NIS software testing (OS X vs SGI)

testing using nisdiff and matlab
Using 48 time series tap dataset taken from /usr/ccn/pkg/nis-3.4/demo/images on tesla(SGI)

nis: airalign
Generated images were binary identical.
nisthresh
All but thresh_00043.img were same.
The generated threshold for thresh_00043.img was 1098 on mac and 1099 on tesla.
Mean (mac) : 439.1308   Mean(SGI)  : 439.0973
Voxel num different: 37793 and 39247
Max diff : 1098
Results are almost identical on SGI and MAC and the small difference is due to rounding of floating point values.
nismaps
mean : Generated images were same same.
Variance : Generated images were same.
nisnorm
All images were same.
niscontrast

niscorrect
The generated images were same when baseline correction, detrend and outlier correction was done separately.
nisfilter
20 out of 48  generated images were different
12 of those 20 images  had 1 voxel different.
7 of those 20 images had 2 voxels different.
1 of those 20 images had 3 voxels different.
The difference was 1 for all cases.
This difference is too small to affect the output is is possibly caused due to rounding of floatinf point values.
nisanova
The generated images were same.
nisroi
The generated images were same.
nisoverlay
The generated images were same.
niscrit
The generated files were same.
nisdiff
The results were same.
SliceApp
The results were same.
RoiApp
The results were same.
RoiDesign
The results were same.

Group Transform Testing
The Detrend images from krcd01 krcd03 krcd06 krcd15 from /krcd/imaging on tesla(SGI) were used for testing.
Krcd15 was the reference subject.

results
CrossRegister
The values in the air_files were smiliar upto the last floating point digit.
airdiff could not be used as the airfiles were compiled with 64 bits on SGI's and  32 bits on Mac. The resulting airfile size were different on the two OS.
Reslice
The results were binary identical
Norm
The results were binary identical
Smooth
The results wree binary identical