]> Cypherpunks repositories - gostls13.git/commit
crypto/internal/fips140/bigmod: disable race detector on tight loops
authorRuss Cox <rsc@golang.org>
Mon, 2 Dec 2024 22:13:49 +0000 (17:13 -0500)
committerRuss Cox <rsc@golang.org>
Tue, 3 Dec 2024 18:15:37 +0000 (18:15 +0000)
commitbdedc5c1d7facb2e6c803e0a319734e60025cbf5
tree717e64d832df91c38fd2df6ed844a35ceb854fd4
parent9d76157e2dc91dff82ed7c8bd2598be6e418c175
crypto/internal/fips140/bigmod: disable race detector on tight loops

These kinds of loops over all words of a Nat
were in assembly in math/big, so the race detector
did not instrument them. Now that they are in Go,
they dramatically slow down crypto code under the
race detector. Disable instrumenting them, just like
if they were still in assembly.

goos: darwin
goarch: arm64
cpu: Apple M3 Pro
race: on

pkg: crypto/internal/fips140/bigmod
                  │     old      │                 new                 │
                  │    sec/op    │   sec/op     vs base                │
ModAdd-12           1191.0n ± 1%   164.7n ± 2%  -86.18% (p=0.000 n=30)
ModSub-12           1179.5n ± 5%   145.6n ± 2%  -87.66% (p=0.000 n=30)
MontgomeryRepr-12    2.022µ ± 0%   1.300µ ± 1%  -35.71% (p=0.000 n=30)
MontgomeryMul-12     2.111µ ± 1%   1.302µ ± 0%  -38.34% (p=0.000 n=30)
ModMul-12            4.160µ ± 1%   2.868µ ± 1%  -31.06% (p=0.000 n=30)
ExpBig-12            2.780m ± 0%   2.767m ± 0%   -0.49% (p=0.000 n=30)
Exp-12               8.751m ± 4%   3.451m ± 1%  -60.56% (p=0.000 n=30)
geomean              17.98µ        7.306µ       -59.36%

pkg: crypto/rsa
                             │      old      │                 new                  │
                             │    sec/op     │    sec/op     vs base                │
DecryptPKCS1v15/2048-12         5.537m ±  4%   1.754m ±  1%  -68.32% (p=0.000 n=30)
DecryptPKCS1v15/3072-12        12.153m ±  1%   3.827m ±  1%  -68.51% (p=0.000 n=30)
DecryptPKCS1v15/4096-12        21.889m ±  1%   7.251m ±  1%  -66.87% (p=0.000 n=30)
EncryptPKCS1v15/2048-12         87.98µ ±  0%   45.47µ ±  1%  -48.33% (p=0.000 n=30)
DecryptOAEP/2048-12             5.583m ±  5%   1.777m ±  2%  -68.17% (p=0.000 n=30)
EncryptOAEP/2048-12             98.99µ ±  0%   54.57µ ±  0%  -44.87% (p=0.000 n=30)
SignPKCS1v15/2048-12            5.542m ±  3%   1.756m ±  2%  -68.32% (p=0.000 n=30)
VerifyPKCS1v15/2048-12          84.22µ ±  1%   44.80µ ±  1%  -46.81% (p=0.000 n=30)
SignPSS/2048-12                 5.595m ±  5%   1.799m ±  2%  -67.84% (p=0.000 n=30)
VerifyPSS/2048-12               92.06µ ±  1%   52.72µ ±  1%  -42.73% (p=0.000 n=30)
GenerateKey/2048-12            2744.0m ± 38%   304.1m ± 11%  -88.92% (p=0.000 n=30)
ParsePKCS8PrivateKey/2048-12    888.0µ ±  1%   101.8µ ±  0%  -88.54% (p=0.000 n=30)
geomean                         2.428m         778.5µ        -67.93%

goos: linux
goarch: amd64
cpu: AMD Ryzen 9 7950X 16-Core Processor
race: on

pkg: crypto/internal/fips140/bigmod
                  │     old     │                 new                 │
                  │   sec/op    │   sec/op     vs base                │
ModAdd-32           900.7n ± 3%   124.2n ± 2%  -86.21% (p=0.000 n=30)
ModSub-32           895.5n ± 8%   117.1n ± 1%  -86.92% (p=0.000 n=30)
MontgomeryRepr-32   1.669µ ± 3%   1.038µ ± 1%  -37.82% (p=0.000 n=30)
MontgomeryMul-32    1.646µ ± 4%   1.033µ ± 1%  -37.25% (p=0.000 n=30)
ModMul-32           3.384µ ± 5%   2.124µ ± 1%  -37.22% (p=0.000 n=30)
ExpBig-32           2.573m ± 2%   2.561m ± 1%        ~ (p=0.146 n=30)
Exp-32              7.188m ± 2%   2.746m ± 1%  -61.80% (p=0.000 n=30)
geomean             14.56µ        5.844µ       -59.86%

pkg: crypto/rsa
                             │      old      │                 new                  │
                             │    sec/op     │    sec/op     vs base                │
DecryptPKCS1v15/2048-32         4.111m ±  1%   1.233m ±  1%  -70.00% (p=0.000 n=30)
DecryptPKCS1v15/3072-32         9.443m ±  1%   2.955m ±  1%  -68.70% (p=0.000 n=30)
DecryptPKCS1v15/4096-32        17.261m ±  1%   5.744m ±  1%  -66.72% (p=0.000 n=30)
EncryptPKCS1v15/2048-32         66.31µ ±  1%   36.16µ ±  1%  -45.46% (p=0.000 n=30)
DecryptOAEP/2048-32             4.107m ±  2%   1.241m ±  1%  -69.80% (p=0.000 n=30)
EncryptOAEP/2048-32             72.19µ ±  0%   42.35µ ±  1%  -41.34% (p=0.000 n=30)
SignPKCS1v15/2048-32            4.145m ±  1%   1.257m ±  2%  -69.68% (p=0.000 n=30)
VerifyPKCS1v15/2048-32          65.14µ ±  1%   34.95µ ±  1%  -46.36% (p=0.000 n=30)
SignPSS/2048-32                 4.005m ±  2%   1.271m ±  1%  -68.26% (p=0.000 n=30)
VerifyPSS/2048-32               70.76µ ±  0%   40.72µ ±  1%  -42.46% (p=0.000 n=30)
GenerateKey/2048-32            1946.4m ± 19%   236.3m ± 18%  -87.86% (p=0.000 n=30)
ParsePKCS8PrivateKey/2048-32   713.94µ ±  1%   85.89µ ±  1%  -87.97% (p=0.000 n=30)
geomean                         1.829m         591.5µ        -67.66%

Change-Id: I49cbb6d14b187100bf5e6002e30096667689d852
Reviewed-on: https://go-review.googlesource.com/c/go/+/632978
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Roland Shoemaker <roland@golang.org>
src/crypto/internal/fips140/bigmod/nat.go