http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54796
Bug #: 54796 Summary: [4.8 Regression] Non-addressable stack parameter debug quality regression Classification: Unclassified Product: gcc Version: 4.8.0 Status: UNCONFIRMED Severity: normal Priority: P3 Component: debug AssignedTo: unassig...@gcc.gnu.org ReportedBy: ja...@gcc.gnu.org CC: aol...@gcc.gnu.org Target: i686-linux /* PR debug/NNNNN */ /* { dg-do run } */ /* { dg-options "-g" } */ __attribute__((noinline, noclone)) void bar (char *a, int b) { __asm volatile ("" : "+r" (a), "+r" (b) : : "memory"); } __attribute__((noinline, noclone)) void foo (int a, int b) { int c = a; char d[b];/* { dg-final { gdb-test 17 "a" "5" } } */ bar (d, 2);/* { dg-final { gdb-test 17 "b" "6" } } */ bar (d, 4);/* { dg-final { gdb-test 17 "c" "5" } } */ } int main () { foo (5, 6); return 0; } (line 17 is the second call to bar) regressed for -m32 -Os -g on {i686,x86_64}-linux supposedly with the http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=188871 change. I've mentioned this already in http://gcc.gnu.org/ml/gcc-patches/2012-10/msg00262.html but filing it now on a testcase that is not dependent on any GCC patches. The problem is that vt_canon_true_dep/canon_true_dependence doesn't figure out that the stack stores (after prologue where hard fp is initialized) can't alias argp. Without the VLA or other reasons forcing use of frame pointer var-tracking replaces sp accesses with argp + offset accesses and canon_true_dependence then disambiguates it using the offsets.