Re: Coupling vs secondary voltage chart

["Tesla list" <tesla@xxxxxxxxxx>]

Original poster: Terry Fritz <teslalist@xxxxxxxxxxxxxxxxxxxxxxx>

Hi Dave,

I "tried" single precision, but the problem is that the program needs to 
"integrate".  It adds really tiny number to the rather large totals.  Ie. 
1.00000 + 0.000000056887467586....  You just need the bug long numbers in 
that case.

Right now, the execution speed is fine.  The code is not very super 
optimized like a real speed programmer would do, but the key is that "it 
works fine" ;-))  The code is also fairly understandable at this point and 
I would not want a fancy optimization to make it unreadable for anyone that 
wants to change it.  So we are giving up speed for the sake of simplicity 
and ease of modification.

If the program ever gets super stable where nobody wants to change it, then 
playing with optimization gets far easier.  But right now, I change it 
about every 15 minutes ;-))

Cheers,

        Terry

At 09:01 PM 6/16/2005, you wrote:
> In a message dated 6/16/2005 7:20:06 P.M. Mountain Daylight Time, 
> teslalist@xxxxxxxxxxxxxxxxxxxxxxx [Terry] writes:
> >The run parameters were:
> >=================
> >ScanTesla-Special  V-6.10  June 14, 2005  Terry Fritz
> C1 2.770000e-008 2.770000e-008 1.000000e-009
> R1 5.000000e-001 5.000000e-001 1.000000e-001
> L1 5.000000e-005 2.500000e-004 1.000000e-007
> L2 7.540000e-002 7.540000e-002 5.000000e-003
> K12 1.000000e-002 9.900000e-001 1.000000e-003
> R2 3.910000e+002 3.910000e+002 1.000000e+000
> C2 4.420000e-011 4.420000e-011 1.000000e-012
> C3 3.830000e-012 3.830000e-012 1.000000e-012
> R3 2.200000e+005 2.200000e+005 1.000000e+003
> T1 0.000000e+000 1.000000e-003 -1.000000e-007
> Vrail 3.400000e+002
> VCpri_init 0.000000e+000
> DwellTime 0.300000e-003
> Current_Limit 5.000000e+002
> BangEnergy_Limit 1.250000e+001
> Goal Type 0
> >=================
> >
>
>     Just a suggestion or two...
>
>
>     Looking at the output here, I see X.YY with exponent. E.g., two 
> digits of precision in floating point.
>
> I am curious if you could speed up the run time by specifying single 
> precision floating point. If I had the tools here I'd give it a shot. Of 
> course, the floating point hardware may be very pleased with Doubles @ 
> 64-bit, and it might not work.
>
> Another possibility would be using something like GCC's fast integer 
> library which emulates floating point by allocating N bits to the regular 
> data and 64-N bits to exponent. It looks to me like this might have 
> trouble, with the range of data (10^ -12 to 10^+5 looks like a range of 
> 10^17).
>
> A final possibility is to trade off memory for speed by pre-calculating 
> what can be pre-calculated. When I was writing a magnetic / electric field 
> plotter this helped a great deal; I precalculed square root lookups for 
> the SQRT [ (X1-X2)^2 + SQRT (Y1-Y2)^2 ], and this helped a great deal.
>
> I hope these suggestions might help, or perhaps trigger a suggestion that 
> does work?
>
>     Thanks,
>
>     Dave