GPU: __cosf(5992552) = 3.826833069324493e-01
GPU: cos(5992552) = -1.411203294992447e-01
CPU: cosf(5992552) = -1.411203444004059e-01
CPU: cos(5992552) = -1.411203444004059e-01The result for the __cosf function on the GPU is different from the result for both cos functions and the cosf on the CPU. This is due to the input number being too large for the __cosf function to handle. The error of this function is small within the -π to π range. However, the error increases as the input number increases. The __cosf function is not suitable for large input numbers. The reason for this behavior is due to the fact that the __cosf function is a single-precision intrinsic function and compared to the cosf function on the CPU the main difference is that on a GPU, generally there are no additional instructions related to a floating point operations such as there are on i.e. an x86 architecture of a CPU.
GPU: __cosf(1) = 5.403023362159729e-01
GPU: cos(1) = 5.403022766113281e-01
CPU: cosf(1) = 5.403022766113281e-01
CPU: cos(1) = 5.403022766113281e-01