Forums - performances gap between snpe tools vs snpe sdk (c++)

2 posts / 0 new
or Register
Last post
performances gap between snpe tools vs snpe sdk (c++)
csjaj9772
Join Date: 23 May 23
Posts: 1
Posted: Tue, 2023-05-23 00:09
Top

Hi.

 

I'd like to know any issues between snpe tools and snpe sdk for c++.

Actually, my model is based on yolo architecture.

And, I use model for inference with DSP only.

When I use snpe tools like 'snpe-net-run' or 'snpe-parallel-run', I got 83ms, and after then I got 57~58ms.
For example, 83ms, 57ms, 57ms, 57ms, for each frames.

 

When I use snpe sdk (c++), then I always get 83ms, 83ms, 83ms, ..., etc.

So, I'd like to know this gap (83ms-57ms=26ms).

for details, I use snpe sdk (SNPE-execute) with TensorMap version from below link.
    https://developer.qualcomm.com/forum/qdn-forums/software/qualcomm-neural...

bool execute(const zdl::DlSystem::TensorMap & input,  zdl::DlSystem::TensorMap & output  )
 

 

Thanks.

  • Up0
  • Down0
Re: performances gap between snpe tools vs snpe sdk (c++)
weihuan
Join Date: 12 Apr 20
Posts: 270
Posted: Fri, 2023-05-26 22:03
Top

Dear developer,

It's recommending to check the snpe acclerator data inner inferenece info. This will be the correct data for you. Sometimes, the gap caused by the cDSP bring up which consumes too much time.

BR.

Wei

  • Up0
  • Down0
or Register

Opinions expressed in the content posted here are the personal opinions of the original authors, and do not necessarily reflect those of Qualcomm Incorporated or its subsidiaries (“Qualcomm”). The content is provided for informational purposes only and is not meant to be an endorsement or representation by Qualcomm or any other party. This site may also provide links or references to non-Qualcomm sites and resources. Qualcomm makes no representations, warranties, or other commitments whatsoever about any non-Qualcomm sites or third-party resources that may be referenced, accessible from, or linked to this site.