[Libre-soc-isa] [Bug 532] discuss iterative approach for statistical analysis of effectiveness of Compressed PowerISA encoding

[bugzilla-daemon at libre-soc.org]

https://bugs.libre-soc.org/show_bug.cgi?id=532

--- Comment #18 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
(In reply to Alexandre Oliva from comment #16)
> Created attachment 122 [details]
> python script skeleton to estimate compression for a given insn selection
> 
> This script is a skeleton to experiment with the selection of insns for
> compressed encoding, so far only supporting the 8-bit nop switching, just
> because I thought it was easier to model.  

ah very cool.  and that's very interesting: only a 6% compression ratio.
with 35,000 transition-nops, if those were not there, that would be...
err... *thinks*... 3 bits thrown out on the transition in, 8 on the
transition out, that's the C.10bit which... guestimate you can get
away with reducing by 2 bytes... if you take out the "transition" bytes
it's 87% so a further 7% reduction.

there must be more going on here.

> Next, I'll improve/rewrite mode switching and selection insns to use 10-bit
> maybe-switch insns, and 16-bit insns that switch out of compressed mode for
> 1 or n insns, or remain in compressed mode with or without an extra 16-bit
> immediate.  Wish me luck ;-)

:)

can you do me a favour and cut/paste the program rather than "complicate"
it by putting in switches / logic for a completely different scheme?

it would be preferable that each script is entirely separate, minimal,
contextually-focussed on one job and one job only, and consequently
much shorter and clearer.

btw i've simply added the script to the ikiwiki
https://libre-soc.org/recentchanges/

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