X-Git-Url: https://git.sesse.net/?p=movit;a=blobdiff_plain;f=overlay_effect.frag;h=9e5709d1dad6271535701316d4ff1b60c4533d99;hp=38970eab7e85469943353b85e9690d87ba59e5a8;hb=a770ba7526844f395b08625480befad83b8fcadf;hpb=5e05605a85a15944b95d7000745446756cdcbfa8

diff --git a/overlay_effect.frag b/overlay_effect.frag
index 38970ea..9e5709d 100644
--- a/overlay_effect.frag
+++ b/overlay_effect.frag
@@ -1,26 +1,19 @@
-// If we didn't have to worry about alpha in the bottom layer,
-// this would be a simple mix(). However, since people might
-// compose multiple layers together and we don't really have
-// any control over the order, it's better to do it right.
+// It's actually (but surprisingly) not correct to do a mix() here;
+// it would be if we had postmultiplied alpha and didn't have to worry
+// about alpha in the bottom layer, but given that we use premultiplied
+// alpha all over, top shouldn't actually be multiplied by anything.
 //
 // These formulas come from Wikipedia:
 //
 //   http://en.wikipedia.org/wiki/Alpha_compositing
+//
+// We use the associative version given. However, note that since we want
+// _output_ to be premultiplied, C_o from Wikipedia is not what we want,
+// but rather c_o (which is not explicitly given, but obviously is just
+// C_o without the division by alpha_o).
 
 vec4 FUNCNAME(vec2 tc) {
 	vec4 bottom = INPUT1(tc);
 	vec4 top = INPUT2(tc);
-	float new_alpha = mix(bottom.a, 1.0, top.a);
-	if (new_alpha < 1e-6) {
-		// new_alpha = 0 only if top.a = bottom.a = 0, at least as long as
-		// both alphas are within range. (If they're not, the result is not
-		// meaningful anyway.) And if new_alpha = 0, we don't really have
-		// any meaningful output anyway, so just set it to zero instead of
-		// getting division-by-zero below.
-		return vec4(0.0);
-	} else {
-		vec3 premultiplied_color = mix(bottom.rgb * bottom.aaa, top.rgb, top.a);
-		vec3 color = premultiplied_color / new_alpha;
-		return vec4(color.r, color.g, color.b, new_alpha);
-	}
+	return top + (1.0 - top.a) * bottom;
 }