[Libre-soc-bugs] [Bug 238] POWER Compressed Formal Standard writeup

bugzilla-daemon at libre-soc.org bugzilla-daemon at libre-soc.org
Sun Nov 29 13:24:16 GMT 2020


--- Comment #93 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
(In reply to Alexandre Oliva from comment #89)
> I had not noticed before writing comment 16 in bug 532, but the 16-bit
> immediate compressed encoding is very very similar, in function and
> possibilities, to the extend-next-insn opcode I'd suggested before.

what you propose sounds perfectly reasonable for single-issue designs of
10-20 years ago.

i really, really do not want this proposal to go down the route of
adding yet more complexity at the 16-bit level, re-expanding length
based on deep-dive multi-bit selections:

* first level EXTNNN selection analysis
* second level FSM selects 16-bit
* third level selects bit0=1,bit15=15 to indicate compressed.immediate mode
* FOURTH level BACK to 32-bit (or greater) based on multi-bit brownfield
  encoding patterns?

no.  this is too much.

the complexity is too great for high performance multi-issue designs
to withstand.

length-mode decoding really must be kept to the absolute bare minimum, with
no "bleed-back" between 1st pipeline stage decode and 2nd stage.

SV is already going to require a *THREE* stage instruction decode.

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

More information about the libre-soc-bugs mailing list