1 files changed, 84 insertions, 8 deletions

diff --git a/indra/newview/app_settings/shaders/class1/deferred/sunLightSSAOF.glsl b/indra/newview/app_settings/shaders/class1/deferred/sunLightSSAOF.glsl
index 006a49d9a0..403df87853 100644
--- a/indra/newview/app_settings/shaders/class1/deferred/sunLightSSAOF.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/sunLightSSAOF.glsl
@@ -35,26 +35,102 @@ out vec4 frag_color;
 
 //class 1 -- no shadow, SSAO only
 
+uniform sampler2DRect depthMap;
 uniform sampler2DRect normalMap;
+uniform sampler2D noiseMap;
+
 
 // Inputs
+uniform float ssao_radius;
+uniform float ssao_max_radius;
+uniform float ssao_factor;
+uniform float ssao_factor_inv;
+
 VARYING vec2 vary_fragcoord;
 
+uniform mat4 inv_proj;
+uniform vec2 screen_res;
+
 vec3 decode_normal (vec2 enc);
-vec3 getNorm(vec2 pos_screen);
-vec4 getPosition(vec2 pos_screen);
+
+vec4 getPosition(vec2 pos_screen)
+{
+	float depth = texture2DRect(depthMap, pos_screen.xy).r;
+	vec2 sc = pos_screen.xy*2.0;
+	sc /= screen_res;
+	sc -= vec2(1.0,1.0);
+	vec4 ndc = vec4(sc.x, sc.y, 2.0*depth-1.0, 1.0);
+	vec4 pos = inv_proj * ndc;
+	pos /= pos.w;
+	pos.w = 1.0;
+	return pos;
+}
 
 //calculate decreases in ambient lighting when crowded out (SSAO)
-float calcAmbientOcclusion(vec4 pos, vec3 norm, vec2 pos_screen);
+float calcAmbientOcclusion(vec4 pos, vec3 norm)
+{
+	float ret = 1.0;
+	
+	vec2 kern[8];
+	// exponentially (^2) distant occlusion samples spread around origin
+	kern[0] = vec2(-1.0, 0.0) * 0.125*0.125;
+	kern[1] = vec2(1.0, 0.0) * 0.250*0.250;
+	kern[2] = vec2(0.0, 1.0) * 0.375*0.375;
+	kern[3] = vec2(0.0, -1.0) * 0.500*0.500;
+	kern[4] = vec2(0.7071, 0.7071) * 0.625*0.625;
+	kern[5] = vec2(-0.7071, -0.7071) * 0.750*0.750;
+	kern[6] = vec2(-0.7071, 0.7071) * 0.875*0.875;
+	kern[7] = vec2(0.7071, -0.7071) * 1.000*1.000;
+
+	vec2 pos_screen = vary_fragcoord.xy;
+	vec3 pos_world = pos.xyz;
+	vec2 noise_reflect = texture2D(noiseMap, vary_fragcoord.xy/128.0).xy;
+		
+	float angle_hidden = 0.0;
+	int points = 0;
+		
+	float scale = min(ssao_radius / -pos_world.z, ssao_max_radius);
+		
+	// it was found that keeping # of samples a constant was the fastest, probably due to compiler optimizations unrolling?)
+	for (int i = 0; i < 8; i++)
+	{
+		vec2 samppos_screen = pos_screen + scale * reflect(kern[i], noise_reflect);
+		vec3 samppos_world = getPosition(samppos_screen).xyz; 
+			
+		vec3 diff = pos_world - samppos_world;
+		float dist2 = dot(diff, diff);
+			
+		// assume each sample corresponds to an occluding sphere with constant radius, constant x-sectional area
+		// --> solid angle shrinking by the square of distance
+		//radius is somewhat arbitrary, can approx with just some constant k * 1 / dist^2
+		//(k should vary inversely with # of samples, but this is taken care of later)
+			
+		angle_hidden = angle_hidden + float(dot((samppos_world - 0.05*norm - pos_world), norm) > 0.0) * min(1.0/dist2, ssao_factor_inv);
+			
+		// 'blocked' samples (significantly closer to camera relative to pos_world) are "no data", not "no occlusion" 
+		points = points + int(diff.z > -1.0);
+	}
+		
+	angle_hidden = min(ssao_factor*angle_hidden/float(points), 1.0);
+		
+	ret = (1.0 - (float(points != 0) * angle_hidden));
+	
+	return min(ret, 1.0);
+}
 
 void main() 
 {
-	vec2 pos_screen = vary_fragcoord.xy;	
-	vec4 pos        = getPosition(pos_screen);
-	vec3 norm       = getNorm(pos_screen);
-
+	vec2 pos_screen = vary_fragcoord.xy;
+	
+	//try doing an unproject here
+	
+	vec4 pos = getPosition(pos_screen);
+	
+	vec3 norm = texture2DRect(normalMap, pos_screen).xyz;
+	norm = decode_normal(norm.xy);
+		
 	frag_color[0] = 1.0;
-	frag_color[1] = calcAmbientOcclusion(pos, norm, pos_screen);
+	frag_color[1] = calcAmbientOcclusion(pos, norm);
 	frag_color[2] = 1.0; 
 	frag_color[3] = 1.0;
 }