G729 Codec IPP 8.2 doesn't perform better than non IPP implementation

[santosh_k_]

Hi,

we recently bought Intel® Parallel Studio XE 2015 which has IPP 8.2 version. we are using them with Intel samples for G729 speech codec. I tried using intel 8.0 samples and Intel 6.1 samples. Both of them are not giving better performance than our own implementation. we also have IPP of 6.1 version which is giving better performance than 8.2.

I linked 6.1 G729 samples with 6.1 version and 8.2 version of IPPs out of both 6.1 version gives better performance.

IPP 6.1 version linked with 6.1 version sample  avg per/frame encode time: 21 microsec

IPP 6.1 version linked with 8.2 version sample  avg per/frame encode time: 42 microsec

IPP 8.0 version linked with 8.2 version sample  avg per/frame encode time: 46 microsec

Current platform is:

OS: windriver (3.4.43-WR5.0.1.13_standard)

Hardware:  Intel(R) Xeon(R) CPU E5-2650 v2 @ 2.60GHz

Regards

Santosh kasula

[Igor_A_Intel]

Hi Santosh,

what do you mean under "8.2 version sample"? according to the sticky message at the top of the forum topics you can see that since 8.0 version IPP doesn't support umc samples and has moved them to legacy.

regards, Igor

[Tang_C_]

HI,Santosh

 I meet the same problem, when i use sample 7.0  g729 under ipp7.0, CPU 61%, but use  samples 8.0 under 8.2, CPU 77%

 in addition, by your testing,

IPP 6.1 version linked with 6.1 version sample  avg per/frame encode time: 21 microsec

IPP 6.1 version linked with 8.2 version sample  avg per/frame encode time: 42 microsec

IPP 8.0 version linked with 8.2 version sample  avg per/frame encode time: 46 microsec

but i test the encode time about 110 microsec per/frame, can you tell how to optimize option to compile the sample and application,thank you very much

 CPU model name    : Intel(R) Xeon(R) CPU E5-2650 v2 @ 2.60GHz