comp-chem/cp2k
1
0
Fork 0
mirror of https://github.com/cp2k/cp2k.git synced 2026-07-28 06:05:29 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

Enable hardware support for atomic operations on mi250 (#3943)

Browse source 
Co-authored-by: Mathieu Taillefumier <mathieu.taillefumier@free.fr>
This commit is contained in:
Taillefumier Mathieu 2025-02-06 02:51:03 -05:00 • committed by GitHub
parent 6337c11533
commit 7eb093a14a
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 88 additions and 114 deletions
Download patch file Download diff file Expand all files Collapse all files

179
CMakeLists.txt
View file

@ -5,7 +5,7 @@
#! SPDX-License-Identifier: GPL-2.0-or-later !
#!-------------------------------------------------------------------------------------------------!
cmake_minimum_required(VERSION 3.22)
cmake_minimum_required(VERSION 3.24)
# include our cmake snippets
set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
@ -50,7 +50,8 @@ endif()
# set language and standard
set(CMAKE_CXX_STANDARD 14)
set(CMAKE_C_STANDARD 11)
set(CMAKE_HIP_STANDARD 11)
set(CMAKE_CUDA_STANDARD 11)
# remove NDEBUG flag
string(REPLACE "-DNDEBUG" "" CMAKE_C_FLAGS_RELEASE ${CMAKE_C_FLAGS_RELEASE})
string(REPLACE "-DNDEBUG" "" CMAKE_CXX_FLAGS_RELEASE ${CMAKE_CXX_FLAGS_RELEASE})
@ -96,8 +97,6 @@ option(CMAKE_POSITION_INDEPENDENT_CODE "Enable position independent code" ON)
option(CP2K_ENABLE_CONSISTENCY_CHECKS
"Check that the list of compiled files and files contained in src match"
OFF)
option(CP2K_DEBUG_MODE "Enable several additional options for debugging cp2k."
OFF)
option(CP2K_USE_DFTD4 "Enable dispersion coorections DFTD4" OFF)
option(CP2K_USE_DEEPMD "Enable DeePMD" OFF)
option(CP2K_USE_SIRIUS "Enable plane wave dft calculations with sirius" OFF)
@ -124,12 +123,31 @@ option(CP2K_USE_TREXIO "Enable trexio support" OFF)
option(CP2K_USE_HDF5 "Enable HDF5 support" OFF)
option(CP2K_USE_LIBXSMM "Use libxsmm for small gemms" OFF)
option(CP2K_USE_LIBSMEAGOL "Enable libsmeagol support" OFF)
# option(CP2K_USE_LIBGRPP "Use libgrpp" OFF)
option(CP2K_DBCSR_USE_CPU_ONLY "Enable compilation DBCSR CPU version" OFF)
option(BUILD_SHARED_LIBS "Build cp2k shared library" ON)
option(
CP2K_USE_FFTW3_WITH_MKL
"MKL has its own compatible implementation of the fftw. This option when ON will use the original implementation of the fftw library"
OFF)
option(
CP2K_ENABLE_GRID_GPU
"disable the hardware accelerated backend for grid related functions. It is only effective when general gpu support is enabled."
ON)
option(
CP2K_ENABLE_PW_GPU
"disable the ffts accelerated (mostly GPU) backend. It is only effective when general gpu support is enabled."
OFF)
option(
CP2K_ENABLE_DBM_GPU
"disable the dbm accelerated (mostly GPU) backend. It is only effective when general gpu support is enabled."
ON)
cmake_dependent_option(
CP2K_USE_UNIFIED_MEMORY "Use CPU/GPU unified memory.
requires Mi250x or future AMD architectures to be fully functional" OFF
"CP2K_USE_ACCEL MATCHES \"HIP\"" OFF)
cmake_dependent_option(CP2K_ENABLE_ELPA_OPENMP_SUPPORT
"Enable elpa openmp support" ON "CP2K_USE_ELPA" OFF)
cmake_dependent_option(CP2K_ENABLE_FFTW3_OPENMP_SUPPORT
@ -144,6 +162,14 @@ cmake_dependent_option(
"Use Nvidia gpu accelerated eigensolver. Only active when CUDA is ON" OFF
"CP2K_USE_ACCEL MATCHES \"CUDA\"" OFF)
cmake_dependent_option(CP2K_USE_NVHPC OFF "Enable Nvidia NVHPC kit"
"(NOT CP2K_USE_ACCEL MATCHES \"CUDA\")" OFF)
cmake_dependent_option(
CP2K_USE_SPLA_GEMM_OFFLOADING ON
"Enable SPLA dgemm offloading (only valid with gpu support on)"
"(NOT CP2K_USE_ACCEL MATCHES \"NONE\") AND (CP2K_USE_SPLA)" OFF)
set(CP2K_BLAS_VENDOR
"auto"
CACHE STRING "Blas library for computations on host")
@ -274,50 +300,6 @@ set(CP2K_USE_ACCEL
set_property(CACHE CP2K_USE_ACCEL
PROPERTY STRINGS ${CP2K_SUPPORTED_ACCELERATION_TARGETS})
cmake_dependent_option(CP2K_USE_NVHPC OFF "Enable Nvidia NVHPC kit"
"(NOT CP2K_USE_ACCEL MATCHES \"CUDA\")" OFF)
cmake_dependent_option(
CP2K_USE_SPLA_GEMM_OFFLOADING ON
"Enable SPLA dgemm offloading (only valid with gpu support on)"
"(NOT CP2K_USE_ACCEL MATCHES \"NONE\") AND (CP2K_USE_SPLA)" OFF)
# ##############################################################################
#
# GPU debug options
#
# ##############################################################################
cmake_dependent_option(
CP2K_DISABLE_GRID_GPU
OFF
"disable the hardware accelerated backend for grid related functions. It is only effective when general gpu support is enabled."
"CP2K_DEBUG_MODE"
OFF)
cmake_dependent_option(
CP2K_DISABLE_PW_GPU
OFF
"disable the ffts accelerated (mostly GPU) backend. It is only effective when general gpu support is enabled."
"CP2K_DEBUG_MODE"
OFF)
cmake_dependent_option(
CP2K_DISABLE_DBM_GPU
OFF
"disable the dbm accelerated (mostly GPU) backend. It is only effective when general gpu support is enabled."
"CP2K_DEBUG_MODE"
OFF)
cmake_dependent_option(
CP2K_DBCSR_CPU_ONLY "Use DBCSR compiled without GPU support." OFF
"(NOT CP2K_USE_ACCEL MATCHES \"NONE\") AND (CP2K_DEBUG_MODE)" OFF)
cmake_dependent_option(
CP2K_USE_UNIFIED_MEMORY "Use CPU/GPU unified memory.
requires Mi250x or future AMD architectures to be fully functional" OFF
"CP2K_USE_ACCEL MATCHES \"HIP\"" OFF)
cmake_dependent_option(DBCSR_USE_OPENMP "Enable openmp support in DBCSR" ON
"CP2K_BUILD_DBCSR" OFF)
# ##############################################################################
# specific variables for the regtests. Binaries will be created with an
# extension #
@ -538,14 +520,14 @@ elseif(CP2K_USE_ACCEL MATCHES "HIP")
if(CMAKE_HIP_PLATFORM MATCHES "nvidia")
find_package(CUDAToolkit)
endif()
if(NOT CMAKE_BUILD_TYPE)
set(HIP_RELEASE_OPTIONS "-O3 --std=c++11")
if(NOT CMAKE_BUILD_TYPE AND (CMAKE_HIP_PLATFORM MATCHES "amd"))
set(CMAKE_HIP_FLAGS "-O3")
elseif(${CMAKE_BUILD_TYPE} STREQUAL "RelWithDebInfo")
set(HIP_RELEASE_OPTIONS "-O2 -g --std=c++11")
set(CMAKE_HIP_FLAGS "-O2 -g")
elseif(${CMAKE_BUILD_TYPE} STREQUAL "Release")
set(HIP_RELEASE_OPTIONS "-O3 --std=c++11")
set(CMAKE_HIP_FLAGS "-O3")
elseif(${CMAKE_BUILD_TYPE} STREQUAL "Debug")
set(HIP_RELEASE_OPTIONS "-O0 -g --std=c++11")
set(CMAKE_HIP_FLAGS "-O0 -g")
endif()
set(CMAKE_HIP_ARCHITECTURES "${CP2K_GPU_ARCH_NUMBER_${CP2K_WITH_GPU}}")
@ -554,8 +536,8 @@ elseif(CP2K_USE_ACCEL MATCHES "HIP")
set(CP2K_USE_PW_GPU ON)
# use hardware atomic operations on Mi250X.
if(CMAKE_HIP_ARCGITECTURES MATCHES "gfx90a|gfx942")
set(HIP_RELEASE_OPTIONS "-munsafe-fp-atomics")
if(NOT CMAKE_HIP_PLATFORM OR (CMAKE_HIP_PLATFORM MATCHES "amd"))
set(CMAKE_HIP_FLAGS "${CMAKE_HIP_FLAGS} -munsafe-fp-atomics")
endif()
# add the Mi300A parameters when available
@ -776,17 +758,17 @@ message(
message(
" - BLAS AND LAPACK\n\n"
" - vendor : ${CP2K_BLAS_VENDOR}\n"
" - include directories : ${CP2K_BLAS_INCLUDE_DIR} ${LAPACK_INCLUDE_DIR}\n"
" - libraries : ${CP2K_BLAS_LINK_LIBRARIES} ${CP2K_LAPACK_LINK_LIBRARIES}\n\n"
" - vendor: ${CP2K_BLAS_VENDOR}\n"
" - include directories: ${CP2K_BLAS_INCLUDE_DIR} ${LAPACK_INCLUDE_DIR}\n"
" - libraries: ${CP2K_BLAS_LINK_LIBRARIES} ${CP2K_LAPACK_LINK_LIBRARIES}\n\n"
)
if(CP2K_USE_MPI)
message(" - MPI\n" " - include directories : ${MPI_INCLUDE_DIRS}\n"
" - libraries : ${MPI_LIBRARIES}\n\n")
message(" - MPI\n" " - include directories: ${MPI_INCLUDE_DIRS}\n"
" - libraries: ${MPI_LIBRARIES}\n\n")
if(CP2K_USE_MPI_F08)
message(" - MPI_08 : ON\n")
message(" - MPI_08: ON\n")
endif()
if(MPI_Fortran_HAVE_F08_MODULE AND NOT CP2K_USE_MPI_F08)
@ -811,13 +793,15 @@ if((CP2K_USE_ACCEL MATCHES "CUDA") OR (CP2K_USE_ACCEL MATCHES "HIP"))
if(CP2K_USE_ACCEL MATCHES "HIP")
message(" - HIP:\n" " - GPU target architecture : ${CP2K_WITH_GPU}\n"
" - GPU architecture number : ${CP2K_ACC_ARCH_NUMBER}\n\n")
" - GPU architecture number : ${CP2K_ACC_ARCH_NUMBER}\n"
" - FLAGS: ${CMAKE_HIP_FLAGS}")
endif()
message(
"Note : Enabling hardware acceleration enable acceleration of the grid, pw, and dbm modules by default\n"
" - PW module : ${CP2K_USE_PW_GPU}\n"
"\n")
" - GPU accelerated modules\n"
" - PW module : ${CP2K_ENABLE_PW_GPU}\n"
" - GRID module : ${CP2K_ENABLE_GRID_GPU}\n"
" - DBM module : ${CP2K_ENABLE_DBM_GPU}\n\n")
endif()
if(CP2K_USE_CUSOLVER_MP)
@ -833,19 +817,19 @@ if(CP2K_USE_CUSOLVER_MP)
endif()
if(CP2K_USE_LIBXC)
message(" - LibXC\n" " - Libxc_VERSION: ${Libxc_VERSION}\n"
" - Libxc_INCLUDE_DIRS : ${Libxc_INCLUDE_DIRS}\n"
" - Libxc_LIBRARIES : ${Libxc_LIBRARIES}\n\n")
message(" - LibXC\n" " - VERSION: ${Libxc_VERSION}\n"
" - include directories: ${Libxc_INCLUDE_DIRS}\n"
" - libraries: ${Libxc_LIBRARIES}\n\n")
endif()
if(CP2K_USE_LIBTORCH)
message(" - LIBTORCH\n" " - libraries : ${CP2K_LIBTORCH_LIBRARIES}\n\n")
message(" - LIBTORCH\n" " - libraries: ${CP2K_LIBTORCH_LIBRARIES}\n\n")
endif()
if(CP2K_USE_HDF5)
message(" - HDF5\n" " - HDF5_VERSION : ${HDF5_VERSION}\n"
" - HDF5_INCLUDE_DIRS : ${HDF5_INCLUDE_DIRS}\n"
" - HDF5_LIBRARIES : ${HDF5_LIBRARIES}\n\n")
message(" - HDF5\n" " - version: ${HDF5_VERSION}\n"
" - include directories: ${HDF5_INCLUDE_DIRS}\n"
" - libraries: ${HDF5_LIBRARIES}\n\n")
endif()
if(CP2K_USE_FFTW3)
@ -856,28 +840,28 @@ endif()
if(CP2K_USE_PLUMED)
message(" - PLUMED\n"
" - include directories : ${CP2K_PLUMED_INCLUDE_DIRS}\n"
" - libraries : ${CP2K_PLUMED_LINK_LIBRARIES}\n\n")
" - include directories: ${CP2K_PLUMED_INCLUDE_DIRS}\n"
" - libraries: ${CP2K_PLUMED_LINK_LIBRARIES}\n\n")
endif()
if(CP2K_USE_LIBXSMM)
message(
" - libxsmm\n"
" - include directories : ${CP2K_LIBXSMM_INCLUDE_DIRS}\n"
" - libraries : ${CP2K_LIBXSMMEXT_LINK_LIBRARIES};${CP2K_LIBXSMMF_LINK_LIBRARIES}\n\n"
" - include directories: ${CP2K_LIBXSMM_INCLUDE_DIRS}\n"
" - libraries: ${CP2K_LIBXSMMEXT_LINK_LIBRARIES};${CP2K_LIBXSMMF_LINK_LIBRARIES}\n\n"
)
endif()
if(CP2K_USE_SPLA)
message(" - SPLA\n" " - include directories : ${SPLA_INCLUDE_DIRS}\n"
" - lbraries : ${SPLA_LIBRARIES}\n\n")
message(" - SPLA\n" " - include directories: ${SPLA_INCLUDE_DIRS}\n"
" - libraries: ${SPLA_LIBRARIES}\n\n")
endif()
if(CP2K_USE_DFTD4)
message(
" - DFTD4 :\n"
" - include directories : ${CP2K_DFTD4_INCLUDE_DIR}\n"
" - libraries : ${CP2K_DFTD4_LINK_LIBRARIES};${CP2K_DFTD4_LIB_LINK_LIBRARIES}\n\n"
" - include directories: ${CP2K_DFTD4_INCLUDE_DIR}\n"
" - libraries: ${CP2K_DFTD4_LINK_LIBRARIES};${CP2K_DFTD4_LIB_LINK_LIBRARIES}\n\n"
)
endif()
@ -887,8 +871,8 @@ endif()
if(CP2K_USE_LIBSMEAGOL)
message(" - LIBSMEAGOL :\n"
" - include directories : ${CP2K_LIBSMEAGOL_INCLUDE_DIRS}\n"
" - libraries : ${CP2K_LIBSMEAGOL_LINK_LIBRARIES}\n\n")
" - include directories: ${CP2K_LIBSMEAGOL_INCLUDE_DIRS}\n"
" - libraries: ${CP2K_LIBSMEAGOL_LINK_LIBRARIES}\n\n")
endif()
if(CP2K_USE_SIRIUS)
@ -896,43 +880,42 @@ if(CP2K_USE_SIRIUS)
endif()
if(CP2K_USE_COSMA)
message(" - COSMA\n"
" - include directories : ${CP2K_COSMA_INCLUDE_DIRS}\n"
" - libraries : ${CP2K_COSMA_LINK_LIBRARIES}\n\n")
message(" - COSMA\n" " - include directories: ${CP2K_COSMA_INCLUDE_DIRS}\n"
" - libraries: ${CP2K_COSMA_LINK_LIBRARIES}\n\n")
endif()
if(CP2K_USE_LIBINT2)
message(" - libint2\n"
" - include directories : ${CP2K_LIBINT2_INCLUDE_DIRS}\n"
" - libraries : ${CP2K_LIBINT2_LINK_LIBRARIES}\n\n")
" - include directories: ${CP2K_LIBINT2_INCLUDE_DIRS}\n"
" - libraries: ${CP2K_LIBINT2_LINK_LIBRARIES}\n\n")
endif()
if(CP2K_USE_VORI)
message(" - libvori\n"
" - include directories : ${CP2K_LIBVORI_INCLUDE_DIRS}\n"
" - libraries : ${CP2K_LIBVORI_LINK_LIBRARIES}\n\n")
" - include directories: ${CP2K_LIBVORI_INCLUDE_DIRS}\n"
" - libraries: ${CP2K_LIBVORI_LINK_LIBRARIES}\n\n")
endif()
if(CP2K_USE_ELPA)
message(" - ELPA\n" " - include directories : ${CP2K_ELPA_INCLUDE_DIRS}\n"
" - libraries : ${CP2K_ELPA_LINK_LIBRARIES}\n\n")
message(" - ELPA\n" " - include directories: ${CP2K_ELPA_INCLUDE_DIRS}\n"
" - libraries: ${CP2K_ELPA_LINK_LIBRARIES}\n\n")
endif()
if(CP2K_USE_DLAF)
message(" - DLA-Future\n"
" - include directories : ${CP2K_DLAF_INCLUDE_DIRS}\n"
" - libraries : ${CP2K_DLAF_LINK_LIBRARIES}\n\n")
" - include directories: ${CP2K_DLAF_INCLUDE_DIRS}\n"
" - libraries: ${CP2K_DLAF_LINK_LIBRARIES}\n\n")
endif()
if(CP2K_USE_GRPP)
message(" - grpp\n" " - include directories : ${GRPP_INCLUDE_DIRS}\n"
" - libraries : ${GRPP_LINK_LIBRARIES}\n\n")
message(" - grpp\n" " - include directories: ${GRPP_INCLUDE_DIRS}\n"
" - libraries: ${GRPP_LINK_LIBRARIES}\n\n")
endif()
if(CP2K_USE_TREXIO)
message(" - trexio\n"
" - include directories : ${CP2K_TREXIO_INCLUDE_DIRS}\n"
" - libraries : ${CP2K_TREXIO_LINK_LIBRARIES}\n\n")
" - include directories: ${CP2K_TREXIO_INCLUDE_DIRS}\n"
" - libraries: ${CP2K_TREXIO_LINK_LIBRARIES}\n\n")
endif()
message(

4
cmake/cp2kConfig.cmake.in
View file

@ -36,7 +36,7 @@ if(NOT TARGET cp2k::cp2k)
set(cp2k_VERSION @cp2k_VERSION@)
find_dependency(DBCSR 2.5 REQUIRED)
find_dependency(DBCSR 2.8 REQUIRED)
if(@CP2K_USE_LIBXSMM@)
find_dependency(LibXSMM REQUIRED)
@ -115,8 +115,6 @@ if(NOT TARGET cp2k::cp2k)
find_dependency(libsmeagol REQUIRED)
endif()
find_dependency(dbcsr 2.6 REQUIRED)
if(@CP2K_USE_VORI@)
find_dependency(LibVORI REQUIRED)
endif()

19
src/CMakeLists.txt
View file

@ -1445,17 +1445,17 @@ set(CP2K_PROGS_F
nequip_unittest.F)
if(CP2K_USE_CUDA OR CP2K_USE_HIP)
if(NOT CP2K_DISABLE_PW_GPU)
if(CP2K_ENABLE_PW_GPU)
list(APPEND CP2K_SRCS_GPU "pw/gpu/pw_gpu_kernels.cu")
list(APPEND CP2K_SRCS_C ${CP2K_PW_SRCS_C})
endif()
if(NOT CP2K_DISABLE_DBM_GPU)
if(CP2K_ENABLE_DBM_GPU)
list(APPEND CP2K_DBM_SRCS_C ${CP2K_DBM_SRCS_GPU_C})
list(APPEND CP2K_SRCS_GPU ${CP2K_DBM_SRCS_GPU})
endif()
if(NOT CP2K_DISABLE_GRID_GPU)
if(CP2K_ENABLE_GRID_GPU)
if(CP2K_USE_HIP)
list(APPEND CP2K_SRCS_GPU ${CP2K_GRID_SRCS_HIP})
endif()
@ -1515,9 +1515,6 @@ if(CP2K_USE_HIP)
INTERFACE hip::hipfft roc::hipblas hip::host
$<$<STREQUAL:"${CMAKE_HIP_PLATFORM}","nvidia">:CUDA::toolkit
CUDA::cudart CUDA::cuda_driver>)
# if (CMAKE_HIP_PLATFORM MATCHES "nvidia") include_directories("
# -I$ENV{ROCM_PATH}/include") include_directories("
# -I${CUDAToolkit_INCLUDE_DIRS}") endif()
endif()
add_library(cp2k_linalg_libs INTERFACE)
target_link_libraries(
@ -1585,8 +1582,6 @@ set_target_properties(
if(CP2K_USE_ACCEL MATCHES "CUDA")
set_property(TARGET cp2k PROPERTY CUDA_ARCHITECTURES
${CMAKE_CUDA_ARCHITECTURES})
target_compile_definitions(
cp2k PUBLIC __OFFLOAD_CUDA $<$<COMPILE_LANGUAGE:CUDA>:__OFFLOAD_CUDA>)
include_directories(${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
elseif(CP2K_USE_ACCEL MATCHES "HIP")
set_property(TARGET cp2k PROPERTY HIP_ARCHITECTURES
@ -1697,14 +1692,12 @@ target_compile_definitions(
${CP2K_GPU_DFLAGS})
if(CP2K_USE_CUDA OR CP2K_USE_HIP)
# these checks are only relevant when the debug mode is on
target_compile_definitions(
cp2k
PUBLIC $<$<NOT:$<BOOL:${CP2K_DBCSR_USE_CPU_ONLY}>>:__DBCSR_ACC>
$<$<BOOL:${CP2K_DISABLE_GRID_GPU}>:__NO_OFFLOAD_GRID>
$<$<BOOL:${CP2K_DISABLE_PW_GPU}>:__NO_OFFLOAD_PW>
$<$<BOOL:${CP2K_DISABLE_DBM_GPU}>:__NO_OFFLOAD_DBM>)
$<$<NOT:$<BOOL:${CP2K_ENABLE_GRID_GPU}>>:__NO_OFFLOAD_GRID>
$<$<NOT:$<BOOL:${CP2K_ENABLE_PW_GPU}>>:__NO_OFFLOAD_PW>
$<$<NOT:$<BOOL:${CP2K_ENABLE_DBM_GPU}>>:__NO_OFFLOAD_DBM>)
endif()
# add an alias for the FetchContent. Note that FindPackage will automatically