[Bug middle-end/20479] New: Instruction scheduler falsely reorders insns (ia64)

[grigory dot zagorodnev at intel dot com]

Test listed below fails when compiled with -02 optimization at IA64. It appears 
that instruction scheduler misses dependency between two assignments in 
the 'set_code' routine and falsely reorders them.

$ cat test.c
typedef struct s_t {
  short a;
  short b;
} *ps;
 
void set_code (ps obj, short code) {
    *((int *) obj) = 0;
    obj->a = code;
}
 
int printf(const char *, ...);
int main() {
    struct s_t obj;
    set_code(&obj, 123);
    
    printf("%s: got %d, expected 123\n", 
        ((obj.a == 123)?"PASS":"FAIL"), obj.a);
}

$ gcc test.c -O2 
$ ./a.out
FAIL: got 0, expected 123

Here is the assembly code of function 'set_code' with incorrectly reordered 
instructions.

$ gcc test.c -S -O2 
        .proc set_code#
set_code:
        .prologue
        .body
        st2 [r32] = r33         ;    obj->a = code;
        st4 [r32] = r0          ;    *((int *) obj) = 0;
        br.ret.sptk.many b0

PS: the minimal set of optimizations reproducing the failure is "-O1 -fstrict-
aliasing -fschedule-insns".

-- 
           Summary: Instruction scheduler falsely reorders insns (ia64)
           Product: gcc
           Version: 4.1.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: grigory dot zagorodnev at intel dot com
                CC: gcc-bugs at gcc dot gnu dot org
  GCC host triplet: ia64-redhat-linux


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20479