From c5cf6e536598a3b1e30fce616b771d66a071a54c Mon Sep 17 00:00:00 2001 From: Jean-Luc Duprat Date: Mon, 6 May 2013 16:55:50 +0000 Subject: Provide InstCombines for the following 3 cases: A * (1 - (uitofp i1 C)) -> select C, 0, A B * (uitofp i1 C) -> select C, B, 0 select C, 0, A + select C, B, 0 -> select C, B, A These come up in code that has been hand-optimized from a select to a linear blend, on platforms where that may have mattered. We want to undo such changes with the following transform: A*(1 - uitofp i1 C) + B*(uitofp i1 C) -> select C, A, B git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181216 91177308-0d34-0410-b5e6-96231b3b80d8 --- include/llvm/Support/PatternMatch.h | 6 ++++ lib/Transforms/InstCombine/InstCombineAddSub.cpp | 25 ++++++++++++++++ .../InstCombine/InstCombineMulDivRem.cpp | 28 ++++++++++++++++++ test/Transforms/InstCombine/add4.ll | 34 ++++++++++++++++++++++ 4 files changed, 93 insertions(+) create mode 100644 test/Transforms/InstCombine/add4.ll diff --git a/include/llvm/Support/PatternMatch.h b/include/llvm/Support/PatternMatch.h index 51b66b0dcd..95d9d782ec 100644 --- a/include/llvm/Support/PatternMatch.h +++ b/include/llvm/Support/PatternMatch.h @@ -693,6 +693,12 @@ m_ZExt(const OpTy &Op) { return CastClass_match(Op); } +/// m_UIToFP +template +inline CastClass_match +m_UIToFp(const OpTy &Op) { + return CastClass_match(Op); +} //===----------------------------------------------------------------------===// // Matchers for unary operators diff --git a/lib/Transforms/InstCombine/InstCombineAddSub.cpp b/lib/Transforms/InstCombine/InstCombineAddSub.cpp index b96eb51081..1aa51d06cb 100644 --- a/lib/Transforms/InstCombine/InstCombineAddSub.cpp +++ b/lib/Transforms/InstCombine/InstCombineAddSub.cpp @@ -1232,6 +1232,31 @@ Instruction *InstCombiner::visitFAdd(BinaryOperator &I) { } } + // select C, 0, B + select C, A, 0 -> select C, A, B + { + Value *A1, *B1, *C1, *A2, *B2, *C2; + if (match(LHS, m_Select(m_Value(C1), m_Value(A1), m_Value(B1))) && + match(RHS, m_Select(m_Value(C2), m_Value(A2), m_Value(B2)))) { + if (C1 == C2) { + Constant *Z1=0, *Z2=0; + Value *A, *B, *C=C1; + if (match(A1, m_AnyZero()) && match(B2, m_AnyZero())) { + Z1 = dyn_cast(A1); A = A2; + Z2 = dyn_cast(B2); B = B1; + } else if (match(B1, m_AnyZero()) && match(A2, m_AnyZero())) { + Z1 = dyn_cast(B1); B = B2; + Z2 = dyn_cast(A2); A = A1; + } + + if (Z1 && Z2 && + (I.hasNoSignedZeros() || + (Z1->isNegativeZeroValue() && Z2->isNegativeZeroValue()))) { + return SelectInst::Create(C, A, B); + } + } + } + } + if (I.hasUnsafeAlgebra()) { if (Value *V = FAddCombine(Builder).simplify(&I)) return ReplaceInstUsesWith(I, V); diff --git a/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp b/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp index df7390652f..0ac0ca27ac 100644 --- a/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp +++ b/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp @@ -524,6 +524,34 @@ Instruction *InstCombiner::visitFMul(BinaryOperator &I) { } } + // B * (uitofp i1 C) -> select C, B, 0 + if(I.hasNoNaNs() && I.hasNoInfs() && I.hasNoSignedZeros()) { + Value *LHS=Op0, *RHS=Op1; + Value *B, *C; + if (!match(RHS, m_UIToFp(m_Value(C)))) + std::swap(LHS, RHS); + + if (match(RHS, m_UIToFp(m_Value(C)))) { + B=LHS; + Value *Zero = ConstantFP::getNegativeZero(B->getType()); + return SelectInst::Create(C, B, Zero); + } + } + + // A * (1 - uitofp i1 C) -> select C, 0, A + if(I.hasNoNaNs() && I.hasNoInfs() && I.hasNoSignedZeros()) { + Value *LHS=Op0, *RHS=Op1; + Value *A, *C; + if (!match(RHS, m_FSub(m_FPOne(), m_UIToFp(m_Value(C))))) + std::swap(LHS, RHS); + + if (match(RHS, m_FSub(m_FPOne(), m_UIToFp(m_Value(C))))) { + A=LHS; + Value *Zero = ConstantFP::getNegativeZero(A->getType()); + return SelectInst::Create(C, Zero, A); + } + } + if (!isa(Op1)) std::swap(Opnd0, Opnd1); else diff --git a/test/Transforms/InstCombine/add4.ll b/test/Transforms/InstCombine/add4.ll new file mode 100644 index 0000000000..9cc24e9205 --- /dev/null +++ b/test/Transforms/InstCombine/add4.ll @@ -0,0 +1,34 @@ +; RUN: opt < %s -instcombine -S | grep inttoptr | count 2 + +;; Target triple for gep raising case below. +target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128" +target triple = "i686-apple-darwin8" + +define float @test1(float %A, float %B, i1 %C) { +EntryBlock: + ;; A*(1 - uitofp i1 C) -> select C, 0, A + %cf = uitofp i1 %C to float + %mc = fsub float 1.000000e+00, %cf + %p1 = fmul fast float %A, %mc + ret float %p1 +} + +define float @test2(float %A, float %B, i1 %C) { +EntryBlock: + ;; B*(uitofp i1 C) -> select C, B, 0 + %cf = uitofp i1 %C to float + %p2 = fmul fast float %B, %cf + ret float %p2 +} + +define float @test3(float %A, float %B, i1 %C) { +EntryBlock: + ;; A*(1 - uitofp i1 C) + B*(uitofp i1 C) -> select C, A, B + %cf = uitofp i1 %C to float + %mc = fsub float 1.000000e+00, %cf + %p1 = fmul fast float %A, %mc + %p2 = fmul fast float %B, %cf + %s1 = fadd fast float %p1, %p2 + ret float %s1 +} + -- cgit v1.2.3