Py matthew

libNeonDomain/include/Neon/domain/details/bGrid/bPartition.h

@@ -177,6 +177,26 @@ class bPartition
     helpGetNghIdx(const Idx& idx, const typename Idx::DataBlockIdx* blockConnectivity)
         const -> Idx;
 
+#if !defined(NEON_WARP_COMPILATION)
+    inline static void getOffsets(size_t* offsets, size_t* length) {
+        static std::vector<size_t> cpp_offsets = {
+            offsetof(bPartition, mCardinality),
+            offsetof(bPartition, mMem),
+            offsetof(bPartition, mStencilNghIndex),
+            offsetof(bPartition, mBlockConnectivity),
+            offsetof(bPartition, mMask),
+            offsetof(bPartition, mOrigin),
+            offsetof(bPartition, mSetIdx),
+            offsetof(bPartition, mMultiResDiscreteIdxSpacing),
+            offsetof(bPartition, mDomainSize),
+        };
+
+        *length = cpp_offsets.size();
+        for (size_t i = 0; i < cpp_offsets.size(); ++i) {
+            offsets[i] = cpp_offsets[i];
+        }
+    }
+#endif
 
     int                                             mCardinality;
     T*                                              mMem;

libNeonDomain/include/Neon/domain/details/bGrid/bSpan.h

@@ -44,7 +44,11 @@ class bSpan
     NEON_CUDA_HOST_DEVICE inline auto setAndValidateGPUDevice(
         Idx& bidx) const -> bool;
 
-
+    #if !defined(NEON_WARP_COMPILATION)
+    // Function to get offsets of member variables
+    static void getOffsets(size_t* offsets, size_t* length);
+    #endif
+
     // We don't need to have a count on active blocks
     typename Idx::DataBlockCount                  mFirstDataBlockOffset;
     typename SBlock::BitMask const* NEON_RESTRICT mActiveMask;

libNeonDomain/include/Neon/domain/details/bGrid/bSpan_imp.h

@@ -1,3 +1,5 @@
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Winvalid-offsetof"
 #include "Neon/domain/details/bGrid/bSpan.h"
 
 namespace Neon::domain::details::bGrid {
@@ -29,7 +31,7 @@ bSpan<SBlock>::setAndValidateCPUDevice(Idx&            bidx,
                                        uint32_t const& z) const -> bool
 {
 
-    bidx.mDataBlockIdx = dataBlockIdx + mFirstDataBlockOffset;
+    bidx.mDataBlockIdx = dataBlockIdx;
     bidx.mInDataBlockIdx.x = static_cast<typename Idx::InDataBlockIdx::Integer>(x);
     bidx.mInDataBlockIdx.y = static_cast<typename Idx::InDataBlockIdx::Integer>(y);
     bidx.mInDataBlockIdx.z = static_cast<typename Idx::InDataBlockIdx::Integer>(z);
@@ -47,42 +49,22 @@ bSpan<SBlock>::bSpan(typename Idx::DataBlockCount                  firstDataBloc
 {
 }
 
-
-// #if !defined(NEON_WARP_COMPILATION)
-// template <typename SBlock>
-// // std::vector<size_t> bSpan<SBlock>::getOffsets() {
-// //     return {
-// //         // can't use the `offsetof` macro here since bSpan is a template type
-// //         0,
-// //         sizeof(Idx::DataBlockCount),
-// //         sizeof(Idx::DataBlockCount) + sizeof(SBlock::BitMask*)
-// //     };
-// // }
-// std::vector<size_t> bSpan<SBlock>::getOffsets() {
-//     std::vector<size_t> offsets;
-//     bSpan temp(typename Idx::DataBlockCount(), nullptr, Neon::DataView());
-
-//     offsets.push_back(reinterpret_cast<char*>(&(temp.mFirstDataBlockOffset)) - reinterpret_cast<char*>(&temp));
-//     offsets.push_back(reinterpret_cast<char*>(&(temp.mActiveMask)) - reinterpret_cast<char*>(&temp));
-//     offsets.push_back(reinterpret_cast<char*>(&(temp.mDataView)) - reinterpret_cast<char*>(&temp));
-
-//     return offsets;
-// }
-// #endif
-
 #if !defined(NEON_WARP_COMPILATION)
 template <typename SBlock>
-inline std::vector<size_t> bSpan<SBlock>::getOffsets() {
-    bSpan temp({0}, nullptr, {});
-
-    // Calculate offsets directly
-    std::vector<size_t> offsets;
-    offsets.push_back(reinterpret_cast<size_t>(&(temp.mFirstDataBlockOffset)) - reinterpret_cast<size_t>(&temp));
-    offsets.push_back(reinterpret_cast<size_t>(&(temp.mActiveMask)) - reinterpret_cast<size_t>(&temp));
-    offsets.push_back(reinterpret_cast<size_t>(&(temp.mDataView)) - reinterpret_cast<size_t>(&temp));
+inline void bSpan<SBlock>::getOffsets(size_t* offsets, size_t* length) {
+    static std::vector<size_t> cpp_offsets = {
+        offsetof(bSpan, mFirstDataBlockOffset),
+        offsetof(bSpan, mActiveMask),
+        offsetof(bSpan, mDataView)
+    };
 
-    return offsets;
+    *length = cpp_offsets.size();
+    for (size_t i = 0; i < cpp_offsets.size(); ++i) {
+        offsets[i] = cpp_offsets[i];
+    }
 }
 #endif
 
-}  // namespace Neon::domain::details::bGrid
+}  // namespace Neon::domain::details::bGrid
+
+#pragma GCC diagnostic pop

libNeonDomain/include/Neon/domain/details/dGrid/dPartition.h

@@ -487,6 +487,34 @@ class dPartition
         return s.str();
     }
 
+#if !defined(NEON_WARP_COMPILATION)
+    inline static void getOffsets(size_t* offsets, size_t* length) {
+        // std::cout << "dGrid_dField_dPartition cpp offsets: " << offsetof(dSpan, mDataView) << " " << offsetof(dSpan, mZghostRadius) << " " << offsetof(dSpan, mZboundaryRadius) << " " << offsetof(dSpan, mMaxZInDomain) << " " << offsetof(dSpan, mSpanDim) << " " <<  std::endl;
+        static std::vector<size_t> cpp_offsets = {
+            offsetof(dPartition, mDataView),
+            offsetof(dPartition, mDim),
+            offsetof(dPartition, mMem),
+            offsetof(dPartition, mZHaloRadius),
+            offsetof(dPartition, mZBoundaryRadius),
+            offsetof(dPartition, mPitch.x),
+            offsetof(dPartition, mPitch.y),
+            offsetof(dPartition, mPitch.z),
+            offsetof(dPartition, mPitch.w),
+            offsetof(dPartition, mPrtID),
+            offsetof(dPartition, mOrigin),
+            offsetof(dPartition, mCardinality),
+            offsetof(dPartition, mFullGridSize),
+            offsetof(dPartition, mPeriodicZ),
+            offsetof(dPartition, mStencil),
+        };
+
+        *length = cpp_offsets.size();
+        for (size_t i = 0; i < cpp_offsets.size(); ++i) {
+            offsets[i] = cpp_offsets[i];
+        }
+    }
+#endif
+
    private:
     Neon::DataView        mDataView;
     Neon::index_3d        mDim;

libNeonDomain/include/Neon/domain/details/mGrid/mPartition.h

@@ -1,3 +1,5 @@
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Winvalid-offsetof"
 #pragma once
 
 
@@ -184,6 +186,36 @@ class mPartition : public Neon::domain::details::bGrid::bPartition<T, C, kStatic
      */
     NEON_CUDA_HOST_DEVICE inline Neon::index_3d getGlobalIndex(Idx gidx) const;
 
+#if !defined(NEON_WARP_COMPILATION)
+    inline static void getOffsets(size_t* offsets, size_t* length) {
+        static std::vector<size_t> cpp_offsets = {
+            offsetof(mPartition, mCardinality),
+            offsetof(mPartition, mMem),
+            offsetof(mPartition, mStencilNghIndex),
+            offsetof(mPartition, mBlockConnectivity),
+            offsetof(mPartition, mMask),
+            offsetof(mPartition, mOrigin),
+            offsetof(mPartition, mSetIdx),
+            offsetof(mPartition, mMultiResDiscreteIdxSpacing),
+            offsetof(mPartition, mDomainSize),
+            offsetof(mPartition, mLevel),
+            offsetof(mPartition, mMemParent),
+            offsetof(mPartition, mMemChild),
+            offsetof(mPartition, mParentBlockID),
+            offsetof(mPartition, mMaskLowerLevel),
+            offsetof(mPartition, mMaskUpperLevel),
+            offsetof(mPartition, mChildBlockID),
+            offsetof(mPartition, mParentNeighbourBlocks),
+            offsetof(mPartition, mRefFactors),
+            offsetof(mPartition, mSpacing),
+        };
+
+        *length = cpp_offsets.size();
+        for (size_t i = 0; i < cpp_offsets.size(); ++i) {
+            offsets[i] = cpp_offsets[i];
+        }
+    }
+#endif
 
    private:
     inline NEON_CUDA_HOST_DEVICE auto childID(const Idx& gidx) const -> uint32_t;
@@ -202,4 +234,6 @@ class mPartition : public Neon::domain::details::bGrid::bPartition<T, C, kStatic
 };
 }  // namespace Neon::domain::details::mGrid
 
-#include "Neon/domain/details/mGrid/mPartition_imp.h"
+#include "Neon/domain/details/mGrid/mPartition_imp.h"
+
+#pragma GCC diagnostic pop

libNeonDomain/src/domain/details/dGrid/dGrid.cpp

@@ -32,7 +32,6 @@ auto dGrid::getSpan(Neon::Execution execution,
     return mData->spanTable.getSpan(execution, setIdx, dataView);
 }
 
-
 auto dGrid::helpGetPartitionDim()
     const -> const Neon::set::DataSet<index_3d>
 {

libNeonPy/include/Neon/py/bGrid.h

@@ -5,11 +5,12 @@
 /**
  * Initialize a new grid object on the heap.
  * NOTE: some parameters are still not exposed
- */ /* TODOMATT fix the constructor to have correct arguments */
+ */
 extern "C" auto bGrid_new(
     uint64_t& handle,
-    Neon::Backend* backendPtr,
-    Neon::index_3d dim)
+    uint64_t& backendPtr,
+    const Neon::int32_3d* dim,
+    int* sparsity_pattern)
     -> int;
 
 /**
@@ -19,12 +20,18 @@ extern "C" auto bGrid_delete(
     uint64_t& handle)
     -> int;
 
+extern "C" auto bGrid_get_dimensions(
+    uint64_t& gridHandle,
+    Neon::index_3d* dim)
+    -> int;
+
 /**
  * Generates a new field object on the heap.
  */
 extern "C" auto bGrid_bField_new(
-    uint64_t& handle,
-    uint64_t& grid)
+    uint64_t& fieldHandle,
+    uint64_t& gridHandle,
+    int cardinality)
     -> int;
 
 /**
@@ -60,24 +67,24 @@ extern "C" auto bGrid_bField_partition_size(
 
 extern "C" auto bGrid_get_properties( /* TODOMATT verify what the return of this method should be */
     uint64_t& gridHandle,
-    const Neon::index_3d& idx) 
+    const Neon::index_3d* idx) 
     -> int;
 
 extern "C" auto bGrid_is_inside_domain(
     uint64_t& gridHandle,
-    const Neon::index_3d& idx) 
+    const Neon::index_3d* const idx) 
     -> bool;
 
 extern "C" auto bGrid_bField_read(
     uint64_t& fieldHandle,
-    const Neon::index_3d& idx,
-    const int& cardinality)
+    const Neon::index_3d* idx,
+    const int cardinality)
     -> int;
 
 extern "C" auto bGrid_bField_write(
     uint64_t& fieldHandle,
-    const Neon::index_3d& idx,
-    const int& cardinality,
+    const Neon::index_3d* idx,
+    const int cardinality,
     int newValue)
     -> int;
 
@@ -91,5 +98,9 @@ extern "C" auto bGrid_bField_update_device_data(
     int streamSetId)
     -> int;
 
-extern "C" auto dGrid_dSpan_get_member_field_offsets(std::size_t* length)
-    -> std::size_t*;
+extern "C" auto bGrid_bSpan_get_member_field_offsets(size_t* offsets, size_t* length)
+    -> void;
+
+
+extern "C" auto bGrid_bField_bPartition_get_member_field_offsets(size_t* offsets, size_t* length)
+    -> void;

libNeonPy/include/Neon/py/backend.h

@@ -2,42 +2,23 @@
 #include <cstdint>
 #include "Neon/set/Backend.h"
 
-
-///**
-// * Empty constructor
-// */
-//extern "C" auto dBackend_new1(
-//    uint64_t& handle)
-//    -> int;
-//
-///**
-// * Creating a Backend object with the first nGpus devices.
-// */
-//extern "C" auto dBackend_new2(
-//    uint64_t& handle,
-//    int nGpus /*!   Number of devices. The devices are selected in the order specifies by CUDA */,
-//    int runtime /*! Type of runtime to use */)
-//    -> int;
-
 /**
  *
  */
 extern "C" auto dBackend_new(
     uint64_t& handle,
-      int runtime /*! Type of runtime to use */,
-      int numDecices /*! Number of devices */,
-    const int* devIds /*!  Vectors of device ids. There are CUDA device ids */
-  )
+    int runtime /*! Type of runtime to use */,
+    int numDecices /*! Number of devices */,
+    const int* devIds /*!  Vectors of device ids. There are CUDA device ids */)
     -> int;
 
-
 /**
  * Delete a backend object on the heap.
  */
 extern "C" auto dBackend_delete(
     uint64_t& handle)
     -> int;
 
-
 extern "C" auto dBackend_get_string(uint64_t& handle) -> const char*;
 
+extern "C" auto dBackend_sync(uint64_t& handle) -> int;