Gerrit Bot (Gerrit)

unread,

Aug 10, 2026, 9:10:58 AM (2 days ago) Aug 10

to goph...@pubsubhelper.golang.org, David Teather, golang-co...@googlegroups.com

Gerrit Bot has uploaded the change for review

Commit message

internal/strconv: check for digits first in readFloat

readFloat scans a float literal one byte at a time through a switch whose
arms were ordered rarest first. Every ordinary digit tested two rarer
cases and missed before matching. Reorder the arms so the two digit cases
come first. The case bodies are unchanged.

The package had no benchmark for hex or underscore input, so two are
added, and the existing benchmarks are converted to b.Loop.

goos: darwin
goarch: arm64
pkg: internal/strconv
cpu: Apple M2 Pro
                          │     old     │                 new                 │
                          │   sec/op    │   sec/op     vs base                │
Atof64Decimal-10            18.38n ± 0%   15.12n ± 0%  -17.74% (p=0.000 n=25)
Atof64Float-10              22.39n ± 1%   18.11n ± 0%  -19.12% (p=0.000 n=25)
Atof64FloatExp-10           23.74n ± 0%   22.31n ± 0%   -6.02% (p=0.000 n=25)
Atof64Big-10                58.92n ± 0%   42.75n ± 1%  -27.44% (p=0.000 n=25)
Atof64RandomBits-10         63.19n ± 0%   53.62n ± 1%  -15.14% (p=0.000 n=25)
Atof64RandomFloats-10       57.59n ± 0%   46.36n ± 0%  -19.50% (p=0.000 n=25)
Atof64RandomLongFloats-10   72.98n ± 1%   63.56n ± 0%  -12.91% (p=0.000 n=25)
Atof64Hex-10                38.01n ± 0%   35.51n ± 0%   -6.58% (p=0.000 n=25)
Atof64Underscores-10        55.96n ± 0%   58.54n ± 0%   +4.61% (p=0.000 n=25)
Atof32Decimal-10            17.81n ± 0%   14.53n ± 1%  -18.42% (p=0.000 n=25)
Atof32Float-10              20.87n ± 0%   17.05n ± 0%  -18.30% (p=0.000 n=25)
Atof32FloatExp-10           26.29n ± 0%   23.01n ± 0%  -12.48% (p=0.000 n=25)
Atof32Random-10             37.93n ± 0%   34.25n ± 0%   -9.70% (p=0.000 n=25)
Atof32RandomLong-10         64.19n ± 0%   55.40n ± 1%  -13.69% (p=0.000 n=25)
geomean                     36.60n        31.45n       -14.06%

The linux/amd64 numbers below are from a shared 2-vCPU cloud instance and
are noisy, they should be treated as directional only. The smaller win is
expected because readFloat drops 25 instructions on arm64 against 10 on
amd64, since the reorder also removes a register shuffle the compiler
emits at the loop head.

goos: linux
goarch: amd64
pkg: internal/strconv
cpu: AMD EPYC 7763 64-Core Processor
                         │     old      │                 new                 │
                         │    sec/op    │   sec/op     vs base                │
Atof64Decimal-2             32.41n ± 2%   28.59n ± 3%  -11.79% (p=0.000 n=25)
Atof64Float-2               39.01n ± 3%   36.12n ± 3%   -7.41% (p=0.000 n=25)
Atof64FloatExp-2            39.11n ± 4%   37.67n ± 3%   -3.68% (p=0.003 n=25)
Atof64Big-2                 99.93n ± 2%   79.83n ± 4%  -20.11% (p=0.000 n=25)
Atof64RandomBits-2         100.70n ± 4%   89.12n ± 3%  -11.50% (p=0.000 n=25)
Atof64RandomFloats-2        86.94n ± 5%   80.81n ± 4%   -7.05% (p=0.000 n=25)
Atof64RandomLongFloats-2    121.8n ± 3%   117.3n ± 2%   -3.69% (p=0.000 n=25)
Atof64Hex-2                 67.69n ± 3%   64.36n ± 3%   -4.92% (p=0.000 n=25)
Atof64Underscores-2         83.52n ± 3%   84.34n ± 4%        ~ (p=0.200 n=25)
Atof32Decimal-2             29.56n ± 3%   27.17n ± 4%   -8.09% (p=0.000 n=25)
Atof32Float-2               33.92n ± 3%   33.88n ± 4%        ~ (p=0.229 n=25)
Atof32FloatExp-2            42.65n ± 3%   41.88n ± 4%   -1.81% (p=0.038 n=25)
Atof32Random-2              61.15n ± 2%   58.62n ± 3%   -4.14% (p=0.001 n=25)
Atof32RandomLong-2         107.70n ± 3%   99.94n ± 3%   -7.21% (p=0.000 n=25)
geomean                     60.29n        56.30n        -6.62%

B/op and allocs/op are unchanged. Input that is mostly underscore
separators is about 5% slower, since that arm moved from first to last.

Updates #66327

Change-Id: Ifaaecc9891cbc3d7f107a2fff79a55884e8c8d84

GitHub-Last-Rev: 5f4ae844c7390f31d73ad30ad5cb8191e7588c33

GitHub-Pull-Request: golang/go#80813

Change diff

diff --git a/src/internal/strconv/atof.go b/src/internal/strconv/atof.go
index 4567af3..64acd46 100644
--- a/src/internal/strconv/atof.go
+++ b/src/internal/strconv/atof.go
@@ -213,18 +213,8 @@
 loop:
 	for ; i < len(s); i++ {
 		switch c := s[i]; true {
-		case c == '_':
-			underscores = true
-			continue
-
-		case c == '.':
-			if sawdot {
-				break loop
-			}
-			sawdot = true
-			dp = nd
-			continue
-
+		// Digits are by far the most common character here,
+		// so check for them first.
 		case '0' <= c && c <= '9':
 			sawdigits = true
 			if c == '0' && nd == 0 { // ignore leading zeros
@@ -252,6 +242,18 @@
 				trunc = true
 			}
 			continue
+
+		case c == '.':
+			if sawdot {
+				break loop
+			}
+			sawdot = true
+			dp = nd
+			continue
+
+		case c == '_':
+			underscores = true
+			continue
 		}
 		break
 	}
diff --git a/src/internal/strconv/atof_test.go b/src/internal/strconv/atof_test.go
index 05ac9e6..3c5135a 100644
--- a/src/internal/strconv/atof_test.go
+++ b/src/internal/strconv/atof_test.go
@@ -646,25 +646,25 @@
 }
 
 func BenchmarkAtof64Decimal(b *testing.B) {
-	for i := 0; i < b.N; i++ {
+	for b.Loop() {
 		ParseFloat("33909", 64)
 	}
 }
 
 func BenchmarkAtof64Float(b *testing.B) {
-	for i := 0; i < b.N; i++ {
+	for b.Loop() {
 		ParseFloat("339.7784", 64)
 	}
 }
 
 func BenchmarkAtof64FloatExp(b *testing.B) {
-	for i := 0; i < b.N; i++ {
+	for b.Loop() {
 		ParseFloat("-5.09e75", 64)
 	}
 }
 
 func BenchmarkAtof64Big(b *testing.B) {
-	for i := 0; i < b.N; i++ {
+	for b.Loop() {
 		ParseFloat("123456789123456789123456789", 64)
 	}
 }
@@ -672,16 +672,20 @@
 func BenchmarkAtof64RandomBits(b *testing.B) {
 	initAtof()
 	b.ResetTimer()
-	for i := 0; i < b.N; i++ {
+	i := 0
+	for b.Loop() {
 		ParseFloat(benchmarksRandomBits[i%1024], 64)
+		i++
 	}
 }
 
 func BenchmarkAtof64RandomFloats(b *testing.B) {
 	initAtof()
 	b.ResetTimer()
-	for i := 0; i < b.N; i++ {
+	i := 0
+	for b.Loop() {
 		ParseFloat(benchmarksRandomNormal[i%1024], 64)
+		i++
 	}
 }
 
@@ -693,7 +697,7 @@
 	}
 	b.ResetTimer()
 	idx := 0
-	for i := 0; i < b.N; i++ {
+	for b.Loop() {
 		ParseFloat(samples[idx], 64)
 		idx++
 		if idx == len(samples) {
@@ -702,20 +706,32 @@
 	}
 }
 
+func BenchmarkAtof64Hex(b *testing.B) {
+	for b.Loop() {
+		ParseFloat("0x1.fp-2", 64)
+	}
+}
+
+func BenchmarkAtof64Underscores(b *testing.B) {
+	for b.Loop() {
+		ParseFloat("1_2_3_4_5_6_7_8.9_1", 64)
+	}
+}
+
 func BenchmarkAtof32Decimal(b *testing.B) {
-	for i := 0; i < b.N; i++ {
+	for b.Loop() {
 		ParseFloat("33909", 32)
 	}
 }
 
 func BenchmarkAtof32Float(b *testing.B) {
-	for i := 0; i < b.N; i++ {
+	for b.Loop() {
 		ParseFloat("339.778", 32)
 	}
 }
 
 func BenchmarkAtof32FloatExp(b *testing.B) {
-	for i := 0; i < b.N; i++ {
+	for b.Loop() {
 		ParseFloat("12.3456e32", 32)
 	}
 }
@@ -728,8 +744,10 @@
 		float32strings[i] = FormatFloat(float64(math.Float32frombits(n)), 'g', -1, 32)
 	}
 	b.ResetTimer()
-	for i := 0; i < b.N; i++ {
+	i := 0
+	for b.Loop() {
 		ParseFloat(float32strings[i%4096], 32)
+		i++
 	}
 }
 
@@ -741,7 +759,9 @@
 		float32strings[i] = FormatFloat(float64(math.Float32frombits(n)), 'g', 20, 32)
 	}
 	b.ResetTimer()
-	for i := 0; i < b.N; i++ {
+	i := 0
+	for b.Loop() {
 		ParseFloat(float32strings[i%4096], 32)
+		i++
 	}
 }

Change information

Files:

M src/internal/strconv/atof.go
M src/internal/strconv/atof_test.go

Change size: M

Delta: 2 files changed, 46 insertions(+), 24 deletions(-)

Open in Gerrit

Related details

Attention set is empty

Submit Requirements:

Code-Review
No-Unresolved-Comments
Review-Enforcement
TryBots-Pass

Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings.

Gerrit

unsatisfied_requirement

satisfied_requirement

open

diffy

David Teather (Gerrit)

unread,

Aug 10, 2026, 9:18:26 AM (2 days ago) Aug 10

to Gerrit Bot, goph...@pubsubhelper.golang.org, golang-co...@googlegroups.com

David Teather added 1 comment

Patchset-level comments

File-level comment, Patchset 1 (Latest):

David Teather . unresolved

I want to double check that my benchmarks are after the b.Loop() modernization changes when I get home today. I think at least the Mac comparison was but need to verify the Linux one.

Open in Gerrit

Related details

Attention set is empty

Submit Requirements:

Code-Review

No-Unresolved-Comments

Review-Enforcement
TryBots-Pass

Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings.

Gerrit

unsatisfied_requirement

open

diffy

David Teather (Gerrit)

unread,

Aug 10, 2026, 9:51:21 PM (2 days ago) Aug 10

to Gerrit Bot, goph...@pubsubhelper.golang.org, golang-co...@googlegroups.com

David Teather added 1 comment

Patchset-level comments

File-level comment, Patchset 1 (Latest):

David Teather . resolved

I want to double check that my benchmarks are after the b.Loop() modernization changes when I get home today. I think at least the Mac comparison was but need to verify the Linux one.

David Teather

Verified that they were both taken post-b.Loop() modernization.

Open in Gerrit

Related details

Attention set is empty

Submit Requirements:

Code-Review

No-Unresolved-Comments

Review-Enforcement
TryBots-Pass

Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings.

Gerrit

unsatisfied_requirement

satisfied_requirement

open

diffy

Robert Griesemer (Gerrit)

unread,

Aug 11, 2026, 12:57:20 PM (15 hours ago) Aug 11

to David Teather, Gerrit Bot, goph...@pubsubhelper.golang.org, golang-co...@googlegroups.com

Robert Griesemer voted Commit-Queue+1

Commit-Queue

+1

Open in Gerrit

Related details

Attention set is empty

Submit Requirements:

Code-Review
No-Unresolved-Comments
Review-Enforcement
TryBots-Pass

Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings.

Gerrit

unsatisfied_requirement

satisfied_requirement

open

diffy

Robert Griesemer (Gerrit)

unread,

Aug 11, 2026, 1:26:51 PM (14 hours ago) Aug 11

to David Teather, Gerrit Bot, goph...@pubsubhelper.golang.org, golang...@luci-project-accounts.iam.gserviceaccount.com, golang-co...@googlegroups.com

Robert Griesemer voted and added 2 comments

Votes added by Robert Griesemer

Auto-Submit	+1
Code-Review	+2

2 comments

Patchset-level comments

File-level comment, Patchset 1 (Latest):

Robert Griesemer . resolved

Nice, thanks!
LGTM but please address the minor grammar mistake.

File src/internal/strconv/atof.go

Line 216, Patchset 1 (Latest): // Digits are by far the most common character here,

Robert Griesemer . unresolved

s/character/characters/

also, feel free to make this just one line (no need for line break)

Open in Gerrit

Related details

Attention set is empty

Submit Requirements:

Code-Review
No-Unresolved-Comments

Review-Enforcement

TryBots-Pass

Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings.

Gerrit

satisfied_requirement

unsatisfied_requirement

open

diffy

Gerrit Bot (Gerrit)

unread,

Aug 11, 2026, 8:28:37 PM (7 hours ago) Aug 11

to David Teather, goph...@pubsubhelper.golang.org, golang-co...@googlegroups.com

Attention needed from Robert Griesemer

Gerrit Bot uploaded new patchset

Gerrit Bot uploaded patch set #2 to this change.

Following approvals got outdated and were removed:

Code-Review: +2 by Robert Griesemer
TryBots-Pass: LUCI-TryBot-Result+1 by golang...@luci-project-accounts.iam.gserviceaccount.com

Open in Gerrit

Related details

Attention is currently required from:

Robert Griesemer

Submit Requirements:

Code-Review

No-Unresolved-Comments
Review-Enforcement

TryBots-Pass

Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings.

Gerrit

unsatisfied_requirement

open

diffy

David Teather (Gerrit)

unread,

Aug 11, 2026, 8:29:39 PM (7 hours ago) Aug 11

to Gerrit Bot, goph...@pubsubhelper.golang.org, Robert Griesemer, golang...@luci-project-accounts.iam.gserviceaccount.com, golang-co...@googlegroups.com

Attention needed from Robert Griesemer

David Teather added 2 comments

Patchset-level comments

File-level comment, Patchset 1:

Robert Griesemer . resolved

Nice, thanks!
LGTM but please address the minor grammar mistake.

David Teather

Updated, thanks for the review!

File src/internal/strconv/atof.go

Line 216, Patchset 1: // Digits are by far the most common character here,

Robert Griesemer . resolved

s/character/characters/
also, feel free to make this just one line (no need for line break)

David Teather

Updated to this and made it one line. thanks!
> // Digits are by far the most common characters here, so check for them first.

Open in Gerrit

Related details

Attention is currently required from:

Robert Griesemer

Submit Requirements:

Code-Review

No-Unresolved-Comments

Review-Enforcement
TryBots-Pass

Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings.

Gerrit

unsatisfied_requirement

satisfied_requirement

open

diffy

Reply all

Reply to author

Forward

[go] internal/strconv: check for digits first in readFloat

Gerrit Bot (Gerrit)

Gerrit Bot has uploaded the change for review

Commit message

Change diff

Change information

Related details

David Teather (Gerrit)

David Teather added 1 comment

Related details

David Teather (Gerrit)

David Teather added 1 comment

Related details

Robert Griesemer (Gerrit)

Robert Griesemer voted Commit-Queue+1

Related details

Robert Griesemer (Gerrit)

Robert Griesemer voted and added 2 comments

Votes added by Robert Griesemer

2 comments

Related details

Gerrit Bot (Gerrit)

Gerrit Bot uploaded new patchset

Related details

David Teather (Gerrit)

David Teather added 2 comments

Related details