On Sun, Aug 20, 2017 at 9:26 AM, Dibyendu Majumdar
<mobile@xxxxxxxxxxxxxxx> wrote:
>>
>> No. All pseudos are strictly write-once (after the SSA conversion).
>> Memory operations don't write to pseudos. They only use pseudos for
>> addresses.
>
> Thanks. So does that mean that pseudos are SSA from the initial
> linearization phase, and therefore the baseline linear output is
> already SSA? By SSA conversion are you referring to the subsequent
> phase when memory accesses are converted to pseudos where possible?
I wouldn't guarantee it - I *think* it may be the case that even the
initial linearization is actually proper SSA (just with everything
going through memory), but I don't think it's necessarily a real
design choice. We do have a "non-SSA" mode that turns phi sources into
OP_COPY that is very much not SSA.
So our linearization form does support a non-SSA mode, although
honestly, I'm not sure how useful it is.
Luc knows this code much better than me by now.
Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
