Implement pinned memory for ATI where buffer regions are slow

[Stanislaw Halik]

On 2013-02-17 12:20, Henri Verbeet wrote:
> On 17 February 2013 07:17, Stanislaw Halik <sthalik at misaki.pl> wrote:
>> This one should be better. Sorry about indentation. Much, much FPS, and
>> supposedly no loss of functionality.
>>
> I didn't really review this, but I think any extension that allows
> "... including program or even system termination." from regular user
> space is terrible.

Security impact is nil - Win32 can reference the extension regardless. 
Well-written code won't expose the issue. Patch is a bit messed up WRT 
lack of binding back to 0 in some places, but that'll be amended.

Extension may be messy but gets the job done, unlike present code.

> (And yes, your formatting is also messed up.) I think Stefan has said
> this before, but what really needs to happen is that you (or someone
> else) figure out where the actual issue with ARB_map_buffer_range is,
> and either get that fixed in Wine, or file a bug for fglrx. >

Why is it included for every mapping in the first place, even for 
programs who never need it?

Why not do a provisionary solution if issue persisted since 1.3.x? Not 
to mention performance gains. All I want is to fix a regression so 
there's no need to maintain a local patch set.