kkimj / performancefuzzer Goto Github PK

View Code? Open in Web Editor NEW

2.0 1.0 0.0 7.65 MB

PerformanceFuzzer

Home Page: https://pypi.org/project/PerformanceFuzzer/

License: GNU General Public License v3.0

Shell 0.68% C 11.30% Python 41.38% LLVM 46.64%

llvm-ir llvm clang fuzzing fuzzer

performancefuzzer's Introduction

PerformanceFuzzer

https://pypi.org/project/PerformanceFuzzer/

Getting Started

$ git clone https://github.com/KKimj/PerformanceFuzzer
$ pip install PerformanceFuzzer

Prerequisites

$ sudo apt update && sudo apt upgrade
$ sudo apt-get install libgmp3-dev
$ pip3 install pytest
$ pip3 install pytest-benchmark
$ pip3 install pygal

Using virtualenv

$ pip3 install virtualenv
$ virtualenv venv
$ python3 -m virtualenv venv
$ source venv/bin/activate

$ pip3 install pytest
$ pip3 install pytest-benchmark
$ pip3 install pygal

$ deactivate

LLVM

# .c file to .ll
$ clang -S -emit-llvm {source_file.c} -o {target_file.ll}

# .ll file optimization & .ll file to .S file
$ opt -S -O2 -aa -basicaa -tbaa -licm {target_file.ll}

# .S file compile to executable file
$ clang {target_file.s} -o {target_file} {Clang_compile_option}

# optional : Dump executable file
$ objdump -D {target_file} > {target_file.dump}

# example from test.c
$ opt -S -O2 -aa -basicaa -tbaa -licm test.ll -o test_opt.ll && llc test_opt.ll -o test.s && clang test.s -o test  -fopenmp=libiomp5 -lgmp -lssl -lcrypto

Running the tests

$ sudo python3 main.py cpubench cpubench --round=8 --warmup=2 --cpubench=50000
# same as
$ sudo python3 main.py cpubench cpubench -r 8 -w 2 -cpu 50000

$ python3 -m pytest main.py --benchmark-histogram

$ ./run.sh cpubench
$ sudo ./tests/cpubench 50000 --singlethreaded --printdigits

Examples

Acknowledgments

https://github.com/theblixguy/CPUBench

https://pypi.org/project/pytest-benchmark/

performancefuzzer's People

Contributors

Stargazers

Watchers

performancefuzzer's Issues

Final IR file

[Benchmark Controller] Compare method

    def compare(self, command1, command2, mode = 'perf'):
        return time1 - time2

HelloWorld code

clang -S -emit-llvm ./tests/helloworld/main.c -o ./test/helloworld/main.ll

opt -S -O2 -aa -basicaa -tbaa -licm ./tests/helloworld/main.ll -o ./tests/helloworld/main.ll

llc ./tests/helloworld/main.ll -o ./tests/helloworld/main.s

clang ./tests/helloworld/main.s -o ./tests/helloworld/main

Controller vs Handler vs Manager

https://tenlie10.tistory.com/164

main.py, CLI : option argument handling

def get_arguments():
    parser = argparse.ArgumentParser()
    parser.add_argument(nargs='+' ,help='Example) index.html', dest='filename')
    parser.add_argument('--optional', '-o', nargs='*', help='Example) save', default=[], dest='option')

    filename_list = parser.parse_args().filename
    option_list = parser.parse_args().option

    return filename_list, option_list

(benchmark) round와 warmup을 옵션으로 지정하는 것을 구현한다.

Insert policy

More Benchmark C code

https://people.sc.fsu.edu/~jburkardt/c_src/linpack_bench/linpack_bench.html

PHI nodes not grouped at top of basic block!

실행문

llc ./tests/cpubench/cpubench500_opt_fuzzer.ll

에러메세지

PHI nodes not grouped at top of basic block!
  %11 = phi i32 [ 0, %4 ], [ %9, %5 ]
label %10
llc: ./tests/cpubench/cpubench500_opt_fuzzer.ll: error: input module is broken!

[meeting] 2021-04-12

Use perf for cpu clock
Cpu clock to constant

pytest: error: unrecognized arguments: --benchmark-autosave

https://github.com/mcchae/pytest-benchmark

pytest-benchmark with Fuzzing

pytest benchmark Argument parsing for cpubench argument

check out,
How can parse pytest-benchmark an argument

IR Model & Controller

[meeting] 2021-05-10

Overfitting

Cache
Input coverage

parallel manager to scheduler

spec

[meeting] 2021-03-29

Data Padding

Stack
Global

[meeting] 2021-03-22

min -> avg

min value can be affected by the background service.

o3 vs -04

Check gcc cache optimization

clock option

cpu clock speed set constant NOT to be variable

afl, mutation policy

reference afl's mutation logic

input coverage

So that it doesn't apply only to certain situations.

IR file parse

%{num}

benchmark model & controller

 Performance counter stats for './tests/cpubench/cpubench 50000 --singlethreaded --printdigits' (2 runs):

          4,340.20 msec task-clock                #    0.998 CPUs utilized            ( +-  0.57% )
                32      context-switches          #    0.007 K/sec                    ( +- 10.77% )
                16      cpu-migrations            #    0.004 K/sec                    ( +-  3.03% )
               393      page-faults               #    0.091 K/sec                    ( +-  1.02% )
    21,042,138,266      cycles                    #    4.848 GHz                      ( +-  0.04% )
    53,506,630,089      instructions              #    2.54  insn per cycle           ( +-  0.00% )
     2,977,691,044      branches                  #  686.073 M/sec                    ( +-  0.03% )
        68,304,710      branch-misses             #    2.29% of all branches          ( +-  0.01% )

            4.3479 +- 0.0323 seconds time elapsed  ( +-  0.74% )

parallel

parse and build IR file
running bench

Mutation Model & Controller

[ML] Reinforcement learning

pytorch
tensorflow

[meeting] 2021-04-05

alloca with IR code parsing
Stack & Data memory check with GDB
time measure with cpu clock
parallel programming
Code convention improve

kkimj / performancefuzzer Goto Github PK

performancefuzzer's Introduction

PerformanceFuzzer

Getting Started

Prerequisites

Using virtualenv

LLVM

Running the tests

Examples

Acknowledgments

performancefuzzer's People

Contributors

Stargazers

Watchers

performancefuzzer's Issues

실행문

에러메세지

Overfitting

parallel manager to scheduler

min -> avg

o3 vs -04

clock option

afl, mutation policy

input coverage

Recommend Projects

Recommend Topics

Recommend Org