working buffer compactor

shaders/Makefile

@@ -1,4 +1,4 @@
-all: voxel.spv
+all: voxel.spv compaction.spv
 
 %.spv: %.slang
 	slangc $< -O3 -fvk-use-entrypoint-name -target spirv -o $@

shaders/compaction.slang

@@ -1,13 +1,13 @@
 
 [[vk::binding(0, 0)]]
 RWStructuredBuffer<uint32_t> count_buffer;
+[[vk::binding(1, 0)]]
+RWStructuredBuffer<uint32_t> index_buffer;
 
 [[vk::binding(0, 1)]]
 RWStructuredBuffer<uint32_t> reduced_buffer;
 [[vk::binding(1, 1)]]
 RWStructuredBuffer<uint32_t> sum_buffer;
-[[vk::binding(2, 1)]]
-RWStructuredBuffer<uint32_t> compaction_buffer;
 
 groupshared uint32_t local_data[256 * 2];
 static uint32_t      THREAD_WIDTH = 256;
@@ -50,7 +50,7 @@ void block_sum(
     var width : uint32_t = 2;
     while (width <= DATA_WIDTH)
     {
-        let dest_index = width * (thread_index + 1) - 1;
+        let dest_index = width * (local_thread_index + 1) - 1;
         let get_index = dest_index - (width / 2);
         // println!("{}, {}", get_index, dest_index);
         if (dest_index < DATA_WIDTH)
@@ -61,10 +61,14 @@ void block_sum(
         GroupMemoryBarrierWithGroupSync();
     }
 
+    // Write to reduced buffer
+    reduced_buffer[workgroup_id.x] = local_data[DATA_WIDTH - 1];
+    // reduced_buffer[workgroup_id.x] = local_data[DATA_WIDTH - 1];
+
     local_data[DATA_WIDTH - 1] = 0;
     while (width >= 2)
     {
-        let dest_index = width * (thread_index + 1) - 1;
+        let dest_index = width * (local_thread_index + 1) - 1;
         let get_index = dest_index - (width / 2);
         // println!("{}, {}", get_index, dest_index);
         if (dest_index < DATA_WIDTH)
@@ -81,9 +85,6 @@ void block_sum(
     // Dump back to sum buffer
     sum_buffer[2 * thread_index] = local_data[2 * local_thread_index];
     sum_buffer[2 * thread_index + 1] = local_data[2 * local_thread_index + 1];
-
-    // Write to reduced buffer
-    reduced_buffer[workgroup_id.x] = local_data[DATA_WIDTH - 1];
 }
 
 [numthreads(1, 1, 1)]
@@ -107,7 +108,7 @@ void linear_reduced_sum(
 
 [numthreads(256, 1, 1)]
 [shader("compute")]
-void uniform_add(
+void scatter(
     uint32_t3 workgroup_id: SV_GroupID,
     uint32_t3 local_thread_id: SV_GroupThreadID,
     uint32_t3 global_thread_id: SV_DispatchThreadID)
@@ -116,9 +117,20 @@ void uniform_add(
     let local_thread_index = local_thread_id.x;
 
     // Gather
-    let reduced_value = reduced_buffer[global_thread_id.x];
+    let reduced_value = reduced_buffer[workgroup_id.x];
     // Apply
-    sum_buffer[thread_index * 2] += reduced_value;
-    sum_buffer[thread_index * 2 + 1] += reduced_value;
+    let sum1 = sum_buffer[thread_index * 2] + reduced_value;
+    let sum2 = sum_buffer[thread_index * 2 + 1] + reduced_value;
+
+    // Scatter index to compact index buffer if predicate is true
+    if (count_buffer[thread_index * 2] != 0)
+    {
+        index_buffer[sum1] = thread_index * 2;
+    }
+
+    if (count_buffer[thread_index * 2 + 1] != 0)
+    {
+        index_buffer[sum2] = thread_index * 2 + 1;
+    }
 }