Forums - SNPE on sa8295P

9 posts / 0 new
Last post
SNPE on sa8295P
735213383
Join Date: 13 Jul 22
Posts: 6
Posted: Wed, 2023-01-18 00:38

Hi,  I'm using snpe on sa8295p,  I found that AIP runtime is not working, how can I use NPU for inference?

 

  • Up0
  • Down0
v.bathini
Join Date: 27 Feb 22
Posts: 11
Posted: Mon, 2023-01-23 00:42

Hi 

Did you intialized with all ADSP library path? If not please do it .

Please check below points for refernce 

1. You will need to ensure that the AIP runtime (AI Processing runtime) is properly installed and configured on the device.
 
2. You can check the AIP runtime version by running the command "adb shell cat /sys/class/aipc/aipc_info" on the device.
 
3. If the AIP runtime is not working, you may try to reinstall it using the Qualcomm Neural Processing SDK. Also, you can check for any compatibility issues between the version of the AIP runtime and the version of the SNPE toolkit you are using.
  • Up0
  • Down0
735213383
Join Date: 13 Jul 22
Posts: 6
Posted: Mon, 2023-01-30 04:50

thank you for your reply, I have intialized with all ADSP library path, DSP runtime is available, error code is as follow:

  • open /dev/msm_npu failed
  • NPU User Driver: npu_read_info -1000
  • build engine error : error_code=500; error_message=Target runtime is not available. error_code=500; error_message=Target runtime is not available. No viable runtimes available.; error_component=Host Runtime; line_no=409; thread_id=508650159360; error_component=Host Runtime; line_no=265; thread_id=520099939576

and for your suggestions 1. how to install AI Processing runtime? 2. No such file or directory in my device

  • Up0
  • Down0
ruzhongl
Join Date: 7 Jun 21
Posts: 42
Posted: Tue, 2023-01-31 03:10

As I reply in another case, SNPE is not used in 8295, from Gen4, we use QNN to replace 8295. Actually, QNN is widely used in Auto IVI(Gen4)/ADAS/Mobile/IOT.

  • Up0
  • Down0
735213383
Join Date: 13 Jul 22
Posts: 6
Posted: Tue, 2023-01-31 23:31

Hi, I am trying to use QNN on 8295 right now, I found that the backend shared libraries provided by QNN include cpu/gpu/htp, I found a picture in the QNN document, I want to know how to run my network in  hta hardware, there is no hta shared library in qnn. thank you. 

  • Up0
  • Down0
ruzhongl
Join Date: 7 Jun 21
Posts: 42
Posted: Wed, 2023-02-01 00:28

you can ask this issue via case. No hta hardware in SA8295, you can ask your bsp team to check SA8295 hardware design. 

You can think HTP is a new hardware which integreate HTA + DSP into one hardware core.  This way, no memory gap b/w HTA and DSP. HTP support int8/int16/fp16 with high Tops, I believe it is powerful than SA8155P in all cases .

In all, in 8295, use HTP backend.

  • Up0
  • Down0
735213383
Join Date: 13 Jul 22
Posts: 6
Posted: Wed, 2023-02-01 01:32

Thank you. Now I have a clearer understanding of 8295.

Have a nice day!

  • Up0
  • Down0
735213383
Join Date: 13 Jul 22
Posts: 6
Posted: Wed, 2023-02-01 23:27

Hi, I'm running QNN demo on 8295, cpu and gpu backend is normal, when I switch to htp backend, it's running but output is empty (cpu and gpu do have output)

I use qnn-platform-validator to test my device and env, the result seems normal to me.
*********** Results Summary ***********
Backend = GPU
{
  Backend Hardware  : Supported
  Backend Libraries : Found
  Library Version   : OpenCL 2.0 Adreno(TM) 690
  Core Version      : Adreno(TM) 690
  Unit Test         : Passed
}
Backend = DSP
{
  Backend Hardware  : Supported
  Backend Libraries : Found
  Library Version   : Not Found
  Core Version      : Hexagon Architecture V68
  Unit Test         : Passed
}
the cmd I use is 
./qnn-net-run --backend libQnnHtp.so --input_list target_raw_list.txt --retrieve_context Inception_v3_quantized.serialized.bin
 
 
  • Up0
  • Down0
ruzhongl
Join Date: 7 Jun 21
Posts: 42
Posted: Fri, 2023-02-03 02:49

For SA8295 AI, we have dedicated support team. You can raise case via creatpoint and get support.

you can download the KBA from creatpoint, it has basic debug tips.

KBA-220726022834: How_to_run_model_in_different_HTP_on_SA8540_QNN1.x

 

Regarding the issue, it is most like load wrong libs. For example, you may load  wrong libQnnHtpV68Skel.so into HTP. In all, it is a simple environment issue and can get better support via case system.

 

 

  • 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.