WP-34S N-Queens Benchmark Based on Patrice Torchet's BITMAP Solution

[Claudio L.]

(10-21-2014 10:51 PM)patrice Wrote:  Take the HP-15C: The single indirection swap in the inner loop imply a mostly 33% overhead. For you, I don't know, but for me, it is NOT MINOR.

30% seems a lot for anybody. When I watched your presentation, I understood that 30% was also including all your other optimizations. But each model will have a different impact. I'm with you, if the difference is 30%, it's not minor by any means. But on other models I don't think that it will be a big impact. If I remember correctly for the C/Regvars, the change in speed was less than 10%, and that included disabling all compiler optimizations, not just that indirection.

(10-21-2014 10:51 PM)patrice Wrote:  It just append that a benchmark IS a competition about being the fastest

What I meant is the goals are different. The objective of the benchmark is not to compete, but to measure and compare. People try to compete (naturally), but that's not what the benchmark is about.
On the benchmark you try to eliminate the user factor (to compare machines). On the competition the user factor is all that matters (it's who does it faster).
For newRPL, for example, I'm not trying to "beat" any particular measures or compete against anything or anybody, yet I'm using this same benchmark just to see where it falls in the "market".

(10-21-2014 10:51 PM)patrice Wrote:  Here is an optimized code for the 15C: the algorithm is exactly the same as in the benchmark. It simply use available features more efficiently.

it saves about 35% of runtime. It don't think it is minor.

And that one should go on the list, as long as it represents what a user would do to solve that problem with the machine and as long as it's still the same algorithm. If your optimizations are too obscure (like the bitmap method) they are not representative of normal use, but using "decrease and skip if not zero" style instructions is OK in my opinion (it's common sense).

Claudio