From: Benjamin Kramer Date: Tue, 21 Dec 2010 21:41:44 +0000 (+0000) Subject: Add some x86 specific dagcombines for conditional increments. X-Git-Url: http://plrg.eecs.uci.edu/git/?a=commitdiff_plain;h=7d6fe13efc9bcb69d3479e32f8eace933c5bd2d5;p=oota-llvm.git Add some x86 specific dagcombines for conditional increments. (add Y, (sete X, 0)) -> cmp X, 1; adc 0, Y (add Y, (setne X, 0)) -> cmp X, 1; sbb -1, Y (sub (sete X, 0), Y) -> cmp X, 1; sbb 0, Y (sub (setne X, 0), Y) -> cmp X, 1; adc -1, Y for unsigned foo(unsigned a, unsigned b) { if (a == 0) b++; return b; } we now get: foo: cmpl $1, %edi movl %esi, %eax adcl $0, %eax ret instead of: foo: testl %edi, %edi sete %al movzbl %al, %eax addl %esi, %eax ret git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122364 91177308-0d34-0410-b5e6-96231b3b80d8 --- diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp index ba357ed76cf..5290453f05a 100644 --- a/lib/Target/X86/X86ISelLowering.cpp +++ b/lib/Target/X86/X86ISelLowering.cpp @@ -969,6 +969,8 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM) setTargetDAGCombine(ISD::SRL); setTargetDAGCombine(ISD::OR); setTargetDAGCombine(ISD::AND); + setTargetDAGCombine(ISD::ADD); + setTargetDAGCombine(ISD::SUB); setTargetDAGCombine(ISD::STORE); setTargetDAGCombine(ISD::ZERO_EXTEND); if (Subtarget->is64Bit()) @@ -11513,6 +11515,44 @@ static SDValue PerformADCCombine(SDNode *N, SelectionDAG &DAG, return SDValue(); } +// fold (add Y, (sete X, 0)) -> adc 0, Y +// (add Y, (setne X, 0)) -> sbb -1, Y +// (sub (sete X, 0), Y) -> sbb 0, Y +// (sub (setne X, 0), Y) -> adc -1, Y +static SDValue OptimizeConditonalInDecrement(SDNode *N, SelectionDAG &DAG) { + DebugLoc DL = N->getDebugLoc(); + + // Look through ZExts. + SDValue Ext = N->getOperand(N->getOpcode() == ISD::SUB ? 1 : 0); + if (Ext.getOpcode() != ISD::ZERO_EXTEND || !Ext.hasOneUse()) + return SDValue(); + + SDValue SetCC = Ext.getOperand(0); + if (SetCC.getOpcode() != X86ISD::SETCC || !SetCC.hasOneUse()) + return SDValue(); + + X86::CondCode CC = (X86::CondCode)SetCC.getConstantOperandVal(0); + if (CC != X86::COND_E && CC != X86::COND_NE) + return SDValue(); + + SDValue Cmp = SetCC.getOperand(1); + if (Cmp.getOpcode() != X86ISD::CMP || !Cmp.hasOneUse() || + !X86::isZeroNode(Cmp.getOperand(1))) + return SDValue(); + + SDValue CmpOp0 = Cmp.getOperand(0); + SDValue NewCmp = DAG.getNode(X86ISD::CMP, DL, MVT::i32, CmpOp0, + DAG.getConstant(1, CmpOp0.getValueType())); + + SDValue OtherVal = N->getOperand(N->getOpcode() == ISD::SUB ? 0 : 1); + if (CC == X86::COND_NE) + return DAG.getNode(N->getOpcode() == ISD::SUB ? X86ISD::ADC : X86ISD::SBB, + DL, OtherVal.getValueType(), OtherVal, + DAG.getConstant(-1ULL, OtherVal.getValueType()), NewCmp); + return DAG.getNode(N->getOpcode() == ISD::SUB ? X86ISD::SBB : X86ISD::ADC, + DL, OtherVal.getValueType(), OtherVal, + DAG.getConstant(0, OtherVal.getValueType()), NewCmp); +} SDValue X86TargetLowering::PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const { @@ -11523,6 +11563,8 @@ SDValue X86TargetLowering::PerformDAGCombine(SDNode *N, return PerformEXTRACT_VECTOR_ELTCombine(N, DAG, *this); case ISD::SELECT: return PerformSELECTCombine(N, DAG, Subtarget); case X86ISD::CMOV: return PerformCMOVCombine(N, DAG, DCI); + case ISD::ADD: + case ISD::SUB: return OptimizeConditonalInDecrement(N, DAG); case X86ISD::ADC: return PerformADCCombine(N, DAG, DCI); case ISD::MUL: return PerformMulCombine(N, DAG, DCI); case ISD::SHL: diff --git a/test/CodeGen/X86/conditional-indecrement.ll b/test/CodeGen/X86/conditional-indecrement.ll new file mode 100644 index 00000000000..a3a0c39905a --- /dev/null +++ b/test/CodeGen/X86/conditional-indecrement.ll @@ -0,0 +1,89 @@ +; RUN: llc -march=x86 < %s | FileCheck %s + +define i32 @test1(i32 %a, i32 %b) nounwind readnone { + %not.cmp = icmp ne i32 %a, 0 + %inc = zext i1 %not.cmp to i32 + %retval.0 = add i32 %inc, %b + ret i32 %retval.0 +; CHECK: test1: +; CHECK: cmpl $1 +; CHECK: sbbl $-1 +; CHECK: ret +} + +define i32 @test2(i32 %a, i32 %b) nounwind readnone { + %cmp = icmp eq i32 %a, 0 + %inc = zext i1 %cmp to i32 + %retval.0 = add i32 %inc, %b + ret i32 %retval.0 +; CHECK: test2: +; CHECK: cmpl $1 +; CHECK: adcl $0 +; CHECK: ret +} + +define i32 @test3(i32 %a, i32 %b) nounwind readnone { + %cmp = icmp eq i32 %a, 0 + %inc = zext i1 %cmp to i32 + %retval.0 = add i32 %inc, %b + ret i32 %retval.0 +; CHECK: test3: +; CHECK: cmpl $1 +; CHECK: adcl $0 +; CHECK: ret +} + +define i32 @test4(i32 %a, i32 %b) nounwind readnone { + %not.cmp = icmp ne i32 %a, 0 + %inc = zext i1 %not.cmp to i32 + %retval.0 = add i32 %inc, %b + ret i32 %retval.0 +; CHECK: test4: +; CHECK: cmpl $1 +; CHECK: sbbl $-1 +; CHECK: ret +} + +define i32 @test5(i32 %a, i32 %b) nounwind readnone { + %not.cmp = icmp ne i32 %a, 0 + %inc = zext i1 %not.cmp to i32 + %retval.0 = sub i32 %b, %inc + ret i32 %retval.0 +; CHECK: test5: +; CHECK: cmpl $1 +; CHECK: adcl $-1 +; CHECK: ret +} + +define i32 @test6(i32 %a, i32 %b) nounwind readnone { + %cmp = icmp eq i32 %a, 0 + %inc = zext i1 %cmp to i32 + %retval.0 = sub i32 %b, %inc + ret i32 %retval.0 +; CHECK: test6: +; CHECK: cmpl $1 +; CHECK: sbbl $0 +; CHECK: ret +} + +define i32 @test7(i32 %a, i32 %b) nounwind readnone { + %cmp = icmp eq i32 %a, 0 + %inc = zext i1 %cmp to i32 + %retval.0 = sub i32 %b, %inc + ret i32 %retval.0 +; CHECK: test7: +; CHECK: cmpl $1 +; CHECK: sbbl $0 +; CHECK: ret +} + +define i32 @test8(i32 %a, i32 %b) nounwind readnone { + %not.cmp = icmp ne i32 %a, 0 + %inc = zext i1 %not.cmp to i32 + %retval.0 = sub i32 %b, %inc + ret i32 %retval.0 +; CHECK: test8: +; CHECK: cmpl $1 +; CHECK: adcl $-1 +; CHECK: ret +}