Performance of Haskell (ghc-compiled) applications
Submitted by itkovian on Fri, 06/03/2005 - 2:14am.
I am wondering if there exists any interest in doing some research into the low-level performance of ghc-cmpiled Haskell (duh) applications. It seems to me that getting an idea on cache-miss rates, branch prediction, etc. could be of interest to the Haskell community.
Possible routes are
- performance counters (x86, PowerPC, x86-64)
- DSS simulation
- DIABLO/FIT to instrument the binaries (can we obtain statically compiled Haskell apps?)
- DIOTA to gather data
The main problem atm lies in the benchmarks that would be needed (of course!). People at #haskell have suggested using the ghc regression suite, but I am not sure if these qualify as real-world apps, i.e. the real stuff people use Haskell for.
All input is appreciated.
- itkovian's blog
- Login to post comments
Valgrind includes good tools for low-level profiling on x86, e.g. cacheprof, and they seem to be developing ppc support too. (Valgrind's lead author was part of ghc implementation team a few years ago.)
Sorry about the late reply. The Cache Behaviour of Large Lazy Functional Programs on Stock Hardware might be a good start for your work.