Summary:
Pull Request resolved: #121

UNIX socket has limit on socket name to be 108, there are tupperware task that has longer than 108 characters.

This diff
- truncates the construction of DynoIPCLogger to last 106 bytes of the tw task name, because the region/cluster info is usually not useful for identifying the task
- change the matching logic in GPU and Task module to be sub-string

Reviewed By: jj10306

Differential Revision: D44050720

fbshipit-source-id: be132a2eedd978933b90e470897c9836e5eb5090

Summary:
Pull Request resolved: #123

Update/Add L3 Counters based on the PPR for Milan processors

Reviewed By: jj10306

Differential Revision: D44489085

fbshipit-source-id: b8d378dea29313d8e8dc2bd2e1c535d6b21330d7

Summary:
Pull Request resolved: #124

Update AmdEvents to include Zen4 events based on Geona PPR (https://www.amd.com/en/support/tech-docs/processor-programming-reference-ppr-for-amd-family-19h-model-11h-revision-b1)

Reviewed By: jj10306

Differential Revision: D44594733

fbshipit-source-id: 135eb70d2da837db1b2df4809daaa2639a73e47c

Summary:
Pull Request resolved: #126

* Add unitMasks for L2 Prefetcher code
* Add Bergamo and Genoa CPU models
* Minor code fixups

Reviewed By: jj10306

Differential Revision: D44671393

fbshipit-source-id: 788008a803f850d4283d2ac3f10c72b4a6b3cb15

Summary:
Pull Request resolved: #127

Boost has defintions for likely and unlikely that conflict and can cause compiler errors.  Renaming to __hbt_likely and __hbt_unlikley to resolve the conflict.

Reviewed By: jj10306

Differential Revision: D44730441

fbshipit-source-id: dbab963a7afce2504052916b0e4866b9950023cd

Summary: Pull Request resolved: #129

Reviewed By: jayesh92

Differential Revision: D44000729

fbshipit-source-id: cfccae48c2f1cfe3d872f19a59e2dba63e7249fc

Summary:
Pull Request resolved: #131

aarch64 does not support __rdtsc()

Reviewed By: dmm-fb

Differential Revision: D44916757

fbshipit-source-id: 50ed0dcaec33b6814d1d9d23aee7cd826ec9f742

Summary:
Pull Request resolved: #130

as above

Reviewed By: jayesh92

Differential Revision: D44921379

fbshipit-source-id: 2045696cf3c10058f12b2e519a11f9179cabe8cc

…tform dependencies (#132)

Summary:
Pull Request resolved: #132

As above

Reviewed By: jayesh92

Differential Revision: D45015343

fbshipit-source-id: 7fcb41a4dd94fad54b933bac762af353225414ff

Summary:
Pull Request resolved: #128

Adds thread level counter for hbt
* Set cpu = -1 to make this thread based perf even
https://man7.org/linux/man-pages/man2/perf_event_open.2.html
* Create Thread counter class specializing `CpuEventGroupBase`
* Avoid verbose prints on PMU sysfs discovery reading?

Reviewed By: haowangludx

Differential Revision: D44764690

fbshipit-source-id: 786f90669c4f05fdc73e5a8e4f2b90513bcea385

Summary:
Pull Request resolved: #133

## About
We can define hbt metrics in terms of a multiple events. However, we lack a way to specify the formula to combine these events,
today that is done manually outside in code or by the user directly.
We add an optional "reducer" functor to our metric definition that helps identify how to reduce performance events..

Included
* Reducer functor type definition and adding it to Metric class (Metric.h/.cpp.
* Perf ReadValues leverages the reducer functor to combine multiple events (CpuEventsGroup.h).
* Include a simple addition and rate reducer (BuiltinMetrics.h).
* Use the rate reducer in instructions/cycle.

Reviewed By: haowangludx

Differential Revision: D45092289

fbshipit-source-id: 2e044915d6595820eddbf95ed7dacfc73c38abfd