Developers: Ang Li, Chenhao Xie, Jieyang Chen, Shuaiwen Song, Linghao Song, Jiajia Li, Jesun Firoz
General Area or Topic of Investigation: Implementing and optimizing sparse Basic Linear Algebra Subprograms (BLAS) on modern multi-GPU systems.
Release Number: 0.1
License: MIT License, PNNL-IPID: 31803-E, IR: PNNL-31803
The following sections detail the compilation, packaging, and installation of the software. Also included are test data and scripts to verify the installation was successful.
Programming Language: CUDA C/C++
Operating System & Version: Ubuntu 16.04
Required Disk Space: 2.5MB (additional space is required for storing test input matrix files).
Required Memory: Varies with different tests.
Nodes / Cores Used: One node with one or more Nvidia GPUs. Using NSHMEM (sptrsv_v3) requires the GPUs are directly P2P connected by NVLink/NVSwitch/NV-SLI.
| Name | Version | Download Location | Country of Origin | Special Instructions |
|---|---|---|---|---|
| GCC | 5.4.0 | https://gcc.gnu.org/ | USA | None |
| CUDA | 10.0 or newer | https://developer.nvidia.com/cuda-toolkit | USA | None |
| OpenMP | 3.1 or newer | https://www.openmp.org/ | USA | None |
| NCCL | 2.0 or newer | https://github.com/NVIDIA/nccl | USA | None |
| File | Description |
|---|---|
| config.h | Configure the settings |
| kernel.h | Utility kernels functions |
| matrix.h | CSR, CSC, COO-format sparse matrix, dense matrix and dense vector definition |
| mmio.h | Matrix markert I/O library for ANSI C |
| mmio_highlevel.h | MMIO high-level API |
| sblas.h | S-BLAS unified header file |
| spmm.h | Sparse-matrix-dense-matrix multiplication (SPMM) implementation |
| spmm_test.h | Testing file for SPMM |
| spmv.h | Sparse-matrix-dense-vector multiplication (SPMV) implementation |
| unit_test.cu | Unit test file for matrix.h, spmv.h and spmm.h |
| ash85.mtx | An example matrix-market (.mtx) sparse matrix |
| Makefile | Makefile for s-blas, update it for target platforms |
| LICENSE | MIT License file |
| NOTICE | A notice file for release |
| README.md | This file |
(1) Modify Makefile, update the path variables for CUDA_PATH and CUDA_SAMPLE_PATH. Update GPU compute capability. Add NCCL and cuSparse path if necessary.
(2) Type make in this directory to compile the library.
(3) Update unit_test.cu if performing unit test.
We have two implmentations for SPMM: 1-Partition the dense matrix B, 2-Partition the sparse matrix A. Select the method 1 or 2.
$ ./unit_test
$ ./spmm 1 ./ash85.mtx 64 1.0 1.0 4We verify the running results using the CPU implmentation. Please see the code for details.
This project is licensed under the MIT License, see LICENSE file for details.
This work was supported by the S-BLAS project under PNNL's High-Performance-Data-Analytics (HPDA) program. The Pacific Northwest National Laboratory (PNNL) is operated by Battelle for the U.S. Department of Energy (DOE) under contract DE-AC05-76RL01830.