Merge pull request #31 from lipracer/dev

refine support any linux

.gitmodules

@@ -1,6 +1,3 @@
-[submodule "third_party/nanobind"]
-	path = third_party/nanobind
-	url = https://github.com/wjakob/nanobind
 [submodule "third_party/fmt"]
 	path = third_party/fmt
 	url = https://github.com/fmtlib/fmt.git

CMakeLists.txt

@@ -15,28 +15,6 @@ add_compile_options(-Wall -Wextra -Wno-unused-parameter
 -Werror=return-type -Werror)
 
 
-if (NOT SKBUILD)
-  message(WARNING "\
-  This CMake file is meant to be executed using 'scikit-build-core'.
-  Running it directly will almost certainly not produce the desired
-  result. If you are a user trying to install this package, use the
-  command below, which will install all necessary build dependencies,
-  compile the package in an isolated environment, and then install it.
-  =====================================================================
-   $ pip install .
-  =====================================================================
-  If you are a software developer, and this is your own package, then
-  it is usually much more efficient to install the build dependencies
-  in your environment once and use the following command that avoids
-  a costly creation of a new virtual environment at every compilation:
-  =====================================================================
-   $ pip install nanobind scikit-build-core[pyproject]
-   $ pip install --no-build-isolation -ve .
-  =====================================================================
-  You may optionally add -Ceditable.rebuild=true to auto-rebuild when
-  the package is imported. Otherwise, you need to rerun the above
-  after editing C++ files.")
-endif()
 
 find_package(Python 3.8 COMPONENTS Interpreter Development.Module REQUIRED)
 
@@ -51,14 +29,11 @@ if (NOT CMAKE_BUILD_TYPE AND NOT CMAKE_CONFIGURATION_TYPES)
 endif()
 
 # Detect the installed nanobind package and import it into CMake
-list(APPEND CMAKE_PREFIX_PATH "third_party/nanobind")
-find_package(nanobind CONFIG REQUIRED)
 
 # list(APPEND CMAKE_PREFIX_PATH "third_party/fmt")
 # find_package(fmt)
 add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/third_party/fmt)
 
-add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/third_party/nanobind)
 add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/src)
 add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/lib)
 

include/hook.h

@@ -99,14 +99,19 @@ class HookRuntimeContext {
     }
     map_type& map() { return map_; }
 
+    map_type::const_iterator& current_iter() {
+        thread_local static map_type::const_iterator iter;
+        return iter;
+    }
+
     map_type::const_iterator setCurrentState(size_t UniqueId) {
-        cur_iter_ = map_.begin();
-        std::advance(cur_iter_, UniqueId);
-        return cur_iter_;
+        current_iter() = map_.begin();
+        std::advance(current_iter(), UniqueId);
+        return current_iter();
     }
 
-    const std::string& curLibName() { return cur_iter_->first.lib_name; }
-    const std::string& curSymName() { return cur_iter_->first.sym_name; }
+    const std::string& curLibName() { return current_iter()->first.lib_name; }
+    const std::string& curSymName() { return current_iter()->first.sym_name; }
 
     void dump() {
         LOG(WARN) << "dum context map:";
@@ -119,7 +124,6 @@ class HookRuntimeContext {
 
    private:
     map_type map_;
-    map_type::const_iterator cur_iter_;
 };
 
 struct StringLiteral {

pyproject.toml

@@ -1,7 +1,3 @@
-[build-system]
-requires = ["scikit-build-core >=0.4.3", "nanobind >=1.3.2"]
-build-backend = "scikit_build_core.build"
-
 [project]
 name = "cuda-mock"
 version = "0.0.2"
@@ -39,3 +35,7 @@ test-requires = "pytest"
 # Needed for full C++17 support
 [tool.cibuildwheel.macos.environment]
 MACOSX_DEPLOYMENT_TARGET = "10.14"
+
+[build-system]
+requires = ["setuptools>=42", "wheel", "scikit-build", "cmake>=3.17", "ninja"]
+build-backend = "setuptools.build_meta"

setup.py

@@ -0,0 +1,6 @@
+from setuptools import setup
+import os
+
+os.system("cmake -S . -B build && cmake --build build")
+os.system("cd build/lib && ls | grep -v '\<cuda_mock\>' | xargs -I {} rm -rf {}")
+setup()

src/CMakeLists.txt

@@ -1,6 +1,11 @@
 
-nanobind_add_module(cuda_mock_impl STABLE_ABI cuda_mock_impl.cpp)
+add_library(cuda_mock_impl SHARED cuda_mock_impl.cpp)
 
 target_link_libraries(cuda_mock_impl PRIVATE cuda_mock)
 
-include_directories(PRIVATE ${CMAKE_SOURCE_DIR}/include)
+include_directories(PRIVATE ${CMAKE_SOURCE_DIR}/include)
+
+set_target_properties(cuda_mock_impl
+    PROPERTIES
+    LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/lib/cuda_mock
+)

src/cuda_mock/cuda_mock_impl.py

@@ -0,0 +1,19 @@
+import os
+import ctypes
+
+script_dir = os.path.dirname(os.path.abspath(__file__))
+cuda_mock_impl = ctypes.CDLL(f'{script_dir}/libcuda_mock_impl.so')
+
+def add(lhs, rhs):
+    return lhs + rhs
+
+def initialize():
+    return cuda_mock_impl.initialize()
+
+def internal_install_hook(*args):
+    new_args = [ctypes.c_char_p(arg.encode('utf-8')) for arg in args]
+    return cuda_mock_impl.internal_install_hook(*new_args)
+
+def xpu_initialize():
+    return cuda_mock_impl.xpu_initialize()
+

src/cuda_mock_impl.cpp

@@ -1,24 +1,43 @@
-#include <nanobind/nanobind.h>
 
 #include "cuda_mock.h"
 #include "xpu_mock.h"
 
-namespace nb = nanobind;
-using namespace nb::literals;
-
-NB_MODULE(cuda_mock_impl, m) {
-    m.def(
-        "add", [](int a, int b) { return a + b; }, "a"_a, "b"_a);
-    m.def("initialize", []() { dh_initialize(); });
-    m.def("internal_install_hook", [](const char* srcLib, const char* targetLib,
-                                      const char* symbolName) {
-        dh_internal_install_hook(srcLib, targetLib, symbolName);
-    });
-    m.def("internal_install_hook",
-          [](const char* srcLib, const char* targetLib, const char* symbolName,
-             const char* hookerLibPath, const char* hookerSymbolName) {
-              dh_internal_install_hook(srcLib, targetLib, symbolName,
-                                       hookerLibPath, hookerSymbolName);
-          });
-    m.def("xpu_initialize", []() { xpu_dh_initialize(); });
+// namespace nb = nanobind;
+// using namespace nb::literals;
+
+// NB_MODULE(cuda_mock_impl, m) {
+//     m.def(
+//         "add", [](int a, int b) { return a + b; }, "a"_a, "b"_a);
+//     m.def("initialize", []() { dh_initialize(); });
+//     m.def("internal_install_hook", [](const char* srcLib, const char*
+//     targetLib,
+//                                       const char* symbolName) {
+//         dh_internal_install_hook(srcLib, targetLib, symbolName);
+//     });
+//     m.def("internal_install_hook",
+//           [](const char* srcLib, const char* targetLib, const char*
+//           symbolName,
+//              const char* hookerLibPath, const char* hookerSymbolName) {
+//               dh_internal_install_hook(srcLib, targetLib, symbolName,
+//                                        hookerLibPath, hookerSymbolName);
+//           });
+//     m.def("xpu_initialize", []() { xpu_dh_initialize(); });
+// }
+
+typedef const char* PHStr;
+
+extern "C" {
+
+HOOK_API int add(int lhs, int rhs) { return lhs + rhs; }
+
+HOOK_API void initialize() { dh_initialize(); }
+
+HOOK_API void internal_install_hook(PHStr srcLib, PHStr targetLib,
+                                    PHStr symbolName, PHStr hookerLibPath,
+                                    PHStr hookerSymbolName) {
+    dh_internal_install_hook(srcLib, targetLib, symbolName, hookerLibPath,
+                             hookerSymbolName);
+}
+
+HOOK_API void xpu_initialize() { xpu_dh_initialize(); }
 }