ionut matasaru 26f47b060b [Insights] ...

- Improved performance of memory analysis (processing of live allocations in AllocationsProvider) with up to 50% (i.e. twice as fast). This is achived by using a better hash for TMap for long living allocations. Also switched the short living allocations to use only the linked list, with a reduced size (64 allocs).
  - Delayed initialization of SystemMemory heap to when the allocations provider inits (to save memory usage when traces does not have mem alloc events).
  - Added option to enable an initial "reserve" number for TMap (one used for long living allocations). Benchmarks shows further performance increase when using large reserve numbers, but mem cost becomes too high. So kept this 0 (unset) by default.
  - Added error checking for very large mem traces (when total number of mem events could exceed 32 bit limit).

Benchmarks:
  * analysis of a 32s session: from 17.6s to 8.3s (-53%)
  * analysis of a 7min session: from 88s to 46s (-48%)

#jira UE-135890
#rb Johan.Berg
#preflight 620fab43a96c65b0d81a9566

[CL 19070177 by ionut matasaru in ue5-main branch]

2022-02-22 04:14:31 -05:00

..

Private

[Insights]

2022-02-22 04:14:31 -05:00

Public

Stats Trace: Add stat group tracking

2022-02-11 10:57:43 -05:00

TraceServices.Build.cs

[Insights] RAD symbols resolver implementation

2021-03-31 07:56:27 -04:00