Release 0.98

Reorganisation of codebase in preparation of some additional module coming for the next release. Since the reorganisation leads to an overall much more logical structure that is fully compatible with existing Hybrid Fortran projects, it has been put together in this 0.98 release so everyone can get it early.

Release 0.97

• Reductions (OpenACC backend)
• Support for early returns in kernels (OpenACC backend)
• Full import parsing for scalars using CUDA Fortran backend. If you are importing module scalars (both local and from other modules) into your kernels, Hybrid Fortran will transform it automatically into additional kernel parameters (and even pack them into automatic arrays such that the parameter lists don't exceed the CUDA imposed limits). You don't need to specify any directives anymore inside the kernels, only in the module specification part where the scalars are defined.
• Support for array slices and functions within array accessors
• Support for mixed implementation (OpenACC + CUDA backend used in same codebase), using new @scheme directive
• Symbol analysis: Creates alias tree for each symbol for better static analysis
• Support for line continuations at any point in the codebase (please note that Hybrid Fortran supports arbitrary line lengths as well)
• Much faster build time. 
e.g. for a 25k LOC codebase it went from ~15min to ~3min
• Automatic debug output for specified data point - this code has been cleaned up and is now much more stable and usable for any backend.
• More exact and more flexible automatic accuracy tests.
• Several more test examples, you will now find 19 example codes in the examples directory.

Release 0.95

• OpenACC has now become the default implementation for GPU parallelizations.
• Tracing implementations can be used to automatically get to the root of a computational problem.
• Many new code examples are included.