- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi!
According to
http://software.intel.com/en-us/articles/intel-integrated-performance-primitives-intel-ipp-threading-openmp-faq
and the file "ThreadedFunctionsList.txt" included in IPP-5.3, IPP should support multithreading for the FFT functions on 32bit float arrays. On an Intel Core i7 CPU, I call "ippSetNumThreads(8)", and the function returns successfully (and "ippGetNumThreads" confirms that the number of threads has been set to 8). However, it takes exactly the same time to execute the transform with "ippSetNumThreads(1)" and "ippSetNumThreads(8)", and the Windows task manager constantly displays a CPU usage of 13% (i.e., one core). I tested with different array sizes between 2^8 and 2^24.
Are there additional restrictions for the use of multithreading with IPP, or is it expected to work "out of the box"?
Thanks & kind regards,
Markus
According to
http://software.intel.com/en-us/articles/intel-integrated-performance-primitives-intel-ipp-threading-openmp-faq
and the file "ThreadedFunctionsList.txt" included in IPP-5.3, IPP should support multithreading for the FFT functions on 32bit float arrays. On an Intel Core i7 CPU, I call "ippSetNumThreads(8)", and the function returns successfully (and "ippGetNumThreads" confirms that the number of threads has been set to 8). However, it takes exactly the same time to execute the transform with "ippSetNumThreads(1)" and "ippSetNumThreads(8)", and the Windows task manager constantly displays a CPU usage of 13% (i.e., one core). I tested with different array sizes between 2^8 and 2^24.
Are there additional restrictions for the use of multithreading with IPP, or is it expected to work "out of the box"?
Thanks & kind regards,
Markus
Link Copied
1 Reply
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Questions specific to IPP would be more likely to get answers on th IPP forum.
In general, applications which use floating point intensively can't be expected to gain by multiple threads per core.
In general, applications which use floating point intensively can't be expected to gain by multiple threads per core.
Reply
Topic Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page