PackInfo MPI buffer alignment bugfix

espressomd · Sep 20, 2024 · 960c2ad · 960c2ad
1 parent 8ea7111
commit 960c2ad
Show file tree

Hide file tree

Showing 8 changed files with 28 additions and 8 deletions.
diff --git a/src/walberla_bridge/src/lattice_boltzmann/generated_kernels/PackInfoPdfDoublePrecision.cpp b/src/walberla_bridge/src/lattice_boltzmann/generated_kernels/PackInfoPdfDoublePrecision.cpp
@@ -24,6 +24,8 @@
 #include "core/cell/CellInterval.h"
 #include "stencil/Directions.h"
 
+#include <cstddef>
+
 #if (defined WALBERLA_CXX_COMPILER_IS_GNU) || (defined WALBERLA_CXX_COMPILER_IS_CLANG)
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wfloat-equal"
@@ -519,6 +521,7 @@ static FUNC_PREFIX void unpack_NE(double *RESTRICT const _data_buffer, double *R
 } // namespace internal_unpack_NE
 
 void PackInfoPdfDoublePrecision::pack(Direction dir, unsigned char *byte_buffer, IBlock *block) const {
+  byte_buffer += sizeof(double) - (reinterpret_cast<std::size_t>(byte_buffer) - (reinterpret_cast<std::size_t>(byte_buffer) / sizeof(double)) * sizeof(double));
   double *buffer = reinterpret_cast<double *>(byte_buffer);
 
   auto pdfs = block->getData<field::GhostLayerField<double, 19>>(pdfsID);
@@ -893,6 +896,7 @@ void PackInfoPdfDoublePrecision::pack(Direction dir, unsigned char *byte_buffer,
 }
 
 void PackInfoPdfDoublePrecision::unpack(Direction dir, unsigned char *byte_buffer, IBlock *block) const {
+  byte_buffer += sizeof(double) - (reinterpret_cast<std::size_t>(byte_buffer) - (reinterpret_cast<std::size_t>(byte_buffer) / sizeof(double)) * sizeof(double));
   double *buffer = reinterpret_cast<double *>(byte_buffer);
 
   auto pdfs = block->getData<field::GhostLayerField<double, 19>>(pdfsID);

diff --git a/src/walberla_bridge/src/lattice_boltzmann/generated_kernels/PackInfoPdfDoublePrecision.h b/src/walberla_bridge/src/lattice_boltzmann/generated_kernels/PackInfoPdfDoublePrecision.h
@@ -54,7 +54,7 @@ class PackInfoPdfDoublePrecision
   void unpackData(IBlock *receiver, stencil::Direction dir,
                   mpi::RecvBuffer &buffer) {
     const auto dataSize = size(dir, receiver);
-    unpack(dir, buffer.skip(dataSize), receiver);
+    unpack(dir, buffer.skip(dataSize + sizeof(double)), receiver);
   }
 
   void communicateLocal(const IBlock *sender, IBlock *receiver,
@@ -68,7 +68,8 @@ class PackInfoPdfDoublePrecision
   void packDataImpl(const IBlock *sender, stencil::Direction dir,
                     mpi::SendBuffer &outBuffer) const {
     const auto dataSize = size(dir, sender);
-    pack(dir, outBuffer.forward(dataSize), const_cast<IBlock *>(sender));
+    pack(dir, outBuffer.forward(dataSize + sizeof(double)),
+         const_cast<IBlock *>(sender));
   }
 
   void pack(stencil::Direction dir, unsigned char *buffer, IBlock *block) const;

diff --git a/src/walberla_bridge/src/lattice_boltzmann/generated_kernels/PackInfoPdfSinglePrecision.cpp b/src/walberla_bridge/src/lattice_boltzmann/generated_kernels/PackInfoPdfSinglePrecision.cpp
@@ -24,6 +24,8 @@
 #include "core/cell/CellInterval.h"
 #include "stencil/Directions.h"
 
+#include <cstddef>
+
 #if (defined WALBERLA_CXX_COMPILER_IS_GNU) || (defined WALBERLA_CXX_COMPILER_IS_CLANG)
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wfloat-equal"
@@ -519,6 +521,7 @@ static FUNC_PREFIX void unpack_NE(float *RESTRICT const _data_buffer, float *RES
 } // namespace internal_unpack_NE
 
 void PackInfoPdfSinglePrecision::pack(Direction dir, unsigned char *byte_buffer, IBlock *block) const {
+  byte_buffer += sizeof(float) - (reinterpret_cast<std::size_t>(byte_buffer) - (reinterpret_cast<std::size_t>(byte_buffer) / sizeof(float)) * sizeof(float));
   float *buffer = reinterpret_cast<float *>(byte_buffer);
 
   auto pdfs = block->getData<field::GhostLayerField<float, 19>>(pdfsID);
@@ -893,6 +896,7 @@ void PackInfoPdfSinglePrecision::pack(Direction dir, unsigned char *byte_buffer,
 }
 
 void PackInfoPdfSinglePrecision::unpack(Direction dir, unsigned char *byte_buffer, IBlock *block) const {
+  byte_buffer += sizeof(float) - (reinterpret_cast<std::size_t>(byte_buffer) - (reinterpret_cast<std::size_t>(byte_buffer) / sizeof(float)) * sizeof(float));
   float *buffer = reinterpret_cast<float *>(byte_buffer);
 
   auto pdfs = block->getData<field::GhostLayerField<float, 19>>(pdfsID);

diff --git a/src/walberla_bridge/src/lattice_boltzmann/generated_kernels/PackInfoPdfSinglePrecision.h b/src/walberla_bridge/src/lattice_boltzmann/generated_kernels/PackInfoPdfSinglePrecision.h
@@ -54,7 +54,7 @@ class PackInfoPdfSinglePrecision
   void unpackData(IBlock *receiver, stencil::Direction dir,
                   mpi::RecvBuffer &buffer) {
     const auto dataSize = size(dir, receiver);
-    unpack(dir, buffer.skip(dataSize), receiver);
+    unpack(dir, buffer.skip(dataSize + sizeof(float)), receiver);
   }
 
   void communicateLocal(const IBlock *sender, IBlock *receiver,
@@ -68,7 +68,8 @@ class PackInfoPdfSinglePrecision
   void packDataImpl(const IBlock *sender, stencil::Direction dir,
                     mpi::SendBuffer &outBuffer) const {
     const auto dataSize = size(dir, sender);
-    pack(dir, outBuffer.forward(dataSize), const_cast<IBlock *>(sender));
+    pack(dir, outBuffer.forward(dataSize + sizeof(float)),
+         const_cast<IBlock *>(sender));
   }
 
   void pack(stencil::Direction dir, unsigned char *buffer, IBlock *block) const;

diff --git a/src/walberla_bridge/src/lattice_boltzmann/generated_kernels/PackInfoVecDoublePrecision.cpp b/src/walberla_bridge/src/lattice_boltzmann/generated_kernels/PackInfoVecDoublePrecision.cpp
@@ -24,6 +24,8 @@
 #include "core/cell/CellInterval.h"
 #include "stencil/Directions.h"
 
+#include <cstddef>
+
 #if (defined WALBERLA_CXX_COMPILER_IS_GNU) || (defined WALBERLA_CXX_COMPILER_IS_CLANG)
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wfloat-equal"
@@ -67,6 +69,7 @@ static FUNC_PREFIX void unpack_SW_BW_W_TW_NW_BS_S_TS_B_C_T_BN_N_TN_SE_BE_E_TE_NE
 } // namespace internal_1ccccad4ca561e07a0934cadb07d0fc1
 
 void PackInfoVecDoublePrecision::pack(Direction dir, unsigned char *byte_buffer, IBlock *block) const {
+  byte_buffer += sizeof(double) - (reinterpret_cast<std::size_t>(byte_buffer) - (reinterpret_cast<std::size_t>(byte_buffer) / sizeof(double)) * sizeof(double));
   double *buffer = reinterpret_cast<double *>(byte_buffer);
 
   auto field = block->getData<field::GhostLayerField<double, 3>>(fieldID);
@@ -119,6 +122,7 @@ void PackInfoVecDoublePrecision::pack(Direction dir, unsigned char *byte_buffer,
 }
 
 void PackInfoVecDoublePrecision::unpack(Direction dir, unsigned char *byte_buffer, IBlock *block) const {
+  byte_buffer += sizeof(double) - (reinterpret_cast<std::size_t>(byte_buffer) - (reinterpret_cast<std::size_t>(byte_buffer) / sizeof(double)) * sizeof(double));
   double *buffer = reinterpret_cast<double *>(byte_buffer);
 
   auto field = block->getData<field::GhostLayerField<double, 3>>(fieldID);

diff --git a/src/walberla_bridge/src/lattice_boltzmann/generated_kernels/PackInfoVecDoublePrecision.h b/src/walberla_bridge/src/lattice_boltzmann/generated_kernels/PackInfoVecDoublePrecision.h
@@ -54,7 +54,7 @@ class PackInfoVecDoublePrecision
   void unpackData(IBlock *receiver, stencil::Direction dir,
                   mpi::RecvBuffer &buffer) {
     const auto dataSize = size(dir, receiver);
-    unpack(dir, buffer.skip(dataSize), receiver);
+    unpack(dir, buffer.skip(dataSize + sizeof(double)), receiver);
   }
 
   void communicateLocal(const IBlock *sender, IBlock *receiver,
@@ -68,7 +68,8 @@ class PackInfoVecDoublePrecision
   void packDataImpl(const IBlock *sender, stencil::Direction dir,
                     mpi::SendBuffer &outBuffer) const {
     const auto dataSize = size(dir, sender);
-    pack(dir, outBuffer.forward(dataSize), const_cast<IBlock *>(sender));
+    pack(dir, outBuffer.forward(dataSize + sizeof(double)),
+         const_cast<IBlock *>(sender));
   }
 
   void pack(stencil::Direction dir, unsigned char *buffer, IBlock *block) const;

diff --git a/src/walberla_bridge/src/lattice_boltzmann/generated_kernels/PackInfoVecSinglePrecision.cpp b/src/walberla_bridge/src/lattice_boltzmann/generated_kernels/PackInfoVecSinglePrecision.cpp
@@ -24,6 +24,8 @@
 #include "core/cell/CellInterval.h"
 #include "stencil/Directions.h"
 
+#include <cstddef>
+
 #if (defined WALBERLA_CXX_COMPILER_IS_GNU) || (defined WALBERLA_CXX_COMPILER_IS_CLANG)
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wfloat-equal"
@@ -67,6 +69,7 @@ static FUNC_PREFIX void unpack_SW_BW_W_TW_NW_BS_S_TS_B_C_T_BN_N_TN_SE_BE_E_TE_NE
 } // namespace internal_1ccccad4ca561e07a0934cadb07d0fc1
 
 void PackInfoVecSinglePrecision::pack(Direction dir, unsigned char *byte_buffer, IBlock *block) const {
+  byte_buffer += sizeof(float) - (reinterpret_cast<std::size_t>(byte_buffer) - (reinterpret_cast<std::size_t>(byte_buffer) / sizeof(float)) * sizeof(float));
   float *buffer = reinterpret_cast<float *>(byte_buffer);
 
   auto field = block->getData<field::GhostLayerField<float, 3>>(fieldID);
@@ -119,6 +122,7 @@ void PackInfoVecSinglePrecision::pack(Direction dir, unsigned char *byte_buffer,
 }
 
 void PackInfoVecSinglePrecision::unpack(Direction dir, unsigned char *byte_buffer, IBlock *block) const {
+  byte_buffer += sizeof(float) - (reinterpret_cast<std::size_t>(byte_buffer) - (reinterpret_cast<std::size_t>(byte_buffer) / sizeof(float)) * sizeof(float));
   float *buffer = reinterpret_cast<float *>(byte_buffer);
 
   auto field = block->getData<field::GhostLayerField<float, 3>>(fieldID);

diff --git a/src/walberla_bridge/src/lattice_boltzmann/generated_kernels/PackInfoVecSinglePrecision.h b/src/walberla_bridge/src/lattice_boltzmann/generated_kernels/PackInfoVecSinglePrecision.h
@@ -54,7 +54,7 @@ class PackInfoVecSinglePrecision
   void unpackData(IBlock *receiver, stencil::Direction dir,
                   mpi::RecvBuffer &buffer) {
     const auto dataSize = size(dir, receiver);
-    unpack(dir, buffer.skip(dataSize), receiver);
+    unpack(dir, buffer.skip(dataSize + sizeof(float)), receiver);
   }
 
   void communicateLocal(const IBlock *sender, IBlock *receiver,
@@ -68,7 +68,8 @@ class PackInfoVecSinglePrecision
   void packDataImpl(const IBlock *sender, stencil::Direction dir,
                     mpi::SendBuffer &outBuffer) const {
     const auto dataSize = size(dir, sender);
-    pack(dir, outBuffer.forward(dataSize), const_cast<IBlock *>(sender));
+    pack(dir, outBuffer.forward(dataSize + sizeof(float)),
+         const_cast<IBlock *>(sender));
   }
 
   void pack(stencil::Direction dir, unsigned char *buffer, IBlock *block) const;