LibGfx: Use gamma-corrected interpolation for color gradients

Switch over to gamma-aware interpolation. This causes color gradients
to not look so dark in the middle. SIMD optimized code is provided for
sse1 enabled builds.

Fixes #1342.

3 files changed, 165 insertions, 21 deletions

diff --git a/Libraries/LibGfx/Color.h b/Libraries/LibGfx/Color.h
index 721b5ad02a..e7f5eb4a42 100644
--- a/Libraries/LibGfx/Color.h
+++ b/Libraries/LibGfx/Color.h
@@ -89,10 +89,10 @@ public:
     static constexpr Color from_rgb(unsigned rgb) { return Color(rgb | 0xff000000); }
     static constexpr Color from_rgba(unsigned rgba) { return Color(rgba); }
 
-    u8 red() const { return (m_value >> 16) & 0xff; }
-    u8 green() const { return (m_value >> 8) & 0xff; }
-    u8 blue() const { return m_value & 0xff; }
-    u8 alpha() const { return (m_value >> 24) & 0xff; }
+    constexpr u8 red() const { return (m_value >> 16) & 0xff; }
+    constexpr u8 green() const { return (m_value >> 8) & 0xff; }
+    constexpr u8 blue() const { return m_value & 0xff; }
+    constexpr u8 alpha() const { return (m_value >> 24) & 0xff; }
 
     void set_alpha(u8 value)
     {
diff --git a/Libraries/LibGfx/Gamma.h b/Libraries/LibGfx/Gamma.h
new file mode 100644
index 0000000000..ace2292953
--- /dev/null
+++ b/Libraries/LibGfx/Gamma.h
@@ -0,0 +1,157 @@
+/*
+ * Copyright (c) 2020, the SerenityOS developers.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice, this
+ *    list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+ * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#pragma once
+
+#include "Color.h"
+#include <math.h>
+#include <xmmintrin.h>
+
+#define GAMMA 2.2
+
+// Most computer graphics are stored in the sRGB color space, which stores something close to
+// the square root of the display intensity of each color channel. This is problematic for most
+// operations that we want to perform on colors, since they typically assume that color scales
+// linearly (e.g. rgb(127, 0, 0) is half as bright as rgb(255, 0, 0)). This causes incorrect
+// results that look more gray than they should, to fix this we have to convert colors to the linear
+// color space before performing these operations, then convert back before displaying.
+//
+// Conversion between linear and sRGB spaces are somewhat expensive to do on the CPU, so we instead
+// interpret sRGB colors as gamma2.2 colors, which are close enough in most cases to be indistinguishable.
+// Gamma 2.2 colors follow the simple rule of `display_intensity = pow(stored_intensity, 2.2)`.
+// This module implements some fast color space transforms between the gamma2.2 and linear color spaces, plus
+// some common primitive operations like blending.
+//
+// For a more in-depth overview of how gamma-adjustment works, check out:
+// https://blog.johnnovak.net/2016/09/21/what-every-coder-should-know-about-gamma/
+
+namespace Gfx {
+
+#ifndef NO_FPU
+
+#    ifdef __SSE__
+
+// A vector of 4 floats, aligned for SSE instructions
+typedef float v4sf __attribute__((vector_size(16)));
+
+// Transform v4sf from gamma2.2 space to linear space
+// Assumes x is in range [0, 1]
+constexpr v4sf gamma_to_linear4(v4sf x)
+{
+    return (0.8f + 0.2f * x) * x * x;
+}
+
+constexpr v4sf linear_to_gamma4(v4sf x)
+{
+    // Source for approximation: https://mimosa-pudica.net/fast-gamma/
+    constexpr float a = 0.00279491f;
+    constexpr float b = 1.15907984f;
+    constexpr float c = (b / sqrt(1 + a)) - 1;
+    return ((b * __builtin_ia32_rsqrtps(x + a)) - c) * x;
+}
+
+// Linearize v1 and v2, lerp them by mix factor, then convert back.
+// The output is entirely v1 when mix = 0 and entirely v2 when mix = 1
+constexpr v4sf gamma_accurate_lerp4(v4sf v1, v4sf v2, float mix)
+{
+    return linear_to_gamma4(gamma_to_linear4(v1) * (1 - mix) + gamma_to_linear4(v2) * mix);
+}
+
+// Convert a and b to linear space, blend them by mix factor, then convert back using sse1.
+// The output is entirely a when mix = 0 and entirely b when mix = 1
+constexpr Color gamma_accurate_blend4(Color a, Color b, float mix)
+{
+    v4sf ac = {
+        (float)a.red(),
+        (float)a.green(),
+        (float)a.blue(),
+        0.f,
+    };
+    v4sf bc = {
+        (float)b.red(),
+        (float)b.green(),
+        (float)b.blue(),
+        0.f,
+    };
+    v4sf out = 255.f * gamma_accurate_lerp4(ac / 255.f, bc / 255.f, mix);
+    return Color(out[0], out[1], out[2]);
+}
+
+#    endif
+
+// Transform scalar from gamma2.2 space to linear space
+// Assumes x is in range [0, 1]
+constexpr float gamma_to_linear(float x)
+{
+#    ifdef ACCURATE_GAMMA_ADJUSTMENT
+    // Slower, but more accurate
+    return pow(x, GAMMA);
+#    else
+    return (0.8 + 0.2 * x) * x * x;
+#    endif
+}
+
+// Transform scalar from linear space to gamma2.2 space
+// Assumes x is in range [0, 1]
+constexpr float linear_to_gamma(float x)
+{
+#    ifdef ACCURATE_GAMMA_ADJUSTMENT
+    // Slower, but more accurate
+    return pow(x, 1. / GAMMA);
+#    else
+    // Source for approximation: https://mimosa-pudica.net/fast-gamma/
+    constexpr float a = 0.00279491;
+    constexpr float b = 1.15907984;
+    constexpr float c = (b / sqrt(1 + a)) - 1;
+    return ((b / __builtin_sqrt(x + a)) - c) * x;
+#    endif
+}
+
+// Linearize v1 and v2, lerp them by mix factor, then convert back.
+// The output is entirely v1 when mix = 0 and entirely v2 when mix = 1
+constexpr float gamma_accurate_lerp(float v1, float v2, float mix)
+{
+    return linear_to_gamma(gamma_to_linear(v1) * (1 - mix) + gamma_to_linear(v2) * mix);
+}
+
+// Convert a and b to linear space, blend them by mix factor, then convert back.
+// The output is entirely a when mix = 0 and entirely b when mix = 1
+constexpr Color gamma_accurate_blend(Color a, Color b, float mix)
+{
+#    ifdef __SSE__
+    return gamma_accurate_blend4(a, b, mix);
+#    else
+    return {
+        static_cast<u8>(255. * gamma_accurate_lerp(a.red() / 255., b.red() / 255., mix)),
+        static_cast<u8>(255. * gamma_accurate_lerp(a.green() / 255., b.green() / 255., mix)),
+        static_cast<u8>(255. * gamma_accurate_lerp(a.blue() / 255., b.blue() / 255., mix)),
+    };
+#    endif
+}
+
+#endif
+
+}
diff --git a/Libraries/LibGfx/Painter.cpp b/Libraries/LibGfx/Painter.cpp
index 63015774ad..9af7a136b7 100644
--- a/Libraries/LibGfx/Painter.cpp
+++ b/Libraries/LibGfx/Painter.cpp
@@ -28,6 +28,7 @@
 #include "Bitmap.h"
 #include "Emoji.h"
 #include "Font.h"
+#include "Gamma.h"
 #include <AK/Assertions.h>
 #include <AK/Function.h>
 #include <AK/Memory.h>
@@ -200,24 +201,13 @@ void Painter::fill_rect_with_gradient(Orientation orientation, const IntRect& a_
     RGBA32* dst = m_target->scanline(clipped_rect.top()) + clipped_rect.left();
     const size_t dst_skip = m_target->pitch() / sizeof(RGBA32);
 
-    float increment = (1.0 / ((rect.primary_size_for_orientation(orientation)) / 255.0));
-
-    int r2 = gradient_start.red();
-    int g2 = gradient_start.green();
-    int b2 = gradient_start.blue();
-    int r1 = gradient_end.red();
-    int g1 = gradient_end.green();
-    int b1 = gradient_end.blue();
+    float increment = (1.0 / ((rect.primary_size_for_orientation(orientation))));
 
     if (orientation == Orientation::Horizontal) {
         for (int i = clipped_rect.height() - 1; i >= 0; --i) {
             float c = offset * increment;
             for (int j = 0; j < clipped_rect.width(); ++j) {
-                dst[j] = Color(
-                    r1 / 255.0 * c + r2 / 255.0 * (255 - c),
-                    g1 / 255.0 * c + g2 / 255.0 * (255 - c),
-                    b1 / 255.0 * c + b2 / 255.0 * (255 - c))
-                             .value();
+                dst[j] = gamma_accurate_blend(gradient_start, gradient_end, c).value();
                 c += increment;
             }
             dst += dst_skip;
@@ -225,10 +215,7 @@ void Painter::fill_rect_with_gradient(Orientation orientation, const IntRect& a_
     } else {
         float c = offset * increment;
         for (int i = clipped_rect.height() - 1; i >= 0; --i) {
-            Color color(
-                r1 / 255.0 * c + r2 / 255.0 * (255 - c),
-                g1 / 255.0 * c + g2 / 255.0 * (255 - c),
-                b1 / 255.0 * c + b2 / 255.0 * (255 - c));
+            auto color = gamma_accurate_blend(gradient_start, gradient_end, c);
             for (int j = 0; j < clipped_rect.width(); ++j) {
                 dst[j] = color.value();
             }