Optimizing running time

[SebastienRietteMTO]

Copied from SebastienRietteMTO/PHYEX-fortran-tool#13

Profiling of commit ca993f46 is the following on all PHYEX (sxphynh)

--addIncludes : 0.16s
--applyCPP: 0.26s
--deleteNonColumnCalls : 30.8s
--inlineContainedSubroutines : 99.6s
--changeIfStatementsInIfConstructs : 2.1s
--expandAllArrays : 141.3s
--attachArraySpecToEntity : 0.35s
--removeIJLoops : 82.8
--reDimKlonArrayToScalar : 23.1s
--addStack : 26.4s
Total : 407s

Detailed running time on internal functions are still in SebastienRietteMTO/PHYEX-fortran-tool#13

[SebastienRietteMTO]

A test with --expandAllArrays on turb.F90 shows that one third of the running time for this option is lost in checkInDoWhile because of the getParent usage.
A way to suppress this time is to rewrite the removeArraySyntax to recursively enter the different blocs instead of going directly to the a-stmt nodes. This way, we know which node is the parent and if we are in a if statment or in a do while construct

Edit: done in 5faa98f (execution time for this option divided by 4)

[SebastienRietteMTO]

It might be interesting to be able to call fxtran without writing on disk (createExpr, createExprPart, wrapH)

[SebastienRietteMTO]

Look for all calls to fortran2xml (as in variables.addVar) to replace them by expressions.createExpr and add a cache around it (as for createExprPart)