Merge branch 'PaddlePaddle:develop' into C18

.clang-tidy

@@ -11,9 +11,9 @@ bugprone-copy-constructor-init,
 bugprone-exception-escape,
 -bugprone-fold-init-type,
 -bugprone-forwarding-reference-overload,
--bugprone-inaccurate-erase,
+bugprone-inaccurate-erase,
 bugprone-incorrect-roundings,
--bugprone-infinite-loop,
+bugprone-infinite-loop,
 bugprone-integer-division,
 -bugprone-macro-repeated-side-effects,
 -bugprone-misplaced-operator-in-strlen-in-alloc,
@@ -28,7 +28,7 @@ bugprone-signed-char-misuse,
 -bugprone-sizeof-container,
 -bugprone-sizeof-expression,
 -bugprone-string-constructor,
--bugprone-string-integer-assignment,
+bugprone-string-integer-assignment,
 -bugprone-string-literal-with-embedded-nul,
 -bugprone-suspicious-enum-usage,
 -bugprone-suspicious-memset-usage,
@@ -123,7 +123,7 @@ clang-analyzer-optin.portability.UnixAPI,
 -clang-analyzer-osx.coreFoundation.CFRetainRelease,
 -clang-analyzer-osx.coreFoundation.containers.OutOfBounds,
 -clang-analyzer-osx.coreFoundation.containers.PointerSizedValues,
--clang-analyzer-security.FloatLoopCounter,
+clang-analyzer-security.FloatLoopCounter,
 -clang-analyzer-security.insecureAPI.DeprecatedOrUnsafeBufferHandling,
 -clang-analyzer-security.insecureAPI.SecuritySyntaxChecker,
 -clang-analyzer-security.insecureAPI.UncheckedReturn,
@@ -198,7 +198,7 @@ performance-move-const-arg,
 -performance-move-constructor-init,
 -performance-no-automatic-move,
 performance-noexcept-move-constructor,
--performance-trivially-destructible,
+performance-trivially-destructible,
 -performance-type-promotion-in-math-fn,
 -performance-unnecessary-copy-initialization,
 readability-container-size-empty,

.gitignore

@@ -107,4 +107,3 @@ paddle/phi/kernels/fusion/cutlass/cutlass_kernels/fpA_intB_gemm/autogen/*
 paddle/phi/kernels/fusion/cutlass/cutlass_kernels/fpA_intB_gemm/autogen_tmp/*
 paddle/fluid/pybind/static_op_function.*
 paddle/fluid/pybind/ops_api.cc
-paddle/fluid/pir/drr/src/*_op_factory_generated.*

.gitmodules

@@ -110,3 +110,11 @@
 	path = third_party/cccl
 	url = https://github.com/NVIDIA/cccl.git
 	ignore = dirty
+[submodule "third_party/cryptopp"]
+	path = third_party/cryptopp
+	url = https://github.com/weidai11/cryptopp.git
+	ignore = dirty
+[submodule "third_party/cryptopp-cmake"]
+	path = third_party/cryptopp-cmake
+	url = https://github.com/noloader/cryptopp-cmake.git
+	ignore = dirty

CMakeLists.txt

@@ -323,8 +323,6 @@ option(WITH_FLPS "FL PS mode" OFF)
 option(WITH_RPC "Compile with rpc support" ${WITH_DISTRIBUTE})
 option(WITH_CUDNN_FRONTEND
        "Compile with CUDNN Frontend API support (experimental)" OFF)
-option(WITH_CUDNN_DSO "Compile PaddlePaddle with cuDNN dynamic-link libraries"
-       OFF)
 option(WITH_SHARED_IR "Compile PaddlePaddle with SHARED LIB of IR" ON)
 option(WITH_NVCC_LAZY
        "Compile PaddlePaddle with nvcc lazy mode, used for CI-Inference only."
@@ -420,16 +418,6 @@ if(NOT WITH_GPU AND WITH_NCCL)
       CACHE STRING "Disable NCCL when compiling without GPU" FORCE)
 endif()
 
-if(NOT WITH_GPU AND WITH_CUDNN_DSO)
-  message(
-    WARNING
-      "Can't compile with cuDNN libraries when compiling without GPU. Force WITH_CUDNN_DSO=OFF."
-  )
-  set(WITH_CUDNN_DSO
-      OFF
-      CACHE STRING "Disable cuDNN libraries when compiling without GPU" FORCE)
-endif()
-
 # force WITH_XPU on when WITH_XPU_KP
 if(WITH_XPU_KP AND NOT WITH_XPU)
   message(

README.md

@@ -50,7 +50,7 @@ Now our developers can acquire Tesla V100 online computing resources for free. I
 
 - **High-Performance Inference Engines for Comprehensive Deployment Environments**
 
-   PaddlePaddle is not only compatible with models trained in 3rd party open-source frameworks , but also offers complete inference products for various production scenarios. Our inference product line includes [Paddle Inference](https://paddle-inference.readthedocs.io/en/master/guides/introduction/index_intro.html): Native inference library for high-performance server and cloud inference; [Paddle Serving](https://github.com/PaddlePaddle/Serving): A service-oriented framework suitable for distributed and pipeline productions; [Paddle Lite](https://github.com/PaddlePaddle/Paddle-Lite): Ultra-Lightweight inference engine for mobile and IoT environments; [Paddle.js](https://www.paddlepaddle.org.cn/paddle/paddlejs): A frontend inference engine for browser and mini-apps. Furthermore, by great amounts of optimization with leading hardware in each scenario, Paddle inference engines outperform most of the other mainstream frameworks.
+   PaddlePaddle is not only compatible with models trained in 3rd party open-source frameworks , but also offers complete inference products for various production scenarios. Our inference product line includes [Paddle Inference](https://www.paddlepaddle.org.cn/inference/master/guides/introduction/index_intro.html): Native inference library for high-performance server and cloud inference; [FastDeploy](https://github.com/PaddlePaddle/FastDeploy): Easy-to-use and High Performance AI model deployment toolkit for Cloud, Mobile and Edge without-of-the-box and unified experience; [Paddle Lite](https://github.com/PaddlePaddle/Paddle-Lite): Ultra-Lightweight inference engine for mobile and IoT environments; [Paddle.js](https://www.paddlepaddle.org.cn/paddle/paddlejs): A frontend inference engine for browser and mini-apps. Furthermore, by great amounts of optimization with leading hardware in each scenario, Paddle inference engines outperform most of the other mainstream frameworks.
 
 - **Industry-Oriented Models and Libraries with Open Source Repositories**
 

README_cn.md

@@ -48,7 +48,7 @@ PaddlePaddle用户可领取**免费Tesla V100在线算力资源**，训练模型
 
 - **支持多端多平台的高性能推理部署工具**
 
-    飞桨不仅广泛兼容第三方开源框架训练的模型部署，并且为不同的场景的生产环境提供了完备的推理引擎，包括适用于高性能服务器及云端推理的原生推理库 [Paddle Inference](https://www.paddlepaddle.org.cn/inference/product_introduction/inference_intro.html)，面向分布式、流水线生产环境下自动上云、A/B测试等高阶功能的服务化推理框架 [Paddle Serving](https://github.com/PaddlePaddle/Serving)，针对于移动端、物联网场景的轻量化推理引擎 [Paddle Lite](https://github.com/PaddlePaddle/Paddle-Lite)，以及在浏览器、小程序等环境下使用的前端推理引擎 [Paddle.js](https://www.paddlepaddle.org.cn/paddle/paddlejs)。同时，透过与不同场景下的主流硬件高度适配优化及异构计算的支持, 飞桨的推理性能也领先绝大部分的主流实现。
+    飞桨不仅广泛兼容第三方开源框架训练的模型部署，并且为不同的场景的生产环境提供了完备的推理引擎，包括适用于高性能服务器及云端推理的原生推理库 [Paddle Inference](https://www.paddlepaddle.org.cn/inference/master/guides/introduction/index_intro.html)，全场景、易用灵活、极致高效的AI推理部署工具，支持云边端部署工具 [FastDeploy](https://github.com/PaddlePaddle/FastDeploy)，针对于移动端、物联网场景的轻量化推理引擎 [Paddle Lite](https://github.com/PaddlePaddle/Paddle-Lite)，以及在浏览器、小程序等环境下使用的前端推理引擎 [Paddle.js](https://www.paddlepaddle.org.cn/paddle/paddlejs)。同时，透过与不同场景下的主流硬件高度适配优化及异构计算的支持, 飞桨的推理性能也领先绝大部分的主流实现。
 
 - **面向产业应用，开源开放覆盖多领域的工业级模型库。**
 

README_ja.md

@@ -50,7 +50,7 @@ pip install paddlepaddle-gpu
 
 - **総合的な展開環境に対応した高性能推論エンジン**
 
-   PaddlePaddle は、サードパーティのオープンソースフレームワークで学習されたモデルとの互換性があるだけでなく、様々な生産シナリオに対応した完全な推論エンジン、システム、スイートを提供しています。当社の推論エンジン、システム、スイートには、[Paddle Inference](https://paddle-inference.readthedocs.io/en/master/guides/introduction/index_intro.html) があります： [Paddle Serving](https://github.com/PaddlePaddle/Serving): 高性能なサーバーおよびクラウド推論用のネイティブ推論ライブラリ： [Paddle Serving](https://github.com/PaddlePaddle/Paddle-Lite): 分散型やパイプライン型プロダクションに適したサービス指向フレームワーク; [Paddle Lite](https://github.com/PaddlePaddle/Paddle-Lite)： モバイルや IoT 環境向けの超軽量推論エンジン; [Paddle.js](https://www.paddlepaddle.org.cn/paddle/paddlejs)： ブラウザやミニアプリのためのフロントエンド推論エンジンです。さらに、各シナリオの主要なハードウェアに最適化することで、Paddle の推論エンジンは他の主流フレームワークのほとんどを凌駕しています。
+   PaddlePaddle は、サードパーティのオープンソースフレームワークで学習されたモデルとの互換性があるだけでなく、様々な生産シナリオに対応した完全な推論エンジン、システム、スイートを提供しています。当社の推論エンジン、システム、スイートには、[Paddle Inference](https://www.paddlepaddle.org.cn/inference/master/guides/introduction/index_intro.html) があります：高性能なサーバーおよびクラウド推論用のネイティブ推論ライブラリ; [FastDeploy](https://github.com/PaddlePaddle/FastDeploy): オールシナリオで使いやすく、柔軟で非常に効率的なAI推論デプロイツールです; [Paddle Lite](https://github.com/PaddlePaddle/Paddle-Lite)： モバイルや IoT 環境向けの超軽量推論エンジン; [Paddle.js](https://www.paddlepaddle.org.cn/paddle/paddlejs)： ブラウザやミニアプリのためのフロントエンド推論エンジンです。さらに、各シナリオの主要なハードウェアに最適化することで、Paddle の推論エンジンは他の主流フレームワークのほとんどを凌駕しています。
 
 - **オープンソースリポジトリによる業界指向のモデルやライブラリ**
 

cmake/cinn/core.cmake

@@ -242,7 +242,7 @@ function(cinn_merge_static_libs TARGET_NAME)
       COMMAND ${CMAKE_COMMAND} -E touch ${target_SRCS}
       DEPENDS ${libs})
 
-    # Generate dummy staic lib
+    # Generate dummy static lib
     file(WRITE ${target_SRCS}
          "const char *dummy_${TARGET_NAME} = \"${target_SRCS}\";")
     add_library(${TARGET_NAME} STATIC ${target_SRCS})

cmake/cinn/external/absl.cmake

@@ -50,7 +50,7 @@ ExternalProject_Add(
   BUILD_BYPRODUCTS ${ABSL_INSTALL_DIR}/lib/libabsl_bad_variant_access.a
   BUILD_BYPRODUCTS ${ABSL_INSTALL_DIR}/lib/libabsl_raw_hash_set.a)
 
-# It may be more convinent if we just include all absl libs
+# It may be more convenient if we just include all absl libs
 set(ABSL_LIB_NAMES
     hash
     wyhash

cmake/cuda.cmake

@@ -82,7 +82,7 @@ function(detect_installed_gpus out_variable)
       set(CUDA_gpu_detect_output
           ${nvcc_out}
           CACHE INTERNAL
-                "Returned GPU architetures from detect_installed_gpus tool"
+                "Returned GPU architectures from detect_installed_gpus tool"
                 FORCE)
     endif()
   endif()
@@ -122,7 +122,7 @@ function(select_nvcc_arch_flags out_variable out_arch_bin)
   # set CUDA_ARCH_NAME strings (so it will be seen as dropbox in CMake-Gui)
   set(CUDA_ARCH_NAME
       ${archs_name_default}
-      CACHE STRING "Select target NVIDIA GPU achitecture.")
+      CACHE STRING "Select target NVIDIA GPU architecture.")
   set_property(CACHE CUDA_ARCH_NAME PROPERTY STRINGS "" ${archs_names})
   mark_as_advanced(CUDA_ARCH_NAME)
 
@@ -300,7 +300,7 @@ set(CUDA_PROPAGATE_HOST_FLAGS OFF)
 # So, don't set these flags here.
 set(CMAKE_CUDA_STANDARD 17)
 
-# (Note) For windows, if delete /W[1-4], /W1 will be added defaultly and conflic with -w
+# (Note) For windows, if delete /W[1-4], /W1 will be added defaultly and conflict with -w
 # So replace /W[1-4] with /W0
 if(WIN32)
   string(REGEX REPLACE "/W[1-4]" " /W0 " CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS}")

cmake/external/cryptopp.cmake

@@ -14,12 +14,13 @@
 
 include(ExternalProject)
 
+set(CRYPTOPP_SOURCE_DIR ${PADDLE_SOURCE_DIR}/third_party/cryptopp)
+set(CRYPTOPP_CMAKE_SOURCE_DIR ${PADDLE_SOURCE_DIR}/third_party/cryptopp-cmake)
 set(CRYPTOPP_PREFIX_DIR ${THIRD_PARTY_PATH}/cryptopp)
 set(CRYPTOPP_INSTALL_DIR ${THIRD_PARTY_PATH}/install/cryptopp)
 set(CRYPTOPP_INCLUDE_DIR
     "${CRYPTOPP_INSTALL_DIR}/include"
     CACHE PATH "cryptopp include directory." FORCE)
-set(CRYPTOPP_REPOSITORY ${GIT_URL}/weidai11/cryptopp.git)
 set(CRYPTOPP_TAG CRYPTOPP_8_2_0)
 
 if(WIN32)
@@ -63,17 +64,16 @@ include_directories(${CRYPTOPP_INCLUDE_DIR})
 ExternalProject_Add(
   extern_cryptopp
   ${EXTERNAL_PROJECT_LOG_ARGS} ${SHALLOW_CLONE}
-  GIT_REPOSITORY ${CRYPTOPP_REPOSITORY}
-  GIT_TAG ${CRYPTOPP_TAG}
   PREFIX ${CRYPTOPP_PREFIX_DIR}
+  SOURCE_DIR ${CRYPTOPP_SOURCE_DIR}
   UPDATE_COMMAND ""
   PATCH_COMMAND
-  COMMAND ${CMAKE_COMMAND} -E remove_directory "<SOURCE_DIR>/cmake/"
-  COMMAND git clone ${GIT_URL}/noloader/cryptopp-cmake "<SOURCE_DIR>/cmake"
-  COMMAND cd "<SOURCE_DIR>/cmake" && git checkout tags/${CRYPTOPP_TAG} -b
-          ${CRYPTOPP_TAG}
-  COMMAND ${CMAKE_COMMAND} -E copy_directory "<SOURCE_DIR>/cmake/"
-          "<SOURCE_DIR>/"
+  COMMAND ${CMAKE_COMMAND} -E copy "${CRYPTOPP_CMAKE_SOURCE_DIR}/CMakeLists.txt"
+          "<SOURCE_DIR>/CMakeLists.txt"
+  COMMAND
+    ${CMAKE_COMMAND} -E copy
+    "${CRYPTOPP_CMAKE_SOURCE_DIR}/cryptopp-config.cmake"
+    "<SOURCE_DIR>/cryptopp-config.cmake"
   COMMAND ${CRYPTOPP_PATCH_COMMAND}
   INSTALL_DIR ${CRYPTOPP_INSTALL_DIR}
   CMAKE_ARGS ${CRYPTOPP_CMAKE_ARGS}

cmake/external/cub.cmake

@@ -29,7 +29,7 @@ if(${CMAKE_CUDA_COMPILER_VERSION} GREATER_EQUAL 11.6)
   set(CUB_TAG 1.16.0)
   execute_process(COMMAND git --git-dir=${CUB_SOURCE_DIR}/.git
                           --work-tree=${CUB_SOURCE_DIR} checkout ${CUB_TAG})
-  # cub 1.16.0 is not compitable with current thrust version
+  # cub 1.16.0 is not compatible with current thrust version
   add_definitions(-DTHRUST_IGNORE_CUB_VERSION_CHECK)
 else()
   set(CUB_TAG 1.8.0)

cmake/external/eigen.cmake

@@ -57,7 +57,7 @@ if(CMAKE_COMPILER_IS_GNUCC)
     file(TO_NATIVE_PATH ${PADDLE_SOURCE_DIR}/patches/eigen/Complex.h.patch
          complex_header)
     set(EIGEN_PATCH_COMMAND
-        ${EIGEN_PATCH_COMMAND} && patch -d
+        ${EIGEN_PATCH_COMMAND} && patch -Nd
         ${SOURCE_DIR}/Eigen/src/Core/arch/SSE/ < ${complex_header})
   endif()
 endif()

cmake/external/lapack.cmake

@@ -20,7 +20,7 @@ set(LAPACK_DOWNLOAD_DIR
 set(LAPACK_INSTALL_DIR ${THIRD_PARTY_PATH}/install/lapack)
 set(LAPACK_LIB_DIR ${LAPACK_INSTALL_DIR}/lib)
 
-# Note(zhouwei): lapack need fortan compiler which many machines don't have, so use precompiled library.
+# Note(zhouwei): lapack need fortran compiler which many machines don't have, so use precompiled library.
 # use lapack tag v3.10.0 on 06/28/2021 https://github.com/Reference-LAPACK/lapack
 if(LINUX)
   set(LAPACK_FILE

cmake/external/mklml.cmake

@@ -35,7 +35,7 @@ if(WIN32)
 else()
   #TODO(intel-huying):
   #  Now enable csrmm function in mklml library temporarily,
-  #  it will be updated as offical version later.
+  #  it will be updated as official version later.
   set(MKLML_FILE
       "csrmm_mklml_lnx_2019.0.5.tgz"
       CACHE STRING "" FORCE)

cmake/external/protobuf.cmake

@@ -322,7 +322,7 @@ if(WITH_IPU)
 elseif(WITH_ARM_BRPC)
   set(PROTOBUF_VERSION 21.12-baidu-ee-common)
 elseif(WIN32)
-  #Lower version prootbuf is used for widows
+  #Lower version protobuf is used for widows
   set(PROTOBUF_VERSION 21.12)
 else()
   set(PROTOBUF_VERSION 21.12)

cmake/external/pybind11.cmake

@@ -28,7 +28,7 @@ if(NOT WIN32)
   file(TO_NATIVE_PATH ${PADDLE_SOURCE_DIR}/patches/pybind/cast.h.patch
        native_dst)
   # Note: [Why calling some `git` commands before `patch`?]
-  # Paddle's CI uses cache to accelarate the make process. However, error might raise when patch codes in two scenarios:
+  # Paddle's CI uses cache to accelerate the make process. However, error might raise when patch codes in two scenarios:
   # 1. Patch to the wrong version: the tag version of CI's cache falls behind PYBIND_TAG, use `git checkout ${PYBIND_TAG}` to solve this.
   # 2. Patch twice: the tag version of cache == PYBIND_TAG, but patch has already applied to cache.
   set(PYBIND_PATCH_COMMAND

cmake/external/warpctc.cmake

@@ -33,7 +33,8 @@ if(WIN32)
       "<SOURCE_DIR>/")
 else()
   set(WARPCTC_PATCH_CUDA_COMMAND
-      patch -d ${SOURCE_DIR} <
+      git checkout -- . && git checkout ${WARPCTC_TAG} && patch -Nd
+      ${SOURCE_DIR} <
       ${PADDLE_SOURCE_DIR}/patches/warpctc/CMakeLists.txt.cuda.patch)
 endif()
 
@@ -42,7 +43,8 @@ if(NOT WIN32 AND WITH_GPU)
                                                   VERSION_GREATER 12.0)
     file(TO_NATIVE_PATH
          ${PADDLE_SOURCE_DIR}/patches/warpctc/CMakeLists.txt.patch native_src)
-    set(WARPCTC_PATCH_COMMAND patch -d ${SOURCE_DIR} < ${native_src} &&)
+    set(WARPCTC_PATCH_COMMAND git checkout -- . && git checkout ${WARPCTC_TAG}
+                              && patch -Nd ${SOURCE_DIR} < ${native_src} &&)
     set(WARPCTC_CCBIN_OPTION -DCCBIN_COMPILER=${CCBIN_COMPILER})
   endif()
 endif()

cmake/external/warprnnt.cmake

@@ -31,7 +31,8 @@ if(WIN32)
       "<SOURCE_DIR>/")
 else()
   set(WARPCTC_PATCH_CUDA_COMMAND
-      patch -d ${SOURCE_DIR} <
+      git checkout -- . && git checkout ${WARPRNNT_TAG} && patch -Nd
+      ${SOURCE_DIR} <
       ${PADDLE_SOURCE_DIR}/patches/warprnnt/CMakeLists.txt.cuda.patch)
 endif()
 
@@ -40,7 +41,9 @@ if(NOT WIN32 AND WITH_GPU)
                                                   VERSION_GREATER 12.0)
     file(TO_NATIVE_PATH
          ${PADDLE_SOURCE_DIR}/patches/warprnnt/CMakeLists.txt.patch native_src)
-    set(WARPRNNT_PATCH_COMMAND patch -d ${SOURCE_DIR} < ${native_src})
+    set(WARPRNNT_PATCH_COMMAND
+        git checkout -- . && git checkout ${WARPRNNT_TAG} && patch -Nd
+        ${SOURCE_DIR} < ${native_src})
     set(WARPRNNT_CCBIN_OPTION -DCCBIN_COMPILER=${CCBIN_COMPILER})
   endif()
 endif()

cmake/external/xpu.cmake

@@ -29,7 +29,7 @@ if(NOT DEFINED XPU_BASE_DATE)
   set(XPU_BASE_DATE "20240104")
 endif()
 if(NOT DEFINED XPU_XHPC_BASE_DATE)
-  set(XPU_XHPC_BASE_DATE "20240218")
+  set(XPU_XHPC_BASE_DATE "20240226")
 endif()
 set(XPU_XCCL_BASE_VERSION "1.1.8.1")
 if(NOT DEFINED XPU_XFT_BASE_VERSION)

cmake/flags.cmake

@@ -17,7 +17,7 @@ function(CheckCompilerCXX14Flag)
   elseif(CMAKE_CXX_COMPILER_ID STREQUAL "AppleClang" OR CMAKE_CXX_COMPILER_ID
                                                         STREQUAL "Clang")
     # cmake >= 3.0 compiler id "AppleClang" on Mac OS X, otherwise "Clang"
-    # Apple Clang is a different compiler than upstream Clang which havs different version numbers.
+    # Apple Clang is a different compiler than upstream Clang which has different version numbers.
     # https://gist.github.com/yamaya/2924292
     if(APPLE) # cmake < 3.0 compiler id "Clang" on Mac OS X
       if(${CMAKE_CXX_COMPILER_VERSION} VERSION_LESS 5.1)