[Mono-list] Large (!) methods?

Paolo Molaro lupus@ximian.com
Thu, 5 Jun 2003 20:56:44 +0200

On 06/04/03 Tony Garnock-Jones wrote:
> Yay! I updated from CVS, and sure enough, my extremely large method (it 
> contains 171484 bytes of IL and 5266 local variables!! I really need to 
> make my compiler less dumb) now compiles and runs to completion. Your 
> fix must have fixed my problem.

>From your description it looks like the same bug (I got it by running
the nunit tests with an optimization that increases the number of
symbolic registers used).

> (It takes an *awfully* long time to JIT the method, though, so even 

With default optimizations? Currently when -O=all is used, some methods
may take a long time to compile, but by default it should be quite fast.
Well, mini generates about 1MB of native code per second on my PIII
1.1, and 170Kbytes of IL code should be about 0.5 MB of native code,
so .5 seconds: we need to profile JIT times and improve on that.
I think doubling the JIT speed is doable, we just need some time to
spend on it...


lupus@debian.org                                     debian/rules
lupus@ximian.com                             Monkeys do it better