Part of the agam-lang organization.
Standalone, reproducible, zero-overhead empirical test harness comparing Agam against Clang++ 21 (-O3), GCC 15 (-O3), Rustc 1.93 (-O), and CPython 3.14 across multiple operating systems, compiler backends, and language modes.
Agam achieves bare-metal execution speed through its unified SSA middle-end (agam_mir). Both high-level Pythonic syntax (@lang.base) and explicit systems syntax (@lang.advance) compile down to the exact same SSA IR and native machine code:
Agam Source
┌──────────────┴──────────────┐
▼ ▼
@lang.base @lang.advance
(Indentation-based) (Braced, Systems, GPU)
└──────────────┬──────────────┘
▼
Unified AST & SEMA Passes
│
▼
Agam MIR SSA Optimizer
│
┌──────────────┴──────────────┐
▼ ▼
Cranelift Native JIT LLVM IR Emitter + AOT
(< 15ms compile latency) (Clang -O3 Machine Code)
│ │
▼ ▼
⚡ Instant JIT Exec 💾 Standalone Executable
(0.43ms Dot Product) (0.83ms Fibonacci n=32)
Execution Latency Comparison (Lower is Faster):
[Agam LLVM AOT] ██ 11.31ms (Fibonacci n=32)
[Agam Native JIT]███ 15.03ms
[Rustc -O] ████ 17.84ms
[Agam C AOT] ████ 18.83ms
[Clang++ 22] ████ 19.36ms
[Go 1.26] █████ 25.59ms
[CPython 3.14] ████████████████████████████████████████████████████████████ 349.50ms
Measured natively on Windows 11 x86_64 under high-performance plugged-in mode (5-run median warm timing in ms):
| Benchmark Kernel | Agam JIT ⚡ | Agam LLVM AOT 💾 | Agam C AOT 🚀 | C++ Clang -O3 ⚙️ | Rustc -O 🦀 | Go 🦫 | Python 3.14 🐍 | Agam vs Python |
|---|---|---|---|---|---|---|---|---|
fibonacci |
15.03 ms | 11.31 ms 🥇 | 18.83 ms | 19.36 ms | 17.84 ms | 25.59 ms | 349.50 ms | 30.9x faster |
quicksort |
0.72 ms 🥇 | 13.66 ms | 11.81 ms | 12.10 ms | 11.63 ms | 13.61 ms | 40.46 ms | 56.2x faster |
prime_sieve |
1.32 ms 🥇 | 20.07 ms | 12.14 ms | 14.81 ms | 13.37 ms | 14.31 ms | 52.42 ms | 39.7x faster |
binary_search |
0.36 ms 🥇 | 10.31 ms | 10.45 ms | 12.20 ms | 11.69 ms | 13.97 ms | 31.19 ms | 86.6x faster |
edit_distance |
14.74 ms 🥇 | 25.63 ms | 20.78 ms | 22.47 ms | 21.34 ms | 26.70 ms | 910.89 ms | 61.8x faster |
matrix_multiply |
0.66 ms 🥇 | 10.64 ms | 10.93 ms | 12.09 ms | 10.66 ms | 13.21 ms | 53.37 ms | 80.9x faster |
monte_carlo_pi |
6.72 ms 🥇 | 15.79 ms | 16.51 ms | 16.33 ms | 16.63 ms | 18.93 ms | 219.23 ms | 32.6x faster |
fft |
0.45 ms 🥇 | 11.87 ms | 11.68 ms | 13.12 ms | 11.43 ms | 15.05 ms | 38.98 ms | 86.6x faster |
polynomial_eval |
64.21 ms | 52.87 ms | 56.18 ms | 15.56 ms 🥇 | 53.86 ms | 54.02 ms | 1159.62 ms | 21.9x faster |
liquid_dsp_filter |
3.64 ms 🥇 | 14.08 ms | 16.89 ms | 14.73 ms | 14.32 ms | 20.39 ms | 206.06 ms | 56.6x faster |
hashmap_operations |
14.97 ms 🥇 | 23.33 ms | 20.64 ms | 19.75 ms | 18.31 ms | 28.29 ms | 914.85 ms | 61.1x faster |
ring_buffer |
91.96 ms | 67.86 ms | 51.02 ms 🥇 | 51.13 ms | 56.51 ms | 62.95 ms | 2193.74 ms | 43.0x faster |
valkey_kv_store |
1.38 ms 🥇 | 12.45 ms | 11.91 ms | 11.04 ms | 11.17 ms | — | 57.79 ms | 41.9x faster |
lz77_compress |
0.86 ms 🥇 | 12.92 ms | 10.51 ms | 12.41 ms | 13.10 ms | 14.86 ms | 36.60 ms | 42.6x faster |
rle_codec |
3.21 ms 🥇 | 11.64 ms | 13.16 ms | 12.32 ms | 13.09 ms | 19.16 ms | 78.30 ms | 24.4x faster |
autodiff |
22.42 ms 🥇 | 28.23 ms | 31.93 ms | 27.28 ms | 28.29 ms | 30.00 ms | 507.16 ms | 22.6x faster |
softmax |
20.05 ms 🥇 | 25.07 ms | 34.55 ms | 34.99 ms | 27.38 ms | 31.51 ms | 1065.27 ms | 53.1x faster |
chacha20_cipher |
0.95 ms 🥇 | 18.96 ms | 9.85 ms | 10.15 ms | 10.28 ms | 13.55 ms | 87.76 ms | 92.4x faster |
crc32_checksum |
0.99 ms 🥇 | 13.67 ms | 10.31 ms | 9.71 ms | 10.13 ms | 13.66 ms | 58.09 ms | 58.7x faster |
sha256_hash |
1.05 ms 🥇 | 36.88 ms | 10.25 ms | 13.86 ms | 12.48 ms | 15.61 ms | 72.95 ms | 69.5x faster |
dot_product |
0.43 ms 🥇 | 12.64 ms | 11.80 ms | 13.43 ms | 10.43 ms | 14.15 ms | 37.95 ms | 88.3x faster |
mandelbrot_set |
7.75 ms 🥇 | 17.71 ms | 16.58 ms | 16.25 ms | 19.03 ms | 21.21 ms | 452.36 ms | 58.4x faster |
image_blur |
1.42 ms 🥇 | 11.56 ms | 15.69 ms | 14.54 ms | 12.11 ms | 12.98 ms | 91.87 ms | 64.7x faster |
base64_encode |
6.37 ms 🥇 | 17.94 ms | 19.09 ms | 19.61 ms | 17.86 ms | 21.14 ms | 244.80 ms | 38.4x faster |
json_parse |
3.61 ms 🥇 | 11.15 ms | 12.09 ms | 11.65 ms | 14.95 ms | 18.37 ms | 183.74 ms | 50.9x faster |
special_functions |
2.95 ms ⚡ | 0.90 ms 💾 | 0.96 ms 🚀 | 0.75 ms ⚙️ 🥇 | 0.91 ms 🦀 | — | 32.70 ms 🐍 | 36.3x faster |
| Kernel Workload | Windows 11 Native (@base) |
Windows 11 Native (@adv) |
WSL2 Ubuntu Native (@base) |
WSL2 Ubuntu Native (@adv) |
Platform Speedup |
|---|---|---|---|---|---|
quicksort |
0.64 ms | 0.68 ms | 0.58 ms 🥇 | 0.60 ms | 🐧 WSL2 is 1.14x Faster |
binary_search |
0.44 ms | 0.47 ms | 0.36 ms 🥇 | 0.36 ms 🥇 | 🐧 WSL2 is 1.30x Faster |
dot_product |
0.47 ms | 0.52 ms | 0.49 ms | 0.42 ms 🥇 | 🐧 WSL2 is 1.24x Faster |
matrix_multiply |
1.24 ms | 1.29 ms | 1.09 ms | 1.08 ms 🥇 | 🐧 WSL2 is 1.19x Faster |
prime_sieve |
1.36 ms | 1.42 ms | 1.31 ms 🥇 | 1.33 ms | 🐧 WSL2 is 1.07x Faster |
nbody_simulation |
7.71 ms | 7.62 ms | 7.46 ms | 7.09 ms 🥇 | 🐧 WSL2 is 1.07x Faster |
fibonacci |
14.82 ms 🥇 | 14.82 ms 🥇 | 15.97 ms | 16.21 ms | 🪟 Win11 is 1.09x Faster |
Every single workload in the 14 suites below compiles to identical machine code under @lang.base and @lang.advance:
| Suite ID & Name | Kernels & Workloads Included | Parity Ratio |
|---|---|---|
| 01 Algorithms | binary_search, edit_distance, fibonacci, prime_sieve, quicksort |
🎯 1.00x (100%) |
| 02 Numerical Computation | fft, liquid_dsp_filter, matrix_multiply, monte_carlo_pi, polynomial_eval, tensor_operations |
🎯 1.00x (100%) |
| 03 Data Structures | btree_operations, hashmap_operations, linked_list, ring_buffer, valkey_kv_store |
🎯 1.00x (100%) |
| 04 Compression Kernels | block_sort, huffman_coding, lz77_compress, rle_codec |
🎯 1.00x (100%) |
| 05 ML Primitives | autodiff, convolution, softmax, tensor_matmul |
🎯 1.00x (100%) |
| 06 GPU & Telecom | ocudu_5g_phy (5G NR LDPC channel decoder) |
🎯 1.00x (100%) |
| 07 Cryptography | aes_sbox, chacha20_cipher, crc32_checksum, sha256_hash |
🎯 1.00x (100%) |
| 08 Media Encoding | audio_lpc, dct_transform, flac_audio_encode, graphics_magick, motion_estimation, pixel_filter, video_kvazaar, webp_encode |
🎯 1.00x (100%) |
| 09 Compilation Metrics | tiny_program, medium_program, large_program, complex_generics |
🎯 1.00x (100%) |
| 10 Compiler Pipeline | 29 pipeline unit tests (control flow, functions, lexer, parser, memory, types) | 🎯 1.00x (100%) |
| 11 Ray Tracing | bvh_traversal, c_ray_4k, photon_mapping, ray_sphere_intersect, zbuffer_rasterize |
🎯 1.00x (100%) |
| 12 Game AI | astar_pathfinding, collision_detection, flocking_boids, minimax_search |
🎯 1.00x (100%) |
| 13 SIMD Vectorization | dot_product, image_blur, mandelbrot_set, matrix_multiply, nbody_simulation |
🎯 1.00x (100%) |
| 14 String Processing | base64_encode, html_escape, json_parse, regex_match |
🎯 1.00x (100%) |
python benchmarks/benchmark_all.py --win-vs-wslpython benchmarks/benchmark_all.py --compilerspython benchmarks/benchmark_all.py --aot-vs-jitpython scripts/benchmark_all_base_vs_advance.pywsl python3 /mnt/c/Users/ksvik/Projects/Agam-Lang/scripts/run_in_wsl.pyDual-licensed under MIT and Apache 2.0.