Validate and test gr0 exclude flag for Power

[aviansie-ben]

A number of instructions on Power have special behaviour where placing gr0 in the RA field of an instruction may be invalid or result in use of a constant 0 instead of the actual value stored in gr0. In #5325, an instruction property flag was added to control this behaviour, however we have not yet validated that the existing list of gr0 excludes is correct nor added tests to ensure that this functionality works as intended.

Generally, validation of these excludes should follow the same rules as validation of binary encoding: anything with a current format of FORMAT_UNKNOWN need not be checked, as this validation should be performed as support is added for using these instructions in the backend and new tests should be added as this happens.

[gita-omr]

Maybe we can somehow parse the Power ISA document to make sure we found all instructions that have that property. My main concern is that somebody may rely on this flag already being set on all instructions that need it.

[aviansie-ben]

I don't see any reliable way of automatically parsing the ISA document to determine this. None of the appendices seem to list this information in tabular format and even the instruction descriptions themselves are inconsistent in how they provide this information (i.e. some use the (RA|0) convention, while others simply have special handling in the pseudocode for RA=0, and still others only mention in the description that using 0 for RA is invalid).