[PATCH] D54889: Fiber support for thread sanitizer

[Dmitry Vyukov via Phabricator via llvm-commits]

dvyukov added a comment.

> I added default synchronization and flag to opt-out.

Can you swicth to it in your codebase? I would expect that you need all (or almost all) of this synchronization anyway?

> In standalone thread we create temporary context and switch into into it. Then original thread context is posted into event loop thread, where it runs for some time together with other fibers.

Please add such test (or we will break it in future).

> IMHO it is hardly possible function entry/exit use shadow stack, memory access functions use clock. Both structure are too big to be copied on each fiber switch.

This is a good question.
Shadow stack seems to use an indirection already.
Clock copying is a sequential memory copy. If handling of memory accesses/function entry/exit becomes a bit slower, then a thousand of memory accesses can outweight cost of copying clock.
A good question is: how much slower is the code with the indirection in cur_thread?


Repository:
  rCRT Compiler Runtime

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D54889/new/

https://reviews.llvm.org/D54889