Kokkos-Polybench: Add GEMM kernel

src/apps-kokkos/NODAL_ACCUMULATION_3D-Kokkos.cpp

@@ -22,7 +22,6 @@ namespace apps
 void NODAL_ACCUMULATION_3D::runKokkosVariant(VariantID vid)
 {
   const Index_type run_reps = getRunReps();
-  const Index_type ibegin = 0;
   const Index_type iend = getActualProblemSize(); //m_domain->n_real_zones;
 
   NODAL_ACCUMULATION_3D_DATA_SETUP;

src/lcals-kokkos/GEN_LIN_RECUR-Kokkos.cpp

@@ -17,7 +17,6 @@ namespace lcals {
 
 void GEN_LIN_RECUR::runKokkosVariant(VariantID vid) {
   const Index_type run_reps = getRunReps();
-  const Index_type ibegin = 1;
   const Index_type iend = getActualProblemSize();
 
   GEN_LIN_RECUR_DATA_SETUP;

src/polybench-kokkos/CMakeLists.txt

@@ -10,6 +10,7 @@
 blt_add_library(
   NAME polybench-kokkos
   SOURCES
+          POLYBENCH_GEMM-Kokkos.cpp
           POLYBENCH_HEAT_3D-Kokkos.cpp
           POLYBENCH_JACOBI_1D-Kokkos.cpp
           POLYBENCH_JACOBI_2D-Kokkos.cpp

src/polybench-kokkos/POLYBENCH_GEMM-Kokkos.cpp

@@ -0,0 +1,75 @@
+//~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~//
+// 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 "POLYBENCH_GEMM.hpp"
+
+#if defined(RUN_KOKKOS)
+
+#include "common/KokkosViewUtils.hpp"
+
+#include <iostream>
+
+namespace rajaperf {
+namespace polybench {
+
+void POLYBENCH_GEMM::runKokkosVariant(VariantID vid) {
+  const Index_type run_reps = getRunReps();
+
+  POLYBENCH_GEMM_DATA_SETUP;
+
+  auto A_view = getViewFromPointer(A, ni, nk);
+  auto B_view = getViewFromPointer(B, nk, nj);
+  auto C_view = getViewFromPointer(C, ni, nj);
+
+  switch (vid) {
+    case Kokkos_Lambda: {
+
+      Kokkos::fence();
+      startTimer();
+
+      // Loop counter increment uses macro to quiet C++20 compiler warning
+      for (RepIndex_type irep = 0; irep < run_reps; RP_REPCOUNTINC(irep)) {
+
+        Kokkos::parallel_for(
+            "POLYBENCH_GEMM Kokkos_Lambda",
+            Kokkos::MDRangePolicy<Kokkos::Rank<2>,
+                                  Kokkos::IndexType<Index_type>>({0, 0},
+                                                                 {ni, nj}),
+            KOKKOS_LAMBDA(Index_type i, Index_type j) {
+              Real_type dot = 0.0;
+              C_view(i, j) *= beta;
+              for (Index_type k = 0; k < nk; ++k) {
+                dot += alpha * A_view(i, k) * B_view(k, j);
+              }
+              C_view(i, j) = dot;
+            });
+      }
+
+      Kokkos::fence();
+      stopTimer();
+
+      break;
+    }
+
+    default: {
+      std::cout << "\n  POLYBENCH_GEMM : Unknown variant id = " << vid
+                << std::endl;
+    }
+  }
+
+  moveDataToHostFromKokkosView(A, A_view, ni, nk);
+  moveDataToHostFromKokkosView(B, B_view, nk, nj);
+  moveDataToHostFromKokkosView(C, C_view, ni, nj);
+}
+
+RAJAPERF_DEFAULT_TUNING_DEFINE_BOILERPLATE(POLYBENCH_GEMM, Kokkos, Kokkos_Lambda)
+
+}  // end namespace polybench
+}  // end namespace rajaperf
+#endif  // RUN_KOKKOS

src/polybench-kokkos/POLYBENCH_HEAT_3D-Kokkos.cpp

@@ -72,16 +72,16 @@ void POLYBENCH_HEAT_3D::runKokkosVariant(VariantID vid) {
       Kokkos::fence();
       stopTimer();
 
-      moveDataToHostFromKokkosView(A, A_view, N, N, N);
-      moveDataToHostFromKokkosView(B, B_view, N, N, N);
-
       break;
     }
 
     default: {
       std::cout << "\n  POLYBENCH_HEAT_3D : Unknown variant id = " << vid
                 << std::endl;
     }
+
+    moveDataToHostFromKokkosView(A, A_view, N, N, N);
+    moveDataToHostFromKokkosView(B, B_view, N, N, N);
   }
 }
 

src/polybench-kokkos/POLYBENCH_JACOBI_1D-Kokkos.cpp

@@ -56,9 +56,6 @@ void POLYBENCH_JACOBI_1D::runKokkosVariant(VariantID vid) {
       Kokkos::fence();
       stopTimer();
 
-      moveDataToHostFromKokkosView(A, A_view, N);
-      moveDataToHostFromKokkosView(B, B_view, N);
-
       break;
     }
 
@@ -67,6 +64,9 @@ void POLYBENCH_JACOBI_1D::runKokkosVariant(VariantID vid) {
                 << std::endl;
     }
   }
+
+  moveDataToHostFromKokkosView(A, A_view, N);
+  moveDataToHostFromKokkosView(B, B_view, N);
 }
 
 RAJAPERF_DEFAULT_TUNING_DEFINE_BOILERPLATE(POLYBENCH_JACOBI_1D, Kokkos, Kokkos_Lambda)

src/polybench-kokkos/POLYBENCH_JACOBI_2D-Kokkos.cpp

@@ -62,9 +62,6 @@ void POLYBENCH_JACOBI_2D::runKokkosVariant(VariantID vid) {
       Kokkos::fence();
       stopTimer();
 
-      moveDataToHostFromKokkosView(A, A_view, N, N);
-      moveDataToHostFromKokkosView(B, B_view, N, N);
-
       break;
     }
 
@@ -73,6 +70,9 @@ void POLYBENCH_JACOBI_2D::runKokkosVariant(VariantID vid) {
                 << std::endl;
     }
   }
+
+  moveDataToHostFromKokkosView(A, A_view, N, N);
+  moveDataToHostFromKokkosView(B, B_view, N, N);
 }
 
 RAJAPERF_DEFAULT_TUNING_DEFINE_BOILERPLATE(POLYBENCH_JACOBI_2D, Kokkos, Kokkos_Lambda)

src/polybench/POLYBENCH_GEMM.hpp

@@ -104,13 +104,15 @@ class POLYBENCH_GEMM : public KernelBase
   void defineSeqVariantTunings();
   void defineOpenMPVariantTunings();
   void defineOpenMPTargetVariantTunings();
+  void defineKokkosVariantTunings();
   void defineCudaVariantTunings();
   void defineHipVariantTunings();
   void defineSyclVariantTunings();
 
   void runSeqVariant(VariantID vid);
   void runOpenMPVariant(VariantID vid);
   void runOpenMPTargetVariant(VariantID vid);
+  void runKokkosVariant(VariantID vid);
 
   template < size_t block_size >
   void runCudaVariantImpl(VariantID vid);