sha3: avoid buffer copy

Previously, the package worked by copying the input (or the output) into
a buffer, and then XOR'ing (or copying) it into (or out of) the state.
(Except for an input fast path.) There's no need for that! We can XOR
straight into the state, and copy straight out of it, at least on little
endian machines. This is a bit faster, almost halves the state size, and
will make it easier to implement marshaling, but most importantly look
at how much simpler it makes the code!

go: go1.23.0
goos: linux
goarch: amd64
pkg: golang.org/x/crypto/sha3
cpu: AMD Ryzen 7 PRO 8700GE w/ Radeon 780M Graphics
                      │ v0.27.0-2-g42ee18b9637 │    v0.27.0-2-g42ee18b9637-dirty    │
                      │         sec/op         │   sec/op     vs base               │
PermutationFunction-8              270.8n ± 0%   270.4n ± 0%       ~ (p=0.099 n=10)
Sha3_512_MTU-8                     5.762µ ± 0%   5.658µ ± 0%  -1.80% (p=0.000 n=10)
Sha3_384_MTU-8                     4.179µ ± 0%   4.070µ ± 0%  -2.60% (p=0.000 n=10)
Sha3_256_MTU-8                     3.316µ ± 0%   3.214µ ± 0%  -3.08% (p=0.000 n=10)
Sha3_224_MTU-8                     3.175µ ± 0%   3.061µ ± 0%  -3.61% (p=0.000 n=10)
Shake128_MTU-8                     2.779µ ± 0%   2.681µ ± 0%  -3.51% (p=0.000 n=10)
Shake256_MTU-8                     2.947µ ± 0%   2.957µ ± 0%  +0.32% (p=0.000 n=10)
Shake256_16x-8                     44.15µ ± 0%   44.45µ ± 0%  +0.67% (p=0.000 n=10)
Shake256_1MiB-8                    2.319m ± 0%   2.274m ± 0%  -1.93% (p=0.000 n=10)
Sha3_512_1MiB-8                    4.204m ± 0%   4.219m ± 0%  +0.34% (p=0.000 n=10)
geomean                            13.75µ        13.54µ       -1.55%

                      │ v0.27.0-2-g42ee18b9637 │    v0.27.0-2-g42ee18b9637-dirty     │
                      │          B/s           │     B/s       vs base               │
PermutationFunction-8             704.3Mi ± 0%   705.4Mi ± 0%       ~ (p=0.105 n=10)
Sha3_512_MTU-8                    223.5Mi ± 0%   227.6Mi ± 0%  +1.83% (p=0.000 n=10)
Sha3_384_MTU-8                    308.1Mi ± 0%   316.4Mi ± 0%  +2.67% (p=0.000 n=10)
Sha3_256_MTU-8                    388.2Mi ± 0%   400.5Mi ± 0%  +3.17% (p=0.000 n=10)
Sha3_224_MTU-8                    405.5Mi ± 0%   420.7Mi ± 0%  +3.73% (p=0.000 n=10)
Shake128_MTU-8                    463.4Mi ± 0%   480.2Mi ± 0%  +3.64% (p=0.000 n=10)
Shake256_MTU-8                    436.9Mi ± 0%   435.5Mi ± 0%  -0.32% (p=0.000 n=10)
Shake256_16x-8                    353.9Mi ± 0%   351.5Mi ± 0%  -0.66% (p=0.000 n=10)
Shake256_1MiB-8                   431.2Mi ± 0%   439.7Mi ± 0%  +1.97% (p=0.000 n=10)
Sha3_512_1MiB-8                   237.8Mi ± 0%   237.1Mi ± 0%  -0.33% (p=0.000 n=10)
geomean                           375.7Mi        381.6Mi       +1.57%

Even stronger effect when patched on top of CL 616555 (forced on).

go: go1.23.0
goos: darwin
goarch: arm64
pkg: golang.org/x/crypto/sha3
cpu: Apple M2
                      │    old    │               new                   │
                      │   sec/op  │     sec/op     vs base              │
PermutationFunction-8   154.7n ± 2%   153.8n ± 1%        ~ (p=0.469 n=10)
Sha3_512_MTU-8          3.260µ ± 2%   3.143µ ± 2%   -3.60% (p=0.000 n=10)
Sha3_384_MTU-8          2.389µ ± 2%   2.244µ ± 2%   -6.07% (p=0.000 n=10)
Sha3_256_MTU-8          1.950µ ± 2%   1.758µ ± 1%   -9.87% (p=0.000 n=10)
Sha3_224_MTU-8          1.874µ ± 2%   1.686µ ± 1%  -10.06% (p=0.000 n=10)
Shake128_MTU-8          1.827µ ± 3%   1.447µ ± 1%  -20.80% (p=0.000 n=10)
Shake256_MTU-8          1.665µ ± 3%   1.604µ ± 3%   -3.63% (p=0.003 n=10)
Shake256_16x-8          25.14µ ± 1%   25.23µ ± 2%        ~ (p=0.912 n=10)
Shake256_1MiB-8         1.236m ± 2%   1.243m ± 2%        ~ (p=0.631 n=10)
Sha3_512_1MiB-8         2.296m ± 2%   2.305m ± 1%        ~ (p=0.315 n=10)
geomean                 7.906µ        7.467µ        -5.56%

                      │    old    │               new                   │
                      │    B/op   │      B/op      vs base              │
PermutationFunction-8   1.204Gi ± 2%   1.212Gi ± 1%        ~ (p=0.529 n=10)
Sha3_512_MTU-8          394.9Mi ± 2%   409.7Mi ± 2%   +3.73% (p=0.000 n=10)
Sha3_384_MTU-8          539.0Mi ± 2%   573.8Mi ± 2%   +6.45% (p=0.000 n=10)
Sha3_256_MTU-8          660.3Mi ± 2%   732.6Mi ± 1%  +10.95% (p=0.000 n=10)
Sha3_224_MTU-8          687.1Mi ± 2%   763.9Mi ± 1%  +11.17% (p=0.000 n=10)
Shake128_MTU-8          704.7Mi ± 2%   889.6Mi ± 2%  +26.24% (p=0.000 n=10)
Shake256_MTU-8          773.4Mi ± 3%   802.5Mi ± 3%   +3.76% (p=0.004 n=10)
Shake256_16x-8          621.6Mi ± 1%   619.3Mi ± 2%        ~ (p=0.912 n=10)
Shake256_1MiB-8         809.1Mi ± 2%   804.7Mi ± 2%        ~ (p=0.631 n=10)
Sha3_512_1MiB-8         435.6Mi ± 2%   433.9Mi ± 1%        ~ (p=0.315 n=10)
geomean                 653.6Mi        692.0Mi        +5.88%

Change-Id: I33a0a1ddf305c395f99bf17f81473e2f42c5ce42
Reviewed-on: https://go-review.googlesource.com/c/crypto/+/616575
Reviewed-by: Daniel McCarney <daniel@binaryparadox.net>
Reviewed-by: Michael Pratt <mpratt@google.com>
Reviewed-by: Roland Shoemaker <roland@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Auto-Submit: Filippo Valsorda <filippo@golang.org>
Reviewed-by: Andrew Ekstedt <andrew.ekstedt@gmail.com>
2 files changed
tree: 709a60d78ceaab34f71fbf9bd9dd92679e805b37
  1. acme/
  2. argon2/
  3. bcrypt/
  4. blake2b/
  5. blake2s/
  6. blowfish/
  7. bn256/
  8. cast5/
  9. chacha20/
  10. chacha20poly1305/
  11. cryptobyte/
  12. curve25519/
  13. ed25519/
  14. hkdf/
  15. internal/
  16. md4/
  17. nacl/
  18. ocsp/
  19. openpgp/
  20. otr/
  21. pbkdf2/
  22. pkcs12/
  23. poly1305/
  24. ripemd160/
  25. salsa20/
  26. scrypt/
  27. sha3/
  28. ssh/
  29. tea/
  30. twofish/
  31. x509roots/
  32. xtea/
  33. xts/
  34. .gitattributes
  35. .gitignore
  36. codereview.cfg
  37. CONTRIBUTING.md
  38. go.mod
  39. go.sum
  40. LICENSE
  41. PATENTS
  42. README.md
README.md

Go Cryptography

Go Reference

This repository holds supplementary Go cryptography libraries.

Download/Install

The easiest way to install is to run go get -u golang.org/x/crypto/.... You can also manually git clone the repository to $GOPATH/src/golang.org/x/crypto.

Report Issues / Send Patches

This repository uses Gerrit for code changes. To learn how to submit changes to this repository, see https://golang.org/doc/contribute.html.

The main issue tracker for the crypto repository is located at https://github.com/golang/go/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.