Skip to content

Cleanup multiratespiir memcpy

Timothy Davies requested to merge tdavies__py3_cleanup_memcpy into gstreamer_python_upgrade

I've removed some fixmes and removed the unused multiratespiir path that skips cudaMallocHost.

There were some old comments about trying to improve it once we had gstreamer 1.0, but I've not been able to find anything to help there. I've removed them as it should really be a separate project. I could make a much broader issue along the lines of "Investigate gstreamer buffers that use pinned memory" or something like that.

Tests

I tried a few variants and added a timer across the whole function, but there didn't seem to be a significant difference on each version. Or rather, the noise was more significant than any improvement. In hindsight it'd make more sense to Just time the memcpys, though the unused path was consistently slower anyway. I've made an issue #113 for one plausible improvement.

The usual 8000 inj test is underway.

Merge request reports