Re: One more oddity with gcc and vax floats...

[Paul Koning <paulkoning%comcast.net@localhost>]

> On Jun 10, 2026, at 12:33 PM, Johnny Billquist <bqt%softjar.se@localhost> wrote:
> 
> On 6/10/26 18:27, Mouse wrote:...
> 
>> VAX float also doesn't have subnormals.
> 
> Also true, but I think we could possibly ignore that one. Just as I think the rounding differences can be ignored, except for test programs that explicitly check for those.

Can you?  I know there are people who care about rounding modes.  Even the crazy "round to even" mode, what sort of drugs were involved in inventing that one?

Then again, it may be people care because they think they should, even though in reality it doesn't matter.  If your code is sensitive to rounding it is skating far too close to the edge and either needs wider floats, or not use floats at all.  ("If you don't understand your problem well enough to scale it to integers, you don't understand it well enough" -- von Neumann)

I remember that CDC was very successful with the 6600 supercomputers even though their rounding is quite broken (I think they round at 1/3 LSB, at least Dijkstra said they do).

	paul