Re: Optimization problem on PPC when casting integers to floats

[David Edelsohn <dje at watson dot ibm dot com>]

	This is a fundamental artifact of the way that the int to FP
conversion is designed in the PowerPC port of GCC.  Current the backend
believes that the stack slot is modified in each iteration and that the
value cannot be preserved.

	This is not correct and should be fixed, but it will take time and
study to understand how to restructure the GCC machine description to
avoid this problem.  The machine description needs to assert that the
unchanging part of the stack word truly is this constant value which can
be hoisted out of the loop.

	Thanks for pointing it out.  Don't expect an overnight patch to
solve it.  Hopefully it can be address in the forthcoming gcc-3.0 release.

Thanks, David