Re: LW compiler optimizations
"David McClain (as dbm at refined-audiometrics dot com)" <[email protected]>
| Newsgroups | gmane.lisp.lispworks.general |
|---|---|
| Message-ID | <[email protected]> |
Can you indicate the magic of getting such low allocation? I see puzzling amounts of allocation in all of my FIR dot product benchmark tests. Everything I construct has dynamic extent and yet there are megabytes being allocated for some reason. > On Jan 13, 2025, at 06:05, Paul Werkowski (as pw at snoopy dot qozzy dot com) <[email protected]> wrote: > > The attached chart shows bytes allocated vs FFT length (N) for several values of N. There are three curves. The blue curve is the result of correct optimization. The others, nearly identical, are the results for different methods of getting the W value for the butterfly (prior to optimization). > > I was not interested in any cpu time difference, if any, here. Eliminating the source of allocation can only improve that. > > Paul > > > On 1/13/2025 2:25 AM, Yuri Davidovsky wrote: >> >>> On 12 Jan 2025, at 20:24, Paul Werkowski (as pw at snoopy dot qozzy dot com) <[email protected]> wrote: >>> >>> I am embarrassed to say it was an error of omission of an important type declaration, among several others there, that was responsible for allocating 280 bytes per call where it should have been zero. Apparently, at least for me, it is hard to find that which is not there. >> So how much performance increase did you see from this change? > <Allocation.png> _______________________________________________ Lisp Hug - the mailing list for LispWorks users [email protected] http://www.lispworks.com/support/lisp-hug.html