Hi all,   
      
   I'm currently working on a compiler. I use graph coloring to allocate   
   registers. The compiler targets different platforms, one of them being   
   the x86. I currently generate code to bypass the fpu stack problem,   
   every operation being prefixed by a load and postfixed by a store.   
   The generated code is currently working.   
      
   Now I want to optimize that code by using the fpu stack. I've thought   
   of using the register allocation with pseudo fpu instructions and   
   pseudo fpu registers. After the register allocation, I would convert   
   pseudo-instruction into real fpu stack, allocating stack slot to   
   register. LLVM is using such a technique to allocate stack slot. But I   
   can't find any information on this process. Does anybody have   
   experience on this or have information about it?   
      
   Thanks   
      
   Julien Hamaide   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)