Merge pull request PaddlePaddle#76 from YaoCheng8667/dev_xpu_mempool

merge from paddlebox master; add xpu_mem_check; fix paddle on infer; Support buddy allocator
zmxdream · May 30, 2024 · 83e62b7 · 83e62b7
2 parents 8bc6a52 + a253d83
commit 83e62b7
Show file tree

Hide file tree

Showing 98 changed files with 6,280 additions and 1,203 deletions.
diff --git a/cmake/cuda.cmake b/cmake/cuda.cmake
@@ -6,28 +6,34 @@ if(WITH_NV_JETSON)
   add_definitions(-DWITH_NV_JETSON)
   set(paddle_known_gpu_archs "53 62 72")
   set(paddle_known_gpu_archs10 "53 62 72")
+  set(paddle_known_gpu_archs11 "53 62 72 87")
+  set(paddle_known_gpu_archs12 "53 62 72 87 90")
 elseif(NEW_RELEASE_ALL)
   message("Using New Release Strategy - All Arches Packge")
   add_definitions(-DNEW_RELEASE_ALL)
-  set(paddle_known_gpu_archs "35 50 52 60 61 70 75 80 86")
-  set(paddle_known_gpu_archs10 "35 50 52 60 61 70 75")
+  set(paddle_known_gpu_archs "50 52 60 61 70 75 80 86 90")
+  set(paddle_known_gpu_archs10 "50 52 60 61 70 75")
   set(paddle_known_gpu_archs11 "50 60 61 70 75 80")
+  set(paddle_known_gpu_archs12 "50 60 61 70 75 80 90")
 elseif(NEW_RELEASE_PYPI)
   message("Using New Release Strategy - Cubin Packge")
   add_definitions(-DNEW_RELEASE_PYPI)
-  set(paddle_known_gpu_archs "35 50 52 60 61 70 75 80 86")
+  set(paddle_known_gpu_archs "50 52 60 61 70 75 80 86 90")
   set(paddle_known_gpu_archs10 "")
-  set(paddle_known_gpu_archs11 "60 61 70 75 80")
+  set(paddle_known_gpu_archs11 "61 70 75 80")
+  set(paddle_known_gpu_archs12 "61 70 75 80 90")
 elseif(NEW_RELEASE_JIT)
   message("Using New Release Strategy - JIT Packge")
   add_definitions(-DNEW_RELEASE_JIT)
-  set(paddle_known_gpu_archs "35 50 52 60 61 70 75 80 86")
-  set(paddle_known_gpu_archs10 "35 50 60 70 75")
-  set(paddle_known_gpu_archs11 "35 50 60 70 75 80")
+  set(paddle_known_gpu_archs "50 52 60 61 70 75 80 86 90")
+  set(paddle_known_gpu_archs10 "50 60 70 75")
+  set(paddle_known_gpu_archs11 "50 60 70 75 80")
+  set(paddle_known_gpu_archs12 "50 60 70 75 80 90")
 else()
-  set(paddle_known_gpu_archs "35 50 52 60 61 70 75 80")
-  set(paddle_known_gpu_archs10 "35 50 52 60 61 70 75")
+  set(paddle_known_gpu_archs "70 80")
+  set(paddle_known_gpu_archs10 "50 52 60 61 70 75")
   set(paddle_known_gpu_archs11 "52 60 61 70 75 80")
+  set(paddle_known_gpu_archs12 "70 80")
 endif()
 
 ######################################################################################
@@ -98,12 +104,12 @@ endfunction()
 function(select_nvcc_arch_flags out_variable)
   # List of arch names
   set(archs_names
-      "Kepler"
       "Maxwell"
       "Pascal"
       "Volta"
       "Turing"
       "Ampere"
+      "Hopper"
       "All"
       "Manual")
   set(archs_name_default "Auto")
@@ -142,9 +148,7 @@ function(select_nvcc_arch_flags out_variable)
     unset(CUDA_ARCH_PTX CACHE)
   endif()
 
-  if(${CUDA_ARCH_NAME} STREQUAL "Kepler")
-    set(cuda_arch_bin "30 35")
-  elseif(${CUDA_ARCH_NAME} STREQUAL "Maxwell")
+  if(${CUDA_ARCH_NAME} STREQUAL "Maxwell")
     if(WITH_NV_JETSON)
       set(cuda_arch_bin "53")
     else()
@@ -165,11 +169,17 @@ function(select_nvcc_arch_flags out_variable)
   elseif(${CUDA_ARCH_NAME} STREQUAL "Turing")
     set(cuda_arch_bin "75")
   elseif(${CUDA_ARCH_NAME} STREQUAL "Ampere")
-    if(${CMAKE_CUDA_COMPILER_VERSION} LESS 11.1) # CUDA 11.0
-      set(cuda_arch_bin "80")
-    elseif(${CMAKE_CUDA_COMPILER_VERSION} LESS 12.0) # CUDA 11.1+
-      set(cuda_arch_bin "80 86")
+    if(WITH_NV_JETSON)
+      set(cuda_arch_bin "87")
+    else()
+      if(${CMAKE_CUDA_COMPILER_VERSION} LESS 11.1) # CUDA 11.0
+        set(cuda_arch_bin "80")
+      else()
+        set(cuda_arch_bin "80 86")
+      endif()
     endif()
+  elseif(${CUDA_ARCH_NAME} STREQUAL "Hopper")
+    set(cuda_arch_bin "90")
   elseif(${CUDA_ARCH_NAME} STREQUAL "All")
     set(cuda_arch_bin ${paddle_known_gpu_archs})
   elseif(${CUDA_ARCH_NAME} STREQUAL "Auto")
@@ -186,6 +196,13 @@ function(select_nvcc_arch_flags out_variable)
     set(cuda_arch_bin ${CUDA_ARCH_BIN})
   endif()
 
+  # cuda11.4
+  if(${CMAKE_CUDA_COMPILER_VERSION} LESS 11.6)
+    set(cuda_arch_bin "70 80")
+  else()
+    set(cuda_arch_bin "70 80 90")
+  endif()
+
   if(NEW_RELEASE_JIT)
     set(cuda_arch_ptx "${cuda_arch_ptx}${cuda_arch_bin}")
     set(cuda_arch_bin "")
@@ -249,6 +266,11 @@ elseif(${CMAKE_CUDA_COMPILER_VERSION} LESS 12.0) # CUDA 11.2+
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D_MWAITXINTRIN_H_INCLUDED")
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D__STRICT_ANSI__")
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
+elseif(${CMAKE_CUDA_COMPILER_VERSION} LESS 13.0) # CUDA 12.0+
+  set(paddle_known_gpu_archs "${paddle_known_gpu_archs12} 90")
+  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D_MWAITXINTRIN_H_INCLUDED")
+  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D__STRICT_ANSI__")
+  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
 endif()
 
 if(NOT ${CMAKE_CUDA_COMPILER_VERSION} LESS 10.0)

diff --git a/cmake/external/gloo.cmake b/cmake/external/gloo.cmake
@@ -25,8 +25,18 @@ set(GLOO_LIBRARY_DIR
     "${GLOO_INSTALL_DIR}/lib"
     CACHE PATH "gloo library directory." FORCE)
 # As we add extra features for gloo, we use the non-official repo
-set(GLOO_REPOSITORY ${GIT_URL}/sandyhouse/gloo.git)
-set(GLOO_TAG v0.0.2)
+if(WITH_GPU)
+  if(${CMAKE_CUDA_COMPILER_VERSION} LESS 12.0)
+    set(GLOO_REPOSITORY ${GIT_URL}/sandyhouse/gloo.git)
+    set(GLOO_TAG v0.0.2)
+  else()
+    set(GLOO_REPOSITORY ${GIT_URL}/ziyoujiyi/gloo.git)
+    set(GLOO_TAG v0.0.3)
+  endif()
+else()
+  set(GLOO_REPOSITORY ${GIT_URL}/sandyhouse/gloo.git)
+  set(GLOO_TAG v0.0.2)
+endif()
 set(GLOO_LIBRARIES
     "${GLOO_INSTALL_DIR}/lib/libgloo.a"
     CACHE FILEPATH "gloo library." FORCE)

diff --git a/cmake/external/warpctc.cmake b/cmake/external/warpctc.cmake
@@ -23,7 +23,15 @@ set(WARPCTC_INSTALL_DIR ${THIRD_PARTY_PATH}/install/warpctc)
 # in case of low internet speed
 #set(WARPCTC_REPOSITORY  https://gitee.com/tianjianhe/warp-ctc.git)
 set(WARPCTC_REPOSITORY ${GIT_URL}/baidu-research/warp-ctc.git)
-set(WARPCTC_TAG 37ece0e1bbe8a0019a63ac7e6462c36591c66a5b)
+if(WITH_GPU)
+  if(${CMAKE_CUDA_COMPILER_VERSION} LESS 12.0)
+      set(WARPCTC_TAG 37ece0e1bbe8a0019a63ac7e6462c36591c66a5b)
+  else()
+      set(WARPCTC_TAG bdc2b4550453e0ef2d3b5190f9c6103a84eff184)
+  endif()
+else()
+  set(WARPCTC_TAG 37ece0e1bbe8a0019a63ac7e6462c36591c66a5b)
+endif()
 
 set(WARPCTC_INCLUDE_DIR
     "${WARPCTC_INSTALL_DIR}/include"

diff --git a/cmake/external/xpu.cmake b/cmake/external/xpu.cmake
@@ -8,6 +8,8 @@ set(XPU_API_LIB_NAME "libxpuapi.so")
 set(XPU_API_PLUGIN_NAME  "libxpuplugin.so")
 set(XPU_RT_LIB_NAME "libxpurt.so")
 set(XPU_RT_ALIAS_LIB_NAME "libxpurt.so.1")
+set(XPU_ML_LIB_NAME "libxpuml.so")
+set(XPU_ML_ALIAS_LIB_NAME "libxpuml.so.1")
 
 if(NOT DEFINED XPU_BASE_URL)
   set(XPU_BASE_URL_WITHOUT_DATE
@@ -128,6 +130,8 @@ set(XPU_API_LIB "${XPU_LIB_DIR}/${XPU_API_LIB_NAME}")
 set(XPU_API_PLUGIN "${XPU_LIB_DIR}/${XPU_API_PLUGIN_NAME}")
 set(XPU_RT_LIB "${XPU_LIB_DIR}/${XPU_RT_LIB_NAME}")
 set(XPU_RT_ALIAS_LIB "${XPU_LIB_DIR}/${XPU_RT_ALIAS_LIB_NAME}")
+set(XPU_ML_LIB "${THIRD_PARTY_PATH}/xpu/src/extern_xpu/xre-bdcentos_x86_64/so/${XPU_ML_LIB_NAME}")
+set(XPU_ML_ALIAS_LIB "${THIRD_PARTY_PATH}/xpu/src/extern_xpu/xre-bdcentos_x86_64/so/${XPU_ML_ALIAS_LIB_NAME}")
 
 set(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_RPATH}" "${XPU_INSTALL_DIR}/lib")
 
@@ -173,9 +177,9 @@ if(WITH_XPU_BKCL)
   set(XPU_BKCL_LIB "${XPU_LIB_DIR}/${XPU_BKCL_LIB_NAME}")
   set(XPU_BKCL_INC_DIR "${THIRD_PARTY_PATH}/install/xpu/include")
   include_directories(${XPU_BKCL_INC_DIR})
-  target_link_libraries(xpulib -Wl,--push-state,--no-as-needed ${XPU_API_LIB} ${XPU_API_PLUGIN} ${XPU_RT_LIB} ${XPU_BKCL_LIB} -Wl,--pop-state)
+  target_link_libraries(xpulib -Wl,--push-state,--no-as-needed ${XPU_API_LIB} ${XPU_API_PLUGIN} ${XPU_RT_LIB} ${XPU_BKCL_LIB} ${XPU_ML_LIB} -Wl,--pop-state)
 else()
-  target_link_libraries(xpulib -Wl,--push-state,--no-as-needed ${XPU_API_LIB} ${XPU_API_PLUGIN} ${XPU_RT_LIB} -Wl,--pop-state)
+  target_link_libraries(xpulib -Wl,--push-state,--no-as-needed ${XPU_API_LIB} ${XPU_API_PLUGIN} ${XPU_RT_LIB} ${XPU_ML_LIB} -Wl,--pop-state)
 endif()
 
 add_dependencies(xpulib ${XPU_PROJECT})

diff --git a/cmake/third_party.cmake b/cmake/third_party.cmake
@@ -317,8 +317,7 @@ if(WITH_ONNXRUNTIME)
 endif()
 
 if(WITH_GPU)
-  if(${CMAKE_CUDA_COMPILER_VERSION} LESS 11.0 OR ${CMAKE_CUDA_COMPILER_VERSION}
-                                                 GREATER_EQUAL 11.6)
+  if(${CMAKE_CUDA_COMPILER_VERSION} LESS 11.0)
     include(external/cub) # download cub
     list(APPEND third_party_deps extern_cub)
   endif()

diff --git a/cmake/version.cmake b/cmake/version.cmake
@@ -71,3 +71,27 @@ math(EXPR PADDLE_VERSION_INTEGER "${PADDLE_MAJOR_VER} * 1000000
 add_definitions(-DPADDLE_VERSION=${PADDLE_VERSION})
 add_definitions(-DPADDLE_VERSION_INTEGER=${PADDLE_VERSION_INTEGER})
 message(STATUS "Paddle version is ${PADDLE_VERSION}")
+
+#add git version
+set(COMMIT_HASH "")
+set(BRANCH_NAME "")
+find_package(Git QUIET)
+if(GIT_FOUND)
+execute_process(
+    COMMAND ${GIT_EXECUTABLE} log -1 --pretty=format:%H
+    OUTPUT_VARIABLE COMMIT_HASH
+    OUTPUT_STRIP_TRAILING_WHITESPACE
+    ERROR_QUIET
+    WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
+)
+execute_process(
+    COMMAND ${GIT_EXECUTABLE} symbolic-ref --short -q HEAD
+    OUTPUT_VARIABLE BRANCH_NAME
+    OUTPUT_STRIP_TRAILING_WHITESPACE
+    ERROR_QUIET
+    WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
+)
+endif()
+message(STATUS "Git version is ${BRANCH_NAME}:${COMMIT_HASH}")
+add_definitions(-DPADDLE_BRANCH_NAME="${BRANCH_NAME}")
+add_definitions(-DPADDLE_COMMIT_HASH="${COMMIT_HASH}")