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
ParaMonte is a serial/parallel library of Monte Carlo routines for sampling mathematical objective functions of arbitrary-dimensions

ParaMonte is a serial/parallel library of Monte Carlo routines for sampling mathematical objective functions of arbitrary-dimensions, in particular, the posterior distributions of Bayesian models in

Computational Data Science Lab 182 Dec 31, 2022
Module for statistical learning, with a particular emphasis on time-dependent modelling

Operating system Build Status Linux/Mac Windows tick tick is a Python 3 module for statistical learning, with a particular emphasis on time-dependent

X - Data Science Initiative 410 Dec 14, 2022
ELI5 is a Python package which helps to debug machine learning classifiers and explain their predictions

A library for debugging/inspecting machine learning classifiers and explaining their predictions

154 Dec 17, 2022
Python ML pipeline that showcases mltrace functionality.

mltrace tutorial Date: October 2021 This tutorial builds a training and testing pipeline for a toy ML prediction problem: to predict whether a passeng

Log Labs 28 Nov 09, 2022
PLUR is a collection of source code datasets suitable for graph-based machine learning.

PLUR (Programming-Language Understanding and Repair) is a collection of source code datasets suitable for graph-based machine learning. We provide scripts for downloading, processing, and loading the

Google Research 76 Nov 25, 2022
Turns your machine learning code into microservices with web API, interactive GUI, and more.

Turns your machine learning code into microservices with web API, interactive GUI, and more.

Machine Learning Tooling 2.8k Jan 02, 2023
30 Days Of Machine Learning Using Pytorch

Objective of the repository is to learn and build machine learning models using Pytorch. 30DaysofML Using Pytorch

Mayur 119 Nov 24, 2022
MegFlow - Efficient ML solutions for long-tailed demands.

Efficient ML solutions for long-tailed demands.

旷视天元 MegEngine 371 Dec 21, 2022
TensorFlow implementation of an arbitrary order Factorization Machine

This is a TensorFlow implementation of an arbitrary order (=2) Factorization Machine based on paper Factorization Machines with libFM. It supports: d

Mikhail Trofimov 785 Dec 21, 2022
Penguins species predictor app is used to classify penguins species created using python's scikit-learn, fastapi, numpy and joblib packages.

Penguins Classification App Penguins species predictor app is used to classify penguins species using their island, sex, bill length (mm), bill depth

Siva Prakash 3 Apr 05, 2022
A Python implementation of the Robotics Toolbox for MATLAB

Robotics Toolbox for Python A Python implementation of the Robotics Toolbox for MATLAB® GitHub repository Documentation Wiki (examples and details) Sy

Peter Corke 1.2k Jan 07, 2023
A fast, distributed, high performance gradient boosting (GBT, GBDT, GBRT, GBM or MART) framework based on decision tree algorithms, used for ranking, classification and many other machine learning tasks.

Light Gradient Boosting Machine LightGBM is a gradient boosting framework that uses tree based learning algorithms. It is designed to be distributed a

Microsoft 14.5k Jan 07, 2023
Official code for HH-VAEM

HH-VAEM This repository contains the official Pytorch implementation of the Hierarchical Hamiltonian VAE for Mixed-type Data (HH-VAEM) model and the s

Ignacio Peis 8 Nov 30, 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
moDel Agnostic Language for Exploration and eXplanation

moDel Agnostic Language for Exploration and eXplanation Overview Unverified black box model is the path to the failure. Opaqueness leads to distrust.

Model Oriented 1.2k Jan 04, 2023
Factorization machines in python

Factorization Machines in Python This is a python implementation of Factorization Machines [1]. This uses stochastic gradient descent with adaptive re

Corey Lynch 892 Jan 03, 2023
A machine learning project that predicts the price of used cars in the UK

Car Price Prediction Image Credit: AA Cars Project Overview Scraped 3000 used cars data from AA Cars website using Python and BeautifulSoup. Cleaned t

Victor Umunna 7 Oct 13, 2022
Required for a machine learning pipeline data preprocessing and variable engineering script needs to be prepared

Feature-Engineering Required for a machine learning pipeline data preprocessing and variable engineering script needs to be prepared. When the dataset

kemalgunay 5 Apr 21, 2022
LiuAlgoTrader is a scalable, multi-process ML-ready framework for effective algorithmic trading

LiuAlgoTrader is a scalable, multi-process ML-ready framework for effective algorithmic trading. The framework simplify development, testing, deployment, analysis and training algo trading strategies

Amichay Oren 458 Dec 24, 2022
The code from the Machine Learning Bookcamp book and a free course based on the book

The code from the Machine Learning Bookcamp book and a free course based on the book

Alexey Grigorev 5.5k Jan 09, 2023