[WIP] Add Branson Monte Carlo Kernel

src/CMakeLists.txt

@@ -47,6 +47,8 @@ blt_add_executable(
   NAME raja-perf-omptarget.exe
   SOURCES RAJAPerfSuiteDriver.cpp
   apps/AppsData.cpp
+  apps/TRANSPORT3DMC.cpp
+  apps/TRANSPORT3DMC-Seq.cpp
   apps/CONVECTION3DPA.cpp
   apps/CONVECTION3DPA-Seq.cpp
   apps/DEL_DOT_VEC_2D.cpp

src/apps/CMakeLists.txt

@@ -10,6 +10,8 @@
 blt_add_library(
   NAME apps
   SOURCES AppsData.cpp
+          TRANSPORT3DMC.cpp
+          TRANSPORT3DMC-Seq.cpp
           CONVECTION3DPA.cpp
           CONVECTION3DPA-Cuda.cpp
           CONVECTION3DPA-Hip.cpp

src/apps/TRANSPORT3DMC-Seq.cpp

@@ -0,0 +1,100 @@
+//~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~//
+// Copyright (c) Lawrence Livermore National Security, LLC and other
+// RAJA Project Developers. See top-level LICENSE and COPYRIGHT
+// files for dates and other details. No copyright assignment is required
+// to contribute to RAJA Performance Suite.
+//
+// SPDX-License-Identifier: (BSD-3-Clause)
+//~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~//
+
+#include "TRANSPORT3DMC.hpp"
+
+#include "RAJA/RAJA.hpp"
+
+#include <iostream>
+
+namespace rajaperf
+{
+namespace apps
+{
+
+
+void TRANSPORT3DMC::runSeqVariant(VariantID vid)
+{
+  const Index_type run_reps = getRunReps();
+  const Index_type ibegin = 0;
+  const Index_type iend = getActualProblemSize();
+
+  TRANSPORT3DMC_DATA_SETUP;
+
+#if defined(RUN_RAJA_SEQ)
+  auto transport3dmc_lam = [=](Index_type i) {
+                             TRANSPORT3DMC_BODY;
+                           };
+#endif
+
+  switch ( vid ) {
+
+    case Base_Seq : {
+
+      startTimer();
+      // Loop counter increment uses macro to quiet C++20 compiler warning
+      for (RepIndex_type irep = 0; irep < run_reps; RP_REPCOUNTINC(irep)) {
+
+        for (Index_type i = ibegin; i < iend; ++i ) {
+          TRANSPORT3DMC_BODY;
+        }
+
+      }
+      stopTimer();
+
+      break;
+    }
+
+#if defined(RUN_RAJA_SEQ)
+    case Lambda_Seq : {
+
+      startTimer();
+      // Loop counter increment uses macro to quiet C++20 compiler warning
+      for (RepIndex_type irep = 0; irep < run_reps; RP_REPCOUNTINC(irep)) {
+
+        for (Index_type i = ibegin; i < iend; ++i ) {
+          transport3dmc_lam(i);
+        }
+
+      }
+      stopTimer();
+
+      break;
+    }
+
+    case RAJA_Seq : {
+
+      auto res{getHostResource()};
+
+      startTimer();
+      // Loop counter increment uses macro to quiet C++20 compiler warning
+      for (RepIndex_type irep = 0; irep < run_reps; RP_REPCOUNTINC(irep)) {
+
+        RAJA::forall<RAJA::seq_exec>( res,
+          RAJA::RangeSegment(ibegin, iend), transport3dmc_lam);
+
+      }
+      stopTimer();
+
+      break;
+    }
+#endif // RUN_RAJA_SEQ
+
+    default : {
+      getCout() << "\n  TRANSPORT3DMC : Unknown variant id = " << vid << std::endl;
+    }
+
+  }
+
+}
+
+RAJAPERF_DEFAULT_TUNING_DEFINE_BOILERPLATE(TRANSPORT3DMC, Seq, Base_Seq, Lambda_Seq, RAJA_Seq)
+
+} // end namespace apps
+} // end namespace rajaperf

src/apps/TRANSPORT3DMC.cpp

@@ -0,0 +1,82 @@
+//~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~//
+// Copyright (c) Lawrence Livermore National Security, LLC and other
+// RAJA Project Developers. See top-level LICENSE and COPYRIGHT
+// files for dates and other details. No copyright assignment is required
+// to contribute to RAJA Performance Suite.
+//
+// SPDX-License-Identifier: (BSD-3-Clause)
+//~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~//
+
+#include "TRANSPORT3DMC.hpp"
+
+#include "RAJA/RAJA.hpp"
+
+#include "common/DataUtils.hpp"
+
+namespace rajaperf
+{
+namespace apps
+{
+
+
+TRANSPORT3DMC::TRANSPORT3DMC(const RunParams& params)
+  : KernelBase(rajaperf::Apps_TRANSPORT3DMC, params)
+{
+  setDefaultProblemSize(1000000);
+  setDefaultReps(1000);
+
+  setSize(params.getTargetSize(getDefaultProblemSize()),
+          params.getReps(getDefaultReps()));
+
+  setChecksumConsistency(ChecksumConsistency::ConsistentPerVariantTuning);
+  setChecksumTolerance(ChecksumTolerance::tight);
+
+  setComplexity(Complexity::N);
+
+  setMaxPerfectLoopDimensions(1);
+  setProblemDimensionality(1);
+
+  setUsesFeature(Forall);
+
+  addVariantTunings();
+}
+
+void TRANSPORT3DMC::setSize(Index_type target_size, Index_type target_reps)
+{
+  setActualProblemSize( target_size );
+  setRunReps( target_reps );
+
+  setItsPerRep( getActualProblemSize() );
+  setKernelsPerRep(1);
+
+  setBytesAllocatedPerRep( 2*sizeof(Real_type) * getActualProblemSize() ); // in, out
+  setBytesReadPerRep( 1*sizeof(Real_type) * getActualProblemSize() ); // in
+  setBytesWrittenPerRep( 1*sizeof(Real_type) * getActualProblemSize() ); // out
+  setBytesModifyWrittenPerRep( 0 );
+  setBytesAtomicModifyWrittenPerRep( 0 );
+  setFLOPsPerRep( 0 );
+}
+
+TRANSPORT3DMC::~TRANSPORT3DMC()
+{
+}
+
+void TRANSPORT3DMC::setUp(VariantID vid, size_t RAJAPERF_UNUSED_ARG(tune_idx))
+{
+  allocAndInitData(m_in, getActualProblemSize(), vid);
+  allocAndInitDataConst(m_out, getActualProblemSize(), 0.0, vid);
+}
+
+void TRANSPORT3DMC::updateChecksum(VariantID vid, size_t RAJAPERF_UNUSED_ARG(tune_idx))
+{
+  addToChecksum(m_out, getActualProblemSize(), vid);
+}
+
+void TRANSPORT3DMC::tearDown(VariantID vid, size_t RAJAPERF_UNUSED_ARG(tune_idx))
+{
+  deallocData(m_in, vid);
+  deallocData(m_out, vid);
+}
+
+} // end namespace apps
+} // end namespace rajaperf

src/apps/TRANSPORT3DMC.hpp

@@ -0,0 +1,65 @@
+//~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~//
+// Copyright (c) Lawrence Livermore National Security, LLC and other
+// RAJA Project Developers. See top-level LICENSE and COPYRIGHT
+// files for dates and other details. No copyright assignment is required
+// to contribute to RAJA Performance Suite.
+//
+// SPDX-License-Identifier: (BSD-3-Clause)
+//~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~//
+
+///
+/// TRANSPORT3DMC placeholder reference implementation:
+///
+/// for (Index_type i = ibegin; i < iend; ++i ) {
+///   out[i] = in[i];
+/// }
+///
+
+#ifndef RAJAPerf_Apps_TRANSPORT3DMC_HPP
+#define RAJAPerf_Apps_TRANSPORT3DMC_HPP
+
+
+#define TRANSPORT3DMC_DATA_SETUP \
+  Real_ptr in = m_in; \
+  Real_ptr out = m_out;
+
+// Placeholder body. Replace this statement with the target kernel operation.
+#define TRANSPORT3DMC_BODY \
+  out[i] = in[i];
+
+
+#include "common/KernelBase.hpp"
+
+namespace rajaperf
+{
+class RunParams;
+
+namespace apps
+{
+
+class TRANSPORT3DMC : public KernelBase
+{
+public:
+
+  TRANSPORT3DMC(const RunParams& params);
+
+  ~TRANSPORT3DMC();
+
+  void setSize(Index_type target_size, Index_type target_reps);
+  void setUp(VariantID vid, size_t tune_idx);
+  void updateChecksum(VariantID vid, size_t tune_idx);
+  void tearDown(VariantID vid, size_t tune_idx);
+
+  void defineSeqVariantTunings();
+
+  void runSeqVariant(VariantID vid);
+
+private:
+  Real_ptr m_in;
+  Real_ptr m_out;
+};
+
+} // end namespace apps
+} // end namespace rajaperf
+
+#endif // closing endif for header file include guard

src/common/RAJAPerfSuite.cpp

@@ -83,6 +83,7 @@
 //
 // Apps kernels...
 //
+#include "apps/TRANSPORT3DMC.hpp"
 #include "apps/CONVECTION3DPA.hpp"
 #include "apps/DEL_DOT_VEC_2D.hpp"
 #include "apps/DIFFUSION3DPA.hpp"
@@ -243,6 +244,7 @@ static const std::string KernelNames [] =
 //
 // Apps kernels...
 //
+  std::string("Apps_TRANSPORT3DMC"),
   std::string("Apps_CONVECTION3DPA"),
   std::string("Apps_DEL_DOT_VEC_2D"),
   std::string("Apps_DIFFUSION3DPA"),
@@ -1158,6 +1160,11 @@ KernelBase* getKernelObject(KernelID kid,
 //
 // Apps kernels...
 //
+    case Apps_TRANSPORT3DMC : {
+       kernel = new apps::TRANSPORT3DMC(run_params);
+       break;
+    }
+
     case Apps_CONVECTION3DPA : {
        kernel = new apps::CONVECTION3DPA(run_params);
        break;

src/common/RAJAPerfSuite.hpp

@@ -143,6 +143,7 @@ enum KernelID {
 //
 // Apps kernels...
 //
+  Apps_TRANSPORT3DMC,
   Apps_CONVECTION3DPA,
   Apps_DEL_DOT_VEC_2D,
   Apps_DIFFUSION3DPA,