release(v0.5.0): integrate shared host runtime and source-location fixes

.github/workflows/book.yml

@@ -15,7 +15,8 @@ on:
   # only ever built after merge, so a PR that broke it could not be caught by
   # review or by CI.
   pull_request:
-    branches: [main]
+    # Include release PRs and stacks targeting prerequisite copy-pr-bot mirrors.
+    branches: [main, "release/v*", "pull-request/[0-9]+"]
     paths:
       - 'cuda-oxide/cuda-oxide-book/**'
       - 'cutile-rs/cutile-book/**'

.github/workflows/ci.yml

@@ -12,17 +12,20 @@ on:
       - 'cuda-bindings/**'
       - 'cuda-core/**'
       - 'cuda-core-derive/**'
+      - 'cuda-repr/**'
       - 'cuda-async/**'
       - 'Cargo.toml'
       - 'Cargo.lock'
       - '.github/workflows/**'
   pull_request:
-    branches: [main]
+    # Include release PRs and stacks targeting prerequisite copy-pr-bot mirrors.
+    branches: [main, "release/v*", "pull-request/[0-9]+"]
     paths:
       - 'cuda-oxide/**'
       - 'cuda-bindings/**'
       - 'cuda-core/**'
       - 'cuda-core-derive/**'
+      - 'cuda-repr/**'
       - 'cuda-async/**'
       - 'Cargo.toml'
       - 'Cargo.lock'

.github/workflows/clippy.yml

@@ -19,10 +19,10 @@ jobs:
         uses: actions/checkout@v4
 
       - name: Install CUDA toolkit
-        uses: Jimver/cuda-toolkit@3d45d157f327c09c04b50ee6ccdea2d9d017ec76  # v0.2.35
+        uses: Jimver/cuda-toolkit@0ca15bb027fc96f7f61ddbb9c5221d699da06bdb  # v0.2.36
         id: cuda-toolkit
         with:
-          cuda: "13.1.1"
+          cuda: "13.3.1"
           method: network
           sub-packages: '["nvcc", "cudart-dev"]'
           non-cuda-sub-packages: '["libcurand-dev"]'
@@ -37,7 +37,7 @@ jobs:
         working-directory: cuda-oxide
         run: cargo clippy --workspace --all-targets -- -D warnings
 
-      # `rustc-codegen-cuda` and each example under examples/ are their own
+      # `rustc-codegen-cuda`, the standard and CuTe examples, and CuTe models have
       # standalone workspaces (the codegen backend is swapped in via
       # RUSTFLAGS, so they can't live in the root [workspace]). That's 60+
       # cargo invocations from different workspace roots. By default each
@@ -49,7 +49,7 @@ jobs:
       # Point all of those invocations at one shared CARGO_TARGET_DIR under
       # ${RUNNER_TEMP}. Cargo fingerprints each unit by package + features
       # + workspace_root + toolchain, so co-located workspaces using the
-      # same pinned nightly deduplicate on identical deps and compile each
+      # same pinned toolchain deduplicate on identical deps and compile each
       # one once. This is intentionally scoped *not* to cover the
       # `cargo clippy --workspace` step above, which already has a
       # well-defined `target/` under cuda-oxide.
@@ -60,18 +60,26 @@ jobs:
         working-directory: cuda-oxide/crates/rustc-codegen-cuda
         run: cargo clippy --all-targets -- -D warnings
 
-      - name: Run clippy (rustc-codegen-cuda examples)
+      - name: Run clippy (standard and CuTe examples and models)
         env:
           CUDA_TOOLKIT_PATH: ${{ steps.cuda-toolkit.outputs.CUDA_PATH }}
           CARGO_TARGET_DIR: ${{ runner.temp }}/clippy-shared-target
+          RUSTC_BOOTSTRAP: "1"
         run: |
           set -e
-          for ex in cuda-oxide/crates/rustc-codegen-cuda/examples/*/; do
-            if [ ! -f "$ex/Cargo.toml" ]; then continue; fi
-            name=$(basename "$ex")
-            echo "::group::clippy: $name"
-            (cd "$ex" && cargo clippy --all-targets -- -D warnings)
-            echo "::endgroup::"
+          example_roots=(
+            cuda-oxide/crates/rustc-codegen-cuda/examples
+            cuda-oxide/cute-rs-preview/examples
+            cuda-oxide/cute-rs-preview/models
+          )
+          for example_root in "${example_roots[@]}"; do
+            for ex in "$example_root"/*/; do
+              if [ ! -f "$ex/Cargo.toml" ]; then continue; fi
+              name=$(basename "$ex")
+              echo "::group::clippy: $name"
+              (cd "$ex" && cargo clippy --all-targets -- -D warnings)
+              echo "::endgroup::"
+            done
           done
 
           # The loop above is one level deep. A few examples nest a second
@@ -97,11 +105,11 @@ jobs:
                grep -q "\"${sub}\""; then
               continue
             fi
-            echo "::group::clippy: ${dir#cuda-oxide/crates/rustc-codegen-cuda/examples/}"
+            echo "::group::clippy: ${dir#cuda-oxide/}"
             (cd "${dir}" && cargo clippy --locked --all-targets -- -D warnings)
             echo "::endgroup::"
           done < <(
-            find cuda-oxide/crates/rustc-codegen-cuda/examples -mindepth 3 -name Cargo.toml \
+            find "${example_roots[@]}" -mindepth 3 -name Cargo.toml \
               -not -path '*/target/*' | sort
           )
 

.github/workflows/codeql.yml

@@ -15,7 +15,8 @@ on:
       - '**/*.sh'
       - '**/*.py'
   pull_request:
-    branches: [main]
+    # Include release PRs and stacks targeting prerequisite copy-pr-bot mirrors.
+    branches: [main, "release/v*", "pull-request/[0-9]+"]
     paths:
       - '**/*.rs'
       - '**/Cargo.toml'

.github/workflows/cutile-rs.yml

@@ -64,6 +64,7 @@ jobs:
               - 'cuda-bindings/**'
               - 'cuda-core/**'
               - 'cuda-core-derive/**'
+              - 'cuda-repr/**'
               - 'cuda-async/**'
               - 'Cargo.toml'
               - 'Cargo.lock'
@@ -288,7 +289,7 @@ jobs:
 
       - name: Native f16 DeviceCopy
         working-directory: ${{ github.workspace }}
-        run: cargo test -p cuda-core --features f16 --test simt_device_copy_impls
+        run: cargo test -p cuda-core --features f16 --test device_copy_impls
 
       - name: loom model check (interleavings + weak memory)
         working-directory: ${{ github.workspace }}

.github/workflows/docs.yml

@@ -10,17 +10,20 @@ on:
       - 'cuda-bindings/**'
       - 'cuda-core/**'
       - 'cuda-core-derive/**'
+      - 'cuda-repr/**'
       - 'cuda-async/**'
       - 'Cargo.toml'
       - 'Cargo.lock'
       - '.github/workflows/**'
   pull_request:
-    branches: [main]
+    # Include release PRs and stacks targeting prerequisite copy-pr-bot mirrors.
+    branches: [main, "release/v*", "pull-request/[0-9]+"]
     paths:
       - 'cuda-oxide/**'
       - 'cuda-bindings/**'
       - 'cuda-core/**'
       - 'cuda-core-derive/**'
+      - 'cuda-repr/**'
       - 'cuda-async/**'
       - 'Cargo.toml'
       - 'Cargo.lock'
@@ -50,10 +53,10 @@ jobs:
       - name: Install CUDA toolkit
         # `cargo doc --workspace` compiles cuda-bindings' build-time bindgen,
         # which needs `cuda.h`. Same toolkit pin as the unit-tests job.
-        uses: Jimver/cuda-toolkit@3d45d157f327c09c04b50ee6ccdea2d9d017ec76  # v0.2.35
+        uses: Jimver/cuda-toolkit@0ca15bb027fc96f7f61ddbb9c5221d699da06bdb  # v0.2.36
         id: cuda-toolkit
         with:
-          cuda: "13.1.1"
+          cuda: "13.3.1"
           method: network
           sub-packages: '["nvcc", "cudart-dev"]'
           non-cuda-sub-packages: '["libcurand-dev"]'

.github/workflows/examples-compile.yml

@@ -10,17 +10,20 @@ on:
       - 'cuda-bindings/**'
       - 'cuda-core/**'
       - 'cuda-core-derive/**'
+      - 'cuda-repr/**'
       - 'cuda-async/**'
       - 'Cargo.toml'
       - 'Cargo.lock'
       - '.github/workflows/**'
   pull_request:
-    branches: [main]
+    # Include release PRs and stacks targeting prerequisite copy-pr-bot mirrors.
+    branches: [main, "release/v*", "pull-request/[0-9]+"]
     paths:
       - 'cuda-oxide/**'
       - 'cuda-bindings/**'
       - 'cuda-core/**'
       - 'cuda-core-derive/**'
+      - 'cuda-repr/**'
       - 'cuda-async/**'
       - 'Cargo.toml'
       - 'Cargo.lock'
@@ -61,10 +64,10 @@ jobs:
       # Needed for cuda.h and curand.h: bindgen runs in the shared
       # cuda-bindings build script. Same setup as unit-tests.yml / clippy.yml.
       - name: Install CUDA toolkit
-        uses: Jimver/cuda-toolkit@3d45d157f327c09c04b50ee6ccdea2d9d017ec76  # v0.2.35
+        uses: Jimver/cuda-toolkit@0ca15bb027fc96f7f61ddbb9c5221d699da06bdb  # v0.2.36
         id: cuda-toolkit
         with:
-          cuda: "13.1.1"
+          cuda: "13.3.1"
           method: network
           sub-packages: '["nvcc", "cudart-dev"]'
           # nvJitLink is packaged separately from cuda-nvcc on Linux. Keep it

.github/workflows/fmt.yml

@@ -20,7 +20,7 @@ jobs:
 
       # rustfmt comes from cuda-oxide/rust-toolchain.toml's component list,
       # like clippy,
-      # so `cargo fmt` resolves against the same nightly as the rest of CI.
+      # so `cargo fmt` resolves against the same toolchain as the rest of CI.
 
       - name: Check stable host workspace formatting
         run: cargo fmt --all -- --check
@@ -43,19 +43,25 @@ jobs:
         run: cargo fmt --all -- --check
         working-directory: cuda-oxide/crates/cuda-macros/tests/device-only
 
-      - name: Check example crate formatting
+      - name: Check example and model crate formatting
+        env:
+          RUSTC_BOOTSTRAP: "1"
         run: |
           set -euo pipefail
           # `cargo fmt --all` also formats local path dependencies. Running it
           # per example repeatedly reformats the large shared workspaces.
           shopt -s globstar nullglob
-          manifests=(cuda-oxide/crates/rustc-codegen-cuda/examples/**/Cargo.toml)
+          manifests=(
+            cuda-oxide/crates/rustc-codegen-cuda/examples/**/Cargo.toml
+            cuda-oxide/cute-rs-preview/examples/**/Cargo.toml
+            cuda-oxide/cute-rs-preview/models/**/Cargo.toml
+          )
           ((${#manifests[@]} > 0))
           for manifest in "${manifests[@]}"; do
             example_dir="$(dirname "$manifest")"
             echo "::group::cargo fmt --check ($example_dir)"
             # Run inside the component so rustup discovers CUDA Oxide's
-            # directory-scoped nightly pin.
+            # directory-scoped toolchain pin.
             (cd "$example_dir" && cargo fmt -- --check)
             echo "::endgroup::"
           done

.github/workflows/naming-guard.yml

@@ -16,7 +16,8 @@ on:
       - 'cuda-oxide/scripts/check-reserved-prefixes.sh'
       - '.github/workflows/naming-guard.yml'
   pull_request:
-    branches: [main]
+    # Include release PRs and stacks targeting prerequisite copy-pr-bot mirrors.
+    branches: [main, "release/v*", "pull-request/[0-9]+"]
     paths:
       - 'cuda-oxide/crates/**'
       - 'cuda-oxide/scripts/check-reserved-prefixes.sh'

.github/workflows/status-guard.yml

@@ -5,7 +5,8 @@ on:
   push:
     branches: [main]
   pull_request:
-    branches: [main]
+    # Include release PRs and stacks targeting prerequisite copy-pr-bot mirrors.
+    branches: [main, "release/v*", "pull-request/[0-9]+"]
 
 # Each commit on main gets its own group, so it starts at once. Any other ref
 # still shares one group and cancels the older run.
@@ -60,13 +61,12 @@ jobs:
       - name: Verify README lists every workspace member
         run: bash cuda-oxide/scripts/check-crate-inventory.sh
 
-  # The pin is copied into places no build compares: the nested backend pin
-  # (rustup reads it for the backend's own [workspace], so a mismatch there
-  # is a load failure rather than a warning), the scaffold `cargo oxide new`
-  # writes into user projects, the devcontainer's preinstalled toolchain, the
-  # book's quoted [toolchain] blocks, and the dated rustup/cargo commands in
-  # the book and READMEs. #727 added rustfmt to both real files and left both
-  # book quotes behind.
+  # The pin is copied into places no build compares: the scaffold `cargo oxide
+  # new` writes into user projects, the devcontainer's preinstalled toolchain,
+  # the book's quoted [toolchain] blocks, the explicit rustup/cargo commands in
+  # the book and READMEs, and the `rust-overlay` input in `cuda-oxide/flake.lock`.
+  # #727 added rustfmt to both real files and left both book quotes behind;
+  # #1609 bumped the pin to 1.99.1 and left `flake.lock` behind.
   toolchain-pin-parity:
     name: toolchain pin parity
     runs-on: ubuntu-latest
@@ -130,9 +130,9 @@ jobs:
       - name: Clone repository
         uses: actions/checkout@v4
 
-# rustfmt comes from cuda-oxide/rust-toolchain.toml's component list through the
+      # rustfmt comes from cuda-oxide/rust-toolchain.toml's component list through the
       # rustup proxy, as in fmt.yml, so the order checked is the pinned
-      # nightly's. No cargo, no CUDA toolkit, no network beyond the toolchain.
+      # toolchain's. No cargo, no CUDA toolkit, no network beyond the toolchain.
       - name: Verify the book's Rust blocks use rustfmt's import order
         run: bash cuda-oxide/scripts/check-book-import-order.sh
 

.github/workflows/unit-tests.yml

@@ -55,6 +55,18 @@ jobs:
           - package: dialect-mir
           - package: dialect-nvvm
           - package: mir-importer
+          # CuTe types, semantic IR, and MLIR export need no CUDA toolkit.
+          # Facade macro tests also compile host metadata via cuda-host.
+          - package: cute-layout
+          - package: cute-rs
+            needs_cuda: true
+          - package: ctm-rs
+            needs_cuda: true
+          - package: ctm-rs
+            test_args: --all-targets --features host
+            needs_cuda: true
+          - package: dialect-cute
+          - package: cuda-oxide-mlir-export
           # Includes a real llc -> ptxas spine-kernel validation.
           - package: cuda-oxide-codegen
             needs_cuda: true
@@ -102,29 +114,14 @@ jobs:
 
       - name: Install CUDA toolkit
         if: matrix.needs_cuda
-        uses: Jimver/cuda-toolkit@3d45d157f327c09c04b50ee6ccdea2d9d017ec76  # v0.2.35
+        uses: Jimver/cuda-toolkit@0ca15bb027fc96f7f61ddbb9c5221d699da06bdb  # v0.2.36
         id: cuda-toolkit
         with:
-          cuda: "13.1.1"
+          cuda: "13.3.1"
           method: network
           sub-packages: '["nvcc", "cudart-dev"]'
           non-cuda-sub-packages: '["libcurand-dev"]'
 
-      - name: Install LLVM 21 (llc/opt for the NVPTX backend)
-        if: matrix.package == 'cuda-oxide-codegen'
-        run: |
-          set -euo pipefail
-          source /etc/os-release
-          sudo install -d -m 0755 /etc/apt/keyrings
-          wget -qO- https://apt.llvm.org/llvm-snapshot.gpg.key \
-            | sudo tee /etc/apt/keyrings/apt-llvm-org.asc >/dev/null
-          echo "deb [signed-by=/etc/apt/keyrings/apt-llvm-org.asc] https://apt.llvm.org/${VERSION_CODENAME}/ llvm-toolchain-${VERSION_CODENAME}-21 main" \
-            | sudo tee /etc/apt/sources.list.d/llvm-21.list
-          sudo apt-get update
-          sudo apt-get install -y --no-install-recommends llvm-21
-          llc-21 --version | head -2
-          opt-21 --version | head -2
-
       - name: Run crate unit tests
         env:
           CUDA_TOOLKIT_PATH: ${{ steps.cuda-toolkit.outputs.CUDA_PATH }}

AGENTS.md

@@ -31,8 +31,9 @@ contribution guide. There is no root `Justfile`. The SIMT recipes live in
 - `cuda-bindings/`, `cuda-core/`, `cuda-core-derive/`, and `cuda-async/` — shared
   host crates in the root workspace. Both kernel stacks depend on these by path.
 - `cuda-oxide/` — the SIMT model. Its `Cargo.toml` explicitly lists workspace
-  members under `crates/`; its `rust-toolchain.toml` pins the nightly needed for
-  `rustc_private`. The book is `cuda-oxide/cuda-oxide-book/`, and
+  members under `crates/`; its `rust-toolchain.toml` pins the stable release
+  (`1.99.1`) and `rustc-dev` / `llvm-tools` components needed for `rustc_private`
+  (unlocked via `RUSTC_BOOTSTRAP=1`). The book is `cuda-oxide/cuda-oxide-book/`, and
   `cuda-oxide/intrinsics/` holds the intrinsic catalog and ABI ledger.
   `cuda-oxide/crates/rustc-codegen-cuda/` declares its own workspace and inherits
   the SIMT toolchain pin. Examples and test fixtures can also declare separate
@@ -84,15 +85,16 @@ nested workspaces.
 ## Toolchains
 
 The root `rust-toolchain.toml` and `cutile-rs/rust-toolchain.toml` carry matching
-stable pins. `cuda-oxide/rust-toolchain.toml` supplies the SIMT nightly, including
-for its nested backend workspace. Do not float a pin to work around a failure.
+stable pins. `cuda-oxide/rust-toolchain.toml` supplies the SIMT stable pin and
+compiler components (`rustc-dev`, `llvm-tools`), including for its nested backend
+workspace. Do not float a pin to work around a failure.
 `cuda-oxide/scripts/check-toolchain-parity.sh` checks the SIMT pin against its
 scaffold, development container, and documentation.
 
 rustup normally resolves the toolchain from the working directory; an explicit
 override can change that. `--manifest-path` does not change the working directory,
 so pointing a root Cargo command at the SIMT manifest does not select the SIMT
-nightly. Change directory first or select the required toolchain explicitly.
+toolchain pin. Change directory first or select the required toolchain explicitly.
 
 Nix environments live in `cuda-oxide/flake.nix` and `cutile-rs/flake.nix`.
 Run `nix develop` from the relevant component.
@@ -105,6 +107,9 @@ Run `nix develop` from the relevant component.
   notice plus `Apache-2.0`, in the form shown under "License Headers" in
   `CONTRIBUTING.md`. Preserve existing and third-party notices.
   `cuda-oxide/scripts/check-spdx-headers.sh` enforces the policy.
+- **Create new branches from the user's fork of this repository.** Synchronize
+  the fork's `main` with upstream before branching. For stacked work, base the
+  next branch on the preceding feature branch in that same fork.
 - **Never push to the canonical upstream repository, `NVIDIA/cuda-rust`.
   Treat it as read-only.** Branch and push to a fork. Before pushing, run
   `git remote -v` and confirm the push remote is a fork of the base, comparing