crypto/subtle: use PCALIGN in xorBytes

goos: linux
goarch: amd64
pkg: crypto/subtle
cpu: Intel(R) Core(TM) i5-8350U CPU @ 1.70GHz
                      │   master    │                HEAD                 │
                      │   sec/op    │   sec/op     vs base                │
XORBytes/8Bytes-8       10.90n ± 1%   10.96n ± 5%        ~ (p=0.617 n=10)
XORBytes/128Bytes-8     14.85n ± 2%   12.05n ± 2%  -18.82% (p=0.000 n=10)
XORBytes/2048Bytes-8    88.30n ± 2%   72.64n ± 1%  -17.73% (p=0.000 n=10)
XORBytes/32768Bytes-8   1.489µ ± 2%   1.442µ ± 1%   -3.12% (p=0.000 n=10)
geomean                 67.91n        60.99n       -10.19%

                      │    master    │                 HEAD                 │
                      │     B/s      │     B/s       vs base                │
XORBytes/8Bytes-8       700.5Mi ± 1%   696.5Mi ± 5%        ~ (p=0.631 n=10)
XORBytes/128Bytes-8     8.026Gi ± 2%   9.890Gi ± 2%  +23.22% (p=0.000 n=10)
XORBytes/2048Bytes-8    21.60Gi ± 2%   26.26Gi ± 1%  +21.55% (p=0.000 n=10)
XORBytes/32768Bytes-8   20.50Gi ± 2%   21.16Gi ± 1%   +3.21% (p=0.000 n=10)
geomean                 7.022Gi        7.819Gi       +11.34%

For #63678

Change-Id: I3996873773748a6f78acc6575e70e09bb6aea979
GitHub-Last-Rev: d9129cb8eae363792b6a6236d597465fcf8d4d6b
GitHub-Pull-Request: golang/go#63754
Reviewed-on: https://go-review.googlesource.com/c/go/+/537856
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Keith Randall <khr@google.com>
Auto-Submit: Keith Randall <khr@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

diff --git a/src/crypto/subtle/xor_amd64.s b/src/crypto/subtle/xor_amd64.s
index 8b04b5870291772fea92dfe3fd47c43def2b686f..949424f87ae8ae3d4f14f8ef66ec49f2c839ba75 100644 (file)
--- a/src/crypto/subtle/xor_amd64.s
+++ b/src/crypto/subtle/xor_amd64.s
@@ -18,6 +18,7 @@ TEXT ·xorBytes(SB), NOSPLIT, $0
 aligned:
        MOVQ $0, AX // position in slices
 
+       PCALIGN $16
 loop16b:
        MOVOU (SI)(AX*1), X0   // XOR 16byte forwards.
        MOVOU (CX)(AX*1), X1
@@ -28,6 +29,7 @@ loop16b:
        JNE   loop16b
        RET
 
+       PCALIGN $16
 loop_1b:
        SUBQ  $1, DX           // XOR 1byte backwards.
        MOVB  (SI)(DX*1), DI