General improvements (tippesi#34)

* Vertex color support * Changed caustics and ocean a bit * Fixed issues * Fixed heap corruption issues * Fixed some issues with clouds * Working on pixel depth offset for impostors * Better impostors * Slight performance improvements * Small fix * Working on specular probe prefiltering * Improved the probe filtering further * Some further fixes * Small fixes * Got vegetation rendering to work again * Update build.yml * Update build.yml * Small fixes * Improved some stuff * Working on cloud shadows * Continue to work on cloud shadows * More widely available cloud shadows * Also commiting shader changes * Fixed impostor pixel-depth offset * Improved terrain rendering performance * More advanced shader hot reload * Fixed a few issues * Starting to fix Vulkan synchonization issues * Fixed remaining open world Vulkan synchronization issues * Fixed remaining synchonization issues * Fixed some shaders * Fixed up further * New descriptor set caching * Sharpen changes * Normal encoding to reduce bandwith usage * Added film grain * Fixed issues * Fixed normal encoding --------- Co-authored-by: Simon Tippe <tippex97@gmail.com>
hzqst · Aug 29, 2023 · b45bb09 · b45bb09
1 parent efefdb8
commit b45bb09
Show file tree

Hide file tree

Showing 169 changed files with 3,155 additions and 1,609 deletions.
diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
@@ -45,7 +45,7 @@ jobs:
         uses: humbletim/setup-vulkan-sdk@v1.2.0
         with:
           vulkan-query-version: 1.3.204.0
-          vulkan-components: Vulkan-Headers, Vulkan-Loader
+          vulkan-components: Vulkan-Headers
           vulkan-use-cache: true
 
       - name: Run scripts
@@ -111,7 +111,7 @@ jobs:
         uses: humbletim/setup-vulkan-sdk@v1.2.0
         with:
           vulkan-query-version: 1.3.204.0
-          vulkan-components: Vulkan-Headers, Vulkan-Loader
+          vulkan-components: Vulkan-Headers
           vulkan-use-cache: true
 
       - name: Run scripts
@@ -180,7 +180,7 @@ jobs:
         uses: humbletim/setup-vulkan-sdk@v1.2.0
         with:
           vulkan-query-version: 1.3.204.0
-          vulkan-components: Vulkan-Headers, Vulkan-Loader
+          vulkan-components: Vulkan-Headers
           vulkan-use-cache: true
 
       - name: Run scripts

diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
@@ -46,7 +46,7 @@ jobs:
         uses: humbletim/setup-vulkan-sdk@v1.2.0
         with:
           vulkan-query-version: 1.3.204.0
-          vulkan-components: Vulkan-Headers, Vulkan-Loader
+          vulkan-components: Vulkan-Headers
           vulkan-use-cache: true
 
       - name: Run scripts
@@ -111,7 +111,7 @@ jobs:
         uses: humbletim/setup-vulkan-sdk@v1.2.0
         with:
           vulkan-query-version: 1.3.204.0
-          vulkan-components: Vulkan-Headers, Vulkan-Loader
+          vulkan-components: Vulkan-Headers
           vulkan-use-cache: true
 
       - name: Run scripts

diff --git a/data/chromesphere.gltf b/data/chromesphere.gltf
diff --git a/data/shader/ao/rtao.csh b/data/shader/ao/rtao.csh
@@ -7,6 +7,7 @@
 #include <../common/random.hsh>
 #include <../common/flatten.hsh>
 #include <../common/convert.hsh>
+#include <../common/normalencode.hsh>
 #include <../common/bluenoise.hsh>
 #include <../brdf/brdfSample.hsh>
 
@@ -63,7 +64,7 @@ void main() {
 
         vec2 recontructTexCoord = (2.0 * vec2(pixel) + offset + vec2(0.5)) / (2.0 * vec2(resolution));
         vec3 worldPos = vec3(globalData.ivMatrix * vec4(ConvertDepthToViewSpace(depth, recontructTexCoord), 1.0));
-        vec3 worldNorm = normalize(vec3(globalData.ivMatrix * vec4(2.0 * textureLod(normalTexture, texCoord, 0).rgb - 1.0, 0.0)));
+        vec3 worldNorm = normalize(vec3(globalData.ivMatrix * vec4(DecodeNormal(textureLod(normalTexture, texCoord, 0).rg), 0.0)));
 
         float ao = 0.0;
 

diff --git a/data/shader/ao/ssao.csh b/data/shader/ao/ssao.csh
@@ -6,6 +6,7 @@ layout (local_size_x = 8, local_size_y = 8) in;
 #include <../common/utility.hsh>
 #include <../common/flatten.hsh>
 #include <../common/random.hsh>
+#include <../common/normalencode.hsh>
 
 layout (set = 3, binding = 0, r16f) writeonly uniform image2D textureOut;
 layout(set = 3, binding = 1) uniform sampler2D normalTexture;
@@ -38,10 +39,11 @@ void main() {
     float depth = texelFetch(shadowMap, pixel, 0).r;
 
     // Early exit, also prevents halo
-
     vec3 fragPos = ConvertDepthToViewSpace(depth, texCoord);
-    vec3 norm = 2.0 * textureLod(normalTexture, texCoord, 0).rgb - 1.0;
-    vec3 randomVec = vec3(2.0 * texelFetch(randomTexture, pixel % ivec2(4), 0).xy - 1.0, 0.0);
+    vec3 norm = DecodeNormal(textureLod(normalTexture, texCoord, 0).rg);
+
+    ivec2 randomPixel = pixel + Unflatten2D(uniforms.frameCount % 16, ivec2(4));
+    vec3 randomVec = vec3(2.0 * texelFetch(randomTexture, randomPixel % ivec2(4), 0).xy - 1.0, 0.0);
 
     //Create TBN matrix
     vec3 tang = normalize(randomVec - norm * dot(randomVec, norm));

diff --git a/data/shader/ao/temporal.csh b/data/shader/ao/temporal.csh
@@ -4,6 +4,7 @@
 #include <../common/stencil.hsh>
 #include <../common/flatten.hsh>
 #include <../common/random.hsh>
+#include <../common/normalencode.hsh>
 
 layout (local_size_x = 16, local_size_y = 16) in;
 
@@ -166,7 +167,7 @@ bool SampleHistory(ivec2 pixel, vec2 historyPixel, out float history, out float
     float weights[4] = { (1 - x) * (1 - y), x * (1 - y), (1 - x) * y, x * y };
 
     uint materialIdx = texelFetch(materialIdxTexture, pixel, 0).r;
-    vec3 normal = 2.0 * texelFetch(normalTexture, pixel, 0).rgb - 1.0;
+    vec3 normal = DecodeNormal(texelFetch(normalTexture, pixel, 0).rg);
     float depth = texelFetch(depthTexture, pixel, 0).r;
 
     float linearDepth = ConvertDepthToViewSpaceDepth(depth);
@@ -180,7 +181,7 @@ bool SampleHistory(ivec2 pixel, vec2 historyPixel, out float history, out float
         uint historyMaterialIdx = texelFetch(historyMaterialIdxTexture, offsetPixel, 0).r;
         confidence *= historyMaterialIdx != materialIdx ? 0.0 : 1.0;
 
-        vec3 historyNormal = 2.0 * texelFetch(historyNormalTexture, offsetPixel, 0).rgb - 1.0;
+        vec3 historyNormal = DecodeNormal(texelFetch(historyNormalTexture, offsetPixel, 0).rg);
         confidence *= pow(abs(dot(historyNormal, normal)), 2.0);
 
         float historyDepth = texelFetch(historyDepthTexture, offsetPixel, 0).r;

diff --git a/data/shader/atmosphere.csh b/data/shader/atmosphere.csh
@@ -99,7 +99,7 @@ void main() {
 
     if (LdotV > sunAngle) {
         float sunDisk = clamp(0.5 * (LdotV - sunAngle) / (1.0 - sunAngle), 0.0, 1.0);
-        color += uniforms.sunIntensity * sunDisk;
+        color += uniforms.sunIntensity * sunDisk * 5.0;
     }
 
 #ifndef ENVIRONMENT_PROBE

diff --git a/data/shader/bilateralBlur.csh b/data/shader/bilateralBlur.csh
@@ -1,6 +1,7 @@
 // NOTE: The kernel is limited to a total size of 65 (2 * 32 + 1).
 #include <common/utility.hsh>
 #include <common/convert.hsh>
+#include <common/normalencode.hsh>
 
 #ifdef HORIZONTAL
 layout (local_size_x = 256, local_size_y = 1) in;
@@ -30,7 +31,7 @@ layout(set = 3, binding = 4, std140) uniform  WeightBuffer {
 } weights;
 
 const float normalPhi = 32.0;
-const float depthPhi = 1.0;
+const float depthPhi = 0.5;
 
 #ifdef BLUR_RGB
 shared vec3 inputs[320];
@@ -77,7 +78,7 @@ void LoadGroupSharedData() {
         depths[i] = ConvertDepthToViewSpaceDepth(localDepth);
 #endif
 #ifdef NORMAL_WEIGHT
-        vec3 localNormal = 2.0 * texelFetch(normalTexture, localOffset, 0).rgb - 1.0;
+        vec3 localNormal = DecodeNormal(texelFetch(normalTexture, localOffset, 0).rg);
         normals[i] = localNormal;
 #endif
     }

diff --git a/data/shader/brdf/filterProbe.csh b/data/shader/brdf/filterProbe.csh
@@ -1,12 +1,23 @@
 #include <../globals.hsh>
 #include <../common/random.hsh>
+#include <brdf.hsh>
+#include <preintegrate.hsh>
 #include <filtering.hsh>
 
 layout (local_size_x = 8, local_size_y = 8) in;
 
 layout(set = 3, binding = 0, rgba16f) uniform imageCube filteredCubeMap;
 layout(set = 3, binding = 1) uniform samplerCube cubeMap;
 
+layout(push_constant) uniform constants {
+    int cubeMapMipLevels;
+    float roughness;
+    uint mipLevel;
+} pushConstants;
+
+vec3 FilterDiffuse(vec3 worldDirection, ivec2 cubeMapSize);
+vec3 FilterSpecular(vec3 worldDirection, ivec2 cubeMapSize);
+
 vec3 CubeCoordToWorld(ivec3 cubeCoord, vec2 cubemapSize) {
     vec2 texCoord = (vec2(cubeCoord.xy) + 0.5) / cubemapSize;
     texCoord = texCoord  * 2.0 - 1.0;
@@ -31,19 +42,29 @@ void main() {
     vec3 worldDirection = CubeCoordToWorld(ivec3(gl_GlobalInvocationID),
         vec2(imageSize(filteredCubeMap).xy));
 
-    uint sampleCount = 256u;
-    vec4 acc = vec4(0.0);
-
     ivec2 size = textureSize(cubeMap, 0);
     vec3 N = normalize(worldDirection);
-    for (uint i = 0u; i < sampleCount; i++) {
-        float raySeed = float(gl_LocalInvocationIndex);
-        float curSeed = float(i);
-        float u0 = random(raySeed, curSeed);
-        curSeed += 0.5;
-        float u1 = random(raySeed, curSeed);
 
-        vec2 Xi = vec2(u0, u1);
+#ifdef FILTER_DIFFUSE
+    vec3 color = FilterDiffuse(worldDirection, size);
+#else
+    vec3 color = FilterSpecular(worldDirection, size);
+#endif
+
+    imageStore(filteredCubeMap, ivec3(gl_GlobalInvocationID), vec4(color, 0.0));
+
+}
+
+vec3 FilterDiffuse(vec3 worldDirection, ivec2 cubeMapSize) {
+
+    const uint sampleCount = 256u;
+
+    vec4 color = vec4(0.0);
+
+    vec3 N = normalize(worldDirection);
+
+    for (uint i = 0u; i < sampleCount; i++) {
+        vec2 Xi = Hammersley(i, sampleCount);
 
         vec3 L;
         float NdotL;
@@ -52,23 +73,61 @@ void main() {
 
         NdotL = saturate(NdotL);
         if (NdotL > 0.0) {
-            /*
-            // Only works for specular
-            // https://developer.nvidia.com/gpugems/gpugems3/part-iii-rendering/chapter-20-gpu-based-importance-sampling
-            float pdf = max(0.0, dot(N, L) / PI);
+            color.xyz += min(textureLod(cubeMap, L, 5.0).rgb, 5.0);
+            color.w += 1.0;
+        }
+
+    }
+
+    return color.rgb / color.a;
+
+}
+
+vec3 FilterSpecular(vec3 worldDirection, ivec2 cubeMapSize) {
+
+    const uint maxSampleCount = 256u;
+
+    uint sampleCount = uint(mix(16.0, float(maxSampleCount), pushConstants.roughness));
+    sampleCount = pushConstants.mipLevel == 0u ? 1u : sampleCount;
+
+    vec4 color = vec4(0.0);
+
+    vec3 N = normalize(worldDirection);
+    vec3 V = N;
+
+    for (uint i = 0u; i < sampleCount; i++) {
+        vec2 Xi = Hammersley(i, sampleCount);
+
+        vec3 L;
+        float NdotL;
+        float pdf;
+        float alpha = sqr(pushConstants.roughness);
+        ImportanceSampleGGX(Xi, N, V, alpha, L, pdf);
+
+        NdotL = dot(N, L);
+        vec3 H = normalize(L + V);
+        NdotL = saturate(NdotL);
+
+        if (NdotL > 0.0) {
+            float NdotH = saturate(dot(N, H));
+            float LdotH = saturate(dot(L, H));
+
+            pdf = DistributionGGX(NdotH, alpha) / 4.0;
 
-            float solidAngleTexel = 4.0 * PI / (6.0 * float(size.x * size.y));
+            float solidAngleTexel = 4.0 * PI / (6.0 * float(cubeMapSize.x * cubeMapSize.y));
             float solidAngleSample = 1.0 / (float(sampleCount) * pdf);
-            float lod = clamp(0.5 * log2(float(solidAngleSample / solidAngleTexel)), 0.0, 7.0);
-            */
+            float lod = clamp(0.5 * log2(float(solidAngleSample / solidAngleTexel)) + 1.0,
+                0.0, float(pushConstants.cubeMapMipLevels - 1.0));
+
+            lod = pushConstants.mipLevel == 0u ? 0.0 : lod;
+
+            color.xyz += clamp(textureLod(cubeMap, L, lod).rgb, vec3(0.0), vec3(5.0)) * NdotL;
+            color.w += NdotL;
 
-            acc.xyz += min(textureLod(cubeMap, L, 5.0).rgb, 1.0);
         }
 
     }
 
-    vec4 color = acc / float(sampleCount);
-
-    imageStore(filteredCubeMap, ivec3(gl_GlobalInvocationID), color);
+    return color.rgb / color.a;
 
 }
diff --git a/data/shader/brdf/importanceSample.hsh b/data/shader/brdf/importanceSample.hsh
@@ -58,7 +58,7 @@ vec3 ImportanceSampleGGXVNDF(vec2 Xi, vec3 V, float roughness) {
 }
 
 void ImportanceSampleGGXVNDF(vec2 Xi, vec3 N, vec3 V, float roughness,
-out vec3 L, out float pdf) {
+    out vec3 L, out float pdf) {
 
     vec3 up        = abs(N.z) < 0.999 ? vec3(0.0, 0.0, 1.0) : vec3(1.0, 0.0, 0.0);
     vec3 tangent   = normalize(cross(up, N));