[Toon-members] General TooN aliasing problem

[Ethan Eade]

I've narrowed down the "bug" I had reported before to the  test case 
below (without TooN), which shows that it's actually an aliasing 
problem, and that the compiler is entitled to get it "wrong". 

The problem is with the use of reinterpret_cast<>  to change data access 
patterns (as in v.as_col() or m.T() or even matrix row access).  The 
compiler doesn't know that two different access patterns are referring 
to the same memory, so the data dependencies aren't enforced, and the 
optimiser can put reads before the necessary writes that generate the 
data. 

Unfortunately, the compiler is not required to get these dependencies 
right.  Using reinterpret_cast<> is saying "give me no guarantees about 
data dependencies".  It was only a matter of time before the GCC 
optimiser got aggressive enough to trigger the behaviour.

I don't see an easy fix for this; it requires a redesign of TooN 
internals if optimisation is to be allowed.

- Ethan

This code gives bad output in gcc 4.3.1 with -O3, but not 4.2.1:
------------------------------
#include <iostream>
using namespace std;

template <int N> struct B;

template <int N>
struct A  // Vector
{
   double x[N];
   const B<N>& as_B() const { return reinterpret_cast<const 
B<N>&>(*this); }
};

template <int N>
struct B // column Matrix
{
   double x[N][1];
   const A<1>& row(int i) const  {return reinterpret_cast<const 
A<1>&>(x[i][0]); }
};

template <int N>
ostream& operator<<(ostream& out, const A<N>& a)
{
   for (int i=0; i<N; ++i)
   out << a.x[i] << " ";
   return out;
}

template <int N>
ostream& operator<<(ostream& out, const B<N>& b)
{
   for (int i=0; i<N; ++i)
   out << b.row(i) << endl;
   return out;
}


int main()
{
   A<2> a;

   a.x[0] = 2;
   a.x[1] = 3;

   //cout << a.as_B().row(0).x[0] << endl << endl;  // (*)
  
   // Gives wrong output if line (*) is commented
   cout << a.as_B() << endl;
}