diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index 0f2cbc90a7640..bc92024edf42b 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -27,7 +27,7 @@ jobs:
     if: github.repository_owner == 'taichi-dev'
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
 
       - name: Save new version
         run: |
@@ -70,66 +70,49 @@ jobs:
       matrix: ${{ fromJson(needs.matrix_prep.outputs.matrix) }}
     runs-on: [self-hosted, cuda, vulkan, cn, release]
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
         with:
           submodules: 'recursive'
+          fetch-depth: '0'
 
-      - name: Get sccache cache
-        uses: actions/cache@v2
-        with:
-          path: sccache_cache
-          key: sccache-linux-gpu-${{ github.sha }}
-          restore-keys: |
-            sccache-linux-gpu-
+      - name: Prepare Environment
+        run: |
+          . .github/workflows/scripts/common-utils.sh
+          prepare-build-cache
+          echo CI_DOCKER_RUN_EXTRA_ARGS="-v $(pwd):/home/dev/taichi" >> $GITHUB_ENV
 
       - name: Build
         run: |
-          mkdir -m777 shared
-          docker create --user dev --name taichi_build --gpus all -v /tmp/.X11-unix:/tmp/.X11-unix \
-            -e DISPLAY -e PY -e GPU_BUILD -e TAICHI_CMAKE_ARGS -e PROJECT_NAME \
+          . .github/workflows/scripts/common-utils.sh
+
+          ci-docker-run-gpu --name taichi-build \
             registry.taichigraphics.com/taichidev-ubuntu18.04:v0.3.4 \
-            /home/dev/taichi/.github/workflows/scripts/unix_build.sh
-          tar -cf - ../${{ github.event.repository.name }} --mode u=+rwx,g=+rwx,o=+rwx --owner 1000 --group 1000 | docker cp - taichi_build:/home/dev/
-          docker start -a taichi_build
-          docker cp taichi_build:/home/dev/taichi/dist shared/dist
-          docker cp taichi_build:/home/dev/taichi/build shared/build
+            /home/dev/taichi/.github/workflows/scripts/unix-build.sh
         env:
           PY: ${{ matrix.conda_python }}
-          GPU_BUILD: ON
-          TAICHI_CMAKE_ARGS: -DTI_WITH_OPENGL:BOOL=ON -DTI_WITH_CC:BOOL=OFF -DTI_WITH_VULKAN:BOOL=ON -DTI_BUILD_TESTS:BOOL=ON -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CXX_COMPILER_LAUNCHER=sccache
           PROJECT_NAME: ${{ matrix.name }}
-          DISPLAY: ':1'
+          TAICHI_CMAKE_ARGS: >-
+            -DTI_WITH_OPENGL:BOOL=ON
+            -DTI_WITH_CC:BOOL=OFF
+            -DTI_WITH_VULKAN:BOOL=ON
+            -DTI_BUILD_TESTS:BOOL=ON
 
       - name: Archive Wheel Artifacts
         uses: actions/upload-artifact@v3
         with:
           name: dist
-          path: shared/dist/*.whl
+          path: dist/*.whl
           retention-days: 20
 
       - name: Test
         run: |
-          docker create --user dev --name taichi_test --gpus all -v /tmp/.X11-unix:/tmp/.X11-unix \
-            -e DISPLAY -e PY -e GPU_TEST \
-            registry.taichigraphics.com/taichidev-ubuntu18.04:v0.3.4 \
-            /home/dev/unix_test.sh
-          docker cp .github/workflows/scripts/unix_test.sh taichi_test:/home/dev/unix_test.sh
-          docker cp .github/workflows/scripts/common-utils.sh taichi_test:/home/dev/common-utils.sh
-          docker cp ./requirements_test.txt taichi_test:/home/dev/requirements_test.txt
-          docker cp shared/dist/ taichi_test:/home/dev/
-          docker cp shared/build/ taichi_test:/home/dev/
-          docker cp pyproject.toml taichi_test:/home/dev/
-          docker cp tests/ taichi_test:/home/dev/
-          docker start -a taichi_test
+          . .github/workflows/scripts/common-utils.sh
+
+          ci-docker-run-gpu --name taichi-test \
+             registry.taichigraphics.com/taichidev-ubuntu18.04:v0.3.4 \
+             /home/dev/taichi/.github/workflows/scripts/unix_test.sh
         env:
           PY: ${{ matrix.conda_python }}
-          GPU_TEST: ON
-          DISPLAY: ':1'
-
-      - name: clean docker container
-        if: always()
-        run: |
-          docker rm taichi_build taichi_test -f
 
   build_and_test_manylinux2014:
     name: Build and Upload (manylinux2014)
@@ -139,66 +122,52 @@ jobs:
       matrix: ${{ fromJson(needs.matrix_prep.outputs.matrix) }}
     runs-on: [self-hosted, cuda, cn, release]
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
         with:
-          submodules: "recursive"
+          submodules: 'recursive'
+          fetch-depth: '0'
 
-      - name: Get sccache cache
-        uses: actions/cache@v2
-        with:
-          path: sccache_cache
-          key: sccache-manylinux-${{ github.sha }}
-          restore-keys: |
-            sccache-manylinux-
+      - name: Prepare Environment
+        run: |
+          . .github/workflows/scripts/common-utils.sh
+          prepare-build-cache
+          echo CI_DOCKER_RUN_EXTRA_ARGS="-v $(pwd):/home/dev/taichi" >> $GITHUB_ENV
 
       - name: Build
         run: |
-          mkdir -m777 shared
-          docker create --user dev --name taichi_build --gpus all \
-            -e PY -e GPU_BUILD -e TAICHI_CMAKE_ARGS -e PROJECT_NAME \
-            registry.taichigraphics.com/taichidev-manylinux2014-cuda:v0.0.0 \
-            /home/dev/taichi/.github/workflows/scripts/unix_build.sh
-          tar -cf - ../${{ github.event.repository.name }} --mode u=+rwx,g=+rwx,o=+rwx --owner 1000 --group 1000 | docker cp - taichi_build:/home/dev/
-          docker start -a taichi_build
-          docker cp taichi_build:/home/dev/taichi/dist shared/dist
-          docker cp taichi_build:/home/dev/taichi/build shared/build
+          . .github/workflows/scripts/common-utils.sh
+
+          ci-docker-run-gpu --name taichi-test \
+             registry.taichigraphics.com/taichidev-manylinux2014-cuda:v0.0.0 \
+             /home/dev/taichi/.github/workflows/scripts/unix-build.sh
+
         env:
           PY: ${{ matrix.conda_python }}
-          GPU_BUILD: ON
-          TAICHI_CMAKE_ARGS: -DTI_WITH_OPENGL:BOOL=OFF -DTI_WITH_CC:BOOL=OFF -DTI_WITH_VULKAN:BOOL=OFF -DTI_BUILD_TESTS:BOOL=ON -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CXX_COMPILER_LAUNCHER=sccache
           PROJECT_NAME: ${{ matrix.name }}
+          TAICHI_CMAKE_ARGS: >-
+            -DTI_WITH_OPENGL:BOOL=OFF
+            -DTI_WITH_CC:BOOL=OFF
+            -DTI_WITH_VULKAN:BOOL=OFF
+            -DTI_BUILD_TESTS:BOOL=ON
 
       - name: Archive Wheel Artifacts
         uses: actions/upload-artifact@v3
         with:
           name: dist
-          path: shared/dist/*.whl
+          path: dist/*.whl
           retention-days: 20
 
       - name: Test
         run: |
-          docker create --user dev --name taichi_test --gpus all -e TI_WANTED_ARCHS \
-            -e PY -e GPU_TEST \
-            registry.taichigraphics.com/taichidev-manylinux2014-cuda:v0.0.0 \
-            /home/dev/unix_test.sh
-          docker cp .github/workflows/scripts/unix_test.sh taichi_test:/home/dev/unix_test.sh
-          docker cp .github/workflows/scripts/common-utils.sh taichi_test:/home/dev/common-utils.sh
-          docker cp ./requirements_test.txt taichi_test:/home/dev/requirements_test.txt
-          docker cp shared/dist/ taichi_test:/home/dev/
-          docker cp shared/build/ taichi_test:/home/dev/
-          docker cp pyproject.toml taichi_test:/home/dev/
-          docker cp tests/ taichi_test:/home/dev/
-          docker start -a taichi_test
+          . .github/workflows/scripts/common-utils.sh
+
+          ci-docker-run-gpu --name taichi-test \
+             registry.taichigraphics.com/taichidev-ubuntu18.04:v0.3.4 \
+             /home/dev/taichi/.github/workflows/scripts/unix_test.sh
         env:
           PY: ${{ matrix.conda_python }}
-          GPU_TEST: ON
           TI_WANTED_ARCHS: "cuda,cpu"
 
-      - name: clean docker container
-        if: always()
-        run: |
-          docker rm taichi_build taichi_test -f
-
   build_and_test_mac:
     name: Build and Upload (macOS only)
     needs: matrix_prep
@@ -209,17 +178,15 @@ jobs:
     env:
       PY: ${{ matrix.python }}
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
         with:
           submodules: 'recursive'
+          fetch-depth: '0'
 
-      - name: Get sccache cache
-        uses: actions/cache@v2
-        with:
-          path: sccache_cache
-          key: sccache-mac-${{ github.sha }}
-          restore-keys: |
-            sccache-mac-
+      - name: Prepare Environment
+        run: |
+          . .github/workflows/scripts/common-utils.sh
+          prepare-build-cache
 
       - name: Download Pre-Built LLVM 10.0.0
         run: python misc/ci_download.py
@@ -230,14 +197,15 @@ jobs:
         run: |
           brew install molten-vk
           export PATH=$(pwd)/taichi-llvm/bin/:$PATH
-          export PATH=$(ls -d ~/mini*/envs/$PY/bin):$PATH
-          bash .github/workflows/scripts/unix_build.sh
-          brew uninstall molten-vk
-          echo PATH=$PATH >> $GITHUB_ENV
+          .github/workflows/scripts/unix-build.sh
         env:
-          TAICHI_CMAKE_ARGS: -DTI_WITH_VULKAN:BOOL=ON -DTI_WITH_OPENGL:BOOL=OFF -DTI_WITH_CC:BOOL=OFF -DTI_BUILD_TESTS:BOOL=ON -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CXX_COMPILER_LAUNCHER=sccache
           PROJECT_NAME: ${{ matrix.name }}
           CXX: clang++
+          TAICHI_CMAKE_ARGS: >-
+            -DTI_WITH_VULKAN:BOOL=ON
+            -DTI_WITH_OPENGL:BOOL=OFF
+            -DTI_WITH_CC:BOOL=OFF
+            -DTI_BUILD_TESTS:BOOL=ON
 
       - name: Archive Wheel Artifacts
         uses: actions/upload-artifact@v3
@@ -263,32 +231,30 @@ jobs:
       run:
         shell: '/usr/bin/arch -arch arm64e /bin/bash --noprofile --norc -eo pipefail {0}'
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
         with:
           submodules: 'recursive'
+          fetch-depth: '0'
 
-      - name: Get sccache cache
-        uses: actions/cache@v2
-        with:
-          path: sccache_cache
-          key: sccache-m1-${{ github.sha }}
-          restore-keys: |
-            sccache-m1-
+      - name: Prepare Environment
+        run: |
+          . .github/workflows/scripts/common-utils.sh
+          prepare-build-cache
 
       - name: Build
         run: |
           brew install molten-vk
-          # We hacked here because conda activate in CI won't update python PATH
-          # automatically. So we don't activate and use desired python version
-          # directly.
-          export PATH=/Users/github/miniforge3/envs/$PYTHON/bin:$PATH
-          bash .github/workflows/scripts/unix_build.sh
-          brew uninstall molten-vk
+          .github/workflows/scripts/unix-build.sh
         env:
-          TAICHI_CMAKE_ARGS: -DTI_WITH_OPENGL:BOOL=OFF -DTI_WITH_CUDA:BOOL=OFF -DTI_WITH_CC:BOOL=OFF -DTI_WITH_VULKAN:BOOL=ON -DTI_BUILD_TESTS:BOOL=ON -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CXX_COMPILER_LAUNCHER=sccache
           PROJECT_NAME: ${{ matrix.name }}
-          PYTHON: ${{ matrix.python }}
+          PY: ${{ matrix.python }}
           CXX: clang++
+          TAICHI_CMAKE_ARGS: >-
+            -DTI_WITH_OPENGL:BOOL=OFF
+            -DTI_WITH_CUDA:BOOL=OFF
+            -DTI_WITH_CC:BOOL=OFF
+            -DTI_WITH_VULKAN:BOOL=ON
+            -DTI_BUILD_TESTS:BOOL=ON
 
       - name: Archive Wheel Artifacts
         uses: actions/upload-artifact@v3
@@ -299,11 +265,10 @@ jobs:
 
       - name: Test
         run: |
-          export PATH=/Users/github/miniforge3/envs/$PYTHON/bin:$PATH
           .github/workflows/scripts/unix_test.sh
         env:
           TI_WANTED_ARCHS: 'metal,vulkan,cpu'
-          PYTHON: ${{ matrix.python }}
+          PY: ${{ matrix.python }}
           GPU_TEST: ON
 
   build_and_test_macos_1014:
@@ -314,32 +279,32 @@ jobs:
       matrix: ${{ fromJson(needs.matrix_prep.outputs.matrix) }}
     runs-on: [self-hosted, macos10.14]
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
         with:
           submodules: 'recursive'
+          fetch-depth: '0'
 
-      - name: Get sccache cache
-        uses: actions/cache@v2
-        with:
-          path: sccache_cache
-          key: sccache-1014-${{ github.sha }}
-          restore-keys: |
-            sccache-1014-
+      - name: Prepare Environment
+        run: |
+          . .github/workflows/scripts/common-utils.sh
+          prepare-build-cache
 
       - name: Build
         run: |
-          # We hacked here because conda activate in CI won't update python PATH
-          # automatically. So we don't activate and use desired python version
-          # directly.
-          export PATH=/Users/buildbot6/miniconda3/envs/$PYTHON/bin:$PATH
+          # TODO: should embed LLVM as submodule
           export LLVM_DIR=/Users/buildbot6/taichi-llvm-10.0.0-macos
           export PATH=$LLVM_DIR/bin:$PATH
-          bash .github/workflows/scripts/unix_build.sh
+          .github/workflows/scripts/unix-build.sh
         env:
-          TAICHI_CMAKE_ARGS: -DTI_WITH_OPENGL:BOOL=OFF -DTI_WITH_CUDA:BOOL=OFF -DTI_WITH_CC:BOOL=OFF -DTI_WITH_VULKAN:BOOL=OFF -DTI_BUILD_TESTS:BOOL=ON -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CXX_COMPILER_LAUNCHER=sccache
           PROJECT_NAME: ${{ matrix.name }}
-          PYTHON: ${{ matrix.python }}
+          PY: ${{ matrix.python }}
           CXX: clang++
+          TAICHI_CMAKE_ARGS: >-
+            -DTI_WITH_OPENGL:BOOL=OFF
+            -DTI_WITH_CUDA:BOOL=OFF
+            -DTI_WITH_CC:BOOL=OFF
+            -DTI_WITH_VULKAN:BOOL=OFF
+            -DTI_BUILD_TESTS:BOOL=ON
 
       - name: Archive Wheel Artifacts
         uses: actions/upload-artifact@v3
@@ -350,11 +315,10 @@ jobs:
 
       - name: Test
         run: |
-          export PATH=/Users/buildbot6/miniconda3/envs/$PYTHON/bin:$PATH
           .github/workflows/scripts/unix_test.sh
         env:
           TI_WANTED_ARCHS: 'cpu'
-          PYTHON: ${{ matrix.python }}
+          PY: ${{ matrix.python }}
 
   build_and_test_windows:
     name: Build and Upload (Windows only)
@@ -362,35 +326,24 @@ jobs:
     strategy:
       fail-fast: false
       matrix: ${{ fromJson(needs.matrix_prep.outputs.matrix) }}
-    runs-on: windows-latest
+    runs-on: [self-hosted, windows, cuda, OpenGL]
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
         with:
           submodules: 'recursive'
+          fetch-depth: '0'
 
       - uses: actions/setup-python@v2
         with:
           python-version: ${{ matrix.python }}
 
-      - name: Add Visual Studio Shell to ENV
-        uses: egor-tensin/vs-shell@v2
-        with:
-          arch: x64
-
-      - name: Get sccache cache
-        uses: actions/cache@v2
-        with:
-          path: ccache_cache
-          key: ccache-win64-clang-${{ github.sha }}
-          restore-keys: |
-            ccache-win64-clang-
-
-      - name: Build Python Wheel
-        shell: powershell
+      - name: Build
+        shell: pwsh
+        if: ${{ needs.check_files.outputs.run_job != 'false' }}
         run: |
-          .\.github\workflows\scripts\win_build.ps1 -installVulkan -libsDir C:\
-          venv\Scripts\python -m pip install $(dir dist\*.whl)
+          .\.github\workflows\scripts\win_build.ps1 -llvmVer 10 -installVulkan -libsDir "$env:LocalAppData/buildbot"
         env:
+          PY: ${{ matrix.python }}
           PROJECT_NAME: ${{ matrix.name }}
 
       - name: Archive Wheel Artifacts
@@ -401,16 +354,12 @@ jobs:
           retention-days: 20
 
       - name: Test
-        shell: powershell
+        shell: pwsh
         run: |
-          $env:PATH = ";C:\taichi_llvm\bin;C:\taichi_clang\bin;" + $env:PATH
-          . venv\Scripts\activate.ps1
-          python -c "import taichi"
-          pip install torch
-          pip install -r requirements_test.txt
-          ti diagnose
-          python tests/run_tests.py -vr2 -t2
+          .\.github\workflows\scripts\win_test.ps1 -libsDir "$env:LocalAppData/buildbot"
         env:
+          PY: ${{ matrix.python }}
+          TI_WANTED_ARCHS: "cpu,cuda,opengl,vulkan"
           TI_SKIP_VERSION_CHECK: ON
 
   upload_to_pypi:
@@ -418,7 +367,7 @@ jobs:
     needs: [build_and_test_linux, build_and_test_manylinux2014, build_and_test_mac, build_and_test_m1, build_and_test_macos_1014, build_and_test_windows]
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
       - uses: actions/setup-python@v2
         with:
           python-version: 3.9
@@ -446,7 +395,7 @@ jobs:
     runs-on: ubuntu-latest
     if: ${{ github.event.inputs.version }}
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
       - uses: actions/setup-python@v2
         with:
           python-version: 3.9
diff --git a/.github/workflows/scripts/android-build.sh b/.github/workflows/scripts/android-build.sh
index 2f9df6681601e..339d01e2311cb 100755
--- a/.github/workflows/scripts/android-build.sh
+++ b/.github/workflows/scripts/android-build.sh
@@ -14,6 +14,7 @@ setup-sccache-local
 setup_python
 setup-android-ndk-env
 
+python -m pip install -r requirements_dev.txt
 python setup.py clean
 python setup.py build_ext
 cd build
diff --git a/.github/workflows/scripts/common-utils.ps1 b/.github/workflows/scripts/common-utils.ps1
new file mode 100644
index 0000000000000..e8e5dd9555628
--- /dev/null
+++ b/.github/workflows/scripts/common-utils.ps1
@@ -0,0 +1,279 @@
+function Info($text, $prefix="BUILD") {
+    Write-Host -ForegroundColor Green "[$prefix] $text"
+}
+
+function ProbeVirtualMemory {
+    $MethodDefinition = @"
+using System;
+using System.Runtime.InteropServices;
+
+public class VirtualProbe {
+    [DllImport("kernel32.dll", SetLastError = true)]
+    public static extern IntPtr GetCurrentProcess();
+    [DllImport("kernel32.dll", SetLastError=true, ExactSpelling=true)]
+    public static extern IntPtr VirtualAllocEx(IntPtr hProcess, IntPtr lpAddress, IntPtr dwSize, uint flAllocationType, uint flProtect);
+    [DllImport("kernel32.dll", SetLastError=true, ExactSpelling=true)]
+    public static extern bool VirtualFreeEx(IntPtr hProcess, IntPtr lpAddress, IntPtr dwSize, uint dwFreeType);
+}
+"@
+    if(-not ('VirtualProbe' -as [type])) {
+        Add-Type -TypeDefinition $MethodDefinition -Language CSharp
+    }
+    $totalSz = 0
+    $attempt = [int64](512 * 1024 * 1024 * 1024)
+    $allocated = @()
+    $me = [VirtualProbe]::GetCurrentProcess()
+    while($attempt -gt 256 * 1024 * 1024) {
+        $ptr = [VirtualProbe]::VirtualAllocEx($me, 0, $attempt, 0x3000, 0x01)  # MEM_COMMIT | MEM_RESERVE, PAGE_NOACCESS
+        if($ptr -eq 0) {
+            $attempt = [int64]($attempt / 2)
+        } else {
+            $allocated += $ptr
+            $totalSz += $attempt
+        }
+    }
+
+    foreach($ptr in $allocated) {
+        $_ = [VirtualProbe]::VirtualFreeEx($me, $ptr, 0, 0x8000)  # MEM_RELEASE
+    }
+
+    return $totalSz
+}
+
+function EstimateNumProcs($expectedPhysicalMemoryUsage = 512, $expectedVirtualMemoryUsage = 2048) {
+    # Things on Windows is complicated.
+    # Windows never overcommits, so we can't just use physical memory to estimate
+    # parallelism, pagefile (aka. swap on Linux but with subtle difference) must
+    # also be considered. What's more irratating is that pagefile is not
+    # a fixed size, it can grow on demand, Win32_OperatingSystem.FreeVirtualMemory
+    # is therefore almost always smaller than what you can really use.
+    # Aaaand, there are containers, where you can't even get the disk size or
+    # pagefile status.
+    # So we just probe the available virtual memory size.
+    $os = Get-CimInstance -ClassName Win32_OperatingSystem
+    $phys = $os.TotalVisibleMemorySize / 1024
+    $maxVirt = [int]($(ProbeVirtualMemory) / 1024 / 1024)
+    $procsByVirtMem = [int]($maxVirt / $expectedVirtualMemoryUsage)
+    $procsByPhysMem = [int](($phys - 1024) / $expectedPhysicalMemoryUsage)
+    $procsByProcessors = [int]($env:NUMBER_OF_PROCESSORS)
+    Info "Physical memory: $phys MB" "ESTIMATE"
+    Info "Allocatable (virtual) memory: $maxVirt MB" "ESTIMATE"
+    Info "Processors: $procsByProcessors" "ESTIMATE"
+    Info "Parallelism by physical memory: $procsByPhysMem" "ESTIMATE"
+    Info "Parallelism by allocatable memory: $procsByVirtMem" "ESTIMATE"
+    $procs = [int]([Math]::Min($procsByPhysMem, [Math]::Min($procsByVirtMem, $procsByProcessors)))
+    Info "Parallelism: $procs" "ESTIMATE"
+    # TODO: GPU memory
+    return $procs
+}
+
+function Setup-VS {
+    Info "Setting up Visual Studio"
+    foreach($progRoot in $env:ProgramFiles, ${env:ProgramFiles(x86)}) {
+        $vsBase = Join-Path $progRoot 'Microsoft Visual Studio'
+        foreach($ver in '2022','2019') {
+            foreach($edition in 'Enterprise','Professional','Community','BuildTools') {
+                $vsPath = Join-Path $vsBase $ver $edition
+                $clangPath = Join-Path $vsPath "VC\Tools\Llvm\x64\bin\clang.exe"
+                if (Test-Path $clangPath) {
+                    $devShellModule = Get-Item $(Join-Path $vsPath "Common7\Tools\Microsoft.VisualStudio.DevShell.dll")
+                    Import-Module $devShellModule
+                    Enter-VsDevShell -VsInstallPath $vsPath -SkipAutomaticLocation -DevCmdArguments "-arch=x64"
+                    return
+                }
+            }
+        }
+    }
+
+    throw "Could not find Visual Studio with Clang"
+}
+
+function Setup-Python($libsDir, $version = "3.7") {
+    if ([string]::IsNullOrEmpty($version)) {
+        throw "Should specify Python version"
+    }
+
+    Info "Setting up Python environment $version"
+
+    function PipOps {
+        Invoke python -m pip install -U pip wheel
+        Invoke python -m pip uninstall taichi taichi-nightly -y
+        # These have to be re-installed to avoid strange certificate issue
+        # on CPU docker environment
+        Invoke python -m pip install --upgrade --force-reinstall numpy cmake wheel
+    }
+
+    if (Get-Command python -ErrorAction SilentlyContinue) {
+        $ver = & python --version
+        Info "Found $ver" "Python"
+        if ($ver.Startswith("Python ${version}.")) {
+            Info "Using $ver" "Python"
+            $venv = "$libsDir/taichi-venv-$version"
+            if(-not (Test-Path $venv)) {
+                Invoke python -m venv $venv
+            }
+            . "$libsDir/taichi-venv-$version/Scripts/activate.ps1"
+            PipOps
+            return
+        }
+    }
+
+    if (Get-Command conda -ErrorAction SilentlyContinue) {
+        Info "Using conda environment" "Python"
+        # <Workaround> bad conda in container
+        Invoke conda shell.powershell hook | Out-String | Invoke-Expression
+        # </Workaround>
+        $condaEnv = "$libsDir/taichi-conda-$version"
+        if (-not (Test-Path $condaEnv)) {
+            conda create -y -q --prefix=$condaEnv python=$version
+        }
+        conda activate $condaEnv
+        PipOps
+        return
+    }
+
+    throw "Could not setup Python"
+}
+
+function Resolve-Path-String-Force {
+    <#
+    .SYNOPSIS
+        Calls Resolve-Path but works for files that don't exist.
+    .REMARKS
+        From http://devhawk.net/blog/2010/1/22/fixing-powershells-busted-resolve-path-cmdlet
+    #>
+    param (
+        [string] $FileName
+    )
+
+    $FileName = Resolve-Path $FileName -ErrorAction SilentlyContinue `
+                                       -ErrorVariable _frperror
+    if (-not($FileName)) {
+        $FileName = $_frperror[0].TargetObject
+    }
+
+    return $FileName
+}
+
+function Invoke() {
+    # https://stackoverflow.com/questions/47032005/why-does-a-powershell-script-not-end-when-there-is-a-non-zero-exit-code-using-th
+    # A handy way to run a command, and automatically throw an error if the
+    # exit code is non-zero.
+    #
+    #
+    if ($args.Count -eq 0) {
+        throw "Must supply some arguments."
+    }
+
+    Write-Host -ForegroundColor Blue "[CMD] $args"
+
+    $command = $args[0]
+    $commandArgs = @()
+    if ($args.Count -gt 1) {
+        $commandArgs = $args[1..($args.Count - 1)]
+    }
+
+    & $command $commandArgs
+    $ok = $?
+    $result = $LASTEXITCODE
+
+    if (-not $ok) {
+        throw "$command $commandArgs failed."
+    }
+
+    if ($result -ne 0) {
+        throw "$command $commandArgs exited with code $result."
+    }
+}
+
+
+# SCCache not working on Windows (reports UnknownFlag with -Xclang)
+# Using CCache here
+function SetupCCacheLocal($root) {
+    $root = Resolve-Path-String-Force $root
+
+    New-Item -ItemType Directory -Path "$root\cache" -ErrorAction SilentlyContinue
+
+    $env:CCACHE_DIR = "$root/cache"
+    $env:CCACHE_TEMPDIR = "$env:TEMP/ccache-temp"
+    $env:CCACHE_MAXSIZE= "10G"
+    $env:CCACHE_LOG_FILE = "$root/ccache_error.log"
+
+    if (-not (Test-Path -Path "$root/bin/ccache.exe")) {
+        Info "Downloading sccache"
+        Push-Location "$root"
+
+        Invoke-WebRequest `
+            -Uri "https://github.com/ccache/ccache/releases/download/v4.5.1/ccache-4.5.1-windows-64.zip" `
+            -MaximumRetryCount 10 -RetryIntervalSec 300 `
+            -OutFile ccache.zip
+        Expand-Archive -Force ccache.zip .
+        Rename-Item -Force -Path "ccache-4.5.1-windows-64" -NewName "bin"
+        Pop-Location
+    }
+
+    $env:PATH += ";$root/bin"
+    $env:TAICHI_CMAKE_ARGS += " -DCMAKE_C_COMPILER_LAUNCHER=ccache -DCMAKE_CXX_COMPILER_LAUNCHER=ccache"
+
+    ccache -s -v
+}
+
+function ClearTaichiOfflineCache {
+    # Remove-Item -Force -Path "$env:LocalAppData/build-cache/dot-cache/" -Recurse -ErrorAction SilentlyContinue
+}
+
+function PrepareBuildCache {
+    "git-cache","pip-cache" | % {
+        New-Item -ItemType Directory -Path "$env:LocalAppData/build-cache/$_" -ErrorAction SilentlyContinue
+    }
+    Push-Location $env:LocalAppData/build-cache/git-cache
+    if (Test-Path -Path objects) {
+        Invoke git init --bare
+    }
+    Pop-Location
+}
+
+function CIDockerRun {
+    $containerName = $null
+    $shouldRm = $true
+
+    for($i = 0; $i -lt $args.Count; $i++) {
+        $v = $args[$i]
+        if($v -eq "-n" -or $v -eq "--name") {
+            $containerName = $args[$i+1]
+            $shouldRm = $false
+            $i++
+        }
+    }
+
+    if($containerName) {
+        Invoke docker rm -f $containerName
+    }
+
+    if($shouldRm -and -not $args.Contains("--rm")) {
+        $args = ,("--rm") + $args
+    }
+
+    $TiEnvs = @()
+    Get-ChildItem "env:*" | % {
+        if($_.Key.Startswith("TI_")) {
+            $TiEnvs += "-e", $_.Key
+        }
+    }
+
+    $extraArgs = ($env:CI_DOCKER_RUN_EXTRA_ARGS ?? "").Trim().Split()
+
+    Invoke docker run `
+        -i `
+        -e PY `
+        -e PROJECT_NAME `
+        -e TAICHI_CMAKE_ARGS `
+        -e IN_DOCKER=true `
+        -e PIP_CACHE_DIR=X:/pip-cache `
+        -e GIT_ALTERNATE_OBJECT_DIRECTORIES=X:/git-cache/objects `
+        -e TI_CI=1 `
+        @TiEnvs `
+        -v (($env:LocalAppData -replace "\\", "/") + "/build-cache:X:") `
+        @extraArgs `
+        @args
+}
diff --git a/.github/workflows/scripts/common-utils.sh b/.github/workflows/scripts/common-utils.sh
index 39f59d2f0a933..1b4e5a27b36b3 100644
--- a/.github/workflows/scripts/common-utils.sh
+++ b/.github/workflows/scripts/common-utils.sh
@@ -1,48 +1,17 @@
 #!/bin/bash
 
-check_in_docker() {
-    # This is a temporary solution to detect in a docker, but it should work
-    if [[ $(whoami) == "dev" ]]; then
-        echo "true"
-    else
-        echo "false"
-    fi
-}
-
-setup_sccache() {
-    export SCCACHE_DIR=$(pwd)/sccache_cache
-    export SCCACHE_CACHE_SIZE="128M"
-    export SCCACHE_LOG=error
-    export SCCACHE_ERROR_LOG=$(pwd)/sccache_error.log
-    mkdir -p "$SCCACHE_DIR"
-    echo "sccache dir: $SCCACHE_DIR"
-    ls -la "$SCCACHE_DIR"
-
-    if [[ $OSTYPE == "linux-"* ]]; then
-        wget https://github.com/mozilla/sccache/releases/download/v0.2.15/sccache-v0.2.15-x86_64-unknown-linux-musl.tar.gz
-        tar -xzf sccache-v0.2.15-x86_64-unknown-linux-musl.tar.gz
-        chmod +x sccache-v0.2.15-x86_64-unknown-linux-musl/sccache
-        export PATH=$(pwd)/sccache-v0.2.15-x86_64-unknown-linux-musl:$PATH
-    elif [[ $(uname -m) == "arm64" ]]; then
-        wget https://github.com/mozilla/sccache/releases/download/v0.2.15/sccache-v0.2.15-aarch64-apple-darwin.tar.gz
-        tar -xzf sccache-v0.2.15-aarch64-apple-darwin.tar.gz
-        chmod +x sccache-v0.2.15-aarch64-apple-darwin/sccache
-        export PATH=$(pwd)/sccache-v0.2.15-aarch64-apple-darwin:$PATH
-    else
-        wget https://github.com/mozilla/sccache/releases/download/v0.2.15/sccache-v0.2.15-x86_64-apple-darwin.tar.gz
-        tar -xzf sccache-v0.2.15-x86_64-apple-darwin.tar.gz
-        chmod +x sccache-v0.2.15-x86_64-apple-darwin/sccache
-        export PATH=$(pwd)/sccache-v0.2.15-x86_64-apple-darwin:$PATH
-    fi
-}
+set -x
 
 setup_python() {
-    if [[ "$(check_in_docker)" == "true" ]]; then
-        source $HOME/miniconda/etc/profile.d/conda.sh
-        conda activate "$PY"
-    fi
+    for conda in miniconda miniconda3 miniforge3; do
+        if [[ -d $HOME/$conda ]]; then
+            source $HOME/$conda/bin/activate
+            conda activate "$PY"
+            break
+        fi
+    done
+    python3 -m pip install -U pip
     python3 -m pip uninstall taichi taichi-nightly -y
-    python3 -m pip install -r requirements_dev.txt
 }
 
 function setup-sccache-local {
@@ -81,6 +50,110 @@ function setup-sccache-local {
     export TAICHI_CMAKE_ARGS="$TAICHI_CMAKE_ARGS -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CXX_COMPILER_LAUNCHER=sccache"
 }
 
+function clear-taichi-offline-cache {
+    rm -rf ~/build-cache/dot-cache/taichi  # Clear taichi offline cache
+    rm -rf ~/.cache/taichi  # Clear taichi offline cache
+}
+
+function prepare-build-cache {
+    export CACHE_HOME=${1:-$HOME/build-cache}
+    mkdir -p $CACHE_HOME/{dot-cache/pip,dot-gradle,git-cache,sccache/{bin,cache}}
+    chmod 0777 $CACHE_HOME/ $CACHE_HOME/* || true
+    pushd $CACHE_HOME/git-cache
+    if [ ! -d objects ]; then git init --bare; fi
+    popd
+
+    clear-taichi-offline-cache
+
+    if [ ! -z $GITHUB_ENV ]; then
+        # for bare metal run
+        export SCCACHE_ROOT=$CACHE_HOME/sccache
+        export GIT_ALTERNATE_OBJECT_DIRECTORIES=$CACHE_HOME/git-cache/objects
+        echo SCCACHE_ROOT=$SCCACHE_ROOT >> $GITHUB_ENV
+        echo GIT_ALTERNATE_OBJECT_DIRECTORIES=$GIT_ALTERNATE_OBJECT_DIRECTORIES >> $GITHUB_ENV
+        echo CACHE_HOME=$CACHE_HOME >> $GITHUB_ENV
+    else
+        # container run
+        true
+    fi
+}
+
+function fix-build-cache-permission {
+    if [ "$(uname -s)" = "Linux" ]; then
+        sudo -n chown -R $(id -u):$(id -g) $CACHE_HOME || true
+    fi
+}
+
+function ci-docker-run {
+    ARGS="$@"
+    SHOULD_RM="--rm"
+    while [[ $# > 0 ]]; do
+        case $1 in
+            -n | --name)
+                shift
+                CONTAINER_NAME="$1"
+                SHOULD_RM=""
+                break
+                ;;
+        esac
+        shift
+    done
+
+    if [ ! -z $CONTAINER_NAME ]; then
+        docker rm -f $CONTAINER_NAME
+    fi
+
+    TI_ENVS=""
+    for i in $(env | grep ^TI_); do
+        TI_ENVS="$TI_ENVS -e $i"
+    done
+
+    CACHE_HOME=${CACHE_HOME:-$HOME/build-cache}
+
+    docker run \
+        -i \
+        $SHOULD_RM \
+        --user dev \
+        -e PY \
+        -e PROJECT_NAME \
+        -e TAICHI_CMAKE_ARGS \
+        -e IN_DOCKER=true \
+        -e TI_CI=1 \
+        $TI_ENVS \
+        -e SCCACHE_ROOT=/var/lib/sccache \
+        -e CACHE_HOME=/var/lib/cache-home \
+        -e GIT_ALTERNATE_OBJECT_DIRECTORIES=/var/lib/git-cache/objects \
+        -v $(readlink -f $CACHE_HOME):/var/lib/cache-home \
+        -v $(readlink -f $CACHE_HOME/sccache):/var/lib/sccache \
+        -v $(readlink -f $CACHE_HOME/git-cache):/var/lib/git-cache \
+        -v $(readlink -f $CACHE_HOME/dot-cache):/home/dev/.cache \
+        -v $(readlink -f $CACHE_HOME/dot-gradle):/home/dev/.gradle \
+        $CI_DOCKER_RUN_EXTRA_ARGS \
+        $ARGS
+}
+
+function ci-docker-run-gpu {
+    for i in {0..9}; do
+        if xset -display ":$i" -q >/dev/null 2>&1; then
+            break
+        fi
+    done
+
+    if [ $? -ne 0 ]; then
+        echo "No display!"
+        exit 1
+    fi
+
+    ci-docker-run \
+        --runtime=nvidia \
+        --gpus 'all,"capabilities=graphics,utility,display,video,compute"' \
+        -e DISPLAY=:$i \
+        -e GPU_BUILD=ON \
+        -e GPU_TEST=ON \
+        -v /tmp/.X11-unix:/tmp/.X11-unix \
+        $@
+}
+
 function setup-android-ndk-env {
     export ANDROID_NDK_ROOT=/android-sdk/ndk-bundle
     export ANDROID_CMAKE_ARGS="-DCMAKE_TOOLCHAIN_FILE=${ANDROID_NDK_ROOT}/build/cmake/android.toolchain.cmake -DANDROID_NATIVE_API_LEVEL=29 -DANDROID_ABI=arm64-v8a"
diff --git a/.github/workflows/scripts/unix-build-v2.sh b/.github/workflows/scripts/unix-build.sh
similarity index 84%
rename from .github/workflows/scripts/unix-build-v2.sh
rename to .github/workflows/scripts/unix-build.sh
index 1c64976c18d12..32c8322892807 100755
--- a/.github/workflows/scripts/unix-build-v2.sh
+++ b/.github/workflows/scripts/unix-build.sh
@@ -1,15 +1,15 @@
 #!/bin/bash
 
-# TODO: replace unix_build.sh
-# currently only used in android build job
 set -ex
 
 . $(dirname $0)/common-utils.sh
 
-IN_DOCKER=$(check_in_docker)
 [[ "$IN_DOCKER" == "true" ]] && cd taichi
 
+# TODO: Move llvm installation from container image to here
+
 build_taichi_wheel() {
+    python3 -m pip install -r requirements_dev.txt
     git fetch origin master --tags
     PROJECT_TAGS=""
     EXTRA_ARGS=""
@@ -27,9 +27,11 @@ build_taichi_wheel() {
     python3 misc/make_changelog.py --ver origin/master --repo_dir ./ --save
 
     python3 setup.py $PROJECT_TAGS bdist_wheel $EXTRA_ARGS
-    sccache -s
+    sccache -s || true
 }
 
+fix-build-cache-permission
+
 setup-sccache-local
 setup_python
 
diff --git a/.github/workflows/scripts/unix_build.sh b/.github/workflows/scripts/unix_build.sh
deleted file mode 100755
index d5565de735b15..0000000000000
--- a/.github/workflows/scripts/unix_build.sh
+++ /dev/null
@@ -1,38 +0,0 @@
-#!/bin/bash
-set -ex
-
-. $(dirname $0)/common-utils.sh
-
-IN_DOCKER=$(check_in_docker)
-[[ "$IN_DOCKER" == "true" ]] && cd taichi
-
-build_taichi_wheel() {
-    git fetch origin master --tags
-    PROJECT_TAGS=""
-    EXTRA_ARGS=""
-    if [ "$PROJECT_NAME" = "taichi-nightly" ]; then
-        PROJECT_TAGS="egg_info --tag-date"
-    fi
-
-    if [[ $OSTYPE == "linux-"* ]]; then
-        if [ -f /etc/centos-release ] ; then
-            EXTRA_ARGS="-p manylinux2014_x86_64"
-        else
-            EXTRA_ARGS="-p manylinux_2_27_x86_64"
-        fi
-    fi
-    python3 misc/make_changelog.py --ver origin/master --repo_dir ./ --save
-
-    exec env TAICHI_CMAKE_ARGS="${TAICHI_CMAKE_ARGS}" python3 setup.py $PROJECT_TAGS bdist_wheel $EXTRA_ARGS
-    sccache -s
-}
-
-setup_sccache
-setup_python
-
-build_taichi_wheel
-NUM_WHL=$(ls dist/*.whl | wc -l)
-if [ $NUM_WHL -ne 1 ]; then echo "ERROR: created more than 1 whl." && exit 1; fi
-
-chmod -R 777 "$SCCACHE_DIR"
-rm -f python/CHANGELOG.md
diff --git a/.github/workflows/scripts/unix_test.sh b/.github/workflows/scripts/unix_test.sh
index 0dc325118df66..ea8a0ae4e7a6d 100755
--- a/.github/workflows/scripts/unix_test.sh
+++ b/.github/workflows/scripts/unix_test.sh
@@ -7,15 +7,13 @@ export PYTHONUNBUFFERED=1
 
 export TI_SKIP_VERSION_CHECK=ON
 export TI_CI=1
-export TI_IN_DOCKER=$(check_in_docker)
 export LD_LIBRARY_PATH=$PWD/build/:$LD_LIBRARY_PATH
 export TI_OFFLINE_CACHE_FILE_PATH=$PWD/.cache/taichi
 
+setup_python
+
+[[ "$IN_DOCKER" == "true" ]] && cd taichi
 
-if [[ "$TI_IN_DOCKER" == "true" ]]; then
-    source $HOME/miniconda/etc/profile.d/conda.sh
-    conda activate "$PY"
-fi
 python3 -m pip install dist/*.whl
 if [ -z "$GPU_TEST" ]; then
     python3 -m pip install -r requirements_test.txt
diff --git a/.github/workflows/scripts/win_build.ps1 b/.github/workflows/scripts/win_build.ps1
index dba6d15e0cc9c..13da3ce821219 100644
--- a/.github/workflows/scripts/win_build.ps1
+++ b/.github/workflows/scripts/win_build.ps1
@@ -11,126 +11,105 @@ param (
 
 $ErrorActionPreference = "Stop"
 
-$RepoURL = 'https://github.com/taichi-dev/taichi'
+. $PSScriptRoot\common-utils.ps1
 
-function WriteInfo($text) {
-    Write-Host -ForegroundColor Green "[BUILD] $text"
+$libsDir = Resolve-Path-String-Force $libsDir
+if (-not (Test-Path $libsDir)) {
+    New-Item -ItemType Directory -Path $libsDir
 }
 
-# Get sccache
-$env:CCACHE_DIR="${pwd}/ccache_cache"
-$env:CCACHE_MAXSIZE="128M"
-$env:CCACHE_LOGFILE="${pwd}/ccache_error.log"
-WriteInfo("ccache dir: $Env:CCACHE_DIR")
-md "$Env:CCACHE_DIR" -ea 0
-if (-not (Test-Path "ccache-4.5.1-windows-64")) {
-    curl.exe --retry 10 --retry-delay 5 https://github.com/ccache/ccache/releases/download/v4.5.1/ccache-4.5.1-windows-64.zip -LO
-    7z x ccache-4.5.1-windows-64.zip
-    $env:PATH += ";${pwd}/ccache-4.5.1-windows-64"
-}
-ccache -v -s
+$RepoURL = 'https://github.com/taichi-dev/taichi'
 
-# WriteInfo("Install 7Zip")
-# Install-Module 7Zip4PowerShell -Force -Verbose -Scope CurrentUser
+SetupCCacheLocal "$libsDir/ccache"
 
 if ($clone) {
-    WriteInfo("Clone the repository")
-    git clone --recurse-submodules $RepoURL
+    Info("Clone the repository")
+    Invoke git clone --recurse-submodules $RepoURL
     Set-Location .\taichi
 }
 
-$libsDir = (Resolve-Path $libsDir).Path
+Setup-VS
 
-if (-not (Test-Path $libsDir)) {
-    New-Item -ItemType Directory -Path $libsDir
-}
 Push-Location $libsDir
 
-if (!$llvmVer.CompareTo("10")) {
-    if (-not (Test-Path "taichi_llvm")) {
-        WriteInfo("Download and extract LLVM")
-        curl.exe --retry 10 --retry-delay 5 https://github.com/taichi-dev/taichi_assets/releases/download/llvm10/taichi-llvm-10.0.0-msvc2019.zip -LO
-        if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE; }
-        7z x taichi-llvm-10.0.0-msvc2019.zip -otaichi_llvm
-    }
-	if (-not (Test-Path "taichi_clang")) {
-		WriteInfo("Download and extract Clang")
-		curl.exe --retry 10 --retry-delay 5 https://github.com/taichi-dev/taichi_assets/releases/download/llvm10/clang-10.0.0-win.zip -LO
-		if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE; }
-		7z x clang-10.0.0-win.zip -otaichi_clang
-	}
-} else {
-    if (-not (Test-Path "taichi_llvm_15")) {
-        WriteInfo("Download and extract LLVM")
-        curl.exe --retry 10 --retry-delay 5 https://github.com/python3kgae/taichi_assets/releases/download/llvm15_vs2019_clang/taichi-llvm-15.0.0-msvc2019.zip -LO
-        if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE; }
-        7z x taichi-llvm-15.0.0-msvc2019.zip -otaichi_llvm_15
+function DownloadDep {
+    param (
+        [string]$name,
+        [string]$outfile,
+        [string]$dir,
+        [string]$url
+    )
+    if (-not (Test-Path $dir)) {
+        Info("Download and extract $name")
+        Invoke-WebRequest `
+            -Uri $url `
+            -MaximumRetryCount 10 -RetryIntervalSec 5 `
+            -OutFile $outfile
+        Expand-Archive -Force $outfile $dir
     }
-	if (-not (Test-Path "taichi_clang_15")) {
-		WriteInfo("Download and extract Clang")
-		curl.exe --retry 10 --retry-delay 5 https://github.com/python3kgae/taichi_assets/releases/download/llvm15_vs2022_clang/clang-15.0.0-win.zip -LO
-		if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE; }
-		7z x clang-15.0.0-win.zip -otaichi_clang_15
-	}
 }
 
-
-if (!$llvmVer.CompareTo("10")) {
+if ($llvmVer -eq "10") {
+    DownloadDep LLVM llvm.zip taichi_llvm `
+        https://github.com/taichi-dev/taichi_assets/releases/download/llvm10/taichi-llvm-10.0.0-msvc2019.zip
+    DownloadDep Clang clang.zip taichi_clang `
+        https://github.com/taichi-dev/taichi_assets/releases/download/llvm10/clang-10.0.0-win.zip
     $env:LLVM_DIR = "$libsDir\taichi_llvm"
-} else {
+	$env:TAICHI_CMAKE_ARGS += " -DCLANG_EXECUTABLE=$($libsDir -replace "\\", "\\")\\taichi_clang\\bin\\clang++.exe"
+	$env:TAICHI_CMAKE_ARGS += " -DLLVM_AS_EXECUTABLE=$($libsDir -replace "\\", "\\")\\taichi_llvm\\bin\\llvm-as.exe"
+} elseif ($llvmVer -eq "15") {
+    DownloadDep LLVM-15 llvm-15.zip taichi_llvm_15 `
+        https://github.com/python3kgae/taichi_assets/releases/download/llvm15_vs2019_clang/taichi-llvm-15.0.0-msvc2019.zip
+    DownloadDep Clang-15 clang-15.zip taichi_clang_15 `
+		https://github.com/python3kgae/taichi_assets/releases/download/llvm15_vs2022_clang/clang-15.0.0-win.zip
     $env:LLVM_DIR = "$libsDir\taichi_llvm_15"
+	$env:TAICHI_CMAKE_ARGS += " -DCLANG_EXECUTABLE=$($libsDir -replace "\\", "\\")\\taichi_clang_15\\bin\\clang++.exe"
+	$env:TAICHI_CMAKE_ARGS += " -DLLVM_AS_EXECUTABLE=$($libsDir -replace "\\", "\\")\\taichi_llvm_15\\bin\\llvm-as.exe"
+    $env:TAICHI_CMAKE_ARGS += " -DTI_LLVM_15:BOOL=ON"
+} else {
+    throw "Unsupported LLVM version"
 }
 
 $env:TAICHI_CMAKE_ARGS += " -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_C_COMPILER=clang"
+
 if ($installVulkan) {
-    WriteInfo("Download and install Vulkan")
-    if (-not (Test-Path "VulkanSDK")) {
-        curl.exe --retry 10 --retry-delay 5 https://sdk.lunarg.com/sdk/download/1.2.189.0/windows/VulkanSDK-1.2.189.0-Installer.exe -Lo VulkanSDK.exe
-        if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE; }
-        $installer = Start-Process -FilePath VulkanSDK.exe -Wait -PassThru -ArgumentList @("/S");
+    if (-not (Test-Path $env:VULKAN_SDK)) {
+        Info("Download and install Vulkan")
+        Invoke-WebRequest `
+            -Uri 'https://sdk.lunarg.com/sdk/download/1.2.189.0/windows/VulkanSDK-1.2.189.0-Installer.exe' `
+            -MaximumRetryCount 10 -RetryIntervalSec 5 `
+            -OutFile VulkanSDK.exe
+        $installer = Start-Process -FilePath VulkanSDK.exe -Wait -PassThru -ArgumentList @("/S")
         $installer.WaitForExit();
     }
-    $env:VULKAN_SDK = "$libsDir\VulkanSDK\1.2.189.0"
+    $env:VULKAN_SDK = "C:\VulkanSDK\1.2.189.0"
     $env:PATH += ";$env:VULKAN_SDK\Bin"
     $env:TAICHI_CMAKE_ARGS += " -DTI_WITH_VULKAN:BOOL=ON"
 }
 
-$env:TAICHI_CMAKE_ARGS += " -DTI_WITH_C_API:BOOL=ON"
-$env:TAICHI_CMAKE_ARGS += " -DTI_WITH_DX11:BOOL=ON"
-
 Pop-Location
-clang --version
-
-WriteInfo("Setting up Python environment")
-python -m venv venv
-. venv\Scripts\activate.ps1
-python -m pip install wheel
-python -m pip install -r requirements_dev.txt
-if (-not $?) { exit 1 }
-WriteInfo("Building Taichi")
-if (!$llvmVer.CompareTo("10")) {
-	$env:TAICHI_CMAKE_ARGS += " -DCLANG_EXECUTABLE=$libsDir\\taichi_clang\\bin\\clang++.exe"
-	$env:TAICHI_CMAKE_ARGS += " -DLLVM_AS_EXECUTABLE=$libsDir\\taichi_llvm\\bin\\llvm-as.exe"
-} else {
-	$env:TAICHI_CMAKE_ARGS += " -DCLANG_EXECUTABLE=$libsDir\\taichi_clang_15\\bin\\clang++.exe"
-	$env:TAICHI_CMAKE_ARGS += " -DLLVM_AS_EXECUTABLE=$libsDir\\taichi_llvm_15\\bin\\llvm-as.exe"
-    $env:TAICHI_CMAKE_ARGS += " -DTI_LLVM_15:BOOL=ON"
-}
+Invoke clang --version
+
+Setup-Python $libsDir $env:PY
+
+Invoke python -m pip install -r requirements_dev.txt
+
+Info("Building Taichi")
+
 if ($install) {
     if ($develop) {
-        python setup.py develop
+        Invoke python setup.py develop
     } else {
-        python setup.py install
+        Invoke python setup.py install
     }
-    if (-not $?) { exit 1 }
-    WriteInfo("Build and install finished")
+    Info("Build and install finished")
 } else {
     if ($env:PROJECT_NAME -eq "taichi-nightly") {
-        python setup.py egg_info --tag-date bdist_wheel
+        Invoke python setup.py egg_info --tag-date bdist_wheel
     } else {
-        python setup.py bdist_wheel
+        Invoke python setup.py bdist_wheel
     }
-    if (-not $?) { exit 1 }
-    WriteInfo("Build finished")
+    Info("Build finished")
 }
+
 ccache -s -v
diff --git a/.github/workflows/scripts/win_build_test_cpu.ps1 b/.github/workflows/scripts/win_build_test_cpu.ps1
deleted file mode 100644
index 4aa90582517ed..0000000000000
--- a/.github/workflows/scripts/win_build_test_cpu.ps1
+++ /dev/null
@@ -1,116 +0,0 @@
-# Build script for windows CPU
-# TODO unify this with the other Win scripts
-
-param (
-    [switch]$clone = $false,
-    [switch]$install = $false,
-    [string]$libsDir = "C:\",
-    [string]$llvmVer = "10"
-)
-
-$ErrorActionPreference = "Stop"
-
-$RepoURL = 'https://github.com/taichi-dev/taichi'
-
-function WriteInfo($text) {
-    Write-Host -ForegroundColor Green "[BUILD] $text"
-}
-
-$libsDir = (Resolve-Path $libsDir).Path
-if (-not (Test-Path $libsDir)) {
-    New-Item -ItemType Directory -Path $libsDir
-}
-Set-Location $libsDir
-
-if (!$llvmVer.CompareTo("10")) {
-    if (-not (Test-Path "taichi_llvm")) {
-        WriteInfo("Download and extract LLVM")
-
-
-        curl.exe --retry 10 --retry-delay 5 https://github.com/taichi-dev/taichi_assets/releases/download/llvm10/taichi-llvm-10.0.0-msvc2019.zip -LO
-        if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE; }
-        7z x taichi-llvm-10.0.0-msvc2019.zip -otaichi_llvm
-    }
-} else {
-    if (-not (Test-Path "taichi_llvm_15")) {
-        WriteInfo("Download and extract LLVM")
-        curl.exe --retry 10 --retry-delay 5 https://github.com/python3kgae/taichi_assets/releases/download/llvm15_vs2019_clang_220731/taichi-llvm-15.0.0-msvc2019.zip -LO
-        if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE; }
-        7z x taichi-llvm-15.0.0-msvc2019.zip -otaichi_llvm_15
-    }
-}
-
-if (-not (Test-Path "taichi_clang")) {
-    WriteInfo("Download and extract Clang")
-    curl.exe --retry 10 --retry-delay 5 https://github.com/taichi-dev/taichi_assets/releases/download/llvm10/clang-10.0.0-win.zip -LO
-    if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE; }
-    7z x clang-10.0.0-win.zip -otaichi_clang
-}
-
-WriteInfo("Setting the env vars")
-if (!$llvmVer.CompareTo("10")) {
-    $env:LLVM_DIR = "C://taichi_llvm"
-} else {
-    $env:LLVM_DIR = "C:\\taichi_llvm_15"
-}
-
-#TODO enable build test
-$env:TAICHI_CMAKE_ARGS = "-DTI_WITH_OPENGL:BOOL=OFF -DTI_WITH_CC:BOOL=OFF -DTI_WITH_VULKAN:BOOL=OFF -DTI_WITH_CUDA:BOOL=OFF -DTI_BUILD_TESTS:BOOL=OFF"
-
-#TODO: For now we need to hard code the compiler path from build tools 2019
-$env:TAICHI_CMAKE_ARGS +=' -DCMAKE_CXX_COMPILER=C:/Program\ Files\ (x86)/Microsoft\ Visual\ Studio/2019/BuildTools/vc/Tools/Llvm/x64/bin/clang++.exe -DCMAKE_C_COMPILER=C:/Program\ Files\ (x86)/Microsoft\ Visual\ Studio/2019/BuildTools/vc/Tools/Llvm/x64/bin/clang.exe'
-$env:TAICHI_CMAKE_ARGS += " -DCLANG_EXECUTABLE=C:\\taichi_clang\\bin\\clang++.exe"
-
-if (!$llvmVer.CompareTo("10")) {
-    $env:TAICHI_CMAKE_ARGS += " -DLLVM_AS_EXECUTABLE=C:\\taichi_llvm\\bin\\llvm-as.exe -DTI_WITH_VULKAN:BOOL=OFF"
-} else {
-    $env:TAICHI_CMAKE_ARGS += " -DLLVM_AS_EXECUTABLE=C:\\taichi_llvm_15\\bin\\llvm-as.exe -DTI_WITH_VULKAN:BOOL=OFF"
-    $env:TAICHI_CMAKE_ARGS += " -DTI_LLVM_15:BOOL=ON"
-    $env:TAICHI_CMAKE_ARGS += " -DTI_WITH_DX12:BOOL=ON"
-}
-
-
-WriteInfo("Checking clang compiler")
-clang --version
-
-WriteInfo("Enter the repository")
-Set-Location .\taichi
-
-# Get sccache
-WriteInfo("Downloading sccache")
-$env:CCACHE_DIR="${pwd}/ccache_cache"
-$env:CCACHE_MAXSIZE="128M"
-$env:CCACHE_LOGFILE="${pwd}/ccache_error.log"
-WriteInfo("ccache dir: $Env:CCACHE_DIR")
-md "$Env:CCACHE_DIR" -ea 0
-if (-not (Test-Path "ccache-4.5.1-windows-64")) {
-    curl.exe --retry 10 --retry-delay 5 https://github.com/ccache/ccache/releases/download/v4.5.1/ccache-4.5.1-windows-64.zip -LO
-    if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE; }
-    7z x ccache-4.5.1-windows-64.zip
-    $env:PATH += ";${pwd}/ccache-4.5.1-windows-64"
-}
-ccache -v -s
-
-WriteInfo("Setting up Python environment")
-conda activate py37
-python -m pip install -r requirements_dev.txt
-python -m pip install -r requirements_test.txt
-
-# These have to be re-installed to avoid strange certificate issue
-# on CPU docker environment
-python -m pip install --upgrade --force-reinstall numpy
-python -m pip install --upgrade --force-reinstall cmake
-python -m pip install --upgrade --force-reinstall wheel
-if (-not $?) { exit 1 }
-
-WriteInfo("Building Taichi")
-python setup.py install
-if (-not $?) { exit 1 }
-WriteInfo("Build finished")
-ccache -s -v
-
-# We skip the test for the moment due to the long job execution time.
-#$env:TI_ENABLE_PADDLE = "0"
-#WriteInfo("Testing Taichi")
-#python tests/run_tests.py -vr2 -t4 -k "not torch and not paddle" -a cpu
-#WriteInfo("Test finished")
diff --git a/.github/workflows/scripts/win_test.ps1 b/.github/workflows/scripts/win_test.ps1
index 371ebed695e51..36659f62187e8 100644
--- a/.github/workflows/scripts/win_test.ps1
+++ b/.github/workflows/scripts/win_test.ps1
@@ -1,59 +1,73 @@
+param (
+    [string]$libsDir = "."
+)
+
 $ErrorActionPreference = "Stop"
 
+. $PSScriptRoot\common-utils.ps1
+
+Setup-VS
+
 $env:PYTHONUNBUFFERED = 1
 $env:TI_CI = 1
 $env:TI_OFFLINE_CACHE_FILE_PATH = Join-Path -Path $pwd -ChildPath ".cache\taichi"
 
-. venv\Scripts\activate.ps1
-python -c "import taichi"
-ti diagnose
-ti changelog
-echo wanted arch: $env:TI_WANTED_ARCHS
-pip install -r requirements_test.txt
-# TODO relax this when torch supports 3.10
-if ("$env:TI_WANTED_ARCHS".Contains("cuda")) {
-    pip install "torch==1.10.1+cu113; python_version < '3.10'" -f https://download.pytorch.org/whl/cu113/torch_stable.html
-} else {
-    pip install "torch; python_version < '3.10'"
-    pip install "paddlepaddle==2.3.0; python_version < '3.10'"
-}
+Setup-Python $libsDir $env:PY
 
+$os = Get-CimInstance -Class Win32_OperatingSystem
+Info "Total system memory: $($os.TotalVisibleMemorySize / 1024 / 1024) GB"
 
-if ("$env:TI_RUN_RELEASE_TESTS" -eq "1" -and -not "$env:TI_LITE_TEST") {
-    echo "Running release tests"
-    # release tests
-    pip install PyYAML
-    git clone https://github.com/taichi-dev/taichi-release-tests
-    mkdir -p repos/taichi/python/taichi
-    $EXAMPLES = & python -c 'import taichi.examples as e; print(e.__path__._path[0])' | Select-Object -Last 1
-    New-Item -Target $EXAMPLES -Path repos/taichi/python/taichi/examples -ItemType Junction
-    New-Item -Target taichi-release-tests/truths -Path truths -ItemType Junction
-    python taichi-release-tests/run.py --log=DEBUG --runners 1 taichi-release-tests/timelines
-    if (-not $?) { exit 1 }
-}
+$whl = & Get-ChildItem -Filter '*.whl' -Path dist | Select-Object -First 1
+echo $whl
+Invoke python -m pip install $whl.FullName
+Invoke python -c "import taichi"
+Invoke ti diagnose
+# Invoke ti changelog
+echo wanted arch: $env:TI_WANTED_ARCHS
+Invoke pip install -r requirements_test.txt
+Invoke pip install "paddlepaddle==2.3.0; python_version < '3.10'"
 
 # Run C++ tests
-python tests/run_tests.py --cpp
-if (-not $?) { exit 1 }
+Invoke python tests/run_tests.py --cpp
 
 # Fail fast, give priority to the error-prone tests
-python tests/run_tests.py -vr2 -t1 -k "paddle" -a cpu
-if (-not $?) { exit 1 }
+Invoke python tests/run_tests.py -vr2 -t1 -k "paddle" -a cpu
 
 # Disable paddle for the remaining test
 $env:TI_ENABLE_PADDLE = "0"
 
-if ("$env:TI_WANTED_ARCHS".Contains("cuda")) {
-  python tests/run_tests.py -vr2 -t4 -k "not torch and not paddle" -a cuda
-  if (-not $?) { exit 1 }
-}
 if ("$env:TI_WANTED_ARCHS".Contains("cpu")) {
-  python tests/run_tests.py -vr2 -t6 -k "not torch and not paddle" -a cpu
-  if (-not $?) { exit 1 }
+  # NOTE: Always test CPU with non-CUDA version of PyTorch,
+  #       since CUDA version of PyTorch will load a lot of CUDA libraries,
+  #       which inflates required commited memory usage (not physical memory, but still relevant)
+  #       to 5GiB per test process (compared to 1.4GiB for non-CUDA version).
+  #       This greatly improves test paralllism.
+  #       This is a non-issue on Linux, since Linux overcommits.
+  # TODO relax this when torch supports 3.10
+  Invoke pip install "torch==1.12.1; python_version < '3.10'"
+  Invoke python tests/run_tests.py -vr2 "-t$(EstimateNumProcs)" -k "not torch and not paddle" -a cpu
+}
+if ("$env:TI_WANTED_ARCHS".Contains("cuda")) {
+  # TODO relax this when torch supports 3.10
+  Invoke pip install "torch==1.10.1+cu113; python_version < '3.10'" -f https://download.pytorch.org/whl/cu113/torch_stable.html
+  Invoke python tests/run_tests.py -vr2 -t4 -k "not torch and not paddle" -a cuda
 }
 if ("$env:TI_WANTED_ARCHS".Contains("opengl")) {
-  python tests/run_tests.py -vr2 -t4 -k "not torch and not paddle" -a opengl
-  if (-not $?) { exit 1 }
+  Invoke python tests/run_tests.py -vr2 -t4 -k "not torch and not paddle" -a opengl
+}
+if ("$env:TI_WANTED_ARCHS".Contains("vulkan")) {
+  Invoke python tests/run_tests.py -vr2 -t4 -k "not torch and not paddle" -a vulkan
+}
+Invoke python tests/run_tests.py -vr2 -t1 -k "torch" -a "$env:TI_WANTED_ARCHS"
+
+if ("$env:TI_RUN_RELEASE_TESTS" -eq "1" -and -not "$env:TI_LITE_TEST") {
+    echo "Running release tests"
+    # release tests
+    Invoke pip install PyYAML
+    Invoke git clone https://github.com/taichi-dev/taichi-release-tests
+    mkdir -p repos/taichi/python/taichi
+    $EXAMPLES = & python -c 'import taichi.examples as e; print(e.__path__._path[0])' | Select-Object -Last 1
+    New-Item -Target $EXAMPLES -Path repos/taichi/python/taichi/examples -ItemType Junction
+    New-Item -Target taichi-release-tests/truths -Path truths -ItemType Junction
+    Invoke python taichi-release-tests/run.py --log=DEBUG --runners 1 taichi-release-tests/timelines
 }
-python tests/run_tests.py -vr2 -t1 -k "torch" -a "$env:TI_WANTED_ARCHS"
-if (-not $?) { exit 1 }
diff --git a/.github/workflows/testing.yml b/.github/workflows/testing.yml
index 6ac9cd1927c7a..555662ffa9cf8 100644
--- a/.github/workflows/testing.yml
+++ b/.github/workflows/testing.yml
@@ -40,9 +40,9 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout code
-        uses: actions/checkout@v2
+        uses: actions/checkout@v3
         with:
-          fetch-depth: 2
+          fetch-depth: '0'
 
       - name: check modified files
         id: check_files
@@ -83,9 +83,10 @@ jobs:
     runs-on: ubuntu-latest
     needs: check_files
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
         with:
           submodules: 'recursive'
+          fetch-depth: '0'
 
       - name: clang-tidy
         run: |
@@ -94,8 +95,8 @@ jobs:
           fi
           # https://docs.github.com/en/packages/managing-github-packages-using-github-actions-workflows/publishing-and-installing-a-package-with-github-actions#upgrading-a-workflow-that-accesses-ghcrio
           echo $CR_PAT | docker login ghcr.io -u ${{ github.actor }} --password-stdin
-          docker pull ghcr.io/taichi-dev/taichidev-cpu-ubuntu18.04:v0.3.2
-          docker run -id --user dev --name check_clang_tidy ghcr.io/taichi-dev/taichidev-cpu-ubuntu18.04:v0.3.2 /bin/bash
+          docker pull ghcr.io/taichi-dev/taichidev-cpu-ubuntu18.04:v0.3.3
+          docker run -id --user dev --name check_clang_tidy ghcr.io/taichi-dev/taichidev-cpu-ubuntu18.04:v0.3.3 /bin/bash
           tar -cf - ../${{ github.event.repository.name }} --mode u=+rwx,g=+rwx,o=+rwx --owner 1000 --group 1000 | docker cp - check_clang_tidy:/home/dev/
           docker exec --user root check_clang_tidy apt install -y clang-tidy-10
           docker exec --user dev check_clang_tidy /home/dev/taichi/.github/workflows/scripts/check_clang_tidy.sh "$CI_SETUP_CMAKE_ARGS"
@@ -106,7 +107,7 @@ jobs:
   build_and_test_cpu_linux:
     name: Build and Test linux (CPU)
     needs: check_files
-    timeout-minutes: ${{ github.event.schedule != '0 18 * * *' && 60 || 120 }}
+    timeout-minutes: ${{ github.event.schedule != '0 18 * * *' && 90 || 120 }}
     strategy:
       matrix:
         include:
@@ -123,89 +124,69 @@ jobs:
       packages: read
       contents: read
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
         with:
+          fetch-depth: '0'
           submodules: 'recursive'
 
-      - name: Get sccache cache
+      - name: Get Build Cache
         uses: actions/cache@v2
         with:
-          path: sccache_cache
-          key: sccache-linux-${{matrix.with_cc}}-${{ github.sha }}
+          path: /tmp/build-cache
+          key: build-cache-linux-${{ github.sha }}
           restore-keys: |
-            sccache-linux-${{matrix.with_cc}}-
+            build-cache-linux-
 
-      - name: Get docker images
+      - name: Prepare Environment
+        run: |
+          . .github/workflows/scripts/common-utils.sh
+          prepare-build-cache /tmp/build-cache
+          echo CI_DOCKER_RUN_EXTRA_ARGS="-v $(pwd):/home/dev/taichi" >> $GITHUB_ENV
+          sudo -n chown -R 1000:1000 . || true
+
+      - name: Get Docker Images
+        if: needs.check_files.outputs.run_job == 'true'
         run: |
-          if [[ ${{needs.check_files.outputs.run_job}} == false ]]; then
-            exit 0
-          fi
           # https://docs.github.com/en/packages/managing-github-packages-using-github-actions-workflows/publishing-and-installing-a-package-with-github-actions#upgrading-a-workflow-that-accesses-ghcrio
-          echo $CR_PAT | docker login ghcr.io -u ${{ github.actor }} --password-stdin
-          docker pull ghcr.io/taichi-dev/taichidev-cpu-ubuntu18.04:v0.3.2
-        env:
-          CR_PAT: ${{ secrets.GITHUB_TOKEN }}
+          docker login ghcr.io -u ${{ github.actor }} -p ${{ secrets.GITHUB_TOKEN }}
+          docker pull ghcr.io/taichi-dev/taichidev-cpu-ubuntu18.04:v0.3.3
 
       - name: Build
+        if: needs.check_files.outputs.run_job == 'true'
         run: |
-          if [[ ${{needs.check_files.outputs.run_job}} == false ]]; then
-            exit 0
-          fi
-          mkdir -m777 shared
-          docker create --user dev --name taichi_build \
-            -e TI_LITE_TEST \
-            -e TI_TEST_OFFLINE_CACHE \
-            -e PY -e PROJECT_NAME -e TAICHI_CMAKE_ARGS \
-            ghcr.io/taichi-dev/taichidev-cpu-ubuntu18.04:v0.3.2 \
-            /home/dev/taichi/.github/workflows/scripts/unix_build.sh
-          # A tarball is needed because sccache needs some permissions that only the file owner has.
-          # 1000 is the uid and gid of user "dev" in the container.
-          # If the uid or gid of the user inside the docker changes, please change the uid and gid in the following line.
-          tar -cf - ../${{ github.event.repository.name }} --mode u=+rwx,g=+rwx,o=+rwx --owner 1000 --group 1000 | docker cp - taichi_build:/home/dev/
-          docker start -a taichi_build
-          rm -rf sccache_cache
-          docker cp taichi_build:/home/dev/taichi/sccache_cache sccache_cache
-          docker cp taichi_build:/home/dev/taichi/dist shared/dist
-          docker cp taichi_build:/home/dev/taichi/build shared/build
+          . .github/workflows/scripts/common-utils.sh
+          ci-docker-run \
+            ghcr.io/taichi-dev/taichidev-cpu-ubuntu18.04:v0.3.3 \
+            /home/dev/taichi/.github/workflows/scripts/unix-build.sh
+
         env:
           PY: ${{ matrix.python }}
           PROJECT_NAME: taichi
-          TAICHI_CMAKE_ARGS: -DTI_WITH_OPENGL:BOOL=OFF -DTI_WITH_CC:BOOL=${{ matrix.with_cc }} -DTI_WITH_VULKAN:BOOL=OFF -DTI_BUILD_TESTS:BOOL=ON -DTI_WITH_C_API=ON -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CXX_COMPILER_LAUNCHER=sccache
+          TAICHI_CMAKE_ARGS: >-
+            -DTI_WITH_OPENGL:BOOL=OFF
+            -DTI_WITH_CC:BOOL=${{ matrix.with_cc }}
+            -DTI_WITH_VULKAN:BOOL=OFF
+            -DTI_BUILD_TESTS:BOOL=ON
+            -DTI_WITH_C_API=ON
 
       - name: Test
         id: test
+        if: needs.check_files.outputs.run_job == 'true'
         run: |
-          if [[ ${{needs.check_files.outputs.run_job}} == false ]]; then
-            exit 0
-          fi
-          docker create --user dev --name taichi_test \
-            -e PY -e TI_WANTED_ARCHS \
-            -e TI_LITE_TEST \
-            -e TI_TEST_OFFLINE_CACHE \
-            ghcr.io/taichi-dev/taichidev-cpu-ubuntu18.04:v0.3.2 /home/dev/unix_test.sh
-          docker cp .github/workflows/scripts/unix_test.sh taichi_test:/home/dev/unix_test.sh
-          docker cp .github/workflows/scripts/common-utils.sh taichi_test:/home/dev/common-utils.sh
-          docker cp shared/dist/ taichi_test:/home/dev/
-          docker cp shared/build/ taichi_test:/home/dev/
-          docker cp ./requirements_test.txt taichi_test:/home/dev/requirements_test.txt
-          docker cp tests/ taichi_test:/home/dev/
-          docker cp pyproject.toml taichi_test:/home/dev/
-          docker start -a taichi_test
+          . .github/workflows/scripts/common-utils.sh
+          ci-docker-run \
+            ghcr.io/taichi-dev/taichidev-cpu-ubuntu18.04:v0.3.3 \
+             /home/dev/taichi/.github/workflows/scripts/unix_test.sh
         env:
           PY: ${{ matrix.python }}
           TI_WANTED_ARCHS: ${{ matrix.wanted_archs }}
 
-      - name: clean docker container
-        if: always()
-        run: |
-          docker rm taichi_build taichi_test -f
-
       - name: Save wheel if test failed
         if: failure() && steps.test.conclusion == 'failure'
         uses: actions/upload-artifact@v3
         with:
           name: broken-wheel
-          path: shared/dist/*
+          path: dist/*
           retention-days: 7
 
   build_and_test_cpu_mac:
@@ -226,44 +207,32 @@ jobs:
     env:
       PY: ${{ matrix.python }}
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
         with:
+          fetch-depth: '0'
           submodules: 'recursive'
 
-      - name: Get sccache cache
-        uses: actions/cache@v2
-        with:
-          path: sccache_cache
-          key: sccache-mac-${{ github.sha }}
-          restore-keys: |
-            sccache-mac-
-
-      - name: Setup Python PATH && Download Pre-Built LLVM 10.0.0
+      - name: Prepare Environment
         run: |
-          export PATH=`pwd`/taichi-llvm/bin/:$PATH
-          # miniconda / miniforge
-          export PATH=$(ls -d ~/mini*/envs/$PY/bin):$PATH
-          if [[ "${{needs.check_files.outputs.run_job}}" == "false" ]]; then
-            exit 0
-          fi
+          . .github/workflows/scripts/common-utils.sh
+          prepare-build-cache
           python misc/ci_download.py
-          echo PATH=$PATH >> $GITHUB_ENV
-          #
         env:
           CI_PLATFORM: macos
 
       - name: Build & Install
+        if: needs.check_files.outputs.run_job == 'true'
         run: |
-          if [[ "${{needs.check_files.outputs.run_job}}" == "false" ]]; then
-            exit 0
-          fi
           brew install molten-vk
-          mkdir -p sccache_cache
-          .github/workflows/scripts/unix_build.sh
-          brew uninstall molten-vk
+          .github/workflows/scripts/unix-build.sh
         env:
-          TAICHI_CMAKE_ARGS: -DTI_WITH_OPENGL:BOOL=OFF -DTI_WITH_CC:BOOL=${{ matrix.with_cc }} -DTI_WITH_VULKAN:BOOL=ON -DTI_WITH_C_API=ON -DTI_BUILD_TESTS:BOOL=${{ matrix.with_cpp_tests }} -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CXX_COMPILER_LAUNCHER=sccache
           CXX: clang++
+          TAICHI_CMAKE_ARGS: >-
+            -DTI_WITH_OPENGL:BOOL=OFF
+            -DTI_WITH_CC:BOOL=${{ matrix.with_cc }}
+            -DTI_WITH_VULKAN:BOOL=ON
+            -DTI_WITH_C_API=ON
+            -DTI_BUILD_TESTS:BOOL=${{ matrix.with_cpp_tests }}
 
       # [DEBUG] Copy this step around to enable debugging inside Github Action instances.
       #- name: Setup tmate session
@@ -273,11 +242,8 @@ jobs:
 
       - name: Test
         id: test
-        run: |
-          if [[ "${{needs.check_files.outputs.run_job}}" == "false" ]]; then
-            exit 0
-          fi
-          .github/workflows/scripts/unix_test.sh
+        if: needs.check_files.outputs.run_job == 'true'
+        run: .github/workflows/scripts/unix_test.sh
         env:
           TI_WANTED_ARCHS: ${{ matrix.wanted_archs }}
           TI_SKIP_CPP_TESTS: Disabled because Vulkan is supported but not working on buildbot4
@@ -305,59 +271,68 @@ jobs:
     permissions:
       packages: read
       contents: read
+    env:
+      PY: "3.7"
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
         with:
+          fetch-depth: '0'
           submodules: 'recursive'
 
-      - name: Get sccache cache
-        uses: actions/cache@v2
+      - name: Get Build Cache
+        uses: actions/cache@v3
         if: needs.check_files.outputs.run_job == 'true'
         with:
-          path: ccache_cache
-          key: ccache-win64-cpu-${{ github.sha }}
+          path: ${{ env.LocalAppData }}/build-cache
+          key: build-cache-win64-cpu-${{ github.sha }}
           restore-keys: |
-            ccache-win64-cpu-
+            build-cache-win64-cpu-
 
-      - name: Get docker images
-        shell: bash
+      - name: Prepare Environment
+        shell: pwsh
         if: needs.check_files.outputs.run_job == 'true'
         run: |
-          echo $CR_PAT | docker login ghcr.io -u ${{ github.actor }} --password-stdin
-          docker pull ghcr.io/taichi-dev/taichidev-cpu-windows:v0.0.1
-        env:
-          CR_PAT: ${{ secrets.GITHUB_TOKEN }}
+          . .github/workflows/scripts/common-utils.ps1
+          Invoke docker login ghcr.io -u ${{ github.actor }} -p ${{ secrets.GITHUB_TOKEN }}
+          Invoke docker pull ghcr.io/taichi-dev/taichidev-cpu-windows:v0.0.2
+          PrepareBuildCache
+          echo CI_DOCKER_RUN_EXTRA_ARGS="--entrypoint pwsh.exe -v ${pwd}:D: -w D:/" >> $env:GITHUB_ENV
 
-      # TODO: split build and test
-      - name: Build and Test
-        shell: bash
+      - name: Build
+        shell: pwsh
+        if: needs.check_files.outputs.run_job == 'true'
         run: |
-          if [[ ${{needs.check_files.outputs.run_job}} == false ]]; then
-            exit 0
-          fi
-          docker create --name taichi_build_test \
-            -e TI_LITE_TEST \
-            -e TI_TEST_OFFLINE_CACHE \
-            ghcr.io/taichi-dev/taichidev-cpu-windows:v0.0.1 \
-            C:/taichi/.github/workflows/scripts/win_build_test_cpu.ps1 -llvmVer ${{ matrix.llvmVer }}
-          tar -cf - ../${{ github.event.repository.name }} --mode u=+rwx,g=+rwx,o=+rwx | docker cp - taichi_build_test:C:/
-          docker start -a taichi_build_test
-          rm -rf ccache_cache
-          docker cp taichi_build_test:C:/taichi/ccache_cache ccache_cache
-
-      - name: clean docker container
-        shell: bash
-        if: always()
+          . .github/workflows/scripts/common-utils.ps1
+          CIDockerRun `
+            ghcr.io/taichi-dev/taichidev-cpu-windows:v0.0.2 `
+            .github/workflows/scripts/win_build.ps1 -llvmVer ${{ matrix.llvmVer }} -libsDir X:/
+        env:
+          TAICHI_CMAKE_ARGS: >-
+            -DTI_WITH_OPENGL:BOOL=OFF
+            -DTI_WITH_CC:BOOL=OFF
+            -DTI_WITH_VULKAN:BOOL=OFF
+            -DTI_WITH_CUDA:BOOL=OFF
+            -DTI_BUILD_TESTS:BOOL=OFF
+            -DTI_WITH_DX12:BOOL=ON
+
+      - name: Test
+        shell: pwsh
+        if: needs.check_files.outputs.run_job == 'true'
         run: |
-          docker rm taichi_build_test -f
+          . .github/workflows/scripts/common-utils.ps1
+          CIDockerRun `
+            ghcr.io/taichi-dev/taichidev-cpu-windows:v0.0.2 `
+            .github/workflows/scripts/win_test.ps1 -libsDir X:/
+        env:
+          TI_WANTED_ARCHS: cpu
 
-      # - name: Save wheel if test failed
-      #   if: failure() && steps.build_and_test.conclusion == 'failure'
-      #   uses: actions/upload-artifact@v3
-      #   with:
-      #     name: broken-wheel
-      #     path: dist/*
-      #     retention-days: 7
+      - name: Save wheel if test failed
+        if: failure() && steps.build_and_test.conclusion == 'failure'
+        uses: actions/upload-artifact@v3
+        with:
+          name: broken-wheel
+          path: dist/*
+          retention-days: 7
 
   build_and_test_gpu_linux:
     name: Build and Test (GPU)
@@ -371,108 +346,57 @@ jobs:
 
     runs-on: ${{ matrix.tags }}
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
         with:
           submodules: 'recursive'
+          fetch-depth: '0'
 
-      - name: Get sccache cache
-        uses: actions/cache@v2
-        with:
-          path: sccache_cache
-          key: sccache-linux-gpu-${{ github.sha }}
-          restore-keys: |
-            sccache-linux-gpu-
+      - name: Prepare Environment
+        run: |
+          . .github/workflows/scripts/common-utils.sh
+          prepare-build-cache
+          echo CI_DOCKER_RUN_EXTRA_ARGS="-v $(pwd):/home/dev/taichi" >> $GITHUB_ENV
 
       - name: Build & Install
         run: |
-          if [[ ${{needs.check_files.outputs.run_job}} == false ]]; then
-            exit 0
-          fi
-          mkdir -m777 shared
-          for i in {0..9}; do
-            if xset -display ":$i" -q >/dev/null 2>&1; then
-              break
-            fi
-          done
-          if [ $? -ne 0 ]; then
-            echo "No display!"
-            exit 1
-          fi
-          export DISPLAY=:$i
-          docker create --user dev --name taichi_build \
-            -e TI_LITE_TEST \
-            -e TI_TEST_OFFLINE_CACHE \
-            --gpus 'all,"capabilities=graphics,utility,display,video,compute"' \
-            -v /tmp/.X11-unix:/tmp/.X11-unix \
-            -e PY -e GPU_BUILD -e PROJECT_NAME -e TAICHI_CMAKE_ARGS -e DISPLAY \
+          [[ ${{needs.check_files.outputs.run_job}} == false ]] && exit 0
+          . .github/workflows/scripts/common-utils.sh
+
+          ci-docker-run-gpu --name taichi-build \
             registry.taichigraphics.com/taichidev-ubuntu18.04:v0.3.4 \
-            /home/dev/taichi/.github/workflows/scripts/unix_build.sh
-          # A tarball is needed because sccache needs some permissions that only the file owner has.
-          # 1000 is the uid and gid of user "dev" in the container.
-          # If the uid or gid of the user inside the docker changes, please change the uid and gid in the following line.
-          tar -cf - ../${{ github.event.repository.name }} --mode u=+rwx,g=+rwx,o=+rwx --owner 1000 --group 1000 | docker cp - taichi_build:/home/dev/
-          docker start -a taichi_build
-          rm -rf sccache_cache
-          docker cp taichi_build:/home/dev/taichi/sccache_cache sccache_cache
-          docker cp taichi_build:/home/dev/taichi/dist shared/dist
-          docker cp taichi_build:/home/dev/taichi/build shared/build
+            /home/dev/taichi/.github/workflows/scripts/unix-build.sh
+
         env:
           PY: py38
-          GPU_BUILD: ON
           PROJECT_NAME: taichi
-          TAICHI_CMAKE_ARGS: -DTI_WITH_OPENGL:BOOL=ON -DTI_WITH_CC:BOOL=OFF -DTI_WITH_VULKAN:BOOL=ON -DTI_WITH_C_API=ON -DTI_BUILD_TESTS:BOOL=ON -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CXX_COMPILER_LAUNCHER=sccache
+          TAICHI_CMAKE_ARGS: >-
+            -DTI_WITH_OPENGL:BOOL=ON
+            -DTI_WITH_CC:BOOL=OFF
+            -DTI_WITH_VULKAN:BOOL=ON
+            -DTI_BUILD_TESTS:BOOL=ON
+            -DTI_WITH_C_API=ON
 
       - name: Test
         id: test
         run: |
-          if [[ ${{needs.check_files.outputs.run_job}} == false ]]; then
-            exit 0
-          fi
-          for i in {0..9}; do
-            if xset -display ":$i" -q >/dev/null 2>&1; then
-              break
-            fi
-          done
-          if [ $? -ne 0 ]; then
-            echo "No display!"
-            exit 1
-          fi
-          export DISPLAY=:$i
-          docker create --user dev --name taichi_test \
-            --gpus 'all,"capabilities=graphics,utility,display,video,compute"' \
-            -v /tmp/.X11-unix:/tmp/.X11-unix \
-            -e TI_LITE_TEST \
-            -e TI_TEST_OFFLINE_CACHE \
-            -e DISPLAY -e PY -e GPU_TEST -e TI_WANTED_ARCHS -e TI_RUN_RELEASE_TESTS \
+          [[ ${{needs.check_files.outputs.run_job}} == false ]] && exit 0
+          . .github/workflows/scripts/common-utils.sh
+
+          ci-docker-run-gpu --name taichi-test \
              registry.taichigraphics.com/taichidev-ubuntu18.04:v0.3.4 \
-             /home/dev/unix_test.sh
-          docker cp .github/workflows/scripts/unix_test.sh taichi_test:/home/dev/unix_test.sh
-          docker cp .github/workflows/scripts/common-utils.sh taichi_test:/home/dev/common-utils.sh
-          docker cp shared/dist/ taichi_test:/home/dev/
-          docker cp shared/build/ taichi_test:/home/dev/
-          docker cp pyproject.toml taichi_test:/home/dev/
-          docker cp tests/ taichi_test:/home/dev/
-          docker cp requirements_test.txt taichi_test:/home/dev/requirements_test.txt
-          docker start -a taichi_test
+             /home/dev/taichi/.github/workflows/scripts/unix_test.sh
         env:
           PY: py38
-          GPU_TEST: ON
           TI_WANTED_ARCHS: 'cuda,vulkan,opengl'
           TI_DEVICE_MEMORY_GB: '0.7'
           TI_RUN_RELEASE_TESTS: '1'
 
-
-      - name: clean docker container
-        if: always()
-        run: |
-          docker rm taichi_build taichi_test -f
-
       - name: Save wheel if test failed
         if: failure() && steps.test.conclusion == 'failure'
         uses: actions/upload-artifact@v3
         with:
           name: broken-wheel
-          path: shared/dist/*
+          path: dist/*
           retention-days: 7
 
   build_and_test_windows:
@@ -492,52 +416,39 @@ jobs:
     runs-on: ${{ matrix.runsOn }}
     timeout-minutes: ${{ github.event.schedule != '0 18 * * *' && 90 || 180 }}
     steps:
-      # See also https://github.com/taichi-dev/taichi/issues/4161
-      - name: Cleanup
-        shell: pwsh
-        run: |
-          remove-item '${{ github.workspace }}\*' -recurse -force
-
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
         with:
+          fetch-depth: '0'
           submodules: 'recursive'
 
       - uses: actions/setup-python@v2
         with:
           python-version: 3.7
 
-      - name: Add Visual Studio Shell to ENV
-        uses: egor-tensin/vs-shell@v2
-        with:
-          arch: x64
-
-      - name: Get sccache cache
-        uses: actions/cache@v2
-        with:
-          path: ccache_cache
-          key: ccache-win64-${{ github.sha }}
-          restore-keys: |
-            ccache-win64-
-
       - name: Build
         shell: pwsh
         if: ${{ needs.check_files.outputs.run_job != 'false' }}
         run: |
-          .\.github\workflows\scripts\win_build.ps1  -llvmVer ${{ matrix.llvmVer }} -installVulkan -install -libsDir C:\
+          .\.github\workflows\scripts\win_build.ps1 -llvmVer ${{ matrix.llvmVer }} -installVulkan -libsDir "$env:LocalAppData/buildbot"
         env:
-          TAICHI_CMAKE_ARGS: -DTI_WITH_OPENGL:BOOL=ON -DTI_WITH_DX11:BOOL=ON -DTI_WITH_CC:BOOL=OFF -DTI_WITH_C_API=ON -DTI_BUILD_TESTS:BOOL=ON
+          PY: "3.7"
+          TAICHI_CMAKE_ARGS: >-
+            -DTI_WITH_OPENGL:BOOL=ON
+            -DTI_WITH_DX11:BOOL=ON
+            -DTI_WITH_CC:BOOL=OFF
+            -DTI_BUILD_TESTS:BOOL=ON
+            -DTI_WITH_C_API=ON
 
       - name: Test
         id: test
         shell: pwsh
         if: ${{ needs.check_files.outputs.run_job != 'false' }}
         run: |
-          .\.github\workflows\scripts\win_test.ps1
+          .\.github\workflows\scripts\win_test.ps1 -libsDir "$env:LocalAppData/buildbot"
         env:
+          PY: "3.7"
           TI_WANTED_ARCHS: ${{ matrix.archs }}
           TI_SKIP_VERSION_CHECK: ON
-          TI_CI: 1
-          PYTHON: '3.7'
           TI_DEVICE_MEMORY_GB: '0.7'
           TI_RUN_RELEASE_TESTS: '1'
 
@@ -564,44 +475,38 @@ jobs:
         shell: '/usr/bin/arch -arch arm64e /bin/bash --noprofile --norc -eo pipefail {0}'
     runs-on: [self-hosted, m1]
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
         with:
+          fetch-depth: '0'
           submodules: 'recursive'
 
-      - name: Get sccache cache
-        uses: actions/cache@v2
-        with:
-          path: sccache_cache
-          key: sccache-m1-${{ github.sha }}
-          restore-keys: |
-            sccache-m1-
-
       - name: Build
         run: |
-          if [[ ${{needs.check_files.outputs.run_job}} == false ]]; then
-            exit 0
-          fi
-          export PATH=/Users/github/miniforge3/envs/$PY/bin:$PATH
+          [[ ${{needs.check_files.outputs.run_job}} == false ]] && exit 0
+          . .github/workflows/scripts/common-utils.sh
+          prepare-build-cache
           brew install molten-vk
-          .github/workflows/scripts/unix_build.sh
+          .github/workflows/scripts/unix-build.sh
         env:
-          TAICHI_CMAKE_ARGS: -DTI_WITH_OPENGL:BOOL=OFF -DTI_WITH_CUDA:BOOL=OFF -DTI_WITH_CC:BOOL=OFF -DTI_WITH_VULKAN:BOOL=ON -DTI_WITH_C_API=ON -DTI_BUILD_TESTS:BOOL=ON -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CXX_COMPILER_LAUNCHER=sccache
           PY: ${{ matrix.python }}
           CXX: clang++
+          TAICHI_CMAKE_ARGS: >-
+            -DTI_WITH_OPENGL:BOOL=OFF
+            -DTI_WITH_CUDA:BOOL=OFF
+            -DTI_WITH_CC:BOOL=OFF
+            -DTI_WITH_VULKAN:BOOL=ON
+            -DTI_BUILD_TESTS:BOOL=ON
+            -DTI_WITH_C_API=ON
 
       - name: Test
         id: test
         run: |
-          if [[ ${{needs.check_files.outputs.run_job}} == false ]]; then
-            exit 0
-          fi
-          export PATH=/Users/github/miniforge3/envs/$PY/bin:$PATH
+          [[ ${{needs.check_files.outputs.run_job}} == false ]] && exit 0
           .github/workflows/scripts/unix_test.sh
         env:
           TI_WANTED_ARCHS: 'cpu,metal,vulkan'
           PY: ${{ matrix.python }}
           PLATFORM: 'm1'
-          TI_CI: 1
           TI_RUN_RELEASE_TESTS: '1'
 
       - name: Save wheel if test failed
@@ -626,101 +531,84 @@ jobs:
       REDIS_HOST: 172.16.5.8
       PY: py39
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
         name: Checkout taichi
         with:
+          fetch-depth: '0'
           submodules: "recursive"
 
       - name: Prepare Environment
         run: >-
+          . .github/workflows/scripts/common-utils.sh;
+          prepare-build-cache;
+
           TAICHI_WHEEL_DIR=$(mktemp -d);
           echo TAICHI_WHEEL_DIR=$TAICHI_WHEEL_DIR >> $GITHUB_ENV;
           chmod 0777 $TAICHI_WHEEL_DIR;
 
-          echo DOCKER_RUN_ARGS="
-          -e PY
+          echo CI_DOCKER_RUN_EXTRA_ARGS="
           -e REDIS_HOST
-          -e TI_LITE_TEST
-          -e TAICHI_CMAKE_ARGS
-          -e SCCACHE_ROOT=/var/lib/sccache
           -v $(pwd):/home/dev/taichi
-          -v $(readlink -f ~/sccache):/var/lib/sccache
-          -v $(readlink -f ~/git-cache):/var/lib/git-cache
-          -v $(readlink -f ~/build-cache)/pip:/home/dev/.cache/pip
-          -v $(readlink -f ~/build-cache)/dot-gradle:/home/dev/.gradle
           -v $TAICHI_WHEEL_DIR:/taichi-wheel
           " >> $GITHUB_ENV;
 
-          for i in {0..9}; do if xset -display ":$i" -q >/dev/null 2>&1; then break; fi; done;
-          if [ $? -ne 0 ]; then echo "No display!"; exit 1; fi;
-
-          echo DOCKER_RUN_GPU_ARGS="
-          --runtime=nvidia
-          -e DISPLAY=:$i
-          -e NVIDIA_VISIBLE_DEVICES=all
-          -e NVIDIA_DRIVER_CAPABILITIES=all
-          " >> $GITHUB_ENV;
-
-          mkdir -p ~/build-cache/pip; chmod 0777 ~/build-cache/pip;
-          mkdir -p ~/build-cache/dot-gradle; chmod 0777 ~/build-cache/dot-gradle;
-          mkdir -p ~/build-cache/pip; chmod 0777 ~/build-cache/pip;
-
       - name: Build Host Taichi
         run: |
-          docker rm -f taichi_build_host
-          docker run --user dev --name taichi_build_host \
-            $DOCKER_RUN_ARGS \
+          . .github/workflows/scripts/common-utils.sh
+          ci-docker-run --name taichi-build-host \
             -v $TAICHI_WHEEL_DIR:/home/dev/taichi/dist \
             registry.taichigraphics.com/taichidev-ubuntu18.04:v0.3.4 \
-            /home/dev/taichi/.github/workflows/scripts/unix-build-v2.sh
+            /home/dev/taichi/.github/workflows/scripts/unix-build.sh
         env:
           TAICHI_CMAKE_ARGS: >-
             -DTI_WITH_OPENGL:BOOL=ON
             -DTI_WITH_CC:BOOL=OFF
             -DTI_WITH_VULKAN:BOOL=ON
-            -DTI_WITH_C_API=ON
-            -DCMAKE_C_COMPILER_LAUNCHER=sccache
-            -DCMAKE_CXX_COMPILER_LAUNCHER=sccache
+            -DTI_WITH_C_API=OFF
 
       - name: Build For Android
         run: |
+          . .github/workflows/scripts/common-utils.sh
           git clean -fxd
-          docker rm taichi_build_android -f
           chown -R 1000:1000 .
-          docker run -i --user dev --name taichi_build_android \
-            $DOCKER_RUN_ARGS \
+          ci-docker-run --name taichi-build-android \
             registry.taichigraphics.com/taichidev-androidsdk18.04:v0.0.6 \
-            /home/dev/taichi/.github/workflows/scripts/android-build.sh build-ext
+            /home/dev/taichi/.github/workflows/scripts/android-build.sh
+        env:
+          TAICHI_CMAKE_ARGS: >-
+            -DTI_WITH_OPENGL:BOOL=ON
+            -DTI_WITH_CC:BOOL=OFF
+            -DTI_WITH_VULKAN:BOOL=ON
+            -DTI_WITH_LLVM:BOOL=OFF
+            -DTI_WITH_C_API=ON
 
       - name: Test For Android AOT (export core)
         run: |
-          docker rm taichi_test_android -f
+          . .github/workflows/scripts/common-utils.sh
           chown -R 1000:1000 .
-          docker run -i --user dev --name taichi_test_android \
-            $DOCKER_RUN_ARGS \
-            $DOCKER_RUN_GPU_ARGS \
+          ci-docker-run-gpu --name taichi-test-android \
             registry.taichigraphics.com/taichidev-androidsdk18.04:v0.0.6 \
             /home/dev/taichi/.github/workflows/scripts/aot-demo.sh build-and-smoke-test-android-aot-demo
 
       - name: Prepare Unity Build Environment
         run: |
+          . .github/workflows/scripts/common-utils.sh
           chown -R 1000:1000 .
-          docker run -i --rm --user dev \
-            $DOCKER_RUN_ARGS \
+          ci-docker-run \
             registry.taichigraphics.com/taichidev-androidsdk18.04:v0.0.6 \
             /home/dev/taichi/.github/workflows/scripts/aot-demo.sh prepare-unity-build-env
 
       - name: Build Taichi-UnityExample
         run: |
-          docker run -i --rm --user dev \
-            $DOCKER_RUN_ARGS \
+          . .github/workflows/scripts/common-utils.sh
+          ci-docker-run \
             registry.taichigraphics.com/unityci-editor:ubuntu-2020.3.14f1-android-1-with-secret-sauce \
             /home/dev/taichi/.github/workflows/scripts/aot-demo.sh build-unity-demo
 
       - name: Run Taichi-UnityExample (C-API)
         run: |
+          . .github/workflows/scripts/common-utils.sh
           chown -R 1000:1000 .
-          docker run -i --rm --user dev \
-            $DOCKER_RUN_ARGS \
+          ci-docker-run \
             registry.taichigraphics.com/taichidev-androidsdk18.04:v0.0.6 \
             /home/dev/taichi/.github/workflows/scripts/aot-demo.sh smoke-test-unity-demo
diff --git a/ci/Dockerfile.ubuntu.18.04.cpu b/ci/Dockerfile.ubuntu.18.04.cpu
index 6c545669d22c7..31222f2c0f0a7 100644
--- a/ci/Dockerfile.ubuntu.18.04.cpu
+++ b/ci/Dockerfile.ubuntu.18.04.cpu
@@ -11,12 +11,14 @@ RUN apt-get update && \
     apt-get install -y python3-pip \
                        libtinfo-dev \
                        clang-10 \
+                       sudo \
                        wget \
                        git \
                        unzip \
                        libx11-xcb-dev \
                        zlib1g-dev \
                        && \
+    printf "root ALL=(ALL:ALL) NOPASSWD: ALL\ndev ALL=(ALL:ALL) NOPASSWD: ALL\n" > /etc/sudoers && \
     apt-get autoclean && \
     rm -rf /var/cache/apt/archives
 
diff --git a/ci/windows/Dockerfile_cpu b/ci/windows/Dockerfile_cpu
index 8a62212ce12d4..15f031ee106ed 100644
--- a/ci/windows/Dockerfile_cpu
+++ b/ci/windows/Dockerfile_cpu
@@ -31,5 +31,8 @@ RUN conda create -n py37 python=3.7
 # TODO: replace this when merge into ci
 COPY win_build_test.ps1 C:\win_build_test.ps1
 
+# Powershell 7
+RUN choco install -y pwsh
+
 # MUST use x64 as entry point
 ENTRYPOINT ["C:\\Program Files (x86)\\Microsoft Visual Studio\\2019\\BuildTools\\VC\\Auxiliary\\build\\vcvars64.bat", "&&", "powershell.exe", "-NoLogo", "-ExecutionPolicy", "Bypass"]