In a serial program, the order of the operations during program execution are known. However, when code executes as multiple parallel tasks, an operation can execute before, after, or simultaneously with an operation in the other task. For example, when parallel tasks access or modify a shared memory location, data sharing problems can occur.
The
Intel®
Advisor
Dependencies tool
performs
extensive analysis of your running serial program to help you predict data
sharing problems. Use the
Dependencies Report window and the topics
introduced by this section to help you understand and decide how to fix the
reported data sharing problems.
For each data sharing problem, you can either:
-
Modify the sources to fix incidental or accidental data sharing by privatizing shared data use. This type of data sharing occurs when tasks use the same memory location, but do not communicate about using that memory location. If the data written by one is not needed by the other, each task could use a private copy of the data.
-
Add lock annotations to implement synchronization for independent updates. This type of sharing occurs when multiple tasks contribute to determining the final value of a memory location.
-
Recognize that the order of the operations cannot change, and consider modifying the chosen parallel sites and their tasks. When shared data access must occur in the original sequential order, this is called true dependence.
The following sections explain how to understand and fix sharing problems.