我应该在C++中手动写出"sub-results"吗？

在我的一段代码中，我从 3 const vec3 s 计算出mat4：

const vec3 position{...}, orientation{...}, scale{...};
float mat[4][4]
{
    {
        (2 * scale.x * cos(orientation.y + orientation.z) + scale.x * cos(orientation.x + orientation.y + orientation.z) - scale.x * cos(orientation.x - orientation.y + orientation.z) + 2 * scale.x * cos(orientation.y - orientation.z) - scale.x * cos(orientation.x + orientation.y - orientation.z) + scale.x * cos(orientation.x - orientation.y - orientation.z)) / 4,
        (-scale.x * cos(orientation.x + orientation.y) + scale.x * cos(orientation.x - orientation.y)) / 2,
        (-2 * scale.x * sin(orientation.y + orientation.z) - scale.x * sin(orientation.x + orientation.y + orientation.z) + scale.x * sin(orientation.x - orientation.y + orientation.z) + 2 * scale.x * sin(orientation.y - orientation.z) - scale.x * sin(orientation.x + orientation.y - orientation.z) + scale.x * sin(orientation.x - orientation.y - orientation.z)) / 4,
        0
    },
    {
        (-scale.y * cos(orientation.x + orientation.z) + scale.y * cos(orientation.x - orientation.z)) / 2,
        scale.y * cos(orientation.x),
        (scale.y * sin(orientation.x + orientation.z) + scale.y * sin(orientation.x - orientation.z)) / 2,
        0
    },
    {
        (2 * scale.z * sin(orientation.y + orientation.z) + scale.z * sin(orientation.x + orientation.y + orientation.z) + scale.z * sin(orientation.x - orientation.y + orientation.z) + 2 * scale.z * sin(orientation.y - orientation.z) - scale.z * sin(orientation.x + orientation.y - orientation.z) - scale.z * sin(orientation.x - orientation.y - orientation.z)) / 4,
        (-scale.z * sin(orientation.x + orientation.y) - scale.z * sin(orientation.x - orientation.y)) / 2,
        (2 * scale.z * cos(orientation.y + orientation.z) + scale.z * cos(orientation.x + orientation.y + orientation.z) + scale.z * cos(orientation.x - orientation.y + orientation.z) - 2 * scale.z * cos(orientation.y - orientation.z) + scale.z * cos(orientation.x + orientation.y - orientation.z) + scale.z * cos(orientation.x - orientation.y - orientation.z)) / 4,
        0
    },
    {
        (-2 * position.y * cos(orientation.x + orientation.z) + 2 * position.x * cos(orientation.y + orientation.z) + position.x * cos(orientation.x + orientation.y + orientation.z) - position.x * cos(orientation.x - orientation.y + orientation.z) + 2 * position.y * cos(orientation.x - orientation.z) + 2 * position.x * cos(orientation.y - orientation.z) - position.x * cos(orientation.x + orientation.y - orientation.z) + position.x * cos(orientation.x - orientation.y - orientation.z) + 2 * position.z * sin(orientation.y + orientation.z) + position.z * sin(orientation.x + orientation.y + orientation.z) + position.z * sin(orientation.x - orientation.y + orientation.z) + 2 * position.z * sin(orientation.y - orientation.z) - position.z * sin(orientation.x + orientation.y - orientation.z) - position.z * sin(orientation.x - orientation.y - orientation.z)) / 4,
        (2 * position.y * cos(orientation.x) - position.x * cos(orientation.x + orientation.y) + position.x * cos(orientation.x - orientation.y) - position.z * sin(orientation.x + orientation.y) - position.z * sin(orientation.x - orientation.y)) / 2,
        (2 * position.z * cos(orientation.y + orientation.z) + position.z * cos(orientation.x + orientation.y + orientation.z) + position.z * cos(orientation.x - orientation.y + orientation.z) - 2 * position.z * cos(orientation.y - orientation.z) + position.z * cos(orientation.x + orientation.y - orientation.z) + position.z * cos(orientation.x - orientation.y - orientation.z) + 2 * position.y * sin(orientation.x + orientation.z) - 2 * position.x * sin(orientation.y + orientation.z) - position.x * sin(orientation.x + orientation.y + orientation.z) + position.x * sin(orientation.x - orientation.y + orientation.z) + 2 * position.y * sin(orientation.x - orientation.z) + 2 * position.x * sin(orientation.y - orientation.z) - position.x * sin(orientation.x + orientation.y - orientation.z) + position.x * sin(orientation.x - orientation.y - orientation.z)) / 4,
        1
    }
};

如您所见，计算只包含几个简单的操作(+、-、*、/、sin和cos(，但其中许多操作出现了很多次。此类操作的示例包括：

orientation.y + orientation.z
orientation.x + oy_plus_oz
sin(orientation.x - oy_plus_oz)
//many more...

我不希望处理器多次计算它们，所以我试图用"shortucts"替换它们：

float oy_plus_oz = orientation.y + orientation.z;
float cos_oy_plus_oz = cos(oy_plus_oz);
float ox_plus_oy_plus_oz = orientation.x + oy_plus_oz;
float ox_minus_oy_plus_oz = orientation.x - oy_plus_oz;
float oy_minus_oz = orientation.y - orientation.z;
float ox_plus_oy_minus_oz = orientation.x + oy_minus_oz;
float ox_minus_oy_minus_oz = orientation.x - oy_minus_oz;
float cos_ox_plus_oy_plus_oz = cos(ox_plus_oy_plus_oz);
float cos_ox_minus_oy_plus_oz = cos(ox_minus_oy_plus_oz);
float two_mult_sx = 2 * scale.x;
float cos_oy_minus_oz = cos(oy_minus_oz);
float cos_ox_plus_oy_minus_oz = cos(ox_plus_oy_minus_oz);
float cos_ox_minus_oy_minus_oz = cos(ox_minus_oy_minus_oz);
float ox_plus_oy = orientation.x + orientation.y;
float ox_minus_oy = orientation.x - orientation.y;
float cos_ox_plus_oy = cos(ox_plus_oy);
float cos_ox_minus_oy = cos(ox_minus_oy);

有没有更好/更快的方法来做到这一点，还是完全没有必要？关键是这段代码应该尽可能快(避免多次对同一值进行不必要的计算(。