Improving Speed

[Damm, Edward F. (E. Buddy)]

Dave and Chet, you were right on.  Thanks!   Python/FiPy is giving the more accurate double precision answer.  I'll need to change all the numeric constants in the FORTRAN to "d0" to get a real comparision, but as the 2 models stand I have wide difference in the run times on a small test case.  Running for an elapsed 'real' time of 10e-5 seconds, my transformation front moves nominally 20-nanometers.  FORTRAN runtime is 20 seconds, and FiPy is 3 mins, 35 seconds.  My real case will need to go ~10 micrometers, so run time is going to need to be reduced.   Any tips out there on logical steps to improving efficiency would be appreciated.  I'll get version 1.1 and start the work understanding sweeping, etc.   Thanks, Buddy     From: fipy@nist.gov [mailto:[EMAIL PROTECTED] On Behalf Of E David Huckaby Sent: Thursday, June 08, 2006 4:48 PM To: Multiple recipients of list Subject: RE: FiPy Precision Have you tried putting "d0" after your numeric constants in FORTRAN,   i.e. compare       easy = 5.11*5.234*4.987 vs.      easy = 5.11d0*5.234d0*4.987d0   Dave >>> [EMAIL PROTECTED] 6/8/2006 4:22 PM >>> Hi,   I've added a very simply calculation to both the FORTRAN and Numeric calcs, and called it 'easy'.  I've defined easy as ---  easy = 5.11*5.234*4.987   the files are attached, and the output is below.  The discrepancy always seems to show up at the 8th or 9th significant digit for simple multiplication, or for Numeric functions.   FORTRAN   M**p =   5.170837517496476E-002   easy =   133.381011962891   atan term =   -1.01175767177248   PYTHON   M**p =  0.0517083688947   easy 133.38100538   atan term =  -1.01175769857 Any diagnosis is greatly appreciated.   If anyone out there cares to run these 2 and verify the output on their machine, that would perhaps rule out/in the problem being my pc.   Buddy     From: fipy@nist.gov [mailto:[EMAIL PROTECTED] On Behalf Of Daniel Wheeler Sent: Thursday, June 08, 2006 12:36 PM To: Multiple recipients of list Subject: Re: FiPy Precision It depends on the size of the discrepancy. To diagnose this, we need a very simple calculation that compares FORTRAN with Numeric for some very simple array manipulation (as Dan Lewis suggested). On Jun 8, 2006, at 10:48 AM, Damm, Edward F. (E. Buddy) wrote: Hello,   I'm getting some slight differences in some of my coefficients when comparing FiPy to Fortran benchmark code I'm using.  I have confirmed that the equations are the same.  There are some other potential areas that may be responsible for the discrepency, but I wonder if it may be due to precision.    I'm using Double Precision in the fortran code, and for any numbers in FiPy I am ending them in a decimal point and zero if it does not have values after the decimal, and of course anything that does have values after the decimal are input as such.   Is it reasonable to think that my descrepencies are due to precision?  If so, do you have suggestions of how to reduce/eliminate the descrepancy?   Thanks   Buddy This message and any attachments are intended for the individual or entity named above. If you are not the intended recipient, please do not forward, copy, print, use or disclose this communication to others; also please notify the sender by replying to this message, and then delete it from your system. The Timken Company / The Timken Corporation Daniel Wheeler