Trying to fix everything

res/shaders/TerrainPass.slang

@@ -31,7 +31,7 @@ VertexOutput vert(VertexInput input)
 
     // Which vertex should be read?
     local_vert_id = local_vert_id == 0 ? 2 : (local_vert_id == 2 ? 0 : 1);
-    float3 positionRWS = pParams.currentVertexBuffer[output.triangleID * 3 + local_vert_id];
+    float3 positionRWS = pParams.currentVertexBuffer[output.triangleID * 3 + local_vert_id].xyz;
 
     // Apply the view projection
     output.positionCS = mul(pViewParams.projectionMatrix, mul(pViewParams.viewMatrix, float4(positionRWS, 1.0)));
@@ -53,26 +53,25 @@ float4 frag(PixelInput input) : SV_Target
 
 [shader("compute")]
 [numthreads(64, 1, 1)]
-void deform(uint currentID: SV_DispatchThreadID)
+void EvaluateDeformation(uint currentID : SV_DispatchThreadID)
 {
-    if(currentID >= pParams.indirectDrawBuffer[9] * 4)
+    // This thread doesn't have any work to do, we're done
+    if (currentID >= pParams.indirectDrawBuffer[9] * 4)
         return;
 
+    // Extract the bisector ID and the vertexID
     uint bisectorID = currentID / 4;
     uint localVertexID = currentID % 4;
 
+    // Operate the indirection
     bisectorID = pParams.indexedBisectorBuffer[bisectorID];
 
+    // Evaluate the source vertex
     currentID = localVertexID < 3 ? bisectorID * 3 + localVertexID : 3 * pParams.geometry.totalNumElements + bisectorID;
 
-    float3 positionWS = pParams.lebPositionBuffer[currentID];
+    // Grab the position that we will be displacing
-    float3 positionRWS = positionWS - pViewParams.cameraPosition_WS.xyz;
+    float3 positionWS = pParams.lebPositionBuffer[currentID].xyz;
+    float3 positionRWS = float3(positionWS - pViewParams.cameraPosition_WS.xyz);
 
-    float3 positionPS = positionWS;
+    pParams.currentVertexBuffer[currentID] = float4(positionRWS, 1);
-
+}
-    float3 normalPS = float3(0, 1, 0);
-
-    float2 sampleUV = float2(0, 0);
-
-    pParams.currentVertexBuffer[currentID] = positionRWS;
-}

res/shaders/terrain/Bisector.slang

@@ -16,12 +16,12 @@ const static int MODIFIED_BISECTOR = 0x2;
 
 struct BisectorData
 {
-    // Subvision that should be applied to this bisector
-    uint32_t subdivisionPattern;
-
     // Allocated indices for this bisector
     uint32_t indices[3];
 
+    // Subvision that should be applied to this bisector
+    uint32_t subdivisionPattern;
+
     // Neighbor that should be processed
     uint32_t problematicNeighbor;
 
@@ -35,7 +35,7 @@ struct BisectorData
     uint32_t propagationID;
 };
 
-uint heapIDDepth(uint64_t x)
+uint HeapIDDepth(uint64_t x)
 {
     uint depth = 0;
     while (x > 0u) {

res/shaders/terrain/CBT.slang

@@ -563,11 +563,7 @@ void load_buffer_to_shared_memory(uint groupIndex)
     // Load the bitfield to the LDS
     for (uint e = 0; e < BUFFER_ELEMENT_PER_LANE; ++e)
     {
-#ifdef AMD
         uint target_element = BUFFER_ELEMENT_PER_LANE * groupIndex + e;
-#else
-        uint target_element = groupIndex + WORKGROUP_SIZE * e;
-#endif
         if (target_element < OCBT_TREE_NUM_SLOTS)
             gs_cbtTree[target_element] = pParams.cbtBuffer[target_element];
     }

res/shaders/terrain/CompileTest.slang

@@ -0,0 +1,50 @@
+
+struct ComputeParams
+{
+    RWStructuredBuffer<uint> indirectDrawBuffer;
+    RWStructuredBuffer<uint64_t> heapIDBuffer;
+    RWStructuredBuffer<uint> classificationBuffer;
+    RWStructuredBuffer<int> allocateBuffer;
+    RWStructuredBuffer<int> memoryBuffer;
+    RWStructuredBuffer<int> propagateBuffer;
+    RWStructuredBuffer<uint> simplifyBuffer;
+};
+ParameterBlock<ComputeParams> pParams;
+
+// Split buffer slots
+static const uint64_t SPLIT_COUNTER = 0;
+static const uint64_t SIMPLIFY_COUNTER = 1;
+static const uint64_t CLASSIFY_COUNTER_OFFSET = 2;
+void ResetBuffers()
+{
+    pParams.memoryBuffer[0] = 0;
+    pParams.memoryBuffer[1] = 0;
+
+    pParams.classificationBuffer[SPLIT_COUNTER] = 0;
+    pParams.classificationBuffer[SIMPLIFY_COUNTER] = 0;
+
+    pParams.allocateBuffer[0] = 0;
+
+    pParams.propagateBuffer[0] = 0;
+    pParams.propagateBuffer[1] = 0;
+
+    pParams.simplifyBuffer[0] = 0;
+
+    pParams.indirectDrawBuffer[0] = 0;
+    pParams.indirectDrawBuffer[1] = 1;
+    pParams.indirectDrawBuffer[2] = 0;
+    pParams.indirectDrawBuffer[3] = 0;
+
+    pParams.indirectDrawBuffer[4] = 0;
+    pParams.indirectDrawBuffer[5] = 1;
+    pParams.indirectDrawBuffer[6] = 0;
+    pParams.indirectDrawBuffer[7] = 0;
+
+    pParams.indirectDrawBuffer[8] = 0;
+}
+
+[numthreads(1, 1, 1)]
+void Reset()
+{
+    ResetBuffers();
+}

res/shaders/terrain/LEB.slang

@@ -0,0 +1,332 @@
+import Parameters;
+import CBT;
+import Bisector;
+
+static const uint32_t WORKGROUP_SIZE = 64;
+
+    // Resolution of the cache
+    #define LEB_TABLE_DEPTH 5ULL
+
+    // Cache in shared memory
+    groupshared float3x3 g_MatrixCache[2ULL << LEB_TABLE_DEPTH];
+
+    void load_leb_matrix_cache_to_shared_memory(uint groupIndex)
+    {
+        if (groupIndex < (2ULL << LEB_TABLE_DEPTH))
+                g_MatrixCache[groupIndex] = pParams.lebMatrixCache[groupIndex];
+        GroupMemoryBarrierWithGroupSync();
+    }
+uint leb_depth(uint64_t heapID)
+{
+    uint depth = 0;
+    while (heapID > 0u)
+    {
+        ++depth;
+        heapID >>= 1u;
+    }
+    return depth - 1;
+}
+
+/*******************************************************************************
+ * GetBitValue -- Returns the value of a bit stored in a 64-bit word
+ *
+ */
+
+uint64_t leb__GetBitValue(uint64_t bitField, int64_t bitID)
+{
+    return ((bitField >> bitID) & 1L);
+}
+
+/*******************************************************************************
+ * IdentityMatrix3x3 -- Sets a 3x3 matrix to identity
+ *
+ */
+
+void leb__IdentityMatrix3x3(out float3x3 m)
+{
+    m[0][0] = 1.0f; m[0][1] = 0.0f; m[0][2] = 0.0f;
+    m[1][0] = 0.0f; m[1][1] = 1.0f; m[1][2] = 0.0f;
+    m[2][0] = 0.0f; m[2][1] = 0.0f; m[2][2] = 1.0f;
+}
+
+/*******************************************************************************
+ * SplittingMatrix -- Computes a LEB splitting matrix from a split bit
+ *
+ */
+
+void leb__SplittingMatrix(inout float3x3 mat, uint64_t bitValue)
+{
+    float b = (float)bitValue;
+    float c = 1.0f - b;
+    const float3x3 splitMatrix = {
+        {0.0f, b, c},
+        {0.5f, 0.0f, 0.5f},
+        {b, c, 0.0f}
+    };
+    mat = mul(splitMatrix, mat);
+}
+
+/*******************************************************************************
+ * SplittingMatrix -- Computes a LEB splitting matrix from a split bit
+ *
+ */
+
+float3x3 leb__SplittingMatrix_out(float3x3 mat, uint64_t bitValue)
+{
+    float b = (float)bitValue;
+    float c = 1.0 - b;
+    float3x3 splitMatrix = {
+        {0.0, b, c},
+        {0.5, 0.0, 0.5},
+        {b, c, 0.0}
+    };
+    return mul(splitMatrix, mat);
+}
+
+/*******************************************************************************
+ * DecodeTransformationMatrix -- Computes the matrix associated to a LEB
+ * node
+ *
+*/
+
+void leb__DecodeTransformationMatrix(uint64_t heapID, out float3x3 mat)
+{
+    int depth = leb_depth(heapID);
+    leb__IdentityMatrix3x3(mat);
+    for (int bitID = depth - 1; bitID >= 0; --bitID)
+        leb__SplittingMatrix(mat, leb__GetBitValue(heapID, bitID));
+}
+
+#if defined(LEB_MATRIX_CACHE_BINDING_SLOT)
+void leb__DecodeTransformationMatrix_Tabulated(uint64_t heapID, out float3x3 mat)
+{
+    leb__IdentityMatrix3x3(mat);
+    const uint64_t msb = (1ULL << LEB_TABLE_DEPTH);
+    const uint64_t mask = ~(~0ULL << LEB_TABLE_DEPTH);
+    while (heapID > mask)
+    {
+        uint32_t index = uint32_t((heapID & mask) | msb);
+        mat = mul(mat, g_MatrixCache[index]);
+        heapID >>= LEB_TABLE_DEPTH;
+    }
+    mat = mul(mat, g_MatrixCache[uint32_t(heapID)]);
+}
+#endif
+
+/*******************************************************************************
+ * DecodeTransformationMatrix -- Computes the matrix associated to a LEB
+ * node
+ *
+*/
+
+void leb__DecodeTransformationMatrix_parent_child(uint64_t heapID, out float3x3 parent, out float3x3 child)
+{
+    int depth = leb_depth(heapID);
+    leb__IdentityMatrix3x3(parent);
+
+    // Evaluate the parent matrix
+    int bitID;
+    for (bitID = depth - 1; bitID > 0; --bitID)
+        leb__SplittingMatrix(parent, leb__GetBitValue(heapID, bitID));
+
+    // Evaluate the child
+    if (depth > 0)
+        child = leb__SplittingMatrix_out(parent, leb__GetBitValue(heapID, bitID));
+    else
+        child = parent;
+}
+
+#if defined(LEB_MATRIX_CACHE_BINDING_SLOT)
+void leb__DecodeTransformationMatrix_parent_child_Tabulated(uint64_t heapID, out float3x3 parent, out float3x3 child)
+{
+    int depth = leb_depth(heapID);
+    leb__IdentityMatrix3x3(parent);
+    const uint64_t msb = (1ULL << LEB_TABLE_DEPTH);
+    const uint64_t mask = ~(~0ULL << LEB_TABLE_DEPTH);
+    uint64_t parentHeapID = heapID / 2;
+    while (parentHeapID > mask)
+    {
+        uint32_t index = uint32_t((parentHeapID & mask) | msb);
+        parent = mul(parent, g_MatrixCache[index]);
+        parentHeapID >>= LEB_TABLE_DEPTH;
+    }
+    if (parentHeapID != 0)
+        parent = mul(parent, g_MatrixCache[uint32_t(parentHeapID)]);
+
+    // Evaluate the child
+    if (depth > 0)
+        child = leb__SplittingMatrix_out(parent, leb__GetBitValue(heapID, 0));
+    else
+        child = parent;
+}
+
+#endif
+
+/*******************************************************************************
+ * DecodeNodeAttributeArray -- Compute the triangle attributes at the input node
+ *
+ */
+
+void leb_DecodeNodeAttributeArray(uint64_t heapID, inout float3 attributeArray[2])
+{
+    float3x3 m;
+    leb__DecodeTransformationMatrix(heapID, m);
+    for (int i = 0; i < 2; ++i)
+    {
+        float3 attributeVector = attributeArray[i];
+        attributeArray[i][0] = dot(m[0], attributeVector);
+        attributeArray[i][1] = dot(m[1], attributeVector);
+        attributeArray[i][2] = dot(m[2], attributeVector);
+    }
+}
+
+void leb_DecodeNodeAttributeArray(uint64_t heapID, inout float3 attributeArray[3])
+{
+    float3x3 m;
+#if defined(LEB_MATRIX_CACHE_BINDING_SLOT)
+    leb__DecodeTransformationMatrix_Tabulated(heapID, m);
+#else
+    leb__DecodeTransformationMatrix(heapID, m);
+#endif
+    for (int i = 0; i < 3; ++i)
+    {
+        float3 attributeVector = attributeArray[i];
+        attributeArray[i][0] = dot(m[0], attributeVector);
+        attributeArray[i][1] = dot(m[1], attributeVector);
+        attributeArray[i][2] = dot(m[2], attributeVector);
+    }
+}
+
+/*******************************************************************************
+ * DecodeNodeAttributeArray -- Compute the triangle attributes at the input node
+ *
+ */
+
+void leb_DecodeNodeAttributeArray_parent_child(uint64_t heapID, inout float3 childAttribute[3], out float3 parentAttribute[3])
+{
+    float3x3 child, parent;
+#if defined(LEB_MATRIX_CACHE_BINDING_SLOT)
+    leb__DecodeTransformationMatrix_parent_child_Tabulated(heapID, parent, child);
+#else
+    leb__DecodeTransformationMatrix_parent_child(heapID, parent, child);
+#endif
+    int i;
+    for (i = 0; i < 3; ++i)
+    {
+        float3 attributeVector = childAttribute[i];
+        parentAttribute[i][0] = dot(parent[0], attributeVector);
+        parentAttribute[i][1] = dot(parent[1], attributeVector);
+        parentAttribute[i][2] = dot(parent[2], attributeVector);
+    }
+
+    for (i = 0; i < 3; ++i)
+    {
+        float3 attributeVector = childAttribute[i];
+        childAttribute[i][0] = dot(child[0], attributeVector);
+        childAttribute[i][1] = dot(child[1], attributeVector);
+        childAttribute[i][2] = dot(child[2], attributeVector);
+    }
+}
+
+[numthreads(WORKGROUP_SIZE, 1, 1)]
+void ClearBuffer(uint currentID : SV_DispatchThreadID)
+{
+    // This thread doesn't have any work to do, we're done
+    if (currentID >= pParams.geometry.totalNumVertices)
+        return;
+    pParams.lebPositionBuffer[currentID] = float4(0.0, 0.0, 0.0, 1.0);
+}
+
+float3 TransformToPlanetCoordinate(float3 posPS)
+{
+    // Evaluate the planet position
+    return normalize(posPS);
+}
+
+[[vk::push_constant]]
+ConstantBuffer<uint> preRendering;
+struct Triangle
+{
+    float3 p[3];
+};
+
+void EvaluateElementPosition(uint64_t heapID, uint32_t vertexDataOffset, uint minDepth, RWStructuredBuffer<float4> vertexBuffer, out Triangle parentTri, out Triangle childTri)
+{
+    // Get the depth of the element
+    uint depth = HeapIDDepth(heapID);
+
+    // Compute the required shift to find the original vertices
+    uint64_t subTreeDepth = depth - minDepth;
+
+    // Compute the base heapID
+    uint64_t baseHeapID = 1u << (minDepth - 1);
+    uint primitiveID = uint((heapID >> subTreeDepth) - baseHeapID);
+
+    // Grab the base positions of the element
+    float3 p0 = float3(vertexBuffer[3 * primitiveID + vertexDataOffset].xyz);
+    float3 p1 = float3(vertexBuffer[3 * primitiveID + 1 + vertexDataOffset].xyz);
+    float3 p2 = float3(vertexBuffer[3 * primitiveID + 2 + vertexDataOffset].xyz);
+
+    // Heap ID in the sub triangle
+    uint64_t mask = subTreeDepth != 0uL ? 0xFFFFFFFFFFFFFFFFull >> (64ull - subTreeDepth) : 0ull;
+    uint64_t baseHeap = (1ull << subTreeDepth);
+    uint64_t baseMask = (mask & heapID);
+    uint64_t subHeapID = baseMask + baseHeap;
+
+    // Generate the triangle positions
+    float3 childArray[3] = {{p0.x, p1.x, p2.x}, {p0.y, p1.y, p2.y}, {p0.z, p1.z, p2.z}};
+    float3 parentArray[3];
+
+    // Decode
+    leb_DecodeNodeAttributeArray_parent_child(subHeapID, childArray, parentArray);
+
+    // Fill the parent triangle
+    parentTri.p[0] = float3(parentArray[0][0], parentArray[1][0], parentArray[2][0]);
+    parentTri.p[1] = float3(parentArray[0][1], parentArray[1][1], parentArray[2][1]);
+    parentTri.p[2] = float3(parentArray[0][2], parentArray[1][2], parentArray[2][2]);
+
+    // Fill the child triangle
+    Triangle child;
+    childTri.p[0] = float3(childArray[0][0], childArray[1][0], childArray[2][0]);
+    childTri.p[1] = float3(childArray[0][1], childArray[1][1], childArray[2][1]);
+    childTri.p[2] = float3(childArray[0][2], childArray[1][2], childArray[2][2]);
+}
+[numthreads(WORKGROUP_SIZE, 1, 1)]
+void EvaluateLEB(uint currentID : SV_DispatchThreadID, uint groupIndex: SV_GroupIndex)
+{
+    #if defined(LEB_MATRIX_CACHE_BINDING_SLOT)
+    // Make sure these are loaded to the shared memory
+    load_leb_matrix_cache_to_shared_memory(groupIndex);
+    #endif
+
+    // This thread doesn't have any work to do, we're done
+    uint32_t numBisectors = bool(preRendering) ? (pParams.indirectDrawBuffer[9]) : (pParams.indirectDrawBuffer[8] / 4);
+    if (currentID >= numBisectors)
+        return;
+
+    // Load the bisector for this element
+    currentID = pParams.indexedBisectorBuffer[currentID];
+
+    // Evaluate the depth of the element
+    uint64_t cHeapID = pParams.heapIDBuffer[currentID];
+    uint depth = HeapIDDepth(cHeapID);
+
+    // Evaluate the positions of the current element
+    Triangle parentTri, childTri;
+    EvaluateElementPosition(cHeapID, 0, pParams.geometry.baseDepth, pParams.currentVertexBuffer, parentTri, childTri);
+
+    // Export the child
+    pParams.lebPositionBuffer[3 * currentID] = float4(TransformToPlanetCoordinate(childTri.p[0]), 1.0f);
+    pParams.lebPositionBuffer[3 * currentID + 1] = float4(TransformToPlanetCoordinate(childTri.p[1]), 1.0f);
+    pParams.lebPositionBuffer[3 * currentID + 2] = float4(TransformToPlanetCoordinate(childTri.p[2]), 1.0f);
+
+    // Export the fourth element
+    if (pParams.geometry.baseDepth < depth)
+    {
+        // Offset for the parent buffer
+        const uint parentOffset = 3 * pParams.geometry.totalNumElements;
+
+        // Transform the coordinate to planet space
+        pParams.lebPositionBuffer[parentOffset + currentID] = float4(TransformToPlanetCoordinate(cHeapID % 2 == 0 ? parentTri.p[0] : parentTri.p[2]), 1.0f);
+    }
+}

res/shaders/terrain/Parameters.slang

@@ -20,6 +20,7 @@ struct DeformationCB
 
 struct UpdateCB
 {
+    float4x4 viewProjectionMatrix;
     float triangleSize;
     uint32_t maxSubdivisionDepth;
     float fov;
@@ -28,12 +29,8 @@ struct UpdateCB
 
 struct DebugStruct
 {
-    float4 sourceP[3];
+    uint3 neighbours;
-    float2 areaP[3];
+    int status;
-    float area;
-    int validity;
-    float fDotV;
-    float vDotN;
 }
 
 struct ComputeParams
@@ -41,7 +38,7 @@ struct ComputeParams
     ConstantBuffer<GeometryCB> geometry;
     ConstantBuffer<UpdateCB> update;
 
-	RWStructuredBuffer<float3> currentVertexBuffer;
+	RWStructuredBuffer<float4> currentVertexBuffer;
 	StructuredBuffer<uint> indexedBisectorBuffer;
     RWStructuredBuffer<uint> indirectDrawBuffer;
     RWStructuredBuffer<uint64_t> heapIDBuffer;
@@ -49,8 +46,8 @@ struct ComputeParams
     RWStructuredBuffer<uint> classificationBuffer;
     RWStructuredBuffer<int> allocateBuffer;
     RWStructuredBuffer<uint> indirectDispatchBuffer;
-    RWStructuredBuffer<uint3> neighboursBuffer;
+    RWStructuredBuffer<uint4> neighboursBuffer;
-    RWStructuredBuffer<uint3> neighboursOutputBuffer;
+    RWStructuredBuffer<uint4> neighboursOutputBuffer;
     RWStructuredBuffer<int> memoryBuffer;
     RWStructuredBuffer<uint> cbtBuffer;
     RWStructuredBuffer<uint64_t> bitFieldBuffer;
@@ -60,7 +57,7 @@ struct ComputeParams
     RWStructuredBuffer<uint> bisectorIndicesBuffer;
     RWStructuredBuffer<uint> visibleBisectorIndices;
     RWStructuredBuffer<uint> modifiedBisectorIndices;
-    RWStructuredBuffer<float3> lebPositionBuffer;
+    RWStructuredBuffer<float4> lebPositionBuffer;
     StructuredBuffer<float3x3> lebMatrixCache;
     RWStructuredBuffer<DebugStruct> debugBuffer;
 };

res/shaders/terrain/update_utils.slang

@@ -0,0 +1,966 @@
+import Common;
+import Bisector;
+import CBT;
+import Parameters;
+
+// Needs to be defined before including update_utilities
+struct BisectorGeometry
+{
+    float3 p[4];
+};
+
+// Possible splits
+static const uint64_t NO_SPLIT = 0x00;
+static const uint64_t CENTER_SPLIT = 0x01;
+static const uint64_t RIGHT_SPLIT = 0x02;
+static const uint64_t LEFT_SPLIT = 0x04;
+static const uint64_t RIGHT_DOUBLE_SPLIT = (CENTER_SPLIT | RIGHT_SPLIT);
+static const uint64_t LEFT_DOUBLE_SPLIT = (CENTER_SPLIT | LEFT_SPLIT);
+static const uint64_t TRIPLE_SPLIT = (CENTER_SPLIT | RIGHT_SPLIT | LEFT_SPLIT);
+
+// Split buffer slots
+static const uint64_t SPLIT_COUNTER = 0;
+static const uint64_t SIMPLIFY_COUNTER = 1;
+static const uint64_t CLASSIFY_COUNTER_OFFSET = 2;
+
+int ClassifyBisector(in BisectorGeometry tri, uint depth)
+{
+    // Check the triangle's visibility
+    float3 triNormal = normalize(cross(tri.p[2] - tri.p[1], tri.p[0] - tri.p[1]));
+    float3 triCenter = (tri.p[0] + tri.p[1] + tri.p[2]) / 3.0;
+    float3 viewDir = normalize(-triCenter);
+    float FdotV = dot(viewDir, pViewParams.cameraForward_WS.xyz);
+    float VdotN = dot(viewDir, triNormal);
+
+    // Here we don't use 0 as it introduces stability issues at grazing angles
+    if (FdotV < 0.0 && VdotN < -1e-3)
+        return BACK_FACE_CULLED;
+
+    // Compute the triangle's AABB
+    float3 aabbMin = float3(min(min(tri.p[0].x, tri.p[1].x), tri.p[2].x), min(min(tri.p[0].y, tri.p[1].y), tri.p[2].y), min(min(tri.p[0].z, tri.p[1].z), tri.p[2].z));
+    float3 aabbMax = float3(max(max(tri.p[0].x, tri.p[1].x), tri.p[2].x), max(max(tri.p[0].y, tri.p[1].y), tri.p[2].y), max(max(tri.p[0].z, tri.p[1].z), tri.p[2].z));
+
+    // First we do a frustum culling pass
+    //if (!FrustumAABBIntersect(_FrustumPlanes, aabbMin, aabbMax))
+    //    return FRUSTUM_CULLED;
+
+    // Project the points on screen
+    float4x4 viewProjectionMatrix = mul(pViewParams.projectionMatrix, pViewParams.viewMatrix);
+    float4 p0P = mul(viewProjectionMatrix, float4(tri.p[0], 1.0));
+    p0P.xy = p0P.xy / p0P.w;
+    p0P.xy = (p0P.xy * 0.5 + 0.5);
+
+    float4 p1P = mul(viewProjectionMatrix, float4(tri.p[1], 1.0));
+    p1P.xy = p1P.xy / p1P.w;
+    p1P.xy = (p1P.xy * 0.5 + 0.5);
+
+    float4 p2P = mul(viewProjectionMatrix, float4(tri.p[2], 1.0));
+    p2P.xy = p2P.xy / p2P.w;
+    p2P.xy = (p2P.xy * 0.5 + 0.5);
+
+    // 2D area of the triangle
+    // here we didn't reverse the sign of the y coordinate at projection time, but we simply adapted the area evaluation
+    // The same way we don't multiply by the screen size before, but after which is equivalent
+    float area = 0.5 * abs(p0P.x * (p2P.y - p1P.y) + p1P.x * (p0P.y - p2P.y) + p2P.x * (p1P.y - p0P.y));
+    area *= pViewParams.screenDimensions.x * pViewParams.screenDimensions.y;
+
+    // We over estimate the area at grazing angles
+    float areaOverestimation = lerp(2.0, 1.0, pow(VdotN, 0.2));
+    area *= areaOverestimation;
+
+    // If the triangle's area is bigger than the target size and the depth is not the maximal depth, subdivide
+    if (pParams.update.triangleSize < area && depth < pParams.update.maxSubdivisionDepth)
+    {
+        // If the area is really big, put it in high priority.
+        return BISECT_ELEMENT;
+    }
+    else if ((pParams.update.triangleSize * 0.5 > area) || (depth > pParams.update.maxSubdivisionDepth))
+    {
+        // Transform the parent's point
+        float4 p3P = mul(viewProjectionMatrix, float4(tri.p[3], 1.0));
+        p3P.xy = p3P.xy / p3P.w;
+        p3P.xy = (p3P.xy * 0.5 + 0.5);
+
+        // Evaluate the parent area
+        float areaParent = 0.5 * abs(p0P.x * (p2P.y - p3P.y) + p3P.x * (p0P.y - p2P.y) + p2P.x * (p3P.y - p0P.y));
+        areaParent *= pViewParams.screenDimensions.x * pViewParams.screenDimensions.y;
+        areaParent *= areaOverestimation;
+
+        // If the depth is too high (max depth changed) or the area is too 
+        return ((pParams.update.triangleSize >= areaParent ) || (depth > pParams.update.maxSubdivisionDepth)) ? TOO_SMALL : UNCHANGED_ELEMENT;
+    }
+    return UNCHANGED_ELEMENT;
+}
+
+void ResetBuffers()
+{
+    pParams.memoryBuffer[0] = 0;
+    pParams.memoryBuffer[1] = cbt_size() - bit_count_buffer();
+
+    pParams.classificationBuffer[SPLIT_COUNTER] = 0;
+    pParams.classificationBuffer[SIMPLIFY_COUNTER] = 0;
+
+    pParams.allocateBuffer[0] = 0;
+
+    pParams.propagateBuffer[0] = 0;
+    pParams.propagateBuffer[1] = 0;
+
+    pParams.simplifyBuffer[0] = 0;
+
+    pParams.indirectDrawBuffer[0] = 0;
+    pParams.indirectDrawBuffer[1] = 1;
+    pParams.indirectDrawBuffer[2] = 0;
+    pParams.indirectDrawBuffer[3] = 0;
+
+    pParams.indirectDrawBuffer[4] = 0;
+    pParams.indirectDrawBuffer[5] = 1;
+    pParams.indirectDrawBuffer[6] = 0;
+    pParams.indirectDrawBuffer[7] = 0;
+
+    pParams.indirectDrawBuffer[8] = 0;
+}
+
+void ClassifyElement(uint currentID, BisectorGeometry bis, uint totalNumElements, uint baseDepth)
+{
+    // Evaluate the depth of the element
+    uint64_t heapID = pParams.heapIDBuffer[currentID];
+    uint depth = HeapIDDepth(heapID);
+    BisectorData cbisectorData = pParams.bisectorDataBuffer[currentID];
+
+    // Reset some values
+    cbisectorData.subdivisionPattern = 0;
+    cbisectorData.bisectorState = UNCHANGED_ELEMENT;
+    cbisectorData.problematicNeighbor = INVALID_POINTER;
+    cbisectorData.flags = VISIBLE_BISECTOR;
+    
+    // Does this triangle intersect the circle?
+    int currentValidity = ClassifyBisector(bis, depth);
+    if (currentValidity > UNCHANGED_ELEMENT)
+    {
+        // This element should be bisected
+        uint targetSlot = 0;
+        cbisectorData.bisectorState = BISECT_ELEMENT;
+        InterlockedAdd(pParams.classificationBuffer[SPLIT_COUNTER], 1, targetSlot);
+        pParams.classificationBuffer[CLASSIFY_COUNTER_OFFSET + targetSlot] = currentID;
+
+    }
+    else
+        cbisectorData.flags = currentValidity >= TOO_SMALL ? VISIBLE_BISECTOR : 0;
+
+    // What's the validity of the father?
+    if (baseDepth != depth && currentValidity < UNCHANGED_ELEMENT)
+    {   
+        // Mark that it requires simplification
+        cbisectorData.bisectorState = SIMPLIFY_ELEMENT;
+
+        // Only register it if it has an even heapID, the odd ones will be processed by the even ones
+        if (heapID % 2 == 0)
+        {
+            uint targetSlot = 0;
+            InterlockedAdd(pParams.classificationBuffer[SIMPLIFY_COUNTER], 1, targetSlot);
+            pParams.classificationBuffer[CLASSIFY_COUNTER_OFFSET + totalNumElements + targetSlot] = currentID;
+        }
+    }
+
+    // Update the bisector data
+    pParams.bisectorDataBuffer[currentID] = cbisectorData;
+}
+
+void SplitElement(uint currentID, uint baseDepth)
+{
+    // Get the neighbors information
+    uint3 cNeighbors = pParams.neighboursBuffer[currentID].xyz;
+
+    // If there is a neighbor X
+    if (cNeighbors.x != INVALID_POINTER)
+    {
+        // This is on the path of it's neighbor X
+        uint3 xNeighbors = pParams.neighboursBuffer[cNeighbors.x].xyz;
+        if (xNeighbors.z == currentID && pParams.bisectorDataBuffer[cNeighbors.x].bisectorState != UNCHANGED_ELEMENT)
+            return;
+    }
+
+    // If there is a neighbor Y
+    if (cNeighbors.y != INVALID_POINTER)
+    {
+        // This is on the path of it's neighbor Y
+        uint3 yNeighbors = pParams.neighboursBuffer[cNeighbors.y].xyz;
+        if (yNeighbors.z == currentID && pParams.bisectorDataBuffer[cNeighbors.y].bisectorState != UNCHANGED_ELEMENT)
+            return;
+    }
+
+    // Depth of the current triangle
+    uint64_t heapID = pParams.heapIDBuffer[currentID];
+    uint currentDepth = HeapIDDepth(heapID);
+
+    // Compute the maximal required memory for this subdivision
+    int maxRequiredMemory = 2 * (currentDepth - baseDepth) - 1;
+
+    // Get the twin information
+    uint twinID = cNeighbors.z;
+
+    // This avoid the massive over-reservation and saves a bunch of artifacts
+    if (twinID == INVALID_POINTER)
+        maxRequiredMemory = 1;
+    else if (pParams.neighboursBuffer[twinID].z == currentID)
+        maxRequiredMemory = 2;
+
+    // Try to reserve
+    int remainingMemory;
+    InterlockedAdd(pParams.memoryBuffer[1], -maxRequiredMemory, remainingMemory);
+
+    // Did someone manage to sneak-in while we were trying to pick the memory, add it back and try again
+    if (remainingMemory < maxRequiredMemory)
+    {
+        // Then add back the required memory and stop
+        InterlockedAdd(pParams.memoryBuffer[1], maxRequiredMemory, remainingMemory);
+        return;
+    }
+
+    // Let's actually count the memory that we will be using
+    uint usedMemory = 1;
+    uint prevPattern;
+    InterlockedOr(pParams.bisectorDataBuffer[currentID].subdivisionPattern, CENTER_SPLIT, prevPattern);
+
+    // If this is not zero, it means an other neighbor went faster than us, we restore the memory and leave.
+    if (prevPattern != 0)
+    {
+        InterlockedAdd(pParams.memoryBuffer[1], maxRequiredMemory, remainingMemory);
+        return;
+    }
+
+    // Mark this for allocation
+    uint targetLocation = 0;
+    InterlockedAdd(pParams.allocateBuffer[0], 1, targetLocation);
+    pParams.allocateBuffer[1 + targetLocation] = currentID;
+
+    // While we're not done (up the tree or everything is subdivided properly)
+    bool done = false;
+    while (!done)
+    {
+        // If this neighbor is not allocated, we're done.
+        if (twinID == INVALID_POINTER)
+            break;
+
+        // Grab the bisector of the neighbor
+        uint64_t nHeapID = pParams.heapIDBuffer[twinID];
+        BisectorData nBisectorData = pParams.bisectorDataBuffer[twinID];
+        uint nDepth = HeapIDDepth(nHeapID);
+        uint3 nNeighbors = pParams.neighboursBuffer[twinID].xyz;
+
+        // If both triangles have the same depth
+        if (nDepth == currentDepth)
+        {
+            // Raised the center split
+            InterlockedOr(pParams.bisectorDataBuffer[twinID].subdivisionPattern, CENTER_SPLIT, prevPattern);
+
+            // Only account for it if it was not raised before.
+            if (prevPattern == 0)
+            {
+                // Mark this for allocation
+                uint targetLocation = 0;
+                InterlockedAdd(pParams.allocateBuffer[0], 1, targetLocation);
+                pParams.allocateBuffer[1 + targetLocation] = twinID;
+                usedMemory++;
+            }
+
+            // And we're done
+            done = true;
+        }
+        // If this node has already been subdivided, it means that we need to add the third subdivision and we're done
+        else
+        {
+            if (nNeighbors[0] == currentID)
+                InterlockedOr(pParams.bisectorDataBuffer[twinID].subdivisionPattern, RIGHT_DOUBLE_SPLIT, prevPattern);
+            else // if (nNeighbors[1] == currentID)
+                InterlockedOr(pParams.bisectorDataBuffer[twinID].subdivisionPattern, LEFT_DOUBLE_SPLIT, prevPattern);
+
+            if (prevPattern != 0)
+            {
+                usedMemory++;
+                done = true;
+            }
+            else
+            {
+                // Mark this for allocation
+                uint targetLocation = 0;
+                InterlockedAdd(pParams.allocateBuffer[0], 1, targetLocation);
+                pParams.allocateBuffer[1 + targetLocation] = twinID;
+                
+                // Account for two splits
+                usedMemory += 2;
+
+                // the new bisector that needs to be propagated
+                currentID = twinID;
+                currentDepth = nDepth;
+                twinID = pParams.neighboursBuffer[currentID].z;
+            }
+        }
+    }
+
+    // Add back the unused memory (in case)
+    InterlockedAdd(pParams.memoryBuffer[1], max(maxRequiredMemory - usedMemory, 0), remainingMemory);
+}
+
+void AllocateElement(uint currentID)
+{
+    // Load the bisector for this element
+    BisectorData bisectorData = pParams.bisectorDataBuffer[currentID];
+
+    // Does this guy need to be subdivided
+    if (bisectorData.subdivisionPattern != 0)
+    {
+        // How many bits do we need?
+        int numSlots = countbits(bisectorData.subdivisionPattern);
+
+        // Request the number of bits we need using an interlock add
+        uint firstBitIndex = 0;
+        InterlockedAdd(pParams.memoryBuffer[0], numSlots, firstBitIndex);
+
+        // llocate the bits we need
+        for (uint bitId = 0; bitId < numSlots; ++bitId)
+        {
+            uint index = decode_bit_complement(firstBitIndex + bitId);
+            bisectorData.indices[bitId] = index;
+        }
+
+        // Output 
+        pParams.bisectorDataBuffer[currentID] = bisectorData;
+    }
+}
+
+#define SUBLING0_ID 0
+#define SUBLING1_ID 1
+#define SUBLING2_ID 2
+void evaluate_neighbors(uint currentID, uint bisectorID, out uint resX, out uint resY)
+{
+    BisectorData nBisectorData = pParams.bisectorDataBuffer[bisectorID];
+    uint3 nNeighbors = pParams.neighboursBuffer[bisectorID].xyz;
+    if (nBisectorData.subdivisionPattern == 0x01)
+    {
+        resX = nBisectorData.indices[SUBLING0_ID];
+        resY = bisectorID;
+    }
+    else if (nBisectorData.subdivisionPattern == 0x03)
+    {
+        if (nNeighbors[0] == currentID)
+        {
+            resX = nBisectorData.indices[SUBLING1_ID];
+            resY = bisectorID;
+        }
+        else
+        {
+            resX = nBisectorData.indices[SUBLING0_ID];
+            resY = nBisectorData.indices[SUBLING1_ID];
+        }
+    }
+    else if (nBisectorData.subdivisionPattern == 0x05)
+    {
+        if (nNeighbors[1] == currentID)
+        {
+            resX = nBisectorData.indices[SUBLING1_ID];
+            resY = nBisectorData.indices[SUBLING0_ID];
+        }
+        else
+        {
+            resX = nBisectorData.indices[SUBLING0_ID];
+            resY = bisectorID;
+        }
+    }
+    else
+    {
+        if (nNeighbors[0] == currentID)
+        {
+            resX = nBisectorData.indices[SUBLING1_ID];
+            resY = bisectorID;
+        }
+        else if (nNeighbors[1] == currentID)
+        {
+            resX = nBisectorData.indices[SUBLING2_ID];
+            resY = nBisectorData.indices[SUBLING0_ID];
+        }
+        else
+        {
+            resX = nBisectorData.indices[SUBLING0_ID];
+            resY = nBisectorData.indices[SUBLING1_ID];
+        }
+    }
+}
+
+void BisectElement(uint currentID)
+{
+    // If this bisector is not allocated or not subdivided, stop right away
+    uint64_t baseHeapID = pParams.heapIDBuffer[currentID];
+    BisectorData cBisectorData = pParams.bisectorDataBuffer[currentID];
+    if (baseHeapID == 0 || cBisectorData.subdivisionPattern == NO_SPLIT)
+        return;
+
+    // Load the bisector data of the target triangle
+    uint currentSubdiv = cBisectorData.subdivisionPattern;
+
+    // neighbors of the parent
+    uint3 cNeighbors = pParams.neighboursBuffer[currentID].xyz;
+    uint p_n0 = cNeighbors[0];
+    uint p_n1 = cNeighbors[1];
+    uint p_n2 = cNeighbors[2];
+
+    // Get the main axis subdiv
+    uint siblingID0 = cBisectorData.indices[0];
+    uint siblingID1 = cBisectorData.indices[1];
+    uint siblingID2 = cBisectorData.indices[2];
+
+    // Simple subdivision (along the main axis)
+    if (currentSubdiv == CENTER_SPLIT)
+    {   
+        uint resX = INVALID_POINTER, resY = INVALID_POINTER;
+        if (p_n2 != INVALID_POINTER)
+            evaluate_neighbors(currentID, p_n2, resX, resY);
+
+        // Set the heap IDs
+        pParams.heapIDBuffer[currentID] = 2 * baseHeapID;
+        pParams.heapIDBuffer[siblingID0] = 2 * baseHeapID + 1;
+
+        // Update the neighbors
+        uint3 modifiedNeighbors;
+        modifiedNeighbors[0] = siblingID0;
+        modifiedNeighbors[1] = resX;
+        modifiedNeighbors[2] = p_n0;
+        pParams.neighboursOutputBuffer[currentID] = uint4(modifiedNeighbors, 0);
+        modifiedNeighbors[0] = resY;
+        modifiedNeighbors[1] = currentID;
+        modifiedNeighbors[2] = p_n1;
+        pParams.neighboursOutputBuffer[siblingID0] = uint4(modifiedNeighbors, 0);
+
+        // Keep track of the parent
+        BisectorData modifiedBisector = cBisectorData;
+        modifiedBisector.propagationID = currentID;
+
+        modifiedBisector.problematicNeighbor = INVALID_POINTER;
+        modifiedBisector.flags = (VISIBLE_BISECTOR | MODIFIED_BISECTOR);
+        pParams.bisectorDataBuffer[currentID] = modifiedBisector;
+
+        modifiedBisector.problematicNeighbor = p_n1;
+        modifiedBisector.flags = (VISIBLE_BISECTOR | MODIFIED_BISECTOR);
+        pParams.bisectorDataBuffer[siblingID0] = modifiedBisector;
+
+        // Mark this for propagation
+        uint targetLocation = 0;
+        InterlockedAdd(pParams.propagateBuffer[0], 1, targetLocation);
+        pParams.propagateBuffer[2 + targetLocation] = siblingID0;
+    }
+    else if (currentSubdiv == RIGHT_DOUBLE_SPLIT)
+    {
+        // Grab the bisector of the twin
+        uint res0X = INVALID_POINTER, res0Y = INVALID_POINTER;
+        evaluate_neighbors(currentID, p_n0, res0X, res0Y);
+
+        uint res1X = INVALID_POINTER, res1Y = INVALID_POINTER;
+        if (p_n2 != INVALID_POINTER)
+            evaluate_neighbors(currentID, p_n2, res1X, res1Y);
+
+        // Set the heap IDs
+        pParams.heapIDBuffer[currentID] = 4 * baseHeapID;
+        pParams.heapIDBuffer[siblingID0] = 2 * baseHeapID + 1;
+        pParams.heapIDBuffer[siblingID1] = 4 * baseHeapID + 1;
+
+        uint3 modifiedNeighbors;
+        modifiedNeighbors[0] = siblingID1;
+        modifiedNeighbors[1] = res0X;
+        modifiedNeighbors[2] = siblingID0;
+        pParams.neighboursOutputBuffer[currentID] = uint4(modifiedNeighbors, 0);
+        modifiedNeighbors[0] = res1Y;
+        modifiedNeighbors[1] = currentID;
+        modifiedNeighbors[2] = p_n1;
+        pParams.neighboursOutputBuffer[siblingID0] = uint4(modifiedNeighbors, 0);
+        modifiedNeighbors[0] = res0Y;
+        modifiedNeighbors[1] = currentID;
+        modifiedNeighbors[2] = res1X;
+        pParams.neighboursOutputBuffer[siblingID1] = uint4(modifiedNeighbors, 0);
+
+        // Keep track of the parent
+        BisectorData modifiedBisector = cBisectorData;
+        modifiedBisector.propagationID = currentID;
+
+        // Lower the element down the tree and update it's sibling
+        modifiedBisector.problematicNeighbor = INVALID_POINTER;
+        modifiedBisector.flags = (VISIBLE_BISECTOR | MODIFIED_BISECTOR);
+        pParams.bisectorDataBuffer[currentID] = modifiedBisector;
+
+        // Create the sibling of the current element
+        modifiedBisector.problematicNeighbor = p_n1;
+        modifiedBisector.flags = (VISIBLE_BISECTOR | MODIFIED_BISECTOR);
+        pParams.bisectorDataBuffer[siblingID0] = modifiedBisector;
+
+        // Create the sibling of the current element
+        modifiedBisector.problematicNeighbor = INVALID_POINTER;
+        modifiedBisector.flags = (VISIBLE_BISECTOR | MODIFIED_BISECTOR);
+        pParams.bisectorDataBuffer[siblingID1] = modifiedBisector;
+
+        // Mark this for propagation
+        uint targetLocation = 0;
+        InterlockedAdd(pParams.propagateBuffer[0], 1, targetLocation);
+        pParams.propagateBuffer[2 + targetLocation] = siblingID0;
+    }
+    else if (currentSubdiv == LEFT_DOUBLE_SPLIT)
+    {
+        // Grab the bisector of the twin
+        uint res0X = INVALID_POINTER, res0Y = INVALID_POINTER;
+        evaluate_neighbors(currentID, p_n1, res0X, res0Y);
+
+        uint res1X = INVALID_POINTER, res1Y = INVALID_POINTER;
+        if (p_n2 != INVALID_POINTER)
+            evaluate_neighbors(currentID, p_n2, res1X, res1Y);
+
+        // Set the heap IDs
+        pParams.heapIDBuffer[currentID] = 2 * baseHeapID;
+        pParams.heapIDBuffer[siblingID0] = 4 * baseHeapID + 2;
+        pParams.heapIDBuffer[siblingID1] = 4 * baseHeapID + 3;
+
+        uint3 modifiedNeighbors;
+        modifiedNeighbors[0] = siblingID1;
+        modifiedNeighbors[1] = res1X;
+        modifiedNeighbors[2] = p_n0;
+        pParams.neighboursOutputBuffer[currentID] = uint4(modifiedNeighbors, 0);
+        modifiedNeighbors[0] = siblingID1;
+        modifiedNeighbors[1] = res0X;
+        modifiedNeighbors[2] = res1Y;
+        pParams.neighboursOutputBuffer[siblingID0] = uint4(modifiedNeighbors, 0);
+        modifiedNeighbors[0] = res0Y;
+        modifiedNeighbors[1] = siblingID0;
+        modifiedNeighbors[2] = currentID;
+        pParams.neighboursOutputBuffer[siblingID1] = uint4(modifiedNeighbors, 0);
+
+        // Keep track of the parent
+        BisectorData modifiedBisector = cBisectorData;
+        modifiedBisector.propagationID = currentID;
+
+        // Lower the element down the tree and update it's sibling
+        modifiedBisector.problematicNeighbor = INVALID_POINTER;
+        modifiedBisector.flags = (VISIBLE_BISECTOR | MODIFIED_BISECTOR);
+        pParams.bisectorDataBuffer[currentID] = modifiedBisector;
+
+        // Create the sibling of the current element
+        modifiedBisector.problematicNeighbor = INVALID_POINTER;
+        modifiedBisector.flags = (VISIBLE_BISECTOR | MODIFIED_BISECTOR);
+        pParams.bisectorDataBuffer[siblingID0] = modifiedBisector;
+
+        // Create the sibling of the current element
+        modifiedBisector.problematicNeighbor = INVALID_POINTER;
+        modifiedBisector.flags = (VISIBLE_BISECTOR | MODIFIED_BISECTOR);
+        pParams.bisectorDataBuffer[siblingID1] = modifiedBisector;
+    }
+    else if (currentSubdiv == TRIPLE_SPLIT)
+    {
+        // Grab the bisector of the twin
+        uint res0X = INVALID_POINTER, res0Y = INVALID_POINTER;
+        evaluate_neighbors(currentID, p_n0, res0X, res0Y);
+
+        uint res1X = INVALID_POINTER, res1Y = INVALID_POINTER;
+        evaluate_neighbors(currentID, p_n1, res1X, res1Y);
+
+        uint res2X = INVALID_POINTER, res2Y = INVALID_POINTER;
+        if (p_n2 != INVALID_POINTER)
+            evaluate_neighbors(currentID, p_n2, res2X, res2Y);
+
+        // Set the heap IDs
+        pParams.heapIDBuffer[currentID] = 4 * baseHeapID;
+        pParams.heapIDBuffer[siblingID0] = 4 * baseHeapID + 2;
+        pParams.heapIDBuffer[siblingID1] = 4 * baseHeapID + 1;
+        pParams.heapIDBuffer[siblingID2] = 4 * baseHeapID + 3;
+
+        uint3 modifiedNeighbors;
+        modifiedNeighbors[0] = siblingID1;
+        modifiedNeighbors[1] = res0X;
+        modifiedNeighbors[2] = siblingID2;
+        pParams.neighboursOutputBuffer[currentID] = uint4(modifiedNeighbors, 0);
+        modifiedNeighbors[0] = siblingID2;
+        modifiedNeighbors[1] = res1X;
+        modifiedNeighbors[2] = res2Y;
+        pParams.neighboursOutputBuffer[siblingID0] = uint4(modifiedNeighbors, 0);
+        modifiedNeighbors[0] = res0Y;
+        modifiedNeighbors[1] = currentID;
+        modifiedNeighbors[2] = res2X;
+        pParams.neighboursOutputBuffer[siblingID1] = uint4(modifiedNeighbors, 0);
+        modifiedNeighbors[0] = res1Y;
+        modifiedNeighbors[1] = siblingID0;
+        modifiedNeighbors[2] = currentID;
+        pParams.neighboursOutputBuffer[siblingID2] = uint4(modifiedNeighbors, 0);
+
+        // Keep track of the parent
+        BisectorData modifiedBisector = cBisectorData;
+        modifiedBisector.propagationID = currentID;
+
+        // Lower the element down the tree and update it's sibling
+        modifiedBisector.problematicNeighbor = INVALID_POINTER;
+        modifiedBisector.flags = (VISIBLE_BISECTOR | MODIFIED_BISECTOR);
+        pParams.bisectorDataBuffer[currentID] = modifiedBisector;
+
+        // Create the sibling of the current element
+        modifiedBisector.problematicNeighbor = INVALID_POINTER;
+        modifiedBisector.flags = (VISIBLE_BISECTOR | MODIFIED_BISECTOR);
+        pParams.bisectorDataBuffer[siblingID0] = modifiedBisector;
+
+        // Create the sibling of the current element
+        modifiedBisector.problematicNeighbor = INVALID_POINTER;
+        modifiedBisector.flags = (VISIBLE_BISECTOR | MODIFIED_BISECTOR);
+        pParams.bisectorDataBuffer[siblingID1] = modifiedBisector;
+
+        // Create the sibling of the current element
+        modifiedBisector.problematicNeighbor = INVALID_POINTER;
+        modifiedBisector.flags = (VISIBLE_BISECTOR | MODIFIED_BISECTOR);
+        pParams.bisectorDataBuffer[siblingID2] = modifiedBisector;
+    }
+
+    // How many bits do we need to raise
+    uint numSiblings = countbits(currentSubdiv);
+    for (uint siblingIdx = 0; siblingIdx < numSiblings; ++siblingIdx)
+    {
+        set_bit_atomic_buffer(cBisectorData.indices[siblingIdx], true);
+    }
+}
+
+void PropagateBisectElement(uint currentID)
+{
+    // Load the bisector data of the target triangle
+    BisectorData cBisectorData = pParams.bisectorDataBuffer[currentID];
+
+    // neighbors of the parent
+    uint parentID = cBisectorData.propagationID;
+    uint problematicNeighbor = cBisectorData.problematicNeighbor;
+
+    // Read the neighbor that may have changed
+    BisectorData tBisectorData = pParams.bisectorDataBuffer[problematicNeighbor];
+    uint3 tNeighbors = pParams.neighboursBuffer[problematicNeighbor].xyz;
+    uint targetID = problematicNeighbor;
+    uint sibling1 = tBisectorData.indices[1];
+
+    if (tBisectorData.subdivisionPattern == NO_SPLIT)
+    {
+        if (tNeighbors[0] == parentID)
+            pParams.neighboursBuffer[targetID][0] = currentID;
+        if (tNeighbors[1] == parentID)
+            pParams.neighboursBuffer[targetID][1] = currentID;
+        if (tNeighbors[2] == parentID)
+            pParams.neighboursBuffer[targetID][2] = currentID;
+    }
+    else if (tBisectorData.subdivisionPattern == CENTER_SPLIT)
+    {
+        if (pParams.neighboursBuffer[targetID][2] == parentID)
+            pParams.neighboursBuffer[targetID][2] = currentID;
+        if (pParams.neighboursBuffer[tBisectorData.propagationID][2] == parentID)
+            pParams.neighboursBuffer[tBisectorData.propagationID][2] = currentID;
+    }
+    else if (tBisectorData.subdivisionPattern == RIGHT_DOUBLE_SPLIT)
+    {
+        pParams.neighboursBuffer[sibling1][2] = currentID;
+    } 
+    else if (tBisectorData.subdivisionPattern == LEFT_DOUBLE_SPLIT)
+    {
+        pParams.neighboursBuffer[targetID][2] = currentID;
+    }
+
+    // Reset the problematic neighbor and the bisection state
+    pParams.bisectorDataBuffer[currentID].problematicNeighbor = INVALID_POINTER;
+    pParams.bisectorDataBuffer[currentID].bisectorState = UNCHANGED_ELEMENT;
+}
+
+void PrepareSimplifyElement(uint currentID)
+{
+    // Get the bisector
+    BisectorData cBisectorData = pParams.bisectorDataBuffer[currentID];
+
+    // Grab the current bisector
+    uint64_t cHeapID = pParams.heapIDBuffer[currentID];
+
+    // If this is not an even heap number it will be handeled by it's pair, the twin or the twin's pair
+
+    // Neighbors of this element
+    uint3 cNeighbors = pParams.neighboursBuffer[currentID].xyz;
+
+    // Evaluate the depth of this bisector
+    uint currentDepth = HeapIDDepth(cHeapID);
+
+    // Grab the pair neighbor (it has to exist)
+    uint pairID = cNeighbors[0];
+    uint64_t pHeapID = pParams.heapIDBuffer[pairID];
+    BisectorData pBisectorData = pParams.bisectorDataBuffer[pairID];
+    uint3 pNeighbors = pParams.neighboursBuffer[pairID].xyz;
+
+    // Evaluate the depth of the pair
+    uint pairDepth = HeapIDDepth(pHeapID);
+
+    // If they are not at the same depth or the pair is not to be simplified, we're done
+    if (pairDepth != currentDepth || pBisectorData.bisectorState != SIMPLIFY_ELEMENT)
+        return;
+
+    // We need to identify our twin pair
+    uint twinLowID = pNeighbors[0];
+    uint twinHighID = cNeighbors[1];
+    if (twinLowID != INVALID_POINTER)
+    {
+        // Grab the two bisectors
+        uint64_t twinLowHeapID = pParams.heapIDBuffer[twinLowID];
+        uint64_t twinHighHeapID = pParams.heapIDBuffer[twinHighID];
+        
+        // The current bisector is not the smallest element of the neighborhood, he will be handeled by twinLowBisect if needed
+        if (cHeapID > twinLowHeapID)
+            return;
+
+        // Compute the depth of both neighbors
+        uint lowFacingDepth = HeapIDDepth(twinLowHeapID);
+        uint highFacingDepth = HeapIDDepth(twinHighHeapID);
+
+        // If all four elements are not on the same
+        if (lowFacingDepth != currentDepth || highFacingDepth != currentDepth)
+            return;
+
+        // Grab the two bisectors
+        BisectorData twinLowBisectData = pParams.bisectorDataBuffer[twinLowID];
+        BisectorData twinHighBisectData = pParams.bisectorDataBuffer[twinHighID];
+
+        // This element should not be doing the simplifications if:
+        // - One of the four elements doesn't have the same depth
+        // - One of the four elements isn't flagged for simplification
+        if (twinLowBisectData.bisectorState != SIMPLIFY_ELEMENT
+            || twinHighBisectData.bisectorState != SIMPLIFY_ELEMENT)
+            return;
+    }
+
+    // This element will simplify itself, it's pair and possibilty it's twin and twin-pair.
+    uint bisectorSlot;
+    InterlockedAdd(pParams.simplifyBuffer[0], 1, bisectorSlot);
+
+    // Log the bisector ID
+    pParams.simplifyBuffer[1 + bisectorSlot] = currentID;
+}
+
+void SimplifyElement(uint currentID)
+{
+    // Grab the current bisector
+    BisectorData cBisectorData = pParams.bisectorDataBuffer[currentID];
+    uint3 cNeighbors = pParams.neighboursBuffer[currentID].xyz;
+
+    // Grab the pair neighbor (it has to exist)
+    uint pairID = cNeighbors[0];
+    BisectorData pBisectorData = pParams.bisectorDataBuffer[pairID];
+    uint3 pNeighbors = pParams.neighboursBuffer[pairID].xyz;
+
+    // We need to indentify our twin pair
+    uint twinLowID = pNeighbors[0];
+    uint twinHighID = cNeighbors[1];
+
+    // Set the heap IDs
+    pParams.heapIDBuffer[currentID] = pParams.heapIDBuffer[currentID] / 2;
+    pParams.heapIDBuffer[pairID] = 0;
+
+    // All conditions are met for us to simplify these triangles
+    uint3 newNeighbors;
+    newNeighbors[0] = cNeighbors[2];
+    newNeighbors[1] = pNeighbors[2];
+    newNeighbors[2] = twinLowID;
+    pParams.neighboursBuffer[currentID] = uint4(newNeighbors, 0);
+
+    // Update the bisector data
+    cBisectorData.propagationID = pairID;
+    cBisectorData.problematicNeighbor = pNeighbors[2];
+    cBisectorData.bisectorState = MERGED_ELEMENT;
+    cBisectorData.flags = (VISIBLE_BISECTOR | MODIFIED_BISECTOR);
+    pParams.bisectorDataBuffer[currentID] = cBisectorData;
+
+    // Mark this for propagation
+    if (cBisectorData.problematicNeighbor != INVALID_POINTER)
+    {
+        // Mark this for propagation
+        uint targetLocation = 0;
+        InterlockedAdd(pParams.propagateBuffer[1], 1, targetLocation);
+        pParams.propagateBuffer[2 + targetLocation] = currentID;
+    }
+
+    // Clear the pair's heap for identification
+    pBisectorData.bisectorState = MERGED_ELEMENT;
+    pBisectorData.flags = 0;
+    pParams.bisectorDataBuffer[pairID] = pBisectorData;
+
+    // Don't forget to free the bit
+    set_bit_atomic_buffer(pairID, false);
+
+    // If there was a facing pair, simplify it aswell
+    if (twinLowID != INVALID_POINTER)
+    {
+        // Set the heap IDs
+        pParams.heapIDBuffer[twinLowID] = pParams.heapIDBuffer[twinLowID] / 2;
+        pParams.heapIDBuffer[twinHighID] = 0;
+
+        // Read both bisectors
+        BisectorData lowFacingBst = pParams.bisectorDataBuffer[twinLowID];
+        uint3 lfNeighbors = pParams.neighboursBuffer[twinLowID].xyz;
+        BisectorData highFacingBst = pParams.bisectorDataBuffer[twinHighID];
+        uint3 hfNeighbors = pParams.neighboursBuffer[twinHighID].xyz;
+
+        // Update the lowest ID
+        newNeighbors[0] = lfNeighbors[2];
+        newNeighbors[1] = hfNeighbors[2];
+        newNeighbors[2] = currentID;
+        pParams.neighboursBuffer[twinLowID] = uint4(newNeighbors, 0);
+
+        // Update the twin bisector data
+        lowFacingBst.propagationID = twinHighID;
+        lowFacingBst.problematicNeighbor = hfNeighbors[2];
+        lowFacingBst.bisectorState = MERGED_ELEMENT;
+        lowFacingBst.flags = (VISIBLE_BISECTOR | MODIFIED_BISECTOR);
+        pParams.bisectorDataBuffer[twinLowID] = lowFacingBst;
+
+        if (lowFacingBst.problematicNeighbor != INVALID_POINTER)
+        {
+            // Mark this for propagation
+            uint targetLocation = 0;
+            InterlockedAdd(pParams.propagateBuffer[1], 1, targetLocation);
+            pParams.propagateBuffer[2 + targetLocation] = twinLowID;
+        }
+
+        // Clear the pair's heap for identification 
+        highFacingBst.bisectorState = MERGED_ELEMENT;
+        highFacingBst.flags = 0;
+        pParams.bisectorDataBuffer[twinHighID] = highFacingBst;
+
+        // Don't forget to free the bit
+        set_bit_atomic_buffer(twinHighID, false);
+    }
+}
+
+void PropagateElementSimplify(uint currentID)
+{
+    // Load the bisector data of the target element
+    BisectorData cBisectorData = pParams.bisectorDataBuffer[currentID];
+
+    // Id of the element before the simplification
+    uint deletedPair = cBisectorData.propagationID;
+
+    // neighbors of the parent
+    uint neighborID = cBisectorData.problematicNeighbor;
+
+    // Read the neighbor that may have changed
+    BisectorData nBisectorData = pParams.bisectorDataBuffer[neighborID];
+    uint3 nNeighbors = pParams.neighboursBuffer[neighborID].xyz;
+
+    // The neighbor has not changed, so we just need to make it point on currentID instead of the pair that was deleted
+    if (nBisectorData.bisectorState != MERGED_ELEMENT)
+    {
+        for (uint i = 0; i < 3; ++i)
+        {
+            if (nNeighbors[i] == deletedPair)
+                pParams.neighboursBuffer[neighborID][i] = currentID;
+        }
+    }
+    // The neighbor has had a simplification, so we need to update a different neighbor based on if it went up one depth in the tree or was deleted.
+    else if (nBisectorData.bisectorState == MERGED_ELEMENT)
+    {
+        // He still exist, but was simplified
+        if (pParams.heapIDBuffer[neighborID] != 0)
+        {
+            for (uint i = 0; i < 3; ++i)
+            {
+                if (nNeighbors[i] == deletedPair)
+                    pParams.neighboursBuffer[neighborID][i] = currentID;
+            }
+        }
+        // He is gone, we need to update his pair instead of him.
+        else
+        {
+            uint neighborPair = nNeighbors[1];
+            for (uint i = 0; i < 3; ++i)
+            {
+                if (pParams.neighboursBuffer[neighborPair][i] == deletedPair)
+                    pParams.neighboursBuffer[neighborPair][i] = currentID;
+            }
+        }
+    }
+
+    // Reset the problematic neighbor
+    pParams.bisectorDataBuffer[currentID].problematicNeighbor = INVALID_POINTER;
+}
+
+void BisectorElementIndexation(uint currentID)
+{
+    // Grab the current heap ID
+    uint64_t cHeapID = pParams.heapIDBuffer[currentID];
+
+    // Deallocated element, we don't care
+    if (cHeapID == 0)
+        return;
+
+    // Reserve a slot for this bisector
+    uint bisectorSlot;
+    InterlockedAdd(pParams.indirectDrawBuffer[0], 3, bisectorSlot);
+
+    // Keep track of it's global ID
+    pParams.bisectorIndicesBuffer[bisectorSlot / 3] = currentID;
+
+    // Load the bisector data of the target element
+    BisectorData cBisectorData = pParams.bisectorDataBuffer[currentID];
+    
+    // Is it visible?
+    if ((cBisectorData.flags & VISIBLE_BISECTOR) == 0)
+        return;
+
+    // Reserve a slot for this visible bisector
+    InterlockedAdd(pParams.indirectDrawBuffer[4], 3, bisectorSlot);
+
+    // Keep track of it's global ID
+    pParams.visibleBisectorIndices[bisectorSlot / 3] = currentID;
+
+    // Is it visible?
+    if ((cBisectorData.flags & MODIFIED_BISECTOR) == 0)
+        return;
+
+    // Reserve a slot for this visible bisector
+    InterlockedAdd(pParams.indirectDrawBuffer[8], 4, bisectorSlot);
+
+    // Keep track of it's global ID
+    pParams.modifiedBisectorIndices[bisectorSlot / 4] = currentID;
+}
+
+void ValidateBisector(uint currentID)
+{
+    // Grab the current heap ID
+    uint64_t cHeapID = pParams.heapIDBuffer[currentID];
+
+    // Deallocated element, we don't care
+    if (cHeapID == 0)
+        return;
+
+    // Load the bisector data of the target element
+    uint3 cNeighbors = pParams.neighboursBuffer[currentID].xyz;
+
+    bool failed = false;
+    uint targetNeighbor = INVALID_POINTER;
+    uint targetIdx = INVALID_POINTER;
+    for (uint i = 0; i < 3; ++i)
+    {
+        uint neighborID = cNeighbors[i];
+        if (neighborID != INVALID_POINTER)
+        {
+            bool found = false;
+            uint3 nNeighbors = pParams.neighboursBuffer[neighborID].xyz;
+            for (uint j = 0; j < 3; ++j)
+            {
+                if (nNeighbors[j] == currentID)
+                    found = true;
+            }
+            if (!found)
+            {
+                failed = true;
+                targetNeighbor = neighborID;
+                targetIdx = i;
+                break;
+            }
+        }
+    }
+    
+    // Notify the failure
+    if (failed)
+    {
+        uint prevValue;
+        InterlockedAdd(pParams.validationBuffer[0], 1, prevValue);
+    }    
+}

src/Engine/Graphics/CBT/CBT.cpp

@@ -281,27 +281,27 @@ void MeshUpdater::init(Gfx::PGraphics gfx, Gfx::PDescriptorLayout viewParamsLayo
 
     graphics->beginShaderCompilation(ShaderCompilationInfo{
         .name = "CBTCompute",
-        .modules = {"CBTCompute"},
+        .modules = {"CBTCompute", "LEB"},
         .entryPoints =
             {
-                {"reset", "CBTCompute"},
+                {"Reset", "CBTCompute"},
-                {"classify", "CBTCompute"},
+                {"Classify", "CBTCompute"},
-                {"split", "CBTCompute"},
+                {"Split", "CBTCompute"},
-                {"prepareIndirect", "CBTCompute"},
+                {"PrepareIndirect", "CBTCompute"},
-                {"allocate", "CBTCompute"},
+                {"Allocate", "CBTCompute"},
-                {"bisect", "CBTCompute"},
+                {"Bisect", "CBTCompute"},
-                {"propagateBisect", "CBTCompute"},
+                {"PropagateBisect", "CBTCompute"},
-                {"prepareSimplify", "CBTCompute"},
+                {"PrepareSimplify", "CBTCompute"},
-                {"simplify", "CBTCompute"},
+                {"Simplify", "CBTCompute"},
-                {"propagateSimplify", "CBTCompute"},
+                {"PropagateSimplify", "CBTCompute"},
-                {"reducePrePass", "CBTCompute"},
+                {"ReducePrePass", "CBTCompute"},
-                {"reduceFirstPass", "CBTCompute"},
+                {"ReduceFirstPass", "CBTCompute"},
-                {"reduceSecondPass", "CBTCompute"},
+                {"ReduceSecondPass", "CBTCompute"},
-                {"bisectorIndexation", "CBTCompute"},
+                {"BisectorIndexation", "CBTCompute"},
-                {"prepareBisectorIndirect", "CBTCompute"},
+                {"PrepareBisectorIndirect", "CBTCompute"},
-                {"validate", "CBTCompute"},
+                {"Validate", "CBTCompute"},
-                {"clearLeb", "CBTCompute"},
+                {"ClearBuffer", "LEB"},
-                {"evaluateLeb", "CBTCompute"},
+                {"EvaluateLEB", "LEB"},
             },
         .rootSignature = pipelineLayout,
     });
@@ -412,9 +412,11 @@ void MeshUpdater::evaluateLeb(const BaseMesh& baseMesh, CBTMesh& mesh, Gfx::PDes
         clearCmd->bindPipeline(lebClear);
         clearCmd->bindDescriptor(set);
         clearCmd->dispatch((mesh.totalNumElements * 3 + WORKGROUP_SIZE - 1) / WORKGROUP_SIZE, 1, 1);
+        graphics->executeCommands(std::move(clearCmd));
         mesh.currentVertexBuffer->pipelineBarrier(Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                                                   Gfx::SE_ACCESS_SHADER_READ_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
     }
+    graphics->waitDeviceIdle();
     Gfx::PDescriptorSet set = layout->allocateDescriptorSet();
     set->updateBuffer(GEOMETRY_CB, 0, geometryBuffer);
     set->updateBuffer(UPDATE_CB, 0, updateBuffer);
@@ -432,7 +434,7 @@ void MeshUpdater::evaluateLeb(const BaseMesh& baseMesh, CBTMesh& mesh, Gfx::PDes
     evalCmd->pushConstants(Gfx::SE_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(uint32), &val);
     evalCmd->dispatchIndirect(mesh.indirectDispatchBuffer, complete ? 0 : sizeof(uint32) * 6);
     graphics->executeCommands(std::move(evalCmd));
-    mesh.currentVertexBuffer->pipelineBarrier(Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
+    mesh.lebVertexBuffer->pipelineBarrier(Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                                               Gfx::SE_ACCESS_SHADER_READ_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
 }
 
@@ -456,7 +458,6 @@ void MeshUpdater::update(CBTMesh& mesh, Gfx::PDescriptorSet viewParamsSet, Gfx::
         set->updateBuffer(HEAP_ID_BUFFER, 0, mesh.heapIDBuffer);
         set->updateBuffer(BISECTOR_DATA_BUFFER, 0, mesh.updateBuffer);
         set->updateBuffer(CLASSIFICATION_BUFFER, 0, mesh.classificationBuffer);
-        set->updateBuffer(DEBUG_BUFFER, 0, debugBuffer);
         set->writeChanges();
         Gfx::OComputeCommand classifyCmd = graphics->createComputeCommand("Classify");
         classifyCmd->bindPipeline(classify);
@@ -482,6 +483,7 @@ void MeshUpdater::update(CBTMesh& mesh, Gfx::PDescriptorSet viewParamsSet, Gfx::
                                         Gfx::SE_ACCESS_SHADER_READ_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
     }
 
+    graphics->waitDeviceIdle();
     // split pass
     {
         Gfx::PDescriptorSet set = layout->allocateDescriptorSet();
@@ -494,6 +496,7 @@ void MeshUpdater::update(CBTMesh& mesh, Gfx::PDescriptorSet viewParamsSet, Gfx::
         set->updateBuffer(NEIGHBOURS_BUFFER, 0, currentNeighborsBuffer);
         set->updateBuffer(MEMORY_BUFFER, 0, memoryBuffer);
         set->updateBuffer(ALLOCATE_BUFFER, 0, mesh.allocateBuffer);
+        set->updateBuffer(DEBUG_BUFFER, 0, debugBuffer);
         set->writeChanges();
         Gfx::OComputeCommand splitCmd = graphics->createComputeCommand("Split");
         splitCmd->bindPipeline(split);
@@ -502,7 +505,10 @@ void MeshUpdater::update(CBTMesh& mesh, Gfx::PDescriptorSet viewParamsSet, Gfx::
         graphics->executeCommands(std::move(splitCmd));
         mesh.updateBuffer->pipelineBarrier(Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                                            Gfx::SE_ACCESS_SHADER_READ_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
+        mesh.allocateBuffer->pipelineBarrier(Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
+                                             Gfx::SE_ACCESS_SHADER_READ_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
     }
+    graphics->waitDeviceIdle();
 
     // prepare indirect pass
     {
@@ -518,6 +524,7 @@ void MeshUpdater::update(CBTMesh& mesh, Gfx::PDescriptorSet viewParamsSet, Gfx::
         indirectBuffer->pipelineBarrier(Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                                         Gfx::SE_ACCESS_SHADER_READ_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
     }
+    graphics->waitDeviceIdle();
 
     // Allocate Pass
     {
@@ -539,6 +546,7 @@ void MeshUpdater::update(CBTMesh& mesh, Gfx::PDescriptorSet viewParamsSet, Gfx::
         mesh.updateBuffer->pipelineBarrier(Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                                            Gfx::SE_ACCESS_SHADER_READ_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
     }
+    graphics->waitDeviceIdle();
 
     // copy
     currentNeighborsBuffer->pipelineBarrier(Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
@@ -546,6 +554,7 @@ void MeshUpdater::update(CBTMesh& mesh, Gfx::PDescriptorSet viewParamsSet, Gfx::
     graphics->copyBuffer(currentNeighborsBuffer, nextNeighborsBuffer);
     nextNeighborsBuffer->pipelineBarrier(Gfx::SE_ACCESS_TRANSFER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_TRANSFER_BIT,
                                          Gfx::SE_ACCESS_SHADER_READ_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
+    graphics->waitDeviceIdle();
 
     // bisect
     {
@@ -570,6 +579,7 @@ void MeshUpdater::update(CBTMesh& mesh, Gfx::PDescriptorSet viewParamsSet, Gfx::
         nextNeighborsBuffer->pipelineBarrier(Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                                              Gfx::SE_ACCESS_SHADER_READ_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
     }
+    graphics->waitDeviceIdle();
 
     // Prepare indirect pass
     {
@@ -585,6 +595,7 @@ void MeshUpdater::update(CBTMesh& mesh, Gfx::PDescriptorSet viewParamsSet, Gfx::
         indirectBuffer->pipelineBarrier(Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                                         Gfx::SE_ACCESS_SHADER_READ_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
     }
+    graphics->waitDeviceIdle();
 
     // propagate split pass
     {
@@ -604,6 +615,7 @@ void MeshUpdater::update(CBTMesh& mesh, Gfx::PDescriptorSet viewParamsSet, Gfx::
         mesh.updateBuffer->pipelineBarrier(Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                                            Gfx::SE_ACCESS_SHADER_READ_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
     }
+    graphics->waitDeviceIdle();
 
     // prepare simplify
     {
@@ -625,6 +637,7 @@ void MeshUpdater::update(CBTMesh& mesh, Gfx::PDescriptorSet viewParamsSet, Gfx::
         mesh.simplificationBuffer->pipelineBarrier(Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                                                    Gfx::SE_ACCESS_SHADER_READ_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
     }
+    graphics->waitDeviceIdle();
 
     // Prepare Indirect Simplify
     {
@@ -640,6 +653,7 @@ void MeshUpdater::update(CBTMesh& mesh, Gfx::PDescriptorSet viewParamsSet, Gfx::
         indirectBuffer->pipelineBarrier(Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                                         Gfx::SE_ACCESS_SHADER_READ_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
     }
+    graphics->waitDeviceIdle();
 
     // Simplify Pass
     {
@@ -663,6 +677,7 @@ void MeshUpdater::update(CBTMesh& mesh, Gfx::PDescriptorSet viewParamsSet, Gfx::
         nextNeighborsBuffer->pipelineBarrier(Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                                              Gfx::SE_ACCESS_SHADER_READ_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
     }
+    graphics->waitDeviceIdle();
 
     // Prepare Indirect Propagate Simplify
     {
@@ -678,6 +693,7 @@ void MeshUpdater::update(CBTMesh& mesh, Gfx::PDescriptorSet viewParamsSet, Gfx::
         indirectBuffer->pipelineBarrier(Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                                         Gfx::SE_ACCESS_SHADER_READ_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
     }
+    graphics->waitDeviceIdle();
 
     // Propagate Simplify Pass
     {
@@ -698,6 +714,7 @@ void MeshUpdater::update(CBTMesh& mesh, Gfx::PDescriptorSet viewParamsSet, Gfx::
         nextNeighborsBuffer->pipelineBarrier(Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                                              Gfx::SE_ACCESS_SHADER_READ_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
     }
+    graphics->waitDeviceIdle();
 
     // Update Tree
     {
@@ -730,10 +747,12 @@ void MeshUpdater::update(CBTMesh& mesh, Gfx::PDescriptorSet viewParamsSet, Gfx::
         mesh.gpuCBT.bufferArray[0]->pipelineBarrier(Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                                                     Gfx::SE_ACCESS_SHADER_READ_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
     }
+    graphics->waitDeviceIdle();
 
     mesh.currentNeighborsBufferIdx = nextNeighborsBufferIdx;
 
     prepareIndirection(mesh, geometryCB);
+    graphics->waitDeviceIdle();
 }
 
 void MeshUpdater::validation(const CBTMesh& mesh, Gfx::PUniformBuffer geometryCB) {

src/Engine/Graphics/CBT/CBT.h

@@ -327,6 +327,7 @@ struct GeometryCB {
     uint32 totalNumVertices;
 };
 struct UpdateCB {
+    Matrix4 viewProjectionMatrix;
     float triangleSize;
     uint32_t maxSubdivisionDepth;
     float fov;
@@ -355,10 +356,10 @@ constexpr int64 BISECT_ELEMENT = 1;
 constexpr int64 SIMPLIFY_ELEMENT = 2;
 constexpr int64 MERGED_ELEMENT = 3;
 struct BisectorData {
-    uint32 subdivisionPattern;
-
     UVector indices;
 
+    uint32 subdivisionPattern;
+    
     uint32 problematicNeighbor;
 
     uint32 bisectorState;

src/Engine/Graphics/Initializer.h

@@ -97,7 +97,6 @@ struct ShaderBufferCreateInfo {
     DataSource sourceData = DataSource();
     uint64 numElements = 1;
     uint32 clearValue = 0;
-    uint8 createCleared = 0;
     Gfx::SeBufferUsageFlags usage = 0;
     std::string name = "Unnamed";
 };

src/Engine/Graphics/RenderPass/BasePass.cpp

@@ -99,7 +99,7 @@ void BasePass::beginFrame(const Component::Camera& cam) {
     transparentCulling = lightCullingLayout->allocateDescriptorSet();
 
     //waterRenderer->beginFrame();
-    terrainRenderer->beginFrame(viewParamsSet);
+    terrainRenderer->beginFrame(viewParamsSet, cam);
 
     // Debug vertices
     {

src/Engine/Graphics/RenderPass/TerrainRenderer.cpp

@@ -10,11 +10,19 @@ using namespace Seele;
 TerrainRenderer::TerrainRenderer(Gfx::PGraphics graphics, PScene scene, Gfx::PDescriptorLayout viewParamsLayout,
                                  Gfx::PDescriptorSet viewParamsSet)
     : graphics(graphics), scene(scene) {
+    //Gfx::OPipelineLayout test = graphics->createPipelineLayout();
+    //graphics->beginShaderCompilation(ShaderCompilationInfo{
+    //    .modules = {"CompileTest"},
+    //    .entryPoints = {{"Reset", "CompileTest"}},
+    //    .rootSignature = test,
+    //});
+    //graphics->createComputeShader({0});
     meshUpdater.init(graphics, viewParamsLayout);
     lebCache.init(graphics, 5);
     CBT<18> cbt;
     CPUMesh cpuMesh = generateCPUMesh(cbt.numElements());
     plainMesh.gpuCBT.lastLevelSize = cbt.lastLevelSize();
+    plainMesh.gpuCBT.bufferCount = 2;
     for (uint32 i = 0; i < 2; ++i) {
         uint32 bufferSize = cbt.bufferSize(i);
         uint32 elementSize = cbt.elementSize(i);
@@ -74,7 +82,6 @@ TerrainRenderer::TerrainRenderer(Gfx::PGraphics graphics, PScene scene, Gfx::PDe
             {
                 .size = sizeof(BisectorData) * cpuMesh.totalNumElements,
             },
-        .usage = Gfx::SE_BUFFER_USAGE_INDIRECT_BUFFER_BIT,
         .name = "UpdateBuffer",
     });
     plainMesh.classificationBuffer = graphics->createShaderBuffer(ShaderBufferCreateInfo{
@@ -151,6 +158,8 @@ TerrainRenderer::TerrainRenderer(Gfx::PGraphics graphics, PScene scene, Gfx::PDe
             },
         .name = "LebVertexBuffer",
     });
+    plainMesh.lebVertexBuffer->pipelineBarrier(Gfx::SE_ACCESS_TRANSFER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_TRANSFER_BIT,
+                                               Gfx::SE_ACCESS_SHADER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
     plainMesh.currentVertexBuffer = graphics->createShaderBuffer(ShaderBufferCreateInfo{
         .sourceData =
             {
@@ -213,6 +222,7 @@ TerrainRenderer::TerrainRenderer(Gfx::PGraphics graphics, PScene scene, Gfx::PDe
     geometryBuffer->pipelineBarrier(Gfx::SE_ACCESS_TRANSFER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_TRANSFER_BIT, Gfx::SE_ACCESS_UNIFORM_READ_BIT,
                                     Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
     UpdateCB updateCB = {
+        .viewProjectionMatrix = Matrix4(0),
         .triangleSize = 10.0f,
         .maxSubdivisionDepth = 63,
         .fov = 70.0f,
@@ -234,7 +244,7 @@ TerrainRenderer::TerrainRenderer(Gfx::PGraphics graphics, PScene scene, Gfx::PDe
             {
                 {"vert", "TerrainPass"},
                 {"frag", "TerrainPass"},
-                {"deform", "TerrainPass"},
+                {"EvaluateDeformation", "TerrainPass"},
             },
         .rootSignature = meshUpdater.pipelineLayout,
     });
@@ -245,20 +255,47 @@ TerrainRenderer::TerrainRenderer(Gfx::PGraphics graphics, PScene scene, Gfx::PDe
         .computeShader = deformCS,
         .pipelineLayout = meshUpdater.pipelineLayout,
     });
+    graphics->waitDeviceIdle();
+
     meshUpdater.resetBuffers(plainMesh);
+    graphics->waitDeviceIdle();
     meshUpdater.prepareIndirection(plainMesh, geometryBuffer);
+    graphics->waitDeviceIdle();
     meshUpdater.evaluateLeb(baseMesh, plainMesh, viewParamsSet, geometryBuffer, updateBuffer, lebCache.getLebMatrixBuffer(), true, true);
+    graphics->waitDeviceIdle();
     applyDeformation(viewParamsSet);
+    graphics->waitDeviceIdle();
 }
 
 TerrainRenderer::~TerrainRenderer() {}
 
 static bool first = true;
 
-void TerrainRenderer::beginFrame(Gfx::PDescriptorSet viewParamsSet) {
+void TerrainRenderer::beginFrame(Gfx::PDescriptorSet viewParamsSet, const Component::Camera& cam) {
+    UpdateCB updateCB = {
+        .viewProjectionMatrix = viewport->getProjectionMatrix() * cam.getViewMatrix(),
+        .triangleSize = 10.0f,
+        .maxSubdivisionDepth = 63,
+        .fov = 70.0f,
+        .farPlaneDistance = 1000.0f,
+    };
+    updateBuffer = graphics->createUniformBuffer(UniformBufferCreateInfo{
+        .sourceData =
+            {
+                .size = sizeof(UpdateCB),
+                .data = (uint8*)&updateCB,
+            },
+        .name = "UpdateCB",
+    });
+    updateBuffer->pipelineBarrier(Gfx::SE_ACCESS_TRANSFER_WRITE_BIT, Gfx::SE_PIPELINE_STAGE_TRANSFER_BIT, Gfx::SE_ACCESS_UNIFORM_READ_BIT,
+                                  Gfx::SE_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
+    graphics->waitDeviceIdle();
     meshUpdater.update(plainMesh, viewParamsSet, geometryBuffer, updateBuffer);
+    graphics->waitDeviceIdle();
     meshUpdater.evaluateLeb(baseMesh, plainMesh, viewParamsSet, geometryBuffer, updateBuffer, lebCache.getLebMatrixBuffer(), false, false);
+    graphics->waitDeviceIdle();
     applyDeformation(viewParamsSet);
+    graphics->waitDeviceIdle();
 }
 
 Gfx::ORenderCommand TerrainRenderer::render(Gfx::PDescriptorSet viewParamsSet) {

src/Engine/Graphics/RenderPass/TerrainRenderer.h

@@ -8,7 +8,7 @@ class TerrainRenderer {
   public:
     TerrainRenderer(Gfx::PGraphics graphics, PScene scene, Gfx::PDescriptorLayout viewParamsLayout, Gfx::PDescriptorSet viewParamsSet);
     ~TerrainRenderer();
-    void beginFrame(Gfx::PDescriptorSet viewParamsSet);
+    void beginFrame(Gfx::PDescriptorSet viewParamsSet, const Component::Camera& cam);
     Gfx::ORenderCommand render(Gfx::PDescriptorSet viewParamsSet);
     void setViewport(Gfx::PViewport viewport, Gfx::PRenderPass renderPass);
 

src/Engine/Graphics/RenderPass/VisibilityPass.cpp

@@ -80,7 +80,6 @@ void VisibilityPass::publishOutputs() {
 
     cullingBuffer = graphics->createShaderBuffer(ShaderBufferCreateInfo{
         .clearValue = 0xffffffff,
-        .createCleared = true,
         .name = "CullingBuffer",
     });
     resources->registerBufferOutput("CULLINGBUFFER", cullingBuffer);

src/Engine/Graphics/Vulkan/Buffer.cpp

@@ -314,7 +314,7 @@ void Buffer::createBuffer(uint64 size, uint32 destIndex) {
         command->bindResource(PBufferAllocation(buffers[destIndex]));
         vkCmdFillBuffer(command->getHandle(), buffers[destIndex]->buffer, 0, VK_WHOLE_SIZE, clearValue);
         pipelineBarrier(VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
-                        VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT);
+                        VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT);
     }
 }
 
@@ -417,7 +417,7 @@ void UniformBuffer::executePipelineBarrier(VkAccessFlags srcAccess, VkPipelineSt
 ShaderBuffer::ShaderBuffer(PGraphics graphics, const ShaderBufferCreateInfo& createInfo)
     : Gfx::ShaderBuffer(graphics->getFamilyMapping(), createInfo),
       Vulkan::Buffer(graphics, createInfo.sourceData.size, VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | createInfo.usage,
-                     createInfo.sourceData.owner, true, createInfo.name, createInfo.createCleared, createInfo.clearValue) {
+                     createInfo.sourceData.owner, true, createInfo.name, createInfo.sourceData.data == nullptr, createInfo.clearValue) {
     if (createInfo.sourceData.size > 0 && createInfo.sourceData.data != nullptr) {
         getAlloc()->updateContents(createInfo.sourceData.offset, createInfo.sourceData.size, createInfo.sourceData.data);
     }

src/Engine/Graphics/slang-compile.cpp

@@ -18,7 +18,6 @@ using namespace Seele;
     {                                                                                                                                      \
         if (diagnostics) {                                                                                                                 \
             std::cout << (const char*)diagnostics->getBufferPointer() << std::endl;                                                        \
-            assert(false);                                                                                                                 \
         }                                                                                                                                  \
     }
 
@@ -35,15 +34,15 @@ void Seele::beginCompilation(const ShaderCompilationInfo& info, SlangCompileTarg
     sessionDesc.flags = 0;
     Array<slang::CompilerOptionEntry> option = {
         {
-            .name = slang::CompilerOptionName::IgnoreCapabilities,
+            .name = slang::CompilerOptionName::Capability,
             .value =
                 {
                     .kind = slang::CompilerOptionValueKind::Int,
-                    .intValue0 = 1,
+                    .intValue0 = globalSession->findCapability("GLSL_450"),
                 },
         },
         {
-            .name = slang::CompilerOptionName::EmitSpirvViaGLSL,
+            .name = slang::CompilerOptionName::EmitSpirvDirectly,
             .value =
                 {
                     .kind = slang::CompilerOptionValueKind::Int,
@@ -89,7 +88,7 @@ void Seele::beginCompilation(const ShaderCompilationInfo& info, SlangCompileTarg
     sessionDesc.preprocessorMacroCount = macros.size();
     sessionDesc.preprocessorMacros = macros.data();
     slang::TargetDesc targetDesc;
-    targetDesc.profile = globalSession->findProfile("GLSL_450");
+    targetDesc.profile = globalSession->findProfile("glsl_450");
     targetDesc.format = target;
     sessionDesc.targetCount = 1;
     sessionDesc.targets = &targetDesc;