#2252: Extreme performance degradation on minor code change
----------------------+-----------------------------------------------------
Reporter: simona | Owner:
Type: bug | Status: new
Priority: normal | Milestone:
Component: Compiler | Version: 6.8.2
Severity: normal | Resolution:
Keywords: | Difficulty: Unknown
Testcase: | Architecture: x86_64 (amd64)
Os: Linux |
----------------------+-----------------------------------------------------
Comment (by simona):
I think (and hope) that BadPerform is going bizarrely slow. BadPerform
runs the same loop twice and should thus only be twice as slow as
GoodPerform, not 800 times as slow. I ran more complex examples a million
times and got something in the range of 4-8s which means that each
fixpoint computation (involving 8 iterations around the loop) takes 4-8us
which is actually quite fast since that would mean that some complicated
list operations and a several calls to simplex are done in a few thousand
instructions on this 2.X GHz machine. But the numbers did seem to make
sense in relation to each other.
I tried to add the `-fno-state-hack` flag but that didn't change anything.
I copied stubs of the relevant function into a new module called `Fake.hs`
-- I tried putting the functions directly into `BadPerform.hs` but that it
went as fast as `GoodPerform.hs`. I made the `fixpoint` function go slower
by adding the test `minimum [i+j | i <- [1..100], j <- [1..100]]<3` as you
suggested.
The results:
{{{
1858$ rm -f tests/GoodPerform.o && ghc-6.8.2 --make GoodPerform.hs && time
GoodPerform
real 0m0.008s
user 0m0.004s
sys 0m0.004s
1859$ rm -f tests/BadPerform.o && ghc-6.8.2 --make BadPerform.hs && time
BadPerform
Linking BadPerform ...
real 0m2.923s
user 0m2.911s
sys 0m0.010s
}}}
I will update the two source files and their simplifier dump. I'll also
add `Fake.hs`.
Thanks for looking into this.
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/2252#comment:6>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler_______________________________________________
Glasgow-haskell-bugs mailing list
Glasgow-haskell-bugs@haskell.org
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
