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
We are slower than Julia's stock GC in some benchmarks, mostly due to crazy GC time. One of the possible reasons is realloc, where we keep reallocating new objects rather than resizingexisting allocation (e.g. for the append benchmark).
I compared 6 builds: [immix, nonmoving immix, Julia stock GC] x [current, base]
Current: after #170 and mmtk/julia#64. It moves much more objects than the base build.
Base: before we introduce conservative stack scanning and more pins.
Binding: cab4348 (https://github.com/qinsoon/mmtk-julia/commits/perf-test-base-1009/),
Julia: 88fea475d9639820488f6dd50fcb08c60a011899(https://github.com/qinsoon/julia/commits/perf-test-base-1009/)
The results are as below
Observations:
realloc
, where we keep reallocating new objects rather than resizingexisting allocation (e.g. for theappend
benchmark).Raw numbers
The text was updated successfully, but these errors were encountered: