Use early bailing to improve 3D shadow filtering performance in GLES3

This makes PCF13 much faster than it previously was. The performance
difference between PCF5 and PCF13 is now much lower.

drivers/gles3/shaders/scene.glsl

@@ -1232,7 +1232,20 @@ LIGHT_SHADER_CODE
 float sample_shadow(highp sampler2DShadow shadow, vec2 shadow_pixel_size, vec2 pos, float depth, vec4 clamp_rect) {
 #ifdef SHADOW_MODE_PCF_13
 
-	float avg = textureProj(shadow, vec4(pos, depth, 1.0));
+	float avg = textureProj(shadow, vec4(pos + vec2(shadow_pixel_size.x * 2.0, 0.0), depth, 1.0));
+	avg += textureProj(shadow, vec4(pos + vec2(-shadow_pixel_size.x * 2.0, 0.0), depth, 1.0));
+	avg += textureProj(shadow, vec4(pos + vec2(0.0, shadow_pixel_size.y * 2.0), depth, 1.0));
+	avg += textureProj(shadow, vec4(pos + vec2(0.0, -shadow_pixel_size.y * 2.0), depth, 1.0));
+	// Early bail if distant samples are fully shaded (or none are shaded) to improve performance.
+	if (avg <= 0.000001) {
+		// None shaded at all.
+		return 0.0;
+	} else if (avg >= 3.999999) {
+		// All fully shaded.
+		return 1.0;
+	}
+
+	avg += textureProj(shadow, vec4(pos, depth, 1.0));
 	avg += textureProj(shadow, vec4(pos + vec2(shadow_pixel_size.x, 0.0), depth, 1.0));
 	avg += textureProj(shadow, vec4(pos + vec2(-shadow_pixel_size.x, 0.0), depth, 1.0));
 	avg += textureProj(shadow, vec4(pos + vec2(0.0, shadow_pixel_size.y), depth, 1.0));
@@ -1241,10 +1254,6 @@ float sample_shadow(highp sampler2DShadow shadow, vec2 shadow_pixel_size, vec2 p
 	avg += textureProj(shadow, vec4(pos + vec2(-shadow_pixel_size.x, shadow_pixel_size.y), depth, 1.0));
 	avg += textureProj(shadow, vec4(pos + vec2(shadow_pixel_size.x, -shadow_pixel_size.y), depth, 1.0));
 	avg += textureProj(shadow, vec4(pos + vec2(-shadow_pixel_size.x, -shadow_pixel_size.y), depth, 1.0));
-	avg += textureProj(shadow, vec4(pos + vec2(shadow_pixel_size.x * 2.0, 0.0), depth, 1.0));
-	avg += textureProj(shadow, vec4(pos + vec2(-shadow_pixel_size.x * 2.0, 0.0), depth, 1.0));
-	avg += textureProj(shadow, vec4(pos + vec2(0.0, shadow_pixel_size.y * 2.0), depth, 1.0));
-	avg += textureProj(shadow, vec4(pos + vec2(0.0, -shadow_pixel_size.y * 2.0), depth, 1.0));
 	return avg * (1.0 / 13.0);
 #endif