[Libre-soc-bugs] [Bug 1155] O(n^2) multiplication REMAP mode(s)

[bugzilla-daemon at libre-soc.org]

https://bugs.libre-soc.org/show_bug.cgi?id=1155

--- Comment #21 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
(In reply to Jacob Lifshay from comment #19)

> we can take all the combinations that duplicate something already encodable
> in matrix mode

ah no. not a good idea. the encoding is complex enough as it is.
the point is that this is a RISC encoding not a CISC encoding.
it's right at the critical juncture between decode and issue.

hence why i said *do not* take out the reversing on reduce/sum because
it matches with other inversion, and you actually have to have *extra gates*
to *EXCLUDE* the "i don't see why anyone would want this" logic that
caused you (without consultation) to remove the reverse-direction logic
from remap_preduce.py (and still not yet add it back, as i've reminded
you needs to be done, many times since)


> and just declare them reserved, so at least 000 and 001, icr
> if matrix mode can produce a sequence like 000011112222333300001111...

of course! that's absolutely critical. it's a degenerate case of an X by Y
matrix Schedule-walk, where Y=1. look at remapyield.py - see "skip" mode.

above sequence is produced by

    skip=0b110,x=4,y=1,VL=20 (larger VL for the dots)

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