Repository navigation
Perf: Optimization for GAMs (VTune) #3004
Description
Benchmarking using VTune has found several bottlenecks in GAMS
- Significant CPU cycles spent in Interlocked.CompareExchange.
- Call-stack & source code of CenterGraph shows Interlocked.CompareExchange in the inner loop, contributing to the high CPU cycles in CompareExchange. Look at reducing the calls to this intrinsic.
- Next step: Look at re-jiggering the code in CenterGraph to avoid the CompareExchange in the inner loop
Top HotSpots:
| Function | Module | CPU Time |
|---|---|---|
| COMInterlocked::CompareExchangeDouble | coreclr.dll | 55.754s |
| <>c__DisplayClass46_0::b__0 | Microsoft.ML.FastTree.dll | 35.381s |
| Microsoft::ML::Internal::Utilities::DoubleParser::TryParseCore | Microsoft.ML.Core.dll | 30.808s |
| <>c__DisplayClass49_0::b__0 | Microsoft.ML.FastTree.dll | 26.872s |
Activity
matouskozak commentedon Oct 23, 2026
Triage: This is an old issue and investigating this will require new profiles. Please re-open if this is still a blocker and we can take a look.