Over-eager code elimination?

[Dan Sheridan <djs@xxxxxxxxxxxxxx>]

Running test-linearize on some of my code, I seem to be losing an
assignment after a break.

In this minimal test code, y should have the previous iteration's value
of x, so the value returned is the difference between the final two
values from fetch():

int fetch(int);
int test(int v) {
  int x, y;
  for (int i=0; ; i++) {
    x = fetch(i);
    if (v < x) break;
    y = x;
  }
  return x-y;
}

test-linearize produces:
test:
.L0xb7db300c:
        <entry-point>
        br          .L0xb7db3034

.L0xb7db3034:
        call.32     %r2 <- fetch, $0
        setlt.32    %r5 <- %arg1, %r2
        dead        %r5
        br          %r5, .L0xb7db3084, .L0xb7db3034

.L0xb7db3084:
        dead        %r2
        sub.32      %r11 <- %r2, %r2
        dead        %r11
        ret.32      %r11

So it thinks that both x and y are in %r2. However, if y is initialised,
the assignment happens as expected.

Any ideas?

	Dan.



-
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