| Author |
Topic |
|
|
jjmackay
24 Posts |
 Posted - 13 Feb 2014 : 13:21:25
|
I have a calculation that give two different result on a PC and on a GCX. The result on the GCX is really strange.
On the GCX
(those are all stored in float variable)
float ParkTime;
float CostPerUnit;
float SecondsInHour;
float TotalCost = ParkTime*(CostPerUnit/SecondsInHour);
printf("ParkTime: %f. CostPerUnit: %f. SecondsInHour: %f. TotalCost: %f.", ParkTime, CostPerUnit, SecondsInHour, TotalCost);
ParkTime: 3600.000000 CostPerUnit:60.000000 SecondsInHour: 3600.000000. TotalCost: 60.000004.
On a PC, it gives 60.000000; which is my expactation.
The code is compiled using VS 2005 and am using those options (set using the Project settings):
/ Op and /fp:strict
I get this warning while I compile:
"cl : Command line warning D9002 : ignoring unknown option '/Op'"
I have also tried to use
#pragma float_control(precise, on)
#pragma fenv_access(on)
#pragma float_control(except, on)
before function that uses the float variable.
I understand the imprecision of floating point number, but is there a way to control it? |
|
|
akidder
1519 Posts |
Posted - 13 Feb 2014 : 13:27:24
|
| In this case, printing with %0.2f would probably meet the needs of the application. But this level of precision *is* a bit low. |
 |
|
|
beitman
63 Posts |
Posted - 13 Feb 2014 : 13:29:02
|
A couple of quick suggestions:
1. Try using double
2. Change the calculation to either (but probably a is better):
a. TotalCost = (ParkTime* CostPerUnit)/SecondsInHour;
b. TotalCost = (ParkTime/SecondsInHour) * CostPerUnit;
|
|
|
|
jjmackay
24 Posts |
Posted - 13 Feb 2014 : 14:06:32
|
Using double instead of float did work.
Thank you for this.
Are you able to provide an explanation or a technical document that would explain that difference on the GCX. This would help for us to determines that what extends with need to rerun a set of test.
Thanks,
|
|
|
|
beitman
63 Posts |
Posted - 13 Feb 2014 : 14:18:31
|
I suspect that the difference in data size between your PC and the GCX is where the problem lies. That, and the PXA255 does not have a floating point processor.
This is further complicated by the order of operations in the calculation of TotalCost. By first dividing, which results in 0.01666666666666666666666666, then multiplying you end up with the "weird" result. I find that these results can almost always be changed by changing the order of operations and/or the data size.
Is there a document that compares and contrast the differences between a PXA255 and your x86? No, probably not. |
|
|
| |
Topic |
|
|
|
All Forums
Microsoft Windows CE
Weird result with floating point calculation.
Forum Locked
Send Topic to a Friend
Printer Friendly