------- Comment #4 from rask at sygehus dot dk 2007-07-07 16:08 ------- In reply to comment #1: If you're splitting a multiword subreg (such as %rbx:%rcx) after reload, then the prologue/epilogue code has no way of knowing that all use of %rbx is later optimized away. I don't know if this is what's happening here, but that would be the first thing I'd look for.
In reply to comment #2: For both GCC 4.1.x and 4.2.x, the status is "regression fixes & docs only", so unless some older release produced better code, this won't be fixed in GCC 4.1.x or 4.2.x. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32662
