CODEWITHSUNDEEP
opencl
mmostajab
mmostajab

Reputation: 2037

Defining different function with same name but different argument type in OpenCL 1.1

I want to define two functions with the same name but one applying a transformation on a float4 and the other one apply a transformation on float3 but it says:

conflicting types for 'mult'

Is it impossible or something is wrong in my code?

typedef struct {
    float m[16];        // member elements of the matrix
} mat4;

typedef struct {
    float m[9];
} mat3;

float4 mult(mat4 matrix, float4 p) {
    return 
        matrix.m[0]     * p.x   + matrix.m[1]   * p.y + matrix.m[2]     * p.z + matrix.m[3]     * p.w +
        matrix.m[4]     * p.x   + matrix.m[5]   * p.y + matrix.m[6]     * p.z + matrix.m[7]     * p.w +
        matrix.m[8]     * p.x   + matrix.m[9]   * p.y + matrix.m[10]    * p.z + matrix.m[11]    * p.w +
        matrix.m[12]    * p.x   + matrix.m[13]  * p.y + matrix.m[14]    * p.z + matrix.m[15]    * p.w ;

}

float3 mult(mat4 matrix, float3 p) {
    return mult(matrix, float4(p, 1)).xyz;
}

Upvotes: 2

Views: 1570

Answers (1)

jprice
jprice

Reputation: 9925

Although the OpenCL built-in functions utilise function overloading to provide different variants of the same function with the same name, the OpenCL C specification does not explicitly allow function overloading in user code (and this is not a feature of C99 either).

It might be that some of the Clang-based OpenCL implementations let you use the overloadable function attribute, which would look something like this:

float4 __attribute__((overloadable)) mult(mat4 matrix, float4 p) {
    ...
}

float3 __attribute__((overloadable)) mult(mat4 matrix, float3 p) {
    ...
}

This is not a standard OpenCL feature however, and is not guaranteed to work on all OpenCL platforms. The OpenCL C++ kernel language proposed for OpenCL 2.1 will natively support function overloading.

Upvotes: 3

Related Questions