c++ - Mac 上的超线程——没有提供太多加速

Question

我已经使用 openMP 在 C++ 中实现了一个凸包算法。

代码可以在这里找到:http://codepad.org/VVQdSdfM

Below are the results when tested in my Mac Book Pro:
      Processor Name:   Intel Core i5
      Processor Speed:  2.5 GHz
      Number of Processors: 1
      Total Number of Cores:    2
      L2 Cache (per Core):  256 KB
      L3 Cache: 3 MB
      Memory:   4 GB

处理器运行代码所需的时间:

With two Threads:
(here size represents the number of points in the input and time in Seconds)

Average Sequential Time Elapsed in seconds for size:10=8.29697e-06
Average Parallel Time Elapsed in seconds for size:10=5.0807e-05

Average Sequential Time Elapsed in seconds for size:100=5.18084e-05
Average Parallel Time Elapsed in seconds for size:100=8.13007e-05

Average Sequential Time Elapsed in seconds for size:1000=0.000471377
Average Parallel Time Elapsed in seconds for size:1000=0.000283003

Average Sequential Time Elapsed in seconds for size:10000=0.00483506
Average Parallel Time Elapsed in seconds for size:10000=0.0032198

Average Sequential Time Elapsed in seconds for size:100000=0.0471328
Average Parallel Time Elapsed in seconds for size:100000=0.0333489

Average Sequential Time Elapsed in seconds for size:1000000=0.460131
Average Parallel Time Elapsed in seconds for size:1000000=0.267305


With four threads:

Average Sequential Time Elapsed in seconds for size:10=1.00136e-05
Average Parallel Time Elapsed in seconds for size:10=0.000106597

Average Sequential Time Elapsed in seconds for size:100=5.91993e-05
Average Parallel Time Elapsed in seconds for size:100=0.000114727

Average Sequential Time Elapsed in seconds for size:1000=0.000503755
Average Parallel Time Elapsed in seconds for size:1000=0.000302839

Average Sequential Time Elapsed in seconds for size:10000=0.00478158
Average Parallel Time Elapsed in seconds for size:10000=0.00235724

Average Sequential Time Elapsed in seconds for size:100000=0.0465738
Average Parallel Time Elapsed in seconds for size:100000=0.0223478

Average Sequential Time Elapsed in seconds for size:1000000=0.466074
Average Parallel Time Elapsed in seconds for size:1000000=0.221905

我在我的 CPU 事件监视器中找到了四个插槽，我开始知道这个版本的英特尔处理器支持超线程。

如果是这样的话，我不应该在使用 4 个线程时获得 4 的加速吗？

请提供任何可以帮助我使用英特尔处理器中的超线程功能的指示。

谢谢，维杰

Answer 1

在 HPC(高性能计算)上下文中使用超线程时，您不应期望性能有太大提高。事实上，您最好在 BIOS 级别关闭超线程。超线程给人以额外内核的印象，如果许多进程在同一个 CPU 上运行，可能会提高性能，但不会为 CPU 密集型 MPI 应用程序增加值(value)。