Skip to content

CROSS_HOST_M2_M4_inproc.md

Latest commit

 

History

History
198 lines (158 loc) · 15.3 KB

CROSS_HOST_M2_M4_inproc.md

File metadata and controls

198 lines (158 loc) · 15.3 KB
Raw

Cross-host warm cross-codec benchmark — Apple M2 (Mac14,2) vs Apple M4 (Mac16,10)

Generated 2026-05-14T14:32:28 from Scripts/benchmarks/compare_hosts.py.

What this report measures — plain English

This report shows how fast each codec encodes and decodes test images on two different Apple chips — the M2 and the newer M4 — when the codec is called directly from inside an app, the way a real iOS or macOS application uses it. There is no command-line tool involved and no separate helper process; the codec is doing its work in the same process as the measurement harness.

This is the most relevant report if you are:

  • Building an iOS or macOS app that uses J2KSwift as a library, OR
  • Asking "how fast is J2KSwift compared to alternatives if I used it the same way?"

Two things to know before reading the numbers:

  1. The three competitor codecs (Kakadu, OpenJPH, Grok) don't ship a Swift library, so we still measure them via their command-line tools. Their numbers in this report are the same as in the sustained / isolated companion reports — the fair comparison lives in the J2KSwift column only.
  2. On M4, J2KSwift wins 31 of 38 test images outright; Kakadu wins the remaining 7. OpenJPH and Grok don't win any.

How to read the columns

Term Meaning
ms Milliseconds. Lower is better.
fixture One test image. The corpus contains 38 of them, ranging from small thumbnails to high-resolution medical scans.
Source Where the fixture came from: real = original public medical images, synth = deterministic synthetic, medical-real = additional PHI-safe real medical fixtures.
codec The software doing the encode/decode. J2KSwift+inproc is ours; Kakadu, OpenJPH, Grok are third-party alternatives.
M4 / M2 Which Apple chip the test ran on. M4 is the newer, faster chip.
speedup (e.g. 1.74×) How many times faster the newer chip is. 1.74× ≈ 43 % less time. 1.00× = no change.
wins (in "Winner pattern" section) How many of the 38 fixtures that codec was the fastest on.

Medical-imaging modality abbreviations in fixture names: MR = magnetic resonance, CT = computed tomography, DX = digital chest X-ray, PX = panoramic dental, MG = mammography, XA = angiography, NM = nuclear medicine, CR = computed radiography.

The other two reports (same data, different shape)

Hosts and runs

# Host J2KSwift version Captured Source JSON
0 (baseline) Apple M2 (Mac14,2) J2KSwift version 9.5.2 2026-05-14T06:47:04 Documentation/Benchmarks/data/benchmark-results-Mac142-9.5.2-warm-inproc-20260514.json
1 Apple M4 (Mac16,10) J2KSwift version 9.5.2 2026-05-14T14:27:34 Documentation/Benchmarks/data/benchmark-results-Mac1610-9.5.2-warm-inproc-20260514.json

Bench parameters: median of 7 timed runs after 2 warmups per fixture per codec per direction. j2kd daemon reachable: True.

ENCODE — aggregate Apple M4 vs Apple M2

Median speedup (Apple M4 ms ÷ Apple M2 ms; >1× means Apple M4 is faster).

Codec n Median Apple M4/Apple M2 Min Max
J2KSwift+inproc 38 1.74× 1.46× 2.28×
Grok 38 2.00× 0.98× 2.15×
Kakadu 38 1.00× 0.89× 2.12×
OpenJPH 38 1.00× 0.98× 2.11×

DECODE — aggregate Apple M4 vs Apple M2

Median speedup (Apple M4 ms ÷ Apple M2 ms; >1× means Apple M4 is faster).

Codec n Median Apple M4/Apple M2 Min Max
J2KSwift+inproc 38 1.55× 1.38× 2.67×
Grok 38 0.99× 0.96× 2.13×
Kakadu 38 0.99× 0.96× 2.04×
OpenJPH 38 1.00× 0.97× 2.13×

Winner pattern by host (lowest median wins)

ENCODE:

  • Apple M2 (Mac14,2) (38 fixtures): J2KSwift+inproc 28, Kakadu 10
  • Apple M4 (Mac16,10) (38 fixtures): J2KSwift+inproc 31, Kakadu 7

DECODE:

  • Apple M2 (Mac14,2) (38 fixtures): J2KSwift+inproc 25, Kakadu 7, Grok 6
  • Apple M4 (Mac16,10) (38 fixtures): J2KSwift+inproc 29, Grok 8, Kakadu 1

Per-fixture ENCODE — full ms + speedup matrix

Fixture Source J2KSwift+inproc Apple M2 J2KSwift+inproc Apple M4 J2KSwift+inproc Apple M4/Apple M2 Grok Apple M2 Grok Apple M4 Grok Apple M4/Apple M2 Kakadu Apple M2 Kakadu Apple M4 Kakadu Apple M4/Apple M2 OpenJPH Apple M2 OpenJPH Apple M4 OpenJPH Apple M4/Apple M2
MR-small 180² real 0.77 0.34 2.28× 9.49 4.49 2.11× 4.48 4.45 1.01× 9.49 4.52 2.10×
MR 886² real 2.19 1.20 1.82× 19.53 9.55 2.05× 4.47 4.48 1.00× 9.47 9.49 1.00×
CT 512² (1) real 2.38 1.33 1.79× 9.49 9.59 0.99× 4.46 4.50 0.99× 9.46 9.52 0.99×
CT 512² (2) real 2.23 1.30 1.72× 9.47 9.58 0.99× 4.48 4.60 0.97× 9.48 9.38 1.01×
XA 1024² real 5.52 3.37 1.64× 19.58 9.55 2.05× 9.52 4.53 2.10× 39.53 19.57 2.02×
PX 2459×1316 real 17.71 10.14 1.75× 39.73 19.76 2.01× 19.66 9.60 2.05× 76.64 76.33 1.00×
DX 2800×2288 real 41.88 22.48 1.86× 75.66 43.38 1.74× 39.78 19.99 1.99× 131.76 131.28 1.00×
NM 256² (synth) synth 0.89 0.51 1.72× 9.46 4.46 2.12× 4.47 4.49 1.00× 9.48 4.57 2.07×
MR 256² (synth) synth 0.91 0.53 1.71× 9.48 4.41 2.15× 4.46 4.50 0.99× 9.47 4.48 2.11×
CT 384² (synth) synth 1.58 0.86 1.84× 9.49 9.51 1.00× 4.50 4.49 1.00× 9.48 9.53 0.99×
MR 512² (synth) synth 2.73 1.35 2.03× 9.49 9.58 0.99× 4.49 4.50 1.00× 9.44 9.55 0.99×
CT 768² (synth) synth 3.96 2.13 1.87× 19.56 9.53 2.05× 9.54 4.51 2.11× 19.48 19.54 1.00×
XA 800² (synth) synth 3.80 2.31 1.65× 19.57 9.61 2.04× 9.51 4.54 2.09× 19.57 19.61 1.00×
PX 1024×800 (synth) synth 4.85 2.81 1.72× 19.49 9.59 2.03× 9.51 4.48 2.12× 19.49 19.53 1.00×
CT 1024² (synth) synth 5.91 3.81 1.55× 19.44 9.60 2.03× 9.56 9.61 0.99× 39.56 19.54 2.03×
MR 1024² (synth) synth 7.10 3.84 1.85× 19.43 9.58 2.03× 9.53 9.62 0.99× 39.53 19.67 2.01×
DX 1024² (synth) synth 6.39 3.55 1.80× 19.59 9.62 2.04× 9.53 9.56 1.00× 39.52 19.56 2.02×
CR 1024² (synth) synth 6.23 3.71 1.68× 19.60 9.60 2.04× 9.55 9.53 1.00× 39.54 19.55 2.02×
MG 1024×1280 (synth) synth 7.51 4.22 1.78× 19.57 9.59 2.04× 9.54 9.60 0.99× 39.50 19.75 2.00×
XA 1280² (synth) synth 10.16 5.37 1.89× 19.56 19.70 0.99× 9.60 9.61 1.00× 39.54 39.62 1.00×
MR 174×192 (small, real) medical-real 0.55 0.38 1.46× 9.47 4.59 2.06× 4.46 4.55 0.98× 4.40 4.50 0.98×
CT 512×512 (small, real) medical-real 2.11 1.22 1.73× 9.48 9.52 1.00× 4.48 4.53 0.99× 9.47 9.54 0.99×
CT 512×512 (mid, real) medical-real 2.13 1.28 1.65× 9.47 9.55 0.99× 4.48 4.51 0.99× 9.44 9.53 0.99×
CT 512×512 (large, real) medical-real 2.05 1.24 1.66× 9.48 9.50 1.00× 4.47 4.55 0.98× 9.46 9.56 0.99×
MR 512×512 (mid, real) medical-real 2.29 1.19 1.92× 9.49 9.55 0.99× 4.46 4.56 0.98× 9.46 9.55 0.99×
MR 512×512 (large, real) medical-real 2.22 1.35 1.65× 9.47 9.63 0.98× 4.49 4.63 0.97× 9.47 9.54 0.99×
XA 1024×1024 (small, real) medical-real 4.98 3.05 1.63× 19.51 9.64 2.02× 9.50 4.51 2.11× 19.51 19.52 1.00×
XA 1024×1024 (mid, real) medical-real 4.94 3.18 1.55× 19.56 9.54 2.05× 9.49 4.53 2.10× 19.48 19.48 1.00×
XA 1024×1024 (large, real) medical-real 4.79 3.03 1.58× 19.25 9.60 2.00× 9.49 4.51 2.10× 19.48 19.51 1.00×
PX 2459×1316 (small, real) medical-real 14.35 8.32 1.73× 39.24 19.71 1.99× 9.26 10.36 0.89× 76.62 39.61 1.93×
PX 2793×1316 (mid, real) medical-real 16.57 9.52 1.74× 39.22 19.74 1.99× 20.44 10.63 1.92× 76.61 76.77 1.00×
PX 2812×1316 (large, real) medical-real 16.81 9.41 1.79× 39.72 19.78 2.01× 9.84 10.57 0.93× 76.70 76.76 1.00×
DX 2224×2798 (small, real) medical-real 34.48 17.29 1.99× 76.60 41.66 1.84× 19.83 10.93 1.81× 130.76 76.72 1.70×
DX 2800×2288 (mid, real) medical-real 34.43 19.05 1.81× 75.06 40.14 1.87× 19.76 20.15 0.98× 131.89 76.65 1.72×
DX 2544×3056 (large, real) medical-real 42.90 23.31 1.84× 76.20 39.48 1.93× 19.99 19.89 1.00× 131.73 131.97 1.00×
MG 3516×4784 (small, real) medical-real 62.93 36.62 1.72× 131.67 82.77 1.59× 39.11 20.33 1.92× 185.99 131.72 1.41×
MG 3518×4784 (mid, real) medical-real 64.05 38.76 1.65× 129.70 81.16 1.60× 39.74 19.92 2.00× 131.79 130.67 1.01×
MG 3521×4784 (large, real) medical-real 63.83 36.66 1.74× 126.86 82.79 1.53× 39.82 19.80 2.01× 185.14 131.72 1.41×

Per-fixture DECODE — full ms + speedup matrix

Fixture Source J2KSwift+inproc Apple M2 J2KSwift+inproc Apple M4 J2KSwift+inproc Apple M4/Apple M2 Grok Apple M2 Grok Apple M4 Grok Apple M4/Apple M2 Kakadu Apple M2 Kakadu Apple M4 Kakadu Apple M4/Apple M2 OpenJPH Apple M2 OpenJPH Apple M4 OpenJPH Apple M4/Apple M2
MR-small 180² real 0.70 0.38 1.83× 9.48 4.48 2.12× 4.45 4.51 0.99× 4.40 4.46 0.99×
MR 886² real 4.99 3.38 1.48× 9.51 9.59 0.99× 9.52 9.51 1.00× 9.43 9.48 0.99×
CT 512² (1) real 2.80 1.84 1.52× 9.50 9.55 0.99× 4.42 4.48 0.99× 9.44 9.53 0.99×
CT 512² (2) real 2.62 1.79 1.46× 9.49 9.58 0.99× 4.44 4.54 0.98× 9.50 9.54 1.00×
XA 1024² real 7.94 5.44 1.46× 9.53 9.73 0.98× 9.52 9.59 0.99× 19.54 19.55 1.00×
PX 2459×1316 real 30.41 18.31 1.66× 19.70 19.72 1.00× 19.54 19.85 0.98× 75.23 39.71 1.89×
DX 2800×2288 real 59.45 32.37 1.84× 39.79 19.80 2.01× 39.80 39.82 1.00× 131.95 75.41 1.75×
NM 256² (synth) synth 0.95 0.57 1.67× 9.47 4.48 2.11× 4.46 4.50 0.99× 9.45 4.53 2.09×
MR 256² (synth) synth 0.99 0.61 1.64× 9.48 4.49 2.11× 4.47 4.49 1.00× 9.46 4.50 2.10×
CT 384² (synth) synth 1.75 1.10 1.59× 9.50 4.46 2.13× 4.47 4.47 1.00× 9.45 4.43 2.13×
MR 512² (synth) synth 3.05 1.84 1.66× 9.48 9.54 0.99× 4.40 4.55 0.97× 9.46 9.51 0.99×
CT 768² (synth) synth 5.41 3.30 1.64× 9.49 9.72 0.98× 9.49 9.69 0.98× 19.52 9.63 2.03×
XA 800² (synth) synth 5.59 3.55 1.58× 9.56 9.54 1.00× 9.36 9.66 0.97× 19.50 9.56 2.04×
PX 1024×800 (synth) synth 7.14 4.54 1.57× 9.53 9.72 0.98× 9.53 9.61 0.99× 19.52 19.64 0.99×
CT 1024² (synth) synth 8.50 5.98 1.42× 9.52 9.68 0.98× 9.52 9.77 0.97× 19.52 19.60 1.00×
MR 1024² (synth) synth 9.19 5.69 1.61× 9.48 9.79 0.97× 9.51 9.56 0.99× 19.50 19.55 1.00×
DX 1024² (synth) synth 15.23 5.70 2.67× 9.52 9.95 0.96× 9.52 9.57 0.99× 19.49 19.54 1.00×
CR 1024² (synth) synth 8.87 5.73 1.55× 9.50 9.66 0.98× 9.54 9.79 0.97× 19.51 19.56 1.00×
MG 1024×1280 (synth) synth 10.52 7.25 1.45× 19.45 9.75 1.99× 9.50 9.68 0.98× 19.54 19.62 1.00×
XA 1280² (synth) synth 12.78 8.89 1.44× 19.55 9.69 2.02× 19.59 9.58 2.04× 39.57 19.58 2.02×
MR 174×192 (small, real) medical-real 0.58 0.38 1.52× 9.47 4.52 2.10× 4.46 4.50 0.99× 9.50 4.50 2.11×
CT 512×512 (small, real) medical-real 2.56 1.71 1.50× 9.48 9.59 0.99× 4.47 4.62 0.97× 9.46 9.47 1.00×
CT 512×512 (mid, real) medical-real 2.50 1.72 1.46× 9.46 9.55 0.99× 4.41 4.53 0.97× 9.48 9.50 1.00×
CT 512×512 (large, real) medical-real 2.49 1.78 1.40× 9.47 9.55 0.99× 4.45 4.54 0.98× 9.47 9.59 0.99×
MR 512×512 (mid, real) medical-real 2.57 1.86 1.38× 9.46 9.57 0.99× 4.46 4.54 0.98× 9.45 9.46 1.00×
MR 512×512 (large, real) medical-real 2.66 1.79 1.49× 9.46 9.58 0.99× 4.47 4.49 1.00× 9.53 9.54 1.00×
XA 1024×1024 (small, real) medical-real 7.41 5.12 1.45× 9.51 9.85 0.97× 9.49 9.55 0.99× 19.51 19.64 0.99×
XA 1024×1024 (mid, real) medical-real 7.58 5.28 1.44× 9.53 9.68 0.98× 9.48 9.56 0.99× 19.49 19.52 1.00×
XA 1024×1024 (large, real) medical-real 7.47 5.16 1.45× 9.54 9.72 0.98× 9.50 9.56 0.99× 18.99 19.52 0.97×
PX 2459×1316 (small, real) medical-real 27.81 17.10 1.63× 19.19 19.75 0.97× 19.62 19.68 1.00× 76.47 39.22 1.95×
PX 2793×1316 (mid, real) medical-real 30.75 19.61 1.57× 19.75 19.41 1.02× 19.59 19.78 0.99× 76.74 39.68 1.93×
PX 2812×1316 (large, real) medical-real 31.12 21.51 1.45× 19.72 19.77 1.00× 19.64 19.67 1.00× 76.23 38.75 1.97×
DX 2224×2798 (small, real) medical-real 49.78 30.40 1.64× 39.68 19.78 2.01× 39.46 19.82 1.99× 76.80 76.80 1.00×
DX 2800×2288 (mid, real) medical-real 50.79 33.73 1.51× 39.81 19.78 2.01× 39.66 35.88 1.11× 131.38 76.54 1.72×
DX 2544×3056 (large, real) medical-real 63.73 39.32 1.62× 38.79 19.81 1.96× 39.81 39.15 1.02× 131.23 76.50 1.72×
MG 3516×4784 (small, real) medical-real 144.68 87.59 1.65× 75.10 40.48 1.86× 73.62 76.33 0.96× 182.91 130.78 1.40×
MG 3518×4784 (mid, real) medical-real 136.20 63.96 2.13× 76.17 39.85 1.91× 76.51 76.89 1.00× 131.82 128.51 1.03×
MG 3521×4784 (large, real) medical-real 133.93 90.13 1.49× 76.62 39.07 1.96× 76.81 74.70 1.03× 186.64 127.98 1.46×

Encoded-byte parity

Cross-codec encoded sizes do not depend on host silicon — bytes are deterministic per (codec, source, config). To cross-check, compare the per-fixture .j2k/.jph byte counts in /tmp/j2kswift_warm_bench/ from each host's run; differences > 0 indicate a non-deterministic codec config or version drift.