从内核调用函数时，OpenCL 程序不构建

Question

这个问题来自this问题。

我有一个调用常规函数的内核。当我构建并运行我的代码时，我得到以下输出：

Number of devices: 2
building program failed
-----COULD NOT CREATE KERNEL!!---

我的.cl 文件有问题的部分如下：

    #define  IDCT_INT_MIN   (- IDCT_INT_MAX - 1)
    #define  IDCT_INT_MAX   2147483647
      ....
      ....
      ....
      ....
      ....
      ....

    #define SCALE(x,n)      ((x) << (n))

    #define but(a,b,x,y)    { x = SUB(a,b); y = ADD(a,b); }

    static  int DESCALE (int x, int n)
    {
        return (x + (1 << (n - 1)) - (x < 0)) >> n;
    }

    static  int ADD(int x, int y)
    {
        int r = x + y;

        return r;       
    }

    static  int SUB(int x, int y)
    {
        int r = x - y;

        return r;        
    }

    static  int CMUL(int c, int x)
    {
        int r = c * x;
        r = (r + (1 << (C_BITS - 1))) >> C_BITS;
        return r;
    }

    static  void rot(int f, int k, int x, int y, int *rx, int *ry) {
        int COS[2][8] = {
            {c0_1, c1_1, c2_1, c3_1, c4_1, c5_1, c6_1, c7_1},
            {c0_s2, c1_s2, c2_s2, c3_s2, c4_s2, c5_s2, c6_s2, c7_s2}
        };

        *rx = SUB(CMUL(COS[f][k], x), CMUL(COS[f][8 - k], y));
        *ry = ADD(CMUL(COS[f][8 - k], x), CMUL(COS[f][k], y));
    }

void idct_1D(__private int *Y);

__kernel void IDCT(__global int *input, __global uchar *output) 
{
    int Y[64];
    int k, l;
    int Yc[8];

    for (k = 0; k < 8; k++)
    {
        for (l = 0; l < 8; l++) Y(k, l) = SCALE(input[(k << 3) + l], S_BITS);
        idct_1d(&Y(k, 0));
    }

    for (l = 0; l < 8; l++)
    {

        for (k = 0; k < 8; k++)
    {
            Yc[k] = Y(k, l);
    }

        idct_1d(Yc);

        for (k = 0; k < 8; k++)
        {
            int r = 128 + DESCALE(Yc[k], S_BITS + 3);
            r = r > 0 ? (r < 255 ? r : 255) : 0;
            X(k, l) = r;
        }

    }
}

void idct_1D(__private int *Y) 
{

int z1[8], z2[8], z3[8];


    but(Y[0], Y[4], z1[1], z1[0]);
    rot(1, 6, Y[2], Y[6], &z1[2], &z1[3]);
    but(Y[1], Y[7], z1[4], z1[7]);
    z1[5] = CMUL(sqrt2, Y[3]);
    z1[6] = CMUL(sqrt2, Y[5]);

    but(z1[0], z1[3], z2[3], z2[0]);
    but(z1[1], z1[2], z2[2], z2[1]);
    but(z1[4], z1[6], z2[6], z2[4]);
    but(z1[7], z1[5], z2[5], z2[7]);

    z3[0] = z2[0];
    z3[1] = z2[1];
    z3[2] = z2[2];
    z3[3] = z2[3];
    rot(0, 3, z2[4], z2[7], &z3[4], &z3[7]);
    rot(0, 1, z2[5], z2[6], &z3[5], &z3[6]);

    but(z3[0], z3[7], Y[7], Y[0]);
    but(z3[1], z3[6], Y[6], Y[1]);
    but(z3[2], z3[5], Y[5], Y[2]);
    but(z3[3], z3[4], Y[4], Y[3]);
}   

错误是由我从内核 IDCT 调用的函数 idct_1D 引起的。

如何解决内核函数的错误？

有没有办法从内核调用函数？

编辑：

按照 pmdj 的回答后，我用以下方式编写了构建语句：

ret= clBuildProgram(program, 1, &device_id, NULL, NULL, NULL);

if (ret != CL_SUCCESS) 
{
 printf("building program failed\n");

  size_t log_size;
  char buffer[2048];
clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_LOG, 0, NULL, &log_size);
clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_LOG, sizeof(buffer), buffer, &log_size);
printf("%s\n", buffer);

 }

在构建并运行我的代码后，我得到了以下响应：

Number of devices: 2
building program failed
ptxas application ptx input, line 71; error   : Call has wrong number of parameters
ptxas application ptx input, line 112; error   : Call has wrong number of parameters
ptxas application ptx input, line 153; error   : Call has wrong number of parameters
ptxas application ptx input, line 194; error   : Call has wrong number of parameters
ptxas application ptx input, line 235; error   : Call has wrong number of parameters
ptxas application ptx input, line 276; error   : Call has wrong number of parameters
ptxas application ptx input, line 317; error   : Call has wrong number of parameters
ptxas application ptx input, line 358; error   : Call has wrong number of parameters
ptxas application ptx input, line 392; error   : Call has wrong number of parameters
ptxas application ptx input, line 520; error   : Call has wrong number of parameters
ptxas application ptx input, line 648; error   : Call has wrong number of parameters
ptxas application ptx input, line 776; error   : Call has wrong number of parameters
ptxas application ptx input, line 904; error   : Call has wrong number of parameters
ptxas application ptx input, line 1032; error   : Call has wrong number of parameters
ptxas application ptx input, line 1160; error   : Call has wrong number of parameters
ptxas application ptx input, line 1288; error   : Call has wrong number of parameters
ptxas fatal   : Ptx assembly aborted due to errors

-----COULD NOT CREATE KERNEL!!---

我检查了这个github link 并从我的.cl 文件中删除了所有 cmets 和 printf 语句。尽管如此，在构建和运行代码后，我仍然遇到同样的错误。

编辑：

代码反映了我在遵循 pmdj 的建议后尝试在代码中实现的更改。现在我只收到语法错误。

我现在得到的输出如下：

Number of devices: 2
building program failed
<kernel>:98:70: error: invalid address space for pointee of pointer argument to __kernel function
__kernel void IDCT(__global int *input, __global uchar *output, int *Yc, int *Yin) 
                                                                     ^
<kernel>:98:79: error: invalid address space for pointee of pointer argument to __kernel function
__kernel void IDCT(__global int *input, __global uchar *output, int *Yc, int *Yin) 
                                                                              ^
<kernel>:105:30: error: called object type '__attribute__((address_space(16776963))) int *' is not a function or function pointer
                for (l = 0; l < 8; l++) Yin(k, l) = SCALE(input[(k << 3) + l], S_BITS);
                                        ~~~^
<kernel>:106:3: warning: implicit declaration of function 'idct_1d' is invalid in C99
                idct_1d(&Yin(k, 0));
                ^
<kernel>:106:15: error: called object type '__attribute__((address_space(16776963))) int *' is not a function or function pointer
                idct_1d(&Yin(k, 0));
                         ~~~^
<kernel>:114:15: error: called object type '__attribute__((address_space(16776963))) int *' is not a function or function pointer
                        Yc[k] = Yin(k, l);
                                ~~~^

-----COULD NOT CREATE KERNEL!!---

编辑：

我将__private 放在函数参数之前，我再次收到 ptx 错误。

编辑：

代码现在可以工作了。原来我用错误的名字调用函数（idct_1d 而不是idct_1D）。

我的函数的标题现在如下：

void idct_1D(int *Y);


__kernel void IDCT(__global int *input, __global uchar *output);

Answer 1

clBuildProgram() 失败后，您可以使用clGetProgramBuildInfo() 函数查看详细的编译器输出。像这样的：

size_t len;
char buffer[2048];
clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_LOG, sizeof(buffer), buffer, &len);

https://www.khronos.org/registry/OpenCL/sdk/1.0/docs/man/xhtml/clGetProgramBuildInfo.html

顺便说一句，问题很可能是您的idct_1D 函数需要一个指向__global 内存的指针（默认），而您正试图将__private 内存中的数组传递给它。 在 OpenCL 中，始终使用正确的内存空间标记您的指针。