[Bug tree-optimization/93055] accumulation loops in stepanov_vector benchmark use more instruction level parpallelism

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93055

--- Comment #5 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Created attachment 47563
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=47563&action=edit
fixed testcase

I have verified that building with 

g++ -O3 -march=bdver1 -fno-prefetch-loop-arrays ~/stepanov_vector.ii

gets me the lea into internal loop. This does not happen with gcc9.