reenable the automatic detection of when to don't load the full kernel for the gpu conv conv_patch_stack_reduce. This is just a speed up reenabled.