https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116845
--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> --- For LP64 (including using long), fre1 is able to optimize it. ILP32 (with long rather than int): ``` k_12 = j_11(D) + -1; j.0_1 = (sizetype) j_11(D); _2 = j.0_1 + 1073741823; _3 = _2 * 4; _4 = a_13(D) + _3; _5 = *_4; k.1_6 = (unsigned int) k_12; _7 = k.1_6 * 4; _8 = a_13(D) + _7; _9 = *_8; _10 = _5 == _9; _15 = (long int) _10; return _15; ``` LP64 (using long rather than int): ``` k_12 = j_11(D) + -1; j.0_1 = (long unsigned int) j_11(D); _16 = j.0_1 + 18446744073709551615; _3 = _16 * 8; _4 = a_13(D) + _3; _5 = *_4; k.1_6 = (long unsigned int) k_12; _7 = k.1_6 * 8; _8 = a_13(D) + _7; _9 = *_8; _10 = _5 == _9; _15 = (long int) _10; ``` First issue I see is: TREE_CODE (type) == INTEGER_TYPE That most likely should be INTEGRAL_TYPE_P but I don't think that fixes the issue here.
