Re: Why is Elisp slow?

[Ergus <spacibba@aol.com>]

On Sat, May 04, 2019 at 12:18:29AM +0200, ?scar Fuentes wrote:
>Ergus <spacibba@aol.com> writes:
>
>>>More importantly, the libJIT build failed to show any significant
>>>speed-up wrt byte code, so it sounds like maybe the whole idea was
>>>either wrong or its design couldn't possibly provide any gains.  Or
>>>maybe we just measured the speed-up in wrong scenarios.
>>>
>> This is not surprising. My work is 80% performance measurement and
>> benchmarking and the real improvements with JIT compilation (in my
>> experience) and specially with libJIT is not as good as many people
>> expect in most of the common scenarios. That's because the generated
>> code is usually very generic (so it does not take advantage of
>> architecture specific features), and strategies like vectorization and
>> branch prediction are very difficult to hint (most of the times
>> impossible). So the only real difference with a bytecode interpreter is
>> the bytecode parsing part, but not too much more.
>
>On Emacs case, I'm pretty sure whatever advantages comes from good
>architecture-specific code, accurate branch prediction, etc are below
>the noise level. As you pointed out below, Elisp is a dynamic language
>and for turning this
>
>(let ((acc 0))
>  (dotimes (i 10)
>    (setq acc (+ acc (foo i)))))
>
>into this
>
>int acc = 0;
>for(int i = 0; i < 10; ++i) {
>  acc += foo(i);
>}
>
>you need either sophisticated analysis (that, in practice, only works
>for the "easy" cases) or annotate the code with type declarations (and
>enforce then).
>
>Because otherwise handling variables as containers for generic values is
>incompatible with "C-like" performance.
>
>And an Elisp -> C translator does not magically solve this.
>
That's true. The code quality will be not very good (as what happen in
Android Java native compilers and Cython) but some of the optimizations
in the low level can be applied. That depends of the optimizations
implemented and the information provided to the compilers (both of
them).

A Lisp-C compiler for example can reduce significantly the function call
overheads and callback overheads cause thanks to the Lisp syntax it is
very easy to apply inline optimizations which in C represent a VERY
important improvement. In your example code foo, setq and operator + are
functions called in runtime, which interpreted means go to the symbol
hashtable, find the pointer to the function, interpret the inputs and
execute... compiling that... just think how it can change.

With a simple optimization like having the hardcoded 10 the compiler
will know the number of iterations to execute and the C compiler applies
many good optimizations in those cases.

SO it won't be the same than your second code, but performance could be
in the same order in many cases.