fix(lite): fix lite header

GitOrigin-RevId: e255597121f1fba1920b6909596ba56c7e40f888

lite/CMakeLists.txt

@@ -28,11 +28,13 @@ if(MGE_WITH_MINIMUM_SIZE)
   set(LITE_ENABLE_EXCEPTION OFF)
 endif()
 
+if(LITE_BUILD_WITH_MGE)
+  set(LITE_BUILD_WITH_MGE ON)
+endif()
+
 # Write out lite_build_config.h It defines macros needed by lite
-configure_file(src/lite_build_config.h.in
+configure_file(build_config/lite_build_config.h.in
                ${CMAKE_CURRENT_BINARY_DIR}/genfiles/lite_build_config.h)
-install(FILES ${CMAKE_CURRENT_BINARY_DIR}/genfiles/lite_build_config.h
-        DESTINATION ${CMAKE_INSTALL_PREFIX}/lite/include)
 
 # begin config lite
 if(LITE_BUILD_WITH_MGE
@@ -47,7 +49,6 @@ include_directories($<BUILD_INTERFACE:${CMAKE_CURRENT_BINARY_DIR}/genfiles>)
 
 if(LITE_BUILD_WITH_MGE)
   target_link_libraries(lite_static PRIVATE megbrain megdnn ${MGE_CUDA_LIBS})
-  add_compile_definitions(LITE_BUILD_WITH_MGE=1)
   message(STATUS "build lite with MegEngine.")
 else()
   target_link_libraries(lite_static PUBLIC flatbuffers)
@@ -154,7 +155,10 @@ if(UNIX AND NOT APPLE)
   set_target_properties(lite_shared_whl PROPERTIES LINK_DEPENDS ${LITE_VERSION_SCRIPT})
 endif()
 
-# config install
+# config install, please keep all install file with lite/BUILD:hdrs_public
+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/genfiles/lite_build_config.h
+        DESTINATION ${CMAKE_INSTALL_PREFIX}/lite/build_config)
+
 install(
   TARGETS lite_static
   LIBRARY DESTINATION lite/lib/${MGE_ARCH}
@@ -176,9 +180,6 @@ if(NOT WIN32)
     ARCHIVE DESTINATION lite/lib/${MGE_ARCH})
 endif()
 
-install(FILES ${PROJECT_SOURCE_DIR}/lite/include/lite/common_enum_c.h
-        DESTINATION ${CMAKE_INSTALL_PREFIX}/lite/include/lite-c)
-
 install(
   DIRECTORY ${PROJECT_SOURCE_DIR}/lite/include
   DESTINATION ${CMAKE_INSTALL_PREFIX}/lite
@@ -187,7 +188,7 @@ install(
 
 install(
   DIRECTORY ${PROJECT_SOURCE_DIR}/lite/lite-c/include
-  DESTINATION ${CMAKE_INSTALL_PREFIX}/lite
+  DESTINATION ${CMAKE_INSTALL_PREFIX}/lite/lite-c
   FILES_MATCHING
   PATTERN "*.h")
 

lite/build_config/lite_build_config.h

-/**
- * \file lite/build_config/lite_build_config.h
- *
- * This file is part of MegEngine, a deep learning framework developed by
- * Megvii.
- *
- * \copyright Copyright (c) 2020-2021 Megvii Inc. All rights reserved.
- */
+//! this file always for bazel
+
 #ifndef _HEADER_LITE_BUILD_CONFIG
 #define _HEADER_LITE_BUILD_CONFIG
 
@@ -28,4 +22,12 @@
 #ifndef LITE_ASSERT_LOC
 #define LITE_ASSERT_LOC 1
 #endif
+
+#ifndef LITE_BUILD_WITH_MGE
+#define LITE_BUILD_WITH_MGE 1
+#endif
+
+#ifndef LITE_BUILD_WITH_RKNPU
+#define LITE_BUILD_WITH_RKNPU 0
+#endif
 #endif  // _HEADER_LITE_BUILD_CONFIG

lite/src/lite_build_config.h.in → lite/build_config/lite_build_config.h.in

+//! this file always for cmake
+
 #ifndef _HEADER_LITE_BUILD_CONFIG
 #define _HEADER_LITE_BUILD_CONFIG
 
 #cmakedefine01 LITE_ENABLE_LOGGING
 #cmakedefine01 LITE_ENABLE_EXCEPTION
+#cmakedefine01 LITE_BUILD_WITH_MGE
 #cmakedefine01 LITE_WITH_CUDA
 #cmakedefine01 LITE_ASSERT_LOC
 
@@ -26,4 +29,8 @@
 #ifndef LITE_ASSERT_LOC
 #define LITE_ASSERT_LOC 0
 #endif
+
+#ifndef LITE_BUILD_WITH_MGE
+#define LITE_BUILD_WITH_MGE 0
+#endif
 #endif  // _HEADER_LITE_BUILD_CONFIG

lite/example/cpp_example/helper.h

+#pragma once
+
+//! helper function like define: lite/src/misc.h, but we need example code just
+//! depends on install header, not depends any lite src file
+#include <chrono>
+#include <cstdio>
+#include <stdexcept>
+#include <string>
+
+std::string exampe_ssprintf(const char* fmt = 0, ...)
+        __attribute__((format(printf, 1, 2)));
+
+#define LITE_EXAMPLE_THROW(msg)          \
+    do {                                 \
+        std::string msg_str(msg);        \
+        printf("%s\n", msg_str.c_str()); \
+        __builtin_trap();                \
+    } while (0)
+
+//! branch prediction hint: likely to take
+#define lite_example_likely(v) __builtin_expect(static_cast<bool>(v), 1)
+
+//! branch prediction hint: unlikely to take
+#define lite_example_unlikely(v) __builtin_expect(static_cast<bool>(v), 0)
+
+#define LITE_EXAMPLE_ASSERT(expr, msg...)                                           \
+    do {                                                                            \
+        if (lite_example_unlikely(!(expr))) {                                       \
+            auto info = exampe_ssprintf(msg);                                       \
+            LITE_EXAMPLE_THROW(exampe_ssprintf(                                     \
+                    "Assert \' %s \' failed at file : %s \n"                        \
+                    "line %d : %s,\nextra "                                         \
+                    "message: %s",                                                  \
+                    #expr, __FILE__, __LINE__, __PRETTY_FUNCTION__, info.c_str())); \
+        }                                                                           \
+    } while (0)
+
+#define LITE_EXAMPLE_MARK_USED_VAR(var) ((void)var)
+
+namespace lite_example_helper {
+class ScopedTimer {
+public:
+    typedef std::chrono::system_clock Clock;
+    typedef std::chrono::nanoseconds Nsec;
+
+    ScopedTimer(std::string name) : m_name(name) { m_start = Clock::now(); }
+    ~ScopedTimer() {
+        m_stop = Clock::now();
+        std::chrono::duration<double> elapsed = m_stop - m_start;
+        Nsec u = std::chrono::duration_cast<Nsec>(elapsed);
+        auto msg = exampe_ssprintf(
+                "%s used time %fms.", m_name.c_str(),
+                static_cast<double>(u.count()) / 1000000.f);
+        printf("%s", msg.c_str());
+    }
+
+private:
+    std::chrono::time_point<std::chrono::system_clock> m_start, m_stop;
+    const std::string m_name;
+};
+
+class Timer {
+public:
+    typedef std::chrono::system_clock Clock;
+    typedef std::chrono::nanoseconds Nsec;
+
+    Timer(std::string name) : m_name(name) { m_start = Clock::now(); }
+    double get_used_time() {
+        m_stop = Clock::now();
+        std::chrono::duration<double> elapsed = m_stop - m_start;
+        Nsec u = std::chrono::duration_cast<Nsec>(elapsed);
+        return static_cast<double>(u.count()) / 1000000.0;
+    }
+    void print_used_time(int iter) {
+        m_stop = Clock::now();
+        std::chrono::duration<double> elapsed = m_stop - m_start;
+        Nsec u = std::chrono::duration_cast<Nsec>(elapsed);
+        printf("%s used time %f ms\n", (m_name + std::to_string(iter)).c_str(),
+               static_cast<double>(u.count()) / 1000000.0);
+    }
+    void reset_start() { m_start = Clock::now(); }
+
+private:
+    std::chrono::time_point<std::chrono::system_clock> m_start, m_stop;
+    const std::string m_name;
+};
+
+inline void mark_used_variable() {}
+template <typename T, typename... Arg>
+inline void mark_used_variable(T firstArg, Arg... args) {
+    LITE_EXAMPLE_MARK_USED_VAR(firstArg);
+    mark_used_variable(args...);
+}
+}  // namespace lite_example_helper
+
+#if defined(_WIN32)
+#include <io.h>
+#include <windows.h>
+#undef CONST
+#define F_OK      0
+#define RTLD_LAZY 0
+// On the windows platform we use a lib_filename without a full path so
+// the win-api "LoadLibrary" would uses a standard search strategy to
+// find the lib module. As we cannot access to the lib_filename without a
+// full path, we should not use "access(a, b)" to verify it.
+#define access(a, b) false
+static inline void* dlopen(const char* file, int) {
+    return static_cast<void*>(LoadLibrary(file));
+}
+
+static inline char* dlerror() {
+    const char* errmsg = "dlerror not aviable in windows";
+    return const_cast<char*>(errmsg);
+}
+
+static inline void* dlsym(void* handle, const char* name) {
+    FARPROC symbol = GetProcAddress((HMODULE)handle, name);
+    return reinterpret_cast<void*>(symbol);
+}
+#elif __linux__ || __unix__ || __APPLE__
+#include <dlfcn.h>
+#include <unistd.h>
+#endif
+
+#if __DEPLOY_ON_XP_SP2__
+//! refer to
+//! https://docs.microsoft.com/en-us/cpp/build/configuring-programs-for-windows-xp?view=msvc-160
+//! xp sp2 do not support vc runtime fully, casused by KERNEL32.dll do not
+//! implement some base apis for c++ std function, for example,
+//! std::mutex/std::thread/std::condition_variable as a workround, we will
+//! disable some MegEngine feature on xp sp2 env, for exampe, multi-thread etc!
+#define LITE_MUTEX                  size_t
+#define LITE_RECURSIVE_MUTEX        size_t
+#define LITE_LOCK_GUARD(mtx)        LITE_EXAMPLE_MARK_USED_VAR(mtx)
+#define LITE_LOCK_GUARD_UNIQUE(mtx) LITE_EXAMPLE_MARK_USED_VAR(mtx)
+#define LITE_LOCK_GUARD_SHARED(mtx) \
+    LITE_EXAMPLE_MARK_USED_VAR(LITE_EXAMPLE_MARK_USED_VAR)
+#else
+#define LITE_MUTEX           std::mutex
+#define LITE_RECURSIVE_MUTEX std::recursive_mutex
+#define LITE_LOCK_GUARD(mtx) std::lock_guard<decltype(mtx)> LITE_LOCK_GUARD_CTOR(mtx)
+
+#define LITE_LOCK_GUARD_UNIQUE(mtx) \
+    std::unique_lock<decltype(mtx)> LITE_LOCK_GUARD_CTOR(mtx)
+
+#define LITE_LOCK_GUARD_SHARED(mtx) \
+    std::shared_lock<decltype(mtx)> LITE_LOCK_GUARD_CTOR(mtx)
+#endif
+
+// vim: syntax=cpp.doxygen foldmethod=marker foldmarker=f{{{,f}}}

lite/example/cpp_example/mge/basic.cpp

@@ -3,7 +3,7 @@
 #if LITE_BUILD_WITH_MGE
 #include <cstdio>
 
-#include "misc.h"
+#include "helper.h"
 
 using namespace lite;
 using namespace example;
@@ -96,12 +96,12 @@ bool basic_load_from_path(const Args& args) {
 
     //! forward
     {
-        lite::Timer ltimer("warmup");
+        lite_example_helper::Timer ltimer("warmup");
         network->forward();
         network->wait();
         ltimer.print_used_time(0);
     }
-    lite::Timer ltimer("forward_iter");
+    lite_example_helper::Timer ltimer("forward_iter");
     for (int i = 0; i < 10; i++) {
         network->forward();
         network->wait();
@@ -110,7 +110,7 @@ bool basic_load_from_path(const Args& args) {
 
     //! forward
     {
-        lite::Timer ltimer("warmup");
+        lite_example_helper::Timer ltimer("warmup");
         network->forward();
         network->wait();
         ltimer.print_used_time(0);
@@ -167,12 +167,12 @@ bool basic_load_from_path_with_loader(const Args& args) {
 
     //! forward
     {
-        lite::Timer ltimer("warmup");
+        lite_example_helper::Timer ltimer("warmup");
         network->forward();
         network->wait();
         ltimer.print_used_time(0);
     }
-    lite::Timer ltimer("forward_iter");
+    lite_example_helper::Timer ltimer("forward_iter");
     for (int i = 0; i < 10; i++) {
         ltimer.reset_start();
         network->forward();
@@ -481,12 +481,12 @@ bool load_from_path_run_cuda(const Args& args) {
 
     //! forward
     {
-        lite::Timer ltimer("warmup");
+        lite_example_helper::Timer ltimer("warmup");
         network->forward();
         network->wait();
         ltimer.print_used_time(0);
     }
-    lite::Timer ltimer("forward_iter");
+    lite_example_helper::Timer ltimer("forward_iter");
     for (int i = 0; i < 10; i++) {
         ltimer.reset_start();
         network->forward();

lite/example/cpp_example/mge/cv/detect_yolox.cpp

@@ -3,8 +3,6 @@
 #if LITE_BUILD_WITH_MGE
 #include <cstdio>
 
-#include "misc.h"
-
 #define STB_IMAGE_STATIC
 #define STB_IMAGE_IMPLEMENTATION
 #include "stb_image.h"

lite/example/cpp_example/mge/cv/picture_classification.cpp

@@ -3,8 +3,6 @@
 #if LITE_BUILD_WITH_MGE
 #include <cstdio>
 
-#include "misc.h"
-
 #define STB_IMAGE_STATIC
 #define STB_IMAGE_IMPLEMENTATION
 #include "stb_image.h"

lite/example/cpp_example/mge/device_io.cpp

 #include <thread>
 #include "example.h"
 #if LITE_BUILD_WITH_MGE
-#include "misc.h"
 
 using namespace lite;
 using namespace example;

lite/example/cpp_example/mge/lite_c_interface.cpp

 #include "example.h"
-#include "misc.h"
+#include "helper.h"
 #if LITE_BUILD_WITH_MGE
 #include "lite-c/global_c.h"
 #include "lite-c/network_c.h"
@@ -11,7 +11,7 @@
     do {                                               \
         int _ret = (_expr);                            \
         if (_ret) {                                    \
-            LITE_THROW(LITE_get_last_error()); \
+            LITE_EXAMPLE_THROW(LITE_get_last_error()); \
         }                                              \
     } while (0)
 
@@ -95,7 +95,7 @@ bool device_io_c_interface(const lite::example::Args& args) {
     LITE_CAPI_CHECK(
             LITE_get_tensor_total_size_in_byte(c_input_tensor, &length_tensor_in));
     if (length_read_in != length_tensor_in) {
-        LITE_THROW(
+        LITE_EXAMPLE_THROW(
                 "The input data size is not match the network input tensro "
                 "size,\n");
     }