From: Sanjay Patel Date: Mon, 24 Jun 2019 22:35:26 +0000 (+0000) Subject: [InstCombine] squash is-not-power-of-2 using ctpop X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=1c6a395c06fe7ee95bfe28a06647d7028e6ef1e4;p=llvm [InstCombine] squash is-not-power-of-2 using ctpop This is the Demorgan'd 'not' of the pattern handled in: D63660 / rL364153 This is another intermediate IR step towards solving PR42314: https://bugs.llvm.org/show_bug.cgi?id=42314 We can test if a value is not a power-of-2 using ctpop(X) > 1, so combining that with an is-zero check of the input is the same as testing if not exactly 1 bit is set: (X == 0) || (ctpop(X) u> 1) --> ctpop(X) != 1 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@364246 91177308-0d34-0410-b5e6-96231b3b80d8 --- diff --git a/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp b/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp index 63f67416f1e..3a8ec1ecd38 100644 --- a/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp +++ b/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp @@ -1022,22 +1022,32 @@ static Value *foldSignedTruncationCheck(ICmpInst *ICmp0, ICmpInst *ICmp1, } /// Reduce a pair of compares that check if a value has exactly 1 bit set. -static Value *foldIsPowerOf2(ICmpInst *Cmp0, ICmpInst *Cmp1, +static Value *foldIsPowerOf2(ICmpInst *Cmp0, ICmpInst *Cmp1, bool JoinedByAnd, InstCombiner::BuilderTy &Builder) { - // Handle 'and' commutation: make the not-equal compare the first operand. - if (Cmp1->getPredicate() == ICmpInst::ICMP_NE) + // Handle 'and' / 'or' commutation: make the equality check the first operand. + if (JoinedByAnd && Cmp1->getPredicate() == ICmpInst::ICMP_NE) + std::swap(Cmp0, Cmp1); + else if (!JoinedByAnd && Cmp1->getPredicate() == ICmpInst::ICMP_EQ) std::swap(Cmp0, Cmp1); // (X != 0) && (ctpop(X) u< 2) --> ctpop(X) == 1 CmpInst::Predicate Pred0, Pred1; Value *X; - if (match(Cmp0, m_ICmp(Pred0, m_Value(X), m_ZeroInt())) && + if (JoinedByAnd && match(Cmp0, m_ICmp(Pred0, m_Value(X), m_ZeroInt())) && match(Cmp1, m_ICmp(Pred1, m_Intrinsic(m_Specific(X)), m_SpecificInt(2))) && Pred0 == ICmpInst::ICMP_NE && Pred1 == ICmpInst::ICMP_ULT) { Value *CtPop = Cmp1->getOperand(0); return Builder.CreateICmpEQ(CtPop, ConstantInt::get(CtPop->getType(), 1)); } + // (X == 0) || (ctpop(X) u> 1) --> ctpop(X) != 1 + if (!JoinedByAnd && match(Cmp0, m_ICmp(Pred0, m_Value(X), m_ZeroInt())) && + match(Cmp1, m_ICmp(Pred1, m_Intrinsic(m_Specific(X)), + m_SpecificInt(1))) && + Pred0 == ICmpInst::ICMP_EQ && Pred1 == ICmpInst::ICMP_UGT) { + Value *CtPop = Cmp1->getOperand(0); + return Builder.CreateICmpNE(CtPop, ConstantInt::get(CtPop->getType(), 1)); + } return nullptr; } @@ -1083,7 +1093,7 @@ Value *InstCombiner::foldAndOfICmps(ICmpInst *LHS, ICmpInst *RHS, if (Value *V = foldSignedTruncationCheck(LHS, RHS, CxtI, Builder)) return V; - if (Value *V = foldIsPowerOf2(LHS, RHS, Builder)) + if (Value *V = foldIsPowerOf2(LHS, RHS, true /* JoinedByAnd */, Builder)) return V; // This only handles icmp of constants: (icmp1 A, C1) & (icmp2 B, C2). @@ -2169,6 +2179,9 @@ Value *InstCombiner::foldOrOfICmps(ICmpInst *LHS, ICmpInst *RHS, if (Value *V = foldAndOrOfEqualityCmpsWithConstants(LHS, RHS, false, Builder)) return V; + if (Value *V = foldIsPowerOf2(LHS, RHS, false /* JoinedByAnd */, Builder)) + return V; + // This only handles icmp of constants: (icmp1 A, C1) | (icmp2 B, C2). if (!LHSC || !RHSC) return nullptr; diff --git a/test/Transforms/InstCombine/ispow2.ll b/test/Transforms/InstCombine/ispow2.ll index ba2867363e7..095e1260794 100644 --- a/test/Transforms/InstCombine/ispow2.ll +++ b/test/Transforms/InstCombine/ispow2.ll @@ -312,10 +312,8 @@ define i1 @is_pow2_ctpop_wrong_pred2(i32 %x) { define i1 @isnot_pow2_ctpop(i32 %x) { ; CHECK-LABEL: @isnot_pow2_ctpop( ; CHECK-NEXT: [[T0:%.*]] = tail call i32 @llvm.ctpop.i32(i32 [[X:%.*]]), !range !0 -; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i32 [[T0]], 1 -; CHECK-NEXT: [[ISZERO:%.*]] = icmp eq i32 [[X]], 0 -; CHECK-NEXT: [[R:%.*]] = or i1 [[ISZERO]], [[CMP]] -; CHECK-NEXT: ret i1 [[R]] +; CHECK-NEXT: [[TMP1:%.*]] = icmp ne i32 [[T0]], 1 +; CHECK-NEXT: ret i1 [[TMP1]] ; %t0 = tail call i32 @llvm.ctpop.i32(i32 %x) %cmp = icmp ugt i32 %t0, 1 @@ -333,8 +331,8 @@ define i1 @isnot_pow2_ctpop_extra_uses(i32 %x) { ; CHECK-NEXT: call void @use_i1(i1 [[CMP]]) ; CHECK-NEXT: [[ISZERO:%.*]] = icmp eq i32 [[X]], 0 ; CHECK-NEXT: call void @use_i1(i1 [[ISZERO]]) -; CHECK-NEXT: [[R:%.*]] = or i1 [[ISZERO]], [[CMP]] -; CHECK-NEXT: ret i1 [[R]] +; CHECK-NEXT: [[TMP1:%.*]] = icmp ne i32 [[T0]], 1 +; CHECK-NEXT: ret i1 [[TMP1]] ; %t0 = tail call i32 @llvm.ctpop.i32(i32 %x) %cmp = icmp ugt i32 %t0, 1 @@ -350,10 +348,8 @@ define i1 @isnot_pow2_ctpop_extra_uses(i32 %x) { define <2 x i1> @isnot_pow2_ctpop_commute_vec(<2 x i8> %x) { ; CHECK-LABEL: @isnot_pow2_ctpop_commute_vec( ; CHECK-NEXT: [[T0:%.*]] = tail call <2 x i8> @llvm.ctpop.v2i8(<2 x i8> [[X:%.*]]) -; CHECK-NEXT: [[CMP:%.*]] = icmp ugt <2 x i8> [[T0]], -; CHECK-NEXT: [[ISZERO:%.*]] = icmp eq <2 x i8> [[X]], zeroinitializer -; CHECK-NEXT: [[R:%.*]] = or <2 x i1> [[CMP]], [[ISZERO]] -; CHECK-NEXT: ret <2 x i1> [[R]] +; CHECK-NEXT: [[TMP1:%.*]] = icmp ne <2 x i8> [[T0]], +; CHECK-NEXT: ret <2 x i1> [[TMP1]] ; %t0 = tail call <2 x i8> @llvm.ctpop.v2i8(<2 x i8> %x) %cmp = icmp ugt <2 x i8> %t0,