diff --git a/lib/Transforms/Vectorize/BBVectorize.cpp b/lib/Transforms/Vectorize/BBVectorize.cpp index ba033183758..290d82dc77a 100644 --- a/lib/Transforms/Vectorize/BBVectorize.cpp +++ b/lib/Transforms/Vectorize/BBVectorize.cpp @@ -858,16 +858,33 @@ namespace { std::vector &PairableInsts, std::multimap &ConnectedPairs, ValuePair P) { + StoreInst *SI, *SJ; + // For each possible pairing for this variable, look at the uses of // the first value... for (Value::use_iterator I = P.first->use_begin(), E = P.first->use_end(); I != E; ++I) { + if (isa(*I)) { + // A pair cannot be connected to a load because the load only takes one + // operand (the address) and it is a scalar even after vectorization. + continue; + } else if ((SI = dyn_cast(*I)) && + P.first == SI->getPointerOperand()) { + // Similarly, a pair cannot be connected to a store through its + // pointer operand. + continue; + } + VPIteratorPair IPairRange = CandidatePairs.equal_range(*I); // For each use of the first variable, look for uses of the second // variable... for (Value::use_iterator J = P.second->use_begin(), E2 = P.second->use_end(); J != E2; ++J) { + if ((SJ = dyn_cast(*J)) && + P.second == SJ->getPointerOperand()) + continue; + VPIteratorPair JPairRange = CandidatePairs.equal_range(*J); // Look for : @@ -883,6 +900,10 @@ namespace { // Look for cases where just the first value in the pair is used by // both members of another pair (splatting). for (Value::use_iterator J = P.first->use_begin(); J != E; ++J) { + if ((SJ = dyn_cast(*J)) && + P.first == SJ->getPointerOperand()) + continue; + if (isSecondInIteratorPair(*J, IPairRange)) ConnectedPairs.insert(VPPair(P, ValuePair(*I, *J))); } @@ -893,9 +914,19 @@ namespace { // both members of another pair (splatting). for (Value::use_iterator I = P.second->use_begin(), E = P.second->use_end(); I != E; ++I) { + if (isa(*I)) + continue; + else if ((SI = dyn_cast(*I)) && + P.second == SI->getPointerOperand()) + continue; + VPIteratorPair IPairRange = CandidatePairs.equal_range(*I); for (Value::use_iterator J = P.second->use_begin(); J != E; ++J) { + if ((SJ = dyn_cast(*J)) && + P.second == SJ->getPointerOperand()) + continue; + if (isSecondInIteratorPair(*J, IPairRange)) ConnectedPairs.insert(VPPair(P, ValuePair(*I, *J))); } diff --git a/test/Transforms/BBVectorize/no-ldstr-conn.ll b/test/Transforms/BBVectorize/no-ldstr-conn.ll new file mode 100644 index 00000000000..ada2a71e36e --- /dev/null +++ b/test/Transforms/BBVectorize/no-ldstr-conn.ll @@ -0,0 +1,23 @@ +target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128" +; RUN: opt < %s -bb-vectorize -bb-vectorize-req-chain-depth=2 -instcombine -gvn -S | FileCheck %s + +; Make sure that things (specifically getelementptr) are not connected to loads +; and stores via the address operand (which would be bad because the address +; is really a scalar even after vectorization) +define i64 @test2(i64 %a) nounwind uwtable readonly { +entry: + %a1 = inttoptr i64 %a to i64* + %a2 = getelementptr i64* %a1, i64 1 + %a3 = getelementptr i64* %a1, i64 2 + %v2 = load i64* %a2, align 8 + %v3 = load i64* %a3, align 8 + %v2a = add i64 %v2, 5 + %v3a = add i64 %v3, 7 + store i64 %v2a, i64* %a2, align 8 + store i64 %v3a, i64* %a3, align 8 + %r = add i64 %v2, %v3 + ret i64 %r +; CHECK: @test2 +; CHECK-NOT: getelementptr <2 x i64*> +} +