Merge branch 'main' into dcu

bbbf4207 · guchaoyang · GitHub · 8f4628e0 · 5eb30a4f · bbbf4207
Unverified Commit bbbf4207 authored Nov 14, 2025 by guchaoyang Committed by GitHub Nov 14, 2025
20 changed files
--- a/.clang-tidy
+++ b/.clang-tidy
 ---
 InheritParentConfig: true
-ExtraArgs: ['-v']
+ExtraArgs: []
 FormatStyle: file
 UseColor: true
 WarningsAsErrors: '*'

--- a/.github/ISSUE_TEMPLATE/release-plan.yml
+++ b/.github/ISSUE_TEMPLATE/release-plan.yml
+name: "Release Plan"
+description: "Plan the next release"
+title: "[Release Plan] vX.Y.Z"
+labels:
+  - release-plan
+  - tracking
+assignees: []
+body:
+  - type: input
+    id: version
+    attributes:
+      label: "Version"
+      placeholder: "v0.2.0"
+    validations:
+      required: true
+
+  - type: input
+    id: milestone
+    attributes:
+      label: "Milestone"
+      description: "Link or name of the milestone for this release"
+      placeholder: "https://github.com/tile-ai/tilelang/milestone/XX"
+
+  - type: textarea
+    id: scope
+    attributes:
+      label: "Scope"
+      description: "Goals and non-goals (brief)"
+      placeholder: |
+        - Goals: ...
+        - Non-goals: ...
+
+  - type: textarea
+    id: tasks
+    attributes:
+      label: "Tasks"
+      description: "Task list; link issues/PRs"
+      value: |
+        - [ ] Features
+        - [ ] Fixes
+        - [ ] Docs
+        - [ ] API/Breaking changes
+        - [ ] Benchmarks
+        - [ ] Release notes
+
+  - type: checkboxes
+    id: readiness
+    attributes:
+      label: "Readiness"
+      options:
+        - label: "All planned issues closed or deferred"
+        - label: "Docs updated"
+        - label: "CI green; artifacts verified"
+        - label: "Release notes drafted"
+
+  - type: textarea
+    id: notes
+    attributes:
+      label: "Notes"
+      description: "Risks or communications (optional)"
+      placeholder: |
+        - Risk: ...
+        - Communication: ...
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -22,10 +22,12 @@ env:
  PYTHONDEVMODE: "1"
  PYTHONUNBUFFERED: "1"
  PYTHONPATH: "" # explicit cleanup
+  PIP_USER: "" # explicit cleanup
  COLUMNS: "100"
  FORCE_COLOR: "1"
  CLICOLOR_FORCE: "1"
  UV_INDEX_STRATEGY: "unsafe-best-match"
+  UV_HTTP_TIMEOUT: "600"
  XDG_CACHE_HOME: "${{ github.workspace }}/.cache" # to be updated
  PIP_CACHE_DIR: "${{ github.workspace }}/.cache/pip" # to be updated
  UV_CACHE_DIR: "${{ github.workspace }}/.cache/uv" # to be updated
@@ -44,7 +46,7 @@ jobs:
          submodules: recursive

      - name: Setup Python 3.8
-        id: setup-py38
+        id: setup-pylowest
        uses: actions/setup-python@v6
        with:
          python-version: "3.8" # use lowest supported version for linting
@@ -52,12 +54,12 @@ jobs:

      - name: Check AST with Python 3.8
        run: |
-          "${{ steps.setup-py38.outputs.python-path }}" -m compileall -q -f tilelang
+          "${{ steps.setup-pylowest.outputs.python-path }}" -m compileall -q -f tilelang

-      - name: Setup Python 3.12
+      - name: Setup Python 3.9
        uses: actions/setup-python@v6
        with:
-          python-version: "3.12"
+          python-version: "3.9"
          update-environment: true
          cache: pip
          cache-dependency-path: |
@@ -127,13 +129,6 @@ jobs:
          echo "UV_CACHE_DIR=${XDG_CACHE_HOME}/uv" | tee -a "${GITHUB_ENV}"
          echo "PRE_COMMIT_HOME=${XDG_CACHE_HOME}/pip/.pre-commit" | tee -a "${GITHUB_ENV}"

-      - name: Set environment (GitHub-hosted runners)
-        if: ${{ !startsWith(matrix.runner.name, 'self-hosted') }}
-        run: |
-          # Enable ccache on GitHub-hosted runners to speed up builds
-          echo "CMAKE_C_COMPILER_LAUNCHER=ccache" | tee -a "${GITHUB_ENV}"
-          echo "CMAKE_CXX_COMPILER_LAUNCHER=ccache" | tee -a "${GITHUB_ENV}"
-
      # Do not use ccache on self-hosted runners, as it will download/upload caches which is slow.
      # Self-hosted runners usually have more CPU power to compile without ccache.
      - name: Setup ccache (GitHub-hosted runners)
@@ -142,8 +137,13 @@ jobs:
        uses: hendrikmuhs/ccache-action@v1
        with:
          create-symlink: true
-          key: ccache-${{ runner.os }}-${{ runner.arch }}-${{ matrix.python-version }}-${{ matrix.runner.name }}-${{ matrix.runner.toolkit }}
          evict-old-files: "7d"
+          append-timestamp: false
+          key: ${{ runner.os }}-${{ runner.arch }}-${{ matrix.runner.toolkit }}-${{ hashFiles('**/*.cc') }}
+          restore-keys: |
+            ${{ runner.os }}-${{ runner.arch }}-${{ matrix.runner.toolkit }}-${{ hashFiles('**/*.cc') }}
+            ${{ runner.os }}-${{ runner.arch }}-${{ matrix.runner.toolkit }}
+            ${{ runner.os }}-${{ runner.arch }}

      - name: Set environment (CUDA)
        if: contains(matrix.runner.toolkit, 'CUDA')
@@ -298,8 +298,9 @@ jobs:
            echo "Using run-clang-tidy from $(command -v run-clang-tidy)"
            CLANG_TIDY=(run-clang-tidy)
          else
-            echo "Downloading run-clang-tidy script"
-            wget -O run-clang-tidy.py https://raw.githubusercontent.com/llvm/llvm-project/refs/heads/release/21.x/clang-tools-extra/clang-tidy/tool/run-clang-tidy.py
+            RCT_URL=https://raw.githubusercontent.com/llvm/llvm-project/refs/heads/release/21.x/clang-tools-extra/clang-tidy/tool/run-clang-tidy.py
+            echo "Downloading run-clang-tidy script from ${RCT_URL}"
+            echo "import urllib.request; url = '${RCT_URL}'.rstrip('/'); urllib.request.urlretrieve(url, url.split('/')[-1])" | uv run --no-project --script -
            CLANG_TIDY=(uv run --no-project --script -- run-clang-tidy.py)
          fi
          if [[ -x "$(command -v clang-apply-replacements)" ]]; then
@@ -378,7 +379,7 @@ jobs:
            pytest --verbose --color=yes --durations=0 --showlocals --cache-clear
          )
          "${PYTEST[@]}" --maxfail=3 --numprocesses=4 \
-            ./python/amd/test_tilelang_test_amd.py
+            ./python/amd

      # Apple Metal tests
      - name: Run Metal tests with Python ${{ matrix.python-version }} (${{ matrix.runner.toolkit }})

--- a/.github/workflows/dist.yml
+++ b/.github/workflows/dist.yml
@@ -44,12 +44,11 @@ jobs:
    runs-on: macos-latest
    timeout-minutes: 30
    env:
-      NO_VERSION_LABEL: ${{ github.event_name == 'release' && 'OFF' || 'ON' }}
-      # NO_GIT_VERSION disables embedding the git commit hash in version metadata.
+      # `NO_VERSION_LABEL=ON` disables embedding the toolchain / git commit hash in version metadata.
      # Otherwise, the version of the SDist has a git hash suffix (e.g., 0.1.0+gitabcdef12),
      # but the package built from the SDist has no way to get the git hash (it is not a git repo),
      # leading to inconsistent versions between SDist and built packages (+gitabcdef12 vs. +gitunknown).
-      NO_GIT_VERSION: "ON"
+      NO_VERSION_LABEL: 'ON'

    steps:
      - name: Checkout repository
@@ -73,24 +72,26 @@ jobs:
        uses: hendrikmuhs/ccache-action@v1
        with:
          create-symlink: true
-          key: ccache-${{ runner.os }}-${{ runner.arch }}
          evict-old-files: "7d"
+          append-timestamp: false
+          key: sdist-${{ runner.os }}-${{ runner.arch }}-${{ hashFiles('**/*.cc') }}
+          restore-keys: |
+            sdist-${{ runner.os }}-${{ runner.arch }}-${{ hashFiles('**/*.cc') }}
+            sdist-${{ runner.os }}-${{ runner.arch }}
+            ${{ runner.os }}-${{ runner.arch }}

      - name: Test SDist buildable
        run: |
          TEMP_DIR="$(mktemp -d -t tilelang-sdist-test)"
          cp -r dist "${TEMP_DIR}/dist"
-          uv venv --seed "${TEMP_DIR}/venv"
-          source "${TEMP_DIR}/venv/bin/activate"
          cd "${TEMP_DIR}"
-          python3 -m pip install --upgrade pip setuptools wheel
-          python3 -m pip install -v dist/*.tar.gz
+          uv pip install -v dist/*.tar.gz
          python3 -c "import tilelang; print(tilelang.__version__)"

      - name: Upload SDist
        # Not PR to save artifact storage, as SDist is only needed for releases.
-        if: github.event_name != 'pull_request'
-        uses: actions/upload-artifact@v4
+        if: github.event_name != 'pull_request' || contains(github.event.pull_request.title, '[Release]')
+        uses: actions/upload-artifact@v5
        with:
          name: sdist
          path: dist/*.tar.gz
@@ -108,14 +109,11 @@ jobs:
          - { runner: ubuntu-24.04-arm, toolkit: "CUDA-12.8" }
          - { runner: macos-latest, toolkit: "Metal" }
        python-version:
-          - "3.8"
-          # TVM is built with Python 3.8 Limited API, it should work with all Python >= 3.8.
-          # - "3.9"
-          # - "3.10"
-          # - "3.11"
-          # - "3.12"
-          # - "3.13"
-          # - "3.14"
+          # Wheels are built with Python 3.8 Limited API, they should work with all Python >= 3.8.
+          # Only build wheels against Python 3.8 Limited API to save CI resources.
+          # FIXME: Here we use Python 3.9 because our dependency `apache-tvm-ffi` claims to support
+          #        Python 3.8 but it depends on a version of `ml-dtypes` that requires Python >= 3.9.
+          - "3.9"
      fail-fast: false
    timeout-minutes: 120
    runs-on: ${{ matrix.target.runner }}
@@ -129,14 +127,19 @@ jobs:
          fetch-depth: 1
          submodules: recursive

-      # NB: CIBW builds wheels in containers on Linux
-      - name: Setup ccache (macOS only)
-        if: runner.os == 'macOS'
+      - name: Setup ccache
        uses: hendrikmuhs/ccache-action@v1
        with:
          create-symlink: true
-          key: ccache-${{ runner.os }}-${{ runner.arch }}-${{ matrix.python-version }}-${{ matrix.target.toolkit }}
          evict-old-files: "7d"
+          append-timestamp: false
+          key: wheel-${{ runner.os }}-${{ runner.arch }}-${{ matrix.target.toolkit }}-${{ hashFiles('**/*.cc') }}
+          restore-keys: |
+            wheel-${{ runner.os }}-${{ runner.arch }}-${{ matrix.target.toolkit }}-${{ hashFiles('**/*.cc') }}
+            wheel-${{ runner.os }}-${{ runner.arch }}-${{ matrix.target.toolkit }}
+            wheel-${{ runner.os }}-${{ runner.arch }}
+            ${{ runner.os }}-${{ runner.arch }}-${{ matrix.target.toolkit }}
+            ${{ runner.os }}-${{ runner.arch }}

      - name: Set CIBW_BUILD
        run: |
@@ -151,6 +154,11 @@ jobs:
            echo "CUDA_VERSION=${CUDA_VERSION}" | tee -a "${GITHUB_ENV}"
          fi

+          if [[ "${{ runner.os }}" == "Linux" ]]; then
+            HOST_CCACHE_DIR="$(ccache --get-config cache_dir)"
+            echo "CIBW_BEFORE_BUILD_LINUX=yum install -y ccache && ccache -o cache_dir=/host${HOST_CCACHE_DIR}" | tee -a "${GITHUB_ENV}"
+          fi
+
      - name: Build wheels
        uses: pypa/cibuildwheel@v3.2
        with:
@@ -160,7 +168,7 @@ jobs:

      - name: Upload wheels
        # Not PR to save artifact storage, as wheels are only needed for releases.
-        if: github.event_name != 'pull_request'
+        if: github.event_name != 'pull_request' || contains(github.event.pull_request.title, '[Release]')
        uses: actions/upload-artifact@v5
        with:
          name: wheels-${{ matrix.python-version }}-${{ runner.os }}-${{ runner.arch }}-${{ matrix.target.toolkit }}
@@ -170,13 +178,13 @@ jobs:
  list-artifacts:
    name: List artifacts
    # Not PR to save artifact storage, as artifacts are only needed for releases.
-    if: github.event_name != 'pull_request'
+    if: github.event_name != 'pull_request' || contains(github.event.pull_request.title, '[Release]')
    runs-on: ubuntu-latest
    needs: [build-sdist, build-wheels]
    timeout-minutes: 15
    steps:
      - name: Download built SDist
-        uses: actions/download-artifact@v5
+        uses: actions/download-artifact@v6
        with:
          # unpacks default artifact into dist/
          # if `name: artifact` is omitted, the action will create extra parent dir

--- a/.github/workflows/pr-perfbench-bot.yml
+++ b/.github/workflows/pr-perfbench-bot.yml
@@ -12,6 +12,17 @@ concurrency:
  group: "${{ github.workflow }}-${{ github.ref }}"
  cancel-in-progress: true # always cancel in-progress

+env:
+  PYTHONDEVMODE: "1"
+  PYTHONUNBUFFERED: "1"
+  PYTHONPATH: "" # explicit cleanup
+  PIP_USER: "" # explicit cleanup
+  COLUMNS: "100"
+  FORCE_COLOR: "1"
+  CLICOLOR_FORCE: "1"
+  XDG_CACHE_HOME: "${{ github.workspace }}/.cache" # to be updated
+  PIP_CACHE_DIR: "${{ github.workspace }}/.cache/pip" # to be updated
+
 jobs:
  perfbench:
    name: Benchmark between PR and main
@@ -31,7 +42,12 @@ jobs:
      - name: Setup Python
        uses: actions/setup-python@v6
        with:
-          python-version: "3.9"
+          python-version: "3.12"
+          update-environment: true
+          cache: pip
+          cache-dependency-path: |
+            pyproject.toml
+            requirements*.txt

      - name: Install merged version
        run: |

--- a/.gitignore
+++ b/.gitignore
@@ -20,6 +20,7 @@
 debug/
 build/
 *dist/
+dist*/
 wheelhouse/
 __pycache__
 nnfusion.tar.gz
@@ -101,3 +102,9 @@ tilelang/jit/adapter/cython/.cycache
 # CMake
 cmake-build/
 cmake-build-*/
+
+# Git version for sdist
+.git_commit.txt
+
+# pre-commit cache
+.pre-commit-cache/*
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -41,7 +41,7 @@ repos:
            ^.+\.json$
          )
  - repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.14.1  # sync with requirements-lint.txt
+    rev: v0.14.3  # sync with requirements-lint.txt
    hooks:
      - id: ruff-check
        args: [--fix, --exit-non-zero-on-fix]

--- a/tvm @ 093b2cdb
+++ b/tvm @ 093b2cdb
-Subproject commit 5bf17a34602931e7d7e01cbccf358a21fe972779
+Subproject commit 093b2cdb2187140b197336496d65d61ace89e8ff
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -8,6 +8,11 @@ set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)

+if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU" AND "$ENV{CIBUILDWHEEL}")
+  # Warning came from tvm submodule
+  string(APPEND CMAKE_CXX_FLAGS " -Wno-dangling-reference")
+endif()
+
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${CMAKE_CURRENT_SOURCE_DIR}/cmake)

 if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/.gitmodules" AND EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/.git")
@@ -36,15 +41,74 @@ endif()

 find_program(CCACHE_PROGRAM ccache)
 if(CCACHE_PROGRAM)
+  message(STATUS "Using ccache: ${CCACHE_PROGRAM} with base_dir=${CMAKE_SOURCE_DIR}")
+  if(APPLE)
+    # Passing configs like `ccache base_dir=/xxx cc ...` is supported
+    # (likely) since ccache 4.x, which has been provided by homebrew.
+    # Our Linux builder image (manylinux2014 & manylinux_2_28) still
+    # provides ccache 3.x and do not support this form.
+    # `cibuildwheel` uses fixed folder on Linux (`/project`) as working directory,
+    # so cache would work without setting `base_dir`.
+    set(CCACHE_PROGRAM "${CCACHE_PROGRAM};base_dir=${CMAKE_SOURCE_DIR}")
+  endif()
  set(CMAKE_C_COMPILER_LAUNCHER "${CCACHE_PROGRAM}" CACHE STRING "C compiler launcher")
  set(CMAKE_CXX_COMPILER_LAUNCHER "${CCACHE_PROGRAM}" CACHE STRING "CXX compiler launcher")
  set(CMAKE_CUDA_COMPILER_LAUNCHER "${CCACHE_PROGRAM}" CACHE STRING "CUDA compiler launcher")
+else()
+  find_program(SCCACHE_PROGRAM sccache)
+  if(SCCACHE_PROGRAM)
+    message(STATUS "Using sccache: ${SCCACHE_PROGRAM}")
+    set(CMAKE_C_COMPILER_LAUNCHER "${SCCACHE_PROGRAM}" CACHE STRING "C compiler launcher")
+    set(CMAKE_CXX_COMPILER_LAUNCHER "${SCCACHE_PROGRAM}" CACHE STRING "CXX compiler launcher")
+    set(CMAKE_CUDA_COMPILER_LAUNCHER "${SCCACHE_PROGRAM}" CACHE STRING "CUDA compiler launcher")
+  endif()
 endif()

 # Configs
-set(USE_CUDA OFF)
-set(USE_ROCM OFF)
-set(USE_METAL OFF)
+set(TILELANG_BACKENDS CUDA ROCM METAL)
+
+set(TILELANG_BACKEND_DOC_CUDA "Enable CUDA backend (ON/OFF/or CUDA SDK path)")
+set(TILELANG_BACKEND_DOC_ROCM "Enable ROCm backend (ON/OFF/or ROCm SDK path)")
+set(TILELANG_BACKEND_DOC_METAL "Enable Metal backend")
+
+# TVM's config.cmake redefines USE_* options later, so we cache the user's choice
+# (including explicit -DUSE_XXX arguments) before we include TVM and restore it
+# afterwards.
+
+macro(tilelang_define_backend_option BACKEND)
+  set(_backend_var "USE_${BACKEND}")
+  set(_doc "${TILELANG_BACKEND_DOC_${BACKEND}}")
+  set(_user_override_var "TILELANG_USER_OVERRIDE_${_backend_var}")
+
+  set(_user_override OFF)
+  if(DEFINED ${_user_override_var})
+    set(_user_override "${${_user_override_var}}")
+  endif()
+
+  if(DEFINED CACHE{${_backend_var}})
+    get_property(_cache_type CACHE ${_backend_var} PROPERTY TYPE)
+    if(_cache_type STREQUAL "UNINITIALIZED")
+      set(_user_override ON)
+    endif()
+  endif()
+
+  set(_default OFF)
+  if(DEFINED ${_backend_var})
+    set(_default "${${_backend_var}}")
+  endif()
+
+  option(${_backend_var} "${_doc}" "${_default}")
+  # Remember if the user explicitly set this option so that later logic
+  # won't auto-toggle backends they configured on the command line.
+  set(${_user_override_var} ${_user_override} CACHE INTERNAL
+    "User explicitly set ${_backend_var} during configuration" FORCE)
+  set(TILELANG_OPTION_${_backend_var} "${${_backend_var}}")
+endmacro()
+
+foreach(BACKEND IN LISTS TILELANG_BACKENDS)
+  tilelang_define_backend_option(${BACKEND})
+endforeach()
+
 set(PREBUILD_CYTHON ON)
 # Configs end

@@ -55,6 +119,14 @@ if(EXISTS ${TVM_SOURCE}/cmake/config.cmake)
 else()
  message(FATAL_ERROR "Nor tvm provided or submodule checkout-ed.")
 endif()
+# Re-apply TileLang's preferred backend settings after TVM's config may have
+# overridden the USE_* cache entries.
+foreach(BACKEND IN LISTS TILELANG_BACKENDS)
+  set(_backend_var "USE_${BACKEND}")
+  set(_doc "${TILELANG_BACKEND_DOC_${BACKEND}}")
+  set(${_backend_var} ${TILELANG_OPTION_${_backend_var}} CACHE STRING "${_doc}" FORCE)
+  set(${_backend_var} ${TILELANG_OPTION_${_backend_var}})
+endforeach()

 # Include directories for TileLang
 set(TILE_LANG_INCLUDES ${TVM_INCLUDES})
@@ -68,29 +140,39 @@ file(GLOB TILE_LANG_SRCS
  src/target/utils.cc
  src/target/codegen_cpp.cc
  src/target/rt_mod_cpp.cc
-  # webgpu doesn't have system dependency
-  src/target/codegen_webgpu.cc
  # intrin_rule doesn't have system dependency
  src/target/intrin_rule*.cc
 )

-# Backend-specific checks and configs
-if($ENV{USE_METAL})
-  set(USE_METAL ON)
-elseif(APPLE)
-  message(STATUS "Enable Metal support by default.")
-  set(USE_METAL ON)
-elseif($ENV{USE_ROCM})
-  set(USE_ROCM ON)
-else()
-  if($ENV{USE_CUDA})
-    set(USE_CUDA ON)
-  elseif(DEFINED ENV{USE_CUDA} AND NOT $ENV{USE_CUDA})
-    # Build CPU-only when we explicitly disable CUDA
-    set(USE_CUDA OFF)
+# Track if the user explicitly selected a backend via cache options.
+set(TILELANG_BACKEND_USER_SELECTED OFF)
+foreach(BACKEND IN LISTS TILELANG_BACKENDS)
+  set(_backend_var "USE_${BACKEND}")
+  set(_override_var "TILELANG_USER_OVERRIDE_${_backend_var}")
+  if(${_backend_var} OR ${_override_var})
+    set(TILELANG_BACKEND_USER_SELECTED ON)
+  endif()
+endforeach()
+
+# Only auto-select a backend when the user didn't specify one explicitly.
+if(NOT TILELANG_BACKEND_USER_SELECTED)
+  if($ENV{USE_METAL})
+    set(USE_METAL ON)
+  elseif(APPLE)
+    message(STATUS "Enable Metal support by default.")
+    set(USE_METAL ON)
+  elseif($ENV{USE_ROCM})
+    set(USE_ROCM ON)
  else()
-    message(STATUS "Enable CUDA support by default.")
-    set(USE_CUDA ON)
+    if($ENV{USE_CUDA})
+      set(USE_CUDA ON)
+    elseif(DEFINED ENV{USE_CUDA} AND NOT $ENV{USE_CUDA})
+      # Build CPU-only when we explicitly disable CUDA
+      set(USE_CUDA OFF)
+    else()
+      message(STATUS "Enable CUDA support by default.")
+      set(USE_CUDA ON)
+    endif()
  endif()
 endif()

@@ -104,7 +186,7 @@ if(USE_METAL)
 elseif(USE_ROCM)
  set(CMAKE_HIP_STANDARD 17)
  include(${TVM_SOURCE}/cmake/utils/FindROCM.cmake)
-  find_rocm($ENV{USE_ROCM})
+  find_rocm(${USE_ROCM})
  add_compile_definitions(__HIP_PLATFORM_AMD__ __HIP_PLATFORM_HCC__=1)

  file(GLOB TILE_LANG_HIP_SRCS
@@ -148,12 +230,8 @@ target_include_directories(tilelang_objs PRIVATE ${TILE_LANG_INCLUDES})

 add_library(tilelang SHARED $<TARGET_OBJECTS:tilelang_objs>)
 add_library(tilelang_module SHARED $<TARGET_OBJECTS:tilelang_objs>)
-target_link_libraries(tilelang PUBLIC tvm_runtime)
+target_link_libraries(tilelang PUBLIC tvm_runtime tvm)
 target_link_libraries(tilelang_module PUBLIC tvm)
-if(APPLE)
-  # FIXME: libtilelang should only link against tvm runtime
-  target_link_libraries(tilelang PUBLIC tvm)
-endif()
 # Build cython extension
 find_package(Python REQUIRED COMPONENTS Interpreter Development.Module ${SKBUILD_SABI_COMPONENT})

@@ -181,18 +259,18 @@ install(TARGETS tilelang_cython_wrapper

 # let libtilelang to search tvm/tvm_runtime in same dir
 if(APPLE)
-  set_target_properties(tilelang PROPERTIES INSTALL_RPATH "@loader_path")
-  set_target_properties(tilelang_module PROPERTIES INSTALL_RPATH "@loader_path")
-else()
-  set_target_properties(tilelang PROPERTIES INSTALL_RPATH "\$ORIGIN")
-  set_target_properties(tilelang_module PROPERTIES INSTALL_RPATH "\$ORIGIN")
+  set_target_properties(tilelang PROPERTIES INSTALL_RPATH "@loader_path;@loader_path/../../tvm_ffi/lib")
+  set_target_properties(tilelang_module PROPERTIES INSTALL_RPATH "@loader_path;@loader_path/../../tvm_ffi/lib")
+  set_target_properties(tvm PROPERTIES INSTALL_RPATH "@loader_path;@loader_path/../../tvm_ffi/lib")
+  set_target_properties(tvm_runtime PROPERTIES INSTALL_RPATH "@loader_path;@loader_path/../../tvm_ffi/lib")
+elseif(UNIX)
+  set_target_properties(tilelang PROPERTIES INSTALL_RPATH "\$ORIGIN:\$ORIGIN/../../tvm_ffi/lib")
+  set_target_properties(tilelang_module PROPERTIES INSTALL_RPATH "\$ORIGIN:\$ORIGIN/../../tvm_ffi/lib")
+  set_target_properties(tvm PROPERTIES INSTALL_RPATH "\$ORIGIN:\$ORIGIN/../../tvm_ffi/lib")
+  set_target_properties(tvm_runtime PROPERTIES INSTALL_RPATH "\$ORIGIN:\$ORIGIN/../../tvm_ffi/lib")
 endif()

-install(TARGETS tvm tvm_runtime tilelang_module tilelang LIBRARY DESTINATION tilelang/lib)
-
-# Copy tvm cython ext for wheels
-# TODO: not necessary for editable builds
-if(TVM_BUILD_FROM_SOURCE)
-  add_dependencies(tilelang tvm_cython)
-  install(FILES "${CMAKE_CURRENT_SOURCE_DIR}/3rdparty/tvm/python/tvm/ffi/core.abi3.so" DESTINATION tilelang/3rdparty/tvm/python/tvm/ffi/)
-endif()
+install(
+  TARGETS tvm tvm_runtime tilelang_module tilelang
+  LIBRARY DESTINATION tilelang/lib
+)
--- a/MANIFEST.in
+++ b/MANIFEST.in
-# Reference: https://setuptools.pypa.io/en/latest/userguide/miscellaneous.html
-
-# Include licenses
-include VERSION
-include LICENSE
-include THIRDPARTYNOTICES.txt
-
-# Version and dependency files
-include version_provider.py
-include requirements*.txt
-include tilelang/jit/adapter/cython/cython_wrapper.pyx
-
-# Include source files in SDist
-include CMakeLists.txt
-graft src
-graft cmake
-graft 3rdparty
-
-# Include test suites in SDist
-graft testing
-graft examples
-global-exclude .coverage .coverage.* coverage.xml coverage-*.xml coverage.*.xml
-global-exclude .junit .junit.* junit.xml junit-*.xml junit.*.xml
-
-# Exclude unneeded files and directories
-prune .git
-prune .github
-prune */.git
-prune */.github
-prune 3rdparty/clang*
-prune 3rdparty/llvm*
-
-# Prune compiled files
-prune */__pycache__
-global-exclude *~ *.py[cod] *.so *.a *.dylib *.pxd *.dll *.lib *.o *.obj
--- a/README.md
+++ b/README.md
@@ -13,6 +13,7 @@ Tile Language (**tile-lang**) is a concise domain-specific language designed to
 <img src=./images/MatmulExample.png />

 ## Latest News
+- 10/30/2025 📦: We have released v0.1.6.post2, which is the last version compatible with Python 3.8.
 - 10/07/2025 🍎: Added Apple Metal Device support, check out [Pull Request #799](https://github.com/tile-ai/tilelang/pull/799) for details.
 - 09/29/2025  🎉: Thrilled to announce that AscendC and AscendNPU IR backends targeting Huawei Ascend chips are now supported!
 Check out the preview here:

--- a/VERSION
+++ b/VERSION
-0.1.6.post1
+0.1.6.post2
--- a/cmake/load_tvm.cmake
+++ b/cmake/load_tvm.cmake
@@ -11,8 +11,17 @@ endif()

 set(TVM_INCLUDES
  ${TVM_SOURCE}/include
-  ${TVM_SOURCE}/ffi/include
  ${TVM_SOURCE}/src
  ${TVM_SOURCE}/3rdparty/dlpack/include
  ${TVM_SOURCE}/3rdparty/dmlc-core/include
 )
+
+if(EXISTS ${TVM_SOURCE}/ffi/include)
+  list(APPEND TVM_INCLUDES ${TVM_SOURCE}/ffi/include)
+elseif(EXISTS ${TVM_SOURCE}/3rdparty/tvm-ffi/include)
+  list(APPEND TVM_INCLUDES ${TVM_SOURCE}/3rdparty/tvm-ffi/include)
+endif()
+
+if(EXISTS ${TVM_SOURCE}/3rdparty/tvm-ffi/3rdparty/dlpack/include)
+  list(APPEND TVM_INCLUDES ${TVM_SOURCE}/3rdparty/tvm-ffi/3rdparty/dlpack/include)
+endif()
--- a/docker/Dockerfile.cu118
+++ b/docker/Dockerfile.cu118
@@ -23,6 +23,6 @@ RUN conda install pip cmake && conda install -c conda-forge libstdcxx-ng=12 && c
 RUN apt-get install -y python3 python3-dev python3-setuptools gcc libtinfo-dev zlib1g-dev build-essential cmake libedit-dev libxml2-dev

 RUN git clone https://github.com/tile-ai/tilelang.git --recursive -b main TileLang \
-  && cd TileLang && ./install_cuda.sh
+  && cd TileLang && USE_CUDA=1 pip install -e . -v

 CMD bash
--- a/docker/Dockerfile.cu120
+++ b/docker/Dockerfile.cu120
@@ -23,6 +23,6 @@ RUN conda install pip cmake && conda install -c conda-forge libstdcxx-ng=12 && c
 RUN apt-get install -y python3 python3-dev python3-setuptools gcc libtinfo-dev zlib1g-dev build-essential cmake libedit-dev libxml2-dev

 RUN git clone https://github.com/tile-ai/tilelang.git --recursive -b main TileLang \
-  && cd TileLang && ./install_cuda.sh
+  && cd TileLang && USE_CUDA=1 pip install -e . -v

 CMD bash
--- a/docker/Dockerfile.cu121
+++ b/docker/Dockerfile.cu121
@@ -23,6 +23,6 @@ RUN conda install pip cmake && conda install -c conda-forge libstdcxx-ng=12 && c
 RUN apt-get install -y python3 python3-dev python3-setuptools gcc libtinfo-dev zlib1g-dev build-essential cmake libedit-dev libxml2-dev

 RUN git clone https://github.com/tile-ai/tilelang.git --recursive -b main TileLang \
-  && cd TileLang && ./install_cuda.sh
+  && cd TileLang && USE_CUDA=1 pip install -e . -v

 CMD bash
--- a/docker/Dockerfile.cu123
+++ b/docker/Dockerfile.cu123
@@ -23,6 +23,6 @@ RUN conda install pip cmake && conda install -c conda-forge libstdcxx-ng=12 && c
 RUN apt-get install -y python3 python3-dev python3-setuptools gcc libtinfo-dev zlib1g-dev build-essential cmake libedit-dev libxml2-dev

 RUN git clone https://github.com/tile-ai/tilelang.git --recursive -b main TileLang \
-  && cd TileLang && ./install_cuda.sh
+  && cd TileLang && USE_CUDA=1 pip install -e . -v

 CMD bash
--- a/docker/Dockerfile.cu124
+++ b/docker/Dockerfile.cu124
@@ -23,6 +23,6 @@ RUN conda install pip cmake && conda install -c conda-forge libstdcxx-ng=12 && c
 RUN apt-get install -y python3 python3-dev python3-setuptools gcc libtinfo-dev zlib1g-dev build-essential cmake libedit-dev libxml2-dev

 RUN git clone https://github.com/tile-ai/tilelang.git --recursive -b main TileLang \
-  && cd TileLang && ./install_cuda.sh
+  && cd TileLang && USE_CUDA=1 pip install -e . -v

 CMD bash
--- a/docker/Dockerfile.cu125
+++ b/docker/Dockerfile.cu125
@@ -23,6 +23,6 @@ RUN conda install pip cmake && conda install -c conda-forge libstdcxx-ng=12 && c
 RUN apt-get install -y python3 python3-dev python3-setuptools gcc libtinfo-dev zlib1g-dev build-essential cmake libedit-dev libxml2-dev

 RUN git clone https://github.com/tile-ai/tilelang.git --recursive -b main TileLang \
-  && cd TileLang && ./install_cuda.sh
+  && cd TileLang && USE_CUDA=1 pip install -e . -v

 CMD bash
--- a/docker/Dockerfile.cu126
+++ b/docker/Dockerfile.cu126
@@ -23,6 +23,6 @@ RUN conda install pip cmake && conda install -c conda-forge libstdcxx-ng=12 && c
 RUN apt-get install -y python3 python3-dev python3-setuptools gcc libtinfo-dev zlib1g-dev build-essential cmake libedit-dev libxml2-dev

 RUN git clone https://github.com/tile-ai/tilelang.git --recursive -b main TileLang \
-  && cd TileLang && ./install_cuda.sh
+  && cd TileLang && USE_CUDA=1 pip install -e . -v

 CMD bash