Yukinary (YKN) Performance & Benchmark Suite

Comprehensive systems benchmarks and industrial workload comparisons for the Yukinary Programming Language, measuring Yukinary (direct x86_64 machine code) against C (GCC -O3), Rust (rustc -O3), Go, and Python 3.13.

---

Visual Overview

---

Verified Benchmark Summary

1. Compiler Build Throughput (performance.json, Median of 7 Runs)

Yukinary emits PE32+ and ELF64 binaries directly from AST without LLVM passes or external linkers:

Compiler / Toolchain	Workload	Median Build Time	Relative Speed
Yukinary (ykn)	optimizer (loop opt & regalloc)	18.6 ms	Baseline (1.0x)
C (gcc -O3)	rivals/workload.c	219.4 ms	11.8x slower
Rust (rustc -O3)	rivals/workload.rs	272.2 ms	14.6x slower
Go (go build)	rivals/workload.go	313.1 ms	16.8x slower
C++ (g++ -O3)	rivals/workload.cpp	583.0 ms	31.3x slower

2. Systems B-Tree & Heavy Allocation Workload (industrial/run_systems_compare.ps1)

100,000 node allocations + 50,000 random lookups + [50,000 .. 150,000] range scan aggregation (verified exact match across all runtimes: Found: 19526, RangeSum: 6363996222):

Implementation	Wall-Clock Time	Checksum Verification	Memory Model
C (gcc -O3)	133 ms	19526 / 6363996222	Manual malloc/free (UCRT Heap)
Yukinary (Native)	213 ms	19526 / 6363996222	Zero-libc Arena / Freestanding
Rust (rustc -O3)	221 ms	19526 / 6363996222	Box<Node> RAII Ownership
Python (3.13)	327 ms	19526 / 6363996222	Reference-counted PyObject

3. BitNet b1.58 Ternary AI Kernel (industrial/bitnet_results.json)

$100 \text{ iterations} \times (64 \times 512 \times 512 \times 2) = 3.355 \text{ GOP}$ integer ternary matrix multiplication + RMSNorm (verified checksum -175 across all runtimes):

Implementation	Execution Time	Throughput	Peak RSS	Notes
C (gcc -O3)	1.874 s	1.79 GOP/s	4.8 MB	Uses GCC SIMD AVX2 auto-vectorization
Rust (rustc -O3)	1.925 s	1.74 GOP/s	5.2 MB	Uses LLVM SIMD AVX2 auto-vectorization
Yukinary (Native)	4.714 s	0.71 GOP/s	2.6 MB	Custom scalar x86_64 codegen (lowest RSS, 34x faster than Python)
Python (3.13)	160.884 s	0.02 GOP/s	22.4 MB	Baseline CPython interpreter

---

Reproducing Locally

# Run the heavy systems B-Tree comparison
powershell -ExecutionPolicy Bypass -File industrial\run_systems_compare.ps1

# Run the BitNet b1.58 AI kernel comparison
powershell -ExecutionPolicy Bypass -File industrial\run_bitnet_compare.ps1

# Generate summary report from performance.json
python report.py

Full methodology and hardware details: industrial/YUKINARY_ENDUSTRIYEL_BENCHMARK_RAPORU.md.

---

License

Dual-licensed under either of:

• Apache License, Version 2.0 (LICENSE-APACHE)
• MIT license (LICENSE-MIT)

at your option.