The patch is addressing following LLVM warning:
gcc/ipa-fnsummary.c:2745:54:Value Conversion Issue: implicit conversion from
'double' to 'int64_t' (aka 'long') changes value from 0.99 to 0:
-Wliteral-conversion
Taking look at GIMPLE IR we end up with:
<bb 45> :
D.107433 ={v} {CLOBBER};
sreal::sreal (&D.107437, -1, 0);
sreal::sreal (&D.107434, 0, 0);
D.107435 = sreal::operator* (&time, &D.107434);
D.107436 = sreal::operator- (&nonspecialized_time, &D.107435);
_65 = operator>= (&D.107436, &D.107437);
_66 = ~_65;
if (_66 != 0)
goto <bb 46>; [INV]
else
goto <bb 47>; [INV]
<bb 46> :
fancy_abort ("/home/marxin/Programming/gcc/gcc/ipa-fnsummary.c", 2745,
&__FUNCTION__);
which confirms the warning.
gcc/ChangeLog:
2018-09-23 Martin Liska <mli...@suse.cz>
* ipa-fnsummary.c (estimate_node_size_and_time): Scale by two
integers and not by a float value.
---
gcc/ipa-fnsummary.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/gcc/ipa-fnsummary.c b/gcc/ipa-fnsummary.c
index 62095c6cf6f..9cb7d41ccc5 100644
--- a/gcc/ipa-fnsummary.c
+++ b/gcc/ipa-fnsummary.c
@@ -2742,7 +2742,7 @@ estimate_node_size_and_time (struct cgraph_node *node,
gcc_checking_assert (time >= 0);
/* nonspecialized_time should be always bigger than specialized time.
Roundoff issues however may get into the way. */
- gcc_checking_assert ((nonspecialized_time - time * 0.99) >= -1);
+ gcc_checking_assert ((nonspecialized_time - time * 99 / 100) >= -1);
/* Roundoff issues may make specialized time bigger than nonspecialized
time. We do not really want that to happen because some heurstics
