transformKernel.cl

const char *kernelSource = 
"void matrixByVector(__global const float* matrix, __private float* vector){"
    "float x = vector[0];"
    "float y = vector[1];"
    "float z = vector[2];"
    "float w = vector[3];"
    "vector[0] = matrix[0] * x + matrix[1] * y + matrix[2] * z + matrix[3] * w;"
    "vector[1] = matrix[4] * x + matrix[5] * y + matrix[6] * z + matrix[7] * w;"
    "vector[2] = matrix[8] * x + matrix[9] * y + matrix[10] * z + matrix[11] * w;"
    "vector[3] = matrix[12] * x + matrix[13] * y + matrix[14] * z + matrix[15] * w;"
"}"

"float dotProduct(__private float* vec1, __private float* vec2){"
    "return vec1[0]*vec2[0] + vec1[1]*vec2[1] + vec1[2]*vec2[2];"
"}"

"int RGBClamp(float val){"
    "if(val < 0) return 0;"
    "if(val > 255) return 255;"
    "return (int)val;"
"}"

"__kernel void transforms(__global const float* vertices,"
                                "__global const float* colors,"
                                "__global const float* normals,"
                                "__global const float* matrices,"
                                "__global float* outputVertices,"
                                "__global float* outputColors," 
                                "const int matrixCount,"
                                "const int height,"
                                "const int width) { "
    "int globalIndex = get_global_id(0) * 3;"
    "float vertex[4];"
    "float normal[4];"
    "for(int i = 0; i < 3; i++){ vertex[i] = vertices[globalIndex + i]; }"
    "for(int i = 0; i < 3; i++){ normal[i] = normals[globalIndex + i]; }" 
    "vertex[3] = 1.0;"
    "normal[3] = 1.0;"
    "//NORMAL TRANSFORMS"
    "for(int i = 0; i < 64; i += 16){"
    "   matrixByVector(&matrices[i], normal);"
    "}"
    "//VERTEX TRANSFORMS"
    "int matrixBufferLength = matrixCount * 16;"
    "for(int i = 0; i < matrixBufferLength; i += 16){"
    "   matrixByVector(&matrices[i], vertex);"
    "}"
    "float hardcodedLight[3] = {0.0, 0.0, 1.0};"
    "float lightScalar = dotProduct(normal, hardcodedLight);"
    "outputColors[globalIndex] = RGBClamp(colors[globalIndex] * lightScalar);"
    "outputColors[globalIndex + 1] = RGBClamp(colors[globalIndex + 1] * lightScalar);"
    "outputColors[globalIndex + 2] = RGBClamp(colors[globalIndex + 2] * lightScalar);"
    "//perspective division"
    "if(vertex[3] >= 1.0){"
    "   vertex[0] = vertex[0] / vertex[3];"
    "   vertex[1] = vertex[1] / vertex[3];"
    "}"
    "//screen space conversion"
    "vertex[0] *= width;"
    "vertex[1] *= height;"
    "vertex[2] *= height;"
    "for(int i = 0; i < 3; i++) { outputVertices[globalIndex + i] = vertex[i]; }"
"}";