Skip to content

Commit

Permalink
Merge branch 'develop' into group_norm_opt
Browse files Browse the repository at this point in the history
  • Loading branch information
@Zjq9409
Zjq9409 committed Feb 22, 2022
2 parents 641ab69 + 496aadf commit 58d7046
Show file tree
Hide file tree
Showing 2,211 changed files with 45,265 additions and 25,010 deletions.
18 changes: 10 additions & 8 deletions .gitignore
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2,16 +2,17 @@ paddle/fluid/operators/distributed/send_recv.proto
paddle/fluid/API.spec
paddle/fluid/API_DEV.spec
paddle/fluid/API_PR.spec
paddle/fluid/eager/api/generated/*
paddle/fluid/op_use_default_grad_maker_DEV.spec
paddle/fluid/op_use_default_grad_maker_PR.spec
paddle/pten/api/include/api.h
paddle/pten/api/lib/api.cc
paddle/pten/api/backward/backward_api.h
paddle/pten/api/lib/backward_api.cc
paddle/pten/include/*
paddle/pten/infermeta/generated.*
paddle/pten/extension.h
paddle/fluid/eager/api/generated/*
paddle/phi/api/backward/backward_api.h
paddle/phi/api/include/api.h
paddle/phi/api/lib/api.cc
paddle/phi/api/lib/dygraph_api.*
paddle/phi/api/lib/backward_api.cc
paddle/phi/extension.h
paddle/phi/include/*
paddle/phi/infermeta/generated.*

*.DS_Store
*.vs
Expand Down Expand Up @@ -51,5 +52,6 @@ paddle/infrt/dialect/pd_ops_info.h
.lit_test_times.txt
paddle/infrt/tests/dialect/Output
paddle/infrt/tests/lit.cfg.py
paddle/infrt/kernel/phi/infershaped/infershaped_kernel_launchers.cc
paddle/fluid/pybind/eager_final_state_op_function_impl.h
paddle/fluid/pybind/tmp_eager_final_state_op_function_impl.h
5 changes: 5 additions & 0 deletions CMakeLists.txt
View file
Original file line number Diff line number Diff line change
Expand Up @@ -243,6 +243,7 @@ option(NEW_RELEASE_JIT "PaddlePaddle next-level release strategy for backup ji
option(WITH_ASCEND_INT64 "Compile with int64 kernel for ascend NPU" OFF)
option(WITH_POCKETFFT "Compile with pocketfft support" ON)
option(WITH_RECORD_BUILDTIME "Compile PaddlePaddle with record all targets build time" OFF)
option(WITH_CUSTOM_DEVICE "Compile with custom device support" OFF)

if(WITH_RECORD_BUILDTIME)
set_property(GLOBAL PROPERTY RULE_LAUNCH_COMPILE "${CMAKE_CURRENT_SOURCE_DIR}/tools/get_build_time.sh")
Expand All @@ -265,6 +266,10 @@ if(SANITIZER_TYPE AND NOT "${SANITIZER_TYPE}" MATCHES "^(Address|Leak|Memory|Thr
return()
endif()

if (LINUX AND NOT WITH_CUSTOM_DEVICE AND NOT ON_INFER)
set(WITH_CUSTOM_DEVICE ON)
endif()

if(WIN32)
if(WITH_DISTRIBUTE)
MESSAGE(WARNING
Expand Down
4 changes: 4 additions & 0 deletions cmake/configure.cmake
View file
Original file line number Diff line number Diff line change
Expand Up @@ -219,3 +219,7 @@ endif(ON_INFER)
if(WITH_CRYPTO)
add_definitions(-DPADDLE_WITH_CRYPTO)
endif(WITH_CRYPTO)

if(WITH_CUSTOM_DEVICE AND NOT WIN32)
add_definitions(-DPADDLE_WITH_CUSTOM_DEVICE)
endif()
32 changes: 20 additions & 12 deletions cmake/external/poplar.cmake
View file
Original file line number Diff line number Diff line change
Expand Up @@ -11,31 +11,35 @@
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

if(WITH_IPU)
set(POPLAR_DIR CACHE PATH "Path to a Poplar install")
set(POPART_DIR CACHE PATH "Path to a Popart install")
set(POPLAR_SDK_DIR CACHE PATH "Path to an extracted SDK archive or to a Poplar & Popart install directory (Will populate POPLAR_DIR and POPART_DIR)")

# support setting SDK both from environment variable or command line arguments

if(DEFINED ENV{POPLAR_SDK_DIR})
set(POPLAR_SDK_DIR $ENV{POPLAR_SDK_DIR})
endif()
if(EXISTS ${POPLAR_SDK_DIR})
execute_process(COMMAND find ${POPLAR_SDK_DIR}/ -maxdepth 1 -type d -name "popart*"
OUTPUT_VARIABLE POPART_DIR OUTPUT_STRIP_TRAILING_WHITESPACE)
execute_process(COMMAND find ${POPLAR_SDK_DIR}/ -maxdepth 1 -type d -name "poplar-*" -o -name "poplar"
OUTPUT_VARIABLE POPLAR_DIR OUTPUT_STRIP_TRAILING_WHITESPACE)
if(NOT IS_DIRECTORY "${POPLAR_DIR}")
message(FATAL_ERROR "Couldn't find a \"poplar\" or \"poplar-*\" folder in '${POPLAR_SDK_DIR}'")
endif()
if(NOT IS_DIRECTORY "${POPART_DIR}")
message(FATAL_ERROR "Couldn't find a \"popart*\" folder in '${POPLAR_SDK_DIR}'")
endif()
else()
message(FATAL_ERROR "You must provide a path to a Poplar install using export POPLAR_SDK_DIR=/path/to/poplar_sdk")
endif()

message("POPLAR_DIR is ${POPLAR_DIR}")
message("POPART_DIR is ${POPART_DIR}")
if(DEFINED ENV{POPLAR_DIR})
set(POPLAR_DIR $ENV{POPLAR_DIR})
endif()
if(DEFINED ENV{POPART_DIR})
set(POPART_DIR $ENV{POPART_DIR})
endif()

if(EXISTS ${POPLAR_DIR})
message("POPLAR_DIR is ${POPLAR_DIR}")
if(NOT IS_DIRECTORY "${POPLAR_DIR}")
message(FATAL_ERROR "Couldn't find a \"poplar\" or \"poplar-*\" folder in '${POPLAR_SDK_DIR}'")
endif()
list(APPEND CMAKE_PREFIX_PATH ${POPLAR_DIR})
set(ENABLE_POPLAR_CMD "source ${POPLAR_DIR}/enable.sh")
find_package(poplar REQUIRED)
Expand All @@ -45,8 +49,11 @@ if(WITH_IPU)
if(NOT poplar_FOUND)
message(FATAL_ERROR "You must provide a path to a Poplar install using -DPOPLAR_DIR=/path/to/popart/build/install")
endif()

if(EXISTS ${POPART_DIR})
message("POPART_DIR is ${POPART_DIR}")
if(NOT IS_DIRECTORY "${POPART_DIR}")
message(FATAL_ERROR "Couldn't find a \"popart*\" folder in '${POPLAR_SDK_DIR}'")
endif()
list(APPEND CMAKE_PREFIX_PATH ${POPART_DIR})
set(ENABLE_POPART_CMD "source ${POPART_DIR}/enable.sh")
find_package(popart REQUIRED COMPONENTS popart-only)
Expand All @@ -56,6 +63,7 @@ if(WITH_IPU)
if(NOT popart_FOUND)
message(FATAL_ERROR "You must provide a path to a Popart build using -DPOPART_DIR=/path/to/popart/build")
endif()

add_definitions(-DONNX_NAMESPACE=onnx)
add_custom_target(extern_poplar DEPENDS poplar popart-only)
endif()
2 changes: 1 addition & 1 deletion cmake/external/xpu.cmake
View file
Original file line number Diff line number Diff line change
Expand Up @@ -36,7 +36,7 @@ ENDIF()

if(NOT DEFINED XPU_BASE_URL)
SET(XPU_BASE_URL_WITHOUT_DATE "https://baidu-kunlun-product.cdn.bcebos.com/KL-SDK/klsdk-dev")
SET(XPU_BASE_URL "${XPU_BASE_URL_WITHOUT_DATE}/20220119")
SET(XPU_BASE_URL "${XPU_BASE_URL_WITHOUT_DATE}/20220215")
else()
SET(XPU_BASE_URL "${XPU_BASE_URL}")
endif()
Expand Down
41 changes: 40 additions & 1 deletion cmake/generic.cmake
View file
Original file line number Diff line number Diff line change
Expand Up @@ -122,7 +122,7 @@ set_property(GLOBAL PROPERTY PTEN_MODULES "")
function(find_pten_modules TARGET_NAME)
get_filename_component(__target_path ${TARGET_NAME} ABSOLUTE)
string(REGEX REPLACE "^${PADDLE_SOURCE_DIR}/" "" __target_path ${__target_path})
string(FIND "${__target_path}" "pten" pos)
string(FIND "${__target_path}" "phi" pos)
if(pos GREATER 1)
get_property(pten_modules GLOBAL PROPERTY PTEN_MODULES)
set(pten_modules ${pten_modules} ${TARGET_NAME})
Expand Down Expand Up @@ -1036,3 +1036,42 @@ function(generate_dummy_static_lib)
add_library(${dummy_LIB_NAME} STATIC ${dummy_FILE_PATH})
endfunction()

function(math_library TARGET)
# math_library is a function to create math library.
# The interface is the same as cc_library.
# But it handle split GPU/CPU code and link some common library.
set(cc_srcs)
set(cu_srcs)
set(hip_srcs)
set(math_common_deps device_context framework_proto enforce)
if (WITH_GPU)
if (${CMAKE_CUDA_COMPILER_VERSION} LESS 11.0)
list(APPEND math_common_deps cub)
else()
list(APPEND math_common_deps)
endif()
endif()
set(multiValueArgs DEPS)
cmake_parse_arguments(math_library "${options}" "${oneValueArgs}"
"${multiValueArgs}" ${ARGN})

if (EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/${TARGET}.cc)
list(APPEND cc_srcs ${TARGET}.cc)
endif()
if (EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/${TARGET}.cu)
list(APPEND cu_srcs ${TARGET}.cu)
endif()
if (EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/${TARGET}.cu.cc)
list(APPEND cu_srcs ${TARGET}.cu.cc)
endif()

list(LENGTH cc_srcs cc_srcs_len)
if (WITH_GPU)
nv_library(${TARGET} SRCS ${cc_srcs} ${cu_srcs} DEPS ${math_library_DEPS} ${math_common_deps})
elseif (WITH_ROCM)
hip_library(${TARGET} SRCS ${cc_srcs} ${cu_srcs} DEPS ${math_library_DEPS} ${math_common_deps})
elseif(${cc_srcs_len} GREATER 0)
cc_library(${TARGET} SRCS ${cc_srcs} DEPS ${math_library_DEPS} ${math_common_deps})
endif()
endfunction()

16 changes: 8 additions & 8 deletions cmake/inference_lib.cmake
View file
Original file line number Diff line number Diff line change
Expand Up @@ -226,17 +226,17 @@ include_directories(${CMAKE_BINARY_DIR}/../paddle/fluid/framework/io)

# copy api headers for pten & custom op
copy(inference_lib_dist
SRCS ${PADDLE_SOURCE_DIR}/paddle/pten/api/ext/*.h
DSTS ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/pten/api/ext/)
SRCS ${PADDLE_SOURCE_DIR}/paddle/phi/api/ext/*.h
DSTS ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/phi/api/ext/)
copy(inference_lib_dist
SRCS ${PADDLE_SOURCE_DIR}/paddle/pten/api/include/*.h
DSTS ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/pten/api/include/)
SRCS ${PADDLE_SOURCE_DIR}/paddle/phi/api/include/*.h
DSTS ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/phi/api/include/)
copy(inference_lib_dist
SRCS ${PADDLE_SOURCE_DIR}/paddle/pten/api/all.h
DSTS ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/pten/api/)
SRCS ${PADDLE_SOURCE_DIR}/paddle/phi/api/all.h
DSTS ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/phi/api/)
copy(inference_lib_dist
SRCS ${PADDLE_SOURCE_DIR}/paddle/pten/common/*.h
DSTS ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/pten/common/)
SRCS ${PADDLE_SOURCE_DIR}/paddle/phi/common/*.h
DSTS ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/phi/common/)
copy(inference_lib_dist
SRCS ${PADDLE_SOURCE_DIR}/paddle/utils/any.h
DSTS ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/utils/)
Expand Down
23 changes: 13 additions & 10 deletions cmake/pten.cmake
View file
Original file line number Diff line number Diff line change
Expand Up @@ -33,7 +33,7 @@ function(generate_unify_header DIR_NAME)

# generate target header file
set(header_file ${CMAKE_CURRENT_SOURCE_DIR}/include/${header_name}.h)
file(WRITE ${header_file} "// Header file generated by paddle/pten/CMakeLists.txt for external users,\n// DO NOT edit or include it within paddle.\n\n#pragma once\n\n")
file(WRITE ${header_file} "// Header file generated by paddle/phi/CMakeLists.txt for external users,\n// DO NOT edit or include it within paddle.\n\n#pragma once\n\n")

# get all top-level headers and write into header file
file(GLOB HEADERS "${CMAKE_CURRENT_SOURCE_DIR}\/${DIR_NAME}\/*.h")
Expand Down Expand Up @@ -114,6 +114,9 @@ function(kernel_library TARGET)
if (EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/gpu/${TARGET}.cu)
list(APPEND gpu_srcs ${CMAKE_CURRENT_SOURCE_DIR}/gpu/${TARGET}.cu)
endif()
if (EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/gpu/${TARGET}.cu.cc)
list(APPEND gpu_srcs ${CMAKE_CURRENT_SOURCE_DIR}/gpu/${TARGET}.cu.cc)
endif()
endif()
if (WITH_XPU)
if (EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/xpu/${TARGET}.cc)
Expand All @@ -134,17 +137,17 @@ function(kernel_library TARGET)
list(APPEND all_srcs ${xpu_srcs})
foreach(src ${all_srcs})
file(READ ${src} target_content)
string(REGEX MATCHALL "#include \"paddle\/pten\/kernels\/[a-z0-9_]+_kernel.h\"" include_kernels ${target_content})
string(REGEX MATCHALL "#include \"paddle\/phi\/kernels\/[a-z0-9_]+_kernel.h\"" include_kernels ${target_content})
if ("${kernel_library_SUB_DIR}" STREQUAL "")
string(REGEX MATCHALL "#include \"paddle\/pten\/kernels\/[a-z0-9_]+_kernel.h\"" include_kernels ${target_content})
string(REGEX MATCHALL "#include \"paddle\/phi\/kernels\/[a-z0-9_]+_kernel.h\"" include_kernels ${target_content})
else()
string(REGEX MATCHALL "#include \"paddle\/pten\/kernels\/${kernel_library_SUB_DIR}\/[a-z0-9_]+_kernel.h\"" include_kernels ${target_content})
string(REGEX MATCHALL "#include \"paddle\/phi\/kernels\/${kernel_library_SUB_DIR}\/[a-z0-9_]+_kernel.h\"" include_kernels ${target_content})
endif()
foreach(include_kernel ${include_kernels})
if ("${kernel_library_SUB_DIR}" STREQUAL "")
string(REGEX REPLACE "#include \"paddle\/pten\/kernels\/" "" kernel_name ${include_kernel})
string(REGEX REPLACE "#include \"paddle\/phi\/kernels\/" "" kernel_name ${include_kernel})
else()
string(REGEX REPLACE "#include \"paddle\/pten\/kernels\/${kernel_library_SUB_DIR}\/" "" kernel_name ${include_kernel})
string(REGEX REPLACE "#include \"paddle\/phi\/kernels\/${kernel_library_SUB_DIR}\/" "" kernel_name ${include_kernel})
endif()
string(REGEX REPLACE ".h\"" "" kernel_name ${kernel_name})
list(APPEND kernel_deps ${kernel_name})
Expand All @@ -161,7 +164,7 @@ function(kernel_library TARGET)

# Build Target according different src organization
if((${cpu_srcs_len} GREATER 0 OR ${gpu_srcs_len} GREATER 0 OR
${xpu_srcs_len} GREATER 0) AND (${common_srcs_len} GREATER 0 OR
${xpu_srcs_len} GREATER 0) AND (${common_srcs_len} GREATER 0 OR
${selected_rows_srcs_len} GREATER 0))
# If the common_srcs/selected_rows_srcs depends on specific device srcs, build target using this rule.
if (WITH_GPU)
Expand Down Expand Up @@ -225,11 +228,11 @@ function(kernel_library TARGET)
cc_library(${TARGET} SRCS ${selected_rows_srcs} DEPS ${kernel_library_DEPS} ${kernel_deps})
endif()
else()
message(FATAL_ERROR "Cannot find any implementation for ${TARGET}")
message(FATAL_ERROR "Cannot find any implementation for ${TARGET}")
endif()

if (${common_srcs_len} GREATER 0 OR ${cpu_srcs_len} GREATER 0 OR
${gpu_srcs_len} GREATER 0 OR ${xpu_srcs_len} GREATER 0 OR
${gpu_srcs_len} GREATER 0 OR ${xpu_srcs_len} GREATER 0 OR
${selected_rows_srcs_len} GREATER 0)
# append target into PTEN_KERNELS property
get_property(pten_kernels GLOBAL PROPERTY PTEN_KERNELS)
Expand Down Expand Up @@ -285,7 +288,7 @@ function(append_op_util_declare TARGET)
string(REGEX MATCH "(PT_REGISTER_BASE_KERNEL_NAME|PT_REGISTER_ARG_MAPPING_FN)\\([ \t\r\n]*[a-z0-9_]*" util_registrar "${target_content}")
string(REPLACE "PT_REGISTER_ARG_MAPPING_FN" "PT_DECLARE_ARG_MAPPING_FN" util_declare "${util_registrar}")
string(REPLACE "PT_REGISTER_BASE_KERNEL_NAME" "PT_DECLARE_BASE_KERNEL_NAME" util_declare "${util_declare}")
string(APPEND util_declare ");")
string(APPEND util_declare ");\n")
file(APPEND ${op_utils_header} "${util_declare}")
endfunction()

Expand Down
10 changes: 5 additions & 5 deletions cmake/pten_header.cmake
View file
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,7 @@ if (pos GREATER 1)
foreach(header ${HEADERS})
if (${header} MATCHES ".*.h$")
file(READ ${header} HEADER_CONTENT)
string(REPLACE "paddle/pten/" "paddle/include/experimental/pten/" HEADER_CONTENT "${HEADER_CONTENT}")
string(REPLACE "paddle/phi/" "paddle/include/experimental/phi/" HEADER_CONTENT "${HEADER_CONTENT}")
string(REPLACE "paddle/utils/" "paddle/include/experimental/utils/" HEADER_CONTENT "${HEADER_CONTENT}")
file(WRITE ${header} "${HEADER_CONTENT}")
message(STATUS "pten header path compat processing complete: ${header}")
Expand All @@ -32,10 +32,10 @@ endif()
endfunction()

pten_header_path_compat(${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental)
pten_header_path_compat(${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/pten/api)
pten_header_path_compat(${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/pten/api/ext)
pten_header_path_compat(${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/pten/api/include)
pten_header_path_compat(${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/pten/common)
pten_header_path_compat(${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/phi/api)
pten_header_path_compat(${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/phi/api/ext)
pten_header_path_compat(${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/phi/api/include)
pten_header_path_compat(${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/phi/common)

# In order to be compatible with the original behavior, the header file name needs to be changed
file(RENAME ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/include/experimental/extension.h
Expand Down
2 changes: 1 addition & 1 deletion paddle/CMakeLists.txt
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,6 @@ add_subdirectory(utils)
add_subdirectory(scripts)
add_subdirectory(testing)
set(PYTHON_TESTS_DIR ${PADDLE_BINARY_DIR}/python/paddle/fluid/tests CACHE INTERNAL "python tests directory")
add_subdirectory(pten)
add_subdirectory(phi)
add_subdirectory(infrt)
add_subdirectory(fluid)
2 changes: 1 addition & 1 deletion paddle/extension.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -15,4 +15,4 @@ limitations under the License. */
#pragma once

// All paddle apis in C++ frontend
#include "paddle/pten/api/all.h"
#include "paddle/phi/api/all.h"
2 changes: 1 addition & 1 deletion paddle/fluid/distributed/common/sparse_sharding_merge.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -24,11 +24,11 @@
#include "glog/logging.h"
#include "paddle/fluid/distributed/common/utils.h"
#include "paddle/fluid/framework/blocking_queue.h"
#include "paddle/fluid/framework/dim.h"
#include "paddle/fluid/framework/framework.pb.h"
#include "paddle/fluid/framework/tensor.h"
#include "paddle/fluid/framework/tensor_util.h"
#include "paddle/fluid/string/split.h"
#include "paddle/phi/core/utils/dim.h"

constexpr int FG = 256 * 1024 * 1024;
constexpr int Q_SIZE = 10000;
Expand Down
8 changes: 3 additions & 5 deletions paddle/fluid/distributed/common/utils.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -24,18 +24,16 @@
#include <utility>
#include <vector>

#include "paddle/fluid/operators/math/blas.h"
#include "paddle/fluid/platform/device_context.h"
#include "paddle/phi/kernels/funcs/blas/blas.h"

namespace paddle {
namespace distributed {

template <typename T>
inline paddle::operators::math::BlasT<paddle::platform::CPUDeviceContext, T>
GetBlas() {
inline phi::funcs::BlasT<paddle::platform::CPUDeviceContext, T> GetBlas() {
paddle::platform::CPUDeviceContext cpu_ctx;
return paddle::operators::math::GetBlas<paddle::platform::CPUDeviceContext,
T>(cpu_ctx);
return phi::funcs::GetBlas<paddle::platform::CPUDeviceContext, T>(cpu_ctx);
}

template <typename T>
Expand Down
Loading

0 comments on commit 58d7046

Please sign in to comment.