[Mono-devel-list] patch for StringReader.Read() performance improvement

[Atsushi Eno]

Ben Maurer wrote:
>>(It is low-tech solution. Reduced function call
> 
> Really, we should be taking advantage of -O=inline. There is no reason to
> do an optimization that the jit will be doing later.

Ok, agreed. People who wanted to improve text processing would like to
use it (either the patch, or -O=inline).

>>to be able to optimize ABCREM).
> 
> Exactly how does this work?
> 
> The only difference i can see is that you aovid the foo [a++]. Doing that
> generates an extra variable in the IL code, which hurts regalloc. copyprop
> if you do -O=copyprop,deadce this may get solved (copyprop is another opt
> we want to enable for 1.2)

Oh, true. I thought ABCREM was prevented by ++ but was not and
intended to work it, but was not the source of the problem.

Atsushi Eno