NORSPEC-290 Allow shader compiler to unroll loop in multiPointLightF

6 files changed, 69 insertions, 82 deletions

diff --git a/indra/newview/app_settings/shaders/class1/deferred/multiPointLightF.glsl b/indra/newview/app_settings/shaders/class1/deferred/multiPointLightF.glsl
index 868526d457..6612892f6e 100755
--- a/indra/newview/app_settings/shaders/class1/deferred/multiPointLightF.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/multiPointLightF.glsl
@@ -45,9 +45,8 @@ uniform float sun_wash;
 
 uniform int light_count;
 
-#define MAX_LIGHT_COUNT		16
-uniform vec4 light[MAX_LIGHT_COUNT];
-uniform vec4 light_col[MAX_LIGHT_COUNT];
+uniform vec4 light[LIGHT_COUNT];
+uniform vec4 light_col[LIGHT_COUNT];
 
 VARYING vec4 vary_fragcoord;
 uniform vec2 screen_res;
@@ -122,73 +121,59 @@ void main()
 	vec3 npos = normalize(-pos);
 
 	// As of OSX 10.6.7 ATI Apple's crash when using a variable size loop
-	for (int i = 0; i < MAX_LIGHT_COUNT; ++i)
+	for (int i = 0; i < LIGHT_COUNT; ++i)
 	{
-		bool light_contrib = (i < light_count);
-		
 		vec3 lv = light[i].xyz-pos;
 		float dist = length(lv);
 		dist /= light[i].w;
-		if (dist > 1.0)
+		if (dist <= 1.0)
 		{
-			light_contrib = false;
-		}
-		
-		float da = dot(norm, lv);
-		if (da < 0.0)
-		{
-			light_contrib = false;
-		}
-		
-		if (light_contrib)
-		{
-			lv = normalize(lv);
-			da = dot(norm, lv);
+			float da = dot(norm, lv);
+			if (da > 0.0)
+			{
+				lv = normalize(lv);
+				da = dot(norm, lv);
 			
-			float fa = light_col[i].a+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;
+				float fa = light_col[i].a+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;
 			
-			dist_atten *= noise;
+				dist_atten *= noise;
 
-			float lit = da * dist_atten;
+				float lit = da * dist_atten;
 						
-			vec3 col = light_col[i].rgb*lit*diff;
+				vec3 col = light_col[i].rgb*lit*diff;
 			
-			//vec3 col = vec3(dist2, light_col[i].a, lit);
+				//vec3 col = vec3(dist2, light_col[i].a, lit);
 			
-			if (spec.a > 0.0)
-			{
-				lit = min(da*6.0, 1.0) * dist_atten;
-				//vec3 ref = dot(pos+lv, norm);
-				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)
+				if (spec.a > 0.0)
 				{
-					float scol = fres*texture2D(lightFunc, vec2(nh, spec.a)).r*gt/(nh*da);
-					col += lit*scol*light_col[i].rgb*spec.rgb;
-					//col += spec.rgb;
+					lit = min(da*6.0, 1.0) * dist_atten;
+					//vec3 ref = dot(pos+lv, norm);
+					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*light_col[i].rgb*spec.rgb;
+						//col += spec.rgb;
+					}
 				}
-			}
 			
-			out_col += col;
+				out_col += col;
+			}
 		}
 	}
 	
-	if (dot(out_col, out_col) <= 0.0)
-	{
-		discard;
-	}
-	
 	frag_color.rgb = out_col;
 	frag_color.a = 0.0;
 }
diff --git a/indra/newview/featuretable_mac.txt b/indra/newview/featuretable_mac.txt
index 1c0d45c11b..0bdd425504 100755
--- a/indra/newview/featuretable_mac.txt
+++ b/indra/newview/featuretable_mac.txt
@@ -309,8 +309,8 @@ RenderVolumeLODFactor		1	2.0
 VertexShaderEnable			1	1
 WindLightUseAtmosShaders	1	1
 WLSkyDetail					1	128
-RenderDeferred				1	0
-RenderDeferredSSAO			1	0
+RenderDeferred				1	1
+RenderDeferredSSAO			1	1
 RenderShadowDetail			1	2
 RenderFSAASamples			1	2
 
diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp
index f021f4ed0f..700b31f8d3 100755
--- a/indra/newview/llface.cpp
+++ b/indra/newview/llface.cpp
@@ -817,12 +817,6 @@ BOOL LLFace::genVolumeBBoxes(const LLVolume &volume, S32 f,
 			size.mul(scale);
 		}
 
-		// Catch potential badness from normalization before it happens
-		//
-		llassert(mat_normal.mMatrix[0].isFinite3() && (mat_normal.mMatrix[0].dot3(mat_normal.mMatrix[0]).getF32() > F_APPROXIMATELY_ZERO));
-		llassert(mat_normal.mMatrix[1].isFinite3() && (mat_normal.mMatrix[1].dot3(mat_normal.mMatrix[1]).getF32() > F_APPROXIMATELY_ZERO));
-		llassert(mat_normal.mMatrix[2].isFinite3() && (mat_normal.mMatrix[2].dot3(mat_normal.mMatrix[2]).getF32() > F_APPROXIMATELY_ZERO));
-
 		mat_normal.mMatrix[0].normalize3fast();
 		mat_normal.mMatrix[1].normalize3fast();
 		mat_normal.mMatrix[2].normalize3fast();
diff --git a/indra/newview/llviewershadermgr.cpp b/indra/newview/llviewershadermgr.cpp
index e24237522a..e4f04a414d 100755
--- a/indra/newview/llviewershadermgr.cpp
+++ b/indra/newview/llviewershadermgr.cpp
@@ -199,7 +199,7 @@ LLGLSLShader			gDeferredTreeShadowProgram;
 LLGLSLShader			gDeferredAvatarProgram;
 LLGLSLShader			gDeferredAvatarAlphaProgram;
 LLGLSLShader			gDeferredLightProgram;
-LLGLSLShader			gDeferredMultiLightProgram;
+LLGLSLShader			gDeferredMultiLightProgram[16];
 LLGLSLShader			gDeferredSpotLightProgram;
 LLGLSLShader			gDeferredMultiSpotLightProgram;
 LLGLSLShader			gDeferredSunProgram;
@@ -1128,7 +1128,10 @@ BOOL LLViewerShaderMgr::loadShadersDeferred()
 		gDeferredImpostorProgram.unload();
 		gDeferredTerrainProgram.unload();
 		gDeferredLightProgram.unload();
-		gDeferredMultiLightProgram.unload();
+		for (U32 i = 0; i < LL_DEFERRED_MULTI_LIGHT_COUNT; ++i)
+		{
+			gDeferredMultiLightProgram[i].unload();
+		}
 		gDeferredSpotLightProgram.unload();
 		gDeferredMultiSpotLightProgram.unload();
 		gDeferredSunProgram.unload();
@@ -1373,17 +1376,21 @@ BOOL LLViewerShaderMgr::loadShadersDeferred()
 		success = gDeferredLightProgram.createShader(NULL, NULL);
 	}
 
-	if (success)
+	for (U32 i = 0; i < LL_DEFERRED_MULTI_LIGHT_COUNT; i++)
 	{
-		gDeferredMultiLightProgram.mName = "Deferred MultiLight Shader";
-		gDeferredMultiLightProgram.mShaderFiles.clear();
-		gDeferredMultiLightProgram.mShaderFiles.push_back(make_pair("deferred/multiPointLightV.glsl", GL_VERTEX_SHADER_ARB));
-		gDeferredMultiLightProgram.mShaderFiles.push_back(make_pair("deferred/multiPointLightF.glsl", GL_FRAGMENT_SHADER_ARB));
-		gDeferredMultiLightProgram.mShaderLevel = mVertexShaderLevel[SHADER_DEFERRED];
+		if (success)
+		{
+			gDeferredMultiLightProgram[i].mName = llformat("Deferred MultiLight Shader %d", i);
+			gDeferredMultiLightProgram[i].mShaderFiles.clear();
+			gDeferredMultiLightProgram[i].mShaderFiles.push_back(make_pair("deferred/multiPointLightV.glsl", GL_VERTEX_SHADER_ARB));
+			gDeferredMultiLightProgram[i].mShaderFiles.push_back(make_pair("deferred/multiPointLightF.glsl", GL_FRAGMENT_SHADER_ARB));
+			gDeferredMultiLightProgram[i].mShaderLevel = mVertexShaderLevel[SHADER_DEFERRED];
+			gDeferredMultiLightProgram[i].addPermutation("LIGHT_COUNT", llformat("%d", i+1));
 
-		SINGLE_FP_PERMUTATION(gDeferredMultiLightProgram);
+			SINGLE_FP_PERMUTATION(gDeferredMultiLightProgram[i]);
 
-		success = gDeferredMultiLightProgram.createShader(NULL, NULL);
+			success = gDeferredMultiLightProgram[i].createShader(NULL, NULL);
+		}
 	}
 
 	if (success)
diff --git a/indra/newview/llviewershadermgr.h b/indra/newview/llviewershadermgr.h
index 438853cd6f..ed8167947d 100755
--- a/indra/newview/llviewershadermgr.h
+++ b/indra/newview/llviewershadermgr.h
@@ -30,6 +30,8 @@
 #include "llshadermgr.h"
 #include "llmaterial.h"
 
+#define LL_DEFERRED_MULTI_LIGHT_COUNT 16
+
 class LLViewerShaderMgr: public LLShaderMgr
 {
 public:
@@ -341,7 +343,7 @@ extern LLGLSLShader			gDeferredTerrainProgram;
 extern LLGLSLShader			gDeferredTreeProgram;
 extern LLGLSLShader			gDeferredTreeShadowProgram;
 extern LLGLSLShader			gDeferredLightProgram;
-extern LLGLSLShader			gDeferredMultiLightProgram;
+extern LLGLSLShader			gDeferredMultiLightProgram[LL_DEFERRED_MULTI_LIGHT_COUNT];
 extern LLGLSLShader			gDeferredSpotLightProgram;
 extern LLGLSLShader			gDeferredMultiSpotLightProgram;
 extern LLGLSLShader			gDeferredSunProgram;
diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp
index 4801c52209..2eeedc8214 100755
--- a/indra/newview/pipeline.cpp
+++ b/indra/newview/pipeline.cpp
@@ -8672,10 +8672,6 @@ void LLPipeline::renderDeferredLighting()
 			vert[2].set(3,1,0);
 
 			{
-				bindDeferredShader(gDeferredMultiLightProgram);
-			
-				mDeferredVB->setBuffer(LLVertexBuffer::MAP_VERTEX);
-
 				LLGLDepthTest depth(GL_FALSE);
 
 				//full screen blit
@@ -8687,7 +8683,7 @@ void LLPipeline::renderDeferredLighting()
 
 				U32 count = 0;
 
-				const U32 max_count = 8;
+				const U32 max_count = LL_DEFERRED_MULTI_LIGHT_COUNT;
 				LLVector4 light[max_count];
 				LLVector4 col[max_count];
 
@@ -8710,17 +8706,20 @@ void LLPipeline::renderDeferredLighting()
 					count++;
 					if (count == max_count || fullscreen_lights.empty())
 					{
-						gDeferredMultiLightProgram.uniform1i(LLShaderMgr::MULTI_LIGHT_COUNT, count);
-						gDeferredMultiLightProgram.uniform4fv(LLShaderMgr::MULTI_LIGHT, count, (GLfloat*) light);
-						gDeferredMultiLightProgram.uniform4fv(LLShaderMgr::MULTI_LIGHT_COL, count, (GLfloat*) col);
-						gDeferredMultiLightProgram.uniform1f(LLShaderMgr::MULTI_LIGHT_FAR_Z, far_z);
+						U32 idx = count-1;
+						bindDeferredShader(gDeferredMultiLightProgram[idx]);
+						gDeferredMultiLightProgram[idx].uniform1i(LLShaderMgr::MULTI_LIGHT_COUNT, count);
+						gDeferredMultiLightProgram[idx].uniform4fv(LLShaderMgr::MULTI_LIGHT, count, (GLfloat*) light);
+						gDeferredMultiLightProgram[idx].uniform4fv(LLShaderMgr::MULTI_LIGHT_COL, count, (GLfloat*) col);
+						gDeferredMultiLightProgram[idx].uniform1f(LLShaderMgr::MULTI_LIGHT_FAR_Z, far_z);
 						far_z = 0.f;
 						count = 0; 
+						mDeferredVB->setBuffer(LLVertexBuffer::MAP_VERTEX);
 						mDeferredVB->drawArrays(LLRender::TRIANGLES, 0, 3);
 					}
 				}
 				
-				unbindDeferredShader(gDeferredMultiLightProgram);
+				unbindDeferredShader(gDeferredMultiLightProgram[0]);
 
 				bindDeferredShader(gDeferredMultiSpotLightProgram);