Spatial sorting of particles

.gitignore

@@ -59,3 +59,4 @@ tags
 action-token
 *.vim
 ignore-*
+tombi/

cmake/benchmark.cmake

@@ -12,4 +12,7 @@ if(${output})
   list(APPEND libs ntt_output)
 endif()
 add_dependencies(${exec} ${libs})
-target_link_libraries(${exec} PRIVATE ${libs} stdc++fs)
+if (NOT "${CMAKE_CXX_COMPILER_ID}" MATCHES "AppleClang")
+  list(APPEND libs stdc++fs)
+endif()
+target_link_libraries(${exec} PRIVATE ${libs})

cmake/tests.cmake

@@ -5,16 +5,11 @@ set(SRC_DIR ${CMAKE_CURRENT_SOURCE_DIR}/src)
 
 set(TEST_DIRECTORIES "")
 
-if(NOT ${mpi})
-  list(APPEND TEST_DIRECTORIES global)
-  list(APPEND TEST_DIRECTORIES metrics)
-  list(APPEND TEST_DIRECTORIES kernels)
-  list(APPEND TEST_DIRECTORIES archetypes)
-  list(APPEND TEST_DIRECTORIES framework)
-elseif(${mpi} AND ${output})
-  list(APPEND TEST_DIRECTORIES framework)
-endif()
-
+list(APPEND TEST_DIRECTORIES global)
+list(APPEND TEST_DIRECTORIES metrics)
+list(APPEND TEST_DIRECTORIES kernels)
+list(APPEND TEST_DIRECTORIES archetypes)
+list(APPEND TEST_DIRECTORIES framework)
 list(APPEND TEST_DIRECTORIES output)
 
 foreach(test_dir IN LISTS TEST_DIRECTORIES)

input.example.toml

@@ -107,7 +107,7 @@
 
     [grid.boundaries.match]
       # Size of the matching layer in each direction for fields in physical (code) units
-      #   @type: float | array<tuple<float>> 
+      #   @type: float | array<tuple<float>>
       #   @default: 1% of the domain size (in shortest dimension)
       #   @note: In spherical, this is the size of the layer in `r` from the outer wall
       #   @example: `ds = 1.5` (will set the same for all directions)
@@ -390,11 +390,16 @@
   #   @type: bool
   #   @default: false
   use_weights = ""
-  # Timesteps between particle re-sorting (removing dead particles)
+  # Timesteps between particle re-sorting by tags (removing dead particles)
   #   @type: uint
   #   @default: 100
   #   @note: Set to 0 to disable re-sorting
   clear_interval = ""
+  # Timesteps between spatial sorting of particles (for better cache performance)
+  #   @type: uint
+  #   @default: 0
+  #   @note: Set to 0 to disable spatial sorting
+  spatial_sorting_interval = ""
 
   # @inferred:
   # - nspec
@@ -453,6 +458,18 @@
     #   @note: Only one emission mechanism allowed
     #   @note: Appropriate radiation drag flag will be applied automatically (unless explicitly set to "None")
     emission = ""
+    # Timesteps between spatial sorting of particles for given species
+    #   @type: uint
+    #   @default: 0
+    #   @note: Set to 0 to disable spatial sorting
+    #   @note: Overrides `particles.spatial_sorting_interval` for the given species
+    spatial_sorting_interval = ""
+    # Timesteps between particle re-sorting by tags (removing dead particles)
+    #   @type: uint
+    #   @default: 100
+    #   @note: Set to 0 to disable re-sorting
+    #   @note: Overrides `particles.clear_interval` for the given species
+    clear_interval = ""
 
 # Parameters for specific problem generators and setups
 [setup]
@@ -634,7 +651,7 @@
   #   @note: 0 = disable checkpointing
   #   @note: -1 = keep all checkpoints
   keep = ""
-  # Write a checkpoint once after a fixed walltime 
+  # Write a checkpoint once after a fixed walltime
   #   @type: string
   #   @default: "00:00:00"
   #   @note: The format is "HH:MM:SS"

minimal/CMakeLists.txt

@@ -83,7 +83,10 @@ if("KOKKOS" IN_LIST MODES)
 endif()
 
 if("ADIOS2_NOMPI" IN_LIST MODES)
-  set(libs stdc++fs)
+  set(libs "")
+  if (NOT "${CMAKE_CXX_COMPILER_ID}" MATCHES "AppleClang")
+    list(APPEND libs stdc++fs)
+  endif()
   set(exec adios2-nompi.xc)
   set(src ${CMAKE_CURRENT_SOURCE_DIR}/adios2.cpp)
 
@@ -97,7 +100,10 @@ if("ADIOS2_NOMPI" IN_LIST MODES)
 endif()
 
 if("ADIOS2_MPI" IN_LIST MODES)
-  set(libs stdc++fs)
+  set(libs "")
+  if (NOT "${CMAKE_CXX_COMPILER_ID}" MATCHES "AppleClang")
+    list(APPEND libs stdc++fs)
+  endif()
   set(exec adios2-mpi.xc)
   set(src ${CMAKE_CURRENT_SOURCE_DIR}/adios2.cpp)
 

src/engines/engine.hpp

@@ -22,7 +22,6 @@
 #include "utils/diag.h"
 #include "utils/reporter.h"
 #include "utils/timer.h"
-#include <toml11/toml.hpp>
 
 #include "archetypes/field_setter.h"
 #include "archetypes/traits.h"
@@ -35,6 +34,8 @@
 
 #include "pgen.hpp"
 
+#include <toml11/toml.hpp>
+
 #if defined(OUTPUT_ENABLED)
   #include <adios2.h>
 #endif
@@ -248,7 +249,7 @@ namespace ntt {
        "ParticlePusher", "FieldBoundaries",
        "ParticleBoundaries", "Communications",
        "Injector", "Custom",
-       "PrtlClear", "Output",
+       "ParticleSort", "Output",
        "Checkpoint" },
       []() {
         Kokkos::fence();

src/engines/grpic.hpp

@@ -19,7 +19,6 @@
 #include "utils/log.h"
 #include "utils/numeric.h"
 #include "utils/timer.h"
-#include <toml11/toml.hpp>
 
 #include "framework/domain/domain.h"
 #include "framework/parameters/parameters.h"
@@ -36,6 +35,7 @@
 
 #include <Kokkos_Core.hpp>
 #include <Kokkos_ScatterView.hpp>
+#include <toml11/toml.hpp>
 
 #include <string>
 #include <utility>
@@ -532,11 +532,9 @@ namespace ntt {
         timers.stop("FieldBoundaries");
       }
 
-      if (clear_interval > 0 and step % clear_interval == 0 and step > 0) {
-        timers.start("PrtlClear");
-        m_metadomain.RemoveDeadParticles(dom);
-        timers.stop("PrtlClear");
-      }
+      timers.start("ParticleSort");
+      m_metadomain.SortParticles(time, step, m_params, dom);
+      timers.stop("ParticleSort");
 
       /**
        * Finally: em0::B   at n-1/2

src/engines/srpic/srpic.hpp

@@ -18,7 +18,6 @@
 
 #include "utils/numeric.h"
 #include "utils/timer.h"
-#include <toml11/toml.hpp>
 
 #include "engines/srpic/currents.h"
 #include "engines/srpic/fields_bcs.h"
@@ -34,6 +33,7 @@
 
 #include <Kokkos_Core.hpp>
 #include <Kokkos_ScatterView.hpp>
+#include <toml11/toml.hpp>
 
 namespace ntt {
 
@@ -186,11 +186,9 @@ namespace ntt {
         timers.stop("Injector");
       }
 
-      if (clear_interval > 0 and step % clear_interval == 0 and step > 0) {
-        timers.start("PrtlClear");
-        m_metadomain.RemoveDeadParticles(dom);
-        timers.stop("PrtlClear");
-      }
+      timers.start("ParticleSort");
+      m_metadomain.SortParticles(time, step, m_params, dom);
+      timers.stop("ParticleSort");
     }
   };
 

src/framework/CMakeLists.txt

@@ -13,12 +13,16 @@
 # * simulation.cpp
 # * domain/grid.cpp
 # * domain/metadomain.cpp
-# * domain/communications.cpp
-# * domain/checkpoint.cpp
+# * domain/metadomain_sort.cpp
+# * domain/metadomain_comm.cpp
+# * domain/metadomain_chckpt.cpp
+# * domain/metadomain_stats.cpp
+# * domain/metadomain_io.cpp
 # * containers/particles.cpp
+# * containers/particles_comm.cpp
+# * containers/particles_io.cpp
+# * containers/particles_sort.cpp
 # * containers/fields.cpp
-# * domain/stats.cpp
-# * domain/output.cpp
 #
 # @includes:
 #
@@ -50,13 +54,15 @@ set(SOURCES
     ${SRC_DIR}/parameters/extra.cpp
     ${SRC_DIR}/domain/grid.cpp
     ${SRC_DIR}/domain/metadomain.cpp
-    ${SRC_DIR}/domain/communications.cpp
-    ${SRC_DIR}/domain/stats.cpp
+    ${SRC_DIR}/domain/metadomain_comm.cpp
+    ${SRC_DIR}/domain/metadomain_sort.cpp
+    ${SRC_DIR}/domain/metadomain_stats.cpp
     ${SRC_DIR}/containers/particles.cpp
+    ${SRC_DIR}/containers/particles_sort.cpp
     ${SRC_DIR}/containers/fields.cpp)
 if(${output})
-  list(APPEND SOURCES ${SRC_DIR}/domain/output.cpp)
-  list(APPEND SOURCES ${SRC_DIR}/domain/checkpoint.cpp)
+  list(APPEND SOURCES ${SRC_DIR}/domain/metadomain_io.cpp)
+  list(APPEND SOURCES ${SRC_DIR}/domain/metadomain_chckpt.cpp)
   list(APPEND SOURCES ${SRC_DIR}/containers/fields_io.cpp)
   list(APPEND SOURCES ${SRC_DIR}/containers/particles_io.cpp)
 endif()
@@ -68,7 +74,9 @@ add_library(ntt_framework ${SOURCES})
 set(libs ntt_global ntt_metrics ntt_kernels ntt_output)
 add_dependencies(ntt_framework ${libs})
 target_link_libraries(ntt_framework PUBLIC ${libs})
-target_link_libraries(ntt_framework PRIVATE stdc++fs)
+if (NOT "${CMAKE_CXX_COMPILER_ID}" MATCHES "AppleClang")
+  target_link_libraries(ntt_framework PRIVATE stdc++fs)
+endif()
 
 target_include_directories(
   ntt_framework