From e3a804ba21634f835a4bbfba0ca44dbb1c21ea9d Mon Sep 17 00:00:00 2001 From: Richard Sandiford Date: Fri, 6 Dec 2013 09:53:09 +0000 Subject: [PATCH] [SystemZ] Optimize selects between 0 and -1 Since z has no setcc instruction as such, the choice of setBooleanContents is a bit arbitrary. Currently it's set to ZeroOrOneBooleanContent, so we produced a branch-free form when selecting between 0 and 1, but not when selecting between 0 and -1. This patch handles the latter case too. At some point I'd like to measure whether it's better to use conditional moves for constant selects on z196, but that's future work. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@196578 91177308-0d34-0410-b5e6-96231b3b80d8 --- lib/Target/SystemZ/SystemZISelLowering.cpp | 58 +++++-- test/CodeGen/SystemZ/selectcc-01.ll | 178 ++++++++++++++++++++ test/CodeGen/SystemZ/selectcc-02.ll | 178 ++++++++++++++++++++ test/CodeGen/SystemZ/selectcc-03.ll | 187 +++++++++++++++++++++ 4 files changed, 587 insertions(+), 14 deletions(-) create mode 100644 test/CodeGen/SystemZ/selectcc-01.ll create mode 100644 test/CodeGen/SystemZ/selectcc-02.ll create mode 100644 test/CodeGen/SystemZ/selectcc-03.ll diff --git a/lib/Target/SystemZ/SystemZISelLowering.cpp b/lib/Target/SystemZ/SystemZISelLowering.cpp index f6e18530f4a..0340a876ee6 100644 --- a/lib/Target/SystemZ/SystemZISelLowering.cpp +++ b/lib/Target/SystemZ/SystemZISelLowering.cpp @@ -1486,16 +1486,11 @@ static void lowerGR128Binary(SelectionDAG &DAG, SDLoc DL, EVT VT, Odd = DAG.getTargetExtractSubreg(SystemZ::odd128(Is32Bit), DL, VT, Result); } -SDValue SystemZTargetLowering::lowerSETCC(SDValue Op, - SelectionDAG &DAG) const { - SDValue CmpOp0 = Op.getOperand(0); - SDValue CmpOp1 = Op.getOperand(1); - ISD::CondCode CC = cast(Op.getOperand(2))->get(); - SDLoc DL(Op); - - unsigned CCValid, CCMask; - SDValue Glue = emitCmp(TM, DAG, DL, CmpOp0, CmpOp1, CC, CCValid, CCMask); - +// Return an i32 value that is 1 if the CC value produced by Glue is +// in the mask CCMask and 0 otherwise. CC is known to have a value +// in CCValid, so other values can be ignored. +static SDValue emitSETCC(SelectionDAG &DAG, SDLoc DL, SDValue Glue, + unsigned CCValid, unsigned CCMask) { IPMConversion Conversion = getIPMConversion(CCValid, CCMask); SDValue Result = DAG.getNode(SystemZISD::IPM, DL, MVT::i32, Glue); @@ -1516,6 +1511,18 @@ SDValue SystemZTargetLowering::lowerSETCC(SDValue Op, return Result; } +SDValue SystemZTargetLowering::lowerSETCC(SDValue Op, + SelectionDAG &DAG) const { + SDValue CmpOp0 = Op.getOperand(0); + SDValue CmpOp1 = Op.getOperand(1); + ISD::CondCode CC = cast(Op.getOperand(2))->get(); + SDLoc DL(Op); + + unsigned CCValid, CCMask; + SDValue Glue = emitCmp(TM, DAG, DL, CmpOp0, CmpOp1, CC, CCValid, CCMask); + return emitSETCC(DAG, DL, Glue, CCValid, CCMask); +} + SDValue SystemZTargetLowering::lowerBR_CC(SDValue Op, SelectionDAG &DAG) const { SDValue Chain = Op.getOperand(0); ISD::CondCode CC = cast(Op.getOperand(1))->get(); @@ -1525,10 +1532,10 @@ SDValue SystemZTargetLowering::lowerBR_CC(SDValue Op, SelectionDAG &DAG) const { SDLoc DL(Op); unsigned CCValid, CCMask; - SDValue Flags = emitCmp(TM, DAG, DL, CmpOp0, CmpOp1, CC, CCValid, CCMask); + SDValue Glue = emitCmp(TM, DAG, DL, CmpOp0, CmpOp1, CC, CCValid, CCMask); return DAG.getNode(SystemZISD::BR_CCMASK, DL, Op.getValueType(), Chain, DAG.getConstant(CCValid, MVT::i32), - DAG.getConstant(CCMask, MVT::i32), Dest, Flags); + DAG.getConstant(CCMask, MVT::i32), Dest, Glue); } SDValue SystemZTargetLowering::lowerSELECT_CC(SDValue Op, @@ -1541,14 +1548,37 @@ SDValue SystemZTargetLowering::lowerSELECT_CC(SDValue Op, SDLoc DL(Op); unsigned CCValid, CCMask; - SDValue Flags = emitCmp(TM, DAG, DL, CmpOp0, CmpOp1, CC, CCValid, CCMask); + SDValue Glue = emitCmp(TM, DAG, DL, CmpOp0, CmpOp1, CC, CCValid, CCMask); + + // Special case for handling -1/0 results. The shifts we use here + // should get optimized with the IPM conversion sequence. + ConstantSDNode *TrueC = dyn_cast(TrueOp); + ConstantSDNode *FalseC = dyn_cast(FalseOp); + if (TrueC && FalseC) { + int64_t TrueVal = TrueC->getSExtValue(); + int64_t FalseVal = FalseC->getSExtValue(); + if ((TrueVal == -1 && FalseVal == 0) || (TrueVal == 0 && FalseVal == -1)) { + // Invert the condition if we want -1 on false. + if (TrueVal == 0) + CCMask ^= CCValid; + SDValue Result = emitSETCC(DAG, DL, Glue, CCValid, CCMask); + EVT VT = Op.getValueType(); + // Extend the result to VT. Upper bits are ignored. + if (!is32Bit(VT)) + Result = DAG.getNode(ISD::ANY_EXTEND, DL, VT, Result); + // Sign-extend from the low bit. + SDValue ShAmt = DAG.getConstant(VT.getSizeInBits() - 1, MVT::i32); + SDValue Shl = DAG.getNode(ISD::SHL, DL, VT, Result, ShAmt); + return DAG.getNode(ISD::SRA, DL, VT, Shl, ShAmt); + } + } SmallVector Ops; Ops.push_back(TrueOp); Ops.push_back(FalseOp); Ops.push_back(DAG.getConstant(CCValid, MVT::i32)); Ops.push_back(DAG.getConstant(CCMask, MVT::i32)); - Ops.push_back(Flags); + Ops.push_back(Glue); SDVTList VTs = DAG.getVTList(Op.getValueType(), MVT::Glue); return DAG.getNode(SystemZISD::SELECT_CCMASK, DL, VTs, &Ops[0], Ops.size()); diff --git a/test/CodeGen/SystemZ/selectcc-01.ll b/test/CodeGen/SystemZ/selectcc-01.ll new file mode 100644 index 00000000000..a57444c831a --- /dev/null +++ b/test/CodeGen/SystemZ/selectcc-01.ll @@ -0,0 +1,178 @@ +; Test an i32 0/-1 SELECTCCC for every floating-point condition. +; +; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s + +; Test CC in { 0 } +define i32 @f1(float %a, float %b) { +; CHECK-LABEL: f1: +; CHECK: ipm %r2 +; CHECK-NEXT: afi %r2, -268435456 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp oeq float %a, %b + %res = select i1 %cond, i32 -1, i32 0 + ret i32 %res +} + +; Test CC in { 1 } +define i32 @f2(float %a, float %b) { +; CHECK-LABEL: f2: +; CHECK: ipm %r2 +; CHECK-NEXT: xilf %r2, 268435456 +; CHECK-NEXT: afi %r2, -268435456 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp olt float %a, %b + %res = select i1 %cond, i32 -1, i32 0 + ret i32 %res +} + +; Test CC in { 0, 1 } +define i32 @f3(float %a, float %b) { +; CHECK-LABEL: f3: +; CHECK: ipm %r2 +; CHECK-NEXT: afi %r2, -536870912 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp ole float %a, %b + %res = select i1 %cond, i32 -1, i32 0 + ret i32 %res +} + +; Test CC in { 2 } +define i32 @f4(float %a, float %b) { +; CHECK-LABEL: f4: +; CHECK: ipm %r2 +; CHECK-NEXT: xilf %r2, 268435456 +; CHECK-NEXT: afi %r2, 1342177280 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp ogt float %a, %b + %res = select i1 %cond, i32 -1, i32 0 + ret i32 %res +} + +; Test CC in { 0, 2 } +define i32 @f5(float %a, float %b) { +; CHECK-LABEL: f5: +; CHECK: ipm %r2 +; CHECK-NEXT: xilf %r2, 4294967295 +; CHECK-NEXT: sll %r2, 3 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp oge float %a, %b + %res = select i1 %cond, i32 -1, i32 0 + ret i32 %res +} + +; Test CC in { 1, 2 } +define i32 @f6(float %a, float %b) { +; CHECK-LABEL: f6: +; CHECK: ipm %r2 +; CHECK-NEXT: afi %r2, 268435456 +; CHECK-NEXT: sll %r2, 2 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp one float %a, %b + %res = select i1 %cond, i32 -1, i32 0 + ret i32 %res +} + +; Test CC in { 0, 1, 2 } +define i32 @f7(float %a, float %b) { +; CHECK-LABEL: f7: +; CHECK: ipm %r2 +; CHECK-NEXT: afi %r2, -805306368 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp ord float %a, %b + %res = select i1 %cond, i32 -1, i32 0 + ret i32 %res +} + +; Test CC in { 3 } +define i32 @f8(float %a, float %b) { +; CHECK-LABEL: f8: +; CHECK: ipm %r2 +; CHECK-NEXT: afi %r2, 1342177280 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp uno float %a, %b + %res = select i1 %cond, i32 -1, i32 0 + ret i32 %res +} + +; Test CC in { 0, 3 } +define i32 @f9(float %a, float %b) { +; CHECK-LABEL: f9: +; CHECK: ipm %r2 +; CHECK-NEXT: afi %r2, -268435456 +; CHECK-NEXT: sll %r2, 2 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp ueq float %a, %b + %res = select i1 %cond, i32 -1, i32 0 + ret i32 %res +} + +; Test CC in { 1, 3 } +define i32 @f10(float %a, float %b) { +; CHECK-LABEL: f10: +; CHECK: ipm %r2 +; CHECK-NEXT: sll %r2, 3 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp ult float %a, %b + %res = select i1 %cond, i32 -1, i32 0 + ret i32 %res +} + +; Test CC in { 0, 1, 3 } +define i32 @f11(float %a, float %b) { +; CHECK-LABEL: f11: +; CHECK: ipm %r2 +; CHECK-NEXT: xilf %r2, 268435456 +; CHECK-NEXT: afi %r2, -805306368 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp ule float %a, %b + %res = select i1 %cond, i32 -1, i32 0 + ret i32 %res +} + +; Test CC in { 2, 3 } +define i32 @f12(float %a, float %b) { +; CHECK-LABEL: f12: +; CHECK: ipm %r2 +; CHECK-NEXT: sll %r2, 2 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp ugt float %a, %b + %res = select i1 %cond, i32 -1, i32 0 + ret i32 %res +} + +; Test CC in { 0, 2, 3 } +define i32 @f13(float %a, float %b) { +; CHECK-LABEL: f13: +; CHECK: ipm %r2 +; CHECK-NEXT: xilf %r2, 268435456 +; CHECK-NEXT: afi %r2, 1879048192 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp uge float %a, %b + %res = select i1 %cond, i32 -1, i32 0 + ret i32 %res +} + +; Test CC in { 1, 2, 3 } +define i32 @f14(float %a, float %b) { +; CHECK-LABEL: f14: +; CHECK: ipm %r2 +; CHECK-NEXT: afi %r2, 1879048192 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp une float %a, %b + %res = select i1 %cond, i32 -1, i32 0 + ret i32 %res +} diff --git a/test/CodeGen/SystemZ/selectcc-02.ll b/test/CodeGen/SystemZ/selectcc-02.ll new file mode 100644 index 00000000000..b1081a0621d --- /dev/null +++ b/test/CodeGen/SystemZ/selectcc-02.ll @@ -0,0 +1,178 @@ +; Test an i32 0/-1 SELECTCCC for every floating-point condition. +; +; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s + +; Test CC in { 1, 2, 3 } +define i32 @f1(float %a, float %b) { +; CHECK-LABEL: f1: +; CHECK: ipm %r2 +; CHECK-NEXT: afi %r2, 1879048192 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp oeq float %a, %b + %res = select i1 %cond, i32 0, i32 -1 + ret i32 %res +} + +; Test CC in { 0, 2, 3 } +define i32 @f2(float %a, float %b) { +; CHECK-LABEL: f2: +; CHECK: ipm %r2 +; CHECK-NEXT: xilf %r2, 268435456 +; CHECK-NEXT: afi %r2, 1879048192 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp olt float %a, %b + %res = select i1 %cond, i32 0, i32 -1 + ret i32 %res +} + +; Test CC in { 2, 3 } +define i32 @f3(float %a, float %b) { +; CHECK-LABEL: f3: +; CHECK: ipm %r2 +; CHECK-NEXT: sll %r2, 2 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp ole float %a, %b + %res = select i1 %cond, i32 0, i32 -1 + ret i32 %res +} + +; Test CC in { 0, 1, 3 } +define i32 @f4(float %a, float %b) { +; CHECK-LABEL: f4: +; CHECK: ipm %r2 +; CHECK-NEXT: xilf %r2, 268435456 +; CHECK-NEXT: afi %r2, -805306368 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp ogt float %a, %b + %res = select i1 %cond, i32 0, i32 -1 + ret i32 %res +} + +; Test CC in { 1, 3 } +define i32 @f5(float %a, float %b) { +; CHECK-LABEL: f5: +; CHECK: ipm %r2 +; CHECK-NEXT: sll %r2, 3 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp oge float %a, %b + %res = select i1 %cond, i32 0, i32 -1 + ret i32 %res +} + +; Test CC in { 0, 3 } +define i32 @f6(float %a, float %b) { +; CHECK-LABEL: f6: +; CHECK: ipm %r2 +; CHECK-NEXT: afi %r2, -268435456 +; CHECK-NEXT: sll %r2, 2 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp one float %a, %b + %res = select i1 %cond, i32 0, i32 -1 + ret i32 %res +} + +; Test CC in { 3 } +define i32 @f7(float %a, float %b) { +; CHECK-LABEL: f7: +; CHECK: ipm %r2 +; CHECK-NEXT: afi %r2, 1342177280 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp ord float %a, %b + %res = select i1 %cond, i32 0, i32 -1 + ret i32 %res +} + +; Test CC in { 0, 1, 2 } +define i32 @f8(float %a, float %b) { +; CHECK-LABEL: f8: +; CHECK: ipm %r2 +; CHECK-NEXT: afi %r2, -805306368 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp uno float %a, %b + %res = select i1 %cond, i32 0, i32 -1 + ret i32 %res +} + +; Test CC in { 1, 2 } +define i32 @f9(float %a, float %b) { +; CHECK-LABEL: f9: +; CHECK: ipm %r2 +; CHECK-NEXT: afi %r2, 268435456 +; CHECK-NEXT: sll %r2, 2 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp ueq float %a, %b + %res = select i1 %cond, i32 0, i32 -1 + ret i32 %res +} + +; Test CC in { 0, 2 } +define i32 @f10(float %a, float %b) { +; CHECK-LABEL: f10: +; CHECK: ipm %r2 +; CHECK-NEXT: xilf %r2, 4294967295 +; CHECK-NEXT: sll %r2, 3 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp ult float %a, %b + %res = select i1 %cond, i32 0, i32 -1 + ret i32 %res +} + +; Test CC in { 2 } +define i32 @f11(float %a, float %b) { +; CHECK-LABEL: f11: +; CHECK: ipm %r2 +; CHECK-NEXT: xilf %r2, 268435456 +; CHECK-NEXT: afi %r2, 1342177280 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp ule float %a, %b + %res = select i1 %cond, i32 0, i32 -1 + ret i32 %res +} + +; Test CC in { 0, 1 } +define i32 @f12(float %a, float %b) { +; CHECK-LABEL: f12: +; CHECK: ipm %r2 +; CHECK-NEXT: afi %r2, -536870912 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp ugt float %a, %b + %res = select i1 %cond, i32 0, i32 -1 + ret i32 %res +} + +; Test CC in { 1 } +define i32 @f13(float %a, float %b) { +; CHECK-LABEL: f13: +; CHECK: ipm %r2 +; CHECK-NEXT: xilf %r2, 268435456 +; CHECK-NEXT: afi %r2, -268435456 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp uge float %a, %b + %res = select i1 %cond, i32 0, i32 -1 + ret i32 %res +} + +; Test CC in { 0 } +define i32 @f14(float %a, float %b) { +; CHECK-LABEL: f14: +; CHECK: ipm %r2 +; CHECK-NEXT: afi %r2, -268435456 +; CHECK-NEXT: sra %r2, 31 +; CHECK: br %r14 + %cond = fcmp une float %a, %b + %res = select i1 %cond, i32 0, i32 -1 + ret i32 %res +} diff --git a/test/CodeGen/SystemZ/selectcc-03.ll b/test/CodeGen/SystemZ/selectcc-03.ll new file mode 100644 index 00000000000..cafb4a2f184 --- /dev/null +++ b/test/CodeGen/SystemZ/selectcc-03.ll @@ -0,0 +1,187 @@ +; Test an i64 0/-1 SELECTCCC for every floating-point condition. +; +; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s + +; Test CC in { 0 } +define i64 @f1(float %a, float %b) { +; CHECK-LABEL: f1: +; CHECK: ipm [[REG:%r[0-5]]] +; CHECK-NEXT: afi [[REG]], -268435456 +; CHECK-NEXT: sllg [[REG]], [[REG]], 32 +; CHECK-NEXT: srag %r2, [[REG]], 63 +; CHECK: br %r14 + %cond = fcmp oeq float %a, %b + %res = select i1 %cond, i64 -1, i64 0 + ret i64 %res +} + +; Test CC in { 1 } +define i64 @f2(float %a, float %b) { +; CHECK-LABEL: f2: +; CHECK: ipm [[REG:%r[0-5]]] +; CHECK-NEXT: xilf [[REG]], 268435456 +; CHECK-NEXT: afi [[REG]], -268435456 +; CHECK-NEXT: sllg [[REG]], [[REG]], 32 +; CHECK-NEXT: srag %r2, [[REG]], 63 +; CHECK: br %r14 + %cond = fcmp olt float %a, %b + %res = select i1 %cond, i64 -1, i64 0 + ret i64 %res +} + +; Test CC in { 0, 1 } +define i64 @f3(float %a, float %b) { +; CHECK-LABEL: f3: +; CHECK: ipm [[REG:%r[0-5]]] +; CHECK-NEXT: afi [[REG]], -536870912 +; CHECK-NEXT: sllg [[REG]], [[REG]], 32 +; CHECK-NEXT: srag %r2, [[REG]], 63 +; CHECK: br %r14 + %cond = fcmp ole float %a, %b + %res = select i1 %cond, i64 -1, i64 0 + ret i64 %res +} + +; Test CC in { 2 } +define i64 @f4(float %a, float %b) { +; CHECK-LABEL: f4: +; CHECK: ipm [[REG:%r[0-5]]] +; CHECK-NEXT: xilf [[REG]], 268435456 +; CHECK-NEXT: afi [[REG]], 1342177280 +; CHECK-NEXT: sllg [[REG]], [[REG]], 32 +; CHECK-NEXT: srag %r2, [[REG]], 63 +; CHECK: br %r14 + %cond = fcmp ogt float %a, %b + %res = select i1 %cond, i64 -1, i64 0 + ret i64 %res +} + +; Test CC in { 0, 2 } +define i64 @f5(float %a, float %b) { +; CHECK-LABEL: f5: +; CHECK: ipm [[REG:%r[0-5]]] +; CHECK-NEXT: xilf [[REG]], 4294967295 +; CHECK-NEXT: sllg [[REG]], [[REG]], 35 +; CHECK-NEXT: srag %r2, [[REG]], 63 +; CHECK: br %r14 + %cond = fcmp oge float %a, %b + %res = select i1 %cond, i64 -1, i64 0 + ret i64 %res +} + +; Test CC in { 1, 2 } +define i64 @f6(float %a, float %b) { +; CHECK-LABEL: f6: +; CHECK: ipm [[REG:%r[0-5]]] +; CHECK-NEXT: afi [[REG]], 268435456 +; CHECK-NEXT: sllg [[REG]], [[REG]], 34 +; CHECK-NEXT: srag %r2, [[REG]], 63 +; CHECK: br %r14 + %cond = fcmp one float %a, %b + %res = select i1 %cond, i64 -1, i64 0 + ret i64 %res +} + +; Test CC in { 0, 1, 2 } +define i64 @f7(float %a, float %b) { +; CHECK-LABEL: f7: +; CHECK: ipm [[REG:%r[0-5]]] +; CHECK-NEXT: afi [[REG]], -805306368 +; CHECK-NEXT: sllg [[REG]], [[REG]], 32 +; CHECK-NEXT: srag %r2, [[REG]], 63 +; CHECK: br %r14 + %cond = fcmp ord float %a, %b + %res = select i1 %cond, i64 -1, i64 0 + ret i64 %res +} + +; Test CC in { 3 } +define i64 @f8(float %a, float %b) { +; CHECK-LABEL: f8: +; CHECK: ipm [[REG:%r[0-5]]] +; CHECK-NEXT: afi [[REG]], 1342177280 +; CHECK-NEXT: sllg [[REG]], [[REG]], 32 +; CHECK-NEXT: srag %r2, [[REG]], 63 +; CHECK: br %r14 + %cond = fcmp uno float %a, %b + %res = select i1 %cond, i64 -1, i64 0 + ret i64 %res +} + +; Test CC in { 0, 3 } +define i64 @f9(float %a, float %b) { +; CHECK-LABEL: f9: +; CHECK: ipm [[REG:%r[0-5]]] +; CHECK-NEXT: afi [[REG]], -268435456 +; CHECK-NEXT: sllg [[REG]], [[REG]], 34 +; CHECK-NEXT: srag %r2, [[REG]], 63 +; CHECK: br %r14 + %cond = fcmp ueq float %a, %b + %res = select i1 %cond, i64 -1, i64 0 + ret i64 %res +} + +; Test CC in { 1, 3 } +define i64 @f10(float %a, float %b) { +; CHECK-LABEL: f10: +; CHECK: ipm [[REG:%r[0-5]]] +; CHECK-NEXT: sllg [[REG]], [[REG]], 35 +; CHECK-NEXT: srag %r2, [[REG]], 63 +; CHECK: br %r14 + %cond = fcmp ult float %a, %b + %res = select i1 %cond, i64 -1, i64 0 + ret i64 %res +} + +; Test CC in { 0, 1, 3 } +define i64 @f11(float %a, float %b) { +; CHECK-LABEL: f11: +; CHECK: ipm [[REG:%r[0-5]]] +; CHECK-NEXT: xilf [[REG]], 268435456 +; CHECK-NEXT: afi [[REG]], -805306368 +; CHECK-NEXT: sllg [[REG]], [[REG]], 32 +; CHECK-NEXT: srag %r2, [[REG]], 63 +; CHECK: br %r14 + %cond = fcmp ule float %a, %b + %res = select i1 %cond, i64 -1, i64 0 + ret i64 %res +} + +; Test CC in { 2, 3 } +define i64 @f12(float %a, float %b) { +; CHECK-LABEL: f12: +; CHECK: ipm [[REG:%r[0-5]]] +; CHECK-NEXT: sllg [[REG]], [[REG]], 34 +; CHECK-NEXT: srag %r2, [[REG]], 63 +; CHECK: br %r14 + %cond = fcmp ugt float %a, %b + %res = select i1 %cond, i64 -1, i64 0 + ret i64 %res +} + +; Test CC in { 0, 2, 3 } +define i64 @f13(float %a, float %b) { +; CHECK-LABEL: f13: +; CHECK: ipm [[REG:%r[0-5]]] +; CHECK-NEXT: xilf [[REG]], 268435456 +; CHECK-NEXT: afi [[REG]], 1879048192 +; CHECK-NEXT: sllg [[REG]], [[REG]], 32 +; CHECK-NEXT: srag %r2, [[REG]], 63 +; CHECK: br %r14 + %cond = fcmp uge float %a, %b + %res = select i1 %cond, i64 -1, i64 0 + ret i64 %res +} + +; Test CC in { 1, 2, 3 } +define i64 @f14(float %a, float %b) { +; CHECK-LABEL: f14: +; CHECK: ipm [[REG:%r[0-5]]] +; CHECK-NEXT: afi [[REG]], 1879048192 +; CHECK-NEXT: sllg [[REG]], [[REG]], 32 +; CHECK-NEXT: srag %r2, [[REG]], 63 +; CHECK: br %r14 + %cond = fcmp une float %a, %b + %res = select i1 %cond, i64 -1, i64 0 + ret i64 %res +} -- 2.34.1