Discrepancy between cluster and local about niftyreg

[mmodat]

In gitlab by @mmodat on May 19, 2017, 13:51

I am able to reproduce it.
The cluster uses gcc 4.4.7 be default whereas I use clang, gcc 4.9, 5.4 or 6.3.

Checking if the difference comes from there.

[mmodat]

In gitlab by @mmodat on May 19, 2017, 13:54

Adding @emolteni to keep her up to date

[mmodat]

In gitlab by @mmodat on May 19, 2017, 15:00

So on, my laptop, I always have the same result whether I use gcc 4.9, 5.4 or 6.3.

On the cluster, I have the same result whether I use gcc 4.9, 5.3 or 6.2.

[mmodat]

In gitlab by @mmodat on May 19, 2017, 15:29

Verdict: It's coming from SSE

[mmodat]

In gitlab by @mranzini on May 19, 2017, 16:31

Assuming that I am looking at the right path... I am using gcc 4.2.1 locally. Could this be an issue, since it is older than the ones you checked?
If not, we can stick with your verdict.

[mmodat]

In gitlab by @mmodat on May 19, 2017, 17:55

Doesn't look like it should, I'll run more test to make sure I understand exactly where it comes from

[mmodat]

In gitlab by @mmodat on May 22, 2017, 12:40

mentioned in commit 91059d8

[mmodat]

In gitlab by @mmodat on May 22, 2017, 15:23

So ... the difference is the use of SSE2 on one computer (cluster) and SSE3 on another (local).

[mmodat]

In gitlab by @mmodat on May 22, 2017, 15:44

I now obtain the exact same result with SSE and OpenMP enabled on both machines when running the following command:

reg_f3d -ref ori/target.nii.gz -flo ori/source.nii.gz -res test_f3d_res.nii \
        -cpp test_f3d_cpp.nii -lp 1 -maxit 1000

I am now testing the groupwise that Erika tested to extract longitudinal Jacobian determinant maps. It uses f3d2. Will keep you posted

[mmodat]

In gitlab by @mmodat on Jul 18, 2017, 11:50

Been fixed!

[mmodat]

In gitlab by @mmodat on Jul 18, 2017, 11:50

closed