How to build xfDNN project?

[semchan]

Dear All,

I want build xfdnn for my project and run on ZCU102 platform but I have not got the introduction. I do the steps as below:

step1: use the platform ZCU102 Production Silicon reVISION Package which provided in revision as my project platform.

step2: then use xfDNN-SDx-2017.2-Lounge-2017-09-07.zip the smaples into the plaftrom, included the following libs:

opencv_core
opencv_imgproc
opencv_imgcodecs
opencv_features2d
opencv_calib3d
opencv_flann
lzma
tiff
png16
z
jpeg
dl
rt
webp
protobuf 
xlnxdnn
parser_arm
xstack

step3: code main.cpp as below only to try test the demo. And build success

int main(int argc, char **argv)

{

 fprintf(stderr, "Test demo\n");
 return 0;

}

step4: run in ZCU102 board, and the message pop out as below:

Test demo
ERROR: application performed illegal memory access and is being terminated
root@xilinx-zcu102-2017_2:/mnt#

Why there is a error about "ERROR: application performed illegal memory access and is being terminated"?

Is that I can not use the platform ZCU102 Production Silicon reVISION Package to build xfdnn?How can I build the xfdnn that I can only modify the input and output which not modify the network implementation?

Thanks so much.

[semchan]

I keep on do the task to build the application of "googlenet_ex.cpp", and added the library as below:

opencv_core
opencv_imgproc
opencv_imgcodecs
lzma
tiff
png16
z
jpeg
dl
rt
webp
protobuf
xlnxdnn
parser_arm
xstack
pthread

build success and got the .elf file. copy the .elf file into SD card with the prebuild librarys and run on ZCU102. The system runs well, but when I run the .elf file, the messages pops as below:

....

[IG001] Saving /mnt/models/GoogleNet/data/inception_5b_5x5_reduce_weights (156 KB)
[IG001] Saving /mnt/models/GoogleNet/data/inception_5b_5x5_reduce_bias (192 B)
[IG001] Saving /mnt/models/GoogleNet/data/inception_5b_pool_proj_weights (416 KB)
[IG001] Saving /mnt/models/GoogleNet/data/inception_5b_pool_proj_bias (0.5 KB)
[IG001] Saving /mnt/models/GoogleNet/data/loss3_classifier_weights (3.91 MB)
[IG001] Saving /mnt/models/GoogleNet/data/loss3_classifier_bias (3.91 KB)
[INFOx] Graph generated
[INFOx] Generating JobQueue
[INFOx] JobQueue generated
[INFOx] Creating Memory

[INFOx] Memory created

[INFOx] Init Start : This may take a while ............................................................................

[INFOx] Init Done
[IMRDx] Image read : /mnt/models/GoogleNet/input/camel1.jpg
[IMRDx] Image read : /mnt/models/GoogleNet/input/camel2.jpg
ERROR: Attempting to write on an axi-lite port that lacks a write status register
ERROR: application performed illegal memory access and is being terminated
Segmentation fault
root@xilinx-zcu102-2017_2:/mnt#

It seems that the internal axi port register bettween PL and PS is not right. But I think I only build the application (.elf), not modify any parts in PL, why have the error? Who can guide me? thanks very much.

[akshayva]

Hi semchan,

               I am looking at the same error message. However, I have been successful in getting past this using the Makefile provided with the examples. I am looking at why the application reports this illegal access error. Will get back to you shortly.

[akshayva]

The xfDNN lounge examples have not been tested to work with SDx GUI. It is expected that the user generates executable using Makefile

[shengscum]

Hi, guy

I am newer to SdSoc, I want to down the file 'xfDNN-SDx-2017.2-Lounge-2017-09-07.zip', but I cannot find the url, can you share me  the download files? 

[akshayva · Accepted solution]

https://www.xilinx.com/applications/megatrends/machine-learning.html
Hit “Download Demos” button and accept the agreement. That should get you started.

[semchan]

Dear akshayva,

Yes, you are right, I have used the Makefile build in Linux system and the result running well. But the build from SDx GUI can not. Wish SDx GUI can also provide the same function and fix the bug soon. Thanks a lot.

[southernduck]

Hello,

Can I use xfDNN on Zedboard?

Thanks

[haojin]

I still got the error. Any updates?

[vandenplas]

Did you get it working?