Skip to content

Add built-in on-device vision, language, and inference#5467

Open
shai-almog wants to merge 6 commits into
masterfrom
builtin-on-device-ai
Open

Add built-in on-device vision, language, and inference#5467
shai-almog wants to merge 6 commits into
masterfrom
builtin-on-device-ai

Conversation

@shai-almog

@shai-almog shai-almog commented Jul 25, 2026

Copy link
Copy Markdown
Collaborator

Summary

  • add vendor-neutral built-in APIs for on-device vision, language, and LiteRT/Core ML inference
  • add Android ML Kit/LiteRT implementations, iOS Vision/NaturalLanguage/Core ML implementations, and Mac Catalyst Vision support
  • centralize granular class- and method-level source/dependency selection in PlatformFeatureCatalog, shared with BuildDaemon
  • retire the small ML Kit and TFLite cn1libs while retaining large-model libraries such as Whisper and Stable Diffusion
  • document API availability, build hints, migration, and platform limitations
  • add focused core/catalog/builder tests plus HelloCodenameOne and Port Status coverage without screenshot requirements

Validation

  • Java 8 core compile: 1,615 sources
  • focused core API tests: 9 passed
  • platform feature catalog tests: 20 passed
  • focused builder selection tests: 24 passed before rebasing; the only rebase conflict was in Port Status coverage
  • Port Status validator: 170 tests, 57 features, 11 ports, 303 golden names
  • Port Status conformance tests: 9 passed
  • HelloCodenameOne shared-source Java compile: 196 sources
  • Android adapters compiled against the official ML Kit and LiteRT dependencies
  • iOS all-feature device Xcode build passed
  • Mac Catalyst default Apple Vision build passed

Platform note

Google ML Kit's MLImage dependency currently contains a device-only slice, so an arm64 iOS Simulator build that explicitly selects the ML Kit vision backend cannot link. Device builds pass, and the default Apple Vision path does not pull that pod into simulator builds.

Copilot AI review requested due to automatic review settings July 25, 2026 11:31
Copilot AI reviewed Jul 25, 2026

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot wasn't able to review this pull request because it exceeds the maximum number of files (300). Try reducing the number of changed files and requesting a review from Copilot again.

Copilot AI review requested due to automatic review settings July 25, 2026 11:46
Copilot AI reviewed Jul 25, 2026

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot wasn't able to review this pull request because it exceeds the maximum number of files (300). Try reducing the number of changed files and requesting a review from Copilot again.

@github-actions

github-actions Bot commented Jul 25, 2026

Copy link
Copy Markdown
Contributor

Cloudflare Preview

@github-actions

github-actions Bot commented Jul 25, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

Developer Guide build artifacts are available for download from this workflow run:

Developer Guide quality checks:

  • AsciiDoc linter: No issues found (report)
  • Vale: No alerts found (report)
  • Paragraph capitalization: No paragraph capitalization issues (report)
  • LanguageTool: No grammar matches (report)
  • Image references: No unused images detected (report)

Copilot AI review requested due to automatic review settings July 25, 2026 12:01
Copilot AI reviewed Jul 25, 2026

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot wasn't able to review this pull request because it exceeds the maximum number of files (300). Try reducing the number of changed files and requesting a review from Copilot again.

Copilot AI review requested due to automatic review settings July 25, 2026 12:20
Copilot AI reviewed Jul 25, 2026

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot wasn't able to review this pull request because it exceeds the maximum number of files (300). Try reducing the number of changed files and requesting a review from Copilot again.

Copilot AI review requested due to automatic review settings July 25, 2026 12:40
Copilot AI reviewed Jul 25, 2026

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot wasn't able to review this pull request because it exceeds the maximum number of files (300). Try reducing the number of changed files and requesting a review from Copilot again.

@shai-almog

shai-almog commented Jul 25, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator Author

Compared 151 screenshots: 151 matched.

Native Android coverage

  • 📊 Line coverage: 7.73% (7476/96721 lines covered) [HTML preview] (artifact android-coverage-report, jacocoAndroidReport/html/index.html)
    • Other counters: instruction 7.61% (39007/512565), branch 2.74% (1333/48583), complexity 3.07% (1600/52127), method 4.73% (1303/27567), class 9.61% (353/3672)
    • Lowest covered classes
      • kotlin.collections.kotlin.collections.ArraysKt___ArraysKt – 0.00% (0/6367 lines covered)
      • kotlin.collections.unsigned.kotlin.collections.unsigned.UArraysKt___UArraysKt – 0.00% (0/2384 lines covered)
      • org.jacoco.agent.rt.internal_0e20598.asm.org.jacoco.agent.rt.internal_0e20598.asm.ClassReader – 0.00% (0/1524 lines covered)
      • kotlin.collections.kotlin.collections.CollectionsKt___CollectionsKt – 0.00% (0/1187 lines covered)
      • org.jacoco.agent.rt.internal_0e20598.asm.org.jacoco.agent.rt.internal_0e20598.asm.MethodWriter – 0.00% (0/922 lines covered)
      • kotlin.sequences.kotlin.sequences.SequencesKt___SequencesKt – 0.00% (0/736 lines covered)
      • com.google.common.cache.com.google.common.cache.LocalCache$Segment – 0.00% (0/726 lines covered)
      • okio.okio.Buffer – 0.00% (0/687 lines covered)
      • kotlin.text.kotlin.text.StringsKt___StringsKt – 0.00% (0/625 lines covered)
      • org.jacoco.agent.rt.internal_0e20598.asm.org.jacoco.agent.rt.internal_0e20598.asm.Frame – 0.00% (0/570 lines covered)

✅ Native Android screenshot tests passed.

Native Android coverage

  • 📊 Line coverage: 7.73% (7476/96721 lines covered) [HTML preview] (artifact android-coverage-report, jacocoAndroidReport/html/index.html)
    • Other counters: instruction 7.61% (39007/512565), branch 2.74% (1333/48583), complexity 3.07% (1600/52127), method 4.73% (1303/27567), class 9.61% (353/3672)
    • Lowest covered classes
      • kotlin.collections.kotlin.collections.ArraysKt___ArraysKt – 0.00% (0/6367 lines covered)
      • kotlin.collections.unsigned.kotlin.collections.unsigned.UArraysKt___UArraysKt – 0.00% (0/2384 lines covered)
      • org.jacoco.agent.rt.internal_0e20598.asm.org.jacoco.agent.rt.internal_0e20598.asm.ClassReader – 0.00% (0/1524 lines covered)
      • kotlin.collections.kotlin.collections.CollectionsKt___CollectionsKt – 0.00% (0/1187 lines covered)
      • org.jacoco.agent.rt.internal_0e20598.asm.org.jacoco.agent.rt.internal_0e20598.asm.MethodWriter – 0.00% (0/922 lines covered)
      • kotlin.sequences.kotlin.sequences.SequencesKt___SequencesKt – 0.00% (0/736 lines covered)
      • com.google.common.cache.com.google.common.cache.LocalCache$Segment – 0.00% (0/726 lines covered)
      • okio.okio.Buffer – 0.00% (0/687 lines covered)
      • kotlin.text.kotlin.text.StringsKt___StringsKt – 0.00% (0/625 lines covered)
      • org.jacoco.agent.rt.internal_0e20598.asm.org.jacoco.agent.rt.internal_0e20598.asm.Frame – 0.00% (0/570 lines covered)

Benchmark Results

Detailed Performance Metrics

Metric Duration
SIMD kernel backend scalar fallback (no native SIMD)
SIMD int-add (64K x300) java 162ms / native 137ms = 1.1x speedup
SIMD float-mul (64K x300) java 235ms / native 124ms = 1.8x speedup
SIMD kernel correctness PASS (native result == scalar reference)
Base64 payload size 8192 bytes
Base64 benchmark iterations 6000
Base64 SIMD byte path gated to scalar (CPU autovectorizes scalar; explicit SIMD not beneficial here)
Base64 CN1 encode 78.000 ms
Base64 CN1 decode 88.000 ms
Base64 native encode 366.000 ms
Base64 encode ratio (CN1/native) 0.213x (78.7% faster)
Base64 native decode 279.000 ms
Base64 decode ratio (CN1/native) 0.315x (68.5% faster)
Image encode benchmark status skipped (SIMD unsupported)

@shai-almog

shai-almog commented Jul 25, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator Author

Compared 144 screenshots: 144 matched.
✅ Native Apple TV (tvOS, Metal) screenshot tests passed.

@shai-almog

shai-almog commented Jul 25, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator Author

Compared 146 screenshots: 146 matched.
Native Windows port (x64 / Intel-AMD): full hellocodenameone screenshot suite rendered offscreen with Direct2D/DirectWrite, plus the real benchmarks (base64 native/CN1/SIMD, image createMask/applyMask/modifyAlpha/PNG/JPEG, SSE2 SIMD kernels). Compared against the in-repo baseline in scripts/windows/screenshots.

Benchmark Results

Detailed Performance Metrics

Metric Duration
SIMD kernel backend SSE2 (x64) / NEON (arm64) native kernels
SIMD int-add (64K x300) java 62ms / native 5ms = 12.4x speedup
SIMD float-mul (64K x300) java 63ms / native 4ms = 15.7x speedup
SIMD kernel correctness PASS (native result == scalar reference)
Base64 native bridge unavailable (CN1 + SIMD + image benchmarks only)
Base64 payload size 8192 bytes
Base64 benchmark iterations 6000
Base64 SIMD byte path gated to scalar (CPU autovectorizes scalar; explicit SIMD not beneficial here)
Base64 CN1 encode 255.000 ms
Base64 CN1 decode 134.000 ms
Base64 SIMD encode 102.000 ms
Base64 encode ratio (SIMD/CN1) 0.400x (60.0% faster)
Base64 SIMD decode 93.000 ms
Base64 decode ratio (SIMD/CN1) 0.694x (30.6% faster)
Image encode benchmark iterations 100
Image createMask (SIMD off) 27.000 ms
Image createMask (SIMD on) 21.000 ms
Image createMask ratio (SIMD on/off) 0.778x (22.2% faster)
Image applyMask (SIMD off) 59.000 ms
Image applyMask (SIMD on) 163.000 ms
Image applyMask ratio (SIMD on/off) 2.763x (176.3% slower)
Image modifyAlpha (SIMD off) 65.000 ms
Image modifyAlpha (SIMD on) 57.000 ms
Image modifyAlpha ratio (SIMD on/off) 0.877x (12.3% faster)
Image modifyAlpha removeColor (SIMD off) 73.000 ms
Image modifyAlpha removeColor (SIMD on) 66.000 ms
Image modifyAlpha removeColor ratio (SIMD on/off) 0.904x (9.6% faster)

@shai-almog

shai-almog commented Jul 25, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator Author

Compared 146 screenshots: 146 matched.
Native Windows port, REAL shipping pipeline: the hellocodenameone screenshot suite rendered by a binary CROSS-COMPILED on Linux (clang-cl + xwin, WebView2 linked) and RUN on a Windows x64 runner. Compared against the in-repo baseline in scripts/windows/screenshots.

Benchmark Results

Detailed Performance Metrics

Metric Duration
SIMD kernel backend SSE2 (x64) / NEON (arm64) native kernels
SIMD int-add (64K x300) java 46ms / native 3ms = 15.3x speedup
SIMD float-mul (64K x300) java 45ms / native 3ms = 15.0x speedup
SIMD kernel correctness PASS (native result == scalar reference)
Base64 native bridge unavailable (CN1 + SIMD + image benchmarks only)
Base64 payload size 8192 bytes
Base64 benchmark iterations 6000
Base64 SIMD byte path gated to scalar (CPU autovectorizes scalar; explicit SIMD not beneficial here)
Base64 CN1 encode 147.000 ms
Base64 CN1 decode 103.000 ms
Base64 SIMD encode 78.000 ms
Base64 encode ratio (SIMD/CN1) 0.531x (46.9% faster)
Base64 SIMD decode 63.000 ms
Base64 decode ratio (SIMD/CN1) 0.612x (38.8% faster)
Image encode benchmark iterations 100
Image createMask (SIMD off) 17.000 ms
Image createMask (SIMD on) 15.000 ms
Image createMask ratio (SIMD on/off) 0.882x (11.8% faster)
Image applyMask (SIMD off) 33.000 ms
Image applyMask (SIMD on) 137.000 ms
Image applyMask ratio (SIMD on/off) 4.152x (315.2% slower)
Image modifyAlpha (SIMD off) 34.000 ms
Image modifyAlpha (SIMD on) 33.000 ms
Image modifyAlpha ratio (SIMD on/off) 0.971x (2.9% faster)
Image modifyAlpha removeColor (SIMD off) 49.000 ms
Image modifyAlpha removeColor (SIMD on) 37.000 ms
Image modifyAlpha removeColor ratio (SIMD on/off) 0.755x (24.5% faster)

@shai-almog

shai-almog commented Jul 25, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator Author

Compared 146 screenshots: 146 matched.
Native Windows port (arm64 / Apple Silicon - Arm): full hellocodenameone screenshot suite rendered offscreen with Direct2D/DirectWrite, plus the real benchmarks (base64 native/CN1/SIMD, image createMask/applyMask/modifyAlpha/PNG/JPEG, NEON SIMD kernels). Compared against the in-repo baseline in scripts/windows/screenshots.

Benchmark Results

Detailed Performance Metrics

Metric Duration
SIMD kernel backend SSE2 (x64) / NEON (arm64) native kernels
SIMD int-add (64K x300) java 56ms / native 3ms = 18.6x speedup
SIMD float-mul (64K x300) java 55ms / native 3ms = 18.3x speedup
SIMD kernel correctness PASS (native result == scalar reference)
Base64 native bridge unavailable (CN1 + SIMD + image benchmarks only)
Base64 payload size 8192 bytes
Base64 benchmark iterations 6000
Base64 SIMD byte path gated to scalar (CPU autovectorizes scalar; explicit SIMD not beneficial here)
Base64 CN1 encode 246.000 ms
Base64 CN1 decode 127.000 ms
Base64 SIMD encode 64.000 ms
Base64 encode ratio (SIMD/CN1) 0.260x (74.0% faster)
Base64 SIMD decode 59.000 ms
Base64 decode ratio (SIMD/CN1) 0.465x (53.5% faster)
Image encode benchmark iterations 100
Image createMask (SIMD off) 11.000 ms
Image createMask (SIMD on) 8.000 ms
Image createMask ratio (SIMD on/off) 0.727x (27.3% faster)
Image applyMask (SIMD off) 26.000 ms
Image applyMask (SIMD on) 20.000 ms
Image applyMask ratio (SIMD on/off) 0.769x (23.1% faster)
Image modifyAlpha (SIMD off) 16.000 ms
Image modifyAlpha (SIMD on) 12.000 ms
Image modifyAlpha ratio (SIMD on/off) 0.750x (25.0% faster)
Image modifyAlpha removeColor (SIMD off) 18.000 ms
Image modifyAlpha removeColor (SIMD on) 12.000 ms
Image modifyAlpha removeColor ratio (SIMD on/off) 0.667x (33.3% faster)

@shai-almog

shai-almog commented Jul 25, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator Author

Compared 181 screenshots: 181 matched.
✅ JavaScript-port screenshot tests passed.

Copilot AI review requested due to automatic review settings July 25, 2026 13:27
Copilot AI reviewed Jul 25, 2026

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot wasn't able to review this pull request because it exceeds the maximum number of files (300). Try reducing the number of changed files and requesting a review from Copilot again.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

Copilot code review Copilot Copilot left review comments

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@shai-almog