Add memory profiler

[GindaChen]

Change overview

• Add mem_profiler.template.cpp - a simple memory profiler that uses tcmalloc to log alloc/dealloc event.
• Add memory profiling workflow (run_memprofile.sh). Usage: sh run_memprofiler.sh <ScaleFactor>.

Cavets

• Must run on lab07. The script still assume we are on lab07 machine.
• Must manually copy profiling code for memory profiling. So far we haven't found a good way to separate the profiler code from source code.

Example Usage

cd /home/junda/Scad/runtime/test/src/tpcds_1_cpp
sh run_memprofiler.sh 1 # test on scale factor 1 dataset
cd /home/junda/Scad/runtime/test/src/tpcds_1_cpp/analysis_memory
# see the output files:
# - tpcds_1.sf_1.lifetime.csv
#     Variable's lifetime. Schema: (variable, start_line, end_line)
# - tpcds_1.sf_1.line_activity.csv
#     Mem activity. Schema: (index, event, timestamp, size, line_num)
# - tpcds_1.sf_1.final.csv
#     Final result

[GindaChen]

def get_used_node_types(target):
    node_types = set()

    def get_all_used_node_types(node):
        node_types.add(node.kind)
        for child in node.get_children():
            get_all_used_node_types(child)

    get_all_used_node_types(target)
    return node_types

[GindaChen]

    f = node.extent.start.file.name