How you can help improve Haskell implementations!
Just a quick note to say that the Haskell implementation shootout is progressing, now supporting jhc, fixing a range of bugs, and providing more benchmark programs. Nice average numbers are also reported for the relative performance of each compiler or interpreter. On x86: http://www.cse.unsw.edu.au/~dons/nobench/i686/results.html On amd64: http://www.cse.unsw.edu.au/~dons/nobench/x86_64/results.html Several bugs in our beloved Haskell compilers have been spotted, and most of them already fixed! ** I'd like to invite contributions of benchmark programs**. If you have a nice benchmarky program, that's reasonably portable (mostly h98 + base libs is ok), and freely licensed, let me know, and we can add it to the benchmark suite. Programs that show up interesting results between systems or that were difficult to get to perform well are particuarly welcome. 3 kinds of programs are welcome: * small artificial programs that illustrate particular issues e.g. (prime sieves, FFT, great language shootout programs) * 'kernels' from larger programs, that illustrate some functionality e.g. a mandelbrot generator, or some other core algorithm * larger real programs. These are particularly welcome. Limited to around 25 modules maximum, and capable of being run with simple stdin/args, producing simple output. Raytracers, chess programs, and small theorem provers are existing examples. So send me your unwashed code. Whatever code doesn't kill them will only make our compilers stronger! :-) The darcs repo for nobench/nofib is at: http://www.cse.unsw.edu.au/~dons/nobench.html Happy hacking! Don
Donald Bruce Stewart wrote:
Just a quick note to say that the Haskell implementation shootout is progressing, now supporting jhc, fixing a range of bugs, and providing more benchmark programs. Nice average numbers are also reported for the relative performance of each compiler or interpreter.
On x86: http://www.cse.unsw.edu.au/~dons/nobench/i686/results.html
On amd64: http://www.cse.unsw.edu.au/~dons/nobench/x86_64/results.html
Several bugs in our beloved Haskell compilers have been spotted, and most of them already fixed!
And also I guess the compilers will do more optimisations, etc. So this suggests an obvious extra feature for nobench which would be the ability to view a graph of each compiler's performance over a period of time, obviously this probably wouldn't be useful for at least a few months. However in particular it would be very good for the compiler writers to be aware if the performance goes *down* that is a further release of a compiler causes a benchmarked program to run slower. regards allan
Hi
And also I guess the compilers will do more optimisations, etc. So this suggests an obvious extra feature for nobench which would be the ability to view a graph of each compiler's performance over a period of time, obviously this probably wouldn't be useful for at least a few months.
We hope to add experimental -O to Yhc within the next week or so, very soon would help us :) Thanks Neil
ndmitchell:
Hi
And also I guess the compilers will do more optimisations, etc. So this suggests an obvious extra feature for nobench which would be the ability to view a graph of each compiler's performance over a period of time, obviously this probably wouldn't be useful for at least a few months.
We hope to add experimental -O to Yhc within the next week or so, very soon would help us :)
Its easy enough to run two or more versions of a compiler. So we can check ghc against ghc-head , or yhc against yhc-old. Just ping me if you want something checked like that. -- Don
participants (3)
-
Allan Clark -
dons＠cse.unsw.edu.au -
Neil Mitchell