Fix SSAO visual artifacts when MSAA is enabled. Increase performance of SSAO, allow passing effect kernel size (#105)

* shaders/pbr : output the NDC depth to intermediate (color) resolved texture and use it

* Optimize SSAO (reduce kernel size, precompute inverse projection matrix)

* posteffects : Allow setting SSAO kernel size

* multibody/RobotScene : pass ssao kernel size option

* multibody/Visualizer : pass ssao kernel size option

* Update CHANGELOG

Author: ManifoldFR

CHANGELOG.md

@@ -10,10 +10,16 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Added
 
 - Expose `Visualizer.toggleGui(value=None)` (https://github.com/Simple-Robotics/candlewick/pull/102)
+- Add option to set SSAO effect kernel size in Visualizer and RobotScene (C++/Python) (https://github.com/Simple-Robotics/candlewick/pull/105)
 
 ### Changed
 
 - Move cpp examples to their own directory (https://github.com/Simple-Robotics/candlewick/pull/103)
+- Improved performance of SSAO effect (https://github.com/Simple-Robotics/candlewick/pull/105)
+
+### Fixed
+
+- Fixed visual effects with SSAO + MSAA. Have SSAO sample from resolved multisampled depth texture (https://github.com/Simple-Robotics/candlewick/pull/105)
 
 ## [0.10.1] - 2026-01-19
 

bindings/python/src/expose-visualizer.cpp

@@ -36,6 +36,10 @@ void exposeVisualizer() {
   bp::class_<Visualizer::Config>("VisualizerConfig", bp::no_init)
       .def_readwrite("width", &Visualizer::Config::width)
       .def_readwrite("height", &Visualizer::Config::height)
+      .def_readwrite("sampleCount", &Visualizer::Config::sampleCount,
+                     "MSAA sample count.")
+      .def_readwrite("ssaoKernelSize", &Visualizer::Config::ssaoKernelSize,
+                     "Kernel size for the SSAO effect.")
       .def(bp::init<>("self"_a))
       .def(bp::init<Uint32, Uint32>(("self"_a, "width", "height")));
 

examples/cpp/Ur5WithSystems.cpp

@@ -231,13 +231,16 @@ static void initialize() {
 
 int main(int argc, char **argv) {
   CLI::App app{"Ur5 example"};
-  bool performRecording{false};
+  bool performRecording = false;
   RobotScene::Config robot_scene_config;
   robot_scene_config.triangle_has_prepass = true;
 
   argv = app.ensure_utf8(argv);
   app.add_flag("-r,--record", performRecording, "Record output");
   app.add_option("--dims", wDims, "Window dimensions.")->capture_default_str();
+  app.add_option("--ssao-kernel-size", robot_scene_config.ssao_kernel_size,
+                 "Number of SSAO kernel samples (max 64).")
+      ->capture_default_str();
   CLI11_PARSE(app, argc, argv);
 
   if (!SDL_Init(SDL_INIT_VIDEO))

examples/cpp/Visualizer.cpp

@@ -54,6 +54,7 @@ int main(int argc, char *argv[]) {
       {"2", SDL_GPU_SAMPLECOUNT_2},
       {"4", SDL_GPU_SAMPLECOUNT_4},
       {"8", SDL_GPU_SAMPLECOUNT_8}};
+  Uint32 ssaoKernelSize = 16u;
   const auto transform_validator =
       CLI::IsMember(sample_count_map) & CLI::Transformer(sample_count_map);
 
@@ -67,6 +68,8 @@ int main(int argc, char *argv[]) {
       })
       ->transform(transform_validator)
       ->capture_default_str();
+  app.add_option("--ssao-kernel-size", ssaoKernelSize,
+                 "Size of the SSAO effect kernel.");
 
   CLI11_PARSE(app, argc, argv);
 

shaders/compiled/PbrBasic.frag.json

@@ -1 +1 @@
-{ "samplers": 2, "storage_textures": 0, "storage_buffers": 0, "uniform_buffers": 4 }
+{ "samplers": 2, "storage_textures": 0, "storage_buffers": 0, "uniform_buffers": 4, "inputs": [{ "name": "fragLightPos", "type": "float3", "location": 2 }, { "name": "fragViewNormal", "type": "float3", "location": 1 }, { "name": "fragViewPos", "type": "float3", "location": 0 }], "outputs": [{ "name": "fragColor", "type": "float4", "location": 0 }, { "name": "outNormal", "type": "float2", "location": 1 }, { "name": "outDepth", "type": "float", "location": 2 }] }

shaders/compiled/PbrBasic.frag.msl

@@ -87,6 +87,7 @@ struct main0_out
 {
     float4 fragColor [[color(0)]];
     float2 outNormal [[color(1)]];
+    float outDepth [[color(2)]];
 };
 
 struct main0_in
@@ -330,5 +331,6 @@ fragment main0_out main0(main0_in in [[stage_in]], constant Material& _572 [[buf
     color = powr(color, float3(0.4545454680919647216796875));
     out.fragColor = float4(color, _572.material.baseColor.w);
     out.outNormal = in.fragViewNormal.xy;
+    out.outDepth = gl_FragCoord.z;
     return out;
 }

shaders/compiled/PbrBasic.frag.spv

@@ -1 +1 @@
-{ "samplers": 3, "storage_textures": 0, "storage_buffers": 0, "uniform_buffers": 2 }
+{ "samplers": 3, "storage_textures": 0, "storage_buffers": 0, "uniform_buffers": 2, "inputs": [{ "name": "inUV", "type": "float2", "location": 0 }], "outputs": [{ "name": "aoValue", "type": "float", "location": 0 }] }

shaders/compiled/SSAO.frag.json

@@ -5,59 +5,11 @@
 
 using namespace metal;
 
-// Returns the determinant of a 2x2 matrix.
-static inline __attribute__((always_inline))
-float spvDet2x2(float a1, float a2, float b1, float b2)
-{
-    return a1 * b2 - b1 * a2;
-}
-
-// Returns the determinant of a 3x3 matrix.
-static inline __attribute__((always_inline))
-float spvDet3x3(float a1, float a2, float a3, float b1, float b2, float b3, float c1, float c2, float c3)
-{
-    return a1 * spvDet2x2(b2, b3, c2, c3) - b1 * spvDet2x2(a2, a3, c2, c3) + c1 * spvDet2x2(a2, a3, b2, b3);
-}
-
-// Returns the inverse of a matrix, by using the algorithm of calculating the classical
-// adjoint and dividing by the determinant. The contents of the matrix are changed.
-static inline __attribute__((always_inline))
-float4x4 spvInverse4x4(float4x4 m)
-{
-    float4x4 adj;   // The adjoint matrix (inverse after dividing by determinant)
-
-    // Create the transpose of the cofactors, as the classical adjoint of the matrix.
-    adj[0][0] =  spvDet3x3(m[1][1], m[1][2], m[1][3], m[2][1], m[2][2], m[2][3], m[3][1], m[3][2], m[3][3]);
-    adj[0][1] = -spvDet3x3(m[0][1], m[0][2], m[0][3], m[2][1], m[2][2], m[2][3], m[3][1], m[3][2], m[3][3]);
-    adj[0][2] =  spvDet3x3(m[0][1], m[0][2], m[0][3], m[1][1], m[1][2], m[1][3], m[3][1], m[3][2], m[3][3]);
-    adj[0][3] = -spvDet3x3(m[0][1], m[0][2], m[0][3], m[1][1], m[1][2], m[1][3], m[2][1], m[2][2], m[2][3]);
-
-    adj[1][0] = -spvDet3x3(m[1][0], m[1][2], m[1][3], m[2][0], m[2][2], m[2][3], m[3][0], m[3][2], m[3][3]);
-    adj[1][1] =  spvDet3x3(m[0][0], m[0][2], m[0][3], m[2][0], m[2][2], m[2][3], m[3][0], m[3][2], m[3][3]);
-    adj[1][2] = -spvDet3x3(m[0][0], m[0][2], m[0][3], m[1][0], m[1][2], m[1][3], m[3][0], m[3][2], m[3][3]);
-    adj[1][3] =  spvDet3x3(m[0][0], m[0][2], m[0][3], m[1][0], m[1][2], m[1][3], m[2][0], m[2][2], m[2][3]);
-
-    adj[2][0] =  spvDet3x3(m[1][0], m[1][1], m[1][3], m[2][0], m[2][1], m[2][3], m[3][0], m[3][1], m[3][3]);
-    adj[2][1] = -spvDet3x3(m[0][0], m[0][1], m[0][3], m[2][0], m[2][1], m[2][3], m[3][0], m[3][1], m[3][3]);
-    adj[2][2] =  spvDet3x3(m[0][0], m[0][1], m[0][3], m[1][0], m[1][1], m[1][3], m[3][0], m[3][1], m[3][3]);
-    adj[2][3] = -spvDet3x3(m[0][0], m[0][1], m[0][3], m[1][0], m[1][1], m[1][3], m[2][0], m[2][1], m[2][3]);
-
-    adj[3][0] = -spvDet3x3(m[1][0], m[1][1], m[1][2], m[2][0], m[2][1], m[2][2], m[3][0], m[3][1], m[3][2]);
-    adj[3][1] =  spvDet3x3(m[0][0], m[0][1], m[0][2], m[2][0], m[2][1], m[2][2], m[3][0], m[3][1], m[3][2]);
-    adj[3][2] = -spvDet3x3(m[0][0], m[0][1], m[0][2], m[1][0], m[1][1], m[1][2], m[3][0], m[3][1], m[3][2]);
-    adj[3][3] =  spvDet3x3(m[0][0], m[0][1], m[0][2], m[1][0], m[1][1], m[1][2], m[2][0], m[2][1], m[2][2]);
-
-    // Calculate the determinant as a combination of the cofactors of the first row.
-    float det = (adj[0][0] * m[0][0]) + (adj[0][1] * m[1][0]) + (adj[0][2] * m[2][0]) + (adj[0][3] * m[3][0]);
-
-    // Divide the classical adjoint matrix by the determinant.
-    // If determinant is zero, matrix is not invertable, so leave it unchanged.
-    return (det != 0.0f) ? (adj * (1.0f / det)) : m;
-}
-
 struct Camera
 {
     float4x4 projection;
+    float4x4 projectionInverse;
+    uint kernelSize;
 };
 
 struct SSAOParams
@@ -79,7 +31,7 @@ static inline __attribute__((always_inline))
 float3 getViewPos(thread const float& depth, thread const float2& uv, constant Camera& camera)
 {
     float4 clipPos = float4((uv * 2.0) - float2(1.0), depth, 1.0);
-    float4 viewPos = spvInverse4x4(camera.projection) * clipPos;
+    float4 viewPos = camera.projectionInverse * clipPos;
     return viewPos.xyz / float3(viewPos.w);
 }
 
@@ -111,28 +63,28 @@ float calculatePixelAO(thread const float2& uv, constant Camera& camera, texture
     float3 bitangent = cross(tangent, viewNormal);
     float3x3 TBN = float3x3(float3(tangent), float3(bitangent), float3(viewNormal));
     float occlusion = 0.0;
-    for (int i = 0; i < 64; i++)
+    for (int i = 0; i < int(camera.kernelSize); i++)
     {
         float3 samplePos = TBN * kernel0.samples[i].xyz;
         samplePos = viewPos + (samplePos * 1.0);
         float4 offset = camera.projection * float4(samplePos, 1.0);
-        float _197 = offset.w;
-        float4 _198 = offset;
-        float2 _201 = _198.xy / float2(_197);
-        offset.x = _201.x;
-        offset.y = _201.y;
-        float4 _206 = offset;
-        float2 _211 = (_206.xy * 0.5) + float2(0.5);
-        offset.x = _211.x;
-        offset.y = _211.y;
+        float _201 = offset.w;
+        float4 _202 = offset;
+        float2 _205 = _202.xy / float2(_201);
+        offset.x = _205.x;
+        offset.y = _205.y;
+        float4 _210 = offset;
+        float2 _215 = (_210.xy * 0.5) + float2(0.5);
+        offset.x = _215.x;
+        offset.y = _215.y;
         float sampleDepth = depthTex.sample(depthTexSmplr, offset.xy).x;
         float param_3 = sampleDepth;
         float2 param_4 = offset.xy;
         float3 sampleViewPos = getViewPos(param_3, param_4, camera);
         float rangeCheck = smoothstep(0.0, 1.0, 1.0 / abs((viewPos.z - sampleViewPos.z) - 0.00999999977648258209228515625));
         occlusion += (float(sampleViewPos.z >= (samplePos.z + 0.00999999977648258209228515625)) * rangeCheck);
     }
-    occlusion = 1.0 - ((occlusion / 64.0) * 1.5);
+    occlusion = 1.0 - ((occlusion / float(camera.kernelSize)) * 1.5);
     return occlusion;
 }