ci: add chipStar presubmit on self-hosted Intel Arc

.ci/exclude.txt

@@ -0,0 +1,32 @@
+# CTest regex patterns (one per line) for tests excluded from CI.
+# Blank lines and '#' comments are ignored. Every entry should link to an
+# upstream issue — fix the test, don't register failures indefinitely.
+#
+# Source: /space/pvelesko/2026.04.14-lib-fixes/bug-summary.md (2026-04-14 sweep).
+
+# Pre-existing chipStar kernel-name lookup failure (not a rocPRIM bug).
+# Tracked in bug-summary.md §1 "test_device_reduce_by_key — unchanged —
+# pre-existing unrelated kernel-name lookup failure".
+rocprim\.device_reduce_by_key
+
+# IGC bug #403 (SIMD32 finalizer silently drops kernels after register-pressure
+# retry). Upstream: https://github.com/intel/intel-graphics-compiler/issues/403
+# Mitigation: CHIP_L0_FORCE_SIMD16 or wait for IGC fix.
+rocprim\.device_run_length_encode
+
+# IGC bug #402 (fp16 subgroup-shuffle narrowing miscompile).
+# Upstream: https://github.com/intel/intel-graphics-compiler/issues/402
+# Fix pending in PR [CHIP-SPV/rocPRIM#3]; remove these two lines when #3 merges.
+rocprim\.warp_reduce
+rocprim\.warp_scan
+
+# fp16/bf16 block_shuffle failures on PoCL CPU (pastrami). Related to the same
+# sub-int warp_shuffle roundtrip addressed by PR [CHIP-SPV/rocPRIM#3]; remove
+# once the fp16 fix lands and is verified on CPU.
+rocprim\.block_shuffle
+
+# Mali GPU (salami) lacks fp64 + SubgroupDispatch SPIR-V capabilities.
+# These tests require hardware features the ARM Mali doesn't expose.
+rocprim\.thread_algos
+rocprim\.intrinsics
+rocprim\.warp_load

.github/workflows/presubmit.yml

@@ -0,0 +1,129 @@
+name: Presubmit
+
+on:
+  pull_request:
+    paths-ignore: "docs/**"
+  push:
+    branches: [main]
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  meatloaf-x64-intel:
+    runs-on: [self-hosted, Linux, X64]
+    env:
+      CHIP_MODULE_CACHE_DIR: ""
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.head.sha }}
+          fetch-depth: 1
+      - name: Build and test (chip-spv-gpu)
+        shell: bash
+        run: |
+          set -e
+          if [ -f "$HOME/.local/init/bash" ]; then
+            export MODULESHOME=$HOME/.local
+            source "$MODULESHOME/init/bash"
+          elif [ -f /etc/profile.d/lmod.sh ]; then
+            source /etc/profile.d/lmod.sh
+          else
+            source /etc/profile.d/modules.sh
+          fi
+          module use /space/pvelesko/modulefiles
+          module load oneapi/2025.0.4 llvm/22.0-native HIP/chipStar/main level-zero/dgpu
+          module list
+          rm -rf ~/.cache/chipStar build
+          cmake -S . -B build \
+            -DCMAKE_CXX_COMPILER="$(which hipcc)" \
+            -DCMAKE_BUILD_TYPE=Release \
+            -DBUILD_TEST=ON -DBUILD_BENCHMARK=OFF \
+            -DAMDGPU_TEST_TARGETS="" \
+            -DCMAKE_POLICY_VERSION_MINIMUM=3.5
+          cmake --build build -j 8
+          exclude=$(grep -Ev '^\s*(#|$)' .ci/exclude.txt | paste -sd '|' -)
+          cd build
+          ctest --output-on-failure --timeout 300 -L chip-spv-gpu ${exclude:+-E "$exclude"}
+      - uses: actions/upload-artifact@v4
+        if: failure()
+        with:
+          name: ctest-logs-meatloaf
+          path: build/Testing/**/LastTest*.log
+          if-no-files-found: ignore
+
+  salami-arm64:
+    runs-on: [self-hosted, Linux, ARM64]
+    env:
+      CHIP_MODULE_CACHE_DIR: ""
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.head.sha }}
+          fetch-depth: 1
+      - name: Build and test (chip-spv-cpu)
+        shell: bash
+        run: |
+          set -e
+          if [ -f "$HOME/.local/init/bash" ]; then
+            export MODULESHOME=$HOME/.local
+            source "$MODULESHOME/init/bash"
+          else
+            source /etc/profile.d/modules.sh
+          fi
+          module use ~/modulefiles
+          module load llvm/22.0-native HIP/chipStar/main
+          module list
+          rm -rf ~/.cache/chipStar build
+          cmake -S . -B build \
+            -DCMAKE_CXX_COMPILER="$(which hipcc)" \
+            -DCMAKE_BUILD_TYPE=Release \
+            -DBUILD_TEST=ON -DBUILD_BENCHMARK=OFF \
+            -DAMDGPU_TEST_TARGETS="" \
+            -DCMAKE_POLICY_VERSION_MINIMUM=3.5
+          cmake --build build -j 2
+          exclude=$(grep -Ev '^\s*(#|$)' .ci/exclude.txt | paste -sd '|' -)
+          cd build
+          ctest --output-on-failure --timeout 300 -L chip-spv-cpu ${exclude:+-E "$exclude"}
+      - uses: actions/upload-artifact@v4
+        if: failure()
+        with:
+          name: ctest-logs-salami
+          path: build/Testing/**/LastTest*.log
+          if-no-files-found: ignore
+
+  pastrami-macos:
+    runs-on: [self-hosted, ARM64, MacOS]
+    env:
+      CHIP_MODULE_CACHE_DIR: ""
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.head.sha }}
+          fetch-depth: 1
+      - name: Build and test (chip-spv-cpu)
+        shell: bash
+        run: |
+          set -e
+          source /opt/homebrew/Cellar/modules/5.6.0/init/bash
+          module use ~/modulefiles
+          module load llvm/22.0-native pocl HIP/chipStar/main
+          module list
+          rm -rf ~/.cache/chipStar build
+          cmake -S . -B build \
+            -DCMAKE_CXX_COMPILER="$(which hipcc)" \
+            -DCMAKE_BUILD_TYPE=Release \
+            -DBUILD_TEST=ON -DBUILD_BENCHMARK=OFF \
+            -DAMDGPU_TEST_TARGETS="" \
+            -DCMAKE_POLICY_VERSION_MINIMUM=3.5
+          cmake --build build -j 4
+          exclude=$(grep -Ev '^\s*(#|$)' .ci/exclude.txt | paste -sd '|' -)
+          cd build
+          ctest --output-on-failure --timeout 300 -L chip-spv-cpu ${exclude:+-E "$exclude"}
+      - uses: actions/upload-artifact@v4
+        if: failure()
+        with:
+          name: ctest-logs-pastrami
+          path: build/Testing/**/LastTest*.log
+          if-no-files-found: ignore