Files
crypto/chacha20
Jayanth Krishnamurthy 8d0d405eed x/crypto/chacha20: cleanup chacha_ppc64le.s
- Adding PCALIGN before the loops
- Changing WORD directive with corresponding Vector Merge EVEN/ODD word  instructions
- Replacing Branch Conditional (BC) with its extended mnemonic form  BDNZ
- VPERMXOR instruction usage in place of VXOR instructions followed by
  VRLW (rotate left) for cases of rotating in multiples of 8. This
  replacements give performace improvement both in time and space  of around 7%-8% as listed below
  using benchstat tool.

goos: linux
goarch: ppc64le
pkg: golang.org/x/crypto/chacha20
cpu: POWER10
                 | chacha20.prev.out |       chacha20.new.out            |
                 |      sec/op       |   sec/op     vs base              |
ChaCha20/64              171.9n ± 0%   156.6n ± 1%  -8.90% (p=0.002 n=6)
ChaCha20/256             165.5n ± 0%   152.4n ± 0%  -7.92% (p=0.002 n=6)
ChaCha20/10x25           505.8n ± 0%   504.3n ± 2%  -0.32% (p=0.589 n=6)
ChaCha20/4096            2.265µ ± 0%   2.052µ ± 0%  -9.40% (p=0.002 n=6)
ChaCha20/100x40          5.359µ ± 3%   5.018µ ± 2%  -6.37% (p=0.002 n=6)
ChaCha20/65536           35.71µ ± 0%   32.29µ ± 0%  -9.57% (p=0.002 n=6)
ChaCha20/1000x65         44.63µ ± 0%   41.05µ ± 0%  -8.02% (p=0.002 n=6)
geomean                  2.235µ        2.073µ       -7.26%

                 | chacha20.prev.out |          chacha20.new.out         |
                 |       B/s         |     B/s       vs base             |
ChaCha20/64             355.1Mi ± 0%   389.8Mi ± 1%   +9.78% (p=0.002 n=6)
ChaCha20/256            1.440Gi ± 0%   1.565Gi ± 0%   +8.62% (p=0.002 n=6)
ChaCha20/10x25          471.3Mi ± 0%   472.8Mi ± 2%   +0.31% (p=0.589 n=6)
ChaCha20/4096           1.684Gi ± 0%   1.859Gi ± 0%  +10.38% (p=0.002 n=6)
ChaCha20/100x40         711.8Mi ± 3%   760.3Mi ± 2%   +6.80% (p=0.002 n=6)
ChaCha20/65536          1.709Gi ± 0%   1.890Gi ± 0%  +10.59% (p=0.002 n=6)
ChaCha20/1000x65        1.356Gi ± 0%   1.475Gi ± 0%   +8.72% (p=0.002 n=6)
geomean                 957.3Mi        1.008Gi        +7.83%
Change-Id: Ib31cb10a2a11eacdacf0272fbfd887eb5ccd8bcb
Reviewed-on: https://go-review.googlesource.com/c/crypto/+/564797
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
Run-TryBot: Paul Murphy <murp@ibm.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
2024-03-26 21:57:38 +00:00
..
2019-11-11 21:39:47 +00:00