Re: [Gems-users] MessageBuffer ordering

[Niket <niketa@xxxxxxxxxxxxx>]

My understanding of the MessageBuffer is that it is a priority queue 
with "arrival time" as the key that determines the priority. The arrival 
time is the time at which the message can leave the buffer and is = 
current time + delta time (latency).
So, in your case say the second message enters 2 cycles after the first 
message, it still can leave early than the first message. Take a look at 
the ISCA tutorial of Gems in which they describe the priority queue.
If you want to implement your condition, you can modify the 
MessageBuffer code in ruby/buffers/MessageBuffer.C
Niket

Marco Solinas wrote:
> Hi list!
>
> I have a doubt regarding MessageBuffer ordering.
> If a MessageBuffer belongs to an ordered virtual network, I'm sure that 
> two outgoing messages are served in the same order they are insered in 
> the buffer (for this reason, Ruby checks if a message is enqueued with a 
> correct latency).
> My doubt is about unordered virtual networks. If I enqueue a first 
> message that, for some reason, has a latency of (say) 10 cycles, and 
> then I enqueue a second message that belongs to the SAME block of the 
> first one, with a latency of 3 cycles, does the second message leave the 
> buffer and enter the network BEFORE the first one? The problem is that I 
> need that even if I specify two different latencies for two messages for 
> the same block, it is important for me that the second message leaves 
> the buffer after the first one, doesn't matter whether or not the second 
> latency is smaller than the first one. I want to be sure of this.
> Hope you can help me!
> Thanks a lot for your support.
>
> Regards,
> Marco
>
> _______________________________________________
> Gems-users mailing list
> Gems-users@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/gems-users
> Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/"; to your search.
>
>