Neither autovectorization-fail nor Cuda executing kernels in serial is a "fail-fail", both are fall-back actions that accomplish a given task in a bit longer th

Neither autovectorization-fail nor Cuda executing kernels in serial is a "fail-fail", both are fall-back actions that accomplish a given task in a bit longer than the explicit instructions imply. That said, executing kernels in serial can supposedly create problems if a programmer creates logic that assumes kernels are always moving in lock-step.