FSR2: fix motion vector jitter, add bicubic anti-ringing, depth-dilated MVs

- Motion shader: unjitter NDC before reprojection (ndc+jitter, not ndc-jitter),
  compute motion against unjittered UV so static scenes produce zero motion
- Pass jitter offset to motion shader (push constant 80→96 bytes)
- Accumulate shader: restore Catmull-Rom bicubic with anti-ringing clamp to
  prevent negative-lobe halos at edges while maintaining sharpness
- Add depth-dilated motion vectors (3x3 nearest-to-camera) to prevent
  background MVs bleeding over foreground edges
- Widen neighborhood clamp gamma to 3.0, uniform 5% blend with
  disocclusion/velocity reactive boosting

assets/shaders/fsr2_accumulate.comp.glsl

@@ -29,20 +29,84 @@ vec3 yCoCgToRgb(vec3 ycocg) {
     return vec3(y + co - cg, y + cg, y - co - cg);
 }
 
+// Catmull-Rom bicubic (9 bilinear taps) with anti-ringing clamp.
+// Sharper than bilinear; anti-ringing prevents edge halos that shift with jitter.
+vec3 sampleBicubic(sampler2D tex, vec2 uv, vec2 texSize) {
+    vec2 invTexSize = 1.0 / texSize;
+    vec2 iTc = uv * texSize;
+    vec2 tc = floor(iTc - 0.5) + 0.5;
+    vec2 f = iTc - tc;
+
+    // Catmull-Rom weights
+    vec2 w0 = f * (-0.5 + f * (1.0 - 0.5 * f));
+    vec2 w1 = 1.0 + f * f * (-2.5 + 1.5 * f);
+    vec2 w2 = f * (0.5 + f * (2.0 - 1.5 * f));
+    vec2 w3 = f * f * (-0.5 + 0.5 * f);
+
+    vec2 s12 = w1 + w2;
+    vec2 offset12 = w2 / s12;
+
+    vec2 tc0  = (tc - 1.0) * invTexSize;
+    vec2 tc3  = (tc + 2.0) * invTexSize;
+    vec2 tc12 = (tc + offset12) * invTexSize;
+
+    // 3x3 bilinear taps covering 4x4 texel support
+    vec3 result =
+        (texture(tex, vec2(tc0.x,  tc0.y)).rgb  * w0.x +
+         texture(tex, vec2(tc12.x, tc0.y)).rgb  * s12.x +
+         texture(tex, vec2(tc3.x,  tc0.y)).rgb  * w3.x) * w0.y +
+        (texture(tex, vec2(tc0.x,  tc12.y)).rgb * w0.x +
+         texture(tex, vec2(tc12.x, tc12.y)).rgb * s12.x +
+         texture(tex, vec2(tc3.x,  tc12.y)).rgb * w3.x) * s12.y +
+        (texture(tex, vec2(tc0.x,  tc3.y)).rgb  * w0.x +
+         texture(tex, vec2(tc12.x, tc3.y)).rgb  * s12.x +
+         texture(tex, vec2(tc3.x,  tc3.y)).rgb  * w3.x) * w3.y;
+
+    // Anti-ringing: clamp to range of the 4 nearest texels.
+    // Prevents Catmull-Rom negative lobe overshoots at high-contrast edges.
+    vec2 tcNear = tc * invTexSize;
+    vec3 t00 = texture(tex, tcNear).rgb;
+    vec3 t10 = texture(tex, tcNear + vec2(invTexSize.x, 0.0)).rgb;
+    vec3 t01 = texture(tex, tcNear + vec2(0.0, invTexSize.y)).rgb;
+    vec3 t11 = texture(tex, tcNear + invTexSize).rgb;
+    vec3 minC = min(min(t00, t10), min(t01, t11));
+    vec3 maxC = max(max(t00, t10), max(t01, t11));
+    return clamp(result, minC, maxC);
+}
+
 void main() {
     ivec2 outPixel = ivec2(gl_GlobalInvocationID.xy);
     ivec2 outSize = ivec2(pc.displaySize.xy);
     if (outPixel.x >= outSize.x || outPixel.y >= outSize.y) return;
 
     vec2 outUV = (vec2(outPixel) + 0.5) * pc.displaySize.zw;
-    vec3 currentColor = texture(sceneColor, outUV).rgb;
+
+    // Bicubic upsampling with anti-ringing: sharp without edge halos
+    vec3 currentColor = sampleBicubic(sceneColor, outUV, pc.internalSize.xy);
 
     if (pc.params.x > 0.5) {
         imageStore(historyOutput, outPixel, vec4(currentColor, 1.0));
         return;
     }
 
-    vec2 motion = texture(motionVectors, outUV).rg;
+    // Depth-dilated motion vector: pick the MV from the nearest-to-camera
+    // pixel in a 3x3 neighborhood. Prevents background MVs from bleeding
+    // over foreground edges.
+    vec2 texelSize = pc.internalSize.zw;
+    float closestDepth = texture(depthBuffer, outUV).r;
+    vec2 closestOffset = vec2(0.0);
+    for (int y = -1; y <= 1; y++) {
+        for (int x = -1; x <= 1; x++) {
+            vec2 off = vec2(float(x), float(y)) * texelSize;
+            float d = texture(depthBuffer, outUV + off).r;
+            if (d < closestDepth) {
+                closestDepth = d;
+                closestOffset = off;
+            }
+        }
+    }
+    vec2 motion = texture(motionVectors, outUV + closestOffset).rg;
+
     vec2 historyUV = outUV + motion;
 
     float historyValid = (historyUV.x >= 0.0 && historyUV.x <= 1.0 &&
@@ -50,8 +114,9 @@ void main() {
 
     vec3 historyColor = texture(historyInput, historyUV).rgb;
 
-    // Neighborhood clamping in YCoCg space
-    vec2 texelSize = pc.internalSize.zw;
+    // Neighborhood clamping in YCoCg space with wide gamma.
+    // Wide gamma (3.0) prevents jitter-chasing: the clamp box only catches
+    // truly stale history (disocclusion), not normal jitter variation.
     vec3 s0 = rgbToYCoCg(currentColor);
     vec3 s1 = rgbToYCoCg(texture(sceneColor, outUV + vec2(-texelSize.x, 0.0)).rgb);
     vec3 s2 = rgbToYCoCg(texture(sceneColor, outUV + vec2( texelSize.x, 0.0)).rgb);
@@ -68,7 +133,7 @@ void main() {
     vec3 variance = max(m2 / 9.0 - mean * mean, vec3(0.0));
     vec3 stddev = sqrt(variance);
 
-    float gamma = 1.5;
+    float gamma = 3.0;
     vec3 boxMin = mean - gamma * stddev;
     vec3 boxMax = mean + gamma * stddev;
 
@@ -77,7 +142,19 @@ void main() {
     historyColor = yCoCgToRgb(clampedHistory);
 
     float clampDist = length(historyYCoCg - clampedHistory);
-    float blendFactor = mix(0.05, 0.30, clamp(clampDist * 2.0, 0.0, 1.0));
+
+    // Uniform 5% blend: ~45 frames for 90% convergence.
+    // Simpler than edge-aware; the anti-ringing bicubic handles edge stability.
+    float blendFactor = 0.05;
+
+    // Disocclusion: large clamp distance → rapidly replace stale history
+    blendFactor = mix(blendFactor, 0.60, clamp(clampDist * 5.0, 0.0, 1.0));
+
+    // Velocity: higher blend during motion reduces ghosting
+    float motionMag = length(motion * pc.displaySize.xy);
+    blendFactor = max(blendFactor, clamp(motionMag * 0.15, 0.0, 0.35));
+
+    // Full current frame when history is out of bounds
     blendFactor = mix(blendFactor, 1.0, 1.0 - historyValid);
 
     vec3 result = mix(historyColor, currentColor, blendFactor);

assets/shaders/fsr2_motion.comp.glsl

@@ -8,6 +8,7 @@ layout(set = 0, binding = 1, rg16f) uniform writeonly image2D motionVectors;
 layout(push_constant) uniform PushConstants {
     mat4 reprojMatrix;      // prevUnjitteredVP * inverse(currentUnjitteredVP)
     vec4 resolution;        // xy = internal size, zw = 1/internal size
+    vec4 jitterOffset;      // xy = current jitter (NDC), zw = unused
 } pc;
 
 void main() {
@@ -15,21 +16,30 @@ void main() {
     ivec2 imgSize = ivec2(pc.resolution.xy);
     if (pixelCoord.x >= imgSize.x || pixelCoord.y >= imgSize.y) return;
 
-    // Sample depth (Vulkan: 0 = near, 1 = far)
     float depth = texelFetch(depthBuffer, pixelCoord, 0).r;
 
-    // Pixel center in UV [0,1] and NDC [-1,1]
+    // Pixel center UV and NDC
     vec2 uv = (vec2(pixelCoord) + 0.5) * pc.resolution.zw;
     vec2 ndc = uv * 2.0 - 1.0;
 
-    // Clip-to-clip reprojection: current unjittered clip → previous unjittered clip
-    vec4 clipPos = vec4(ndc, depth, 1.0);
+    // Unjitter the NDC: the scene was rendered with jitter applied to
+    // projection[2][0/1]. For RH perspective (P[2][3]=-1, clip.w=-vz):
+    //   jittered_ndc = unjittered_ndc - jitter
+    //   unjittered_ndc = ndc + jitter
+    vec2 unjitteredNDC = ndc + pc.jitterOffset.xy;
+
+    // Reproject to previous frame via unjittered VP matrices
+    vec4 clipPos = vec4(unjitteredNDC, depth, 1.0);
     vec4 prevClip = pc.reprojMatrix * clipPos;
     vec2 prevNdc = prevClip.xy / prevClip.w;
     vec2 prevUV = prevNdc * 0.5 + 0.5;
 
-    // Motion = previous position - current position (both unjittered, in UV space)
-    vec2 motion = prevUV - uv;
+    // Current unjittered UV for this pixel's world content
+    vec2 currentUnjitteredUV = unjitteredNDC * 0.5 + 0.5;
+
+    // Motion between unjittered positions — jitter-free.
+    // For a static scene (identity reprojMatrix), this is exactly zero.
+    vec2 motion = prevUV - currentUnjitteredUV;
 
     imageStore(motionVectors, pixelCoord, vec4(motion, 0.0, 0.0));
 }

src/rendering/renderer.cpp

@@ -3785,7 +3785,7 @@ bool Renderer::initFSR2Resources() {
         VkPushConstantRange pc{};
         pc.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT;
         pc.offset = 0;
-        pc.size = sizeof(glm::mat4) + sizeof(glm::vec4);  // 80 bytes
+        pc.size = sizeof(glm::mat4) + 2 * sizeof(glm::vec4);  // 96 bytes
 
         VkPipelineLayoutCreateInfo plCI{VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO};
         plCI.setLayoutCount = 1;
@@ -4086,17 +4086,20 @@ void Renderer::dispatchMotionVectors() {
     vkCmdBindDescriptorSets(currentCmd, VK_PIPELINE_BIND_POINT_COMPUTE,
         fsr2_.motionVecPipelineLayout, 0, 1, &fsr2_.motionVecDescSet, 0, nullptr);
 
-    // Single reprojection matrix: prevUnjitteredVP * inv(currentUnjitteredVP)
-    // Both matrices are unjittered — jitter only affects sub-pixel sampling,
-    // not motion vector computation. This avoids numerical instability from
-    // jitter amplification through large world coordinates.
+    // Reprojection: prevUnjitteredVP * inv(currentUnjitteredVP)
+    // Using unjittered VPs avoids numerical instability from jitter amplification
+    // through large world coordinates. The shader corrects NDC by subtracting
+    // current jitter before reprojection (depth was rendered at jittered position).
     struct {
-        glm::mat4 reprojMatrix;   // prevUnjitteredVP * inv(currentUnjitteredVP)
+        glm::mat4 reprojMatrix;
         glm::vec4 resolution;
+        glm::vec4 jitterOffset;   // xy = current jitter (NDC), zw = unused
     } pc;
 
     glm::mat4 currentUnjitteredVP = camera->getUnjitteredViewProjectionMatrix();
     pc.reprojMatrix = fsr2_.prevViewProjection * glm::inverse(currentUnjitteredVP);
+    glm::vec2 jitter = camera->getJitter();
+    pc.jitterOffset = glm::vec4(jitter.x, jitter.y, 0.0f, 0.0f);
     pc.resolution = glm::vec4(
         static_cast<float>(fsr2_.internalWidth),
         static_cast<float>(fsr2_.internalHeight),