On 09/08/2023 13:15, Konrad Dybcio wrote:
Hmm.
Well from earlier in the thread the question "why do we have these compat strings" is because we can have any combination of encoder/decoder assigned.
If there's a cogent argument_still_ to be made to transition to some new way of assignment then fine so long as we don't break that basic flexibility.
Though my own €0.02 is that a module parameter is more of a PITA than a compat string.
OTOH I could make the argument, that the high probability is most people - probably all, just instantiate a single encoder and decoder and aren't aware of or using the inbuilt flexibility.
@stan probably has the right idea what to do.
Actually..
Has anybody tested this, ever, with the mainline driver?
I assume Stan has.
Do we have anyone using this?
Can't say.
Is anybody willing to maintain that, test for regressions and
fix them in a reasonable amount of time?
If we don't have at least 2x "yes" here, I don't think it makes sense
to worry about it..
Hmm.
We decide if we are encoding or decoding when we init a session and the
blocks are symmetrical. The hw blocks themselves are not bound to a
particular encode/decode mode.
Having two parallel encoders or decoders is exactly the same effort as
having a parallel encoder/decoder.
We don't test parallel encoding/decoding but we should. I'd not be
surprised to find there are bugs but, that's not a reason to exclude
rather to find and fix bugs.
---
bod
