bytes: simpler and faster FieldsFunc (apply same changes as for strings)

This change applies the recent changes to the strings package
(https://golang.org/cl/229765) to this package, with relevant
local adjustments.

In contrast to the changes in strings, for the bytes package
this change leads to a nice performance improvement of >10%.

Benchmarks run on a "quiet" MacBook Pro, 3.3GHz Dual-Core Intel Core i7,
with 16GB 2133MHz LPDDR3 RAM running macOS 10.15.4.

name                        old time/op    new time/op     delta
FieldsFunc/ASCII/16-4          191ns ± 9%      163ns ± 2%  -14.66%  (p=0.008 n=5+5)
FieldsFunc/ASCII/256-4        2.08µs ± 5%     1.80µs ± 8%  -13.51%  (p=0.008 n=5+5)
FieldsFunc/ASCII/4096-4       36.1µs ± 2%     31.7µs ± 7%  -12.34%  (p=0.008 n=5+5)
FieldsFunc/ASCII/65536-4       584µs ± 3%      517µs ± 3%  -11.52%  (p=0.008 n=5+5)
FieldsFunc/ASCII/1048576-4    9.45ms ± 3%     8.11ms ± 8%  -14.19%  (p=0.008 n=5+5)
FieldsFunc/Mixed/16-4          202ns ± 2%      181ns ± 2%  -10.02%  (p=0.008 n=5+5)
FieldsFunc/Mixed/256-4        2.12µs ± 6%     1.78µs ± 3%  -16.16%  (p=0.008 n=5+5)
FieldsFunc/Mixed/4096-4       40.4µs ± 2%     36.1µs ± 1%  -10.66%  (p=0.008 n=5+5)
FieldsFunc/Mixed/65536-4       700µs ± 5%      612µs ± 1%  -12.59%  (p=0.008 n=5+5)
FieldsFunc/Mixed/1048576-4    11.2ms ± 8%     10.3ms ± 4%   -8.15%  (p=0.008 n=5+5)

name                        old speed      new speed       delta
FieldsFunc/ASCII/16-4       84.0MB/s ± 9%   98.3MB/s ± 2%  +17.04%  (p=0.008 n=5+5)
FieldsFunc/ASCII/256-4       123MB/s ± 5%    143MB/s ± 8%  +15.74%  (p=0.008 n=5+5)
FieldsFunc/ASCII/4096-4      113MB/s ± 2%    130MB/s ± 6%  +14.20%  (p=0.008 n=5+5)
FieldsFunc/ASCII/65536-4     112MB/s ± 2%    127MB/s ± 3%  +13.05%  (p=0.008 n=5+5)
FieldsFunc/ASCII/1048576-4   111MB/s ± 3%    130MB/s ± 8%  +16.66%  (p=0.008 n=5+5)
FieldsFunc/Mixed/16-4       79.3MB/s ± 2%   88.2MB/s ± 2%  +11.15%  (p=0.008 n=5+5)
FieldsFunc/Mixed/256-4       121MB/s ± 5%    144MB/s ± 3%  +19.19%  (p=0.008 n=5+5)
FieldsFunc/Mixed/4096-4      101MB/s ± 2%    113MB/s ± 1%  +11.92%  (p=0.008 n=5+5)
FieldsFunc/Mixed/65536-4    93.7MB/s ± 5%  107.1MB/s ± 1%  +14.31%  (p=0.008 n=5+5)
FieldsFunc/Mixed/1048576-4  93.6MB/s ± 8%  101.8MB/s ± 4%   +8.77%  (p=0.008 n=5+5)

name                        old alloc/op   new alloc/op    delta
FieldsFunc/ASCII/16-4          80.0B ± 0%      80.0B ± 0%     ~     (all equal)
FieldsFunc/ASCII/256-4          768B ± 0%       768B ± 0%     ~     (all equal)
FieldsFunc/ASCII/4096-4       24.8kB ± 0%     24.8kB ± 0%     ~     (all equal)
FieldsFunc/ASCII/65536-4       497kB ± 0%      497kB ± 0%     ~     (p=0.444 n=5+5)
FieldsFunc/ASCII/1048576-4    9.61MB ± 0%     9.61MB ± 0%     ~     (p=0.500 n=5+5)
FieldsFunc/Mixed/16-4          96.0B ± 0%      96.0B ± 0%     ~     (all equal)
FieldsFunc/Mixed/256-4          768B ± 0%       768B ± 0%     ~     (all equal)
FieldsFunc/Mixed/4096-4       24.8kB ± 0%     24.8kB ± 0%     ~     (all equal)
FieldsFunc/Mixed/65536-4       497kB ± 0%      497kB ± 0%     ~     (all equal)
FieldsFunc/Mixed/1048576-4    9.61MB ± 0%     9.61MB ± 0%     ~     (p=0.881 n=5+5)

name                        old allocs/op  new allocs/op   delta
FieldsFunc/ASCII/16-4           1.00 ± 0%       1.00 ± 0%     ~     (all equal)
FieldsFunc/ASCII/256-4          1.00 ± 0%       1.00 ± 0%     ~     (all equal)
FieldsFunc/ASCII/4096-4         5.00 ± 0%       5.00 ± 0%     ~     (all equal)
FieldsFunc/ASCII/65536-4        12.0 ± 0%       12.0 ± 0%     ~     (all equal)
FieldsFunc/ASCII/1048576-4      24.0 ± 0%       24.0 ± 0%     ~     (all equal)
FieldsFunc/Mixed/16-4           1.00 ± 0%       1.00 ± 0%     ~     (all equal)
FieldsFunc/Mixed/256-4          1.00 ± 0%       1.00 ± 0%     ~     (all equal)
FieldsFunc/Mixed/4096-4         5.00 ± 0%       5.00 ± 0%     ~     (all equal)
FieldsFunc/Mixed/65536-4        12.0 ± 0%       12.0 ± 0%     ~     (all equal)
FieldsFunc/Mixed/1048576-4      24.0 ± 0%       24.0 ± 0%     ~     (all equal)

Change-Id: I59a2ed52563851c693b2c8dfce7e3cde640f62a9
Reviewed-on: https://go-review.googlesource.com/c/go/+/231120
Reviewed-by: Ian Lance Taylor <iant@golang.org>

1 files changed, 11 insertions, 10 deletions

diff --git a/src/bytes/bytes.go b/src/bytes/bytes.go
index 16d1854c44..aa07b9fbc1 100644
--- a/src/bytes/bytes.go
+++ b/src/bytes/bytes.go
@@ -458,8 +458,10 @@ func FieldsFunc(s []byte, f func(rune) bool) [][]byte {
 	spans := make([]span, 0, 32)
 
 	// Find the field start and end indices.
-	wasField := false
-	fromIndex := 0
+	// Doing this in a separate pass (rather than slicing the string s
+	// and collecting the result substrings right away) is significantly
+	// more efficient, possibly due to cache effects.
+	start := -1 // valid span start if >= 0
 	for i := 0; i < len(s); {
 		size := 1
 		r := rune(s[i])
@@ -467,22 +469,21 @@ func FieldsFunc(s []byte, f func(rune) bool) [][]byte {
 			r, size = utf8.DecodeRune(s[i:])
 		}
 		if f(r) {
-			if wasField {
-				spans = append(spans, span{start: fromIndex, end: i})
-				wasField = false
+			if start >= 0 {
+				spans = append(spans, span{start, i})
+				start = -1
 			}
 		} else {
-			if !wasField {
-				fromIndex = i
-				wasField = true
+			if start < 0 {
+				start = i
 			}
 		}
 		i += size
 	}
 
 	// Last field might end at EOF.
-	if wasField {
-		spans = append(spans, span{fromIndex, len(s)})
+	if start >= 0 {
+		spans = append(spans, span{start, len(s)})
 	}
 
 	// Create subslices from recorded field indices.