[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Discuss-gnuradio] which architecture is used for SIMD optimized mat
From: |
Josh Blum |
Subject: |
Re: [Discuss-gnuradio] which architecture is used for SIMD optimized math blocks of gr-blocks |
Date: |
Fri, 27 Jan 2012 10:55:37 -0800 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:8.0) Gecko/20111124 Thunderbird/8.0 |
>
> I see that there are some work on the SIMD optimized math blocks in
> gr-blocks. I am wondering what kind of computer is used for your test
> regarding this SIMD optimization.
>
>
Code has been re-based and moved to this branch, BTW:
http://gnuradio.org/cgit/jblum.git/log/?h=new_blocks
The for the floating point version of the multiplier, and adder blocks,
and decimating FIR filter: the work function calls into a libvolk math
kernel. Therefore, the new blocks are not architecture specific.
Libvolk handles architecture selection and architecture-specific
dispatching. In other words, libvolk is responsible for calling into a
vectorized SIMD implementation that will operate on your architecture.
For a given kernel, there is a generic implementation, and optionally,
orc, sse*, neon, etc... To see if an optimized kernel is supported on
you architecture, see the kernel's header in volk/include/volk/*.h
If you have an x86 machine, most likely, there is already an sse*
implementation of a particular kernel.
-Josh