crypto/aes,crypto/cipher: improve gcm performance on ppc64x

This improves performance for AES-GCM. The function
counterCrypt is written in assembler so the loop can
be unrolled and the stitched approach used for the
encryption.

This implementation works on ppc64le and ppc64. The use
of GOPPC64=power9 generates the best performance,
goos: linux
goarch: ppc64le
pkg: crypto/cipher
cpu: POWER10
                     │ gcmx8.cpu1.out │         gcmx8.new.cpu1.out         │
                     │     sec/op     │   sec/op     vs base               │
AESGCM/Open-128-64        180.5n ± 0%   152.7n ± 1%  -15.43% (p=0.000 n=8)
AESGCM/Seal-128-64        166.8n ± 0%   147.2n ± 0%  -11.72% (p=0.000 n=8)
AESGCM/Open-256-64        194.9n ± 0%   156.6n ± 1%  -19.65% (p=0.000 n=8)
AESGCM/Seal-256-64        183.7n ± 0%   157.0n ± 0%  -14.51% (p=0.000 n=8)
AESGCM/Open-128-1350     1769.5n ± 0%   454.5n ± 0%  -74.31% (p=0.000 n=8)
AESGCM/Seal-128-1350     1759.0n ± 0%   453.7n ± 0%  -74.21% (p=0.000 n=8)
AESGCM/Open-256-1350     2104.0n ± 0%   504.4n ± 1%  -76.03% (p=0.000 n=8)
AESGCM/Seal-256-1350     2092.0n ± 0%   503.0n ± 0%  -75.96% (p=0.000 n=8)
AESGCM/Open-128-8192     10.054µ ± 0%   1.961µ ± 0%  -80.50% (p=0.000 n=8)
AESGCM/Seal-128-8192     10.050µ ± 0%   1.965µ ± 0%  -80.45% (p=0.000 n=8)
AESGCM/Open-256-8192     12.080µ ± 0%   2.232µ ± 0%  -81.52% (p=0.000 n=8)
AESGCM/Seal-256-8192     12.069µ ± 0%   2.238µ ± 0%  -81.46% (p=0.000 n=8)
geomean                   1.566µ        535.5n       -65.80%

                     │ gcmx8.cpu1.out │          gcmx8.new.cpu1.out           │
                     │      B/s       │      B/s       vs base                │
AESGCM/Open-128-64       338.1Mi ± 0%    399.8Mi ± 1%   +18.27% (p=0.000 n=8)
AESGCM/Seal-128-64       366.1Mi ± 0%    414.6Mi ± 0%   +13.28% (p=0.000 n=8)
AESGCM/Open-256-64       313.1Mi ± 0%    389.7Mi ± 0%   +24.47% (p=0.000 n=8)
AESGCM/Seal-256-64       332.3Mi ± 0%    388.5Mi ± 0%   +16.93% (p=0.000 n=8)
AESGCM/Open-128-1350     727.6Mi ± 0%   2832.8Mi ± 0%  +289.33% (p=0.000 n=8)
AESGCM/Seal-128-1350     732.0Mi ± 0%   2837.8Mi ± 0%  +287.70% (p=0.000 n=8)
AESGCM/Open-256-1350     611.9Mi ± 0%   2552.6Mi ± 0%  +317.18% (p=0.000 n=8)
AESGCM/Seal-256-1350     615.3Mi ± 0%   2559.6Mi ± 0%  +315.97% (p=0.000 n=8)
AESGCM/Open-128-8192     777.1Mi ± 0%   3983.5Mi ± 0%  +412.63% (p=0.000 n=8)
AESGCM/Seal-128-8192     777.3Mi ± 0%   3975.9Mi ± 0%  +411.47% (p=0.000 n=8)
AESGCM/Open-256-8192     646.7Mi ± 0%   3500.6Mi ± 0%  +441.27% (p=0.000 n=8)
AESGCM/Seal-256-8192     647.3Mi ± 0%   3491.1Mi ± 0%  +439.30% (p=0.000 n=8)
geomean                  542.7Mi         1.550Gi       +192.42%

Change-Id: I3600831a263ec8a99b5e3bdd495eb36e966d8075
Reviewed-on: https://go-review.googlesource.com/c/go/+/484575
Reviewed-by: Roland Shoemaker <roland@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Paul Murphy <murp@ibm.com>
Reviewed-by: Than McIntosh <thanm@google.com>

0 files changed, 0 insertions, 0 deletions