gentype half_recip(gentype x);
- gentype native_recip(gentype x);
native_recip computes reciprocal over an implementation-defined range. The maximum error is implementation-defined.
This built-in math function takes scalar or vector arguments. The generic type name gentype is used to indicate that the function can take float, float2, float4, float8, or float16 as the type for the arguments. For any specific use of these function, the actual type has to be the same for all arguments and the return type.
Functions with the half_ prefix are implemented with a minimum of 10-bits of accuracy i.e. an ULP value less than or equal to 8192 ulp.
Functions with the native__ prefix may map to one or more native device instructions and will typically have better performance compared to the corresponding functions (without the native__ prefix). The accuracy (and in some cases the input range(s)) of these functions is implementation-defined.
The vector versions of the math functions operate component-wise. The description is per-component.
The built-in math functions are not affected by the prevailing rounding mode in the calling environment, and always return the same value as they would if called with the round to nearest even rounding mode.
The Khronos Group
Copyright © 2007-2011 The Khronos Group Inc.
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and/or associated documentation files (the "Materials"), to deal in the Materials without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Materials, and to permit persons to whom the Materials are furnished to do so, subject to the condition that this copyright notice and permission notice shall be included in all copies or substantial portions of the Materials.
- page 244, section 6.12.2 - Math Functions