-
Notifications
You must be signed in to change notification settings - Fork 225
Replies: 2 comments · 5 replies
-
The only thing I've had to add to my image recipes (I'm assuming you're using
The other needed bits should be picked up automatically through the bbappends in recipes-core/packagegroups. |
Beta Was this translation helpful? Give feedback.
All reactions
-
We have a custom recipe for making the SDK, and it works similar to this:
However, this is not working, the gcc-8.5 cross compiler does not get added. I am trying to make a minimal example which creates an SDK with CUDA and gcc-8.5. We are defining our own custom tx2-based MACHINE, could that be causing any problems? |
Beta Was this translation helpful? Give feedback.
All reactions
-
It's more likely to be your recipe than the custom machine definition, but some debugging will tell you. You might need to start using In particular, make sure your SDK recipe is dependent on |
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 1
-
Okay, this is new territory for me, but here we go. I tried bitbake -e, and get the following hits for
And finally that
Found from:
Continuing on for
But that does not look all that relevant. How does this look? |
Beta Was this translation helpful? Give feedback.
All reactions
-
Does this clarify anything? I notice that |
Beta Was this translation helpful? Give feedback.
All reactions
-
Yep, there's your problem. The bbappend in meta-boot2qt is removing what we add in our bbappend:
Looking at the commit message for their bbappend, it looks like Qt had reasons to do this that probably don't apply to your situation. You might consider removing that layer from your builds altogether. If you can't do that, you'll need to work around the problem it's introducing, maybe by explicitly adding |
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 1
-
Yes! That did it! Thanks very much for helping out! :D |
Beta Was this translation helpful? Give feedback.
-
We are transitioning from dunfell to a kirkstone based yocto build targetting a TX2. To get the correct host code compiler (gcc 8.5) for CUDA (10.2), I have understood that making recipies inherit cuda should be enough. However, we also have a SDK recipe which should include that compiler. Currently, I have not been able to get that to work; only the default compiler (gcc 11) gets included. What is the simplest way of getting everything needed for cuda development to show up in the SDK?
Beta Was this translation helpful? Give feedback.
All reactions