Merge branch 'release-2.7.0-rc2'

Author: alazzaro

.ci/daint.cscs.ch/Jenkinsfile renamed to .ci/daint.cscs.ch/Jenkinsfile.off

@@ -27,7 +27,7 @@ if [ ! -d "${HOME}/libxsmm" ]; then
 fi
 cd "${HOME}/libxsmm"
 git fetch
-git checkout 05705477183444a82c8d9be8d7c2627efd6d67fa
+git checkout 2fe2b1a7077ddfbc9ab3b3f7ba1f5a45d52549cb
 make -j
 cd ..
 

.ci/daint.cscs.ch/ocl.build.sh

@@ -32,7 +32,7 @@ export OMP_PROC_BIND=TRUE # set thread affinity
 # OMP_NUM_THREADS is set by cmake
 
 # use default parameters (omit loading tuned parameters)
-export OPENCL_LIBSMM_SMM_PARAMS=0
+#export OPENCL_LIBSMM_SMM_PARAMS=0
 
 # document the current environment
 env |& tee -a "${STAGE_NAME}.out"

.ci/daint.cscs.ch/ocl.test.sh

@@ -57,6 +57,7 @@ jobs:
           -DMPI_EXECUTABLE_SUFFIX=.${{ matrix.mpi_suffix }} \
           -DMPIEXEC_PREFLAGS="$([ "${{ matrix.mpi_suffix }}" = "openmpi" ] && echo "-mca btl ^openib --allow-run-as-root --oversubscribe")" \
           -DLCOV_ARGS="--test-name;${{ matrix.use_mpi }}-${{ matrix.use_openmp }}-${{ matrix.use_smm }}-cpu" \
+          -DTEST_MPI_RANKS=auto \
           ..
 
     - name: Build

.github/workflows/testing-linux.yml

@@ -6,10 +6,6 @@ on:
     - 'develop'
   pull_request:
 
-# Workaround issue in Xcode 14.1/2
-env:
-  DEVELOPER_DIR: /Applications/Xcode_14.0.1.app/Contents/Developer
-
 jobs:
   build-and-test:
     runs-on: macos-latest
@@ -45,15 +41,15 @@ jobs:
         mkdir -p build
         cd build
         env \
-          CC=gcc-12 CXX=g++-12 FC=gfortran-12 \
+          CC=gcc-14 CXX=g++-14 FC=gfortran-14 \
           cmake -G Ninja \
           -DCMAKE_BUILD_TYPE=Release \
           -DUSE_${{ matrix.use_mpi }} \
           -DUSE_${{ matrix.use_openmp }} \
           -DUSE_${{ matrix.use_smm }} \
           $([ "${{ matrix.blas_impl }}" = "openblas" ] && echo '-DCMAKE_PREFIX_PATH=/usr/local/opt/openblas') \
           -DMPIEXEC_PREFLAGS="$([ "${{ matrix.mpi_suffix }}" = "openmpi" ] && echo "-mca btl ^openib --allow-run-as-root")" \
-          -DTEST_MPI_RANKS=1 \
+          -DTEST_MPI_RANKS=auto \
           ..
 
     - name: Build

.github/workflows/testing-macos.yml

@@ -6,7 +6,7 @@ fail_fast: false
 minimum_pre_commit_version: 3.2.0
 repos:
 - repo: https://github.com/astral-sh/ruff-pre-commit
-  rev: 'v0.3.2'
+  rev: 'v0.4.10'
   hooks:
   - id: ruff
     args: [ --fix, --exit-non-zero-on-fix ]
@@ -15,13 +15,13 @@ repos:
         .cp2k/.*|
       )$
 - repo: https://github.com/psf/black
-  rev: 24.2.0
+  rev: 24.4.2
   hooks:
   - id: black
     name: Reformat Python files with the black code formatter
     files: '^.*(/PACKAGE)|(\.py)$'
 - repo: https://github.com/pre-commit/pre-commit-hooks
-  rev: v4.5.0
+  rev: v4.6.0
   hooks:
   - id: check-ast
   - id: check-yaml

.pre-commit-config.yaml

@@ -0,0 +1,8 @@
+/*------------------------------------------------------------------------------------------------*/
+/* Copyright (C) by the DBCSR developers group - All rights reserved                              */
+/* This file is part of the DBCSR library.                                                        */
+/*                                                                                                */
+/* For information on the license, see the LICENSE file.                                          */
+/* For further information please visit https://dbcsr.cp2k.org                                    */
+/* SPDX-License-Identifier: BSD-3-Clause                                                          */
+/*------------------------------------------------------------------------------------------------*/

.pre-commit/headers/c_cpp.3

@@ -82,7 +82,7 @@ cmake_dependent_option(WITH_EXAMPLES "Build the examples" ON "USE_MPI" OFF
 )# all examples require MPI
 
 set(TEST_MPI_RANKS
-    "auto"
+    2
     CACHE STRING "Number of MPI ranks for testing")
 set(TEST_OMP_THREADS
     2

CMakeLists.txt

@@ -1,8 +1,8 @@
 MAJOR = 2
 MINOR = 7
-PATCH = 0-rc1
+PATCH = 0-rc2
 # A specific DATE (YYYY-MM-DD) fixes an official release, otherwise
 # it is considered Development version.
-DATE  = 2024-03-13
+DATE  = 2024-06-27
 
 

VERSION

@@ -1,9 +1,12 @@
 if (CMAKE_Fortran_COMPILER_ID STREQUAL "GNU")
-  set(CMAKE_Fortran_FLAGS          "${CMAKE_Fortran_FLAGS} -ffree-form -std=f2008ts -fimplicit-none -Werror=aliasing -Werror=ampersand -Werror=c-binding-type -Werror=intrinsic-shadow -Werror=intrinsics-std -Werror=line-truncation -Werror=tabs -Werror=target-lifetime -Werror=underflow -Werror=unused-but-set-parameter -Werror=unused-but-set-variable -Werror=unused-variable -Werror=unused-dummy-argument -Werror=conversion -Werror=zerotrip -Werror=uninitialized -Wno-maybe-uninitialized -Werror=unused-parameter")
-  if (CMAKE_CXX_COMPILER_VERSION VERSION_LESS 10)
-    set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} -Werror=argument-mismatch")  # gcc 10+ has this automatically
+  set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} -ffree-form -std=f2008ts -fimplicit-none -Werror=aliasing -Werror=ampersand -Werror=c-binding-type -Werror=intrinsic-shadow -Werror=intrinsics-std -Werror=line-truncation -Werror=tabs -Werror=target-lifetime -Werror=underflow -Werror=unused-but-set-parameter -Werror=unused-but-set-variable -Werror=unused-variable -Werror=unused-dummy-argument -Werror=conversion -Werror=zerotrip -Wno-maybe-uninitialized -Werror=unused-parameter")
+  if (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 10) # comparison against CXX version rather than GFortran version
+    set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} -fallow-argument-mismatch") # required for 10+ (MPI wrap)
    else ()
-    set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} -fallow-argument-mismatch") # requires for 10+ for the MPI wrap module
+    set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} -Werror=argument-mismatch")  # gcc 10+ has this automatically
+  endif ()
+  if (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 13) # comparison against CXX version rather than GFortran version
+    set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} -Wno-error=uninitialized") # false positive (allocatable array)
   endif ()
   set(CMAKE_Fortran_FLAGS_RELEASE  "-O3 -g -funroll-loops")
   set(CMAKE_Fortran_FLAGS_COVERAGE "-O0 -g --coverage -fno-omit-frame-pointer -fcheck=all,no-array-temps -ffpe-trap=invalid,zero,overflow -fbacktrace -finit-real=snan -finit-integer=-42 -finit-derived -Werror=realloc-lhs -finline-matmul-limit=0 -Werror")
@@ -48,6 +51,9 @@ if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
   if ((NOT (USE_MPI)) OR (NOT ("${MPI_Fortran_LIBRARY_VERSION_STRING}" MATCHES "Open MPI")))
     set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -fsanitize=leak")
   endif ()
+  if (USE_ACCEL MATCHES "hip" AND  hip_VERSION GREATER_EQUAL 6.0.0) # Remove deprecated function error with ROCm v6+
+     set(CMAKE_CXX_FLAGS            "${CMAKE_CXX_FLAGS} -Wno-error=deprecated-declarations")
+  endif	()
 elseif (CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
   set(CMAKE_CXX_FLAGS_RELEASE      "-O3 -funroll-loops")
   set(CMAKE_CXX_FLAGS_COVERAGE     "-O0 -g --coverage")

cmake/CompilerConfiguration.cmake

@@ -76,8 +76,8 @@ make
 -DWITH_GPU=<P100|K20X|K40|K80|V100|Mi50|Mi100|Mi250>
 -DCMAKE_BUILD_TYPE=<Release|Debug|Coverage>
 -DBUILD_TESTING=<ON|OFF>
--DTEST_MPI_RANKS=<auto,N>
--DTEST_OMP_THREADS=<2,N>
+-DTEST_MPI_RANKS=<2|auto|N>
+-DTEST_OMP_THREADS=<2|N>
 ```
 
 When providing a build of LIBXSMM, make sure the `lib` directory is added to the `PKG_CONFIG_PATH` variable prior

docs/guide/2-user-guide/1-installation/index.md

@@ -8,6 +8,9 @@ foreach (dbcsr_program_src ${DBCSR_PROGRAM_SRCS_FTN})
   get_filename_component(dbcsr_program_name ${dbcsr_program_src} NAME_WE)
   add_executable(${dbcsr_program_name} ${dbcsr_program_src})
   target_link_libraries(${dbcsr_program_name} dbcsr)
+  if (OpenMP_FOUND)
+    target_link_libraries(${dbcsr_program_name} OpenMP::OpenMP_Fortran)
+  endif ()
 
   # with the Intel compiler CMake 3.12 seems to forget that the source is
   # actually Fortran and needs to be told explicitly:
@@ -29,6 +32,10 @@ if (WITH_C_API)
     set(dbcsr_program_name ${dbcsr_program_name}_cpp)
     add_executable(${dbcsr_program_name} ${dbcsr_program_src})
     target_link_libraries(${dbcsr_program_name} dbcsr_c MPI::MPI_CXX)
+    set_target_properties(${dbcsr_program_name} PROPERTIES LINKER_LANGUAGE CXX)
+    if (OpenMP_FOUND)
+      target_link_libraries(${dbcsr_program_name} OpenMP::OpenMP_CXX)
+    endif ()
 
     if (CMAKE_CXX_COMPILER_ID STREQUAL "Cray")
       # for recent Cray compiler versions CMake doesn't know

examples/CMakeLists.txt

@@ -53,8 +53,8 @@ int main(int argc, char* argv[]) {
 
   for (int i = 0; i != mpi_size; ++i) {
     if (mpi_rank == i) {
-      std::cout << "I'm processor " << mpi_rank << " over " << mpi_size << " proc" << ", (" << coord[0] << ", " << coord[1]
-                << ") in the 2D grid" << std::endl;
+      std::cout << "I'm processor " << mpi_rank << " over " << mpi_size << " proc"
+                << ", (" << coord[0] << ", " << coord[1] << ") in the 2D grid" << std::endl;
     }
     MPI_Barrier(MPI_COMM_WORLD);
   }

examples/dbcsr_example_3.cpp

@@ -25,6 +25,10 @@
 #  define INLINE
 #endif
 
+#if !defined(MAX_KERNEL_DIM)
+#  define MAX_KERNEL_DIM 80
+#endif
+
 #define INIT_MAT(ELEM_TYPE, SEED, MAT, M, N, SCALE) \
   do { \
     const double init_mat_seed1_ = (SCALE) * (SEED) + (SCALE); \

src/acc/acc_bench.h

@@ -39,7 +39,7 @@
     STRIDE_C, INDEX_STRIDE, INDEX_BASE, BATCHSIZE) \
     ACC_BENCH_USEOMP(libxsmm_gemm_batch) \
     (IPREC, OPREC, TRANSA, TRANSB, M, N, K, ALPHA, A, LDA, STRIDE_A, B, LDB, STRIDE_B, BETA, C, LDC, STRIDE_C, INDEX_STRIDE, \
-      INDEX_BASE, BATCHSIZE, 0 /*batchcheck*/)
+      INDEX_BASE, BATCHSIZE)
 #  define PRINTF(...) \
     do { \
       const size_t print_buffer_size = sizeof(print_buffer) - print_offset; \
@@ -227,9 +227,7 @@ int main(int argc, char* argv[]) {
     int ndevices = 0;
     result = c_dbcsr_acc_get_ndevices(&ndevices);
     if (0 < ndevices && (0 == device || EXIT_SUCCESS == c_dbcsr_acc_set_active_device(device))) {
-#if defined(_DEBUG)
-      fprintf(stderr, "Activated device %i of %i (device%i).\n", device + 1, ndevices, device);
-#endif
+      printf("Activated device%i (ndevices=%i)\n", device, ndevices);
     }
     else {
       if (0 >= ndevices) {

src/acc/acc_bench_smm.c

@@ -123,9 +123,7 @@ int main(int argc, char* argv[]) {
   if (EXIT_SUCCESS == result) {
     result = c_dbcsr_acc_get_ndevices(&ndevices);
     if (0 < ndevices && (0 == device || EXIT_SUCCESS == c_dbcsr_acc_set_active_device(device))) {
-#if defined(_DEBUG)
-      fprintf(stderr, "Activated device %i of %i (device%i).\n", device + 1, ndevices, device);
-#endif
+      printf("Activated device%i (ndevices=%i)\n", device, ndevices);
     }
     else {
       if (0 >= ndevices) {

src/acc/acc_bench_trans.c

@@ -22,7 +22,7 @@ OBJSMM := $(SRCSMM:.cpp=.o)
 
 INCALL := $(INCACC) $(INCSMM)
 
-LIBXSMMROOT := $(wildcard $(ACCDIR)/../../../libxsmm)
+LIBXSMMROOT := $(wildcard $(ACCDIR)/../../../../../libxsmm)
 ifeq (,$(LIBXSMMROOT))
   LIBXSMMROOT := $(wildcard $(HOME)/libxsmm)
 endif
@@ -33,6 +33,9 @@ INTEL ?= 0
 GNU ?= 0
 DEV ?= 0
 
+# C++ baseline standard
+CXXSTD ?= -std=c++14
+
 # select from set of predefined triplet specifications
 SPECID ?= 0
 # limit shape in tests (zero or negative for unlimited)
@@ -118,7 +121,7 @@ else ifneq (0,$(GNU))
   else
     override AR := ar
   endif
-  override LD_LIBRARY_DIRS := $(NULL)
+  #override LD_LIBRARY_DIRS := $(NULL)
 else
   CXX := g++
   CC := gcc
@@ -211,7 +214,7 @@ LD_LIBSTUB_PATH := $(wildcard $(patsubst %,%/stubs,$(LD_LIBRARY_DIRS)))
 LIBPATHS := $(foreach DIR,$(LD_LIBRARY_DIRS),$(if $(filter -L$(DIR),$(LDFLAGS)),$(NULL),-L$(DIR)))
 LIBSTUBS := $(foreach DIR,$(LD_LIBSTUB_PATH),$(if $(filter -L$(DIR),$(LDFLAGS)),$(NULL),-L$(DIR)))
 LDFLAGS += $(LIBPATHS) $(LIBSTUBS) -lcudart -lcublas -lnvrtc -lcuda
-CXXFLAGS += -std=c++11 $(CFLAGS)
+CXXFLAGS += $(CXXSTD) $(CFLAGS)
 
 .PHONY: bench
 bench: $(ACCDIR)/acc_bench_smm $(ACCDIR)/acc_bench_trans
@@ -296,15 +299,16 @@ libsmm: $(ACCDIR)/dbcsr_acc_smm.a
 $(ACCDIR)/dbcsr_acc_smm.a: $(OBJSMM)
 	$(AR) -rs $@ $^
 
-%.o: %.cu $(INCALL) $(MAKDIR)/Makefile
-	$(NVCC) $(DFLAGS) -allow-unsupported-compiler --compiler-options="$(CXXFLAGS) $(CFLAGS_XSMM)" -c $< -o $@
-
 %.o: %.cpp $(INCALL) $(MAKDIR)/Makefile
 	$(CXX) $(DFLAGS) $(CXXFLAGS) $(CFLAGS_XSMM) -c $< -o $@
 
+%.o: %.cu $(INCALL) $(MAKDIR)/Makefile
+	$(NVCC) $(DFLAGS) -allow-unsupported-compiler $(CXXSTD) \
+		--compiler-options="$(filter-out $(CXXSTD),$(CXXFLAGS)) $(CFLAGS_XSMM)" -c $< -o $@
+
 $(ACCDIR)/cuda_hip/calculate_norms.o: $(ACCDIR)/cuda_hip/calculate_norms.cpp $(INCALL) $(MAKDIR)/Makefile
-	$(NVCC) $(DFLAGS) -x cu -allow-unsupported-compiler \
-		--compiler-options="$(filter-out -pedantic,$(CXXFLAGS)) $(CFLAGS_XSMM)" -c $< -o $@
+	$(NVCC) $(DFLAGS) -allow-unsupported-compiler $(CXXSTD) -x cu \
+		--compiler-options="$(filter-out $(CXXSTD) -pedantic,$(CXXFLAGS)) $(CFLAGS_XSMM)" -c $< -o $@
 
 $(MAKDIR)/acc_bench_smm.o: $(ACCDIR)/acc_bench_smm.c $(MAKDIR)/Makefile
 ifneq (0,$(LIBXSMM))

src/acc/cuda/Makefile

@@ -20,11 +20,6 @@
 #include <stdio.h>
 #include <math.h>
 
-// for debug purpose
-#if defined(__HIP_PLATFORM_NVCC__)
-static const int verbose_print = 1;
-#endif
-
 /****************************************************************************/
 extern "C" int c_dbcsr_acc_get_ndevices(int* n_devices) {
   ACC_API_CALL(GetDeviceCount, (n_devices));
@@ -49,9 +44,11 @@ extern "C" int c_dbcsr_acc_set_active_device(int device_id) {
   // establish context
   ACC_API_CALL(Free, (0));
 
-#if defined(__HIP_PLATFORM_NVCC__)
-  if (verbose_print) {
+#if defined(__CUDA) || defined(__HIP_PLATFORM_NVCC__)
+  static bool once = false;
+  if (!once) {
     ACC_API_CALL(DeviceSetLimit, (ACC(LimitPrintfFifoSize), (size_t)1000000000));
+    once = true;
   }
 #endif
 

src/acc/cuda_hip/acc_dev.cpp

@@ -51,7 +51,7 @@ extern "C" int c_dbcsr_acc_stream_create(void** stream_p, const char* name, int
     cErr = ACC(StreamCreate)(acc_stream);
   }
 
-  if (verbose_print) printf("StreamCreate : %p -> %p \n", *stream_p, *acc_stream);
+  if (verbose_print) printf("StreamCreate : %p -> %p \n", *stream_p, (const void*)*acc_stream);
   if (acc_error_check(cErr)) return -1;
   if (acc_error_check(ACC(GetLastError)())) return -1;
 

src/acc/cuda_hip/acc_stream.cpp

@@ -50,6 +50,8 @@ The performance of the matrix-matrix multiplication kernels is highly dependent
 
 ## Contributing to libsmm_acc
 
+We expect users to contribute to the library by providing new optimized kernels and support for new GPUs.
+
 #### Autotuning procedure
 
 Follow the [autotuning procedure](https://github.com/cp2k/dbcsr/blob/develop/src/acc/libsmm_acc/tune/README.md)