Bend's codegen is still abysmal, but these are all low-hanging fruits. Most of the work went into making the parallel evaluator correct (which is extremely hard!). I know that sounds "trust me", but the single-thread performance will get much better once we start compiling procedures, generating loops, etc. It just hasn't been done.
(I wonder if I should have waited a little bit more before actually posting it)
To illustrate that, many people on here have been losing their mind over Kolomogorov-Arnold Networks, which are almost identically positioned; interesting idea, kind of cool, does what the paper claims, potentially useful in the future, definitely not any use at all for any real use-case right now.
(In part that's probably because the average understanding of ML here is _not_ strong, so there's more deference and credulousness around those claims.)
No. You built something that’s pretty cool. It’s not done yet, but you’ve accomplished a lot! I’m glad you posted it. Thank you. Ignore the noise and keep cooking!
I've never understood the fascination with tail calls and recursion among computer science folks. Just write a loop, it's what it optimises to anyway.
Good point about the signed integer overflow, though!
(though note that I mention jemalloc because I've had "huh, this C code now magically runs faster" experiences with it, make no claim it's the right one to look at, and am very sure that I don't know what I'm talking about sufficiently wrt allocators to be able to recognise the right one if it bit me on the leg)
I just wrote a simple loop in C++ to sum up 0 to 2^30. With a single thread without any optimizations it runs in 1.7s on my laptop -- matching Bend's performance on an RTX 4090! With -O3 it vectorizes the loop to run in less than 80ms.