Add double precision build support.

This commit is contained in:
bruvzg
2022-03-20 18:19:27 +02:00
parent f262ae9a6f
commit e06d5cd414
5 changed files with 49 additions and 24 deletions

View File

@@ -159,7 +159,7 @@ public:
inline Color blend(const Color &p_over) const {
Color res;
float sa = (real_t)1.0 - p_over.a;
float sa = 1.0 - p_over.a;
res.a = a * sa + p_over.a;
if (res.a == 0) {
return Color(0, 0, 0, 0);
@@ -173,16 +173,16 @@ public:
inline Color srgb_to_linear() const {
return Color(
r < (real_t)0.04045 ? r * (real_t)(1.0 / 12.92) : Math::pow((r + (real_t)0.055) * (real_t)(1.0 / (1.0 + 0.055)), (real_t)2.4),
g < (real_t)0.04045 ? g * (real_t)(1.0 / 12.92) : Math::pow((g + (real_t)0.055) * (real_t)(1.0 / (1.0 + 0.055)), (real_t)2.4),
b < (real_t)0.04045 ? b * (real_t)(1.0 / 12.92) : Math::pow((b + (real_t)0.055) * (real_t)(1.0 / (1.0 + 0.055)), (real_t)2.4),
r < 0.04045 ? r * (1.0 / 12.92) : Math::pow((r + 0.055f) * (1.0 / (1.0 + 0.055)), 2.4),
g < 0.04045 ? g * (1.0 / 12.92) : Math::pow((g + 0.055f) * (1.0 / (1.0 + 0.055)), 2.4),
b < 0.04045 ? b * (1.0 / 12.92) : Math::pow((b + 0.055f) * (1.0 / (1.0 + 0.055)), 2.4),
a);
}
inline Color linear_to_srgb() const {
return Color(
r < (real_t)0.0031308 ? (real_t)12.92 * r : (real_t)(1.0 + 0.055) * Math::pow(r, (real_t)(1.0 / 2.4)) - (real_t)0.055,
g < (real_t)0.0031308 ? (real_t)12.92 * g : (real_t)(1.0 + 0.055) * Math::pow(g, (real_t)(1.0 / 2.4)) - (real_t)0.055,
b < (real_t)0.0031308 ? (real_t)12.92 * b : (real_t)(1.0 + 0.055) * Math::pow(b, (real_t)(1.0 / 2.4)) - (real_t)0.055, a);
r < 0.0031308f ? 12.92f * r : (1.0f + 0.055f) * Math::pow(r, 1.0f / 2.4f) - 0.055f,
g < 0.0031308f ? 12.92f * g : (1.0f + 0.055f) * Math::pow(g, 1.0f / 2.4f) - 0.055f,
b < 0.0031308f ? 12.92f * b : (1.0f + 0.055f) * Math::pow(b, 1.0f / 2.4f) - 0.055f, a);
}
static Color hex(uint32_t p_hex);
@@ -204,14 +204,14 @@ public:
operator String() const;
// For the binder.
inline void set_r8(int32_t r8) { r = (Math::clamp(r8, 0, 255) / (real_t)255.0); }
inline int32_t get_r8() const { return int32_t(Math::clamp(r * (real_t)255.0, (real_t)0.0, (real_t)255.0)); }
inline void set_g8(int32_t g8) { g = (Math::clamp(g8, 0, 255) / (real_t)255.0); }
inline int32_t get_g8() const { return int32_t(Math::clamp(g * (real_t)255.0, (real_t)0.0, (real_t)255.0)); }
inline void set_b8(int32_t b8) { b = (Math::clamp(b8, 0, 255) / (real_t)255.0); }
inline int32_t get_b8() const { return int32_t(Math::clamp(b * (real_t)255.0, (real_t)0.0, (real_t)255.0)); }
inline void set_a8(int32_t a8) { a = (Math::clamp(a8, 0, 255) / (real_t)255.0); }
inline int32_t get_a8() const { return int32_t(Math::clamp(a * (real_t)255.0, (real_t)0.0, (real_t)255.0)); }
inline void set_r8(int32_t r8) { r = (Math::clamp(r8, 0, 255) / 255.0); }
inline int32_t get_r8() const { return int32_t(Math::clamp(r * 255.0, 0.0, 255.0)); }
inline void set_g8(int32_t g8) { g = (Math::clamp(g8, 0, 255) / 255.0); }
inline int32_t get_g8() const { return int32_t(Math::clamp(g * 255.0, 0.0, 255.0)); }
inline void set_b8(int32_t b8) { b = (Math::clamp(b8, 0, 255) / 255.0); }
inline int32_t get_b8() const { return int32_t(Math::clamp(b * 255.0, 0.0, 255.0)); }
inline void set_a8(int32_t a8) { a = (Math::clamp(a8, 0, 255) / 255.0); }
inline int32_t get_a8() const { return int32_t(Math::clamp(a * 255.0, 0.0, 255.0)); }
inline void set_h(float p_h) { set_hsv(p_h, get_s(), get_v()); }
inline void set_s(float p_s) { set_hsv(get_h(), p_s, get_v()); }