Re: Initial E500 backport

[Aldy Hernandez <aldyh at redhat dot com>]

On Monday, April 21, 2003, at 04:58  PM, Zack Weinberg wrote:

> I've committed the first cut of a backport of the E500 (PowerPC
> subtarget) to the gcc-3_3-e500-branch.  Appended to this message is
> the change log; I'm omitting the diff as it is very long and consists
> entirely of patches mined out of the mainline.
>
> Aldy, I'd appreciate it if you could cast a glance over the change log
> and tell me if I missed anything.

Wow, those are more changes than I remember ;-).

On a glance, it seems you have all the changes I recall.

Question... The e500 ABI specifies that long doubles are either 128 
bits or not implemented at all.  On the FSF tree I have left them as 
64-bits.  Setting rs6000_long_double_type_size to 128 doesn't work 
because, last I checked, there was no libgcc* bits for 128 bit 
emulation (for PPC anyhow).  On an internal Red Hat tree, I have set 
rs6000_long_double_type_size to 128 (see eabispe.h) and tweaked 
fp-bit.c from t-spe to disable the TF stuff.  This causes a gazillion 
regressions, but achieves the functionality Motorola wants (long 
doubles are 128 bit, but are not implemented).

You may opt to uncomment the relevant bits in eabispe.h to be ABI 
compliant.  I have chosen not to do so on mainline.

Aldy