Adding GPU target capability and creating parent test workflow (ROCm#193)

geomin12 · web-flow · commit 2a605a2aeb33 · 2025-03-21T13:06:42.000-07:00
Organizing GH action setup to enable multiple tests to run in parallel rocPRIM and hipBLASlt will be added shortly after this PR goes through this PR will default all GH action builds to gfx942 as we only have mi300s so far, and certain tests require specific GFX matching targets Closes ROCm#225
diff --git a/.github/workflows/build_linux_packages.yml b/.github/workflows/build_linux_packages.yml
@@ -6,12 +6,18 @@ on:
       package_version:
         type: string
         default: ADHOCBUILD
+      amdgpu_families:
+        type: string
+        default: gfx94X-dcgpu
 
   workflow_call:
     inputs:
       package_version:
         type: string
         default: ADHOCBUILD
+      amdgpu_families:
+        type: string
+        default: gfx94X-dcgpu
 
 jobs:
   build_linux_packages:
@@ -64,12 +70,13 @@ jobs:
         run: |
           # Generate a new build id.
           package_version="${{ inputs.package_version }}"
+          amdgpu_families="${{ inputs.amdgpu_families }}"
           echo "Building package ${package_version}"
 
           # Build.
           cmake -B build -GNinja . \
             -DCMAKE_C_COMPILER_LAUNCHER=ccache -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \
-            -DTHEROCK_AMDGPU_FAMILIES=gfx110X-dgpu \
+            -DTHEROCK_AMDGPU_FAMILIES=${amdgpu_families} \
             -DTHEROCK_PACKAGE_VERSION="${package_version}" \
             -DTHEROCK_VERBOSE=ON
           ./build_tools/watch_top_processes.sh &
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
@@ -7,6 +7,9 @@ on:
       - main
   workflow_dispatch:
     inputs:
+      amdgpu_families:
+        type: string
+        default: gfx94X-dcgpu
 
 permissions:
   contents: read
@@ -25,6 +28,8 @@ jobs:
       id-token: write
     name: Build Linux Packages
     uses: ./.github/workflows/build_linux_packages.yml
+    with:
+      amdgpu_families: "${{ inputs.amdgpu_families != '' && inputs.amdgpu_families || 'gfx94X-dcgpu' }}"
 
   build_windows_packages:
     name: Build Windows Packages
@@ -34,6 +39,8 @@ jobs:
     needs: build_linux_packages
     name: Test Linux Packages
     uses: ./.github/workflows/test_linux_packages.yml
+    with:
+      amdgpu_families: "${{ inputs.amdgpu_families != '' && inputs.amdgpu_families || 'gfx94X-dcgpu' }}"
 
   # build_python_packages:
   #   name: Build Python Packages
diff --git a/.github/workflows/test_linux_packages.yml b/.github/workflows/test_linux_packages.yml
@@ -1,9 +1,3 @@
-# Copyright 2025 Advanced Micro Devices, Inc.
-#
-# Licensed under the Apache License v2.0 with LLVM Exceptions.
-# See https://llvm.org/LICENSE.txt for license information.
-# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-
 name: Test Linux Packages
 
 on:
@@ -12,75 +6,21 @@ on:
       artifact_run_id:
         type: string
         default: ""
+      amdgpu_families:
+        type: string
+        default: gfx94X-dcgpu
+  workflow_call:
+    inputs:
+      amdgpu_families:
+        type: string
+        default: gfx94X-dcgpu
   push:
     branches:
       - ADHOCBUILD
-  workflow_call:
 
 jobs:
-  test_rocm_info:
-    name: "Sanity ROCM Test"
-    runs-on: linux-mi300-1gpu-ossci-rocm
-    strategy:
-      fail-fast: true
-    defaults:
-      run:
-        shell: bash
-    env:
-      VENV_DIR: ${{ github.workspace }}/.venv
-      ARTIFACT_RUN_ID: "${{ inputs.artifact_run_id != '' && inputs.artifact_run_id || github.run_id }}"
-
-    # MOSTLY BOILER PLATE ABOVE.
-    steps:
-      - name: Checkout Repository
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
-
-      - name: Install the AWS tool
-        run: ./dockerfiles/cpubuilder/install_awscli.sh
-
-      - name: "Setting up Python"
-        id: setup_python
-        uses: actions/setup-python@42375524e23c412d93fb67b49958b491fce71c38 # v5.4.0
-        with:
-          python-version: 3.11
-
-      - name: Create Python venv
-        run: |
-          python -m venv ${VENV_DIR}
-          source ${VENV_DIR}/bin/activate
-          pip install -r requirements-test.txt
-
-      - name: Download and Unpack Artifacts
-        run: |
-          source ${VENV_DIR}/bin/activate
-
-          BUILD_ARTIFACTS_DIR="build/artifacts"
-          # TODO(sosa): We should infer this from the runner. For Libs and Above it should be based on the
-          # runner, everything else is generic. Use just generic since this is all core/below.
-          VARIANT="_generic"
-          echo "Making ${BUILD_ARTIFACTS_DIR} directory"
-          mkdir -p "${BUILD_ARTIFACTS_DIR}"
-
-          echo "Downloading artifacts"
-          aws s3 cp s3://therock-artifacts/${ARTIFACT_RUN_ID}/core-runtime_run"${VARIANT}".tar.xz "${BUILD_ARTIFACTS_DIR}" --no-sign-request
-          aws s3 cp s3://therock-artifacts/${ARTIFACT_RUN_ID}/core-runtime_lib"${VARIANT}".tar.xz "${BUILD_ARTIFACTS_DIR}" --no-sign-request
-          aws s3 cp s3://therock-artifacts/${ARTIFACT_RUN_ID}/sysdeps_lib"${VARIANT}".tar.xz "${BUILD_ARTIFACTS_DIR}" --no-sign-request
-          aws s3 cp s3://therock-artifacts/${ARTIFACT_RUN_ID}/base_lib"${VARIANT}".tar.xz "${BUILD_ARTIFACTS_DIR}" --no-sign-request
-
-          echo "Unpacking artifacts"
-          pushd "${BUILD_ARTIFACTS_DIR}"
-          mkdir output_dir
-          python ${{ github.workspace }}/build_tools/fileset_tool.py artifact-flatten *.tar.xz -o output_dir --verbose
-          sudo rsync --archive --keep-dirlinks --verbose output_dir/* /
-          popd
-
-      - name: Run rocminfo
-        run: |
-          echo "Running rocminfo"
-          /bin/rocminfo
-
-      - name: Run ROCm Sanity Tests
-        run: |
-          source ${VENV_DIR}/bin/activate
-          pytest tests/ \
-          --log-cli-level=info
+  test_sanity_check:
+    name: 'Test Sanity Check'
+    uses: './.github/workflows/test_sanity_check.yml'
+    with:
+      artifact_run_id: ${{ inputs.artifact_run_id }}
diff --git a/.github/workflows/test_sanity_check.yml b/.github/workflows/test_sanity_check.yml
@@ -0,0 +1,83 @@
+name: TheRock Sanity Check
+
+on:
+  workflow_dispatch:
+    inputs:
+      artifact_run_id:
+        type: string
+        default: ""
+  workflow_call:
+    inputs:
+      artifact_run_id:
+        type: string
+        default: ""
+  push:
+    branches:
+      - ADHOCBUILD
+
+jobs:
+  test_sanity_check:
+    name: "Sanity ROCM Test"
+    runs-on: linux-mi300-1gpu-ossci-rocm
+    strategy:
+      fail-fast: false
+    defaults:
+      run:
+        shell: bash
+    env:
+      VENV_DIR: ${{ github.workspace }}/.venv
+      ARTIFACT_RUN_ID: "${{ inputs.artifact_run_id != '' && inputs.artifact_run_id || github.run_id }}"
+      BUILD_ARTIFACTS_DIR: "${{ github.workspace }}/build/artifacts"
+      THEROCK_BIN_DIR: "${{ github.workspace }}/build/artifacts/output_dir/bin"
+
+    steps:
+      - name: Checkout Repository
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+
+      - name: Install the AWS tool
+        run: ./dockerfiles/cpubuilder/install_awscli.sh
+
+
+      - name: "Setting up Python"
+        id: setup_python
+        uses: actions/setup-python@42375524e23c412d93fb67b49958b491fce71c38 # v5.4.0
+        with:
+          python-version: 3.11
+
+      - name: Create Python venv
+        run: |
+          python -m venv ${VENV_DIR}
+          source ${VENV_DIR}/bin/activate
+          pip install -r requirements-test.txt
+
+      - name: Download and Unpack Artifacts
+        run: |
+          source ${VENV_DIR}/bin/activate
+
+          # TODO(sosa): We should infer this from the runner. For Libs and Above it should be based on the
+          # runner, everything else is generic. Use just generic since this is all core/below.
+          VARIANT="_generic"
+          echo "Making ${BUILD_ARTIFACTS_DIR} directory"
+          mkdir -p "${BUILD_ARTIFACTS_DIR}"
+
+          echo "Downloading artifacts"
+          aws s3 cp s3://therock-artifacts/${ARTIFACT_RUN_ID}/core-runtime_run"${VARIANT}".tar.xz "${BUILD_ARTIFACTS_DIR}" --no-sign-request
+          aws s3 cp s3://therock-artifacts/${ARTIFACT_RUN_ID}/core-runtime_lib"${VARIANT}".tar.xz "${BUILD_ARTIFACTS_DIR}" --no-sign-request
+          aws s3 cp s3://therock-artifacts/${ARTIFACT_RUN_ID}/sysdeps_lib"${VARIANT}".tar.xz "${BUILD_ARTIFACTS_DIR}" --no-sign-request
+          aws s3 cp s3://therock-artifacts/${ARTIFACT_RUN_ID}/base_lib"${VARIANT}".tar.xz "${BUILD_ARTIFACTS_DIR}" --no-sign-request
+          aws s3 cp s3://therock-artifacts/${ARTIFACT_RUN_ID}/amd-llvm_run"${VARIANT}".tar.xz "${BUILD_ARTIFACTS_DIR}" --no-sign-request
+          aws s3 cp s3://therock-artifacts/${ARTIFACT_RUN_ID}/amd-llvm_lib"${VARIANT}".tar.xz "${BUILD_ARTIFACTS_DIR}" --no-sign-request
+
+          echo "Unpacking artifacts"
+          mkdir ${BUILD_ARTIFACTS_DIR}/output_dir
+          python ${{ github.workspace }}/build_tools/fileset_tool.py artifact-flatten ${BUILD_ARTIFACTS_DIR}/*.tar.xz -o ${BUILD_ARTIFACTS_DIR}/output_dir --verbose
+
+      - name: Run rocminfo
+        run: |
+          ${THEROCK_BIN_DIR}/rocminfo
+
+      - name: Run ROCm Sanity Tests
+        run: |
+          source ${VENV_DIR}/bin/activate
+          pytest tests/ \
+          --log-cli-level=info
diff --git a/tests/hipcc_check.cpp b/tests/hipcc_check.cpp
@@ -13,4 +13,5 @@ int main() {
   hipMalloc(&d_buf, size * sizeof(int));
   hipLaunchKernelGGL(squares, gridsize, blocksize, 0, 0, d_buf);
   hipDeviceSynchronize();
+  return 0;
 }
diff --git a/tests/test_rocm_sanity.py b/tests/test_rocm_sanity.py
@@ -4,30 +4,24 @@
 from pathlib import Path
 from pytest_check import check
 import logging
+import os
 
 THIS_DIR = Path(__file__).resolve().parent
 
 logger = logging.getLogger(__name__)
 
+THEROCK_BIN_DIR = os.getenv("THEROCK_BIN_DIR")
 
-def run_command(command):
-    process = subprocess.run(command, capture_output=True)
-    return str(process.stdout)
 
-
-@pytest.fixture(scope="session")
-def rocm_info_output():
-    try:
-        return run_command(["rocminfo"])
-    except Exception as e:
-        logger.info(str(e))
-        return None
+def run_command(command, cwd=None):
+    process = subprocess.run(command, capture_output=True, cwd=cwd)
+    return process
 
 
 @pytest.fixture(scope="session")
-def clinfo_info_output():
+def rocm_info_output():
     try:
-        return run_command(["clinfo"])
+        return str(run_command([f"{THEROCK_BIN_DIR}/rocminfo"]).stdout)
     except Exception as e:
         logger.info(str(e))
         return None
@@ -55,38 +49,19 @@ def test_rocm_output(self, rocm_info_output, to_search):
             f"Failed to search for {to_search} in rocminfo output",
         )
 
-    @pytest.mark.parametrize(
-        "to_search",
-        [
-            (r"Device(\s|\\t)*Type:(\s|\\t)*CL_DEVICE_TYPE_GPU"),
-            (r"Name:(\s|\\t)*gfx"),
-            (r"Vendor:(\s|\\t)*Advanced Micro Devices, Inc."),
-        ],
-        ids=[
-            "clinfo - GPU Device Type Search",
-            "clinfo - GFX Name Search",
-            "clinfo - AMD Vendor Name Search",
-        ],
-    )
-    def test_clinfo_output(self, clinfo_info_output, to_search):
-        if not clinfo_info_output:
-            pytest.fail("Command clinfo failed to run")
-        check.is_not_none(
-            re.search(to_search, clinfo_info_output),
-            f"Failed to search for {to_search} in clinfo output",
-        )
-
     def test_hip_printf(self):
         # Compiling .cpp file using hipcc
         run_command(
             [
-                "hipcc",
-                str(THIS_DIR / "hip_printf.cpp"),
+                "./hipcc",
+                str(THIS_DIR / "hipcc_check.cpp"),
                 "-o",
-                str(THIS_DIR / "hip_printf"),
-            ]
+                str(THIS_DIR / "hipcc_check"),
+            ],
+            cwd=str(THEROCK_BIN_DIR),
         )
 
-        # Running the executable
-        output = run_command([str(THIS_DIR / "hip_printf")])
-        check.is_not_none(re.search(r"Thread.*is\swriting", output))
+        # Running and checking the executable
+        process = run_command(["./hipcc_check"], cwd=str(THIS_DIR))
+        check.equal(process.returncode, 0)
+        check.greater(os.path.getsize(str(THIS_DIR / "hipcc_check")), 0)

Original file line number	Diff line number	Diff line change
`@@ -13,4 +13,5 @@ int main() {`
`13`	`13`	`hipMalloc(&d_buf, size * sizeof(int));`
`14`	`14`	`hipLaunchKernelGGL(squares, gridsize, blocksize, 0, 0, d_buf);`
`15`	`15`	`hipDeviceSynchronize();`
	`16`	`+ return 0;`
`16`	`17`	`}`