Bug in handling of init-spin block in parallel
Bug #1648053 reported by
Alberto Garcia
This bug affects 1 person
| Affects | Status | Importance | Assigned to | Milestone | |
|---|---|---|---|---|---|
| Siesta |
Fix Released
|
Low
|
Nick Papior | ||
| 4.1 |
Fix Released
|
Low
|
Nick Papior | ||
Bug Description
The attached example shows that assigning opposite spins to two atoms in the init-spin block
gives different "initial spin" values depending on the number of processors. The right answer (0.0)
is obtained with 1 processor, or when the (spin-polarizable) 4s orbitals of the two atoms involved are handled by the same processor.
(Run with 1, 2, 3, and 4 procs and look at the "blocksize". The number of orbitals on each Cu atom is 15).
(Bug exhibited by 4.1-630. Not by 4.0)
Compiler: gfortran-5.2.0 + openmpi-1.10.1 on Mac OSX 10.9; also gfortran-4.6.2 + openmpi-1.6.1 on linux
Revision history for this message
| Alberto Garcia (albertog) wrote | #1 |
| Changed in siesta: | |
| assignee: | nobody → Nick Papior (nickpapior) |
| importance: | Undecided → Low |
Revision history for this message
| Nick Papior (nickpapior) wrote | #2 |
| Changed in siesta: | |
| status: | New → Fix Committed |
| milestone: | none → 4.1-b3 |
Revision history for this message
| Nick Papior (nickpapior) wrote | #3 |
| Changed in siesta: | |
| status: | Fix Committed → Fix Released |
To post a comment you must log in.
Fixed in r631.