GZip : Segfault when no awsxclbin or xclbin is present

[adentinger]

It makes sense that the execution does not work when no AWSXCLBIN nor XCLBIN is found, but, in my case, this situation causes a segfault on the call to cl::Program, which is difficult to diagnose.

Maybe find_binary_file in xcl2.cpp should at least specify whether we found an .awsxclbin, an .xclbin, or neither and at least warn that finding neither may cause errors?

[kaliudayxilinx]

Hi @AnthonyD973

Can you explain the steps that you have tried to run this application.

[adentinger]

I essentially :

1. Logged in on a Z1D instance (Lots of RAM while being cheaper than an F1 instance) ;
2. Cloned this GitHub repository and the aws-fpga repository into /home/centos/src/project-data ;
3. Ran sdaccel_setup.sh ;
4. Ran make all TARGETS=hw DEVICES="${AWS_PLATFORM}" inside the GZip/ directory ;
5. Created the AFI using the SDAccel/tools/create_sdaccel_afi.sh script under the aws-fpga repository and waited until the AFI creation is done on AWS's side ;
6. Logged on an F1 instance (obviously copying the entire built GZip/ directory over to that instance) ;
7. Cloned the aws-fpga repository into /home/centos/src/project-data ;
8. Waited until the XOCL driver was running on the F1 instance ;
9. Ran the sdaccel_setup.sh and sdaccel_runtime_setup.sh scripts ;
10. Ran the xil_gzip, passing it some random text file (I don't remember the exact flag, maybe -i?).

I might have missed a few steps, but that's about it.

Since the SDAccel/tools/create_sdaccel_afi.sh script creates the .awsxclbin file inside the current directory and I use an automated script to do all of this, my .awsxclbin was not placed in the same directory as the xil_gzip file, and this situation causes the xil_gzip executable to segfault when running the cl::Program constructor. As I said in my OP, I suggest that the find_binary_file in xcl2.cpp should at least print out a warning, saying that it did not find the .awsxclbin file, which may lead to an error or a segfault.

[kaliudayxilinx]

Hi @AnthonyD973 ,

Looks like an infrastructure related issue.
Can you please post this query on AWS Forum.

AWS Forum Link:
https://forums.aws.amazon.com/forum.jspa?forumID=243&start=0