Hi! If delete_trivially_dead_insns deletes some insn, DF state might be out of date, and, what's worse, if there are pseudos set multiple times referenced in debug_insns, where the debug_insn references were ok before the deletions, but the deletions shortened lifetime of such a pseudo in certain location, nothing resets the debug insns afterwards and we end up with -fcompare-debug failures because the lifetime of the pseudos is different between -g0 and -g.
Fixed by rerunning df_analyze () if delete_trivially_dead_insns removed anything, bootstrapped/regtested on x86_64-linux and i686-linux. Ok for trunk? 2011-03-07 Jakub Jelinek <ja...@redhat.com> PR debug/47881 * ira.c (ira): Call df_analyze again if delete_trivially_dead_insns removed anything. * gcc.dg/pr47881.c: New test. --- gcc/ira.c.jj 2011-02-21 15:37:42.000000000 +0100 +++ gcc/ira.c 2011-03-07 12:33:59.000000000 +0100 @@ -3232,7 +3232,8 @@ ira (FILE *f) check_allocation (); #endif - delete_trivially_dead_insns (get_insns (), max_reg_num ()); + if (delete_trivially_dead_insns (get_insns (), max_reg_num ())) + df_analyze (); init_reg_equiv_memory_loc (); --- gcc/testsuite/gcc.dg/pr47881.c.jj 2011-03-08 14:12:04.000000000 +0100 +++ gcc/testsuite/gcc.dg/pr47881.c 2011-03-08 14:11:46.000000000 +0100 @@ -0,0 +1,24 @@ +/* PR debug/47881 */ +/* { dg-do compile } */ +/* { dg-options "-O -fcompare-debug -fno-dce -funroll-loops -fno-web" } */ + +extern int data[]; + +int +foo (int *t, int *f, int n) +{ + int i = 0, a, b, c, d; + while (data[*f] && n) + n--; + for (; i < n; i += 4) + { + a = data[*(f++) & 0x7f]; + c = data[*(f++) & 0x7f]; + c = data[*(f++) & 0x7f]; + d = data[*(f++) & 0x7f]; + if ((a & 0x80) || (b & 0x80) || (c & 0x80) || (d & 0x80)) + return 1; + *(t++) = 16; + } + return 0; +} Jakub