add initial version of exastencils

.gitignore

+target
+.cache
+.idea

AUTHORS.txt

+Main authors
+====================
+
+Christian Schmitt
+Sebastian Kuckuk
+Stefan Kronawitter
+
+List of contributors
+====================
+
+Alexander Grebhahn
+Armin Größlinger
+Christoph Wollner
+Daniel Zint
+David Spitzberg
+Denis Ribica
+Ewald Flad
+Georg Altmann
+Hannah Rittich
+Harald Köstler
+Jonas Schmitt
+Lisa Claus
+Maik Haase
+Markus Obereisenbuchner
+Max Gerecke
+Meike Blöcher
+Richard Angersbach
+Thomas Lang

Compiler/.gitignore

+bin
+build
+target
+.cache
+.settings

Compiler/build.sbt

+
+
+name := "Compiler"
+
+version := "1.0"
+
+sbtVersion := "1.3.2"
+
+scalaVersion := "2.12.10"
+
+scalaSource in Compile := baseDirectory.value / "src"
+
+resourceDirectory in Compile := baseDirectory.value / "res"
+
+unmanagedBase in Compile := baseDirectory.value / "lib"
+
+libraryDependencies += "org.scala-lang" % "scala-reflect" % scalaVersion.value
+libraryDependencies += "org.scala-lang.modules" %% "scala-parser-combinators" % "1.1.2"
+libraryDependencies += "io.github.kostaskougios" % "cloning" % "1.10.0"
+libraryDependencies += "org.objenesis" % "objenesis" % "3.1"
+
+lazy val downloadCImg = taskKey[Unit]("Download CImg header and store it in Compiler/res")
+
+downloadCImg := {
+  val sourceUrl = "https://framagit.org/dtschump/CImg/raw/master/CImg.h"
+  val targetFileName = "Compiler/res/CImg.h"
+  val overwrite = true
+
+  if (overwrite || java.nio.file.Files.notExists(new File(targetFileName).toPath)) {
+    println(s"Downloading CImg header from $sourceUrl")
+
+    val targetFile = new java.io.File(targetFileName)
+    if (!targetFile.getParentFile.exists()) {
+      println("Setting up directory")
+      targetFile.getParentFile.mkdirs()
+    }
+
+    try {
+      val src = scala.io.Source.fromURL(sourceUrl)
+      val out = new java.io.FileWriter(targetFile)
+      out.write(src.mkString)
+      out.close()
+    } catch {
+      case e : java.io.IOException => println(s"Error downloading: $e")
+    }
+
+    println("Successful")
+  } else {
+    println("CImg header already exists")
+  }
+}

Compiler/src/Main.scala

+//=============================================================================
+//
+//  This file is part of the ExaStencils code generation framework. ExaStencils
+//  is free software: you can redistribute it and/or modify it under the terms
+//  of the GNU General Public License as published by the Free Software
+//  Foundation, either version 3 of the License, or (at your option) any later
+//  version.
+//
+//  ExaStencils is distributed in the hope that it will be useful, but WITHOUT
+//  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+//  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+//  more details.
+//
+//  You should have received a copy of the GNU General Public License along
+//  with ExaStencils. If not, see <http://www.gnu.org/licenses/>.
+//
+//=============================================================================
+
+import scala.collection.mutable.ListBuffer
+
+import exastencils.app._
+import exastencils.base.ExaRootNode
+import exastencils.config._
+import exastencils.core._
+import exastencils.core.logger.Logger_HTML
+import exastencils.datastructures._
+import exastencils.logger._
+import exastencils.parsers.config.Settings_Parser
+import exastencils.polyhedron.IR_PolyOpt
+import exastencils.prettyprinting._
+import exastencils.util._
+
+object Main {
+  def localInitialize(args : Array[String]) = {
+    //if (Settings.timeStrategies) -> right now this Schroedinger flag is neither true nor false
+    StrategyTimer.startTiming("Initializing")
+
+    StateManager.setRoot(ExaRootNode)
+
+    // check from where to read input
+    val settingsParser = new Settings_Parser(Settings)
+    val knowledgeParser = new Settings_Parser(Knowledge)
+    val platformParser = new Settings_Parser(Platform)
+    if (args.length >= 1)
+      settingsParser.parseFile(args(0))
+    if (Settings.produceHtmlLog) Logger_HTML.init() // allows emitting errors and warning in knowledge and platform parsers
+    if (args.length >= 2)
+      knowledgeParser.parseFile(args(1))
+    if (args.length >= 3)
+      platformParser.parseFile(args(2))
+    if (args.length >= 4)
+      IR_PolyOpt.polyOptExplIDs = args(3)
+
+    // validate knowledge, etc.
+    Knowledge.update()
+    Settings.update()
+    Platform.update()
+
+    // resolve aliases in knowledge, settings and platform
+    ResolveAlias.apply()
+
+    // begin writing log to file after potential alias resolution in filename
+    if (Settings.produceHtmlLog) Logger_HTML.beginFileWrite()
+
+    if (Settings.cancelIfOutFolderExists) {
+      if (new java.io.File(Settings.getOutputPath).exists()) {
+        Logger.error(s"Output path ${ Settings.getOutputPath } already exists but cancelIfOutFolderExists is set to true. Shutting down now...")
+        sys.exit(0)
+      }
+    }
+
+    // init buildfile generator, overrides settings file
+    if ("MSVC" == Platform.targetCompiler)
+      Settings.buildfileGenerators = ListBuffer("ProjectfileGenerator")
+
+    if (Settings.timeStrategies)
+      StrategyTimer.stopTiming("Initializing")
+  }
+
+  def localShutdown() = {
+    StateManager.root.nodes.clear()
+    ExaRootNode.clear()
+
+    if (Settings.timeStrategies) {
+      StrategyTimer.print()
+      StrategyTimer.clear()
+    }
+
+    if (Settings.produceHtmlLog)
+      Logger_HTML.finish()
+
+    ObjectWithStateCollection.clearAll()
+  }
+
+  def print() = {
+    Logger.dbg("Prettyprinting to folder " + new java.io.File(Settings.getOutputPath).getAbsolutePath)
+    ExaLayerHandler.ir_handler.print()
+    PrettyprintingManager.finish()
+  }
+
+  def main(args : Array[String]) : Unit = {
+    try {
+      // for runtime measurement
+      val start : Long = System.nanoTime()
+
+      localInitialize(args)
+      ExaLayerHandler.initializeAllLayers()
+
+      ExaLayerHandler.handleAllLayers()
+
+      print()
+
+      Logger.dbg("Done!")
+
+      Logger.dbg("Runtime:\t" + math.round((System.nanoTime() - start) / 1e8) / 10.0 + " seconds")
+      new CountNodes("number of printed nodes").apply()
+
+      ExaLayerHandler.shutdownAllLayers()
+      localShutdown()
+    } catch {
+      case e : LoggerError =>
+        throw e
+      case e : Throwable =>
+        Logger.warn(s"Critical error: ${ e.getMessage }")
+        Logger.warn(s"Stack trace:\n${ e.getStackTrace.mkString("\n\tat ") }")
+
+        if (Settings.produceHtmlLog)
+          Logger_HTML.finish()
+
+        throw e
+    }
+  }
+}

Compiler/src/exastencils/app/ExaLayerHandler.scala

+//=============================================================================
+//
+//  This file is part of the ExaStencils code generation framework. ExaStencils
+//  is free software: you can redistribute it and/or modify it under the terms
+//  of the GNU General Public License as published by the Free Software
+//  Foundation, either version 3 of the License, or (at your option) any later
+//  version.
+//
+//  ExaStencils is distributed in the hope that it will be useful, but WITHOUT
+//  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+//  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+//  more details.
+//
+//  You should have received a copy of the GNU General Public License along
+//  with ExaStencils. If not, see <http://www.gnu.org/licenses/>.
+//
+//=============================================================================
+
+package exastencils.app
+
+import exastencils.app.ir._
+import exastencils.app.l1._
+import exastencils.app.l2._
+import exastencils.app.l3._
+import exastencils.app.l4._
+import exastencils.config.Settings
+
+/// ExaLayerHandler
+
+object ExaLayerHandler {
+  var l1_handler : L1_LayerHandler = L1_DefaultLayerHandler
+  var l2_handler : L2_LayerHandler = L2_DefaultLayerHandler
+  var l3_handler : L3_LayerHandler = L3_DefaultLayerHandler
+  var l4_handler : L4_LayerHandler = L4_DefaultLayerHandler
+  var ir_handler : IR_LayerHandler = IR_DefaultLayerHandler
+
+  def allLayers = Array(l1_handler, l2_handler, l3_handler, l4_handler, ir_handler)
+
+  def maskUnusedLayers() : Unit = {
+    if (Settings.minLayerFileProvided > 1) l1_handler = L1_DummyLayerHandler
+    if (Settings.minLayerFileProvided > 2) l2_handler = L2_DummyLayerHandler
+    if (Settings.minLayerFileProvided > 3) l3_handler = L3_DummyLayerHandler
+    if (Settings.minLayerFileProvided > 4) l4_handler = L4_DummyLayerHandler
+    if (Settings.minLayerFileProvided > 5) ir_handler = IR_DummyLayerHandler
+  }
+
+  def initializeAllLayers() : Unit = {
+    allLayers.foreach(_.initialize())
+  }
+
+  def handleAllLayers() : Unit = {
+    allLayers.foreach(_.handle())
+  }
+
+  def shutdownAllLayers() : Unit = {
+    allLayers.foreach(_.shutdown())
+  }
+}
\ No newline at end of file

Compiler/src/exastencils/app/LayerHandler.scala

+//=============================================================================
+//
+//  This file is part of the ExaStencils code generation framework. ExaStencils
+//  is free software: you can redistribute it and/or modify it under the terms
+//  of the GNU General Public License as published by the Free Software
+//  Foundation, either version 3 of the License, or (at your option) any later
+//  version.
+//
+//  ExaStencils is distributed in the hope that it will be useful, but WITHOUT
+//  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+//  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+//  more details.
+//
+//  You should have received a copy of the GNU General Public License along
+//  with ExaStencils. If not, see <http://www.gnu.org/licenses/>.
+//
+//=============================================================================
+
+package exastencils.app
+
+/// LayerHandler
+
+trait LayerHandler {
+  def initialize() : Unit
+  def handle() : Unit
+  def print() : Unit
+  def shutdown() : Unit
+}

Compiler/src/exastencils/app/ir/IR_LayerHandler.scala

+//=============================================================================
+//
+//  This file is part of the ExaStencils code generation framework. ExaStencils
+//  is free software: you can redistribute it and/or modify it under the terms
+//  of the GNU General Public License as published by the Free Software
+//  Foundation, either version 3 of the License, or (at your option) any later
+//  version.
+//
+//  ExaStencils is distributed in the hope that it will be useful, but WITHOUT
+//  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+//  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+//  more details.
+//
+//  You should have received a copy of the GNU General Public License along
+//  with ExaStencils. If not, see <http://www.gnu.org/licenses/>.
+//
+//=============================================================================
+
+package exastencils.app.ir
+
+import exastencils.app.LayerHandler
+import exastencils.applications.ir.IR_HandleMainApplication
+import exastencils.applications.swe.ir.IR_ResolveStationFunctions
+import exastencils.base.ExaRootNode
+import exastencils.base.ir._
+import exastencils.baseExt.ir._
+import exastencils.boundary.ir.IR_ResolveBoundaryFunctions
+import exastencils.communication.DefaultNeighbors
+import exastencils.communication.ir._
+import exastencils.config._
+import exastencils.domain.ir._
+import exastencils.experimental.ir.IR_ResolveGismoFunctions
+import exastencils.field.ir._
+import exastencils.globals.ir._
+import exastencils.grid.ir._
+import exastencils.interfacing.ir._
+import exastencils.knowledge.ir._
+import exastencils.layoutTransformation.ir.IR_LayoutTansformation
+import exastencils.operator.ir.IR_ApplyOffsetToStencilFieldAccess
+import exastencils.optimization.ir._
+import exastencils.parallelization.api.cuda._
+import exastencils.parallelization.api.mpi._
+import exastencils.parallelization.api.omp._
+import exastencils.performance.ir.IR_AddPerformanceEstimates
+import exastencils.polyhedron._
+import exastencils.prettyprinting.PrintToFile
+import exastencils.solver.ir._
+import exastencils.stencil.ir._
+import exastencils.timing.ir._
+import exastencils.util._
+import exastencils.util.ir._
+import exastencils.visualization.ir._
+
+/// IR_LayerHandler
+
+trait IR_LayerHandler extends LayerHandler
+
+/// IR_DummyLayerHandler
+
+object IR_DummyLayerHandler extends IR_LayerHandler {
+  def initialize() : Unit = {}
+  def handle() : Unit = {}
+  def print() : Unit = {}
+  def shutdown() : Unit = {}
+}
+
+/// IR_DefaultLayerHandler
+
+object IR_DefaultLayerHandler extends IR_LayerHandler {
+  override def initialize() : Unit = {
+    // TODO: use KnowledgeContainer structure
+  }
+
+  override def shutdown() : Unit = {
+    // TODO: use KnowledgeContainer structure; remove IR_ClearKnowledge
+    IR_ClearKnowledge.apply()
+  }
+
+  override def print() : Unit = {
+    PrintToFile.apply()
+  }
+
+  override def handle() : Unit = {
+    IR_ProcessInlineKnowledge.apply()
+
+    // add globals - init mpi before cuda since cuda might need mpiRank to choose device
+    if (Knowledge.mpi_enabled)
+      MPI_AddGlobals.apply()
+    if (Knowledge.cuda_enabled)
+      CUDA_AddGlobals.apply()
+
+    DefaultNeighbors.setup()
+    IR_GlobalCollection.get += IR_AllocateDataFunction(IR_FieldCollection.objects, DefaultNeighbors.neighbors)
+    IR_ExternalFieldCollection.generateCopyFunction().foreach(IR_UserFunctions.get += _)
+
+    // setup transformations for communication
+    if (Knowledge.comm_enableCommTransformations)
+      IR_CommTransformationCollection.setup()
+
+    // add remaining nodes
+    ExaRootNode.ir_root.nodes ++= List(
+      // FunctionCollections
+      IR_UtilFunctions(),
+      IR_DomainFunctions(),
+      IR_CommunicationFunctions(),
+
+      // Util
+      IR_Stopwatch(),
+      IR_TimerFunctions(),
+      CImg() // TODO: only if required
+    )
+
+    if (Knowledge.cuda_enabled)
+      ExaRootNode.ir_root.nodes += CUDA_KernelFunctions()
+
+    if (Knowledge.experimental_mergeCommIntoLoops)
+      IR_MergeCommunicateAndLoop.apply()
+    IR_GeneralSimplify.doUntilDone() // removes (conditional) calls to communication functions that are not possible
+    IR_SetupCommunication.firstCall = true
+    IR_SetupCommunication.apply()
+
+    IR_InferDiagAndInverseCallDataTypes.doUntilDone()
+    IR_HandleMainApplication.apply()
+    IR_ResolveBoundaryFunctions.apply()
+    IR_ResolveReadParameters.apply()
+    IR_ResolveStationFunctions.apply()
+    IR_ResolveCImgFunctions.apply()
+    IR_ResolveCharacteristicsFunctions.apply()
+    IR_ResolveBenchmarkFunctions.apply()
+    IR_ResolveGismoFunctions.apply()
+    IR_ResolveVtkPrinters.apply()
+    IR_ResolvePrintWithReducedPrec.apply()
+    IR_AdaptTimerFunctions.apply()
+
+    if (Knowledge.useFasterExpand)
+      IR_ExpandInOnePass.apply()
+    else
+      IR_Expand.doUntilDone()
+
+    // HACK: create discr_h* again if there are no multigrid level and the field size was defined explicitly
+    //   currently this works only if all fields are equally sized
+    if (Knowledge.domain_rect_generate && Knowledge.maxLevel <= 0) {
+      def globalSize = IR_DomainCollection.getByIdentifier("global").get.asInstanceOf[IR_DomainFromAABB].aabb
+
+      val fLayout : Array[IR_FieldLayoutPerDim] = IR_FieldCollection.objects.head.layout.layoutsPerDim
+      Knowledge.discr_hx = Array[Double](globalSize.width(0) /
+        (Knowledge.domain_rect_numFragsTotal_x * Knowledge.domain_fragmentLength_x * fLayout(0).numInnerLayers))
+      if (Knowledge.dimensionality > 1)
+        Knowledge.discr_hy = Array[Double](globalSize.width(1) /
+          (Knowledge.domain_rect_numFragsTotal_y * Knowledge.domain_fragmentLength_y * fLayout(1).numInnerLayers))
+      if (Knowledge.dimensionality > 2)
+        Knowledge.discr_hz = Array[Double](globalSize.width(2) /
+          (Knowledge.domain_rect_numFragsTotal_z * Knowledge.domain_fragmentLength_z * fLayout(2).numInnerLayers))
+    }
+
+    IR_ResolveIntegrateOnGrid.apply()
+    IR_ResolveEvaluateOnGrid.apply()
+    IR_ResolveVirtualFieldAccesses.apply()
+
+    IR_ResolveLoopOverPoints.apply()
+    IR_ResolveIntergridIndices.apply()
+    IR_ApplyOffsetToFieldAccess.apply()
+    IR_ApplyOffsetToStencilFieldAccess.apply()
+    // simplify indices modified just now, otherwise equality checks will not work later on
+    IR_GeneralSimplify.apply()
+
+    var convChanged = false
+    do {
+      IR_FindStencilConvolutions.changed = false
+      IR_FindStencilConvolutions.apply()
+      convChanged = IR_FindStencilConvolutions.changed
+
+      IR_WrapStencilConvolutions.apply()
+
+      if (Knowledge.useFasterExpand)
+        IR_ExpandInOnePass.apply()
+      else
+        IR_Expand.doUntilDone()
+    } while (convChanged)
+
+    IR_ResolveStencilFunction.apply()
+
+    if (Knowledge.experimental_visit_enable)
+      IR_SetupVisit.apply()
+
+    // resolve new virtual field accesses
+    IR_ResolveIntegrateOnGrid.apply()
+    IR_ResolveEvaluateOnGrid.apply()
+    IR_ResolveVirtualFieldAccesses.apply()
+    IR_ApplyOffsetToFieldAccess.apply()
+    IR_ApplyOffsetToStencilFieldAccess.apply()
+
+    IR_ResolveLoopOverPointsInOneFragment.apply()
+
+    IR_ResolveLocalSolve.apply()
+    IR_GeneralSimplify.doUntilDone()
+
+    IR_ExtractMatrices.apply()
+    IR_SetupMatrixExpressions.apply()
+    var sthChanged = true
+    while (sthChanged) {
+      // TODO: move matrix and vector specific parts of IR_GeneralSimplify to specialized strategy
+      IR_GeneralSimplify.doUntilDone()
+      IR_ResolveMatrixFunctions.apply()
+      sthChanged = IR_ResolveMatrixFunctions.results.last._2.matches > 0
+    }
+    IR_GeneralSimplify.doUntilDone()
+    IR_ResolveMatrixAssignments.apply()
+    IR_LinearizeMatrices.apply()
+
+    IR_SetupCommunication.apply() // handle communication statements generated by loop resolution
+
+    IR_TypeInference.warnMissingDeclarations = false
+    IR_TypeInference.apply() // first sweep to allow for VariableAccess extraction in SplitLoopsForHostAndDevice
+
+    if (Knowledge.performance_addEstimation)
+      IR_AddPerformanceEstimates.apply()
+
+    // Prepare all suitable LoopOverDimensions and ContractingLoops. This transformation is applied before resolving
+    // ContractingLoops to guarantee that memory transfer statements appear only before and after a resolved
+    // ContractingLoop (required for temporal blocking). Leads to better device memory occupancy.
+    if (Knowledge.cuda_enabled) {
+      CUDA_PrepareHostCode.apply()
+      CUDA_PrepareMPICode.apply()
+    }
+
+    IR_ResolveContractingLoop.apply()
+
+    IR_SetupCommunication.apply() // handle communication statements generated by loop resolution
+
+    IR_MapStencilAssignments.apply()
+    IR_ResolveFieldAccess.apply()
+
+    if (Knowledge.useFasterExpand)
+      IR_ExpandInOnePass.apply()
+    else
+      IR_Expand.doUntilDone()
+
+    IR_ResolveLoopOverFragments.apply()
+
+    // resolve constant IVs before applying poly opt
+    IR_ResolveConstIVs.apply()
+    IR_SimplifyFloatExpressions.apply()
+    IR_GeneralSimplify.doUntilDone()
+
+    if (Knowledge.opt_conventionalCSE || Knowledge.opt_loopCarriedCSE) {
+      DuplicateNodes.instances.clear()
+      DuplicateNodes.printStack = false
+      DuplicateNodes.apply() // FIXME: only debug
+      IR_Inlining.apply(true)
+      IR_CommonSubexpressionElimination.apply()
+    }
+
+    IR_MergeConditions.apply()
+    if (Knowledge.poly_optLevel_fine > 0)
+      IR_PolyOpt.apply()
+    IR_ResolveLoopOverDimensions.apply()
+
+    IR_TypeInference.apply() // second sweep for any newly introduced nodes - TODO: check if this is necessary
+
+    // Apply CUDA kernel extraction after polyhedral optimizations to work on optimized ForLoopStatements
+    if (Knowledge.cuda_enabled) {
+      CUDA_AnnotateLoop.apply()
+      CUDA_ExtractHostAndDeviceCode.apply()
+      CUDA_AdaptKernelDimensionality.apply()
+      CUDA_HandleReductions.apply()
+    }
+
+    IR_LayoutTansformation.apply()
+
+    // before converting kernel functions -> requires linearized accesses
+    IR_LinearizeDirectFieldAccess.apply()
+    IR_LinearizeExternalFieldAccess.apply()
+    IR_LinearizeTempBufferAccess.apply()
+    CUDA_LinearizeReductionDeviceDataAccess.apply()
+    IR_LinearizeLoopCarriedCSBufferAccess.apply()
+
+    IR_SimplifyModulo.apply()
+