HIP updates

arch/eurohpc/lumi/cray-gpu/16.0.1/env.sh

@@ -0,0 +1,50 @@
+# (C) Copyright 1988- ECMWF.
+#
+# This software is licensed under the terms of the Apache Licence Version 2.0
+# which can be obtained at http://www.apache.org/licenses/LICENSE-2.0.
+# In applying this licence, ECMWF does not waive the privileges and immunities
+# granted to it by virtue of its status as an intergovernmental organisation
+# nor does it submit to any jurisdiction.
+
+# Source me to get the correct configure/build/run environment
+
+# Store tracing and disable (module is *way* too verbose)
+{ tracing_=${-//[^x]/}; set +x; } 2>/dev/null
+
+module_load() {
+  echo "+ module load $1"
+  module load $1
+}
+module_unload() {
+  echo "+ module unload $1"
+  module unload $1
+}
+
+# Unload to be certain
+module reset
+
+# Load modules
+module_load LUMI/23.09
+module_load partition/G
+module_load PrgEnv-cray/8.4.0
+module_load cce/16.0.1
+module_load cray-mpich/8.1.27
+module_load craype-network-ofi
+module_load rocm/5.2.3
+module_load buildtools/23.09
+module_load Boost/1.82.0-cpeCray-23.09
+module_load cray-python/3.10.10
+module_load cray-hdf5/1.12.2.7
+module_load craype-x86-trento
+module_load craype-accel-amd-gfx90a
+
+module list
+
+set -x
+
+export CC=cc CXX=CC FC=ftn
+
+# Restore tracing to stored setting
+{ if [[ -n "$tracing_" ]]; then set -x; else set +x; fi } 2>/dev/null
+
+export ECBUILD_TOOLCHAIN="./toolchain.cmake"

arch/eurohpc/lumi/cray-gpu/16.0.1/toolchain.cmake

@@ -0,0 +1,49 @@
+# (C) Copyright 1988- ECMWF.
+#
+# This software is licensed under the terms of the Apache Licence Version 2.0
+# which can be obtained at http://www.apache.org/licenses/LICENSE-2.0.
+# In applying this licence, ECMWF does not waive the privileges and immunities
+# granted to it by virtue of its status as an intergovernmental organisation
+# nor does it submit to any jurisdiction.
+
+####################################################################
+# COMPILER
+####################################################################
+
+set( ECBUILD_FIND_MPI OFF )
+set( ENABLE_USE_STMT_FUNC ON CACHE STRING "" )
+
+####################################################################
+# OpenMP FLAGS
+####################################################################
+
+set( OpenMP_C_FLAGS           "-fopenmp" CACHE STRING "" )
+set( OpenMP_CXX_FLAGS         "-fopenmp" CACHE STRING "" )
+set( OpenMP_Fortran_FLAGS     "-homp -hlist=aimd" CACHE STRING "" )
+set( OpenMP_C_LIB_NAMES       "craymp" CACHE STRING "" )
+set( OpenMP_CXX_LIB_NAMES     "craymp" CACHE STRING "" )
+set( OpenMP_Fortran_LIB_NAMES "craymp" CACHE STRING "" )
+set( OpenMP_craymp_LIBRARY    "/opt/cray/pe/cce/16.0.1/cce/x86_64/lib/libcraymp.so" CACHE STRING "" )
+
+####################################################################
+# OpenACC FLAGS
+####################################################################
+
+set( OpenACC_C_FLAGS "-hacc" CACHE STRING "" )
+set( OpenACC_CXX_FLAGS "-hacc" CACHE STRING "" )
+set( OpenACC_Fortran_FLAGS "-hacc" CACHE STRING "" )
+
+####################################################################
+# Compiler FLAGS
+####################################################################
+
+# General Flags (add to default)
+set(ECBUILD_Fortran_FLAGS "-hcontiguous")
+set(ECBUILD_Fortran_FLAGS "${ECBUILD_Fortran_FLAGS} -hbyteswapio")
+set(ECBUILD_Fortran_FLAGS "${ECBUILD_Fortran_FLAGS} -Wl, --as-needed")
+
+set(ECBUILD_Fortran_FLAGS_BIT "-O3 -hfp1 -hscalar3 -hvector3 -G2 -haggress -DNDEBUG")
+
+set( GPU_TARGETS "gfx90a" CACHE STRING "" )
+# select OpenMP pragma to be used 
+set( HAVE_OMP_TARGET_LOOP_CONSTRUCT_BIND_PARALLEL OFF CACHE BOOL "" )

cmake/features/OMP.cmake

@@ -1,5 +1,7 @@
 if( HAVE_OMP )
 
+  if( NOT DEFINED HAVE_OMP_TARGET_TEAMS_DISTRIBUTE )
+
     try_compile(
         HAVE_OMP_TARGET_TEAMS_DISTRIBUTE
         ${CMAKE_CURRENT_BINARY_DIR}
@@ -11,6 +13,10 @@ if( HAVE_OMP )
     ecbuild_debug_var( HAVE_OMP_TARGET_TEAMS_DISTRIBUTE )
     ecbuild_debug_var( _HAVE_OMP_TARGET_TEAMS_DISTRIBUTE_OUTPUT )
 
+  endif()
+
+  if( NOT DEFINED HAVE_OMP_TARGET_LOOP_CONSTRUCT_BIND_PARALLEL )
+
     try_compile(
         HAVE_OMP_TARGET_LOOP_CONSTRUCT_BIND_PARALLEL
         ${CMAKE_CURRENT_BINARY_DIR}
@@ -22,6 +28,10 @@ if( HAVE_OMP )
     ecbuild_debug_var( HAVE_OMP_TARGET_LOOP_CONSTRUCT_BIND_PARALLEL )
     ecbuild_debug_var( _HAVE_OMP_TARGET_LOOP_CONSTRUCT_BIND_PARALLEL_OUTPUT )
 
+  endif()
+
+  if( NOT DEFINED HAVE_OMP_TARGET_LOOP_CONSTRUCT_BIND_THREAD )
+
     try_compile(
         HAVE_OMP_TARGET_LOOP_CONSTRUCT_BIND_THREAD
         ${CMAKE_CURRENT_BINARY_DIR}
@@ -32,6 +42,8 @@ if( HAVE_OMP )
 
     ecbuild_debug_var( HAVE_OMP_TARGET_LOOP_CONSTRUCT_BIND_THREAD )
     ecbuild_debug_var( _HAVE_OMP_TARGET_LOOP_CONSTRUCT_BIND_THREAD_OUTPUT )
+
+  endif()
 
     if( HAVE_OMP_TARGET_LOOP_CONSTRUCT_BIND_PARALLEL OR HAVE_OMP_TARGET_LOOP_CONSTRUCT_BIND_THREAD )
         set( HAVE_OMP_TARGET_LOOP_CONSTRUCT ON CACHE BOOL "OpenMP target teams loop is supported" )

src/cloudsc_hip/cloudsc/cloudsc_driver.cpp

@@ -457,9 +457,9 @@ void cloudsc_driver(int numthreads, int numcols, int nproma) {
   double t2 = omp_get_wtime();
 
   printf("     NUMOMP=%d, NGPTOT=%d, NPROMA=%d, NGPBLKS=%d\n", numthreads, numcols, nproma, nblocks);
-  printf(" %+10s%+10s%+10s%+10s%+10s %+4s : %+10s%+10s\n",
-    "NUMOMP", "NGPTOT", "#GP-cols", "#BLKS", "NPROMA", "tid#", "Time(msec)", "MFlops/s");
-  double zfrac, zmflops;
+  printf(" %+10s%+10s%+10s%+10s%+10s %+4s : %+10s%+10s%+10s\n",
+    "NUMOMP", "NGPTOT", "#GP-cols", "#BLKS", "NPROMA", "tid#", "Time(msec)", "MFlops/s", "col/s");
+  double zfrac, zmflops, zthrput;
   for (int t = 0; t < numthreads; t++) {
     const double tloc = zinfo[0][t];
     const int coreid = (int) zinfo[1][t];
@@ -468,21 +468,25 @@ void cloudsc_driver(int numthreads, int numcols, int nproma) {
     zfrac = (double)igpc / (double)numcols;
     if (tloc > 0.0) {
       zmflops = 1.0e-06 * zfrac * zhpm * ((double)numcols / 100.) / tloc;
+      zthrput = (double)numcols/tloc;
     } else {
       zmflops = 0.;
+      zthrput = 0.0;
     }
-    printf(" %10d%10d%10d%10d%10d %4d : %10d%10d @ core#\n",
-	   numthreads, numcols, igpc, icalls, nproma, t, (int)(tloc * 1000.), (int)zmflops);
+    printf(" %10d%10d%10d%10d%10d %4d : %10d%10d%10d @ core#\n",
+	   numthreads, numcols, igpc, icalls, nproma, t, (int)(tloc * 1000.), (int)zmflops, (int)zthrput);
   }
   double tdiff = t2 - t1;
   zfrac = 1.0;
   if (tdiff > 0.0) {
     zmflops = 1.0e-06 * zfrac * zhpm * ((double)numcols / 100.) / tdiff;
+    zthrput = (double)numcols/tdiff;
   } else {
     zmflops = 0.0;
+    zthrput = 0.0;
   }
-  printf(" %10d%10d%10d%10d%10d %4d : %10d%10d TOTAL\n",
-	 numthreads, numcols, numcols, nblocks, nproma, -1, (int)(tdiff * 1000.), (int)zmflops);
+  printf(" %10d%10d%10d%10d%10d %4d: %10d%10d%10d TOTAL\n",
+	 numthreads, numcols, numcols, nblocks, nproma, -1, (int)(tdiff * 1000.), (int)zmflops, (int)zthrput);
 
   cloudsc_validate(klon, nlev, nclv, numcols, nproma,
 		   plude, pcovptot, prainfrac_toprfz, pfsqlf, pfsqif,

src/cloudsc_hip/cloudsc/cloudsc_driver_hoist.cpp

@@ -498,9 +498,9 @@ void cloudsc_driver(int numthreads, int numcols, int nproma) {
   double t2 = omp_get_wtime();
 
   printf("     NUMOMP=%d, NGPTOT=%d, NPROMA=%d, NGPBLKS=%d\n", numthreads, numcols, nproma, nblocks);
-  printf(" %+10s%+10s%+10s%+10s%+10s %+4s : %+10s%+10s\n",
-    "NUMOMP", "NGPTOT", "#GP-cols", "#BLKS", "NPROMA", "tid#", "Time(msec)", "MFlops/s");
-  double zfrac, zmflops;
+  printf(" %+10s%+10s%+10s%+10s%+10s %+4s : %+10s%+10s%+10s\n",
+    "NUMOMP", "NGPTOT", "#GP-cols", "#BLKS", "NPROMA", "tid#", "Time(msec)", "MFlops/s", "col/s");
+  double zfrac, zmflops, zthrput;
   for (int t = 0; t < numthreads; t++) {
     const double tloc = zinfo[0][t];
     const int coreid = (int) zinfo[1][t];
@@ -509,21 +509,25 @@ void cloudsc_driver(int numthreads, int numcols, int nproma) {
     zfrac = (double)igpc / (double)numcols;
     if (tloc > 0.0) {
       zmflops = 1.0e-06 * zfrac * zhpm * ((double)numcols / 100.) / tloc;
+      zthrput = (double)numcols/tloc;
     } else {
       zmflops = 0.;
+      zthrput = 0.0;
     }
-    printf(" %10d%10d%10d%10d%10d %4d : %10d%10d @ core#\n",
-	   numthreads, numcols, igpc, icalls, nproma, t, (int)(tloc * 1000.), (int)zmflops);
+    printf(" %10d%10d%10d%10d%10d %4d : %10d%10d%10d @ core#\n",
+           numthreads, numcols, igpc, icalls, nproma, t, (int)(tloc * 1000.), (int)zmflops, (int)zthrput);
   }
   double tdiff = t2 - t1;
   zfrac = 1.0;
   if (tdiff > 0.0) {
     zmflops = 1.0e-06 * zfrac * zhpm * ((double)numcols / 100.) / tdiff;
+    zthrput = (double)numcols/tdiff;
   } else {
     zmflops = 0.0;
+    zthrput = 0.0;
   }
-  printf(" %10d%10d%10d%10d%10d %4d : %10d%10d TOTAL\n",
-	 numthreads, numcols, numcols, nblocks, nproma, -1, (int)(tdiff * 1000.), (int)zmflops);
+  printf(" %10d%10d%10d%10d%10d %4d: %10d%10d%10d TOTAL\n",
+         numthreads, numcols, numcols, nblocks, nproma, -1, (int)(tdiff * 1000.), (int)zmflops, (int)zthrput);
 
   cloudsc_validate(klon, nlev, nclv, numcols, nproma,
 		   plude, pcovptot, prainfrac_toprfz, pfsqlf, pfsqif,

src/cloudsc_hip/cloudsc/load_state.cpp

@@ -1,5 +1,14 @@
+/*
+ * (C) Copyright 1988- ECMWF.
+ *
+ * This software is licensed under the terms of the Apache Licence Version 2.0
+ * which can be obtained at http://www.apache.org/licenses/LICENSE-2.0.
+ * In applying this licence, ECMWF does not waive the privileges and immunities
+ * granted to it by virtue of its status as an intergovernmental organisation
+ * nor does it submit to any jurisdiction.
+ */
+
 #include "load_state.h"
-//#include "yomcst_c.hpp"
 #include <iostream>
 
 #include <math.h>