Add fine-grained step timing and trim BH RHS overhead

This commit is contained in:
2026-04-13 14:50:55 +08:00
parent f3988ac8ca
commit 968522995b
4 changed files with 497 additions and 235 deletions

View File

@@ -144,6 +144,34 @@ def generate_macrodef_h():
print( "#define REGLEV 0", file=file1 )
print( file=file1 )
# Define fine-grained timestep timing macros
# These default to enabled profiling without requiring AMSS_NCKU_Input.py edits.
fine_timing = getattr(input_data, "Fine_Timing",
getattr(input_data, "Finegrained_Timing", "yes"))
timing_report_every = max(1, int(getattr(
input_data, "Timing_Every_Steps",
getattr(input_data, "Timing_Report_Every", 1))))
timing_top_hotspots = max(1, int(getattr(
input_data, "Timing_Top_Hotspots", 8)))
if ( fine_timing == "yes" ):
print( "#define BSSN_FINE_TIMING 1", file=file1 )
print( file=file1 )
elif ( fine_timing == "no" ):
print( "#define BSSN_FINE_TIMING 0", file=file1 )
print( file=file1 )
else:
print( "Fine_Timing setting error!!!" )
print()
print( "# Fine_Timing setting error!!!", file=file1 )
print( file=file1 )
print( f"#define BSSN_FINE_TIMING_EVERY {timing_report_every}", file=file1 )
print( file=file1 )
print( f"#define BSSN_FINE_TIMING_TOPN {timing_top_hotspots}", file=file1 )
print( file=file1 )
# Define macro USE_GPU
# use GPU or not
@@ -224,6 +252,15 @@ def generate_macrodef_h():
print( "// 0: for every level;", file=file1 )
print( "// 1: for all", file=file1 )
print( "//", file=file1 )
print( "// define BSSN_FINE_TIMING", file=file1 )
print( "// enable fine-grained per-timestep timing monitor", file=file1 )
print( "//", file=file1 )
print( "// define BSSN_FINE_TIMING_EVERY", file=file1 )
print( "// report timing every N coarse timesteps", file=file1 )
print( "//", file=file1 )
print( "// define BSSN_FINE_TIMING_TOPN", file=file1 )
print( "// number of hottest timing buckets shown in stdout", file=file1 )
print( "//", file=file1 )
print( "// define USE_GPU", file=file1 )
print( "// use gpu or not", file=file1 )
print( "//", file=file1 )