Feature benchmark search

CMakeLists.txt

@@ -54,6 +54,7 @@ option( T8CODE_ENABLE_MPI "Enable t8code's features which rely on MPI" ON )
 option( T8CODE_ENABLE_VTK "Enable t8code's features which rely on VTK" OFF )
 option( T8CODE_ENABLE_OCC "Enable t8code's features which rely on OpenCASCADE" OFF )
 option( T8CODE_ENABLE_NETCDF "Enable t8code's features which rely on netCDF" OFF )
+option( T8CODE_ENABLE_STANDALONE_LOOP_UNROLL "Enables t8code's loop unroll in standalone scheme" OFF )
 
 option( T8CODE_USE_SYSTEM_SC "Use system-installed sc library" OFF )
 option( T8CODE_USE_SYSTEM_P4EST "Use system-installed p4est library" OFF )
@@ -64,6 +65,8 @@ cmake_dependent_option( T8CODE_BUILD_DOCUMENTATION_SPHINX "Build t8code's docume
 set(T8CODE_CUSTOM_PARALLEL_TEST_COMMAND "" CACHE STRING "Define a custom command for parallel tests , e.g.: mpirun -np 8 (overwrites standard mpirun -np 4 if build with mpi)")
 set(T8CODE_CUSTOM_SERIAL_TEST_COMMAND "" CACHE STRING "Define a custom command for serial tests.")
 
+option( T8CODE_ENABLE_PROFILING_BARRIERS "Use MPI-Barriers for profiling" OFF )
+
 # Reading the option T8CODE_TEST_LEVEL: Possible choices are T8_TEST_LEVEL_FULL, T8_TEST_LEVEL_MEDIUM, or T8_TEST_LEVEL_BASIC. Default is T8_TEST_LEVEL_FULL.
 set(T8CODE_TEST_LEVEL "T8_TEST_LEVEL_FULL" CACHE STRING "Test level: T8_TEST_LEVEL_FULL for full tests, T8_TEST_LEVEL_MEDIUM for less thorough tests, T8_TEST_LEVEL_BASIC for minimal tests. (WARNING: Use with care.)")
 set_property(CACHE T8CODE_TEST_LEVEL PROPERTY STRINGS "T8_TEST_LEVEL_FULL" "T8_TEST_LEVEL_MEDIUM" "T8_TEST_LEVEL_BASIC")

benchmarks/CMakeLists.txt

@@ -44,6 +44,9 @@ function( add_t8_benchmark )
     if( T8CODE_EXPORT_COMPILE_COMMANDS )
         set_target_properties( ${ADD_T8_BENCHMARK_NAME} PROPERTIES EXPORT_COMPILE_COMMANDS ON )
     endif( T8CODE_EXPORT_COMPILE_COMMANDS )
+    if( T8CODE_ENABLE_PROFILING_BARRIERS )
+        target_compile_definitions( ${ADD_T8_BENCHMARK_NAME} PUBLIC T8_ENABLE_PROFILE_BARRIER=1 )
+    endif( T8CODE_ENABLE_PROFILING_BARRIERS )
 
     install( TARGETS ${ADD_T8_BENCHMARK_NAME} DESTINATION ${CMAKE_INSTALL_BINDIR} )
 endfunction()
@@ -54,3 +57,4 @@ add_t8_benchmark( NAME t8_time_prism_adapt SOURCES t8_time_prism_adapt.cxx )
 add_t8_benchmark( NAME t8_time_fractal SOURCES t8_time_fractal.cxx )
 add_t8_benchmark( NAME t8_time_set_join_by_vertices SOURCES t8_time_set_join_by_vertices.cxx )
 add_t8_benchmark( NAME t8_time_new_refine SOURCES t8_time_new_refine.c )
+add_t8_benchmark( NAME t8_time_search SOURCES t8_time_search.cxx )

benchmarks/search_benchmark.txt

@@ -0,0 +1,14 @@
+The search benchmark constructs a partitioned uniform refinement of a single coarse mesh element.
+Afterwards, generates a single search point in the center of each local element, and finally searches for all these points
+in the local part of the mesh.
+Finally, each cell that had at least one point found, is refined (here uniform refinement, since each element has exactly one search point).
+This cycle is repeated for num_repetitions with a new mesh each time.
+
+Instructions for running the search benchmark:
+
+It has 4 relevant parameters: scheme, eclass, level, repititions
+An example call might look like:
+
+mpirun -n 8 ./t8_time_search -s1 -e4 -l6 -r20
+
+The file search_benchmark_runscript.py provides the python script used to create the benchmarks in the paper.

benchmarks/search_benchmark_runscript.py

@@ -0,0 +1,40 @@
+from pathlib import Path
+import subprocess
+
+eclasses = [4, 5, 6, 7]
+schemes = [0, 1]
+max_proc_log = 14
+
+max_elem_log=20
+min_elem_log=5
+repititions = 20
+
+maxlevel=18
+dim=3
+
+#CARO has 128 =2^^7 procs per node
+log_procs_per_node=7
+
+out_directory = "out"
+
+for eclass in eclasses:
+    for scheme in schemes:
+        for proc_log in range(max_proc_log+1):
+                n_nodes = max(1, int(2**(proc_log- log_procs_per_node)))
+                string="#!/bin/bash\n"
+                string += "#SBATCH --time=00:15:00\n"
+                string += f"#SBATCH -N {n_nodes}\n"
+                string += f"#SBATCH --ntasks={2**proc_log}\n"
+                for level in range(maxlevel+1,-1,-1):
+                    if (max_elem_log >= dim*level - proc_log >= min_elem_log):
+                        string += f"echo srun ./t8_time_search -s{scheme} -e{eclass} -l{level} -r{repititions}\n"
+                        string += f"srun -o {out_directory}/e{eclass}/s{scheme}/p{proc_log}_l{level}_%j ./t8_time_search -s{scheme} -e{eclass} -l{level} -r{repititions}\n"
+
+                file = Path(f"{out_directory}/e{eclass}/s{scheme}/scripts/p{proc_log}.sh")
+                file.parent.mkdir(exist_ok=True, parents=True)
+                file.write_text(string)
+
+for eclass in eclasses:
+    for scheme in schemes:
+        for proc_log in range(max_proc_log+1):
+            subprocess.call(f"sbatch {out_directory}/e{eclass}/s{scheme}/scripts/p{proc_log}.sh", shell=True)