strings: simpler and slightly faster implementation of FieldsFunc

Removed the need for maintaining an extra variable in one of the inner loops,
leading to a slight speed-up for short strings.

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          147ns ± 0%     144ns ± 1%  -2.04%  (p=0.000 n=4+5)
FieldsFunc/ASCII/256-4        1.63µs ± 0%    1.59µs ± 1%  -2.50%  (p=0.008 n=5+5)
FieldsFunc/ASCII/4096-4       30.0µs ± 0%    29.3µs ± 2%    ~     (p=0.190 n=4+5)
FieldsFunc/ASCII/65536-4       491µs ± 5%     473µs ± 2%    ~     (p=0.095 n=5+5)
FieldsFunc/ASCII/1048576-4    8.02ms ± 7%    7.85ms ± 4%    ~     (p=0.548 n=5+5)
FieldsFunc/Mixed/16-4          182ns ± 1%     181ns ± 4%    ~     (p=0.357 n=5+5)
FieldsFunc/Mixed/256-4        1.74µs ± 1%    1.74µs ± 1%    ~     (p=0.881 n=5+5)
FieldsFunc/Mixed/4096-4       34.9µs ± 2%    34.7µs ± 0%    ~     (p=0.310 n=5+5)
FieldsFunc/Mixed/65536-4       595µs ± 1%     589µs ± 2%    ~     (p=0.095 n=5+5)
FieldsFunc/Mixed/1048576-4    10.1ms ± 3%     9.8ms ± 2%    ~     (p=0.095 n=5+5)

name                        old speed      new speed      delta
FieldsFunc/ASCII/16-4        109MB/s ± 1%   111MB/s ± 1%  +2.33%  (p=0.008 n=5+5)
FieldsFunc/ASCII/256-4       157MB/s ± 0%   161MB/s ± 1%  +2.57%  (p=0.008 n=5+5)
FieldsFunc/ASCII/4096-4      137MB/s ± 0%   140MB/s ± 2%    ~     (p=0.190 n=4+5)
FieldsFunc/ASCII/65536-4     134MB/s ± 4%   139MB/s ± 2%    ~     (p=0.095 n=5+5)
FieldsFunc/ASCII/1048576-4   131MB/s ± 6%   134MB/s ± 4%    ~     (p=0.548 n=5+5)
FieldsFunc/Mixed/16-4       87.8MB/s ± 1%  88.3MB/s ± 4%    ~     (p=0.421 n=5+5)
FieldsFunc/Mixed/256-4       147MB/s ± 1%   147MB/s ± 1%    ~     (p=0.841 n=5+5)
FieldsFunc/Mixed/4096-4      117MB/s ± 2%   118MB/s ± 0%    ~     (p=0.310 n=5+5)
FieldsFunc/Mixed/65536-4     110MB/s ± 1%   111MB/s ± 2%    ~     (p=0.095 n=5+5)
FieldsFunc/Mixed/1048576-4   104MB/s ± 3%   107MB/s ± 2%    ~     (p=0.095 n=5+5)

name                        old alloc/op   new alloc/op   delta
FieldsFunc/ASCII/16-4          32.0B ± 0%     32.0B ± 0%    ~     (all equal)
FieldsFunc/ASCII/256-4          352B ± 0%      352B ± 0%    ~     (all equal)
FieldsFunc/ASCII/4096-4       21.9kB ± 0%    21.9kB ± 0%    ~     (all equal)
FieldsFunc/ASCII/65536-4       448kB ± 0%     448kB ± 0%    ~     (all equal)
FieldsFunc/ASCII/1048576-4    8.85MB ± 0%    8.85MB ± 0%    ~     (p=0.738 n=5+5)
FieldsFunc/Mixed/16-4          48.0B ± 0%     48.0B ± 0%    ~     (all equal)
FieldsFunc/Mixed/256-4          416B ± 0%      416B ± 0%    ~     (all equal)
FieldsFunc/Mixed/4096-4       21.5kB ± 0%    21.5kB ± 0%    ~     (all equal)
FieldsFunc/Mixed/65536-4       448kB ± 0%     448kB ± 0%    ~     (all equal)
FieldsFunc/Mixed/1048576-4    8.85MB ± 0%    8.85MB ± 0%    ~     (p=0.690 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: I06828d798ca1a624a26edd7f7b68c3bf2fc28f84
Reviewed-on: https://go-review.googlesource.com/c/go/+/229765
Reviewed-by: Martin Möhrmann <moehrmann@google.com>

diff --git a/src/strings/strings.go b/src/strings/strings.go
index 5f244d6e204906af42f932ac6ba1a27fde679e73..88fbeecc6f2c95213978d07b9cf01f33f25a2a14 100644 (file)
--- a/src/strings/strings.go
+++ b/src/strings/strings.go
@@ -379,25 +379,29 @@ func FieldsFunc(s string, f func(rune) bool) []string {
        spans := make([]span, 0, 32)
 
        // Find the field start and end indices.
-       wasField := false
-       fromIndex := 0
-       for i, rune := range s {
+       // 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 end, rune := range s {
                if f(rune) {
-                       if wasField {
-                               spans = append(spans, span{start: fromIndex, end: i})
-                               wasField = false
+                       if start >= 0 {
+                               spans = append(spans, span{start, end})
+                               // Set start to a negative value.
+                               // Note: using -1 here consistently and reproducibly
+                               // slows down this code by a several percent on amd64.
+                               start = ^start
                        }
                } else {
-                       if !wasField {
-                               fromIndex = i
-                               wasField = true
+                       if start < 0 {
+                               start = end
                        }
                }
        }
 
        // 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 strings from recorded field indices.