Memory Instructions for Manycore
Posted by Bryan Catanzaro on February 2, 2007
Steven Ericsson-Zenith responds to Dave's presentation at Stanford:
I enjoyed your lecture yesterday (I am one of the transputer/occam architects and spoke to you at the end).
I have quickly reviewed the RAMP materials and find them interesting. I have my attention on the theoretical foundations of logic, recognition and motility currently but I would be interested in participating at some level.
I can offer a solution for you to consider in the memory/storage, synchronization, instruction set - that arose from my consideration of the issues at INMOS and YALE (I worked with Gelernter too).
In my 1992 Ph.D. thesis I propose extending the typing system of parallel programming systems (consistent with your suggestion yesterday) and added four complementary primitive instructions :
put and get, read and write
These are aimed at resolving the copy penalty that will occur when you move data around in a parallel system. Simplistically, it is a way of encapsulating pointers and access synchronization. The proposal has the advantage of a formal basis.
You can find my thesis online at:
http://process-interaction-models.info
I enjoyed your lecture yesterday (I am one of the transputer/occam architects and spoke to you at the end).
I have quickly reviewed the RAMP materials and find them interesting. I have my attention on the theoretical foundations of logic, recognition and motility currently but I would be interested in participating at some level.
I can offer a solution for you to consider in the memory/storage, synchronization, instruction set - that arose from my consideration of the issues at INMOS and YALE (I worked with Gelernter too).
In my 1992 Ph.D. thesis I propose extending the typing system of parallel programming systems (consistent with your suggestion yesterday) and added four complementary primitive instructions :
put and get, read and write
These are aimed at resolving the copy penalty that will occur when you move data around in a parallel system. Simplistically, it is a way of encapsulating pointers and access synchronization. The proposal has the advantage of a formal basis.
You can find my thesis online at:
http://process-interaction-models.info