CUDA C linear & cubic interpolation kernels

Cubic_interp.cu

__global__ void cubic_interp(float* x, float* y, int n, float* x_new, float* y_new, int m) {

    int idx = blockIdx.x * blockDim.x + threadIdx.x;

    if (idx >= m) return;

    int i = 1;

    while (i < n && x[i] <= x_new[idx]) {

        i++;

    }

    i--;

    float h = x[i + 1] - x[i];

    float a = (x[i + 1] - x_new[idx]) / h;

    float b = (x_new[idx] - x[i]) / h;

    float c = ((y[i + 1] - y[i]) / h - (2 * (y[i + 1] - y[i])) / (h * h) * h * h) / h;

    float d = ((2 * (y[i + 1] - y[i])) / (h * h) * h * h - (y[i + 1] - y[i]) / h) / h;

    y_new[idx] = y[i] + a * (y[i + 1] - y[i]) + b * ((1.0 / 3.0) * c * h * h * h + (1.0 / 2.0) * d * h * h);

}

Linear_interpolation.cu

__global__ void linear_interp(float* x, float* y, int n, float* x_new, float* y_new, int m) {
    int idx = blockIdx.x * blockDim.x + threadIdx.x;
    if (idx >= m) return;

    int i = 0;
    while (i < n - 1 && x_new[idx] > x[i + 1]) {
        i++;
    }

    float slope = (y[i + 1] - y[i]) / (x[i + 1] - x[i]);
    y_new[idx] = y[i] + slope * (x_new[idx] - x[i]);
}