Use the mask to only enqueue tasks that we want everywhere and not just in scheduler_start()

That fixes bug #59 and is compatible with the new version of scheduler_done() that is in the parallel_sort branch.

I have also made scheduler_unlock() closer to scheduler_done().

See merge request !58

Corrected the IC scripts. 

The initial smoothing lengths were set-up for the GADGET convention not SWIFT. 

That should reduce the time spent in the first time step of the uniform box and perturbed box. 
The Sedov blast script is just modified for consistency with the rest.

See merge request !59

Corrected the IC scripts. The initial smoothing lengths were set-up for the GADGET convention not SWIFT.

Use the mask to only enqueue tasks that we want even in scheduler_done() and scheduler_unlock(). If a dependency is unlocked that is not in the task mask it is simply ignored.

Removed an unecessary condition that GCC optimizes out and produces a warning blocking compilation.

GCC 5.x.y returns a lethal warning when compiling the code:

`space.c: In function 'parts_sort':
space.c:595:12: error: assuming signed overflow does not occur when assuming that (X + c) < X is always false [-Werror=strict-overflow]
         if (jj + 1 < j && pivot + 1 < max) {
            ^
space.c: In function 'gparts_sort':
space.c:735:12: error: assuming signed overflow does not occur when assuming that (X + c) < X is always false [-Werror=strict-overflow]
         if (jj + 1 < j && pivot + 1 < max) {`

After some research it turns out that the compiler notices that the condition "jj + 1<j" is always true and attempts to optimize it out. But it is only true if j and jj don't overflow (which is what we want anyway). So since it is an "unsafe" optimisation, the compiler returns a warning and we die on warnings... 

In this patch I completely remove that condition. Do you agree with that fix ? 

Note that the parallel_sort branch suffers from the same issue.

See merge request !55

Handle running with one MPI rank

Just keep going without a redistribute (which fails) as that is 
not necessary. Issue a warning as this is inefficient.

See merge request !57

Some minor tidying up.

METIS doesn't work in this situation, but there is nothing to do anyway, so just return

Fix up many and varied spelling issues, the main one being "Coypright".

Ran the comments and strings through a spelling check and came up with
all these.

See merge request !54

Cleaning up of the main file

I have:

- Moved the tests in example/test.c to a separate file
- Renamed test.c to main.c for clarity
- Renamed the compilation target to swift_* to avoid confusion

No real code change, just re-organisation and renaming.

See merge request !52

Updated the script fetching the cosmo volume ICs to point towards the updated ones.

I have modified the cosmoVolume ICs to be the output_001 file. They are also stored on the permanent stargate storage.

See merge request !53

Matthieu Schaller authored Oct 27, 2015 and Peter W. Draper committed Oct 27, 2015

(cherry picked from commit 7131dc13 on master)

Issuws with MPICH having task_lock argument in API and multiple inclusions

Re-aassert copyrights

Moved the functions that dump kernels for cross-check to a new file src/kernel.c and the factorisation routine to src/tools.c

Minor fix for GCC 5.2

The newer GCC versions are more strict about signed/unsigned correctness and raise warnings.

See merge request !51

Conflicts:
	examples/test.c

Io fixes

That fixes some issues #52 that appeared on the BlueGene when reading more than 1>>32 particles in parallel or serial mode.

I have also modified the uniform box script to be much faster and lightweight. 

See merge request !48

Ensure GIT_BRANCH non-empty when not on a branch

If HEAD is detached e.g. during `git bisect`, GIT_BRANCH ends up empty, which causes an error at runtime. This change grabs the description from `git branch`, which is either the branch name or something like `(HEAD detached at <hash>)` on recent versions or `(no branch)` in older versions. Works with BSD and GNU sed, and git >= 1.0.0b (at least) although detached HEAD only introduced in 1.5.0.

See merge request !47

By default the code does not dump the task graph any more. This is dumped only hen the option '-y' is passed to the command line.