monolish: MONOlithic Liner equation Solvers for Highly-parallel architecture

Related tags

Machine Learningcpuhpcgpumatrixlinear-algebracpp14openmpcudascientific-computingblaslapacklinear-algebra-librarysparse-matrixmklmatrix-structures
Overview

monolish: MONOlithic LIner equation Solvers for Highly-parallel architecture

monolish is a linear equation solver library that monolithically fuses variable data type, matrix structures, matrix data format, vendor specific data transfer APIs, and vendor specific numerical algebra libraries.

monolish let developer forget about:

  • Performance tuning
  • Processor differences which execute library (Intel CPU, NVIDIA GPU, AMD CPU, ARM CPU, NEC SX-Aurora TSUBASA, etc.)
  • Vendor specific data transfer APIs (host RAM to Device RAM)
  • Finding bottlenecks and performance benchmarks
  • The argument data type of matrix/vector operations
  • Matrix structures and storage formats
  • Cumbersome package dependency

License

Copyright 2021 RICOS Co. Ltd.

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

Comments
  • It seems that the `Dense(M, N, min, max)` constructor is not completely random.

    It seems that the `Dense(M, N, min, max)` constructor is not completely random.

    Running the following simple program

    #include <iostream>
#include <monolish_blas.hpp>

int main() {
  monolish::matrix::Dense<double>x(2, 3, 0.0, 10.0);
  x.print_all();
  return 0;
}

    $ g++ -O3 main.cpp -o main.out -lmonolish_cpu

    will produce results like this.

    [email protected]:/$ ./main.out
1 1 5.27196
1 2 2.82358 <--
1 3 2.13893 <--
2 1 9.72054
2 2 2.82358 <--
2 3 2.13893 <--
[email protected]:/$ ./main.out
1 1 5.3061
1 2 9.75236
1 3 7.15652
2 1 5.28961
2 2 2.05967
2 3 0.59838
[email protected]:/$ ./main.out
1 1 9.33149 <--
1 2 4.75639 <--
1 3 8.71093 <--
2 1 9.33149 <--
2 2 4.75639 <--
2 3 8.71093 <--

    The arrows (<--) indicate that the number is repeating.

    This is probably due to that the pseudo-random number generator does not split well when it is parallelized by OpenMP.

    https://github.com/ricosjp/monolish/blob/1b89942e869b7d0acd2d82b4c47baeba2fbdf3e6/src/utils/dense_constructor.cpp#L120-L127

    This may happen not only with Dense, but also with random constructors of other data structures.

    I tested this on docker image ghcr.io/ricosjp/monolish/mkl:0.14.1.

    opened by lotz84 5
  • impl. transpose matvec, matmul

    impl. transpose matvec, matmul

    I want to give modern and intuitive transposition information. But I have no idea how to implement it easily.

    First, we create the following function as a prototype

    matmul(A,B,C) // C=AB
matmul_TNN(A, B, C); // C=A^TB
matvec(A,x,y); // y = Ax
matvec_T(A, x, y); // y=A^Tx

    This interface is not beautiful. However, it has the following advantages

    • It does not affect other functions.
    • Easy to trace with logger
    • Simple to implement FFI in the future.
    • When beautiful ideas appear in the future, these functions can be implemented wrapping it.
    opened by t-hishinuma 2
  • try -fopenmp-cuda-mode flag

    try -fopenmp-cuda-mode flag

    memo:

    Clang supports two data-sharing models for Cuda devices: Generic and Cuda modes. The default mode is Generic. Cuda mode can give an additional performance and can be activated using the -fopenmp-cuda-mode flag. In Generic mode all local variables that can be shared in the parallel regions are stored in the global memory. In Cuda mode local variables are not shared between the threads and it is user responsibility to share the required data between the threads in the parallel regions.

    https://clang.llvm.org/docs/OpenMPSupport.html#basic-support-for-cuda-devices

    opened by t-hishinuma 2
  • Reserch the effect of the level information of the performance of cusparse ILU precondition

    Reserch the effect of the level information of the performance of cusparse ILU precondition

    The level information may not improve the performance but spend extra time doing analysis. For example, a tridiagonal matrix has no parallelism. In this case, CUSPARSE_SOLVE_POLICY_NO_LEVEL performs better than CUSPARSE_SOLVE_POLICY_USE_LEVEL. If the user has an iterative solver, the best approach is to do csrsv2_analysis() with CUSPARSE_SOLVE_POLICY_USE_LEVEL once. Then do csrsv2_solve() with CUSPARSE_SOLVE_POLICY_NO_LEVEL in the first run and with CUSPARSE_SOLVE_POLICY_USE_LEVEL in the second run, picking faster one to perform the remaining iterations.

    https://docs.nvidia.com/cuda/cusparse/index.html#csric02

    opened by t-hishinuma 2
  • ignoring return value in test

    ignoring return value in test 

    matrix_transpose.cpp:60:3: warning: ignoring return value of 'monolish::matrix::COO<Float>& monolish::matrix::COO<Float>::transpose() [with Float = double]', declared with attribute nodiscard [-Wunused-result]
   60 |   A.transpose();
    opened by t-hishinuma 2
  • Automatic deploy at release

    Automatic deploy at release

    impl. in github actions

    • [x] generate Doxyben (need to chenge version name)
    • [x] generate deb file
    • [x] generate monolish docker

    need to get version number...

    opened by t-hishinuma 2
  • write how to install nvidia-docker

    write how to install nvidia-docker 

    distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee > /etc/apt/sources.list.d/nvidia-docker.list
sudo apt update -y
sudo apt install -y nvidia-docker2
sudo systemctl restart docker
    opened by t-hishinuma 1
  • Resolve conflict of libmonolish-cpu and libmonolish-nvidia-gpu deb package

    Resolve conflict of libmonolish-cpu and libmonolish-nvidia-gpu deb package

    What conflicts?

    libomp.so is contained in both package

    How to resolve?

    • (a) Use libomp5-12 distributed by ubuntu
    • (b) Create another package of libomp in allgebra stage (libomp-allgebra)
    opened by termoshtt 1
  • resolve curse of type name in src/

    resolve curse of type name in src/

    In src/, int and size_t are written. When change the class or function declarations in include/, I don't want to rewrite src/. Use auto or decltype() to remove them.

    opened by t-hishinuma 1
  • LLVM OpenMP Offloading can be installed by apt?

    LLVM OpenMP Offloading can be installed by apt? 

    docker run -it --gpus all -v $PWD:/work nvidia/cuda:11.7.0-devel-ubuntu22.04
==
apt update -y
apt install -y git intel-mkl cmake ninja-build ccache clang clang-tools libomp-14-dev gcc gfortran
git config --global --add safe.directory /work
cd /work; make gpu

    pass??

    opened by t-hishinuma 0
  • cusparse IC / ILU functions is deprecated

    cusparse IC / ILU functions is deprecated

    but, sample code of cusparse is not updated

    https://docs.nvidia.com/cuda/cusparse/index.html#csric02

    I dont like trial and error, so wait for the sample code to be updated.

    opened by t-hishinuma 0
Releases(0.17.0)
Owner
RICOS Co. Ltd.
株式会社科学計算総合研究所 / Research Institute for Computational Science Co. Ltd.
RICOS Co. Ltd.
GitHub Repository
This repo implements a Topological SLAM: Deep Visual Odometry with Long Term Place Recognition (Loop Closure Detection)

This repo implements a topological SLAM system. Deep Visual Odometry (DF-VO) and Visual Place Recognition are combined to form the topological SLAM system.

Best of Australian Centre for Robotic Vision (ACRV) 32 Jun 23, 2022
Tool for producing high quality forecasts for time series data that has multiple seasonality with linear or non-linear growth.

Prophet: Automatic Forecasting Procedure Prophet is a procedure for forecasting time series data based on an additive model where non-linear trends ar

Facebook 15.4k Jan 07, 2023
A Python implementation of GRAIL, a generic framework to learn compact time series representations.

GRAIL A Python implementation of GRAIL, a generic framework to learn compact time series representations. Requirements Python 3.6+ numpy scipy tslearn

3 Nov 24, 2021
Transform ML models into a native code with zero dependencies

m2cgen (Model 2 Code Generator) - is a lightweight library which provides an easy way to transpile trained statistical models into a native code

Bayes' Witnesses 2.3k Jan 03, 2023
A data preprocessing and feature engineering script for a machine learning pipeline is prepared.

FEATURE ENGINEERING Business Problem: A data preprocessing and feature engineering script for a machine learning pipeline needs to be prepared. It is

Pinar Oner 7 Dec 18, 2021
Pyomo is an object-oriented algebraic modeling language in Python for structured optimization problems.

Pyomo is a Python-based open-source software package that supports a diverse set of optimization capabilities for formulating and analyzing optimization models. Pyomo can be used to define symbolic p

Pyomo 1.4k Dec 28, 2022
An open source framework that provides a simple, universal API for building distributed applications. Ray is packaged with RLlib, a scalable reinforcement learning library, and Tune, a scalable hyperparameter tuning library.

Ray provides a simple, universal API for building distributed applications. Ray is packaged with the following libraries for accelerating machine lear

23.3k Dec 31, 2022
🎛 Distributed machine learning made simple.

🎛 lazycluster Distributed machine learning made simple. Use your preferred distributed ML framework like a lazy engineer. Getting Started • Highlight

Machine Learning Tooling 44 Nov 27, 2022
Simple and flexible ML workflow engine.

This is a simple and flexible ML workflow engine. It helps to orchestrate events across a set of microservices and create executable flow to handle requests. Engine is designed to be configurable wit

Katana ML 295 Jan 06, 2023
A Streamlit demo to interactively visualize Uber pickups in New York City

Streamlit Demo: Uber Pickups in New York City A Streamlit demo written in pure Python to interactively visualize Uber pickups in New York City. View t

Streamlit 230 Dec 28, 2022
A basic Ray Tracer that exploits numpy arrays and functions to work fast.

Python-Fast-Raytracer A basic Ray Tracer that exploits numpy arrays and functions to work fast. The code is written keeping as much readability as pos

Rafael de la Fuente 393 Dec 27, 2022
Implementation of the Object Relation Transformer for Image Captioning

Object Relation Transformer This is a PyTorch implementation of the Object Relation Transformer published in NeurIPS 2019. You can find the paper here

Yahoo 158 Dec 24, 2022
Random Forest Classification for Neural Subtypes

Random Forest classifier for neural subtypes extracted from extracellular recordings from human brain organoids.

Michael Zabolocki 1 Jan 31, 2022
Responsible AI Workshop: a series of tutorials & walkthroughs to illustrate how put responsible AI into practice

Responsible AI Workshop Responsible innovation is top of mind. As such, the tech industry as well as a growing number of organizations of all kinds in

Microsoft 9 Sep 14, 2022
Can a machine learning project be implemented to estimate the salaries of baseball players whose salary information and career statistics for 1986 are shared?

END TO END MACHINE LEARNING PROJECT ON HITTERS DATASET Can a machine learning project be implemented to estimate the salaries of baseball players whos

Pinar Oner 7 Dec 18, 2021
nn-Meter is a novel and efficient system to accurately predict the inference latency of DNN models on diverse edge devices

A DNN inference latency prediction toolkit for accurately modeling and predicting the latency on diverse edge devices.

Microsoft 241 Dec 26, 2022
Simulate & classify transient absorption spectroscopy (TAS) spectral features for bulk semiconducting materials (Post-DFT)

PyTASER PyTASER is a Python (3.9+) library and set of command-line tools for classifying spectral features in bulk materials, post-DFT. The goal of th

Materials Design Group 4 Dec 27, 2022
Optimal Randomized Canonical Correlation Analysis

ORCCA Optimal Randomized Canonical Correlation Analysis This project is for the python version of ORCCA algorithm. It depends on Numpy for matrix calc

Yinsong Wang 1 Nov 21, 2021
Adaptive: parallel active learning of mathematical functions

adaptive Adaptive: parallel active learning of mathematical functions. adaptive is an open-source Python library designed to make adaptive parallel fu

741 Dec 27, 2022
scikit-fem is a lightweight Python 3.7+ library for performing finite element assembly.

scikit-fem is a lightweight Python 3.7+ library for performing finite element assembly. Its main purpose is the transformation of bilinear forms into sparse matrices and linear forms into vectors.

Tom Gustafsson 297 Dec 13, 2022