[Libre-soc-isa] [Bug 701] document Matrix REMAP in SVP64

[bugzilla-daemon at libre-soc.org]

https://bugs.libre-soc.org/show_bug.cgi?id=701

--- Comment #13 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
andrey it looks good, esp. the images. must convert them soe time.

 however the explanation of svshape is unnecessarily
long, duplicating content that is already in thespec. it is a distraction
(and is "spec information" style not "readable" style writing)

also this is wrong

   For Matrix Multiply, SHAPE0 SPR is used:

*three* SVSHAPE SPRs are used and they are named SVSHAPEn not SHAPEn
(i corrected this earlier last week)

you can see the instruction source code, you do not need me
to see that:

https://git.libre-soc.org/?p=openpower-isa.git;a=blob;f=openpower/isa/simplev.mdwn;h=33a02e#l137

notice how SVSHAPE0-2 are set up?
one for A, one for B, one for C  in C=AxB...

otherwise how can we ever do the schedule for A B and C
unless there is an SVSHAPE for each?

hmmm i thought i had created a yield-demo which uses remapyield.py
to shw that more clearly, but i think (over 2!) years ago i did
a demo for DCT and FFT.

https://git.libre-soc.org/?p=openpower-isa.git;a=blob;f=src/openpower/decoder/isa/remap_fft_yield.py;hb=HEAD

yyeah there it is. that is standalone executable (designed as such).
sorry, didn't do a similar one for matrix, rempyield.py is much more
basic it was the first REMAP subsystem i ever did.

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