On Wed, Aug 22, 2012 at 3:48 PM, Gianni. wrote:
> On Wednesday 22 August 2012 15:39:59 Mike Frysinger wrote:
>> On Wed, Aug 22, 2012 at 3:22 PM, William Brana wrote:
>> > I have never used 64-bit Windows.
>> > Size of long isn't important, but register count/size and pointer size,
>> > which is same for Windows and Linux.
>>
>> and register/count size is exactly the same. only the pointer size is
>> different, but even that is irrelevant. your logic makes no sense --
>> nvidia isn't going to be held up doing a port to Linux/x32 by
>> *anything* microsoft does (or does not do as is the case here).
>>
>> if you want to see an nvidia/x32 port, then go post to nvidia's linux
>> forums asking for it.
>
> I'm sorry, but what does a *driver* has to do to support x32? After all,
> kernel-space is still x86-64, no?
binary graphics drivers usually have two components -- the kernel and
userspace. you're correct the kernel part needs no changes at all,
but the userland part does. specifically, nvidia's driver package
includes (among other things):
libcuda.so.1
libOpenCL.so.1
libglx.so.1
libGL.so.1
it ships both x86 (32bit) and x86_64 (64bit) libraries so that the
driver works with either userland. without these pieces, you can't
get accelerated hardware graphics with an x32 userland :(.
nvidia isn't unique in this setup -- pretty much everyone doing closed
binary graphics drivers splits it this way. the arm community went
through a painful transition when they changed the userland ABI from
soft floating point (emulating in software) to hard floating point
(using hardware instructions).
-mike