English
资源说明：Catalog of floating point software and hardware implementations.
Floating Cat
A catalog of floating point software and hardware implementations.
Copyright 2009 Nathan Whitehead
Released under the GNU General Public License v3 or any later version
CONTENTS
Hardware
Goals
Methodology notes
Stack
Normalization
Rounding
Floating point addition
Hardware : Goals
The goal of this catalog on the hardware side is to be a nice
collection of usable blocks that can be connected together to quickly
create interesting FPGA designs that involve numerical processing.
Floating point is already hard enough to use, so the number one design
goal of the catalog is to make things work correctly all the time,
even at a small cost in performance. The goal of each block is to be
as easy to use as possible and reusable in larger designs for
different purposes.
My personal goal is to experiment with new architectures for high
performance floating point on FPGAs.
Hardware : Methodology notes
All the hardware designs are done in MyHDL and extracted to Verilog or
VHDL. This lets us use Python to do test case generation and
verification. For numerical applications, this is a big win. Many of
the test case generation algorithms are complicated and probably
impossible to express in Verilog directly. I'm a fan of unit testing
and having powerful regression tests to find errors.
The cost is that you have to take some time to understand MyHDL syntax
and Python before you can change the RTL implementations. Note that
you don't have to understand MyHDL and Python to use the RTL itself,
it's only if you want to make changes and have them be integrated into
the project and reverify the result using the Python testbench.
If you have not heard of MyHDL, visit http://www.myhdl.org
Both Python and MyHDL are not hard to learn. For most FPGA projects
it is probably overkill to use MyHDL and do full Python unittests for
every little detail, since you can try things in the Verilog simulator
and on the FPGA to quickly see if it works. MyHDL and Python really
shine when manual inspection isn't enough to verify a module. For
numeric development, you really need a good testing setup; Python
is ideal.
To simplify unit testing in Python, I use the py.test tool. To get
this tool, on Ubuntu install package "python-codespeak-lib".
For other platforms see http://codespeak.net/py/dist/test/
本源码包内暂不包含可直接显示的源代码文件，请下载源码包。
