Add deisa plugin

AUTHORS

@@ -11,8 +11,12 @@ Julien Bigot - CEA ([email protected])
 * Maintainer (Dec. 2014 - ...)
 * Design and initial implementation
 
+Amal Gueroudji - CEA ([email protected])
+* Deisa plugin creation
+
 Benoit Martin - CEA ([email protected])
 * CMake maintainer (Oct. 2024 - ...)
+* Deisa plugin refactoring
 
 Tomasz Paluszkiewicz - PSNC ([email protected])
 The scientific/academic work is financed from financial resources for science

CHANGELOG.md

@@ -13,6 +13,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
 ### Added
 * NetCDF plugin now support the size_of attribute, fixes
   [#446](https://gitlab.maisondelasimulation.fr/pdidev/pdi/-/issues/446)
+* Deisa plugin for in-situ analysis using Dask
 
 ### Changed
 * Replaced the astyle based indentation by a clang-format one, fixes

CMakeLists.txt

@@ -77,6 +77,7 @@ option(BUILD_SERIALIZE_PLUGIN   "Build Serialize plug-in" ON)
 option(BUILD_SHARED_LIBS        "Build shared libraries rather than static ones" ON)
 option(BUILD_TRACE_PLUGIN       "Build Trace plugin" ON)
 option(BUILD_USER_CODE_PLUGIN   "Build User-code plugin" ON)
+option(BUILD_DEISA_PLUGIN   	"Build Deisa plug-in" OFF)
 
 
 
@@ -113,6 +114,9 @@ endif()
 if("${BUILD_DECL_NETCDF_PLUGIN}")
 	list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/plugins/decl_netcdf/cmake")
 endif()
+if("${BUILD_DEISA_PLUGIN}")
+	list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/plugins/deisa/cmake")
+endif()
 if("${BUILD_MPI_PLUGIN}")
 	list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/plugins/mpi/cmake")
 endif()
@@ -175,6 +179,14 @@ if("${BUILD_DECL_NETCDF_PLUGIN}" AND "${BUILD_NETCDF_PARALLEL}" AND NOT "${BUILD
 	)
 endif()
 
+if("${BUILD_DEISA_PLUGIN}" AND NOT "${BUILD_PYTHON}")
+	message(FATAL_ERROR
+			"Deisa plugin is enabled but Python support is disabled.\n"
+			"    you have the following options:\n"
+			"     * enable Python support => pass `-DBUILD_PYTHON=ON' to cmake\n"
+			"     * do not build the Deisa plugin => pass `-DBUILD_DEISA_PLUGIN=OFF' to cmake"
+	)
+endif()
 
 
 ### Dependencies
@@ -501,6 +513,13 @@ sbuild_add_module(USER_CODE_PLUGIN
 )
 
 
+sbuild_add_module(DEISA_PLUGIN
+	ENABLE_BUILD_FLAG BUILD_DEISA_PLUGIN
+	SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/plugins/deisa"
+	DEPENDS PDI
+)
+
+
 sbuild_add_module(PDI_EXAMPLE
 	ENABLE_BUILD_FLAG BUILD_TESTING
 	SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/example"

README.md

@@ -42,6 +42,7 @@ PDI distribution is made of the following submodules:
 * `pdi/` : the PDI library,
 * `plugins/decl_hdf5/`: the Decl'HDF5 plugin,
 * `plugins/decl_netcdf/`: the Decl'NetCDF plugin,
+* `plugins/deisa/`: the Deisa plugin,
 * `plugins/mpi/`: the MPI plugin,
 * `plugins/pycall/`: the Pycall plugin,
 * `plugins/serialize/`: the serialize plugin,

example/deisa.yml

@@ -0,0 +1,33 @@
+# duration in seconds
+duration: 0.75
+# global [height, width] (excluding boundary conditions or ghosts)
+datasize: [60, 12]
+# degree of parallelism
+parallelism: { height: 3, width: 1 }
+
+# only the following config is passed to PDI
+pdi:
+  metadata: # type of small values for which PDI keeps a copy
+    iter:   int                    # current iteration id
+    dsize:  { size: 2, type: array, subtype: int } # local data size including ghosts/boundary
+    psize:  { size: 2, type: array, subtype: int } # number of processes in each dimension
+    pcoord: { size: 2, type: array, subtype: int } # coordinate of the process
+  data: # type of values for which PDI does not keep a copy
+    main_field: { size: [ '$dsize[0]', '$dsize[1]' ], type: array, subtype: double }
+
+  plugins:
+    mpi:
+    deisa:
+      scheduler_info: /tmp/scheduler.json
+      init_on: init
+      time_step: $iter
+      deisa_arrays:
+        global_t: # Name of the Deisa array
+          type: array
+          subtype: double
+          size: [ 10, '$psize[0]*($dsize[0]-2)', '$psize[1]*($dsize[1]-2)' ] # first dimension corresponds to time steps
+          subsize: [ 1, '$dsize[0]-2', '$dsize[1]-2' ]
+          start: [ $iter, '($dsize[0]-2)*$pcoord[0]', '($dsize[1]-2)*$pcoord[1]' ]
+          +timedim: 0
+      map_in:
+        main_field: global_t            # `main_field` is mapped to `global_t`

example/deisa_client.py

@@ -0,0 +1,26 @@
+from deisa import Deisa
+
+# Scheduler file name and configuration file
+scheduler_info = '/tmp/scheduler.json'
+
+# Initialize Deisa
+deisa = Deisa(scheduler_info, nb_workers=1, use_ucx=False)
+
+# either: Get data descriptor as a list of Deisa arrays object
+arrays = deisa.get_deisa_arrays()
+
+# Select data
+gt = arrays["global_t"][...]
+
+# Check contract
+arrays.check_contract()
+
+
+##########################
+# Perform analysis here
+##########################
+
+
+arrays.validate_contract()
+
+deisa.wait_for_last_bridge_and_shutdown()

example/example.c

@@ -173,7 +173,8 @@ int main(int argc, char* argv[])
 
 	PDI_expose("mpi_rank", &pcoord_1d, PDI_OUT);
 	PDI_expose("mpi_size", &psize_1d, PDI_OUT);
-	PDI_event("init");
+	int ii = 0;
+	PDI_expose("iter", &ii, PDI_OUT);
 
 	long longval;
 
@@ -209,6 +210,7 @@ int main(int argc, char* argv[])
 	PDI_expose("dsize", dsize, PDI_OUT);
 	PDI_expose("psize", psize, PDI_OUT);
 	PDI_expose("pcoord", pcoord, PDI_OUT);
+	PDI_event("init");
 
 	double(*cur)[dsize[1]] = malloc(sizeof(double) * dsize[1] * dsize[0]);
 	double(*next)[dsize[1]] = malloc(sizeof(double) * dsize[1] * dsize[0]);
@@ -217,7 +219,6 @@ int main(int argc, char* argv[])
 
 	PDI_event("main_loop");
 	double start = MPI_Wtime();
-	int ii;
 	int next_reduce = 0;
 	for (ii = 0;; ++ii) {
 		PDI_multi_expose("newiter", "iter", &ii, PDI_INOUT, "main_field", cur, PDI_INOUT, NULL);

pdi/CHANGELOG.md

@@ -18,6 +18,7 @@ and this project adheres to
 #### Added
 * Add the `Context::find()` method.
   [#445](https://gitlab.maisondelasimulation.fr/pdidev/pdi/-/issues/445)
+* Add the `pybind11::dtype to_python(const std::shared_ptr<const Scalar_datatype>& scalar_type)` helper function.
 
 #### Changed
 * Update the version of dependencies according to our policy: oldest supported

pdi/docs/Plugins.md

@@ -4,6 +4,7 @@
 |:----------------------------------------------------|:------------------------------------------------------------------|
 |\subpage Decl_HDF5_plugin "Decl'HDF5 plugin"         |Read and write data from HDF5 files in a declarative way.          |
 |\subpage Decl_NetCDF_plugin "Decl'NetCDF plugin"     |Read and write data from NetCDF files in a declarative way.        |
+|Deisa plugin                                         |Expose data to a Dask analysis script.                             |
 |\subpage mpi_plugin "MPI plugin"                     |Enables MPI support in %PDI and plugins.                           |
 |\subpage pycall_plugin "Pycall plugin"               |Call python scripts from C application                             |
 |\subpage serialize_plugin "Serialize plugin"         |Serializes and deserializes shared data.                           |

pdi/docs/Source_installation.md

@@ -105,6 +105,7 @@ The following flags define which features of the distribution to enable or not.
 |`BUILD_TESTING`            |`ON`   |Build the tests.|
 |`BUILD_TRACE_PLUGIN`       |`ON`   |Build the Trace plug-in.|
 |`BUILD_USER_CODE_PLUGIN`   |`ON`   |Build the User-code plug-in.|
+|`BUILD_DEISA_PLUGIN`       |`OFF`  |Build the Deisa plug-in.|
 |`BUILD_DOCUMENTATION`      |`OFF`  |Build the documentation website. (devel profile)|
 |`BUILD_PYCALL_PLUGIN`      |`OFF`  |Build Pycall plug-in. (unstable)|
 |`BUILD_PYTHON`             |`OFF`  |Build the Python interface. (unstable)|

pdi/include/pdi/python/tools.h

@@ -49,6 +49,13 @@ pybind11::object PDI_EXPORT to_python(Ref r, bool force_const = false);
  */
 Datatype_sptr PDI_EXPORT python_type(const pybind11::array& a);
 
+/** Convert a Scalar_datatype to a pybind11 data type.
+ *
+ * \param scalar_type to convert to pybind11 data type
+ * \return pybind11 data type for the given scalar type
+ */
+pybind11::dtype PDI_EXPORT to_python(const std::shared_ptr<const Scalar_datatype>& scalar_type);
+
 } // namespace PDI
 
 #endif // PDI_PYTHON_TOOLS_H_

pdi/src/python/tools.cxx

@@ -26,6 +26,7 @@
 #include "config.h"
 
 #include <cstdint>
+#include <memory>
 #include <vector>
 
 #include "pdi/array_datatype.h"
@@ -92,59 +93,7 @@ pybind11::object to_python(Ref r, bool force_const)
 	auto&& scalar_type = dynamic_pointer_cast<const Scalar_datatype>(subtype);
 	if (ndim) strides.emplace_back(scalar_type->buffersize());
 
-	pybind11::dtype pytype;
-	switch (scalar_type->kind()) {
-	case Scalar_kind::FLOAT: {
-		switch (scalar_type->datasize()) {
-		case sizeof(float):
-			pytype = pybind11::dtype::of<float>();
-			break;
-		case sizeof(double):
-			pytype = pybind11::dtype::of<double>();
-			break;
-		default:
-			throw Type_error{"Unable to pass {} bytes floating point value to python", scalar_type->datasize()};
-		}
-	} break;
-	case Scalar_kind::SIGNED: {
-		switch (scalar_type->datasize()) {
-		case sizeof(int8_t):
-			pytype = pybind11::dtype::of<int8_t>();
-			break;
-		case sizeof(int16_t):
-			pytype = pybind11::dtype::of<int16_t>();
-			break;
-		case sizeof(int32_t):
-			pytype = pybind11::dtype::of<int32_t>();
-			break;
-		case sizeof(int64_t):
-			pytype = pybind11::dtype::of<int64_t>();
-			break;
-		default:
-			throw Type_error{"Unable to pass {} bytes integer value to python", scalar_type->datasize()};
-		}
-	} break;
-	case Scalar_kind::UNSIGNED: {
-		switch (scalar_type->datasize()) {
-		case sizeof(uint8_t):
-			pytype = pybind11::dtype::of<uint8_t>();
-			break;
-		case sizeof(uint16_t):
-			pytype = pybind11::dtype::of<uint16_t>();
-			break;
-		case sizeof(uint32_t):
-			pytype = pybind11::dtype::of<uint32_t>();
-			break;
-		case sizeof(uint64_t):
-			pytype = pybind11::dtype::of<uint64_t>();
-			break;
-		default:
-			throw Type_error{"Unable to pass {} bytes unsigned integer value to python", scalar_type->datasize()};
-		}
-	} break;
-	default:
-		throw Type_error{"Unable to pass value of unexpected type to python"};
-	}
+	pybind11::dtype pytype = to_python(scalar_type);
 
 	ssize_t cumulated_stride = 1;
 	for (auto&& stride = strides.rbegin(); stride != strides.rend(); ++stride) {
@@ -221,4 +170,51 @@ Datatype_sptr python_type(const pybind11::array& a)
 	return result;
 }
 
+pybind11::dtype to_python(const std::shared_ptr<const Scalar_datatype>& scalar_type)
+{
+	switch (scalar_type->kind()) {
+	case Scalar_kind::FLOAT: {
+		switch (scalar_type->datasize()) {
+		case sizeof(float):
+			return pybind11::dtype::of<float>();
+		case sizeof(double):
+			return pybind11::dtype::of<double>();
+		default:
+			throw Type_error{"Unable to pass {} bytes floating point value to python", scalar_type->datasize()};
+		}
+	} break;
+	case Scalar_kind::SIGNED: {
+		switch (scalar_type->datasize()) {
+		case sizeof(int8_t):
+			return pybind11::dtype::of<int8_t>();
+		case sizeof(int16_t):
+			return pybind11::dtype::of<int16_t>();
+		case sizeof(int32_t):
+			return pybind11::dtype::of<int32_t>();
+		case sizeof(int64_t):
+			return pybind11::dtype::of<int64_t>();
+		default:
+			throw Type_error{"Unable to pass {} bytes integer value to python", scalar_type->datasize()};
+		}
+	} break;
+	case Scalar_kind::UNSIGNED: {
+		switch (scalar_type->datasize()) {
+		case sizeof(uint8_t):
+			return pybind11::dtype::of<uint8_t>();
+		case sizeof(uint16_t):
+			return pybind11::dtype::of<uint16_t>();
+		case sizeof(uint32_t):
+			return pybind11::dtype::of<uint32_t>();
+		case sizeof(uint64_t):
+			return pybind11::dtype::of<uint64_t>();
+		default:
+			throw Type_error{"Unable to pass {} bytes unsigned integer value to python", scalar_type->datasize()};
+		}
+	} break;
+	default:
+		throw Type_error{"Unable to pass value of unexpected type to python"};
+	}
+}
+
+
 } // namespace PDI

plugins/deisa/CMakeLists.txt

@@ -0,0 +1,39 @@
+#=============================================================================
+# Copyright (c) 2020-2024 Centre national de la recherche scientifique (CNRS)
+# Copyright (c) 2020-2024 Commissariat a l'énergie atomique et aux énergies alternatives (CEA)
+# Copyright (c) 2020-2023 Institut national de recherche en informatique et en automatique (Inria)
+# Copyright (c) 2020-2024 Université Paris-Saclay
+# Copyright (c) 2020-2024 Université de Versailles Saint-Quentin-en-Yvelines
+#
+# SPDX-License-Identifier: MIT
+#=============================================================================
+
+cmake_minimum_required(VERSION 3.16...3.29)
+project(pdi_deisa_plugin LANGUAGES C CXX)
+list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_LIST_DIR}/cmake")
+
+include(GNUInstallDirs)
+
+find_package(Python3Path 3.8.2 REQUIRED COMPONENTS Interpreter Development)
+
+# PyBind11
+set(Python_ADDITIONAL_VERSIONS "${Python3_VERSION}" CACHE STRING "Python version found by FindPython3 for coherency" FORCE)
+set(PYBIND11_PYTHON_VERSION "${Python3_VERSION}" CACHE STRING "Python version to use for compiling modules" FORCE)
+find_package(pybind11 2.4.3 REQUIRED)
+
+# PDI
+find_package(PDI REQUIRED COMPONENTS plugins pysupport)
+find_package(MPI REQUIRED COMPONENTS CXX)
+
+# The plugin
+add_library(pdi_deisa_plugin MODULE deisa.cxx)
+target_link_libraries(pdi_deisa_plugin PUBLIC PDI::PDI_plugins PDI::PDI_pysupport pybind11::embed MPI::MPI_CXX)
+set_target_properties(pdi_deisa_plugin PROPERTIES
+    CXX_STANDARD_REQUIRED TRUE
+    CXX_VISIBILITY_PRESET hidden)
+
+# installation
+set(INSTALL_PDIPLUGINDIR "${PDI_DEFAULT_PLUGINDIR}" CACHE PATH "PDI plugins (${PDI_DEFAULT_PLUGINDIR})")
+install(TARGETS pdi_deisa_plugin
+    LIBRARY DESTINATION "${INSTALL_PDIPLUGINDIR}"
+)