Move the cache for the marshaler methods to be stored on the RuntimeType instead of in a ConditionalWeakTable by jkoritzinsky · Pull Request #126621 · dotnet/runtime

jkoritzinsky · 2026-04-07T23:32:12Z

Move the marshalling info to be cached on the type itself instead of in a ConditionalWeakTable.

Move some of the validation (in particular the "has layout" validation) to be done only at "create" time so we don't have to rerun it for every time someone calls the Marshal methods with the same (valid) type.

Use the blittable fallbacks on StructureMarshaler<T> and LayoutClassMarshaler<T> instead of inlining the calls. The cost may go up slightly (delegate call to a method that calls SpanHelpers.Memmove instead of direct call to SpanHelpers.Memmove) but still much faster than the existing implementation that does the "is blittable" checks each time.

…ype instead of in a ConditionalWeakTable Should fixdotnet#126608

jkoritzinsky · 2026-04-07T23:32:46Z

@MihuBot benchmark Interop.StructureToPtr

dotnet-policy-service · 2026-04-07T23:33:33Z

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

Copilot

Pull request overview

This PR moves the cache for Marshal.LayoutTypeMarshalerMethods from a ConditionalWeakTable<Type, …> to the per-RuntimeType generic cache infrastructure to address the reported interop performance regressions (issue #126608).

Changes:

Adds LayoutTypeMarshalerMethods storage to RuntimeType.CompositeCacheEntry so it can live in the existing RuntimeType generic cache.
Refactors LayoutTypeMarshalerMethods to implement RuntimeType.IGenericCacheEntry<T> and retrieves it via RuntimeType.GetOrCreateCacheEntry<T>().
Switches from delegate creation (CreateDelegate) to caching raw function pointers (MethodHandle.GetFunctionPointer) for the marshaler methods.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
`src/coreclr/System.Private.CoreLib/src/System/RuntimeType.GenericCache.cs`	Adds a new composite-cache field to hold the marshaler-methods cache on `RuntimeType`.
`src/coreclr/System.Private.CoreLib/src/System/Runtime/InteropServices/Marshal.CoreCLR.cs`	Reworks `LayoutTypeMarshalerMethods` to use `RuntimeType` cache entries and cached function pointers instead of a `ConditionalWeakTable`.

MihuBot · 2026-04-07T23:51:28Z

Interop.StructureToPtr

BenchmarkDotNet v0.16.0-nightly.20260320.467, Linux Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74 2.60GHz, 1 CPU, 8 logical and 4 physical cores
Memory: 31.34 GB Total, 2.07 GB Available
Job-TPEJOW : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
Job-HKHXHK : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
EvaluateOverhead=False OutlierMode=Default PowerPlanMode=
IterationTime=250ms MaxIterationCount=20 MemoryRandomization=Default
MinIterationCount=15 WarmupCount=1

Method	Toolchain	Mean	Error	Ratio	Allocated	Alloc Ratio
MarshalStructureToPtr	Main	53.46 ns	0.045 ns	1.00	24 B	1.00
MarshalStructureToPtr	PR	41.83 ns	0.073 ns	0.78	24 B	1.00

MarshalPtrToStructure	Main	57.27 ns	0.111 ns	1.00	24 B	1.00
MarshalPtrToStructure	PR	43.72 ns	0.169 ns	0.76	24 B	1.00

MarshalDestroyStructure	Main	109.19 ns	0.141 ns	1.00	96 B	1.00
MarshalDestroyStructure	PR	87.02 ns	0.241 ns	0.80	96 B	1.00

…ayout info for the type.

jkoritzinsky · 2026-04-08T18:00:31Z

@MihuBot benchmark Interop.StructureToPtr

MihuBot · 2026-04-08T18:17:49Z

Interop.StructureToPtr

BenchmarkDotNet v0.16.0-nightly.20260320.467, Linux Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74 2.60GHz, 1 CPU, 8 logical and 4 physical cores
Memory: 31.34 GB Total, 1.78 GB Available
Job-TPEJOW : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
Job-HKHXHK : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
EvaluateOverhead=False OutlierMode=Default PowerPlanMode=
IterationTime=250ms MaxIterationCount=20 MemoryRandomization=Default
MinIterationCount=15 WarmupCount=1

Method	Toolchain	Mean	Error	Ratio	Allocated	Alloc Ratio
MarshalStructureToPtr	Main	64.20 ns	0.578 ns	1.00	24 B	1.00
MarshalStructureToPtr	PR	30.70 ns	0.567 ns	0.48	24 B	1.00

MarshalPtrToStructure	Main	62.55 ns	0.117 ns	1.00	24 B	1.00
MarshalPtrToStructure	PR	37.14 ns	0.026 ns	0.59	24 B	1.00

MarshalDestroyStructure	Main	109.97 ns	0.152 ns	1.00	96 B	1.00
MarshalDestroyStructure	PR	68.63 ns	0.158 ns	0.62	96 B	1.00

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

jkoritzinsky · 2026-04-07T23:32:12Z

Fixes#126608

Move the marshalling info to be cached on the type itself instead of in a ConditionalWeakTable.

Move some of the validation (in particular the "has layout" validation) to be done only at "create" time so we don't have to rerun it for every time someone calls the Marshal methods with the same (valid) type.

Use the blittable fallbacks on StructureMarshaler<T> and LayoutClassMarshaler<T> instead of inlining the calls. The cost may go up slightly (delegate call to a method that calls SpanHelpers.Memmove instead of direct call to SpanHelpers.Memmove) but still much faster than the existing implementation that does the "is blittable" checks each time.

…ype instead of in a ConditionalWeakTable Should fixdotnet#126608

jkoritzinsky · 2026-04-07T23:32:46Z

@MihuBot benchmark Interop.StructureToPtr

dotnet-policy-service · 2026-04-07T23:33:33Z

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

Copilot

Pull request overview

This PR moves the cache for Marshal.LayoutTypeMarshalerMethods from a ConditionalWeakTable<Type, …> to the per-RuntimeType generic cache infrastructure to address the reported interop performance regressions (issue #126608).

Changes:

Adds LayoutTypeMarshalerMethods storage to RuntimeType.CompositeCacheEntry so it can live in the existing RuntimeType generic cache.
Refactors LayoutTypeMarshalerMethods to implement RuntimeType.IGenericCacheEntry<T> and retrieves it via RuntimeType.GetOrCreateCacheEntry<T>().
Switches from delegate creation (CreateDelegate) to caching raw function pointers (MethodHandle.GetFunctionPointer) for the marshaler methods.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
`src/coreclr/System.Private.CoreLib/src/System/RuntimeType.GenericCache.cs`	Adds a new composite-cache field to hold the marshaler-methods cache on `RuntimeType`.
`src/coreclr/System.Private.CoreLib/src/System/Runtime/InteropServices/Marshal.CoreCLR.cs`	Reworks `LayoutTypeMarshalerMethods` to use `RuntimeType` cache entries and cached function pointers instead of a `ConditionalWeakTable`.

MihuBot · 2026-04-07T23:51:28Z

Interop.StructureToPtr

BenchmarkDotNet v0.16.0-nightly.20260320.467, Linux Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74 2.60GHz, 1 CPU, 8 logical and 4 physical cores
Memory: 31.34 GB Total, 2.07 GB Available
Job-TPEJOW : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
Job-HKHXHK : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
EvaluateOverhead=False OutlierMode=Default PowerPlanMode=
IterationTime=250ms MaxIterationCount=20 MemoryRandomization=Default
MinIterationCount=15 WarmupCount=1

Method	Toolchain	Mean	Error	Ratio	Allocated	Alloc Ratio
MarshalStructureToPtr	Main	53.46 ns	0.045 ns	1.00	24 B	1.00
MarshalStructureToPtr	PR	41.83 ns	0.073 ns	0.78	24 B	1.00

MarshalPtrToStructure	Main	57.27 ns	0.111 ns	1.00	24 B	1.00
MarshalPtrToStructure	PR	43.72 ns	0.169 ns	0.76	24 B	1.00

MarshalDestroyStructure	Main	109.19 ns	0.141 ns	1.00	96 B	1.00
MarshalDestroyStructure	PR	87.02 ns	0.241 ns	0.80	96 B	1.00

…ayout info for the type.

jkoritzinsky · 2026-04-08T18:00:31Z

@MihuBot benchmark Interop.StructureToPtr

MihuBot · 2026-04-08T18:17:49Z

Interop.StructureToPtr

BenchmarkDotNet v0.16.0-nightly.20260320.467, Linux Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74 2.60GHz, 1 CPU, 8 logical and 4 physical cores
Memory: 31.34 GB Total, 1.78 GB Available
Job-TPEJOW : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
Job-HKHXHK : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
EvaluateOverhead=False OutlierMode=Default PowerPlanMode=
IterationTime=250ms MaxIterationCount=20 MemoryRandomization=Default
MinIterationCount=15 WarmupCount=1

Method	Toolchain	Mean	Error	Ratio	Allocated	Alloc Ratio
MarshalStructureToPtr	Main	64.20 ns	0.578 ns	1.00	24 B	1.00
MarshalStructureToPtr	PR	30.70 ns	0.567 ns	0.48	24 B	1.00

MarshalPtrToStructure	Main	62.55 ns	0.117 ns	1.00	24 B	1.00
MarshalPtrToStructure	PR	37.14 ns	0.026 ns	0.59	24 B	1.00

MarshalDestroyStructure	Main	109.97 ns	0.152 ns	1.00	96 B	1.00
MarshalDestroyStructure	PR	68.63 ns	0.158 ns	0.62	96 B	1.00

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

jkoritzinsky · 2026-04-07T23:32:12Z

Fixes#126608

Move the marshalling info to be cached on the type itself instead of in a ConditionalWeakTable.

Move some of the validation (in particular the "has layout" validation) to be done only at "create" time so we don't have to rerun it for every time someone calls the Marshal methods with the same (valid) type.

Use the blittable fallbacks on StructureMarshaler<T> and LayoutClassMarshaler<T> instead of inlining the calls. The cost may go up slightly (delegate call to a method that calls SpanHelpers.Memmove instead of direct call to SpanHelpers.Memmove) but still much faster than the existing implementation that does the "is blittable" checks each time.

…ype instead of in a ConditionalWeakTable Should fixdotnet#126608

jkoritzinsky · 2026-04-07T23:32:46Z

@MihuBot benchmark Interop.StructureToPtr

dotnet-policy-service · 2026-04-07T23:33:33Z

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

Copilot

Pull request overview

This PR moves the cache for Marshal.LayoutTypeMarshalerMethods from a ConditionalWeakTable<Type, …> to the per-RuntimeType generic cache infrastructure to address the reported interop performance regressions (issue #126608).

Changes:

Adds LayoutTypeMarshalerMethods storage to RuntimeType.CompositeCacheEntry so it can live in the existing RuntimeType generic cache.
Refactors LayoutTypeMarshalerMethods to implement RuntimeType.IGenericCacheEntry<T> and retrieves it via RuntimeType.GetOrCreateCacheEntry<T>().
Switches from delegate creation (CreateDelegate) to caching raw function pointers (MethodHandle.GetFunctionPointer) for the marshaler methods.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
`src/coreclr/System.Private.CoreLib/src/System/RuntimeType.GenericCache.cs`	Adds a new composite-cache field to hold the marshaler-methods cache on `RuntimeType`.
`src/coreclr/System.Private.CoreLib/src/System/Runtime/InteropServices/Marshal.CoreCLR.cs`	Reworks `LayoutTypeMarshalerMethods` to use `RuntimeType` cache entries and cached function pointers instead of a `ConditionalWeakTable`.

MihuBot · 2026-04-07T23:51:28Z

Interop.StructureToPtr

BenchmarkDotNet v0.16.0-nightly.20260320.467, Linux Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74 2.60GHz, 1 CPU, 8 logical and 4 physical cores
Memory: 31.34 GB Total, 2.07 GB Available
Job-TPEJOW : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
Job-HKHXHK : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
EvaluateOverhead=False OutlierMode=Default PowerPlanMode=
IterationTime=250ms MaxIterationCount=20 MemoryRandomization=Default
MinIterationCount=15 WarmupCount=1

Method	Toolchain	Mean	Error	Ratio	Allocated	Alloc Ratio
MarshalStructureToPtr	Main	53.46 ns	0.045 ns	1.00	24 B	1.00
MarshalStructureToPtr	PR	41.83 ns	0.073 ns	0.78	24 B	1.00

MarshalPtrToStructure	Main	57.27 ns	0.111 ns	1.00	24 B	1.00
MarshalPtrToStructure	PR	43.72 ns	0.169 ns	0.76	24 B	1.00

MarshalDestroyStructure	Main	109.19 ns	0.141 ns	1.00	96 B	1.00
MarshalDestroyStructure	PR	87.02 ns	0.241 ns	0.80	96 B	1.00

…ayout info for the type.

jkoritzinsky · 2026-04-08T18:00:31Z

@MihuBot benchmark Interop.StructureToPtr

MihuBot · 2026-04-08T18:17:49Z

Interop.StructureToPtr

BenchmarkDotNet v0.16.0-nightly.20260320.467, Linux Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74 2.60GHz, 1 CPU, 8 logical and 4 physical cores
Memory: 31.34 GB Total, 1.78 GB Available
Job-TPEJOW : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
Job-HKHXHK : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
EvaluateOverhead=False OutlierMode=Default PowerPlanMode=
IterationTime=250ms MaxIterationCount=20 MemoryRandomization=Default
MinIterationCount=15 WarmupCount=1

Method	Toolchain	Mean	Error	Ratio	Allocated	Alloc Ratio
MarshalStructureToPtr	Main	64.20 ns	0.578 ns	1.00	24 B	1.00
MarshalStructureToPtr	PR	30.70 ns	0.567 ns	0.48	24 B	1.00

MarshalPtrToStructure	Main	62.55 ns	0.117 ns	1.00	24 B	1.00
MarshalPtrToStructure	PR	37.14 ns	0.026 ns	0.59	24 B	1.00

MarshalDestroyStructure	Main	109.97 ns	0.152 ns	1.00	96 B	1.00
MarshalDestroyStructure	PR	68.63 ns	0.158 ns	0.62	96 B	1.00

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

jkoritzinsky · 2026-04-07T23:32:12Z

Fixes#126608

Move the marshalling info to be cached on the type itself instead of in a ConditionalWeakTable.

Move some of the validation (in particular the "has layout" validation) to be done only at "create" time so we don't have to rerun it for every time someone calls the Marshal methods with the same (valid) type.

Use the blittable fallbacks on StructureMarshaler<T> and LayoutClassMarshaler<T> instead of inlining the calls. The cost may go up slightly (delegate call to a method that calls SpanHelpers.Memmove instead of direct call to SpanHelpers.Memmove) but still much faster than the existing implementation that does the "is blittable" checks each time.

…ype instead of in a ConditionalWeakTable Should fixdotnet#126608

jkoritzinsky · 2026-04-07T23:32:46Z

@MihuBot benchmark Interop.StructureToPtr

dotnet-policy-service · 2026-04-07T23:33:33Z

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

Copilot

Pull request overview

This PR moves the cache for Marshal.LayoutTypeMarshalerMethods from a ConditionalWeakTable<Type, …> to the per-RuntimeType generic cache infrastructure to address the reported interop performance regressions (issue #126608).

Changes:

Adds LayoutTypeMarshalerMethods storage to RuntimeType.CompositeCacheEntry so it can live in the existing RuntimeType generic cache.
Refactors LayoutTypeMarshalerMethods to implement RuntimeType.IGenericCacheEntry<T> and retrieves it via RuntimeType.GetOrCreateCacheEntry<T>().
Switches from delegate creation (CreateDelegate) to caching raw function pointers (MethodHandle.GetFunctionPointer) for the marshaler methods.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
`src/coreclr/System.Private.CoreLib/src/System/RuntimeType.GenericCache.cs`	Adds a new composite-cache field to hold the marshaler-methods cache on `RuntimeType`.
`src/coreclr/System.Private.CoreLib/src/System/Runtime/InteropServices/Marshal.CoreCLR.cs`	Reworks `LayoutTypeMarshalerMethods` to use `RuntimeType` cache entries and cached function pointers instead of a `ConditionalWeakTable`.

MihuBot · 2026-04-07T23:51:28Z

Interop.StructureToPtr

BenchmarkDotNet v0.16.0-nightly.20260320.467, Linux Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74 2.60GHz, 1 CPU, 8 logical and 4 physical cores
Memory: 31.34 GB Total, 2.07 GB Available
Job-TPEJOW : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
Job-HKHXHK : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
EvaluateOverhead=False OutlierMode=Default PowerPlanMode=
IterationTime=250ms MaxIterationCount=20 MemoryRandomization=Default
MinIterationCount=15 WarmupCount=1

Method	Toolchain	Mean	Error	Ratio	Allocated	Alloc Ratio
MarshalStructureToPtr	Main	53.46 ns	0.045 ns	1.00	24 B	1.00
MarshalStructureToPtr	PR	41.83 ns	0.073 ns	0.78	24 B	1.00

MarshalPtrToStructure	Main	57.27 ns	0.111 ns	1.00	24 B	1.00
MarshalPtrToStructure	PR	43.72 ns	0.169 ns	0.76	24 B	1.00

MarshalDestroyStructure	Main	109.19 ns	0.141 ns	1.00	96 B	1.00
MarshalDestroyStructure	PR	87.02 ns	0.241 ns	0.80	96 B	1.00

…ayout info for the type.

jkoritzinsky · 2026-04-08T18:00:31Z

@MihuBot benchmark Interop.StructureToPtr

MihuBot · 2026-04-08T18:17:49Z

Interop.StructureToPtr

BenchmarkDotNet v0.16.0-nightly.20260320.467, Linux Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74 2.60GHz, 1 CPU, 8 logical and 4 physical cores
Memory: 31.34 GB Total, 1.78 GB Available
Job-TPEJOW : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
Job-HKHXHK : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
EvaluateOverhead=False OutlierMode=Default PowerPlanMode=
IterationTime=250ms MaxIterationCount=20 MemoryRandomization=Default
MinIterationCount=15 WarmupCount=1

Method	Toolchain	Mean	Error	Ratio	Allocated	Alloc Ratio
MarshalStructureToPtr	Main	64.20 ns	0.578 ns	1.00	24 B	1.00
MarshalStructureToPtr	PR	30.70 ns	0.567 ns	0.48	24 B	1.00

MarshalPtrToStructure	Main	62.55 ns	0.117 ns	1.00	24 B	1.00
MarshalPtrToStructure	PR	37.14 ns	0.026 ns	0.59	24 B	1.00

MarshalDestroyStructure	Main	109.97 ns	0.152 ns	1.00	96 B	1.00
MarshalDestroyStructure	PR	68.63 ns	0.158 ns	0.62	96 B	1.00

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

jkoritzinsky · 2026-04-07T23:32:12Z

Fixes#126608

Move the marshalling info to be cached on the type itself instead of in a ConditionalWeakTable.

Move some of the validation (in particular the "has layout" validation) to be done only at "create" time so we don't have to rerun it for every time someone calls the Marshal methods with the same (valid) type.

Use the blittable fallbacks on StructureMarshaler<T> and LayoutClassMarshaler<T> instead of inlining the calls. The cost may go up slightly (delegate call to a method that calls SpanHelpers.Memmove instead of direct call to SpanHelpers.Memmove) but still much faster than the existing implementation that does the "is blittable" checks each time.

…ype instead of in a ConditionalWeakTable Should fixdotnet#126608

jkoritzinsky · 2026-04-07T23:32:46Z

@MihuBot benchmark Interop.StructureToPtr

dotnet-policy-service · 2026-04-07T23:33:33Z

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

Copilot

Pull request overview

This PR moves the cache for Marshal.LayoutTypeMarshalerMethods from a ConditionalWeakTable<Type, …> to the per-RuntimeType generic cache infrastructure to address the reported interop performance regressions (issue #126608).

Changes:

Adds LayoutTypeMarshalerMethods storage to RuntimeType.CompositeCacheEntry so it can live in the existing RuntimeType generic cache.
Refactors LayoutTypeMarshalerMethods to implement RuntimeType.IGenericCacheEntry<T> and retrieves it via RuntimeType.GetOrCreateCacheEntry<T>().
Switches from delegate creation (CreateDelegate) to caching raw function pointers (MethodHandle.GetFunctionPointer) for the marshaler methods.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
`src/coreclr/System.Private.CoreLib/src/System/RuntimeType.GenericCache.cs`	Adds a new composite-cache field to hold the marshaler-methods cache on `RuntimeType`.
`src/coreclr/System.Private.CoreLib/src/System/Runtime/InteropServices/Marshal.CoreCLR.cs`	Reworks `LayoutTypeMarshalerMethods` to use `RuntimeType` cache entries and cached function pointers instead of a `ConditionalWeakTable`.

MihuBot · 2026-04-07T23:51:28Z

Interop.StructureToPtr

BenchmarkDotNet v0.16.0-nightly.20260320.467, Linux Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74 2.60GHz, 1 CPU, 8 logical and 4 physical cores
Memory: 31.34 GB Total, 2.07 GB Available
Job-TPEJOW : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
Job-HKHXHK : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
EvaluateOverhead=False OutlierMode=Default PowerPlanMode=
IterationTime=250ms MaxIterationCount=20 MemoryRandomization=Default
MinIterationCount=15 WarmupCount=1

Method	Toolchain	Mean	Error	Ratio	Allocated	Alloc Ratio
MarshalStructureToPtr	Main	53.46 ns	0.045 ns	1.00	24 B	1.00
MarshalStructureToPtr	PR	41.83 ns	0.073 ns	0.78	24 B	1.00

MarshalPtrToStructure	Main	57.27 ns	0.111 ns	1.00	24 B	1.00
MarshalPtrToStructure	PR	43.72 ns	0.169 ns	0.76	24 B	1.00

MarshalDestroyStructure	Main	109.19 ns	0.141 ns	1.00	96 B	1.00
MarshalDestroyStructure	PR	87.02 ns	0.241 ns	0.80	96 B	1.00

…ayout info for the type.

jkoritzinsky · 2026-04-08T18:00:31Z

@MihuBot benchmark Interop.StructureToPtr

MihuBot · 2026-04-08T18:17:49Z

Interop.StructureToPtr

BenchmarkDotNet v0.16.0-nightly.20260320.467, Linux Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74 2.60GHz, 1 CPU, 8 logical and 4 physical cores
Memory: 31.34 GB Total, 1.78 GB Available
Job-TPEJOW : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
Job-HKHXHK : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
EvaluateOverhead=False OutlierMode=Default PowerPlanMode=
IterationTime=250ms MaxIterationCount=20 MemoryRandomization=Default
MinIterationCount=15 WarmupCount=1

Method	Toolchain	Mean	Error	Ratio	Allocated	Alloc Ratio
MarshalStructureToPtr	Main	64.20 ns	0.578 ns	1.00	24 B	1.00
MarshalStructureToPtr	PR	30.70 ns	0.567 ns	0.48	24 B	1.00

MarshalPtrToStructure	Main	62.55 ns	0.117 ns	1.00	24 B	1.00
MarshalPtrToStructure	PR	37.14 ns	0.026 ns	0.59	24 B	1.00

MarshalDestroyStructure	Main	109.97 ns	0.152 ns	1.00	96 B	1.00
MarshalDestroyStructure	PR	68.63 ns	0.158 ns	0.62	96 B	1.00

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

jkoritzinsky · 2026-04-07T23:32:12Z

Fixes#126608

Move the marshalling info to be cached on the type itself instead of in a ConditionalWeakTable.

Move some of the validation (in particular the "has layout" validation) to be done only at "create" time so we don't have to rerun it for every time someone calls the Marshal methods with the same (valid) type.

Use the blittable fallbacks on StructureMarshaler<T> and LayoutClassMarshaler<T> instead of inlining the calls. The cost may go up slightly (delegate call to a method that calls SpanHelpers.Memmove instead of direct call to SpanHelpers.Memmove) but still much faster than the existing implementation that does the "is blittable" checks each time.

…ype instead of in a ConditionalWeakTable Should fixdotnet#126608

jkoritzinsky · 2026-04-07T23:32:46Z

@MihuBot benchmark Interop.StructureToPtr

dotnet-policy-service · 2026-04-07T23:33:33Z

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

Copilot

Pull request overview

This PR moves the cache for Marshal.LayoutTypeMarshalerMethods from a ConditionalWeakTable<Type, …> to the per-RuntimeType generic cache infrastructure to address the reported interop performance regressions (issue #126608).

Changes:

Adds LayoutTypeMarshalerMethods storage to RuntimeType.CompositeCacheEntry so it can live in the existing RuntimeType generic cache.
Refactors LayoutTypeMarshalerMethods to implement RuntimeType.IGenericCacheEntry<T> and retrieves it via RuntimeType.GetOrCreateCacheEntry<T>().
Switches from delegate creation (CreateDelegate) to caching raw function pointers (MethodHandle.GetFunctionPointer) for the marshaler methods.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
`src/coreclr/System.Private.CoreLib/src/System/RuntimeType.GenericCache.cs`	Adds a new composite-cache field to hold the marshaler-methods cache on `RuntimeType`.
`src/coreclr/System.Private.CoreLib/src/System/Runtime/InteropServices/Marshal.CoreCLR.cs`	Reworks `LayoutTypeMarshalerMethods` to use `RuntimeType` cache entries and cached function pointers instead of a `ConditionalWeakTable`.

MihuBot · 2026-04-07T23:51:28Z

Interop.StructureToPtr

BenchmarkDotNet v0.16.0-nightly.20260320.467, Linux Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74 2.60GHz, 1 CPU, 8 logical and 4 physical cores
Memory: 31.34 GB Total, 2.07 GB Available
Job-TPEJOW : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
Job-HKHXHK : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
EvaluateOverhead=False OutlierMode=Default PowerPlanMode=
IterationTime=250ms MaxIterationCount=20 MemoryRandomization=Default
MinIterationCount=15 WarmupCount=1

Method	Toolchain	Mean	Error	Ratio	Allocated	Alloc Ratio
MarshalStructureToPtr	Main	53.46 ns	0.045 ns	1.00	24 B	1.00
MarshalStructureToPtr	PR	41.83 ns	0.073 ns	0.78	24 B	1.00

MarshalPtrToStructure	Main	57.27 ns	0.111 ns	1.00	24 B	1.00
MarshalPtrToStructure	PR	43.72 ns	0.169 ns	0.76	24 B	1.00

MarshalDestroyStructure	Main	109.19 ns	0.141 ns	1.00	96 B	1.00
MarshalDestroyStructure	PR	87.02 ns	0.241 ns	0.80	96 B	1.00

…ayout info for the type.

jkoritzinsky · 2026-04-08T18:00:31Z

@MihuBot benchmark Interop.StructureToPtr

MihuBot · 2026-04-08T18:17:49Z

Interop.StructureToPtr

BenchmarkDotNet v0.16.0-nightly.20260320.467, Linux Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74 2.60GHz, 1 CPU, 8 logical and 4 physical cores
Memory: 31.34 GB Total, 1.78 GB Available
Job-TPEJOW : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
Job-HKHXHK : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
EvaluateOverhead=False OutlierMode=Default PowerPlanMode=
IterationTime=250ms MaxIterationCount=20 MemoryRandomization=Default
MinIterationCount=15 WarmupCount=1

Method	Toolchain	Mean	Error	Ratio	Allocated	Alloc Ratio
MarshalStructureToPtr	Main	64.20 ns	0.578 ns	1.00	24 B	1.00
MarshalStructureToPtr	PR	30.70 ns	0.567 ns	0.48	24 B	1.00

MarshalPtrToStructure	Main	62.55 ns	0.117 ns	1.00	24 B	1.00
MarshalPtrToStructure	PR	37.14 ns	0.026 ns	0.59	24 B	1.00

MarshalDestroyStructure	Main	109.97 ns	0.152 ns	1.00	96 B	1.00
MarshalDestroyStructure	PR	68.63 ns	0.158 ns	0.62	96 B	1.00

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

jkoritzinsky · 2026-04-07T23:32:12Z

Fixes#126608

Move the marshalling info to be cached on the type itself instead of in a ConditionalWeakTable.

Move some of the validation (in particular the "has layout" validation) to be done only at "create" time so we don't have to rerun it for every time someone calls the Marshal methods with the same (valid) type.

Use the blittable fallbacks on StructureMarshaler<T> and LayoutClassMarshaler<T> instead of inlining the calls. The cost may go up slightly (delegate call to a method that calls SpanHelpers.Memmove instead of direct call to SpanHelpers.Memmove) but still much faster than the existing implementation that does the "is blittable" checks each time.

…ype instead of in a ConditionalWeakTable Should fixdotnet#126608

jkoritzinsky · 2026-04-07T23:32:46Z

@MihuBot benchmark Interop.StructureToPtr

dotnet-policy-service · 2026-04-07T23:33:33Z

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

Copilot

Pull request overview

This PR moves the cache for Marshal.LayoutTypeMarshalerMethods from a ConditionalWeakTable<Type, …> to the per-RuntimeType generic cache infrastructure to address the reported interop performance regressions (issue #126608).

Changes:

Adds LayoutTypeMarshalerMethods storage to RuntimeType.CompositeCacheEntry so it can live in the existing RuntimeType generic cache.
Refactors LayoutTypeMarshalerMethods to implement RuntimeType.IGenericCacheEntry<T> and retrieves it via RuntimeType.GetOrCreateCacheEntry<T>().
Switches from delegate creation (CreateDelegate) to caching raw function pointers (MethodHandle.GetFunctionPointer) for the marshaler methods.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
`src/coreclr/System.Private.CoreLib/src/System/RuntimeType.GenericCache.cs`	Adds a new composite-cache field to hold the marshaler-methods cache on `RuntimeType`.
`src/coreclr/System.Private.CoreLib/src/System/Runtime/InteropServices/Marshal.CoreCLR.cs`	Reworks `LayoutTypeMarshalerMethods` to use `RuntimeType` cache entries and cached function pointers instead of a `ConditionalWeakTable`.

MihuBot · 2026-04-07T23:51:28Z

Interop.StructureToPtr

BenchmarkDotNet v0.16.0-nightly.20260320.467, Linux Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74 2.60GHz, 1 CPU, 8 logical and 4 physical cores
Memory: 31.34 GB Total, 2.07 GB Available
Job-TPEJOW : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
Job-HKHXHK : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
EvaluateOverhead=False OutlierMode=Default PowerPlanMode=
IterationTime=250ms MaxIterationCount=20 MemoryRandomization=Default
MinIterationCount=15 WarmupCount=1

Method	Toolchain	Mean	Error	Ratio	Allocated	Alloc Ratio
MarshalStructureToPtr	Main	53.46 ns	0.045 ns	1.00	24 B	1.00
MarshalStructureToPtr	PR	41.83 ns	0.073 ns	0.78	24 B	1.00

MarshalPtrToStructure	Main	57.27 ns	0.111 ns	1.00	24 B	1.00
MarshalPtrToStructure	PR	43.72 ns	0.169 ns	0.76	24 B	1.00

MarshalDestroyStructure	Main	109.19 ns	0.141 ns	1.00	96 B	1.00
MarshalDestroyStructure	PR	87.02 ns	0.241 ns	0.80	96 B	1.00

…ayout info for the type.

jkoritzinsky · 2026-04-08T18:00:31Z

@MihuBot benchmark Interop.StructureToPtr

MihuBot · 2026-04-08T18:17:49Z

Interop.StructureToPtr

BenchmarkDotNet v0.16.0-nightly.20260320.467, Linux Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74 2.60GHz, 1 CPU, 8 logical and 4 physical cores
Memory: 31.34 GB Total, 1.78 GB Available
Job-TPEJOW : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
Job-HKHXHK : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
EvaluateOverhead=False OutlierMode=Default PowerPlanMode=
IterationTime=250ms MaxIterationCount=20 MemoryRandomization=Default
MinIterationCount=15 WarmupCount=1

Method	Toolchain	Mean	Error	Ratio	Allocated	Alloc Ratio
MarshalStructureToPtr	Main	64.20 ns	0.578 ns	1.00	24 B	1.00
MarshalStructureToPtr	PR	30.70 ns	0.567 ns	0.48	24 B	1.00

MarshalPtrToStructure	Main	62.55 ns	0.117 ns	1.00	24 B	1.00
MarshalPtrToStructure	PR	37.14 ns	0.026 ns	0.59	24 B	1.00

MarshalDestroyStructure	Main	109.97 ns	0.152 ns	1.00	96 B	1.00
MarshalDestroyStructure	PR	68.63 ns	0.158 ns	0.62	96 B	1.00

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

jkoritzinsky · 2026-04-07T23:32:12Z

Fixes#126608

Move the marshalling info to be cached on the type itself instead of in a ConditionalWeakTable.

Move some of the validation (in particular the "has layout" validation) to be done only at "create" time so we don't have to rerun it for every time someone calls the Marshal methods with the same (valid) type.

Use the blittable fallbacks on StructureMarshaler<T> and LayoutClassMarshaler<T> instead of inlining the calls. The cost may go up slightly (delegate call to a method that calls SpanHelpers.Memmove instead of direct call to SpanHelpers.Memmove) but still much faster than the existing implementation that does the "is blittable" checks each time.

…ype instead of in a ConditionalWeakTable Should fixdotnet#126608

jkoritzinsky · 2026-04-07T23:32:46Z

@MihuBot benchmark Interop.StructureToPtr

dotnet-policy-service · 2026-04-07T23:33:33Z

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

Copilot

Pull request overview

This PR moves the cache for Marshal.LayoutTypeMarshalerMethods from a ConditionalWeakTable<Type, …> to the per-RuntimeType generic cache infrastructure to address the reported interop performance regressions (issue #126608).

Changes:

Adds LayoutTypeMarshalerMethods storage to RuntimeType.CompositeCacheEntry so it can live in the existing RuntimeType generic cache.
Refactors LayoutTypeMarshalerMethods to implement RuntimeType.IGenericCacheEntry<T> and retrieves it via RuntimeType.GetOrCreateCacheEntry<T>().
Switches from delegate creation (CreateDelegate) to caching raw function pointers (MethodHandle.GetFunctionPointer) for the marshaler methods.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
`src/coreclr/System.Private.CoreLib/src/System/RuntimeType.GenericCache.cs`	Adds a new composite-cache field to hold the marshaler-methods cache on `RuntimeType`.
`src/coreclr/System.Private.CoreLib/src/System/Runtime/InteropServices/Marshal.CoreCLR.cs`	Reworks `LayoutTypeMarshalerMethods` to use `RuntimeType` cache entries and cached function pointers instead of a `ConditionalWeakTable`.

MihuBot · 2026-04-07T23:51:28Z

Interop.StructureToPtr

BenchmarkDotNet v0.16.0-nightly.20260320.467, Linux Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74 2.60GHz, 1 CPU, 8 logical and 4 physical cores
Memory: 31.34 GB Total, 2.07 GB Available
Job-TPEJOW : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
Job-HKHXHK : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
EvaluateOverhead=False OutlierMode=Default PowerPlanMode=
IterationTime=250ms MaxIterationCount=20 MemoryRandomization=Default
MinIterationCount=15 WarmupCount=1

Method	Toolchain	Mean	Error	Ratio	Allocated	Alloc Ratio
MarshalStructureToPtr	Main	53.46 ns	0.045 ns	1.00	24 B	1.00
MarshalStructureToPtr	PR	41.83 ns	0.073 ns	0.78	24 B	1.00

MarshalPtrToStructure	Main	57.27 ns	0.111 ns	1.00	24 B	1.00
MarshalPtrToStructure	PR	43.72 ns	0.169 ns	0.76	24 B	1.00

MarshalDestroyStructure	Main	109.19 ns	0.141 ns	1.00	96 B	1.00
MarshalDestroyStructure	PR	87.02 ns	0.241 ns	0.80	96 B	1.00

…ayout info for the type.

jkoritzinsky · 2026-04-08T18:00:31Z

@MihuBot benchmark Interop.StructureToPtr

MihuBot · 2026-04-08T18:17:49Z

Interop.StructureToPtr

BenchmarkDotNet v0.16.0-nightly.20260320.467, Linux Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74 2.60GHz, 1 CPU, 8 logical and 4 physical cores
Memory: 31.34 GB Total, 1.78 GB Available
Job-TPEJOW : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
Job-HKHXHK : .NET 11.0.0 (11.0.0-dev, 42.42.42.42424), X64 RyuJIT x86-64-v4
EvaluateOverhead=False OutlierMode=Default PowerPlanMode=
IterationTime=250ms MaxIterationCount=20 MemoryRandomization=Default
MinIterationCount=15 WarmupCount=1

Method	Toolchain	Mean	Error	Ratio	Allocated	Alloc Ratio
MarshalStructureToPtr	Main	64.20 ns	0.578 ns	1.00	24 B	1.00
MarshalStructureToPtr	PR	30.70 ns	0.567 ns	0.48	24 B	1.00

MarshalPtrToStructure	Main	62.55 ns	0.117 ns	1.00	24 B	1.00
MarshalPtrToStructure	PR	37.14 ns	0.026 ns	0.59	24 B	1.00

MarshalDestroyStructure	Main	109.97 ns	0.152 ns	1.00	96 B	1.00
MarshalDestroyStructure	PR	68.63 ns	0.158 ns	0.62	96 B	1.00

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

Copilot

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

Move the cache for the marshaler methods to be stored on the RuntimeT…
20eb643
…ype instead of in a ConditionalWeakTable Should fixdotnet#126608

CopilotAI review requested due to automatic review settings April 7, 2026 23:32

github-actionsBot added the area-System.Reflection label Apr 7, 2026

dotnet-policy-serviceBot assigned jkoritzinsky Apr 7, 2026

github-project-automationBot added this to AppModel Apr 7, 2026

Copilotstarted reviewing on behalf of jkoritzinsky April 7, 2026 23:33 View session

MihuBot mentioned this pull request Apr 7, 2026
[Benchmark X64] [jkoritzinsky] Move the cache for the marshaler methods to b ... MihuBot/runtime-utils#1843
Open

CopilotAI reviewed Apr 7, 2026
View reviewed changes

Comment threadsrc/coreclr/System.Private.CoreLib/src/System/Runtime/InteropServices/Marshal.CoreCLR.cs Outdated

jkoritzinsky added 2 commits April 8, 2026 10:52

Only validate that the type has layout the first time we set up the l…
8cc2f74
…ayout info for the type.

Keep the delegates (less need for unsafe)
d6dc96a

MihuBot mentioned this pull request Apr 8, 2026
[Benchmark X64] [jkoritzinsky] Move the cache for the marshaler methods to b ... MihuBot/runtime-utils#1844
Open

jkoritzinsky marked this pull request as ready for review April 8, 2026 18:18

CopilotAI review requested due to automatic review settings April 8, 2026 18:18

Copilotstarted reviewing on behalf of jkoritzinsky April 8, 2026 18:19 View session

CopilotAI reviewed Apr 8, 2026
View reviewed changes

AaronRobinsonMSFT approved these changes Apr 8, 2026
View reviewed changes

jkoritzinsky enabled auto-merge (squash) April 8, 2026 18:36

This was referenced Apr 9, 2026
Unable to pull image from mcr.microsoft.com #117164
Closed
System.Net.NameResolution.Tests DNS failures: Name or service not known #126641
Closed

Fix test failures.
7b317a7

AaronRobinsonMSFT approved these changes Apr 10, 2026
View reviewed changes

Remove trailing whitespace
eefb95d

CopilotAI review requested due to automatic review settings April 10, 2026 19:18

Copilotstarted reviewing on behalf of jkoritzinsky April 10, 2026 19:19 View session

CopilotAI reviewed Apr 10, 2026
View reviewed changes

Comment threadsrc/coreclr/System.Private.CoreLib/src/System/Runtime/InteropServices/Marshal.CoreCLR.cs Outdated
Comment threadsrc/coreclr/System.Private.CoreLib/src/System/Runtime/InteropServices/Marshal.CoreCLR.cs

Fix failures
288cd77

jkoritzinsky merged commit e45ecce into dotnet:mainApr 11, 2026
100 of 111 checks passed

github-project-automationBot moved this to Done in AppModel Apr 11, 2026

dotnet-maestroBot mentioned this pull request Apr 11, 2026
[main] Source code updates from dotnet/runtime dotnet/dotnet#5987
Merged

jkotas mentioned this pull request Apr 11, 2026
System.Runtime.InteropServices.Tests.DestroyStructureTests.DestroyStructure_Blittable_Success failing in CI #126788
Closed

CopilotAI mentioned this pull request Apr 11, 2026
Fix NullReferenceException in DestroyStructure for blittable types #126795
Merged

github-actionsBot locked and limited conversation to collaborators May 11, 2026