https://github.com/danieldk/gemm-benchmark
Simple [sd]gemm benchmark, similar to ACES dgemm
https://github.com/danieldk/gemm-benchmark
Last synced: about 1 month ago
JSON representation
Simple [sd]gemm benchmark, similar to ACES dgemm
- Host: GitHub
- URL: https://github.com/danieldk/gemm-benchmark
- Owner: danieldk
- License: apache-2.0
- Created: 2021-07-06T17:58:51.000Z (almost 4 years ago)
- Default Branch: main
- Last Pushed: 2024-05-19T11:51:39.000Z (about 1 year ago)
- Last Synced: 2024-11-28T20:12:00.324Z (6 months ago)
- Language: Rust
- Size: 52.7 KB
- Stars: 9
- Watchers: 2
- Forks: 3
- Open Issues: 2
-
Metadata Files:
- Readme: README.md
- License: LICENSE-APACHE
Awesome Lists containing this project
README
# `[sd]gemm` benchmark
## Introduction
This is a small `[sd]gemm` benchmark based, similar to
[ACES DGEMM](https://www.lanl.gov/projects/crossroads/benchmarks-performance-analysis.php),
implemented in Rust. It supports the following BLAS libraries:
- Accelerate (macOS)
- Intel MKL
- OpenBLAS
## Building
### Build with Accelerate (macOS)
```
$ cargo install gemm-benchmark --features accelerate
```
### Build with BLIS
```
$ cargo install gemm-benchmark --features blis
```
### Build with Intel MKL
To build the benchmark with Intel MKL statically linked, use:
```
$ cargo install gemm-benchmark --features intel-mkl
```
Intel MKL uses Zen-specific `[sd]gemm`kernels on AMD Zen CPUs.
However, these kernels are slower on many Zen CPUs than the AVX2
kernels. You can build the benchmark to override Intel CPU
detection, so that MKL uses AVX2 kernels on Zen CPUs as well. This
does require dynamic linking, since it is not permitted to modify
MKL binaries. To enable this override, use the `intel-mkl-amd`
feature:
```
$ cargo install gemm-benchmark --features intel-mkl-amd
```
### Build with OpenBLAS
```shell
$ cargo install gemm-benchmark --features openblas
```
Set `OPENBLAS_NUM_THREADS=1` before running.
## Benchmarking
By default, `sgemm` is benchmarked using _256 x 256_ matrices, for
_1,000_ iterations and _1_ thread. The dimensionality (`-d`), number
of iterations (`-i`), and the number of threads (`-t`) can be set
with command-line flags. For example:
```shell
$ gemm-benchmark -d 1024 -i 2000 -t 4
```
Runs the benchmark using _1024 x 1024_ matrices, for _1,000_ iterations,
and _4_ threads. It is also possible to benchmark `dgem,` using the
`--dgemm` option:
```shell
$ gemm-benchmark -d 1024 -i 2000 -t 4 --dgemm
```
## Example results
The following table shows GFLOPS for various CPUs using 1 to 16 threads on
matrix size 768, tested for 1000 iterations (`gemm-benchmark -d 768 -t NTHREADS`).
| Threads | M1 Accelerate | M1 Pro Accelerate | M1 Ultra Accelerate | M2 Accelerate | i7-13700K |
| ------- | ------------- | ----------------- | ------------------- | ------------- | --------- |
| 1 | 1340 | 2061 | 2177 | 1475 | 165 |
| 2 | 1226 | 2583 | 3427 | 1639 | 323 |
| 4 | 1102 | 2685 | 3788 | 1730 | 646 |
| 8 | 1253 | 2381 | 4344 | 1601 | 1279 |
| 12 | 1225 | 2248 | 4261 | 1456 | 1148 |
| 16 | 1217 | 2254 | 4376 | 1388 | 1524 |