#5749: GHC 7.0.4 Performance Regression (Possibly Vector)
-------------------------------------+--------------------------------------
Reporter: sanketr | Owner:
Type: bug | Status: new
Priority: normal | Component: Compiler
Version: 7.0.4 | Keywords: performance, vector
Os: Linux | Architecture: x86_64 (amd64)
Failure: Runtime performance bug | Testcase:
Blockedby: | Blocking:
Related: 5623 |
-------------------------------------+--------------------------------------
Comment(by sanketr):
I just ran some more tests, including criterion benchmarking and full
profiling, and they show 7.0.4 performs about the same as 6.12.3 actually.
Which contradicts the RTS data I used to diagnose this issue, and this
ticket! I ran basic timing test on Linux which too confirmed it:
{{{
GHC 7.0.4: ghc -O2 --make test.hs -fforce-recomp -rtsopts -fasm && time
./test
[1 of 1] Compiling Main ( test.hs, test.o )
Linking test ...
real 0m1.194s
user 0m0.484s
sys 0m0.649s
GHC 6.12.3: ghc -O2 --make test.hs -fforce-recomp -rtsopts -fasm && time
./test
[1 of 1] Compiling Main ( test.hs, test.o )
Linking test ...
real 0m1.331s
user 0m0.467s
sys 0m0.679s
}}}
So, actually GHC 7.0.4 seems to perform about the same in basic timing
tests (and benchmarking and profiling outputs back it up). But, RTS output
shows a different story as attached in the original description. I am now
confused whether I should be trusting RTS output. I understand that
profiling can introduce distortions, but if I understand it correctly, RTS
is a basic timekeeping option, not a full-on profiling.
In case there is any doubt, all data here has been collected on the same
box, and validated for consistency across multiple iterations.
I will leave the ticket open for now - please resolve as appropriate. As
part of the resolution, I will appreciate some clarity into why RTS data
contradicts the benchmarks and timing tests.
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/5749#comment:2>
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
