1*9880d681SAndroid Build Coastguard Worker; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s 2*9880d681SAndroid Build Coastguard Worker 3*9880d681SAndroid Build Coastguard Worker; When the extractelement is converted to a load the store can be re-used. 4*9880d681SAndroid Build Coastguard Worker; This will, however, introduce a cycle into the selection DAG (the load 5*9880d681SAndroid Build Coastguard Worker; of the extractelement index is dependent on the store, and so after the 6*9880d681SAndroid Build Coastguard Worker; conversion it becomes dependent on the new load, which is dependent on 7*9880d681SAndroid Build Coastguard Worker; the index). Make sure we skip the store, and conservatively instead 8*9880d681SAndroid Build Coastguard Worker; use a store to the stack. 9*9880d681SAndroid Build Coastguard Worker 10*9880d681SAndroid Build Coastguard Workerdefine float @foo(i32* %i, <4 x float>* %v) { 11*9880d681SAndroid Build Coastguard Worker; CHECK-LABEL: foo: 12*9880d681SAndroid Build Coastguard Worker; CHECK: movaps %xmm0, -[[OFFSET:[0-9]+]](%rsp) 13*9880d681SAndroid Build Coastguard Worker; CHECK: movss -[[OFFSET]](%rsp,{{.*}}), %xmm0 {{.*}} 14*9880d681SAndroid Build Coastguard Worker; CHECK-NEXT: retq 15*9880d681SAndroid Build Coastguard Worker %1 = load <4 x float>, <4 x float>* %v, align 16 16*9880d681SAndroid Build Coastguard Worker %mul = fmul <4 x float> %1, %1 17*9880d681SAndroid Build Coastguard Worker store <4 x float> %mul, <4 x float>* %v, align 16 18*9880d681SAndroid Build Coastguard Worker %2 = load i32, i32* %i, align 4 19*9880d681SAndroid Build Coastguard Worker %vecext = extractelement <4 x float> %mul, i32 %2 20*9880d681SAndroid Build Coastguard Worker ret float %vecext 21*9880d681SAndroid Build Coastguard Worker} 22