On Wed, Jul 28, 2010 at 4:37 PM, Hite, Christopher
<[email protected]> wrote:
>
>
>
> I'm writing a decoder using a meta programming techniques alla
> boost::mpl and I'd like to know if I'm asking too much of the compiler.
>
> Basically I've got lots of packet types with different ids. The
> classical way to write these would be a switch case
>
> int id;
> switch(id){
> case Id1: decode1() break;
> case Id2: decode2() break;
> ...
> }
>
> I'm tring to use the template compiler to generate equivalent code. I
> have a mpl::list of packet descriptions like this:
> struct Packet1{
> static const int id=1;
> void decode();
> };
>
> I then use boost::mpl::fold to generate a decode function which might
> look sort of like this:
>
> void decode1(int id){
> if(id==1)
> Packet1::decode();
> else
> decode2(id);
> }
>
> What I'm hoping is that the compiler is smart enough
> * to inline all decode*() calls into one function
> * to notice I compare id to N constants and use switch case optimization
> (binary search or lookup table)
>
> Am I asking too much?
>
> Do I have to watch for any pitfalls that willl break the optimization,
> like making id a member of a class or leaving out the 'else' ?
>
>
> I could write more complicated meta-code which sorts by id and does a
> runtime binary search, that would probably prevent the optimizer from
> doing anything smarter.
Generally without knowing the compiler version you are using
it is hard to tell. The same is true without a complete compilable
testcase.
You can use the flatten attribute to tell the compiler to inline all
calls in a given function, like
void __attribute__((flatten)) foo(void)
{
...
decode1();
...
}
and it will inline decode1() (and recursively all its callees).
Richard.
> Chris Hite
>
>
