Merge branch 'develop' into PolynomialDecay

.flake8

@@ -7,9 +7,6 @@ exclude =
     # Exclude third-party libraries
     ./third_party/**,
     ./python/paddle/utils/gast/**,
-    # Temporarily ignore CINN files, it will fix later
-    ./python/cinn/**,
-    ./test/cinn/**,
 ignore =
     # Whitespace before ‘,’, ‘;’, or ‘:’, it is not compatible with black
     E203,
@@ -30,3 +27,21 @@ ignore =
 per-file-ignores =
     # These files need tabs for testing.
     test/dygraph_to_static/test_error.py:E101,W191
+    # Temporarily ignore CINN files, it will fix later
+    python/cinn/**:
+        E265,
+    test/cinn/**:
+        E126,
+        E231,
+        E251,
+        E265,
+        E266,
+        E401,
+        E711,
+        W291,
+        W504,
+    paddle/cinn/**:
+        E265,
+    tools/cinn/**:
+        E265,
+        E401,

.pre-commit-config.yaml

@@ -98,6 +98,11 @@ repos:
             - --extensions=c,cc,cxx,cpp,cu,cuh,h,hpp,hxx,kps
             - --filter=-readability/fn_size,-build/include_what_you_use,-build/c++11,-whitespace/parens
             - --quiet
+        exclude:  |
+            (?x)^(
+                paddle/cinn/.+|
+                test/cpp/cinn/.+
+            )$
 # For CMake files
 -   repo: local
     hooks:

CMakeLists.txt

@@ -240,7 +240,6 @@ else()
   )
 endif()
 
-
 find_package(Threads REQUIRED)
 
 include(simd)
@@ -429,6 +428,14 @@ if(NOT WITH_XPU AND WITH_XPU_XFT)
       CACHE STRING "Enable WITH_XPU when compiling with WITH_XPU_XFT" FORCE)
 endif()
 
+if(NOT WITH_XPU AND WITH_XPTI)
+  message(
+    WARNING "Disable XPTI when compiling without XPU. Force WITH_XPTI=OFF.")
+  set(WITH_XPTI
+      OFF
+      CACHE STRING "Disable XPTI when compiling without XPU" FORCE)
+endif()
+
 if(NOT WITH_XPU AND WITH_XPU_BKCL)
   message(
     WARNING "Disable BKCL when compiling without XPU. Force WITH_XPU_BKCL=OFF.")
@@ -575,15 +582,11 @@ include(flags) # set paddle compile flags
 
 #------------- cinn cmake config start --------------
 
-set(WITH_MKL_CBLAS ${WITH_MKL})
-set(WITH_CUDA ${WITH_GPU})
-set(WITH_CUDNN ${WITH_GPU})
 if(WITH_CINN)
   message(STATUS "Compile Paddle with CINN.")
-  include(cmake/cinn.cmake)
-  add_definitions(-DPADDLE_WITH_CINN)
   # TODO(6clc): Use CINN_WITH_CUDNN to completely replace WITH_CUDNN in CINN.
   #             Use WITH_GPU to completely replace WITH_CUDA in CINN.
+  set(WITH_MKL_CBLAS ${WITH_MKL})
   if(WITH_GPU)
     set(WITH_CUDA ${WITH_GPU})
     add_definitions(-DCINN_WITH_CUDA)
@@ -592,6 +595,8 @@ if(WITH_CINN)
       add_definitions(-DCINN_WITH_CUDNN)
     endif()
   endif()
+  include(cmake/cinn.cmake)
+  add_definitions(-DPADDLE_WITH_CINN)
 
   if(CINN_ONLY)
     if(WITH_PYTHON)

cmake/cinn.cmake

@@ -3,18 +3,25 @@ set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 set(DOWNLOAD_MODEL_DIR "${CINN_THIRD_PARTY_PATH}/model")
 
 string(REGEX MATCH "-std=(c\\+\\+[^ ]+)" STD_FLAG "${CMAKE_CXX_FLAGS}")
-if (NOT STD_FLAG)
-  if (NOT CMAKE_CXX_STANDARD)
-    message(STATUS "STD_FLAG and CMAKE_CXX_STANDARD not found, using default flag: -std=c++17")
+if(NOT STD_FLAG)
+  if(NOT CMAKE_CXX_STANDARD)
+    message(
+      STATUS
+        "STD_FLAG and CMAKE_CXX_STANDARD not found, using default flag: -std=c++17"
+    )
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++17")
     set(CMAKE_CXX_STANDARD 17)
   else()
-    message(STATUS "Got CMAKE_CXX_STANDARD=${CMAKE_CXX_STANDARD}, append -std=c++${CMAKE_CXX_STANDARD} to CMAKE_CXX_FLAGS")
+    message(
+      STATUS
+        "Got CMAKE_CXX_STANDARD=${CMAKE_CXX_STANDARD}, append -std=c++${CMAKE_CXX_STANDARD} to CMAKE_CXX_FLAGS"
+    )
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++${CMAKE_CXX_STANDARD}")
   endif()
 else()
   string(REGEX MATCH "([0-9]+)" STD_VALUE "${STD_FLAG}")
-  message(STATUS "Got STD_FLAG=${STD_FLAG}, set CMAKE_CXX_STANDARD=${STD_VALUE}")
+  message(
+    STATUS "Got STD_FLAG=${STD_FLAG}, set CMAKE_CXX_STANDARD=${STD_VALUE}")
   set(CMAKE_CXX_STANDARD ${STD_VALUE})
 endif()
 
@@ -34,7 +41,6 @@ if(WITH_DEBUG)
   add_definitions(-DCINN_WITH_DEBUG)
 endif()
 
-
 # TODO(zhhsplendid): CINN has lots of warnings during early development.
 # They will be treated as errors under paddle. We set no-error now and we will
 # clean the code in the future.
@@ -43,13 +49,15 @@ add_definitions(-w)
 include(cmake/cinn/version.cmake)
 # include the customized configures
 if(NOT EXISTS ${CMAKE_BINARY_DIR}/cmake/cinn/config.cmake)
-  file(COPY ${PROJECT_SOURCE_DIR}/cmake/cinn/config.cmake DESTINATION ${CMAKE_BINARY_DIR}/cmake/cinn)
+  file(COPY ${PROJECT_SOURCE_DIR}/cmake/cinn/config.cmake
+       DESTINATION ${CMAKE_BINARY_DIR}/cmake/cinn)
 endif()
 include(${CMAKE_BINARY_DIR}/cmake/cinn/config.cmake)
 
 if(WITH_MKL)
   generate_dummy_static_lib(LIB_NAME "cinn_mklml" GENERATOR "mklml.cmake")
   target_link_libraries(cinn_mklml ${MKLML_LIB} ${MKLML_IOMP_LIB})
+  add_dependencies(cinn_mklml ${MKLML_PROJECT})
   add_definitions(-DCINN_WITH_MKL_CBLAS)
 endif()
 if(WITH_MKLDNN)
@@ -59,8 +67,10 @@ endif()
 if(WITH_GPU)
   message(STATUS "Enable CINN CUDA")
   add_definitions(-DCINN_WITH_CUDA)
-  message(STATUS "Enable CINN CUDNN")
-  add_definitions(-DCINN_WITH_CUDNN)
+  if(WITH_CUDNN)
+    message(STATUS "Enable CINN CUDNN")
+    add_definitions(-DCINN_WITH_CUDNN)
+  endif()
   enable_language(CUDA)
   find_package(CUDA REQUIRED)
   include_directories(${CUDA_INCLUDE_DIRS})
@@ -81,10 +91,14 @@ if(WITH_GPU)
 
   find_library(CUDASTUB libcuda.so HINTS ${CUDA_TOOLKIT_ROOT_DIR}/lib64/stubs/
                                          REQUIRED)
-  find_library(CUBLAS libcublas.so HINTS ${CUDA_TOOLKIT_ROOT_DIR}/lib64 /usr/lib /usr/lib64 REQUIRED)
-  find_library(CUDNN libcudnn.so HINTS ${CUDA_TOOLKIT_ROOT_DIR}/lib64 /usr/lib /usr/lib64 REQUIRED)
-  find_library(CURAND libcurand.so HINTS ${CUDA_TOOLKIT_ROOT_DIR}/lib64 /usr/lib /usr/lib64 REQUIRED)
-  find_library(CUSOLVER libcusolver.so HINTS ${CUDA_TOOLKIT_ROOT_DIR}/lib64 /usr/lib /usr/lib64 REQUIRED)
+  find_library(CUBLAS libcublas.so HINTS ${CUDA_TOOLKIT_ROOT_DIR}/lib64
+                                         /usr/lib /usr/lib64 REQUIRED)
+  find_library(CUDNN libcudnn.so HINTS ${CUDA_TOOLKIT_ROOT_DIR}/lib64 /usr/lib
+                                       /usr/lib64 REQUIRED)
+  find_library(CURAND libcurand.so HINTS ${CUDA_TOOLKIT_ROOT_DIR}/lib64
+                                         /usr/lib /usr/lib64 REQUIRED)
+  find_library(CUSOLVER libcusolver.so HINTS ${CUDA_TOOLKIT_ROOT_DIR}/lib64
+                                             /usr/lib /usr/lib64 REQUIRED)
 endif()
 
 set(cinnapi_src CACHE INTERNAL "" FORCE)
@@ -108,7 +122,7 @@ include(cmake/cinn/external/openmp.cmake)
 include(cmake/cinn/external/jitify.cmake)
 
 if(CINN_ONLY)
-  LINK_LIBRARIES(gflags)
+  link_libraries(gflags)
 endif()
 
 set(LINK_FLAGS
@@ -269,15 +283,18 @@ if(PUBLISH_LIBS)
     POST_BUILD
     COMMAND cmake -E copy ${CMAKE_BINARY_DIR}/libcinnapi.so
             ${CMAKE_BINARY_DIR}/dist/cinn/lib/libcinnapi.so
-            COMMAND cmake -E copy_directory ${CINN_THIRD_PARTY_PATH}/install
+    COMMAND cmake -E copy_directory ${CINN_THIRD_PARTY_PATH}/install
             ${CMAKE_BINARY_DIR}/dist/third_party DEPENDS cinnapi)
   add_custom_command(
     TARGET cinncore_static
     POST_BUILD
-    COMMAND cmake -E copy ${PROJECT_SOURCE_DIR}/tools/cinn/tutorials_demo/demo.cc
-            ${CMAKE_BINARY_DIR}/dist/demo.cc
-    COMMAND cmake -E copy ${PROJECT_SOURCE_DIR}/tools/cinn/tutorials_demo/build_demo.sh
-            ${CMAKE_BINARY_DIR}/dist/build_demo.sh
+    COMMAND
+      cmake -E copy ${PROJECT_SOURCE_DIR}/tools/cinn/tutorials_demo/demo.cc
+      ${CMAKE_BINARY_DIR}/dist/demo.cc
+    COMMAND
+      cmake -E copy
+      ${PROJECT_SOURCE_DIR}/tools/cinn/tutorials_demo/build_demo.sh
+      ${CMAKE_BINARY_DIR}/dist/build_demo.sh
     COMMAND cmake -E copy ${CMAKE_BINARY_DIR}/libcinncore_static.a
             ${CMAKE_BINARY_DIR}/dist/cinn/lib/libcinncore_static.a
     COMMAND

cmake/cinn/external/absl.cmake

@@ -63,6 +63,9 @@ set(ABSL_LIB_NAMES
     bad_optional_access
     bad_variant_access
     raw_hash_set)
+if(CINN_ONLY)
+  list(APPEND ABSL_LIB_NAMES strings_internal raw_logging_internal)
+endif()
 set(ABSL_LIBS "")
 
 add_library(absl STATIC IMPORTED GLOBAL)

cmake/cupti.cmake

@@ -2,9 +2,15 @@ if(NOT WITH_GPU AND NOT WITH_ROCM)
   return()
 endif()
 
-set(CUPTI_ROOT
-    "/usr"
-    CACHE PATH "CUPTI ROOT")
+if(WITH_ROCM)
+  set(CUPTI_ROOT
+      "${ROCM_PATH}/CUPTI"
+      CACHE PATH "CUPTI ROOT")
+else()
+  set(CUPTI_ROOT
+      "/usr"
+      CACHE PATH "CUPTI ROOT")
+endif()
 find_path(
   CUPTI_INCLUDE_DIR cupti.h
   PATHS ${CUPTI_ROOT}

cmake/external/gtest.cmake

@@ -56,14 +56,9 @@ else()
       "${GTEST_INSTALL_DIR}/${CMAKE_INSTALL_LIBDIR}/libgmock.a"
       CACHE FILEPATH "gmock libraries." FORCE)
   set(GTEST_CMAKE_C_FLAGS "${CMAKE_C_FLAGS}")
-  if(CINN_ONLY)
-    set(GTEST_CMAKE_CXX_FLAGS "-std=c++17 ${CMAKE_CXX_FLAGS}")
-  else()
-    set(GTEST_CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}")
-  endif()
+  set(GTEST_CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}")
 endif()
 
-
 if(WITH_MKLML)
   # wait for mklml downloading completed
   set(GTEST_DEPENDS ${MKLML_PROJECT})

cmake/external/xpu.cmake

@@ -10,7 +10,7 @@ set(XPU_XFT_LIB_NAME "libxft.so")
 set(XPU_XPTI_LIB_NAME "libxpti.so")
 
 if(NOT DEFINED XPU_BASE_DATE)
-  set(XPU_BASE_DATE "20230602")
+  set(XPU_BASE_DATE "20230620")
 endif()
 set(XPU_XCCL_BASE_VERSION "1.0.49.2")
 if(NOT DEFINED XPU_XFT_BASE_VERSION)
@@ -137,7 +137,7 @@ ExternalProject_Add(
     pack_paddle_depence.sh ${XPU_XRE_URL} ${XPU_XRE_DIR_NAME} ${XPU_XDNN_URL}
     ${XPU_XDNN_DIR_NAME} ${XPU_XCCL_URL} ${XPU_XCCL_DIR_NAME} && wget
     ${XPU_XFT_GET_DEPENCE_URL} && bash get_xft_dependence.sh ${XPU_XFT_URL}
-    ${XPU_XFT_DIR_NAME} [ -n "$WITH_XPTI" ] && bash
+    ${XPU_XFT_DIR_NAME} && bash
     ${CMAKE_SOURCE_DIR}/tools/xpu/get_xpti_dependence.sh ${XPU_XPTI_URL}
     ${XPU_XPTI_DIR_NAME}
   DOWNLOAD_NO_PROGRESS 1

cmake/hip.cmake

@@ -106,7 +106,11 @@ list(APPEND HIP_CXX_FLAGS -Wno-duplicate-decl-specifier)
 list(APPEND HIP_CXX_FLAGS -Wno-implicit-int-float-conversion)
 list(APPEND HIP_CXX_FLAGS -Wno-pass-failed)
 list(APPEND HIP_CXX_FLAGS -DTHRUST_DEVICE_SYSTEM=THRUST_DEVICE_SYSTEM_HIP)
-list(APPEND HIP_CXX_FLAGS -std=c++14)
+if(WITH_CINN)
+  list(APPEND HIP_CXX_FLAGS -std=c++14)
+else()
+  list(APPEND HIP_CXX_FLAGS -std=c++17)
+endif()
 
 if(CMAKE_BUILD_TYPE MATCHES Debug)
   list(APPEND HIP_CXX_FLAGS -g2)

cmake/third_party.cmake

@@ -263,6 +263,7 @@ endif()
 
 # cinn_only includes third-party libraries separately
 if(CINN_ONLY)
+  set(CMAKE_CXX_FLAGS "-std=c++17 ${CMAKE_CXX_FLAGS}")
   include(external/zlib)
   include(external/gflags)
   include(external/glog)
@@ -289,7 +290,6 @@ if(WITH_CINN)
   endif()
 endif()
 
-
 include(external/zlib) # download, build, install zlib
 include(external/gflags) # download, build, install gflags
 include(external/glog) # download, build, install glog

paddle/cinn/backends/llvm/codegen_llvm.cc

@@ -1086,9 +1086,7 @@ llvm::Value *CodeGenLLVM::Visit(const ir::Sum *op) {
 
 #undef __IR_EMITTER_CINN_NOT_IMPLEMENTED
 
-void CodeGenLLVM::Compile(const ir::Module &module) {
-  Visit(module.self());
-}
+void CodeGenLLVM::Compile(const ir::Module &module) { Visit(module.self()); }
 
 llvm::Value *CodeGenLLVM::EmitCall_buffer_malloc(const ir::Call *op) { return nullptr; }
 

paddle/cinn/backends/llvm/simple_jit.cc

@@ -111,25 +111,18 @@ SimpleJIT::SimpleJIT() : context_(std::make_unique<llvm::LLVMContext>()) {
 
 template <typename CodeGenT>
 void SimpleJIT::Link(ir::Module module, bool optimize) {
-  VLOG(-1) << "dddddd";
   std::string runtime_ir(backends::kRuntimeLlvmIr);
   llvm::SMDiagnostic error;
   auto m = llvm::parseAssemblyString(runtime_ir, error, context());
   m->setDataLayout(jit_->getDataLayout());
   auto b = std::make_unique<llvm::IRBuilder<>>(context());
 
   auto ir_emitter = std::make_unique<CodeGenT>(m.get(), b.get());
-  VLOG(-1) << "dddddd";
   ir_emitter->Compile(module);
-  VLOG(-1) << "dddddd";
 
-  VLOG(-1) << "dddddd";
   CHECK(!llvm::verifyModule(*m, &llvm::errs())) << "Invalid module found";
-  VLOG(-1) << "dddddd";
 
-  VLOG(-1) << "dddddd";
   AddModule(std::move(m), optimize);
-  VLOG(-1) << "dddddd";
 }
 
 template void SimpleJIT::Link<CodeGenLLVM>(ir::Module module, bool optimize);