meeting minutes - 6-15-06

Attendees:
   Per Bojsen
   Edmund Fong
   Shabtay Matalon
   Jason Rothfuss
   Bryan Sniderman
   John Stickley

-----------------------------------------
Meeting attendance

- JohnS stated he will not be able to attend June 22nd or 29th.
   but that he had a number of issues and IMs he wanted to discuss
   by e-mails.

- Shabtay and Jason will not be able to attend Jul. 6th due to
   plant shutdown.

-----------------------------------------
Copyright issues

- JohnS and Per to try to get written permission for Cadence to use
   files they created for pipes reference model implementation - according
   to internal use agreement sent by Shabtay.

-----------------------------------------
Shabtay's proposal for multi-channel, object oriented pipes.

- JohnS asked what the differences are between multi-channel object
   oriented pipes and the existing function based pipes.

- Shabtay and Jason indicated the only difference is the function vs.
   module based pipes use model - pipe ID handling flexibility
   and limitations is identical for both.

-----------------------------------------
Discussion of bytes_per_element

- JohnS and Per clarified that bytes_per_element is fixed at first
   use and error checked for consistency thereafter.

- Shabtay stated that although the reference model does it this way
   the spec is not clear about fixing bytes/element with 1st use.
   Spec. should be updated to at least make this clear.

- Shabtay and Jason both felt that bytes_per_element should be
   compile time static.

- JohnS and Per felt that init-time dynamic affords more flexibility.

- Jason pointed out that this can only be checked at run-time and
   that he feels this is a liability for designs that are compiled
   for emulation.

- JohnS stated opinion that these issues are typically worked
   out during prototyping that that the disadvantage of deferred
   error checking is more than offset by the flexibility of
   run-time bytes/element specification.

- Bryan Sniderman also expressed preference for init-time dynamic
   specification of bytes/element.

- Philosophical differences were discussed as to more flexibility
   of dynamic parameter specification but error checking deferred
   to run-time vs. less flexibility but error checking caught up front
   at compile time. So far these differences of opinion have affected
   both pipe IDs and bytes/element settings for pipes.

-----------------------------------------
Byte vs. bit granularity for element size for pipes

- JohnS stated preference for byte granularity for reasons of
   more efficient C-side manipulation.

- Shabtay stated that bit granularity offers more flexibility.

- Bryan S. expressed agreement with this but said that bit granularity
   on the H/W side would be useful with the understanding that internally
   elements are sized to the next byte size. And that for the C side,
   possibly passing a bit mask for the residual byte of an element transfer
   would be useful.

- Per stated that bit manipulation on C side is problematic and difficult.

-- johnS
______________________________/\/            \     \
John Stickley                   \             \     \
Mgr., Acceleration Methodologies \             \________________
Mentor Graphics - MED             \_
________________________________________________________________