[Libre-soc-dev] [RFC] Matrix and DCT/FFT SVP64 REMAP
Luke Kenneth Casson Leighton
lkcl at lkcl.net
Tue Jul 6 14:28:31 BST 2021
folks let's take openpower-hdl-cores off of cc for now.
when the OPF ISA WG kicks into gear we can always have these kinds of
additional uses for REMAP include simply being able to access
registers sequentially in large regular jumps, without needing a
r0 r3 r6 r9 r12
can be accessed with VL=5 and a 2D structure X=3,Y=5. this without
having to MV the registers first.
using a predicate would be wasteful because VL=15 and
predicate=0b001001001001001, 66% of the operations issued would be
the caveat here: REMAP is between *issue and execution* and it is not
necessarily the case that the execution engine has the ability to
efficiently or optimally move register data around.
the execution efficiency is *not* part of the spec in other words.
just as the efficiency of scalar execution is not part of an ISA spec.
More information about the Libre-soc-dev