .LBB7_9 +; RV32IBB-NEXT: srl a4, a7, a5 +; RV32IBB-NEXT: or a1, a1, a4 +; RV32IBB-NEXT: srl a4, a0, a5 +; RV32IBB-NEXT: sub a3, a6, a3 +; RV32IBB-NEXT: slli a5, a7, 1 +; RV32IBB-NEXT: sll a3, a5, a3 +; RV32IBB-NEXT: or a3, a4, a3 +; RV32IBB-NEXT: bgez t0, .LBB7_7 ; RV32IBB-NEXT: .LBB7_6: ; RV32IBB-NEXT: sll a0, a0, a2 -; RV32IBB-NEXT: or a0, a0, a4 -; RV32IBB-NEXT: ret +; RV32IBB-NEXT: or a3, a3, a0 ; RV32IBB-NEXT: .LBB7_7: -; RV32IBB-NEXT: srl t0, a1, a4 -; RV32IBB-NEXT: bltz a3, .LBB7_5 -; RV32IBB-NEXT: .LBB7_8: -; RV32IBB-NEXT: srl a4, a1, a3 -; RV32IBB-NEXT: or a1, a7, t0 -; RV32IBB-NEXT: bltz t1, .LBB7_6 -; RV32IBB-NEXT: .LBB7_9: -; RV32IBB-NEXT: or a0, zero, a4 +; RV32IBB-NEXT: mv a0, a3 ; RV32IBB-NEXT: ret ; ; RV32IBP-LABEL: rol_i64: ; RV32IBP: # %bb.0: -; RV32IBP-NEXT: andi a3, a2, 63 -; RV32IBP-NEXT: addi t1, a3, -32 +; RV32IBP-NEXT: mv a7, a1 +; RV32IBP-NEXT: andi a1, a2, 63 +; RV32IBP-NEXT: addi t0, a1, -32 ; RV32IBP-NEXT: addi a6, zero, 31 -; RV32IBP-NEXT: bltz t1, .LBB7_2 +; RV32IBP-NEXT: bltz t0, .LBB7_2 ; RV32IBP-NEXT: # %bb.1: -; RV32IBP-NEXT: sll a7, a0, t1 +; RV32IBP-NEXT: sll a1, a0, t0 ; RV32IBP-NEXT: j .LBB7_3 ; RV32IBP-NEXT: .LBB7_2: -; RV32IBP-NEXT: sll a4, a1, a2 -; RV32IBP-NEXT: sub a3, a6, a3 -; RV32IBP-NEXT: srli a5, a0, 1 -; RV32IBP-NEXT: srl a3, a5, a3 -; RV32IBP-NEXT: or a7, a4, a3 +; RV32IBP-NEXT: sll a3, a7, a2 +; RV32IBP-NEXT: sub a1, a6, a1 +; RV32IBP-NEXT: srli a4, a0, 1 +; RV32IBP-NEXT: srl a1, a4, a1 +; RV32IBP-NEXT: or a1, a3, a1 ; RV32IBP-NEXT: .LBB7_3: -; RV32IBP-NEXT: neg a4, a2 -; RV32IBP-NEXT: andi a5, a4, 63 -; RV32IBP-NEXT: addi a3, a5, -32 -; RV32IBP-NEXT: bltz a3, .LBB7_7 +; RV32IBP-NEXT: neg a5, a2 +; RV32IBP-NEXT: andi a3, a5, 63 +; RV32IBP-NEXT: addi a4, a3, -32 +; RV32IBP-NEXT: bltz a4, .LBB7_5 ; RV32IBP-NEXT: # %bb.4: -; RV32IBP-NEXT: mv t0, zero -; RV32IBP-NEXT: bgez a3, .LBB7_8 +; RV32IBP-NEXT: srl a3, a7, a4 +; RV32IBP-NEXT: bltz t0, .LBB7_6 +; RV32IBP-NEXT: j .LBB7_7 ; RV32IBP-NEXT: .LBB7_5: -; RV32IBP-NEXT: srl a3, a0, a4 -; RV32IBP-NEXT: sub a4, a6, a5 -; RV32IBP-NEXT: slli a1, a1, 1 -; RV32IBP-NEXT: sll a1, a1, a4 -; RV32IBP-NEXT: or a4, a3, a1 -; RV32IBP-NEXT: or a1, a7, t0 -; RV32IBP-NEXT: bgez t1, .LBB7_9 +; RV32IBP-NEXT: srl a4, a7, a5 +; RV32IBP-NEXT: or a1, a1, a4 +; RV32IBP-NEXT: srl a4, a0, a5 +; RV32IBP-NEXT: sub a3, a6, a3 +; RV32IBP-NEXT: slli a5, a7, 1 +; RV32IBP-NEXT: sll a3, a5, a3 +; RV32IBP-NEXT: or a3, a4, a3 +; RV32IBP-NEXT: bgez t0, .LBB7_7 ; RV32IBP-NEXT: .LBB7_6: ; RV32IBP-NEXT: sll a0, a0, a2 -; RV32IBP-NEXT: or a0, a0, a4 -; RV32IBP-NEXT: ret +; RV32IBP-NEXT: or a3, a3, a0 ; RV32IBP-NEXT: .LBB7_7: -; RV32IBP-NEXT: srl t0, a1, a4 -; RV32IBP-NEXT: bltz a3, .LBB7_5 -; RV32IBP-NEXT: .LBB7_8: -; RV32IBP-NEXT: srl a4, a1, a3 -; RV32IBP-NEXT: or a1, a7, t0 -; RV32IBP-NEXT: bltz t1, .LBB7_6 -; RV32IBP-NEXT: .LBB7_9: -; RV32IBP-NEXT: or a0, zero, a4 +; RV32IBP-NEXT: mv a0, a3 ; RV32IBP-NEXT: ret %or = tail call i64 @llvm.fshl.i64(i64 %a, i64 %a, i64 %b) ret i64 %or @@ -431,48 +416,43 @@ 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: andi a3, a2, 63 -; RV32I-NEXT: addi t1, a3, -32 +; 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 t1, .LBB9_2 +; RV32I-NEXT: bltz a7, .LBB9_2 ; RV32I-NEXT: # %bb.1: -; RV32I-NEXT: srl a7, a1, t1 +; RV32I-NEXT: srl a0, a1, a7 ; RV32I-NEXT: j .LBB9_3 ; RV32I-NEXT: .LBB9_2: -; RV32I-NEXT: srl a4, a0, a2 -; RV32I-NEXT: sub a3, a6, a3 -; RV32I-NEXT: slli a5, a1, 1 -; RV32I-NEXT: sll a3, a5, a3 -; RV32I-NEXT: or a7, a4, a3 +; 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 a4, a2 -; RV32I-NEXT: andi a5, a4, 63 -; RV32I-NEXT: addi a3, a5, -32 -; RV32I-NEXT: bltz a3, .LBB9_7 +; 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: mv t0, zero -; RV32I-NEXT: bgez a3, .LBB9_8 +; 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, a1, a4 -; RV32I-NEXT: sub a4, a6, a5 -; RV32I-NEXT: srli a0, a0, 1 -; RV32I-NEXT: srl a0, a0, a4 -; RV32I-NEXT: or a4, a3, a0 -; RV32I-NEXT: or a0, a7, t0 -; RV32I-NEXT: bgez t1, .LBB9_9 +; 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 a1, a1, a2 -; RV32I-NEXT: or a1, a1, a4 -; RV32I-NEXT: ret +; RV32I-NEXT: or a3, a3, a1 ; RV32I-NEXT: .LBB9_7: -; RV32I-NEXT: sll t0, a0, a4 -; RV32I-NEXT: bltz a3, .LBB9_5 -; RV32I-NEXT: .LBB9_8: -; RV32I-NEXT: sll a4, a0, a3 -; RV32I-NEXT: or a0, a7, t0 -; RV32I-NEXT: bltz t1, .LBB9_6 -; RV32I-NEXT: .LBB9_9: -; RV32I-NEXT: or a1, zero, a4 +; RV32I-NEXT: mv a1, a3 ; RV32I-NEXT: ret ; ; RV32IB-LABEL: ror_i64: @@ -512,94 +492,84 @@ define i64 @ror_i64(i64 %a, i64 %b) nounwind { ; ; RV32IBB-LABEL: ror_i64: ; RV32IBB: # %bb.0: -; RV32IBB-NEXT: andi a3, a2, 63 -; RV32IBB-NEXT: addi t1, a3, -32 +; RV32IBB-NEXT: mv t0, a0 +; RV32IBB-NEXT: andi a0, a2, 63 +; RV32IBB-NEXT: addi a7, a0, -32 ; RV32IBB-NEXT: addi a6, zero, 31 -; RV32IBB-NEXT: bltz t1, .LBB9_2 +; RV32IBB-NEXT: bltz a7, .LBB9_2 ; RV32IBB-NEXT: # %bb.1: -; RV32IBB-NEXT: srl a7, a1, t1 +; RV32IBB-NEXT: srl a0, a1, a7 ; RV32IBB-NEXT: j .LBB9_3 ; RV32IBB-NEXT: .LBB9_2: -; RV32IBB-NEXT: srl a4, a0, a2 -; RV32IBB-NEXT: sub a3, a6, a3 -; RV32IBB-NEXT: slli a5, a1, 1 -; RV32IBB-NEXT: sll a3, a5, a3 -; RV32IBB-NEXT: or a7, a4, a3 +; RV32IBB-NEXT: srl a3, t0, a2 +; RV32IBB-NEXT: sub a0, a6, a0 +; RV32IBB-NEXT: slli a4, a1, 1 +; RV32IBB-NEXT: sll a0, a4, a0 +; RV32IBB-NEXT: or a0, a3, a0 ; RV32IBB-NEXT: .LBB9_3: -; RV32IBB-NEXT: neg a4, a2 -; RV32IBB-NEXT: andi a5, a4, 63 -; RV32IBB-NEXT: addi a3, a5, -32 -; RV32IBB-NEXT: bltz a3, .LBB9_7 +; RV32IBB-NEXT: neg a5, a2 +; RV32IBB-NEXT: andi a4, a5, 63 +; RV32IBB-NEXT: addi a3, a4, -32 +; RV32IBB-NEXT: bltz a3, .LBB9_5 ; RV32IBB-NEXT: # %bb.4: -; RV32IBB-NEXT: mv t0, zero -; RV32IBB-NEXT: bgez a3, .LBB9_8 +; RV32IBB-NEXT: sll a3, t0, a3 +; RV32IBB-NEXT: bltz a7, .LBB9_6 +; RV32IBB-NEXT: j .LBB9_7 ; RV32IBB-NEXT: .LBB9_5: -; RV32IBB-NEXT: sll a3, a1, a4 -; RV32IBB-NEXT: sub a4, a6, a5 -; RV32IBB-NEXT: srli a0, a0, 1 -; RV32IBB-NEXT: srl a0, a0, a4 -; RV32IBB-NEXT: or a4, a3, a0 -; RV32IBB-NEXT: or a0, a7, t0 -; RV32IBB-NEXT: bgez t1, .LBB9_9 +; RV32IBB-NEXT: sll a3, t0, a5 +; RV32IBB-NEXT: or a0, a0, a3 +; RV32IBB-NEXT: sll a3, a1, a5 +; RV32IBB-NEXT: sub a4, a6, a4 +; RV32IBB-NEXT: srli a5, t0, 1 +; RV32IBB-NEXT: srl a4, a5, a4 +; RV32IBB-NEXT: or a3, a3, a4 +; RV32IBB-NEXT: bgez a7, .LBB9_7 ; RV32IBB-NEXT: .LBB9_6: ; RV32IBB-NEXT: srl a1, a1, a2 -; RV32IBB-NEXT: or a1, a1, a4 -; RV32IBB-NEXT: ret +; RV32IBB-NEXT: or a3, a3, a1 ; RV32IBB-NEXT: .LBB9_7: -; RV32IBB-NEXT: sll t0, a0, a4 -; RV32IBB-NEXT: bltz a3, .LBB9_5 -; RV32IBB-NEXT: .LBB9_8: -; RV32IBB-NEXT: sll a4, a0, a3 -; RV32IBB-NEXT: or a0, a7, t0 -; RV32IBB-NEXT: bltz t1, .LBB9_6 -; RV32IBB-NEXT: .LBB9_9: -; RV32IBB-NEXT: or a1, zero, a4 +; RV32IBB-NEXT: mv a1, a3 ; RV32IBB-NEXT: ret ; ; RV32IBP-LABEL: ror_i64: ; RV32IBP: # %bb.0: -; RV32IBP-NEXT: andi a3, a2, 63 -; RV32IBP-NEXT: addi t1, a3, -32 +; RV32IBP-NEXT: mv t0, a0 +; RV32IBP-NEXT: andi a0, a2, 63 +; RV32IBP-NEXT: addi a7, a0, -32 ; RV32IBP-NEXT: addi a6, zero, 31 -; RV32IBP-NEXT: bltz t1, .LBB9_2 +; RV32IBP-NEXT: bltz a7, .LBB9_2 ; RV32IBP-NEXT: # %bb.1: -; RV32IBP-NEXT: srl a7, a1, t1 +; RV32IBP-NEXT: srl a0, a1, a7 ; RV32IBP-NEXT: j .LBB9_3 ; RV32IBP-NEXT: .LBB9_2: -; RV32IBP-NEXT: srl a4, a0, a2 -; RV32IBP-NEXT: sub a3, a6, a3 -; RV32IBP-NEXT: slli a5, a1, 1 -; RV32IBP-NEXT: sll a3, a5, a3 -; RV32IBP-NEXT: or a7, a4, a3 +; RV32IBP-NEXT: srl a3, t0, a2 +; RV32IBP-NEXT: sub a0, a6, a0 +; RV32IBP-NEXT: slli a4, a1, 1 +; RV32IBP-NEXT: sll a0, a4, a0 +; RV32IBP-NEXT: or a0, a3, a0 ; RV32IBP-NEXT: .LBB9_3: -; RV32IBP-NEXT: neg a4, a2 -; RV32IBP-NEXT: andi a5, a4, 63 -; RV32IBP-NEXT: addi a3, a5, -32 -; RV32IBP-NEXT: bltz a3, .LBB9_7 +; RV32IBP-NEXT: neg a5, a2 +; RV32IBP-NEXT: andi a4, a5, 63 +; RV32IBP-NEXT: addi a3, a4, -32 +; RV32IBP-NEXT: bltz a3, .LBB9_5 ; RV32IBP-NEXT: # %bb.4: -; RV32IBP-NEXT: mv t0, zero -; RV32IBP-NEXT: bgez a3, .LBB9_8 +; RV32IBP-NEXT: sll a3, t0, a3 +; RV32IBP-NEXT: bltz a7, .LBB9_6 +; RV32IBP-NEXT: j .LBB9_7 ; RV32IBP-NEXT: .LBB9_5: -; RV32IBP-NEXT: sll a3, a1, a4 -; RV32IBP-NEXT: sub a4, a6, a5 -; RV32IBP-NEXT: srli a0, a0, 1 -; RV32IBP-NEXT: srl a0, a0, a4 -; RV32IBP-NEXT: or a4, a3, a0 -; RV32IBP-NEXT: or a0, a7, t0 -; RV32IBP-NEXT: bgez t1, .LBB9_9 +; RV32IBP-NEXT: sll a3, t0, a5 +; RV32IBP-NEXT: or a0, a0, a3 +; RV32IBP-NEXT: sll a3, a1, a5 +; RV32IBP-NEXT: sub a4, a6, a4 +; RV32IBP-NEXT: srli a5, t0, 1 +; RV32IBP-NEXT: srl a4, a5, a4 +; RV32IBP-NEXT: or a3, a3, a4 +; RV32IBP-NEXT: bgez a7, .LBB9_7 ; RV32IBP-NEXT: .LBB9_6: ; RV32IBP-NEXT: srl a1, a1, a2 -; RV32IBP-NEXT: or a1, a1, a4 -; RV32IBP-NEXT: ret +; RV32IBP-NEXT: or a3, a3, a1 ; RV32IBP-NEXT: .LBB9_7: -; RV32IBP-NEXT: sll t0, a0, a4 -; RV32IBP-NEXT: bltz a3, .LBB9_5 -; RV32IBP-NEXT: .LBB9_8: -; RV32IBP-NEXT: sll a4, a0, a3 -; RV32IBP-NEXT: or a0, a7, t0 -; RV32IBP-NEXT: bltz t1, .LBB9_6 -; RV32IBP-NEXT: .LBB9_9: -; RV32IBP-NEXT: or a1, zero, a4 +; RV32IBP-NEXT: mv a1, a3 ; RV32IBP-NEXT: ret %or = tail call i64 @llvm.fshr.i64(i64 %a, i64 %a, i64 %b) ret i64 %or diff --git a/test/CodeGen/RISCV/rv32zbs.ll b/test/CodeGen/RISCV/rv32zbs.ll index df832744653..049fa6a6c7e 100644 --- a/test/CodeGen/RISCV/rv32zbs.ll +++ b/test/CodeGen/RISCV/rv32zbs.ll @@ -63,17 +63,14 @@ define i64 @sbclr_i64(i64 %a, i64 %b) nounwind { ; RV32I-NEXT: addi a3, zero, 1 ; RV32I-NEXT: bltz a4, .LBB2_2 ; RV32I-NEXT: # %bb.1: -; RV32I-NEXT: mv a2, zero -; RV32I-NEXT: sll a4, a3, a4 -; RV32I-NEXT: j .LBB2_3 -; RV32I-NEXT: .LBB2_2: -; RV32I-NEXT: mv a4, zero -; RV32I-NEXT: sll a2, a3, a2 -; RV32I-NEXT: .LBB2_3: -; RV32I-NEXT: not a3, a4 +; RV32I-NEXT: sll a2, a3, a4 ; RV32I-NEXT: not a2, a2 -; RV32I-NEXT: and a0, a2, a0 -; RV32I-NEXT: and a1, a3, a1 +; RV32I-NEXT: and a1, a1, a2 +; RV32I-NEXT: ret +; RV32I-NEXT: .LBB2_2: +; RV32I-NEXT: sll a2, a3, a2 +; RV32I-NEXT: not a2, a2 +; RV32I-NEXT: and a0, a0, a2 ; RV32I-NEXT: ret ; ; RV32IB-LABEL: sbclr_i64: @@ -96,17 +93,10 @@ define i64 @sbclr_i64(i64 %a, i64 %b) nounwind { ; RV32IBS-NEXT: addi a3, a3, -32 ; RV32IBS-NEXT: bltz a3, .LBB2_2 ; RV32IBS-NEXT: # %bb.1: -; RV32IBS-NEXT: mv a2, zero -; RV32IBS-NEXT: bset a3, zero, a3 -; RV32IBS-NEXT: j .LBB2_3 +; RV32IBS-NEXT: bclr a1, a1, a3 +; RV32IBS-NEXT: ret ; RV32IBS-NEXT: .LBB2_2: -; RV32IBS-NEXT: mv a3, zero -; RV32IBS-NEXT: bset a2, zero, a2 -; RV32IBS-NEXT: .LBB2_3: -; RV32IBS-NEXT: not a3, a3 -; RV32IBS-NEXT: not a2, a2 -; RV32IBS-NEXT: and a0, a2, a0 -; RV32IBS-NEXT: and a1, a3, a1 +; RV32IBS-NEXT: bclr a0, a0, a2 ; RV32IBS-NEXT: ret %and = and i64 %b, 63 %shl = shl nuw i64 1, %and diff --git a/test/CodeGen/RISCV/rv32zbt.ll b/test/CodeGen/RISCV/rv32zbt.ll index 24b814d67a7..9dd68466e07 100644 --- a/test/CodeGen/RISCV/rv32zbt.ll +++ b/test/CodeGen/RISCV/rv32zbt.ll @@ -444,51 +444,45 @@ 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 t1, a5, -32 +; RV32I-NEXT: addi a7, a5, -32 ; RV32I-NEXT: addi a6, zero, 31 -; RV32I-NEXT: bltz t1, .LBB13_2 +; RV32I-NEXT: bltz a7, .LBB13_2 ; RV32I-NEXT: # %bb.1: -; RV32I-NEXT: sll a7, a0, t1 +; RV32I-NEXT: sll a1, a0, a7 ; RV32I-NEXT: j .LBB13_3 ; RV32I-NEXT: .LBB13_2: -; RV32I-NEXT: sll a7, a1, a4 +; 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 a7, a7, a1 +; RV32I-NEXT: or a1, t0, a1 ; RV32I-NEXT: .LBB13_3: -; RV32I-NEXT: not a1, a4 -; RV32I-NEXT: andi t3, a1, 63 -; RV32I-NEXT: addi a5, t3, -32 -; RV32I-NEXT: srli t2, a3, 1 -; RV32I-NEXT: bltz a5, .LBB13_7 +; 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: mv t0, zero -; RV32I-NEXT: bgez a5, .LBB13_8 +; 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: srli a2, a2, 1 ; RV32I-NEXT: or a2, a2, a3 -; RV32I-NEXT: srl a1, a2, a1 -; RV32I-NEXT: sub a2, a6, t3 -; RV32I-NEXT: slli a3, t2, 1 -; RV32I-NEXT: sll a2, a3, a2 -; RV32I-NEXT: or a2, a1, a2 -; RV32I-NEXT: or a1, a7, t0 -; RV32I-NEXT: bgez t1, .LBB13_9 +; 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: .LBB13_6: ; RV32I-NEXT: sll a0, a0, a4 -; RV32I-NEXT: or a0, a0, a2 -; RV32I-NEXT: ret +; RV32I-NEXT: or a2, a2, a0 ; RV32I-NEXT: .LBB13_7: -; RV32I-NEXT: srl t0, t2, a1 -; RV32I-NEXT: bltz a5, .LBB13_5 -; RV32I-NEXT: .LBB13_8: -; RV32I-NEXT: srl a2, t2, a5 -; RV32I-NEXT: or a1, a7, t0 -; RV32I-NEXT: bltz t1, .LBB13_6 -; RV32I-NEXT: .LBB13_9: -; RV32I-NEXT: or a0, zero, a2 +; RV32I-NEXT: mv a0, a2 ; RV32I-NEXT: ret ; ; RV32IB-LABEL: fshl_i64: @@ -605,54 +599,48 @@ 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: andi a5, a4, 63 -; RV32I-NEXT: addi t1, a5, -32 -; RV32I-NEXT: addi a6, zero, 31 -; RV32I-NEXT: bltz t1, .LBB15_2 +; 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: # %bb.1: -; RV32I-NEXT: srl a7, a3, t1 +; RV32I-NEXT: srl a0, a3, a6 ; RV32I-NEXT: j .LBB15_3 ; RV32I-NEXT: .LBB15_2: -; RV32I-NEXT: srl a7, a2, a4 -; RV32I-NEXT: sub a5, a6, a5 -; RV32I-NEXT: slli a2, a3, 1 -; RV32I-NEXT: sll a2, a2, a5 -; RV32I-NEXT: or a7, a7, a2 +; 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 a2, a4 -; RV32I-NEXT: andi t2, a2, 63 -; RV32I-NEXT: addi a5, t2, -32 -; RV32I-NEXT: slli t3, a0, 1 -; RV32I-NEXT: bltz a5, .LBB15_7 +; 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: mv t0, zero -; RV32I-NEXT: bgez a5, .LBB15_8 +; RV32I-NEXT: sll a1, t1, a2 +; RV32I-NEXT: bltz a6, .LBB15_6 +; RV32I-NEXT: j .LBB15_7 ; RV32I-NEXT: .LBB15_5: -; RV32I-NEXT: lui a5, 524288 -; RV32I-NEXT: addi a5, a5, -1 -; RV32I-NEXT: and t3, a0, a5 -; RV32I-NEXT: sub a5, a6, t2 -; RV32I-NEXT: srl a5, t3, a5 -; RV32I-NEXT: srli a0, a0, 31 +; 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 a0, a1, a0 -; RV32I-NEXT: sll a0, a0, a2 -; RV32I-NEXT: or a1, a0, a5 -; RV32I-NEXT: or a0, t0, a7 -; RV32I-NEXT: bgez t1, .LBB15_9 +; 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: .LBB15_6: ; RV32I-NEXT: srl a2, a3, a4 ; RV32I-NEXT: or a1, a1, a2 -; RV32I-NEXT: ret ; RV32I-NEXT: .LBB15_7: -; RV32I-NEXT: sll t0, t3, a2 -; RV32I-NEXT: bltz a5, .LBB15_5 -; RV32I-NEXT: .LBB15_8: -; RV32I-NEXT: sll a1, t3, a5 -; RV32I-NEXT: or a0, t0, a7 -; RV32I-NEXT: bltz t1, .LBB15_6 -; RV32I-NEXT: .LBB15_9: -; RV32I-NEXT: or a1, a1, zero ; RV32I-NEXT: ret ; ; RV32IB-LABEL: fshr_i64: diff --git a/test/CodeGen/RISCV/select-binop-identity.ll b/test/CodeGen/RISCV/select-binop-identity.ll index 79a2da033da..9739583b3f2 100644 --- a/test/CodeGen/RISCV/select-binop-identity.ll +++ b/test/CodeGen/RISCV/select-binop-identity.ll @@ -7,27 +7,26 @@ ; InstCombine canonicalizes (c ? x | y : x) to (x | (c ? y : 0)) similar for ; other binary operations using their identity value as the constant. -; TODO: We can reverse this for and/or/xor. Allowing us to pull the binop into +; We can reverse this for and/or/xor. Allowing us to pull the binop into ; the basic block we create when we expand select. define signext i32 @and_select_all_ones_i32(i1 zeroext %c, i32 signext %x, i32 %y) { ; RV32I-LABEL: and_select_all_ones_i32: ; RV32I: # %bb.0: -; RV32I-NEXT: bnez a0, .LBB0_2 +; RV32I-NEXT: beqz a0, .LBB0_2 ; RV32I-NEXT: # %bb.1: -; RV32I-NEXT: addi a1, zero, -1 +; RV32I-NEXT: and a2, a2, a1 ; RV32I-NEXT: .LBB0_2: -; RV32I-NEXT: and a0, a1, a2 +; RV32I-NEXT: mv a0, a2 ; RV32I-NEXT: ret ; ; RV64I-LABEL: and_select_all_ones_i32: ; RV64I: # %bb.0: -; RV64I-NEXT: bnez a0, .LBB0_2 +; RV64I-NEXT: beqz a0, .LBB0_2 ; RV64I-NEXT: # %bb.1: -; RV64I-NEXT: addi a1, zero, -1 +; RV64I-NEXT: and a2, a2, a1 ; RV64I-NEXT: .LBB0_2: -; RV64I-NEXT: and a0, a1, a2 -; RV64I-NEXT: sext.w a0, a0 +; RV64I-NEXT: sext.w a0, a2 ; RV64I-NEXT: ret %a = select i1 %c, i32 %x, i32 -1 %b = and i32 %a, %y @@ -37,25 +36,22 @@ define signext i32 @and_select_all_ones_i32(i1 zeroext %c, i32 signext %x, i32 % define i64 @and_select_all_ones_i64(i1 zeroext %c, i64 %x, i64 %y) { ; RV32I-LABEL: and_select_all_ones_i64: ; RV32I: # %bb.0: -; RV32I-NEXT: addi a6, zero, -1 -; RV32I-NEXT: addi a5, zero, -1 ; RV32I-NEXT: bnez a0, .LBB1_2 ; RV32I-NEXT: # %bb.1: -; RV32I-NEXT: mv a6, a2 -; RV32I-NEXT: mv a5, a1 +; RV32I-NEXT: and a3, a3, a1 +; RV32I-NEXT: and a4, a4, a2 ; RV32I-NEXT: .LBB1_2: -; RV32I-NEXT: and a0, a3, a5 -; RV32I-NEXT: and a1, a4, a6 +; RV32I-NEXT: mv a0, a3 +; RV32I-NEXT: mv a1, a4 ; RV32I-NEXT: ret ; ; RV64I-LABEL: and_select_all_ones_i64: ; RV64I: # %bb.0: -; RV64I-NEXT: addi a3, zero, -1 ; RV64I-NEXT: bnez a0, .LBB1_2 ; RV64I-NEXT: # %bb.1: -; RV64I-NEXT: mv a3, a1 +; RV64I-NEXT: and a2, a2, a1 ; RV64I-NEXT: .LBB1_2: -; RV64I-NEXT: and a0, a2, a3 +; RV64I-NEXT: mv a0, a2 ; RV64I-NEXT: ret %a = select i1 %c, i64 -1, i64 %x %b = and i64 %y, %a @@ -65,20 +61,20 @@ define i64 @and_select_all_ones_i64(i1 zeroext %c, i64 %x, i64 %y) { define signext i32 @or_select_all_zeros_i32(i1 zeroext %c, i32 signext %x, i32 signext %y) { ; RV32I-LABEL: or_select_all_zeros_i32: ; RV32I: # %bb.0: -; RV32I-NEXT: bnez a0, .LBB2_2 +; RV32I-NEXT: beqz a0, .LBB2_2 ; RV32I-NEXT: # %bb.1: -; RV32I-NEXT: mv a1, zero +; RV32I-NEXT: or a2, a2, a1 ; RV32I-NEXT: .LBB2_2: -; RV32I-NEXT: or a0, a2, a1 +; RV32I-NEXT: mv a0, a2 ; RV32I-NEXT: ret ; ; RV64I-LABEL: or_select_all_zeros_i32: ; RV64I: # %bb.0: -; RV64I-NEXT: bnez a0, .LBB2_2 +; RV64I-NEXT: beqz a0, .LBB2_2 ; RV64I-NEXT: # %bb.1: -; RV64I-NEXT: mv a1, zero +; RV64I-NEXT: or a2, a2, a1 ; RV64I-NEXT: .LBB2_2: -; RV64I-NEXT: or a0, a2, a1 +; RV64I-NEXT: mv a0, a2 ; RV64I-NEXT: ret %a = select i1 %c, i32 %x, i32 0 %b = or i32 %y, %a @@ -88,25 +84,22 @@ define signext i32 @or_select_all_zeros_i32(i1 zeroext %c, i32 signext %x, i32 s define i64 @or_select_all_zeros_i64(i1 zeroext %c, i64 %x, i64 %y) { ; RV32I-LABEL: or_select_all_zeros_i64: ; RV32I: # %bb.0: -; RV32I-NEXT: mv a6, zero -; RV32I-NEXT: mv a5, zero ; RV32I-NEXT: bnez a0, .LBB3_2 ; RV32I-NEXT: # %bb.1: -; RV32I-NEXT: mv a6, a2 -; RV32I-NEXT: mv a5, a1 +; RV32I-NEXT: or a3, a3, a1 +; RV32I-NEXT: or a4, a4, a2 ; RV32I-NEXT: .LBB3_2: -; RV32I-NEXT: or a0, a5, a3 -; RV32I-NEXT: or a1, a6, a4 +; RV32I-NEXT: mv a0, a3 +; RV32I-NEXT: mv a1, a4 ; RV32I-NEXT: ret ; ; RV64I-LABEL: or_select_all_zeros_i64: ; RV64I: # %bb.0: -; RV64I-NEXT: mv a3, zero ; RV64I-NEXT: bnez a0, .LBB3_2 ; RV64I-NEXT: # %bb.1: -; RV64I-NEXT: mv a3, a1 +; RV64I-NEXT: or a2, a2, a1 ; RV64I-NEXT: .LBB3_2: -; RV64I-NEXT: or a0, a3, a2 +; RV64I-NEXT: mv a0, a2 ; RV64I-NEXT: ret %a = select i1 %c, i64 0, i64 %x %b = or i64 %a, %y @@ -116,22 +109,20 @@ define i64 @or_select_all_zeros_i64(i1 zeroext %c, i64 %x, i64 %y) { define signext i32 @xor_select_all_zeros_i32(i1 zeroext %c, i32 signext %x, i32 signext %y) { ; RV32I-LABEL: xor_select_all_zeros_i32: ; RV32I: # %bb.0: -; RV32I-NEXT: mv a3, zero ; RV32I-NEXT: bnez a0, .LBB4_2 ; RV32I-NEXT: # %bb.1: -; RV32I-NEXT: mv a3, a1 +; RV32I-NEXT: xor a2, a2, a1 ; RV32I-NEXT: .LBB4_2: -; RV32I-NEXT: xor a0, a2, a3 +; RV32I-NEXT: mv a0, a2 ; RV32I-NEXT: ret ; ; RV64I-LABEL: xor_select_all_zeros_i32: ; RV64I: # %bb.0: -; RV64I-NEXT: mv a3, zero ; RV64I-NEXT: bnez a0, .LBB4_2 ; RV64I-NEXT: # %bb.1: -; RV64I-NEXT: mv a3, a1 +; RV64I-NEXT: xor a2, a2, a1 ; RV64I-NEXT: .LBB4_2: -; RV64I-NEXT: xor a0, a2, a3 +; RV64I-NEXT: mv a0, a2 ; RV64I-NEXT: ret %a = select i1 %c, i32 0, i32 %x %b = xor i32 %y, %a @@ -141,22 +132,22 @@ define signext i32 @xor_select_all_zeros_i32(i1 zeroext %c, i32 signext %x, i32 define i64 @xor_select_all_zeros_i64(i1 zeroext %c, i64 %x, i64 %y) { ; RV32I-LABEL: xor_select_all_zeros_i64: ; RV32I: # %bb.0: -; RV32I-NEXT: bnez a0, .LBB5_2 +; RV32I-NEXT: beqz a0, .LBB5_2 ; RV32I-NEXT: # %bb.1: -; RV32I-NEXT: mv a2, zero -; RV32I-NEXT: mv a1, zero +; RV32I-NEXT: xor a3, a3, a1 +; RV32I-NEXT: xor a4, a4, a2 ; RV32I-NEXT: .LBB5_2: -; RV32I-NEXT: xor a0, a1, a3 -; RV32I-NEXT: xor a1, a2, a4 +; RV32I-NEXT: mv a0, a3 +; RV32I-NEXT: mv a1, a4 ; RV32I-NEXT: ret ; ; RV64I-LABEL: xor_select_all_zeros_i64: ; RV64I: # %bb.0: -; RV64I-NEXT: bnez a0, .LBB5_2 +; RV64I-NEXT: beqz a0, .LBB5_2 ; RV64I-NEXT: # %bb.1: -; RV64I-NEXT: mv a1, zero +; RV64I-NEXT: xor a2, a2, a1 ; RV64I-NEXT: .LBB5_2: -; RV64I-NEXT: xor a0, a1, a2 +; RV64I-NEXT: mv a0, a2 ; RV64I-NEXT: ret %a = select i1 %c, i64 %x, i64 0 %b = xor i64 %a, %y