revised x86_64-pc-mingw32 patch

[Richard Henderson]

On Fri, Mar 23, 2007 at 10:02:37AM +0100, Kai Tietz wrote:
> 1) in function_arg_advance_ms_64: The XF registers are not treated proper.

The MSDN documentation doesn't mention XFmode arguments at all.
You'll notice all the ??? comments to that effect in the code.

Are we mistaken in defining "long double" as XFmode?  Should it
in fact be DFmode, just like plain double?  Certainly the documented
default 64-bit rounding mode in the fpcr makes an XFmode long double
less than useful...

>   It seems, that the reuse of examine_argument for the ms abi - as I did - 
> does it better.  the SSE register treating isn't that easy for x86_64 ms
> abi case as assumed.

Nonsense.

> 2) legitimize_pic_address: The case for TARGET_64BIT_MS_ABI is wrong.

Yes, that's a hack.  I plan to rewrite dllimport handling for both
32 and 64-bit from scratch.

> 3) Defaulting for x86_regparm has always to be REGPARM_MAX for this target

... which I did for all 64-bit targets?  Why is this a bullet point?



r~