Intel® 64 and IA-32 Architectures Optimization Reference Manual

POWER OPTIMIZATION FOR MOBILE USAGESActiveSleepThread 1(core 1)ActiveSleepThread 2(core 2)ActiveSleepCPUDeeperSleepFigure 10-6. Progression to Deeper SleepEnabling Both Cores to Take Advantage of Intel Enhanced Deeper Sleep.To best utilize processor-specific C-state (e.g., Intel Enhanced Deeper Sleep) toconserve battery life in multithreaded applications, a multi-threaded applicationshould synchronize threads to work simultaneously and sleep simultaneously usingOS synchronization primitives. By keeping the package in a fully idle state longer(satisfying ACPI C3 requirement), the physical processor can transparently takeadvantage of processor-specific Deep C4 state if it is available.Multi-threaded applications need to identify and correct load-imbalances of itsthreaded execution before implementing coordinated thread synchronization. Identifyingthread imbalance can be accomplished using performance monitoring events.Intel Core Duo processor provides an event for this purpose. The event(Serial_Execution_Cycle) increments under the following conditions:• Core actively executing code in C0 state• Second core in physical processor in idle state (C1-C4)This event enables software developers to find code that is executing serially, bycomparing Serial_Execution_Cycle and Unhalted_Ref_Cycles. Changing sections ofserialized code to execute into two parallel threads enables coordinated threadsynchronization to achieve better power savings.Although Serial_Execution_Cycle is available only on Intel Core Duo processors,application thread with load-imbalance situations usually remains the same forsymmetric application threads and on symmetrically configured multicore processors,irrespective of differences in their underlying microarchitecture. For thisreason, the technique to identify load-imbalance situations can be applied to multithreadedapplications in general, and not specific to Intel Core Duo processors.10-13