GCC FP fix

Nick Burrett nick at sqrt.co.uk
Fri Dec 3 06:49:32 PST 2004


Peter Naulls wrote:
> 
> Nick (or anyone) could you comment on this?
> 
> I've been looking at bug #119.   John's done a fix whereby it no longer
> takes out the machine on FP crashes, which of course is rather useful.

> The fix appears to be to revert this patch hunk in arm.h:
> 
> @@ -1159,8 +1176,8 @@ extern const char * structure_size_strin
>  #define HARD_REGNO_NREGS(REGNO, MODE)  	\
>    ((TARGET_ARM 				\
>      && REGNO >= FIRST_ARM_FP_REGNUM	\
> -    && REGNO != FRAME_POINTER_REGNUM	\
> -    && REGNO != ARG_POINTER_REGNUM)	\
> +    && (! TARGET_APCS_STACK && REGNO != FRAME_POINTER_REGNUM)	\
> +    && (! TARGET_APCS_STACK && REGNO != ARG_POINTER_REGNUM))	\
>     ? 1 : ARM_NUM_REGS (MODE))
>  
>  /* Return true if REGNO is suitable for holding a quantity of type MODE.  */

You were right about this, it's a silly logic error.  I will commit a 
patch to fix this shortly.  I'm just building and about to test an 
updated compiler with the libstdc++ updates included.  All patches will 
be committed together.


Nick.



More information about the gcc mailing list