Improve inlining
Currently the contruct
case (x# ># 0#) of ...
attracts no argument discount for x#, which is silly.
The comment in CoreUnfold says:
PrimOpId op -> primOpSize op (valArgCount args)
-- foldr addSize (primOpSize op) (map arg_discount args)
-- At one time I tried giving an arg-discount if a primop
-- is applied to one of the function's arguments, but it's
-- not good. At the moment, any unlifted-type arg gets a
-- 'True' for 'yes I'm evald', so we collect the discount even
-- if we know nothing about it. And just having it in a primop
-- doesn't help at all if we don't know something more.
But the right thing to do seems to be to fix interestingArg in SimplUtils so that it only thinks a primitive-typed thing is interesting if it knows its value (or some structure).
Here's the program that triggered this thought:
import GHC.Word
import GHC.Base
import GHC.Prim
a `shiftRLT` b | b >=# 32# = int2Word# 0#
| otherwise = a `uncheckedShiftRL#` b
(W32# x#) `shift` (I# i#) =
{- we do an actual case analysis on i# to try to give us a discount -}
case i# of
{- For some bizzare reason removing the `shiftRLT` 0# makes the
inlining fail again -}
0# -> W32# (x# `shiftRLT` 0#)
_ ->
if i# >=# 0# then W32# (narrow32Word# (x# `shiftL#` i#))
else W32# (x# `shiftRLT` negateInt# i#)
x `shiftR` y = x `shift` (-y)
shift7 x = x `shiftR` 7
roconnor@theorem.ca initiated the thread
Trac metadata
Trac field | Value |
---|---|
Version | 6.4.2 |
Type | Task |
TypeOfFailure | OtherFailure |
Priority | normal |
Resolution | Unresolved |
Component | Compiler |
Test case | |
Differential revisions | |
BlockedBy | |
Related | |
Blocking | |
CC | |
Operating system | Unknown |
Architecture | Unknown |