On Sat, Aug 19, 2017 at 02:24:59AM +0100, Dibyendu Majumdar wrote:
> Hi,
>
> This test program appears to generate incorrect IR.
>
> extern int printf(const char *s, ...);
>
> int main(void) {
> struct { char a:4; char b:4; } x = { 2, 4 };
> printf("a=%d b=%d\n", (int)x.a, (int)x.b);
> return 0;
> }
It's a very surprising bug. It's not a linearization or
an optimization bug as the AST is already wrong.
With a simpler test case, like:
struct s {
char a:4;
char b:4;
};
int foo(void)
{
struct s x = { .a = 2, .b = 4 };
return x.b;
}
you can see that the linearization produce correct
code for the initializer.
You can also see that the return statement to be
linearized is something like
STMT_RETURN
ret_value: EXPR_VALUE (value = 2)
The EXPR_VALUE means that the expression is a constant
(which is indeed the case but I would expect that the
optimization has to deduce this) but it's value is wrong
(a quick check seems to indicate that whatever selected
by the x.<member> expression, it's always the value
of the first member that we get here.
It needs real investigation, though.
It would guess for a bug in the expansion or something.
Nice catch!
-- Luc
--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
