blob: 0e6e87121547fed49d7c7f90b647d73be0469a60 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
|
module ShouldCompile where
postInlineUnconditionally
= case Just "Hey" of
-- The point of examining occ_info here is that for *non-values*
-- that occur outside a lambda, the call-site inliner won't have
-- a chance (because it doesn't know that the thing
-- only occurs once). The pre-inliner won't have gotten
-- it either, if the thing occurs in more than one branch
-- So the main target is things like
-- let x = f y in
-- case v of
-- True -> case x of ...
-- False -> case x of ...
-- I'm not sure how important this is in practice
Just a -- OneOcc => no work-duplication issue
-> True -- Small enough to dup
-- ToDo: consider discount on smallEnoughToInline if int_cxt is true
--
-- NB: Do NOT inline arbitrarily big things, even if one_br is True
-- Reason: doing so risks exponential behaviour. We simplify a big
-- expression, inline it, and simplify it again. But if the
-- very same thing happens in the big expression, we get
-- exponential cost!
-- PRINCIPLE: when we've already simplified an expression once,
-- make sure that we only inline it if it's reasonably small.
_ -> False
-- Here's an example that we don't handle well:
-- let f = if b then Left (\x.BIG) else Right (\y.BIG)
-- in \y. ....case f of {...} ....
-- Here f is used just once, and duplicating the case work is fine (exprIsCheap).
-- But
-- * We can't preInlineUnconditionally because that woud invalidate
-- the occ info for b.
-- * We can't postInlineUnconditionally because the RHS is big, and
-- that risks exponential behaviour
-- * We can't call-site inline, because the rhs is big
-- Alas!
where
x = id
|