IGNOREME: distance functions with golang simd (v1.26)

[cpegeric]

User description

What type of PR is this?

• API-change
• BUG
• Improvement
• Documentation
• Feature
• Test and CI
• Code Refactoring

Which issue(s) this PR fixes:

issue #23502

What this PR does / why we need it:

distance functions implemented by simd instruction with golang v1.26.

---

PR Type

Enhancement

---

Description

• Implement SIMD-optimized distance functions for x86_64 architecture

  • L2 distance, L1 distance, inner product, cosine distance/similarity
  • Spherical distance for sparse vectors
  • Support AVX-512, AVX2, and AVX instruction sets with scalar fallback

• Refactor distance function resolution into separate module

  • Extract kmeans and general distance resolution functions to resolve.go

• Add float32 test cases for L2 distance squared calculation

• Update Makefile to support GOEXPERIMENT=simd and GOAMD64 configuration

---

Diagram Walkthrough

flowchart LR
  A["distance_func.go<br/>Generic implementations"] -->|amd64 build| B["distance_func_amd64.go<br/>SIMD optimized"]
  A -->|non-amd64 build| C["distance_func.go<br/>Scalar fallback"]
  D["resolve.go<br/>Distance resolution"] -->|uses| B
  D -->|uses| C
  E["Makefile<br/>Build config"] -->|GOEXPERIMENT=simd| B
  E -->|GOAMD64 flag| B

Loading

File Walkthrough

	Relevant files
Refactoring	distance_func.go Mark as non-amd64 fallback, remove resolution functions    pkg/vectorindex/metric/distance_func.go Added build constraint //go:build !amd64 to mark as non-amd64 fallback Removed distance function resolution functions (moved to resolve.go) Kept generic scalar implementations for non-amd64 platforms +2/-84    resolve.go Extract distance function resolution to separate module    pkg/vectorindex/metric/resolve.go New file containing distance function resolution logic Extracted ResolveKmeansDistanceFn, ResolveKmeansDistanceFnForDense, ResolveKmeansDistanceFnForSparse Extracted ResolveDistanceFn for general distance function selection Supports metric type resolution for both dense and sparse vectors +104/-0
Enhancement	distance_func_amd64.go SIMD-optimized distance functions for x86_64                          pkg/vectorindex/metric/distance_func_amd64.go New file with SIMD-optimized implementations for amd64 architecture Implements L2 distance squared with AVX-512, AVX2, and scalar paths Implements L1 distance with multi-level SIMD optimization Implements inner product with AVX-512 and AVX2 support Implements cosine distance and similarity with SIMD acceleration Implements spherical distance for sparse vector operations Helper functions for horizontal sum reduction across vector types Support for both float32 and float64 data types +942/-0
Tests	distance_func_test.go Add float32 L2 distance squared tests                                        pkg/vectorindex/metric/distance_func_test.go Added Test_L2DistanceSqFp32 test function with 6 test cases Tests float32 L2 distance squared calculation with various vector sizes Covers edge cases and typical vector dimensions +68/-0
Configuration changes	Makefile Add GOEXPERIMENT=simd and GOAMD64 build configuration        Makefile Added GO variable support for custom Go binary path Added GOEXPERIMENT_OPT to enable SIMD experiment for x86_64 Added GOAMD64 flag support for specifying AMD64 microarchitecture level Refactored TAGS to GOTAGS for proper tag formatting Updated build commands to use $(GO) and $(GOEXPERIMENT_OPT) variables Updated musl build target to use new tag variables +29/-11

---