[pypy-dev] JIT issues on x86_64
sh at lutzhaase.com
Sat Dec 12 20:35:12 CET 2009
On 12.12.2009 20:25, Leonardo Santagada wrote:
> On Dec 12, 2009, at 5:03 PM, Sven-Hendrik Haase wrote:
>> I have issues translating JIT on x86_64 architecture as opposed to i686.
>> I know this issue from Psyco which was never meant to work on other
>> architectures due to low-level optimizations. However, I think it was
>> one of PyPy's goals to make the JIT available to more architecture
>> (wasn't it?). Should I be able to successfully translate the JIT at this
>> point or is it supposed to break due to missing implementation?
> There isn't a JIT backend for x86_64 yet, so there is no way to get a PyPy with JIT in 64 bit right now, unfortunately. But you are right one of the goals is to make it available in more architectures, and that is already made simpler by separating the backend code of the jit from the rest of the implementation. No one is currently working on it as the focus for the next release is getting the jit solid and fast on 32 bit machines.
> If you want to port is to x86_64 PyPy people are probably going to help you to do it, but if you are not you can follow the developments on the blog because it is going to be announced there when it is at least ready for testing :)
>  http://morepypy.blogspot.com/
> ps: Psyco doesn't actually have anything prohibiting it to run on 64bit it is just way more difficult to work with it than with PyPy.
> ps2: A better version of this answer must be in the site somewhere, as more and more people will be asking this when anything about the JIT is posted on the blog... maybe a faq that gets linked to in every post.
> Leonardo Santagada
> santagada at gmail.com
Thanks for the answers (also Jan). I haven't looked at it at all yet but
I imagine it will involve a lot of assembly or very low-level C?
And having this in the FAQ till then would be a good idea I imagine.
Also have the translator fail on different architectures for JIT with a
clear error that it isn't implemented yet.
More information about the Pypy-dev