|
Message
From: virtual_timoliver@l...
Date: Wed, 16 Jul 2003 08:06:16 -0100
Subject: Re: [openrisc] Custom Unit Compiler
Hi,
So the custom unit compiler converts ANSI C to verilog, ok.
I can kind of see what it is doing, something along the lines
of taking a data flow graph and scheduling it in parallel hardware?
It seems to have the intermediate step of converting it to the or1k ISA?
And you say you can choose functions to be converted.
Would it be possible for you to give me an example piece of
ANSI C code that you converted and a quick set of instructions
on how to convert it to verilog using or1ksim's cuc?
Also could you give me some references (books, papers, websites)
about the techniques you used in the cuc?
thanks and regards
Tim Oliver
----- Original Message -----
From: Marko Mlinar <markom@o... >
To: openrisc@o...
Date: Thu, 17 Apr 2003 13:31:52 +0200
Subject: Re: [openrisc] Custom Unit Compiler
>
>
>
> Hi!
>
> On Thursday 17 April 2003 12:16, virtual_timoliver@l...
> wrote:
> > Ok thanks I will have look at that source code.
> > I was also wondering what was the exact intended purpose of
> the
> > custom unit compiler in the architectural simulator.
> It is fully capable ANSI C to verilog converter. It does many
> really heavy
> optimizations (more than modern C compilers) on the source code and
> can build
> very smart verilog models, which can do bursts, etc. At the same
> time it is a
> profiler, so basically you can choose functions which takes most
> time.
> Estimations for hardware resources, timings etc. are calculated.
> Speedup is of course application specific and can range from few
> times to
> several thousands or even more.
> units can be used asynchronously, synchronously or pipelined.
> Waiting is fully
> automated.
> Anyway code is still beta, but I made parts of MPEG4 and test
> programs run on
> it in verilog simulator.
>
> > A brief introduction to its operation would help me greatly to
> > further understand the code :-)
> If you look at the code, you will find even that highly unlikely ;)
> Although
> the code is IMHO well documented is too much complex to explain it
> in just a
> few sentences.
>
> best regards,
> Marko
>
|
 |
Browse
