Dev 0.4.2

.github/workflows/android-arm-cpu.yml

@@ -13,19 +13,19 @@ jobs:
     outputs:
       CONDITION: ${{ steps.preflight.outputs.CONDITION }}
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v3
     - name: Preflight
       id: preflight
       run: |
-        echo ::set-output name=CONDITION::0
-        ./scripts/.ci/preflight.sh android || ret=$? && echo $ret && echo ::set-output name=CONDITION::$ret
+        echo "CONDITION=0" >> $GITHUB_OUTPUT
+        ./scripts/.ci/preflight.sh android || ret=$? && echo $ret && echo "CONDITION=$ret" >> $GITHUB_OUTPUT
 
   android:
     needs: [setup]
     if: ${{ needs.setup.outputs.CONDITION != '11' }}
     runs-on: ubuntu-latest
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v3
     - name: configure
       run: sudo apt-get install attr
     - name: build

.github/workflows/ios-cpu.yml

@@ -13,18 +13,18 @@ jobs:
     outputs:
       CONDITION: ${{ steps.preflight.outputs.CONDITION }}
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v3
     - name: Preflight
       id: preflight
       run: |
-        echo ::set-output name=CONDITION::0
-        ./scripts/.ci/preflight.sh ios || ret=$? && echo $ret && echo ::set-output name=CONDITION::$ret
+        echo "CONDITION=0" >> $GITHUB_OUTPUT
+        ./scripts/.ci/preflight.sh ios || ret=$? && echo $ret && echo "CONDITION=$ret" >> $GITHUB_OUTPUT
 
   ios-iphone-os:
     needs: [setup]
     if: ${{ needs.setup.outputs.CONDITION != '11' }}
     runs-on: macos-latest
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v3
     - name: build
       run: ./scripts/build_framework_ios.sh

.github/workflows/linux-x86-cpu-gcc.yml

@@ -9,22 +9,22 @@ on:
 
 jobs:
   setup:
-    runs-on: ubuntu-18.04
+    runs-on: ubuntu-22.04
     outputs:
       CONDITION: ${{ steps.preflight.outputs.CONDITION }}
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v3
     - name: Preflight
       id: preflight
       run: |
-        echo ::set-output name=CONDITION::0
-        ./scripts/.ci/preflight.sh x86 || ret=$? && echo $ret && echo ::set-output name=CONDITION::$ret
+        echo "CONDITION=0" >> $GITHUB_OUTPUT
+        ./scripts/.ci/preflight.sh x86 || ret=$? && echo $ret && echo "CONDITION=$ret" >> $GITHUB_OUTPUT
 
   linux-gcc:
     needs: [setup]
     if: ${{ needs.setup.outputs.CONDITION != '11' }}
-    runs-on: ubuntu-18.04
+    runs-on: ubuntu-22.04
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v3
     - name: build
       run: ./scripts/build_x86_linux.sh

.github/workflows/macos-x64-cpu.yml

@@ -13,19 +13,19 @@ jobs:
     outputs:
       CONDITION: ${{ steps.preflight.outputs.CONDITION }}
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v3
     - name: Preflight
       id: preflight
       run: |
-        echo ::set-output name=CONDITION::0
-        ./scripts/.ci/preflight.sh x86 || ret=$? && echo $ret && echo ::set-output name=CONDITION::$ret
+        echo "CONDITION=0" >> $GITHUB_OUTPUT
+        ./scripts/.ci/preflight.sh x86 || ret=$? && echo $ret && echo "CONDITION=$ret" >> $GITHUB_OUTPUT
 
   macos-clang:
     needs: [setup]
     if: ${{ needs.setup.outputs.CONDITION != '11' }}
     runs-on: macos-11
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v3
     - name: protobuf
       run: brew install protobuf opencv3
     - name: build

.gitignore

@@ -494,6 +494,10 @@ model/
 # opencl generated code
 opencl_program.cc
 
+# cache
+*.cache
+*.cache~
+
 # opencl generated code
 opencl_program.cc
 platforms/mac/tnn.xcodeproj/project.xcworkspace/xcuserdata/darrenyao.xcuserdatad/UserInterfaceState.xcuserstate

CMakeLists.txt

@@ -1,5 +1,6 @@
 cmake_minimum_required(VERSION 3.1)
-
+#set(CMAKE_CXX_STANDARD 14)
+#set(CMAKE_CXX_STANDARD_REQUIRED ON)
 # https://cmake.org/cmake/help/latest/policy/CMP0068.html
 if(POLICY CMP0068)
     cmake_policy(SET CMP0068 NEW)
@@ -15,9 +16,9 @@ project(TNN)
 ENABLE_LANGUAGE(ASM)
 
 set(TNN_MAJOR_VERSION 0)
-set(TNN_MINOR_VERSION 3)
-set(TNN_PATCH_VERSION 0)
-set(TNN_BUILD_VERSION 0)
+set(TNN_MINOR_VERSION 4)
+set(TNN_PATCH_VERSION 2)
+set(TNN_BUILD_VERSION 11)
 set(TNN_VERSION "${TNN_MAJOR_VERSION}.${TNN_MINOR_VERSION}.${TNN_PATCH_VERSION}.${TNN_BUILD_VERSION}")
 
 option(TNN_CPU_ENABLE "Enable Cpu" ON)
@@ -27,14 +28,16 @@ option(TNN_ARM82_ENABLE "Enable Arm82" OFF)
 option(TNN_METAL_ENABLE "Enable Metal" OFF)
 option(TNN_OPENCL_ENABLE "Enable OpenCL" OFF)
 option(TNN_CUDA_ENABLE "Enable CUDA" OFF)
-option(TNN_DSP_ENABLE "Enable DSP" OFF)
+option(TNN_SNPE_ENABLE "Enable Qualcomm SNPE DSP" OFF)
 option(TNN_ATLAS_ENABLE "Enable Atlas" OFF)
 option(TNN_TENSORRT_ENABLE "Enable TensorRT" OFF)
 option(TNN_OPENVINO_ENABLE  "Enable OPENVINO" OFF)
 option(TNN_APPLE_NPU_ENABLE "Enable NPU" OFF)
 option(TNN_HUAWEI_NPU_ENABLE "Enable NPU" OFF)
 option(TNN_RK_NPU_ENABLE "Enable RKNPU" OFF)
-option(TNN_JETSON_NANO_ENABLE "Enable Jetson Nano" OFF)
+option(TNN_TNNTORCH_ENABLE "Enable TNNTorch" OFF)
+option(TNN_ZIXIAO_ENABLE "Enable TNNTorch" OFF)
+option(TNN_TORCHVISION_ENABLE "Enable TorchVision" OFF)
 option(TNN_SYMBOL_HIDE "Enable Hide Symbol Visibility" ON)
 option(TNN_OPENMP_ENABLE "Enable OpenMP" OFF)
 option(TNN_BUILD_SHARED "Build Shared Library" ON)
@@ -52,12 +55,18 @@ option(TNN_ONNX2TNN_ENABLE "Enable ONNX2TNN Converter" OFF)
 option(TNN_TNN2MEM_ENABLE "Enable tnn2mem" OFF)
 option(TNN_BUILD_BENCHMARK_TEST_LIB_ENABLE "Enable Build Benchmark Test Lib" OFF)
 option(TNN_GLIBCXX_USE_CXX11_ABI_ENABLE "Enable Use CXX11 ABI" ON)
+option(TNN_PYBIND_ENABLE  "Enable Pybind" OFF)
 option(TNN_METAL_FLOAT32 "Enable Metal Float32" OFF)
 option(TNN_COREML_FLOAT32 "Enable Float32 CoreML Model" ON)
 option(TNN_DYNAMIC_RANGE_QUANTIZATION_ENABLE "Enable Dynamic Range Quantization" OFF)
+option(TNN_PACK_TORCH_LIB "Enable Torch Lib Pack in release" ON)
+option(TNN_CUDA_JETSON_ENABLE "Enable CUDA build for Nvidia Jetson Driving Chips like Orin, Thor etc." OFF)
 
 set(TNN_USE_GFLAGS OFF)
 
+set(CMAKE_INSTALL_RPATH_USE_LINK_PATH TRUE)
+set(CMAKE_SKIP_BUILD_RPATH FALSE)
+
 message(${CMAKE_SOURCE_DIR})
 message(${CMAKE_CURRENT_SOURCE_DIR})
 
@@ -76,6 +85,10 @@ if(TNN_PROFILER_ENABLE)
     set(TNN_SYMBOL_HIDE OFF)
 endif()
 
+if(TNN_TORCHVISION_ENABLE)
+    add_definitions(-DTNN_TORCHVISION)
+endif()
+
 if(TNN_BENCHMARK_MODE)
     add_definitions(-DGENERATE_RESOURCE)
 endif()
@@ -131,12 +144,17 @@ if(TNN_UNIT_TEST_ENABLE)
     add_definitions(-DGENERATE_RESOURCE)
 endif()
 
+if(TNN_MATCHER_TEST_ENABLE)
+    set(TNN_SYMBOL_HIDE OFF)
+endif()
+
 if(TNN_CONVERTER_ENABLE)
     set(TNN_ONNX2TNN_ENABLE ON)
 endif()
 
 if(TNN_CONVERTER_ENABLE OR TNN_ONNX2TNN_ENABLE)
     set(TNN_SYMBOL_HIDE OFF)
+    set(TNN_PYBIND_ENABLE ON)
     add_definitions(-DTNN_CONVERTER_RUNTIME)
 endif()
 
@@ -220,7 +238,11 @@ if(UNIX)
     endif()
 endif()
 
-set(CMAKE_CXX_STANDARD 11)
+if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU" AND CMAKE_CXX_COMPILER_VERSION VERSION_LESS 9.0)
+    set(CMAKE_CXX_STANDARD 11)
+else()
+    set(CMAKE_CXX_STANDARD 17)
+endif()
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 if(TNN_METAL_ENABLE OR TNN_APPLE_NPU_ENABLE)
@@ -256,14 +278,16 @@ message(STATUS "\tArm82:\t${TNN_ARM82_ENABLE}")
 message(STATUS "\tMetal:\t${TNN_METAL_ENABLE}")
 message(STATUS "\tOpenCL:\t${TNN_OPENCL_ENABLE}")
 message(STATUS "\tCUDA:\t${TNN_CUDA_ENABLE}")
-message(STATUS "\tDSP:\t${TNN_DSP_ENABLE}")
+message(STATUS "\tSNPE:\t${TNN_SNPE_ENABLE}")
 message(STATUS "\tAtlas:\t${TNN_ATLAS_ENABLE}")
 message(STATUS "\tTensorRT:\t${TNN_TENSORRT_ENABLE}")
 message(STATUS "\tAppleNPU:\t${TNN_APPLE_NPU_ENABLE}")
 message(STATUS "\tHuaweiNPU:\t${TNN_HUAWEI_NPU_ENABLE}")
 message(STATUS "\tRKNPU:\t${TNN_RK_NPU_ENABLE}")
 message(STATUS "\tJetson Nano:\t${TNN_JETSON_NANO_ENABLE}")
 message(STATUS "\tOpenVINO:\t${TNN_OPENVINO_ENABLE}")
+message(STATUS "\tTNNTorch:\t${TNN_TNNTORCH_ENABLE}")
+message(STATUS "\tZIXIAO:\t${TNN_ZIXIAO_ENABLE}")
 message(STATUS "\tOpenMP:\t${TNN_OPENMP_ENABLE}")
 message(STATUS "\tTEST:\t${TNN_TEST_ENABLE}")
 message(STATUS "\t--Unit Test:\t${TNN_UNIT_TEST_ENABLE}")
@@ -279,6 +303,7 @@ message(STATUS "\tTNN2MEM:\t${TNN_TNN2MEM_ENABLE}")
 message(STATUS "\tBENCHMARK Test Lib:\t${TNN_BUILD_BENCHMARK_TEST_LIB_ENABLE}")
 message(STATUS "\tDynamic Range Quantization:\t${TNN_DYNAMIC_RANGE_QUANTIZATION_ENABLE}")
 message(STATUS "\tSHARING_MEM_WITH_OPENGL:\t${SHARING_MEM_WITH_OPENGL}")
+message(STATUS "\tCuda Build Jetson Chips:\t${TNN_CUDA_JETSON_ENABLE}")
 
 include_directories(include)
 include_directories(source)
@@ -385,19 +410,58 @@ if(TNN_CUDA_ENABLE)
     set(TARGET_OBJECTS ${TARGET_OBJECTS} "$<TARGET_OBJECTS:TNNCuda>")
 endif()
 
+if(TNN_SNPE_ENABLE)
+    if(ANDROID_ABI STREQUAL "armeabi-v7a")
+        # SNPE 2.11+ no longer support ARMv7
+        message(STATUS "TNN SNPE not available on Android ARMv7")
+    else()
+        link_directories(third_party/snpe/lib/aarch64-android/)
+        add_subdirectory(source/tnn/device/snpe)
+        set(TARGET_OBJECTS ${TARGET_OBJECTS} "$<TARGET_OBJECTS:TNNSNPE>")
+    endif()
+endif()
+
 if(TNN_HUAWEI_NPU_ENABLE)
     if(ANDROID_ABI STREQUAL "armeabi-v7a")
         link_directories(
-                third_party/huawei_npu/hiai_ddk_latest/armeabi-v7a/
+                third_party/huawei_npu/hiai_ddk_latest/ddk/ai_ddk_lib/lib/
         )
     else()
         link_directories(
-                third_party/huawei_npu/hiai_ddk_latest/arm64-v8a/
+                third_party/huawei_npu/hiai_ddk_latest/ddk/ai_ddk_lib/lib64/
         )
     endif()
     add_subdirectory(source/tnn/device/huawei_npu)
     set(TARGET_OBJECTS ${TARGET_OBJECTS} "$<TARGET_OBJECTS:TNNNPU>")
 endif()
+if(TNN_ATLAS_ENABLE)
+    add_definitions(-DGET_NETWORK_ENABLE)
+    add_subdirectory(source/tnn/device/atlas)
+    set(TARGET_OBJECTS ${TARGET_OBJECTS} "$<TARGET_OBJECTS:TNNAtlas>")
+    include_directories(${CMAKE_SOURCE_DIR}/source/tnn/device/atlas)
+    set(ASCEND_PATH $ENV{DDK_PATH})
+    if (NOT DEFINED ENV{DDK_PATH})
+        set(ASCEND_PATH "/usr/local/Ascend/ascend-toolkit/latest")
+        message(STATUS "set default ASCEND_PATH: ${ASCEND_PATH}")
+    else ()
+        message(STATUS "env ASCEND_PATH: ${ASCEND_PATH}")
+    endif()
+    set(ACL_LIB_PATH $ENV{NPU_HOST_LIB})
+    if (NOT DEFINED ENV{NPU_HOST_LIB})
+        set(ACL_LIB_PATH "/usr/local/Ascend/ascend-toolkit/latest/lib64")
+        message(STATUS "set default ACL_LIB_PATH: ${ACL_LIB_PATH}")
+    else ()
+        message(STATUS "env ACL_LIB_PATH: ${ACL_LIB_PATH}")
+    endif()
+    # Header path
+    include_directories(
+        ${ASCEND_PATH}/acllib/include/
+        )
+    # add host lib path
+    link_directories(
+        ${ACL_LIB_PATH}
+        )
+endif()
 
 if(TNN_RK_NPU_ENABLE)
     if(CMAKE_SIZEOF_VOID_P EQUAL 8)
@@ -413,9 +477,23 @@ if(TNN_RK_NPU_ENABLE)
     set(TARGET_OBJECTS ${TARGET_OBJECTS} "$<TARGET_OBJECTS:TNNRKNPU>")
 endif()
 
+if(TNN_TNNTORCH_ENABLE)
+    add_subdirectory(source/tnn/network/torch)
+    set(TARGET_OBJECTS ${TARGET_OBJECTS} "$<TARGET_OBJECTS:TNNTorch>")
+endif()
+
+if(TNN_ZIXIAO_ENABLE)
+    add_subdirectory(source/tnn/device/zixiao)
+    set(TARGET_OBJECTS ${TARGET_OBJECTS} "$<TARGET_OBJECTS:TNNZIXIAO>")
+endif()
+
 if(TNN_BUILD_SHARED)
     add_library(TNN SHARED ${SRC} ${TARGET_OBJECTS})
     set_target_properties(TNN PROPERTIES VERSION ${TNN_VERSION} SOVERSION ${TNN_MAJOR_VERSION})
+    set_target_properties(TNN PROPERTIES LINK_FLAGS "-Wl,-rpath,$ORIGIN")
+    if (NOT TNN_PACK_TORCH_LIB)
+        set_target_properties(TNN PROPERTIES LINK_FLAGS "-Wl,-rpath,$ORIGIN -Wl,-rpath,$ORIGIN/../torch/lib")
+    endif()
     if(SHARING_MEM_WITH_OPENGL)
         if(SYSTEM.Windows)
             target_link_libraries(TNN opengl32)
@@ -456,6 +534,19 @@ elseif(SYSTEM.Windows)
     include(platforms/windows/CMakeLists.txt)
 endif()
 
+if(TNN_PYBIND_ENABLE)
+    set(CMAKE_CXX_STANDARD 17)
+    include_directories(third_party/pybind11/include)
+    add_subdirectory(third_party/pybind11)
+    file(GLOB_RECURSE TORCH_SRC "source/pytnn/*.cc")
+    add_library(_pytnn SHARED ${TORCH_SRC})
+    target_link_libraries(_pytnn pybind11::module)
+    set_target_properties(_pytnn PROPERTIES PREFIX "${PYTHON_MODULE_PREFIX}"
+                                         SUFFIX "${PYTHON_MODULE_EXTENSION}")
+    set_target_properties(_pytnn PROPERTIES LINK_FLAGS "-Wl,-rpath,\$ORIGIN")
+    target_link_libraries(_pytnn TNN)
+endif()
+
 if (TNN_TEST_ENABLE OR TNN_CONVERTER_ENABLE OR TNN_MODEL_CHECK_ENABLE OR TNN_DYNAMIC_RANGE_QUANTIZATION_ENABLE)
     set(TNN_USE_GFLAGS ON)
 endif ()
@@ -499,6 +590,7 @@ endif()
 if(TNN_DYNAMIC_RANGE_QUANTIZATION_ENABLE)
     add_subdirectory(tools/dynamic_range_quantization)
 endif()
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DPYBIND11_COMPILER_TYPE=\\\"_gcc\\\" -DPYBIND11_STDLIB=\\\"_libstdcpp\\\" -DPYBIND11_BUILD_ABI=\\\"_cxxabi1011\\\"")
 
 if (MSVC)
     target_compile_options(TNN PUBLIC "/Zc:__cplusplus")