[Libre-soc-bugs] [Bug 1028] implement integer-versions of fft/dct "butterfly" instructions in ISACaller Simulator

[bugzilla-daemon at libre-soc.org]

https://bugs.libre-soc.org/show_bug.cgi?id=1028

--- Comment #7 from Jacob Lifshay <programmerjake at gmail.com> ---
(In reply to Konstantinos Margaritis (markos) from comment #5)
> Latest commit fixes the XLEN != 64 cases.
> 
> https://git.libre-soc.org/?p=openpower-isa.git;a=commit;
> h=189a461f1c86d21a6813ed8df892c0cf0b6d8acb

yay! looks good for XLEN<64

you should be able to delete both XLEN=64 branches and just keep the else
blocks, since then the whole function is independent of how many bits all the
values have. this will make the XLEN=64 case not drop bits -- e.g. right now
the XLEN=64 case will return the wrong result for SH=31 with RT=2^32, RA=0, and
RB=2^32 (essentially fixed-point 2.0 * 2.0 for 33.31 fixed-point format)
whereas afaict the else block would return the correct result, if only you used
it for XLEN=64 too.

so, basically you can delete lines 24-40 and 74-91 in
https://git.libre-soc.org/?p=openpower-isa.git;a=blob;f=openpower/isa/butterfly.mdwn;h=b6b7efe523c4ee9c9e61dd1fc6ce5ab96d3d7121;hb=189a461f1c86d21a6813ed8df892c0cf0b6d8acb#l24

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