https://github.com/malstraem/generic-simd-numerics

https://github.com/malstraem/generic-simd-numerics

benchmarks dotnet intrinsics linear-algebra numerics simd

Last synced: about 1 month ago
JSON representation

• Host: GitHub
• URL: https://github.com/malstraem/generic-simd-numerics
• Owner: malstraem
• License: mit
• Created: 2025-11-11T18:16:38.000Z (8 months ago)
• Default Branch: main
• Last Pushed: 2026-06-04T14:43:53.000Z (about 1 month ago)
• Last Synced: 2026-06-04T15:12:41.440Z (about 1 month ago)
• Topics: benchmarks, dotnet, intrinsics, linear-algebra, numerics, simd
• Language: C#
• Homepage:
• Size: 396 KB
• Stars: 1
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: readme.md
  • License: license

Awesome Lists containing this project

README

          
prototype of generic linear algebraic objects using current state of

[System.Runtime.Intrinsics](https://learn.microsoft.com/ru-ru/dotnet/api/system.runtime.intrinsics?view=net-10.0) SIMD vectors

started as try to outperform [Silk.NET](https://github.com/dotnet/Silk.NET), especially on matrix operations with `double` precision

this is primarily for in-house engine, but possibly helpful draft for https://github.com/dotnet/runtime/issues/24168

next step is hardware accelerated ways for type conversions like `Vec4` <-> `Vec4` 🚧

> [!NOTE]

check [benchmark reports](./reports) for generated asm and timings

tested and benchmarked on AMD Ryzen 9 7900X that have

`AVX512 - BITALG+VBMI2+VNNI+VPOPCNTDQ+IFMA+VBMI+F+BW+CD+DQ+VL`

`AVX2 - BMI1+BMI2+F16C+FMA+LZCNT+MOVBE`

`AVX+SSE3+SSSE3+SSE4.1+SSE4.2`