https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111560
--- Comment #4 from Yi <652023330028 at smail dot nju.edu.cn> --- (In reply to Andrew Pinski from comment #1) > The problem is in this case: > e=a+b+c; //line 5 > f=d+b+c; //"b+c" can be replaced with the value at line 5 > > at the gimple level, b+c could introduce an signed integer overflow. And we > don't want to introduce that. > > > If you had used unsigned, rather than int, the optimization would have been > done as we do reassociate unsigned integers just fine. Thank you very much for your quick reply! But example 1 in Bug 111561 does not seem to be well resolved. https://godbolt.org/z/cM4n8qvrh Given the following code: ```c++ extern unsigned var_26; extern unsigned var_29; void test(unsigned var_1, unsigned var_2, unsigned var_3) { var_29 = (var_1 + var_2) ? var_1 : var_2; var_26 = var_1 + (var_2 + var_3); } ``` ```asm test(unsigned int, unsigned int, unsigned int): mov eax, edi add eax, esi mov eax, edi cmove eax, esi add esi, edx add esi, edi mov DWORD PTR var_29[rip], eax mov DWORD PTR var_26[rip], esi ret ``` So does Bug 111562 .
