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
Flask app to predict daily radiation from the time series of Solcast from Islamabad, Pakistan

Solar-radiation-ISB-MLOps - Flask app to predict daily radiation from the time series of Solcast from Islamabad, Pakistan.

Abid Ali Awan 1 Dec 31, 2021
Painless Machine Learning for python based on scikit-learn

PlainML Painless Machine Learning Library for python based on scikit-learn. Install pip install plainml Example from plainml import KnnModel, load_ir

1 Aug 06, 2022
whylogs: A Data and Machine Learning Logging Standard

whylogs: A Data and Machine Learning Logging Standard whylogs is an open source standard for data and ML logging whylogs logging agent is the easiest

WhyLabs 2k Jan 06, 2023
LibRerank is a toolkit for re-ranking algorithms. There are a number of re-ranking algorithms, such as PRM, DLCM, GSF, miDNN, SetRank, EGRerank, Seq2Slate.

LibRerank LibRerank is a toolkit for re-ranking algorithms. There are a number of re-ranking algorithms, such as PRM, DLCM, GSF, miDNN, SetRank, EGRer

126 Dec 28, 2022
ML Optimizers from scratch using JAX

Toy implementations of some popular ML optimizers using Python/JAX

Shreyansh Singh 38 Jul 29, 2022
50% faster, 50% less RAM Machine Learning. Numba rewritten Sklearn. SVD, NNMF, PCA, LinearReg, RidgeReg, Randomized, Truncated SVD/PCA, CSR Matrices all 50+% faster

[Due to the time taken @ uni, work + hell breaking loose in my life, since things have calmed down a bit, will continue commiting!!!] [By the way, I'm

Daniel Han-Chen 1.4k Jan 01, 2023
Primitives for machine learning and data science.

An Open Source Project from the Data to AI Lab, at MIT MLPrimitives Pipelines and primitives for machine learning and data science. Documentation: htt

MLBazaar 65 Dec 29, 2022
A simple application that calculates the probability distribution of a normal distribution

probability-density-function General info An application that calculates the probability density and cumulative distribution of a normal distribution

1 Oct 25, 2022
Coursera Machine Learning - Python code

Coursera Machine Learning This repository contains python implementations of certain exercises from the course by Andrew Ng. For a number of assignmen

Jordi Warmenhoven 859 Dec 10, 2022
Made in collaboration with Chris George for Art + ML Spring 2019.

Deepdream Eyes Made in collaboration with Chris George for Art + ML Spring 2019.

Francisco Cabrera 1 Jan 12, 2022
XGBoost-Ray is a distributed backend for XGBoost, built on top of distributed computing framework Ray.

XGBoost-Ray is a distributed backend for XGBoost, built on top of distributed computing framework Ray.

92 Dec 14, 2022
Machine learning that just works, for effortless production applications

Machine learning that just works, for effortless production applications

Elisha Yadgaran 16 Sep 02, 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
A unified framework for machine learning with time series

Welcome to sktime A unified framework for machine learning with time series We provide specialized time series algorithms and scikit-learn compatible

The Alan Turing Institute 6k Jan 06, 2023
Data science, Data manipulation and Machine learning package.

duality Data science, Data manipulation and Machine learning package. Use permitted according to the terms of use and conditions set by the attached l

David Kundih 3 Oct 19, 2022
This handbook accompanies the course: Machine Learning with Hung-Yi Lee

This handbook accompanies the course: Machine Learning with Hung-Yi Lee

RenChu Wang 472 Dec 31, 2022
My project contrasts K-Nearest Neighbors and Random Forrest Regressors on Real World data

kNN-vs-RFR My project contrasts K-Nearest Neighbors and Random Forrest Regressors on Real World data In many areas, rental bikes have been launched to

1 Oct 28, 2021
Evaluate on three different ML model for feature selection using Breast cancer data.

Anomaly-detection-Feature-Selection Evaluate on three different ML model for feature selection using Breast cancer data. ML models: SVM, KNN and MLP.

Tarek idrees 1 Mar 17, 2022
A simple machine learning python sign language detection project.

SST Coursework 2022 About the app A python application that utilises the tensorflow object detection algorithm to achieve automatic detection of ameri

Xavier Koh 2 Jun 30, 2022
Scalable, Portable and Distributed Gradient Boosting (GBDT, GBRT or GBM) Library, for Python, R, Java, Scala, C++ and more. Runs on single machine, Hadoop, Spark, Dask, Flink and DataFlow

eXtreme Gradient Boosting Community | Documentation | Resources | Contributors | Release Notes XGBoost is an optimized distributed gradient boosting l

Distributed (Deep) Machine Learning Community 23.6k Jan 03, 2023