Snoop-valid may be better called snoop-invalid. Every processor starts with it on (set to 1), and once every processor turns it off, then the final "or" value will be 0 as stated in the slide.
This comment was marked helpful 0 times.
sjoyner
Every line is just a wire, so it just ORs the signals returned by each processor to get the result.
Shared line: If a processor has an address in it's cache, it signals the other processors on this line.
Dirty line: Processor signals that it has the address in the dirty state.
Snoop-valid line: Processor signals that it is done. Another idea was to just wait a given period of time to make sure every processor has enough time to respond. However if all processors finish before time is up, it is better if we return right away than wait for the timer. The snoop-valid line makes this easy since we just wait until the line is set to 0.
This comment was marked helpful 1 times.
akashr
Could we not also make this a snoop-valid line by just and'ing all the processor responses. The only way that line would switch to valid was if all the processors are done and send a "on" signal. If any of the processors are not done, then the entire valid line will be off.
Could architects chose either or is there a better choice of the two. Snoop valid or Snoop invalid?
This comment was marked helpful 0 times.
Xiao
Both ANDing and ORing solutions are logically equivalent. However, implementation wise they could be different. For example, one catch might be the behavior under faulty situation. If a open circuit occurs from one CPU, is the line connecting to that CPU more likely to be on or off? Would you prefer the system to make progress under such situation or deadlock?
Another problem you might consider is which logic gate takes the larger number of transistors.
Snoop-valid may be better called snoop-invalid. Every processor starts with it on (set to 1), and once every processor turns it off, then the final "or" value will be 0 as stated in the slide.
This comment was marked helpful 0 times.
Every line is just a wire, so it just ORs the signals returned by each processor to get the result. Shared line: If a processor has an address in it's cache, it signals the other processors on this line. Dirty line: Processor signals that it has the address in the dirty state. Snoop-valid line: Processor signals that it is done. Another idea was to just wait a given period of time to make sure every processor has enough time to respond. However if all processors finish before time is up, it is better if we return right away than wait for the timer. The snoop-valid line makes this easy since we just wait until the line is set to 0.
This comment was marked helpful 1 times.
Could we not also make this a snoop-valid line by just and'ing all the processor responses. The only way that line would switch to valid was if all the processors are done and send a "on" signal. If any of the processors are not done, then the entire valid line will be off.
Could architects chose either or is there a better choice of the two. Snoop valid or Snoop invalid?
This comment was marked helpful 0 times.
Both ANDing and ORing solutions are logically equivalent. However, implementation wise they could be different. For example, one catch might be the behavior under faulty situation. If a open circuit occurs from one CPU, is the line connecting to that CPU more likely to be on or off? Would you prefer the system to make progress under such situation or deadlock? Another problem you might consider is which logic gate takes the larger number of transistors.
This comment was marked helpful 0 times.