UI rewrite 3rd attempt

examples/CustomShader/inverse-circle.comp.glsl

@@ -0,0 +1,61 @@
+// Custom UI compute shader. Demonstrates the Tier 1 dispatch path:
+// the user defines their own item struct, writes their own GLSL alongside
+// the standard shaders (sharing the same UIDispatchHeader contract via
+// ui-shared.glsl), and dispatches it via UIRenderer::Dispatch.
+//
+// What it does: each item is a circle. For every pixel the workgroup tile
+// owns, if the pixel falls inside any item-circle, the pixel's RGB is
+// inverted (1 - rgb). Composes naturally with whatever previous dispatches
+// drew into the swapchain image — works on top of standard quads, custom
+// effects, anything.
+#version 460
+#extension GL_GOOGLE_include_directive : enable
+#include "ui-shared.glsl"
+
+// Application-defined item: just (cx, cy, radius, _pad). Layout matches the
+// C++ InverseCircleItem struct in main.cpp byte-for-byte under std430.
+struct InverseCircleItem {
+    vec4 centerRadius;
+};
+
+layout(descriptor_heap, std430) readonly buffer InvCircleBuf {
+    InverseCircleItem items[];
+} invCircleHeap[];
+
+// NVIDIA workaround — same per-member-load pattern the library shaders use
+// for the descriptor-heap'd SSBO composite-load issue.
+InverseCircleItem LoadInverseCircleItem(uint heap, uint i) {
+    InverseCircleItem it;
+    it.centerRadius = invCircleHeap[heap].items[i].centerRadius;
+    return it;
+}
+
+layout(push_constant) uniform PC {
+    UIDispatchHeader hdr;
+} pc;
+
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+
+void main() {
+    ivec2 screenPx;
+    if (!uiResolveScreenPixel(pc.hdr, screenPx)) return;
+
+    vec2 sp = vec2(screenPx) + 0.5;
+
+    // Find the strongest circle coverage at this pixel — using max instead
+    // of literal per-item invert so overlapping circles don't double-invert
+    // back to the original.
+    float coverage = 0.0;
+    for (uint i = 0u; i < pc.hdr.itemCount; ++i) {
+        InverseCircleItem it = LoadInverseCircleItem(pc.hdr.itemBuffer, i);
+        vec2  c = it.centerRadius.xy;
+        float r = it.centerRadius.z;
+        float d = length(sp - c) - r;
+        coverage = max(coverage, clamp(0.5 - d, 0.0, 1.0));
+    }
+    if (coverage <= 0.0) return;
+
+    vec4 dst = imageLoad(uiImages[pc.hdr.outImage], screenPx);
+    dst.rgb  = mix(dst.rgb, vec3(1.0) - dst.rgb, coverage);
+    imageStore(uiImages[pc.hdr.outImage], screenPx, dst);
+}