Use cached strings for ToString on integer values 0 to 9 #18383

stephentoub · 2018-06-08T21:29:06Z

When doing ToString on single-digit integers, just return a cached string. This helps avoid string allocation/formatting costs for this very common case, while only costing one comparison for other cases.

Microbenchmark:

using System;
using System.Diagnostics;
using System.Runtime.CompilerServices;
using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Running;
using BenchmarkDotNet.Attributes.Jobs;

[MemoryDiagnoser]
[InProcess]
public class Benchmarks
{
    private static void Main() => BenchmarkRunner.Run<Benchmarks>();

    [Benchmark]
    public static string ToString1() => 1.ToString();

    [Benchmark]
    public static string ToString11() => 11.ToString();
}

Before:

Method	Mean	Error	StdDev	Gen 0	Allocated
ToString1	20.13 ns	0.4800 ns	0.5527 ns	0.0057	24 B
ToString11	23.45 ns	0.5625 ns	1.0970 ns	0.0076	32 B

After:

Method	Mean	Error	StdDev	Gen 0	Allocated
ToString1	14.79 ns	0.3820 ns	0.3752 ns	-	0 B
ToString11	23.11 ns	0.5537 ns	0.6154 ns	0.0076	32 B

cc: @jkotas, @mjsabby, @ahsonkhan

When doing ToString on single-digit integers, just return a cached string. This helps avoid string allocation/formatting costs for this very common case, while only costing one comparison for other cases.

jkotas

Nice

briansull · 2018-06-08T21:50:38Z

The value 10 is also a commonly used, if you wan to add one more to the cache.

briansull · 2018-06-08T21:53:13Z

src/System.Private.CoreLib/shared/System/Number.Formatting.cs

@@ -1095,6 +1097,13 @@ private static unsafe void UInt32ToNumber(uint value, ref NumberBuffer number)
        private static unsafe string UInt32ToDecStr(uint value, int digits)
        {
            int bufferLength = Math.Max(digits, FormattingHelpers.CountDigits(value));


Why do we have to call CountDigits, when we have a small 'value' 0-10 ?
Can't we first test the value and see it is range. [0-10]

CountDigits is generally quite fast already and in the 0-10 case, it would be just 2 comparison instead of one (without adding the overhead of the 0-10 check for all other values):

coreclr/src/System.Private.CoreLib/shared/System/Buffers/Text/FormattingHelpers.CountDigits.cs

Line 69 in d2e2299

public static int CountDigits(uint value)

stephentoub · 2018-06-08T21:54:23Z

The value 10 is also a commonly used, if you wan to add one more to the cache.

We could experiment with that. It would entail not only adding one more string but also changing the bufferLength == 1 comparison to something more like digits <= 1 && (uint)value <= 10 (which would skip the fast path for the niche case where digits is 2 and value is 10), or bufferLength == 1 || (digits <= 2 && value == 10), or something like that.

briansull · 2018-06-08T22:03:11Z

src/System.Private.CoreLib/shared/System/Number.Formatting.cs

+            // For single-digit values that are very common, especially 0 and 1, just return cached strings.
+            if (bufferLength == 1)
+            {
+                return s_singleDigitStringCache[value];


if (value <= 10) { return s_smallNumberStringCache[value]; }

That also needs to factor in digits. This can be called with something like digits == 5 and value == 2, and it needs to return "00002".

In that case, should we be validating that digits is -1/etc., or is it safe to always return single digits as "digit" if bufferLength is 1?

In that case, should we be validating that digits is -1/etc., or is it safe to always return single digits as "digit" if bufferLength is 1?

No matter what the digits value is, there will always be at least one digit rendered.

benaadams · 2018-06-08T22:15:43Z

src/System.Private.CoreLib/shared/System/Number.Formatting.cs

@@ -1339,6 +1348,13 @@ private static unsafe string UInt64ToDecStr(ulong value, int digits)
                digits = 1;

            int bufferLength = Math.Max(digits, FormattingHelpers.CountDigits(value));
+
+            // For single-digit values that are very common, especially 0 and 1, just return cached strings.
+            if (bufferLength == 1)


Skip range check for this one (moving bufferLength after loop) and use

string[] ditgitCache = s_singleDigitStringCache; if ((uint)value < (uint)ditgitCache.Length) { return ditgitCache[value]; }

Would also allow you to have any number in cache? (within reason)

Would also allow you to have any number in cache? (within reason)

As noted earlier, it would also need to check digits. That's why I did the comparison against bufferLength.

Ahh... leading zeros...

Then we can use this:

if ((value <= s_smallNumberStringCache.Length) && (digits <= FormattingHelpers.CountDigits(value))) { return s_smallNumberStringCache[value]; }

Note value is already unsigned (a least in this variant of the method)

Right, and I've been simply noting it would need to be measured.

ahsonkhan

Awesome!

stephentoub · 2018-06-09T02:28:30Z

I'm looking into the commonality of other values, and will follow-up if it looks fruitful. For now, my goal was really just focused on 0 and 1, and 2-9 came along for the ride (cheaper to support them than to not support them).

When doing ToString on single-digit integers, just return a cached string. This helps avoid string allocation/formatting costs for this very common case, while only costing one comparison for other cases. Signed-off-by: dotnet-bot-corefx-mirror <[email protected]>

When doing ToString on single-digit integers, just return a cached string. This helps avoid string allocation/formatting costs for this very common case, while only costing one comparison for other cases. Signed-off-by: dotnet-bot <[email protected]>

When doing ToString on single-digit integers, just return a cached string. This helps avoid string allocation/formatting costs for this very common case, while only costing one comparison for other cases. Commit migrated from dotnet/coreclr@12f8a27

Use cached strings for values 0 to 9

3f4006e

When doing ToString on single-digit integers, just return a cached string. This helps avoid string allocation/formatting costs for this very common case, while only costing one comparison for other cases.

jkotas approved these changes Jun 8, 2018

View reviewed changes

briansull reviewed Jun 8, 2018

View reviewed changes

benaadams reviewed Jun 8, 2018

View reviewed changes

ahsonkhan approved these changes Jun 8, 2018

View reviewed changes

stephentoub merged commit 12f8a27 into dotnet:master Jun 9, 2018

stephentoub deleted the singledigitstring branch June 9, 2018 02:53

ahsonkhan mentioned this pull request Jul 16, 2018

Cleanup and move Utf8Parser tests to corefx dotnet/corefxlab#2322

Closed

ahsonkhan mentioned this pull request Sep 5, 2018

Add UTF8 parser performance tests dotnet/corefx#32026

Merged

lewurm mentioned this pull request Feb 1, 2019

[2018-08] Bump corert mono/mono#12721

Merged

stephentoub mentioned this pull request May 9, 2019

Remove unnecessary string allocation from ZipFileUtils dotnet/corefx#37546

Merged

Use cached strings for ToString on integer values 0 to 9 #18383

Use cached strings for ToString on integer values 0 to 9 #18383

Uh oh!

Conversation

stephentoub commented Jun 8, 2018

Uh oh!

jkotas left a comment

Choose a reason for hiding this comment

Uh oh!

briansull commented Jun 8, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

stephentoub commented Jun 8, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ahsonkhan Jun 8, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

briansull Jun 8, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ahsonkhan left a comment

Choose a reason for hiding this comment

Uh oh!

stephentoub commented Jun 9, 2018

Uh oh!

Uh oh!

briansull commented Jun 8, 2018 •

edited

Loading

stephentoub commented Jun 8, 2018 •

edited

Loading

ahsonkhan Jun 8, 2018 •

edited

Loading

briansull Jun 8, 2018 •

edited

Loading