FPGA (opencl) code works in emulation (with strict channel depths) but does not work on FPGA (wrong results)

Your code does not compile on my environment due to missing channel connections for “A_feeder_channel”. I am using aoc directly on your kernel. Furthermore, you have two writes to the “B_loader_channel” channel. Multiple call sites were not allowed until v17.0 and I would definitely avoid it even now that it is allowed. You are using very complex channeling operations, seemingly with multiple call sites in some places, and you are not enforcing channel ordering as far as I can see. This could very well be the reason for the incorrect output. The emulator fails at emulating concurrency properly; hence, you should not expect it to show concurrency issues that might exists in your kernel correctly.