neo-ai · wweic · Jun 27, 2019 · May 13, 2019 · May 13, 2019 · May 14, 2019
diff --git a/.gitignore b/.gitignore
@@ -216,3 +216,14 @@ patched.txt
 # Python type checking
 .mypy_cache/
 .pyre/
+
+# pipenv file
+Pipfile
+Pipfile.lock
+
+# conda package artifacts
+conda/Dockerfile.cuda*
+conda/pkg
+
+.envrc
+*.nix
diff --git a/3rdparty/HalideIR b/3rdparty/HalideIR
diff --git a/3rdparty/bfloat16/bfloat16.cc b/3rdparty/bfloat16/bfloat16.cc
@@ -0,0 +1,80 @@
+/*
+    Copyright (c) 2019 by Contributors
+   \file tvm/src/codegen/custom_datatypes/mybfloat16.cc
+   \brief Small bfloat16 library for use in unittests
+
+  Code originally from TensorFlow; taken and simplified. Original license:
+
+  Licensed under the Apache License, Version 2.0 (the "License");
+  you may not use this file except in compliance with the License.
+  You may obtain a copy of the License at
+  http://www.apache.org/licenses/LICENSE-2.0
+  Unless required by applicable law or agreed to in writing, software
+  distributed under the License is distributed on an "AS IS" BASIS,
+  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+  See the License for the specific language governing permissions and
+  limitations under the License.
+  ==============================================================================*/
+
+#include <tvm/runtime/c_runtime_api.h>
+#include <cstddef>
+#include <cstdint>
+
+void FloatToBFloat16(const float* src, uint16_t* dst, size_t size) {
+  const uint16_t* p = reinterpret_cast<const uint16_t*>(src);
+  uint16_t* q = reinterpret_cast<uint16_t*>(dst);
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  for (; size != 0; p += 2, q++, size--) {
+    *q = p[0];
+  }
+#else
+  for (; size != 0; p += 2, q++, size--) {
+    *q = p[1];
+  }
+#endif
+}
+
+void BFloat16ToFloat(const uint16_t* src, float* dst, size_t size) {
+  const uint16_t* p = reinterpret_cast<const uint16_t*>(src);
+  uint16_t* q = reinterpret_cast<uint16_t*>(dst);
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  for (; size != 0; p++, q += 2, size--) {
+    q[0] = *p;
+    q[1] = 0;
+  }
+#else
+  for (; size != 0; p++, q += 2, size--) {
+    q[0] = 0;
+    q[1] = *p;
+  }
+#endif
+}
+
+void BFloat16Add(const uint16_t* a, const uint16_t* b, uint16_t* dst,
+                 size_t size) {
+  float a_f, b_f;
+  BFloat16ToFloat(a, &a_f, 1);
+  BFloat16ToFloat(b, &b_f, 1);
+  float out_f = a_f + b_f;
+  FloatToBFloat16(&out_f, dst, 1);
+}
+
+extern "C" {
+TVM_DLL TVM_DLL uint16_t FloatToBFloat16_wrapper(float in) {
+  uint16_t out;
+  FloatToBFloat16(&in, &out, 1);
+  return out;
+}
+
+TVM_DLL float BFloat16ToFloat_wrapper(uint16_t in) {
+  float out;
+  BFloat16ToFloat(&in, &out, 1);
+  return out;
+}
+
+TVM_DLL uint16_t BFloat16Add_wrapper(uint16_t a, uint16_t b) {
+  uint16_t out;
+  BFloat16Add(&a, &b, &out, 1);
+  return out;
+}
+}
diff --git a/3rdparty/dlpack b/3rdparty/dlpack
diff --git a/3rdparty/dmlc-core b/3rdparty/dmlc-core
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -37,6 +37,13 @@ tvm_option(USE_SGX "Build with SGX" OFF)
 tvm_option(USE_RTTI "Build with RTTI" ON)
 tvm_option(USE_MSVC_MT "Build with MT" OFF)
 tvm_option(INSTALL_DEV "Install compiler infrastructure" OFF)
+tvm_option(HIDE_PRIVATE_SYMBOLS "Compile with -fvisibility=hidden." OFF)
+
+# 3rdparty libraries
+tvm_option(DLPACK_PATH "Path to DLPACK" "3rdparty/dlpack/include")
+tvm_option(DMLC_PATH "Path to DMLC" "3rdparty/dmlc-core/include")
+tvm_option(RANG_PATH "Path to RANG" "3rdparty/rang/include")
+tvm_option(COMPILER_RT_PATH "Path to COMPILER-RT" "3rdparty/compiler-rt")
 
 # Contrib library options
 tvm_option(USE_BLAS "The blas library to be linked" none)
@@ -52,11 +59,12 @@ tvm_option(USE_TENSORRT "Build with TensorRT, must have CUDA and CUDNN enabled"
 tvm_option(USE_ANTLR "Build with ANTLR for Relay parsing" OFF)
 
 # include directories
+include_directories(${CMAKE_INCLUDE_PATH})
 include_directories("include")
-include_directories("3rdparty/dlpack/include")
-include_directories("3rdparty/dmlc-core/include")
-include_directories("3rdparty/rang/include")
-include_directories("3rdparty/compiler-rt")
+include_directories(${DLPACK_PATH})
+include_directories(${DMLC_PATH})
+include_directories(${RANG_PATH})
+include_directories(${COMPILER_RT_PATH})
 
 # initial variables
 set(TVM_LINKER_LIBS "")
@@ -90,8 +98,13 @@ else(MSVC)
     set(CMAKE_C_FLAGS "-O0 -g -Wall -fPIC ${CMAKE_C_FLAGS} -rdynamic")
     set(CMAKE_CXX_FLAGS "-O0 -g -Wall -fPIC -std=c++11 ${CMAKE_CXX_FLAGS} -rdynamic")
   else()
-    set(CMAKE_C_FLAGS "-O2 -Wall -fPIC -fvisibility=hidden ${CMAKE_C_FLAGS}")
-    set(CMAKE_CXX_FLAGS "-O2 -Wall -fPIC -fvisibility=hidden -std=c++11 ${CMAKE_CXX_FLAGS}")
+    set(CMAKE_C_FLAGS "-O2 -Wall -fPIC ${CMAKE_C_FLAGS}")
+    set(CMAKE_CXX_FLAGS "-O2 -Wall -fPIC -std=c++11 ${CMAKE_CXX_FLAGS}")
+    if (HIDE_PRIVATE_SYMBOLS)
+      message("Hide private symbols...")
+      set(CMAKE_C_FLAGS "-fvisibility=hidden ${CMAKE_C_FLAGS}")
+      set(CMAKE_CXX_FLAGS "-fvisibility=hidden ${CMAKE_CXX_FLAGS}")
+    endif(HIDE_PRIVATE_SYMBOLS)
   endif ()
   if (CMAKE_CXX_COMPILER_ID MATCHES "GNU" AND
       CMAKE_CXX_COMPILER_VERSION VERSION_GREATER 7.0)
@@ -123,6 +136,8 @@ file(GLOB_RECURSE RELAY_SRCS
     )
 list(APPEND COMPILER_SRCS ${RELAY_SRCS})
 
+file(GLOB DATATYPE_SRCS src/codegen/datatype/*.cc)
+list(APPEND COMPILER_SRCS ${DATATYPE_SRCS})
 
 if(NOT MSVC)
   file(GLOB COMPILER_VERILOG_SRCS src/codegen/verilog/*.cc)
@@ -152,6 +167,8 @@ if(NOT USE_RTTI)
   add_definitions(-DDMLC_ENABLE_RTTI=0)
 endif()
 
+list(APPEND RUNTIME_SRCS 3rdparty/bfloat16/bfloat16.cc)
+
 if(USE_RPC)
   message(STATUS "Build with RPC support...")
   file(GLOB RUNTIME_RPC_SRCS src/runtime/rpc/*.cc)
@@ -209,6 +226,7 @@ add_library(tvm_runtime_static STATIC ${RUNTIME_SRCS})
 if(USE_RELAY_DEBUG)
   message(STATUS "Building Relay in debug mode...")
   set_target_properties(tvm PROPERTIES COMPILE_DEFINITIONS "USE_RELAY_DEBUG")
+else()
   set_target_properties(tvm PROPERTIES COMPILE_DEFINITIONS "NDEBUG")
 endif(USE_RELAY_DEBUG)
 

diff --git a/CONTRIBUTORS.md b/CONTRIBUTORS.md
@@ -55,7 +55,8 @@ We do encourage everyone to work anything they are interested in.
 - [Zhixun Tan](https://github.com/phisiart): @phisiart - opengl, web
 - [Leyuan Wang](https://github.com/Laurawly): @Laurawly: - topi
 - [Yao Wang](https://github.com/kevinthesun): @kevinthesun: - topi, vision
-- [Eddie Yan](https://github.com/eqy): @eqy - runtime, autotvm, rpc, topi
+- [Jian Weng](https://github.com/were): @were: - hybrid script
+- [Eddie Yan](https://github.com/eqy) (PMC): @eqy - runtime, autotvm, rpc, topi
 - [Lianmin Zheng](https://github.com/merrymercy) (PMC): @merrymercy - autotvm, topi, relay
 
 ## Reviewers
@@ -82,6 +83,7 @@ We do encourage everyone to work anything they are interested in.
 - [Kazutaka Morita](https://github.com/kazum): @kazum
 - [Tatsuya Nishiyama](https://github.com/nishi-t): @nishi-t
 - [Pariksheet Pinjari](https://github.com/PariksheetPinjari909): @PariksheetPinjari909
+- [Josh Pollock](https://github.com/joshpoll): @joshpoll
 - [Jared Roesch](https://github.com/jroesch): @jroesch
 - [Siva](https://github.com/srkreddy1238): @srkreddy1238
 - [Siju Samuel](https://github.com/siju-samuel): @siju-samuel
@@ -94,6 +96,7 @@ We do encourage everyone to work anything they are interested in.
 - [Leyuan Wang](https://github.com/Laurawly): @Laurawly
 - [Jian Weng](https://github.com/were): @were
 - [Zhao Wu](https://github.com/FrozenGene): @FrozenGene
+- [Bing Xu](https://github.com/antinucleon): @antinucleon
 - [Eddie Yan](https://github.com/eqy): @eqy
 - [Joshua Z. Zhang](https://github.com/zhreshold): @zhreshold
 - [Lianmin Zheng](https://github.com/merrymercy): @merrymercy

diff --git a/Jenkinsfile b/Jenkinsfile
@@ -39,7 +39,7 @@
 // - Periodically cleanup the old versions on local workers
 //
 ci_lint = "tvmai/ci-lint:v0.51"
-ci_gpu = "tvmai/ci-gpu:v0.51"
+ci_gpu = "tvmai/ci-gpu:v0.52"
 ci_cpu = "tvmai/ci-cpu:v0.50"
 ci_i386 = "tvmai/ci-i386:v0.50"
 
@@ -135,6 +135,7 @@ stage('Build') {
            echo set\\(USE_CUDNN ON\\) >> config.cmake
            echo set\\(USE_CUDA ON\\) >> config.cmake
            echo set\\(USE_OPENGL ON\\) >> config.cmake
+           echo set\\(USE_MICRO ON\\) >> config.cmake
            echo set\\(USE_LLVM llvm-config-6.0\\) >> config.cmake
            echo set\\(USE_NNPACK ON\\) >> config.cmake
            echo set\\(NNPACK_PATH /NNPACK/build/\\) >> config.cmake
@@ -157,6 +158,7 @@ stage('Build') {
            echo set\\(USE_OPENCL ON\\) >> config.cmake
            echo set\\(USE_ROCM ON\\) >> config.cmake
            echo set\\(USE_VULKAN ON\\) >> config.cmake
+           echo set\\(USE_MICRO ON\\) >> config.cmake
            echo set\\(USE_GRAPH_RUNTIME_DEBUG ON\\) >> config.cmake
            echo set\\(CMAKE_CXX_COMPILER clang-6.0\\) >> config.cmake
            echo set\\(CMAKE_CXX_FLAGS -Werror\\) >> config.cmake
@@ -174,12 +176,14 @@ stage('Build') {
            cd build
            cp ../cmake/config.cmake .
            echo set\\(USE_SORT ON\\) >> config.cmake
+           echo set\\(USE_MICRO ON\\) >> config.cmake
            echo set\\(USE_GRAPH_RUNTIME_DEBUG ON\\) >> config.cmake
            echo set\\(USE_LLVM llvm-config-4.0\\) >> config.cmake
            echo set\\(USE_NNPACK ON\\) >> config.cmake
            echo set\\(NNPACK_PATH /NNPACK/build/\\) >> config.cmake
            echo set\\(CMAKE_CXX_COMPILER g++\\) >> config.cmake
            echo set\\(CMAKE_CXX_FLAGS -Werror\\) >> config.cmake
+           echo set\\(HIDE_PRIVATE_SYMBOLS ON\\) >> config.cmake
            """
         make(ci_cpu, 'build', '-j4')
         pack_lib('cpu', tvm_lib)
@@ -202,6 +206,7 @@ stage('Build') {
            cd build
            cp ../cmake/config.cmake .
            echo set\\(USE_SORT ON\\) >> config.cmake
+           echo set\\(USE_MICRO ON\\) >> config.cmake
            echo set\\(USE_RPC ON\\) >> config.cmake
            echo set\\(USE_GRAPH_RUNTIME_DEBUG ON\\) >> config.cmake
            echo set\\(USE_LLVM llvm-config-5.0\\) >> config.cmake
@@ -277,6 +282,17 @@ stage('Integration Test') {
       }
     }
   },
+  'legacy: GPU': {
+    node('GPU') {
+      ws('workspace/tvm/legacy-python-gpu') {
+        init_git()
+        unpack_lib('gpu', tvm_multilib)
+        timeout(time: max_time, unit: 'MINUTES') {
+          sh "${docker_run} ${ci_gpu} ./tests/scripts/task_python_legacy.sh"
+        }
+      }
+    }
+  },
   'docs: GPU': {
     node('GPU') {
       ws('workspace/tvm/docs-python-gpu') {

diff --git a/apps/android_rpc/README.md b/apps/android_rpc/README.md
@@ -141,7 +141,7 @@ export TVM_NDK_CC=/opt/android-toolchain-arm64/bin/aarch64-linux-android-g++
 python android_rpc_test.py
 ```
 
-This will compile TVM IR to shared libraries (CPU, OpenCL and Vulkan) and run vector addition on your Android device. To verify compiled TVM IR shared libraries on OpenCL target set [`'test_opencl = True'`](https://github.com/dmlc/tvm/blob/master/apps/android_rpc/tests/android_rpc_test.py#L25) and on Vulkan target set [`'test_vulkan = False'`](https://github.com/dmlc/tvm/blob/master/apps/android_rpc/tests/android_rpc_test.py#L27) in  [tests/android_rpc_test.py](https://github.com/dmlc/tvm/blob/master/apps/android_rpc/tests/android_rpc_test.py), by default on CPU target will execute.
+This will compile TVM IR to shared libraries (CPU, OpenCL and Vulkan) and run vector addition on your Android device. To verify compiled TVM IR shared libraries on OpenCL target set `'test_opencl = True'` and on Vulkan target set `'test_vulkan = True'` in  [tests/android_rpc_test.py](https://github.com/dmlc/tvm/blob/master/apps/android_rpc/tests/android_rpc_test.py), by default on CPU target will execute.
 On my test device, it gives following results.
 
 ```bash

diff --git a/cmake/config.cmake b/cmake/config.cmake
@@ -135,9 +135,6 @@ set(USE_TENSORRT OFF)
 # Build ANTLR parser for Relay text format
 set(USE_ANTLR OFF)
 
-# Build TSIM for VTA
-set(USE_VTA_TSIM OFF)
-
 # Whether use Relay debug mode
 set(USE_RELAY_DEBUG OFF)
 
diff --git a/cmake/modules/VTA.cmake b/cmake/modules/VTA.cmake
@@ -29,8 +29,7 @@ elseif(PYTHON)
       --use-cfg=${CMAKE_CURRENT_BINARY_DIR}/vta_config.json)
   endif()
 
-  execute_process(COMMAND ${VTA_CONFIG} --target OUTPUT_VARIABLE __vta_target)
-  string(STRIP ${__vta_target} VTA_TARGET)
+  execute_process(COMMAND ${VTA_CONFIG} --target OUTPUT_VARIABLE VTA_TARGET OUTPUT_STRIP_TRAILING_WHITESPACE)
 
   message(STATUS "Build VTA runtime with target: " ${VTA_TARGET})
 
@@ -44,6 +43,13 @@ elseif(PYTHON)
 
   add_library(vta SHARED ${VTA_RUNTIME_SRCS})
 
+  if(${VTA_TARGET} STREQUAL "tsim")
+    target_compile_definitions(vta PUBLIC USE_TSIM)
+    include_directories("vta/include")
+    file(GLOB RUNTIME_DPI_SRCS vta/src/dpi/module.cc)
+    list(APPEND RUNTIME_SRCS ${RUNTIME_DPI_SRCS})
+  endif()
+
   target_include_directories(vta PUBLIC vta/include)
 
   foreach(__def ${VTA_DEFINITIONS})
@@ -55,18 +61,12 @@ elseif(PYTHON)
     set_target_properties(vta PROPERTIES LINK_FLAGS "-undefined dynamic_lookup")
   endif(APPLE)
 
-  # PYNQ rules for Pynq v2.3
+  # PYNQ rules for Pynq v2.4
   if(${VTA_TARGET} STREQUAL "pynq")
     find_library(__cma_lib NAMES cma PATH /usr/lib)
     target_link_libraries(vta ${__cma_lib})
   endif()
 
-  if(NOT USE_VTA_TSIM STREQUAL "OFF")
-    include_directories("vta/include")
-    file(GLOB RUNTIME_DPI_SRCS vta/src/dpi/module.cc)
-    list(APPEND RUNTIME_SRCS ${RUNTIME_DPI_SRCS})
-  endif()
-
 else()
   message(STATUS "Cannot found python in env, VTA build is skipped..")
 endif()
diff --git a/cmake/modules/contrib/BLAS.cmake b/cmake/modules/contrib/BLAS.cmake
@@ -27,7 +27,11 @@ elseif(USE_BLAS STREQUAL "mkl")
   if(NOT IS_DIRECTORY ${USE_MKL_PATH})
     set(USE_MKL_PATH /opt/intel/mkl)
   endif()
-  find_library(BLAS_LIBRARY NAMES mkl_rt mklml_gnu HINTS ${USE_MKL_PATH}/lib/ ${USE_MKL_PATH}/lib/intel64)
+  if(APPLE)
+    find_library(BLAS_LIBRARY NAMES mklml HINTS ${USE_MKL_PATH}/lib/ ${USE_MKL_PATH}/lib/intel64)
+  elseif(UNIX)
+    find_library(BLAS_LIBRARY NAMES mkl_rt mklml_gnu HINTS ${USE_MKL_PATH}/lib/ ${USE_MKL_PATH}/lib/intel64)
+  endif()
   include_directories(${USE_MKL_PATH}/include)
   list(APPEND TVM_RUNTIME_LINKER_LIBS ${BLAS_LIBRARY})
   list(APPEND RUNTIME_SRCS ${CBLAS_CONTRIB_SRC})

diff --git a/conda/Dockerfile.cuda100 → conda/Dockerfile.template b/conda/Dockerfile.cuda100 → conda/Dockerfile.template
@@ -15,7 +15,13 @@
 # specific language governing permissions and limitations
 # under the License.
 
-FROM nvidia/cuda:10.0-devel-centos6
+FROM nvidia/cuda:{{ cuda_version }}-devel-centos6
+
+RUN curl -fsSL http://developer.download.nvidia.com/compute/redist/cudnn/v{{ cudnn_short_version }}/cudnn-{{ cuda_version }}-linux-x64-v{{ cudnn_version }}.tgz -O && \
+    tar --no-same-owner -xzf cudnn-{{ cuda_version }}-linux-x64-v{{ cudnn_version }}.tgz -C /usr/local && \
+    rm cudnn-{{ cuda_version }}-linux-x64-v{{ cudnn_version }}.tgz && \
+    ldconfig
+
 
 RUN curl -o ~/miniconda.sh -O https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh  && \
      chmod +x ~/miniconda.sh && \
@@ -30,4 +36,4 @@ ENV LD_LIBRARY_PATH /usr/local/nvidia/lib:/usr/local/nvidia/lib64
 WORKDIR /workspace
 RUN chmod -R a+w /workspace
 
-CMD conda build --output-folder /workspace/conda/pkg --variants '{cuda: True, cuda_version: 10.0}' /workspace/conda/tvm-libs
+CMD conda build --output-folder /workspace/conda/pkg --variants '{cuda: True, cuda_version: {{ cuda_version }}}' /workspace/conda/tvm-libs
diff --git a/conda/build_cuda.sh → conda/Makefile b/conda/build_cuda.sh → conda/Makefile
@@ -5,22 +5,18 @@
 # to you under the Apache License, Version 2.0 (the
 # "License"); you may not use this file except in compliance
 # with the License.  You may obtain a copy of the License at
-# 
+#
 #   http://www.apache.org/licenses/LICENSE-2.0
-# 
+#
 # Unless required by applicable law or agreed to in writing,
 # software distributed under the License is distributed on an
 # "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
 # KIND, either express or implied.  See the License for the
 # specific language governing permissions and limitations
 # under the License.
-#/bin/sh
-condadir=`dirname $0`
-condadir=`readlink -f $condadir`
-srcdir=`dirname $condadir`
 
-docker build -t tvm-cuda100-forge $condadir -f $condadir/Dockerfile.cuda100
-docker run --rm -v $srcdir:/workspace tvm-cuda100-forge
-docker build -t tvm-cuda92-forge $condadir -f $condadir/Dockerfile.cuda92
-docker run --rm -v $srcdir:/workspace tvm-cuda92-forge
-sudo chown -R `whoami` $condadir/pkg
+packages:
+	conda build tvm-libs
+	conda build tvm
+	conda build topi
+	conda built nnvm
+6 −3		CMakeLists.txt
+2 −0		apps/from_numpy/Makefile
+103 −0		apps/from_numpy/main.py
+45 −0		apps/from_numpy/numpy_dlpack.c
+8 −9		CMakeLists.txt
+11 −0		include/dmlc/base.h
+93 −71		include/dmlc/logging.h
+3 −0		include/dmlc/omp.h
+13 −0		include/dmlc/parameter.h
+1 −1		scripts/lint.py
+2 −2		scripts/packages.mk
+25 −0		src/data/libsvm_parser.h
+7 −4		src/io/filesys.cc
+34 −13		test/unittest/unittest_parser.cc
+1 −1		tracker/dmlc-submit
+1 −1		tracker/dmlc_tracker/kubernetes.py
+1 −1		tracker/dmlc_tracker/launcher.py
+1 −1		tracker/dmlc_tracker/mesos.py
+1 −1		tracker/dmlc_tracker/ssh.py
+0 −1		tracker/yarn/src/main/java/org/apache/hadoop/yarn/dmlc/ApplicationMaster.java