On Thu, Sep 17, 2015 at 1:56 AM, West, Nathan
wrote:
>
>
>> I've nagged people about this before, but I'd like to make this an
>> official thing: Put this into the VOLK docs (i.e. state in the contract
>> that in- and output buffers may be the same) and then include that in
>> the unit tests, so w
On Wed, Sep 16, 2015 at 8:20 PM, Martin Braun
wrote:
> On 16.09.2015 13:29, West, Nathan wrote:
> > There is a volk_32f_s32f_multiply_32f. It doesn't operate in-place, but
> > almost none of the VOLK kernels do. I think it's safe to give the same
> > output buffer as input buffer. (I've heard tha
On 16.09.2015 13:29, West, Nathan wrote:
> There is a volk_32f_s32f_multiply_32f. It doesn't operate in-place, but
> almost none of the VOLK kernels do. I think it's safe to give the same
> output buffer as input buffer. (I've heard that doing stuff in-place is
> noticeably better, but I've never t
On Wed, 2015-09-16 at 16:29 -0400, West, Nathan wrote:
> On Wed, Sep 16, 2015 at 3:58 PM, Dennis Glatting
> wrote:
> >
> > It would also be nice to have a VOLK kernel that multiples a vector
> > by a constant:
> >
> >void
> >volk_32f_s32f_multiply( float* vecbuffer,
> >
On Wed, Sep 16, 2015 at 3:58 PM, Dennis Glatting wrote:
>
>
> It would also be nice to have a VOLK kernel that multiples a vector by a
> constant:
>
>void
>volk_32f_s32f_multiply( float* vecbuffer,
>const float scalar,
>unsigned int n
On Wed, 2015-09-16 at 12:49 -0400, mle...@ripnet.com wrote:
> Sure, but if the flow-graph basically is decimated 10:1 by the time
> it reaches the log10, improvements in log10 are going to have very
> little impact. It is, I would assert, boldly and perhaps brashly,
> that log10 operations almost
Sure, but if the flow-graph basically is decimated 10:1 by the time it
reaches the log10, improvements in log10 are going to have very little
impact. It is, I would assert, boldly and perhaps brashly,
that log10 operations almost never need to be done at "line rate",
since they are an artifice
On 15.09.2015 20:35, Marcus D. Leech wrote:
> Ordinarily, one does a log10 to convert into engineering units at the
> back of, for example, a power-measurement chain.
>
> There's usually no reason to do that in the middle of a flow-graph,
> where things can stay in linear units.
That's true, but
This is the use case I had in mind when we were working on the log2 kernel.
Binary logs are the fastest log to compute (for floating point numbers)
because of Marcus' explanation. I even wrote the NEON log2 proto-kernel so
that the computation can be in-lined in other kernels in the future to
avoid
By the way, with a bit of approximation and a little acceptance for
errors in corner cases, log2 of a IEEE-754 float is simply it's exponent
+ log2(mantissa), with the later potentially being implementable using
lookup tables.
Also, VOLK does have a log2, kernel, volk_32f_log2_32f
Best regards,
Ma
On Wed, 2015-09-16 at 00:32 -0400, Brian Padalino wrote:
> On a side note, this is extremely impressive and I wish I knew about
> it sooner:
>
> https://software.intel.com/sites/landingpage/IntrinsicsGuide/
>
> There's a few different log10_ps (packed single) functions in there.
>
That's real
On Wed, Sep 16, 2015 at 12:24 AM, Dennis Glatting wrote:
>
> 1) Working with VOLK to learn VOLK.
> 2) Having fun with vectors.
> 3) Generating power data points for plotting across a selected
>set of samples.
Lastly, another alternative, if you wanted to utilize what is already
there in VOLK,
On a side note, this is extremely impressive and I wish I knew about it sooner:
https://software.intel.com/sites/landingpage/IntrinsicsGuide/
There's a few different log10_ps (packed single) functions in there.
Brian
On Wed, Sep 16, 2015 at 12:24 AM, Dennis Glatting wrote:
> On Tue, 2015-09-
On Tue, 2015-09-15 at 23:35 -0400, Marcus D. Leech wrote:
> On 09/15/2015 11:15 PM, Dennis Glatting wrote:
> > With the VOLK library, is there a way to compute the log10() of
> > each
> > 32f in a buffer?
> >
> > That is:
> >
> > for( int i = 0; i < num; ++i )
> > buf[i] = std::log10( buf[i
It sounded like a fun and interesting question, so apparently there is
a little post from 2007 about a new implementation in C licensed under
GPLv2:
https://sourceware.org/ml/libc-alpha/2007-06/msg00124.html
It would be interesting to compare and see if the results are
beneficial even if no VOL
On 09/15/2015 11:15 PM, Dennis Glatting wrote:
With the VOLK library, is there a way to compute the log10() of each
32f in a buffer?
That is:
for( int i = 0; i < num; ++i )
buf[i] = std::log10( buf[i]);
I only see log2() in the library but don't know if there is an easy way
to compute lo
With the VOLK library, is there a way to compute the log10() of each
32f in a buffer?
That is:
for( int i = 0; i < num; ++i )
buf[i] = std::log10( buf[i]);
I only see log2() in the library but don't know if there is an easy way
to compute log10().
Thanks.
__
17 matches
Mail list logo
