Ruby Fiddle和全局解释器锁(GIL) [英] Ruby Fiddle and Global Interpreter Lock (GIL)

问题描述

我正在使用小提琴进行呼叫用于某些线性代数计算(Mac OSX).

I am using Fiddle to call an external C Function for some heavy linear algebra calculations (Mac OSX).

模块定义类似于

  module BLAS
    extend Fiddle::Importer
    dlload $BLAS_PATH
    extern 'void cblas_dgemm(int, int, int, int, int, int, double, double*, int, double*, int, double, double*, int)'
    extern 'void cblas_dgemv(int, int, int, int, double, double*, int, double*, int, double, double*, int)'
.......

这些库针对Mac osx进行了优化，可以使用所有可用的内核.不幸的是，当我从Ruby中使用Fiddle调用这些函数时，仅使用了一个内核，我的理解是这是由于GIL造成的.

The libraries have been optimized for mac osx and can use all the cores available. Unfortunately when I call these functions from Ruby with Fiddle only one core is used and my understanding is that this happens because of the GIL.

是否可以在执行外部功能期间释放锁?

Is there a way to release the lock during the execution of external functions?

我正在使用ruby 2.0.0-p247(不是系统Ruby).

I am using ruby 2.0.0-p247 (not the system Ruby).

推荐答案

在ruby 2.3中，这是默认设置(请参见

In ruby 2.3 this is the default (see the issue on the ruby bug tracker).

在此之前，它看起来像不是在Fiddle api中公开的.在ruby c API中有一个名为rb_thread_call_without_gvl的函数，其功能与在锡罐上所说的完全一样:释放gvl，调用传递的函数，重新获取gvl并返回.使用起来有点棘手-请务必阅读 thread.c中的注释.

Prior to that, it doesn't look like it was exposed in the Fiddle api. There is a function in the ruby c API called rb_thread_call_without_gvl that does pretty much exactly what it says on the tin: it releases the gvl, calls the passed function, reacquires the gvl and returns. It's a little tricky to use - be sure to read the comments in thread.c.

如果您无法升级，我怀疑您最好的选择是编写一个非常薄的C层，用rb_thread_call_without_gvl包装调用，然后使用小提琴来调用该包装器库.

If you can't upgrade, I suspect your best bet is to write a very thin C layer that wraps calls with rb_thread_call_without_gvl and then use fiddle to call that wrapper library.

这篇关于Ruby Fiddle和全局解释器锁(GIL)的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！