IPP 6.1 Performance Tests for Linux* segfault on SNB

[Christopher_Dietsch]

I am running ./ps_ippiem64t on SNB architecture to get some FLOP #s out of it but am getting a segfault, at least one after ippiResize_32f_C1R. I'm running on other machines, but it appears it correctly sees the architecture as having AVX support when the program starts. I need to know what is wrong here.

Thanks.

| CPU : Processor supporting Advanced Vector Extensions instruction set, 8x2 GHz, Max cache size 8192 K

| OS : Linux (2.6.31-14-generic, x86_64) |

| Computer : Wolverine |

| Library : libippie9.so.6.1, 6.1 build 137.20, Jun 10 2009 |

| Test : Timing

| CPU : Processor supporting Advanced Vector Extensions instruction set, 8x2 GHz, Max cache size 8192 K| OS : Linux (2.6.31-14-generic, x86_64) || Computer : Wolverine || Library : libippie9.so.6.1, 6.1 build 137.20, Jun 10 2009 || Test : Timing

...

| Function : ippiResize_32f_C1R Mon Aug 16 10:26:13 2010|

Segmentation fault

[igorastakhov]

Hi,

AVX code for 6.1 (you see - its release date is June 2009) was developed under simulator/emulator and could not be tested on the real HW. First of all if you need AVX support - you should switch to IPP version that has been tested on the real HW - 6.1 update 6 or 7.0 beta/gold (of course the latest - 7.0 gold - is the best variant). Also Resize_ is deprecated functionality - it is reccomended to use ResizeSqrPixel instead.

[Naveen_G_Intel]

Hi,
To get IPP 7.0 Beta - http://software.intel.com/en-us/articles/intel-integrated-performance-primitives-intel-ipp-70-beta-program-homepage/

Regards,
Naveen Gv

[Christopher_Dietsch]

Thank you. IPP ver 7.0 beta appears to have fixed these problems on the actual SNB architecture.