Update docs and optimzie the convdw of rv64 (#631)

* add the vs compile ug, update docs

* optimize the convdw kernel of rv64

Author: BUG1989

doc/compile.md

@@ -236,6 +236,7 @@ popd
 @ENDLOCAL
 ```
 
-## 6. 总结
+## 6.  使用 Microsoft Visual Studio 编译
+
+请参考 [Visual Studio 使用说明](visual_sudio_user_manual.md)。
 
-本文档只是简单指导如何编译对应的 Tengine Lite 版本，有需要可以参考 ` Tengine-Lite/build.sh` 文件。

doc/faq.md

@@ -4,7 +4,7 @@
 
 ### 支持模型
 
-已支持：TensorFlow、TFLite、PyTorch ( ONNX )、MXNet、Caffe、Darknet
+已支持：TensorFlow、TFLite、PyTorch ( ONNX )、MXNet、Caffe、Darknet、OneFlow、PaddlePaddle
 
 计划中：TensorFlow2 ( MLIR )、PyTorch ( TorchScript )
 
@@ -14,7 +14,14 @@
 
 ## 如何获取模型中间结果？
 
-创建 cmake 工程时添加 `-DTENGINE_DEBUG_DATA=ON` 变量，将在每一层的 input、output tensor 数据保存到 `./output` 中，以 mobilenet_v1.tmfile 为例：
+### 使用方法
+
+- 程序执行前，添加环境变量 `export TG_DEBUG_DATA=1`，启用精度 Profiler 功能；
+- 删除环境变量 `unset TG_DEBUG_DATA`， 关闭精度 Profiler 功能。
+
+### logo 信息
+
+数据导出后，在程序执行的当前路径下生成 `./output` 文件夹。
 
 ```bash
 $ ./tm_classification -m models/squeezenet.tmfile -i images/cat.jpg
@@ -93,53 +100,51 @@ pool6_out_blob_data.txt
 
 ## 如何获取模型各个layer耗时？
 
-创建 cmake 工程时添加 `-DTENGINE_DEBUG_TIME=ON` 变量，将在网络模型运行过程中打印每一层的耗时情况，以 mobilenet_v1.tmfile 为例：
+### 使用方法
 
-```bash
-$ ./tm_classification -m models/squeezenet.tmfile -i images/cat.jpg
-Convolution              13.29 ms  conv1-conv1/bn-conv1/scale-relu1
-Convolution              16.78 ms  conv2_1/dw-conv2_1/dw/bn-conv2_1/dw/scale-relu2_1/dw
-Convolution              32.15 ms  conv2_1/sep-conv2_1/sep/bn-conv2_1/sep/scale-relu2_1/sep
-Convolution               7.85 ms  conv2_2/dw-conv2_2/dw/bn-conv2_2/dw/scale-relu2_2/dw
-Convolution              26.75 ms  conv2_2/sep-conv2_2/sep/bn-conv2_2/sep/scale-relu2_2/sep
-Convolution              15.02 ms  conv3_1/dw-conv3_1/dw/bn-conv3_1/dw/scale-relu3_1/dw
-Convolution              59.70 ms  conv3_1/sep-conv3_1/sep/bn-conv3_1/sep/scale-relu3_1/sep
-Convolution               3.48 ms  conv3_2/dw-conv3_2/dw/bn-conv3_2/dw/scale-relu3_2/dw
-Convolution              28.39 ms  conv3_2/sep-conv3_2/sep/bn-conv3_2/sep/scale-relu3_2/sep
-Convolution               8.57 ms  conv4_1/dw-conv4_1/dw/bn-conv4_1/dw/scale-relu4_1/dw
-Convolution              61.16 ms  conv4_1/sep-conv4_1/sep/bn-conv4_1/sep/scale-relu4_1/sep
-Convolution               2.21 ms  conv4_2/dw-conv4_2/dw/bn-conv4_2/dw/scale-relu4_2/dw
-Convolution              31.55 ms  conv4_2/sep-conv4_2/sep/bn-conv4_2/sep/scale-relu4_2/sep
-Convolution               4.19 ms  conv5_1/dw-conv5_1/dw/bn-conv5_1/dw/scale-relu5_1/dw
-Convolution              63.83 ms  conv5_1/sep-conv5_1/sep/bn-conv5_1/sep/scale-relu5_1/sep
-Convolution               3.96 ms  conv5_2/dw-conv5_2/dw/bn-conv5_2/dw/scale-relu5_2/dw
-Convolution              65.00 ms  conv5_2/sep-conv5_2/sep/bn-conv5_2/sep/scale-relu5_2/sep
-Convolution               4.95 ms  conv5_3/dw-conv5_3/dw/bn-conv5_3/dw/scale-relu5_3/dw
-Convolution              65.26 ms  conv5_3/sep-conv5_3/sep/bn-conv5_3/sep/scale-relu5_3/sep
-Convolution               4.02 ms  conv5_4/dw-conv5_4/dw/bn-conv5_4/dw/scale-relu5_4/dw
-Convolution              64.07 ms  conv5_4/sep-conv5_4/sep/bn-conv5_4/sep/scale-relu5_4/sep
-Convolution               4.21 ms  conv5_5/dw-conv5_5/dw/bn-conv5_5/dw/scale-relu5_5/dw
-Convolution              69.94 ms  conv5_5/sep-conv5_5/sep/bn-conv5_5/sep/scale-relu5_5/sep
-Convolution               1.89 ms  conv5_6/dw-conv5_6/dw/bn-conv5_6/dw/scale-relu5_6/dw
-Convolution              31.51 ms  conv5_6/sep-conv5_6/sep/bn-conv5_6/sep/scale-relu5_6/sep
-Convolution               3.37 ms  conv6/dw-conv6/dw/bn-conv6/dw/scale-relu6/dw
-Convolution              63.23 ms  conv6/sep-conv6/sep/bn-conv6/sep/scale-relu6/sep
-Pooling                   0.23 ms  pool6
-Convolution               1.53 ms  fc7
+- 程序执行前，添加环境变量 `export TG_DEBUG_TIME=1`，启用性能 Profiler 功能；
+- 删除环境变量 `unset TG_DEBUG_TIME`， 关闭性能 Profiler 功能。
+
+### logo 信息
 
-model file : models/squeezenet.tmfile
-image file : images/cat.jpg
-label_file : (null)
-img_h, img_w, scale[3], mean[3] : 227 227 , 1.000 1.000 1.000, 104.0 116.7 122.7
-Repeat 1 times, thread 1, avg time 480.62 ms, max_time 480.62 ms, min_time 480.62 ms
---------------------------------------
-0.273199, 281
-0.267552, 282
-0.181004, 278
-0.081799, 285
-0.072407, 151
---------------------------------------
 ```
+ 0 [ 7.48% :  0.7 ms]   Convolution idx:  5 shape: {1   3 100 100} -> {1   8  50  50}     int8 K: 3x3 | S: 2x2 | P: 0 1 0 1         MFLOPS:  1.08 Rate:1519
+ 1 [ 6.66% :  0.6 ms]   Convolution idx:  8 shape: {1   8  50  50} -> {1   8  50  50}     int8 K: 3x3 | S: 1x1 | P: 1 1 1 1 DW(  8) MFLOPS:  0.36 Rate:569
+ 2 [ 9.54% :  0.9 ms]   Convolution idx: 11 shape: {1   8  50  50} -> {1  16  50  50}     int8 K: 1x1 | S: 1x1 | P: 0 0 0 0         MFLOPS:  0.64 Rate:706
+ 3 [ 3.99% :  0.4 ms]   Convolution idx: 14 shape: {1  16  50  50} -> {1  16  25  25}     int8 K: 3x3 | S: 2x2 | P: 0 1 0 1 DW( 16) MFLOPS:  0.18 Rate:475
+ 4 [ 6.77% :  0.6 ms]   Convolution idx: 17 shape: {1  16  25  25} -> {1  32  25  25}     int8 K: 1x1 | S: 1x1 | P: 0 0 0 0         MFLOPS:  0.64 Rate:995
+ 5 [ 6.90% :  0.7 ms]   Convolution idx: 20 shape: {1  32  25  25} -> {1  32  25  25}     int8 K: 3x3 | S: 1x1 | P: 1 1 1 1 DW( 32) MFLOPS:  0.36 Rate:549
+ 6 [ 4.20% :  0.4 ms]   Convolution idx: 23 shape: {1  32  25  25} -> {1  32  25  25}     int8 K: 1x1 | S: 1x1 | P: 0 0 0 0         MFLOPS:  1.28 Rate:3207
+ 7 [ 1.42% :  0.1 ms]   Convolution idx: 26 shape: {1  32  25  25} -> {1  32  13  13}     int8 K: 3x3 | S: 2x2 | P: 1 1 1 1 DW( 32) MFLOPS:  0.10 Rate:721
+ 8 [ 2.36% :  0.2 ms]   Convolution idx: 29 shape: {1  32  13  13} -> {1  64  13  13}     int8 K: 1x1 | S: 1x1 | P: 0 0 0 0         MFLOPS:  0.69 Rate:3092
+ 9 [ 3.43% :  0.3 ms]   Convolution idx: 32 shape: {1  64  13  13} -> {1  64  13  13}     int8 K: 3x3 | S: 1x1 | P: 1 1 1 1 DW( 64) MFLOPS:  0.19 Rate:597
+10 [ 3.98% :  0.4 ms]   Convolution idx: 35 shape: {1  64  13  13} -> {1  64  13  13}     int8 K: 1x1 | S: 1x1 | P: 0 0 0 0         MFLOPS:  1.38 Rate:3663
+11 [ 0.80% :  0.1 ms]   Convolution idx: 38 shape: {1  64  13  13} -> {1  64   7   7}     int8 K: 3x3 | S: 2x2 | P: 1 1 1 1 DW( 64) MFLOPS:  0.06 Rate:741
+12 [ 2.24% :  0.2 ms]   Convolution idx: 41 shape: {1  64   7   7} -> {1 128   7   7}     int8 K: 1x1 | S: 1x1 | P: 0 0 0 0         MFLOPS:  0.80 Rate:3771
+13 [ 1.59% :  0.2 ms]   Convolution idx: 44 shape: {1 128   7   7} -> {1 128   7   7}     int8 K: 3x3 | S: 1x1 | P: 1 1 1 1 DW(128) MFLOPS:  0.11 Rate:747
+14 [ 4.21% :  0.4 ms]   Convolution idx: 47 shape: {1 128   7   7} -> {1 128   7   7}     int8 K: 1x1 | S: 1x1 | P: 0 0 0 0         MFLOPS:  1.61 Rate:4015
+15 [ 1.53% :  0.1 ms]   Convolution idx: 50 shape: {1 128   7   7} -> {1 128   7   7}     int8 K: 3x3 | S: 1x1 | P: 1 1 1 1 DW(128) MFLOPS:  0.11 Rate:778
+16 [ 4.41% :  0.4 ms]   Convolution idx: 53 shape: {1 128   7   7} -> {1 128   7   7}     int8 K: 1x1 | S: 1x1 | P: 0 0 0 0         MFLOPS:  1.61 Rate:3833
+17 [ 1.66% :  0.2 ms]   Convolution idx: 56 shape: {1 128   7   7} -> {1 128   7   7}     int8 K: 3x3 | S: 1x1 | P: 1 1 1 1 DW(128) MFLOPS:  0.11 Rate:715
+18 [ 4.16% :  0.4 ms]   Convolution idx: 59 shape: {1 128   7   7} -> {1 128   7   7}     int8 K: 1x1 | S: 1x1 | P: 0 0 0 0         MFLOPS:  1.61 Rate:4065
+19 [ 1.52% :  0.1 ms]   Convolution idx: 62 shape: {1 128   7   7} -> {1 128   7   7}     int8 K: 3x3 | S: 1x1 | P: 1 1 1 1 DW(128) MFLOPS:  0.11 Rate:784
+20 [ 4.46% :  0.4 ms]   Convolution idx: 65 shape: {1 128   7   7} -> {1 128   7   7}     int8 K: 1x1 | S: 1x1 | P: 0 0 0 0         MFLOPS:  1.61 Rate:3786
+21 [ 1.59% :  0.2 ms]   Convolution idx: 68 shape: {1 128   7   7} -> {1 128   7   7}     int8 K: 3x3 | S: 1x1 | P: 1 1 1 1 DW(128) MFLOPS:  0.11 Rate:748
+22 [ 4.37% :  0.4 ms]   Convolution idx: 71 shape: {1 128   7   7} -> {1 128   7   7}     int8 K: 1x1 | S: 1x1 | P: 0 0 0 0         MFLOPS:  1.61 Rate:3869
+23 [ 0.54% :  0.1 ms]   Convolution idx: 74 shape: {1 128   7   7} -> {1 128   4   4}     int8 K: 3x3 | S: 2x2 | P: 1 1 1 1 DW(128) MFLOPS:  0.04 Rate:722
+24 [ 2.88% :  0.3 ms]   Convolution idx: 77 shape: {1 128   4   4} -> {1 256   4   4}     int8 K: 1x1 | S: 1x1 | P: 0 0 0 0         MFLOPS:  1.05 Rate:3825
+25 [ 1.02% :  0.1 ms]   Convolution idx: 80 shape: {1 256   4   4} -> {1 256   4   4}     int8 K: 3x3 | S: 1x1 | P: 1 1 1 1 DW(256) MFLOPS:  0.07 Rate:761
+26 [ 5.54% :  0.5 ms]   Convolution idx: 81 shape: {1 256   4   4} -> {1 256   4   4}     int8 K: 1x1 | S: 1x1 | P: 0 0 0 0         MFLOPS:  2.10 Rate:3986
+27 [ 0.11% :  0.0 ms]       Pooling idx: 84 shape: {1 256   4   4} -> {1 256   1   1}     int8 K: 4x4 | S: 1x1 | P: 0 0 0 0         Avg
+28 [ 0.27% :  0.0 ms] FullyConnected idx: 85 shape: {1 256   1   1} -> {1 131   1   1}    int8
+29 [ 0.40% :  0.0 ms]       Softmax idx: 86 shape: {1 131   1   1} -> {1 131   1   1}     int8
+```
+
+## 如何屏蔽性能算子？
 
-## Todo......
+Naive Profiler，用于关闭 CPU 性能算子，后端计算只使用 Naive C 实现的 reference op，用于对比分析性能算子的计算结果。
+ 
+### 使用方法
 
+- 程序执行前，添加环境变量 `export TG_DEBUG_REF=1`，启用 Naive Profiler 功能；
+- 删除环境变量 `unset TG_DEBUG_REF`， 关闭精度 Naive Profiler 功能。

doc/gpu_cuda_user_manual.md

@@ -10,18 +10,13 @@ Todo
 
 On Ubuntu
 
-### setup nvcc enva
-```
-$ export CUDACXX=/usr/local/cuda/bin/nvcc
-```
 ### build
 ```bash
 $ cd <tengine-lite-root-dir>
 $ mkdir -p build-linux-cuda
-$ cmake -DTENGINE_ENABLE_CUDABACKEND=ON ..
+$ cmake -DTENGINE_ENABLE_CUDA=ON ..
 
 $ make -j4
-$ make install
 ```
 
 ## Demo

doc/gpu_opencl_user_manual.md

@@ -12,7 +12,7 @@ $ export ROOT_PATH={Path of tengine-lite}
 ```
 ### Build
 
-`-DOPENCL_LIBRARY: libOpenCL.so 路径。可通过 <sudo find /usr -name "libOpenCL.so"> 命令查询`
+`-DOPENCL_LIBRARY: libOpenCL.so 文件夹路径。可通过 <sudo find /usr -name "libOpenCL.so"> 命令查询`
 
 `-DOPENCL_INCLUDE_DIRS：指定CL/cl.h 路径。可通过 <sudo find /usr -name "cl.h"> 命令查询`
 
@@ -21,7 +21,7 @@ $ cd <tengine-lite-root-dir>
 $ mkdir -p build-linux-opencl
 $ cmake \
 -DTENGINE_ENABLE_OPENCL=ON \
--DOPENCL_LIBRARY=/usr/lib/aarch64-linux-gnu/libOpenCL.so \
+-DOPENCL_LIBRARY=/usr/lib/aarch64-linux-gnu \
 -DOPENCL_INCLUDE_DIRS=/usr/include ..
 
 $ make -j4

doc/npu_tim-vx_user_manual.md

@@ -27,15 +27,6 @@ $ cd tengine-lite
 
 **non-cross-compilation**
 
-```bash
-$ cd <TIM-VX-root-dir>
-$ mkdir build && cd build
-$ cmake ..
-$ make -j4
-```
-
-**Create depend files**
-
 ```bash
 $ cd <tengine-lite-root-dir>
 $ mkdir -p ./3rdparty/tim-vx/lib/x86_64
@@ -45,10 +36,17 @@ $ cp -rf ../TIM-VX/src    ./src/dev/tim-vx/
 $ cp -rf ../TIM-VX/prebuilt-sdk/x86_64_linux/include/*    ./3rdparty/tim-vx/include/
 $ cp -rf ../TIM-VX/prebuilt-sdk/x86_64_linux/lib/*    ./3rdparty/tim-vx/lib/x86_64/
 $ rm ./src/dev/tim-vx/src/tim/vx/*_test.cc
+```
 
-$ cp -rf ../TIM-VX/build/src/tim/vx/libtim-vx.so    ./3rdparty/tim-vx/lib/x86_64/
+Build Tengine
 
+```bash
 $ export LD_LIBRARY_PATH=<tengine-lite-root-dir>/3rdparty/tim-vx/lib/x86_64
+
+$ cd <tengine-lite-root-dir>
+$ mkdir build && cd build
+$ cmake -DTENGINE_ENABLE_TIM_VX=ON ..
+$ make -j4
 ```
 
 #### 2.2 Prepare for Khadas VIM3 platform
@@ -59,14 +57,26 @@ Prepare for VIM3 prebuild sdk:
 $ wget -c https://github.com/VeriSilicon/TIM-VX/releases/download/v1.1.28/aarch64_A311D_D312513_A294074_R311680_T312233_O312045.tgz
 $ tar zxvf aarch64_A311D_D312513_A294074_R311680_T312233_O312045.tgz
 $ mv aarch64_A311D_D312513_A294074_R311680_T312233_O312045 prebuild-sdk-a311d
+
+$ cd <tengine-lite-root-dir>
+$ mkdir -p ./3rdparty/tim-vx/lib/aarch64
+$ mkdir -p ./3rdparty/tim-vx/include
+$ cp -rf ../TIM-VX/include/*    ./3rdparty/tim-vx/include/
+$ cp -rf ../TIM-VX/src    ./src/dev/tim-vx/
+$ cp -rf ../prebuild-sdk-a311d/include/*    ./3rdparty/tim-vx/include/
+$ cp -rf ../prebuild-sdk-a311d/lib/*    ./3rdparty/tim-vx/lib/aarch64/
+$ rm ./src/dev/tim-vx/src/tim/vx/*_test.cc
 ```
 
 **2.2.1 cross-compilation**
 
+TOOLCHAIN_FILE in the <tengine-lite-root-dir>/toolchains
 ```bash
-$ cd <TIM-VX-root-dir>
+$ export LD_LIBRARY_PATH=<tengine-lite-root-dir>/3rdparty/tim-vx/lib/aarch64
+
+$ cd <tengine-lite-root-dir>
 $ mkdir build && cd build
-$ cmake -DCONFIG=A311D ..
+$ cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/aarch64-linux-gnu.toolchain.cmake -DTENGINE_ENABLE_TIM_VX=ON ..
 $ make -j4
 ```
 
@@ -100,48 +110,19 @@ $ mv /usr/lib/libOpenVX.so* ./Backup
 $ cp -rf ../prebuild-sdk-a311d/lib/libOpenVX.so* /usr/lib
 ```
 
-build for libtim-vx.so:
+Build Tengine
 
 ```bash
-$ cd <TIM-VX-root-dir>
+$ cd <tengine-lite-root-dir>
 $ mkdir build && cd build
-$ cmake .. 
+$ cmake -DTENGINE_ENABLE_TIM_VX=ON ..
 $ make -j4
 ```
 
-##### Create depend files
-
-```bash
-$ cd <tengine-lite-root-dir>
-$ mkdir -p ./3rdparty/tim-vx/lib/aarch64
-$ cp -rf ../TIM-VX/build/src/tim/vx/libtim-vx.so    ./3rdparty/tim-vx/lib/aarch64/
-
-$ export LD_LIBRARY_PATH=<tengine-lite-root-dir>/3rdparty/tim-vx/lib/aarch64
-```
-
 #### 2.3 Prepare for NXP platform
 
 **non-cross-compilation**
 
-```bash
-$ cd <TIM-VX-root-dir>
-$ mkdir build && cd build
-$ cmake ..
-$ make -j4
-```
-
-**Create depend files**
-
-```bash
-$ cd <tengine-lite-root-dir>
-$ mkdir -p ./3rdparty/tim-vx/lib/aarch64
-$ cp -rf ../TIM-VX/build/src/tim/vx/libtim-vx.so    ./3rdparty/tim-vx/lib/aarch64/
-
-$ export LD_LIBRARY_PATH=<tengine-lite-root-dir>/3rdparty/tim-vx/lib/aarch64
-```
-
-#### 2.4 Build Tengine Lite with TIM-VX
-
 ```bash
 $ cd <tengine-lite-root-dir>
 $ mkdir build && cd build
@@ -154,9 +135,6 @@ $ make -j4
 #### 3.1 Depned librarys
 
 ```
-3rdparty/tim-vx/lib/
-├── libtim-vx.so
-
 build-tim-vx-arm64/install/lib/
 └── libtengine-lite.so
 ```
@@ -202,8 +180,9 @@ Repeat 10 times, thread 1, avg time 2.95 ms, max_time 3.42 ms, min_time 2.76 ms
 ### 4. Support list
 | Vendor  | Devices      |
 | ------- | ------------ |
-| Amlogic | A311D        |
+| Amlogic | A311D, S905D3|
 | NXP     | i.MX 8M Plus |
+| JLQ     | JA310        |
 | X86-64  | Simulator    |
 
 ### 5. The uint8 quantization model

doc/visual_sudio_user_manual.md

@@ -0,0 +1,37 @@
+# Tengine Microsoft Visual Studio User Manual
+
+## Brief
+
+The Visual Studio dev tools & services make app development easy for any platform & language. Tengine support building on windows now.
+
+
+## Prepare
+CMake >= 3.13, Visual Studio >= 2015
+
+Before the very begging, please check CMake and Visual Studio already has been installed. CMake >= 3.13, Visual Studio Version 2017 or 2019 is recommended.
+For CUDA or TensorRT backend user, CMake >= 3.18 is needed. CUDA or TensorRT needs to be installed or unpackaged.
+
+
+## Build
+
+### Download
+Download https://github.com/OAID/Tengine.git from GitHub first of all. 
+
+#### CMD shell user
+Open "x86 Native Tools Command Prompt for VS 201x" or "x64 Native Tools Command Prompt for VS 201x", "201x" is your installed version. Suppose VS2017 was installed, then:
+
+```bash
+set PATH=X:/your/cmake/bin;%PATH%
+
+cd /d X:/your/downloaded/Tengine
+md build
+cd build
+cmake.exe -G "Visual Studio 15 2017 Win64" -DTENGINE_OPENMP=OFF -DTENGINE_BUILD_EXAMPLES=OFF ..
+::cmake.exe -G "Visual Studio 16 2019" -A x64 -DTENGINE_OPENMP=OFF ..
+cmake.exe --build . --parallel %NUMBER_OF_PROCESSORS%
+cmake.exe --build . --target install
+```
+
+## Demo
+
+TODO

scripts/build.sh

@@ -35,6 +35,13 @@ cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/aarch64-linux-gnu.toolchain.cmake ..
 cmake --build . --parallel `nproc` && cmake --build . --target install
 popd
 
+##### linux for rv64-c906 toolchain
+mkdir -p build-aarch64-linux-gnu
+pushd build-aarch64-linux-gnu
+cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/rv64-c906.toolchain.cmake ..
+cmake --build . --parallel `nproc` && cmake --build . --target install
+popd
+
 ##### linux of hisiv200
 mkdir -p build-hisiv200-linux
 pushd build-hisiv200-linux

source/device/cpu/cpu_node.h

@@ -82,14 +82,6 @@ struct node_ops
 
     /* score */
     int (*score)(struct node_ops*, struct exec_graph*, struct node*);
-
-#ifdef CONFIG_AUTH_DEVICE
-    void (*InitTimeLimited)(struct node_ops*);
-    unsigned long time_limited;
-    bool skip_run;
-    int run_count;
-    unsigned long tv_start;
-#endif
 };
 
 int init_exec_node(struct exec_graph* exec_graph, struct exec_node* exec_node, struct node* ir_node, struct node_ops* node_ops);