Merge remote-tracking branch 'tvm-source/main' into SYSOL-584-pytorch…

…-conv-transpose-pad-fix

.gitignore

@@ -233,3 +233,8 @@ conda/pkg
 # nix files
 .envrc
 *.nix
+
+# Downloaded models/datasets
+.tvm_test_data
+.dgl
+.caffe2

CMakeLists.txt

@@ -34,9 +34,9 @@ tvm_option(USE_RPC "Build with RPC" ON)
 tvm_option(USE_THREADS "Build with thread support" ON)
 tvm_option(USE_LLVM "Build with LLVM, can be set to specific llvm-config path" OFF)
 tvm_option(USE_STACKVM_RUNTIME "Include stackvm into the runtime" OFF)
-tvm_option(USE_GRAPH_RUNTIME "Build with tiny graph runtime" ON)
-tvm_option(USE_GRAPH_RUNTIME_CUDA_GRAPH "Build with tiny graph runtime with CUDA Graph for GPUs" OFF)
-tvm_option(USE_PROFILER "Build profiler for the VM and graph runtime" ON)
+tvm_option(USE_GRAPH_EXECUTOR "Build with tiny graph executor" ON)
+tvm_option(USE_GRAPH_EXECUTOR_CUDA_GRAPH "Build with tiny graph executor with CUDA Graph for GPUs" OFF)
+tvm_option(USE_PROFILER "Build profiler for the VM and graph executor" ON)
 tvm_option(USE_OPENMP "Build with OpenMP thread pool implementation" OFF)
 tvm_option(USE_RELAY_DEBUG "Building Relay in debug mode..." OFF)
 tvm_option(USE_RTTI "Build with RTTI" ON)
@@ -79,7 +79,7 @@ tvm_option(USE_COREML "Build with coreml support" OFF)
 tvm_option(USE_BNNS "Build with BNNS support" OFF)
 tvm_option(USE_TARGET_ONNX "Build with ONNX Codegen support" OFF)
 tvm_option(USE_ARM_COMPUTE_LIB "Build with Arm Compute Library" OFF)
-tvm_option(USE_ARM_COMPUTE_LIB_GRAPH_RUNTIME "Build with Arm Compute Library graph runtime" OFF)
+tvm_option(USE_ARM_COMPUTE_LIB_GRAPH_EXECUTOR "Build with Arm Compute Library graph executor" OFF)
 tvm_option(USE_TENSORRT_CODEGEN "Build with TensorRT Codegen support" OFF)
 tvm_option(USE_TENSORRT_RUNTIME "Build with TensorRT runtime" OFF)
 tvm_option(USE_RUST_EXT "Build with Rust based compiler extensions, STATIC, DYNAMIC, or OFF" OFF)
@@ -307,16 +307,30 @@ else()
   list(APPEND COMPILER_SRCS ${STACKVM_RUNTIME_SRCS})
 endif(USE_STACKVM_RUNTIME)
 
-if(USE_GRAPH_RUNTIME)
-  message(STATUS "Build with Graph runtime support...")
-  file(GLOB RUNTIME_GRAPH_SRCS src/runtime/graph/*.cc)
-  list(APPEND RUNTIME_SRCS ${RUNTIME_GRAPH_SRCS})
+# NOTE(areusch): USE_GRAPH_RUNTIME will be deleted in a future release
+if(USE_GRAPH_RUNTIME AND NOT DEFINED USE_GRAPH_EXECUTOR)
+  message(WARNING "USE_GRAPH_RUNTIME renamed to USE_GRAPH_EXECUTOR. Please update your config.cmake")
+  set(USE_GRAPH_EXECUTOR ${USE_GRAPH_RUNTIME})
+  unset(USE_GRAPH_RUNTIME CACHE)
+endif(USE_GRAPH_RUNTIME AND NOT DEFINED USE_GRAPH_EXECUTOR)
+
+# NOTE(areusch): USE_GRAPH_RUNTIME_DEBUG will be deleted in a future release
+if(USE_GRAPH_RUNTIME_DEBUG AND NOT DEFINED USE_GRAPH_EXECUTOR_DEBUG)
+  message(WARNING "USE_GRAPH_RUNTIME_DEBUG renamed to USE_GRAPH_EXECUTOR_DEBUG. Please update your config.cmake")
+  set(USE_GRAPH_EXECUTOR_DEBUG ${USE_GRAPH_RUNTIME_DEBUG})
+  unset(USE_GRAPH_RUNTIME_DEBUG CACHE)
+endif(USE_GRAPH_RUNTIME_DEBUG AND NOT DEFINED USE_GRAPH_EXECUTOR_DEBUG)
+
+if(USE_GRAPH_EXECUTOR)
+  message(STATUS "Build with Graph Executor support...")
+  file(GLOB RUNTIME_GRAPH_EXECUTOR_SRCS src/runtime/graph_executor/*.cc)
+  list(APPEND RUNTIME_SRCS ${RUNTIME_GRAPH_EXECUTOR_SRCS})
 
-endif(USE_GRAPH_RUNTIME)
+endif(USE_GRAPH_EXECUTOR)
 
 # convert old options for profiler
-if(USE_GRAPH_RUNTIME_DEBUG)
-  unset(USE_GRAPH_RUNTIME_DEBUG CACHE)
+if(USE_GRAPH_EXECUTOR_DEBUG)
+  unset(USE_GRAPH_EXECUTOR_DEBUG CACHE)
   set(USE_PROFILER ON)
 endif()
 if(USE_VM_PROFILER)
@@ -327,10 +341,10 @@ endif()
 if(USE_PROFILER)
   message(STATUS "Build with profiler...")
 
-  file(GLOB RUNTIME_GRAPH_DEBUG_SRCS src/runtime/graph/debug/*.cc)
-  list(APPEND RUNTIME_SRCS ${RUNTIME_GRAPH_DEBUG_SRCS})
-  set_source_files_properties(${RUNTIME_GRAPH_SRCS}
-    PROPERTIES COMPILE_DEFINITIONS "TVM_GRAPH_RUNTIME_DEBUG")
+  file(GLOB RUNTIME_GRAPH_EXECUTOR_DEBUG_SRCS src/runtime/graph_executor/debug/*.cc)
+  list(APPEND RUNTIME_SRCS ${RUNTIME_GRAPH_EXECUTOR_DEBUG_SRCS})
+  set_source_files_properties(${RUNTIME_GRAPH_EXECUTOR_SRCS}
+    PROPERTIES COMPILE_DEFINITIONS "TVM_GRAPH_EXECUTOR_DEBUG")
 
   file(GLOB RUNTIME_VM_PROFILER_SRCS src/runtime/vm/profiler/*.cc)
   list(APPEND RUNTIME_SRCS ${RUNTIME_VM_PROFILER_SRCS})
@@ -435,12 +449,6 @@ if(BUILD_FOR_HEXAGON)
   # Wrap pthread_create to allow setting custom stack size.
   set_property(TARGET tvm_runtime APPEND PROPERTY LINK_FLAGS
                         "-Wl,--wrap=pthread_create")
-
-  target_include_directories(tvm_runtime SYSTEM
-    PUBLIC "${USE_HEXAGON_SDK}/libs/common/qurt/ADSPv62MP/include/posix"
-    PUBLIC "${USE_HEXAGON_SDK}/libs/common/qurt/ADSPv62MP/include/qurt"
-    PUBLIC "${USE_HEXAGON_SDK}/incs"
-    PUBLIC "${USE_HEXAGON_SDK}/incs/stddef")
 endif()
 
 if(USE_THREADS AND NOT BUILD_FOR_HEXAGON)

CONTRIBUTORS.md

@@ -46,7 +46,7 @@ We do encourage everyone to work anything they are interested in.
 - [Liangfu Chen](https://github.com/liangfu): @liangfu - vta, chisel, intel FPGA, c runtime
 - [Wei Chen](https://github.com/wweic): @wweic - runtime, relay, vm
 - [Zhi Chen](https://github.com/zhiics) (PMC): @zhiics - relay, quantization, pass manager
-- [Chenfan](https://github.com/jcf94): @jcf94 - autoscheduling
+- [Chenfan](https://github.com/jcf94): @jcf94 - auto_scheduler
 - [Josh Fromm](https://github.com/jwfromm): @jwfromm - frontends, quantization, topi
 - [Yuwei Hu](https://github.com/Huyuwei): @Huyuwei - topi, frontends
 - [Nick Hynes](https://github.com/nhynes): @nhynes: - sgx, rust
@@ -70,12 +70,12 @@ We do encourage everyone to work anything they are interested in.
 - [Andrew Tulloch](https://github.com/ajtulloch): @ajtulloch - topi, compiler, runtime
 - [Luis Vega](https://github.com/vegaluisjose): @vegaluisjose - vta, chisel
 - [Leyuan Wang](https://github.com/Laurawly) (PMC): @Laurawly: - topi
-- [Yao Wang](https://github.com/kevinthesun): @kevinthesun: - topi, vision
+- [Yao Wang](https://github.com/kevinthesun): @kevinthesun (PMC): - topi, vision
 - [Jian Weng](https://github.com/were): @were: - hybrid script
 - [Zhao Wu](https://github.com/FrozenGene): @FrozenGene - runtime, topi, frontends
 - [Eddie Yan](https://github.com/eqy) (PMC): @eqy - runtime, autotvm, rpc, topi
-- [Hao Yu](https://github.com/comaniac): @comaniac - relay, byoc, ansor
-- [Lianmin Zheng](https://github.com/merrymercy) (PMC): @merrymercy - autotvm, topi, relay
+- [Hao Yu](https://github.com/comaniac): @comaniac - relay, byoc, auto_scheduler
+- [Lianmin Zheng](https://github.com/merrymercy) (PMC): @merrymercy - autotvm, auto_scheduler, topi, relay
 
 ## Reviewers
 
@@ -92,8 +92,10 @@ We do encourage everyone to work anything they are interested in.
 - [Balint Cristian](https://github.com/cbalint13): @cbalint13
 - [Haozheng Fan](https://github.com/hzfan): @hzfan
 - [Josh Fromm](https://github.com/jwfromm): @jwfromm
+- [Siyuan Feng](https://github.com/Hzfengsy): @Hzfengsy
 - [Sergei Grechanik](https://github.com/sgrechanik-h): @sgrechanik-h
 - [Hao Lu](https://github.com/hlu1): @hlu1
+- [Bohan Hou](https://github.com/spectrometerHBH): @spectrometerHBH
 - [Nick Hynes](https://github.com/nhynes): @nhynes
 - [Yuwei Hu](https://github.com/Huyuwei): @Huyuwei
 - [Luke Hutton](https://github.com/lhutton1): @lhutton1

Jenkinsfile

@@ -46,11 +46,11 @@
 // NOTE: these lines are scanned by docker/dev_common.sh. Please update the regex as needed. -->
 ci_lint = "tlcpack/ci-lint:v0.62"
 ci_gpu = "tlcpack/ci-gpu:v0.72"
-ci_cpu = "tlcpack/ci-cpu:v0.72-t0"
+ci_cpu = "tlcpack/ci-cpu:v0.73"
 ci_wasm = "tlcpack/ci-wasm:v0.70"
 ci_i386 = "tlcpack/ci-i386:v0.72-t0"
-ci_qemu = "tlcpack/ci-qemu:v0.02"
-ci_arm = "tlcpack/ci-arm:v0.02"
+ci_qemu = "tlcpack/ci-qemu:v0.03"
+ci_arm = "tlcpack/ci-arm:v0.03"
 // <--- End of regex-scanned config.
 
 // tvm libraries
@@ -187,7 +187,8 @@ stage('Build') {
           sh "${docker_run} ${ci_cpu} ./tests/scripts/task_python_vta_fsim.sh"
           sh "${docker_run} ${ci_cpu} ./tests/scripts/task_python_vta_tsim.sh"
           // sh "${docker_run} ${ci_cpu} ./tests/scripts/task_golang.sh"
-          sh "${docker_run} ${ci_cpu} ./tests/scripts/task_rust.sh"
+          // TODO(@jroesch): need to resolve CI issue will turn back on in follow up patch
+          // sh "${docker_run} ${ci_cpu} ./tests/scripts/task_rust.sh"
           junit "build/pytest-results/*.xml"
         }
       }

Makefile

@@ -52,10 +52,10 @@ cpptest:
 	@mkdir -p $(OUTPUTDIR) && cd $(OUTPUTDIR) && cmake .. && $(MAKE) cpptest
 
 crttest:
-	@mkdir -p build && cd build && cmake .. && $(MAKE) crttest
+	@mkdir -p $(OUTPUTDIR) && cd $(OUTPUTDIR) && cmake .. && $(MAKE) crttest
 
 # EMCC; Web related scripts
-EMCC_FLAGS= -std=c++11 -DDMLC_LOG_STACK_TRACE=0\
+EMCC_FLAGS= -std=c++11\
 	-Oz -s RESERVED_FUNCTION_POINTERS=2 -s MAIN_MODULE=1 -s NO_EXIT_RUNTIME=1\
 	-s TOTAL_MEMORY=1073741824\
 	-s EXTRA_EXPORTED_RUNTIME_METHODS="['addFunction','cwrap','getValue','setValue']"\

NEWS.md

@@ -1635,7 +1635,7 @@ Rust language support in TVM includes two parts. 1. The frontend wraps the curre
 * [topi] enable fp16 sort for arm (#4084)
 * Add OpenOCD Low-Level Device (RISC-V Support) (#3756)
 * Add wave 32 bc for AMD ROCm backend (#3984)
-* [RUTNIME] Support C++ RPC (#4281)
+* [RUNTIME] Support C++ RPC (#4281)
 * [TOPI][OP] Support Faster-RCNN Proposal OP on CPU (#4297)
 * [TVM][RUNTIME] A minimum example to generate external library wrappers for DSOModule (#4280)
 

apps/android_camera/app/src/main/java/org/apache/tvm/android/androidcamerademo/Camera2BasicFragment.java

@@ -57,7 +57,7 @@
 import org.apache.tvm.Function;
 import org.apache.tvm.Module;
 import org.apache.tvm.NDArray;
-import org.apache.tvm.TVMContext;
+import org.apache.tvm.Device;
 import org.apache.tvm.TVMType;
 import org.apache.tvm.TVMValue;
 import org.json.JSONException;
@@ -111,7 +111,7 @@ public class Camera2BasicFragment extends Fragment {
     private AppCompatTextView mInfoView;
     private ListView mModelView;
     private AssetManager assetManager;
-    private Module graphRuntimeModule;
+    private Module graphExecutorModule;
     private JSONObject labels;
     private ListenableFuture<ProcessCameraProvider> cameraProviderFuture;
     private PreviewView previewView;
@@ -187,21 +187,21 @@ private String[] getModels() {
     private String[] inference(float[] chw) {
         NDArray inputNdArray = NDArray.empty(new long[]{1, IMG_CHANNEL, MODEL_INPUT_SIZE, MODEL_INPUT_SIZE}, new TVMType("float32"));
         inputNdArray.copyFrom(chw);
-        Function setInputFunc = graphRuntimeModule.getFunction("set_input");
+        Function setInputFunc = graphExecutorModule.getFunction("set_input");
         setInputFunc.pushArg(INPUT_NAME).pushArg(inputNdArray).invoke();
         // release tvm local variables
         inputNdArray.release();
         setInputFunc.release();
 
         // get the function from the module(run it)
-        Function runFunc = graphRuntimeModule.getFunction("run");
+        Function runFunc = graphExecutorModule.getFunction("run");
         runFunc.invoke();
         // release tvm local variables
         runFunc.release();
 
         // get the function from the module(get output data)
         NDArray outputNdArray = NDArray.empty(new long[]{1, 1000}, new TVMType("float32"));
-        Function getOutputFunc = graphRuntimeModule.getFunction("get_output");
+        Function getOutputFunc = graphExecutorModule.getFunction("get_output");
         getOutputFunc.pushArg(OUTPUT_INDEX).pushArg(outputNdArray).invoke();
         float[] output = outputNdArray.asFloatArray();
         // release tvm local variables
@@ -272,8 +272,8 @@ public void onActivityCreated(Bundle savedInstanceState) {
     @Override
     public void onDestroy() {
         // release tvm local variables
-        if (null != graphRuntimeModule)
-            graphRuntimeModule.release();
+        if (null != graphExecutorModule)
+            graphExecutorModule.release();
         super.onDestroy();
     }
 
@@ -516,7 +516,7 @@ private void setInputName(String modelName) {
     }
 
     /*
-       Load precompiled model on TVM graph runtime and init the system.
+       Load precompiled model on TVM graph executor and init the system.
     */
     private class LoadModelAsyncTask extends AsyncTask<Void, Void, Integer> {
 
@@ -571,36 +571,36 @@ protected Integer doInBackground(Void... args) {
                 return -1;//failure
             }
 
-            Log.i(TAG, "creating java tvm context...");
-            // create java tvm context
-            TVMContext tvmCtx = EXE_GPU ? TVMContext.opencl() : TVMContext.cpu();
+            Log.i(TAG, "creating java tvm device...");
+            // create java tvm device
+            Device tvmDev = EXE_GPU ? Device.opencl() : Device.cpu();
 
             Log.i(TAG, "loading compiled functions...");
             Log.i(TAG, libCacheFilePath);
             // tvm module for compiled functions
             Module modelLib = Module.load(libCacheFilePath);
 
 
-            // get global function module for graph runtime
-            Log.i(TAG, "getting graph runtime create handle...");
+            // get global function module for graph executor
+            Log.i(TAG, "getting graph executor create handle...");
 
-            Function runtimeCreFun = Function.getFunction("tvm.graph_runtime.create");
-            Log.i(TAG, "creating graph runtime...");
+            Function runtimeCreFun = Function.getFunction("tvm.graph_executor.create");
+            Log.i(TAG, "creating graph executor...");
 
-            Log.i(TAG, "ctx type: " + tvmCtx.deviceType);
-            Log.i(TAG, "ctx id: " + tvmCtx.deviceId);
+            Log.i(TAG, "device type: " + tvmDev.deviceType);
+            Log.i(TAG, "device id: " + tvmDev.deviceId);
 
             TVMValue runtimeCreFunRes = runtimeCreFun.pushArg(modelGraph)
                     .pushArg(modelLib)
-                    .pushArg(tvmCtx.deviceType)
-                    .pushArg(tvmCtx.deviceId)
+                    .pushArg(tvmDev.deviceType)
+                    .pushArg(tvmDev.deviceId)
                     .invoke();
 
             Log.i(TAG, "as module...");
-            graphRuntimeModule = runtimeCreFunRes.asModule();
-            Log.i(TAG, "getting graph runtime load params handle...");
+            graphExecutorModule = runtimeCreFunRes.asModule();
+            Log.i(TAG, "getting graph executor load params handle...");
             // get the function from the module(load parameters)
-            Function loadParamFunc = graphRuntimeModule.getFunction("load_params");
+            Function loadParamFunc = graphExecutorModule.getFunction("load_params");
             Log.i(TAG, "loading params...");
             loadParamFunc.pushArg(modelParams).invoke();
             // release tvm local variables

apps/android_camera/app/src/main/jni/Application.mk

@@ -31,7 +31,7 @@ include $(config)
 APP_ABI ?= all
 APP_STL := c++_shared
 
-APP_CPPFLAGS += -DTVM4J_ANDROID=1 -std=c++14 -Oz -frtti
+APP_CPPFLAGS += -DTVM_LOG_STACK_TRACE=0 -DTVM4J_ANDROID=1 -std=c++14 -Oz -frtti
 ifeq ($(USE_OPENCL), 1)
     APP_CPPFLAGS += -DTVM_OPENCL_RUNTIME=1
 endif

apps/android_camera/app/src/main/jni/tvm_runtime.h

@@ -26,7 +26,7 @@
 #include <fstream>
 
 #define DMLC_USE_LOGGING_LIBRARY <tvm/runtime/logging.h>
-#define TVM_BACKTRACE_DISABLED 1
+#define TVM_USE_LIBBACKTRACE 0
 /* Enable custom logging - this will cause TVM to use a custom implementation
  * of tvm::runtime::detail::LogMessage. We use this to pass TVM log messages to
  * Android logcat.
@@ -37,8 +37,9 @@
 #include "../src/runtime/cpu_device_api.cc"
 #include "../src/runtime/dso_library.cc"
 #include "../src/runtime/file_utils.cc"
-#include "../src/runtime/graph/graph_runtime.cc"
+#include "../src/runtime/graph_executor/graph_executor.cc"
 #include "../src/runtime/library_module.cc"
+#include "../src/runtime/logging.cc"
 #include "../src/runtime/module.cc"
 #include "../src/runtime/ndarray.cc"
 #include "../src/runtime/object.cc"

apps/android_camera/models/prepare_model.py

@@ -25,7 +25,7 @@
 
 import tvm
 import tvm.relay as relay
-from tvm.contrib import utils, ndk, graph_runtime as runtime
+from tvm.contrib import utils, ndk, graph_executor as runtime
 from tvm.contrib.download import download_testdata, download
 
 target = "llvm -mtriple=arm64-linux-android"