Efficient Synchronization for Multithreaded Processors

cycle multithreading. Eliminating this source of synchronization In this paper we describe a source of synchronization overhead by allowing synchronization to be used in places where its unique to multithreaded processors, and present a novel generaloverhead previously precluded its use. We give an example in purpose synchronization scheme that addresses this overhead. Section 2, after we introduce our proposed mechanism and Unlike previous schemes, we reduce the cost associated with compare it to previous implementations. Additionally, it provides synchronization by eliminating the need for busy-waiting or for a performance advantage for existing code. In [WOT95], the software intervention while blocking. This is achieved via a authors measured the percent of the time programs in the combination of multithreading and semaphore operations impleSPLASH-2 benchmark suite spent in synchronization to range up mented in hardware. Once this synchronization mechanism is in to 40% of the execution time (although some of this idle time place, we propose using it to simplify the customary implementacould be addressed by having multiple computation threads per tion of interrupts and exceptions. Lastly, we provide simulation processor). Thus, in Section 3 we present simulation results results from four programs from the SPLASH2 benchmark suite, showing the performance advantage of our scheme. Section 4 is showing performance increases between 2% and 150%. the conclusion.