ICODE-ADC/lammps
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #3127 from akohlmey/gpu-opencl-updates

Browse Source 
Update compiling OpenCL loader lib
This commit is contained in:
Axel Kohlmeyer
2022-02-14 16:10:53 -05:00
committed by GitHub
parent 5366621947 78d149f118
commit 5f47ff770c
12 changed files with 35 additions and 89 deletions
Download Patch File Download Diff File Expand all files Collapse all files

52
cmake/Modules/OpenCLLoader.cmake
View File

@ -1,50 +1,12 @@
message(STATUS "Downloading and building OpenCL loader library")
set(OPENCL_LOADER_URL "${LAMMPS_THIRDPARTY_URL}/opencl-loader-2021.09.18.tar.gz" CACHE STRING "URL for OpenCL loader tarball")
set(OPENCL_LOADER_MD5 "3b3882627964bd02e5c3b02065daac3c" CACHE STRING "MD5 checksum of OpenCL loader tarball")
set(OPENCL_LOADER_URL "${LAMMPS_THIRDPARTY_URL}/opencl-loader-2022.01.04.tar.gz" CACHE STRING "URL for OpenCL loader tarball")
set(OPENCL_LOADER_MD5 "8d3a801e87a2c6653bf0e27707063914" CACHE STRING "MD5 checksum of OpenCL loader tarball")
mark_as_advanced(OPENCL_LOADER_URL)
mark_as_advanced(OPENCL_LOADER_MD5)
include(ExternalProject)
ExternalProject_Add(opencl_loader
URL ${OPENCL_LOADER_URL}
URL_MD5 ${OPENCL_LOADER_MD5}
SOURCE_DIR "${CMAKE_BINARY_DIR}/opencl_loader-src"
BINARY_DIR "${CMAKE_BINARY_DIR}/opencl_loader-build"
CMAKE_ARGS ${CMAKE_REQUEST_PIC} ${CMAKE_EXTRA_OPENCL_LOADER_OPTS}
-DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
-DCMAKE_INSTALL_PREFIX=<INSTALL_DIR>
-DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}
-DCMAKE_MAKE_PROGRAM=${CMAKE_MAKE_PROGRAM}
-DCMAKE_TOOLCHAIN_FILE=${CMAKE_TOOLCHAIN_FILE}
BUILD_BYPRODUCTS <BINARY_DIR>/libOpenCL${CMAKE_STATIC_LIBRARY_SUFFIX}
LOG_DOWNLOAD ON
LOG_CONFIGURE ON
LOG_BUILD ON
INSTALL_COMMAND ""
TEST_COMMAND "")
ExternalProject_Get_Property(opencl_loader SOURCE_DIR)
set(OPENCL_LOADER_INCLUDE_DIR ${SOURCE_DIR}/inc)
# workaround for CMake 3.10 on ubuntu 18.04
file(MAKE_DIRECTORY ${OPENCL_LOADER_INCLUDE_DIR})
ExternalProject_Get_Property(opencl_loader BINARY_DIR)
set(OPENCL_LOADER_LIBRARY_PATH "${BINARY_DIR}/libOpenCL${CMAKE_STATIC_LIBRARY_SUFFIX}")
find_package(Threads QUIET)
if(NOT WIN32)
set(OPENCL_LOADER_DEP_LIBS "Threads::Threads;${CMAKE_DL_LIBS}")
else()
set(OPENCL_LOADER_DEP_LIBS "cfgmgr32;runtimeobject")
endif()
add_library(OpenCL::OpenCL UNKNOWN IMPORTED)
add_dependencies(OpenCL::OpenCL opencl_loader)
set_target_properties(OpenCL::OpenCL PROPERTIES
IMPORTED_LOCATION ${OPENCL_LOADER_LIBRARY_PATH}
INTERFACE_INCLUDE_DIRECTORIES ${OPENCL_LOADER_INCLUDE_DIR}
INTERFACE_LINK_LIBRARIES "${OPENCL_LOADER_DEP_LIBS}")
set(INSTALL_LIBOPENCL OFF CACHE BOOL "" FORCE)
set(BUILD_SHARED_LIBS OFF)
include(ExternalCMakeProject)
ExternalCMakeProject(opencl_loader ${OPENCL_LOADER_URL} ${OPENCL_LOADER_MD5} opencl-loader . "")
add_library(OpenCL::OpenCL ALIAS OpenCL)

2
lib/gpu/lal_answer.cpp
View File

@ -248,7 +248,7 @@ double AnswerT::energy_virial(double *eatom, double **vatom,
return energy_virial(eatom,vatom,virial);
double evdwl=0.0;
int ii, vstart=0, iend=_ev_stride;
int vstart=0, iend=_ev_stride;
if (_eflag) {
iend=_ev_stride*2;
#if (LAL_USE_OMP_SIMD == 1)

16
lib/gpu/lal_base_atomic.cpp
View File

@ -239,14 +239,14 @@ void BaseAtomicT::compute(const int f_ago, const int inum_full,
// Reneighbor on GPU if necessary and then compute forces, virials, energies
// ---------------------------------------------------------------------------
template <class numtyp, class acctyp>
int ** BaseAtomicT::compute(const int ago, const int inum_full,
const int nall, double **host_x, int *host_type,
double *sublo, double *subhi, tagint *tag,
int **nspecial, tagint **special,
const bool eflag_in, const bool vflag_in,
const bool eatom, const bool vatom,
int &host_start, int **ilist, int **jnum,
const double cpu_time, bool &success) {
int **BaseAtomicT::compute(const int ago, const int inum_full,
const int nall, double **host_x, int *host_type,
double *sublo, double *subhi, tagint *tag,
int **nspecial, tagint **special,
const bool eflag_in, const bool vflag_in,
const bool eatom, const bool vatom,
int &host_start, int **ilist, int **jnum,
const double cpu_time, bool &success) {
acc_timers();
int eflag, vflag;
if (eatom) eflag=2;

12
lib/gpu/lal_base_atomic.h
View File

@ -133,20 +133,12 @@ class BaseAtomic {
int &host_start, const double cpu_time, bool &success);
/// Pair loop with device neighboring
int * compute(const int ago, const int inum_full,
int **compute(const int ago, const int inum_full,
const int nall, double **host_x, int *host_type, double *sublo,
double *subhi, tagint *tag, int **nspecial,
tagint **special, const bool eflag, const bool vflag,
const bool eatom, const bool vatom, int &host_start,
const double cpu_time, bool &success);
/// Pair loop with device neighboring
int ** compute(const int ago, const int inum_full,
const int nall, double **host_x, int *host_type, double *sublo,
double *subhi, tagint *tag, int **nspecial,
tagint **special, const bool eflag, const bool vflag,
const bool eatom, const bool vatom, int &host_start,
int **ilist, int **numj, const double cpu_time, bool &success);
int **ilist, int **numj, const double cpu_time, bool &success);
// -------------------------- DEVICE DATA -------------------------

2
lib/gpu/lal_base_ellipsoid.cpp
View File

@ -69,7 +69,7 @@ BaseEllipsoidT::~BaseEllipsoid() {
}
template <class numtyp, class acctyp>
int BaseEllipsoidT::bytes_per_atom(const int max_nbors) const {
int BaseEllipsoidT::bytes_per_atom_ellipsoid(const int max_nbors) const {
return device->atom.bytes_per_atom()+ans->bytes_per_atom()+
nbor->bytes_per_atom(max_nbors);
}

21
lib/gpu/lal_base_ellipsoid.h
View File

@ -108,7 +108,7 @@ class BaseEllipsoid {
void output_times();
/// Returns memory usage on device per atom
int bytes_per_atom(const int max_nbors) const;
int bytes_per_atom_ellipsoid(const int max_nbors) const;
/// Total host memory used by library for pair style
double host_memory_usage_base() const;
@ -173,18 +173,13 @@ class BaseEllipsoid {
const double cpu_time, bool &success, double **quat);
/// Pair loop with device neighboring
int** compute(const int ago, const int inum_full, const int nall,
double **host_x, int *host_type, double *sublo,
double *subhi, tagint *tag, int **nspecial,
tagint **special, const bool eflag, const bool vflag,
const bool eatom, const bool vatom, int &host_start,
int **ilist, int **numj, const double cpu_time, bool &success,
double **host_quat);
/// Build neighbor list on accelerator
void build_nbor_list(const int inum, const int host_inum, const int nall,
double **host_x, int *host_type, double *sublo,
double *subhi, bool &success);
int**compute(const int ago, const int inum_full, const int nall,
double **host_x, int *host_type, double *sublo,
double *subhi, tagint *tag, int **nspecial,
tagint **special, const bool eflag, const bool vflag,
const bool eatom, const bool vatom, int &host_start,
int **ilist, int **numj, const double cpu_time, bool &success,
double **host_quat);
// -------------------------- DEVICE DATA -------------------------

2
lib/gpu/lal_device.cpp
View File

@ -198,7 +198,7 @@ int DeviceT::init_device(MPI_Comm world, MPI_Comm replica, const int ngpu,
// Find deviceID with most CUs (priority given to the accelerator type)
if (_first_device < 0) {
int best_device = 0;
int best_cus = gpu->cus(0);
unsigned best_cus = gpu->cus(0);
bool type_match = (gpu->device_type(0) == type);
for (int i = 1; i < gpu->num_devices(); i++) {
if (type_match==true && gpu->device_type(i)!=type)

3
lib/gpu/lal_eam.h
View File

@ -62,9 +62,6 @@ class EAM : public BaseAtomic<numtyp, acctyp> {
/** \note This is called at the beginning of the init() routine **/
void clear();
/// Returns memory usage on device per atom
int bytes_per_atom(const int max_nbors) const;
/// Total host memory used by library for pair style
double host_memory_usage() const;

2
lib/gpu/lal_gayberne.cpp
View File

@ -43,7 +43,7 @@ GayBerneT::~GayBerne() {
template <class numtyp, class acctyp>
int GayBerneT::bytes_per_atom(const int max_nbors) const {
return this->bytes_per_atom(max_nbors);
return this->bytes_per_atom_ellipsoid(max_nbors);
}
template <class numtyp, class acctyp>

2
lib/gpu/lal_re_squared.cpp
View File

@ -43,7 +43,7 @@ RESquaredT::~RESquared() {
template <class numtyp, class acctyp>
int RESquaredT::bytes_per_atom(const int max_nbors) const {
return this->bytes_per_atom(max_nbors);
return this->bytes_per_atom_ellipsoid(max_nbors);
}
template <class numtyp, class acctyp>

6
src/GPU/pair_lj_expand_coul_long_gpu.cpp
View File

@ -51,9 +51,9 @@ int ljecl_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
int &gpu_mode, FILE *screen, double **host_cut_ljsq,
double host_cut_coulsq, double *host_special_coul,
const double qqrd2e, const double g_ewald);
int ljecl_gpu_reinit(const int ntypes, double **cutsq, double **host_lj1,
double **host_lj2, double **host_lj3, double **host_lj4,
double **offset, double **shift, double **host_lj_cutsq);
void ljecl_gpu_reinit(const int ntypes, double **cutsq, double **host_lj1,
double **host_lj2, double **host_lj3, double **host_lj4,
double **offset, double **shift, double **host_lj_cutsq);
void ljecl_gpu_clear();
int ** ljecl_gpu_compute_n(const int ago, const int inum,
const int nall, double **host_x, int *host_type,

4
src/GPU/pair_ufm_gpu.cpp
View File

@ -42,8 +42,8 @@ int ufml_gpu_init(const int ntypes, double **cutsq, double **host_uf1,
const int nall, const int max_nbors, const int maxspecial,
const double cell_size, int &gpu_mode, FILE *screen);
int ufml_gpu_reinit(const int ntypes, double **cutsq, double **host_uf1,
double **host_uf2, double **host_uf3, double **offset);
void ufml_gpu_reinit(const int ntypes, double **cutsq, double **host_uf1,
double **host_uf2, double **host_uf3, double **offset);
void ufml_gpu_clear();
int ** ufml_gpu_compute_n(const int ago, const int inum, const int nall,