#4913: Make event tracing conditional on an RTS flag only
---------------------------------+------------------------------------------
Reporter: tibbe | Owner:
Type: feature request | Status: new
Priority: normal | Component: Runtime System
Version: 7.0.1 | Keywords:
Testcase: | Blockedby:
Os: Unknown/Multiple | Blocking:
Architecture: Unknown/Multiple | Failure: None/Unknown
---------------------------------+------------------------------------------
Comment(by tibbe):
I've just benchmarked the current and the proposed approach. The proposed
approach is about five times faster. Here's the benchmark:
`trace.c`:
{{{
int tracelevel = 0;
}}}
`Test.hs`:
{{{
{-# LANGUAGE ForeignFunctionInterface #-}
module Main (main) where
import Criterion.Main
import Foreign.C.Types
import Foreign.Ptr
import Foreign.Storable
import GHC.Exts
foreign import ccall unsafe "&tracelevel" tracelevel :: Ptr CUInt
traceEventIfEnabled :: String -> IO ()
traceEventIfEnabled s = do
enabled <- peek tracelevel
if enabled > 0
then traceEvent s
else return ()
main = defaultMain
[ bench "load and branch" (loadAndBranch "")
, bench "function call" (functionCall "")
]
loadAndBranch :: String -> IO ()
loadAndBranch s = traceEventIfEnabled s
functionCall :: String -> IO ()
functionCall s = traceEvent s
}}}
Compile and run:
{{{
ghc -O2 -threaded Test.hs trace.c
./Test
}}}
And the results:
{{{
warming up
estimating clock resolution...
mean is 8.629898 us (80001 iterations)
found 1650 outliers among 79999 samples (2.1%)
1479 (1.8%) high severe
estimating cost of a clock call...
mean is 84.59817 ns (63 iterations)
found 4 outliers among 63 samples (6.3%)
3 (4.8%) high mild
1 (1.6%) high severe
benchmarking load and branch
collecting 100 samples, 982316 iterations each, in estimated 862.9905 ms
bootstrapping with 100000 resamples
mean: 8.819932 ns, lb 8.779123 ns, ub 8.940035 ns, ci 0.950
std dev: 323.7500 ps, lb 135.2691 ps, ub 692.9337 ps, ci 0.950
found 9 outliers among 100 samples (9.0%)
5 (5.0%) high mild
4 (4.0%) high severe
variance introduced by outliers: 0.999%
variance is unaffected by outliers
benchmarking function call
collecting 100 samples, 169890 iterations each, in estimated 862.9919 ms
bootstrapping with 100000 resamples
mean: 51.03444 ns, lb 50.98904 ns, ub 51.10433 ns, ci 0.950
std dev: 286.2063 ps, lb 206.6244 ps, ub 408.2373 ps, ci 0.950
found 9 outliers among 100 samples (9.0%)
3 (3.0%) high mild
6 (6.0%) high severe
variance introduced by outliers: 0.990%
variance is unaffected by outliers
}}}
The current approach takes about 51ns on average while the proposed
approach takes only 9ns on average.
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/4913#comment:1>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler
_______________________________________________
Glasgow-haskell-bugs mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
