Adding AlexNet v2 Benchmark

llvm/test/VISC/DNN_Benchmarks/benchmarks/alexnet2/Makefile

+DNN_BENCHMARK_ROOT = $(LLVM_SRC_ROOT)/test/VISC/DNN_Benchmarks
+# NOTE: can configure build directory
+HPVM_BUILD_DIR = $(LLVM_SRC_ROOT)/../build_hpvm/
+
+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 = alexnet2
+
+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
+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 -lOpenCL
+
+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
+
+
+QUANT_FILE_PATH=/home/hsharif3/Gitlab/hpvm/llvm/test/VISC/DNN_Benchmarks/benchmarks/alexnet2/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  $(HPVM_LIB_DIR)/LLVMClearDFG.so   -inplace -hpvm-fuse -dfg2llvm-promise  -quantization-levels-filename=$(QUANT_FILE_PATH) -dfg2llvm-cudnn  -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
+
+
+$(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) $(VISC_OPTFLAGS)  $(BUILD_DIR)/$(APP).visc.ll  -o  $(BUILD_DIR)/$(APP)_cudnn.bc
+	$(OPT) $(VISC_OPTFLAGS2) $(BUILD_DIR)/$(APP)_promise.visc.ll  -o  $(BUILD_DIR)/$(APP)_promise.bc
+	$(LLVM_LINK) $(BUILD_DIR)/$(APP)_cudnn.bc $(VISC_RT_PATH) -o $(BUILD_DIR)/$(APP)_cudnn_linked.bc
+	$(LLVM_LINK) $(BUILD_DIR)/$(APP)_promise.bc $(VISC_RT_PATH) -o $(BUILD_DIR)/$(APP)_promise_linked.bc
+	$(CC) $(BUILD_DIR)/$(APP)_cudnn_linked.bc $(TENSOR_LIB_DIR) -o $(BUILD_DIR)/$(APP)_cudnn_linked $(LINKER_FLAGS)
+	$(CC) $(BUILD_DIR)/$(APP)_promise_linked.bc $(TENSOR_LIB_DIR) -o $(BUILD_DIR)/$(APP)_promise_linked $(LINKER_FLAGS)
+	#$(CC) $(BUILD_DIR)/$(APP)_cudnn_linked.bc $(TENSOR_AUTOTUNER_DIR) -o $(BUILD_DIR)/lenet_tune $(LINKER_FLAGS)
+
+$(BUILD_DIR):
+	mkdir -p $@
+
+clean:
+	rm -rf $(BUILD_DIR)

llvm/test/VISC/DNN_Benchmarks/benchmarks/alexnet2/data/quant_ranges.txt

+-1.8816435 2.0934134 -0.5421946 0.3710851 -0.06697306 0.040868897 -0.775027394891 0.779944300652 
+-0.775027394891 0.779944300652 -0.42474225 0.31460348 -0.3557253 -0.17281663 -0.808667064309 0.983953297734 
+-0.808667064309 0.983953297734 -0.44134507 0.79587924 -0.80424446 0.75330096 -0.995678424835 0.998566448689 
+-0.995678424835 0.998566448689 -0.2883836 0.31025785 -0.6353164 0.29015934 -0.993219196796 0.992379009724 
+-0.993219196796 0.992379009724 -0.2792431 0.37689754 -1.1379756 1.2391574 -0.999901354313 0.999910891056 
+-0.999901354313 0.999910891056 -0.27078503 0.27942517 -0.503003 0.12762362 -0.991036117375 0.971404970288 
+-0.991036117375 0.971404970288 -0.24273404 0.5845544 -0.53745 0.558251 -119.27973732 -25.2262819576
+0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

llvm/test/VISC/DNN_Benchmarks/benchmarks/alexnet2/src/alexnet2.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_tanh(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_tanh(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_tanh(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_tanh(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_tanh(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_tanh(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_20_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_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_mul(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_22_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_23_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* dense_1_w, size_t dense_1_w_bytes, 
+	  void* dense_1_b, size_t dense_1_b_bytes){ 
+
+
+  __visc__hint(visc::CPU_TARGET); 
+  __visc__attributes(15, 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, dense_1_w, dense_1_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); 
+
+  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, 26, 2, 0); 
+  __visc__bindIn(var_21, 27, 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); 
+  __visc__bindIn(var_22, 28, 2, 0); 
+  __visc__bindIn(var_22, 29, 3, 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); 
+
+  __visc__bindOut(var_23, 0, 0, 0); 
+  __visc__bindOut(var_23, 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* dense_1_w; 
+  size_t dense_1_w_bytes; 
+  void* dense_1_b; 
+  size_t dense_1_b_bytes; 
+
+  struct ret_t r; 
+}
+RootIn;
+
+int main(){ 
+
+  std::string dir_prefix = std::string("../../../../../../projects/hpvm-tensor-rt/model_params/alexnet2_cifar10_test/");
+  
+  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(), 0,32,3,3,3); 
+  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(), 0,1,32,1,1); 
+  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(), 0,32,32,3,3); 
+  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(), 0,1,32,1,1); 
+  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(), 0,64,32,3,3); 
+  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(), 0,1,64,1,1); 
+  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(), 0,64,64,3,3); 
+  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(), 0,1,64,1,1); 
+  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(), 0,128,64,3,3); 
+  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(), 0,1,128,1,1); 
+  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(), 0,128,128,3,3); 
+  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(), 0,1,128,1,1); 
+  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(), 0,1,1,2048,10); 
+  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(), 0,1,10,1,1); 
+  void* input = readTrainedWeights(input_path.c_str(), 0,10000,3,32,32); 
+  uint8_t* labels = readLabels(labels_path.c_str(),10000); 
+
+  __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->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; 
+
+  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(); 
+  computeAccuracy2(labels, 10000, result); 
+  return 0; 
+
+} 

llvm/test/VISC/DNN_Benchmarks/benchmarks/alexnet2/src/alexnet2_promise.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::PROMISE_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::PROMISE_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::PROMISE_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_tanh(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::PROMISE_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::PROMISE_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::PROMISE_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_tanh(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_6_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::PROMISE_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::PROMISE_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::PROMISE_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::PROMISE_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_tanh(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::PROMISE_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::PROMISE_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::PROMISE_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_tanh(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_13_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::PROMISE_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::PROMISE_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::PROMISE_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::PROMISE_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_tanh(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::PROMISE_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::PROMISE_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::PROMISE_TARGET); 
+  __visc__attributes(1, t1, 0); 
+
+  void* r = __visc__tensor_tanh(t1); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_20_node(void* t1, size_t bytes_t1) { 
+  __visc__hint(visc::PROMISE_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_21_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::PROMISE_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_mul(t1, t2); 
+  __visc__return(2, r, (size_t) 0); 
+}
+
+void var_22_node(void* t1, size_t bytes_t1, void* t2, size_t bytes_t2) { 
+  __visc__hint(visc::PROMISE_TARGET); 
+  __visc__attributes(2, t1, t2, 0); 
+
+  void *r = __visc__tensor_add(t1, t2); 
+  __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_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* dense_1_w, size_t dense_1_w_bytes, 
+	  void* dense_1_b, size_t dense_1_b_bytes){ 
+
+
+  __visc__hint(visc::CPU_TARGET); 
+  __visc__attributes(15, 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, dense_1_w, dense_1_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); 
+
+  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, 26, 2, 0); 
+  __visc__bindIn(var_21, 27, 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); 
+  __visc__bindIn(var_22, 28, 2, 0); 
+  __visc__bindIn(var_22, 29, 3, 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); 
+
+  __visc__bindOut(var_23, 0, 0, 0); 
+  __visc__bindOut(var_23, 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* dense_1_w; 
+  size_t dense_1_w_bytes; 
+  void* dense_1_b; 
+  size_t dense_1_b_bytes; 
+
+  struct ret_t r; 
+}
+RootIn;
+
+int main(){ 
+
+  std::string dir_prefix = std::string("../../../../../../projects/hpvm-tensor-rt/model_params/alexnet2_cifar10_test/");
+  
+ 
+  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(), 0,32,3,3,3); 
+  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(), 0,1,32,1,1); 
+  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(), 0,32,32,3,3); 
+  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(), 0,1,32,1,1); 
+  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(), 0,64,32,3,3); 
+  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(), 0,1,64,1,1); 
+  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(), 0,64,64,3,3); 
+  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(), 0,1,64,1,1); 
+  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(), 0,128,64,3,3); 
+  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(), 0,1,128,1,1); 
+  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(), 0,128,128,3,3); 
+  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(), 0,1,128,1,1); 
+  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(), 0,1,1,2048,10); 
+  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(), 0,1,10,1,1); 
+  void* input = readTrainedWeights(input_path.c_str(), 0,10000,3,32,32); 
+  uint8_t* labels = readLabels(labels_path.c_str(),10000); 
+
+  __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->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; 
+
+  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(); 
+  computeAccuracy2(labels, 10000, result); 
+  return 0; 
+
+}