Sent: Monday, October 01, 2001 5:37 PM
To: lance@leland. Stanford. EDU; Ujval Kapasi; Billmark@Lambert.
Stanford. Edu


  Good meeting on Friday.  Here is our schedule for this week.

  By end of day Wednesday 10/3:  Send out email regarding language thoughts

  Friday 10/5 3:30:  Meeting Gates 359 (Note different room)

 The email should outline some of the ideas and issues that you feel is
important for the language.  We should then have a chance to read each
others emails and talk about them on Friday.  Also you should probably try
not to read each others emails until you have posted so we can gather an
honest consensus.

 Some issues talked about on Friday that probably need more discussion.  (I
didn't exactly write everything down so feel free to email more out.)

 What Stream I/O operations should be allowed?
   - Relative Addressing (arrays)?
   - Full Indexing?
   - Strides?
   How useful are these for stream computing?
   How bad/hard are these for hardware?

 Compiler:
  What should the compiler be responsible for?
  Should the compiler do things like:
	Stream memory management
	Kernel collapsing
	Recognize temporary intermediate streams
	Blocking of computation
	General scheduling of kernels
  Where do each of these fall in the easy-hard scale?
  If the compiler doesn't do it, how is it expressed?

 Semantics:
  There are still all sorts of semantic questions.
    Do we want allow conditional reads or use the function metaphor?
    What can a kernel routine do?  Function calls, Conditionals, etc.

 Hardware Model:
  What are the underlying assumptions about the hardware model?
  MIMD/SIMD may not be that important but what should the programmer think
about when writing optimal code with the language?

 Applications:
  What applications are critical for the streaming language?

 Feel free to email more topics out, I'm sure I missed a few.  Lets all try
to email out responses to some of these questions by Wednesday.

  Ian.