[Oberon] Fighting a dragon ...

[Hans Klaver]

Hi Jörg,

Thanks for this multiplication test.

The outcomes are quite remarkable:

PdW's emulator without patch indeed gives one wrong* and two correct answers:
   1.9999998
   1.0000001*
   2.0000002

But Michael's patched emulator gives three wrong* answers (all around 1 instead of 2):
    .9999999*
   1.0000000*
   1.0000001*

For me this last result is unexpected, because apart from the fact that the problems with the even dragon curves are gone, I also did the test that I posted on this mailing list on 8 Oct 2023 (TestRealMul.Mod) and that gave all correct results using Michael's patched emulator (and many wrong results using the unpatched emulator).

I did NOT also apply the change that you suggested in file risc-fp.c. 
Will that make any difference?

Regards,

Hans


On 15 July 2024, at 15:40 Jörg Straube wrote:

> Below you find three different REAL multiplications.
> As the original Verilog files for the FPGA board had an error, PdW’s emulator had the same error as well.
> Without patch, the second multiplications will give a wrong result.
> After Michael’s patch, all three multiplications should work.
>  
> MODULE Multiplication;
> IMPORT Out;
> PROCEDURE Test*;
>                 VAR r: REAL;
>                 BEGIN
>                                r := 1.45052123;
>                                Out.RealFix( r*1.3788146, 12, 7); Out.Ln;
>                                Out.RealFix( r*1.3788147, 12, 7); Out.Ln;
>                                Out.RealFix( r*1.3788148, 12, 7); Out.Ln
>                 END Test;
> END Multiplication.Test