[PATCH] - Use of powerpc 64bit instructions in 32bit ABI

[Eric Botcazou]

> Does the appended patch work better for you?

Definitely. I couldn't find any obvious cases for which the code still does 
dumb things. If it suits Fariborz' needs and yours, please commit it (a 
clarified comment would be nice too).

> 	The SPARC ABI, especially for records, is complex and I am trying
> to understand how to mitigate the problems you are seeing.

Yes, the 64-bit ABI is a bit intricate, to say the least. Thanks for your 
patience with it :-)

> You need to meet me half way and consider what this change is trying to
> accomplish on other architectures.

I understand. Simply, until very recently (your previous message), I didn't 
know where you really had started from.

> If we make the test too tight to match one specific port or architecture,
> then we no longer have a portable, retargetable compiler.

Note that the opposite statement is also true: if we make the test not tight 
enough, we risk to inadvertently catch unwanted patterns on other 
architectures, as the SPARC64 example clearly demonstrates.

-- 
Eric Botcazou