[Libre-soc-bugs] [Bug 644] MP3: Basic SV impl

[bugzilla-daemon at libre-soc.org]

https://bugs.libre-soc.org/show_bug.cgi?id=644

--- Comment #10 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
(In reply to cand from comment #8)
> I have not received several mails in that thread. Possibly others missing
> too. Not in spam.

very weird, they're going out. i checked server logs.

(In reply to cand from comment #9)
> On the questions such as "for temporaries t0 t1 etc were
> you planning to drop those into stack, with an offset 16, then
> *reread* them back with an offset of 4?" or the last mail's five loop, I
> can't answer those until I have time to look at the thing properly.

ok.

well i spent several days staring at imdct36, i am confident now that no more
needs to be added to get the "basic" version running. (please ignore the FFT
work i am currently doing, that is for the optimised version)

* sv.add/mrr/m=r30 you can use for adding in *reverse* order, setting
r30=0b010101

this will cover the for-loop for (i=17; i >= ... i-=2)

the first big procesing loop, for i 0 to 1, VL can be set to 2

the second, VL can be set to *five* but the 2nd of each group of operations, a
predicate mask can be set to 0b01111.

my feeling is, it may be best to actually morph the c code a little, to make it
clearer.

if you can add a small test Makefile plus main.c like for the mp3_0 test,
reading and comparing data, i am happy to do some code-morphs that remove the
very last block (out of the forloop 0..3) and increase the loop to 0..4 with
conditional code.

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