Intel® 64 and IA-32 Architectures Optimization Reference Manual

MULTICORE AND HYPER-THREADING TECHNOLOGY8.2.3 Specialized Programming ModelsIntel Core Duo processor and processors based on Intel Core microarchitecture offera second-level cache shared by two processor cores in the same physical package.This provides opportunities for two application threads to access some applicationdata while minimizing the overhead of bus traffic.Multi-threaded applications may need to employ specialized programming models totake advantage of this type of hardware feature. One such scenario is referred to asproducer-consumer. In this scenario, one thread writes data into some destination(hopefully in the second-level cache) and another thread executing on the other corein the same physical package subsequently reads data produced by the first thread.The basic approach for implementing a producer-consumer model is to create twothreads; one thread is the producer and the other is the consumer. Typically, theproducer and consumer take turns to work on a buffer and inform each other whenthey are ready to exchange buffers. In a producer-consumer model, there is somethread synchronization overhead when buffers are exchanged between the producerand consumer. To achieve optimal scaling with the number of cores, the synchronizationoverhead must be kept low. This can be done by ensuring the producer andconsumer threads have comparable time constants for completing each incrementaltask prior to exchanging buffers.Example 8-1 illustrates the coding structure of single-threaded execution of asequence of task units, where each task unit (either the producer or consumer)executes serially (shown in Figure 8-2). In the equivalent scenario under multithreadedexecution, each producer-consumer pair is wrapped as a thread functionand two threads can be scheduled on available processor resources simultaneously.Example 8-1. Serial Execution of Producer and Consumer Work Itemsfor (i = 0; i < number_of_iterations; i++) {producer (i, buff); // pass buffer index and buffer addressconsumer (i, buff);}(MainThreadP(1)C(1)P(1)C(1)P(1)Figure 8-2. Single-threaded Execution of Producer-consumer Threading Model8-6