Open Mingku-Chang opened 6 months ago
You are right vwsll
with SEW=64-bit (vsew=3
) would mean an effective element width of 128-bit (EEW=2*SEW) which is not currently supported so this behaviour is reserved.
Note that they are other possible restrictions, for example if ELEN=32 then SEW=32-bit is also reserved (Since Zvbb
could be supported alongside Zve32x
for example). In that case I am not sure a specific mention of SEW==64
really make sense since it is not the only possible limitation. We could (should ?) mention that this instructions inherits element width constraint from RVV.
@kdockser for your consideration
Is this recommendation still valid? If yes, please transfer issue to riscv-isa-manual repo or let me know to do it. Thanks.
I guess that the VWSLL operation with (SEW == 64) is not allowed. Vector Crypto follows a lot of regulations from Vector. Therefore, results can't exceed 64-bit width. I recommend adding some statements to show readers that Vector has such a regulation.