Optimize Array.GetFlattenedIndex (all runtimes) and InternalGetValue (CoreCLR)

[Sergio0694]

This is only used when the element type is a reference type.
Idea came up while chatting with @jakobbotsch on Discord about array covariance being bad 😛
Also cc. @GrabYourPitchforks as this seems like his area as well.
I'm guessing we could also add a managed fast path for SetValue by using CastHelpers, but haven't done that yet.
Do we want to explore that in this same PR or should I publish this and then we can eventually do that separately?

EDIT: seems like the logic for SetValue is (at least in part) in CastHelpers.StelemRef, but that API is... Private? 🤔
Not sure I understand how those are meant to be used, I'm sure there's a reason why they're all private...

API additions (CoreCLR-only)

namespaceSystem.Runtime.CompilerServices{publicstaticclassRuntimeHelpers{internalstaticunsafeobject?Box(MethodTable*methodTable,refbytedata);}internalstructMethodTable{publicboolIsValueType{get;}publicTypeHandleGetArrayElementTypeHandle();}internalstructTypeHandle{publicboolIsTypeDesc{get;}publicMethodTable*AsMethodTable();}}

[ghost]

I couldn't figure out the best area label to add to this PR. If you have write-permissions please help me learn by adding exactly one area label.

[ghost]

Tagging subscribers to this area: @dotnet/area-system-runtime
See info in area-owners.md if you want to be subscribed.

Issue Details

---

This is only used when the element type is a reference type.
Idea came up while chatting with @jakobbotsch on Discord about array covariance being bad 😛
Also cc. @GrabYourPitchforks as this seems like his area as well.
I'm guessing we could also add a managed fast path for SetValue by using CastHelpers, but haven't done that yet.
Do we want to explore that in this same PR or should I publish this and then we can eventually do that separately?

EDIT: seems like the logic for SetValue is (at least in part) in CastHelpers.StelemRef, but that API is... Private? 🤔
Not sure I understand how those are meant to be used, I'm sure there's a reason why they're all private...

Author:	Sergio0694
Assignees:	-
Labels:	area-System.Runtime, community-contribution
Milestone:	-

[Sergio0694]

Sure! Currently reading docs to get started with the performance repo, will share some numbers as soon as I have them 🙂

[Sergio0694]

Benchmark results:

| Method | Job | Toolchain | Mean | Error | StdDev | Ratio | RatioSD | Code Size ||---------------- |----------- |---------------------------- |----------:|----------:|----------:|------:|--------:|----------:|| GetValue_Object | Job-NBCGUL | \Core_Root\corerun.exe | 1.295 ns | 0.0818 ns | 0.1035 ns | 0.23 | 0.02 | 188 B || GetValue_Object | Job-IEBNQH | \Core_Root_base\corerun.exe | 5.748 ns | 0.1714 ns | 0.2669 ns | 1.00 | 0.00 | 231 B |||||||||||| GetValueInt | Job-NBCGUL | \Core_Root\corerun.exe | 44.603 ns | 0.8945 ns | 0.8367 ns | 0.96 | 0.02 | 188 B || GetValueInt | Job-IEBNQH | \Core_Root_base\corerun.exe | 46.517 ns | 0.5497 ns | 0.5142 ns | 1.00 | 0.00 | 231 B |

For:

[GroupBenchmarksBy(BenchmarkLogicalGroupRule.ByCategory)]publicclassArrayGetValueBenchmark{privatereadonlyArrayarrayOfStrings;privatereadonlyArrayarrayOfInts;publicArrayGetValueBenchmark(){arrayOfStrings=new[]{"Hello","World","Bob","Alice","Fred","Tanner","Egor","Single"};arrayOfInts=Enumerable.Range(0,20).ToArray();}[Benchmark][BenchmarkCategory("REF_TYPE")]publicobject?GetValue_Object(){returnarrayOfStrings.GetValue(4);}[Benchmark][BenchmarkCategory("VALUE_TYPE")]publicobject?GetValueInt(){returnarrayOfInts.GetValue(4);}}

Using the benchmark script kindly provided by @EgorBohere.

Results seem in line with what was expected:

• Reference type is much faster due to no internal call and pretty much all inlined.
• Value type case is almost the same or barely faster just due to slightly more efficient flattened index logic.

[jkotas]

Thank you!

[Sergio0694]

This is only used when the element type is a reference type.
Idea came up while chatting with @jakobbotsch on Discord about array covariance being bad 😛
Also cc. @GrabYourPitchforks as this seems like his area as well.
I'm guessing we could also add a managed fast path for SetValue by using CastHelpers, but haven't done that yet.
Do we want to explore that in this same PR or should I publish this and then we can eventually do that separately?

EDIT: seems like the logic for SetValue is (at least in part) in CastHelpers.StelemRef, but that API is... Private? 🤔
Not sure I understand how those are meant to be used, I'm sure there's a reason why they're all private...

API additions (CoreCLR-only)

namespaceSystem.Runtime.CompilerServices{publicstaticclassRuntimeHelpers{internalstaticunsafeobject?Box(MethodTable*methodTable,refbytedata);}internalstructMethodTable{publicboolIsValueType{get;}publicTypeHandleGetArrayElementTypeHandle();}internalstructTypeHandle{publicboolIsTypeDesc{get;}publicMethodTable*AsMethodTable();}}

[ghost]

I couldn't figure out the best area label to add to this PR. If you have write-permissions please help me learn by adding exactly one area label.

[ghost]

Tagging subscribers to this area: @dotnet/area-system-runtime
See info in area-owners.md if you want to be subscribed.

Issue Details

---

This is only used when the element type is a reference type.
Idea came up while chatting with @jakobbotsch on Discord about array covariance being bad 😛
Also cc. @GrabYourPitchforks as this seems like his area as well.
I'm guessing we could also add a managed fast path for SetValue by using CastHelpers, but haven't done that yet.
Do we want to explore that in this same PR or should I publish this and then we can eventually do that separately?

EDIT: seems like the logic for SetValue is (at least in part) in CastHelpers.StelemRef, but that API is... Private? 🤔
Not sure I understand how those are meant to be used, I'm sure there's a reason why they're all private...

Author:	Sergio0694
Assignees:	-
Labels:	area-System.Runtime, community-contribution
Milestone:	-

[Sergio0694]

Sure! Currently reading docs to get started with the performance repo, will share some numbers as soon as I have them 🙂

[Sergio0694]

Benchmark results:

| Method | Job | Toolchain | Mean | Error | StdDev | Ratio | RatioSD | Code Size ||---------------- |----------- |---------------------------- |----------:|----------:|----------:|------:|--------:|----------:|| GetValue_Object | Job-NBCGUL | \Core_Root\corerun.exe | 1.295 ns | 0.0818 ns | 0.1035 ns | 0.23 | 0.02 | 188 B || GetValue_Object | Job-IEBNQH | \Core_Root_base\corerun.exe | 5.748 ns | 0.1714 ns | 0.2669 ns | 1.00 | 0.00 | 231 B |||||||||||| GetValueInt | Job-NBCGUL | \Core_Root\corerun.exe | 44.603 ns | 0.8945 ns | 0.8367 ns | 0.96 | 0.02 | 188 B || GetValueInt | Job-IEBNQH | \Core_Root_base\corerun.exe | 46.517 ns | 0.5497 ns | 0.5142 ns | 1.00 | 0.00 | 231 B |

For:

[GroupBenchmarksBy(BenchmarkLogicalGroupRule.ByCategory)]publicclassArrayGetValueBenchmark{privatereadonlyArrayarrayOfStrings;privatereadonlyArrayarrayOfInts;publicArrayGetValueBenchmark(){arrayOfStrings=new[]{"Hello","World","Bob","Alice","Fred","Tanner","Egor","Single"};arrayOfInts=Enumerable.Range(0,20).ToArray();}[Benchmark][BenchmarkCategory("REF_TYPE")]publicobject?GetValue_Object(){returnarrayOfStrings.GetValue(4);}[Benchmark][BenchmarkCategory("VALUE_TYPE")]publicobject?GetValueInt(){returnarrayOfInts.GetValue(4);}}

Using the benchmark script kindly provided by @EgorBohere.

Results seem in line with what was expected:

• Reference type is much faster due to no internal call and pretty much all inlined.
• Value type case is almost the same or barely faster just due to slightly more efficient flattened index logic.

[jkotas]

Thank you!

[Sergio0694]

This is only used when the element type is a reference type.
Idea came up while chatting with @jakobbotsch on Discord about array covariance being bad 😛
Also cc. @GrabYourPitchforks as this seems like his area as well.
I'm guessing we could also add a managed fast path for SetValue by using CastHelpers, but haven't done that yet.
Do we want to explore that in this same PR or should I publish this and then we can eventually do that separately?

EDIT: seems like the logic for SetValue is (at least in part) in CastHelpers.StelemRef, but that API is... Private? 🤔
Not sure I understand how those are meant to be used, I'm sure there's a reason why they're all private...

API additions (CoreCLR-only)

namespaceSystem.Runtime.CompilerServices{publicstaticclassRuntimeHelpers{internalstaticunsafeobject?Box(MethodTable*methodTable,refbytedata);}internalstructMethodTable{publicboolIsValueType{get;}publicTypeHandleGetArrayElementTypeHandle();}internalstructTypeHandle{publicboolIsTypeDesc{get;}publicMethodTable*AsMethodTable();}}

[ghost]

I couldn't figure out the best area label to add to this PR. If you have write-permissions please help me learn by adding exactly one area label.

[ghost]

Tagging subscribers to this area: @dotnet/area-system-runtime
See info in area-owners.md if you want to be subscribed.

Issue Details

---

This is only used when the element type is a reference type.
Idea came up while chatting with @jakobbotsch on Discord about array covariance being bad 😛
Also cc. @GrabYourPitchforks as this seems like his area as well.
I'm guessing we could also add a managed fast path for SetValue by using CastHelpers, but haven't done that yet.
Do we want to explore that in this same PR or should I publish this and then we can eventually do that separately?

EDIT: seems like the logic for SetValue is (at least in part) in CastHelpers.StelemRef, but that API is... Private? 🤔
Not sure I understand how those are meant to be used, I'm sure there's a reason why they're all private...

Author:	Sergio0694
Assignees:	-
Labels:	area-System.Runtime, community-contribution
Milestone:	-

[Sergio0694]

Sure! Currently reading docs to get started with the performance repo, will share some numbers as soon as I have them 🙂

[Sergio0694]

Benchmark results:

| Method | Job | Toolchain | Mean | Error | StdDev | Ratio | RatioSD | Code Size ||---------------- |----------- |---------------------------- |----------:|----------:|----------:|------:|--------:|----------:|| GetValue_Object | Job-NBCGUL | \Core_Root\corerun.exe | 1.295 ns | 0.0818 ns | 0.1035 ns | 0.23 | 0.02 | 188 B || GetValue_Object | Job-IEBNQH | \Core_Root_base\corerun.exe | 5.748 ns | 0.1714 ns | 0.2669 ns | 1.00 | 0.00 | 231 B |||||||||||| GetValueInt | Job-NBCGUL | \Core_Root\corerun.exe | 44.603 ns | 0.8945 ns | 0.8367 ns | 0.96 | 0.02 | 188 B || GetValueInt | Job-IEBNQH | \Core_Root_base\corerun.exe | 46.517 ns | 0.5497 ns | 0.5142 ns | 1.00 | 0.00 | 231 B |

For:

[GroupBenchmarksBy(BenchmarkLogicalGroupRule.ByCategory)]publicclassArrayGetValueBenchmark{privatereadonlyArrayarrayOfStrings;privatereadonlyArrayarrayOfInts;publicArrayGetValueBenchmark(){arrayOfStrings=new[]{"Hello","World","Bob","Alice","Fred","Tanner","Egor","Single"};arrayOfInts=Enumerable.Range(0,20).ToArray();}[Benchmark][BenchmarkCategory("REF_TYPE")]publicobject?GetValue_Object(){returnarrayOfStrings.GetValue(4);}[Benchmark][BenchmarkCategory("VALUE_TYPE")]publicobject?GetValueInt(){returnarrayOfInts.GetValue(4);}}

Using the benchmark script kindly provided by @EgorBohere.

Results seem in line with what was expected:

• Reference type is much faster due to no internal call and pretty much all inlined.
• Value type case is almost the same or barely faster just due to slightly more efficient flattened index logic.

[jkotas]

Thank you!

[Sergio0694]

This is only used when the element type is a reference type.
Idea came up while chatting with @jakobbotsch on Discord about array covariance being bad 😛
Also cc. @GrabYourPitchforks as this seems like his area as well.
I'm guessing we could also add a managed fast path for SetValue by using CastHelpers, but haven't done that yet.
Do we want to explore that in this same PR or should I publish this and then we can eventually do that separately?

EDIT: seems like the logic for SetValue is (at least in part) in CastHelpers.StelemRef, but that API is... Private? 🤔
Not sure I understand how those are meant to be used, I'm sure there's a reason why they're all private...

API additions (CoreCLR-only)

namespaceSystem.Runtime.CompilerServices{publicstaticclassRuntimeHelpers{internalstaticunsafeobject?Box(MethodTable*methodTable,refbytedata);}internalstructMethodTable{publicboolIsValueType{get;}publicTypeHandleGetArrayElementTypeHandle();}internalstructTypeHandle{publicboolIsTypeDesc{get;}publicMethodTable*AsMethodTable();}}

[ghost]

I couldn't figure out the best area label to add to this PR. If you have write-permissions please help me learn by adding exactly one area label.

[ghost]

Tagging subscribers to this area: @dotnet/area-system-runtime
See info in area-owners.md if you want to be subscribed.

Issue Details

---

This is only used when the element type is a reference type.
Idea came up while chatting with @jakobbotsch on Discord about array covariance being bad 😛
Also cc. @GrabYourPitchforks as this seems like his area as well.
I'm guessing we could also add a managed fast path for SetValue by using CastHelpers, but haven't done that yet.
Do we want to explore that in this same PR or should I publish this and then we can eventually do that separately?

EDIT: seems like the logic for SetValue is (at least in part) in CastHelpers.StelemRef, but that API is... Private? 🤔
Not sure I understand how those are meant to be used, I'm sure there's a reason why they're all private...

Author:	Sergio0694
Assignees:	-
Labels:	area-System.Runtime, community-contribution
Milestone:	-

[Sergio0694]

Sure! Currently reading docs to get started with the performance repo, will share some numbers as soon as I have them 🙂

[Sergio0694]

Benchmark results:

| Method | Job | Toolchain | Mean | Error | StdDev | Ratio | RatioSD | Code Size ||---------------- |----------- |---------------------------- |----------:|----------:|----------:|------:|--------:|----------:|| GetValue_Object | Job-NBCGUL | \Core_Root\corerun.exe | 1.295 ns | 0.0818 ns | 0.1035 ns | 0.23 | 0.02 | 188 B || GetValue_Object | Job-IEBNQH | \Core_Root_base\corerun.exe | 5.748 ns | 0.1714 ns | 0.2669 ns | 1.00 | 0.00 | 231 B |||||||||||| GetValueInt | Job-NBCGUL | \Core_Root\corerun.exe | 44.603 ns | 0.8945 ns | 0.8367 ns | 0.96 | 0.02 | 188 B || GetValueInt | Job-IEBNQH | \Core_Root_base\corerun.exe | 46.517 ns | 0.5497 ns | 0.5142 ns | 1.00 | 0.00 | 231 B |

For:

[GroupBenchmarksBy(BenchmarkLogicalGroupRule.ByCategory)]publicclassArrayGetValueBenchmark{privatereadonlyArrayarrayOfStrings;privatereadonlyArrayarrayOfInts;publicArrayGetValueBenchmark(){arrayOfStrings=new[]{"Hello","World","Bob","Alice","Fred","Tanner","Egor","Single"};arrayOfInts=Enumerable.Range(0,20).ToArray();}[Benchmark][BenchmarkCategory("REF_TYPE")]publicobject?GetValue_Object(){returnarrayOfStrings.GetValue(4);}[Benchmark][BenchmarkCategory("VALUE_TYPE")]publicobject?GetValueInt(){returnarrayOfInts.GetValue(4);}}

Using the benchmark script kindly provided by @EgorBohere.

Results seem in line with what was expected:

• Reference type is much faster due to no internal call and pretty much all inlined.
• Value type case is almost the same or barely faster just due to slightly more efficient flattened index logic.

[jkotas]

Thank you!

[Sergio0694]

This is only used when the element type is a reference type.
Idea came up while chatting with @jakobbotsch on Discord about array covariance being bad 😛
Also cc. @GrabYourPitchforks as this seems like his area as well.
I'm guessing we could also add a managed fast path for SetValue by using CastHelpers, but haven't done that yet.
Do we want to explore that in this same PR or should I publish this and then we can eventually do that separately?

EDIT: seems like the logic for SetValue is (at least in part) in CastHelpers.StelemRef, but that API is... Private? 🤔
Not sure I understand how those are meant to be used, I'm sure there's a reason why they're all private...

API additions (CoreCLR-only)

namespaceSystem.Runtime.CompilerServices{publicstaticclassRuntimeHelpers{internalstaticunsafeobject?Box(MethodTable*methodTable,refbytedata);}internalstructMethodTable{publicboolIsValueType{get;}publicTypeHandleGetArrayElementTypeHandle();}internalstructTypeHandle{publicboolIsTypeDesc{get;}publicMethodTable*AsMethodTable();}}

[ghost]

I couldn't figure out the best area label to add to this PR. If you have write-permissions please help me learn by adding exactly one area label.

[ghost]

Tagging subscribers to this area: @dotnet/area-system-runtime
See info in area-owners.md if you want to be subscribed.

Issue Details

---

This is only used when the element type is a reference type.
Idea came up while chatting with @jakobbotsch on Discord about array covariance being bad 😛
Also cc. @GrabYourPitchforks as this seems like his area as well.
I'm guessing we could also add a managed fast path for SetValue by using CastHelpers, but haven't done that yet.
Do we want to explore that in this same PR or should I publish this and then we can eventually do that separately?

EDIT: seems like the logic for SetValue is (at least in part) in CastHelpers.StelemRef, but that API is... Private? 🤔
Not sure I understand how those are meant to be used, I'm sure there's a reason why they're all private...

Author:	Sergio0694
Assignees:	-
Labels:	area-System.Runtime, community-contribution
Milestone:	-

[Sergio0694]

Sure! Currently reading docs to get started with the performance repo, will share some numbers as soon as I have them 🙂

[Sergio0694]

Benchmark results:

| Method | Job | Toolchain | Mean | Error | StdDev | Ratio | RatioSD | Code Size ||---------------- |----------- |---------------------------- |----------:|----------:|----------:|------:|--------:|----------:|| GetValue_Object | Job-NBCGUL | \Core_Root\corerun.exe | 1.295 ns | 0.0818 ns | 0.1035 ns | 0.23 | 0.02 | 188 B || GetValue_Object | Job-IEBNQH | \Core_Root_base\corerun.exe | 5.748 ns | 0.1714 ns | 0.2669 ns | 1.00 | 0.00 | 231 B |||||||||||| GetValueInt | Job-NBCGUL | \Core_Root\corerun.exe | 44.603 ns | 0.8945 ns | 0.8367 ns | 0.96 | 0.02 | 188 B || GetValueInt | Job-IEBNQH | \Core_Root_base\corerun.exe | 46.517 ns | 0.5497 ns | 0.5142 ns | 1.00 | 0.00 | 231 B |

For:

[GroupBenchmarksBy(BenchmarkLogicalGroupRule.ByCategory)]publicclassArrayGetValueBenchmark{privatereadonlyArrayarrayOfStrings;privatereadonlyArrayarrayOfInts;publicArrayGetValueBenchmark(){arrayOfStrings=new[]{"Hello","World","Bob","Alice","Fred","Tanner","Egor","Single"};arrayOfInts=Enumerable.Range(0,20).ToArray();}[Benchmark][BenchmarkCategory("REF_TYPE")]publicobject?GetValue_Object(){returnarrayOfStrings.GetValue(4);}[Benchmark][BenchmarkCategory("VALUE_TYPE")]publicobject?GetValueInt(){returnarrayOfInts.GetValue(4);}}

Using the benchmark script kindly provided by @EgorBohere.

Results seem in line with what was expected:

• Reference type is much faster due to no internal call and pretty much all inlined.
• Value type case is almost the same or barely faster just due to slightly more efficient flattened index logic.

[jkotas]

Thank you!

[Sergio0694]

This is only used when the element type is a reference type.
Idea came up while chatting with @jakobbotsch on Discord about array covariance being bad 😛
Also cc. @GrabYourPitchforks as this seems like his area as well.
I'm guessing we could also add a managed fast path for SetValue by using CastHelpers, but haven't done that yet.
Do we want to explore that in this same PR or should I publish this and then we can eventually do that separately?

EDIT: seems like the logic for SetValue is (at least in part) in CastHelpers.StelemRef, but that API is... Private? 🤔
Not sure I understand how those are meant to be used, I'm sure there's a reason why they're all private...

API additions (CoreCLR-only)

namespaceSystem.Runtime.CompilerServices{publicstaticclassRuntimeHelpers{internalstaticunsafeobject?Box(MethodTable*methodTable,refbytedata);}internalstructMethodTable{publicboolIsValueType{get;}publicTypeHandleGetArrayElementTypeHandle();}internalstructTypeHandle{publicboolIsTypeDesc{get;}publicMethodTable*AsMethodTable();}}

[ghost]

I couldn't figure out the best area label to add to this PR. If you have write-permissions please help me learn by adding exactly one area label.

[ghost]

Tagging subscribers to this area: @dotnet/area-system-runtime
See info in area-owners.md if you want to be subscribed.

Issue Details

---

This is only used when the element type is a reference type.
Idea came up while chatting with @jakobbotsch on Discord about array covariance being bad 😛
Also cc. @GrabYourPitchforks as this seems like his area as well.
I'm guessing we could also add a managed fast path for SetValue by using CastHelpers, but haven't done that yet.
Do we want to explore that in this same PR or should I publish this and then we can eventually do that separately?

EDIT: seems like the logic for SetValue is (at least in part) in CastHelpers.StelemRef, but that API is... Private? 🤔
Not sure I understand how those are meant to be used, I'm sure there's a reason why they're all private...

Author:	Sergio0694
Assignees:	-
Labels:	area-System.Runtime, community-contribution
Milestone:	-

[Sergio0694]

Sure! Currently reading docs to get started with the performance repo, will share some numbers as soon as I have them 🙂

[Sergio0694]

Benchmark results:

| Method | Job | Toolchain | Mean | Error | StdDev | Ratio | RatioSD | Code Size ||---------------- |----------- |---------------------------- |----------:|----------:|----------:|------:|--------:|----------:|| GetValue_Object | Job-NBCGUL | \Core_Root\corerun.exe | 1.295 ns | 0.0818 ns | 0.1035 ns | 0.23 | 0.02 | 188 B || GetValue_Object | Job-IEBNQH | \Core_Root_base\corerun.exe | 5.748 ns | 0.1714 ns | 0.2669 ns | 1.00 | 0.00 | 231 B |||||||||||| GetValueInt | Job-NBCGUL | \Core_Root\corerun.exe | 44.603 ns | 0.8945 ns | 0.8367 ns | 0.96 | 0.02 | 188 B || GetValueInt | Job-IEBNQH | \Core_Root_base\corerun.exe | 46.517 ns | 0.5497 ns | 0.5142 ns | 1.00 | 0.00 | 231 B |

For:

[GroupBenchmarksBy(BenchmarkLogicalGroupRule.ByCategory)]publicclassArrayGetValueBenchmark{privatereadonlyArrayarrayOfStrings;privatereadonlyArrayarrayOfInts;publicArrayGetValueBenchmark(){arrayOfStrings=new[]{"Hello","World","Bob","Alice","Fred","Tanner","Egor","Single"};arrayOfInts=Enumerable.Range(0,20).ToArray();}[Benchmark][BenchmarkCategory("REF_TYPE")]publicobject?GetValue_Object(){returnarrayOfStrings.GetValue(4);}[Benchmark][BenchmarkCategory("VALUE_TYPE")]publicobject?GetValueInt(){returnarrayOfInts.GetValue(4);}}

Using the benchmark script kindly provided by @EgorBohere.

Results seem in line with what was expected:

• Reference type is much faster due to no internal call and pretty much all inlined.
• Value type case is almost the same or barely faster just due to slightly more efficient flattened index logic.

[jkotas]

Thank you!

[Sergio0694]

This is only used when the element type is a reference type.
Idea came up while chatting with @jakobbotsch on Discord about array covariance being bad 😛
Also cc. @GrabYourPitchforks as this seems like his area as well.
I'm guessing we could also add a managed fast path for SetValue by using CastHelpers, but haven't done that yet.
Do we want to explore that in this same PR or should I publish this and then we can eventually do that separately?

EDIT: seems like the logic for SetValue is (at least in part) in CastHelpers.StelemRef, but that API is... Private? 🤔
Not sure I understand how those are meant to be used, I'm sure there's a reason why they're all private...

API additions (CoreCLR-only)

namespaceSystem.Runtime.CompilerServices{publicstaticclassRuntimeHelpers{internalstaticunsafeobject?Box(MethodTable*methodTable,refbytedata);}internalstructMethodTable{publicboolIsValueType{get;}publicTypeHandleGetArrayElementTypeHandle();}internalstructTypeHandle{publicboolIsTypeDesc{get;}publicMethodTable*AsMethodTable();}}

[ghost]

I couldn't figure out the best area label to add to this PR. If you have write-permissions please help me learn by adding exactly one area label.

[ghost]

Tagging subscribers to this area: @dotnet/area-system-runtime
See info in area-owners.md if you want to be subscribed.

Issue Details

---

This is only used when the element type is a reference type.
Idea came up while chatting with @jakobbotsch on Discord about array covariance being bad 😛
Also cc. @GrabYourPitchforks as this seems like his area as well.
I'm guessing we could also add a managed fast path for SetValue by using CastHelpers, but haven't done that yet.
Do we want to explore that in this same PR or should I publish this and then we can eventually do that separately?

EDIT: seems like the logic for SetValue is (at least in part) in CastHelpers.StelemRef, but that API is... Private? 🤔
Not sure I understand how those are meant to be used, I'm sure there's a reason why they're all private...

Author:	Sergio0694
Assignees:	-
Labels:	area-System.Runtime, community-contribution
Milestone:	-

[Sergio0694]

Sure! Currently reading docs to get started with the performance repo, will share some numbers as soon as I have them 🙂

[Sergio0694]

Benchmark results:

| Method | Job | Toolchain | Mean | Error | StdDev | Ratio | RatioSD | Code Size ||---------------- |----------- |---------------------------- |----------:|----------:|----------:|------:|--------:|----------:|| GetValue_Object | Job-NBCGUL | \Core_Root\corerun.exe | 1.295 ns | 0.0818 ns | 0.1035 ns | 0.23 | 0.02 | 188 B || GetValue_Object | Job-IEBNQH | \Core_Root_base\corerun.exe | 5.748 ns | 0.1714 ns | 0.2669 ns | 1.00 | 0.00 | 231 B |||||||||||| GetValueInt | Job-NBCGUL | \Core_Root\corerun.exe | 44.603 ns | 0.8945 ns | 0.8367 ns | 0.96 | 0.02 | 188 B || GetValueInt | Job-IEBNQH | \Core_Root_base\corerun.exe | 46.517 ns | 0.5497 ns | 0.5142 ns | 1.00 | 0.00 | 231 B |

For:

[GroupBenchmarksBy(BenchmarkLogicalGroupRule.ByCategory)]publicclassArrayGetValueBenchmark{privatereadonlyArrayarrayOfStrings;privatereadonlyArrayarrayOfInts;publicArrayGetValueBenchmark(){arrayOfStrings=new[]{"Hello","World","Bob","Alice","Fred","Tanner","Egor","Single"};arrayOfInts=Enumerable.Range(0,20).ToArray();}[Benchmark][BenchmarkCategory("REF_TYPE")]publicobject?GetValue_Object(){returnarrayOfStrings.GetValue(4);}[Benchmark][BenchmarkCategory("VALUE_TYPE")]publicobject?GetValueInt(){returnarrayOfInts.GetValue(4);}}

Using the benchmark script kindly provided by @EgorBohere.

Results seem in line with what was expected:

• Reference type is much faster due to no internal call and pretty much all inlined.
• Value type case is almost the same or barely faster just due to slightly more efficient flattened index logic.

[jkotas]

Thank you!

[Sergio0694]

This is only used when the element type is a reference type.
Idea came up while chatting with @jakobbotsch on Discord about array covariance being bad 😛
Also cc. @GrabYourPitchforks as this seems like his area as well.
I'm guessing we could also add a managed fast path for SetValue by using CastHelpers, but haven't done that yet.
Do we want to explore that in this same PR or should I publish this and then we can eventually do that separately?

EDIT: seems like the logic for SetValue is (at least in part) in CastHelpers.StelemRef, but that API is... Private? 🤔
Not sure I understand how those are meant to be used, I'm sure there's a reason why they're all private...

API additions (CoreCLR-only)

namespaceSystem.Runtime.CompilerServices{publicstaticclassRuntimeHelpers{internalstaticunsafeobject?Box(MethodTable*methodTable,refbytedata);}internalstructMethodTable{publicboolIsValueType{get;}publicTypeHandleGetArrayElementTypeHandle();}internalstructTypeHandle{publicboolIsTypeDesc{get;}publicMethodTable*AsMethodTable();}}

[ghost]

I couldn't figure out the best area label to add to this PR. If you have write-permissions please help me learn by adding exactly one area label.

[ghost]

Tagging subscribers to this area: @dotnet/area-system-runtime
See info in area-owners.md if you want to be subscribed.

Issue Details

---

This is only used when the element type is a reference type.
Idea came up while chatting with @jakobbotsch on Discord about array covariance being bad 😛
Also cc. @GrabYourPitchforks as this seems like his area as well.
I'm guessing we could also add a managed fast path for SetValue by using CastHelpers, but haven't done that yet.
Do we want to explore that in this same PR or should I publish this and then we can eventually do that separately?

EDIT: seems like the logic for SetValue is (at least in part) in CastHelpers.StelemRef, but that API is... Private? 🤔
Not sure I understand how those are meant to be used, I'm sure there's a reason why they're all private...

Author:	Sergio0694
Assignees:	-
Labels:	area-System.Runtime, community-contribution
Milestone:	-

[Sergio0694]

Sure! Currently reading docs to get started with the performance repo, will share some numbers as soon as I have them 🙂

[Sergio0694]

Benchmark results:

| Method | Job | Toolchain | Mean | Error | StdDev | Ratio | RatioSD | Code Size ||---------------- |----------- |---------------------------- |----------:|----------:|----------:|------:|--------:|----------:|| GetValue_Object | Job-NBCGUL | \Core_Root\corerun.exe | 1.295 ns | 0.0818 ns | 0.1035 ns | 0.23 | 0.02 | 188 B || GetValue_Object | Job-IEBNQH | \Core_Root_base\corerun.exe | 5.748 ns | 0.1714 ns | 0.2669 ns | 1.00 | 0.00 | 231 B |||||||||||| GetValueInt | Job-NBCGUL | \Core_Root\corerun.exe | 44.603 ns | 0.8945 ns | 0.8367 ns | 0.96 | 0.02 | 188 B || GetValueInt | Job-IEBNQH | \Core_Root_base\corerun.exe | 46.517 ns | 0.5497 ns | 0.5142 ns | 1.00 | 0.00 | 231 B |

For:

[GroupBenchmarksBy(BenchmarkLogicalGroupRule.ByCategory)]publicclassArrayGetValueBenchmark{privatereadonlyArrayarrayOfStrings;privatereadonlyArrayarrayOfInts;publicArrayGetValueBenchmark(){arrayOfStrings=new[]{"Hello","World","Bob","Alice","Fred","Tanner","Egor","Single"};arrayOfInts=Enumerable.Range(0,20).ToArray();}[Benchmark][BenchmarkCategory("REF_TYPE")]publicobject?GetValue_Object(){returnarrayOfStrings.GetValue(4);}[Benchmark][BenchmarkCategory("VALUE_TYPE")]publicobject?GetValueInt(){returnarrayOfInts.GetValue(4);}}

Using the benchmark script kindly provided by @EgorBohere.

Results seem in line with what was expected:

• Reference type is much faster due to no internal call and pretty much all inlined.
• Value type case is almost the same or barely faster just due to slightly more efficient flattened index logic.

[jkotas]

Thank you!