The parallel concepts of threads and processes are introduced and their impacts on performance are outlined.

• Detect race conditions and use synchronization mechanisms to avoid them.
• Estimate the problems that may result from erroneous use of synchronization mechanisms (e.g. deadlocks).
• Assess data dependency situations, i.e. an instruction reading the data written by a preceding instruction in the source code, and anti dependencies, i.e. an instruction having to read data before a succeeding instruction overwrites it, and output dependencies, i.e. instructions writing to the same memory location.
• Use data parallelism, e.g. applying parallel streams of identical instructions to different elements of appropriate data structures such as arrays.
• Apply the concept of functional parallelism, i.e. executing a set of distinct functions possibly using the same data.
• Assess the applicability of parallel language extensions like OpenMP.
• Assess concepts like software pipelining, e.g. to optimize loops by out-of-order execution.
• Assess the applicability of parallel language extensions like CUDA as well as their interoperability (e.g. combining OpenACC and CUDA).
• Assess parallel concepts typically used for shared memory systems, e.g. to exploit temporal locality by data reuse with efficient utilization of the memory hierarchy.
• Assess concepts like software pipelining and vectorization principles.
• Assess the influence of control dependencies by jumps, branches, and function calls, e.g. on pipeline filling.
• Assess the applicability of parallel language extensions like OpenACC, and C++ AMP.
• Understand Single Instruction Multiple Data (SIMD)
• Understand Synchronization
• Understand Memory Hierarchy and Data Reuse
• Understand Software Pipelining
• Understand Dependency Pattern and remember
  • Data Dependencies
  • Control Dependencies
• Understand Language Extensions and remember
  • Processes
  • CUDA
  • OpenACC
  • Threading with OpenMP and Pthreads
  • C++ AMP
  • Vectorization