Skip to content

Conversation

@JRPan
Copy link

@JRPan JRPan commented Sep 4, 2025

#127 is causing SEGF on PTX RTX 3070 simulations.

Two problems here

at set_kernel_done, the kernel is cleared. But get_stream_id is getting is checking m_kernel of the shader core to get stream id. Since m_kernel is cleared, this caused SEGF.

  1. get_streamID should be getting from the warp, not the shader core one.
  2. Launching new CTAs will set m_kernel again. Even though I think this is kinda fishy but this was the original behavior before introducing the queue, so I'm keeping t his.

@JRPan JRPan requested review from a team and William-An and removed request for a team September 4, 2025 22:12
@JRPan
Copy link
Author

JRPan commented Sep 5, 2025

Can we make this higher priority? This is breaking dev. btw the reviewer was assigned by algorithm :) so don't blame me.

@JRPan JRPan requested a review from LAhmos September 7, 2025 22:50
@JRPan
Copy link
Author

JRPan commented Sep 7, 2025

This needs to be merged first

@JRPan JRPan merged commit b1a8f71 into accel-sim:dev Sep 7, 2025
15 checks passed
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