Re: [fpu] fdiv problem

[Jamil Khatib <jamilkhatib75@yahoo.com>]

> >> The test pattern generator generates proper
> floating
> >> point
> >> numbers and results. It appears that you don't
> >> understand how
> >> to integrate it into a test bench and how to
> write a
> >> test
> >> bench.
> > 
> > yes I do not know how to use it in the testbench
> > forgive me I am very new to verilog.
> 
> It is irrelevant what tool you use, VHDL or verilog,
> it appears to me that you do not understand the
> basics.

Yes the language is irrelevent if I do not know the
inputs format????

> 
> > 
> > You need to learn to look at floating point
> >> numbers in
> >> their native IEEE representation and understand
> what
> >> it all
> >> means.
> > 
> > what is the reason for that?
> 
> Because that is the standard. That is what every
> IEEE
> compliant fpu uses.
> 

Correct, but humen are created by God not by IEEE
 
> > Finally when I ran the program in verbose mode and
> > after following the code I found that you are
> > generating the exceptions and the results
> > 
> > You only provide the result from library and
> ignore
> > the native result
> 
> If you would have read my previous emails, I would
> not
> have to waste my time and explain it to you again.
> 
> The reason I am ignoring the host results are:
> 1) I do not know how to set the rounding mode in C
> 2) I do not know how to catch exceptions in C
> 3) The author claims that SoftFloat is 100% IEEE
>    compliant. At this point I believe it.
> 

Thanks for your time. this is because of miss of
documentation

> > The code seems to generate the same test vectors
> again
> > and again for the same parameteres, why???
> 
> Read the help screen. When you use -p option, it
> will
> use pre defined values from a table. If you want
> random
> code you should omit -p, then it will generate
> semi-random
> code. I do not change the seed, so the random code
> will
> always be based on the same seed. That's something
> that
> should be an option in the program. I will add it
> later.

thats right but I used the following command line
pg -n 10 -v -o pattern.vec -r 0 -m 8

see no -p !!!!!!

> 
> > Do you depend on the test bench to decide if the
> > result of the core is correct or on the software?
> 
> The test bench. It has to compare the result from
> the core
> against what was provided by the program. In some
> cases,
> when I'm not sure who is correct, I do the math with
> a
> pencil and paper, and reread the IEEE standard. So
> far the
> program was always correct.
> 
I know that
you have to dig in (i do not know how many digits )
hex numbers

> > I hope so, but still I have to do rounding that
> you
> > said you are working on, I do not want to have
> > different results
> 
> Beyond rounding, you need to make sure that INF, and
> NAN
> operators are handled properly. You also must
> generate
> all exceptions, they are generated differently for
> each
> block.
If you read also my emails carfuly then you should
know that the code handles NAN and INF ( invalid and
div by zero exceptions) what I do not have is the
overflow,underflow and inexact, have you tested them?

The rounding should be also the same for all results
using also the same internal representations

> 
> > Anyhow I give you always the status of my work but
> no
> > one give his work status, is not it a cooperative
> > work?
> 
> I'm in constant communication with Dmajan, at this
> point

So that's fine if the Opencores is Damjan's work
project only. if this kind of work will lead to
private work (or advantage) for anyone I'll not
continue working here

It seems that from the begining that you want to work
alone on this project

I should know that from the begining so as not to
waste my time

> I feel it is more important to ensure that he can
> use the
> fpu.
So he is more important than the opencores community

I just want to tell you that you lost the spirit of
the open hardware design.

> 
> 
> > Can you please be kind and resend the test bench
> > interface again?
> 
> Attached is my entire test bench.
> 

Thanks

> >>> Jamil Khatib
> >> 
> >> rudi
> 
> rudi
> 


__________________________________________________
Do You Yahoo!?
Yahoo! Mail - Free email you can access from anywhere!
http://mail.yahoo.com/