[Mono-list] callvirt instruction performance penalty?

[Nigel Delaney]

Hi All,

 

I had a quick question I was hoping someone could answer about the always
emit callvirt instruction pattern.  I had always understood that one of the
great advantages of C# over Java was, as Miguel put it on his blog about
porting Android from java, "virtual methods were made opt-in, instead of
opt-out which made for simpler VMs."  Which makes a lot of sense to me, no
virtual call overhead.

 

However, I was surprised to learn when disassembling code that both Mono and
Microsoft seem to ignore this optimization, and emit all method calls as
callvirt instructions in IL regardless of whether or not they are actually
virtual method calls, which seems to defeat the whole point of not being
java.

 

A Microsoft employee blogged about this
(http://blogs.msdn.com/b/ericgu/archive/2008/07/02/why-does-c-always-use-cal
lvirt.aspx)  and it seems that made this change to ensure that instance
methods could not be called on null instance references (a debatable
decision perhaps).  

 

I gather/suspect that Microsoft still optimizes the callvirt instruction
during the JIT stage to call the method and check for a null reference,
though they never say how large this performance penalty is.  Does anyone
know what the cost of this null reference check is?  And if Mono also is
able to optimize callvirt calls that are not actually virtual calls?

 

Someone posted earlier today about a performance issue when comparing
Microsoft versus mono virtual machines on windows, and it got me thinking
about the expense of method calls, so I would be curious to know if anyone
has the answer.

 

Cheers,

Nigel

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ximian.com/pipermail/mono-list/attachments/20130311/5b6467db/attachment.html>