Port RegularExpressions benchmarks

[adamsitnik]

Fixes #71 and #76

TestData:

To consume the TestData I added new NuGet feed and configured the project to copy the files:

<add key="dotnet-core" value="https://dotnet.myget.org/F/dotnet-core/api/v3/index.json" />

  <ItemGroup>
    <Content Include="$(NuGetPackageRoot)\system.text.regularexpressions.testdata\1.0.2\content\regexredux\*.*">
      <Link>corefx\System.Text.RegularExpressions\content\%(Filename)%(Extension)</Link>
      <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
    </Content>

However, I am not sure if I should not simply copy-paste the single file used by benchmarks. It's 2 MB file, it would simplify it.

@jorive @ViktorHofer any thoughts on this?

[adamsitnik]

I have also removed the managed heap allocations from TestData of Perf_Regex. Every iteration through Match_TestData was causing managed array allocation (new object[] ) and boxing (RegexOptions is an enum, if we store it in object array we box it).

public static IEnumerable<object[]> Match_TestData()
 {
     yield return new object[] { "[abcd-[d]]+", "dddaabbccddd", RegexOptions.None };
}

The difference:

Before:

Method	Mean	Error	StdDev	Median	Min	Max	Gen 0	Allocated
Match	177.9 ms	9.831 ms	11.32 ms	175.1 ms	161.4 ms	202.4 ms	30000.0000	181.99 MB

After:

Method	Mean	Error	StdDev	Median	Min	Max	Gen 0	Allocated
Match	158.7 ms	2.299 ms	2.038 ms	158.5 ms	155.8 ms	163.2 ms	29000.0000	179.63 MB

[ViktorHofer]

We copied that over from our Match unit test. Should we remove some permutations?

[adamsitnik]

Today if this benchmark fails we don't know which test case was regressed.

In the perfect scenario, we would have a single benchmark per most commonly used regex expression to monitor its performance.

@ViktorHofer Do we have that data/knowledge somewhere to split it into many benchmarks?

[jorive]

Should this path be read from some sort of config file for this benchmark, so we do not need to rebuild to test against different content?

[adamsitnik]

It could be, however I decided to port it as is, the source benchmark has the file name hardcoded https://github.com/dotnet/corefx/blob/master/src/System.Text.RegularExpressions/tests/Performance/Perf.RegexRedux.cs#L19

[jorive]

Categories.CoreFX [](start = 23, length = 17)

Do we know if there is any duplication compared with the regex-redux benchmarks from Benchmarks Game in CoreClr?

[adamsitnik]

@ViktorHofer you are the person that is similar with both benchmarks and the regexp code. Could you answer this question?

[jorive]

Perf_Regex2 [](start = 17, length = 11)

Does changing the name here mean new benchmarks?

[adamsitnik]

If I change it the ID exported for BenchView integration purpose is going to change too ;(

[jorive]

Perf_Regex2 was the original name then?

[adamsitnik]

I double checked: no, it's my mistake. Will revert now. Good catch!

[adamsitnik]

@jorive reverted!

[jorive]

LGTM