Problem with xf_sw_utils.h file

[anjaliagarwal8]

Hi I am trying to port Canny edge detection function in my code for hardware acceleration. I am getting errors related to xf_sw_utils.h header file.

In file included from /home/ml1/workspace/image_registration/src/image_registration.cpp:12: /home/ml1/zcu104_rv_ss/sw/a53_linux/inc/xfopencv/common/xf_sw_utils.h:120:98: error: expected '>' int cv_pix = cv_img.at<ap_uint<PixelType<DataType<_PTYPE,_NPC>::pixeldepth>::bitdepth>>(i,j);

I also checked this line in the header file many times, but the line there is different from what is specified in the error. The line is something like this: int cv_pix = cv_img.at<ap_uint<XF_PIXELDEPTH ( XF_DEPTH (_PTYPE, _NPC))>>(i,j);

[bgouthamb]

@Anjali-iit8 Unable to replicate this error. Can you please let me know these details: OS, SDx version, Platform, along with the flags and paths used in Makefile? Are you facing this issue when building the function Canny standalone also (individual build, not as a part of your project)?

[bgouthamb]

@Anjali-iit8 XF_PIXELDEPTH(...), XF_DEPTH(...) are macros which resolve to the corresponding template specialization calls (see include/common/xf_types.h). The line in your error log file:

int cv_pix = cv_img.at<ap_uint<PixelType<DataType<_PTYPE,_NPC>::pixeldepth>::bitdepth>>(i,j);

is the expanded form of the macros in the original code:

int cv_pix = cv_img.at<ap_uint<XF_PIXELDEPTH ( XF_DEPTH (_PTYPE, _NPC))>>(i,j);

[anjaliagarwal8]

@bgouthamb OS: Ubuntu 16.04 SDx version: 2017.4 Platform: zcu104 I am building it using the GUI. I have build the Canny standalone and it is working but when I am building it as a part of my project then it is giving errors. I have attached the header file where I have defined the function. canny_edge.txt

and here in my code I have called the function: cv::Mat imgInput1; imgInput1 = cv::imread(argv[1],0); xf::Mat<XF_8UC1, 640, 480, XF_NPPC1> imgInput(imgInput1.rows, imgInput1.cols); imgInput = xf::imread<XF_8UC1, 640, 480, XF_NPPC1>(argv[1],0);

     cann(imgInput);

Also, I am able to port it to the PS side. This error is encountered when porting to PL side.

[anjaliagarwal8]

@bgouthamb I just checked, when I am running the Canny standalone using Makefile provided then it is working but when I am building it using SDx GUI then it is giving errors. So, I think problem is with working on GUI. Is there any extra steps required for building it using GUI? I have included the necessary libraries and header files which were specified in the Makefile.

[bgouthamb]

@Anjali-iit8 Please have a look at the UG1233 document, page 12 to 20. Check if you have followed the steps mentioned there, based on whether you are using zcu104 or _zcu104_rvss platform.

[anjaliagarwal8]

@bgouthamb Yes I have followed that process only, but then also I am getting errors while building. This is the error I am getting: /home/ml1/workspace/canny/src/xf_canny_accel.cpp:43:13: error: array type 'typename DataType<8, 32>::name [64800]' is not assignable _dst1.data = (ap_uint<64>*)_dst.data;

[bgouthamb]

@Anjali-iit8 This is a different error you have posted now. Are you seeing this in the standalone build? If yes, please check whether you have marked "xf::Canny<...>" and "xf::EdgeTracing<...>" to move to hardware, not "canny_accel".

[anjaliagarwal8]

@bgouthamb When I marked the above functions for acceleration then I am not getting this error but still I am not able to build: make: *** No rule to make target 'xf_canny_accel.o', needed by 'canny.elf'. Stop.

Also when I gave xf::Canny<...> function for the hardware acceleration in my code I am getting the same errors: /home/ml1/zcu104_rv_ss/sw/a53_linux/inc/xfopencv/common/xf_sw_utils.h:120:98: error: expected '>' int cv_pix = cv_img.at<ap_uint<PixelType<DataType<_PTYPE,_NPC>::pixeldepth>::bitdepth>>(i,j);

[bgouthamb]

@Anjali-iit8 If it is building with Makefile and not building with GUI, might be that you are missing setting up something(?). I just had it built in my system in GUI with no issues. Recheck your settings and try doing a clean build or new SDx project.

[anjaliagarwal8]

@bgouthamb Thank you. The Canny standalone is building now with the GUI. I also made a new SDx project of my own project also, and I gave the function xf::Canny<..> for hardware acceleration but when I am building it, it is not accelerating the function. It is not giving any errors but it is not accelerating the above function also. It seems like it is porting the whole code to the PS side. Also when I am trying to open the hardware function with Vivado standalone, it is giving error: No Vivado HLS project exist for the given hardware function.

[anjaliagarwal8]

@bgouthamb If I understand right, we cannot give our own function for hardware acceleration, even if it comprises of only xfopencv functions?

[bgouthamb]

@Anjali-iit8 Of course, you can! See the color detection example in UG1233. Not accelerating a function, even after marking for hardware, seems to be a problem with the SDx GUI. Please post in SDSoC forum about this. Now that, your original issue about _xf_swutils.h is resolved, can you please close this?

[anjaliagarwal8]

Yes. Thanks