diff options
author | Jean-Luc Duprat <jduprat@apple.com> | 2013-05-22 18:29:31 +0000 |
---|---|---|
committer | Jean-Luc Duprat <jduprat@apple.com> | 2013-05-22 18:29:31 +0000 |
commit | 5e6cabd163c9c207df3778f3508b0de445578057 (patch) | |
tree | 1c7d2157422f09ffd9ad2df66a07ec67f2162597 | |
parent | e0f26bc008bf7eb176595ab170cca86e21a6eae8 (diff) | |
download | external_llvm-5e6cabd163c9c207df3778f3508b0de445578057.zip external_llvm-5e6cabd163c9c207df3778f3508b0de445578057.tar.gz external_llvm-5e6cabd163c9c207df3778f3508b0de445578057.tar.bz2 |
This is an update to a previous commit (r181216).
The earlier change list introduced the following inst combines:
B * (uitofp i1 C) —> select C, B, 0
A * (1 - uitofp i1 C) —> select C, 0, A
select C, 0, B + select C, A, 0 —> select C, A, B
Together these 3 changes would simplify :
A * (1 - uitofp i1 C) + B * uitofp i1 C
down to :
select C, B, A
In practice we found that the first two substitutions can have a
negative effect on performance, because they reduce opportunities to
use FMA contractions; between the two options FMAs are often the
better choice. This change list amends the previous one to enable
just these inst combines:
select C, B, 0 + select C, 0, A —> select C, B, A
A * (1 - uitofp i1 C) + B * uitofp i1 C —> select C, B, A
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@182499 91177308-0d34-0410-b5e6-96231b3b80d8
-rw-r--r-- | lib/Transforms/InstCombine/InstCombineAddSub.cpp | 43 | ||||
-rw-r--r-- | lib/Transforms/InstCombine/InstCombineMulDivRem.cpp | 29 | ||||
-rw-r--r-- | test/Transforms/InstCombine/add4.ll | 40 |
3 files changed, 56 insertions, 56 deletions
diff --git a/lib/Transforms/InstCombine/InstCombineAddSub.cpp b/lib/Transforms/InstCombine/InstCombineAddSub.cpp index 166f8df..a2c545f 100644 --- a/lib/Transforms/InstCombine/InstCombineAddSub.cpp +++ b/lib/Transforms/InstCombine/InstCombineAddSub.cpp @@ -1262,6 +1262,49 @@ Instruction *InstCombiner::visitFAdd(BinaryOperator &I) { } } + // A * (1 - uitofp i1 C) + B * (uitofp i1 C) -> select C, B, A + { + if (I.hasNoNaNs() && I.hasNoInfs() && I.hasNoSignedZeros()) { + Value *M1L, *M1R, *M2L, *M2R; + if (match(LHS, m_FMul(m_Value(M1L), m_Value(M1R))) && + match(RHS, m_FMul(m_Value(M2L), m_Value(M2R)))) { + + Value *A, *B, *C1, *C2; + if (!match(M1R, m_FSub(m_FPOne(), m_UIToFp(m_Value(C1))))) + std::swap(M1L, M1R); + if (!match(M2R, m_UIToFp(m_Value(C2)))) + std::swap(M2L, M2R); + + if (match(M1R, m_FSub(m_FPOne(), m_UIToFp(m_Value(C1)))) && + match(M2R, m_UIToFp(m_Value(C2))) && + C2->getType()->isIntegerTy(1) && + C1 == C2) { + A = M1L; + B = M2L; + return SelectInst::Create(C1, B, A); + } + + std::swap(M1L, M2L); + std::swap(M1R, M2R); + + if (!match(M1R, m_FSub(m_FPOne(), m_UIToFp(m_Value(C1))))) + std::swap(M1L, M1R); + if (!match(M2R, m_UIToFp(m_Value(C2)))) + std::swap(M2L, M2R); + + if (match(M1R, m_FSub(m_FPOne(), m_UIToFp(m_Value(C1)))) && + match(M2R, m_UIToFp(m_Value(C2))) && + C2->getType()->isIntegerTy(1) && + C1 == C2) { + A = M1L; + B = M2L; + return SelectInst::Create(C1, B, A); + } + } + } + } + + 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 51ca29b..2628f4b 100644 --- a/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp +++ b/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp @@ -524,35 +524,6 @@ 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))) && C->getType()->isIntegerTy(1)) { - 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)))) && - C->getType()->isIntegerTy(1)) { - A = LHS; - Value *Zero = ConstantFP::getNegativeZero(A->getType()); - return SelectInst::Create(C, Zero, A); - } - } - if (!isa<Constant>(Op1)) std::swap(Opnd0, Opnd1); else diff --git a/test/Transforms/InstCombine/add4.ll b/test/Transforms/InstCombine/add4.ll index 0fc0a6c..1047e16 100644 --- a/test/Transforms/InstCombine/add4.ll +++ b/test/Transforms/InstCombine/add4.ll @@ -6,30 +6,34 @@ target triple = "i686-apple-darwin8" define float @test1(float %A, float %B, i1 %C) { EntryBlock: - ;; A*(1 - uitofp i1 C) -> select C, 0, A + ;; select C, 0, B + select C, A, 0 -> select C, A, B %cf = uitofp i1 %C to float - %mc = fsub float 1.000000e+00, %cf - %p1 = fmul fast float %A, %mc - ret float %p1 + %s1 = select i1 %C, float 0.000000e+00, float %B + %s2 = select i1 %C, float %A, float 0.000000e+00 + %sum = fadd fast float %s1, %s2 + ret float %sum ; CHECK: @test1 -; CHECK: select i1 %C, float -0.000000e+00, float %A +; CHECK: select i1 %C, float %A, float %B } define float @test2(float %A, float %B, i1 %C) { EntryBlock: - ;; B*(uitofp i1 C) -> select C, B, 0 + ;; B*(uitofp i1 C) + A*(1 - uitofp i1 C) -> select C, A, B %cf = uitofp i1 %C to float + %mc = fsub fast float 1.000000e+00, %cf + %p1 = fmul fast float %A, %mc %p2 = fmul fast float %B, %cf - ret float %p2 + %s1 = fadd fast float %p2, %p1 + ret float %s1 ; CHECK: @test2 -; CHECK: select i1 %C, float %B, float -0.000000e+00 +; CHECK: select i1 %C, float %B, float %A } 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 + %mc = fsub fast float 1.000000e+00, %cf %p1 = fmul fast float %A, %mc %p2 = fmul fast float %B, %cf %s1 = fadd fast float %p1, %p2 @@ -38,21 +42,3 @@ EntryBlock: ; CHECK: select i1 %C, float %B, float %A } -; PR15952 -define float @test4(float %A, float %B, i32 %C) { - %cf = uitofp i32 %C to float - %mc = fsub float 1.000000e+00, %cf - %p1 = fmul fast float %A, %mc - ret float %p1 -; CHECK: @test4 -; CHECK: uitofp -} - -define float @test5(float %A, float %B, i32 %C) { - %cf = uitofp i32 %C to float - %p2 = fmul fast float %B, %cf - ret float %p2 -; CHECK: @test5 -; CHECK: uitofp -} - |