Function five_bit_pixel

src/fl/gfx/five_bit_hd_gamma.cpp.hpp:36–77

Core per-pixel transform, fully inlined. All uniform values are precomputed by the caller and passed in to avoid per-pixel branches.

// Core per-pixel transform, fully inlined. All uniform values are
// precomputed by the caller and passed in to avoid per-pixel branches.
FL_ALWAYS_INLINE void five_bit_pixel(
    u16 r16, u16 g16, u16 b16, u8 brightness,
    // Precomputed: (1 + brightness), or 0 if brightness==0xff (skip scaling)
    u16 bright_p1, bool apply_brightness,
    CRGB *out, u8 *out_power_5bit) {

    // All-zero fast path (rare but worth checking — writes are cheap).
    if ((r16 | g16 | b16) == 0) {
        *out = CRGB(0, 0, 0);
        *out_power_5bit = (brightness <= 31) ? brightness : 31;
        return;
    }

    // Apply brightness: branchless multiply, skip only if brightness==0xff.
    if (apply_brightness) {
        r16 = scale16by8_nozero(r16, bright_p1);
        g16 = scale16by8_nozero(g16, bright_p1);
        b16 = scale16by8_nozero(b16, bright_p1);
    }

    // max3 via branchless ternary (compiler emits cmov on x86, csel on ARM).
    u16 scale = r16;
    if (g16 > scale) scale = g16;
    if (b16 > scale) scale = b16;

    if (scale == 0) {
        *out = CRGB(0, 0, 0);
        *out_power_5bit = 0;
        return;
    }

    // 5-bit quantized scale at or above maximum value.
    scale = (scale + (2047 - (scale >> 5))) >> 11;

    u32 scalef = FL_PGM_READ_DWORD_ALIGNED(&BRIGHT_SCALE[scale]);
    u8 r8 = static_cast<u8>((r16 * scalef + 0x808000) >> 24);
    u8 g8 = static_cast<u8>((g16 * scalef + 0x808000) >> 24);
    u8 b8 = static_cast<u8>((b16 * scalef + 0x808000) >> 24);

    *out = CRGB(r8, g8, b8);
    *out_power_5bit = static_cast<u8>(scale);
}

} // namespace five_bit_impl
} // anonymous namespace