Re: register-compiled-module performance
On Sat, Jan 11, 2020 at 8:47 AM megane wrote: > > Matt Welland writes: > > > With Chicken 4 I'm seeing a pretty high impact on startup time from > > (declare (uses foo.import)). I think that will be less of a problem when > I > > switch from using * for my export lists to only the procedures that need > to > > be exported. However I was very surprised to see readline in the top > three > > when I ran a profile. I'm thinking somehow I've created a genuine problem > > but I've no idea what to look for. > > > > Suggestions? > > I'd guess this is from the O(n^2) behaviour of merge-se in modules.scm. > This has been fixed in C5. > Ok, that is good to know. I'll not worry about startup time for now and will put the energy into getting to chicken 5. Thanks. > Here's a version for C4 that I used to use: > > (define (merge-se . ses*) ; later occurrences take precedence to earlier > ones > (let ([table (make-hash-table)] > [rses (reverse ses*)]) > (let loop ([ses (cdr rses)] >[last-se #f] >[se2 (car rses)]) > (cond >[(null? ses) se2] >[(eq? last-se (car ses)) > (loop (cdr ses) last-se se2)] >[(null? (car ses)) > (loop (cdr ses) last-se se2)] >[(not last-se) > (for-each (lambda (e) (hash-table-set! table (car e) #t)) se2) > (loop ses se2 se2)] >[else > (let lp ([se (car ses)] > [se2 se2]) > (cond >[(null? se) (loop (cdr ses) (car ses) se2)] >[(hash-table-exists? table (caar se)) > (lp (cdr se) se2)] >[else > (hash-table-set! table (caar se) #t) > (lp (cdr se) (cons (car se) se2))]))] > > > > Note - migration is a work in progress but we are not quite at a point > > where I can test this in Chicken 5. > > > > procedure calls > > seconds average percent > > > --- > > 29 > > 1.8800.064 12.384 > > megamod.import.scm:71: ##sys#register-compiled-module 1 > > 1.6601.660 10.935 > > readline.scm:74: ##sys#register-compiled-module1 > > 1.5801.580 10.408 > > runsmod.import.scm:43: ##sys#register-compiled-module 1 > > 1.3501.3508.893 > > -- -- Complexity is your enemy. Any fool can make something complicated. It is hard to keep things simple. - Richard Branson.
Re: register-compiled-module performance
Matt Welland writes: > With Chicken 4 I'm seeing a pretty high impact on startup time from > (declare (uses foo.import)). I think that will be less of a problem when I > switch from using * for my export lists to only the procedures that need to > be exported. However I was very surprised to see readline in the top three > when I ran a profile. I'm thinking somehow I've created a genuine problem > but I've no idea what to look for. > > Suggestions? I'd guess this is from the O(n^2) behaviour of merge-se in modules.scm. This has been fixed in C5. Here's a version for C4 that I used to use: (define (merge-se . ses*) ; later occurrences take precedence to earlier ones (let ([table (make-hash-table)] [rses (reverse ses*)]) (let loop ([ses (cdr rses)] [last-se #f] [se2 (car rses)]) (cond [(null? ses) se2] [(eq? last-se (car ses)) (loop (cdr ses) last-se se2)] [(null? (car ses)) (loop (cdr ses) last-se se2)] [(not last-se) (for-each (lambda (e) (hash-table-set! table (car e) #t)) se2) (loop ses se2 se2)] [else (let lp ([se (car ses)] [se2 se2]) (cond [(null? se) (loop (cdr ses) (car ses) se2)] [(hash-table-exists? table (caar se)) (lp (cdr se) se2)] [else (hash-table-set! table (caar se) #t) (lp (cdr se) (cons (car se) se2))]))] > > Note - migration is a work in progress but we are not quite at a point > where I can test this in Chicken 5. > > procedure calls > seconds average percent > --- > 29 > 1.8800.064 12.384 > megamod.import.scm:71: ##sys#register-compiled-module 1 > 1.6601.660 10.935 > readline.scm:74: ##sys#register-compiled-module1 > 1.5801.580 10.408 > runsmod.import.scm:43: ##sys#register-compiled-module 1 > 1.3501.3508.893
register-compiled-module performance
With Chicken 4 I'm seeing a pretty high impact on startup time from (declare (uses foo.import)). I think that will be less of a problem when I switch from using * for my export lists to only the procedures that need to be exported. However I was very surprised to see readline in the top three when I ran a profile. I'm thinking somehow I've created a genuine problem but I've no idea what to look for. Suggestions? Note - migration is a work in progress but we are not quite at a point where I can test this in Chicken 5. procedure calls seconds average percent --- 29 1.8800.064 12.384 megamod.import.scm:71: ##sys#register-compiled-module 1 1.6601.660 10.935 readline.scm:74: ##sys#register-compiled-module1 1.5801.580 10.408 runsmod.import.scm:43: ##sys#register-compiled-module 1 1.3501.3508.893 -- Complexity is your enemy. Any fool can make something complicated. It is hard to keep things simple. - Richard Branson.
