Perf/optimize encoder

.github/workflows/ci.yml

@@ -11,10 +11,9 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-latest, windows-latest, macos-latest]
+        os: [ubuntu-latest, windows-latest]
     runs-on: ${{ matrix.os }}
     timeout-minutes: 15
-    continue-on-error: ${{ matrix.os == 'macos-latest' }}
     steps:
       - uses: actions/checkout@v4
       - uses: mlugg/setup-zig@v2

src/core/color_space.zig

@@ -5,15 +5,30 @@ const RGBAPixel = types.RGBAPixel;
 const YCbCrPixel = types.YCbCrPixel;
 const YCbCrImage = types.YCbCrImage;
 
-pub fn rgbToYCbCr(r: u8, g: u8, b: u8) YCbCrPixel {
-    const rf: f64 = @floatFromInt(r);
-    const gf: f64 = @floatFromInt(g);
-    const bf: f64 = @floatFromInt(b);
+const Y_R: [256]f64 = buildLut(0.299);
+const Y_G: [256]f64 = buildLut(0.587);
+const Y_B: [256]f64 = buildLut(0.114);
+const Cb_R: [256]f64 = buildLut(-0.168736);
+const Cb_G: [256]f64 = buildLut(-0.331264);
+const Cb_B: [256]f64 = buildLut(0.5);
+const Cr_R: [256]f64 = buildLut(0.5);
+const Cr_G: [256]f64 = buildLut(-0.418688);
+const Cr_B: [256]f64 = buildLut(-0.081312);
 
+fn buildLut(comptime coeff: f64) [256]f64 {
+    var t: [256]f64 = undefined;
+    var i: usize = 0;
+    inline while (i < 256) : (i += 1) {
+        t[i] = coeff * @as(f64, @floatFromInt(i));
+    }
+    return t;
+}
+
+pub fn rgbToYCbCr(r: u8, g: u8, b: u8) YCbCrPixel {
     return .{
-        .y = 0.299 * rf + 0.587 * gf + 0.114 * bf,
-        .cb = -0.168736 * rf - 0.331264 * gf + 0.5 * bf + 128.0,
-        .cr = 0.5 * rf - 0.418688 * gf - 0.081312 * bf + 128.0,
+        .y = Y_R[r] + Y_G[g] + Y_B[b],
+        .cb = Cb_R[r] + Cb_G[g] + Cb_B[b] + 128.0,
+        .cr = Cr_R[r] + Cr_G[g] + Cr_B[b] + 128.0,
     };
 }