[Enhancement] Improve CUDA path detection (#157)

* [Typo] Fix formatting in installation instructions in README.md

* [Enhancement] Improve CUDA path detection and update configuration handling

* fix typo

* remove IS_WINDOWS constant

* lint fix

* Improve error messages for CUDA detection failure

* lint fix

* lint fix

* Fix .gitignore to correctly include venv directory

install_cuda.sh

@@ -93,7 +93,11 @@ cd build
 
 
 echo "Configuring TVM build with LLVM and CUDA paths..."
-echo "set(USE_LLVM $LLVM_CONFIG_PATH)" >> config.cmake && echo "set(USE_CUDA /usr/local/cuda)" >> config.cmake
+echo "set(USE_LLVM \"$LLVM_CONFIG_PATH\")" >> config.cmake && \
+    CUDA_HOME=$(python -c "import sys; sys.path.append('../tilelang'); from env import CUDA_HOME; print(CUDA_HOME)") || \
+    { echo "ERROR: Failed to retrieve CUDA_HOME via Python script." >&2; exit 1; } && \
+    { [ -n "$CUDA_HOME" ] || { echo "ERROR: CUDA_HOME is empty, check CUDA installation or _find_cuda_home() in setup.py" >&2; exit 1; }; } && \
+    echo "set(USE_CUDA \"$CUDA_HOME\")" >> config.cmake
 
 echo "Running CMake for TileLang..."
 cmake ..

setup.py

@@ -20,12 +20,27 @@ from distutils.version import LooseVersion
 import platform
 import multiprocessing
 from setuptools.command.build_ext import build_ext
+import importlib
 
 # Environment variables False/True
 PYPI_BUILD = os.environ.get("PYPI_BUILD", "False").lower() == "true"
 PACKAGE_NAME = "tilelang"
 ROOT_DIR = os.path.dirname(__file__)
 
+
+def load_module_from_path(module_name, path):
+    spec = importlib.util.spec_from_file_location(module_name, path)
+    module = importlib.util.module_from_spec(spec)
+    sys.modules[module_name] = module
+    spec.loader.exec_module(module)
+    return module
+
+
+envs = load_module_from_path('env', os.path.join(ROOT_DIR, PACKAGE_NAME, 'env.py'))
+CUDA_HOME = envs.CUDA_HOME
+
+assert CUDA_HOME, "Failed to automatically detect CUDA installation. Please set the CUDA_HOME environment variable manually (e.g., export CUDA_HOME=/usr/local/cuda)."
+
 # TileLang only supports Linux platform
 assert sys.platform.startswith("linux"), "TileLang only supports Linux platform (including WSL)."
 
@@ -193,7 +208,7 @@ def build_csrc(llvm_config_path):
     # Set LLVM path and enable CUDA in config.cmake
     with open("config.cmake", "a") as config_file:
         config_file.write(f"set(USE_LLVM {llvm_config_path})\n")
-        config_file.write("set(USE_CUDA /usr/local/cuda)\n")
+        config_file.write(f"set(USE_CUDA {CUDA_HOME})\n")
     # Run CMake and make
     try:
         subprocess.check_call(["cmake", ".."])
@@ -519,7 +534,7 @@ class CMakeBuild(build_ext):
         # Here, we set USE_LLVM and USE_CUDA, for example.
         with open(dst_config_cmake, "a") as config_file:
             config_file.write(f"set(USE_LLVM {llvm_config_path})\n")
-            config_file.write("set(USE_CUDA /usr/local/cuda)\n")
+            config_file.write(f"set(USE_CUDA {CUDA_HOME})\n")
 
         # Run CMake to configure the project with the given arguments.
         subprocess.check_call(["cmake", ext.sourcedir] + cmake_args, cwd=build_temp)

tilelang/contrib/nvcc.py

@@ -8,6 +8,7 @@ from __future__ import absolute_import as _abs
 import os
 import subprocess
 import warnings
+from ..env import CUDA_HOME
 
 import tvm._ffi
 from tvm.target import Target
@@ -132,18 +133,11 @@ def find_cuda_path():
     path : str
         Path to cuda root.
     """
-    if "CUDA_PATH" in os.environ:
-        return os.environ["CUDA_PATH"]
-    cmd = ["which", "nvcc"]
-    proc = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
-    (out, _) = proc.communicate()
-    out = py_str(out)
-    if proc.returncode == 0:
-        return os.path.realpath(os.path.join(str(out).strip(), "../.."))
-    cuda_path = "/usr/local/cuda"
-    if os.path.exists(os.path.join(cuda_path, "bin/nvcc")):
-        return cuda_path
-    raise RuntimeError("Cannot find cuda path")
+    if CUDA_HOME:
+        return CUDA_HOME
+    raise RuntimeError(
+        "Failed to automatically detect CUDA installation. Please set the CUDA_HOME environment variable manually (e.g., export CUDA_HOME=/usr/local/cuda)."
+    )
 
 
 def get_cuda_version(cuda_path=None):

tilelang/env.py

@@ -2,9 +2,38 @@ import sys
 import os
 import pathlib
 import logging
+import shutil
+import glob
 
 logger = logging.getLogger(__name__)
 
+
+def _find_cuda_home() -> str:
+    """Find the CUDA install path.
+    
+    Adapted from https://github.com/pytorch/pytorch/blob/main/torch/utils/cpp_extension.py
+    """
+    # Guess #1
+    cuda_home = os.environ.get('CUDA_HOME') or os.environ.get('CUDA_PATH')
+    if cuda_home is None:
+        # Guess #2
+        nvcc_path = shutil.which("nvcc")
+        if nvcc_path is not None and "cuda" in nvcc_path.lower():
+            cuda_home = os.path.dirname(os.path.dirname(nvcc_path))
+        else:
+            # Guess #3
+            if sys.platform == 'win32':
+                cuda_homes = glob.glob('C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v*.*')
+                cuda_home = '' if len(cuda_homes) == 0 else cuda_homes[0]
+            else:
+                cuda_home = '/usr/local/cuda'
+            if not os.path.exists(cuda_home):
+                cuda_home = None
+    return cuda_home if cuda_home is not None else ""
+
+
+CUDA_HOME = _find_cuda_home()
+
 CUTLASS_INCLUDE_DIR: str = os.environ.get("TL_CUTLASS_PATH", None)
 TVM_PYTHON_PATH: str = os.environ.get("TVM_IMPORT_PYTHON_PATH", None)
 TVM_LIBRARY_PATH: str = os.environ.get("TVM_LIBRARY_PATH", None)
@@ -85,4 +114,5 @@ __all__ = [
     "TVM_PYTHON_PATH",
     "TVM_LIBRARY_PATH",
     "TILELANG_TEMPLATE_PATH",
+    "CUDA_HOME",
 ]