OpenCL™ API offers two basic ways to trade precision for speed:
- native_* and half_* math built-ins, which have lower precision, but are faster than their un-prefixed variants
- Compiler optimization options that enable optimizations for floating-point arithmetic for the whole OpenCL program (for example, the -cl-fast-relaxed-math flag).
For the list of other compiler options and their description please refer to the Intel® Code Builder for OpenCL™ API - User Manual. In general, while the -cl-fast-relaxed-math flag is a quick way to get potentially large performance gains for kernels with many math operations, it does not permit fine control of numeric accuracy. Consider experimenting with native_* equivalents separately for each specific case, keeping track of the resulting accuracy.
The native_ versions of math built-ins are generally supported in hardware and run substantially faster, while offering lower accuracy. Use native trigonometry and transcendental functions, such as sin, cos, exp or log, when performance is more important than precision.
The list of functions that have optimized versions support is provided in "Working with cl-fast-relaxed-math Flag" section of the OpenCL Code Builder - User’s Guide.
OpenCL™ Build and Linking Options chapter of the Intel® Code Builder for OpenCL™ API - User Manual