From 18cc6abbb65f8ce17ac062f66447422003d0576b Mon Sep 17 00:00:00 2001
From: perseusdg <f20180523@goa.bits-pilani.ac.in>
Date: Tue, 7 Sep 2021 23:57:52 +0530
Subject: [PATCH] DeformableConvRT.h fixes ,cmake cuda arch auto detection

---
 .gitignore                                 |  3 +-
 CMakeLists.txt                             | 68 +++++++++++++++++++---
 cmake/getCudaArch.cu                       | 20 +++++++
 include/tkDNN/pluginsRT/DeformableConvRT.h |  2 +-
 4 files changed, 82 insertions(+), 11 deletions(-)
 create mode 100644 cmake/getCudaArch.cu

diff --git a/.gitignore b/.gitignore
index c1d362c..683834c 100644
--- a/.gitignore
+++ b/.gitignore
@@ -19,4 +19,5 @@ demo/BDD100K_val
 cmake-build-minsizerel/*
 scripts/COCO_val2017/*
 scripts/COCO_val2017.zip
-scripts/all_labels.txt
\ No newline at end of file
+scripts/all_labels.txt
+/cmake/cuda_script
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 9191adc..c10f6ac 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1,15 +1,67 @@
 cmake_minimum_required(VERSION 3.15)
-
-project (tkDNN)
+project(tkDNN)
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
-if(UNIX)
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14 -fPIC -Wno-deprecated-declarations -Wno-unused-variable -O3")
+set(CMAKE_CXX_STANDARD 14)
+
+find_package(CUDA 9.0 REQUIRED)
+if (CUDA_FOUND)
+    #Get CUDA compute capability
+    set(OUTPUTFILE ${CMAKE_CURRENT_SOURCE_DIR}/cmake/cuda_script) # No suffix required
+    set(CUDAFILE ${CMAKE_CURRENT_SOURCE_DIR}/cmake/getCudaArch.cu)
+    execute_process(COMMAND nvcc -lcuda ${CUDAFILE} -o ${OUTPUTFILE})
+    execute_process(COMMAND ${OUTPUTFILE}
+            RESULT_VARIABLE CUDA_RETURN_CODE
+            OUTPUT_VARIABLE ARCH)
+
+    if(${CUDA_RETURN_CODE} EQUAL 0)
+        set(CUDA_SUCCESS "TRUE")
+    else()
+        set(CUDA_SUCCESS "FALSE")
+    endif()
+
+    if (${CUDA_SUCCESS})
+        message(STATUS "CUDA Architecture: ${ARCH}")
+        message(STATUS "CUDA Version: ${CUDA_VERSION_STRING}")
+        message(STATUS "CUDA Path: ${CUDA_TOOLKIT_ROOT_DIR}")
+        message(STATUS "CUDA Libararies: ${CUDA_LIBRARIES}")
+        message(STATUS "CUDA Performance Primitives: ${CUDA_npp_LIBRARY}")
+
+        set(CUDA_NVCC_FLAGS "${ARCH}")
+        add_definitions(-DGPU) #You may not require this
+
+    else()
+        message(WARNING ${ARCH})
+    endif()
 endif()
+
+SET(CUDA_SEPARABLE_COMPILATION ON)
+
+
+if(UNIX)
+    if(CMAKE_BUILD_TYPE MATCHES Release)
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}  -fPIC -Wno-deprecated-declarations -Wno-unused-variable -O3")
+        set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32)
+    endif()
+
+    if(CMAKE_BUILD_TYPE MATCHES Debug)
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}  -fPIC -Wno-deprecated-declarations -Wno-unused-variable -g3")
+        set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32 -G -g)
+    endif()
+endif()
+
 if(WIN32)
-set(CMAKE_CXX_STANDARD 11)
-set(CMAKE_CXX_FLAGS "/O2 /FS /EHsc")
+    if(CMAKE_BUILD_TYPE MATCHES Release)
+        set(CMAKE_CXX_FLAGS "/O2 /FS /EHsc")
+        set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32 -arch=sm_61)
+    endif()
+
+    if(CMAKE_BUILD_TYPE MATCHES Debug)
+        set(CMAKE_CXX_FLAGS "/Od /FS /EHsc")
+        set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32 -arch=sm_61 -G -g)
+    endif()
 set(CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS ON)
 endif(WIN32)
+
 include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include/tkDNN)
 
 # project specific flags
@@ -28,10 +80,8 @@ endif()
 #-------------------------------------------------------------------------------
 # CUDA
 #-------------------------------------------------------------------------------
-find_package(CUDA 9.0 REQUIRED)
-SET(CUDA_SEPARABLE_COMPILATION ON)
+
 #set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -arch=sm_30 --compiler-options '-fPIC'")
-set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32 -arch=sm_61)
 
 
 find_package(CUDNN REQUIRED)
diff --git a/cmake/getCudaArch.cu b/cmake/getCudaArch.cu
new file mode 100644
index 0000000..1d66199
--- /dev/null
+++ b/cmake/getCudaArch.cu
@@ -0,0 +1,20 @@
+#include <stdio.h>
+
+int main(int argc, char **argv){
+    cudaDeviceProp dP;
+    float min_cc = 5.0;
+
+    int rc = cudaGetDeviceProperties(&dP, 0);
+    if(rc != cudaSuccess) {
+        cudaError_t error = cudaGetLastError();
+        printf("CUDA error: %s", cudaGetErrorString(error));
+        return rc; /* Failure */
+    }
+    if((dP.major+(dP.minor/10)) < min_cc) {
+        printf("Min Compute Capability of %2.1f required:  %d.%d found\n Not Building CUDA Code", min_cc, dP.major, dP.minor);
+        return 1; /* Failure */
+    } else {
+        printf("-arch=sm_%d%d", dP.major, dP.minor);
+        return 0; /* Success */
+    }
+}
\ No newline at end of file
diff --git a/include/tkDNN/pluginsRT/DeformableConvRT.h b/include/tkDNN/pluginsRT/DeformableConvRT.h
index adc5554..1cbf55d 100644
--- a/include/tkDNN/pluginsRT/DeformableConvRT.h
+++ b/include/tkDNN/pluginsRT/DeformableConvRT.h
@@ -110,6 +110,6 @@ namespace nvinfer1 {
         static std::vector<PluginField> mPluginAttributes;
         std::string mPluginNamespace;
     };
-
+    REGISTER_TENSORRT_PLUGIN(DeformableConvRTPluginCreator);
 };
 #endif