https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97950
Bug ID: 97950
Summary: Unoptimal code generation with
__builtin_*_overflow{,_p} for short and __int128
Product: gcc
Version: 11.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: denis.campredon at gmail dot com
Target Milestone: ---
For the following code, the generation is unoptimal on x86-64.
For most of the functions with `short` a jump is generated.
For the functions with `__int128` all but `mul_overflow_{,un}signed___int128`
seems to have extra `mov` produced.
The same problems apply to all the `__builtin_*_overflow_p`
-------------
#define TEST_OVERFLOW(type) \
bool mul_overflow_signed_##type(signed type a, signed type b, signed type c)
{return __builtin_mul_overflow(a,b,&c);} \
bool add_overflow_signed_##type(signed type a, signed type b, signed type c)
{return __builtin_add_overflow(a,b,&c);} \
bool sub_overflow_signed_##type(signed type a, signed type b, signed type c)
{return __builtin_sub_overflow(a,b,&c);} \
bool mul_overflow_unsigned_##type(unsigned type a, unsigned type b, unsigned
type c) {return __builtin_mul_overflow(a,b,&c);} \
bool add_overflow_unsigned_##type(unsigned type a, unsigned type b, unsigned
type c) {return __builtin_add_overflow(a,b,&c);} \
bool sub_overflow_unsigned_##type(unsigned type a, unsigned type b, unsigned
type c) {return __builtin_sub_overflow(a,b,&c);} \
TEST_OVERFLOW(short)
TEST_OVERFLOW(__int128)
-------------
