Ryan Newton:
But, anyway, it turns out that my example above is easily transformed from a bad GHC performance story into a good one. If you'll bear with me, I'll show how below. First, Manuel makes a good point about the LLVM backend. My "6X" anecdote was from a while ago and I didn't use llvm [1]. I redid it just now with 7.4.1+LLVM, results below. (The below table should read correctly in fixed width font, but you can also see the data in the spreadsheet here.)
Time (ms) Compiled File size Comple+Runtime (ms) GHC 7.4.1 O0 2444 1241K GHC 7.4.1 O2 925 1132K 1561 GHC 7.4.1 O2 llvm 931 1133K GHC 7.0.4 O2 via-C 684 974K
So LLVM didn't help [1]. And in fact the deprecated via-C backend did the best!
I would classify that as a bug.
[1] P.P.S. Most concerning to me about Haskell/C++ comparisons are David Peixotto's findings that LLVM optimizations are not very effective on Haskell-generated LLVM compared with typical clang-generated LLVM.
There is some work underway to improve the situation, but I am sure more could be done. Manuel