[LLVMdev] mips16 whitepaper

[reed kotler]

On 01/21/2013 03:28 PM, Sean Silva wrote:
> Please add this to <http://llvm.org/docs/CompilerWriterInfo.html>
> (docs/CompilerWriterInfo.rst)
>
> -- Sean Silva
I will. I owe a very long mips16 write up , especially for floating 
point when I'm done.
It's very complicated the way they implemented hard float in gcc and it 
is not documented.
It took me some thinking and study to unravel it all. They do different 
things for pic/non pic
and the interoperability with mips32 is what makes in complicated.

I have come up with another totally different way to implement floating 
point in mips16 which I hope
to have time to do in the future. It's probably about 2 months work but 
management wants me
to do this gcc way first.

My alternate design is a more true floating point implementation as 
opposed to the gcc method which is more or less a variant of soft-float 
which uses different runtime routines but it has a lot of complications 
with mips32 interoperability whereas my scheme has none of these issues 
and does not need any special helper functions.