[pvrusb2] HVR-1950 driver loading problem

[storkus at storkus.com]

You're wrong on that: the compiler generates code for the architecture
and, if given, specific processor.

On 32 bit, there are a slew of optimizations and code generating options
for different processors from Intel, AMD, the old Cyrix processors, etc.
 Most distros have their compilations set at a generic default so the
code will run on any processor family and higher.  However, if a
something specific is being thrown in there like "-march=xxx", that
could break something when going from one processor to another.  If
Fedora did something like that, it's their fault.

On 64 bit things are much simpler (for now): as long as you specify
something like "-mcpu=x86_64" (I think this flag name may have changed),
you'll see code that should work on either Intel's EM-64T or AMD's
AMD-64.  OTOH, specifying "-march=amdk8" (if I remember right) will
throw in AMD-specific optimizations that may/will break on Intel.

Finally, I've heard that using different compiler versions can break
ABI's, so be careful.

I use Slackware myself, so I can't comment on anything directly
Fedora-related.

Mike

On Wed, 17 Jun 2009 11:41:07 -0500, "JE Geiger"
<james.e.geiger at gmail.com> said:
> I thought the compiler generated a "family" of code and each processor just executed it
> based on it's instruction set.
> 
> Do you think it would make sense to recompile my kernel using an earlier
> version of gcc, like the one shipped with Fedora 10?