chacha20: add loong64 SIMD implementation
The performance of chacha20 has been greatly improved on 3A6000 and 3A5000.
goos: linux
goarch: loong64
pkg: golang.org/x/crypto/chacha20
cpu: Loongson-3A6000 @ 2500.00MHz
| bench.old | bench.new |
| sec/op | sec/op vs base |
ChaCha20/64 171.9n ± 0% 159.3n ± 0% -7.33% (p=0.000 n=20)
ChaCha20/256 592.2n ± 0% 142.8n ± 0% -75.89% (p=0.000 n=20)
ChaCha20/10x25 981.5n ± 0% 518.8n ± 0% -47.14% (p=0.000 n=20)
ChaCha20/4096 8.991µ ± 0% 1.732µ ± 0% -80.74% (p=0.000 n=20)
ChaCha20/100x40 10.651µ ± 0% 5.135µ ± 0% -51.79% (p=0.000 n=20)
ChaCha20/65536 143.43µ ± 0% 28.76µ ± 0% -79.95% (p=0.000 n=20)
ChaCha20/1000x65 146.17µ ± 0% 37.13µ ± 0% -74.60% (p=0.000 n=20)
geomean 5.721µ 1.962µ -65.70%
| bench.old | bench.new |
| B/s | B/s vs base |
ChaCha20/64 355.1Mi ± 0% 383.1Mi ± 0% +7.89% (p=0.000 n=20)
ChaCha20/256 412.2Mi ± 0% 1710.2Mi ± 0% +314.86% (p=0.000 n=20)
ChaCha20/10x25 242.9Mi ± 0% 459.6Mi ± 0% +89.19% (p=0.000 n=20)
ChaCha20/4096 434.5Mi ± 0% 2255.8Mi ± 0% +419.22% (p=0.000 n=20)
ChaCha20/100x40 358.1Mi ± 0% 742.9Mi ± 0% +107.44% (p=0.000 n=20)
ChaCha20/65536 435.8Mi ± 0% 2173.2Mi ± 0% +398.72% (p=0.000 n=20)
ChaCha20/1000x65 424.1Mi ± 0% 1669.4Mi ± 0% +293.64% (p=0.000 n=20)
geomean 373.9Mi 1.065Gi +191.55%
goos: linux
goarch: loong64
pkg: golang.org/x/crypto/chacha20
cpu: Loongson-3A5000 @ 2500.00MHz
| bench.old | bench.new |
| sec/op | sec/op vs base |
ChaCha20/64 234.5n ± 0% 295.8n ± 0% +26.14% (p=0.000 n=20)
ChaCha20/256 782.0n ± 0% 274.6n ± 0% -64.88% (p=0.000 n=20)
ChaCha20/10x25 1340.0n ± 0% 752.7n ± 0% -43.83% (p=0.000 n=20)
ChaCha20/4096 11.744µ ± 0% 3.455µ ± 0% -70.58% (p=0.000 n=20)
ChaCha20/100x40 14.151µ ± 0% 7.435µ ± 0% -47.46% (p=0.000 n=20)
ChaCha20/65536 188.05µ ± 0% 54.33µ ± 0% -71.11% (p=0.000 n=20)
ChaCha20/1000x65 191.44µ ± 0% 66.29µ ± 0% -65.37% (p=0.000 n=20)
geomean 7.604µ 3.436µ -54.81%
| bench.old | bench.new |
| B/s | B/s vs base |
ChaCha20/64 260.3Mi ± 0% 206.3Mi ± 0% -20.73% (p=0.000 n=20)
ChaCha20/256 312.2Mi ± 0% 888.9Mi ± 0% +184.75% (p=0.000 n=20)
ChaCha20/10x25 177.9Mi ± 0% 316.8Mi ± 0% +78.08% (p=0.000 n=20)
ChaCha20/4096 332.6Mi ± 0% 1130.8Mi ± 0% +239.95% (p=0.000 n=20)
ChaCha20/100x40 269.6Mi ± 0% 513.1Mi ± 0% +90.34% (p=0.000 n=20)
ChaCha20/65536 332.4Mi ± 0% 1150.5Mi ± 0% +246.16% (p=0.000 n=20)
ChaCha20/1000x65 323.8Mi ± 0% 935.2Mi ± 0% +188.81% (p=0.000 n=20)
geomean 281.3Mi 622.6Mi +121.31%
Change-Id: I5386f2029122076c1d22a04610567e3df23877cd
Reviewed-on: https://go-review.googlesource.com/c/crypto/+/636257
Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: abner chenc <chenguoqi@loongson.cn>
Reviewed-by: Carlos Amedee <carlos@golang.org>
This repository holds supplementary Go cryptography packages.
This repository uses Gerrit for code changes. To learn how to submit changes to this repository, see https://go.dev/doc/contribute.
The git repository is https://go.googlesource.com/crypto.
The main issue tracker for the crypto repository is located at https://go.dev/issues. Prefix your issue with “x/crypto:” in the subject line, so it is easy to find.
Note that contributions to the cryptography package receive additional scrutiny due to their sensitive nature. Patches may take longer than normal to receive feedback.