Merge pull request #40 from CortexFoundation/wlt

Update features

Author: wlt-cortex

.gitignore

@@ -34,3 +34,14 @@ Thumbs.db
 # lib
 *.a
 *.so
+python/cvm/_cython/libcvm.c
+
+compile_commands.json
+.ycm_extra_conf.py
+
+# binary
+tests/test_model*
+tests/test_op*
+tests/ipy
+
+config.cmake

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "z3/3rdparty/z3"]
+	path = z3/3rdparty/z3
+	url = [email protected]:Z3Prover/z3.git

CMakeLists.txt

@@ -1,123 +1,130 @@
 cmake_minimum_required(VERSION 3.2)
 project(cvm C CXX)
 
-set( CMAKE_EXPORT_COMPILE_COMMANDS ON )
+set(CMAKE_EXPORT_COMPILE_COMMANDS ON )
+
 # Utility functions
 include(cmake/util/Util.cmake)
 include(cmake/util/FindCUDA.cmake)
 include(cmake/util/FindOpenmp.cmake)
 
-if(EXISTS ${CMAKE_CURRENT_BINARY_DIR}/config.cmake)
-  include(${CMAKE_CURRENT_BINARY_DIR}/config.cmake)
-else()
-  if(EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/config.cmake)
-    include(${CMAKE_CURRENT_SOURCE_DIR}/config.cmake)
-  endif()
+if(EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/config.cmake)
+  include(${CMAKE_CURRENT_SOURCE_DIR}/config.cmake)
 endif()
 
-# NOTE: do not modify this file to change option values.
-# You can create a config.cmake at build folder
-# and add set(OPTION VALUE) to override these build options.
-# Alernatively, use cmake -DOPTION=VALUE through command-line.
-cvm_option(USE_CUDA "Build with CUDA" ON)
-cvm_option(USE_LLVM "Build with LLVM, can be set to specific llvm-config path" ON)
-cvm_option(USE_CVM_RUNTIME "Build with tiny cvm runtime" ON)
-cvm_option(USE_RTTI "Build with RTTI" ON)
-cvm_option(INSTALL_DEV "Install compiler infrastructure" OFF)
-
 # include directories
 include_directories("include")
 
-# initial variables
-set(CVM_LINKER_LIBS "")
-set(CVM_RUNTIME_LINKER_LIBS "")
-set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
-
 # Generic compilation options
-if(MSVC)
-  add_definitions(-DWIN32_LEAN_AND_MEAN)
-  add_definitions(-D_CRT_SECURE_NO_WARNINGS)
-  add_definitions(-D_SCL_SECURE_NO_WARNINGS)
-  add_definitions(-D_ENABLE_EXTENDED_ALIGNED_STORAGE)
-  add_definitions(-DHalide_SHARED)
-  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /EHsc")
-  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /MP")
-  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /bigobj")
-  if(USE_MSVC_MT)
-    foreach(flag_var
-        CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_DEBUG CMAKE_CXX_FLAGS_RELEASE
-        CMAKE_CXX_FLAGS_MINSIZEREL CMAKE_CXX_FLAGS_RELWITHDEBINFO)
-      if(${flag_var} MATCHES "/MD")
-        string(REGEX REPLACE "/MD" "/MT" ${flag_var} "${${flag_var}}")
-      endif(${flag_var} MATCHES "/MD")
-    endforeach(flag_var)
-  endif()
-else(MSVC)
-  include(CheckCXXCompilerFlag)
-  check_cxx_compiler_flag("-std=c++11"    SUPPORT_CXX11)
-  if ("${CMAKE_BUILD_TYPE}" STREQUAL "Debug")
-    set(CMAKE_C_FLAGS "-mavx2 -Og -g -Wall -fPIC ${CMAKE_C_FLAGS}")
-    set(CMAKE_CXX_FLAGS "-mavx2 -Og -g -Wall -fPIC -std=c++11 ${CMAKE_CXX_FLAGS}")
-  else()
-    set(CMAKE_C_FLAGS "-mavx2 -O3 -Wall -fPIC ${CMAKE_C_FLAGS}")
-    set(CMAKE_CXX_FLAGS "-mavx2 -O3 -Wall -fPIC -std=c++11 ${CMAKE_CXX_FLAGS}")
-    # set(CMAKE_CXX_FLAGS "-mavx2 -O2 -fno-strict-aliasing -Wall -fPIC -std=c++11 ${CMAKE_CXX_FLAGS}")}"
-  endif ()
-  if (CMAKE_CXX_COMPILER_ID MATCHES "GNU" AND
-      CMAKE_CXX_COMPILER_VERSION VERSION_GREATER 7.0)
-    set(CMAKE_CXX_FLAGS "-faligned-new ${CMAKE_CXX_FLAGS}")
+set(CMAKE_C_FLAGS "-fsigned-char -Wall -fPIC ${CMAKE_C_FLAGS}")
+set(CMAKE_CXX_FLAGS "-fsigned-char -Wall -fPIC -std=c++11 ${CMAKE_CXX_FLAGS}")
+
+if (DEBUG)
+  set(CMAKE_C_FLAGS "-Og -g ${CMAKE_C_FLAGS}")
+  set(CMAKE_CXX_FLAGS "-Og -g ${CMAKE_CXX_FLAGS}")
+else()
+  set(CMAKE_C_FLAGS "-O3 ${CMAKE_C_FLAGS}")
+  set(CMAKE_CXX_FLAGS "-O3 ${CMAKE_CXX_FLAGS}")
+endif ()
+
+if(PROFILE)
+  add_definitions(-DPROFILE)
+  if (ENABLE_CUDA)
+    add_definitions(-DCUDA_PROFILE)
   endif()
-endif(MSVC)
+endif()
 
-# add source group
-FILE(GLOB_RECURSE GROUP_INCLUDE "src/*.h" "include/*.h")
-assign_source_group("Include" ${GROUP_INCLUDE})
+# compile source files
+set (CVM_RUNTIME_LINKER_LIBS "")
+set (EXECUTE_NAME cvm_runtime)
+set (LIBRARY_NAME cvm)
 
 # Source file lists
-file(GLOB RUNTIME_SRCS src/*.cc)
+file(GLOB RUNTIME_SRCS src/core/*.cc
+                       src/runtime/*.cc
+                       src/runtime/cpu/devapi/*.cc)
+
+file(GLOB CVM_SRCS src/compiler/*.cc
+                   src/pass/*.cc)
+
+if(ENABLE_CPU)
+    message(STATUS "ENABLE CPU")
+    file(GLOB CVM_OPS_CPU_SRCS src/runtime/cpu/ops/*.cc)
+    list(APPEND RUNTIME_SRCS ${CVM_OPS_CPU_SRCS})
+endif()
 
-# Package runtime rules
-if(NOT USE_RTTI)
-  add_definitions(-DDMLC_ENABLE_RTTI=0)
+if(ENABLE_FORMAL)
+    message(STATUS "ENABLE FORMALIZATION")
+    file(GLOB RUNTIME_FORMAL_SRCS src/runtime/formal/devapi/*.cc)
+    list(APPEND RUNTIME_SRCS ${RUNTIME_FORMAL_SRCS})
+
+    file(GLOB CVM_OPS_FORMAL_SRCS src/runtime/formal/ops/*.cc)
+    list(APPEND RUNTIME_SRCS ${CVM_OPS_FORMAL_SRCS})
 endif()
 
-message(STATUS "Build with CVM runtime support...")
-file(GLOB RUNTIME_CVM_SRCS src/cvm/*.cc) 
-if(${USE_CUDA} STREQUAL "ON")
-    message("use cuda")
-    project(cvm CUDA)
-    set (EXECUTE_NAME cvm_runtime_cuda)
+if(ENABLE_CUDA)
+    message(STATUS "ENABLE CUDA")
+    # project(cvm CUDA)
+
+    find_cuda(${ENABLE_CUDA})
+    if (CUDA_FOUND)
+      include_directories(${CUDA_INCLUDE_DIRS})
+    else()
+      message(FATAL_ERROR "Cannot find CUDA, ENABLE_CUDA=" ${ENABLE_CUDA})
+    endif()
+
+    list(APPEND CVM_RUNTIME_LINKER_LIBS ${CUDA_CUDART_LIBRARY})
+    list(APPEND CVM_RUNTIME_LINKER_LIBS ${CUDA_CUDA_LIBRARY})
+    list(APPEND CVM_RUNTIME_LINKER_LIBS ${CUDA_NVRTC_LIBRARY})
+
+    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --expt-extended-lambda)
+
     add_definitions(-DCVM_RUNTIME_CUDA)
-    file(GLOB RUNTIME_CVM_CUDA_SRCS src/cuda/*.cc)
-    file(GLOB CVM_OPS_CUDA_SRCS src/cvm/ops/cuda/*.cc
-      src/cvm/ops/cuda/*.cu)
-    list(APPEND RUNTIME_CVM_SRCS ${CVM_OPS_CUDA_SRCS})
-    list(APPEND RUNTIME_SRCS ${RUNTIME_CVM_CUDA_SRCS})
-    list(APPEND CVM_RUNTIME_LINKER_LIBS "cudart")
-    list(APPEND CVM_RUNTIME_LINKER_LIBS "cuda")
-    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -std=c++11 --expt-extended-lambda")
-else()
-    file(GLOB CVM_OPS_CPU_SRCS src/cvm/ops/cpu/*.cc)
-    list(APPEND RUNTIME_CVM_SRCS ${CVM_OPS_CPU_SRCS})
-    set (EXECUTE_NAME cvm_runtime_cpu)
+    if(USE_NANO)
+      add_definitions(-DNANO)
+    endif()
+      
+    file(GLOB RUNTIME_CUDA_SRCS src/runtime/cuda/devapi/*.cc)
+    list(APPEND RUNTIME_SRCS ${RUNTIME_CUDA_SRCS})
+
+    file(GLOB CVM_OPS_CUDA_SRCS src/runtime/cuda/ops/*.cc)
+    list(APPEND RUNTIME_SRCS ${CVM_OPS_CUDA_SRCS})
+
+    file(GLOB CUDA_SRCS src/runtime/cuda/ops/*.cu)
+    cuda_compile(CUDA_LIBS ${CUDA_SRCS})
+    list(APPEND RUNTIME_SRCS ${CUDA_LIBS})
 endif()
-list(APPEND RUNTIME_SRCS ${RUNTIME_CVM_SRCS})
-file(GLOB CVM_TOP_SRCS src/cvm/top/*.cc
-          src/cvm/top/nn/*.cc
-          src/cvm/top/tensor/*.cc)
-list(APPEND RUNTIME_SRCS ${CVM_TOP_SRCS})
 
-# Module rules
-include(cmake/modules/CUDA.cmake)
+if(ENABLE_OPENCL)
+  message(STATUS "ENABLE OPENCL")
+
+# list(APPEND RUNTIME_SRCS ${RUNTIME_CUDA_SRCS})
+#  list(APPEND CVM_RUNTIME_LINKER_LIBS ${CUDA_CUDART_LIBRARY})
+#  list(APPEND CVM_RUNTIME_LINKER_LIBS ${CUDA_CUDA_LIBRARY})
+
+  set(CMAKE_C_FLAGS "-lOpenCL ${CMAKE_C_FLAGS}")
+  set(CMAKE_CXX_FLAGS "-lOpenCL ${CMAKE_CXX_FLAGS}")
+
+  file(GLOB RUNTIME_OPENCL_SRCS src/runtime/opencl/devapi/*.cc)
+  list(APPEND RUNTIME_SRCS ${RUNTIME_OPENCL_SRCS})
+
+  file(GLOB CVM_OPS_OPENCL_SRCS src/runtime/opencl/ops/*.cc)
+  list(APPEND RUNTIME_SRCS ${CVM_OPS_OPENCL_SRCS})
+
+endif()
+
+list(APPEND RUNTIME_SRCS ${CVM_SRCS})
+
+file(GLOB CVM_TOP_SRCS src/top/*.cc
+                       src/top/nn/*.cc
+                       src/top/tensor/*.cc)
+list(APPEND RUNTIME_SRCS ${CVM_TOP_SRCS})
 
 add_library(${EXECUTE_NAME} SHARED ${RUNTIME_SRCS})
 target_link_libraries(${EXECUTE_NAME} ${CVM_RUNTIME_LINKER_LIBS})
 
-# Installation rules
-install(TARGETS ${EXECUTE_NAME} DESTINATION lib${LIB_SUFFIX})
+list(APPEND CVM_SRCS ${RUNTIME_SRCS})
+add_library(${LIBRARY_NAME} SHARED ${CVM_SRCS})
+target_link_libraries(${LIBRARY_NAME} ${CVM_RUNTIME_LINKER_LIBS})
+
 
-# More target definitions
-if(MSVC)
-    target_compile_definitions(${EXECUTE_NAME} PRIVATE -DHalide_EXPORTS)
-    target_compile_definitions(${EXECUTE_NAME} PRIVATE -DCVM_EXPORTS)
-endif()

Makefile

@@ -1,20 +1,45 @@
-ROOTDIR = $(CURDIR)
+.PHONY: clean all dep test_cpu test_gpu test_formal python
+# .PHONY: test_model_cpu test_model_gpu test_model_formal
+# .PHONY: test_op_cpu test_op_gpu test_op_formal
 
-.PHONY: clean all test
+BUILD := build
+INCLUDE := include
+TESTS := tests
 
-INCLUDE_FLAGS = -Iinclude
-PKG_CFLAGS = -std=c++11 -Wall -O2 $(INCLUDE_FLAGS) -fPIC
-PKG_LDFLAGS =
+all: lib python test_cpu test_gpu test_formal
+	echo ${TEST_CPUS}
 
-all: cpu gpu
+# Mac OS should install libomp with brew
+dep:
+	python3 install/deps.py
 
-cpu:
-		 @mkdir -p build/cpu && cd build/cpu && cmake ../.. -DUSE_CUDA=OFF && $(MAKE)
-#		@mkdir -p build/cpu && cd build/cpu && cmake ../.. -DUSE_CUDA=OFF -DCMAKE_BUILD_TYPE=Debug && $(MAKE)
+lib: dep
+	@cd ${BUILD} && cmake ../ && $(MAKE)
 
-gpu:
-		 @mkdir -p build/gpu && cd build/gpu && cmake ../.. -DUSE_CUDA=ON && $(MAKE)
-# @mkdir -p build/gpu && cd build/gpu && cmake ../.. -DUSE_CUDA=ON -DCMAKE_BUILD_TYPE=Debug && $(MAKE)
+TEST_SRCS := $(wildcard ${TESTS}/*.cc)
+TEST_EXES := $(patsubst ${TESTS}/%.cc,%,${TEST_SRCS})
+
+TEST_CPUS := $(patsubst %,%_cpu,${TEST_EXES})
+TEST_GPUS := $(patsubst %,%_gpu,${TEST_EXES})
+TEST_FORMALS := $(patsubst %,%_formal,${TEST_EXES})
+TEST_OPENCL := $(patsubst %,%_opencl,${TEST_EXES})
+
+test_cpu: ${TEST_CPUS}
+test_gpu: ${TEST_GPUS}
+test_formal: ${TEST_FORMALS}
+test_opencl: ${TEST_OPENCL}
+
+%_cpu: ${TESTS}/%.cc lib
+	g++ -o ${BUILD}/${TESTS}/$@ $< -DDEVICE=0 -std=c++11 -I${INCLUDE} -L${BUILD} -lcvm -fopenmp -fsigned-char -pthread -Wl,-rpath=${BUILD}
+
+%_gpu: ${TESTS}/%.cc lib
+	g++ -o ${BUILD}/${TESTS}/$@ $< -DDEVICE=1 -std=c++11 -I${INCLUDE} -L${BUILD} -lcvm -fopenmp -fsigned-char -pthread -Wl,-rpath=${BUILD}
+
+%_formal: ${TESTS}/%.cc lib
+	g++ -o ${BUILD}/${TESTS}/$@ $< -DDEVICE=2 -std=c++11 -I${INCLUDE} -L${BUILD} -lcvm -fopenmp -fsigned-char -pthread -Wl,-rpath=${BUILD}
+
+%_opencl: ${TESTS}/%.cc lib
+	g++ -o ${BUILD}/${TESTS}/$@ $< -DDEVICE=3 -std=c++11 -I${INCLUDE} -L${BUILD} -lcvm_runtime -fopenmp -L/usr/local/cuda/lib64/ -lOpenCL -fsigned-char -pthread -Wl,-rpath=${BUILD}
 
 clean:
-	  rm -rf ./build/*
+	rm -rf ./build/*

README.md

@@ -0,0 +1,29 @@
+# cvm-runtime
+CVM Runtime
+
+
+
+## Latency
+
+model|  Jetson Nano \- Cortex\-A57(s) | Intel E5\-2650(s) |  Jetson Nano \- GPU(128 CUDA Cores)(s) | 1080Ti(3584 CUDA Cores)(s)
+-|-|-|-|-
+yolo_tfm | | | 1.076 | 0.043
+resnet50_mxg | 1.2076| 0.3807| 0.147 | 0.009
+resnet18_v1 | |  | 0.055 | 0.004
+qd10_resnet20_v2 ||  | 0.064 | 0.010
+resnet50_v2 |1.4674| 0.5005 | 0.185 | 0.010
+qd10_resnet20_v2|0.2944|0.1605 | 0.065 | 0.012
+trec | 0.0075| 0.0028 | 0.002 | 0.001
+dcnet_mnist_v1|0.0062|0.0057 | 0.002 | 0.001
+mobilenetv1.0_imagenet|0.3508| 0.1483| 0.039  | 0.002
+resnet50_v1_imagenet|1.2453| 0.3429 | 0.150 | 0.009
+animal10 | 0.3055 | 0.1466 | 0.065 | 0.010
+vgg16_gcv|4.3787| 0.6092 | 0.713 | 0.021
+sentiment_trec|0.0047| 0.0022 |  0.002 | 0.001
+vgg19_gcv|5.1753| 0.7513 | 0.788 | 0.023
+squeezenet_gcv1.1|0.3889|  0.0895 |  0.044 | 0.002
+squeezenet_gcv1.0|0.1987| 0.1319 | 0.064 | 0.003
+shufflenet|1.4575| 0.7697 | 0.140 | 0.004
+ssd| | |0.773 | 0.030
+ssd_512_mobilenet1.0_coco_tfm| | | 0.311 | 0.016
+ssd_512_mobilenet1.0_voc_tfm| | | 0.220 | 0.014