Double indirection causes bad loop performance

The attached test case contains this loop.

    for (i=0; i<10; i++)
    {
        p1 = *list;
        p2 = *list++;
        *p2 = *p1 + 2;
    }

Build it ...

% cl6x -mv6740 -o3 -s -mw testcase.c

Then inspect the resulting assembly file.

All of the pointers are restrict qualified. Including list and *list. Despite that, the compiler generated comments indicate a dependence that is carried over from one loop to the next. The II (initiation interval) is 10, because each loop has to finish before the next one starts.