taskbench

A C++ Library implementing different common tasks for benchmarking purposes

Build

Taskbenchs desired build system is CMake. This has several reasons but the most important is the convenience that all used external libraries build using CMake. Thus, all dependencies are included into this repository and build together with taskbench.

Requirements

• A C++ compiler supporting C++20 features (including concepts):
  • g++>=10
  • clang>=10
  • msvc>=16.3
• CMake (Version 3.9)
• OpenCL (Your GPU drivers already provide OpenCL. Only install OpenCL if you have no external GPU.)
  • Windows: OpenCL™ and OpenGL® Compatibility Pack
  • Linux: ocl-icd-opencl-dev - Search for your specific distribution and package manage...

Dependencies

taskbench depends on the following libraries:

• ZStandard
• lfreist/AES (fork of SergeyBel/AES)
• fmtlib
• json
• miss-ocl (An OpenCL wrapper including the OpenCL headers)

All these dependencies are included into this repository. No need to install anything!

Build

git submodule update --init --recursive
cmake -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build --config Release

This builds all target libraries as shared and static libraries.

Include taskbench to your CMake project

1. Add taskbench to your project. Choose one of these options:
   1. Clone it from GitHub into external/taskbench: git clone https://github.com/lfreist/taskbench.git
   2. Include it as submodule (preferred): git submodule add https://github.com/lfreist/taskbench.git external/taskbench
2. Add the following to your projects root CMakeLists.txt:

   # Your setup ...
   include_directories(external/taskbench/include)
   add_subdirectory(external/taskbench)
   # ...

3. Include taskbench to your source files:

   // main.cpp
   
   #include <taskbench/taskbench.h>
   
   int main(int argc, char** argv) {
     // run all CPU benchmarks
     taskbench::cpu::Benchmark cpu_benchmark;
     cpu_benchmark.run_all(taskbench::seconds(3));
     return 0;
   }

4. Link against taskbench or taskbench_static:

   # ...
   add_executable(main main.cpp)
   target_link_libraries(main taskbench::taskbench_static)

Run BlitzBench

Taskbench provides a command line tool BlitzBench that runs the implemented benchmark tasks and prints out the results. BlitzBench is built together with the library as described above.

./build/taskbench/BlitzBench # Linux/MacOS
.\build\taskbench\BlitzBench # Windows

Example Output

RAM Benchmarks:
    read                 (0.018 +/- 0.001) s [59.91 GiB/s]
    write                (0.017 +/- 0.001) s [66.17 GiB/s]
    read/write           (0.020 +/- 0.003) s [67.54 GiB/s]
CPU Benchmarks:
  AES Benchmarks:
    AES encryption       (1.330 +/- 0.005) s 
    AES decryption       (1.417 +/- 0.001) s 
  ZStandard Benchmarks:
    ZSTD compression     (0.874 +/- 0.001) s 
    ZSTD decompression   (0.769 +/- 0.000) s 
  FFT Benchmarks:
    FFT                  (0.722 +/- 0.006) s 
    Inverse FFT          (0.762 +/- 0.003) s 
  Matrix Multiplication Benchmarks:
    mmul (int)           (1.407 +/- 0.001) s [1.53 G op/s]
    mmul (double)        (1.653 +/- 0.004) s [1.30 G op/s]
  Sort Benchmarks:
    sort (int)           (2.057 +/- 0.001) s 
    sort (double)        (2.371 +/- 0.001) s 
    sort (string)        (1.188 +/- 0.000) s 
  Synthetic Benchmarks:
    add/sub (int)        (1.264 +/- 0.009) s [3.18 G op/s]
    mul (int)            (1.901 +/- 0.000) s [2.10 G op/s]
    div (int)            (1.096 +/- 0.000) s [365.23 M op/s]
    add/sub (double)     (1.567 +/- 0.000) s [2.55 G op/s]
    mul (double)         (3.216 +/- 0.000) s [1.24 G op/s]
    div (int)            (0.901 +/- 0.169) s [530.09 M op/s]
GPU Benchmarks:
  Matrix Multiplication Benchmarks:
    mmul                 (1.471 +/- 0.009) s [11.75 G op/s]
  Synthetic Benchmarks:
    integer ops          (0.982 +/- 0.076) s [596.29 G op/s]
    float ops            (0.328 +/- 0.002) s [1.69 T op/s]
  Memory Benchmarks:
    memory write         (0.020 +/- 0.000) s [51.35 GiB/s]
    memory read          (0.016 +/- 0.000) s [61.86 GiB/s]
  Transfer Benchmarks:
    write to device      (0.074 +/- 0.000) s [13.51 GiB/s]
    read from device     (0.074 +/- 0.000) s [13.50 GiB/s]