Comment 1:

FTR, here is the profile for 'go test regexp'. The first part is race detector itself
(23%) and the second part is various thunks (37%). If we implement custom asm thunks for
raceread/write/funcenter/funcexit, the second part can be cut to >5% I think. The custom
thunks must do the following:
0. For read/write check that the addr is not on stack.
1. Extract racectx from current g.
2. Extract caller pc from stack. 
3. Switch stack to g0.
4. Do static call of __tsan_read/write/..., pass racectx, pc, addr.
5. Switch stack back.
    11.34%  regexp.test  regexp.test        [.] __tsan_read
     4.03%  regexp.test  regexp.test        [.] __tsan_write
     1.50%  regexp.test  regexp.test        [.] __tsan_func_enter
     1.36%  regexp.test  regexp.test        [.] __tsan_func_exit
     1.01%  regexp.test  regexp.test        [.] __tsan::Mutex::ReadLock()
     0.79%  regexp.test  regexp.test        [.] __tsan::Mutex::ReadUnlock()
     0.47%  regexp.test  regexp.test        [.] __tsan::SyncTab::GetAndLock(__tsan::ThreadState*, unsigned long, unsigned long, bool, bool)
     0.40%  regexp.test  regexp.test        [.] __tsan::ThreadClock::acquire(__tsan::SyncClock const*)
     0.38%  regexp.test  regexp.test        [.] __tsan_write_range
     0.38%  regexp.test  regexp.test        [.] __tsan_read_range
     0.36%  regexp.test  regexp.test        [.] __tsan::Mutex::Lock()
     0.31%  regexp.test  regexp.test        [.] __tsan::ThreadClock::release(__tsan::SyncClock*) const
     0.18%  regexp.test  regexp.test        [.] __tsan::Mutex::Unlock()
     0.17%  regexp.test  regexp.test        [.] __tsan::Acquire(__tsan::ThreadState*, unsigned long, unsigned long)
     0.12%  regexp.test  regexp.test        [.] __tsan::Release(__tsan::ThreadState*, unsigned long, unsigned long)
     0.12%  regexp.test  regexp.test        [.] __tsan::ThreadClock::ReleaseStore(__tsan::SyncClock*) const
     0.10%  regexp.test  regexp.test        [.] __tsan::SyncTab::PartIdx(unsigned long)
23% total
    11.23%  regexp.test  regexp.test        [.] runtime.asmcgocall
     6.34%  regexp.test  regexp.test        [.] runtime.raceread
     4.24%  regexp.test  regexp.test        [.] runtime.cgocall
     2.73%  regexp.test  regexp.test        [.] runtime.racewrite
     2.55%  regexp.test  regexp.test        [.] runtime/race.Read
     1.73%  regexp.test  regexp.test        [.] runtime/race._Cfunc___tsan_read
     1.32%  regexp.test  regexp.test        [.] runtime.racefuncenter
     1.21%  regexp.test  regexp.test        [.] runtime/race.Write
     0.82%  regexp.test  regexp.test        [.] _cgo_12e5d96e021c_Cfunc___tsan_read
     0.75%  regexp.test  regexp.test        [.] runtime.racefuncexit
     0.69%  regexp.test  regexp.test        [.] runtime.getcallerpc
     0.67%  regexp.test  regexp.test        [.] runtime/race._Cfunc___tsan_write
     0.66%  regexp.test  regexp.test        [.] runtime/race.FuncEnter
     0.45%  regexp.test  regexp.test        [.] runtime/race._Cfunc___tsan_func_enter
     0.44%  regexp.test  regexp.test        [.] runtime/race.FuncExit
     0.38%  regexp.test  regexp.test        [.] _cgo_12e5d96e021c_Cfunc___tsan_write
     0.37%  regexp.test  regexp.test        [.] runtime/race._Cfunc___tsan_func_exit
     0.15%  regexp.test  regexp.test        [.] runtime.gentraceback
     0.14%  regexp.test  regexp.test        [.] runtime.raceacquireg
     0.12%  regexp.test  regexp.test        [.] _cgo_12e5d96e021c_Cfunc___tsan_func_exit
     0.11%  regexp.test  regexp.test        [.] _cgo_12e5d96e021c_Cfunc___tsan_func_enter
     0.07%  regexp.test  regexp.test        [.] runtime.RaceAcquire
37% total

Comment 2:

Labels changed: added go1.3maybe.

Comment 3:

Labels changed: added release-none, removed go1.3maybe.

Comment 4:

Labels changed: added repo-main.

Comment 5:

This issue was closed by revision a1695d2.

Status changed to Fixed.