Multi sched read #55
Multi sched read #55
Conversation
marioskogias
force-pushed
the
multi-sched-read
branch
4 times, most recently
from
c74300f to
f8ae235
Compare
| @@ -39,6 +39,99 @@ void test_body() | |||
| }); | |||
| } | |||
|
|
|||
There was a problem hiding this comment.
I wonder if we should have some more racy tests. Where there are concurrent attempts to schedule, and hence concurrent potential wake and schedule, and sleep. We don't seem to cover those cases here.
src/rt/sched/behaviourcore.h
Outdated
| { | ||
| Logging::cout() << " Previous slot is a writer or blocked reader cown " | ||
| << *new_slot << Logging::endl; | ||
| << *chain_first_slot << Logging::endl; | ||
| yield(); | ||
| goto fn_out; |
There was a problem hiding this comment.
I wonder if direct returns in this code would be clearer.
| goto fn_out; | |
| return {0, 0}; |
I'm also not sure I get the usage of ex_count here. This is that all the reads in this chain have seen a READ_AVAILABLE, so they can all decrease there execution count by 1?
src/rt/sched/behaviourcore.h
Outdated
| yield(); | ||
| cown->next_writer = first_body; | ||
| } | ||
| } | ||
|
|
||
| // Process execution count | ||
| ec[first_body_index] += ex_count; | ||
|
|
||
| for (int i = 1; i < first_consecutive_readers_count; i++) | ||
| ec[first_body_index + i] += 1; |
There was a problem hiding this comment.
Why is ex_count not used here? It feels like the whole chain should get to go if the read available was seen during scheduling, but otherwise it shouldn't? Doesn't this increase the count on the chain apart from the first entry, if read available wasn't seen?
src/rt/sched/behaviourcore.h
Outdated
| { | ||
| Logging::cout() << " Writer at head of queue and got the cown " | ||
| << *curr_slot << Logging::endl; | ||
| << *chain_first_slot << Logging::endl; | ||
| ex_count++; | ||
| yield(); | ||
| } |
There was a problem hiding this comment.
Can ex_count be anything other than 0 in this branch?
| { | |
| Logging::cout() << " Writer at head of queue and got the cown " | |
| << *curr_slot << Logging::endl; | |
| << *chain_first_slot << Logging::endl; | |
| ex_count++; | |
| yield(); | |
| } | |
| { | |
| Logging::cout() << " Writer at head of queue and got the cown " | |
| << *chain_first_slot << Logging::endl; | |
| ec[first_body_index] += 1; | |
| yield(); | |
| continue; | |
| } |
src/rt/sched/behaviourcore.h
Outdated
|
|
||
| // Process execution count | ||
| ec[first_body_index] += ex_count; | ||
|
|
||
| for (int i = 1; i < first_consecutive_readers_count; i++) | ||
| ec[first_body_index + i] += 1; | ||
| } | ||
|
|
There was a problem hiding this comment.
If we can treat ex_count as a read chain that observed the read was available. Could we do this? Perhaps with a rename of ex_count?
| // Process execution count | |
| ec[first_body_index] += ex_count; | |
| for (int i = 1; i < first_consecutive_readers_count; i++) | |
| ec[first_body_index + i] += 1; | |
| } | |
| // Process execution count | |
| if (ex_count != 0) | |
| { | |
| for (int i = 0; i < first_consecutive_readers_count; i++) | |
| ec[first_body_index + i] += 1; | |
| } | |
| } | |
marioskogias
force-pushed
the
multi-sched-read
branch
from
f8ae235 to
30f0896
Compare
|
@vishalgupta97 could you take a look at this? Thanks |
|
The changes looks good. Maybe a test case can be added to test longer changes with different combinations of reads and writes. |
No description provided.