summaryrefslogtreecommitdiff
path: root/compiler/GHC/Cmm
diff options
context:
space:
mode:
authorMatthew Pickering <matthewtpickering@gmail.com>2021-02-19 18:58:22 +0000
committerMarge Bot <ben+marge-bot@smart-cactus.org>2021-02-26 16:26:49 -0500
commit24777bb334a49f6bd6c0df2d5ddb371f98436888 (patch)
tree4bea47a4d8f4922426d226326aebcab5f90f70df /compiler/GHC/Cmm
parent8d1fb46da8883b03f9f3f664a9085ff4fda76e7f (diff)
downloadhaskell-24777bb334a49f6bd6c0df2d5ddb371f98436888.tar.gz
Reimplement Stream in "yoneda" style for efficiency
'Stream' is implemented in the "yoneda" style for efficiency. By representing a stream in this manner 'fmap' and '>>=' operations are accumulated in the function parameters before being applied once when the stream is destroyed. In the old implementation each usage of 'mapM' and '>>=' would traverse the entire stream in order to apply the substitution at the leaves. It is well-known for free monads that this representation can improve performance, and the test results demonstrate this for GHC as well. The operation mapAccumL is not used in the compiler and can't be implemented efficiently because it requires destroying and rebuilding the stream. I removed one use of mapAccumL_ which has similar problems but the other use was difficult to remove. In the future it may be worth exploring whether the 'Stream' encoding could be modified further to capture the mapAccumL pattern, and likewise defer the passing of accumulation parameter until the stream is finally consumed. The >>= operation for 'Stream' was a hot-spot in the ticky profile for the "ManyConstructors" test which called the 'cg' function many times in "StgToCmm.hs" Metric Decrease: ManyConstructors
Diffstat (limited to 'compiler/GHC/Cmm')
-rw-r--r--compiler/GHC/Cmm/Info.hs19
1 files changed, 10 insertions, 9 deletions
diff --git a/compiler/GHC/Cmm/Info.hs b/compiler/GHC/Cmm/Info.hs
index 9298df2544..eaa3c2a923 100644
--- a/compiler/GHC/Cmm/Info.hs
+++ b/compiler/GHC/Cmm/Info.hs
@@ -72,19 +72,20 @@ mkEmptyContInfoTable info_lbl
cmmToRawCmm :: Logger -> DynFlags -> Stream IO CmmGroupSRTs a
-> IO (Stream IO RawCmmGroup a)
cmmToRawCmm logger dflags cmms
- = do { uniqs <- mkSplitUniqSupply 'i'
- ; let do_one :: UniqSupply -> [CmmDeclSRTs] -> IO (UniqSupply, [RawCmmDecl])
- do_one uniqs cmm =
+ = do {
+ ; let do_one :: [CmmDeclSRTs] -> IO [RawCmmDecl]
+ do_one cmm = do
+ uniqs <- mkSplitUniqSupply 'i'
-- NB. strictness fixes a space leak. DO NOT REMOVE.
withTimingSilent logger dflags (text "Cmm -> Raw Cmm")
- forceRes $
- case initUs uniqs $ concatMapM (mkInfoTable dflags) cmm of
- (b,uniqs') -> return (uniqs',b)
- ; return (snd <$> Stream.mapAccumL_ do_one uniqs cmms)
+ (\x -> seqList x ())
+ -- TODO: It might be better to make `mkInfoTable` run in
+ -- IO as well so we don't have to pass around
+ -- a UniqSupply (see #16843)
+ (return $ initUs_ uniqs $ concatMapM (mkInfoTable dflags) cmm)
+ ; return (Stream.mapM do_one cmms)
}
- where forceRes (uniqs, rawcmms) =
- uniqs `seq` foldr (\decl r -> decl `seq` r) () rawcmms
-- Make a concrete info table, represented as a list of CmmStatic
-- (it can't be simply a list of Word, because the SRT field is