Merge branch 'develop' into apply_tensor

PaddlePaddle · Dec 19, 2023 · 4931c16 · 4931c16
2 parents bd99a1d + 5d32c61
commit 4931c16
Show file tree

Hide file tree

Showing 560 changed files with 28,222 additions and 5,661 deletions.
diff --git a/.gitignore b/.gitignore
@@ -108,5 +108,13 @@ paddle/fluid/pir/dialect/operator/ir/pd_api.*
 paddle/fluid/pir/dialect/operator/ir/op_decomp.cc
 paddle/fluid/pir/dialect/operator/ir/pd_op_vjp.cc
 paddle/fluid/pir/dialect/operator/ir/pd_op.*
+paddle/fluid/pir/dialect/operator/ir/pd_op_bwd.*
+paddle/fluid/pir/dialect/operator/ir/pd_op_fused.*
+paddle/fluid/pir/dialect/operator/ir/pd_op_fused_bwd.*
+paddle/fluid/pir/dialect/operator/ir/pd_pir_op.*
+paddle/fluid/pir/dialect/operator/ir/pd_pir_op_bwd.*
+paddle/fluid/pir/dialect/operator/ir/pd_pir_op_update.*
+paddle/fluid/pir/dialect/operator/ir/pd_op_info.*
 paddle/cinn/hlir/dialect/generated/ops.parsed.yaml
 paddle/cinn/hlir/dialect/operator/ir/cinn_op.*
+paddle/cinn/hlir/dialect/operator/ir/cinn_op_info.*
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -54,6 +54,7 @@ option(WITH_XPU "Compile PaddlePaddle with BAIDU KUNLUN XPU" OFF)
 option(WITH_XPU_KP "Compile PaddlePaddle with BAIDU XPU compiler " OFF)
 option(WITH_XPU_XFT "Compile PaddlePaddle with BAIDU XPU-XFT" OFF)
 option(WITH_XPU_PLUGIN "Compile PaddlePaddle with BAIDU XPU plugin" OFF)
+option(WITH_XPU_XHPC "Compile PaddlePaddle with BAIDU XPU-HPC library" OFF)
 option(WITH_WIN_DUMP_DBG "Compile with windows core dump debug mode" OFF)
 option(WITH_ROCM "Compile PaddlePaddle with ROCM platform" OFF)
 option(WITH_IPU "Compile PaddlePaddle with Graphcore IPU" OFF)
@@ -86,6 +87,9 @@ endif()
 if(WITH_GPU AND WITH_XPU_XFT)
   message(FATAL_ERROR "Error when compile GPU and XPU-XFT at the same time")
 endif()
+if(WITH_GPU AND WITH_XPU_XHPC)
+  message(FATAL_ERROR "Error when compile GPU and XPU-HPC at the same time")
+endif()
 if(WITH_GPU AND WITH_ROCM)
   message(FATAL_ERROR "Error when compile CUDA and ROCM at the same time")
 endif()
@@ -451,6 +455,14 @@ if(NOT WITH_XPU AND WITH_XPU_BKCL)
       CACHE STRING "Disable BKCL when compiling without XPU" FORCE)
 endif()
 
+if(NOT WITH_XPU AND WITH_XPU_XHPC)
+  message(
+    WARNING "Disable XHPC when compiling without XPU. Force WITH_XPU_XHPC=OFF.")
+  set(WITH_XPU_XHPC
+      OFF
+      CACHE STRING "Disable XHPC when compiling without XPU" FORCE)
+endif()
+
 if(WITH_NCCL)
   add_definitions("-DPADDLE_WITH_NCCL")
   include(nccl)

diff --git a/cmake/cinn/external/llvm.cmake b/cmake/cinn/external/llvm.cmake
@@ -88,7 +88,6 @@ llvm_map_components_to_libnames(
 message(STATUS "LLVM libs: ${llvm_libs}")
 
 get_property(mlir_libs GLOBAL PROPERTY MLIR_ALL_LIBS)
-message(STATUS "MLIR libs: ${mlir_libs}")
 add_definitions(${LLVM_DEFINITIONS})
 
 # The minimum needed libraries for MLIR IR parse and transform.

diff --git a/cmake/cinn/llvm.cmake b/cmake/cinn/llvm.cmake
@@ -54,7 +54,6 @@ llvm_map_components_to_libnames(
 message(STATUS "LLVM libs: ${llvm_libs}")
 
 get_property(mlir_libs GLOBAL PROPERTY MLIR_ALL_LIBS)
-message(STATUS "MLIR libs: ${mlir_libs}")
 add_definitions(${LLVM_DEFINITIONS})
 
 # The minimum needed libraries for MLIR IR parse and transform.

diff --git a/cmake/external/xpu.cmake b/cmake/external/xpu.cmake
@@ -22,11 +22,16 @@ set(XPU_API_LIB_NAME "libxpuapi.so")
 set(XPU_RT_LIB_NAME "libxpurt.so")
 set(XPU_XFT_LIB_NAME "libxft.so")
 set(XPU_XPTI_LIB_NAME "libxpti.so")
+set(XPU_XBLAS_LIB_NAME "libxpu_blas.so")
+set(XPU_XFA_LIB_NAME "libxpu_flash_attention.so")
 
 if(NOT DEFINED XPU_BASE_DATE)
   set(XPU_BASE_DATE "20231203")
 endif()
-set(XPU_XCCL_BASE_VERSION "1.1.6.1")
+if(NOT DEFINED XPU_XHPC_BASE_DATE)
+  set(XPU_XHPC_BASE_DATE "20231215")
+endif()
+set(XPU_XCCL_BASE_VERSION "1.1.7.1")
 if(NOT DEFINED XPU_XFT_BASE_VERSION)
   set(XPU_XFT_BASE_VERSION "20230602")
 endif()
@@ -68,18 +73,20 @@ if(WITH_AARCH64)
 elseif(WITH_SUNWAY)
   set(XPU_XRE_DIR_NAME "xre-deepin_sw6_64")
   set(XPU_XDNN_DIR_NAME "xdnn-deepin_sw6_64")
-  set(XPU_XCCL_DIR_NAME "${XPU_XCCL_PREFIX}-deepin_sw6_64")
+  set(XPU_XCCL_DIR_NAME "") # TODO: xccl has no deepin output at now.
   set(XPU_XFT_DIR_NAME "") # TODO: xft has no deepin output at now.
 elseif(WITH_BDCENTOS)
   set(XPU_XRE_DIR_NAME "xre-bdcentos_x86_64")
   set(XPU_XDNN_DIR_NAME "xdnn-bdcentos_x86_64")
   set(XPU_XCCL_DIR_NAME "${XPU_XCCL_PREFIX}-bdcentos_x86_64")
   set(XPU_XFT_DIR_NAME "xft_bdcentos6u3_x86_64_gcc82")
+  set(XPU_XHPC_DIR_NAME "xhpc-bdcentos_x86_64")
 elseif(WITH_UBUNTU)
   set(XPU_XRE_DIR_NAME "xre-ubuntu_x86_64")
   set(XPU_XDNN_DIR_NAME "xdnn-ubuntu_x86_64")
   set(XPU_XCCL_DIR_NAME "${XPU_XCCL_PREFIX}-ubuntu_x86_64")
   set(XPU_XFT_DIR_NAME "xft_ubuntu1604_x86_64")
+  set(XPU_XHPC_DIR_NAME "xhpc-ubuntu_x86_64")
 elseif(WITH_CENTOS)
   set(XPU_XRE_DIR_NAME "xre-centos7_x86_64")
   set(XPU_XDNN_DIR_NAME "xdnn-centos7_x86_64")
@@ -90,6 +97,7 @@ else()
   set(XPU_XDNN_DIR_NAME "xdnn-ubuntu_x86_64")
   set(XPU_XCCL_DIR_NAME "${XPU_XCCL_PREFIX}-ubuntu_x86_64")
   set(XPU_XFT_DIR_NAME "xft_ubuntu1604_x86_64")
+  set(XPU_XHPC_DIR_NAME "xhpc-ubuntu_x86_64")
 endif()
 set(XPU_XPTI_DIR_NAME "xpti")
 
@@ -110,14 +118,22 @@ set(XPU_XFT_GET_DEPENCE_URL
     "https://baidu-kunlun-public.su.bcebos.com/paddle_depence/get_xft_dependence.sh"
     CACHE STRING "" FORCE)
 
+if(WITH_XPU_XHPC)
+  set(XPU_XHPC_URL
+      "https://klx-sdk-release-public.su.bcebos.com/xhpc/dev/${XPU_XHPC_BASE_DATE}/${XPU_XHPC_DIR_NAME}.tar.gz"
+      CACHE STRING "" FORCE)
+endif()
+
 set(SNAPPY_PREFIX_DIR "${THIRD_PARTY_PATH}/xpu")
 set(XPU_DOWNLOAD_DIR "${SNAPPY_PREFIX_DIR}/src/${XPU_PROJECT}")
 set(XPU_INSTALL_DIR "${THIRD_PARTY_PATH}/install/xpu")
 set(XPU_INC_DIR "${THIRD_PARTY_PATH}/install/xpu/include")
 set(XPU_LIB_DIR "${THIRD_PARTY_PATH}/install/xpu/lib")
 
 set(XPU_API_LIB "${XPU_LIB_DIR}/${XPU_API_LIB_NAME}")
+set(XPU_XBLAS_LIB "${XPU_LIB_DIR}/${XPU_XBLAS_LIB_NAME}")
 set(XPU_RT_LIB "${XPU_LIB_DIR}/${XPU_RT_LIB_NAME}")
+set(XPU_XFA_LIB "${XPU_LIB_DIR}/${XPU_XFA_LIB_NAME}")
 
 set(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_RPATH}" "${XPU_INSTALL_DIR}/lib")
 
@@ -144,19 +160,20 @@ ExternalProject_Add(
   DOWNLOAD_DIR ${XPU_DOWNLOAD_DIR}
   DOWNLOAD_COMMAND
     bash ${CMAKE_SOURCE_DIR}/tools/xpu/check_xpu_dependence.sh ${XPU_BASE_URL}
-    ${XPU_XCCL_BASE_URL} && bash
+    ${XPU_XCCL_BASE_URL} && WITH_XPU_XHPC=${WITH_XPU_XHPC} bash
     ${CMAKE_SOURCE_DIR}/tools/xpu/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} ${XPU_XFT_DIR_NAME} &&
-    WITH_XPTI=${WITH_XPTI} bash
+    ${XPU_XCCL_DIR_NAME} ${XPU_XHPC_URL} ${XPU_XHPC_DIR_NAME} && wget
+    ${XPU_XFT_GET_DEPENCE_URL} && bash get_xft_dependence.sh ${XPU_XFT_URL}
+    ${XPU_XFT_DIR_NAME} && WITH_XPTI=${WITH_XPTI} bash
     ${CMAKE_SOURCE_DIR}/tools/xpu/get_xpti_dependence.sh ${XPU_XPTI_URL}
     ${XPU_XPTI_DIR_NAME}
   DOWNLOAD_NO_PROGRESS 1
   UPDATE_COMMAND ""
   CMAKE_ARGS -DCMAKE_INSTALL_PREFIX=${XPU_INSTALL_ROOT}
   CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${XPU_INSTALL_ROOT}
-  BUILD_BYPRODUCTS ${XPU_API_LIB}
+  BUILD_BYPRODUCTS ${XPU_API_LIB} BUILD_BYPORDUCTS ${XPU_XBLAS_LIB}
+  BUILD_BYPRODUCTS ${XPU_XFA_LIB}
   BUILD_BYPRODUCTS ${XPU_RT_LIB}
   BUILD_BYPRODUCTS ${XPU_BKCL_LIB})
 
@@ -179,6 +196,19 @@ if(WITH_XPU_XFT)
   set(XPU_XFT_LIB "${XPU_LIB_DIR}/${XPU_XFT_LIB_NAME}")
 endif()
 
+if(WITH_XPU_XHPC)
+  message(STATUS "Compile with XPU XHPC!")
+  add_definitions(-DPADDLE_WITH_XPU_XHPC)
+
+  set(XPU_XHPC_INC_DIR "${XPU_INC_DIR}/xhpc")
+  include_directories(${XPU_XHPC_INC_DIR})
+  set(XPU_XBLAS_INC_DIR "${XPU_INC_DIR}/xhpc/xblas")
+  include_directories(${XPU_XBLAS_INC_DIR})
+
+  set(XPU_XFA_INC_DIR "${XPU_INC_DIR}/xhpc/xfa")
+  include_directories(${XPU_XFA_INC_DIR})
+endif()
+
 if(WITH_XPTI)
   message(STATUS "Compile with XPU XPTI!")
   add_definitions(-DPADDLE_WITH_XPTI)
@@ -191,21 +221,18 @@ if(WITH_XPU_PLUGIN)
   include_directories(${CMAKE_SOURCE_DIR}/paddle/phi/kernels/xpu/plugin/include)
 endif()
 
-if(WITH_XPU_BKCL AND WITH_XPU_XFT)
-  target_link_libraries(xpulib ${XPU_API_LIB} ${XPU_RT_LIB} ${XPU_BKCL_LIB}
-                        ${XPU_XFT_LIB})
-elseif(WITH_XPU_BKCL)
-  target_link_libraries(xpulib ${XPU_API_LIB} ${XPU_RT_LIB} ${XPU_BKCL_LIB})
-elseif(WITH_XPU_XFT)
-  target_link_libraries(xpulib ${XPU_API_LIB} ${XPU_RT_LIB} ${XPU_XFT_LIB})
-else()
-  target_link_libraries(xpulib ${XPU_API_LIB} ${XPU_RT_LIB})
-endif()
+target_link_libraries(xpulib ${XPU_API_LIB} ${XPU_RT_LIB} ${XPU_BKCL_LIB}
+                      ${XPU_XFT_LIB})
 
 if(WITH_XPTI)
   target_link_libraries(xpulib ${XPU_XPTI_LIB})
 endif()
 
+if(WITH_XPU_XHPC)
+  target_link_libraries(xpulib ${XPU_API_LIB} ${XPU_RT_LIB} ${XPU_XBLAS_LIB}
+                        ${XPU_XFA_LIB})
+endif()
+
 add_dependencies(xpulib ${XPU_PROJECT})
 
 # Ensure that xpu/api.h can be included without dependency errors.

diff --git a/cmake/generic.cmake b/cmake/generic.cmake
@@ -598,6 +598,10 @@ function(paddle_test_build TARGET_NAME)
                           ${paddle_test_DEPS} common paddle_gtest_main_new)
     add_dependencies(${TARGET_NAME} ${paddle_lib} ${paddle_test_DEPS} common
                      paddle_gtest_main_new)
+    if(WITH_MKLDNN)
+      target_link_libraries(${TARGET_NAME} mkldnn)
+      add_dependencies(${TARGET_NAME} mkldnn)
+    endif()
     if(WITH_SHARED_PHI)
       target_link_libraries(${TARGET_NAME} $<TARGET_LINKER_FILE:phi>)
       add_dependencies(${TARGET_NAME} phi)
@@ -833,7 +837,6 @@ function(hip_test TARGET_NAME)
       ${hip_test_DEPS}
       paddle_gtest_main
       lod_tensor
-      memory
       gtest
       glog
       phi
@@ -843,7 +846,6 @@ function(hip_test TARGET_NAME)
       ${hip_test_DEPS}
       paddle_gtest_main
       lod_tensor
-      memory
       gtest
       phi
       glog)
@@ -940,7 +942,6 @@ function(xpu_test TARGET_NAME)
       ${xpu_test_DEPS}
       paddle_gtest_main
       lod_tensor
-      memory
       gtest
       phi
       glog
@@ -950,7 +951,6 @@ function(xpu_test TARGET_NAME)
       ${xpu_test_DEPS}
       paddle_gtest_main
       lod_tensor
-      memory
       gtest
       phi
       glog)

diff --git a/paddle/cinn/api/CMakeLists.txt b/paddle/cinn/api/CMakeLists.txt
@@ -1,5 +1,3 @@
 core_gather_headers()
 
 gather_srcs(cinnapi_src SRCS op_node.cc tensor_node.cc)
-
-message(STATUS "srcs: ${cinnapi_src}")
diff --git a/paddle/cinn/common/CMakeLists.txt b/paddle/cinn/common/CMakeLists.txt
@@ -20,9 +20,9 @@ gather_srcs(
   cas.cc
   union_find.cc
   python_interpreter_guard.cc
-  nvgpu_dev_info.cc)
-
-message(STATUS "srcs: ${cinnapi_src}")
+  nvgpu_dev_info.cc
+  integer_set.cc
+  dim_expr_simplify.cc)
 
 cinn_cc_test(test_equation_graph_topo_walker SRCS
              equation_graph_topo_walker_test.cc DEPS gtest glog)
@@ -38,10 +38,15 @@ cinn_cc_test(test_arithmatic SRCS arithmatic_test.cc DEPS cinncore)
 cinn_cc_test(test_cas SRCS cas_test.cc DEPS cinncore)
 cinn_cc_test(test_type SRCS type_test.cc DEPS cinncore)
 cinn_cc_test(test_axis SRCS axis_test.cc DEPS cinncore)
+cinn_cc_test(test_integer_set SRCS integer_set_test.cc DEPS cinncore)
 
 cinn_cc_test(test_fp16_bf16_host SRCS float16_bfloat16_host_test.cc DEPS gtest
              glog)
 if(WITH_CUDA)
   cinn_nv_test(test_fp16_bf16_cuda SRCS float16_bfloat16_cuda_test.cu DEPS
                gtest glog)
 endif()
+if(NOT CINN_ONLY)
+  cinn_cc_test(dim_expr_simplify_test SRCS dim_expr_simplify_test.cc DEPS
+               cinncore)
+endif()