summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorPeter Trommler <ptrommler@acm.org>2021-07-18 17:28:48 +0200
committerMarge Bot <ben+marge-bot@smart-cactus.org>2021-07-29 13:12:10 -0400
commitde2629304aa2df1783f099e1da2cd04ce0423790 (patch)
tree45a0ffd540eb60860651a302e60c4cf393ac24ab
parent4a2ef3dd38adf613f425bb1ef8a8614bea558ff0 (diff)
downloadhaskell-de2629304aa2df1783f099e1da2cd04ce0423790.tar.gz
Delete ToDo about incorrect optimisation [skip ci]
On big-endian systems a narrow after a load cannot be replaced with a narrow load.
-rw-r--r--compiler/GHC/Cmm/Opt.hs4
1 files changed, 0 insertions, 4 deletions
diff --git a/compiler/GHC/Cmm/Opt.hs b/compiler/GHC/Cmm/Opt.hs
index 8035d7b2bd..676c1270b5 100644
--- a/compiler/GHC/Cmm/Opt.hs
+++ b/compiler/GHC/Cmm/Opt.hs
@@ -108,10 +108,6 @@ cmmMachOpFoldM platform conv_outer [CmmMachOp conv_inner [x]]
intconv True = MO_SS_Conv
intconv False = MO_UU_Conv
--- ToDo: a narrow of a load can be collapsed into a narrow load, right?
--- but what if the architecture only supports word-sized loads, should
--- we do the transformation anyway?
-
cmmMachOpFoldM platform mop [CmmLit (CmmInt x xrep), CmmLit (CmmInt y _)]
= case mop of
-- for comparisons: don't forget to narrow the arguments before