#include "kernels.h"

const char* matrix_mul_kernel_source =
"__kernel void mmul (                                       \n"
"    __global float* A,                                \n"
"    __global float* B,                                \n"
"    __global float* C)                                \n"
"{                                                     \n"
"    int k;                                            \n"
"    int i = get_global_id(0);                         \n"
"    int j = get_global_id(1);                         \n"
"    float tmp;                                        \n"
"    if( (i < 4) && (j < 4) )                          \n"
"    {                                                 \n"
"        tmp = 0.0;                                    \n"
"        for(k = 0; k < 4; k++)                        \n"
"            tmp += A[i*4+k] * B[k*4+j];               \n"
"        C[i*4+j] = tmp;                               \n"
"    }                                                 \n"
"}                                                     \n" ;

const char* vertex_shader_kernel_source =
"__kernel                                                           \n"
"void vertex_shader( __global float* z_matrix,                      \n"
"                    __global float* VertexBuffer,                  \n"
"                    __global float* NewVertexBuffer)               \n"
"{                                                                  \n"
"    int index = 3 * get_global_id(0);                              \n"
"    float new_x = 0;                                               \n"
"    float new_y = 0;                                               \n"
"    float new_z = 0;                                               \n"
"    float new_w = 0;                                               \n"
"    for(int i = 0; i < 3; i++) {                                   \n"
"        new_x = new_x + VertexBuffer[index+i]*z_matrix[0*4 + i];   \n"
"        new_y = new_y + VertexBuffer[index+i]*z_matrix[1*4 + i];   \n"
"        new_z = new_z + VertexBuffer[index+i]*z_matrix[2*4 + i];   \n"
"        new_w = new_w + VertexBuffer[index+i]*z_matrix[3*4 + i];   \n"
"    }                                                              \n"
"    new_x = new_x + z_matrix[0*4 +3];                              \n"
"    new_y = new_y + z_matrix[1*4 +3];                              \n"
"    new_z = new_z + z_matrix[2*4 +3];                              \n"
"    new_w = new_w + z_matrix[3*4 +3];                              \n"
"    NewVertexBuffer[index+0] = new_x / new_w;                      \n"
"    NewVertexBuffer[index+1] = new_y / new_w;                      \n"
"    NewVertexBuffer[index+2] = new_z / new_w;                      \n"
"}                                                                  \n";