Add Z-Compression shader to better use position buffer's FP16 (scp-fs2open#7245)

Author: BMagnu

code/def_files/data/effects/deferred-f.sdr

@@ -2,6 +2,7 @@
 #include "lighting.sdr" //! #include "lighting.sdr"
 #include "gamma.sdr" //! #include "gamma.sdr"
 #include "shadows.sdr" //! #include "shadows.sdr"
+#include "z-compress.sdr" //! #include "z-compress.sdr"
 
 out vec4 fragOut0;
 
@@ -219,7 +220,7 @@ void main()
 {
 	vec2 screenPos = gl_FragCoord.xy * vec2(invScreenWidth, invScreenHeight);
 	vec4 position_buffer = texture(PositionBuffer, screenPos);
-	vec3 position = position_buffer.xyz;
+	vec3 position = vec3(position_buffer.xy, uncompress_depth_value(position_buffer.z));
 
 	if(abs(dot(position, position)) < nearPlane * nearPlane)
 		discard;

code/def_files/data/effects/effect-f.sdr

@@ -1,5 +1,6 @@
 
 #include "gamma.sdr"
+#include "z-compress.sdr"
 
 in float fragRadius;
 in vec4 fragPosition;
@@ -38,7 +39,7 @@ void main()
 	float sceneDepthLinear;
 	float fragDepthLinear;
 	if ( linear_depth == 1 ) {
-		sceneDepthLinear = -sceneDepth.z;
+		sceneDepthLinear = -uncompress_depth_value(sceneDepth.z);
 		fragDepthLinear = -fragPosition.z;
 	} else {
 		sceneDepthLinear = ( 2.0 * farZ * nearZ ) / ( farZ + nearZ - sceneDepth.x * (farZ-nearZ) );

code/def_files/data/effects/main-f.sdr

@@ -2,6 +2,7 @@
 #include "shadows.sdr" //! #include "shadows.sdr"
 #include "lighting.sdr" //! #include "lighting.sdr"
 #include "gamma.sdr" //! #include "gamma.sdr"
+#include "z-compress.sdr" //! #include "z-compress.sdr"
 
 #conditional_include +"LARGE_SHADER" "main_large.sdr"
 #conditional_include -"LARGE_SHADER" "main_small.sdr"
@@ -428,7 +429,7 @@ void main()
 	fragOut0 = baseColor;
 
 	#prereplace IF_FLAG MODEL_SDR_FLAG_DEFERRED
-		fragOut1 = vec4(vertIn.position.xyz, aoFactors.x);
+		fragOut1 = vec4(vertIn.position.xy, compress_depth_value(vertIn.position.z), aoFactors.x);
 		fragOut2 = vec4(normal, glossData);
 		fragOut3 = vec4(specColor.rgb, fresnelFactor);
 		fragOut4 = emissiveColor;

code/def_files/data/effects/msaa-f.sdr

@@ -1,3 +1,5 @@
+#include "z-compress.sdr"
+
 in vec4 fragTexCoord;
 out vec4 fragOut0;
 out vec4 fragOut1;
@@ -32,7 +34,7 @@ float getMedianDist(ivec2 texel) {
 	float dists[4];
 
 	for(int i = 0; i < 4; i++) {
-		dists[i] = texelFetch(texPos, texel, i).z;
+		dists[i] = uncompress_depth_value(texelFetch(texPos, texel, i).z);
 	}
 
 	SORT(0, 2)
@@ -51,7 +53,7 @@ float getMedianDist(ivec2 texel) {
 	float dists[8];
 
 	for(int i = 0; i < 8; i++) {
-		dists[i] = texelFetch(texPos, texel, i).z;
+		dists[i] = uncompress_depth_value(texelFetch(texPos, texel, i).z);
 	}
 
 	SORT(0, 4)
@@ -82,7 +84,7 @@ float getMedianDist(ivec2 texel) {
 	float dists[16];
 
 	for(int i = 0; i < 16; i++) {
-		dists[i] = texelFetch(texPos, texel, i).z;
+		dists[i] = uncompress_depth_value(texelFetch(texPos, texel, i).z);
 	}
 
 	SORT(0, 1)
@@ -145,7 +147,7 @@ float getMedianDist(ivec2 texel) {
 float getMedianDist(ivec2 texel) {
 	float minDist = -1000000;
 	for(int i = 0; i < samples; i++) {
-		minDist = max(minDist, texelFetch(texPos, texel, i).z);
+		minDist = max(minDist, uncompress_depth_value(texelFetch(texPos, texel, i).z));
 	}
 	return minDist;
 }
@@ -173,6 +175,7 @@ void main()
 
 	for(int i = 0; i < samples; i++) {
 		vec4 localPos = texelFetch(texPos, texel, i);
+		localPos.z = uncompress_depth_value(localPos.z);
 		//Calculate local weight from distance Voxel, but if the distance is 0 (i.e. no model at all), set weight to 1 to allow stuff like background emissive
 		//However, if the median distance is 0, only deal with current texel if it's local distance is 0 as well
 		float localWeight = max(step(-0.001, dist) * step(-0.001, localPos.z),
@@ -189,8 +192,11 @@ void main()
 		weight += localWeight;
 	}
 
+    pos /= weight;
+    pos.z = compress_depth_value(pos.z);
+
 	fragOut0 = color / weight;
-	fragOut1 = pos / weight;
+	fragOut1 = pos;
 	fragOut2 = vec4(normalize(normal.xyz), normal.a / weight);
 	fragOut3 = specular / weight;
 	fragOut4 = emissive / weight;

code/def_files/data/effects/z-compress.sdr

@@ -0,0 +1,89 @@
+float compress_depth_value(float linear) {
+	float linear_mult;
+	float exponent_mult;	
+	float offset;
+
+	linear = max(-linear, 0.0);
+
+	if(linear < 10.0) {
+		linear_mult = 0.00000999908;
+		exponent_mult = 1.32878452580;
+		offset = 0.0;
+	}
+	else if (linear < 100.0) {
+		linear_mult = 0.05550473078;
+		exponent_mult = 0.08493173544;
+		offset = 0.0;
+	}
+	else if (linear < 1570.0) {
+		linear_mult = 11.53589613947;
+		exponent_mult = 0.00793869919;
+		offset = 0.0;
+	}
+	else if (linear < 3000.0) {
+		linear_mult = -0.10000000000;
+		exponent_mult = 0.00928910863;
+		offset = 3000.0;
+	}
+	else if (linear < 10000.0) {
+		linear_mult = -0.10000000000;
+		exponent_mult = 0.00094912231;
+		offset = 3000.0;
+	}
+	else if (linear < 100000.0) {
+		linear_mult = -6.98947320727;
+		exponent_mult = 0.00007382062;
+		offset = 3000.0;
+	}
+	else {
+		linear_mult = -1000.0;
+		exponent_mult = log2(1000.0 / 65000.0) / (100000.0 - GLOBAL_FAR_Z);
+		offset = 100000.0;
+	}
+
+	return linear_mult * pow(2.0, (linear - offset) * exponent_mult);
+}
+
+float uncompress_depth_value(float compressed) {
+	float linear_mult;
+	float exponent_mult;	
+	float offset;
+
+	if(compressed < -1000.0) {
+		linear_mult = -1000.0;
+		exponent_mult = log2(1000.0 / 65000.0) / (100000.0 - GLOBAL_FAR_Z);
+		offset = 100000.0;
+	}
+	else if (compressed < -10.0) {
+		linear_mult = -6.98947320727;
+		exponent_mult = 0.00007382062;
+		offset = 3000.0;
+	}
+	else if (compressed < -0.1) {
+		linear_mult = -0.10000000000;
+		exponent_mult = 0.00094912231;
+		offset = 3000.0;
+	}
+	else if (compressed < 0.0) {
+		linear_mult = -0.10000000000;
+		exponent_mult = 0.00928910863;
+		offset = 3000.0;
+	}
+	else if (compressed < 0.1) {
+		linear_mult = 0.00000999908;
+		exponent_mult = 1.32878452580;
+		offset = 0.0;
+	}
+	else if (compressed < 20.0) {
+		linear_mult = 0.05550473078;
+		exponent_mult = 0.08493173544;
+		offset = 0.0;
+	}
+	else {
+		linear_mult = 11.53589613947;
+		exponent_mult = 0.00793869919;
+		offset = 0.0;
+	}
+
+	return -(log2(compressed / linear_mult) / exponent_mult + offset);
+}

code/graphics/2d.cpp

@@ -283,7 +283,8 @@ int gr_stencil_mode = 0;
 
 // Default clipping distances
 const float Default_min_draw_distance = 1.0f;
-const float Default_max_draw_distance = 1e10;
+// Reduced from 1e10 to 1e6, as beyond that FSO's physics precision is horrendous anyways, and it allows reasonable lighting and particle clipping all the way out until that point, unlike 1e7 or above where the depth precision just is not enough
+const float Default_max_draw_distance = 1e6f;
 float Min_draw_distance_cockpit = 0.02f;
 float Min_draw_distance = Default_min_draw_distance;
 float Max_draw_distance = Default_max_draw_distance;

code/graphics/opengl/gropenglshader.cpp

@@ -503,6 +503,9 @@ static SCP_string handle_predefines(const char* filename, const SCP_string& orig
 	SCP_stringstream output;
 	SCP_unordered_map<SCP_string, SCP_string> defines;
 
+	//In any shader, define GLOBAL_FAR_Z
+	output << "#define GLOBAL_FAR_Z " << std::fixed << std::setprecision(2) << Max_draw_distance << std::defaultfloat << '\n';
+
 	const char* PREDEFINE_STRING = "#predefine";
 	const char* PREREPLACE_STRING = "#prereplace";
 

code/source_groups.cmake

@@ -278,6 +278,7 @@ add_file_folder("Default files\\\\data\\\\effects"
 	def_files/data/effects/video-f.sdr
 	def_files/data/effects/video-v.sdr
 	def_files/data/effects/volumetric-f.sdr
+	def_files/data/effects/z-compress.sdr
 )
 
 add_file_folder("Default files\\\\data\\\\maps"