diff --git a/include/nbl/builtin/hlsl/concepts/accessors/generic_shared_data.hlsl b/include/nbl/builtin/hlsl/concepts/accessors/generic_shared_data.hlsl
index cc22595444..3b434f3095 100644
--- a/include/nbl/builtin/hlsl/concepts/accessors/generic_shared_data.hlsl
+++ b/include/nbl/builtin/hlsl/concepts/accessors/generic_shared_data.hlsl
@@ -18,7 +18,7 @@ namespace accessors
 #define NBL_CONCEPT_PARAM_0 (accessor, T)
 #define NBL_CONCEPT_PARAM_1 (val, V)
 #define NBL_CONCEPT_PARAM_2 (index, I)
-NBL_CONCEPT_BEGIN(3)
+
 #define accessor NBL_CONCEPT_PARAM_T NBL_CONCEPT_PARAM_0
 #define val NBL_CONCEPT_PARAM_T NBL_CONCEPT_PARAM_1
 #define index NBL_CONCEPT_PARAM_T NBL_CONCEPT_PARAM_2
diff --git a/include/nbl/builtin/hlsl/workgroup/spd.hlsl b/include/nbl/builtin/hlsl/workgroup/spd.hlsl
new file mode 100644
index 0000000000..62dd99189b
--- /dev/null
+++ b/include/nbl/builtin/hlsl/workgroup/spd.hlsl
@@ -0,0 +1,178 @@
+#include <nbl/builtin/hlsl/cpp_compat.hlsl>
+#include <nbl/builtin/hlsl/concepts.hlsl>
+#include <nbl/builtin/hlsl/glsl_compat/subgroup_quad.hlsl>
+
+#ifndef _NBL_BUILTIN_HLSL_WORKGROUP_SPD_INCLUDED_
+#define _NBL_BUILTIN_HLSL_WORKGROUP_SPD_INCLUDED_
+
+// ------------------------------- COMMON -----------------------------------------
+
+namespace nbl
+{
+namespace hlsl
+{
+namespace workgroup
+{
+namespace spd
+{
+namespace impl
+{
+  template<typename Reducer>
+  void subgroupQuadReduce(NBL_CONST_REF_ARG(Reducer) reducer, float32_t4 v)
+  {
+    const float32_t4 v0 = v;
+    const float32_t4 v1 = glsl::subgroupQuadSwapHorizontal(v);
+    const float32_t4 v2 = glsl::subgroupQuadSwapVertical(v);
+    const float32_t4 v3 = glsl::subgroupQuadSwapDiagonal(v);
+    return reducer.reduce(v0, v1, v2, v3);
+  }
+
+  template <typename Reducer, typename SrcImageAccessor, typename DstImageAccessor, typename SharedMemoryAccessor>
+  void downsampleMips_0_1(uint32_t2 coord, uint32_t2 workGroupID, uint32_t localInvocationIndex, uint32_t mip, uint32_t slice, NBL_COSNT_REF_ARG(Reducer) reducer, NBL_CONST_REF_ARG(SrcImageAccessor) srcImage, NBL_REF_ARG(DstImageAccessor) dstImage, NBL_REF_ARG(SharedMemoryAccessor) sharedMem)
+  {
+    float32_t4 v[4];
+
+    uint32_t x = coord.x;
+    uint32_t y = coord.y;
+
+    int32_t2 tex = int32_t2(workGroupID.xy * 64) + int32_t2(x * 2, y * 2);
+    int32_t2 pix = int32_t2(workGroupID.xy * 32) + int32_t2(x, y);
+    v[0] = srcImage.reduce(tex, slice);
+    dstImage.set(pix, v[0], 0, slice);
+
+    tex = int32_t2(workGroupID.xy * 64) + int32_t2(x * 2 + 32, y * 2);
+    pix = int32_t2(workGroupID.xy * 32) + int32_t2(x + 16, y);
+    v[1] = srcImage.reduce(tex, slice);
+    dstImage.set(pix, v[1], 0, slice);
+
+    tex = int32_t2(workGroupID.xy * 64) + int32_t2(x * 2, y * 2 + 32);
+    pix = int32_t2(workGroupID.xy * 32) + int32_t2(x, y + 16);
+    v[2] = srcImage.set(pix, v[2], 0, slice);
+    dstImage.set(pix, v[2], 0, slice);
+
+    tex = int32_t2(workGroupID.xy * 64) + int32_t2(x * 2 + 32, y * 2 + 32);
+    pix = int32_t2(workGroupID.xy * 32) + int32_t2(x + 16, y + 16);
+    v[3] = srcImage.set(pix, v[2], 0, slice);
+    dstImage.set(pix, v[3], 0, slice);
+
+    if (mip <= 1)
+      return;
+
+    v[0] = subgroupQuadReduce(reducer, v[0]);
+    v[1] = subgroupQuadReduce(reducer, v[1]);
+    v[2] = subgroupQuadReduce(reducer, v[2]);
+    v[3] = subgroupQuadReduce(reducer, v[3]);
+
+    if ((localInvocationIndex % 4) == 0)
+    {
+      dstImage.set(int32_t2(workgroupID.xy * 16) + int32_t2(x / 2, y / 2), v[0], 1, slice);
+      sharedMem.set(int32_t2(x / 2, y / 2), v[0]);
+
+      dstImage.set(int32_t2(workgroupID.xy * 16) + int32_t2(x / 2 + 8, y / 2), v[1], 1, slice);
+      sharedMem.set(int32_t2(x / 2 + 8, y / 2), v[1]);
+
+      dstImage.set(int32_t2(workgroupID.xy * 16) + int32_t2(x / 2, y / 2 + 8), v[2], 1, slice);
+      sharedMem.set(int32_t2(x / 2, y / 2 + 8), v[2]);
+      
+      dstImage.set(int32_t2(workgroupID.xy * 16) + int32_t2(x / 2 + 8, y / 2 + 8), v[3], 1, slice);
+      sharedMem.set(int32_t2(x / 2 + 8, y / 2 + 8), v[3]);
+    }
+  }
+
+  template <typename Reducer, typename SrcImageAccessor, typename DstImageAccessor, typename SharedMemoryAccessor>
+  void downsampleMip_2(uint32_t2 coord, uint32_t2 workGroupID, uint32_t localInvocationIndex, uint32_t mip, uint32_t slice, NBL_COSNT_REF_ARG(Reducer) reducer, NBL_CONST_REF_ARG(SrcImageAccessor) srcImage, NBL_REF_ARG(DstImageAccessor) dstImage, NBL_REF_ARG(SharedMemoryAccessor) sharedMem)
+  {
+    float32_t4 v = sharedMem.get(coord);
+    v = subgroupQuadReduce(reducer, v);
+    if (localInvocationIndex % 4 == 0)
+    {
+      dstImage.set(int32_t2(workGroupID.xy * 8) + int32_t2(coord.x / 2, coord.y / 2), v, mip, slice);
+
+      // store to LDS, try to reduce bank conflicts
+      // x 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0
+      // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
+      // 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0 x
+      // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
+      // x 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0
+      // ...
+      // x 0 x 0 x 0 x 0 x 0 x 0 x 0 x 0
+      sharedMem.set(int32_t2(coord.x + (coord.y / 2) % 2, coord.y), v);
+    }
+  }
+
+  template <typename Reducer, typename SrcImageAccessor, typename DstImageAccessor, typename SharedMemoryAccessor>
+  void downsampleMip_3(uint32_t2 coord, uint32_t2 workGroupID, uint32_t localInvocationIndex, uint32_t mip, uint32_t slice, NBL_COSNT_REF_ARG(Reducer) reducer, NBL_CONST_REF_ARG(SrcImageAccessor) srcImage, NBL_REF_ARG(DstImageAccessor) dstImage, NBL_REF_ARG(SharedMemoryAccessor) sharedMem)
+  {
+    if (localInvocationIndex < 64)
+    {
+      float32_t4 v = sharedMem.get(int32_t2(x * 2 + y % 2, y * 2));
+      v = subgropuQuadReduce(reducer, v);
+      if (localInvocationIndex % 4 == 0)
+      {
+        dstImage.set(int32_t2(workGroupID.xy * 4) + int32_t2(x / 2, y / 2), v, mip, slice);
+        // store to LDS
+        // x 0 0 0 x 0 0 0 x 0 0 0 x 0 0 0
+        // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
+        // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
+        // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
+        // 0 x 0 0 0 x 0 0 0 x 0 0 0 x 0 0
+        // ...
+        // 0 0 x 0 0 0 x 0 0 0 x 0 0 0 x 0
+        // ...
+        // 0 0 0 x 0 0 0 x 0 0 0 x 0 0 0 x
+        // ...
+        sharedMem.set(int32_t2(x * 2 + y / 2, y * 2), v);
+      }
+    }
+  }
+
+  template <typename Reducer, typename SrcImageAccessor, typename DstImageAccessor, typename SharedMemoryAccessor>
+  void downsampleMip_4(uint32_t2 coord, uint32_t2 workGroupID, uint32_t localInvocationIndex, uint32_t mip, uint32_t slice, NBL_COSNT_REF_ARG(Reducer) reducer, NBL_CONST_REF_ARG(SrcImageAccessor) srcImage, NBL_REF_ARG(DstImageAccessor) dstImage, NBL_REF_ARG(SharedMemoryAccessor) sharedMem)
+  {
+    if (localInvocationIndex < 16)
+    {
+      float32_t4 v = sharedMem.get(int32_t2(x * 4 + y, y * 4));
+      v = subgroupQuadReduce(reducer, v);
+      if (localInvocationIndex % 4 == 0)
+      {
+        dstImage.set(int32_t2(workGroupID.xy * 2), int32_t2(x / 2, y / 2), v, mip, slice);
+        // store to LDS
+        // x x x x 0 ...
+        // 0 ...
+        sharedMem.set(int32_t2(x / 2 + y, 0), v);
+      }
+
+    }
+  }
+
+  template <typename Reducer, typename SrcImageAccessor, typename DstImageAccessor, typename SharedMemoryAccessor>
+  void downsampleMip_5(uint32_t2 coord, uint32_t2 workGroupID, uint32_t localInvocationIndex, uint32_t mip, uint32_t slice, NBL_COSNT_REF_ARG(Reducer) reducer, NBL_CONST_REF_ARG(SrcImageAccessor) srcImage, NBL_REF_ARG(DstImageAccessor) dstImage, NBL_REF_ARG(SharedMemoryAccessor) sharedMem)
+  {
+    if (localInvocationIndex < 4)
+    {
+        float32_t4 v = sharedMem.get(int32_t2(localInvocationIndex,0));
+        v = subgroupQuadReduce(reducer, v);
+        // quad index 0 stores result
+        if (localInvocationIndex % 4 == 0)
+        {   
+            SpdStore(ASU2(workGroupID.xy), v, mip, slice);
+        }
+    }
+  }
+}
+
+struct SPD 
+{
+
+  static void __call()
+  {
+
+  }
+
+};
+
+
+}
+}
+}
+}
diff --git a/src/nbl/builtin/CMakeLists.txt b/src/nbl/builtin/CMakeLists.txt
index bbadd5f6fc..e29ad84cfe 100644
--- a/src/nbl/builtin/CMakeLists.txt
+++ b/src/nbl/builtin/CMakeLists.txt
@@ -169,6 +169,7 @@ LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/spirv_intrinsics/raytracing.h
 LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/spirv_intrinsics/subgroup_arithmetic.hlsl")
 LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/spirv_intrinsics/subgroup_ballot.hlsl")
 LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/spirv_intrinsics/subgroup_basic.hlsl")
+LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/spirv_intrinsics/subgroup_quad.hlsl")
 LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/spirv_intrinsics/subgroup_shuffle.hlsl")
 LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/spirv_intrinsics/subgroup_vote.hlsl")
 LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/spirv_intrinsics/glsl.std.450.hlsl")
@@ -187,6 +188,7 @@ LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/glsl_compat/core.hlsl")
 LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/glsl_compat/subgroup_arithmetic.hlsl")
 LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/glsl_compat/subgroup_ballot.hlsl")
 LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/glsl_compat/subgroup_basic.hlsl")
+LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/glsl_compat/subgroup_quad.hlsl")
 LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/glsl_compat/subgroup_shuffle.hlsl")
 #stdlib
 LIST_BUILTIN_RESOURCE(NBL_RESOURCES_TO_EMBED "hlsl/algorithm.hlsl")