Re: RFC: failure of gcc.dg/funcorder.c on hppa64

[Jim Wilson <wilson at specifixinc dot com>]

John David Anglin wrote:
> Is there any hope of enhancing cse/gcse to deal with promoted variables
> in a better way?

I think the right answer here is tree-ssa.  We should perform the 
optimization before we expose low-level RTL details, so that we don't 
have to worry about the sign/zero extension.

I can reproduce the same problem with a mipsisa64-elf target, except 
that I get a sign-extension instead of a zero-extension.  When I try 
this on the tree-ssa branch with -O2, the function q is reduced to "j 
e", i.e. a tail call to e, which is as good as we can expect without 
function inlining.  We also get the right result with -O, though the 
function is not optimized to a tail call in this case.
--
Jim Wilson, GNU Tools Support, http://www.SpecifixInc.com