The proposed fix for the part number extraction logic from platforminfo

[salehjg]

The proposed fix for the part number extraction logic from platforminfo utility output on AWS F1 (Ubuntu AMI/2024.1).

The proposed logic is to skip the extraction step if PROGRAM_PLATFORM_PART is defined beforehand. For example: set(PROGRAM_PLATFORM_PART "xcvu9p-flga2104-2-e" CACHE INTERNAL "F1 part number")

To confirm that this does not break anything else that depends on the part number variable:

1. A kernel should be defined and hw_emu/hw builds should be run.
2. HLS synthesis should be run for a given kernel target to be sure it works as intended.

My information is a bit rusty (I used to work with SDAccel 2019.1 and many things have been changed/depreciated).

Also, see #72 for more details on the problem.

[definelicht]

Did you give this a try using a non-F1 platform as well? 🙂

[definelicht]

Is the DEFINED necessary? I thought both undefined and empty would fail the plain check.

IIUC this is effectively turning the OR into an AND?

[salehjg]

Exactly. I agree.
The only difference of NOT DEFINED VAR and NOT VAR is when VAR is defined, but as an empty string. ORing them will be identical to NOT VAR.

[definelicht]

I don't think we should print the internal CMake variable name

[salehjg]

Unfortunately, at the moment I only have access to the aws f1 instances. I need a day or two to test it on F1 to make sure it doesn't break other stuff.

…

On Fri, 21 Mar 2025, 17:19 Johannes de Fine Licht, ***@***.***> wrote: ***@***.**** commented on this pull request. Did you give this a try using a non-F1 platform as well? 🙂 ------------------------------ In cmake/FindVitis.cmake <#73 (comment)>: > @@ -476,7 +476,7 @@ function(add_vitis_program set(PROGRAM_DEPENDS ${PROGRAM_DEPENDS} ${_PROGRAM_DEPENDS}) # Recover the part name used by the given platform - if(NOT PROGRAM_PLATFORM_PART OR NOT "${${PROGRAM_TARGET}_PLATFORM}" STREQUAL "${PROGRAM_PLATFORM}") + if(((NOT DEFINED PROGRAM_PLATFORM_PART) OR (NOT PROGRAM_PLATFORM_PART)) AND (NOT "${${PROGRAM_TARGET}_PLATFORM}" STREQUAL "${PROGRAM_PLATFORM}")) Is the DEFINED necessary? I thought both undefined and empty would fail the plain check. IIUC this is effectively turning the OR into an AND? ------------------------------ In cmake/FindVitis.cmake <#73 (comment)>: > @@ -485,6 +485,7 @@ function(add_vitis_program set(PROGRAM_PLATFORM_PART ${PLATFORM_PART} CACHE INTERNAL "") endif() endif() + message(STATUS "Deduced PROGRAM_PLATFORM_PART: ${PROGRAM_PLATFORM_PART}") I don't think we should print the internal CMake variable name — Reply to this email directly, view it on GitHub <#73 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB5T7A6I4PABVQHTICKJC3T2VQ3ZJAVCNFSM6AAAAABZP2V6SSVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZDOMBWGUZDEMBZHA> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[definelicht]

Unfortunately, at the moment I only have access to the aws f1 instances. I need a day or two to test it on F1 to make sure it doesn't break other stuff.

No worries! To be honest with I haven't had (easy) access to FPGA development platform tools for years now, so it's not quick for me to test either. I wonder if @derpda is still active?

[derpda]

Ohh a little blast from the past!
I'm not currently active in FPGA development either, but hopefully will be later this year or early next year...

[salehjg]

Today, I ran hw_emu for a small kernel and it generated the xclbin file without a complaint.
Unfortunately, the hw run crashed since the instance did not have enough ram/swap.
Now, I only need to test to see if I can launch Vivado HLS for a specific kernel and see if everything's set.