You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
But the difference in behavior is just due to RADV using DMA ops some of the time to fill buffers, and AMDVLK always using CS which gets shown by RGP just fine. The case in point was a vkCmdResetQueryPool call, but it probably also happens with vkCmdFillBuffer and others.
My suggestion is to show these DMA operations in the Events view at least. I understand they won't be first-class citizens and they can't show any wavefronts, but not having them represented at all can leave gaps in the timeline.
The text was updated successfully, but these errors were encountered:
Initially reported for RADV at https://gitlab.freedesktop.org/mesa/mesa/-/issues/9720, includes traces captured from AMDVLK & RADV showing the difference
But the difference in behavior is just due to RADV using DMA ops some of the time to fill buffers, and AMDVLK always using CS which gets shown by RGP just fine. The case in point was a
vkCmdResetQueryPool
call, but it probably also happens withvkCmdFillBuffer
and others.My suggestion is to show these DMA operations in the Events view at least. I understand they won't be first-class citizens and they can't show any wavefronts, but not having them represented at all can leave gaps in the timeline.
The text was updated successfully, but these errors were encountered: