feat: make tensors resizeable + release v0.2.0

.github/workflows/ci.yaml

@@ -0,0 +1,166 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+env:
+  CARGO_TERM_COLOR: always
+  RUSTFLAGS: --deny warnings
+  RUSTDOCFLAGS: --deny warnings
+  SLANG_TAG: 2025.18.2
+
+jobs:
+  # Check formatting.
+  format:
+    name: Format
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Install Rust toolchain
+        uses: dtolnay/rust-toolchain@stable
+        with:
+          components: rustfmt
+
+      - name: Run cargo fmt
+        run: cargo fmt --all -- --check
+  setup-slang:
+    strategy:
+      matrix:
+        os: [ubuntu-latest]
+    runs-on: ${{ matrix.os }}
+    outputs:
+      slang-dir: ${{ steps.setup.outputs.slang-dir }} # Pass SLANG_DIR to dependent jobs
+      slang-cache-key: ${{ steps.setup.outputs.slang-cache-key }} # Pass SLANG_DIR to dependent jobs
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Cache Slang
+        id: cache-slang
+        uses: actions/cache/restore@v4 # Restore first
+        with:
+          path: |
+            ~/.cache/slang  # Matches script's default OUTPUT_DIR
+          key: slang-v$SLANG_TAG-${{ runner.os }}-${{ runner.arch }}
+
+      - name: Setup Slang
+        id: setup
+        run: |
+          echo "version=$SLANG_TAG" >> $GITHUB_OUTPUT  # Output for cache key
+          SLANG_DIR=$(./.github/workflows/download_slang.sh --version $SLANG_TAG | grep '^SLANG_DIR=' | cut -d'=' -f2-)
+          echo "slang-dir=$SLANG_DIR" >> $GITHUB_OUTPUT  # Output for dependents
+          echo "slang-cache-key=slang-v$SLANG_TAG-${{ runner.os }}-${{ runner.arch }}" >> $GITHUB_OUTPUT
+          echo "SLANG_DIR=$SLANG_DIR" >> $GITHUB_ENV  # For this job if needed
+
+      - name: Save Slang Cache
+        if: steps.cache-slang.outputs.cache-hit != 'true' # Only save on miss
+        uses: actions/cache/save@v4
+        with:
+          path: ~/.cache/slang
+          key: ${{ steps.setup.outputs.slang-cache-key }}
+  # Run clippy lints.
+  clippy:
+    needs: setup-slang # Depends on setup-slang
+    name: Clippy
+    runs-on: ubuntu-latest
+    env:
+      SLANG_DIR: ${{ needs.setup-slang.outputs.slang-dir }}
+    timeout-minutes: 30
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Install Rust toolchain
+        uses: dtolnay/rust-toolchain@stable
+        with:
+          components: clippy
+
+      - name: Install dependencies
+        run: sudo apt-get update; sudo apt-get install --no-install-recommends build-essential curl wget file libssl-dev
+
+      - name: Retrieve Cache for Slang
+        uses: actions/cache/restore@v4
+        with:
+          path: ~/.cache/slang
+          key: ${{ needs.setup-slang.outputs.slang-cache-key }}
+
+      - name: Populate target directory from cache
+        uses: Leafwing-Studios/cargo-cache@v2
+        with:
+          sweep-cache: true
+
+      - name: Run clippy lints
+        run: SLANG_DIR=$SLANG_DIR cargo clippy --locked --workspace --all-targets -- --deny warnings
+
+  # Check documentation.
+  doc:
+    needs: setup-slang # Depends on setup-slang
+    name: Docs
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+    env:
+      SLANG_DIR: ${{ needs.setup-slang.outputs.slang-dir }}
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Install Rust toolchain
+        uses: dtolnay/rust-toolchain@stable
+
+      - name: Install dependencies
+        run: sudo apt-get update; sudo apt-get install --no-install-recommends build-essential curl wget file libssl-dev
+
+      - name: Retrieve Cache for Slang
+        uses: actions/cache/restore@v4
+        with:
+          path: ~/.cache/slang
+          key: ${{ needs.setup-slang.outputs.slang-cache-key }}
+
+      - name: Populate target directory from cache
+        uses: Leafwing-Studios/cargo-cache@v2
+        with:
+          sweep-cache: true
+
+      - name: Check documentation
+        run: SLANG_DIR=$SLANG_DIR cargo doc --locked --workspace --document-private-items --no-deps
+    # Testing.
+  test:
+    needs: setup-slang # Depends on setup-slang
+    name: Tests
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+    env:
+      SLANG_DIR: ${{ needs.setup-slang.outputs.slang-dir }}
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Install Rust toolchain
+        uses: dtolnay/rust-toolchain@stable
+
+      - name: Install dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install --no-install-recommends -y \
+            build-essential curl wget file libssl-dev \
+            libegl1-mesa-dev libgl1-mesa-dri libxcb-xfixes0-dev mesa-vulkan-drivers
+
+      - name: Retrieve Cache for Slang
+        uses: actions/cache/restore@v4
+        with:
+          path: ~/.cache/slang
+          key: ${{ needs.setup-slang.outputs.slang-cache-key }}
+
+      - name: Populate target directory from cache
+        uses: Leafwing-Studios/cargo-cache@v2
+        with:
+          sweep-cache: true
+      - name: Run Cargo Tests
+        run: |
+          SLANG_DIR=$SLANG_DIR LIBGL_ALWAYS_SOFTWARE=1 cargo test --verbose

.github/workflows/download_slang.sh

@@ -0,0 +1,109 @@
+#!/bin/bash
+
+# Default values
+OS=""
+OUTPUT_DIR="$HOME/.cache/slang"
+SLANG_VERSION=""
+SLANG_TAG=""
+ASSET_SUFFIX=""
+SLANG_URL_BASE="https://github.com/shader-slang/slang/releases/download"
+
+# Help message
+usage() {
+    echo "Usage: $0 [--os <linux|macos|macos-arm64|windows>] [--output-dir <path>] [--version <version>]"
+    echo "  --os: Target OS (default: auto-detect from current platform)"
+    echo "  --output-dir: Directory to extract Slang (default: ~/.cache/slang)"
+    echo "  --version: Slang version (e.g., 2025.18.2, default: latest)"
+    echo "Example: $0 --os linux --output-dir /tmp/slang"
+}
+
+# Parse arguments
+while [[ "$#" -gt 0 ]]; do
+    case $1 in
+        --os) OS="$2"; shift ;;
+        --output-dir) export OUTPUT_DIR="$2"; shift ;;
+        --version) export SLANG_VERSION="$2"; shift ;;
+        *) usage ; exit 1 ;;
+    esac
+    shift
+done
+
+# Detect OS if not specified
+if [[ -z "$OS" ]]; then
+    case "$(uname -s)" in
+        Linux*) OS="linux" ;;
+        Darwin*)
+            if [[ "$(uname -m)" == "arm64" ]]; then
+                OS="macos-aarch64"
+            else
+                OS="macos"
+            fi
+            ;;
+        CYGWIN*|MINGW*|MSYS*) OS="windows" ;;
+        *) echo "Error: Unable to detect OS. Specify --os (linux, macos, macos-arm64, windows)"; exit 1 ;;
+    esac
+fi
+
+# Determine asset suffix based on OS
+case "$OS" in
+    linux) ASSET_SUFFIX="linux-x86_64.zip" ;;
+    macos) ASSET_SUFFIX="macos-x86_64.zip" ;;
+    macos-aarch64) ASSET_SUFFIX="macos-aarch64.zip" ;;
+    windows) ASSET_SUFFIX="windows-x86_64.zip" ;;
+    *) echo "Error: Unsupported OS: $OS"; exit 1 ;;
+esac
+
+# Get Slang version if not specified
+if [[ -z "$SLANG_VERSION" ]]; then
+    export SLANG_TAG=$(curl -s https://api.github.com/repos/shader-slang/slang/releases/latest | grep '"tag_name"' | sed -E 's/.*"([^"]+)".*/\1/')
+    export SLANG_VERSION=$(echo "$SLANG_TAG" | sed 's/v//')  # e.g., v2025.18.2 -> 2025.18.2
+else
+    export SLANG_TAG="v$SLANG_VERSION"
+fi
+
+if [[ -z "$SLANG_VERSION" ]]; then
+    echo "Error: Could not determine Slang version"
+    exit 1
+fi
+
+# Set up paths
+SLANG_DIR="$OUTPUT_DIR/slang-v$SLANG_VERSION-$OS"
+ZIP_URL="$SLANG_URL_BASE/$SLANG_TAG/slang-$SLANG_VERSION-$ASSET_SUFFIX"
+TEMP_ZIP="/tmp/slang-$SLANG_VERSION.zip"
+
+# Check if Slang is already extracted
+if [[ -d "$SLANG_DIR" ]] && [[ -f "$SLANG_DIR/bin/slangc" || -f "$SLANG_DIR/bin/slangc.exe" ]]; then
+    echo "Using existing Slang at $SLANG_DIR"
+    echo "SLANG_DIR=$SLANG_DIR"
+    exit 0
+fi
+
+# Download Slang release
+echo "Downloading Slang v$SLANG_VERSION for $OS from $ZIP_URL..."
+mkdir -p "$OUTPUT_DIR"
+curl -L -o "$TEMP_ZIP" "$ZIP_URL" || { echo "Error: Download failed for $ZIP_URL"; exit 1; }
+
+# Extract based on OS
+echo "Extracting to $SLANG_DIR..."
+if [[ "$OS" == "windows" ]]; then
+    # Windows: Assume 7z is available (or adjust for PowerShell/Expand-Archive)
+    7z x "$TEMP_ZIP" -o"$SLANG_DIR" -y > /dev/null || { echo "Error: Extraction failed"; rm -f "$TEMP_ZIP"; exit 1; }
+else
+    # Linux/macOS: Use unzip
+    unzip -q "$TEMP_ZIP" -d "$SLANG_DIR" || { echo "Error: Extraction failed"; rm -f "$TEMP_ZIP"; exit 1; }
+fi
+
+# Clean up
+rm -f "$TEMP_ZIP"
+
+# Verify extraction
+if [[ ! -f "$SLANG_DIR/bin/slangc" && ! -f "$SLANG_DIR/bin/slangc.exe" ]]; then
+    echo "Error: Extraction incomplete, slangc not found in $SLANG_DIR/bin"
+    exit 1
+fi
+
+echo "Slang v$SLANG_VERSION extracted to $SLANG_DIR"
+echo "SLANG_DIR=$SLANG_DIR"
+
+# For use in calling script
+export SLANG_DIR

CHANGELOG.md

@@ -0,0 +1,4 @@
+# v0.2.0 (27 Oct. 2025)
+- Update to slang-hal 0.2.
+- Make rank-1 tensors resizeable.
+- Fix svd3.slang retuning incorrect results for near-identity matrices.

Cargo.toml

@@ -3,7 +3,7 @@ name = "stensor"
 authors = ["Sébastien Crozet <sebcrozet@dimforge.com>"]
 description = "Cross-platform GPU tensor library with Slang and Rust."
 repository = "https://github.com/dimforge/stensor"
-version = "0.1.1"
+version = "0.2.0"
 edition = "2024"
 license = "Apache-2.0"
 
@@ -12,15 +12,15 @@ cuda = [ "cudarc", "slang-hal/cuda" ]
 cublas = [ "slang-hal/cublas" ]
 
 [dependencies]
-wgpu = "26"
+wgpu = "27"
 encase = "0.12"
 bytemuck = "1"
 nalgebra = { version = "0.34", features = ["encase"] }
 
 cudarc = {  version = "0.16", optional = true }
 
-minislang = "0.1"
-slang-hal = { version = "0.1", features = ["derive"] }
+minislang = "0.2"
+slang-hal = { version = "0.2", features = ["derive"] }
 include_dir = "0.7"
 
 [dev-dependencies]

README.md

@@ -5,15 +5,15 @@ gpu". It aims (but it isn’t there yet) to expose linear algebra operations (in
 operations) as well as geometric types (quaternions, similarities, etc.) as Slang shaders and kernels.
 
 > **Warning**
-**stensor** is still very incomplete and under heavy development and is lacking many features.
+> **stensor** is still very incomplete and under heavy development and is lacking many features.
 
 See also the README of [slang-hal](https://github.com/dimforge/slang-hal/blob/main/README.md) for information on
 supported platforms.
 
 ### Using Slang
 
 In order to compile and run any slang project, be sure to define the `SLANG_DIR` environment variable:
-1. Download the Slang compiler libraries for your platform: https://github.com/shader-slang/slang/releases/tag/v2025.16
+1. Download the Slang compiler libraries for your platform: <https://github.com/shader-slang/slang/releases/tag/v2025.16>
 2. Unzip the downloaded directory, and use its path as value to the `SLANG_DIR` environment variable: `SLANG_DIR=/path/to/slang`.
    Note that the variable must point to the root of the slang installation (i.e. the directory that contains `bin` and `lib`).
    We recommend adding that as a system-wide environment variables so that it also becomes available to your IDE.

build.rs

@@ -15,4 +15,3 @@ pub fn main() {
         slang.compile_all(target, "../shaders", "./src/autogen", &[]);
     }
 }
-

examples/gemm_bench.rs

@@ -1,13 +1,12 @@
-use approx::assert_relative_eq;
 use indexmap::IndexMap;
 use minislang::SlangCompiler;
 use nalgebra::DMatrix;
 use slang_hal::Shader;
 use slang_hal::backend::WebGpu;
 use slang_hal::backend::{Backend, Encoder};
-use slang_hal::shapes::ViewShapeBuffers;
-use slang_hal::tensor::{GpuTensor, TensorBuilder};
 use stensor::linalg::{Gemm, GemmVariant};
+use stensor::shapes::ViewShapeBuffers;
+use stensor::tensor::GpuTensor;
 use wgpu::{BufferUsages, Features, Limits};
 
 #[async_std::main]
@@ -105,7 +104,7 @@ async fn run_gemm<B: Backend>(
         drop(pass); // Ensure the pass is ended before the encoder is borrowed again.
 
         backend.submit(encoder)?;
-        backend.synchronize();
+        backend.synchronize()?;
         timing[i] = t0.elapsed().as_secs_f32();
         backend
             .slow_read_buffer(result.buffer(), gpu_result.as_mut_slice())

shaders/stensor/geometry/svd3.slang

@@ -49,8 +49,10 @@ public struct Svd3 {
 
 // Constants used for calculation of givens quaternions
 static const float GAMMA = 5.828427124; // sqrt(8)+3;
-static const float CSTAR = 0.923879532; // cos(pi/8)
-static const float SSTAR = 0.3826834323; // sin(p/8)
+static const float CSTAR = 1.0; // TODO: using no-identity values (below) breaks the SVD for near-identity matrices.
+static const float SSTAR = 0.0; // TODO: using no-identity values (below) breaks the SVD for near-identity matrices.
+//static const float CSTAR = 0.923879532; // cos(pi/8)
+//static const float SSTAR = 0.3826834323; // sin(p/8)
 // Threshold value
 static const float SVD_EPSILON = 1e-6;
 // Iteration counts for Jacobi Eigenanalysis and reciprocal square root functions, influence precision

shaders/stensor/linalg/gemm.slang

@@ -26,8 +26,6 @@ void gemm_fast(
 ) {
     let local_id = local_id.y;
 
-    out[0] = 1.0;
-
     for (var k = 0u; k < shape_m2.ncols; k += 4u) {
         var sum = float4x4(0.0);