If you're on the JVM, and if you're compiling J down to JVM bytecode (much as April compiles APL -> CL -> machine code), and if your data is stored in primitive Java arrays, then you'd get pretty close to C speeds. Or at least J speeds (where J is implemented in C).

That's a lot of conditions, but it's not completely crazy. :)

Maybe. Java is quite fast, but for very different reasons than APL; I don't think it would transfer that easily. Part of the reason APLs are so fast is because the interpreter and all the code fit in cache; my copy of k is 76kb. I don't think there's a chance you could get that with the jvm. APL also relies heavily on SIMD, and I don't know how well-supported that is on the JVM.

Granted, those are fair points. My copy of J is 42KB, so I understand where you're coming from. :)

There's some support for vectorization in the JVM, e.g. [1, 2], but I suspect it's hit-or-miss.

[1] https://bugs.java.com/bugdatabase/view_bug.do?bug_id=6340864

[2] https://bugs.java.com/bugdatabase/view_bug.do?bug_id=7192383

Well - Maybe you're right. I hope to find out some day