Add test script to ensure no leaks in PC queue

- Still quite far from a state where it can be merged
- Reduce 5x5 to 3x3
- Reduce 3x3 to 2x2

Open issues:

- [ ] Can we generalize it? 
- [ ] Which reductions are needed
- [ ] Naming

CMakeLists.txt

@@ -368,6 +368,7 @@ set(PUBLICHEADERS
 
 
 set(SourceFiles
+    ${CMAKE_CURRENT_SOURCE_DIR}/src/calibration.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/src/CtbRawFile.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/src/decode.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/src/defs.cpp
@@ -437,6 +438,7 @@ endif()
 if(AARE_TESTS)
     set(TestSources
             ${CMAKE_CURRENT_SOURCE_DIR}/src/algorithm.test.cpp
+            ${CMAKE_CURRENT_SOURCE_DIR}/src/calibration.test.cpp
             ${CMAKE_CURRENT_SOURCE_DIR}/src/defs.test.cpp
             ${CMAKE_CURRENT_SOURCE_DIR}/src/decode.test.cpp
             ${CMAKE_CURRENT_SOURCE_DIR}/src/Dtype.test.cpp

RELEASE.md

@@ -1,7 +1,22 @@
 # Release notes
 
 
-### 2025.07.18
+### 2025.8.22
+
+Features:
+
+- Apply calibration works in G0 if passes a 2D calibration and pedestal
+- count pixels that switch
+- calculate pedestal (also g0 version)
+- NDArray::view() needs an lvalue to reduce issues with the view outliving the array
+
+
+Bugfixes:
+
+- Now using glibc 2.17 in conda builds (was using the host)
+- Fixed shifted pixels in clusters close to the edge of a frame
+
+### 2025.7.18
 
 Features:
 
@@ -15,7 +30,7 @@ Bugfixes:
 - Removed unused file: ClusterFile.cpp 
 
 
-### 2025.05.22
+### 2025.5.22
 
 Features:
 

VERSION

@@ -1 +1 @@
-2025.7.18
+2025.8.22

benchmarks/CMakeLists.txt

@@ -15,7 +15,7 @@ FetchContent_MakeAvailable(benchmark)
 
 add_executable(benchmarks)
 
-target_sources(benchmarks PRIVATE ndarray_benchmark.cpp calculateeta_benchmark.cpp)
+target_sources(benchmarks PRIVATE ndarray_benchmark.cpp calculateeta_benchmark.cpp reduce_benchmark.cpp)
 
 # Link Google Benchmark and other necessary libraries
 target_link_libraries(benchmarks PRIVATE benchmark::benchmark aare_core aare_compiler_flags)

benchmarks/reduce_benchmark.cpp

@@ -0,0 +1,168 @@
+#include "aare/Cluster.hpp"
+#include <benchmark/benchmark.h>
+
+using namespace aare;
+
+class ClustersForReduceFixture : public benchmark::Fixture {
+  public:
+    Cluster<int, 5, 5> cluster_5x5{};
+    Cluster<int, 3, 3> cluster_3x3{};
+
+  private:
+    using benchmark::Fixture::SetUp;
+
+    void SetUp([[maybe_unused]] const benchmark::State &state) override {
+        int temp_data[25] = {1, 1, 1, 1, 1, 1, 1, 2, 1, 1,
+                             1, 2, 3, 1, 2, 1, 1, 1, 1, 2};
+        std::copy(std::begin(temp_data), std::end(temp_data),
+                  std::begin(cluster_5x5.data));
+
+        cluster_5x5.x = 5;
+        cluster_5x5.y = 5;
+
+        int temp_data2[9] = {1, 1, 1, 2, 3, 1, 2, 2, 1};
+        std::copy(std::begin(temp_data2), std::end(temp_data2),
+                  std::begin(cluster_3x3.data));
+
+        cluster_3x3.x = 5;
+        cluster_3x3.y = 5;
+    }
+
+    // void TearDown(::benchmark::State& state) {
+    // }
+};
+
+template <typename T>
+Cluster<T, 3, 3, int16_t> reduce_to_3x3(const Cluster<T, 5, 5, int16_t> &c) {
+    Cluster<T, 3, 3, int16_t> result;
+
+    // Write out the sums in the hope that the compiler can optimize this
+    std::array<T, 9> sum_3x3_subclusters;
+
+    // Write out the sums in the hope that the compiler can optimize this
+    sum_3x3_subclusters[0] = c.data[0] + c.data[1] + c.data[2] + c.data[5] +
+                             c.data[6] + c.data[7] + c.data[10] + c.data[11] +
+                             c.data[12];
+    sum_3x3_subclusters[1] = c.data[1] + c.data[2] + c.data[3] + c.data[6] +
+                             c.data[7] + c.data[8] + c.data[11] + c.data[12] +
+                             c.data[13];
+    sum_3x3_subclusters[2] = c.data[2] + c.data[3] + c.data[4] + c.data[7] +
+                             c.data[8] + c.data[9] + c.data[12] + c.data[13] +
+                             c.data[14];
+    sum_3x3_subclusters[3] = c.data[5] + c.data[6] + c.data[7] + c.data[10] +
+                             c.data[11] + c.data[12] + c.data[15] + c.data[16] +
+                             c.data[17];
+    sum_3x3_subclusters[4] = c.data[6] + c.data[7] + c.data[8] + c.data[11] +
+                             c.data[12] + c.data[13] + c.data[16] + c.data[17] +
+                             c.data[18];
+    sum_3x3_subclusters[5] = c.data[7] + c.data[8] + c.data[9] + c.data[12] +
+                             c.data[13] + c.data[14] + c.data[17] + c.data[18] +
+                             c.data[19];
+    sum_3x3_subclusters[6] = c.data[10] + c.data[11] + c.data[12] + c.data[15] +
+                             c.data[16] + c.data[17] + c.data[20] + c.data[21] +
+                             c.data[22];
+    sum_3x3_subclusters[7] = c.data[11] + c.data[12] + c.data[13] + c.data[16] +
+                             c.data[17] + c.data[18] + c.data[21] + c.data[22] +
+                             c.data[23];
+    sum_3x3_subclusters[8] = c.data[12] + c.data[13] + c.data[14] + c.data[17] +
+                             c.data[18] + c.data[19] + c.data[22] + c.data[23] +
+                             c.data[24];
+
+    auto index = std::max_element(sum_3x3_subclusters.begin(),
+                                  sum_3x3_subclusters.end()) -
+                 sum_3x3_subclusters.begin();
+
+    switch (index) {
+    case 0:
+        result.x = c.x - 1;
+        result.y = c.y + 1;
+        result.data = {c.data[0], c.data[1],  c.data[2],  c.data[5], c.data[6],
+                       c.data[7], c.data[10], c.data[11], c.data[12]};
+        break;
+    case 1:
+        result.x = c.x;
+        result.y = c.y + 1;
+        result.data = {c.data[1], c.data[2],  c.data[3],  c.data[6], c.data[7],
+                       c.data[8], c.data[11], c.data[12], c.data[13]};
+        break;
+    case 2:
+        result.x = c.x + 1;
+        result.y = c.y + 1;
+        result.data = {c.data[2], c.data[3],  c.data[4],  c.data[7], c.data[8],
+                       c.data[9], c.data[12], c.data[13], c.data[14]};
+        break;
+    case 3:
+        result.x = c.x - 1;
+        result.y = c.y;
+        result.data = {c.data[5],  c.data[6],  c.data[7],
+                       c.data[10], c.data[11], c.data[12],
+                       c.data[15], c.data[16], c.data[17]};
+        break;
+    case 4:
+        result.x = c.x + 1;
+        result.y = c.y;
+        result.data = {c.data[6],  c.data[7],  c.data[8],
+                       c.data[11], c.data[12], c.data[13],
+                       c.data[16], c.data[17], c.data[18]};
+        break;
+    case 5:
+        result.x = c.x + 1;
+        result.y = c.y;
+        result.data = {c.data[7],  c.data[8],  c.data[9],
+                       c.data[12], c.data[13], c.data[14],
+                       c.data[17], c.data[18], c.data[19]};
+        break;
+    case 6:
+        result.x = c.x + 1;
+        result.y = c.y - 1;
+        result.data = {c.data[10], c.data[11], c.data[12],
+                       c.data[15], c.data[16], c.data[17],
+                       c.data[20], c.data[21], c.data[22]};
+        break;
+    case 7:
+        result.x = c.x + 1;
+        result.y = c.y - 1;
+        result.data = {c.data[11], c.data[12], c.data[13],
+                       c.data[16], c.data[17], c.data[18],
+                       c.data[21], c.data[22], c.data[23]};
+        break;
+    case 8:
+        result.x = c.x + 1;
+        result.y = c.y - 1;
+        result.data = {c.data[12], c.data[13], c.data[14],
+                       c.data[17], c.data[18], c.data[19],
+                       c.data[22], c.data[23], c.data[24]};
+        break;
+    }
+    return result;
+}
+
+BENCHMARK_F(ClustersForReduceFixture, Reduce2x2)(benchmark::State &st) {
+    for (auto _ : st) {
+        // This code gets timed
+        benchmark::DoNotOptimize(reduce_to_2x2<int, 3, 3, int16_t>(
+            cluster_3x3)); // make sure compiler evaluates the expression
+    }
+}
+
+BENCHMARK_F(ClustersForReduceFixture, SpecificReduce2x2)(benchmark::State &st) {
+    for (auto _ : st) {
+        // This code gets timed
+        benchmark::DoNotOptimize(reduce_to_2x2<int>(cluster_3x3));
+    }
+}
+
+BENCHMARK_F(ClustersForReduceFixture, Reduce3x3)(benchmark::State &st) {
+    for (auto _ : st) {
+        // This code gets timed
+        benchmark::DoNotOptimize(
+            reduce_to_3x3<int, 5, 5, int16_t>(cluster_5x5));
+    }
+}
+
+BENCHMARK_F(ClustersForReduceFixture, SpecificReduce3x3)(benchmark::State &st) {
+    for (auto _ : st) {
+        // This code gets timed
+        benchmark::DoNotOptimize(reduce_to_3x3<int>(cluster_5x5));
+    }
+}

conda-recipe/conda_build_config.yaml

@@ -3,3 +3,14 @@ python:
     - 3.12
     - 3.13
 
+c_compiler:
+  - gcc                        # [linux]
+
+c_stdlib:
+  - sysroot                    # [linux]
+
+cxx_compiler:
+  - gxx                        # [linux]
+
+c_stdlib_version:              # [linux]
+  - 2.17                       # [linux]

conda-recipe/meta.yaml

@@ -16,6 +16,8 @@ build:
 
 requirements:
   build:
+    - {{ compiler('c') }}
+    - {{ stdlib("c") }}
     - {{ compiler('cxx') }}
     - cmake
     - ninja

docs/src/ClusterVector.rst

@@ -12,4 +12,11 @@ ClusterVector
    :members:
    :undoc-members:
    :private-members:
+   
+
+**Free Functions:** 
+
+.. doxygenfunction:: aare::reduce_to_3x3(const ClusterVector<Cluster<T, ClusterSizeX, ClusterSizeY, CoordType>>&)
+   
+.. doxygenfunction:: aare::reduce_to_2x2(const ClusterVector<Cluster<T, ClusterSizeX, ClusterSizeY, CoordType>>&)
    

docs/src/pyClusterVector.rst

@@ -33,4 +33,17 @@ C++ functions that support the ClusterVector or to view it as a numpy array.
     :members:
     :undoc-members:
     :show-inheritance:
-    :inherited-members:
+    :inherited-members:
+
+
+**Free Functions:** 
+
+.. autofunction:: reduce_to_3x3
+   :noindex:
+
+   Reduce a single Cluster to 3x3 by taking the 3x3 subcluster with highest photon energy.
+
+.. autofunction:: reduce_to_2x2
+   :noindex:
+
+   Reduce a single Cluster to 2x2 by taking the 2x2 subcluster with highest photon energy.

docs/src/pycalibration.rst

@@ -17,8 +17,24 @@ Functions for applying calibration to data.
     # Apply calibration to raw data to convert from raw ADC values to keV
     data = aare.apply_calibration(raw_data, pd=pedestal, cal=calibration)
 
+    # If you pass a 2D pedestal and calibration only G0 will be used for the conversion
+    # Pixels that switched to G1 or G2 will be set to 0
+    data = aare.apply_calibration(raw_data, pd=pedestal[0], cal=calibration[0])
+
+    
+
 .. py:currentmodule:: aare
 
 .. autofunction:: apply_calibration
 
 .. autofunction:: load_calibration
+
+.. autofunction:: calculate_pedestal
+
+.. autofunction:: calculate_pedestal_float
+
+.. autofunction:: calculate_pedestal_g0
+
+.. autofunction:: calculate_pedestal_g0_float
+
+.. autofunction:: count_switching_pixels

include/aare/CalculateEta.hpp

@@ -28,7 +28,7 @@ enum class pixel : int {
 template <typename T> struct Eta2 {
     double x;
     double y;
-    int c;
+    int c{0};
     T sum;
 };
 
@@ -70,6 +70,8 @@ calculate_eta2(const Cluster<T, ClusterSizeX, ClusterSizeY, CoordType> &cl) {
     size_t index_bottom_left_max_2x2_subcluster =
         (int(c / (ClusterSizeX - 1))) * ClusterSizeX + c % (ClusterSizeX - 1);
 
+    // calculate direction of gradient
+
     // check that cluster center is in max subcluster
     if (cluster_center_index != index_bottom_left_max_2x2_subcluster &&
         cluster_center_index != index_bottom_left_max_2x2_subcluster + 1 &&
@@ -128,12 +130,15 @@ Eta2<T> calculate_eta2(const Cluster<T, 2, 2, int16_t> &cl) {
     Eta2<T> eta{};
 
     if ((cl.data[0] + cl.data[1]) != 0)
-        eta.x = static_cast<double>(cl.data[1]) / (cl.data[0] + cl.data[1]);
+        eta.x = static_cast<double>(cl.data[1]) /
+                (cl.data[0] + cl.data[1]); // between (0,1) the closer to zero
+                                           // left value probably larger
     if ((cl.data[0] + cl.data[2]) != 0)
-        eta.y = static_cast<double>(cl.data[2]) / (cl.data[0] + cl.data[2]);
+        eta.y = static_cast<double>(cl.data[2]) /
+                (cl.data[0] + cl.data[2]); // between (0,1) the closer to zero
+                                           // bottom value probably larger
     eta.sum = cl.sum();
-    eta.c = static_cast<int>(corner::cBottomLeft); // TODO! This is not correct,
-                                                   // but need to put something
+
     return eta;
 }
 
@@ -150,13 +155,11 @@ template <typename T> Eta2<T> calculate_eta3(const Cluster<T, 3, 3> &cl) {
 
     eta.sum = sum;
 
-    eta.c = corner::cBottomLeft;
-
     if ((cl.data[3] + cl.data[4] + cl.data[5]) != 0)
 
         eta.x = static_cast<double>(-cl.data[3] + cl.data[3 + 2]) /
 
-                (cl.data[3] + cl.data[4] + cl.data[5]);
+                (cl.data[3] + cl.data[4] + cl.data[5]); // (-1,1)
 
     if ((cl.data[1] + cl.data[4] + cl.data[7]) != 0)
 

include/aare/Cluster.hpp

@@ -8,6 +8,7 @@
 
 #pragma once
 
+#include "logger.hpp"
 #include <algorithm>
 #include <array>
 #include <cstdint>
@@ -74,6 +75,163 @@ struct Cluster {
     }
 };
 
+/**
+ * @brief Reduce a cluster to a 2x2 cluster by selecting the 2x2 block with the
+ * highest sum.
+ * @param c Cluster to reduce
+ * @return reduced cluster
+ */
+template <typename T, uint8_t ClusterSizeX, uint8_t ClusterSizeY,
+          typename CoordType = int16_t>
+Cluster<T, 2, 2, CoordType>
+reduce_to_2x2(const Cluster<T, ClusterSizeX, ClusterSizeY, CoordType> &c) {
+
+    static_assert(ClusterSizeX >= 2 && ClusterSizeY >= 2,
+                  "Cluster sizes must be at least 2x2 for reduction to 2x2");
+
+    // TODO maybe add sanity check and check that center is in max subcluster
+    Cluster<T, 2, 2, CoordType> result;
+
+    auto [sum, index] = c.max_sum_2x2();
+
+    int16_t cluster_center_index =
+        (ClusterSizeX / 2) + (ClusterSizeY / 2) * ClusterSizeX;
+
+    int16_t index_bottom_left_max_2x2_subcluster =
+        (int(index / (ClusterSizeX - 1))) * ClusterSizeX +
+        index % (ClusterSizeX - 1);
+
+    result.x =
+        c.x + (index_bottom_left_max_2x2_subcluster - cluster_center_index) %
+                  ClusterSizeX;
+
+    result.y =
+        c.y - (index_bottom_left_max_2x2_subcluster - cluster_center_index) /
+                  ClusterSizeX;
+    result.data = {
+        c.data[index_bottom_left_max_2x2_subcluster],
+        c.data[index_bottom_left_max_2x2_subcluster + 1],
+        c.data[index_bottom_left_max_2x2_subcluster + ClusterSizeX],
+        c.data[index_bottom_left_max_2x2_subcluster + ClusterSizeX + 1]};
+    return result;
+}
+
+template <typename T>
+Cluster<T, 2, 2, int16_t> reduce_to_2x2(const Cluster<T, 3, 3, int16_t> &c) {
+    Cluster<T, 2, 2, int16_t> result;
+
+    auto [s, i] = c.max_sum_2x2();
+    switch (i) {
+    case 0:
+        result.x = c.x - 1;
+        result.y = c.y + 1;
+        result.data = {c.data[0], c.data[1], c.data[3], c.data[4]};
+        break;
+    case 1:
+        result.x = c.x;
+        result.y = c.y + 1;
+        result.data = {c.data[1], c.data[2], c.data[4], c.data[5]};
+        break;
+    case 2:
+        result.x = c.x - 1;
+        result.y = c.y;
+        result.data = {c.data[3], c.data[4], c.data[6], c.data[7]};
+        break;
+    case 3:
+        result.x = c.x;
+        result.y = c.y;
+        result.data = {c.data[4], c.data[5], c.data[7], c.data[8]};
+        break;
+    }
+
+    return result;
+}
+
+template <typename T, uint8_t ClusterSizeX, uint8_t ClusterSizeY,
+          typename CoordType = int16_t>
+inline std::pair<T, uint16_t>
+max_3x3_sum(const Cluster<T, ClusterSizeX, ClusterSizeY, CoordType> &cluster) {
+
+    if constexpr (ClusterSizeX == 3 && ClusterSizeY == 3) {
+        return std::make_pair(cluster.sum(), 0);
+    } else {
+
+        size_t index = 0;
+        T max_3x3_subcluster_sum = 0;
+        for (size_t i = 0; i < ClusterSizeY - 2; ++i) {
+            for (size_t j = 0; j < ClusterSizeX - 2; ++j) {
+
+                T sum = cluster.data[i * ClusterSizeX + j] +
+                        cluster.data[i * ClusterSizeX + j + 1] +
+                        cluster.data[i * ClusterSizeX + j + 2] +
+                        cluster.data[(i + 1) * ClusterSizeX + j] +
+                        cluster.data[(i + 1) * ClusterSizeX + j + 1] +
+                        cluster.data[(i + 1) * ClusterSizeX + j + 2] +
+                        cluster.data[(i + 2) * ClusterSizeX + j] +
+                        cluster.data[(i + 2) * ClusterSizeX + j + 1] +
+                        cluster.data[(i + 2) * ClusterSizeX + j + 2];
+                if (sum > max_3x3_subcluster_sum) {
+                    max_3x3_subcluster_sum = sum;
+                    index = i * (ClusterSizeX - 2) + j;
+                }
+            }
+        }
+
+        return std::make_pair(max_3x3_subcluster_sum, index);
+    }
+}
+
+/**
+ * @brief Reduce a cluster to a 3x3 cluster by selecting the 3x3 block with the
+ * highest sum.
+ * @param c Cluster to reduce
+ * @return reduced cluster
+ */
+template <typename T, uint8_t ClusterSizeX, uint8_t ClusterSizeY,
+          typename CoordType = int16_t>
+Cluster<T, 3, 3, CoordType>
+reduce_to_3x3(const Cluster<T, ClusterSizeX, ClusterSizeY, CoordType> &c) {
+
+    static_assert(ClusterSizeX >= 3 && ClusterSizeY >= 3,
+                  "Cluster sizes must be at least 3x3 for reduction to 3x3");
+
+    Cluster<T, 3, 3, CoordType> result;
+
+    // TODO maybe add sanity check and check that center is in max subcluster
+
+    auto [sum, index] = max_3x3_sum(c);
+
+    int16_t cluster_center_index =
+        (ClusterSizeX / 2) + (ClusterSizeY / 2) * ClusterSizeX;
+
+    int16_t index_center_max_3x3_subcluster =
+        (int(index / (ClusterSizeX - 2))) * ClusterSizeX + ClusterSizeX +
+        index % (ClusterSizeX - 2) + 1;
+
+    int16_t index_3x3_subcluster_cluster_center =
+        int((cluster_center_index - 1 - ClusterSizeX) / ClusterSizeX) *
+            (ClusterSizeX - 2) +
+        (cluster_center_index - 1 - ClusterSizeX) % ClusterSizeX;
+
+    result.x =
+        c.x + (index % (ClusterSizeX - 2) -
+               (index_3x3_subcluster_cluster_center % (ClusterSizeX - 2)));
+    result.y =
+        c.y - (index / (ClusterSizeX - 2) -
+               (index_3x3_subcluster_cluster_center / (ClusterSizeX - 2)));
+
+    result.data = {c.data[index_center_max_3x3_subcluster - ClusterSizeX - 1],
+                   c.data[index_center_max_3x3_subcluster - ClusterSizeX],
+                   c.data[index_center_max_3x3_subcluster - ClusterSizeX + 1],
+                   c.data[index_center_max_3x3_subcluster - 1],
+                   c.data[index_center_max_3x3_subcluster],
+                   c.data[index_center_max_3x3_subcluster + 1],
+                   c.data[index_center_max_3x3_subcluster + ClusterSizeX - 1],
+                   c.data[index_center_max_3x3_subcluster + ClusterSizeX],
+                   c.data[index_center_max_3x3_subcluster + ClusterSizeX + 1]};
+    return result;
+}
+
 // Type Traits for is_cluster_type
 template <typename T>
 struct is_cluster : std::false_type {}; // Default case: Not a Cluster

include/aare/ClusterFinder.hpp

@@ -144,9 +144,9 @@ class ClusterFinder {
                                     static_cast<CT>(
                                         m_pedestal.mean(iy + ir, ix + ic));
                                 cluster.data[i] =
-                                    tmp; // Watch for out of bounds access
-                                i++;
+                                    tmp; // Watch for out of bounds access                                
                             }
+                            i++;
                         }
                     }
 

include/aare/ClusterVector.hpp

@@ -32,8 +32,7 @@ class ClusterVector; // Forward declaration
  */
 template <typename T, uint8_t ClusterSizeX, uint8_t ClusterSizeY,
           typename CoordType>
-class ClusterVector<Cluster<T, ClusterSizeX, ClusterSizeY, CoordType>> 
-{
+class ClusterVector<Cluster<T, ClusterSizeX, ClusterSizeY, CoordType>> {
 
     std::vector<Cluster<T, ClusterSizeX, ClusterSizeY, CoordType>> m_data{};
     int32_t m_frame_number{0}; // TODO! Check frame number size and type
@@ -173,4 +172,40 @@ class ClusterVector<Cluster<T, ClusterSizeX, ClusterSizeY, CoordType>>
     }
 };
 
+/**
+ * @brief Reduce a cluster to a 2x2 cluster by selecting the 2x2 block with the
+ * highest sum.
+ * @param cv Clustervector containing clusters to reduce
+ * @return Clustervector with reduced clusters
+ */
+template <typename T, uint8_t ClusterSizeX, uint8_t ClusterSizeY,
+          typename CoordType = uint16_t>
+ClusterVector<Cluster<T, 2, 2, CoordType>> reduce_to_2x2(
+    const ClusterVector<Cluster<T, ClusterSizeX, ClusterSizeY, CoordType>>
+        &cv) {
+    ClusterVector<Cluster<T, 2, 2, CoordType>> result;
+    for (const auto &c : cv) {
+        result.push_back(reduce_to_2x2(c));
+    }
+    return result;
+}
+
+/**
+ * @brief Reduce a cluster to a 3x3 cluster by selecting the 3x3 block with the
+ * highest sum.
+ * @param cv Clustervector containing clusters to reduce
+ * @return Clustervector with reduced clusters
+ */
+template <typename T, uint8_t ClusterSizeX, uint8_t ClusterSizeY,
+          typename CoordType = uint16_t>
+ClusterVector<Cluster<T, 3, 3, CoordType>> reduce_to_3x3(
+    const ClusterVector<Cluster<T, ClusterSizeX, ClusterSizeY, CoordType>>
+        &cv) {
+    ClusterVector<Cluster<T, 3, 3, CoordType>> result;
+    for (const auto &c : cv) {
+        result.push_back(reduce_to_3x3(c));
+    }
+    return result;
+}
+
 } // namespace aare

include/aare/Frame.hpp

@@ -105,7 +105,7 @@ class Frame {
      * @tparam T type of the pixels
      * @return NDView<T, 2>
      */
-    template <typename T> NDView<T, 2> view() {
+    template <typename T> NDView<T, 2> view() & {
         std::array<ssize_t, 2> shape = {static_cast<ssize_t>(m_rows),
                                         static_cast<ssize_t>(m_cols)};
         T *data = reinterpret_cast<T *>(m_data);

include/aare/NDArray.hpp

@@ -25,7 +25,7 @@ template <typename T, ssize_t Ndim = 2>
 class NDArray : public ArrayExpr<NDArray<T, Ndim>, Ndim> {
     std::array<ssize_t, Ndim> shape_;
     std::array<ssize_t, Ndim> strides_;
-    size_t size_{};
+    size_t size_{}; //TODO! do we need to store size when we have shape?
     T *data_;
 
   public:
@@ -33,7 +33,7 @@ class NDArray : public ArrayExpr<NDArray<T, Ndim>, Ndim> {
      * @brief Default constructor. Will construct an empty NDArray.
      *
      */
-    NDArray() : shape_(), strides_(c_strides<Ndim>(shape_)), data_(nullptr){};
+    NDArray() : shape_(), strides_(c_strides<Ndim>(shape_)), data_(nullptr) {};
 
     /**
      * @brief Construct a new NDArray object with a given shape.
@@ -43,8 +43,7 @@ class NDArray : public ArrayExpr<NDArray<T, Ndim>, Ndim> {
      */
     explicit NDArray(std::array<ssize_t, Ndim> shape)
         : shape_(shape), strides_(c_strides<Ndim>(shape_)),
-          size_(std::accumulate(shape_.begin(), shape_.end(), 1,
-                                std::multiplies<>())),
+          size_(num_elements(shape_)),
           data_(new T[size_]) {}
 
     /**
@@ -79,6 +78,24 @@ class NDArray : public ArrayExpr<NDArray<T, Ndim>, Ndim> {
         other.reset(); // TODO! is this necessary?
     }
 
+
+    //Move constructor from an an array with Ndim + 1
+    template <ssize_t M, typename = std::enable_if_t<(M == Ndim + 1)>>
+    NDArray(NDArray<T, M> &&other) 
+        : shape_(drop_first_dim(other.shape())),
+          strides_(c_strides<Ndim>(shape_)), size_(num_elements(shape_)),
+          data_(other.data()) {
+
+            // For now only allow move if the size matches, to avoid unreachable data
+            // if the use case arises we can remove this check
+            if(size() != other.size()) {
+                data_ = nullptr; // avoid double free, other will clean up the memory in it's destructor
+                throw std::runtime_error(LOCATION +
+                                         "Size mismatch in move constructor of NDArray<T, Ndim-1>");
+            }
+        other.reset();
+    }
+
     // Copy constructor
     NDArray(const NDArray &other)
         : shape_(other.shape_), strides_(c_strides<Ndim>(shape_)),
@@ -380,12 +397,6 @@ NDArray<T, Ndim> NDArray<T, Ndim>::operator*(const T &value) {
     result *= value;
     return result;
 }
-// template <typename T, ssize_t Ndim> void NDArray<T, Ndim>::Print() {
-//     if (shape_[0] < 20 && shape_[1] < 20)
-//         Print_all();
-//     else
-//         Print_some();
-// }
 
 template <typename T, ssize_t Ndim>
 std::ostream &operator<<(std::ostream &os, const NDArray<T, Ndim> &arr) {
@@ -437,4 +448,23 @@ NDArray<T, Ndim> load(const std::string &pathname,
     return img;
 }
 
+template <typename RT, typename NT, typename DT, ssize_t Ndim>
+NDArray<RT, Ndim> safe_divide(const NDArray<NT, Ndim> &numerator,
+                              const NDArray<DT, Ndim> &denominator) {
+    if (numerator.shape() != denominator.shape()) {
+        throw std::runtime_error(
+            "Shapes of numerator and denominator must match");
+    }
+    NDArray<RT, Ndim> result(numerator.shape());
+    for (ssize_t i = 0; i < numerator.size(); ++i) {
+        if (denominator[i] != 0) {
+            result[i] =
+                static_cast<RT>(numerator[i]) / static_cast<RT>(denominator[i]);
+        } else {
+            result[i] = RT{0}; // or handle division by zero as needed
+        }
+    }
+    return result;
+}
+
 } // namespace aare

include/aare/NDView.hpp

@@ -26,6 +26,33 @@ Shape<Ndim> make_shape(const std::vector<size_t> &shape) {
     return arr;
 }
 
+
+/**
+ * @brief Helper function to drop the first dimension of a shape.
+ * This is useful when you want to create a 2D view from a 3D array.
+ * @param shape The shape to drop the first dimension from.
+ * @return A new shape with the first dimension dropped.
+ */
+template<size_t Ndim>
+Shape<Ndim-1> drop_first_dim(const Shape<Ndim> &shape) {
+    static_assert(Ndim > 1, "Cannot drop first dimension from a 1D shape");
+    Shape<Ndim - 1> new_shape;
+    std::copy(shape.begin() + 1, shape.end(), new_shape.begin());
+    return new_shape;
+}
+
+/**
+ * @brief Helper function when constructing NDArray/NDView. Calculates the number
+ * of elements in the resulting array from a shape.
+ * @param shape The shape to calculate the number of elements for.
+ * @return The number of elements in and NDArray/NDView of that shape.
+ */
+template <size_t Ndim>
+size_t num_elements(const Shape<Ndim> &shape) {
+    return std::accumulate(shape.begin(), shape.end(), 1,
+                           std::multiplies<size_t>());
+}
+
 template <ssize_t Dim = 0, typename Strides>
 ssize_t element_offset(const Strides & /*unused*/) {
     return 0;
@@ -66,17 +93,28 @@ class NDView : public ArrayExpr<NDView<T, Ndim>, Ndim> {
         : buffer_(buffer), strides_(c_strides<Ndim>(shape)), shape_(shape),
           size_(std::accumulate(std::begin(shape), std::end(shape), 1,
                                 std::multiplies<>())) {}
-
+                   
     template <typename... Ix>
     std::enable_if_t<sizeof...(Ix) == Ndim, T &> operator()(Ix... index) {
         return buffer_[element_offset(strides_, index...)];
     }
 
     template <typename... Ix>
-    const std::enable_if_t<sizeof...(Ix) == Ndim, T &> operator()(Ix... index) const {
+    std::enable_if_t<sizeof...(Ix) == 1 && (Ndim > 1), NDView<T, Ndim - 1>> operator()(Ix... index) {
+        // return a view of the next dimension
+        std::array<ssize_t, Ndim - 1> new_shape{};
+        std::copy_n(shape_.begin() + 1, Ndim - 1, new_shape.begin());
+        return NDView<T, Ndim - 1>(&buffer_[element_offset(strides_, index...)],
+                                   new_shape);
+        
+    }
+
+    template <typename... Ix>
+    std::enable_if_t<sizeof...(Ix) == Ndim, const T &> operator()(Ix... index) const {
         return buffer_[element_offset(strides_, index...)];
     }
 
+
     ssize_t size() const { return static_cast<ssize_t>(size_); }
     size_t total_bytes() const { return size_ * sizeof(T); }
     std::array<ssize_t, Ndim> strides() const noexcept { return strides_; }
@@ -85,9 +123,19 @@ class NDView : public ArrayExpr<NDView<T, Ndim>, Ndim> {
     T *end() { return buffer_ + size_; }
     T const *begin() const { return buffer_; }
     T const *end() const { return buffer_ + size_; }
-    T &operator()(ssize_t i)  { return buffer_[i]; }
+    
+    
+
+
+
+    /**
+     * @brief Access element at index i.
+     */
     T &operator[](ssize_t i)  { return buffer_[i]; }
-    const T &operator()(ssize_t i) const { return buffer_[i]; }
+
+    /**
+     * @brief Access element at index i.
+     */
     const T &operator[](ssize_t i) const { return buffer_[i]; }
 
     bool operator==(const NDView &other) const {
@@ -157,6 +205,22 @@ class NDView : public ArrayExpr<NDView<T, Ndim>, Ndim> {
     const T *data() const { return buffer_; }
     void print_all() const;
 
+    /**
+     * @brief Create a subview of a range of the first dimension. 
+     * This is useful for splitting a batches of frames in parallel processing.
+     * @param first The first index of the subview (inclusive).
+     * @param last The last index of the subview (exclusive).
+     * @return A new NDView that is a subview of the current view.
+     * @throws std::runtime_error if the range is invalid.
+     */
+    NDView sub_view(ssize_t first, ssize_t last) const {
+        if (first < 0 || last > shape_[0] || first >= last)
+            throw std::runtime_error(LOCATION + "Invalid sub_view range");
+        auto new_shape = shape_;
+        new_shape[0] = last - first;
+        return NDView(buffer_ + first * strides_[0], new_shape);
+    }
+
   private:
     T *buffer_{nullptr};
     std::array<ssize_t, Ndim> strides_{};

include/aare/ProducerConsumerQueue.hpp

@@ -45,19 +45,63 @@ template <class T> struct ProducerConsumerQueue {
     ProducerConsumerQueue(const ProducerConsumerQueue &) = delete;
     ProducerConsumerQueue &operator=(const ProducerConsumerQueue &) = delete;
 
-    ProducerConsumerQueue(ProducerConsumerQueue &&other) {
+    // ProducerConsumerQueue(ProducerConsumerQueue &&other) {
+    //     size_ = other.size_;
+    //     records_ = other.records_;
+    //     other.records_ = nullptr;
+    //     readIndex_ = other.readIndex_.load(std::memory_order_acquire);
+    //     writeIndex_ = other.writeIndex_.load(std::memory_order_acquire);
+    // }
+
+    ProducerConsumerQueue(ProducerConsumerQueue&& other) noexcept {
         size_ = other.size_;
         records_ = other.records_;
+        readIndex_.store(other.readIndex_.load(std::memory_order_acquire),
+                        std::memory_order_relaxed);
+        writeIndex_.store(other.writeIndex_.load(std::memory_order_acquire),
+                        std::memory_order_relaxed);
+
         other.records_ = nullptr;
-        readIndex_ = other.readIndex_.load(std::memory_order_acquire);
-        writeIndex_ = other.writeIndex_.load(std::memory_order_acquire);
+        other.size_ = 0;
+        other.readIndex_.store(0, std::memory_order_relaxed);
+        other.writeIndex_.store(0, std::memory_order_relaxed);
     }
-    ProducerConsumerQueue &operator=(ProducerConsumerQueue &&other) {
+
+    // ProducerConsumerQueue &operator=(ProducerConsumerQueue &&other) {
+    //     size_ = other.size_;
+    //     records_ = other.records_;
+    //     other.records_ = nullptr;
+    //     readIndex_ = other.readIndex_.load(std::memory_order_acquire);
+    //     writeIndex_ = other.writeIndex_.load(std::memory_order_acquire);
+    //     return *this;
+    // }
+
+    ProducerConsumerQueue& operator=(ProducerConsumerQueue&& other) {
+        if (this == &other) return *this;
+
+        //Destroy existing elements and free old storage
+        if (records_ && !std::is_trivially_destructible<T>::value) {
+            size_t r = readIndex_.load(std::memory_order_relaxed);
+            size_t w = writeIndex_.load(std::memory_order_relaxed);
+            while (r != w) {
+                records_[r].~T();
+                if (++r == size_) r = 0;
+            }
+        }
+        std::free(records_);
+        
+        //Steal other's state
         size_ = other.size_;
         records_ = other.records_;
+        readIndex_.store( other.readIndex_.load(std::memory_order_acquire), std::memory_order_relaxed );
+        writeIndex_.store( other.writeIndex_.load(std::memory_order_acquire), std::memory_order_relaxed );
+
+        //leave 'other' empty and harmless
         other.records_ = nullptr;
-        readIndex_ = other.readIndex_.load(std::memory_order_acquire);
-        writeIndex_ = other.writeIndex_.load(std::memory_order_acquire);
+        other.size_ = 0;
+        other.readIndex_.store(0, std::memory_order_relaxed);
+        other.writeIndex_.store(0, std::memory_order_relaxed);
+
         return *this;
     }
 

include/aare/VarClusterFinder.hpp

@@ -240,14 +240,14 @@ template <typename T> void VarClusterFinder<T>::first_pass() {
 
     for (ssize_t i = 0; i < original_.size(); ++i) {
         if (use_noise_map)
-            threshold_ = 5 * noiseMap(i);
-        binary_(i) = (original_(i) > threshold_);
+            threshold_ = 5 * noiseMap[i];
+        binary_[i] = (original_[i] > threshold_);
     }
 
     for (int i = 0; i < shape_[0]; ++i) {
         for (int j = 0; j < shape_[1]; ++j) {
 
-            // do we have someting to process?
+            // do we have something to process?
             if (binary_(i, j)) {
                 auto tmp = check_neighbours(i, j);
                 if (tmp != 0) {

include/aare/calibration.hpp

@@ -1,6 +1,9 @@
 #pragma once
 
+#include "aare/NDArray.hpp"
+#include "aare/NDView.hpp"
 #include "aare/defs.hpp"
+#include "aare/utils/par.hpp"
 #include "aare/utils/task.hpp"
 #include <cstdint>
 #include <future>
@@ -55,32 +58,152 @@ ALWAYS_INLINE std::pair<uint16_t, int16_t> get_value_and_gain(uint16_t raw) {
 
 template <class T>
 void apply_calibration_impl(NDView<T, 3> res, NDView<uint16_t, 3> raw_data,
-                       NDView<T, 3> ped, NDView<T, 3> cal, int start,
-                       int stop) {
+                            NDView<T, 3> ped, NDView<T, 3> cal, int start,
+                            int stop) {
 
     for (int frame_nr = start; frame_nr != stop; ++frame_nr) {
         for (int row = 0; row != raw_data.shape(1); ++row) {
             for (int col = 0; col != raw_data.shape(2); ++col) {
-                auto [value, gain] = get_value_and_gain(raw_data(frame_nr, row, col));
+                auto [value, gain] =
+                    get_value_and_gain(raw_data(frame_nr, row, col));
+
+                // Using multiplication does not seem to speed up the code here
+                // ADU/keV is the standard unit for the calibration which
+                // means rewriting the formula is not worth it.
                 res(frame_nr, row, col) =
-                    (value - ped(gain, row, col)) / cal(gain, row, col); //TODO! use multiplication
+                    (value - ped(gain, row, col)) / cal(gain, row, col);
             }
         }
     }
 }
 
 template <class T>
+void apply_calibration_impl(NDView<T, 3> res, NDView<uint16_t, 3> raw_data,
+                            NDView<T, 2> ped, NDView<T, 2> cal, int start,
+                            int stop) {
+
+    for (int frame_nr = start; frame_nr != stop; ++frame_nr) {
+        for (int row = 0; row != raw_data.shape(1); ++row) {
+            for (int col = 0; col != raw_data.shape(2); ++col) {
+                auto [value, gain] =
+                    get_value_and_gain(raw_data(frame_nr, row, col));
+
+                // Using multiplication does not seem to speed up the code here
+                // ADU/keV is the standard unit for the calibration which
+                // means rewriting the formula is not worth it.
+
+                // Set the value to 0 if the gain is not 0
+                if (gain == 0)
+                    res(frame_nr, row, col) =
+                        (value - ped(row, col)) / cal(row, col);
+                else
+                    res(frame_nr, row, col) = 0;
+            }
+        }
+    }
+}
+
+template <class T, ssize_t Ndim = 3>
 void apply_calibration(NDView<T, 3> res, NDView<uint16_t, 3> raw_data,
-                       NDView<T, 3> ped, NDView<T, 3> cal,
+                       NDView<T, Ndim> ped, NDView<T, Ndim> cal,
                        ssize_t n_threads = 4) {
     std::vector<std::future<void>> futures;
     futures.reserve(n_threads);
     auto limits = split_task(0, raw_data.shape(0), n_threads);
     for (const auto &lim : limits)
-        futures.push_back(std::async(&apply_calibration_impl<T>, res, raw_data, ped, cal,
-                                     lim.first, lim.second));
+        futures.push_back(std::async(
+            static_cast<void (*)(NDView<T, 3>, NDView<uint16_t, 3>,
+                                 NDView<T, Ndim>, NDView<T, Ndim>, int, int)>(
+                apply_calibration_impl),
+            res, raw_data, ped, cal, lim.first, lim.second));
     for (auto &f : futures)
         f.get();
 }
 
+template <bool only_gain0>
+std::pair<NDArray<size_t, 3>, NDArray<size_t, 3>>
+sum_and_count_per_gain(NDView<uint16_t, 3> raw_data) {
+    constexpr ssize_t num_gains = only_gain0 ? 1 : 3;
+    NDArray<size_t, 3> accumulator(
+        std::array<ssize_t, 3>{num_gains, raw_data.shape(1), raw_data.shape(2)},
+        0);
+    NDArray<size_t, 3> count(
+        std::array<ssize_t, 3>{num_gains, raw_data.shape(1), raw_data.shape(2)},
+        0);
+    for (int frame_nr = 0; frame_nr != raw_data.shape(0); ++frame_nr) {
+        for (int row = 0; row != raw_data.shape(1); ++row) {
+            for (int col = 0; col != raw_data.shape(2); ++col) {
+                auto [value, gain] =
+                    get_value_and_gain(raw_data(frame_nr, row, col));
+                if (gain != 0 && only_gain0)
+                    continue;
+                accumulator(gain, row, col) += value;
+                count(gain, row, col) += 1;
+            }
+        }
+    }
+
+    return {std::move(accumulator), std::move(count)};
+}
+
+template <typename T, bool only_gain0 = false>
+NDArray<T, 3 - static_cast<ssize_t>(only_gain0)>
+calculate_pedestal(NDView<uint16_t, 3> raw_data, ssize_t n_threads) {
+
+    constexpr ssize_t num_gains = only_gain0 ? 1 : 3;
+    std::vector<std::future<std::pair<NDArray<size_t, 3>, NDArray<size_t, 3>>>>
+        futures;
+    futures.reserve(n_threads);
+
+    auto subviews = make_subviews(raw_data, n_threads);
+
+    for (auto view : subviews) {
+        futures.push_back(std::async(
+            static_cast<std::pair<NDArray<size_t, 3>, NDArray<size_t, 3>> (*)(
+                NDView<uint16_t, 3>)>(&sum_and_count_per_gain<only_gain0>),
+            view));
+    }
+    Shape<3> shape{num_gains, raw_data.shape(1), raw_data.shape(2)};
+    NDArray<size_t, 3> accumulator(shape, 0);
+    NDArray<size_t, 3> count(shape, 0);
+
+    // Combine the results from the futures
+    for (auto &f : futures) {
+        auto [acc, cnt] = f.get();
+        accumulator += acc;
+        count += cnt;
+    }
+
+
+    // Will move to a NDArray<T, 3 - static_cast<ssize_t>(only_gain0)>
+    // if only_gain0 is true
+    return safe_divide<T>(accumulator, count);
+
+}
+
+/**
+ * @brief Count the number of switching pixels in the raw data.
+ * This function counts the number of pixels that switch between G1 and G2 gain.
+ * It returns an NDArray with the number of switching pixels per pixel.
+ * @param raw_data The NDView containing the raw data
+ * @return An NDArray with the number of switching pixels per pixel
+ */
+NDArray<int, 2> count_switching_pixels(NDView<uint16_t, 3> raw_data);
+
+/**
+ * @brief Count the number of switching pixels in the raw data.
+ * This function counts the number of pixels that switch between G1 and G2 gain.
+ * It returns an NDArray with the number of switching pixels per pixel.
+ * @param raw_data The NDView containing the raw data
+ * @param n_threads The number of threads to use for parallel processing
+ * @return An NDArray with the number of switching pixels per pixel
+ */
+NDArray<int, 2> count_switching_pixels(NDView<uint16_t, 3> raw_data,
+                                       ssize_t n_threads);
+
+template <typename T>
+auto calculate_pedestal_g0(NDView<uint16_t, 3> raw_data, ssize_t n_threads) {
+    return calculate_pedestal<T, true>(raw_data, n_threads);
+}
+
 } // namespace aare

include/aare/utils/par.hpp

@@ -1,7 +1,10 @@
+#pragma once
 #include <thread>
 #include <utility>
 #include <vector>
 
+#include "aare/utils/task.hpp"
+
 namespace aare {
 
 template <typename F>
@@ -15,4 +18,17 @@ void RunInParallel(F func, const std::vector<std::pair<int, int>> &tasks) {
         thread.join();
     }
 }
+
+
+template <typename T>
+std::vector<NDView<T,3>> make_subviews(NDView<T, 3> &data, ssize_t n_threads) {
+    std::vector<NDView<T, 3>> subviews;
+    subviews.reserve(n_threads);
+    auto limits = split_task(0, data.shape(0), n_threads);
+    for (const auto &lim : limits) {
+        subviews.push_back(data.sub_view(lim.first, lim.second));
+    }
+    return subviews;
+}
+
 } // namespace aare

include/aare/utils/task.hpp

@@ -1,4 +1,4 @@
-
+#pragma once
 #include <utility>
 #include <vector>
 

python/aare/ClusterFinder.py

@@ -46,14 +46,13 @@ def ClusterFinderMT(image_size, cluster_size = (3,3), dtype=np.int32, n_sigma=5,
     return cls(image_size, n_sigma=n_sigma, capacity=capacity, n_threads=n_threads)
 
 
-
-def ClusterCollector(clusterfindermt, cluster_size = (3,3), dtype=np.int32): 
+def ClusterCollector(clusterfindermt, dtype=np.int32): 
     """ 
     Factory function to create a ClusterCollector object. Provides a cleaner syntax for 
     the templated ClusterCollector in C++.
     """
-
-    cls = _get_class("ClusterCollector", cluster_size, dtype)
+    
+    cls = _get_class("ClusterCollector", clusterfindermt.cluster_size, dtype)
     return cls(clusterfindermt)
 
 def ClusterFileSink(clusterfindermt, cluster_file, dtype=np.int32): 

python/aare/__init__.py

@@ -17,7 +17,7 @@ from .ClusterVector import ClusterVector
 from ._aare import fit_gaus, fit_pol1, fit_scurve, fit_scurve2
 from ._aare import Interpolator
 from ._aare import calculate_eta2
-
+from ._aare import reduce_to_2x2, reduce_to_3x3
 
 from ._aare import apply_custom_weights
 
@@ -32,6 +32,7 @@ from .utils import random_pixels, random_pixel, flat_list, add_colorbar
 from .func import *
 
 from .calibration import *
-from ._aare import apply_calibration
+from ._aare import apply_calibration, count_switching_pixels
+from ._aare import calculate_pedestal, calculate_pedestal_float, calculate_pedestal_g0, calculate_pedestal_g0_float
 
 from ._aare import VarClusterFinder

python/src/bind_Cluster.hpp

@@ -24,7 +24,8 @@ void define_Cluster(py::module &m, const std::string &typestr) {
     py::class_<Cluster<Type, ClusterSizeX, ClusterSizeY, CoordType>>(
         m, class_name.c_str(), py::buffer_protocol())
 
-        .def(py::init([](uint8_t x, uint8_t y, py::array_t<Type> data) {
+        .def(py::init([](uint8_t x, uint8_t y,
+                         py::array_t<Type, py::array::forcecast> data) {
             py::buffer_info buf_info = data.request();
             Cluster<Type, ClusterSizeX, ClusterSizeY, CoordType> cluster;
             cluster.x = x;
@@ -34,31 +35,58 @@ void define_Cluster(py::module &m, const std::string &typestr) {
                 cluster.data[i] = r(i);
             }
             return cluster;
-        }));
+        }))
 
-    /*
-    //TODO! Review if to keep or not
-    .def_property(
-        "data",
-        [](ClusterType &c) -> py::array {
-            return py::array(py::buffer_info(
-                c.data, sizeof(Type),
-                py::format_descriptor<Type>::format(), // Type
-                                                       // format
-                1, // Number of dimensions
-                {static_cast<ssize_t>(ClusterSizeX *
-                                      ClusterSizeY)}, // Shape (flattened)
-                {sizeof(Type)} // Stride (step size between elements)
-                ));
+        // TODO! Review if to keep or not
+        .def_property_readonly(
+            "data",
+            [](Cluster<Type, ClusterSizeX, ClusterSizeY, CoordType> &c)
+                -> py::array {
+                return py::array(py::buffer_info(
+                    c.data.data(), sizeof(Type),
+                    py::format_descriptor<Type>::format(), // Type
+                                                           // format
+                    2, // Number of dimensions
+                    {static_cast<ssize_t>(ClusterSizeX),
+                     static_cast<ssize_t>(ClusterSizeY)}, // Shape (flattened)
+                    {sizeof(Type) * ClusterSizeY, sizeof(Type)}
+                    // Stride (step size between elements)
+                    ));
+            })
+
+        .def_readonly("x",
+                      &Cluster<Type, ClusterSizeX, ClusterSizeY, CoordType>::x)
+
+        .def_readonly("y",
+                      &Cluster<Type, ClusterSizeX, ClusterSizeY, CoordType>::y);
+}
+
+template <typename T, uint8_t ClusterSizeX, uint8_t ClusterSizeY,
+          typename CoordType = int16_t>
+void reduce_to_3x3(py::module &m) {
+
+    m.def(
+        "reduce_to_3x3",
+        [](const Cluster<T, ClusterSizeX, ClusterSizeY, CoordType> &cl) {
+            return reduce_to_3x3(cl);
         },
-        [](ClusterType &c, py::array_t<Type> arr) {
-            py::buffer_info buf_info = arr.request();
-            Type *ptr = static_cast<Type *>(buf_info.ptr);
-            std::copy(ptr, ptr + ClusterSizeX * ClusterSizeY,
-                      c.data); // TODO dont iterate over centers!!!
+        py::return_value_policy::move,
+        "Reduce cluster to 3x3 subcluster by taking the 3x3 subcluster with "
+        "the highest photon energy.");
+}
 
-        });
-    */
+template <typename T, uint8_t ClusterSizeX, uint8_t ClusterSizeY,
+          typename CoordType = int16_t>
+void reduce_to_2x2(py::module &m) {
+
+    m.def(
+        "reduce_to_2x2",
+        [](const Cluster<T, ClusterSizeX, ClusterSizeY, CoordType> &cl) {
+            return reduce_to_2x2(cl);
+        },
+        py::return_value_policy::move,
+        "Reduce cluster to 2x2 subcluster by taking the 2x2 subcluster with "
+        "the highest photon energy.");
 }
 
 #pragma GCC diagnostic pop

python/src/bind_ClusterVector.hpp

@@ -104,4 +104,47 @@ void define_ClusterVector(py::module &m, const std::string &typestr) {
           });
 }
 
+template <typename Type, uint8_t ClusterSizeX, uint8_t ClusterSizeY,
+          typename CoordType = uint16_t>
+void define_2x2_reduction(py::module &m) {
+    m.def(
+        "reduce_to_2x2",
+        [](const ClusterVector<
+            Cluster<Type, ClusterSizeX, ClusterSizeY, CoordType>> &cv) {
+            return new ClusterVector<Cluster<Type, 2, 2, CoordType>>(
+                reduce_to_2x2(cv));
+        },
+        R"(
+        
+        Reduce cluster to 2x2 subcluster by taking the 2x2 subcluster with 
+        the highest photon energy."
+        Parameters
+        ----------
+        cv : ClusterVector   
+        )",
+        py::arg("clustervector"));
+}
+
+template <typename Type, uint8_t ClusterSizeX, uint8_t ClusterSizeY,
+          typename CoordType = uint16_t>
+void define_3x3_reduction(py::module &m) {
+
+    m.def(
+        "reduce_to_3x3",
+        [](const ClusterVector<
+            Cluster<Type, ClusterSizeX, ClusterSizeY, CoordType>> &cv) {
+            return new ClusterVector<Cluster<Type, 3, 3, CoordType>>(
+                reduce_to_3x3(cv));
+        },
+        R"(
+        
+        Reduce cluster to 3x3 subcluster by taking the 3x3 subcluster with 
+        the highest photon energy."
+        Parameters
+        ----------
+        cv : ClusterVector   
+        )",
+        py::arg("clustervector"));
+}
+
 #pragma GCC diagnostic pop

python/src/bind_calibration.hpp

@@ -17,27 +17,137 @@ py::array_t<DataType> pybind_apply_calibration(
         calibration,
     int n_threads = 4) {
 
-    auto data_span = make_view_3d(data);
-    auto ped = make_view_3d(pedestal);
-    auto cal = make_view_3d(calibration);
-
+    auto data_span = make_view_3d(data); // data is always 3D
     /* No pointer is passed, so NumPy will allocate the buffer */
     auto result = py::array_t<DataType>(data_span.shape());
     auto res = make_view_3d(result);
-
-    aare::apply_calibration<DataType>(res, data_span, ped, cal, n_threads);
-
+    if (data.ndim() == 3 && pedestal.ndim() == 3 && calibration.ndim() == 3) {
+        auto ped = make_view_3d(pedestal);
+        auto cal = make_view_3d(calibration);
+        aare::apply_calibration<DataType, 3>(res, data_span, ped, cal,
+                                             n_threads);
+    } else if (data.ndim() == 3 && pedestal.ndim() == 2 &&
+               calibration.ndim() == 2) {
+        auto ped = make_view_2d(pedestal);
+        auto cal = make_view_2d(calibration);
+        aare::apply_calibration<DataType, 2>(res, data_span, ped, cal,
+                                             n_threads);
+    } else {
+        throw std::runtime_error(
+            "Invalid number of dimensions for data, pedestal or calibration");
+    }
     return result;
 }
 
+py::array_t<int> pybind_count_switching_pixels(
+    py::array_t<uint16_t, py::array::c_style | py::array::forcecast> data,
+    ssize_t n_threads = 4) {
+
+    auto data_span = make_view_3d(data);
+    auto arr = new NDArray<int, 2>{};
+    *arr = aare::count_switching_pixels(data_span, n_threads);
+    return return_image_data(arr);
+}
+
+template <typename T>
+py::array_t<T> pybind_calculate_pedestal(
+    py::array_t<uint16_t, py::array::c_style | py::array::forcecast> data,
+    ssize_t n_threads) {
+
+    auto data_span = make_view_3d(data);
+    auto arr = new NDArray<T, 3>{};
+    *arr = aare::calculate_pedestal<T, false>(data_span, n_threads);
+    return return_image_data(arr);
+}
+
+template <typename T>
+py::array_t<T> pybind_calculate_pedestal_g0(
+    py::array_t<uint16_t, py::array::c_style | py::array::forcecast> data,
+    ssize_t n_threads) {
+
+    auto data_span = make_view_3d(data);
+    auto arr = new NDArray<T, 2>{};
+    *arr = aare::calculate_pedestal<T, true>(data_span, n_threads);
+    return return_image_data(arr);
+}
+
 void bind_calibration(py::module &m) {
+    m.def("apply_calibration", &pybind_apply_calibration<double>,
+          py::arg("raw_data").noconvert(), py::kw_only(),
+          py::arg("pd").noconvert(), py::arg("cal").noconvert(),
+          py::arg("n_threads") = 4);
+
     m.def("apply_calibration", &pybind_apply_calibration<float>,
           py::arg("raw_data").noconvert(), py::kw_only(),
           py::arg("pd").noconvert(), py::arg("cal").noconvert(),
           py::arg("n_threads") = 4);
 
-    m.def("apply_calibration", &pybind_apply_calibration<double>,
+    m.def("count_switching_pixels", &pybind_count_switching_pixels,
+          R"(
+        Count the number of time each pixel switches to G1 or G2.
+
+        Parameters
+        ----------
+        raw_data : array_like
+            3D array of shape (frames, rows, cols) to count the switching pixels from.
+        n_threads : int 
+            The number of threads to use for the calculation.
+        )",
           py::arg("raw_data").noconvert(), py::kw_only(),
-          py::arg("pd").noconvert(), py::arg("cal").noconvert(),
           py::arg("n_threads") = 4);
+
+    m.def("calculate_pedestal", &pybind_calculate_pedestal<double>,
+          R"(
+        Calculate the pedestal for all three gains and return the result as a 3D array of doubles.
+
+        Parameters
+        ----------
+        raw_data : array_like
+            3D array of shape (frames, rows, cols) to calculate the pedestal from.
+            Needs to contain data for all three gains (G0, G1, G2).
+        n_threads : int 
+            The number of threads to use for the calculation.
+        )",
+          py::arg("raw_data").noconvert(), py::arg("n_threads") = 4);
+
+    m.def("calculate_pedestal_float", &pybind_calculate_pedestal<float>,
+          R"(
+        Same as `calculate_pedestal` but returns a 3D array of floats.
+
+        Parameters
+        ----------
+        raw_data : array_like
+            3D array of shape (frames, rows, cols) to calculate the pedestal from.
+            Needs to contain data for all three gains (G0, G1, G2).
+        n_threads : int 
+            The number of threads to use for the calculation.
+        )",
+          py::arg("raw_data").noconvert(), py::arg("n_threads") = 4);
+
+    m.def("calculate_pedestal_g0", &pybind_calculate_pedestal_g0<double>,
+          R"(
+        Calculate the pedestal for G0 and return the result as a 2D array of doubles.
+        Pixels in G1 and G2 are ignored.
+
+        Parameters
+        ----------
+        raw_data : array_like
+            3D array of shape (frames, rows, cols) to calculate the pedestal from.
+        n_threads : int 
+            The number of threads to use for the calculation.
+        )",
+          py::arg("raw_data").noconvert(), py::arg("n_threads") = 4);
+
+    m.def("calculate_pedestal_g0_float", &pybind_calculate_pedestal_g0<float>,
+          R"(
+        Same as `calculate_pedestal_g0` but returns a 2D array of floats.
+
+        Parameters
+        ----------
+        raw_data : array_like
+            3D array of shape (frames, rows, cols) to calculate the pedestal from.
+        n_threads : int 
+            The number of threads to use for the calculation.
+        )",
+          py::arg("raw_data").noconvert(), py::arg("n_threads") = 4);
 }

python/src/module.cpp

@@ -47,7 +47,9 @@ double, 'f' for float)
     define_ClusterFileSink<T, N, M, U>(m, "Cluster" #N "x" #M #TYPE_CODE);     \
     define_ClusterCollector<T, N, M, U>(m, "Cluster" #N "x" #M #TYPE_CODE);    \
     define_Cluster<T, N, M, U>(m, #N "x" #M #TYPE_CODE);                       \
-    register_calculate_eta<T, N, M, U>(m);
+    register_calculate_eta<T, N, M, U>(m);                                     \
+    define_2x2_reduction<T, N, M, U>(m);                                       \
+    reduce_to_2x2<T, N, M, U>(m);
 
 PYBIND11_MODULE(_aare, m) {
     define_file_io_bindings(m);
@@ -84,4 +86,30 @@ PYBIND11_MODULE(_aare, m) {
     DEFINE_CLUSTER_BINDINGS(int, 9, 9, uint16_t, i);
     DEFINE_CLUSTER_BINDINGS(double, 9, 9, uint16_t, d);
     DEFINE_CLUSTER_BINDINGS(float, 9, 9, uint16_t, f);
+
+    define_3x3_reduction<int, 3, 3, uint16_t>(m);
+    define_3x3_reduction<double, 3, 3, uint16_t>(m);
+    define_3x3_reduction<float, 3, 3, uint16_t>(m);
+    define_3x3_reduction<int, 5, 5, uint16_t>(m);
+    define_3x3_reduction<double, 5, 5, uint16_t>(m);
+    define_3x3_reduction<float, 5, 5, uint16_t>(m);
+    define_3x3_reduction<int, 7, 7, uint16_t>(m);
+    define_3x3_reduction<double, 7, 7, uint16_t>(m);
+    define_3x3_reduction<float, 7, 7, uint16_t>(m);
+    define_3x3_reduction<int, 9, 9, uint16_t>(m);
+    define_3x3_reduction<double, 9, 9, uint16_t>(m);
+    define_3x3_reduction<float, 9, 9, uint16_t>(m);
+
+    reduce_to_3x3<int, 3, 3, uint16_t>(m);
+    reduce_to_3x3<double, 3, 3, uint16_t>(m);
+    reduce_to_3x3<float, 3, 3, uint16_t>(m);
+    reduce_to_3x3<int, 5, 5, uint16_t>(m);
+    reduce_to_3x3<double, 5, 5, uint16_t>(m);
+    reduce_to_3x3<float, 5, 5, uint16_t>(m);
+    reduce_to_3x3<int, 7, 7, uint16_t>(m);
+    reduce_to_3x3<double, 7, 7, uint16_t>(m);
+    reduce_to_3x3<float, 7, 7, uint16_t>(m);
+    reduce_to_3x3<int, 9, 9, uint16_t>(m);
+    reduce_to_3x3<double, 9, 9, uint16_t>(m);
+    reduce_to_3x3<float, 9, 9, uint16_t>(m);
 }

python/tests/test_Cluster.py

@@ -101,6 +101,27 @@ def test_cluster_finder():
     assert clusters.size == 0
 
 
+def test_2x2_reduction(): 
+    """Test 2x2 Reduction"""
+    cluster = _aare.Cluster3x3i(5,5,np.array([1, 1, 1, 2, 3, 1, 2, 2, 1], dtype=np.int32))
+
+    reduced_cluster = _aare.reduce_to_2x2(cluster) 
+
+    assert reduced_cluster.x == 4
+    assert reduced_cluster.y == 5
+    assert (reduced_cluster.data == np.array([[2, 3], [2, 2]], dtype=np.int32)).all()
+    
+    
+def test_3x3_reduction(): 
+    """Test 3x3 Reduction"""
+    cluster = _aare.Cluster5x5d(5,5,np.array([1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 2.0, 1.0, 1.0, 1.0, 2.0, 2.0, 3.0,
+                                   1.0, 1.0, 1.0, 2.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], dtype=np.double))
+
+    reduced_cluster = _aare.reduce_to_3x3(cluster) 
+
+    assert reduced_cluster.x == 4
+    assert reduced_cluster.y == 5
+    assert (reduced_cluster.data == np.array([[1.0, 2.0, 1.0], [2.0, 2.0, 3.0], [1.0, 2.0, 1.0]], dtype=np.double)).all()
 
 
 

python/tests/test_ClusterVector.py

@@ -5,7 +5,7 @@ import time
 from pathlib import Path
 import pickle
 
-from aare import ClusterFile
+from aare import ClusterFile, ClusterVector
 from aare import _aare
 from conftest import test_data_path
 
@@ -51,4 +51,36 @@ def test_make_a_hitmap_from_cluster_vector():
     # print(img)
     # print(ref)
     assert (img == ref).all()
-    
+
+
+def test_2x2_reduction(): 
+    cv = ClusterVector((3,3))
+
+    cv.push_back(_aare.Cluster3x3i(5, 5, np.array([1, 1, 1, 2, 3, 1, 2, 2, 1], dtype=np.int32)))
+    cv.push_back(_aare.Cluster3x3i(5, 5, np.array([2, 2, 1, 2, 3, 1, 1, 1, 1], dtype=np.int32)))
+
+    reduced_cv = np.array(_aare.reduce_to_2x2(cv), copy=False) 
+
+    assert reduced_cv.size == 2
+    assert reduced_cv[0]["x"] == 4
+    assert reduced_cv[0]["y"] == 5
+    assert (reduced_cv[0]["data"] == np.array([[2, 3], [2, 2]], dtype=np.int32)).all()
+    assert reduced_cv[1]["x"] == 4
+    assert reduced_cv[1]["y"] == 6
+    assert (reduced_cv[1]["data"] == np.array([[2, 2], [2, 3]], dtype=np.int32)).all()
+    
+    
+def test_3x3_reduction(): 
+    cv = _aare.ClusterVector_Cluster5x5d()
+    
+    cv.push_back(_aare.Cluster5x5d(5,5,np.array([1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 2.0, 1.0, 1.0, 1.0, 2.0, 2.0, 3.0,
+                                   1.0, 1.0, 1.0, 2.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], dtype=np.double)))
+    cv.push_back(_aare.Cluster5x5d(5,5,np.array([1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 2.0, 1.0, 1.0, 1.0, 2.0, 2.0, 3.0,
+                                   1.0, 1.0, 1.0, 2.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], dtype=np.double)))
+    
+    reduced_cv = np.array(_aare.reduce_to_3x3(cv), copy=False)  
+
+    assert reduced_cv.size == 2
+    assert reduced_cv[0]["x"] == 4
+    assert reduced_cv[0]["y"] == 5
+    assert (reduced_cv[0]["data"] == np.array([[1.0, 2.0, 1.0], [2.0, 2.0, 3.0], [1.0, 2.0, 1.0]], dtype=np.double)).all()

python/tests/test_calibration.py

@@ -1,6 +1,7 @@
 import pytest 
 import numpy as np
-from aare import apply_calibration
+
+import aare
 
 def test_apply_calibration_small_data():
     # The raw data consists of 10 4x5 images
@@ -27,7 +28,7 @@ def test_apply_calibration_small_data():
 
     
 
-    data = apply_calibration(raw, pd = pedestal, cal = calibration)
+    data = aare.apply_calibration(raw, pd = pedestal, cal = calibration)
 
 
     # The formula that is applied is:
@@ -41,3 +42,94 @@ def test_apply_calibration_small_data():
     assert data[2,2,2] == 0
     assert data[0,1,1] == 0
     assert data[1,3,0] == 0
+
+
+@pytest.fixture
+def raw_data_3x2x2():
+    raw = np.zeros((3, 2, 2), dtype=np.uint16)
+    raw[0, 0, 0] = 100
+    raw[1,0, 0] =  200
+    raw[2, 0, 0] = 300
+
+    raw[0, 0, 1] = (1<<14) + 100
+    raw[1, 0, 1] = (1<<14) + 200
+    raw[2, 0, 1] = (1<<14) + 300
+
+    raw[0, 1, 0] = (1<<14) + 37
+    raw[1, 1, 0] =  38
+    raw[2, 1, 0] = (3<<14) + 39
+
+    raw[0, 1, 1] = (3<<14) + 100
+    raw[1, 1, 1] = (3<<14) + 200
+    raw[2, 1, 1] = (3<<14) + 300
+    return raw
+
+def test_calculate_pedestal(raw_data_3x2x2):
+    # Calculate the pedestal
+    pd = aare.calculate_pedestal(raw_data_3x2x2)
+    assert pd.shape == (3, 2, 2)
+    assert pd.dtype == np.float64
+    assert pd[0, 0, 0] == 200
+    assert pd[1, 0, 0] == 0
+    assert pd[2, 0, 0] == 0
+
+    assert pd[0, 0, 1] == 0
+    assert pd[1, 0, 1] == 200
+    assert pd[2, 0, 1] == 0
+
+    assert pd[0, 1, 0] == 38
+    assert pd[1, 1, 0] == 37
+    assert pd[2, 1, 0] == 39
+
+    assert pd[0, 1, 1] == 0
+    assert pd[1, 1, 1] == 0
+    assert pd[2, 1, 1] == 200
+
+def test_calculate_pedestal_float(raw_data_3x2x2):
+    #results should be the same for float
+    pd2 = aare.calculate_pedestal_float(raw_data_3x2x2)
+    assert pd2.shape == (3, 2, 2)
+    assert pd2.dtype == np.float32
+    assert pd2[0, 0, 0] == 200
+    assert pd2[1, 0, 0] == 0
+    assert pd2[2, 0, 0] == 0
+
+    assert pd2[0, 0, 1] == 0
+    assert pd2[1, 0, 1] == 200
+    assert pd2[2, 0, 1] == 0
+
+    assert pd2[0, 1, 0] == 38
+    assert pd2[1, 1, 0] == 37
+    assert pd2[2, 1, 0] == 39
+
+    assert pd2[0, 1, 1] == 0
+    assert pd2[1, 1, 1] == 0
+    assert pd2[2, 1, 1] == 200
+
+def test_calculate_pedestal_g0(raw_data_3x2x2):
+    pd = aare.calculate_pedestal_g0(raw_data_3x2x2)
+    assert pd.shape == (2, 2)
+    assert pd.dtype == np.float64
+    assert pd[0, 0] == 200
+    assert pd[1, 0] == 38
+    assert pd[0, 1] == 0
+    assert pd[1, 1] == 0
+
+def test_calculate_pedestal_g0_float(raw_data_3x2x2):
+    pd = aare.calculate_pedestal_g0_float(raw_data_3x2x2)
+    assert pd.shape == (2, 2)
+    assert pd.dtype == np.float32
+    assert pd[0, 0] == 200
+    assert pd[1, 0] == 38
+    assert pd[0, 1] == 0
+    assert pd[1, 1] == 0
+
+def test_count_switching_pixels(raw_data_3x2x2):
+    # Count the number of pixels that switched gain
+    count = aare.count_switching_pixels(raw_data_3x2x2)
+    assert count.shape == (2, 2)
+    assert count.sum() == 8
+    assert count[0, 0] == 0 
+    assert count[1, 0] == 2 
+    assert count[0, 1] == 3 
+    assert count[1, 1] == 3

src/Cluster.test.cpp

@@ -18,4 +18,86 @@ TEST_CASE("Test sum of Cluster", "[.cluster]") {
     Cluster<int, 2, 2> cluster{0, 0, {1, 2, 3, 4}};
 
     CHECK(cluster.sum() == 10);
+}
+
+using ClusterTypes = std::variant<Cluster<int, 2, 2>, Cluster<int, 3, 3>,
+                                  Cluster<int, 5, 5>, Cluster<int, 2, 3>>;
+
+using ClusterTypesLargerThan2x2 =
+    std::variant<Cluster<int, 3, 3>, Cluster<int, 4, 4>, Cluster<int, 5, 5>>;
+
+TEST_CASE("Test reduce to 2x2 Cluster", "[.cluster]") {
+    auto [cluster, expected_reduced_cluster] = GENERATE(
+        std::make_tuple(ClusterTypes{Cluster<int, 2, 2>{5, 5, {1, 2, 3, 4}}},
+                        Cluster<int, 2, 2>{4, 6, {1, 2, 3, 4}}),
+        std::make_tuple(
+            ClusterTypes{Cluster<int, 3, 3>{5, 5, {1, 1, 1, 1, 3, 2, 1, 2, 2}}},
+            Cluster<int, 2, 2>{5, 5, {3, 2, 2, 2}}),
+        std::make_tuple(
+            ClusterTypes{Cluster<int, 3, 3>{5, 5, {1, 1, 1, 2, 3, 1, 2, 2, 1}}},
+            Cluster<int, 2, 2>{4, 5, {2, 3, 2, 2}}),
+        std::make_tuple(
+            ClusterTypes{Cluster<int, 3, 3>{5, 5, {2, 2, 1, 2, 3, 1, 1, 1, 1}}},
+            Cluster<int, 2, 2>{4, 6, {2, 2, 2, 3}}),
+        std::make_tuple(
+            ClusterTypes{Cluster<int, 3, 3>{5, 5, {1, 2, 2, 1, 3, 2, 1, 1, 1}}},
+            Cluster<int, 2, 2>{5, 6, {2, 2, 3, 2}}),
+        std::make_tuple(ClusterTypes{Cluster<int, 5, 5>{
+                            5, 5, {1, 1, 1, 1, 1, 1, 1, 2, 2, 1, 1, 1, 3,
+                                   2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}}},
+                        Cluster<int, 2, 2>{5, 6, {2, 2, 3, 2}}),
+        std::make_tuple(ClusterTypes{Cluster<int, 5, 5>{
+                            5, 5, {1, 1, 1, 1, 1, 1, 2, 2, 1, 1, 1, 2, 3,
+                                   1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}}},
+                        Cluster<int, 2, 2>{4, 6, {2, 2, 2, 3}}),
+        std::make_tuple(
+            ClusterTypes{Cluster<int, 2, 3>{5, 5, {2, 2, 3, 2, 1, 1}}},
+            Cluster<int, 2, 2>{4, 6, {2, 2, 3, 2}}));
+
+    auto reduced_cluster = std::visit(
+        [](const auto &clustertype) { return reduce_to_2x2(clustertype); },
+        cluster);
+
+    CHECK(reduced_cluster.x == expected_reduced_cluster.x);
+    CHECK(reduced_cluster.y == expected_reduced_cluster.y);
+    CHECK(std::equal(reduced_cluster.data.begin(),
+                     reduced_cluster.data.begin() + 4,
+                     expected_reduced_cluster.data.begin()));
+}
+
+TEST_CASE("Test reduce to 3x3 Cluster", "[.cluster]") {
+    auto [cluster, expected_reduced_cluster] = GENERATE(
+        std::make_tuple(ClusterTypesLargerThan2x2{Cluster<int, 3, 3>{
+                            5, 5, {1, 1, 1, 1, 3, 1, 1, 1, 1}}},
+                        Cluster<int, 3, 3>{5, 5, {1, 1, 1, 1, 3, 1, 1, 1, 1}}),
+        std::make_tuple(
+            ClusterTypesLargerThan2x2{Cluster<int, 4, 4>{
+                5, 5, {2, 2, 1, 1, 2, 2, 1, 1, 1, 1, 3, 1, 1, 1, 1, 1}}},
+            Cluster<int, 3, 3>{4, 6, {2, 2, 1, 2, 2, 1, 1, 1, 3}}),
+        std::make_tuple(
+            ClusterTypesLargerThan2x2{Cluster<int, 4, 4>{
+                5, 5, {1, 1, 2, 2, 1, 1, 2, 2, 1, 1, 3, 1, 1, 1, 1, 1}}},
+            Cluster<int, 3, 3>{5, 6, {1, 2, 2, 1, 2, 2, 1, 3, 1}}),
+        std::make_tuple(
+            ClusterTypesLargerThan2x2{Cluster<int, 4, 4>{
+                5, 5, {1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 3, 2, 1, 1, 2, 2}}},
+            Cluster<int, 3, 3>{5, 5, {1, 1, 1, 1, 3, 2, 1, 2, 2}}),
+        std::make_tuple(
+            ClusterTypesLargerThan2x2{Cluster<int, 4, 4>{
+                5, 5, {1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 3, 1, 2, 2, 1, 1}}},
+            Cluster<int, 3, 3>{4, 5, {1, 1, 1, 2, 2, 3, 2, 2, 1}}),
+        std::make_tuple(ClusterTypesLargerThan2x2{Cluster<int, 5, 5>{
+                            5, 5, {1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 2, 2, 3,
+                                   1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1}}},
+                        Cluster<int, 3, 3>{4, 5, {1, 2, 1, 2, 2, 3, 1, 2, 1}}));
+
+    auto reduced_cluster = std::visit(
+        [](const auto &clustertype) { return reduce_to_3x3(clustertype); },
+        cluster);
+
+    CHECK(reduced_cluster.x == expected_reduced_cluster.x);
+    CHECK(reduced_cluster.y == expected_reduced_cluster.y);
+    CHECK(std::equal(reduced_cluster.data.begin(),
+                     reduced_cluster.data.begin() + 9,
+                     expected_reduced_cluster.data.begin()));
 }

src/ClusterFinderMT.test.cpp

@@ -57,6 +57,7 @@ class ClusterFinderMTWrapper
     size_t m_sink_size() const { return this->m_sink.sizeGuess(); }
 };
 
+
 TEST_CASE("multithreaded cluster finder", "[.with-data]") {
     auto fpath =
         test_data_path() / "raw/moench03/cu_half_speed_master_4.json";
@@ -81,7 +82,8 @@ TEST_CASE("multithreaded cluster finder", "[.with-data]") {
     CHECK(cf.m_input_queues_are_empty() == true);
 
     for (size_t i = 0; i < n_frames_pd; ++i) {
-        cf.find_clusters(file.read_frame().view<uint16_t>());
+        auto frame = file.read_frame();
+        cf.find_clusters(frame.view<uint16_t>());
     }
 
     cf.stop();

src/Makefile

@@ -0,0 +1,20 @@
+# Makefile
+CXX      := g++
+CXXFLAGS := -std=c++17 -O0 -g
+INCLUDE  := -I../include
+
+SRC  := ProducerConsumerQueue.test.cpp
+BIN  := test_pcq
+
+.PHONY: all clean run
+
+all: $(BIN)
+
+$(BIN): $(SRC)
+	$(CXX) $(CXXFLAGS) $(INCLUDE) $< -o $@
+
+run: $(BIN)
+	./$(BIN)
+
+clean:
+	$(RM) $(BIN)

src/NDArray.test.cpp

@@ -25,13 +25,13 @@ TEST_CASE("Construct from an NDView") {
     REQUIRE(image.data() != view.data());
 
     for (uint32_t i = 0; i < image.size(); ++i) {
-        REQUIRE(image(i) == view(i));
+        REQUIRE(image[i] == view[i]);
     }
 
     // Changing the image doesn't change the view
     image = 43;
     for (uint32_t i = 0; i < image.size(); ++i) {
-        REQUIRE(image(i) != view(i));
+        REQUIRE(image[i] != view[i]);
     }
 }
 
@@ -427,4 +427,30 @@ TEST_CASE("Construct an NDArray from an std::array") {
     for (uint32_t i = 0; i < a.size(); ++i) {
         REQUIRE(a(i) == b[i]);
     }
-}
+}
+
+
+
+TEST_CASE("Move construct from an array with Ndim + 1") {
+    NDArray<int, 3> a({{1,2,2}}, 0);
+    a(0, 0, 0) = 1;
+    a(0, 0, 1) = 2;
+    a(0, 1, 0) = 3;
+    a(0, 1, 1) = 4;
+
+
+    NDArray<int, 2> b(std::move(a));
+    REQUIRE(b.shape() == Shape<2>{2,2});
+    REQUIRE(b.size() == 4);
+    REQUIRE(b(0, 0) == 1);
+    REQUIRE(b(0, 1) == 2);
+    REQUIRE(b(1, 0) == 3);
+    REQUIRE(b(1, 1) == 4);
+
+}
+
+TEST_CASE("Move construct from an array with Ndim + 1 throws on size mismatch") {
+    NDArray<int, 3> a({{2,2,2}}, 0);
+    REQUIRE_THROWS(NDArray<int, 2>(std::move(a)));
+}
+

src/ProducerConsumerQueue.test.cpp

@@ -0,0 +1,83 @@
+#include <iostream>
+#include <atomic>
+#include <string>
+#include <vector>
+#include "aare/ProducerConsumerQueue.hpp"
+
+struct Tracker {
+    static std::atomic<int> ctors;
+    static std::atomic<int> dtors;
+    static std::atomic<int> moves;
+    static std::atomic<int> live;
+
+    std::string tag;
+    std::vector<int> buf;
+
+    Tracker() = delete;
+    explicit Tracker(int id)
+        : tag("T" + std::to_string(id)), buf(1 << 18, id) 
+    {
+        ++ctors; ++live;
+    }
+    
+    Tracker(Tracker&& other) noexcept
+        : tag(std::move(other.tag)), buf(std::move(other.buf)) 
+    {
+        ++moves; 
+        ++ctors; 
+        ++live;
+    }
+
+    Tracker& operator=(Tracker&&) = delete;
+    Tracker(const Tracker&) = delete;
+    Tracker& operator=(const Tracker&) = delete;
+
+    ~Tracker() 
+    {
+        ++dtors; --live;
+    }
+};
+
+std::atomic<int> Tracker::ctors{0};
+std::atomic<int> Tracker::dtors{0};
+std::atomic<int> Tracker::moves{0};
+std::atomic<int> Tracker::live{0};
+
+int main() {
+    using Queue = aare::ProducerConsumerQueue<Tracker>;
+
+    // Scope make sure destructors have ran before we check the counters.
+    {
+        Queue q1(8);
+        Queue q2(8);
+
+        for (int i = 0; i < 3; ++i) q2.write(Tracker(100 + i));
+        for (int i = 0; i < 5; ++i) q1.write(Tracker(200 + i));
+
+        q2 = std::move(q1);
+
+        Tracker tmp(9999);
+        if (auto* p = q2.frontPtr()) 
+        {
+            (void)p;
+        }
+    }
+
+    std::cout << "ctors=" << Tracker::ctors.load()
+            << " dtors=" << Tracker::dtors.load()
+            << " moves=" << Tracker::moves.load()
+            << " live="  << Tracker::live.load()
+            << "\n";
+
+    bool ok = (Tracker::ctors.load() == Tracker::dtors.load()) && (Tracker::live.load() == 0);
+
+    if (!ok) 
+    {
+        std::cerr << "Leak or skipped destructors detected (move-assignment bug)\n";
+        return 1;
+    }
+
+    std::cout << "No leaks; move-assignment cleans up correctly\n";
+    
+    return 0;
+}

src/calibration.cpp

@@ -0,0 +1,44 @@
+#include "aare/calibration.hpp"
+
+namespace aare {
+
+NDArray<int, 2> count_switching_pixels(NDView<uint16_t, 3> raw_data) {
+    NDArray<int, 2> switched(
+        std::array<ssize_t, 2>{raw_data.shape(1), raw_data.shape(2)}, 0);
+    for (int frame_nr = 0; frame_nr != raw_data.shape(0); ++frame_nr) {
+        for (int row = 0; row != raw_data.shape(1); ++row) {
+            for (int col = 0; col != raw_data.shape(2); ++col) {
+                auto [value, gain] =
+                    get_value_and_gain(raw_data(frame_nr, row, col));
+                if (gain != 0) {
+                    switched(row, col) += 1;
+                }
+            }
+        }
+    }
+    return switched;
+}
+
+NDArray<int, 2> count_switching_pixels(NDView<uint16_t, 3> raw_data,
+                                       ssize_t n_threads) {
+    NDArray<int, 2> switched(
+        std::array<ssize_t, 2>{raw_data.shape(1), raw_data.shape(2)}, 0);
+    std::vector<std::future<NDArray<int, 2>>> futures;
+    futures.reserve(n_threads);
+
+    auto subviews = make_subviews(raw_data, n_threads);
+
+    for (auto view : subviews) {
+        futures.push_back(
+            std::async(static_cast<NDArray<int, 2> (*)(NDView<uint16_t, 3>)>(
+                           &count_switching_pixels),
+                       view));
+    }
+
+    for (auto &f : futures) {
+        switched += f.get();
+    }
+    return switched;
+}
+
+} // namespace aare

src/calibration.test.cpp

@@ -0,0 +1,49 @@
+/************************************************
+ * @file test-Cluster.cpp
+ * @short test case for generic Cluster, ClusterVector, and calculate_eta2
+ ***********************************************/
+
+#include "aare/calibration.hpp"
+
+// #include "catch.hpp"
+#include <array>
+#include <catch2/catch_all.hpp>
+#include <catch2/catch_test_macros.hpp>
+
+using namespace aare;
+
+TEST_CASE("Test Pedestal Generation", "[.calibration]") {
+    NDArray<uint16_t, 3> raw(std::array<ssize_t, 3>{3, 2, 2}, 0);
+
+    // gain 0
+    raw(0, 0, 0) = 100;
+    raw(1, 0, 0) = 200;
+    raw(2, 0, 0) = 300;
+
+    // gain 1
+    raw(0, 0, 1) = (1 << 14) + 100;
+    raw(1, 0, 1) = (1 << 14) + 200;
+    raw(2, 0, 1) = (1 << 14) + 300;
+
+    raw(0, 1, 0) = (1 << 14) + 37;
+    raw(1, 1, 0) = 38;
+    raw(2, 1, 0) = (3 << 14) + 39;
+
+    // gain 2
+    raw(0, 1, 1) = (3 << 14) + 100;
+    raw(1, 1, 1) = (3 << 14) + 200;
+    raw(2, 1, 1) = (3 << 14) + 300;
+
+    auto pedestal = calculate_pedestal<double>(raw.view(), 4);
+
+    REQUIRE(pedestal.size() == raw.size());
+    CHECK(pedestal(0, 0, 0) == 200);
+    CHECK(pedestal(1, 0, 0) == 0);
+    CHECK(pedestal(1, 0, 1) == 200);
+
+    auto pedestal_gain0 = calculate_pedestal_g0<double>(raw.view(), 4);
+
+    REQUIRE(pedestal_gain0.size() == 4);
+    CHECK(pedestal_gain0(0, 0) == 200);
+    CHECK(pedestal_gain0(1, 0) == 38);
+}

update_version.py

@@ -7,6 +7,7 @@ Script to update VERSION file with semantic versioning if provided as an argumen
 import sys
 import os
 import re
+from datetime import datetime
 
 from packaging.version import Version, InvalidVersion
 
@@ -26,9 +27,9 @@ def get_version():
 
     # Check at least one argument is passed
     if len(sys.argv) < 2:
-        return "0.0.0"
-
-    version = sys.argv[1]
+        version =   datetime.today().strftime('%Y.%-m.%-d')
+    else: 
+        version = sys.argv[1]
 
     try:
         v = Version(version)  # normalize check if version follows PEP 440 specification
@@ -54,4 +55,4 @@ def write_version_to_file(version):
 if __name__ == "__main__":
 
     version = get_version()
-    write_version_to_file(version)
+    write_version_to_file(version)