03-17-2019 10:48 PM - edited 03-19-2019 10:08 PM
03-22-2019 08:37 AM
What's the OS and SDx version you are using?
03-25-2019 02:26 AM - edited 03-25-2019 02:26 AM
This error pops up when there is a memory out of bound access.
Steps to debug,
04-03-2019 09:47 PM
What do you mean by enable line by line in kernel. I can't add any print statements in the .cl Kernel file. I have tried disabling the clEnqueueNDRangeKernel and I get this warning and the program exits.
WARNING: Profiling may contain incomplete information. Please ensure all OpenCL objects are released by your host code (e.g., clReleaseProgram()).
terminate called without an active exception
terminate called recursively
The code works fine on nvidia and amd gpu, it fails only here in the SDAccel tool.
04-04-2019 02:16 AM - edited 04-04-2019 02:18 AM
1. I assume you are in sw_emu mode, printf is supported for .cl kernels for debugging. Track down the section of code that is failing in iterative manner with print messages in your kernel.
2. Don't disable the NDRange to debug this issue, you must enable the kernel launch.
3. Work Group/Work Item path is not the right approach for FPGA based OpenCL kernels, unless you are trying to baseport existing .cl kernel to SDAccel environment.
The direct portability of a .cl kernel from GPU to FPGA doesnt exist. It results in worst performance or even design itself may not be built/run on the FPGA card.
Link below helps you to understand OpenCL FPGA flow better,
I hope this is useful.