| Re: [AD] 4.2 final cleanup |
[ Thread Index |
Date Index
| More lists.liballeg.org/allegro-developers Archives
]
On 2005-09-25, Peter Wang <tjaden@xxxxxxxxxx> wrote:
> The documentation for fixmul specifically talks about overflow, so we
> need to restore that behaviour, or change the semantics and update the
> documentation.
>
> The following patch:
>
> 1. For Intel machines (and compilers without long long) revert to the
> original implementation of fixmul (i.e. fixmulf). I don't know how to
> detect an overflow with only 32-bit registers, and when I checked it
> again fixmulf was faster (on my P4) than fixmuli anyway!
>
> 2. For everything else, I made fixmull detect overflows. It's much
> slower as a result, but still faster than fixmulf on an AMD64.
Commited.