fixed #616

[davideschiavone]

Fix issue #616

As this is a bug in the doc, it may lead to erros in the compiler porting as well. Thus, pinging @jeremybennett as well

[davideschiavone]

hi @jeremybennett , did you have time to have a look at it?

[jeremybennett]

This much more needs advice from @jessicamills and the GNU tools team. In particular the work on the PULP load/store extensions has been done by @NBruschi and @EnricoTabanelli assisted by @gtagliavini

My immediate thoughts are that the work to date has been primarily on the assembler and builtins. For the assembler the semantics of the instruction don't matter, but we need to ensure that signed literals are permitted in the syntax for the assembler. For the builtins it will affect whether signed or unsigned types are used for parameters.

There may be impacts on the linker if there are any relocations impacted - for example if the index were symbolic, then potentially we need a signed rather than unsigned relocation.

I don't believe we have any committed work on GCC, where this will be more significant. However I think @NBruschi and @EnricoTabanelli have been working on this, and should comment on the impact.

I'm happy to approve, since this comment is about ensuring that those impacted in the tool chain have an opportunity to consider the implications for their existing and future work.