[libre-riscv-dev] [Bug 132] SIMD-like nmigen signal for partitioning

bugzilla-daemon at libre-riscv.org bugzilla-daemon at libre-riscv.org
Thu Aug 15 11:03:27 BST 2019


--- Comment #26 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
So the takeaway summary is: anything that violates the Stage API (uses sync)
can no longer use any of the (many) derivatives from ControlBase.

The multiply unit is not conforming to the API and that severely curtails and
restricts its usefulness, and *will* require *massive* duplication of already
existing proven and tested pipeline classes.

I will do some diagrams and a video, later to show how MaskCancellable and
MultiIn and MultiOut are the core basis for doing the entire ALU: earlyin,
earlyout, loopbacking, ptedication, speculative execution, everything.

Also if we have partitionable Signals, we may even be able to *split* early in
and early out partial results (route part of a mask down one path and the rest
down another) then recombine them back in a MultiIn.

Have to think that one through carefully though. It may be the case that we
have to keep partitioned data together, no matter what.

You are receiving this mail because:
You are on the CC list for the bug.

More information about the libre-riscv-dev mailing list