From 53624a2df557b4a24f2ee98cfce1a69bf83243af Mon Sep 17 00:00:00 2001 From: Bob Wilson Date: Fri, 3 Aug 2012 23:29:17 +0000 Subject: [PATCH] Refactor and check "onlyReadsMemory" before optimizing builtins. This patch is mostly just refactoring a bunch of copy-and-pasted code, but it also adds a check that the call instructions are readnone or readonly. That check was already present for sin, cos, sqrt, log2, and exp2 calls, but it was missing for the rest of the builtins being handled in this code. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@161282 91177308-0d34-0410-b5e6-96231b3b80d8 --- .../SelectionDAG/SelectionDAGBuilder.cpp | 112 +++++------------- .../SelectionDAG/SelectionDAGBuilder.h | 1 + test/CodeGen/ARM/fabss.ll | 2 +- test/CodeGen/ARM/fcopysign.ll | 6 +- test/CodeGen/ARM/fparith.ll | 4 +- test/CodeGen/ARM/vfp.ll | 4 +- test/CodeGen/CellSPU/fcmp32.ll | 4 +- test/CodeGen/CellSPU/fneg-fabs.ll | 4 +- test/CodeGen/Hexagon/opt-fabs.ll | 2 +- test/CodeGen/PowerPC/fabs.ll | 2 +- test/CodeGen/PowerPC/fnabs.ll | 2 +- test/CodeGen/X86/fabs.ll | 4 +- test/CodeGen/X86/stack-align.ll | 4 +- 13 files changed, 49 insertions(+), 102 deletions(-) diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp index a6c28587946..ba5bd79722c 100644 --- a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp +++ b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp @@ -5511,6 +5511,22 @@ bool SelectionDAGBuilder::visitMemCmpCall(const CallInst &I) { return false; } +/// visitUnaryFloatCall - If a call instruction is a unary floating-point +/// operation (as expected), translate it to an SDNode with the specified opcode +/// and return true. +bool SelectionDAGBuilder::visitUnaryFloatCall(const CallInst &I, + unsigned Opcode) { + // Sanity check that it really is a unary floating-point call. + if (I.getNumArgOperands() != 1 || + !I.getArgOperand(0)->getType()->isFloatingPointTy() || + I.getType() != I.getArgOperand(0)->getType() || + !I.onlyReadsMemory()) + return false; + + SDValue Tmp = getValue(I.getArgOperand(0)); + setValue(&I, DAG.getNode(Opcode, getCurDebugLoc(), Tmp.getValueType(), Tmp)); + return true; +} void SelectionDAGBuilder::visitCall(const CallInst &I) { // Handle inline assembly differently. @@ -5553,7 +5569,8 @@ void SelectionDAGBuilder::visitCall(const CallInst &I) { if (I.getNumArgOperands() == 2 && // Basic sanity checks. I.getArgOperand(0)->getType()->isFloatingPointTy() && I.getType() == I.getArgOperand(0)->getType() && - I.getType() == I.getArgOperand(1)->getType()) { + I.getType() == I.getArgOperand(1)->getType() && + I.onlyReadsMemory()) { SDValue LHS = getValue(I.getArgOperand(0)); SDValue RHS = getValue(I.getArgOperand(1)); setValue(&I, DAG.getNode(ISD::FCOPYSIGN, getCurDebugLoc(), @@ -5564,139 +5581,68 @@ void SelectionDAGBuilder::visitCall(const CallInst &I) { case LibFunc::fabs: case LibFunc::fabsf: case LibFunc::fabsl: - if (I.getNumArgOperands() == 1 && // Basic sanity checks. - I.getArgOperand(0)->getType()->isFloatingPointTy() && - I.getType() == I.getArgOperand(0)->getType()) { - SDValue Tmp = getValue(I.getArgOperand(0)); - setValue(&I, DAG.getNode(ISD::FABS, getCurDebugLoc(), - Tmp.getValueType(), Tmp)); + if (visitUnaryFloatCall(I, ISD::FABS)) return; - } break; case LibFunc::sin: case LibFunc::sinf: case LibFunc::sinl: - if (I.getNumArgOperands() == 1 && // Basic sanity checks. - I.getArgOperand(0)->getType()->isFloatingPointTy() && - I.getType() == I.getArgOperand(0)->getType() && - I.onlyReadsMemory()) { - SDValue Tmp = getValue(I.getArgOperand(0)); - setValue(&I, DAG.getNode(ISD::FSIN, getCurDebugLoc(), - Tmp.getValueType(), Tmp)); + if (visitUnaryFloatCall(I, ISD::FSIN)) return; - } break; case LibFunc::cos: case LibFunc::cosf: case LibFunc::cosl: - if (I.getNumArgOperands() == 1 && // Basic sanity checks. - I.getArgOperand(0)->getType()->isFloatingPointTy() && - I.getType() == I.getArgOperand(0)->getType() && - I.onlyReadsMemory()) { - SDValue Tmp = getValue(I.getArgOperand(0)); - setValue(&I, DAG.getNode(ISD::FCOS, getCurDebugLoc(), - Tmp.getValueType(), Tmp)); + if (visitUnaryFloatCall(I, ISD::FCOS)) return; - } break; case LibFunc::sqrt: case LibFunc::sqrtf: case LibFunc::sqrtl: - if (I.getNumArgOperands() == 1 && // Basic sanity checks. - I.getArgOperand(0)->getType()->isFloatingPointTy() && - I.getType() == I.getArgOperand(0)->getType() && - I.onlyReadsMemory()) { - SDValue Tmp = getValue(I.getArgOperand(0)); - setValue(&I, DAG.getNode(ISD::FSQRT, getCurDebugLoc(), - Tmp.getValueType(), Tmp)); + if (visitUnaryFloatCall(I, ISD::FSQRT)) return; - } break; case LibFunc::floor: case LibFunc::floorf: case LibFunc::floorl: - if (I.getNumArgOperands() == 1 && // Basic sanity checks. - I.getArgOperand(0)->getType()->isFloatingPointTy() && - I.getType() == I.getArgOperand(0)->getType()) { - SDValue Tmp = getValue(I.getArgOperand(0)); - setValue(&I, DAG.getNode(ISD::FFLOOR, getCurDebugLoc(), - Tmp.getValueType(), Tmp)); + if (visitUnaryFloatCall(I, ISD::FFLOOR)) return; - } break; case LibFunc::nearbyint: case LibFunc::nearbyintf: case LibFunc::nearbyintl: - if (I.getNumArgOperands() == 1 && // Basic sanity checks. - I.getArgOperand(0)->getType()->isFloatingPointTy() && - I.getType() == I.getArgOperand(0)->getType()) { - SDValue Tmp = getValue(I.getArgOperand(0)); - setValue(&I, DAG.getNode(ISD::FNEARBYINT, getCurDebugLoc(), - Tmp.getValueType(), Tmp)); + if (visitUnaryFloatCall(I, ISD::FNEARBYINT)) return; - } break; case LibFunc::ceil: case LibFunc::ceilf: case LibFunc::ceill: - if (I.getNumArgOperands() == 1 && // Basic sanity checks. - I.getArgOperand(0)->getType()->isFloatingPointTy() && - I.getType() == I.getArgOperand(0)->getType()) { - SDValue Tmp = getValue(I.getArgOperand(0)); - setValue(&I, DAG.getNode(ISD::FCEIL, getCurDebugLoc(), - Tmp.getValueType(), Tmp)); + if (visitUnaryFloatCall(I, ISD::FCEIL)) return; - } break; case LibFunc::rint: case LibFunc::rintf: case LibFunc::rintl: - if (I.getNumArgOperands() == 1 && // Basic sanity checks. - I.getArgOperand(0)->getType()->isFloatingPointTy() && - I.getType() == I.getArgOperand(0)->getType()) { - SDValue Tmp = getValue(I.getArgOperand(0)); - setValue(&I, DAG.getNode(ISD::FRINT, getCurDebugLoc(), - Tmp.getValueType(), Tmp)); + if (visitUnaryFloatCall(I, ISD::FRINT)) return; - } break; case LibFunc::trunc: case LibFunc::truncf: case LibFunc::truncl: - if (I.getNumArgOperands() == 1 && // Basic sanity checks. - I.getArgOperand(0)->getType()->isFloatingPointTy() && - I.getType() == I.getArgOperand(0)->getType()) { - SDValue Tmp = getValue(I.getArgOperand(0)); - setValue(&I, DAG.getNode(ISD::FTRUNC, getCurDebugLoc(), - Tmp.getValueType(), Tmp)); + if (visitUnaryFloatCall(I, ISD::FTRUNC)) return; - } break; case LibFunc::log2: case LibFunc::log2f: case LibFunc::log2l: - if (I.getNumArgOperands() == 1 && // Basic sanity checks. - I.getArgOperand(0)->getType()->isFloatingPointTy() && - I.getType() == I.getArgOperand(0)->getType() && - I.onlyReadsMemory()) { - SDValue Tmp = getValue(I.getArgOperand(0)); - setValue(&I, DAG.getNode(ISD::FLOG2, getCurDebugLoc(), - Tmp.getValueType(), Tmp)); + if (visitUnaryFloatCall(I, ISD::FLOG2)) return; - } break; case LibFunc::exp2: case LibFunc::exp2f: case LibFunc::exp2l: - if (I.getNumArgOperands() == 1 && // Basic sanity checks. - I.getArgOperand(0)->getType()->isFloatingPointTy() && - I.getType() == I.getArgOperand(0)->getType() && - I.onlyReadsMemory()) { - SDValue Tmp = getValue(I.getArgOperand(0)); - setValue(&I, DAG.getNode(ISD::FEXP2, getCurDebugLoc(), - Tmp.getValueType(), Tmp)); + if (visitUnaryFloatCall(I, ISD::FEXP2)) return; - } break; case LibFunc::memcmp: if (visitMemCmpCall(I)) diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h index d0fde6f01d6..40900023140 100644 --- a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h +++ b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h @@ -520,6 +520,7 @@ private: void visitPHI(const PHINode &I); void visitCall(const CallInst &I); bool visitMemCmpCall(const CallInst &I); + bool visitUnaryFloatCall(const CallInst &I, unsigned Opcode); void visitAtomicLoad(const LoadInst &I); void visitAtomicStore(const StoreInst &I); diff --git a/test/CodeGen/ARM/fabss.ll b/test/CodeGen/ARM/fabss.ll index 45c322dce8b..5c0db88407a 100644 --- a/test/CodeGen/ARM/fabss.ll +++ b/test/CodeGen/ARM/fabss.ll @@ -6,7 +6,7 @@ define float @test(float %a, float %b) { entry: %dum = fadd float %a, %b - %0 = tail call float @fabsf(float %dum) + %0 = tail call float @fabsf(float %dum) readnone %dum1 = fadd float %0, %b ret float %dum1 } diff --git a/test/CodeGen/ARM/fcopysign.ll b/test/CodeGen/ARM/fcopysign.ll index 27fa2b093d8..5511d24cb28 100644 --- a/test/CodeGen/ARM/fcopysign.ll +++ b/test/CodeGen/ARM/fcopysign.ll @@ -11,7 +11,7 @@ entry: ; HARD: test1: ; HARD: vmov.i32 [[REG1:(d[0-9]+)]], #0x80000000 ; HARD: vbsl [[REG1]], d - %0 = tail call float @copysignf(float %x, float %y) nounwind + %0 = tail call float @copysignf(float %x, float %y) nounwind readnone ret float %0 } @@ -25,7 +25,7 @@ entry: ; HARD: vmov.i32 [[REG2:(d[0-9]+)]], #0x80000000 ; HARD: vshl.i64 [[REG2]], [[REG2]], #32 ; HARD: vbsl [[REG2]], d1, d0 - %0 = tail call double @copysign(double %x, double %y) nounwind + %0 = tail call double @copysign(double %x, double %y) nounwind readnone ret double %0 } @@ -36,7 +36,7 @@ entry: ; SOFT: vshl.i64 [[REG3]], [[REG3]], #32 ; SOFT: vbsl [[REG3]], %0 = fmul double %x, %y - %1 = tail call double @copysign(double %0, double %z) nounwind + %1 = tail call double @copysign(double %0, double %z) nounwind readnone ret double %1 } diff --git a/test/CodeGen/ARM/fparith.ll b/test/CodeGen/ARM/fparith.ll index ce6d6b29e9d..a8bae3b9519 100644 --- a/test/CodeGen/ARM/fparith.ll +++ b/test/CodeGen/ARM/fparith.ll @@ -84,7 +84,7 @@ define float @f11(float %a) { ;CHECK: f11: ;CHECK: bic entry: - %tmp1 = call float @fabsf( float %a ) ; [#uses=1] + %tmp1 = call float @fabsf( float %a ) readnone ; [#uses=1] ret float %tmp1 } @@ -94,7 +94,7 @@ define double @f12(double %a) { ;CHECK: f12: ;CHECK: vabs.f64 entry: - %tmp1 = call double @fabs( double %a ) ; [#uses=1] + %tmp1 = call double @fabs( double %a ) readnone ; [#uses=1] ret double %tmp1 } diff --git a/test/CodeGen/ARM/vfp.ll b/test/CodeGen/ARM/vfp.ll index 49a69827bc0..bf0f3680c13 100644 --- a/test/CodeGen/ARM/vfp.ll +++ b/test/CodeGen/ARM/vfp.ll @@ -17,11 +17,11 @@ define void @test_abs(float* %P, double* %D) { ;CHECK: test_abs: %a = load float* %P ; [#uses=1] ;CHECK: vabs.f32 - %b = call float @fabsf( float %a ) ; [#uses=1] + %b = call float @fabsf( float %a ) readnone ; [#uses=1] store float %b, float* %P %A = load double* %D ; [#uses=1] ;CHECK: vabs.f64 - %B = call double @fabs( double %A ) ; [#uses=1] + %B = call double @fabs( double %A ) readnone ; [#uses=1] store double %B, double* %D ret void } diff --git a/test/CodeGen/CellSPU/fcmp32.ll b/test/CodeGen/CellSPU/fcmp32.ll index c14fd7ba4a4..b2a7317c129 100644 --- a/test/CodeGen/CellSPU/fcmp32.ll +++ b/test/CodeGen/CellSPU/fcmp32.ll @@ -15,8 +15,8 @@ define i1 @fcmp_eq(float %arg1, float %arg2) { define i1 @fcmp_mag_eq(float %arg1, float %arg2) { ; CHECK: fcmeq ; CHECK: bi $lr - %1 = call float @fabsf(float %arg1) - %2 = call float @fabsf(float %arg2) + %1 = call float @fabsf(float %arg1) readnone + %2 = call float @fabsf(float %arg2) readnone %3 = fcmp oeq float %1, %2 ret i1 %3 } diff --git a/test/CodeGen/CellSPU/fneg-fabs.ll b/test/CodeGen/CellSPU/fneg-fabs.ll index 1e5e3b34144..6e01906dae6 100644 --- a/test/CodeGen/CellSPU/fneg-fabs.ll +++ b/test/CodeGen/CellSPU/fneg-fabs.ll @@ -32,11 +32,11 @@ declare double @fabs(double) declare float @fabsf(float) define double @fabs_dp(double %X) { - %Y = call double @fabs( double %X ) + %Y = call double @fabs( double %X ) readnone ret double %Y } define float @fabs_sp(float %X) { - %Y = call float @fabsf( float %X ) + %Y = call float @fabsf( float %X ) readnone ret float %Y } diff --git a/test/CodeGen/Hexagon/opt-fabs.ll b/test/CodeGen/Hexagon/opt-fabs.ll index 1cf0dd0cd99..f1a66f4e65c 100644 --- a/test/CodeGen/Hexagon/opt-fabs.ll +++ b/test/CodeGen/Hexagon/opt-fabs.ll @@ -8,7 +8,7 @@ entry: %x.addr = alloca float, align 4 store float %x, float* %x.addr, align 4 %0 = load float* %x.addr, align 4 - %call = call float @fabsf(float %0) + %call = call float @fabsf(float %0) readnone ret float %call } diff --git a/test/CodeGen/PowerPC/fabs.ll b/test/CodeGen/PowerPC/fabs.ll index 156e00b4e57..ddcce745084 100644 --- a/test/CodeGen/PowerPC/fabs.ll +++ b/test/CodeGen/PowerPC/fabs.ll @@ -2,6 +2,6 @@ define double @fabs(double %f) { entry: - %tmp2 = tail call double @fabs( double %f ) ; [#uses=1] + %tmp2 = tail call double @fabs( double %f ) readnone ; [#uses=1] ret double %tmp2 } diff --git a/test/CodeGen/PowerPC/fnabs.ll b/test/CodeGen/PowerPC/fnabs.ll index bbd5c7159ed..9fa2dcb2909 100644 --- a/test/CodeGen/PowerPC/fnabs.ll +++ b/test/CodeGen/PowerPC/fnabs.ll @@ -3,7 +3,7 @@ declare double @fabs(double) define double @test(double %X) { - %Y = call double @fabs( double %X ) ; [#uses=1] + %Y = call double @fabs( double %X ) readnone ; [#uses=1] %Z = fsub double -0.000000e+00, %Y ; [#uses=1] ret double %Z } diff --git a/test/CodeGen/X86/fabs.ll b/test/CodeGen/X86/fabs.ll index 07a8d91185b..2074f04d405 100644 --- a/test/CodeGen/X86/fabs.ll +++ b/test/CodeGen/X86/fabs.ll @@ -11,7 +11,7 @@ declare x86_fp80 @fabsl(x86_fp80) ; UNSAFE: test1: ; NOOPT: test1: define float @test1(float %X) { - %Y = call float @fabsf(float %X) + %Y = call float @fabsf(float %X) readnone ret float %Y } ; CHECK: {{^[ \t]+fabs$}} @@ -42,7 +42,7 @@ define double @test2(double %X) { ; UNSAFE: test3: ; NOOPT: test3: define x86_fp80 @test3(x86_fp80 %X) { - %Y = call x86_fp80 @fabsl(x86_fp80 %X) + %Y = call x86_fp80 @fabsl(x86_fp80 %X) readnone ret x86_fp80 %Y } ; CHECK: {{^[ \t]+fabs$}} diff --git a/test/CodeGen/X86/stack-align.ll b/test/CodeGen/X86/stack-align.ll index f6c13ec0adf..0ddb2378ef2 100644 --- a/test/CodeGen/X86/stack-align.ll +++ b/test/CodeGen/X86/stack-align.ll @@ -10,11 +10,11 @@ target triple = "i686-apple-darwin8" define void @test({ double, double }* byval %z, double* %P) nounwind { entry: %tmp3 = load double* @G, align 16 ; [#uses=1] - %tmp4 = tail call double @fabs( double %tmp3 ) ; [#uses=1] + %tmp4 = tail call double @fabs( double %tmp3 ) readnone ; [#uses=1] store volatile double %tmp4, double* %P %tmp = getelementptr { double, double }* %z, i32 0, i32 0 ; [#uses=1] %tmp1 = load volatile double* %tmp, align 8 ; [#uses=1] - %tmp2 = tail call double @fabs( double %tmp1 ) ; [#uses=1] + %tmp2 = tail call double @fabs( double %tmp1 ) readnone ; [#uses=1] ; CHECK: andpd{{.*}}4(%esp), %xmm %tmp6 = fadd double %tmp4, %tmp2 ; [#uses=1] store volatile double %tmp6, double* %P, align 8 -- 2.34.1