Optimize the implementation of scale.

[Xreki]

优化scale算子生成的代码，避免在代码中使用立即数。编译器默认立即数为double类型，影响整体性能。

• 优化前

  // Note: need to support vectorized operation
  __forceinline__ __host__ __device__
  T operator()(T x, const Arguments& args, const MatrixCoord& coord) const {
    T out;
    float op1_out0 = static_cast<float>(0.100000 * x + 0.000000);
    out = op1_out0;
    return out;
  }

• 优化后

    T out;
    float op1_scale = (0.100000);
    float op1_bias = (0.000000);
    float op1_out0 = static_cast<float>(op1_scale * x + op1_bias);
    out = op1_out0;
    return out;