- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
We have a signal processing application built & running well under x86_em64t using IPP & ICC. I would like to port and test this application to take advantage of AVX. What is the build environment needed (OS, Compiler & IPP version)?
Thanks,
Hari Tadepalli
Performance Engineer
ECG/Intel/Chandler
Link Copied
2 Replies
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Hari,
As it is a Signal Processing app, i assume it is using floating point data and is a good candidate for AVX. I wrote many signal processing algorithm using AVX.
Intel compiler 11.1+ is good. Win7 (or linux) will be right OS. As Win7 is not available with AVX support SDE canhelp you emulate the code.Latest IPP has many kernels ported to AVX.
One question, are you going to write AVX? or are you relying on compiler to generate AVX? Compiler can auto vectorize the code but it depends on the algorithms too.
As it is a Signal Processing app, i assume it is using floating point data and is a good candidate for AVX. I wrote many signal processing algorithm using AVX.
Intel compiler 11.1+ is good. Win7 (or linux) will be right OS. As Win7 is not available with AVX support SDE canhelp you emulate the code.Latest IPP has many kernels ported to AVX.
One question, are you going to write AVX? or are you relying on compiler to generate AVX? Compiler can auto vectorize the code but it depends on the algorithms too.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
AVX support on Win7 was scheduled for a service pack 17 months or more away, last we were informed. Fedora 12 and 13 currently support AVX, and RHEL6 is nearing release. In order to compile and emulate with SDE, as Brijender said, you don't need an OS which supports AVX.
Performance libraries such as IPP and MKL carry a greater advantage over compiled code with AVX. Optimized settings of affinity and number of threads carry increased importance.
Performance libraries such as IPP and MKL carry a greater advantage over compiled code with AVX. Optimized settings of affinity and number of threads carry increased importance.

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