On Jan 20, 2015, at 6:32 AM, Mike Stump <mikest...@comcast.net> wrote:
> On Jan 19, 2015, at 1:28 PM, Maxim Kuvyrkov <maxim.kuvyr...@linaro.org> wrote: >> >> Yes, the ordering relation is screwed, as you put it. With the number of >> independent heuristics that rank_for_schedule has to consider there is no >> other way then to have a "screwed" ordering. > > I have not seen a proof that an working comparison routine is impossible. I > can’t fathom why anyone would think that. Still perplexed. > >> I don't think we can afford sacrificing code quality to gain independency >> from number of elements being sorted. > > A working comparator has nothing to do with the number of elements, so, I > can’t fathom why you mention the number of elements. A working comparator > should, in theory improve code quality. However, even if it didn’t, that's > immaterial. We’d fix the bug irrespective of code quality, just to fix the > bug. Do you have a specific constructive suggestion how to structure haifa-sched.c:rank_for_schedule() so that it can never return a cyclic ordering like A < B < C < A? Thank you, -- Maxim Kuvyrkov www.linaro.org
