Hi Armin:
________________________________
From: Armin Rigo <ar...@tunes.org>
To: PyPy Developer Mailing List <pypy-dev@python.org>
Sent: Thursday, February 16, 2012 3:58 PM
Subject: [pypy-dev] STM status
>An update for STM: today I managed to build a pypy using the new "stm
>gc". It runs richards.py on tannit:
>in 1 thread: 2320 ms per iteration
>in 2 threads: 1410 ms per iteration
>in 4 threads: 785 ms per iteration
>in 8 threads: 685 ms per iteration
>The small gap between 4 and 8 threads is due to tannit having only 4
>"real" cpus, each one hyperthreaded. The additional gain is thus
>smaller than expected.
This is really exciting stuff. As always I have a few questions.
1) I have been looking at the transaction module and its dependent modules.
In rstm.perform_transaction, I see a comment to a "custom GIL." So the GIL is
still there? Or will it be eventually removed?
2) I have glanced over the transaction and rstm module (it would help if I
understood PyPy's architecture better but I am working on that). I figured it
would be best to focus on translator.transform.py and translator.llstminterp.py
first to understand what is happening at a high level.
I admit I am not comfortable with reading the code. However I see places where
the code distinguishes between mutable and immutable, local and non local
variables. I will assume mutable and non-local variables will be the subject of
transactional memory.
Although the STM implementation will be transparent to the programmer (there is
only transaction.run() and transaction.add() available to the programmer), can
I assume that one can "help" the STM if a) somehow more variables can be marked
as immutable? 2)"Threads of execution" do not share state. Essentially a more
functional programming approach.
Or am I reading too much into this?
3) Could the stackless module work with STM?
Cheers,
Andrew
_______________________________________________
pypy-dev mailing list
pypy-dev@python.org
http://mail.python.org/mailman/listinfo/pypy-dev
