optimize radixsort with simd

[kazeburo]

PR Type

Enhancement, Tests, Documentation

---

Description

• Adds adaptive, allocation-aware radix sorting

• Introduces SIMD scans for amd64 and ARM64

• Adds extensive correctness and fuzz testing

• Documents benchmarks, portability, and usage

---

Diagram Walkthrough

flowchart LR
  input["Non-negative float64 input"] --> scan["SIMD or generic key scan"]
  scan --> ordered["Ordered or descending fast path"]
  scan --> adaptive["Adaptive radix strategy"]
  adaptive --> lsd["Six-pass 11-bit LSD sort"]
  adaptive --> msd["Sampled MSD partition sort"]
  ordered --> result["In-place sorted output"]
  lsd --> result
  msd --> result

Loading

File Walkthrough

	Relevant files
Tests	3 files benchmark_test.go Expand sorting and scan benchmarks                                              +28/-5    radix_test.go Add comprehensive radix correctness tests                                +186/-0  sorted_test.go Extend radix input test distributions                                        +8/-2
Enhancement	6 files radix.go Add adaptive in-place radix sorting                                            +163/-0  radix_scan.go Add portable radix scan implementation                                      +21/-0    radix_scan_amd64.go Declare amd64 SIMD radix scan                                                        +8/-0      radix_scan_arm64.go Declare ARM64 SIMD radix scan                                                        +8/-0      radix_scan_amd64.s Implement amd64 SSE2 key scanning                                                +85/-0    radix_scan_arm64.s Implement ARM64 NEON key scanning                                                +72/-0
Configuration changes	2 files radix_scan_generic.go Provide portable scan fallback selection                                  +7/-0      Makefile Validate pure-Go build configuration                                          +1/-0
Miscellaneous	1 files sorted.go Move radix implementation into dedicated file                        +0/-53
Documentation	2 files BENCHMARKS.md Document radix performance measurements                                    +186/-0  README.md Document radix implementation and benchmarks                          +39/-2

---

[github-actions]

Benchmark Comparison

Base: main
PR: feat/simd-radix-sort

BenchmarkSampdoAll results

goos: linux
goarch: amd64
pkg: github.com/monitoring-forge/sampdo
cpu: AMD EPYC 7763 64-Core Processor                
                                        │ base-bench.txt │             pr-bench.txt             │
                                        │     sec/op     │    sec/op     vs base                │
SampdoAll/sampdo/100000/duplicates-4        1.534m ± 11%   1.170m ±  6%  -23.72% (p=0.000 n=10)
SampdoAll/sampdo/100000/random-4            3.253m ±  2%   2.184m ±  1%  -32.86% (p=0.000 n=10)
SampdoAll/sampdo/100000/wide-4              3.363m ±  2%   2.220m ±  1%  -33.98% (p=0.000 n=10)
SampdoAll/sampdo/100000/sorted-4            245.0µ ±  4%   220.5µ ±  2%  -10.03% (p=0.000 n=10)
SampdoAll/sampdo/100000/response_time-4     2.410m ±  1%   1.478m ± 16%  -38.68% (p=0.000 n=10)
geomean                                     1.582m         1.131m        -28.52%

                                        │ base-bench.txt │            pr-bench.txt             │
                                        │      B/op      │     B/op      vs base               │
SampdoAll/sampdo/100000/duplicates-4        1.532Mi ± 0%   1.532Mi ± 0%       ~ (p=0.695 n=10)
SampdoAll/sampdo/100000/random-4            1.532Mi ± 0%   1.532Mi ± 0%       ~ (p=0.669 n=10)
SampdoAll/sampdo/100000/wide-4              1.532Mi ± 0%   1.532Mi ± 0%       ~ (p=0.354 n=10)
SampdoAll/sampdo/100000/sorted-4            785.1Ki ± 0%   785.1Ki ± 0%       ~ (p=0.098 n=10)
SampdoAll/sampdo/100000/response_time-4     1.532Mi ± 0%   1.532Mi ± 0%       ~ (p=0.052 n=10)
geomean                                     1.334Mi        1.334Mi       -0.00%

                                        │ base-bench.txt │            pr-bench.txt             │
                                        │   allocs/op    │ allocs/op   vs base                 │
SampdoAll/sampdo/100000/duplicates-4          5.000 ± 0%   5.000 ± 0%       ~ (p=1.000 n=10) ¹
SampdoAll/sampdo/100000/random-4              5.000 ± 0%   5.000 ± 0%       ~ (p=1.000 n=10) ¹
SampdoAll/sampdo/100000/wide-4                5.000 ± 0%   5.000 ± 0%       ~ (p=1.000 n=10) ¹
SampdoAll/sampdo/100000/sorted-4              4.000 ± 0%   4.000 ± 0%       ~ (p=1.000 n=10) ¹
SampdoAll/sampdo/100000/response_time-4       5.000 ± 0%   5.000 ± 0%       ~ (p=1.000 n=10) ¹
geomean                                       4.782        4.782       +0.00%
¹ all samples are equal

[github-actions]

PR Reviewer Guide 🔍

(Review updated until commit e091567)

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 4 🔵🔵🔵🔵⚪

🧪 PR contains tests

🔒 No security concerns identified

⚡ No major issues detected

[github-actions]

PR Code Suggestions ✨

No code suggestions found for the PR.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[github-actions]

Persistent review updated to latest commit e091567