Starpu/alloc on the fly

Modify the way the workspaces are allocated to be able to allocate them on the fly.

The objectives are:

• Allocate less memory in QR like algorithms, as only the useful T tiles are allocated.
• Be more asynchronous in some algorithm as QR again, or norms by avoiding the required sequence_wait at the end of the call before freeing the allocated workspaces. This is also used in the upcoming SUMMA algorithms.

The changes are:

• switch geadd to axpy in norm computations as they may be optimized with mkl
• switch workspaces from global allocation to tile allocation
• update QR kernels that generating the T tiles to set it to 0 first. This can not be done through global memset anymore, and to avoid an complete allocation of the matrix, this is moved in the codelets to initialized only the touched tiles.

changed milestone to %Chameleon 1.0.0

added Feature StarPU labels

@all is some of you can make a quick review that would be nice.

@viroulea I noticed that some of the depends clause in the OpenMP codelets are incorrect. They need to be all checked and in some way 'standardized' because right now, they all have a different writing.

mentioned in issue #34 (closed)

added 1 commit

• 16b8c4f3 - Fix openmp codelets

Compare with previous version

added 1 commit

• d031e74a - Cleanup depend clauses in openmp

Compare with previous version

added 1 commit

• 16b8c4f3 - Fix openmp codelets

Compare with previous version

changed the description

Thanks a lot @faverge for noticing this!

Indeed the dependencies length were very wrong, I thought I caught them all when rebasing my original PR... Thanks for fixing them!

added 1 commit

• c100ef0f - Factozize LQ/QR kernels as suggested by @viroulea

Compare with previous version

resolved all discussions

enabled an automatic merge when the pipeline for c100ef0f succeeds

merged

mentioned in commit fa6d78a3