Skip to content

Conversation

rankaiyx
Copy link

@rankaiyx rankaiyx commented Jul 3, 2025

mlc-ai/mlc-llm#3231

When I use mlc-llm, I encounter error in linking.
I solved the problem with this patch.
This patch may not solve the root cause. But for my scene it is suitable, the performance loss is at the noise level.
The root cause may be the lack of relevant instructions in Pascal, or a bug in Nvidia's Thrust library.
This patch is intended as an emergency mitigation. Looking forward to a better way.

@rankaiyx rankaiyx closed this Jul 4, 2025
@rankaiyx rankaiyx force-pushed the rankaiyx-patch-fix-pascal-error branch from ecd8615 to 6620fe2 Compare July 4, 2025 01:00
@rankaiyx rankaiyx reopened this Jul 4, 2025
@rankaiyx
Copy link
Author

rankaiyx commented Jul 4, 2025

When using pascal,
can_use_thrust(target, "tvm.contrib.thrust.sum_scan") returns True,
but will actually fail.

@gatepoet
Copy link

Could you include sm_61 as well. NVIDIA Tesla P40 has the same issue

@rankaiyx rankaiyx changed the title Update scan.py to fix pascal error. Update scan.py to fix pascal error Aug 26, 2025
gatepoet added a commit to gatepoet/relax that referenced this pull request Sep 13, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants