Forum Discussion

Björne2's avatar
Björne2
Icon for New Contributor rankNew Contributor
2 years ago
Solved

What causes OpenCL to insert arbitration for local memory accesses?

I know FPGA OpenCL is deprecated in favor of OneAPI, but I hope you can help me anyway. I've created a MWE of a kernel for which the compiler inserts arbitration: __attribute__((uses_global_work_o...
Board Debug
High-level Design Tools
aikeu's avatar
aikeu
Icon for Regular Contributor rankRegular Contributor
2 years ago

Hi Björne2,


I did get some feedback from the team related to your OpenCL question, hope if will also help:

"

It’s arbitrating since because has 3 stores and 2 loads.

Some things to try:

1. Disable pipelining on loop at line 14 (n_tics) -> this should make the inner loops execute consecutively and in theory the LDs and STs can be shared instead of arbitrated.

2. don’t pre-load frontier. This will remove one of the ST sites and simplify the control flow. Add some logic to the loop at line 15 (i = 0:100) to detect if it’s the first iteration and re-use that store site to pre-load frontier.

"


Thanks.

Regards,

Aik Eu