Question

我正在尝试在openCL内核中进行一些双精度数学运算。我启用了cl_khr_fp64，我可以在double中执行简单的+ - * /操作，但是当我尝试使用内置的数学函数（例如，exp）时，代码无法编译。如果我切换到浮动它可以工作。

我在Khronos网站上读到，如果你启用了cl_khr_fp64，那么数学函数会被重载以支持double。（http://www.khronos.org/registry/cl/sdk/1.0/docs/man/xhtml/mathFunctions.html）

我正在使用D700 GPU开发新的MacPro，我读到只有一些数学函数被重载以包含double（http://developer.amd.com/knowledge-base/），但是指数在包含的函数列表中。此外，当我将此代码发送到CPU而不是GPU时，它也会失败。

这是我内核的代码。它只是Hello World的延伸！ Apple开发者网站上的示例。如果我将u切换为浮点数，或者如果我将exp（u）替换为u（如条件的第一部分），它也可以。这只是一个玩具问题，我在开始实现我的真实代码之前试图开始工作，但是我必须在移动之前让它工作。我还尝试过double_exp，expd和native_exp。它们都导致代码编译，但后来我遇到了运行时错误。

#pragma OPENCL EXTENSION cl_khr_fp64 : enable
kernel void square5(global double* input, global double* output, double mul)
{
    size_t i = get_global_id(0);
    double u = 5.0;
    if(i==0) {
        output[i] = mul*u*input[i]*input[i];
    } else if (i==1023) {
        output[i] = mul*u*input[i]*input[i];
    } else {
        output[i] = 0.25*mul*exp(u)*(input[i-1] + input[i+1])*(input[i-1] + input[i+1]);
    } 
}

这是错误日志

/Users/me/Downloads/OpenCL_Hello_World_Example/mykernel.cl:12:30: error: call to '__fast_relax_exp' is ambiguous
    output[i] = 0.25*mul*exp(u)*(input[i-1] + input[i+1])*(input[i-1] + input[i+1]);
                         ^~~~~~
/System/Library/Frameworks/OpenCL.framework/Versions/A/Libraries/../lib/clang/3.2/include/cl_kernel.h:4496:22: note: expanded from macro 'exp'
#define exp(__x) __fast_relax_exp(__x)
                 ^~~~~~~~~~~~~~~~
/System/Library/Frameworks/OpenCL.framework/Versions/A/Libraries/../lib/clang/3.2/include/cl_kernel.h:4494:30: note: candidate function
__CLFN_FD_1FD_FAST_RELAX(__fast_relax_exp, native_exp, __cl_exp);
                         ^
/System/Library/Frameworks/OpenCL.framework/Versions/A/Libraries/../lib/clang/3.2/include/cl_kernel.h:416:27: note: expanded from macro '__CLFN_FD_1FD_FAST_RELAX'
inline float __OVERLOAD__ _name(float x) { return _default_name(x); } \
                      ^
/System/Library/Frameworks/OpenCL.framework/Versions/A/Libraries/../lib/clang/3.2/include/cl_kernel.h:4494:30: note: candidate function
__CLFN_FD_1FD_FAST_RELAX(__fast_relax_exp, native_exp, __cl_exp);
                         ^
/System/Library/Frameworks/OpenCL.framework/Versions/A/Libraries/../lib/clang/3.2/include/cl_kernel.h:417:28: note: expanded from macro '__CLFN_FD_1FD_FAST_RELAX'
inline float2 __OVERLOAD__ _name(float2 x) { return _default_name(x); } \
                       ^
/System/Library/Frameworks/OpenCL.framework/Versions/A/Libraries/../lib/clang/3.2/include/cl_kernel.h:4494:30: note: candidate function
__CLFN_FD_1FD_FAST_RELAX(__fast_relax_exp, native_exp, __cl_exp);
                         ^
/System/Library/Frameworks/OpenCL.framework/Versions/A/Libraries/../lib/clang/3.2/include/cl_kernel.h:418:28: note: expanded from macro '__CLFN_FD_1FD_FAST_RELAX'
inline float3 __OVERLOAD__ _name(float3 x) { return _default_name(x); } \
                       ^
/System/Library/Frameworks/OpenCL.framework/Versions/A/Libraries/../lib/clang/3.2/include/cl_kernel.h:4494:30: note: candidate function
__CLFN_FD_1FD_FAST_RELAX(__fast_relax_exp, native_exp, __cl_exp);
                         ^
/System/Library/Frameworks/OpenCL.framework/Versions/A/Libraries/../lib/clang/3.2/include/cl_kernel.h:419:28: note: expanded from macro '__CLFN_FD_1FD_FAST_RELAX'
    inline float4 __OVERLOAD__ _name(float4 x) { return _default_name(x); } \
                       ^
/System/Library/Frameworks/OpenCL.framework/Versions/A/Libraries/../lib/clang/3.2/include/cl_kernel.h:4494:30: note: candidate function
__CLFN_FD_1FD_FAST_RELAX(__fast_relax_exp, native_exp, __cl_exp);
                         ^
/System/Library/Frameworks/OpenCL.framework/Versions/A/Libraries/../lib/clang/3.2/include/cl_kernel.h:420:28: note: expanded from macro '__CLFN_FD_1FD_FAST_RELAX'
inline float8 __OVERLOAD__ _name(float8 x) { return _default_name(x); } \
                       ^
/System/Library/Frameworks/OpenCL.framework/Versions/A/Libraries/../lib/clang/3.2/include/cl_kernel.h:4494:30: note: candidate function
__CLFN_FD_1FD_FAST_RELAX(__fast_relax_exp, native_exp, __cl_exp);
                         ^
/System/Library/Frameworks/OpenCL.framework/Versions/A/Libraries/../lib/clang/3.2/include/cl_kernel.h:421:29: note: expanded from macro '__CLFN_FD_1FD_FAST_RELAX'
inline float16 __OVERLOAD__ _name(float16 x){ return _default_name(x); }
                        ^
1 error generated.
Command /System/Library/Frameworks/OpenCL.framework/Libraries/openclc failed with exit code 1

在openCL中尝试双精度数学不能编译

0 个答案: