diff --git a/.github/workflows/linux_matrix_e2e_on_nightly.yml b/.github/workflows/linux_matrix_e2e_on_nightly.yml index 5351c0fd635c1..7911f31ba069a 100644 --- a/.github/workflows/linux_matrix_e2e_on_nightly.yml +++ b/.github/workflows/linux_matrix_e2e_on_nightly.yml @@ -25,6 +25,7 @@ jobs: extra_cmake_args: -DHIP_PLATFORM="AMD" -DAMD_ARCH="gfx1031" extra_image_opts: --device=/dev/kfd target_devices: all + reset_gpu: false - name: Intel runner: '["Linux", "gen9"]' @@ -32,6 +33,7 @@ jobs: extra_cmake_args: extra_image_opts: -u 1001 target_devices: all + reset_gpu: true - name: ESIMD Emu runner: '["Linux", "x86-cpu"]' @@ -39,6 +41,7 @@ jobs: extra_cmake_args: extra_image_opts: -u 1001 target_devices: ext_intel_esimd_emulator:gpu + reset_gpu: false uses: ./.github/workflows/linux_single_e2e_on_nightly.yml with: name: ${{ matrix.name }} @@ -48,6 +51,7 @@ jobs: extra_image_opts: ${{ matrix.extra_image_opts }} target_devices: ${{ matrix.target_devices }} ref: ${{ inputs.ref }} + reset_gpu: ${{ matrix.reset_gpu }} aws_start: name: AWS Start @@ -69,6 +73,7 @@ jobs: extra_image_opts: --gpus all target_devices: all ref: ${{ inputs.ref }} + reset_gpu: false aws_stop: name: AWS Stop diff --git a/.github/workflows/linux_single_e2e_on_nightly.yml b/.github/workflows/linux_single_e2e_on_nightly.yml index ab42972c5938c..49fd542f02f84 100644 --- a/.github/workflows/linux_single_e2e_on_nightly.yml +++ b/.github/workflows/linux_single_e2e_on_nightly.yml @@ -17,6 +17,8 @@ on: type: string ref: type: string + reset_gpu: + type: string jobs: lin_e2e_only: @@ -26,6 +28,11 @@ jobs: image: ${{ inputs.image }} options: --device=/dev/dri --privileged --cap-add SYS_ADMIN ${{ inputs.extra_image_opts }} steps: + - name: Reset GPU + if: inputs.reset_gpu == 'true' + run: | + sudo mount -t debugfs none /sys/kernel/debug + sudo bash -c 'echo 1 > /sys/kernel/debug/dri/0/i915_wedged' - uses: actions/checkout@v3 with: path: llvm diff --git a/.github/workflows/sycl_linux_build_and_test.yml b/.github/workflows/sycl_linux_build_and_test.yml index 0ea7e37434608..c023be7de4902 100644 --- a/.github/workflows/sycl_linux_build_and_test.yml +++ b/.github/workflows/sycl_linux_build_and_test.yml @@ -211,7 +211,7 @@ jobs: options: ${{ matrix.container_options }} steps: - name: Reset GPU - if: ${{ contains(matrix.config, 'gen9') }} + if: ${{ contains(matrix.runs-on, 'gen9') && contains(matrix.runs-on, 'Linux') }} run: | sudo mount -t debugfs none /sys/kernel/debug sudo bash -c 'echo 1 > /sys/kernel/debug/dri/0/i915_wedged'