Re: Sequence test

new topic     » goto parent     » topic index » view thread      » older message » newer message
RStowasser said...

Forked from Re: Edita on 4.0a3

petelomax said...

Actually, my tests show it 13,800 times slower (0.03 vs 414 seconds), probably an exponential slowdown due to messed up refcounts. It has nothing to do with the new and bizarre include system, try this:

atom t 
    t = time() 
 
sequence text 
integer lineno, count 
 
function Clone(sequence intxt) 
 
 
George 
integer lt 
    lt = length(intxt) 
    return intxt[1..lt] 
end function 
 
    text = {repeat("1234",60000)} 
    count = 0 
    lineno = 0 
    while 1 do 
	if lineno>=length(text[1]) then exit end if 
	lineno += 1 
	text[1][lineno] = Clone(text[1][lineno]) 
	count += 1 
	if count>2000 then 
	    printf(1,"unpacking %d\n",{lineno}) 
	    count = 0 
	end if 
    end while 
 
    printf(1,"done, t=%3.2f\n",time()-t) 

I tried this code in Windows (Vista) with exwc of rev 1351 and 1352.
My result for 1351 is t=0.13 and for 1352 t=155.99. There must be a change in rev 1352 (introduces inline) which creates this strange behaviour. This must not necessarily be because of inlining.
If I add: without inline in the code and run it with rev 3025 I get the same result with or without this line: t=194.5.
Btw if I translate with euc -con the result is: t=0.09

Roland

I'm still running 2.4 and t is .04 running interpretive.

George

new topic     » goto parent     » topic index » view thread      » older message » newer message

Search



Quick Links

User menu

Not signed in.

Misc Menu