Attached patch converts SUB with its output unused to CMP. This can happen in double-word subtraction when lowpart of the result is unused (but highpart subtraction still needs to borrow carry flag).
2017-10-19 Uros Bizjak <ubiz...@gmail.com> PR target/82618 * config/i386/i386.md (sub to cmp): New peephole2 pattern. testsuite/ChangeLog: 2017-10-18 Uros Bizjak <ubiz...@gmail.com> Jakub Jelinek <ja...@redhat.com> PR target/82618 * gcc.target/i386/pr82618.c: New test. Bootstrapped and regression tested on x86_64-linux-gnu {,-m32}. Committed to mainline SVN. Uros.
Index: config/i386/i386.md =================================================================== --- config/i386/i386.md (revision 253899) +++ config/i386/i386.md (working copy) @@ -6766,6 +6766,17 @@ [(set_attr "type" "alu") (set_attr "mode" "<MODE>")]) +(define_peephole2 + [(parallel + [(set (reg:CC FLAGS_REG) + (compare:CC (match_operand:SWI 0 "general_reg_operand") + (match_operand:SWI 1 "general_gr_operand"))) + (set (match_dup 0) + (minus:SWI (match_dup 0) (match_dup 1)))])] + "find_regno_note (peep2_next_insn (0), REG_UNUSED, REGNO (operands[0])) != 0" + [(set (reg:CC FLAGS_REG) + (compare:CC (match_dup 0) (match_dup 1)))]) + (define_insn "*subsi_3_zext" [(set (reg FLAGS_REG) (compare (match_operand:SI 1 "register_operand" "0") Index: testsuite/gcc.target/i386/pr82618.c =================================================================== --- testsuite/gcc.target/i386/pr82618.c (nonexistent) +++ testsuite/gcc.target/i386/pr82618.c (working copy) @@ -0,0 +1,18 @@ +/* PR target/82618 */ +/* { dg-do compile } */ +/* { dg-options "-O2" } */ + +#ifdef __SIZEOF_INT128__ +typedef unsigned __int128 U; +typedef unsigned long long H; +#else +typedef unsigned long long U; +typedef unsigned int H; +#endif + +H f0 (U x, U y) +{ + return (x - y) >> (__CHAR_BIT__ * sizeof (H)); +} + +/* { dg-final { scan-assembler {\mcmp} } } */