diff --git a/CMakeLists.txt b/CMakeLists.txt
index 2a3c20f34557077e042f61d3cf6a76a1f760fc02..e15cb00b421267ad3f17310e0b1c313d6fd4290a 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -19,11 +19,15 @@ set(ENV{HDF5_ROOT} ${HDF5_ROOT})
set(OPENFPM_PDATA_DIR CACHE PATH "OpenFPM_pdata dirs")
set(PETSC_ROOT CACHE PATH "If compiling with linear algebra indicate the PETSC root directory")
set(ENV{PETSC_DIR} ${PETSC_ROOT})
+set(HIP_ENABLE CACHE BOOL "Enable HIP compiler")
+set(AMD_ARCH_COMPILE "gfx900" CACHE STRING "AMD gpu architecture used to compile kernels")
if (ENABLE_GPU)
- set(CUDA_ON_CPU OFF)
- enable_language(CUDA)
- find_package(CUDA)
+ set(CUDA_ON_CPU OFF)
+ if (NOT HIP_ENABLE)
+ enable_language(CUDA)
+ find_package(CUDA)
+ endif()
endif()
set (CMAKE_CXX_STANDARD 14)
@@ -41,6 +45,16 @@ find_package(MPI REQUIRED)
find_package(HDF5 REQUIRED)
find_package(TinyObjLoader)
find_package(PETSc)
+find_package(HIP)
+
+if(HIP_FOUND)
+ set(DEFINE_HIP_GPU "#define HIP_GPU")
+ set(DEFINE_CUDIFY_USE_HIP "#define CUDIFY_USE_HIP")
+endif()
+
+if(HIP_FOUND)
+ set(DEFINE_CUDA_GPU "#define CUDA_GPU")
+endif()
if(CUDA_FOUND)
set(OPENFPM_INIT_FILE "initialize/initialize_wrapper_cuda.cu")
diff --git a/build.sh b/build.sh
index 55b11b81cdec43c08949b9b4fb5ea370a3c7c813..d03d3ef014785d4c63e5adbea9606615cfb03913 100755
--- a/build.sh
+++ b/build.sh
@@ -1,6 +1,5 @@
#! /bin/bash
-
workspace=$1
hostname=$(hostname)
branch=$3
diff --git a/configure b/configure
index b34c8afb1697fe5f827dd7ce3f712409fe65d6ea..31f1735d38c3980de785b285dde79423e54d748f 100755
--- a/configure
+++ b/configure
@@ -122,7 +122,10 @@ with_petsc
with_eigen
with_vcdevel
enable_gpu
+enable_hip
enable_asan
+enable_garbageinj
+enable_garbageinjv
'
rm -rf build
@@ -251,11 +254,25 @@ do
else
conf_options="$conf_options -DCMAKE_CUDA_HOST_COMPILER=$(which $CXX)"
fi
+ if [ x"$CXXCUDA" == x"" ]; then
+ conf_options="$conf_options"
+ else
+ conf_options="$conf_options -DCMAKE_CUDA_COMPILER=$(which $CXXCUDA)"
+ fi
conf_options="$conf_options -DENABLE_GPU=ON"
;;
+ hip)
+ conf_options="$conf_options -DHIP_ENABLE=ON"
+ ;;
asan)
conf_options="$conf_options -DENABLE_ASAN=ON"
;;
+ garbageinj)
+ conf_options="$conf_options -DENABLE_GARBAGE_INJECTOR=ON"
+ ;;
+ garbageinjv)
+ conf_options="$conf_options -DENABLE_VCLUSTER_GARBAGE_INJECTOR=ON"
+ ;;
*) ac_unrecognized_opts="$ac_unrecognized_opts$ac_unrecognized_sep--enable-$ac_useropt_orig"
ac_unrecognized_sep=', '
;;
@@ -573,6 +590,7 @@ cd build
## remove enerything
echo "Calling cmake ../. $conf_options"
+printf "cmake ../. $conf_options" > cmake_build_options
rm ../error_code
DYLD_LIBRARY_PATH=$ld_lib_pathopt cmake ../. $conf_options
if [ $? != 0 ]; then
@@ -601,6 +619,7 @@ clean:
install:
\$(MAKE) -C build \$@
+ script/install_parallel_debugger
pdata:
\$(MAKE) -C build \$@
diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
index f8b87de0d1856c8a21e9ee1f1d8881481c13d72d..eb50c2f1c660f526c88f34e0efbdef3b4ba5d9c3 100644
--- a/src/CMakeLists.txt
+++ b/src/CMakeLists.txt
@@ -6,12 +6,41 @@ if (CUDA_ON_CPU)
add_definitions(-DCUDA_ON_CPU)
endif()
+if (CUDA_FOUND OR CUDA_ON_CPU OR HIP_FOUND)
-add_executable(io main.cpp
+ set(CUDA_SOURCES HDF5_wr/HDF5_writer_cuda.cu)
+
+endif()
+
+if ( HIP_ENABLE AND HIP_FOUND )
+
+ list(APPEND HIP_HIPCC_FLAGS ${CMAKE_CXX_FLAGS_DEBUG})
+
+ if (CMAKE_BUILD_TYPE STREQUAL "Debug")
+ list(APPEND HIP_HIPCC_FLAGS -O0)
+ endif()
+
+ set(CMAKE_CXX_COMPILER ${HIP_HIPCC_EXECUTABLE})
+
+ list(APPEND HIP_HIPCC_FLAGS -D__NVCC__ -D__HIP__ -DCUDART_VERSION=11000 -D__CUDACC__ -D__CUDACC_VER_MAJOR__=11 -D__CUDACC_VER_MINOR__=0 -D__CUDACC_VER_BUILD__=0)
+
+ set_source_files_properties(${CUDA_SOURCES} PROPERTIES LANGUAGE CXX)
+
+ hip_add_executable(io main.cpp
+ MetaParser/MetaParser_unit_test.cpp
+ ${CUDA_SOURCES}
+ ObjReader/ObjReader_unit_test.cpp)
+
+
+else()
+
+ add_executable(io main.cpp
MetaParser/MetaParser_unit_test.cpp
- HDF5_wr/HDF5_writer_cuda.cu
+ ${CUDA_SOURCES}
ObjReader/ObjReader_unit_test.cpp)
+endif()
+
add_dependencies(io ofpmmemory)
add_dependencies(io vcluster)
@@ -71,7 +100,7 @@ endif()
# will also build with -std=c++11
target_compile_features(io PUBLIC cxx_std_11)
target_link_libraries(io ${MPI_C_LIBRARIES})
-
+target_link_libraries(io ${MPI_CXX_LIBRARIES})
if (TEST_COVERAGE)
target_link_libraries(io -lgcov)
endif()
diff --git a/src/RawReader/InitGridWithPixel.hpp b/src/RawReader/InitGridWithPixel.hpp
index bd53a1ce70550f7a1c50de06000aa8c88dc4e2f8..1bd71814ab9e460047aa00cfe4b77f04e2ffbcec 100644
--- a/src/RawReader/InitGridWithPixel.hpp
+++ b/src/RawReader/InitGridWithPixel.hpp
@@ -9,7 +9,7 @@
*
* @details Can be run in parallel.
*
- * @author Justina Stark
+ * @author Justina Stark & Pietro Incardona
* @date November 2019 - August 2020
*/
#ifndef IMAGE_BASED_RECONSTRUCTION_GETINITIALGRID_HPP
@@ -18,6 +18,7 @@
#include <iostream>
#include <typeinfo>
#include <cmath>
+#include <sys/stat.h>
#include "Vector/vector_dist.hpp"
#include "Grid/grid_dist_id.hpp"
@@ -26,6 +27,14 @@
#include "level_set/redistancing_Sussman/HelpFunctionsForGrid.hpp"
typedef signed char BYTE;
+
+inline bool exists_test (const std::string& name) {
+ struct stat buffer;
+ return (stat (name.c_str(), &buffer) == 0);
+}
+
+
+
/**@brief Read the number of pixels per dimension from a csv-file in order to create a grid with the same size.
*
* @param path_to_file Std::string containing the path to the csv file that holds the image/volume size in
@@ -35,13 +44,15 @@ typedef signed char BYTE;
std::vector<size_t> get_size(const std::string & path_to_file)
{
std::vector<size_t> stack_dimst_1d;
- // stream input csv file and error check
- std::ifstream file(path_to_file);
- if (!file)
- {
- std::cout << "Error opening file" << std::endl;
- exit (EXIT_FAILURE);
+ // check if file exists and stream input csv file
+ if(!exists_test(path_to_file)){
+ std::cout << "------------------------------------------------------------------------" << std::endl;
+ std::cout << "Error: file " << path_to_file << " does not exist. Aborting..." << std::endl;
+ std::cout << "------------------------------------------------------------------------" << std::endl;
+ abort();
}
+ std::ifstream file(path_to_file);
+
// get its size
std::streampos fileSize;
@@ -83,7 +94,14 @@ void load_pixel_onto_grid(grid_type & grid, std::string file_name, std::vector <
constexpr size_t x = 0;
constexpr size_t y = 1;
constexpr size_t z = 2;
-
+
+ // check if file exists and stream input file
+ if(!exists_test(file_name)){
+ std::cout << "------------------------------------------------------------------------" << std::endl;
+ std::cout << "Error: file " << file_name << " does not exist. Aborting..." << std::endl;
+ std::cout << "------------------------------------------------------------------------" << std::endl;
+ abort();
+ }
std::ifstream file_stream (file_name, std::ifstream::binary);
auto & v_cl = create_vcluster();
@@ -121,17 +139,24 @@ void load_pixel_onto_grid(grid_type & grid, std::string file_name, std::vector <
auto key = dom.get();
auto gkey = grid.getGKey(key);
+ // In case a patch starts within a group of nodes to which same pixel-value should be assigned, get the
+ // respective rest-offset
+ size_t rest_offset = (size_t) (fmod(gkey.get(0), refinement[x])); // get the remainder
+
+
// get l as the length of one x-line of the original image stack for the specific patch on the processor
- auto & gbox = grid.getLocalGridsInfo();
- auto & DomBox = gbox.get(key.getSub()).Dbox;
- int l = (size_t) std::round((DomBox.getHigh(0) - DomBox.getLow(0) + 1) / refinement[x]);
+ auto & gbox = grid.getLocalGridsInfo();
+ auto & DomBox = gbox.get(key.getSub()).Dbox;
+ size_t patch_size = DomBox.getHigh(0) - DomBox.getLow(0) + 1;
+
+ size_t l = (size_t) ceil( (patch_size + rest_offset) / refinement[x]);
// in case that the grid has a different resolution than the underlying image stack:
// create a key which is used to get the offset for the file reading
// the indices in this key are corrected by the refinement factor
for (size_t d = 0; d < grid_type::dims; d++)
{
- gkey.set_d(d, std::round(gkey.get(d) / refinement[d]));
+ gkey.set_d(d, floor(gkey.get(d) / refinement[d]));
}
// the offset matches the pixel from the image stack to the corresponding current position of the iterator
@@ -147,10 +172,13 @@ void load_pixel_onto_grid(grid_type & grid, std::string file_name, std::vector <
// run over a whole grid-line in x and assign pixel values from pixel_line to grid nodes
// if the grid is finer in x as the image stack, the same pixel value from pixel_line is
// assigned refinement[x] times
- for (size_t i = 0; i < l * refinement[x]; i++)
+ for (size_t k = 0; k < patch_size; ++k)
{
auto key = dom.get();
- grid.template get<Phi_0>(key) = (double) pixel_line[(size_t)floor(i / refinement[x])];
+ // get the correct index of the pixel to be read from pixel_line by considering a potential rest-offset,
+ // when the patch divides group of nodes that belong to the same pixel
+ size_t i = (size_t) floor((k + rest_offset) / refinement[x]);
+ grid.template get<Phi_0>(key) = (double) pixel_line[i];
++dom;
}
// now one grid line in x is finished and the iterator dom has advanced accordingly s.t. next loop continues
diff --git a/src/config/config_cmake.h.in b/src/config/config_cmake.h.in
index 85840022fcec2c75463a60a05aa6c09b76ea448e..89dc2f504342560b87f489780adc1563f3f4fed3 100644
--- a/src/config/config_cmake.h.in
+++ b/src/config/config_cmake.h.in
@@ -1,3 +1,9 @@
+/* HIP GPU support */
+${DEFINE_HIP_GPU}
+
+/* HIP Cudify GPU support */
+${DEFINE_CUDIFY_USE_HIP}
+
/* Coverty scan */
${DEFINE_COVERTY_SCAN}
diff --git a/src/main.cpp b/src/main.cpp
index 176bf030ab86272271f3feea368596579bb02788..d73982ffad94993a966141815b8e09a216f0651c 100755
--- a/src/main.cpp
+++ b/src/main.cpp
@@ -8,7 +8,6 @@
#define BOOST_TEST_DYN_LINK
#include <boost/test/unit_test.hpp>
-
// initialization function:
bool init_unit_test()
{