Курнат Андрей
43 lines
1.8 KiB
Markdown
43 lines
1.8 KiB
Markdown
# 2025 revival
|
|
|
|
For IEEE checksums AVX512 can be used to speed up CRC32 checksums by approximately 2x.
|
|
|
|
Castagnoli checksums (CRC32C) can also be computer with AVX512,
|
|
but the performance gain is not as significant enough for the downsides of using it at this point.
|
|
|
|
# crc32
|
|
|
|
This package is a drop-in replacement for the standard library `hash/crc32` package,
|
|
that features AVX 512 optimizations on x64 platforms, for a 2x speedup for IEEE CRC32 checksums.
|
|
|
|
# usage
|
|
|
|
Install using `go get github.com/klauspost/crc32`. This library is based on Go 1.24
|
|
|
|
Replace `import "hash/crc32"` with `import "github.com/klauspost/crc32"` and you are good to go.
|
|
|
|
# changes
|
|
* 2025: Revived and updated to Go 1.24, with AVX 512 optimizations.
|
|
|
|
# performance
|
|
|
|
AVX512 are enabled above 1KB input size. This rather high limit is due to AVX512 may be slower to ramp up than
|
|
the regular SSE4 implementation for smaller inputs. This is not reflected in the benchmarks below.
|
|
|
|
| Benchmark | Old MB/s | New MB/s | Speedup |
|
|
|-----------------------------------------------|----------|----------|---------|
|
|
| BenchmarkCRC32/poly=IEEE/size=512/align=0-32 | 17996.39 | 17969.94 | 1.00x |
|
|
| BenchmarkCRC32/poly=IEEE/size=512/align=1-32 | 18021.48 | 17945.55 | 1.00x |
|
|
| BenchmarkCRC32/poly=IEEE/size=1kB/align=0-32 | 19921.70 | 45613.77 | 2.29x |
|
|
| BenchmarkCRC32/poly=IEEE/size=1kB/align=1-32 | 19946.60 | 46819.09 | 2.35x |
|
|
| BenchmarkCRC32/poly=IEEE/size=4kB/align=0-32 | 21538.65 | 48600.93 | 2.26x |
|
|
| BenchmarkCRC32/poly=IEEE/size=4kB/align=1-32 | 21449.20 | 48477.84 | 2.26x |
|
|
| BenchmarkCRC32/poly=IEEE/size=32kB/align=0-32 | 21785.49 | 46013.10 | 2.11x |
|
|
| BenchmarkCRC32/poly=IEEE/size=32kB/align=1-32 | 21946.47 | 45954.10 | 2.09x |
|
|
|
|
cpu: AMD Ryzen 9 9950X 16-Core Processor
|
|
|
|
# license
|
|
|
|
Standard Go license. See [LICENSE](LICENSE) for details.
|