LonelyCat124

> That doesn't resolve it because the merging of the symbol tables isn't happening in lowering but during code generation. I think it would be quite difficult to regenerate clauses...

Also my "fix" here only works for code with tasking, if we inlined and then applied other OpenMP directives we'd get the same issue, so this should probably be done...

I also found a weird case when attempting to create a test: ``` def test_failure(fortran_reader, fortran_writer): '''Test with OMPParallelTrans and OMPLoopTrans''' code = '''subroutine to_inline(i, j, k) integer :: i...

I guess then we need to decide which fix we prefer. Either we say this is InlineTrans job and it has to search upwards for a Routine to merge the...

Well its complicated because if we're being language-independent it shouldn't have to, as by the OpenMP standard variables declared inside an OpenMP region are considered to be private automatically (since...

I've been trying a bit to do inlining + OpenMP loop transformation (weird performance results) and I cannot get this to work at all. I attempted to have InlineTrans merge...

> > I attempted to have InlineTrans merge the scopes at a higher level (to avoid the dependence analysis failures), but this fails due to it believing one of the...

@arporter I realised my error when talking to Rupert yesterday. I was trying to merge the symbol tables early (before inlining was complete), which was failing as it was attempting...

@arporter I think this is the issue I mentioned yesterday.