benchmarks

AutoMap.Generator Benchmarks

BenchmarkDotNet comparison of AutoMap.Generator vs Mapperly, AutoMapper, and a hand-written baseline.

Running

cd benchmarks/AutoMap.Benchmarks
dotnet run -c Release

Requires .NET 9 SDK. First run downloads BenchmarkDotNet, AutoMapper, and Mapperly packages (~30 s).

Latest results

Mapping a 5-property class (Order → OrderDto) on Windows 11, AMD Ryzen 9 5900X, .NET 9.0.18:

Method	Mean	Ratio	Alloc
Hand-written	7.23 ns	1.00	64 B
AutoMap.Generator	6.64 ns	0.93	64 B
Mapperly	6.68 ns	0.93	64 B
AutoMapper	53.40 ns	7.45x	64 B

Mapping a nested/collection scenario (BigOrder → BigOrderDto, one nested Customer object plus a 10-item List<LineItem>) on the same machine:

Method	Mean	Ratio	Alloc
Hand-written	113.5 ns	1.00	824 B
AutoMap.Generator	105.8 ns	0.94	824 B
Mapperly	118.6 ns	1.05	792 B
AutoMapper	234.2 ns	2.07x	944 B

AutoMap.Generator and Mapperly both produce code equivalent to hand-written property assignment — differences are within measurement noise. AutoMapper's reflection overhead is ~2×–7.5× higher depending on mapping complexity, and its per-call allocations don't shrink with AOT/trimming since they route through cached delegate reflection.

Regenerate these numbers yourself: cd benchmarks/AutoMap.Benchmarks && dotnet run -c Release -- --filter '*'. Raw BenchmarkDotNet output (including full statistics) is written to BenchmarkDotNet.Artifacts/results/.

What is being measured

• Hand-written — a static Map(Order) → OrderDto method with explicit property assignments (baseline).
• AutoMap.Generator — the generated order.ToOrderDto() / order.ToBigOrderDto() extension method.
• Mapperly — a [Mapper] partial class with an auto-generated Map(Order) → OrderDto method.
• AutoMapper — IMapper.Map<OrderDto>(order) with a pre-configured MapperConfiguration.
• The nested/collection scenario additionally measures a Customer → CustomerDto object reference and a List<LineItem> → List<LineItemDto> collection copy, to show overhead beyond flat property copies.