Fix for CUDA communication: Consider neighbor fragment accesses with "dity...

Fix for CUDA communication: Consider neighbor fragment accesses with "dity data" for memory transfers.

Compiler/src/exastencils/parallelization/api/cuda/CUDA_GatherFieldAccess.scala

@@ -23,6 +23,7 @@ import scala.collection.mutable._
 import exastencils.base.ir._
 import exastencils.core.collectors.Collector
 import exastencils.datastructures._
+import exastencils.domain.ir.IR_IV_NeighborFragmentIdx
 import exastencils.field.ir._
 import exastencils.logger.Logger
 
@@ -85,6 +86,12 @@ class CUDA_GatherFieldAccess extends Collector {
           }
         }
 
+        // also consider neighbor fragment accesses
+        access.fragIdx match {
+          case neigh : IR_IV_NeighborFragmentIdx => identifier += s"_n${ neigh.neighIdx }"
+          case _                                 =>
+        }
+
         if (isRead)
           fieldAccesses.put("read_" + identifier, access)
         if (isWrite)

Compiler/src/exastencils/parallelization/api/cuda/CUDA_HandleFieldAccessLike.scala

@@ -25,6 +25,7 @@ import exastencils.base.ir._
 import exastencils.config._
 import exastencils.datastructures.Transformation._
 import exastencils.datastructures._
+import exastencils.domain.ir.IR_IV_NeighborFragmentIdx
 import exastencils.field.ir._
 import exastencils.optimization.ir.IR_SimplifyExpression
 
@@ -49,6 +50,12 @@ object CUDA_GatherFieldAccessLike extends QuietDefaultStrategy("Gather local Fie
       }
     }
 
+    // also consider neighbor fragment accesses
+    access.fragIdx match {
+      case neigh : IR_IV_NeighborFragmentIdx => identifier += s"_n${ neigh.neighIdx }"
+      case _                                 =>
+    }
+
     identifier
   }
 

Compiler/src/exastencils/parallelization/api/cuda/CUDA_MemoryTransfer.scala

@@ -66,7 +66,7 @@ case class CUDA_UpdateHostData(var fieldData : IR_IV_FieldData) extends CUDA_Hos
     val field = fieldData.field
 
     IR_IfCondition(
-      CUDA_DeviceDataUpdated(field, Duplicate(fieldData.slot)),
+      CUDA_DeviceDataUpdated(field, Duplicate(fieldData.slot), fieldData.fragmentIdx),
       ListBuffer[IR_Statement](
         CUDA_TransferUtil.genTransfer(
           IR_IV_FieldData(field, Duplicate(fieldData.slot)),
@@ -92,7 +92,7 @@ case class CUDA_UpdateDeviceData(var fieldData : IR_IV_FieldData) extends CUDA_H
     val field = fieldData.field
 
     IR_IfCondition(
-      CUDA_HostDataUpdated(field, Duplicate(fieldData.slot)),
+      CUDA_HostDataUpdated(field, Duplicate(fieldData.slot), fieldData.fragmentIdx),
       ListBuffer[IR_Statement](
         CUDA_TransferUtil.genTransfer(
           IR_IV_FieldData(field, Duplicate(fieldData.slot)),