Vitis HLS Puzzle

By Rishov Sarkar

Here’s a “fun” little puzzle I recently solved. In the following screenshot, why does Vitis HLS think that the C testbench failed and refuse to run cosimulation?

• The testbench runs to completion and exits with return code 0. No segfaults or anything like that.
• The kernel code is correct, i.e., no issues with #pragma HLS interface depth=….
• This error occurs during the C testbench pre-check, before the RTL simulation runs at all, so it doesn’t have anything to do with the simulation itself.

Still stumped? Here’s a hint. In the following screenshot, everything works: Vitis HLS doesn’t complain and runs cosimulation to completion.

Do you see the difference?

Vitis HLS saw the string Error: in the output Average Error: 0.015153 and assumed the testbench failed!

Moral of the story: before working with Vitis HLS, learn some synonyms for the word “error”. Maybe “deviation” or “discrepancy”?!