如何在Android JNI中使用OpenCl rand实现双指针

Question

I am currently using an implementation of OpenCL bound with JNI interface to my android app. 我目前正在使用与JNI接口绑定到我的android应用程序的OpenCL实现。

When the kernel is loaded 加载内核时

__kernel void someFunction( __global double* out,

I get this error : 我收到此错误：

 error: use of type 'double' requires cl_khr_fp64 extension to be enabled
    __global double* weight_coef,

I tried to replace it by float and it works but I want to avoid that ! 我试图用float代替它，但是它可以工作，但是我想避免这种情况！

Answer 1

You need to enable the 'cl_khr_fp64' extension. 您需要启用“ cl_khr_fp64”扩展名。

To do that, you need to do 为此，您需要做

#pragma OPENCL EXTENSION cl_khr_fp64 : enable

before any double-precision data type is initialized in the kernel code. 在内核代码中初始化任何双精度数据类型之前。

For further information, feel free to refer to the Khronos OpenCL specification: https://www.khronos.org/registry/cl/sdk/1.0/docs/man/xhtml/cl_khr_fp64.html 有关更多信息，请随时参考Khronos OpenCL规范： https : //www.khronos.org/registry/cl/sdk/1.0/docs/man/xhtml/cl_khr_fp64.html