Libjava failures status

[Chris Lattner]

On 12 Mar 2002, Tom Tromey wrote:

> Chris> No, it would be perfectly reasonable to represent this (safely)
> Chris> with explicit checks, which would not break any

> However, if you assume that the checks usually pass, and you care more
> about the performance of the normal case, and not so much about the
> performance of generating the exception, and your platform has support
> for it, then it is better to omit the tests and let the trap be
> converted into an exception by the runtime.

That's perfectly fine, and incredibly important.  My point is just that
the control flow edges should be explicit for most of the compilation
process... and then dropped late in the back end after (f.e.) reordering
has been finished.  It would seem to be fairly straightforward to have the
machine description describe how to lower these things.

Not doing this may give you better performance, but if it generates
incorrect code (or obfuscates the compiler), then high performance isn't
very useful...

-Chris

http://www.nondot.org/~sabre/os/
http://www.nondot.org/MagicStats/