Merge branch 'develop' into add_pd_op

_typos.toml

@@ -12,6 +12,12 @@ extend-exclude = [
 anc = 'anc'
 arange = "arange"
 astroid = 'astroid'
+ba = 'ba'
+cacl = 'cacl'
+CANN = 'CANN'
+Clas = 'Clas'
+clen = 'clen'
+cll = 'cll'
 dout = "dout"
 eles = 'eles'
 grad = "grad"
@@ -24,84 +30,17 @@ UE = "UE"
 unpacket = "unpacket"
 
 # These words need to be fixed
-Acceses = 'Acceses'
-accessable = 'accessable'
-accesss = 'accesss'
-accoding = 'accoding'
-accurary = 'accurary'
-Accuarcy = 'Accuarcy'
-actived = 'actived'
-acitve = 'acitve'
-actualy = 'actualy'
-actural = 'actural'
-Actural = 'Actural'
-assigend = 'assigend'
-assined = 'assined'
-assgin = 'assgin'
-axises = 'axises'
-Axises = 'Axises'
-aixs = 'aixs'
-beacuse = 'beacuse'
-becasue = 'becasue'
-Becasue = 'Becasue'
-becuase = 'becuase'
 blokc = 'blokc'
 blcok = 'blcok'
 bootom = 'bootom'
 bondary = 'bondary'
 branchs = 'branchs'
-Broardcast = 'Broardcast'
-Bradcast = 'Bradcast'
-Boardcast = 'Boardcast'
-Buitin = 'Buitin'
-buitlin = 'buitlin'
-buitin = 'buitin'
-builded = 'builded'
-ba = 'ba'
 cahe = 'cahe'
 Caculate = 'Caculate'
 caculate = 'caculate'
 calcualtion = 'calcualtion'
-cacl = 'cacl'
-cll = 'cll'
-candiate = 'candiate'
-cadidate = 'cadidate'
-connot = 'connot'
-CANN = 'CANN'
-Cann = 'Cann'
-cann = 'cann'
-vart = 'vart'
 checkings = 'checkings'
 childs = 'childs'
-Chunck = 'Chunck'
-clen = 'clen'
-Clas = 'Clas'
-clas = 'clas'
-compability = 'compability'
-compatiblity = 'compatiblity'
-Compitable = 'Compitable'
-compatable = 'compatable'
-compitable = 'compitable'
-compling = 'compling'
-comple = 'comple'
-complition = 'complition'
-complext = 'complext'
-compsite = 'compsite'
-configurated = 'configurated'
-configed = 'configed'
-confict = 'confict'
-conjuction = 'conjuction'
-conect = 'conect'
-consequtive = 'consequtive'
-consistant = 'consistant'
-contraints = 'contraints'
-contruction = 'contruction'
-consructor = 'consructor'
-comsume = 'comsume'
-Continer = 'Continer'
-contenst = 'contenst'
-conter = 'conter'
-ontext = 'ontext'
 Continous = 'Continous'
 contibute = 'contibute'
 controled = 'controled'
@@ -149,13 +88,6 @@ defind = 'defind'
 defeine = 'defeine'
 defition = 'defition'
 defination = 'defination'
-delet = 'delet'
-dependecies = 'dependecies'
-dependecy = 'dependecy'
-decprecated = 'decprecated'
-derivated = 'derivated'
-descripor = 'descripor'
-deserailize = 'deserailize'
 Destory = 'Destory'
 DEIVCE = 'DEIVCE'
 dictionnary = 'dictionnary'
@@ -426,7 +358,6 @@ Normlized = 'Normlized'
 normlize = 'normlize'
 noraml = 'noraml'
 numer = 'numer'
-Numberic = 'Numberic'
 occured = 'occured'
 Ocurred = 'Ocurred'
 occures = 'occures'
@@ -435,7 +366,6 @@ fo = 'fo'
 offets = 'offets'
 offseted = 'offseted'
 OLT = 'OLT'
-olny = 'olny'
 pn = 'pn'
 Operants = 'Operants'
 operants = 'operants'
@@ -454,13 +384,11 @@ orginal = 'orginal'
 onces = 'onces'
 outter = 'outter'
 outpus = 'outpus'
-outputing = 'outputing'
 outout = 'outout'
 ouput = 'ouput'
 outpout = 'outpout'
 ouptut = 'ouptut'
 Ouput = 'Ouput'
-ouside = 'ouside'
 overriden = 'overriden'
 Overide = 'Overide'
 overide = 'overide'

cmake/cinn.cmake

@@ -160,6 +160,7 @@ cinn_cc_library(
   ${cinnapi_src}
   DEPS
   glog
+  python
   ${llvm_libs}
   param_proto
   auto_schedule_proto

cmake/cinn/core.cmake

@@ -19,10 +19,6 @@ function(cinn_cc_library TARGET_NAME)
     endif()
 
     if(cinn_cc_library_DEPS)
-      if("${cinn_cc_library_DEPS};" MATCHES "python;")
-        list(REMOVE_ITEM cinn_cc_library_DEPS python)
-        add_dependencies(${TARGET_NAME} python)
-      endif()
       target_link_libraries(${TARGET_NAME} ${cinn_cc_library_DEPS})
       add_dependencies(${TARGET_NAME} ${cinn_cc_library_DEPS})
     endif()

cmake/external/brpc.cmake

@@ -48,13 +48,23 @@ set(prefix_path
     "${THIRD_PARTY_PATH}/install/gflags|${THIRD_PARTY_PATH}/install/leveldb|${THIRD_PARTY_PATH}/install/snappy|${THIRD_PARTY_PATH}/install/gtest|${THIRD_PARTY_PATH}/install/protobuf|${THIRD_PARTY_PATH}/install/zlib|${THIRD_PARTY_PATH}/install/glog"
 )
 
+if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU" AND ${CMAKE_CXX_COMPILER_VERSION}
+                                            VERSION_GREATER_EQUAL 13.0)
+  file(TO_NATIVE_PATH ${PADDLE_SOURCE_DIR}/patches/brpc/http2.h.patch
+       http2_h_patch)
+  set(BRPC_PATCH_COMMAND_GCC13 git apply ${http2_h_patch})
+endif()
+
 # If minimal .a is need, you can set  WITH_DEBUG_SYMBOLS=OFF
 ExternalProject_Add(
   extern_brpc
   ${EXTERNAL_PROJECT_LOG_ARGS}
   SOURCE_DIR ${BRPC_SOURCE_DIR}
   PREFIX ${BRPC_PREFIX_DIR}
   UPDATE_COMMAND ""
+  PATCH_COMMAND
+  COMMAND git checkout -- . && git checkout ${BRPC_TAG}
+  COMMAND ${BRPC_PATCH_COMMAND_GCC13}
   CMAKE_ARGS -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
              -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
              -DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS}

cmake/external/flashattn.cmake

@@ -80,12 +80,20 @@ if(WITH_ROCM)
 else()
 
   add_definitions(-DPADDLE_WITH_FLASHATTN)
+  option(FA_BUILD_WITH_CACHE "Download cache so files from bos" ON)
 
   set(FLASHATTN_PREFIX_DIR ${THIRD_PARTY_PATH}/flashattn)
   set(FLASHATTN_SOURCE_SUBDIR csrc)
   set(FLASHATTN_INSTALL_DIR ${THIRD_PARTY_PATH}/install/flashattn)
   set(SOURCE_DIR ${PADDLE_SOURCE_DIR}/third_party/flashattn)
-  set(FLASHATTN_TAG 5fc132ac11e78d26471ca09e5ba0cd817c3424d8)
+
+  # get FA git commit
+  execute_process(
+    COMMAND git rev-parse HEAD
+    WORKING_DIRECTORY ${SOURCE_DIR}
+    OUTPUT_VARIABLE FLASHATTN_TAG
+    OUTPUT_STRIP_TRAILING_WHITESPACE)
+  message(STATUS "flashattn git commit: ${FLASHATTN_TAG}")
 
   set(FLASHATTN_INCLUDE_DIR
       "${FLASHATTN_INSTALL_DIR}/include"
@@ -166,6 +174,99 @@ else()
     endif()
   endforeach()
 
+  set(BASE_URL
+      "https://xly-devops.bj.bcebos.com/gpups/flash-attention/cu${FA_NVCC_ARCH_BIN}"
+  )
+  set(TAR_FILE_NAME "flashattn_libs_${FLASHATTN_TAG}.tar")
+  set(TAR_FILE_URL "${BASE_URL}/${TAR_FILE_NAME}")
+  set(FA_BUILD_DIR "${FLASHATTN_PREFIX_DIR}/src/extern_flashattn-build/")
+  set(CACHE_TAR_PATH "${FA_BUILD_DIR}/${TAR_FILE_NAME}")
+  set(CACHE_TAR_DIR "${FA_BUILD_DIR}/flashattn_libs_${FLASHATTN_TAG}")
+
+  set(SKIP_BUILD_FA OFF)
+  if(FA_BUILD_WITH_CACHE)
+
+    message(STATUS "Downloading ${TAR_FILE_URL} to ${CACHE_TAR_PATH}")
+    file(
+      DOWNLOAD "${TAR_FILE_URL}" "${CACHE_TAR_PATH}"
+      STATUS DOWNLOAD_STATUS
+      LOG DOWNLOAD_LOG)
+    list(GET DOWNLOAD_STATUS 0 DOWNLOAD_RESULT)
+
+    if(DOWNLOAD_RESULT EQUAL 0)
+      message(STATUS "Download Successful")
+
+      file(MAKE_DIRECTORY ${FA_BUILD_DIR})
+
+      execute_process(
+        COMMAND ${CMAKE_COMMAND} -E tar xf ${CACHE_TAR_PATH}
+        WORKING_DIRECTORY ${FA_BUILD_DIR}
+        RESULT_VARIABLE TAR_RESULT)
+
+      if(NOT TAR_RESULT EQUAL 0)
+        message(FATAL_ERROR "Failed to extract ${CACHE_TAR_PATH}")
+      endif()
+
+      file(STRINGS ${CACHE_TAR_DIR}/MD5.txt FILE_MD5)
+
+      # Strip any leading or trailing whitespace
+      string(STRIP ${FILE_MD5} FILE_MD5)
+
+      file(MD5 ${CACHE_TAR_DIR}/fa_libs.tar FILE_MD5_ACTUAL)
+
+      message(STATUS "Expected MD5: ${FILE_MD5}")
+      message(STATUS "Actual MD5:   ${FILE_MD5_ACTUAL}")
+
+      if(NOT "${FILE_MD5}" STREQUAL "${FILE_MD5_ACTUAL}")
+        message(
+          FATAL_ERROR "MD5 checksum mismatch! The download may be corrupted.")
+      else()
+        message(STATUS "MD5 checksum verified successfully.")
+      endif()
+
+      execute_process(
+        COMMAND ${CMAKE_COMMAND} -E tar xf ${CACHE_TAR_DIR}/fa_libs.tar
+        WORKING_DIRECTORY ${CACHE_TAR_DIR}
+        RESULT_VARIABLE TAR_RESULT)
+
+      if(NOT TAR_RESULT EQUAL 0)
+        message(FATAL_ERROR "Failed to extract ${CACHE_TAR_PATH}/fa_libs.tar")
+      endif()
+
+      file(GLOB_RECURSE SO_FILES "${CACHE_TAR_DIR}/fa_libs/*.so")
+      foreach(so_file ${SO_FILES})
+        message(STATUS "Copy ${so_file} to ${FA_BUILD_DIR}")
+        message(STATUS "Copy ${so_file} to ${FLASHATTN_LIB_DIR}")
+        file(COPY "${so_file}" DESTINATION "${FA_BUILD_DIR}")
+        file(COPY "${so_file}" DESTINATION "${FLASHATTN_LIB_DIR}")
+      endforeach()
+
+      file(REMOVE_RECURSE ${CACHE_TAR_DIR})
+      message(STATUS "Extraction completed in ${FA_BUILD_DIR}")
+
+      set(SKIP_BUILD_FA ON)
+
+    elseif(DOWNLOAD_RESULT EQUAL 6)
+      message(
+        STATUS
+          "Could not resolve host. The given remote host was not resolvable.")
+    elseif(DOWNLOAD_RESULT EQUAL 7)
+      message(STATUS "Failed to connect to host.")
+    elseif(DOWNLOAD_RESULT EQUAL 22)
+      message(
+        STATUS
+          "HTTP page not retrieved. The requested URL was not found or a server returned a 4xx (client error) or 5xx (server error) response."
+      )
+    elseif(DOWNLOAD_RESULT EQUAL 28)
+      message(
+        STATUS
+          "Operation timeout. The specified time-out period was reached according to the conditions."
+      )
+    else()
+      message(STATUS "An error occurred. Error code: ${DOWNLOAD_RESULT}")
+    endif()
+  endif()
+
   ExternalProject_Add(
     extern_flashattn
     ${EXTERNAL_PROJECT_LOG_ARGS}
@@ -196,13 +297,13 @@ else()
                -DCMAKE_JOB_POOLS:STRING=compile=${FA_JOB_POOLS_COMPILE}
                -DNVCC_ARCH_BIN=${FA_NVCC_ARCH_BIN}
                -DWITH_FLASHATTN_V3=${WITH_FLASHATTN_V3}
+               -DSKIP_BUILD_FA=${SKIP_BUILD_FA}
                ${EXTERNAL_OPTIONAL_ARGS}
     CMAKE_CACHE_ARGS
       -DCMAKE_BUILD_TYPE:STRING=${THIRD_PARTY_BUILD_TYPE}
       -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON
       -DCMAKE_INSTALL_PREFIX:PATH=${FLASHATTN_INSTALL_DIR}
     BUILD_BYPRODUCTS ${BUILD_BYPRODUCTS_LIST})
-
 endif()
 
 message(STATUS "flash-attn library: ${FLASHATTN_LIBRARIES}")

cmake/external/openblas.cmake

@@ -19,16 +19,11 @@ set(CBLAS_INSTALL_DIR ${THIRD_PARTY_PATH}/install/openblas)
 set(CBLAS_SOURCE_DIR ${PADDLE_SOURCE_DIR}/third_party/openblas)
 set(CBLAS_TAG v0.3.7)
 
-# Why use v0.3.18?  The IDG business line encountered a random openblas error,
-# which can be resolved after upgrading openblas.
-# And why compile when gcc>8.2? Please refer to
-# https://github.com/spack/spack/issues/19932#issuecomment-733452619
-# v0.3.18 only support gcc>=8.3 or gcc>=7.4
-if((CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
-   AND CMAKE_CXX_COMPILER_VERSION VERSION_GREATER 8.2
+if(UNIX
+   AND NOT APPLE
+   AND NOT WITH_ROCM
    AND NOT WITH_XPU)
-  # We only compile with openblas 0.3.18 when gcc >= 8.3
-  set(CBLAS_TAG v0.3.18)
+  set(CBLAS_TAG v0.3.28)
 endif()
 
 if(APPLE AND WITH_ARM)