[Bug tree-optimization/27743] New: Wrong code for ((unsigned) ((a) >> 2)) >> 15

[Erwin dot Unruh at fujitsu-siemens dot com]

A double shift with both constants and cast in between is illegally folded into
a single shift. My first checks indicate that the change made for PR
tree-optimization/14796 on 2005-06-12 clashes with a piece of code earlier in
the function fold_binary. The comment starting with "Strip any conversions that
don't change the mode" is not correct for shifts. The signedness of the left
operand of a right shift is relevant. It determines whether the sign bit or
zero is shifted in.
The problem occurs only if the shifts are within a single expression. Split
accross different expression the behaviour is correct.

Here is a test case showing the different behaviour:

int foo(int a)
{
        int t2 = a >> 2;
        unsigned t3 = t2;
        unsigned t4 = t3 >> 15;
        return t4;
}

int bar(int a)
{
        return ((unsigned) ((a) >> 2)) >> 15;
}

main()
{
        int a = 0xffff3000;
        int b = foo(a);
        int c = bar(a);
        printf ("%x, %x\n", b, c);
        if (b != c)
                printf ("failed\n");
}


-- 
           Summary: Wrong code for ((unsigned) ((a) >> 2)) >> 15
           Product: gcc
           Version: 4.1.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: Erwin dot Unruh at fujitsu-siemens dot com
GCC target triplet: sparc-sun-solaris2.8


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27743