Add non subgroup based prefix sum

src/main/java/me/cortex/voxy/client/core/rendering/section/MDICSectionRenderer.java

@@ -67,7 +67,8 @@ public class MDICSectionRenderer extends AbstractSectionRenderer<MDICViewport, B
             .compile();
 
     private final Shader prefixSumShader = Shader.make()
-            .add(ShaderType.COMPUTE, "voxy:util/prefixsum/inital3.comp")
+            //Use subgroup prefix sum if possible otherwise use dodgy... slow prefix sum
+            .add(ShaderType.COMPUTE, Capabilities.INSTANCE.subgroup?"voxy:util/prefixsum/inital3.comp":"voxy:util/prefixsum/simple.comp")
             .define("IO_BUFFER", 0)
             .compile();
 

src/main/resources/assets/voxy/shaders/util/prefixsum/simple.comp

@@ -0,0 +1,56 @@
+#version 460
+
+#extension GL_KHR_shader_subgroup_basic : require
+#extension GL_KHR_shader_subgroup_arithmetic: require
+
+#define WORK_SIZE 256
+
+//Does inital parralel prefix sum on batches of WORK_SIZE
+layout(local_size_x=WORK_SIZE) in;
+
+layout(binding = IO_BUFFER, std430) buffer InputBuffer {
+    uvec4[] ioCount;
+};
+
+shared uint prefixSum[256];
+
+#define STEP(i)  if ((I&(1u<<i))!=0) prefixSum[I] += prefixSum[(I&~((1u<<(i+1))-1))+((1u<<i)-1)];
+uint prefixSumExclusive(uint value) {
+    uint I = gl_LocalInvocationID.x;
+    prefixSum[I] = value;
+    barrier();
+    STEP(0)
+    barrier();
+    STEP(1)
+    barrier();
+    STEP(2)
+    barrier();
+    STEP(3)
+    barrier();
+    STEP(4)
+    barrier();
+    STEP(5)
+    barrier();
+    STEP(6)
+    barrier();
+    STEP(7)
+    barrier();
+    return prefixSum[I]-value;
+}
+
+void main() {
+    uint gid = gl_LocalInvocationID.x;
+    prefixSum[gid] = 0;
+    uvec4 count = uvec4(0);
+    uint sum = 0;
+    {
+        uvec4 dat = ioCount[gid];
+        count.yzw = dat.xyz;
+        count.z += count.y;
+        count.w += count.z;
+        sum = count.w + dat.w;
+    }
+    barrier();
+    count += prefixSumExclusive(sum);
+    ioCount[gid] = count;
+}