Test nvcc and clang-cuda in workflows/ci.yml

* Test nvcc and clang-cuda in workflows/ci.yml
* Fix clang-cuda tests
* Ensure /tmp/sccache_*.txt files are included in failed job artifacts on Windows

.github/actions/artifact_failure/action.yml

@@ -16,7 +16,8 @@ runs:
         lsof +D `pwd` || true
         killall sccache || true
         killall sccache-dist || true
-
+        # possible temp dirs for either linux or windows
+        cp "${TMP:-${TEMP:-${TMPDIR:-/tmp}}}"/sccache_*.txt . 2>/dev/null || true
         tar --exclude='target' \
             --exclude='docs' \
             --exclude='bins' \
@@ -25,6 +26,4 @@ runs:
     - uses: actions/upload-artifact@v3
       with:
         name: ${{ inputs.name }}
-        path: |
-          target/failure-${{ inputs.name }}.tar.gz
-          /tmp/sccache_*.txt
+        path: target/failure-${{ inputs.name }}.tar.gz

.github/actions/nvcc-toolchain/action.yml

@@ -0,0 +1,16 @@
+name: nvcc-toolchain
+inputs:
+  cuda-version:
+    description: CUDA Toolkit version
+    required: true
+
+runs:
+  using: composite
+  steps:
+    - if: runner.os == 'Linux'
+      shell: bash
+      run: .github/actions/nvcc-toolchain/install-cuda.sh ${{ inputs.cuda-version }}
+
+    - if: runner.os == 'Windows'
+      shell: powershell
+      run: .\.github\actions\nvcc-toolchain\install-cuda.ps1 -cudaVersion ${{ inputs.cuda-version }}

.github/actions/nvcc-toolchain/install-cuda.ps1

@@ -0,0 +1,50 @@
+Param(
+    [Parameter(Mandatory=$false)]
+    [string]
+    $cudaVersion="12.6.0"
+)
+
+# Use System.Version to tokenize version
+$version = [Version]$cudaVersion
+
+$major = $version.Major
+$minor = $version.Minor
+$build = $version.Build
+
+# Minimum build is 0, not -1 as default in case "12.5" is passed
+if ($build -lt 0) {
+    $build = 0
+}
+
+# mmb == major minor build
+$mmbVersionTag = "${major}.${minor}.${build}"
+# mm = major minor
+$mmVersionTag = "${major}.${minor}"
+
+# `cuda_${mmbVersionTag}_windows_network.exe` name only valid back to CUDA v11.5.1.
+# Before that it was named `cuda_${mmbVersionTag}_win10_network.exe`.
+$cudaVersionUrl = "https://developer.download.nvidia.com/compute/cuda/${mmbVersionTag}/network_installers/cuda_${mmbVersionTag}_windows_network.exe"
+$cudaComponents =
+    "nvcc_$mmVersionTag",
+    "curand_$mmVersionTag",
+    "curand_dev_$mmVersionTag",
+    "cudart_$mmVersionTag",
+    "cupti_$mmVersionTag",
+    "nvrtc_$mmVersionTag",
+    "nvrtc_dev_$mmVersionTag",
+    "nvml_dev_$mmVersionTag",
+    "nvtx_$mmVersionTag"
+
+Invoke-WebRequest -Uri "$cudaVersionUrl" -OutFile "./cuda_network.exe" -UseBasicParsing
+Start-Process -Wait -PassThru -FilePath .\cuda_network.exe -ArgumentList "-s $cudaComponents"
+
+$ENV:PATH="$ENV:PATH;C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v$mmVersionTag\bin"
+$ENV:CUDA_PATH="C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v$mmVersionTag"
+
+$PATH_STR="PATH=$ENV:PATH"
+$PATH_STR | Out-File -Append $ENV:GITHUB_ENV
+
+$CUDA_PATH_STR="CUDA_PATH=$ENV:CUDA_PATH"
+$CUDA_PATH_STR | Out-File -Append $ENV:GITHUB_ENV
+
+Remove-Item .\cuda_network.exe

.github/actions/nvcc-toolchain/install-cuda.sh

@@ -0,0 +1,72 @@
+#! /usr/bin/env bash
+set -eu
+
+export DEBIAN_FRONTEND=noninteractive
+
+get_cuda_deb() {
+    local deb="$(                                 \
+        wget --no-hsts -q -O- "${1}/Packages"     \
+    | grep -P "^Filename: \./${2}(.*)\.deb$"      \
+    | sort -Vr | head -n1 | cut -d' ' -f2         \
+    )";
+    if [ -z "$deb" ]; then
+        echo "Error: No matching .deb found for '${1}' and '${2}'" >&2
+        return 1
+    fi
+    wget --no-hsts -q -O "/tmp/${deb#./}" "${1}/${deb#./}";
+    echo -n "/tmp/${deb#./}";
+}
+
+VERSION="$1";
+
+NVARCH="$(uname -p)";
+
+if test "$NVARCH" = aarch64; then
+    NVARCH="sbsa";
+fi
+
+OSNAME="$(
+    . /etc/os-release;
+    major="$(cut -d'.' -f1 <<< "${VERSION_ID}")";
+    minor="$(cut -d'.' -f2 <<< "${VERSION_ID}")";
+    echo "$ID$((major - (major % 2)))${minor}";
+)";
+
+CUDA_HOME="/usr/local/cuda";
+
+cuda_repo_base="https://developer.download.nvidia.com/compute/cuda/repos";
+cuda_repo="${cuda_repo_base}/${OSNAME}/${NVARCH}";
+
+cuda_ver="$VERSION";
+cuda_ver="$(grep -Po '^[0-9]+\.[0-9]+' <<< "${cuda_ver}")";
+cuda_ver="${cuda_ver/./-}";
+
+if ! dpkg -s cuda-keyring; then
+    sudo apt-get install -y --no-install-recommends   \
+        "$(get_cuda_deb "${cuda_repo}" cuda-keyring)" \
+        ;
+fi
+
+PKGS=();
+PKGS+=("cuda-toolkit-${cuda_ver}");
+
+sudo apt-get update;
+sudo apt-get install -y --no-install-recommends "${PKGS[@]}";
+
+if ! test -L "${CUDA_HOME}"; then
+    # Create /usr/local/cuda symlink
+    sudo ln -s "${CUDA_HOME}-${cuda_ver}" "${CUDA_HOME}";
+fi
+
+export PATH="$PATH:$CUDA_HOME/bin"
+
+which -a nvcc
+nvcc --version
+
+cat <<EOF | tee -a "$GITHUB_ENV"
+CUDA_HOME=$CUDA_HOME
+CUDA_PATH=$CUDA_HOME
+PATH=$PATH
+EOF
+
+rm /tmp/*.deb

.github/workflows/ci.yml

@@ -86,21 +86,53 @@ jobs:
             extra_args: --no-default-features
             allow_failure: true
           - os: ubuntu-22.04
-          - os: macos-13
-          # M1 CPU
-          - os: macos-14
+            cuda: "11.8"
+            extra_desc: cuda11.8
+          - os: ubuntu-24.04
+            cuda: "12.6"
+            # Oldest supported version, keep in sync with README.md
+            rustc: "1.75.0"
+            extra_desc: cuda12.6
+          # - os: macos-13
+          # # M1 CPU
+          # - os: macos-14
           - os: windows-2019
+            cuda: "11.8"
             # Oldest supported version, keep in sync with README.md
             rustc: "1.75.0"
+            extra_args: --no-fail-fast
+            extra_desc: cuda11.8
           - os: windows-2019
+            cuda: "11.8"
             rustc: nightly
             allow_failure: true
             extra_args: --features=unstable
+            extra_desc: cuda11.8
           - os: windows-2019
+            cuda: "11.8"
             rustc: beta
+            extra_desc: cuda11.8
+          - os: windows-2022
+            cuda: "12.6"
+            # Oldest supported version, keep in sync with README.md
+            rustc: "1.75.0"
+            extra_args: --no-fail-fast
+            extra_desc: cuda12.6
+          - os: windows-2022
+            cuda: "12.6"
+            rustc: nightly
+            allow_failure: true
+            extra_args: --features=unstable
+            extra_desc: cuda12.6
+          - os: windows-2022
+            cuda: "12.6"
+            rustc: beta
+            extra_desc: cuda12.6
     env:
       RUST_BACKTRACE: 1
     steps:
+      - uses: ilammy/msvc-dev-cmd@v1
+
       - name: Clone repository
         uses: actions/checkout@v4
 
@@ -109,9 +141,28 @@ jobs:
         with:
           toolchain: ${{ matrix.rustc }}
 
-      - name: Install gcc & clang for tests
-        run: sudo apt-get install -y clang gcc
-        if: ${{ matrix.os == 'ubuntu-20.04' || matrix.os == 'ubuntu-22.04' }}
+      - if: ${{ contains(matrix.os, 'ubuntu') }}
+        name: Install gcc & clang for tests
+        env:
+          DEBIAN_FRONTEND: noninteractive
+        run: |
+          set -x
+          # Conflicts with clang-cuda
+          if dpkg -s gcc-14 >/dev/null 2>&1; then
+            sudo apt remove -y gcc-14 g++-14
+            sudo apt autoremove -y
+          fi
+          sudo apt install -y --no-install-recommends clang gcc
+          echo 'gcc version:'
+          gcc --version
+          echo 'clang version:'
+          clang --version
+
+      - if: matrix.cuda != '' && contains(fromJSON('["Linux", "Windows"]'), runner.os)
+        name: Install nvcc
+        uses: ./.github/actions/nvcc-toolchain
+        with:
+          cuda-version: ${{ matrix.cuda }}
 
       - name: Build tests
         run: cargo test --no-run --locked --all-targets ${{ matrix.extra_args }}
@@ -206,7 +257,8 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - os: ubuntu-20.04
+          - os: ubuntu-22.04
+            cuda: "11.8"
             rustc: nightly
             allow_failure: true
             extra_args: --features=unstable
@@ -231,6 +283,12 @@ jobs:
         run: sudo apt-get install -y clang gcc
         if: ${{ matrix.os == 'ubuntu-20.04' }}
 
+      - if: matrix.cuda != '' && contains(fromJSON('["Linux", "Windows"]'), runner.os)
+        name: Install nvcc
+        uses: ./.github/actions/nvcc-toolchain
+        with:
+          cuda-version: ${{ matrix.cuda }}
+
       - name: "`grcov` ~ install"
         run: cargo install grcov
 

src/compiler/clang.rs

@@ -140,6 +140,7 @@ impl CCompilerImpl for Clang {
             self.kind(),
             rewrite_includes_only,
             ignorable_whitespace_flags,
+            language_to_clang_arg,
         )
         .await
     }
@@ -168,13 +169,31 @@ impl CCompilerImpl for Clang {
             env_vars,
             self.kind(),
             rewrite_includes_only,
+            language_to_clang_arg,
         )
         .map(|(command, dist_command, cacheable)| {
             (CCompileCommand::new(command), dist_command, cacheable)
         })
     }
 }
 
+pub fn language_to_clang_arg(lang: Language) -> Option<&'static str> {
+    match lang {
+        Language::C => Some("c"),
+        Language::CHeader => Some("c-header"),
+        Language::Cxx => Some("c++"),
+        Language::CxxHeader => Some("c++-header"),
+        Language::ObjectiveC => Some("objective-c"),
+        Language::ObjectiveCxx => Some("objective-c++"),
+        Language::Cuda => Some("cuda"),
+        Language::Ptx => None,
+        Language::Cubin => None,
+        Language::Rust => None, // Let the compiler decide
+        Language::Hip => Some("hip"),
+        Language::GenericHeader => None, // Let the compiler decide
+    }
+}
+
 counted_array!(pub static ARGS: [ArgInfo<gcc::ArgData>; _] = [
     take_arg!("--dependent-lib", OsString, Concatenated('='), PassThrough),
     take_arg!("--hip-device-lib-path", PathBuf, Concatenated('='), PassThroughPath),