Re: [PATCH] Hexagon: move GETPC() calls to top level helpers

[Richard Henderson <richard.henderson@linaro.org>]

On 7/5/23 14:45, Matheus Tavares Bernardino wrote:
>> Taylor <ltaylorsimpson@gmail.com> wrote:
>> Note that check_noshuf calls HELPER(probe_noshuf_load) and
>> HELPER(commit_store).  Both of those call GETPC() from within.  So, you'll
>> need to pull the contents into separate functions that take ra as an
>> argument.
> 
> Ah, good point. It was my understanding that, in case of a memory
> exception in one of those nested helper calls, the GETPC() we would want
> to use for unwinding was the one from the most recent helper. I'm still
> trying to wrap my head around these concepts, though, so I might have
> misunderstood it. Is this not the case?

No, it is not the case.

GETPC fetches the return address from the current function.

The unwinder which uses this value needs the return address *into* the generated code.

Therefore the HELPER function that is directly called from generated code is the place at 
which GETPC must be used, and nowhere else.  The corollary is that one HELPER should avoid 
calling another HELPER, Just In Case -- use separate intermediate functions instead, as 
Taylor suggests above.


r~