Re: Why are intepreters so slow today
Larry Wall (lwall@netlabs.com)
Sun, 17 Apr 1994 09:38:36 GMT
In article <766522873@sheol.UUCP> throopw%sheol@concert.net (Wayne Throop) writes:
: : From: lwall@netlabs.com (Larry Wall)
: : Everything is still done in floating point, however. Ask me
: : again after I've implemented "pragma integer". :-)
:
: Larry's comment is ensmilied, but seriously: my investigations into
: interpreter performance bottlenecks indicate that pragma integer won't
: get you much at all if the basic expression overhead is still 1/30 or
: so. Out of that 30-times-slower, the actual float-that-could-be-int
: calculations are (I estimate) maybe 2 or 3 of the 30 times.
That's about right. It turned out to save about 10%. By the way, instead
of "pragma integer" it turned out to be "hint integer". Ah, well...
: Of course, if in reacting to "pragma integer" Larry meant that he'd use
: a completely different, leaner, meaner, embedded expression interpreter
: for *all* operations, including looping and such... well then, as Emily
: Litella said..... "Nevermind".
For my next trick I will add
hint optimize_the_heck_out_of_this_no_matter_how_long_it_takes;
:-)
Larry