[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Emacs Lisp JIT Compiler

From: Tom Tromey
Subject: Re: Emacs Lisp JIT Compiler
Date: Wed, 12 Sep 2018 22:32:06 -0600
User-agent: Gnus/5.13 (Gnus v5.13) Emacs/26.1.50 (gnu/linux)

>> Yes, el-compilador is pretty unfinished.  I think it should support
>> all of elisp in the front end and in the optimizers; but the C back end
>> is quite incomplete.

Robert> OK. Iʼm assuming completing the C back end is just a SMP? :-)

Yeah, for some value of S.  The github repo has some bugs if you're
interested.  Also you can pretty much pick a random, real elisp program,
compile it, and find failures, usually in the C back end.  I've got an
Emacs branch where I tried rewriting bits of C to elisp and then compile
them back to C; and IIRC I had to skip some functions for this reason.

Not to talk down el-compilador.  It has some nice things that are more
of a pain to do in something like the bytecode->C compiler.  For example
it can apply type inference along one side of a conditional.

>> I wrote a second compiler, though, that just compiles from bytecode to
>> C.  This one is more complete, see my post about it from sometime in the
>> last month or so.  IIRC wiring it up to the build was the remaining
>> difficulty.

Robert> I saw that, but I donʼt remember if you said where youʼd put it. I
Robert> couldnʼt see a likely branch in the savannah emacs git repository.

It's in my github in the feature/bytecode-c branch.

The C generation is ok but the rest isn't finished.  It is tricky to
associate the compiled C code with the Lisp defun.  Also see my other
post about it, it doubles (IIRC) the size of the executable.

You may recall that I was also looking at gcc-jit.  I have partially
done this translation but I got stalled by my illness.  I'm not sure if
I'll return to it, mostly because libjit is sounding perhaps more
attractive here on emacs-devel.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]