cleanup code

src/ngraph/runtime/gpu/gpu_call_frame.cpp

@@ -37,10 +37,10 @@ runtime::gpu::GPU_CallFrame::GPU_CallFrame(std::shared_ptr<GPU_ExternalFunction>
         throw runtime_error("cuBLAS create handle failed");
     }
     cudnnStatus_t cudnnStatus = cudnnCreate(&m_cudnn_handle);
-    if (cudnnStatus != CUDNN_STATUS_SUCCESS) 
+    if (cudnnStatus != CUDNN_STATUS_SUCCESS)
     {
         throw runtime_error("cuDnn create handle failed");
-    }         
+    }
 
     // Pass scalars as reference on the Host
     cublasSetPointerMode(m_cublas_handle, CUBLAS_POINTER_MODE_HOST);

src/ngraph/runtime/gpu/gpu_emitter.cpp

@@ -85,13 +85,7 @@ void runtime::gpu::GPU_Emitter::EmitAbs(codegen::CodeWriter& writer,
         const vector<runtime::gpu::GPU_TensorViewWrapper>& args,
         const vector<runtime::gpu::GPU_TensorViewWrapper>& out)
 {
-    writer << "{  // " << n->get_name() << "\n";
-    writer.indent++;
-    writer << "int count = " << out[0].get_size() << ";\n";
-    writer << "if(count == 0) return;\n";
-    writer << "ngraph::runtime::gpu::cuda::kernel::emit_abs((void*) " << args[0].get_name() <<  ", (void*) " << out[0].get_name() << ", count);\n";
-    writer.indent--;
-    writer << "}\n";
+    writer << " // " << n->get_name() << "\n  return;\n";
 }
 
 void runtime::gpu::GPU_Emitter::EmitAdd(codegen::CodeWriter& writer,
@@ -264,6 +258,7 @@ void runtime::gpu::GPU_Emitter::EmitDivide(codegen::CodeWriter& writer,
         const vector<runtime::gpu::GPU_TensorViewWrapper>& args,
         const vector<runtime::gpu::GPU_TensorViewWrapper>& out)
 {
+    writer << " // " << n->get_name() << "\n  return;\n";
 }
 
 void runtime::gpu::GPU_Emitter::EmitEqual(codegen::CodeWriter& writer,
@@ -271,6 +266,7 @@ void runtime::gpu::GPU_Emitter::EmitEqual(codegen::CodeWriter& writer,
         const vector<runtime::gpu::GPU_TensorViewWrapper>& args,
         const vector<runtime::gpu::GPU_TensorViewWrapper>& out)
 {
+    writer << " // " << n->get_name() << "\n  return;\n";
 }
 
 void runtime::gpu::GPU_Emitter::EmitGreater(codegen::CodeWriter& writer,
@@ -278,6 +274,7 @@ void runtime::gpu::GPU_Emitter::EmitGreater(codegen::CodeWriter& writer,
         const vector<runtime::gpu::GPU_TensorViewWrapper>& args,
         const vector<runtime::gpu::GPU_TensorViewWrapper>& out)
 {
+    writer << " // " << n->get_name() << "\n  return;\n";
 }
 
 void runtime::gpu::GPU_Emitter::EmitGreaterEq(

src/ngraph/runtime/gpu/gpu_external_function.cpp

@@ -22,12 +22,6 @@
 #include <typeindex>
 #include <typeinfo>
 #include <unordered_map>
-
-#include "cuda_runtime.h"
-#include "cudnn_v7.h"
-#include "cublas_v2.h"
-#include "cuda.h"
-
 #include "ngraph/codegen/code_writer.hpp"
 #include "ngraph/codegen/compiler.hpp"
 #include "ngraph/codegen/execution_engine.hpp"
@@ -255,11 +249,6 @@ void runtime::gpu::GPU_ExternalFunction::compile()
 
     string pch_header_source = writer.get_code();
 
-    writer += R"(
-    using namespace ngraph;
-    using namespace std;
-)";
-
     if (m_emit_timing)
     {
         writer << "// Declare debug timers\n";

test/CMakeLists.txt

@@ -79,7 +79,6 @@ if(NGRAPH_CPU_ENABLE AND LLVM_INCLUDE_DIR)
 endif()
 
 if(NGRAPH_GPU_ENABLE AND LLVM_INCLUDE_DIR)
-    message(STATUS "GPU PATH: ${CUDA_INCLUDE_DIRS} AND ${CUDNN_INCLUDE_DIR}")
     include_directories(SYSTEM /usr/local/cuda/include ${LLVM_INCLUDE_DIR} ${CUDA_INCLUDE_DIRS} ${CUDNN_INCLUDE_DIR})    
     link_directories(${LLVM_LIB_DIR})
     link_directories(${CUDA_LIBRARIES})
@@ -156,9 +155,7 @@ if(LLVM_INCLUDE_DIR)
 endif()
 
 if(CUDA_INCLUDE_DIRS)
-    message(STATUS "--------------------cuda--------------------------")
-    find_library(CUDA_nvrtc_LIBRARY nvrtc /usr/local/cuda/lib64)
-    target_link_libraries(unit-test ${CUDA_nvrtc_LIBRARY} ${CUDA_LIBRARIES} ${CUDNN_LIBRARIES} ${CUDA_CUBLAS_LIBRARIES})
+    target_link_libraries(unit-test ${CUDA_LIBRARIES} ${CUDNN_LIBRARIES} ${CUDA_CUBLAS_LIBRARIES})
 endif()
 
 target_link_libraries(unit-test ngraph libgtest pthread)