Hacktoberfest 2026：メンテナが10月に向けて印を付けた、オープンで初心者向けの issue。 Hacktoberfest の issue を見る

Dense FP32 (and FP16) matmul on the Android JNI tier: attention runs scalar today

オープン
  • 何か問題がありますか？
#1,275 コメント 1 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

まだ誰も着手していません。

評価

難易度
4/5
見積もり時間
3〜5日
初心者へのやさしさ
35/100
issue の種類
機能追加
明瞭さ
明確に書かれている
活発さ
活発
技術スタック
android, c, kotlin
領域
machine-learning, mobile, performance

調査の方向性

Start by reading fp32_matmul.c, fp16_matmul.c, and the JNI exports in native/skainet_jni.c; then trace how JniKernelProvider registers kernels and review JniKernelParityTest and KernelSupportMatrixTest. Add the requested JNI matmul variants and registrations, verify scalar parity including offset and strided operands, regenerate the support matrix, and measure the whisper-tiny encoder on a device. Done means bit-identical attention-shape results and at least 5× speedup on a Pixel 7 Pro/8a-class phone.

索引モデルが issue の本文から書いたものです。

説明

compute-backend enhancement skill:android skill:native

Context

A fully offline Android transcription app built on LiteRT (Whisper large-v3-turbo split encoder/decoder on OpenCL FP16, Parakeet TDT 0.6B, Silero VAD on ONNX Runtime) was evaluated as a SKaiNET consumer. Its author's own measurements on a Pixel 7 Pro are the bar: LiteRT CPU XNNPACK transcribes a 14 s memo in 82.2 s (RTF 5.9); the GPU path needs 34.4 s.

Since 0.50.0 the Android JNI tier (skainet-backend-jni-cpu) serves every GGML quant format from mapped weights on NEON, and 0.52.0 made the dispatch self-installing. That covers the weight side of a Whisper encoder (Q8_0 from whisper.cpp GGUFs, 874 MB). What it does not cover is the activation side.

Gap

The generated kernel support matrix (docs/.../reference/kernel-support-matrix.adoc, 0.54.0) lists Float32 and BFloat16 as scalar on Android, and native/skainet_jni.c exports only q40/q4k/q50/q51/q5k/q6k/q80 matmul entries plus the ternary gemv. A Whisper large-v3-turbo encoder issues, per layer, 20 heads of QKᵀ ([1500,64]×[64,1500]) and AV ([1500,1500]×[1500,64]) as FP32 activation × activation matmuls, 32 layers deep, plus two conv1d stem layers. On Android all of that runs through the scalar Kotlin path, so the quantized weight kernels cannot make the encoder fast on their own.

fp32_matmul.c already exists in skainet-backend-native-cpu (aarch64-verified, see #920) and the JVM reaches it through FFM; Android does not.

Scope

  • JNI entries for dense FP32 matmul, including a batched variant and a transposed-right-operand variant (A × Bᵀ) so attention does not pay a transpose copy per head, using the existing skainet_row_threads pool.
  • FP16 weight matmul on the JNI tier, since whisper.cpp ships F16 GGUFs and fp16_matmul.c is in tree (#885 tracks the FFM side of the same kernel).
  • JniKernelProvider registrations so KernelDispatch selects them on Android at the native priority.
  • Parity tests against ScalarFp32MatmulKernel in JniKernelParityTest, including offset/strided operands (the #1173 class of bug).
  • KernelSupportMatrixTest regenerated: Float32 and Float16 show native-jni on Android.
  • Device measurement with the M2-A5 harness: whisper-tiny encoder before/after.

Acceptance

  • Bit-identical results to the scalar path on device for the attention shapes above.
  • whisper-tiny.en encoder on a Pixel 7 Pro / 8a class phone at least 5× faster than the scalar Android baseline.

Related

  • #920 — NEON kernels to mobile (the JNI tier this extends)
  • #885 — FP16 native kernel on the FFM tier
  • #949 — the non-matmul Android overhead this issue does not address (see the companion issue on fused elementwise/norm kernels)
主要言語
Kotlin
スター
52
フォーク
15
平均マージ
1日 15時間
マージ済み PR（30日）
36

環境構築

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

SKaiNET-developers/SKaiNET のほかの issue

SKaiNET-developers/SKaiNET の issue をすべて見る

似ている issue

Kotlin の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。