We can reduce the number of idle threads by reducing the total number of threads per block and also using each thread to load multiple pixels into the shared memory.
We can reduce the number of idle threads by reducing the total number of threads per block and also using each thread to load multiple pixels into shared memory.
How to prevent idling threads from clogging the critical resources in the pipeline becomes a must in sustaining desired system performance.
In this paper, we show that, if one can manage to recall instructions of idling threads from the shared Issue Queue (IQ), the system performance is easily enhanced by a significant margin, with up to 20% for some benchmark mixes.
In the opposite case, the remaining threads will idle until the reader thread has filled the passive buffer.
Additionally, large subset sizes can lead to reduced opportunities for parallelization, especially on smaller OUs that are broken into fewer chunks than the available threads on site, thus leaving some threads idle.
In this way, the balancing of tasks in the MB level is achieved dynamically without going any thread to idle state, which solves the load balance problem.
This can be attributed to the fact that for larger subset sizes, the load may not be optimally balanced and some threads that were assigned disproportionately higher load might prolong the overall total execution time while other threads are idle.
Moreover, with q threads, most of the threads will remain idle for many iterations due to the dependency on other cells.
