Christos Zoulas wrote:
> In article <45F714B3.3080707@softjar.se>,
> Johnny Billquist  <bqt@softjar.se> wrote:
> 
>>We really should have GCC always return true for finite() I guess, since 
>>that's what the function does anyway, and then gcc could optimize correctly.
>>
>>But it seems we need someone with knowledge of gcc to fix a number of 
>>issues that we have at the moment...
>>If just gcc would work right for us, then I think we should actually 
>>have a fairly good base at the moment for the VAX.
> 
> In real.c if (exp == 0) in all the decode_vax_?() functions don't put
> anything in the r-> stuff, they just leave it 0'ed. Is that ok?
> I understand exp == 0 means really -128, but...

On the VAX, if exp == 0 and sign == 0, then the value is to be taken as 
literally zero, no matter what the other bits are.
If exp == 0, and sign == 1, it's a NaN.

So I'd say that gcc is probably doing the right thing here. (I don't 
think they every play around with NaNs...)

	Johnny