Cinn trivalop fuse

.github/PULL_REQUEST_TEMPLATE.md

@@ -1,9 +1,13 @@
+<!-- TemplateReference: https://github.com/PaddlePaddle/Paddle/wiki/PULL-REQUEST-TEMPLATE--REFERENCE -->
 <!-- Demo: https://github.com/PaddlePaddle/Paddle/pull/24810 -->
-### PR types
-<!-- One of [ New features | Bug fixes | Function optimization | Performance optimization | Breaking changes | Others ] -->
 
-### PR changes
-<!-- One of [ OPs | APIs | Docs | Others ] -->
+### PR Category
+<!-- One of [ User Experience | Execute Infrastructure | Operator Mechanism | CINN | Custom Device | Performance Optimization | Distributed Strategy | Parameter Server | Communication Library | Auto Parallel | Inference | Environment Adaptation | Others ] -->
+
+
+### PR Types
+<!-- One of [ New features | Bug fixes | Improvements | Performance | BC Breaking | Deprecations | Docs | Devs | Not User Facing | Security | Deprecations | Others ] -->
+
 
 ### Description
 <!-- Describe what you’ve done -->

.pre-commit-config.yaml

@@ -36,7 +36,7 @@ repos:
         # Exclude some unit test files that require tabs.
         exclude: |
             (?x)^(
-                test/dygraph_to_static/test_legacy_error.py
+                test/dygraph_to_static/test_error.py
             )$
 -   repo: local
     hooks:
@@ -56,13 +56,8 @@ repos:
     hooks:
     -   id: black
         files: (.*\.(py|pyi|bzl)|BUILD|.*\.BUILD|WORKSPACE)$
--   repo: https://github.com/PyCQA/flake8
-    rev: 5.0.4
-    hooks:
-    -   id: flake8
-        args: ["--config=.flake8"]
 -   repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.2.0
+    rev: v0.3.0
     hooks:
     -   id: ruff
         args: [--fix, --exit-non-zero-on-fix, --no-cache]

CMakeLists.txt

@@ -63,9 +63,11 @@ option(WITH_ONNXRUNTIME "Compile PaddlePaddle with ONNXRUNTIME" OFF)
 option(WITH_CUSPARSELT "Compile PaddlePaddle with CUSPARSELT" OFF)
 option(WITH_SETUP_INSTALL "Compile PaddlePaddle with setup.py" OFF)
 option(WITH_SHARED_PHI "Compile PaddlePaddle with SHARED LIB of PHI" ON)
-option(CINN_ONLY "Compile CINN only in Paddle" OFF)
 option(CINN_WITH_CUDNN "Compile CINN with CUDNN support" ON)
-
+option(WITH_PIP_CUDA_LIBRARIES
+       "Paddle uses the CUDA library provided by NVIDIA" OFF)
+option(WITH_NIGHTLY_BUILD
+       "Compile nightly paddle whl package of the develop branch" OFF)
 find_package(Git REQUIRED)
 
 # config GIT_URL with github mirrors to speed up dependent repos clone
@@ -97,11 +99,16 @@ endif()
 
 if(WITH_GPU AND NOT APPLE)
   #(Note risemeup1): The cudart dynamic library libcudart.so is used by set CUDA_USE_STATIC_CUDA_RUNTIME and CMAKE_CUDA_FLAGS
-  if(LINUX)
+  if(CMAKE_SYSTEM_NAME STREQUAL "Linux" AND CMAKE_SYSTEM_PROCESSOR STREQUAL
+                                            "x86_64")
     set(CUDA_USE_STATIC_CUDA_RUNTIME
         OFF
         CACHE BOOL "" FORCE)
     set(CMAKE_CUDA_FLAGS "--cudart shared")
+    if(WITH_PIP_CUDA_LIBRARIES)
+      #(Note risemeup1): Flag 'WITH_PIP_CUDA_LIBRARIES' will be used in dynamic_loader.cc to search for CUDA-related .so files through the Python libraries provided by NVIDIA.
+      add_definitions(-DWITH_PIP_CUDA_LIBRARIES)
+    endif()
   endif()
   enable_language(CUDA)
   message(STATUS "CUDA compiler: ${CMAKE_CUDA_COMPILER}, version: "
@@ -135,7 +142,10 @@ endif()
 if(WIN32)
   option(MSVC_STATIC_CRT "use static C Runtime library by default" ON)
   message("Build static library of PHI")
-  set(CMAKE_SUPPRESS_REGENERATION ON)
+  # (Note xuxinyi04): If CMAKE_SUPPRESS_REGENERATION is OFF, which is default, then CMake adds a
+  # special target on which all other targets depend that checks the build system and optionally
+  # re-runs CMake to regenerate the build system when the target specification source changes.
+  set(CMAKE_SUPPRESS_REGENERATION OFF)
   set(CMAKE_STATIC_LIBRARY_PREFIX lib)
   set(WITH_SHARED_PHI
       OFF
@@ -233,6 +243,8 @@ if(WIN32)
         "${${flag_var}} /ignore:4049 /ignore:4217 /ignore:4006 /ignore:4221")
     if(MSVC_STATIC_CRT)
       set(${flag_var} "${${flag_var}} /NODEFAULTLIB:MSVCRT.LIB")
+    else()
+      set(${flag_var} "${${flag_var}} /NODEFAULTLIB:LIBCMT.LIB")
     endif()
   endforeach()
 
@@ -618,18 +630,6 @@ if(WITH_CINN)
 
   include(cmake/cinn.cmake)
   add_definitions(-DPADDLE_WITH_CINN)
-
-  if(CINN_ONLY)
-    add_definitions(-DCINN_WITH_ONLY)
-    if(WITH_PYTHON)
-      add_subdirectory(python)
-    endif()
-    add_subdirectory(test)
-    if(NOT WITH_GFLAGS)
-      add_subdirectory(paddle/utils)
-    endif()
-    return()
-  endif()
 endif()
 
 #------------- cinn cmake config end --------------

cmake/ccache.cmake

@@ -11,8 +11,9 @@ if(NOT WIN32)
     # show statistics summary of ccache
     message("ccache version\t\t\t    " ${ccache_version} "\n"
             ${cache_directory})
-    set_property(GLOBAL PROPERTY RULE_LAUNCH_COMPILE ${CCACHE_PATH})
-    set_property(GLOBAL PROPERTY RULE_LAUNCH_LINK ${CCACHE_PATH})
+    set(CMAKE_C_COMPILER_LAUNCHER ${CCACHE_PATH})
+    set(CMAKE_CXX_COMPILER_LAUNCHER ${CCACHE_PATH})
+    set(CMAKE_CUDA_COMPILER_LAUNCHER ${CCACHE_PATH})
   endif()
 elseif("${CMAKE_GENERATOR}" STREQUAL "Ninja")
   # (Note:zhouwei25) Only Ninja Generator can support sccache now

cmake/cinn.cmake

@@ -164,13 +164,13 @@ cinn_cc_library(
   isl
   ginac
   pybind
+  group_cluster
+  cinn_op_dialect
   ${jitify_deps})
 add_dependencies(cinnapi GEN_LLVM_RUNTIME_IR_HEADER ZLIB::ZLIB)
 add_dependencies(cinnapi GEN_LLVM_RUNTIME_IR_HEADER ${core_deps})
-if(NOT CINN_ONLY)
-  target_link_libraries(cinnapi op_dialect pir phi)
-  add_dependencies(cinnapi op_dialect pir phi)
-endif()
+target_link_libraries(cinnapi op_dialect pir phi)
+add_dependencies(cinnapi op_dialect pir phi)
 
 target_link_libraries(cinnapi ${PYTHON_LIBRARIES})
 
@@ -183,11 +183,6 @@ if(WITH_MKL)
   endif()
 endif()
 
-if(CINN_ONLY)
-  target_link_libraries(cinnapi common)
-  add_dependencies(cinnapi common)
-endif()
-
 if(WITH_GPU)
   target_link_libraries(
     cinnapi
@@ -227,15 +222,17 @@ function(gen_cinncore LINKTYPE)
     schedule_desc_proto
     absl
     isl
-    ginac)
+    ginac
+    pybind
+    group_cluster
+    cinn_op_dialect
+    ${jitify_deps})
   add_dependencies(${CINNCORE_TARGET} GEN_LLVM_RUNTIME_IR_HEADER ZLIB::ZLIB)
   add_dependencies(${CINNCORE_TARGET} GEN_LLVM_RUNTIME_IR_HEADER ${core_deps})
-  if(NOT CINN_ONLY)
-    target_link_libraries(${CINNCORE_TARGET} op_dialect pir phi)
-    add_dependencies(${CINNCORE_TARGET} op_dialect pir phi)
-  endif()
+  target_link_libraries(${CINNCORE_TARGET} op_dialect pir phi)
+  add_dependencies(${CINNCORE_TARGET} op_dialect pir phi)
 
-  add_dependencies(${CINNCORE_TARGET} pybind)
+  # add_dependencies(${CINNCORE_TARGET} pybind)
   target_link_libraries(${CINNCORE_TARGET} ${PYTHON_LIBRARIES})
 
   if(WITH_MKL)
@@ -247,11 +244,6 @@ function(gen_cinncore LINKTYPE)
     endif()
   endif()
 
-  if(CINN_ONLY)
-    target_link_libraries(${CINNCORE_TARGET} common)
-    add_dependencies(${CINNCORE_TARGET} common)
-  endif()
-
   if(WITH_GPU)
     target_link_libraries(
       ${CINNCORE_TARGET}
@@ -261,16 +253,16 @@ function(gen_cinncore LINKTYPE)
       ${CUBLAS}
       ${CUDNN}
       ${CURAND}
-      ${CUSOLVER}
-      ${jitify_deps})
+      ${CUSOLVER})
+    # ${jitify_deps})
     if(NVTX_FOUND)
       target_link_libraries(${CINNCORE_TARGET} ${CUDA_NVTX_LIB})
     endif()
   endif()
 
   if(WITH_CUTLASS)
-    target_link_libraries(cinnapi cutlass)
-    add_dependencies(cinnapi cutlass)
+    target_link_libraries(${CINNCORE_TARGET} cutlass)
+    add_dependencies(${CINNCORE_TARGET} cutlass)
   endif()
 endfunction()
 

cmake/coveralls.cmake

@@ -60,8 +60,8 @@ endfunction()
 
 if(WITH_COVERAGE)
   if(WITH_INCREMENTAL_COVERAGE)
-    # if *.h changed, generate coverage report totaly.
-    # if pybind.cc changed, generate coverage report totaly.
+    # if *.h changed, generate coverage report totally.
+    # if pybind.cc changed, generate coverage report totally.
     # Because if pybind.cc add '-g -O0 -fprofile-arcs -ftest-coverage' only, some testcase will fail.
     if((NOT ("$ENV{PADDLE_GIT_DIFF_H_FILE}" STREQUAL ""))
        OR ("$ENV{PADDLE_GIT_DIFF_CC_FILE}" MATCHES "pybind.cc"))

cmake/coverallsGcovJsons.cmake

@@ -248,7 +248,7 @@ foreach(GCOV_FILE ${GCOV_FILES})
   # Instead of trying to parse the source from the
   # gcov file, simply read the file contents from the source file.
   # (Parsing it from the gcov is hard because C-code uses ; in many places
-  #  which also happens to be the same as the CMake list delimeter).
+  #  which also happens to be the same as the CMake list delimiter).
   file(READ ${GCOV_SRC_PATH} GCOV_FILE_SOURCE)
 
   string(REPLACE "\\" "\\\\" GCOV_FILE_SOURCE "${GCOV_FILE_SOURCE}")

cmake/cuda.cmake

@@ -294,7 +294,7 @@ select_nvcc_arch_flags(NVCC_FLAGS_EXTRA NVCC_ARCH_BIN)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} ${NVCC_FLAGS_EXTRA}")
 message(STATUS "NVCC_FLAGS_EXTRA: ${NVCC_FLAGS_EXTRA}")
 
-# Set C++14 support
+# Set C++17 support
 set(CUDA_PROPAGATE_HOST_FLAGS OFF)
 # Release/Debug flags set by cmake. Such as -O3 -g -DNDEBUG etc.
 # So, don't set these flags here.

cmake/experiments/cuda_module_loading_lazy.cmake

@@ -13,7 +13,7 @@
 # limitations under the License.
 
 # this file contains experimental build options for lazy cuda module loading
-# cuda moduel lazy loading is supported by CUDA 11.7+
+# cuda module lazy loading is supported by CUDA 11.7+
 # this experiment option makes Paddle supports lazy loading before CUDA 11.7.
 
 if(LINUX)

cmake/phi_header.cmake → cmake/export_paddle_header.cmake

@@ -15,33 +15,57 @@
 set(PADDLE_INFERENCE_INSTALL_DIR
     "${CMAKE_BINARY_DIR}/paddle_inference_install_dir")
 
-function(phi_header_path_compat TARGET_PATH)
-  message(STATUS "phi header path compat processing: ${TARGET_PATH}")
+function(header_path_compat TARGET_PATH)
+  message(STATUS "header path compat processing: ${TARGET_PATH}")
   file(GLOB HEADERS "${TARGET_PATH}/*" "*.h")
   foreach(header ${HEADERS})
     if(${header} MATCHES ".*.h$")
       file(READ ${header} HEADER_CONTENT)
       string(REPLACE "paddle/fluid/platform/" "paddle/phi/" HEADER_CONTENT
                      "${HEADER_CONTENT}")
+      string(REPLACE "paddle/pir/include/" "paddle/pir/" HEADER_CONTENT
+                     "${HEADER_CONTENT}")
+      string(REPLACE "paddle/fluid/pir/drr/include/" "paddle/pir/drr/"
+                     HEADER_CONTENT "${HEADER_CONTENT}")
+      string(REPLACE "paddle/fluid/pir/utils/" "paddle/pir/utils/"
+                     HEADER_CONTENT "${HEADER_CONTENT}")
       file(WRITE ${header} "${HEADER_CONTENT}")
-      message(STATUS "phi header path compat processing complete: ${header}")
+      message(STATUS "header path compat processing complete: ${header}")
     endif()
   endforeach()
 endfunction()
 
-phi_header_path_compat(${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle)
-phi_header_path_compat(
-  ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle/phi)
-phi_header_path_compat(
+header_path_compat(${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle)
+header_path_compat(${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle/phi)
+header_path_compat(
   ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle/phi/api)
-phi_header_path_compat(
+header_path_compat(
   ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle/phi/api/ext)
-phi_header_path_compat(
+header_path_compat(
   ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle/phi/api/include)
-phi_header_path_compat(
+header_path_compat(
   ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle/phi/common)
-phi_header_path_compat(
+header_path_compat(
   ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle/phi/core)
+header_path_compat(
+  ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle/pir/core)
+header_path_compat(
+  ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle/pir/core/parser)
+header_path_compat(
+  ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle/pir/dialect/control_flow/ir
+)
+header_path_compat(
+  ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle/pir/dialect/shape/ir)
+header_path_compat(
+  ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle/pir/dialect/shape/utils)
+header_path_compat(
+  ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle/pir/drr)
+header_path_compat(
+  ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle/pir/pass)
+header_path_compat(
+  ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle/pir/pattern_rewrite)
+header_path_compat(
+  ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle/pir/utils)
 
 # NOTE(liuyuanle): In inference lib, no need include paddle/utils/pybind.h, so we delete this.
 file(READ ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/paddle/extension.h

cmake/external/cccl.cmake

@@ -15,12 +15,18 @@ set(CCCL_INCLUDE_DIR ${CCCL_SOURCE_DIR})
 message("CCCL_INCLUDE_DIR is ${CCCL_INCLUDE_DIR}")
 include_directories(${CCCL_INCLUDE_DIR})
 
+file(TO_NATIVE_PATH ${PADDLE_SOURCE_DIR}/patches/cccl/util_device.cuh.patch
+     native_src)
+set(CCCL_PATCH_COMMAND git checkout -- . && git checkout ${CCCL_TAG} && patch
+                       -p1 -Nd ${CCCL_SOURCE_DIR} < ${native_src})
+
 ExternalProject_Add(
   extern_cccl
   ${EXTERNAL_PROJECT_LOG_ARGS}
   SOURCE_DIR ${CCCL_SOURCE_DIR}
   PREFIX ${CCCL_PREFIX_DIR}
   UPDATE_COMMAND ""
+  PATCH_COMMAND ${CCCL_PATCH_COMMAND}
   CONFIGURE_COMMAND ""
   BUILD_COMMAND ""
   INSTALL_COMMAND ""

cmake/external/dirent.cmake

@@ -27,7 +27,9 @@ if((NOT DEFINED DIRENT_NAME) OR (NOT DEFINED DIRENT_URL))
   set(DIRENT_URL
       "${GIT_URL}/tronkko/dirent/archive/refs/tags/1.23.2.tar.gz"
       CACHE STRING "" FORCE)
-  set(DIRENT_CACHE_FILENAME "1.23.2.tar.gz")
+  set(DIRENT_CACHE_FILENAME
+      "1.23.2.tar.gz"
+      CACHE STRING "" FORCE)
 endif()
 
 message(STATUS "DIRENT_NAME: ${DIRENT_NAME}, DIRENT_URL: ${DIRENT_URL}")