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
In a complex program built with warp, with arrays being stored in multiple classes, and some of them being "true" arrays while others are zero-copy views, it can become complicated to keep track of all memory allocations and understand where the memory usage comes from. A great addition would be a debugging tool that tells you what arrays are currently allocated at a given point, and with what size.
Context
I was seeing unexpectedly high memory consumption when profiling my code, and after inspecting the more obvious suspects I was still missing some memory. I could try an experimental version of that feature, and it pointed me directly to the allocations I hadn't been accounting for, so that was very helpful.
Activity
shi-eric commentedon Apr 17, 2026
This has been merged in 26bc6ea and will be a part of the next nightly build and the next Warp release at the start of May.
Memory allocation tracking NVIDIAGH-1269
Document custom allocator tracker integration
Fix memory tracker init, sort stability, and docs [NVIDIAGH-1269]