[PATCH] D54889: Fiber support for thread sanitizer

Tue Jan 8 02:25:52 PST 2019

yuri added a comment.

In D54889#1349383 <https://reviews.llvm.org/D54889#1349383>, @dvyukov wrote:

> "current" is compiler-rt HEAD with no local changes, or with the previous version of this change? This makes clang-compiled runtime 12% faster?

"current" is compiler-rt HEAD without any changes.
"clang" is LLVM and clang HEAD.

In case of clang-compiled library, even previous version of patch was faster then HEAD.

Looks like additional indirection of ThreadState by itself introduce minimal overhead, but affect code generation in unpredictable way, especially for clang.

> Were you able to reproduce the slowdown in  https://reviews.llvm.org/D54889#1343582 with the previous version of the fibers change? What slowdown did you get?

Yes, I can see slowdown with previous version of patch and gcc-compiled library around 3.5%.

> What exactly has changed wrt performance? I see some small tweaks, but I am not sure if they are the main reason behind the speedup or I am missing something important.

You may consider this as "cheating" because changes are general and not related to fibers.

Most speedup is because of change in tsan_rtl.cc:

  - if (!SANITIZER_GO && *shadow_mem == kShadowRodata) {
  + if (!SANITIZER_GO && !kAccessIsWrite && *shadow_mem == kShadowRodata) {

Placing LIKELY/UNLIKELY in code gives additional 1-2%.

Repository:
  rCRT Compiler Runtime

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

https://reviews.llvm.org/D54889