[Bug tree-optimization/94952] New: Possible false positive of uninitialized variable usage during release build in gimple-ssa-store-merging.c

stefansf at linux dot ibm.com gcc-bugzilla@gcc.gnu.org
Tue May 5 10:03:57 GMT 2020


https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94952

            Bug ID: 94952
           Summary: Possible false positive of uninitialized variable
                    usage during release build in
                    gimple-ssa-store-merging.c
           Product: gcc
           Version: 10.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: stefansf at linux dot ibm.com
  Target Milestone: ---

Created attachment 48450
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=48450&action=edit
initialize variables `bitpos`, `bitregion_start`, and `bitregion_end`

While bootstrapping GCC on S/390 with --enable-checking=release the following
warnings are raised:

gcc/gimple-ssa-store-merging.c: In member function 'virtual unsigned int
{anonymous}::pass_store_merging::execute(function*)':                          
                     gcc/gimple-ssa-store-merging.c:1412:24: error:
'bitregion_end.poly_int_pod<1, long unsigned int>::coeffs[0]' may be used
uninitialized in this fu
nction [-Werror=maybe-uninitialized]
 1412 |     , ops { op0r, op1r } 
      |                        ^ 
gcc/gimple-ssa-store-merging.c:4672:32: note: 'bitregion_end.poly_int_pod<1,
long unsigned int>::coeffs[0]' was declared here
 4672 |   poly_uint64 bitregion_start, bitregion_end;
      |                                ^~~~~~~~~~~~~
gcc/gimple-ssa-store-merging.c:1412:24: error: 'bitregion_start.poly_int_pod<1,
long unsigned int>::coeffs[0]' may be used uninitialized in this
function [-Werror=maybe-uninitialized]
 1412 |     , ops { op0r, op1r } 
      |                        ^ 
gcc/gimple-ssa-store-merging.c:4672:15: note: 'bitregion_start.poly_int_pod<1,
long unsigned int>::coeffs[0]' was declared here
 4672 |   poly_uint64 bitregion_start, bitregion_end;
      |               ^~~~~~~~~~~~~~~
In file included from gcc/coretypes.h:449,
                 from gcc/gimple-ssa-store-merging.c:143:
gcc/poly-int.h:2063:21: error: 'bitpos.poly_int<1, long unsigned
int>::<anonymous>.poly_int_pod<1, long unsigned int>::coeffs[0]' may be used
uni
nitialized in this function [-Werror=maybe-uninitialized]
 2063 |     if (a.coeffs[i] % b != 0)
      |         ~~~~~~~~~~~~^~~
gcc/gimple-ssa-store-merging.c:4671:24: note: 'bitpos.poly_int<1, long unsigned
int>::<anonymous>.poly_int_pod<1, long unsigned int>::coeffs[0]'
was declared here
 4671 |   poly_uint64 bitsize, bitpos;
      |                        ^~~~~~

Function `mem_valid_for_store_merging` initializes the aforementioned variables
`bitpos`, `bitregion_start`, and `bitregion_end` if it returns a value
different than `NULL_TREE`.

tree base_addr
  = mem_valid_for_store_merging (lhs, &bitsize, &bitpos,
                                 &bitregion_start, &bitregion_end);

Thus the local variable `invalid` equals `true` in case the aforementioned
variables are uninitialized.

bool invalid = (base_addr == NULL_TREE
                || /* ... */);

What follows is a check on `invalid`:

if (invalid)
  ;
else if (rhs_valid_for_store_merging_p (rhs))
  {
    // ...
  }
else if (TREE_CODE (rhs) != SSA_NAME)
  invalid = true;
else
  {
    // ... (*)
  }

// ...
if (invalid                                                                     
    || /* ... */)
  return terminate_all_aliasing_chains (NULL, stmt);

In total we have that in case `bitpos`, `bitregion_start`, or `bitregion_end`
is uninitialized, then `invalid` equals true and the function returns prior an
access which renders the warnings as false positives.  Interestingly, if (*)
gets removed which cannot be taken in case the variables are uninitialized,
then the warnings disappear which endorses the suspicion of a false positive.

The attached patch fixes the warnings by initializing `bitpos`,
`bitregion_start`, or `bitregion_end` to zero.


More information about the Gcc-bugs mailing list