Safari Books Online is a digital library providing on-demand subscription access to thousands of learning resources.
- Create BookmarkCreate Bookmark
- Create Note or TagCreate Note or Tag
- PrintPrint
1. Overview and C++ AMP Approach > Technologies for CPU Parallelism
Technologies for CPU Parallelism
One way to reduce the amount of time spent in the sequential portion of your application is to make it less sequential—to redesign the application to take advantage of CPU parallelism as well as GPU parallelism. Although the GPU can have thousands of threads at once and the CPU far less, leveraging CPU parallelism as well still contributes to the overall speedup. Ideally, the technologies used for CPU parallelism and GPU parallelism would be compatible. A number of approaches are possible.
An important way to make processing faster is SIMD, which stands for Single Instruction, Multiple Data. In a typical application, instructions must be fetched one at a time and different instructions are executed as control flows through your application. But if you are performing a large data-parallel operation like matrix addition, the instructions (the actual addition of the integers or floating-point numbers that comprise the matrices) are the same over and over again. This means that the cost of fetching an instruction can be spread over a large number of operations, performing the same instruction on different data (for example, different elements of the matrices.) This can amplify your speed tremendously or reduce the power consumed to perform your calculation.
