sliceop: improve performances of Take

name                old time/op    new time/op    delta
Take/Len=2-8          2.95ns ± 2%    2.96ns ± 2%     ~     (p=0.952 n=20+20)
Take/Len=4-8          3.43ns ± 1%    3.67ns ± 1%   +6.96%  (p=0.000 n=20+19)
Take/Len=8-8          3.92ns ± 2%    4.17ns ± 1%   +6.31%  (p=0.000 n=19+20)
Take/Len=128-8        36.4ns ± 2%    35.6ns ± 2%   -2.21%  (p=0.000 n=20+20)
Take/Len=1024-8        278ns ± 5%     271ns ± 2%   -2.29%  (p=0.001 n=17+19)
Take/Len=1048576-8    1.27ms ± 2%    0.67ms ± 6%  -46.97%  (p=0.000 n=19+20)

name                old alloc/op   new alloc/op   delta
Take/Len=2-8           0.00B          0.00B          ~     (all equal)
Take/Len=4-8           0.00B          0.00B          ~     (all equal)
Take/Len=8-8           0.00B          0.00B          ~     (all equal)
Take/Len=128-8         0.00B          0.00B          ~     (all equal)
Take/Len=1024-8        0.00B          0.00B          ~     (all equal)
Take/Len=1048576-8     0.00B          0.00B          ~     (all equal)

name                old allocs/op  new allocs/op  delta
Take/Len=2-8            0.00           0.00          ~     (all equal)
Take/Len=4-8            0.00           0.00          ~     (all equal)
Take/Len=8-8            0.00           0.00          ~     (all equal)
Take/Len=128-8          0.00           0.00          ~     (all equal)
Take/Len=1024-8         0.00           0.00          ~     (all equal)
Take/Len=1048576-8      0.00           0.00          ~     (all equal)

Author: sbinet

sliceop/sliceop.go

@@ -101,8 +101,11 @@ func Take[T any](dst, src []T, indices []int) []T {
 	}
 
 	dst[0] = src[indices[0]]
-	for i := 1; i < len(indices); i++ {
-		v0 := indices[i-1]
+	var (
+		v0 = indices[0]
+		nn = len(indices)
+	)
+	for i := 1; i < nn; i++ {
 		v1 := indices[i]
 		switch {
 		case v0 < v1:
@@ -112,8 +115,10 @@ func Take[T any](dst, src []T, indices []int) []T {
 		case v0 > v1:
 			panic(errSortedIndices)
 		}
-		dst[i] = src[v1]
+		dst[i-1] = src[v0]
+		v0 = v1
 	}
+	dst[nn-1] = src[v0]
 
 	return dst
 }