[Bug tree-optimization/59058] [4.7/4.8/4.9 Regression] wrong code at -O3 on x86_64-linux-gnu (affecting gcc 4.6 to trunk)

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59058

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
I improve this with also using max_stmt_executions I at least get no
vect.exp fail but the testcase in this PR is not vectorized when using
a size_t b.