cmd/compile: use generated loops instead of DUFFCOPY on amd64

goarch: amd64
cpu: 12th Gen Intel(R) Core(TM) i7-12700
                        │     base      │                 exp                 │
                        │    sec/op     │   sec/op     vs base                │
MemmoveKnownSize112-20     1.764n ±  0%   1.247n ± 0%  -29.31% (p=0.000 n=10)
MemmoveKnownSize128-20     1.891n ±  0%   1.405n ± 1%  -25.72% (p=0.000 n=10)
MemmoveKnownSize192-20     2.521n ±  0%   2.114n ± 3%  -16.16% (p=0.000 n=10)
MemmoveKnownSize248-20     4.028n ±  0%   3.877n ± 1%   -3.75% (p=0.000 n=10)
MemmoveKnownSize256-20     3.272n ±  0%   2.961n ± 2%   -9.53% (p=0.000 n=10)
MemmoveKnownSize512-20     6.733n ±  3%   5.936n ± 4%  -11.83% (p=0.000 n=10)
MemmoveKnownSize1024-20   13.905n ±  5%   9.798n ± 9%  -29.54% (p=0.000 n=10)

Change-Id: Icc01cec0d8b072300d749a5ce76f53b3725b5c65
Reviewed-on: https://go-review.googlesource.com/c/go/+/678620
Reviewed-by: Jorropo <jorropo.pgm@gmail.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
Reviewed-by: Jakub Ciolek <jakub@ciolek.dev>
This commit is contained in:
Keith Randall 2025-06-04 10:39:29 -07:00
parent d0a64f7969
commit ec9e1176c3
8 changed files with 258 additions and 226 deletions

View file

@ -31,6 +31,7 @@ const (
removeDeadValues = true
repZeroThreshold = 1408 // size beyond which we use REP STOS for zeroing
repMoveThreshold = 1408 // size beyond which we use REP MOVS for copying
)
// deadcode indicates whether rewrite should try to remove any values that become dead.