PATCH:[darwin] fix load of a misaligned double word

Geoff Keating geoffk@geoffk.org
Mon Jan 19 23:14:00 GMT 2004


David Edelsohn <dje@watson.ibm.com> writes:

> >>>>> Geoff Keating writes:
> 
> Geoff> Why would we want to discourage DFmode move operations that use GPRs?
> 
> 1) Because an FP value is not very useful in a GPR
> 2) Symmetric with discouraging int in FPRs
> 3) Because of more limited addressing available
> 
> 	What is the benefit of DFmode in GPRs other than conversion
> routines?

1. If the FP value is not useful, that's register allocation's job to
   find this out by looking at the operations the value will be used
   for.
2. Symmetric with discouraging int in FPRs: encourage float in GPRs.
3. It's reload's job to notice that there is more limited addressing,
   and this is interfering with it.
4. It's faster to copy a value using GPRs than FPRs in many cases.
5. Better for values that change mode.

-- 
- Geoffrey Keating <geoffk@geoffk.org>



More information about the Gcc-patches mailing list