NORSPEC-290 Shader optimization WIP -- remove some more divides and normalizes from various lighting functions, rework flow control based on profile feedback.

4 files changed, 46 insertions, 54 deletions

diff --git a/indra/newview/app_settings/shaders/class1/deferred/multiPointLightF.glsl b/indra/newview/app_settings/shaders/class1/deferred/multiPointLightF.glsl
index a955ef6e9d..ed51e01a53 100755
--- a/indra/newview/app_settings/shaders/class1/deferred/multiPointLightF.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/multiPointLightF.glsl
@@ -107,14 +107,14 @@ void main()
 	for (int i = 0; i < LIGHT_COUNT; ++i)
 	{
 		vec3 lv = light[i].xyz-pos;
-		float dist = length(lv);
-		dist /= light[i].w;
+		float d = length(lv);
+		float dist = d * light[i].w;
 		if (dist <= 1.0)
 		{
 			float da = dot(norm, lv);
 			if (da > 0.0)
 			{
-				lv = normalize(lv);
+				lv /= d;
 				da = dot(norm, lv);
 			
 				float fa = light_col[i].a+1.0;
diff --git a/indra/newview/app_settings/shaders/class1/deferred/pointLightF.glsl b/indra/newview/app_settings/shaders/class1/deferred/pointLightF.glsl
index 106d48bd71..f162f70592 100755
--- a/indra/newview/app_settings/shaders/class1/deferred/pointLightF.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/pointLightF.glsl
@@ -35,7 +35,7 @@ uniform sampler2DRect diffuseRect;
 uniform sampler2DRect specularRect;
 uniform sampler2DRect normalMap;
 uniform samplerCube environmentMap;
-uniform sampler2D noiseMap;
+//uniform sampler2D noiseMap;
 uniform sampler2D lightFunc;
 uniform sampler2DRect depthMap;
 
@@ -93,64 +93,56 @@ void main()
 	
 	vec3 pos = getPosition(frag.xy).xyz;
 	vec3 lv = trans_center.xyz-pos;
-	float dist = length(lv);
-	dist /= size;
-	if (dist > 1.0)
-	{
-		discard;
-	}
+	float d = length(lv);
 	
-	vec3 norm = texture2DRect(normalMap, frag.xy).xyz;
-	norm = decode_normal(norm.xy); // unpack norm
-	float da = dot(norm, lv);
-	if (da < 0.0)
+	float dist = d*size;
+
+	vec3 col = vec3(0.0);
+	if (dist <= 1.0)
 	{
-		discard;
-	}
+		vec3 norm = texture2DRect(normalMap, frag.xy).xyz;
+		norm = decode_normal(norm.xy); // unpack norm
+		float da = dot(norm, lv);
 	
-	norm = normalize(norm);
-	lv = normalize(lv);
-	da = dot(norm, lv);
+		norm = normalize(norm);
+		lv = normalize(lv);
+		da = max(dot(norm, lv), 0.0);
 	
-	float noise = texture2D(noiseMap, frag.xy/128.0).b;
+		//float noise = texture2D(noiseMap, frag.xy/128.0).b;
 	
-	vec3 col = texture2DRect(diffuseRect, frag.xy).rgb;
-	float fa = falloff+1.0;
-	float dist_atten = clamp(1.0-(dist-1.0*(1.0-fa))/fa, 0.0, 1.0);
-	dist_atten *= dist_atten;
-	dist_atten *= 2.0;
+		col = texture2DRect(diffuseRect, frag.xy).rgb;
+		float fa = falloff+1.0;
+		float dist_atten = clamp(1.0-(dist-1.0+fa)/fa, 0.0, 1.0);
+		dist_atten *= dist_atten;
+		dist_atten *= 2.0;
 	
-	float lit = da * dist_atten * noise;
+		float lit = da * dist_atten; // * noise;
 
-	col = color.rgb*lit*col;
+		col = color.rgb*lit*col;
 
-	vec4 spec = texture2DRect(specularRect, frag.xy);
-	if (spec.a > 0.0)
-	{
-		lit = min(da*6.0, 1.0) * dist_atten;
-
-		vec3 npos = -normalize(pos);
-		vec3 h = normalize(lv+npos);
-		float nh = dot(norm, h);
-		float nv = dot(norm, npos);
-		float vh = dot(npos, h);
-		float sa = nh;
-		float fres = pow(1 - dot(h, npos), 5) * 0.4+0.5;
-		float gtdenom = 2 * nh;
-		float gt = max(0,(min(gtdenom * nv / vh, gtdenom * da / vh)));
-
-		if (nh > 0.0)
+		vec4 spec = texture2DRect(specularRect, frag.xy);
+		if (spec.a > 0.0)
 		{
-			float scol = fres*texture2D(lightFunc, vec2(nh, spec.a)).r*gt/(nh*da);
-			col += lit*scol*color.rgb*spec.rgb;
+			lit = min(da*6.0, 1.0) * dist_atten;
+
+			vec3 npos = -normalize(pos);
+			vec3 h = normalize(lv+npos);
+			float nh = dot(norm, h);
+			float nv = dot(norm, npos);
+			float vh = dot(npos, h);
+			float sa = nh;
+			float fres = pow(1 - dot(h, npos), 5) * 0.4+0.5;
+			float gtdenom = 2 * nh;
+			float gt = max(0,(min(gtdenom * nv / vh, gtdenom * da / vh)));
+
+			if (nh > 0.0)
+			{
+				float scol = fres*texture2D(lightFunc, vec2(nh, spec.a)).r*gt/(nh*da);
+				col += lit*scol*color.rgb*spec.rgb;
+			}
 		}
 	}
 	
-	if (dot(col, col) <= 0.0)
-	{
-		discard;
-	}
-		
 	frag_color.rgb = col;	
 	frag_color.a = 0.0;
 }
diff --git a/indra/newview/app_settings/shaders/class1/deferred/pointLightV.glsl b/indra/newview/app_settings/shaders/class1/deferred/pointLightV.glsl
index a5625fbc16..aeef09cf5f 100755
--- a/indra/newview/app_settings/shaders/class1/deferred/pointLightV.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/pointLightV.glsl
@@ -37,7 +37,7 @@ VARYING vec3 trans_center;
 void main()
 {
 	//transform vertex
-	vec3 p = position*size+center;
+	vec3 p = position*1.f/size+center;
 	vec4 pos = modelview_projection_matrix * vec4(p.xyz, 1.0);
 	vary_fragcoord = pos;
 	trans_center = (modelview_matrix*vec4(center.xyz, 1.0)).xyz;
diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp
index 7072c95c3e..7fa0a972ad 100755
--- a/indra/newview/pipeline.cpp
+++ b/indra/newview/pipeline.cpp
@@ -8680,7 +8680,7 @@ void LLPipeline::renderDeferredLighting()
 							
 							LLFastTimer ftm(FTM_LOCAL_LIGHTS);
 							gDeferredLightProgram.uniform3fv(LLShaderMgr::LIGHT_CENTER, 1, c);
-							gDeferredLightProgram.uniform1f(LLShaderMgr::LIGHT_SIZE, s);
+							gDeferredLightProgram.uniform1f(LLShaderMgr::LIGHT_SIZE, 1.f/s);
 							gDeferredLightProgram.uniform3fv(LLShaderMgr::DIFFUSE_COLOR, 1, col.mV);
 							gDeferredLightProgram.uniform1f(LLShaderMgr::LIGHT_FALLOFF, volume->getLightFalloff()*0.5f);
 							gGL.syncMatrices();
@@ -8701,7 +8701,7 @@ void LLPipeline::renderDeferredLighting()
 						glh::vec3f tc(c);
 						mat.mult_matrix_vec(tc);
 					
-						fullscreen_lights.push_back(LLVector4(tc.v[0], tc.v[1], tc.v[2], s));
+						fullscreen_lights.push_back(LLVector4(tc.v[0], tc.v[1], tc.v[2], 1.f/s));
 						light_colors.push_back(LLVector4(col.mV[0], col.mV[1], col.mV[2], volume->getLightFalloff()*0.5f));
 					}
 				}
@@ -8786,7 +8786,7 @@ void LLPipeline::renderDeferredLighting()
 					col[count].mV[1] = powf(col[count].mV[1], 2.2f);
 					col[count].mV[2] = powf(col[count].mV[2], 2.2f);*/
 					
-					far_z = llmin(light[count].mV[2]-light[count].mV[3], far_z);
+					far_z = llmin(light[count].mV[2]-1.f/light[count].mV[3], far_z);
 					//col[count] = pow4fsrgb(col[count], 2.2f);
 					count++;
 					if (count == max_count || fullscreen_lights.empty())