[pretty-ipa] Fix inliner double accounting problem
Jan Hubicka
hubicka@ucw.cz
Tue Apr 6 11:00:00 GMT 2010
Hi,
I've commited the following patch to pretty-ipa to fix inliner double accounting the savings caused
by eliminating needs for offline copy of a function. For tramp3d I also need to bump unit growht
to 50% to get same results as before.
I've also merged from mainline today to get tramp3d compiling again.
Bootstrapped/regtested x86_64-linux.
* tree-inline.c (cgraph_mark_inline_edge): Do not update overall size for fully
inlined functions.
* params.def (PARAM_INLINE_UNIT_GROWTH): Set to 50%.
Index: ipa-inline.c
===================================================================
--- ipa-inline.c (revision 157985)
+++ ipa-inline.c (working copy)
@@ -336,8 +336,6 @@ cgraph_mark_inline_edge (struct cgraph_e
gcc_assert (what->global.inlined_to == to);
if (new_size > old_size)
overall_size += new_size - old_size;
- if (!duplicate)
- overall_size -= orig_size;
ncalls_inlined++;
if (flag_indirect_inlining)
Index: params.def
===================================================================
--- params.def (revision 157985)
+++ params.def (working copy)
@@ -177,7 +177,7 @@ DEFPARAM(PARAM_LARGE_UNIT_INSNS,
DEFPARAM(PARAM_INLINE_UNIT_GROWTH,
"inline-unit-growth",
"How much can given compilation unit grow because of the inlining (in percent)",
- 30, 0, 0)
+ 50, 0, 0)
DEFPARAM(PARAM_IPCP_UNIT_GROWTH,
"ipcp-unit-growth",
"How much can given compilation unit grow because of the interprocedural constant propagation (in percent)",
More information about the Gcc-patches
mailing list