[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Discuss-gnuradio] 3DNow fir test failing under debian with latest u
From: |
Eric Blossom |
Subject: |
Re: [Discuss-gnuradio] 3DNow fir test failing under debian with latest unstable updates |
Date: |
Mon, 22 Aug 2005 11:58:32 -0700 |
User-agent: |
Mutt/1.5.6i |
On Mon, Aug 22, 2005 at 09:08:39AM -0400, Krzysztof Kamieniecki wrote:
> Quoting Eric Blossom <address@hidden>:
> > The test code has an alignment problem under gcc 4.x (that we should fix...)
> > When run within the normal gnu radio framework, the alignment
> > constraint is met and the right answer is computed.
> >
>
> What would be the prefered way of fixing this, #pragma (which might cause
> problems for anyone who does not use GCC), allocate some extra space and the
> add a pointer to the proper aligned starting point (is there a library for
> this
> somewhere?), or some other way?
Probably ought to allocate extra space and adjust the pointer. We
could also use __attribute__(...) on the allocation of the input and
output buffers in the test code.
The issue is that our ccc SIMD code assumes that the input (and I think
output) gr_complex's are 8-byte aligned, which is more restrictive
than the ABI requires for "struct { float re; float im; }"
The problem doesn't occur in the "real code", since the buffers are
page aligned.
Eric