Add memory profiler

runtime/test/src/tpcds_1_cpp/Makefile

@@ -1,18 +1,32 @@
 PYTHON_CONFIG := python3-config
+# CFLAGS  := $(CFLAGS) -O3
 CFLAGS  := $(CFLAGS) $(shell python3-config --cflags)
 CFLAGS  := $(CFLAGS) -I$(shell python3 -c "import numpy as np; print(np.get_include());")
-LDFLAGS := -lpython3.8 $(shell python3-config --ldflags)
+CFLAGS  := $(CFLAGS) -g -pg -O0 -std=c++17
+LDFLAGS := -lpython3.8 $(shell python3-config --ldflags) 
 
 .PHONY: all clean
 
-all: tpcds_1
+all: tpcds_1 tpcds_1_mem tpcds_1_cpu
 
 clean:
-	rm reference.o tpcds_1
+	rm *.o tpcds_1 tpcds_1_mem tpcds_1_cpu
 
 reference.o: reference.c
-	gcc -c $(CFLAGS) -g -o $@ -fPIC $^
+	g++ -c $(CFLAGS) -g -o $@ -fPIC $^
+
+reference_cpu.o: reference_cpu.cpp
+	g++ -c $(CFLAGS) -g -o $@ -fPIC $^
+
+reference_mem.o: reference_mem.cpp
+	g++ -c $(CFLAGS) -g -o $@ -fPIC $^
 
 tpcds_1: reference.o
-	gcc $^ -o $@ $(LDFLAGS)
+	g++ $^ -o $@ $(LDFLAGS)
+
+tpcds_1_mem: reference_mem.o
+	g++ $^ -o $@ $(LDFLAGS) -ltcmalloc -lprofiler
+
+tpcds_1_cpu: reference_cpu.o
+	g++ $^ -o $@ $(LDFLAGS)
 

runtime/test/src/tpcds_1_cpp/analysis_memory/analyze_tpcds_1.sh

@@ -0,0 +1,3 @@
+python lifetime.py ../reference_mem.cpp --target_function main_ > tpcds.lifetime.csv
+python memusage.py  ./tpcds_1.mem.log --source_path ../reference_mem.cpp --exec_path ../tpcds_1_mem > tpcds.line_activity.csv
+python final.py --lifetime_csv_path tpcds.lifetime.csv --line_activity_csv_path tpcds.line_activity.csv

runtime/test/src/tpcds_1_cpp/analysis_memory/final.py

@@ -0,0 +1,45 @@
+# Calculate the per-symbol lifetime and memory usage.
+
+import argparse 
+
+import pandas as pd
+
+def parse_args():
+    args = argparse.ArgumentParser(
+        usage='python final.py --lifetime_csv_path tpcds.lifetime.csv --line_activity_csv_path tpcds.line_activity.csv'
+    )
+    args.add_argument('--lifetime_csv_path', type=str, help='Path to the lifetime CSV file.', required=True)
+    args.add_argument('--line_activity_csv_path', type=str, help='Path to the line activity CSV file.', required=True)
+    args.add_argument('--display_markdown', action='store_true', help='Print markdown format to stdout.', default=False)
+    return args.parse_args()
+
+def human_readable_size(x):
+    if x < 1024:
+        return f'{x}B'
+    elif x < 1024 * 1024:
+        return f'{x / 1024:.2f}KB'
+    elif x < 1024 * 1024 * 1024:
+        return f'{x / 1024 / 1024:.2f}MB'
+    else:
+        return f'{x / 1024 / 1024 / 1024:.2f}GB'
+
+def main():
+    args = parse_args()
+    lifetime_df = pd.read_csv(args.lifetime_csv_path)
+    line_activity_df = pd.read_csv(args.line_activity_csv_path, index_col=0)
+    line_activity_df = line_activity_df.drop(columns=['timestamp'])
+    line_activity_df = line_activity_df[line_activity_df.event == 'NewHook']
+    line_activity_df = line_activity_df.drop(columns=['event'])
+    line_activity_df = line_activity_df.groupby(['line_num']).sum().reset_index()
+
+    line_activity_df = line_activity_df.rename(columns={'size': 'alloc_footprint'})
+    line_activity_df['alloc_footprint_h'] = line_activity_df['alloc_footprint'].apply(human_readable_size)
+
+    df = lifetime_df.merge(line_activity_df, left_on='start_line', right_on='line_num', how='inner')
+    if args.display_markdown:
+        print(df.to_markdown())
+    else:
+        print(df.to_csv())
+
+if __name__ == '__main__':
+    main()