Inconsistent output

[antonio-fc]

Hi team,

I have an implementation using IRON. I had a previous version that uses float and it works as intended here. Then I reimplemented it but using bfloat16 here. Both applications are executed using the "make run_py" command, the test file used is "test.py", and the npu design file is "aie2.py".

However, I have an issue with the output, where sometimes it prints the correct output, but most times it just prints 0. I am not sure if this is caused by the output itself from the NPU being wrong or just the printing not being interpreted correctly due to the special type. Please ignore the "PASS" output, this is printed regardless of the correctness of the output. The expected output for the bfloat16 application is:
"
Output from distributed input 1/3:
-171
-19
"

Could I get some feedback/insight on this issue?

Thanks in advance.

Best,
Antonio