release: prepare 0.39.1

CHANGELOG.md

@@ -2,13 +2,24 @@
 
 ## [Unreleased]
 
+## [0.39.1] - 2026-08-11
+
+Headline: **the packed Q5 gap is closed, and eager overhead off the JVM is gone.**
+Native Q5_0/Q5_1 matmul kernels land in all three native tiers — FFM (JVM),
+Kotlin/Native, and the Android JNI bridge — unblocking the packed `NATIVE_OPTIMIZED`
+path for Q5_1-quantized checkpoints such as `functiongemma-270m`. On top of that,
+the eager CPU ops gain primitive FP32 fast paths, removing the per-element
+allocation/boxing overhead that dominated on-device LLM decode (83% of end-to-end
+time on a Pixel 8a even with NEON matmul).
+
 ### Documentation
 
 - **README points LLM users to SKaiNET-transformers**
   ([#923](https://github.com/SKaiNET-developers/SKaiNET/issues/923)): a callout under
   "Start in 5 minutes" says plainly that LLM inference lives in the
   SKaiNET-transformers repository — this repo is the engine underneath — and names
   the `sk.ainet.transformers` artifacts and BOM to depend on.
+
 ### Added
 
 - **Native Q5_0 / Q5_1 packed matmul kernels (FFM, Kotlin/Native, JNI).** 0.39.0 shipped
@@ -24,6 +35,7 @@
   `JniKernelProvider`), each with parity tests against the scalar references. Unblocks the
   packed Q5_1 path for `functiongemma-270m` "Q5_K_M" checkpoints (whose attention/FFN
   weights are Q5_1) under `NATIVE_OPTIMIZED` — see SKaiNET-transformers#170. (#708)
+
 ### Performance
 
 - **Primitive FP32 fast paths for the eager CPU ops** (`skainet-backend-cpu`,

README.md

@@ -53,7 +53,7 @@ Add the core dependencies (Gradle Kotlin DSL):
 ```kotlin
 dependencies {
     // Recommended: import the umbrella BOM and drop versions on the engine modules.
-    implementation(platform("sk.ainet:skainet-bom:0.39.0"))
+    implementation(platform("sk.ainet:skainet-bom:0.39.1"))
 
     implementation("sk.ainet.core:skainet-lang-core")
     implementation("sk.ainet.core:skainet-backend-cpu")
@@ -297,7 +297,13 @@ val withoutLabel = dataPipeline<RawDataset>()
 
 ---
 
-## What's New in 0.39.0
+## What's New in 0.39.1
+
+- **Native Q5_0/Q5_1 packed matmul kernels — FFM, Kotlin/Native, JNI.** 0.39.0 could *load* Q5_0/Q5_1 packed but had no native-tier kernels for them: the JVM fell back to Panama and Kotlin/Native/Android to scalar. New plain-NEON C kernels (no dotprod/i8mm requirement, so they run on every AArch64 core) are wired into all three native consumers, each with parity tests against the scalar references. This unblocks the packed `NATIVE_OPTIMIZED` path for Q5_1-quantized checkpoints such as `functiongemma-270m`.
+- **Eager CPU ops run primitive FP32 fast paths.** The generic per-element paths (index-array allocations, boxed accessors, dtype dispatch) dominated on-device LLM decode — 83% of end-to-end SmolLM2-135M decode time on a Pixel 8a was non-matmul overhead even with the NEON backend. Hot ops (arithmetic, activations, unary math, softmax/logSoftmax, reductions, concat, reshape) now run flat primitive loops over the dense `FloatArray` buffer, benefiting every non-JVM target — Android, Kotlin/Native, JS/Wasm. `DirectCpuExecutionContext.ops` is also cached instead of rebuilt per access.
+- **README points LLM users to SKaiNET-transformers.** A callout under "Start in 5 minutes" makes clear that LLM inference lives in the SKaiNET-transformers repository — this repo is the engine underneath.
+
+### Previously, in 0.39.0
 
 - **On-device AI on Android — a NEON kernel backend.** New `skainet-backend-jni-cpu` module: the hand-tuned ARM matmul kernels reach Android through a JNI bridge (ART has no `java.lang.foreign`, so the FFM provider can never run there). Two `.so` tiers are built from the same sources and selected at load time from `/proc/cpuinfo` — a baseline `armv8-a` build that runs on every 64-bit core, and an `armv8.2-a+dotprod` build for the `vdotq_s32` Q4_K/Q6_K paths — so a single artifact is safe from Cortex-A53 up. Measured on a Pixel 8a: **~24 tok/s** SmolLM2-135M Q8_0 decode versus ~3.8 scalar (6.4x), clearing the on-device usability bar. The provider auto-registers via `ServiceLoader`; an app just adds the AAR.
 - **Android GGUF loading no longer OOMs.** `createRandomAccessSource` returned `null` on Android, forcing every model load through a full-file heap read that exhausted the ART heap on real devices. It now streams via positional `FileChannel` reads across `skainet-io-gguf` / `-safetensors` / `-onnx`.
@@ -342,6 +348,10 @@ We love contributions! Whether it's a new operator, documentation, or a bug fix:
 
 Browse the full codebase documentation on [DeepWiki](https://deepwiki.com/SKaiNET-developers/SKaiNET).
 
+### Contributors (0.39.1)
+
+- **Michal Harakal** ([@michalharakal](https://github.com/michalharakal)) — native Q5_0/Q5_1 packed matmul kernels across FFM, Kotlin/Native, and JNI (#708), primitive FP32 fast paths for the eager CPU ops (#949), README pointer to SKaiNET-transformers (#923)
+
 ### Contributors (0.39.0)
 
 - **Michal Harakal** ([@michalharakal](https://github.com/michalharakal)) — Android JNI NEON kernel backend with runtime dotprod dispatch (#943, #945), Android `createRandomAccessSource` streaming loads (#922), cinterop klib archive embedding (#942), Q4_0 NEON kernel (#939), GGUF loader fail-fast (#919), tensor-storage correctness fixes (#927, #928, #929, #930, #931), AAR release publishing (#947)

docs/antora.yml

@@ -15,7 +15,7 @@ asciidoc:
     framework_name: SKaiNET
     # Current SKaiNET release — bump once per release; referenced as
     # {skainet_version} in dependency snippets (blocks need subs="attributes+").
-    skainet_version: 0.39.0
+    skainet_version: 0.39.1
     ksp_version: 2.2.21-2.0.5
     dokka_version: 2.1.0
     asciidoctorj_version: 3.0.0

docs/modules/ROOT/pages/reference/kernel-support-matrix.adoc

@@ -1,7 +1,7 @@
 = Kernel × platform support matrix
 :description: Which compute-kernel provider serves each weight format on each KMP target.
 
-Generated from `kernel-support.json` (version `0.39.0`) by `KernelSupportMatrixTest` — registry introspection of the registered `KernelProvider` implementations. Do not edit by hand; run `./gradlew generateKernelMatrix` to refresh.
+Generated from `kernel-support.json` (version `0.39.1`) by `KernelSupportMatrixTest` — registry introspection of the registered `KernelProvider` implementations. Do not edit by hand; run `./gradlew generateKernelMatrix` to refresh.
 
 Each cell is the best (highest-priority) provider that serves `Float32 × format` `matmul` on that platform: *native-ffm* (100) → *panama-vector* (50) → *scalar* (0). An empty cell (`—`) means no provider carries a kernel there (the format is dequant-to-FP32 only).
 
@@ -16,8 +16,8 @@ Each cell is the best (highest-priority) provider that serves `Float32 × format
 | `Q4_K` | native-ffm | native-jni | scalar | scalar | scalar 
 | `Q6_K` | panama-vector | native-jni | scalar | scalar | scalar 
 | `Q5_K` | native-ffm | native-jni | scalar | scalar | scalar 
-| `Q5_1` | panama-vector | panama-vector | scalar | scalar | scalar 
-| `Q5_0` | panama-vector | panama-vector | scalar | scalar | scalar 
+| `Q5_1` | native-ffm | native-jni | scalar | scalar | scalar 
+| `Q5_0` | native-ffm | native-jni | scalar | scalar | scalar 
 |===
 
 See also the eager backends & kernels mindmap (`docs/eager-execution-backends-and-kernels.md`) for the narrative overview and gaps.

gradle.properties

@@ -1,5 +1,5 @@
 GROUP=sk.ainet.core
-VERSION_NAME=0.39.0
+VERSION_NAME=0.39.1
 POM_DESCRIPTION=SKaiNET
 
 POM_URL=https://github.com/SKaiNET-developers/skainet/