I'm searching somebody to compile ...

[Tennison]

Hi,

In my Toga project I have trouble to have good compiles.

And I can't test my compiles against original ones because too slow !!

I'm searching somebody to help me in my project by compiling differents versions.

[Michel]

Hi,

In my Toga project I have trouble to have good compiles.

And I can't test my compiles against original ones because too slow !!

I'm searching somebody to help me in my project by compiling differents versions.

Don't worry about compiles (except when releasing a final version).

Just test against previous compiles of the same type (e.g. -O2, no other options).

[Tennison]

Yes it was what I thought but when I compile an original version of Fruit or Toga I'm about 10 and 20% slower in nps.

And then results are bad.

For settings I put in GCC :

-O3
-fno-exceptions
-s
-w

But the problem is I want to tests against multiple opponents to see real elo improvment. Because self testing are only a way but not always a good one if the only one.

[Michel]

But the problem is I want to tests against multiple opponents to see real elo improvment. Because self testing are only a way but not always a good one if the only one.

I don't see the difference. Just use the original version and the new version with the same compile options....

[jd1]

Michel is right, and I do it that way too. Just run your base compile against the opponents, and compare development versions to that result.

Jerry

[Tennison]

It doesn't work.

Because Original Compiled Versions are far fatest than mine Original compiled.

And I want to compare actual versions with mine over differents opponents.

[hgm]

I also always do it as Michel suggests. No problem at all. If my own poor compile of a new version is 20 Elo better than the poor compile of the original version, the difference between the fast compiles will also be 20 Elo.

[Michel]

And I want to compare actual versions with mine over differents opponents.

It is simply not clear to me why you would want to do that. If you have 20 elo improvement in your compiled versions against a pool of engines you will also have 20 elo in good compiles.

The speed difference between compiles can be converted into elo with the rule that 1% is (approximately) 1elo.

But first you should make sure that you really have 20 elo.

[jd1]

The idea is that if your slow compile of your development version is better than your slow compile of the original, when you compare fast compiles of both, the elo gain will still be the same,

Two versions can always be compared as long as the compiles are the same.

Jerry

[Tennison]

For the nearly 20 elo improvment it's in very fast games.

I hope the same on longer games. And with other opponents.