Re: LW compiler optimizations
"David McClain (as dbm at refined-audiometrics dot com)" <[email protected]>
| Newsgroups | gmane.lisp.lispworks.general |
|---|---|
| Message-ID | <[email protected]> |
I think your D language is an indication that it is becoming more mainstream. One of the reasons, I suspect, for its being ignored for so long, is that the MMX architecture has different ways of handling SIMD floats, compared to IEEE standard handling. In particular, the treatment of denormal numbers. And so the MMX would produce results that differ in some cases, compared to scalar processing. For DSP you don’t care. When things go wrong, just drop it on the floor and keep running. Realtime is more important than point-wise detail accuracy. And DSP really does not like the slowdown imposed on denormal floats. Just round to zero and move along. For many years we had to constantly add in TPDF noise alongside IIR processes that could run into limit cycles or generate denormals. The MMX removed that problem for us in a big way. But it isn’t kosher IEEE 754. > On Jan 14, 2025, at 04:43, Yuri Davidovsky (as work at disclosure dot ie) <[email protected]> wrote: > > >> On 14 Jan 2025, at 12:24, David McClain <[email protected]> wrote: >> >> And so, I forgot to mention the other use case - Neural Networks, alongside cryptography, DSP, and others. MAC is so useful in so many areas. Today’s languages really ought to be making overt allowances for it. > > Can sign under each word in that sentence. Even if we ignore the DSP stuff as being somewhat niche, the rise of neural nets alone is a clear pointer where the needs of the language users to come are going to placed. > > It is somewhat surprising that this hasn’t been a thing for at least a decade, or even two, ever since MMX was introduced into Pentium processor in I forget which year. > > Was it 1934? > > > > _______________________________________________ > Lisp Hug - the mailing list for LispWorks users > [email protected] > http://www.lispworks.com/support/lisp-hug.html _______________________________________________ Lisp Hug - the mailing list for LispWorks users [email protected] http://www.lispworks.com/support/lisp-hug.html