Re: LW compiler optimizations
"David McClain (as dbm at refined-audiometrics dot com)" <[email protected]>
| Newsgroups | gmane.lisp.lispworks.general |
|---|---|
| Message-ID | <[email protected]> |
er, sorry, I mis-spoke - 1.4 ns/MAC pair. Close to what you got. But still 3x slower than going through the FLI to vDSP. > On Jan 14, 2025, at 16:03, David McClain (as dbm at refined-audiometrics dot com) <[email protected]> wrote: > > Heh! Tim!! No need to get so hot. > > I admit I didn’t read about your macro at all. > > And I really do appreciate your contrary perspective, so please don’t leave. > > But the fact remains that I cannot attain your 1.1 ns timing. The closest I come is 6.7 ns/MAC pair. > > > >> On Jan 14, 2025, at 15:57, Tim Bradshaw <[email protected]> wrote: >> >> On 14 Jan 2025, at 22:21, David McClain <[email protected]> wrote: >>> >>> But this indicates that using a displaced array costs me nearly 5x performance degradation, as opposed to your claim of 1% penalty. >> >> I am sorry I am now going to lose my temper. >> >> LEARN TO FUCKING READ. >> >> I will repeat what I actually wrote which you have repeatedly misquoted: >> >> So let's summarize: the demonstrated performance cost of displaced arrays in an existing CL implementation, after a little macro, is *under one percent*. >> >> No, I know this is hard, but you need to read all of that sentence. All of it. In case that really is too hard, here are two important bits: >> >> "after a little macro": this is the macro which is described in that message, and which is included in some of the code I posted. It turns a displaced array into its underlying simple array and optionally provides a macro to reference that underlying array with a suitable offset. >> "in an existing CL implementation". Being, in this case, as I made clear, SBCL (although LW can be persuaded, as is demonstrated by the tsda/tsoa function in ts-dar.lisp. >> >> I know it's hard to read what people say, but really, it is not that hard. >> >> OK, I feel at this point the LW people will be fully justified from removing me from the list. That's fine. >> >> --tim >> >> >> >