Re: Performance assistance / advice

This discussion, and today’s article on D-Wave in Ars Technica, reminds me…

https://arstechnica.com/science/2025/03/d-wave-quantum-annealers-solve-problems-classical-algorithms-struggle-with/
D-Wave quantum annealers solve problems classical algorithms struggle with
arstechnica.com


I once had a computer that was a SIMD engine with 4096 processors, arranged in a 64x64 pyramidal interconnect. Special purpose computers are much better attuned to certain problems than Von Neumann serial machines (no matter however many cores). Our problem was implementing a target detection system - a virtual retina, if you will. D-Wave solves the Ising Model.

But coupled to all these basic problems is a tower of infrastructure with non-numerical needs - files, queues, lists, GUI’s, etc, etc. And that variety of infrastructure is where Lisp shines above them all. If you really have a hard problem to solve, then choose the best architecture you can, and couple it to a Lisp orchestrator.

I believe Scheme was an outgrowth in the late 80s for the Connection Machine??

- DM



> On Mar 12, 2025, at 09:15, Yuri Davidovsky (as work at disclosure dot ie) <[email protected]> wrote:
> 
> 
> 
>> On 12 Mar 2025, at 16:51, Tim Bradshaw <[email protected]> wrote:
>> 
>> Perhaps there are 100 full-time-equivalents working on Lisp systems world-wide: I expect that is rather generous.  That's why Lisp compilers produce code which is perhaps not quite as fast that produced by C-family compilers.
> 
> That is exactly the reason. There are a lot more man hours going into the C compilers than into the Lisp compilers so Lisp is always going to play catch-up. Personally, it does not bother me, I would use Lisp for business logic and if there are computational bottlenecks, I’d resort to writing asm helpers. 
> 
> It is macros and repl that have kept me coming back to lisp over the years, not its performance. In fact, as been mentioned, people routinely write heavy stuff even in Python, by leveraging foreign interface to offload computations. It is pretty much the standard in machine learning these days and that stuff is as heavy to compute as it gets.
> 
> With that said, there are things like llvm that may be able to help close the gap on the compiler front, it is a bit surprising that there haven’t been any attempts yet. Maybe that is not really something that majority of Lisp people care about at the end of the day.
> 
> _______________________________________________
> Lisp Hug - the mailing list for LispWorks users
> [email protected]
> http://www.lispworks.com/support/lisp-hug.html