Pushing working VGG16_ImageNet - CUDNN Pass

llvm/test/VISC/DNN_Benchmarks/benchmarks/vgg16_imagenet/Makefile

+DNN_BENCHMARK_ROOT = $(LLVM_SRC_ROOT)/test/VISC/DNN_Benchmarks
+HPVM_BUILD_DIR = $(LLVM_BUILD_ROOT)
+
+CC = $(HPVM_BUILD_DIR)/bin/clang++
+OPT = $(HPVM_BUILD_DIR)/bin/opt
+LLVM_DIS = $(HPVM_BUILD_DIR)/bin/llvm-dis
+LLVM_LINK = $(HPVM_BUILD_DIR)/bin/llvm-link
+LLVM_INCLUDE_DIR = $(LLVM_SRC_ROOT)/include
+
+SRC_DIR = src
+BUILD_DIR = build
+APP = vgg16_imagenet
+
+TENSOR_INCLUDE_DIR = $(DNN_BENCHMARK_ROOT)/common/include
+TENSOR_RT_INCLUDE_DIR = $(LLVM_SRC_ROOT)/projects/hpvm-tensor-rt/tensor_runtime/include
+TENSOR_LIB_DIR = $(LLVM_SRC_ROOT)/projects/hpvm-tensor-rt/lib/libtensor_runtime.a
+PROFILER_LIB_DIR = $(LLVM_SRC_ROOT)/projects/gpu_profiler/lib/libgpu_profiler.a
+SOC_SIMULATOR_LIB_DIR = $(LLVM_SRC_ROOT)/projects/soc_simulator/lib/libpromise_profiler.a
+TENSOR_AUTOTUNER_DIR = $(LLVM_SRC_ROOT)/projects/hpvm-tensor-rt/lib/libtensor_autotuner.a
+
+CC_FLAGS = -I $(LLVM_INCLUDE_DIR) -I $(TENSOR_INCLUDE_DIR) -I $(TENSOR_RT_INCLUDE_DIR) -I $(CUDA_INCLUDE_PATH)  -fno-exceptions -ffast-math -std=c++11 -O3
+CCFLAGS += -DDEVICE=CUDNN_TARGET
+LINKER_FLAGS = -lpthread -lcudart -lcurand -lcudnn -lcublas -lcufft -lOpenCL -lstdc++fs
+
+HPVM_LIB_DIR = $(HPVM_BUILD_DIR)/lib
+
+
+VISC_OPTFLAGS = -load  $(HPVM_LIB_DIR)/LLVMBuildDFG.so -load $(HPVM_LIB_DIR)/LLVMInPlaceDFGAnalysis.so -load  $(HPVM_LIB_DIR)/LLVMDFG2LLVM_CUDNN.so -load  $(HPVM_LIB_DIR)/LLVMDFG2LLVM_X86.so -load  $(HPVM_LIB_DIR)/LLVMClearDFG.so -inplace -dfg2llvm-cudnn -dfg2llvm-x86 -clearDFG
+
+
+PROMISE_QUANT_FILE_PATH=$(LLVM_SRC_ROOT)/test/VISC/DNN_Benchmarks/benchmarks/$(APP)/data/quant_ranges.txt
+
+VISC_OPTFLAGS2 = -load  $(HPVM_LIB_DIR)/LLVMBuildDFG.so -load $(HPVM_LIB_DIR)/LLVMInPlaceDFGAnalysis.so -load  $(HPVM_LIB_DIR)/LLVMDFG2LLVM_PROMISE.so  -load  $(HPVM_LIB_DIR)/LLVMDFG2LLVM_CUDNN.so    -load  $(HPVM_LIB_DIR)/LLVMDFG2LLVM_X86.so -load  $(HPVM_LIB_DIR)/LLVMFuseHPVMTensorNodes.so  -load  $(HPVM_LIB_DIR)/LLVMClearDFG.so   -inplace -hpvm-fuse -dfg2llvm-promise  -quantization-levels-filename=$(PROMISE_QUANT_FILE_PATH) -dfg2llvm-cudnn  -dfg2llvm-x86 -clearDFG
+
+WRAPPER_API_QUANT_FILE_PATH=$(LLVM_SRC_ROOT)/test/VISC/DNN_Benchmarks/benchmarks/$(APP)/data/quant_ranges_rt.txt
+CONF_FILE_PATH=$(LLVM_SRC_ROOT)/test/VISC/DNN_Benchmarks/benchmarks/$(APP)/data/tuner_confs.txt
+
+
+VISC_OPTFLAGS3 = -load  $(HPVM_LIB_DIR)/LLVMBuildDFG.so -load $(HPVM_LIB_DIR)/LLVMInPlaceDFGAnalysis.so -load  $(HPVM_LIB_DIR)/LLVMDFG2LLVM_WrapperAPI.so    -load  $(HPVM_LIB_DIR)/LLVMDFG2LLVM_X86.so -load  $(HPVM_LIB_DIR)/LLVMFuseHPVMTensorNodes.so  -load  $(HPVM_LIB_DIR)/LLVMClearDFG.so   -inplace -hpvm-fuse -dfg2llvm-wrapperapi -quantization-levels-filename=$(WRAPPER_API_QUANT_FILE_PATH) -configuration-inputs-filename=$(CONF_FILE_PATH) -dfg2llvm-x86 -clearDFG
+
+
+TARGET = $(BUILD_DIR)/$(APP).opt.bc
+SOURCES = $(SRC_DIR)/$(APP).cpp
+VISC_RT_PATH = $(LLVM_SRC_ROOT)/../build/projects/visc-rt/visc-rt.ll
+
+#OBJS = $(BUILD_DIR)/$(wildcabrd *.ll)
+.PRECIOUS: $(BUILD_DIR)/$(APP).ll $(BUILD_DIR)/$(APP).visc.ll
+default: $(BUILD_DIR) $(TARGET)
+
+
+$(BUILD_DIR)/%.ll: $(SRC_DIR)/%.cpp
+	$(CC) $(CC_FLAGS) -emit-llvm src/$(APP).cpp -S -o  $(BUILD_DIR)/$(APP).ll  
+	#-- $(CC) $(CC_FLAGS) -emit-llvm src/$(APP)_promise.cpp -S -o $(BUILD_DIR)/$(APP)_promise.ll
+	#-- $(CC) $(CC_FLAGS) -emit-llvm src/$(APP)_loop.cpp -S -o  $(BUILD_DIR)/$(APP)_loop.ll  
+
+$(BUILD_DIR)/%.opt.bc: $(BUILD_DIR)/%.ll
+	$(OPT) -load LLVMGenVISC.so -genvisc -globaldce  $(BUILD_DIR)/$(APP).ll -S -o  $(BUILD_DIR)/$(APP).visc.ll
+	#-- $(OPT) -load LLVMGenVISC.so -genvisc -globaldce  $(BUILD_DIR)/$(APP)_promise.ll -S -o  $(BUILD_DIR)/$(APP)_promise.visc.ll
+	#-- $(OPT) -load LLVMGenVISC.so -genvisc -globaldce  $(BUILD_DIR)/$(APP)_loop.ll -S -o  $(BUILD_DIR)/$(APP)_loop.visc.ll
+	$(OPT) $(VISC_OPTFLAGS)  $(BUILD_DIR)/$(APP).visc.ll  -o  $(BUILD_DIR)/$(APP)_cudnn.bc
+	#-- $(OPT) $(VISC_OPTFLAGS3) $(BUILD_DIR)/$(APP)_promise.visc.ll  -o  $(BUILD_DIR)/$(APP)_wrapperapi.bc
+	#-- $(OPT) $(VISC_OPTFLAGS3) $(BUILD_DIR)/$(APP)_loop.visc.ll  -o  $(BUILD_DIR)/$(APP)_loop_wrapperapi.bc
+
+	$(LLVM_LINK) $(BUILD_DIR)/$(APP)_cudnn.bc $(VISC_RT_PATH) -o $(BUILD_DIR)/$(APP)_cudnn_linked.bc
+	#-- $(LLVM_LINK) $(BUILD_DIR)/$(APP)_wrapperapi.bc $(VISC_RT_PATH) -o $(BUILD_DIR)/$(APP)_wrapperapi_linked.bc
+	#-- $(LLVM_LINK) $(BUILD_DIR)/$(APP)_loop_wrapperapi.bc $(VISC_RT_PATH) -o $(BUILD_DIR)/$(APP)_loop_wrapperapi_linked.bc
+	$(CC) $(BUILD_DIR)/$(APP)_cudnn_linked.bc $(TENSOR_LIB_DIR) $(PROFILER_LIB_DIR) $(SOC_SIMULATOR_LIB_DIR) -o $(BUILD_DIR)/$(APP)_cudnn_linked $(LINKER_FLAGS)
+	#- $(CC) $(BUILD_DIR)/$(APP)_wrapperapi_linked.bc $(TENSOR_LIB_DIR) $(PROFILER_LIB_DIR) $(SOC_SIMULATOR_LIB_DIR) -o $(BUILD_DIR)/$(APP)_wrapperapi_linked $(LINKER_FLAGS)
+	#-- $(CC) $(BUILD_DIR)/$(APP)_loop_wrapperapi_linked.bc $(TENSOR_LIB_DIR) $(PROFILER_LIB_DIR) $(SOC_SIMULATOR_LIB_DIR) -o $(BUILD_DIR)/$(APP)_loop_wrapperapi_linked $(LINKER_FLAGS)
+
+
+$(BUILD_DIR):
+	mkdir -p $@
+
+clean:
+	rm -rf $(BUILD_DIR)

llvm/test/VISC/DNN_Benchmarks/benchmarks/vgg16_imagenet/src/vgg16_imagenet.cpp

+
+#include <stdio.h> 
+#include <stdlib.h> 
+#include <unistd.h> 
+#include <fcntl.h> 
+#include <sys/stat.h> 
+#include <cstring> 
+#include <visc.h> 
+#include <tensorTypes.h> 
+#include <tensorUtils.h> 
+
+void var_0_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_convolution(t1, t2, 1, 1, 1, 1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_1_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_add(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_2_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_relu(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_3_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_convolution(t1, t2, 1, 1, 1, 1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_4_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_add(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_5_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_relu(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_6_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_pool_max(t1, 2, 2, 0, 0, 2, 2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_7_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_convolution(t1, t2, 1, 1, 1, 1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_8_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_add(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_9_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_relu(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_10_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_convolution(t1, t2, 1, 1, 1, 1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_11_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_add(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_12_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_relu(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_13_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_pool_max(t1, 2, 2, 0, 0, 2, 2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_14_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_convolution(t1, t2, 1, 1, 1, 1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_15_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_add(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_16_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_relu(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_17_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_convolution(t1, t2, 1, 1, 1, 1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_18_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_add(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_19_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_relu(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_20_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_convolution(t1, t2, 1, 1, 1, 1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_21_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_add(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_22_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_relu(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_23_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_pool_max(t1, 2, 2, 0, 0, 2, 2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_24_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_convolution(t1, t2, 1, 1, 1, 1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_25_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_add(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_26_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_relu(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_27_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_convolution(t1, t2, 1, 1, 1, 1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_28_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_add(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_29_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_relu(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_30_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_convolution(t1, t2, 1, 1, 1, 1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_31_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_add(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_32_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_relu(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_33_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_pool_max(t1, 2, 2, 0, 0, 2, 2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_34_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_convolution(t1, t2, 1, 1, 1, 1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_35_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_add(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_36_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_relu(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_37_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_convolution(t1, t2, 1, 1, 1, 1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_38_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_add(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_39_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_relu(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_40_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_convolution(t1, t2, 1, 1, 1, 1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_41_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_add(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_42_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_relu(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_43_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_pool_max(t1, 2, 2, 0, 0, 2, 2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_44_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_mul(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_45_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_add(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_46_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_relu(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_47_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_mul(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_48_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_add(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_49_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_relu(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_50_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_mul(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_51_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_add(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_52_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::CUDNN_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_softmax(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void root(void* input, size_t input_bytes, 
+	  void* conv2d_1_w, size_t conv2d_1_w_bytes, 
+	  void* conv2d_1_b, size_t conv2d_1_b_bytes, 
+	  void* conv2d_2_w, size_t conv2d_2_w_bytes, 
+	  void* conv2d_2_b, size_t conv2d_2_b_bytes, 
+	  void* conv2d_3_w, size_t conv2d_3_w_bytes, 
+	  void* conv2d_3_b, size_t conv2d_3_b_bytes, 
+	  void* conv2d_4_w, size_t conv2d_4_w_bytes, 
+	  void* conv2d_4_b, size_t conv2d_4_b_bytes, 
+	  void* conv2d_5_w, size_t conv2d_5_w_bytes, 
+	  void* conv2d_5_b, size_t conv2d_5_b_bytes, 
+	  void* conv2d_6_w, size_t conv2d_6_w_bytes, 
+	  void* conv2d_6_b, size_t conv2d_6_b_bytes, 
+	  void* conv2d_7_w, size_t conv2d_7_w_bytes, 
+	  void* conv2d_7_b, size_t conv2d_7_b_bytes, 
+	  void* conv2d_8_w, size_t conv2d_8_w_bytes, 
+	  void* conv2d_8_b, size_t conv2d_8_b_bytes, 
+	  void* conv2d_9_w, size_t conv2d_9_w_bytes, 
+	  void* conv2d_9_b, size_t conv2d_9_b_bytes, 
+	  void* conv2d_10_w, size_t conv2d_10_w_bytes, 
+	  void* conv2d_10_b, size_t conv2d_10_b_bytes, 
+	  void* conv2d_11_w, size_t conv2d_11_w_bytes, 
+	  void* conv2d_11_b, size_t conv2d_11_b_bytes, 
+	  void* conv2d_12_w, size_t conv2d_12_w_bytes, 
+	  void* conv2d_12_b, size_t conv2d_12_b_bytes, 
+	  void* conv2d_13_w, size_t conv2d_13_w_bytes, 
+	  void* conv2d_13_b, size_t conv2d_13_b_bytes, 
+	  void* dense_1_w, size_t dense_1_w_bytes, 
+	  void* dense_1_b, size_t dense_1_b_bytes, 
+	  void* dense_2_w, size_t dense_2_w_bytes, 
+	  void* dense_2_b, size_t dense_2_b_bytes, 
+	  void* dense_3_w, size_t dense_3_w_bytes, 
+	  void* dense_3_b, size_t dense_3_b_bytes){ 
+
+
+  __visc__hint(visc::CPU_TARGET); 
+  __visc__attributes(33, input, conv2d_1_w, conv2d_1_b, conv2d_2_w, conv2d_2_b, conv2d_3_w, conv2d_3_b, conv2d_4_w, conv2d_4_b, conv2d_5_w, conv2d_5_b, conv2d_6_w, conv2d_6_b, conv2d_7_w, conv2d_7_b, conv2d_8_w, conv2d_8_b, conv2d_9_w, conv2d_9_b, conv2d_10_w, conv2d_10_b, conv2d_11_w, conv2d_11_b, conv2d_12_w, conv2d_12_b, conv2d_13_w, conv2d_13_b, dense_1_w, dense_1_b, dense_2_w, dense_2_b, dense_3_w, dense_3_b, 0); 
+
+
+  void* var_0 = __visc__createNodeND(0, var_0_node); 
+
+  __visc__bindIn(var_0, 0, 0, 0); 
+  __visc__bindIn(var_0, 1, 1, 0); 
+  __visc__bindIn(var_0, 2, 2, 0); 
+  __visc__bindIn(var_0, 3, 3, 0); 
+
+  void* var_1 = __visc__createNodeND(0, var_1_node); 
+
+  __visc__edge(var_0, var_1, 1, 0, 0, 0); 
+  __visc__edge(var_0, var_1, 1, 1, 1, 0); 
+  __visc__bindIn(var_1, 4, 2, 0); 
+  __visc__bindIn(var_1, 5, 3, 0); 
+
+  void* var_2 = __visc__createNodeND(0, var_2_node); 
+
+  __visc__edge(var_1, var_2, 1, 0, 0, 0); 
+  __visc__edge(var_1, var_2, 1, 1, 1, 0); 
+
+  void* var_3 = __visc__createNodeND(0, var_3_node); 
+
+  __visc__edge(var_2, var_3, 1, 0, 0, 0); 
+  __visc__edge(var_2, var_3, 1, 1, 1, 0); 
+  __visc__bindIn(var_3, 6, 2, 0); 
+  __visc__bindIn(var_3, 7, 3, 0); 
+
+  void* var_4 = __visc__createNodeND(0, var_4_node); 
+
+  __visc__edge(var_3, var_4, 1, 0, 0, 0); 
+  __visc__edge(var_3, var_4, 1, 1, 1, 0); 
+  __visc__bindIn(var_4, 8, 2, 0); 
+  __visc__bindIn(var_4, 9, 3, 0); 
+
+  void* var_5 = __visc__createNodeND(0, var_5_node); 
+
+  __visc__edge(var_4, var_5, 1, 0, 0, 0); 
+  __visc__edge(var_4, var_5, 1, 1, 1, 0); 
+
+  void* var_6 = __visc__createNodeND(0, var_6_node); 
+
+  __visc__edge(var_5, var_6, 1, 0, 0, 0); 
+  __visc__edge(var_5, var_6, 1, 1, 1, 0); 
+
+  void* var_7 = __visc__createNodeND(0, var_7_node); 
+
+  __visc__edge(var_6, var_7, 1, 0, 0, 0); 
+  __visc__edge(var_6, var_7, 1, 1, 1, 0); 
+  __visc__bindIn(var_7, 10, 2, 0); 
+  __visc__bindIn(var_7, 11, 3, 0); 
+
+  void* var_8 = __visc__createNodeND(0, var_8_node); 
+
+  __visc__edge(var_7, var_8, 1, 0, 0, 0); 
+  __visc__edge(var_7, var_8, 1, 1, 1, 0); 
+  __visc__bindIn(var_8, 12, 2, 0); 
+  __visc__bindIn(var_8, 13, 3, 0); 
+
+  void* var_9 = __visc__createNodeND(0, var_9_node); 
+
+  __visc__edge(var_8, var_9, 1, 0, 0, 0); 
+  __visc__edge(var_8, var_9, 1, 1, 1, 0); 
+
+  void* var_10 = __visc__createNodeND(0, var_10_node); 
+
+  __visc__edge(var_9, var_10, 1, 0, 0, 0); 
+  __visc__edge(var_9, var_10, 1, 1, 1, 0); 
+  __visc__bindIn(var_10, 14, 2, 0); 
+  __visc__bindIn(var_10, 15, 3, 0); 
+
+  void* var_11 = __visc__createNodeND(0, var_11_node); 
+
+  __visc__edge(var_10, var_11, 1, 0, 0, 0); 
+  __visc__edge(var_10, var_11, 1, 1, 1, 0); 
+  __visc__bindIn(var_11, 16, 2, 0); 
+  __visc__bindIn(var_11, 17, 3, 0); 
+
+  void* var_12 = __visc__createNodeND(0, var_12_node); 
+
+  __visc__edge(var_11, var_12, 1, 0, 0, 0); 
+  __visc__edge(var_11, var_12, 1, 1, 1, 0); 
+
+  void* var_13 = __visc__createNodeND(0, var_13_node); 
+
+  __visc__edge(var_12, var_13, 1, 0, 0, 0); 
+  __visc__edge(var_12, var_13, 1, 1, 1, 0); 
+
+  void* var_14 = __visc__createNodeND(0, var_14_node); 
+
+  __visc__edge(var_13, var_14, 1, 0, 0, 0); 
+  __visc__edge(var_13, var_14, 1, 1, 1, 0); 
+  __visc__bindIn(var_14, 18, 2, 0); 
+  __visc__bindIn(var_14, 19, 3, 0); 
+
+  void* var_15 = __visc__createNodeND(0, var_15_node); 
+
+  __visc__edge(var_14, var_15, 1, 0, 0, 0); 
+  __visc__edge(var_14, var_15, 1, 1, 1, 0); 
+  __visc__bindIn(var_15, 20, 2, 0); 
+  __visc__bindIn(var_15, 21, 3, 0); 
+
+  void* var_16 = __visc__createNodeND(0, var_16_node); 
+
+  __visc__edge(var_15, var_16, 1, 0, 0, 0); 
+  __visc__edge(var_15, var_16, 1, 1, 1, 0); 
+
+  void* var_17 = __visc__createNodeND(0, var_17_node); 
+
+  __visc__edge(var_16, var_17, 1, 0, 0, 0); 
+  __visc__edge(var_16, var_17, 1, 1, 1, 0); 
+  __visc__bindIn(var_17, 22, 2, 0); 
+  __visc__bindIn(var_17, 23, 3, 0); 
+
+  void* var_18 = __visc__createNodeND(0, var_18_node); 
+
+  __visc__edge(var_17, var_18, 1, 0, 0, 0); 
+  __visc__edge(var_17, var_18, 1, 1, 1, 0); 
+  __visc__bindIn(var_18, 24, 2, 0); 
+  __visc__bindIn(var_18, 25, 3, 0); 
+
+  void* var_19 = __visc__createNodeND(0, var_19_node); 
+
+  __visc__edge(var_18, var_19, 1, 0, 0, 0); 
+  __visc__edge(var_18, var_19, 1, 1, 1, 0); 
+
+  void* var_20 = __visc__createNodeND(0, var_20_node); 
+
+  __visc__edge(var_19, var_20, 1, 0, 0, 0); 
+  __visc__edge(var_19, var_20, 1, 1, 1, 0); 
+  __visc__bindIn(var_20, 26, 2, 0); 
+  __visc__bindIn(var_20, 27, 3, 0); 
+
+  void* var_21 = __visc__createNodeND(0, var_21_node); 
+
+  __visc__edge(var_20, var_21, 1, 0, 0, 0); 
+  __visc__edge(var_20, var_21, 1, 1, 1, 0); 
+  __visc__bindIn(var_21, 28, 2, 0); 
+  __visc__bindIn(var_21, 29, 3, 0); 
+
+  void* var_22 = __visc__createNodeND(0, var_22_node); 
+
+  __visc__edge(var_21, var_22, 1, 0, 0, 0); 
+  __visc__edge(var_21, var_22, 1, 1, 1, 0); 
+
+  void* var_23 = __visc__createNodeND(0, var_23_node); 
+
+  __visc__edge(var_22, var_23, 1, 0, 0, 0); 
+  __visc__edge(var_22, var_23, 1, 1, 1, 0); 
+
+  void* var_24 = __visc__createNodeND(0, var_24_node); 
+
+  __visc__edge(var_23, var_24, 1, 0, 0, 0); 
+  __visc__edge(var_23, var_24, 1, 1, 1, 0); 
+  __visc__bindIn(var_24, 30, 2, 0); 
+  __visc__bindIn(var_24, 31, 3, 0); 
+
+  void* var_25 = __visc__createNodeND(0, var_25_node); 
+
+  __visc__edge(var_24, var_25, 1, 0, 0, 0); 
+  __visc__edge(var_24, var_25, 1, 1, 1, 0); 
+  __visc__bindIn(var_25, 32, 2, 0); 
+  __visc__bindIn(var_25, 33, 3, 0); 
+
+  void* var_26 = __visc__createNodeND(0, var_26_node); 
+
+  __visc__edge(var_25, var_26, 1, 0, 0, 0); 
+  __visc__edge(var_25, var_26, 1, 1, 1, 0); 
+
+  void* var_27 = __visc__createNodeND(0, var_27_node); 
+
+  __visc__edge(var_26, var_27, 1, 0, 0, 0); 
+  __visc__edge(var_26, var_27, 1, 1, 1, 0); 
+  __visc__bindIn(var_27, 34, 2, 0); 
+  __visc__bindIn(var_27, 35, 3, 0); 
+
+  void* var_28 = __visc__createNodeND(0, var_28_node); 
+
+  __visc__edge(var_27, var_28, 1, 0, 0, 0); 
+  __visc__edge(var_27, var_28, 1, 1, 1, 0); 
+  __visc__bindIn(var_28, 36, 2, 0); 
+  __visc__bindIn(var_28, 37, 3, 0); 
+
+  void* var_29 = __visc__createNodeND(0, var_29_node); 
+
+  __visc__edge(var_28, var_29, 1, 0, 0, 0); 
+  __visc__edge(var_28, var_29, 1, 1, 1, 0); 
+
+  void* var_30 = __visc__createNodeND(0, var_30_node); 
+
+  __visc__edge(var_29, var_30, 1, 0, 0, 0); 
+  __visc__edge(var_29, var_30, 1, 1, 1, 0); 
+  __visc__bindIn(var_30, 38, 2, 0); 
+  __visc__bindIn(var_30, 39, 3, 0); 
+
+  void* var_31 = __visc__createNodeND(0, var_31_node); 
+
+  __visc__edge(var_30, var_31, 1, 0, 0, 0); 
+  __visc__edge(var_30, var_31, 1, 1, 1, 0); 
+  __visc__bindIn(var_31, 40, 2, 0); 
+  __visc__bindIn(var_31, 41, 3, 0); 
+
+  void* var_32 = __visc__createNodeND(0, var_32_node); 
+
+  __visc__edge(var_31, var_32, 1, 0, 0, 0); 
+  __visc__edge(var_31, var_32, 1, 1, 1, 0); 
+
+  void* var_33 = __visc__createNodeND(0, var_33_node); 
+
+  __visc__edge(var_32, var_33, 1, 0, 0, 0); 
+  __visc__edge(var_32, var_33, 1, 1, 1, 0); 
+
+  void* var_34 = __visc__createNodeND(0, var_34_node); 
+
+  __visc__edge(var_33, var_34, 1, 0, 0, 0); 
+  __visc__edge(var_33, var_34, 1, 1, 1, 0); 
+  __visc__bindIn(var_34, 42, 2, 0); 
+  __visc__bindIn(var_34, 43, 3, 0); 
+
+  void* var_35 = __visc__createNodeND(0, var_35_node); 
+
+  __visc__edge(var_34, var_35, 1, 0, 0, 0); 
+  __visc__edge(var_34, var_35, 1, 1, 1, 0); 
+  __visc__bindIn(var_35, 44, 2, 0); 
+  __visc__bindIn(var_35, 45, 3, 0); 
+
+  void* var_36 = __visc__createNodeND(0, var_36_node); 
+
+  __visc__edge(var_35, var_36, 1, 0, 0, 0); 
+  __visc__edge(var_35, var_36, 1, 1, 1, 0); 
+
+  void* var_37 = __visc__createNodeND(0, var_37_node); 
+
+  __visc__edge(var_36, var_37, 1, 0, 0, 0); 
+  __visc__edge(var_36, var_37, 1, 1, 1, 0); 
+  __visc__bindIn(var_37, 46, 2, 0); 
+  __visc__bindIn(var_37, 47, 3, 0); 
+
+  void* var_38 = __visc__createNodeND(0, var_38_node); 
+
+  __visc__edge(var_37, var_38, 1, 0, 0, 0); 
+  __visc__edge(var_37, var_38, 1, 1, 1, 0); 
+  __visc__bindIn(var_38, 48, 2, 0); 
+  __visc__bindIn(var_38, 49, 3, 0); 
+
+  void* var_39 = __visc__createNodeND(0, var_39_node); 
+
+  __visc__edge(var_38, var_39, 1, 0, 0, 0); 
+  __visc__edge(var_38, var_39, 1, 1, 1, 0); 
+
+  void* var_40 = __visc__createNodeND(0, var_40_node); 
+
+  __visc__edge(var_39, var_40, 1, 0, 0, 0); 
+  __visc__edge(var_39, var_40, 1, 1, 1, 0); 
+  __visc__bindIn(var_40, 50, 2, 0); 
+  __visc__bindIn(var_40, 51, 3, 0); 
+
+  void* var_41 = __visc__createNodeND(0, var_41_node); 
+
+  __visc__edge(var_40, var_41, 1, 0, 0, 0); 
+  __visc__edge(var_40, var_41, 1, 1, 1, 0); 
+  __visc__bindIn(var_41, 52, 2, 0); 
+  __visc__bindIn(var_41, 53, 3, 0); 
+
+  void* var_42 = __visc__createNodeND(0, var_42_node); 
+
+  __visc__edge(var_41, var_42, 1, 0, 0, 0); 
+  __visc__edge(var_41, var_42, 1, 1, 1, 0); 
+
+  void* var_43 = __visc__createNodeND(0, var_43_node); 
+
+  __visc__edge(var_42, var_43, 1, 0, 0, 0); 
+  __visc__edge(var_42, var_43, 1, 1, 1, 0); 
+
+  void* var_44 = __visc__createNodeND(0, var_44_node); 
+
+  __visc__edge(var_43, var_44, 1, 0, 0, 0); 
+  __visc__edge(var_43, var_44, 1, 1, 1, 0); 
+  __visc__bindIn(var_44, 54, 2, 0); 
+  __visc__bindIn(var_44, 55, 3, 0); 
+
+  void* var_45 = __visc__createNodeND(0, var_45_node); 
+
+  __visc__edge(var_44, var_45, 1, 0, 0, 0); 
+  __visc__edge(var_44, var_45, 1, 1, 1, 0); 
+  __visc__bindIn(var_45, 56, 2, 0); 
+  __visc__bindIn(var_45, 57, 3, 0); 
+
+  void* var_46 = __visc__createNodeND(0, var_46_node); 
+
+  __visc__edge(var_45, var_46, 1, 0, 0, 0); 
+  __visc__edge(var_45, var_46, 1, 1, 1, 0); 
+
+  void* var_47 = __visc__createNodeND(0, var_47_node); 
+
+  __visc__edge(var_46, var_47, 1, 0, 0, 0); 
+  __visc__edge(var_46, var_47, 1, 1, 1, 0); 
+  __visc__bindIn(var_47, 58, 2, 0); 
+  __visc__bindIn(var_47, 59, 3, 0); 
+
+  void* var_48 = __visc__createNodeND(0, var_48_node); 
+
+  __visc__edge(var_47, var_48, 1, 0, 0, 0); 
+  __visc__edge(var_47, var_48, 1, 1, 1, 0); 
+  __visc__bindIn(var_48, 60, 2, 0); 
+  __visc__bindIn(var_48, 61, 3, 0); 
+
+  void* var_49 = __visc__createNodeND(0, var_49_node); 
+
+  __visc__edge(var_48, var_49, 1, 0, 0, 0); 
+  __visc__edge(var_48, var_49, 1, 1, 1, 0); 
+
+  void* var_50 = __visc__createNodeND(0, var_50_node); 
+
+  __visc__edge(var_49, var_50, 1, 0, 0, 0); 
+  __visc__edge(var_49, var_50, 1, 1, 1, 0); 
+  __visc__bindIn(var_50, 62, 2, 0); 
+  __visc__bindIn(var_50, 63, 3, 0); 
+
+  void* var_51 = __visc__createNodeND(0, var_51_node); 
+
+  __visc__edge(var_50, var_51, 1, 0, 0, 0); 
+  __visc__edge(var_50, var_51, 1, 1, 1, 0); 
+  __visc__bindIn(var_51, 64, 2, 0); 
+  __visc__bindIn(var_51, 65, 3, 0); 
+
+  void* var_52 = __visc__createNodeND(0, var_52_node); 
+
+  __visc__edge(var_51, var_52, 1, 0, 0, 0); 
+  __visc__edge(var_51, var_52, 1, 1, 1, 0); 
+
+  __visc__bindOut(var_52, 0, 0, 0); 
+  __visc__bindOut(var_52, 1, 1, 0); 
+
+}
+
+struct ret_t {
+  void* tensor; 
+  size_t bytes; 
+}; 
+
+typedef struct __attribute__((__packed__)) {
+  void* input; 
+  size_t input_bytes; 
+  void* conv2d_1_w; 
+  size_t conv2d_1_w_bytes; 
+  void* conv2d_1_b; 
+  size_t conv2d_1_b_bytes; 
+  void* conv2d_2_w; 
+  size_t conv2d_2_w_bytes; 
+  void* conv2d_2_b; 
+  size_t conv2d_2_b_bytes; 
+  void* conv2d_3_w; 
+  size_t conv2d_3_w_bytes; 
+  void* conv2d_3_b; 
+  size_t conv2d_3_b_bytes; 
+  void* conv2d_4_w; 
+  size_t conv2d_4_w_bytes; 
+  void* conv2d_4_b; 
+  size_t conv2d_4_b_bytes; 
+  void* conv2d_5_w; 
+  size_t conv2d_5_w_bytes; 
+  void* conv2d_5_b; 
+  size_t conv2d_5_b_bytes; 
+  void* conv2d_6_w; 
+  size_t conv2d_6_w_bytes; 
+  void* conv2d_6_b; 
+  size_t conv2d_6_b_bytes; 
+  void* conv2d_7_w; 
+  size_t conv2d_7_w_bytes; 
+  void* conv2d_7_b; 
+  size_t conv2d_7_b_bytes; 
+  void* conv2d_8_w; 
+  size_t conv2d_8_w_bytes; 
+  void* conv2d_8_b; 
+  size_t conv2d_8_b_bytes; 
+  void* conv2d_9_w; 
+  size_t conv2d_9_w_bytes; 
+  void* conv2d_9_b; 
+  size_t conv2d_9_b_bytes; 
+  void* conv2d_10_w; 
+  size_t conv2d_10_w_bytes; 
+  void* conv2d_10_b; 
+  size_t conv2d_10_b_bytes; 
+  void* conv2d_11_w; 
+  size_t conv2d_11_w_bytes; 
+  void* conv2d_11_b; 
+  size_t conv2d_11_b_bytes; 
+  void* conv2d_12_w; 
+  size_t conv2d_12_w_bytes; 
+  void* conv2d_12_b; 
+  size_t conv2d_12_b_bytes; 
+  void* conv2d_13_w; 
+  size_t conv2d_13_w_bytes; 
+  void* conv2d_13_b; 
+  size_t conv2d_13_b_bytes; 
+  void* dense_1_w; 
+  size_t dense_1_w_bytes; 
+  void* dense_1_b; 
+  size_t dense_1_b_bytes; 
+  void* dense_2_w; 
+  size_t dense_2_w_bytes; 
+  void* dense_2_b; 
+  size_t dense_2_b_bytes; 
+  void* dense_3_w; 
+  size_t dense_3_w_bytes; 
+  void* dense_3_b; 
+  size_t dense_3_b_bytes; 
+
+  struct ret_t r; 
+}
+RootIn;
+
+
+int main(){ 
+
+  std::string dir_prefix = std::string("/home/hsharif3/vgg16_imagenet/"); 
+  std::string input_path =  dir_prefix + std::string("input.bin"); 
+  std::string labels_path =  dir_prefix + std::string("labels.bin"); 
+  std::string conv2d_1_w_path =  dir_prefix + std::string("conv2d_1_w.bin"); 
+  void* conv2d_1_w =  readTrainedWeights(conv2d_1_w_path.c_str(), 0,64,3,3,3); 
+  std::string conv2d_1_b_path =  dir_prefix + std::string("conv2d_1_b.bin"); 
+  void* conv2d_1_b =  readTrainedWeights(conv2d_1_b_path.c_str(), 0,1,64,1,1); 
+  std::string conv2d_2_w_path =  dir_prefix + std::string("conv2d_2_w.bin"); 
+  void* conv2d_2_w =  readTrainedWeights(conv2d_2_w_path.c_str(), 0,64,64,3,3); 
+  std::string conv2d_2_b_path =  dir_prefix + std::string("conv2d_2_b.bin"); 
+  void* conv2d_2_b =  readTrainedWeights(conv2d_2_b_path.c_str(), 0,1,64,1,1); 
+  std::string conv2d_3_w_path =  dir_prefix + std::string("conv2d_3_w.bin"); 
+  void* conv2d_3_w =  readTrainedWeights(conv2d_3_w_path.c_str(), 0,128,64,3,3); 
+  std::string conv2d_3_b_path =  dir_prefix + std::string("conv2d_3_b.bin"); 
+  void* conv2d_3_b =  readTrainedWeights(conv2d_3_b_path.c_str(), 0,1,128,1,1); 
+  std::string conv2d_4_w_path =  dir_prefix + std::string("conv2d_4_w.bin"); 
+  void* conv2d_4_w =  readTrainedWeights(conv2d_4_w_path.c_str(), 0,128,128,3,3); 
+  std::string conv2d_4_b_path =  dir_prefix + std::string("conv2d_4_b.bin"); 
+  void* conv2d_4_b =  readTrainedWeights(conv2d_4_b_path.c_str(), 0,1,128,1,1); 
+  std::string conv2d_5_w_path =  dir_prefix + std::string("conv2d_5_w.bin"); 
+  void* conv2d_5_w =  readTrainedWeights(conv2d_5_w_path.c_str(), 0,256,128,3,3); 
+  std::string conv2d_5_b_path =  dir_prefix + std::string("conv2d_5_b.bin"); 
+  void* conv2d_5_b =  readTrainedWeights(conv2d_5_b_path.c_str(), 0,1,256,1,1); 
+  std::string conv2d_6_w_path =  dir_prefix + std::string("conv2d_6_w.bin"); 
+  void* conv2d_6_w =  readTrainedWeights(conv2d_6_w_path.c_str(), 0,256,256,3,3); 
+  std::string conv2d_6_b_path =  dir_prefix + std::string("conv2d_6_b.bin"); 
+  void* conv2d_6_b =  readTrainedWeights(conv2d_6_b_path.c_str(), 0,1,256,1,1); 
+  std::string conv2d_7_w_path =  dir_prefix + std::string("conv2d_7_w.bin"); 
+  void* conv2d_7_w =  readTrainedWeights(conv2d_7_w_path.c_str(), 0,256,256,3,3); 
+  std::string conv2d_7_b_path =  dir_prefix + std::string("conv2d_7_b.bin"); 
+  void* conv2d_7_b =  readTrainedWeights(conv2d_7_b_path.c_str(), 0,1,256,1,1); 
+  std::string conv2d_8_w_path =  dir_prefix + std::string("conv2d_8_w.bin"); 
+  void* conv2d_8_w =  readTrainedWeights(conv2d_8_w_path.c_str(), 0,512,256,3,3); 
+  std::string conv2d_8_b_path =  dir_prefix + std::string("conv2d_8_b.bin"); 
+  void* conv2d_8_b =  readTrainedWeights(conv2d_8_b_path.c_str(), 0,1,512,1,1); 
+  std::string conv2d_9_w_path =  dir_prefix + std::string("conv2d_9_w.bin"); 
+  void* conv2d_9_w =  readTrainedWeights(conv2d_9_w_path.c_str(), 0,512,512,3,3); 
+  std::string conv2d_9_b_path =  dir_prefix + std::string("conv2d_9_b.bin"); 
+  void* conv2d_9_b =  readTrainedWeights(conv2d_9_b_path.c_str(), 0,1,512,1,1); 
+  std::string conv2d_10_w_path =  dir_prefix + std::string("conv2d_10_w.bin"); 
+  void* conv2d_10_w =  readTrainedWeights(conv2d_10_w_path.c_str(), 0,512,512,3,3); 
+  std::string conv2d_10_b_path =  dir_prefix + std::string("conv2d_10_b.bin"); 
+  void* conv2d_10_b =  readTrainedWeights(conv2d_10_b_path.c_str(), 0,1,512,1,1); 
+  std::string conv2d_11_w_path =  dir_prefix + std::string("conv2d_11_w.bin"); 
+  void* conv2d_11_w =  readTrainedWeights(conv2d_11_w_path.c_str(), 0,512,512,3,3); 
+  std::string conv2d_11_b_path =  dir_prefix + std::string("conv2d_11_b.bin"); 
+  void* conv2d_11_b =  readTrainedWeights(conv2d_11_b_path.c_str(), 0,1,512,1,1); 
+  std::string conv2d_12_w_path =  dir_prefix + std::string("conv2d_12_w.bin"); 
+  void* conv2d_12_w =  readTrainedWeights(conv2d_12_w_path.c_str(), 0,512,512,3,3); 
+  std::string conv2d_12_b_path =  dir_prefix + std::string("conv2d_12_b.bin"); 
+  void* conv2d_12_b =  readTrainedWeights(conv2d_12_b_path.c_str(), 0,1,512,1,1); 
+  std::string conv2d_13_w_path =  dir_prefix + std::string("conv2d_13_w.bin"); 
+  void* conv2d_13_w =  readTrainedWeights(conv2d_13_w_path.c_str(), 0,512,512,3,3); 
+  std::string conv2d_13_b_path =  dir_prefix + std::string("conv2d_13_b.bin"); 
+  void* conv2d_13_b =  readTrainedWeights(conv2d_13_b_path.c_str(), 0,1,512,1,1); 
+  std::string dense_1_w_path =  dir_prefix + std::string("dense_1_w.bin"); 
+  void* dense_1_w =  readTrainedWeights(dense_1_w_path.c_str(), 0,1,1,25088,4096); 
+  std::string dense_1_b_path =  dir_prefix + std::string("dense_1_b.bin"); 
+  void* dense_1_b =  readTrainedWeights(dense_1_b_path.c_str(), 0,1,4096,1,1); 
+  std::string dense_2_w_path =  dir_prefix + std::string("dense_2_w.bin"); 
+  void* dense_2_w =  readTrainedWeights(dense_2_w_path.c_str(), 0,1,1,4096,4096); 
+  std::string dense_2_b_path =  dir_prefix + std::string("dense_2_b.bin"); 
+  void* dense_2_b =  readTrainedWeights(dense_2_b_path.c_str(), 0,1,4096,1,1); 
+  std::string dense_3_w_path =  dir_prefix + std::string("dense_3_w.bin"); 
+  void* dense_3_w =  readTrainedWeights(dense_3_w_path.c_str(), 0,1,1,4096,1000); 
+  std::string dense_3_b_path =  dir_prefix + std::string("dense_3_b.bin"); 
+  void* dense_3_b =  readTrainedWeights(dense_3_b_path.c_str(), 0,1,1000,1,1); 
+  void* input = readTrainedWeights(input_path.c_str(), 0,100,3,224,224);
+  uint32_t* labels = readLabelsBatch3(labels_path.c_str(),0,100); 
+
+  
+  __visc__init(); 
+  RootIn* args = static_cast<RootIn*>(malloc(sizeof(RootIn))); 
+
+  args->input = input; 
+  args->input_bytes = 0; 
+  args->conv2d_1_w = conv2d_1_w; 
+  args->conv2d_1_w_bytes = 0; 
+  args->conv2d_1_b = conv2d_1_b; 
+  args->conv2d_1_b_bytes = 0; 
+  args->conv2d_2_w = conv2d_2_w; 
+  args->conv2d_2_w_bytes = 0; 
+  args->conv2d_2_b = conv2d_2_b; 
+  args->conv2d_2_b_bytes = 0; 
+  args->conv2d_3_w = conv2d_3_w; 
+  args->conv2d_3_w_bytes = 0; 
+  args->conv2d_3_b = conv2d_3_b; 
+  args->conv2d_3_b_bytes = 0; 
+  args->conv2d_4_w = conv2d_4_w; 
+  args->conv2d_4_w_bytes = 0; 
+  args->conv2d_4_b = conv2d_4_b; 
+  args->conv2d_4_b_bytes = 0; 
+  args->conv2d_5_w = conv2d_5_w; 
+  args->conv2d_5_w_bytes = 0; 
+  args->conv2d_5_b = conv2d_5_b; 
+  args->conv2d_5_b_bytes = 0; 
+  args->conv2d_6_w = conv2d_6_w; 
+  args->conv2d_6_w_bytes = 0; 
+  args->conv2d_6_b = conv2d_6_b; 
+  args->conv2d_6_b_bytes = 0; 
+  args->conv2d_7_w = conv2d_7_w; 
+  args->conv2d_7_w_bytes = 0; 
+  args->conv2d_7_b = conv2d_7_b; 
+  args->conv2d_7_b_bytes = 0; 
+  args->conv2d_8_w = conv2d_8_w; 
+  args->conv2d_8_w_bytes = 0; 
+  args->conv2d_8_b = conv2d_8_b; 
+  args->conv2d_8_b_bytes = 0; 
+  args->conv2d_9_w = conv2d_9_w; 
+  args->conv2d_9_w_bytes = 0; 
+  args->conv2d_9_b = conv2d_9_b; 
+  args->conv2d_9_b_bytes = 0; 
+  args->conv2d_10_w = conv2d_10_w; 
+  args->conv2d_10_w_bytes = 0; 
+  args->conv2d_10_b = conv2d_10_b; 
+  args->conv2d_10_b_bytes = 0; 
+  args->conv2d_11_w = conv2d_11_w; 
+  args->conv2d_11_w_bytes = 0; 
+  args->conv2d_11_b = conv2d_11_b; 
+  args->conv2d_11_b_bytes = 0; 
+  args->conv2d_12_w = conv2d_12_w; 
+  args->conv2d_12_w_bytes = 0; 
+  args->conv2d_12_b = conv2d_12_b; 
+  args->conv2d_12_b_bytes = 0; 
+  args->conv2d_13_w = conv2d_13_w; 
+  args->conv2d_13_w_bytes = 0; 
+  args->conv2d_13_b = conv2d_13_b; 
+  args->conv2d_13_b_bytes = 0; 
+  args->dense_1_w = dense_1_w; 
+  args->dense_1_w_bytes = 0; 
+  args->dense_1_b = dense_1_b; 
+  args->dense_1_b_bytes = 0; 
+  args->dense_2_w = dense_2_w; 
+  args->dense_2_w_bytes = 0; 
+  args->dense_2_b = dense_2_b; 
+  args->dense_2_b_bytes = 0; 
+  args->dense_3_w = dense_3_w; 
+  args->dense_3_w_bytes = 0; 
+  args->dense_3_b = dense_3_b; 
+  args->dense_3_b_bytes = 0; 
+
+  void* dfg = __visc__launch(0, root, (void*) args); 
+
+  __visc__wait(dfg); 
+
+  void *result = static_cast<RootIn*>(args)->input; 
+  hpvm_request_tensor(result, 0); 
+
+  __visc__cleanup(); 
+  computeAccuracy3(labels, result); 
+  return 0; 
+
+}