[FastTokenizer] Support FastTokenizer on Android (PaddlePaddle#783)

* [FastTokenizer] Support FastTokenizer on Android * [OMP] Add OMP bind_proc(close) policy * [Android] Add Lite support for UIE * [Android] Add VIS_SEG_OMP_NUM_THREADS
mei-727 · Dec 5, 2022 · 65f270d · 65f270d
1 parent 8b2f231
commit 65f270d
Show file tree

Hide file tree

Showing 6 changed files with 89 additions and 43 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -425,13 +425,6 @@ if(ENABLE_VISION)
   endif()
 endif()
 
-if(ANDROID OR IOS)
-  if(ENABLE_TEXT)
-    set(ENABLE_TEXT OFF CACHE BOOL "Force ENABLE_TEXT OFF" FORCE)
-    message(STATUS "Found Android or IOS, force ENABLE_TEXT OFF. We do not support fast_tokenizer with Android/IOS now.")
-  endif()
-endif()
-
 if(ENABLE_TEXT)
   add_definitions(-DENABLE_TEXT)
   list(APPEND ALL_DEPLOY_SRCS ${DEPLOY_TEXT_SRCS})
@@ -488,7 +481,7 @@ set_target_properties(${LIBRARY_NAME} PROPERTIES VERSION ${FASTDEPLOY_VERSION})
 if(MSVC)
   # disable warnings for dll export
   target_compile_options(${LIBRARY_NAME} PRIVATE "$<$<BUILD_INTERFACE:$<COMPILE_LANGUAGE:CXX>>:/wd4251>$<$<BUILD_INTERFACE:$<COMPILE_LANGUAGE:CUDA>>:-Xcompiler=/wd4251>")
-  file(GLOB FD_FILES_REQUIRE_BIGOBJ ${CSRCS_DIR_NAME}/fastdeploy/function/reduce.cc) 
+  file(GLOB FD_FILES_REQUIRE_BIGOBJ ${CSRCS_DIR_NAME}/fastdeploy/function/reduce.cc)
   set_source_files_properties(${FD_FILES_REQUIRE_BIGOBJ} PROPERTIES COMPILE_FLAGS "/bigobj")
 endif()
 

diff --git a/FastDeploy.cmake.in b/FastDeploy.cmake.in
@@ -210,11 +210,18 @@ endif()
 
 if (ENABLE_TEXT)
   if(ANDROID)
-    message(FATAL_ERROR "Not support fastdeploy text APIs with Android now!")
+    if(NOT ANDROID_TOOLCHAIN MATCHES "clang")
+      message(FATAL_ERROR "Currently, only support clang toolchain while cross compiling FastDeploy for Android with FastTokenizer, but found ${ANDROID_TOOLCHAIN}.")
+    endif()
+    add_library(core_tokenizers STATIC IMPORTED GLOBAL)
+    set_property(TARGET core_tokenizers PROPERTY IMPORTED_LOCATION
+                ${CMAKE_CURRENT_LIST_DIR}/third_libs/install/fast_tokenizer/lib/${ANDROID_ABI}/libcore_tokenizers.so)
+    list(APPEND FASTDEPLOY_LIBS core_tokenizers)
+  else()
+    # Add dependency libs later: Linux/Mac/Win/...
+    find_library(FAST_TOKENIZER_LIB core_tokenizers ${CMAKE_CURRENT_LIST_DIR}/third_libs/install/fast_tokenizer/lib NO_DEFAULT_PATH)
+    list(APPEND FASTDEPLOY_LIBS ${FAST_TOKENIZER_LIB})
   endif()
-  # Add dependency libs later
-  find_library(FAST_TOKENIZER_LIB core_tokenizers ${CMAKE_CURRENT_LIST_DIR}/third_libs/install/fast_tokenizer/lib NO_DEFAULT_PATH)
-  list(APPEND FASTDEPLOY_LIBS ${FAST_TOKENIZER_LIB})
   list(APPEND FASTDEPLOY_INCS ${CMAKE_CURRENT_LIST_DIR}/third_libs/install/fast_tokenizer/include)
   list(APPEND FASTDEPLOY_INCS ${CMAKE_CURRENT_LIST_DIR}/third_libs/install/fast_tokenizer/third_party/include)
 endif()

diff --git a/cmake/fast_tokenizer.cmake b/cmake/fast_tokenizer.cmake
@@ -24,9 +24,16 @@ set(FASTTOKENIZER_INC_DIR
     "${FASTTOKENIZER_INSTALL_DIR}/include"
     "${FASTTOKENIZER_INSTALL_DIR}/third_party/include"
     CACHE PATH "fast_tokenizer include directory." FORCE)
-set(FASTTOKENIZER_LIB_DIR
-    "${FASTTOKENIZER_INSTALL_DIR}/lib/"
-    CACHE PATH "fast_tokenizer lib directory." FORCE)
+if(ANDROID)
+  set(FASTTOKENIZER_LIB_DIR
+      "${FASTTOKENIZER_INSTALL_DIR}/lib/${ANDROID_ABI}"
+      CACHE PATH "fast_tokenizer lib directory." FORCE)
+else()
+  set(FASTTOKENIZER_LIB_DIR
+      "${FASTTOKENIZER_INSTALL_DIR}/lib/"
+      CACHE PATH "fast_tokenizer lib directory." FORCE)
+endif()    
+
 set(FASTTOKENIZER_THIRD_LIB_DIR
     "${FASTTOKENIZER_INSTALL_DIR}/third_party/lib/"
     CACHE PATH "fast_tokenizer lib directory." FORCE)
@@ -37,21 +44,21 @@ include_directories(${FASTTOKENIZER_INC_DIR})
 
 # Set lib path
 if(WIN32)
-set(FASTTOKENIZER_COMPILE_LIB "${FASTTOKENIZER_LIB_DIR}/core_tokenizers.lib"
-    CACHE FILEPATH "fast_tokenizer compile library." FORCE)
-message("FASTTOKENIZER_COMPILE_LIB = ${FASTTOKENIZER_COMPILE_LIB}")
-set(ICUDT_LIB "${FASTTOKENIZER_THIRD_LIB_DIR}/icudt.lib")
-set(ICUUC_LIB "${FASTTOKENIZER_THIRD_LIB_DIR}/icuuc.lib")
-
+  set(FASTTOKENIZER_COMPILE_LIB "${FASTTOKENIZER_LIB_DIR}/core_tokenizers.lib"
+      CACHE FILEPATH "fast_tokenizer compile library." FORCE)
+  set(ICUDT_LIB "${FASTTOKENIZER_THIRD_LIB_DIR}/icudt.lib")
+  set(ICUUC_LIB "${FASTTOKENIZER_THIRD_LIB_DIR}/icuuc.lib")
 elseif(APPLE)
-set(FASTTOKENIZER_COMPILE_LIB "${FASTTOKENIZER_LIB_DIR}/libcore_tokenizers.dylib"
-    CACHE FILEPATH "fast_tokenizer compile library." FORCE)
+  set(FASTTOKENIZER_COMPILE_LIB "${FASTTOKENIZER_LIB_DIR}/libcore_tokenizers.dylib"
+      CACHE FILEPATH "fast_tokenizer compile library." FORCE)
+elseif(ANDROID)
+  set(FASTTOKENIZER_COMPILE_LIB "${FASTTOKENIZER_LIB_DIR}/libcore_tokenizers.so"
+      CACHE FILEPATH "fast_tokenizer compile library." FORCE)
 else()
-
-set(FASTTOKENIZER_COMPILE_LIB "${FASTTOKENIZER_LIB_DIR}/libcore_tokenizers.so"
-    CACHE FILEPATH "fast_tokenizer compile library." FORCE)
-message("FASTTOKENIZER_COMPILE_LIB = ${FASTTOKENIZER_COMPILE_LIB}")
+  set(FASTTOKENIZER_COMPILE_LIB "${FASTTOKENIZER_LIB_DIR}/libcore_tokenizers.so"
+      CACHE FILEPATH "fast_tokenizer compile library." FORCE)
 endif(WIN32)
+message("FASTTOKENIZER_COMPILE_LIB = ${FASTTOKENIZER_COMPILE_LIB}")
 
 set(FASTTOKENIZER_URL_BASE "https://bj.bcebos.com/paddlenlp/fast_tokenizer/")
 set(FASTTOKENIZER_VERSION "1.0.0")
@@ -68,6 +75,15 @@ elseif(APPLE)
   else()
     set(FASTTOKENIZER_FILE "fast_tokenizer-osx-x86_64-${FASTTOKENIZER_VERSION}.tgz")
   endif()
+elseif(ANDROID)
+  # check ABI, toolchain
+  if((NOT ANDROID_ABI MATCHES "armeabi-v7a") AND (NOT ANDROID_ABI MATCHES "arm64-v8a"))
+    message(FATAL_ERROR "FastDeploy with FastTokenizer on Android only support armeabi-v7a, arm64-v8a now.")
+  endif()
+  if(NOT ANDROID_TOOLCHAIN MATCHES "clang")
+     message(FATAL_ERROR "Currently, only support clang toolchain while cross compiling FastDeploy for Android with FastTokenizer, but found ${ANDROID_TOOLCHAIN}.")
+  endif()    
+  set(FASTTOKENIZER_FILE "fast_tokenizer-android-${ANDROID_ABI}-${FASTTOKENIZER_VERSION}.tgz")
 else()
   if(CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "aarch64")
     set(FASTTOKENIZER_FILE "fast_tokenizer-linux-aarch64-${FASTTOKENIZER_VERSION}.tgz")
@@ -77,18 +93,39 @@ else()
 endif()
 set(FASTTOKENIZER_URL "${FASTTOKENIZER_URL_BASE}${FASTTOKENIZER_FILE}")
 
-ExternalProject_Add(
-  ${FASTTOKENIZER_PROJECT}
-  ${EXTERNAL_PROJECT_LOG_ARGS}
-  URL ${FASTTOKENIZER_URL}
-  PREFIX ${FASTTOKENIZER_PREFIX_DIR}
-  DOWNLOAD_NO_PROGRESS 1
-  CONFIGURE_COMMAND ""
-  BUILD_COMMAND ""
-  UPDATE_COMMAND ""
-  INSTALL_COMMAND
-    ${CMAKE_COMMAND} -E copy_directory ${FASTTOKENIZER_SOURCE_DIR} ${FASTTOKENIZER_INSTALL_DIR}
-  BUILD_BYPRODUCTS ${FASTTOKENIZER_COMPILE_LIB})
+if(ANDROID)
+  ExternalProject_Add(
+    ${FASTTOKENIZER_PROJECT}
+    ${EXTERNAL_PROJECT_LOG_ARGS}
+    URL ${FASTTOKENIZER_URL}
+    PREFIX ${FASTTOKENIZER_PREFIX_DIR}
+    DOWNLOAD_NO_PROGRESS 1
+    CONFIGURE_COMMAND ""
+    BUILD_COMMAND ""
+    UPDATE_COMMAND ""
+    INSTALL_COMMAND
+      ${CMAKE_COMMAND} -E remove_directory ${FASTTOKENIZER_INSTALL_DIR} &&
+      ${CMAKE_COMMAND} -E make_directory ${FASTTOKENIZER_INSTALL_DIR} &&  
+      ${CMAKE_COMMAND} -E make_directory ${FASTTOKENIZER_INSTALL_DIR}/lib && 
+      ${CMAKE_COMMAND} -E make_directory ${FASTTOKENIZER_INSTALL_DIR}/third_party &&
+      ${CMAKE_COMMAND} -E rename ${FASTTOKENIZER_SOURCE_DIR}/lib/ ${FASTTOKENIZER_INSTALL_DIR}/lib/${ANDROID_ABI} &&
+      ${CMAKE_COMMAND} -E copy_directory ${FASTTOKENIZER_SOURCE_DIR}/include ${FASTTOKENIZER_INSTALL_DIR}/include && 
+      ${CMAKE_COMMAND} -E copy_directory ${FASTTOKENIZER_SOURCE_DIR}/third_party/include ${FASTTOKENIZER_INSTALL_DIR}/third_party/include 
+    BUILD_BYPRODUCTS ${FASTTOKENIZER_COMPILE_LIB})
+else()
+  ExternalProject_Add(
+    ${FASTTOKENIZER_PROJECT}
+    ${EXTERNAL_PROJECT_LOG_ARGS}
+    URL ${FASTTOKENIZER_URL}
+    PREFIX ${FASTTOKENIZER_PREFIX_DIR}
+    DOWNLOAD_NO_PROGRESS 1
+    CONFIGURE_COMMAND ""
+    BUILD_COMMAND ""
+    UPDATE_COMMAND ""
+    INSTALL_COMMAND
+      ${CMAKE_COMMAND} -E copy_directory ${FASTTOKENIZER_SOURCE_DIR} ${FASTTOKENIZER_INSTALL_DIR}
+    BUILD_BYPRODUCTS ${FASTTOKENIZER_COMPILE_LIB})
+endif()
 
 add_library(fast_tokenizer STATIC IMPORTED GLOBAL)
 set_property(TARGET fast_tokenizer PROPERTY IMPORTED_LOCATION ${FASTTOKENIZER_COMPILE_LIB})

diff --git a/fastdeploy/backends/lite/lite_backend.cc b/fastdeploy/backends/lite/lite_backend.cc
@@ -221,6 +221,10 @@ bool LiteBackend::Infer(std::vector<FDTensor>& inputs,
       tensor->CopyFromCpu<uint8_t, paddle::lite_api::TargetType::kARM>(
         reinterpret_cast<const uint8_t*>(const_cast<void*>(
         inputs[i].CpuData())));
+    } else if (inputs[i].dtype == FDDataType::INT64) {
+      tensor->CopyFromCpu<int64_t, paddle::lite_api::TargetType::kARM>(
+        reinterpret_cast<const int64_t*>(const_cast<void*>(
+        inputs[i].CpuData())));
     } else {
       FDASSERT(false, "Unexpected data type of %d.", inputs[i].dtype);
     }

diff --git a/fastdeploy/text/uie/model.cc b/fastdeploy/text/uie/model.cc
@@ -230,7 +230,7 @@ bool UIEModel::Initialize() {
 
 void UIEModel::SetValidBackend() {
   // TODO(zhoushunjie): Add lite backend in future
-  valid_cpu_backends = {Backend::ORT, Backend::OPENVINO, Backend::PDINFER};
+  valid_cpu_backends = {Backend::ORT, Backend::OPENVINO, Backend::PDINFER, Backend::LITE};
   valid_gpu_backends = {Backend::ORT, Backend::PDINFER, Backend::TRT};
 }
 

diff --git a/fastdeploy/vision/visualize/segmentation.cc b/fastdeploy/vision/visualize/segmentation.cc
@@ -25,6 +25,8 @@ namespace fastdeploy {
 namespace vision {
 
 #ifdef __ARM_NEON  
+static constexpr int VIS_SEG_OMP_NUM_THREADS=2;
+
 static inline void QuantizeBlendingWeight8(
   float weight, uint8_t* old_multi_factor, uint8_t* new_multi_factor) {
   // Quantize the weight to boost blending performance.
@@ -53,7 +55,8 @@ static cv::Mat FastVisSegmentationNEON(
   const uint8_t *im_ptr = static_cast<const uint8_t*>(im.data);
 
   if (!quantize_weight) {
-    #pragma omp parallel for num_threads(2) schedule(static)
+    #pragma omp parallel for proc_bind(close) \
+    num_threads(VIS_SEG_OMP_NUM_THREADS) schedule(static)
     for (int i = 0; i < size - 15; i += 16) {
       uint8x16_t labelx16 = vld1q_u8(label_ptr + i); // 16 bytes
       // e.g 0b00000001 << 7 -> 0b10000000 128;
@@ -87,7 +90,8 @@ static cv::Mat FastVisSegmentationNEON(
 
   if (new_multi_factor == 8) {
     // Only keep mask, no need to blending with origin image.
-    #pragma omp parallel for num_threads(2) schedule(static)
+    #pragma omp parallel for proc_bind(close) \
+    num_threads(VIS_SEG_OMP_NUM_THREADS) schedule(static)
     for (int i = 0; i < size - 15; i += 16) {
       uint8x16_t labelx16 = vld1q_u8(label_ptr + i); // 16 bytes
       // e.g 0b00000001 << 7 -> 0b10000000 128;
@@ -112,7 +116,8 @@ static cv::Mat FastVisSegmentationNEON(
   uint8x16_t old_mulx16 = vdupq_n_u8(old_multi_factor);
   uint8x16_t new_mulx16 = vdupq_n_u8(new_multi_factor);
   // Blend the two colors together with quantize 'weight'.
-  #pragma omp parallel for num_threads(2) schedule(static)
+  #pragma omp parallel for proc_bind(close) \
+  num_threads(VIS_SEG_OMP_NUM_THREADS) schedule(static)
   for (int i = 0; i < size - 15; i += 16) {
     uint8x16x3_t bgrx16x3 = vld3q_u8(im_ptr + i * 3);  // 48 bytes
     uint8x16_t labelx16 = vld1q_u8(label_ptr + i); // 16 bytes