From: terje.mathisen@nospicedham.tmsw.no   
      
   fizz buzz wrote:   
   > I am thinking to use separate thread to fill output buffer with "\n"   
   > characters (perhaps using MOVQ filling 8 bytes at a time), and then   
   > just print words and numbers in main thread without caring about   
   > newlines. For example, I copy word "Fizz" to the buffer, which is 4   
   > bytes, overwriting four newline characters, but add 5 to buffer   
   > pointer, so next word or number won't overwrite newline character   
   > that's located in the buffer right after the "Fizz".   
   >   
   > Not sure if I'm explaining it clear enough.   
   >   
   > There should be no way that additional thread will take longer to   
   > execute, so words and numbers should never be overwritten by newline   
   > characters. It will require more computing power, but it should be   
   > faster. Theoretically. :-)   
      
   If we allow multiple threads, then it becomes trivial to split the   
   output range into N slices, one for each thread. Since you can also   
   trivially calculate exactly how long (in characters) a given range will   
   be, you can even have those threads writing to a common/shared output   
   buffer.   
      
   It is however faster to keep writing to a single L1/L2 sized buffer and   
   then flush that to a file/NULL when nearly full.   
      
   I think for this test we should stay with   
      
   a) No precalculating the output, i.e. the range will be given on the   
   command line, this can be up to 1E9 so it will fit in a 32-bit register.   
      
   b) A single CPU thread.   
      
   c) A flag to flush/save the output.   
      
   Terje   
      
   --   
   -    
   "almost all programming can be viewed as an exercise in caching"   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)