ushort arrays in volatile properties of HDL workers aren't readable for all indices

Bug Summary

The final one or two indices of a ushort array will often return the wrong value.

This wrong value tends to be a value that is near the front of the array.

This is worse when the ushort array starts from a byte offset that is not a multiple of 4 (i.e. it starts in the middle of a 32 bit word).

Workaround

Use a ulong array instead.

Steps to reproduce

A project that causes this issue is available here.

A rough explanation:

• Make a HDL worker with N ushort array volatile properties of increasing length (1, 2, up to N).
  • This covers starting the array in the middle of a 32 bit word.
• Drive a constant to each array index.
  • I wrote the index + 1 to each index to avoid any 0s.
• Drive ctl_out.finished with '1'.
• Create an assembly and application with just this worker.
• Create an application with just this worker.
• Use ocpirun with -d to dump properties.

Environment info

• OpenCPI Version used: release-v2.4.6
• Operating System and version: CentOS 7 + Arch Linux

Possible fixes (Optional)

I've found several potential problems with proptypes.cpp_gen and how the entities it generates interact with the wci infrastructure.

At the bare minimum, the ushort_read_array_property entity uses srl after a width truncation when it should be before. Remedying that fixes the issue when the array aligns with 32 bit words, but not when it starts half way. In that case, the error then moves to break the first index of the array instead.

The reason for that is due to how the byte offset is changed by decoder.vhd. For some reason, when a ushort array starts in the high bytes of a 32 bit word, the byte enable to read the first index is still f when it should be c. This results in the offset losing its lowest two bits, so the ushort_read_array_property is given an incorrect offset to read from.

Whilst I have come up with a rewrite of the aforementioned line from decode.vhd, I think I'm addressing a symptom rather than the root cause.

Hopefully, I'll be able to continue looking into this in the near future.

Acceptance criteria

The example listed functions correctly.