XScale port contributed

[Richard Earnshaw]

nickc@redhat.com said:
>   This is to let you know that I have checked in the patch below to
>   add support for Intel's XScale processor to the ARM port.  The
>   XScale is a v5te ARM with some Intel specific extensions. 


Excellent.  This is great news.

Some initial comments:

+ xscale-*-elf)
+ 	tm_file=arm/xscale-elf.h
+ 	tmake_file=arm/t-xscale-elf

what do these processor-specific headers/makefile fragments give us that 
can't be achieved with --with-cpu=xscale (or by synthesising this up in 
configure)?  The reason I ask is that they make layering other OS support 
on top much harder.

+   if (TARGET_HARD_FLOAT && (tune_flags & FL_XSCALE))
+     {
+       warning ("XScale does not support hardware FP instructions.");
+       target_flags |= ARM_FLAG_SOFT_FLOAT;
+     }
+ 

This is wrong.  The tuning should *never* alter the set of instructions 
that can be selected from.  It must *never never* alter the ABI generated. 
 The above code does both :-(

You are in effect saying that with xscale you can never use a floating 
point emulator.

R.