- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Is there a code example that shows significant speed up just by switching to a processor that supports AVX? IPP 7.0 supports the new AVX, but I am unable to show much improvement. In one of my applications the code that does an FFT/Matrix Mul/IFFT (using the IPPI library) gets about 10% speed up. I have also tried simply using the ippMul functions of IPP in a simple test, but haven't been able to show any significant speed up. Even with using the ippCpuInit dispatcher function and telling it to run with ippCpuAVX vs ippCpuSSE42 doesn't really show any different.
Is there any samples that cans how greater than 25%?
Thanks!
Link Copied
5 Replies
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello,
we were focused on AVX optimization for 64-bit OS, so if you run test on 32-bit there is not that much functions optimized in IPP 7.0 beta.
Regards,
Vladimir
we were focused on AVX optimization for 64-bit OS, so if you run test on 32-bit there is not that much functions optimized in IPP 7.0 beta.
Regards,
Vladimir
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Vladimir,
Thank you for your reply, but I didn't mention running on a 32-bit operating system. I am using a 64-bit version of Linux running on sandy bridge. My intel sponsor and I are just looking for some example code that really takes advantage of AVX through IPP. We would like to show improvements between runs when the dispatcher is allowed to use AVX optimizations and when it is not.
Thank you.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Got it. I would recommend you to check performance of FFT functions in signal processing. For these functions you should seegood performance gain over SSE4.2 optimized code.
The simpliest way actually is to run IPP performance system on your side, on your system and compare performance results. For example, when you run ps_ipps.exe with the following parameters:
-B-r -s -TAES -N1
and
-B -r -s -TAVX -N1
you will be able to get performance data of all signal processing functions in single thread mode with AVX optimization (in the first case) and with SSE4.2 optimization (in thesecond case)
Regards,
Vladimir
The simpliest way actually is to run IPP performance system on your side, on your system and compare performance results. For example, when you run ps_ipps.exe with the following parameters:
-B-r -s -TAES -N1
and
-B -r -s -TAVX -N1
you will be able to get performance data of all signal processing functions in single thread mode with AVX optimization (in the first case) and with SSE4.2 optimization (in thesecond case)
Regards,
Vladimir
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Vladimir,
We aren't working with windows. We are using a 64 bit version of Linux. Does that sample program still exist for that platform. If I remember correctly there is only a makefile for the 32 bit version of that fft sample in IPP 7.0 beta.
Thanks,
Derek
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Derek,
precompiled performance system executableshould also provided for Linux,could you please check if your installation have\tools\perfsys folder?
Regards,
Vladimir
precompiled performance system executableshould also provided for Linux,could you please check if your installation have
Regards,
Vladimir
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