We define "write is observed" by the point at which a read to that value will return the value just written. The difference between the second bullet of TSO and the bullet in PC is that PC doesn't assume anything about what other processors know about the newly written value; it can be the old value or the new value.
TSO does not maintain W$\rightarrow$R on the same processor, but maintains it across other processors. PC does not maintain W$\rightarrow$R across any processor.
This comment was marked helpful 0 times.
alex
Here is a recent blog post where the author generated a short program the behaves differently under TSO than under sequential consistency.
This comment was marked helpful 0 times.
pd43
In other words, TSO is 'stricter'. If a processor reads a value from a given location in memroy, it can assume other processors will read the same value from that location, but PC does not have this guarantee.
This comment was marked helpful 0 times.
mitraraman
If a system uses PC, isn't correctness not guaranteed if processors have the ability to read a value that is not updated?
This comment was marked helpful 0 times.
gbarboza
Guaranteeing correctness is now placed on the programmer.
This comment was marked helpful 0 times.
tpassaro
Is there a reason why processor consistency is around? I could see it useful where a system knows that no two threads directly share the same data, and thus would never need to wait for anyone to complete any writes. This would take care of latency associated with assuring that a processor is done writing a value to memory.
We define "write is observed" by the point at which a read to that value will return the value just written. The difference between the second bullet of TSO and the bullet in PC is that PC doesn't assume anything about what other processors know about the newly written value; it can be the old value or the new value.
TSO does not maintain W$\rightarrow$R on the same processor, but maintains it across other processors. PC does not maintain W$\rightarrow$R across any processor.
This comment was marked helpful 0 times.
Here is a recent blog post where the author generated a short program the behaves differently under TSO than under sequential consistency.
This comment was marked helpful 0 times.
In other words, TSO is 'stricter'. If a processor reads a value from a given location in memroy, it can assume other processors will read the same value from that location, but PC does not have this guarantee.
This comment was marked helpful 0 times.
If a system uses PC, isn't correctness not guaranteed if processors have the ability to read a value that is not updated?
This comment was marked helpful 0 times.
Guaranteeing correctness is now placed on the programmer.
This comment was marked helpful 0 times.
Is there a reason why processor consistency is around? I could see it useful where a system knows that no two threads directly share the same data, and thus would never need to wait for anyone to complete any writes. This would take care of latency associated with assuring that a processor is done writing a value to memory.
This comment was marked helpful 0 times.