I recommend using -ddump-simpl, as it produces more readable output.
Actually, I can't see any effect of that pragma in the core files whatsoever, but it certainly has effect on run time.
How about diffing the whole core output (and using -ddump-simpl). If there's a performance difference then there must be a difference in the core code too.
Ok. $ ghc-6.11.20090421 --make primes-test.hs -threaded -O2 -ddump-simpl &> main.simpl $ time ./primes-test +RTS -N2 4001 real 0m9.636s user 0m18.201s sys 0m0.088s $ ghc-6.11.20090421 --make primes-test.hs -threaded -O2 -ddump-simpl &> imported.simpl $ time ./primes-test +RTS -N24001 real 0m17.547s user 0m17.453s sys 0m0.052s I can't exactly use diff because the generated identifier names are not the same, but after poring over with Emacs ediff I have found only one difference that's not attributable to identifiers: $diff main.simpl imported.simpl ... 223c232 < a_s1rs [ALWAYS Just L] :: GHC.Integer.Internals.Integer ---
a_s1sV [ALWAYS Just S] :: GHC.Integer.Internals.Integer
... Does this S vs. L difference have anything to do with strictness and laziness? That line is a part of the `Main.test' definition: $ diff -C 3 main.simpl imported.simpl *** 217,244 **** [Arity 2 Str: DmdType LL] Main.test = ! \ (n1_ahR :: GHC.Integer.Internals.Integer) ! (n2_ahT :: GHC.Integer.Internals.Integer) -> let { ! a_s1rs [ALWAYS Just L] :: GHC.Integer.Internals.Integer LclId [Str: DmdType] ! a_s1rs = Data.List.prod1 (Main.factors2 (Data.List.prod1 ! (GHC.Num.up_list Main.lvl Main.lvl n1_ahR) Data.List.lvl1)) Data.List.lvl1 } in ! case GHC.Prim.par# @ GHC.Integer.Internals.Integer a_s1rs of _ { __DEFAULT -> case Data.List.prod1 (Main.factors2 (Data.List.prod1 ! (GHC.Num.up_list Main.lvl Main.lvl n2_ahT) Data.List.lvl1)) Data.List.lvl1 ! of x1_aUS { __DEFAULT -> ! case GHC.Real.$wdivMod x1_aUS a_s1rs of _ { (# ww1_aUo, _ #) -> ! ww1_aUo } } }
Or do you mean to say that *your* installation of GHC behaves the same when the function `parallelize' is defined in the same module and when it's imported?
Yes, exactly. With both ghc-6.10.1 and a very recent build of ghc-6.11
As you can see, I'm using the 6.11.20090421 build. I looked at recent ones, but the Linux builds seem to be less stable in May. I have the same results (though I didn't use -ddump-simpl) with 6.8.2. Can you recommend a different version to try?