https://github.com/thilinarmtb/lsbench
Repository for benchmarking linear solvers on GPU.
https://github.com/thilinarmtb/lsbench
amg amgx cholesky cholmod cpu cusparse finite-element-methods gpu hpc hypre linear-solvers spectral-element-method
Last synced: 4 months ago
JSON representation
Repository for benchmarking linear solvers on GPU.
- Host: GitHub
- URL: https://github.com/thilinarmtb/lsbench
- Owner: thilinarmtb
- Created: 2023-04-15T18:22:17.000Z (over 2 years ago)
- Default Branch: main
- Last Pushed: 2023-05-12T01:03:14.000Z (over 2 years ago)
- Last Synced: 2025-04-07T12:48:17.479Z (7 months ago)
- Topics: amg, amgx, cholesky, cholmod, cpu, cusparse, finite-element-methods, gpu, hpc, hypre, linear-solvers, spectral-element-method
- Language: C
- Homepage:
- Size: 9.56 MB
- Stars: 2
- Watchers: 1
- Forks: 2
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
## Introduction
`lsbench` is a repository containing benchmark codes for various high performant
linear solvers on CPUs and GPUs. It can read a sparse matrix and solve it using
`cuSparse`, `AmgX`, `Hypre` or `CHOLMOD`. An example on how to use the API is in
`bin/driver.c`. A few test matrices are available in `tests/` directory. Currently,
all the benchmarks are performend in double precision.
## Building lsbench
`lsbench` by defaults download and builds `CHOLMOD`. You can enable/disable
solvers (`AmgX`, `cuSparse`, etc.) when configuring the `cmake` build using
`ENABLE_=ON|OFF`.
### Build requirements
- cmake (>= 3.18)
- CUDAToolkit (>= 11.0) if using AmgX, cuSparse or Hypre with Cuda backend)
### Build instructions
1. Clone the repo first using `git` and `cd` into the repository:
```sh
git clone https://github.com/thilinarmtb/lsbench.git
cd lsbench
```
2. Then you can use `cmake` to build the benchmarks:
```sh
mkdir build && cd build
cmake .. -DCMAKE_C_COMPILER= -DCMAKE_CXX_COMPILER= \
-DCMAKE_INSTALL_PREFIX= \
-DENABLE_CUSPARSE=ON -DENABLE_AMGX=OFF
make -j8 install
cd -
```
You can add the `bin` directory to the `PATH` variable to access the binary
without the full path:
```sh
export PATH=${PATH}:/bin
```
## Running the benchmarks
Once the benchmarks are built, they can be run using the following command:
Do `driver --help` to see available options (not everything is implemented
right now).
```sh
driver --solver cholmod --test ./tests/I1_05x05.txt --verbose=1
```