Agilex 7 I-Series "aocl diagnose acl0" error following OFS

NOTE: THIS SOLUTION DOES NOT WORK. CONTINUE READING THE THREAD FOR AN UPDATED CONCLUSION

Hi,

I can confirm that AGIB027R29A1E1VB works with oneAPI support when configured for PCIe Gen4 1x16 and a 2x8 memory subsystem.

oneAPI does not work with either of the PCIe Gen 5 configurations even though I can successfully build OFS FIMs without oneAPI support for both Gen 5 1x16 and Bifurcated Gen 5 2x8 PCIe. When I try to build the oneAPI ASP with PCIe Gen 5, I encounter one of the following errors:

1. For Gen 5 1x16, the build fails due to a Verilog error in ofs-platform-afu-bbb
2. For Bifurcated Gen 5 2x8, the build succeeds but aocl diagnose acl0 times out (DMA fail)

Here are the changes that I made to get OFS and oneAPI to work with my card:

1. Follow the instructions to migrate to the AGIB027R29A1E1VB card: PCIe Attach I-Series (2xR-Tile, F-Tile) - OFS
2. Remove the two South I/O Rows. Follow steps 1-4 from PCIe Attach I-Series (2xR-Tile, F-Tile) - OFS and then follow the removal steps from PCIe Attach F-Series (P-Tile/E-Tile) - OFS. The second set of instructions are for removing the HPS, which is not present in the mem_ss ip, but the steps for removing the two North banks are the same. I attached my preset for your reference.
3. Configure PCIe for Gen4 1x16 by specifying pcie_gen = 4 under [settings] in pcie_host_1pf_1vf.ofss. I attached my ofss for your reference.
4. (Optional) Refloorplan the PR region to recover resources that were used by the southern memory banks by following the steps here PCIe Attach I-Series (2xR-Tile, F-Tile) - OFS. I attached my pr_assignments.tcl for your reference.
5. Build the FIM with ./ofs-common/scripts/common/syn/build_top.sh -p --ofss tools/ofss_config/pcie/pcie_host_1pf_1vf.ofss iseries-dk:null_he_lb,null_he_hssi,null_he_mem,null_he_mem_tg,no_hssi work_iseries-dk_minimal_fim
6. Add a device model for AGIB027R29A1E1VB to $INTELFPGAOCLSDKROOT/share/models/dm. I attached mine for your reference.
7. Modify oneAPI ASP board_spec.xml for a 2x8 memory subsystem by following the instructions here: oneAPI Accelerator Support Package(ASP) Reference Manual - OFS. Make sure the number of memory banks is 2, the maximum theoretical bandwidth is 42656 ((1333MHz x 2 x 64 bits / 8 bits) x 2 banks), the Quartus project settings points to the correct device model, and the FPGA resources for kernel region matches your PR region size.  I attached my preset file for your reference. 
8. Set $OFS_ASP_ROOT appropriately and build the ASP by following the instructions here: oneAPI ASP Getting Started User Guide - OFS
9. Program the FIM following the instructions here: OFS for Agilex® 7 FPGA I-Series Development Kit (2xR-Tile,1xF-Tile) - OFS. Do not skip the sudo pci_device <PCIe BDF> unplug step, otherwise your PCIe link speed may be downgraded. 
10. Initialize the board and run aocl diagnose acl0 following oneAPI ASP Getting Started User Guide - OFS

I am a student unaffiliated with Altera. Let me know if this works for you.

Best,
Jason

Hi,

May I know if you follow the guide in https://ofs.github.io/ofs-2024.3-1/hw/common/reference_manual/oneapi_asp/oneapi_asp_ref_mnl/#412-aocl_mmd_get_info?

Can you share the logging of running "aocl diagnose" and "aocl diagnose acl0"?

Hi,

From the log, it looks like the DMA is stuck. There might be some implmentation issue on your PIM or BSP. Are you using creating your own bsp or this is the BSP release from github?

Are you using creating your own bsp or this is the BSP release from github? Can you share how you implement it?

Hi,

Can you check which driver was installed for the FPGA board? Can you share the "lspci -vv" for that specific board?

Thanks.

Following this thread. I am encountering the same issues shown in James's screenshots with the AGIB027R29A1E1VB card