lvqcl wrote:
> Martijn van Beurden wrote:
>
> > I just checked on my Raspberry pi (armv6-hf, GCC 4.6) and it
> > looks like decoding is actually faster with these changes. I
> > benchmarked 1b8af6b against f7c52c8, the results are attached.
>
> Interesting results, thanks.
>
>
> (OTOH, GCC 4.6 was released ~4.5 years ago, so it would be also
> interesting to test it on newer compilers - GCC 4.9.x or 5.x,
> or some new Clang...)
I've added a micro benchmark for this is in the microbench/ directory.
If you pull from git and then:
./autogen.sh
./configure
cd microbench/
make
./benchmark_residual
you will get something like this (armhf with gcc-4.9.3):
shift order 2 : 0.000279 0.000284 0.000285 0.000525
mult order 2 : 0.000240 0.000242 0.000243 0.000247
shift order 3 : 0.000408 0.000413 0.000412 0.000415
mult order 3 : 0.000438 0.000443 0.000443 0.000445
shift order 4 : 0.000510 0.000512 0.000512 0.000518
mult order 4 : 0.000661 0.000663 0.000664 0.000669
where the times are; minimum, median, mean and maximum.
This benchmark code will probably only run on Linux. Patches
accepted to fix that.
Erik
--
----------------------------------------------------------------------
Erik de Castro Lopo
http://www.mega-nerd.com/