09-08-2017 11:43 AM
During the execution of test on the host, observing error similar to this in some cases.
ERROR: Failed to sync 1024/-1 bytes
Initially saw this only bigger buffer size. However it has become more frequent and is now happening even at 1 KB
Debugging showed that this is during rad back of device data into Host. This code has worked realizably so far
q.enqueueReadBuffer(buffer_c, CL_TRUE, 0, vector_size_bytes, result_krnl.data());
What could be causing this? Has anyone seen this?
1. Host memory being corrupted?
2. Device (FPGA) being corrupted/
3. Timing issue on Device (FPGA)?
4. Data Type (int32 vs int?)
5. Buffer size?
09-19-2017 11:10 AM
I have not seen any suggestion for my original post. Unfortunately this issue still exists
Please let me know if anyone has any suggestions to work around or fix this
09-19-2017 11:41 AM
would help if you can provite small testcase and/or :
thank you !
09-19-2017 02:30 PM
I am installing 2017.2,
However the issue is not on all designs, which indicates that the drivers, board and OS, PCIE slot are working
Board is => KCU1500
PCIE is reporting => Gen3, x8
LnkCap: Port #0, Speed 8GT/s, Width x8, ASPM unknown, Latency L0 unlimited, L1 unlimited
ClockPM- Surprise- LLActRep- BwNot-
OS (from uname -r) : 4.1.15.pnotify
SDAccel Version => 2017.1
The OpenCL code has beffers defined for inputs and outputs to kernel. Kenel is a simple MAC (multiply-add). The issue seems to happen when the buffer is filled (result of MAC), i.e readback from kernel
Hope this helps
09-19-2017 03:09 PM