Re: [Discuss-gnuradio] Volk library invalid opcode exception

[Tom Rondeau]

On Sun, Apr 15, 2012 at 11:51 AM, Marcus D. Leech <address@hidden> wrote:
> On 04/15/2012 11:45 AM, Tom Rondeau wrote:
>>
>>
>> Yes, so the vmovss is an AVX instruction (the AVX version of movss),
>> but your processor doesn't have AVX according to your flags above.
>> Except that it does. According to Intel, the i5-2540M processor
>> supports AVX, but your OS isn't recognizing the avx flag in
>> /proc/cpuinfo. The Volk build process asks the processor directly for
>> the flags that it can use.
>>
>> I really think this is a problem with Xen (or at least something in the
>> setup).
>>
>> Tom
>>
> So, how is this going to play out with packaged-binaries?  If the decisions
> about which instruction sets to use are made at compile time,
>  you could end up with packaged binaries that aren't portable, and will blow
> the heck up.  Or am I mis-understanding what you mean
>  by "at build time"?
>
>
> --
> Marcus Leech
> Principal Investigator
> Shirleys Bay Radio Astronomy Consortium
> http://www.sbrac.org

No, you didn't misunderstand, I misspoke. Josh and the other Volk
developers figured this out already. The system builds libraries for
all intrinsics that the compiler supports. It's at run-time that the
correct machine is chosen from the list of what's available and what
your system can support.

I'm not sure if I'm explaining that exactly or well, but the problem
you brought up was considered in the design to allow for us to do
exactly that with Volk. I think in this case, the OS and the actual
processor are at odds with what they can do, causing the problem.

Tom