diff --git a/CMakeLists.txt b/CMakeLists.txt index baf05030..2a30c479 100755 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -56,7 +56,7 @@ if (ENABLE_OPEN_SRC) set(GE_LIB_PATH ${GE_LIB_PATH}/${GE_SYS_ARCH}) set(STATIC_ACL_LIB ${GE_LIB_PATH}) find_module(slog libslog.so ${GE_LIB_PATH}) - find_module(mmpa libmmpa.so ${GE_LIB_PATH}) + find_module(static_mmpa libmmpa.a ${GE_LIB_PATH}) find_module(msprof libmsprof.so ${GE_LIB_PATH}) find_module(hccl libhccl.so ${GE_LIB_PATH}) find_module(adump_server libadump_server.a ${GE_LIB_PATH}) @@ -67,10 +67,10 @@ if (ENABLE_OPEN_SRC) find_module(ascend_hal_stub libascend_hal.so ${GE_LIB_PATH}) find_module(error_manager_static liberror_manager.a ${GE_LIB_PATH}) find_module(msprofiler libmsprofiler.a ${GE_LIB_PATH}) - #find_module(ascendcl_static libascendcl.a ${GE_LIB_PATH}) + #find_module(ascendcl_static libascendcl.a ${GE_LIB_PATH}) else() find_module(slog libslog.so ${ASCEND_ATC_DIR}) - find_module(mmpa libmmpa.so ${ASCEND_ATC_DIR}) + find_module(static_mmpa libmmpa.a ${ASCEND_ATC_DIR}) if(PLATFORM STREQUAL "train") find_module(msprof libmsprof.so ${ASCEND_DRIVER_COMMON_DIR}) find_module(hccl libhccl.so ${ASCEND_RUNTIME_DIR}) @@ -91,7 +91,7 @@ if (ENABLE_OPEN_SRC) find_module(error_manager liberror_manager.so ${ASCEND_ATC_DIR}) find_module(error_manager_static liberror_manager.a ${ASCEND_ACL_DIR}) find_module(msprofiler libmsprofiler.a ${ASCEND_ACL_DIR}) - #find_module(ascendcl_static libascendcl.a ${ASCEND_ACL_DIR}) + #find_module(ascendcl_static libascendcl.a ${ASCEND_ACL_DIR}) if(PRODUCT STREQUAL "flr3") find_module(msprof libmsprof.so ${ASCEND_DRIVER_SHARE_DIR}) elseif(PRODUCT STREQUAL "flr1") @@ -114,7 +114,7 @@ if (ENABLE_OPEN_SRC) find_module(error_manager_static liberror_manager.a ${ASCEND_ACL_DIR}) find_module(msprofiler libmsprofiler.a ${ASCEND_ACL_DIR}) find_module(ascend_hal_stub libascend_hal.so ${ASCEND_DRIVER_DIR}/driver) - #find_module(ascendcl_static libascendcl.a ${ASCEND_ACL_DIR}) + #find_module(ascendcl_static libascendcl.a ${ASCEND_ACL_DIR}) else() message(FATAL_ERROR "PLATFORM param is invalid, should be train or inference, build terminated") endif() @@ -148,16 +148,23 @@ elseif (ENABLE_D OR ENABLE_ACL) # common libraries find_module(slog libslog.so ${ASCEND_MS_DRIVER_PATH}) find_module(error_manager liberror_manager.so ${ASCEND_MS_RUNTIME_PATH} ${ATLAS_MS_RUNTIME_PATH}) + find_module(static_mmpa libmmpa.a ${ASCEND_MS_RUNTIME_PATH} ${ATLAS_MS_RUNTIME_PATH}) if (ENABLE_D) # training - find_module(mmpa libmmpa.so ${ASCEND_MS_DRIVER_PATH}) find_module(runtime libruntime.so ${ASCEND_MS_RUNTIME_PATH} ${ATLAS_MS_RUNTIME_PATH}) find_module(register libregister.so ${ASCEND_MS_RUNTIME_PATH} ${ATLAS_MS_RUNTIME_PATH}) endif () set(METADEF_DIR ${CMAKE_CURRENT_LIST_DIR}/metadef) add_subdirectory(metadef) +elseif(ENABLE_MS_TESTCASE) + include(cmake/external_libs/protobuf_static.cmake) + include(cmake/external_libs/securec.cmake) + include(cmake/intf_pub_linux.cmake) + + set(METADEF_DIR ${CMAKE_CURRENT_LIST_DIR}/metadef) + add_subdirectory(metadef) else() set(METADEF_DIR ${CMAKE_CURRENT_LIST_DIR}/../metadef) set(PARSER_DIR ${CMAKE_CURRENT_LIST_DIR}/../parser) diff --git a/cmake/external_libs/protobuf_static.cmake b/cmake/external_libs/protobuf_static.cmake index 51f4ed25..23630d44 100755 --- a/cmake/external_libs/protobuf_static.cmake +++ b/cmake/external_libs/protobuf_static.cmake @@ -48,7 +48,7 @@ set_target_properties(ascend_protobuf_static_lib PROPERTIES add_library(ascend_protobuf_static INTERFACE) target_include_directories(ascend_protobuf_static INTERFACE ${PROTOBUF_STATIC_PKG_DIR}/include) target_link_libraries(ascend_protobuf_static INTERFACE ascend_protobuf_static_lib) -if (ENABLE_D OR ENABLE_ACL) +if (ENABLE_D OR ENABLE_ACL OR ENABLE_MS_TESTCASES) include_directories(${PROTOBUF_STATIC_PKG_DIR}/include) endif () diff --git a/ge/CMakeLists.txt b/ge/CMakeLists.txt index 6eba7bc4..88a5c52f 100755 --- a/ge/CMakeLists.txt +++ b/ge/CMakeLists.txt @@ -1,4 +1,4 @@ -if (NOT ENABLE_D AND NOT ENABLE_ACL) +if (NOT ENABLE_D AND NOT ENABLE_ACL AND NOT ENABLE_MS_TESTCASES) add_subdirectory(common) add_subdirectory(plugin/engine) add_subdirectory(graph/build/memory) @@ -600,7 +600,7 @@ set(INFER_SRC_LIST "analyzer/analyzer.cc" ) -if (NOT ENABLE_D AND NOT ENABLE_ACL) +if (NOT ENABLE_D AND NOT ENABLE_ACL AND NOT ENABLE_MS_TESTCASES) ############ libge_runner.so ############ add_library(ge_runner SHARED ${TRAIN_SRC_LIST} ${PROTO_SRCS} ${PROTO_CLIENT_SRCS}) @@ -648,6 +648,7 @@ target_link_libraries(ge_runner ge_memory adump_server msprofiler + static_mmpa -Wl,--no-as-needed graph ge_common @@ -655,7 +656,6 @@ target_link_libraries(ge_runner register c_sec slog - mmpa msprof runtime resource @@ -712,6 +712,7 @@ target_include_directories(ge_compiler PRIVATE target_link_libraries(ge_compiler $ ge_memory + static_mmpa -Wl,--no-as-needed graph ge_common @@ -720,7 +721,6 @@ target_link_libraries(ge_compiler c_sec error_manager slog - mmpa runtime_compile resource -Wl,--as-needed @@ -770,6 +770,7 @@ target_link_libraries(opensrc_ascendcl PRIVATE ge_executor ge_common_static graph_static + static_mmpa ascend_protobuf_static register_static error_manager_static @@ -779,11 +780,11 @@ target_link_libraries(opensrc_ascendcl PRIVATE -Wl,--no-as-needed c_sec runtime - mmpa slog msprof ascend_hal_stub -Wl,--as-needed + -lrt -ldl json ) diff --git a/ge/client/ge_api.cc b/ge/client/ge_api.cc index 5619f137..ef8d5622 100644 --- a/ge/client/ge_api.cc +++ b/ge/client/ge_api.cc @@ -177,7 +177,7 @@ Session::Session(const std::map &options) { // check init status sessionId_ = 0; if (!g_ge_initialized) { - GELOGE(GE_CLI_GE_NOT_INITIALIZED); + GELOGE(GE_CLI_GE_NOT_INITIALIZED, "GE is not initialized."); return; } // call Initialize diff --git a/ge/common/CMakeLists.txt b/ge/common/CMakeLists.txt index 444f93a2..728971b6 100755 --- a/ge/common/CMakeLists.txt +++ b/ge/common/CMakeLists.txt @@ -105,6 +105,7 @@ target_include_directories(ge_common PRIVATE target_link_libraries(ge_common PRIVATE $ + static_mmpa -Wl,--no-as-needed graph ascend_protobuf @@ -112,7 +113,6 @@ target_link_libraries(ge_common PRIVATE c_sec error_manager slog - mmpa -Wl,--as-needed json -lrt @@ -210,7 +210,7 @@ target_link_libraries(ge_common PRIVATE c_sec error_manager slog - mmpa + static_mmpa -Wl,--as-needed json -lrt diff --git a/ge/common/auth/file_saver.cc b/ge/common/auth/file_saver.cc index 60d99c0b..7b41397a 100755 --- a/ge/common/auth/file_saver.cc +++ b/ge/common/auth/file_saver.cc @@ -16,9 +16,7 @@ #include "common/auth/file_saver.h" -#include #include -#include #include #include #include @@ -39,12 +37,12 @@ Status FileSaver::OpenFile(int32_t &fd, const std::string &file_path) { return FAILED; } - char real_path[PATH_MAX] = {0}; - GE_IF_BOOL_EXEC(realpath(file_path.c_str(), real_path) == nullptr, + char real_path[MMPA_MAX_PATH] = {0}; + GE_IF_BOOL_EXEC(mmRealPath(file_path.c_str(), real_path, MMPA_MAX_PATH) != EN_OK, GELOGI("File %s is not exist, it will be created.", file_path.c_str())); // Open file - mode_t mode = S_IRUSR | S_IWUSR; - fd = mmOpen2(real_path, O_RDWR | O_CREAT | O_TRUNC, mode); + mmMode_t mode = M_IRUSR | M_IWUSR; + fd = mmOpen2(real_path, M_RDWR | M_CREAT | O_TRUNC, mode); if (fd == EN_INVALID_PARAM || fd == EN_ERROR) { // -1: Failed to open file; - 2: Illegal parameter GELOGE(FAILED, "Open file failed. mmpa_errno = %d, %s", fd, strerror(errno)); @@ -194,7 +192,7 @@ Status FileSaver::SaveToBuffWithFileHeader(const ModelFileHeader &file_header, FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status FileSaver::CheckPath(const std::string &file_path) { // Determine file path length - if (file_path.size() >= PATH_MAX) { + if (file_path.size() >= MMPA_MAX_PATH) { GELOGE(FAILED, "Path is too long:%zu", file_path.size()); return FAILED; } diff --git a/ge/common/cust_aicpu_kernel_store.cc b/ge/common/cust_aicpu_kernel_store.cc index 86881b0e..0bf4d819 100755 --- a/ge/common/cust_aicpu_kernel_store.cc +++ b/ge/common/cust_aicpu_kernel_store.cc @@ -25,7 +25,7 @@ void CustAICPUKernelStore::AddCustAICPUKernel(const CustAICPUKernelPtr &kernel) } void CustAICPUKernelStore::LoadCustAICPUKernelBinToOpDesc(const std::shared_ptr &op_desc) const { - GELOGI("LoadCustAICPUKernelBinToOpDesc in"); + GELOGD("LoadCustAICPUKernelBinToOpDesc in"); if (op_desc != nullptr) { auto kernel_bin = FindKernel(op_desc->GetName()); if (kernel_bin != nullptr) { @@ -34,6 +34,6 @@ void CustAICPUKernelStore::LoadCustAICPUKernelBinToOpDesc(const std::shared_ptr< GELOGI("Load cust aicpu kernel:%s, %zu", kernel_bin->GetName().c_str(), kernel_bin->GetBinDataSize()); } } - GELOGI("LoadCustAICPUKernelBinToOpDesc success"); + GELOGD("LoadCustAICPUKernelBinToOpDesc success"); } } // namespace ge diff --git a/ge/common/debug/memory_dumper.cc b/ge/common/debug/memory_dumper.cc index d2b8d674..872fe1da 100644 --- a/ge/common/debug/memory_dumper.cc +++ b/ge/common/debug/memory_dumper.cc @@ -16,9 +16,6 @@ #include "common/debug/memory_dumper.h" -#include - -#include #include #include "framework/common/debug/log.h" @@ -138,26 +135,26 @@ int MemoryDumper::OpenFile(const char *filename) { } // Get the absolute path string real_path; - char tmp_path[PATH_MAX] = {0}; + char tmp_path[MMPA_MAX_PATH] = {0}; GE_IF_BOOL_EXEC( -1 != path_split_pos, string prefix_path = std::string(filename).substr(0, path_split_pos); string last_path = std::string(filename).substr(path_split_pos, strlen(filename) - 1); - GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(prefix_path.length() >= PATH_MAX, return kInvalidFd, "Prefix path is too long!"); - GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(realpath(prefix_path.c_str(), tmp_path) == nullptr, return kInvalidFd, + GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(prefix_path.length() >= MMPA_MAX_PATH, return kInvalidFd, "Prefix path is too long!"); + GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(mmRealPath(prefix_path.c_str(), tmp_path, MMPA_MAX_PATH) != EN_OK, return kInvalidFd, "Dir %s does not exit.", prefix_path.c_str()); real_path = std::string(tmp_path) + last_path;) GE_IF_BOOL_EXEC( path_split_pos == -1 || path_split_pos == 0, - GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(strlen(filename) >= PATH_MAX, return kInvalidFd, "Prefix path is too long!"); - GE_IF_BOOL_EXEC(realpath(filename, tmp_path) == nullptr, + GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(strlen(filename) >= MMPA_MAX_PATH, return kInvalidFd, "Prefix path is too long!"); + GE_IF_BOOL_EXEC(mmRealPath(filename, tmp_path, MMPA_MAX_PATH) != EN_OK, GELOGI("File %s does not exit, it will be created.", filename)); real_path = std::string(tmp_path);) // Open file, only the current user can read and write, to avoid malicious application access // Using the O_EXCL, if the file already exists,return failed to avoid privilege escalation vulnerability. - mode_t mode = S_IRUSR | S_IWUSR; + mmMode_t mode = M_IRUSR | M_IWUSR; - int32_t fd = mmOpen2(real_path.c_str(), O_RDWR | O_CREAT | O_APPEND, mode); + int32_t fd = mmOpen2(real_path.c_str(), M_RDWR | M_CREAT | O_TRUNC, mode); if (fd == EN_ERROR || fd == EN_INVALID_PARAM) { GELOGE(kInvalidFd, "open file failed. errno = %d, %s", fd, strerror(errno)); return kInvalidFd; diff --git a/ge/common/formats/format_transfers/format_transfer_nchw_fz_c04.cc b/ge/common/formats/format_transfers/format_transfer_nchw_fz_c04.cc index 275d9f10..a66aeeb4 100644 --- a/ge/common/formats/format_transfers/format_transfer_nchw_fz_c04.cc +++ b/ge/common/formats/format_transfers/format_transfer_nchw_fz_c04.cc @@ -118,19 +118,19 @@ Status TransFormatFromNchwToFzC04(const TransArgs &args, TransResult &result) { // data overflow check totally GE_IF_BOOL_EXEC(!CheckInt64MulOverflow(h_o, w_o), - GELOGE(INTERNAL_ERROR, "int64 mul overflow.A[%lld], B[%lld]", h_o, w_o); + GELOGE(INTERNAL_ERROR, "int64 mul overflow.A[%ld], B[%ld]", h_o, w_o); return INTERNAL_ERROR); GE_IF_BOOL_EXEC(!CheckInt64MulOverflow(n_o, c_o), - GELOGE(INTERNAL_ERROR, "int64 mul overflow.A[%lld], B[%lld]", n_o, c_o); + GELOGE(INTERNAL_ERROR, "int64 mul overflow.A[%ld], B[%ld]", n_o, c_o); return INTERNAL_ERROR); auto t1 = h_o * w_o; auto t2 = n_o * c_o; - GE_IF_BOOL_EXEC(!CheckInt64MulOverflow(t1, t2), GELOGE(INTERNAL_ERROR, "int64 mul overflow.A[%lld], B[%lld]", t1, t2); + GE_IF_BOOL_EXEC(!CheckInt64MulOverflow(t1, t2), GELOGE(INTERNAL_ERROR, "int64 mul overflow.A[%ld], B[%ld]", t1, t2); return INTERNAL_ERROR); int64_t total_ele_cnt = n_o * c_o * h_o * w_o; GE_IF_BOOL_EXEC(!CheckInt64MulOverflow(total_ele_cnt, size), - GELOGE(INTERNAL_ERROR, "int64 mul overflow.A[%lld], B[%lld]", total_ele_cnt, size); + GELOGE(INTERNAL_ERROR, "int64 mul overflow.A[%ld], B[%d]", total_ele_cnt, size); return INTERNAL_ERROR); int64_t dst_size = total_ele_cnt * size; if (dst_size == 0) { @@ -205,20 +205,20 @@ Status PaddingNC(const TransArgs &args, TransArgs &args_tmp, std::shared_ptr #include -#include #include #include #include @@ -38,8 +36,10 @@ const char *const kExt = ".so"; // supported extension of shared obje namespace ge { void PluginManager::ClearHandles_() noexcept { for (const auto &handle : handles_) { - if (dlclose(handle.second) != 0) { - GELOGW("Failed to close handle of %s: %s", handle.first.c_str(), dlerror()); + if (mmDlclose(handle.second) != 0) { + const char *error = mmDlerror(); + GE_IF_BOOL_EXEC(error == nullptr, error = ""); + GELOGW("Failed to close handle of %s: %s", handle.first.c_str(), error); } } handles_.clear(); @@ -48,18 +48,18 @@ void PluginManager::ClearHandles_() noexcept { PluginManager::~PluginManager() { ClearHandles_(); } string PluginManager::GetPath() { - Dl_info dl_info; - if (dladdr(reinterpret_cast(&PluginManager::GetPath), &dl_info) == 0) { + mmDlInfo dl_info; + if (mmDladdr(reinterpret_cast(&PluginManager::GetPath), &dl_info) != EN_OK) { GELOGW("Failed to read the shared library file path!"); return string(); } else { std::string so_path = dl_info.dli_fname; - char path[PATH_MAX] = {0}; - if (so_path.length() >= PATH_MAX) { + char path[MMPA_MAX_PATH] = {0}; + if (so_path.length() >= MMPA_MAX_PATH) { GELOGW("The shared library file path is too long!"); return string(); } - if (realpath(so_path.c_str(), path) == nullptr) { + if (mmRealPath(so_path.c_str(), path, MMPA_MAX_PATH) != EN_OK) { GELOGW("Failed to get realpath of %s", so_path.c_str()); return string(); } @@ -93,7 +93,7 @@ Status PluginManager::LoadSo(const string &path, const vector &func_chec std::vector path_vec; SplitPath(path, path_vec); for (const auto &single_path : path_vec) { - GE_IF_BOOL_EXEC(single_path.length() >= PATH_MAX, GELOGE(GE_PLGMGR_PATH_INVALID, + GE_IF_BOOL_EXEC(single_path.length() >= MMPA_MAX_PATH, GELOGE(GE_PLGMGR_PATH_INVALID, "The shared library file path is too long!"); continue); // load break when number of loaded so reach maximum @@ -119,16 +119,18 @@ Status PluginManager::LoadSo(const string &path, const vector &func_chec GELOGI("dlopen the shared library path name: %s.", file_path_dlopen.c_str()); // load continue when dlopen is failed - auto handle = dlopen(file_path_dlopen.c_str(), RTLD_NOW | RTLD_GLOBAL); + auto handle = mmDlopen(file_path_dlopen.c_str(), MMPA_RTLD_NOW | MMPA_RTLD_GLOBAL); if (handle == nullptr) { - GELOGE(GE_PLGMGR_PATH_INVALID, "Failed to dlopen %s!", dlerror()); + const char *error = mmDlerror(); + GE_IF_BOOL_EXEC(error == nullptr, error = ""); + GELOGE(GE_PLGMGR_PATH_INVALID, "Failed to dlopen %s!", error); continue; } // load continue when so is invalid bool is_valid = true; for (const auto &func_name : func_check_list) { - auto real_fn = (void (*)())dlsym(handle, func_name.c_str()); + auto real_fn = (void (*)())mmDlsym(handle, const_cast(func_name.c_str())); if (real_fn == nullptr) { GELOGE(GE_PLGMGR_PATH_INVALID, "%s is skipped since function %s is not existed!", func_name.c_str(), func_name.c_str()); @@ -137,7 +139,7 @@ Status PluginManager::LoadSo(const string &path, const vector &func_chec } } if (!is_valid) { - GE_LOGE_IF(dlclose(handle), "Failed to dlclose."); + GE_LOGE_IF(mmDlclose(handle), "Failed to dlclose."); continue; } @@ -197,22 +199,29 @@ Status PluginManager::Load(const string &path, const vector &func_check_ so_list_.clear(); ClearHandles_(); - char canonical_path[PATH_MAX] = {0}; - GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(path.length() >= PATH_MAX, GELOGW("File path is too long!"); + char canonical_path[MMPA_MAX_PATH] = {0}; + GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(path.length() >= MMPA_MAX_PATH, GELOGW("File path is too long!"); return FAILED, "File path is too long!"); - if (realpath(path.c_str(), canonical_path) == nullptr) { + if (mmRealPath(path.c_str(), canonical_path, MMPA_MAX_PATH) != EN_OK) { GELOGW("Failed to get realpath of %s", path.c_str()); return SUCCESS; } - DIR *dir = opendir(canonical_path); - if (dir == nullptr) { - GELOGW("Invalid path for load: %s", path.c_str()); - return SUCCESS; + INT32 is_dir = mmIsDir(canonical_path); + // Lib plugin path not exist + if (is_dir != EN_OK) { + GELOGW("Invalid path for load: %s", path.c_str()); + return SUCCESS; } - struct dirent *entry = nullptr; - while ((entry = readdir(dir)) != nullptr) { + mmDirent **entries = nullptr; + auto ret = mmScandir(canonical_path, &entries, nullptr, nullptr); + if (ret < EN_OK) { + GELOGW("scan dir failed. path = %s, ret = %d", canonical_path, ret); + return FAILED; + } + for (int i = 0; i < ret; ++i) { + mmDirent *entry = entries[i]; // read fileName and fileType std::string file_name = entry->d_name; unsigned char file_type = entry->d_type; @@ -250,9 +259,11 @@ Status PluginManager::Load(const string &path, const vector &func_check_ GELOGI("Dlopen so path name: %s. ", file_path_dlopen.c_str()); // load continue when dlopen is failed - auto handle = dlopen(file_path_dlopen.c_str(), RTLD_NOW | RTLD_GLOBAL); + auto handle = mmDlopen(file_path_dlopen.c_str(), MMPA_RTLD_NOW | MMPA_RTLD_GLOBAL); if (handle == nullptr) { - GELOGW("Failed in dlopen %s!", dlerror()); + const char *error = mmDlerror(); + GE_IF_BOOL_EXEC(error == nullptr, error = ""); + GELOGW("Failed in dlopen %s!", error); continue; } @@ -261,7 +272,7 @@ Status PluginManager::Load(const string &path, const vector &func_check_ // load continue when so is invalid bool is_valid = true; for (const auto &func_name : func_check_list) { - auto real_fn = (void (*)())dlsym(handle, func_name.c_str()); + auto real_fn = (void (*)())mmDlsym(handle, const_cast(func_name.c_str())); if (real_fn == nullptr) { GELOGW("The %s is skipped since function %s is not existed!", file_name.c_str(), func_name.c_str()); is_valid = false; @@ -269,7 +280,7 @@ Status PluginManager::Load(const string &path, const vector &func_check_ } } if (!is_valid) { - GE_LOGE_IF(dlclose(handle), "Failed to dlclose."); + GE_LOGE_IF(mmDlclose(handle), "Failed to dlclose."); continue; } @@ -279,7 +290,7 @@ Status PluginManager::Load(const string &path, const vector &func_check_ handles_[string(file_name)] = handle; num_of_loaded_so++; } - closedir(dir); + mmScandirFree(entries, ret); if (num_of_loaded_so == 0) { GELOGW("No loadable shared library found in the path: %s", path.c_str()); return SUCCESS; diff --git a/ge/common/ge/plugin_manager.h b/ge/common/ge/plugin_manager.h index 903367a3..7ef0f81a 100755 --- a/ge/common/ge/plugin_manager.h +++ b/ge/common/ge/plugin_manager.h @@ -17,7 +17,6 @@ #ifndef GE_COMMON_GE_PLUGIN_MANAGER_H_ #define GE_COMMON_GE_PLUGIN_MANAGER_H_ -#include #include #include #include @@ -30,6 +29,7 @@ #include "common/ge_inner_error_codes.h" #include "engine/dnnengine.h" #include "framework/common/debug/ge_log.h" +#include "mmpa/mmpa_api.h" namespace ge { using SoToHandleMap = std::map; @@ -57,7 +57,7 @@ class PluginManager { template Status GetAllFunctions(const string &func_name, map> &funcs) { for (const auto &handle : handles_) { - auto real_fn = (R(*)(Types...))dlsym(handle.second, func_name.c_str()); + auto real_fn = (R(*)(Types...))mmDlsym(handle.second, const_cast(func_name.c_str())); if (real_fn == nullptr) { GELOGW("Failed to get function %s in %s!", func_name.c_str(), handle.first.c_str()); return GE_PLGMGR_FUNC_NOT_EXIST; @@ -72,7 +72,7 @@ class PluginManager { Status InvokeAll(const string &func_name, Types... args) { for (const auto &handle : handles_) { // If the funcName is existed, signature of realFn can be casted to any type - auto real_fn = (void (*)(Types...))dlsym(handle.second, func_name.c_str()); + auto real_fn = (void (*)(Types...))mmDlsym(handle.second, const_cast(func_name.c_str())); if (real_fn == nullptr) { GELOGW("Failed to invoke function %s in %s!", func_name.c_str(), handle.first.c_str()); return GE_PLGMGR_INVOKE_FAILED; @@ -87,7 +87,7 @@ class PluginManager { Status InvokeAll(const string &func_name, T arg) { for (const auto &handle : handles_) { // If the funcName is existed, signature of realFn can be casted to any type - auto real_fn = (void (*)(T))dlsym(handle.second, func_name.c_str()); + auto real_fn = (void (*)(T))mmDlsym(handle.second, const_cast(func_name.c_str())); if (real_fn == nullptr) { GELOGW("Failed to invoke function %s in %s!", func_name.c_str(), handle.first.c_str()); return GE_PLGMGR_INVOKE_FAILED; @@ -112,7 +112,7 @@ class PluginManager { Status InvokeAll(const string &func_name, T1 arg) { for (const auto &handle : handles_) { // If the funcName is existed, signature of realFn can be casted to any type - auto real_fn = (T2(*)(T1))dlsym(handle.second, func_name.c_str()); + auto real_fn = (T2(*)(T1))mmDlsym(handle.second, const_cast(func_name.c_str())); if (real_fn == nullptr) { GELOGW("Failed to invoke function %s in %s!", func_name.c_str(), handle.first.c_str()); return GE_PLGMGR_INVOKE_FAILED; @@ -130,7 +130,7 @@ class PluginManager { Status InvokeAll(const string &func_name) { for (const auto &handle : handles_) { // If the funcName is existed, signature of realFn can be casted to any type - auto real_fn = (T(*)())dlsym(handle.second, func_name.c_str()); + auto real_fn = (T(*)())mmDlsym(handle.second, const_cast(func_name.c_str())); if (real_fn == nullptr) { GELOGW("Failed to invoke function %s in %s!", func_name.c_str(), handle.first.c_str()); return GE_PLGMGR_INVOKE_FAILED; diff --git a/ge/common/ge/tbe_plugin_manager.cc b/ge/common/ge/tbe_plugin_manager.cc index 92da8e14..b91f1204 100755 --- a/ge/common/ge/tbe_plugin_manager.cc +++ b/ge/common/ge/tbe_plugin_manager.cc @@ -16,8 +16,6 @@ #include "common/ge/tbe_plugin_manager.h" -#include -#include #include #include #include @@ -50,9 +48,11 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY TBEPluginManager &TBEPluginMana Status TBEPluginManager::ClearHandles_() { Status ret = SUCCESS; for (const auto &handle : handles_vec_) { - if (dlclose(handle) != 0) { + if (mmDlclose(handle) != 0) { ret = FAILED; - GELOGW("Failed to close handle: %s", dlerror()); + const char *error = mmDlerror(); + GE_IF_BOOL_EXEC(error == nullptr, error = ""); + GELOGW("Failed to close handle: %s", error); } } handles_vec_.clear(); @@ -65,18 +65,18 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status TBEPluginManager::Finali } string TBEPluginManager::GetPath() { - Dl_info dl_info; - if (dladdr(reinterpret_cast(&TBEPluginManager::GetPath), &dl_info) == 0) { + mmDlInfo dl_info; + if (mmDladdr(reinterpret_cast(&TBEPluginManager::GetPath), &dl_info) != EN_OK) { GELOGW("Failed to read so path!"); return string(); } else { string so_path = dl_info.dli_fname; - char path[PATH_MAX] = {0}; - if (so_path.length() >= PATH_MAX) { + char path[MMPA_MAX_PATH] = {0}; + if (so_path.length() >= MMPA_MAX_PATH) { GELOGW("File path is too long!"); return string(); } - if (realpath(so_path.c_str(), path) == nullptr) { + if (mmRealPath(so_path.c_str(), path, MMPA_MAX_PATH) != EN_OK) { GELOGW("Failed to get realpath of %s", so_path.c_str()); return string(); } @@ -108,35 +108,36 @@ void TBEPluginManager::FindParserSo(const string &path, vector &file_lis GELOGW("RealPath is empty."); return; } - struct stat stat_buf; - if ((stat(real_path.c_str(), &stat_buf) != 0) || (!S_ISDIR(stat_buf.st_mode))) { - GELOGW("%s is not a dir.", real_path.c_str()); - return; - } - struct dirent *dent(0); - DIR *dir = opendir(real_path.c_str()); - // Plugin path does not exist - if (dir == nullptr) { - GELOGW("Open directory %s failed.", real_path.c_str()); - return; + INT32 is_dir = mmIsDir(real_path.c_str()); + // Lib plugin path not exist + if (is_dir != EN_OK) { + GELOGW("%s is not a dir.", real_path.c_str()); + return; } - while ((dent = readdir(dir)) != nullptr) { - if (strcmp(dent->d_name, ".") == 0 || strcmp(dent->d_name, "..") == 0) continue; - string name = dent->d_name; - string full_name = real_path + "/" + name; - const string so_suff = ".so"; - const string caffe_parser_so_suff = "lib_caffe_parser.so"; - const string aicpu_so_suff = "_aicpu.so"; - const string aicpu_host_so_suff = "_online.so"; - if (name.size() >= so_suff.size() && name.compare(name.size() - so_suff.size(), so_suff.size(), so_suff) == 0) { - ProcessSoFullName(file_list, caffe_parser_path, full_name, caffe_parser_so_suff, aicpu_so_suff, - aicpu_host_so_suff); - } else { - FindParserSo(full_name, file_list, caffe_parser_path); - } + mmDirent **entries = nullptr; + auto ret = mmScandir(real_path.c_str(), &entries, nullptr, nullptr); + if (ret < EN_OK) { + GELOGW("scan dir failed. path = %s, ret = %d", real_path.c_str(), ret); + return; + } + for (int i = 0; i < ret; ++i) { + mmDirent *dent = entries[i]; + if (strcmp(dent->d_name, ".") == 0 || strcmp(dent->d_name, "..") == 0) continue; + string name = dent->d_name; + string full_name = real_path + "/" + name; + const string so_suff = ".so"; + const string caffe_parser_so_suff = "lib_caffe_parser.so"; + const string aicpu_so_suff = "_aicpu.so"; + const string aicpu_host_so_suff = "_online.so"; + if (name.size() >= so_suff.size() && name.compare(name.size() - so_suff.size(), so_suff.size(), so_suff) == 0) { + ProcessSoFullName(file_list, caffe_parser_path, full_name, caffe_parser_so_suff, aicpu_so_suff, + aicpu_host_so_suff); + } else { + FindParserSo(full_name, file_list, caffe_parser_path); + } } - closedir(dir); + mmScandirFree(entries, ret); } void TBEPluginManager::GetPluginSoFileList(const string &path, vector &file_list, string &caffe_parser_path) { @@ -159,8 +160,9 @@ void TBEPluginManager::GetCustomOpPath(std::string &customop_path) { fmk_type = ge::TypeUtils::FmkTypeToSerialString(type); GELOGI("Framework type is %s.", fmk_type.c_str()); - const char *path_env = std::getenv("ASCEND_OPP_PATH"); - if (path_env != nullptr) { + char path_env[MMPA_MAX_PATH] = { 0x00 }; + INT32 res = mmGetEnv("ASCEND_OPP_PATH", path_env, MMPA_MAX_PATH); + if (res == EN_OK) { std::string path = path_env; customop_path = (path + "/framework/custom" + "/:") + (path + "/framework/built-in/" + fmk_type); GELOGI("Get custom so path from env : %s", path_env); @@ -210,9 +212,11 @@ void TBEPluginManager::LoadPluginSo(const std::map &options) { for (auto elem : file_list) { StringUtils::Trim(elem); - void *handle = dlopen(elem.c_str(), RTLD_NOW | RTLD_GLOBAL | RTLD_NODELETE); + void *handle = mmDlopen(elem.c_str(), MMPA_RTLD_NOW | MMPA_RTLD_GLOBAL | MMPA_RTLD_NODELETE); if (handle == nullptr) { - GELOGW("dlopen failed, plugin name:%s. Message(%s).", elem.c_str(), dlerror()); + const char *error = mmDlerror(); + GE_IF_BOOL_EXEC(error == nullptr, error = ""); + GELOGW("dlopen failed, plugin name:%s. Message(%s).", elem.c_str(), error); } else if (find(handles_vec_.begin(), handles_vec_.end(), handle) == handles_vec_.end()) { // Close dl when the program exist, not close here GELOGI("Plugin load %s success.", elem.c_str()); diff --git a/ge/common/ge/tbe_plugin_manager.h b/ge/common/ge/tbe_plugin_manager.h index 41db8ef9..4bd8c6e3 100755 --- a/ge/common/ge/tbe_plugin_manager.h +++ b/ge/common/ge/tbe_plugin_manager.h @@ -17,7 +17,6 @@ #ifndef GE_COMMON_GE_TBE_PLUGIN_MANAGER_H_ #define GE_COMMON_GE_TBE_PLUGIN_MANAGER_H_ -#include #include #include #include diff --git a/ge/common/ge_common.mk b/ge/common/ge_common.mk index 0e3fc972..3fffd203 100755 --- a/ge/common/ge_common.mk +++ b/ge/common/ge_common.mk @@ -110,11 +110,12 @@ LOCAL_SHARED_LIBRARIES := \ libascend_protobuf \ libc_sec \ libslog \ - libmmpa \ libgraph \ libregister \ liberror_manager \ +LOCAL_STATIC_LIBRARIES += libmmpa + LOCAL_LDFLAGS := -lrt -ldl include $(BUILD_HOST_SHARED_LIBRARY) @@ -152,11 +153,12 @@ LOCAL_SHARED_LIBRARIES := \ libascend_protobuf \ libc_sec \ libslog \ - libmmpa \ libgraph \ libregister \ liberror_manager \ +LOCAL_STATIC_LIBRARIES += libmmpa + ifeq ($(device_os),android) LOCAL_LDFLAGS += -ldl LOCAL_LDLIBS += -L$(PWD)/prebuilts/clang/linux-x86/aarch64/android-ndk-r21/sysroot/usr/lib/aarch64-linux-android/29 -llog diff --git a/ge/common/helper/model_cache_helper.cc b/ge/common/helper/model_cache_helper.cc index e84e0077..0b592e11 100755 --- a/ge/common/helper/model_cache_helper.cc +++ b/ge/common/helper/model_cache_helper.cc @@ -14,8 +14,6 @@ * limitations under the License. */ -#include -#include #include #include #include @@ -448,12 +446,12 @@ Status ModelCacheHelper::SaveJsonToFile(const string &file_name, const Json &jso } const string path = cache_path_ + file_name; const int FILE_AUTHORITY = 0600; - int fd = open(path.c_str(), O_WRONLY | O_CREAT | O_TRUNC, FILE_AUTHORITY); + int fd = mmOpen2(path.c_str(), M_WRONLY | M_CREAT | O_TRUNC, FILE_AUTHORITY); if (fd < 0) { GELOGW("Fail to open the file: %s.", path.c_str()); return INTERNAL_ERROR; } - if (close(fd) != 0) { + if (mmClose(fd) != 0) { GELOGW("Fail to close the file: %s.", path.c_str()); return INTERNAL_ERROR; } diff --git a/ge/common/helper/model_helper.cc b/ge/common/helper/model_helper.cc index ab5075e1..fbc6b782 100644 --- a/ge/common/helper/model_helper.cc +++ b/ge/common/helper/model_helper.cc @@ -98,7 +98,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::SaveToOmMod ge::Buffer model_buffer; (void)model_tmp->Save(model_buffer); - GELOGI("MODEL_DEF size is %zu", model_buffer.GetSize()); + GELOGD("MODEL_DEF size is %zu", model_buffer.GetSize()); if (model_buffer.GetSize() > 0) { if (SaveModelPartition(om_file_save_helper, ModelPartitionType::MODEL_DEF, model_buffer.GetData(), model_buffer.GetSize()) != SUCCESS) { @@ -107,7 +107,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::SaveToOmMod } } auto ge_model_weight = ge_model->GetWeight(); - GELOGI("WEIGHTS_DATA size is %zu, %p", ge_model_weight.GetSize(), ge_model_weight.GetData()); + GELOGD("WEIGHTS_DATA size is %zu, %p", ge_model_weight.GetSize(), ge_model_weight.GetData()); // weight is not necessary if (ge_model_weight.GetSize() > 0) { GE_CHK_STATUS_RET(SaveModelPartition(om_file_save_helper, @@ -117,7 +117,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::SaveToOmMod } TBEKernelStore tbe_kernel_store = ge_model->GetTBEKernelStore(); - GELOGI("TBE_KERNELS size is %zu", tbe_kernel_store.DataSize()); + GELOGD("TBE_KERNELS size is %zu", tbe_kernel_store.DataSize()); if (tbe_kernel_store.DataSize() > 0) { GE_CHK_STATUS_RET(SaveModelPartition(om_file_save_helper, ModelPartitionType::TBE_KERNELS, @@ -129,7 +129,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::SaveToOmMod (void)tbe_kernel_store.Load(tbe_kernel_store.Data(), tbe_kernel_store.DataSize()); CustAICPUKernelStore cust_aicpu_kernel_store = ge_model->GetCustAICPUKernelStore(); - GELOGI("cust aicpu kernels size is %zu", cust_aicpu_kernel_store.DataSize()); + GELOGD("cust aicpu kernels size is %zu", cust_aicpu_kernel_store.DataSize()); if (cust_aicpu_kernel_store.DataSize() > 0) { GE_CHK_STATUS_RET(SaveModelPartition(om_file_save_helper, ModelPartitionType::CUST_AICPU_KERNELS, @@ -155,8 +155,8 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::SaveToOmMod } (void)model_task_def->SerializePartialToArray(task_buffer.GetData(), static_cast(partition_task_size)); - GELOGI("TASK_INFO op_size:%d, stream_num:%u", model_task_def->op().size(), model_task_def->stream_num()); - GELOGI("TASK_INFO size is %zu", partition_task_size); + GELOGD("TASK_INFO op_size:%d, stream_num:%u", model_task_def->op().size(), model_task_def->stream_num()); + GELOGD("TASK_INFO size is %zu", partition_task_size); if (SaveModelPartition(om_file_save_helper, ModelPartitionType::TASK_INFO, task_buffer.GetData(), partition_task_size) != SUCCESS) { @@ -168,7 +168,6 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::SaveToOmMod model_header.platform_type = ge_model->GetPlatformType(); model_header.om_ir_version = ge_model->GetVersion(); std::string platform_version = ge_model->GetPlatformVersion(); - GELOGI("Platform version save: %s", platform_version.c_str()); errno_t err; err = memcpy_s(model_header.platform_version, PLATFORM_VERSION_LEN, platform_version.c_str(), @@ -178,7 +177,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::SaveToOmMod return MEMALLOC_FAILED; } string version = reinterpret_cast(model_header.platform_version); - GELOGI("Platform version save: %s", version.c_str()); + GELOGD("Platform version save: %s", version.c_str()); size_t name_size = ge_model->GetName().size(); name_size = name_size > (MODEL_NAME_LENGTH - 1) ? (MODEL_NAME_LENGTH - 1) : name_size; @@ -188,7 +187,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::SaveToOmMod return MEMALLOC_FAILED; } string model_name = reinterpret_cast(model_header.name); - GELOGI("Model name save:%s", model_name.c_str()); + GELOGD("Model name save:%s", model_name.c_str()); Status ret = om_file_save_helper->SaveModel(save_param, output_file.c_str(), model, is_offline_); if (ret != SUCCESS) { @@ -346,7 +345,7 @@ Status ModelHelper::LoadModelData(OmFileLoadHelper &om_load_helper) { ModelPartition partition_model_def; // no need to check value, DATA->NetOutput om_load_helper.GetModelPartition(ModelPartitionType::MODEL_DEF, partition_model_def); - GELOGI("Model_def partition addr:%p,size:%u", partition_model_def.data, partition_model_def.size); + GELOGD("Model_def partition addr:%p,size:%u", partition_model_def.data, partition_model_def.size); ge::Model model; if (ge::Model::Load(partition_model_def.data, partition_model_def.size, model) != SUCCESS) { @@ -376,7 +375,7 @@ Status ModelHelper::LoadWeights(OmFileLoadHelper &om_load_helper) { ge::Buffer weight = ge::Buffer::CopyFrom(partition.data, partition.size); model_->SetWeight(weight); - GELOGI("GetWeight size:%u", partition.size); + GELOGD("GetWeight size:%u", partition.size); return SUCCESS; } @@ -393,7 +392,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::LoadTask(Om GELOGE(INTERNAL_ERROR, "ReadProtoFromArray failed."); return INTERNAL_ERROR; } - GELOGI("TASK_INFO op_size:%zu, stream_num:%u", task->op().size(), task->stream_num()); + GELOGD("TASK_INFO op_size:%d, stream_num:%u", task->op().size(), task->stream_num()); } model_->SetModelTaskDef(task); return SUCCESS; @@ -404,9 +403,9 @@ Status ModelHelper::LoadTBEKernelStore(OmFileLoadHelper &om_load_helper) { ModelPartition partition_kernel_def; TBEKernelStore kernel_store; if (om_load_helper.GetModelPartition(ModelPartitionType::TBE_KERNELS, partition_kernel_def) == SUCCESS) { - GELOGI("Kernels partition size:%u", partition_kernel_def.size); + GELOGD("Kernels partition size:%u", partition_kernel_def.size); if (kernel_store.Load(partition_kernel_def.data, partition_kernel_def.size)) { - GELOGI("Load tbe kernels success"); + GELOGD("Load tbe kernels success"); } else { GELOGW("Load tbe kernels failed"); } @@ -420,11 +419,9 @@ Status ModelHelper::LoadCustAICPUKernelStore(OmFileLoadHelper &om_load_helper) { ModelPartition partition_kernel_def; CustAICPUKernelStore kernel_store; if (om_load_helper.GetModelPartition(ModelPartitionType::CUST_AICPU_KERNELS, partition_kernel_def) == SUCCESS) { - GELOGI("Kernels partition size:%u", partition_kernel_def.size); + GELOGD("Kernels partition size:%u", partition_kernel_def.size); if (kernel_store.Load(partition_kernel_def.data, partition_kernel_def.size)) { GELOGI("Load cust aicpu kernels success"); - } else { - GELOGW("Load cust aicpu kernels failed"); } } model_->SetCustAICPUKernelStore(kernel_store); diff --git a/ge/common/helper/om_file_helper.cc b/ge/common/helper/om_file_helper.cc index 75140e9c..0b076aaf 100644 --- a/ge/common/helper/om_file_helper.cc +++ b/ge/common/helper/om_file_helper.cc @@ -123,7 +123,7 @@ Status OmFileLoadHelper::LoadModelPartitionTable(uint8_t *model_data, const uint return ACL_ERROR_GE_EXEC_MODEL_PARTITION_NUM_INVALID; } size_t mem_offset = SIZE_OF_MODEL_PARTITION_TABLE(*partition_table); - GELOGI("ModelPartitionTable num :%u, ModelFileHeader length :%zu, ModelPartitionTable length :%zu", + GELOGD("ModelPartitionTable num :%u, ModelFileHeader length :%zu, ModelPartitionTable length :%zu", partition_table->num, sizeof(ModelFileHeader), mem_offset); if (model_data_size <= mem_offset) { GELOGE(ACL_ERROR_GE_EXEC_MODEL_DATA_SIZE_INVALID, "invalid model data, partition_table->num:%u, model data size %u", @@ -143,7 +143,7 @@ Status OmFileLoadHelper::LoadModelPartitionTable(uint8_t *model_data, const uint return ACL_ERROR_GE_EXEC_MODEL_DATA_SIZE_INVALID; } mem_offset += partition.size; - GELOGI("Partition, type:%d, size:%u", static_cast(partition.type), partition.size); + GELOGD("Partition, type:%d, size:%u", static_cast(partition.type), partition.size); } return SUCCESS; } @@ -167,7 +167,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY ModelPartitionTable *OmFileSave ModelPartition partition = context_.partition_datas_[i]; partition_table->partition[i] = {partition.type, mem_offset, partition.size}; mem_offset += partition.size; - GELOGI("Partition, type:%d, size:%u", static_cast(partition.type), partition.size); + GELOGD("Partition, type:%d, size:%u", static_cast(partition.type), partition.size); } return partition_table; } @@ -191,7 +191,7 @@ Status OmFileSaveHelper::SaveModel(const SaveParam &save_param, const char *outp (void)save_param.pri_key_file; Status ret = SaveModelToFile(output_file, model, is_offline); if (ret == SUCCESS) { - GELOGI("Generate model with encrypt."); + GELOGD("Generate model with encrypt."); } return ret; } @@ -213,7 +213,7 @@ Status OmFileSaveHelper::SaveModelToFile(const char *output_file, ModelBufferDat FMK_UINT32_ADDCHECK(size_of_table, model_data_len) model_header_.length = size_of_table + model_data_len; - GELOGI("Sizeof(ModelFileHeader):%zu,sizeof(ModelPartitionTable):%u, model_data_len:%u, model_total_len:%zu", + GELOGD("Sizeof(ModelFileHeader):%zu,sizeof(ModelPartitionTable):%u, model_data_len:%u, model_total_len:%zu", sizeof(ModelFileHeader), size_of_table, model_data_len, model_header_.length + sizeof(ModelFileHeader)); std::vector partition_datas = context_.partition_datas_; @@ -224,7 +224,7 @@ Status OmFileSaveHelper::SaveModelToFile(const char *output_file, ModelBufferDat ret = FileSaver::SaveToBuffWithFileHeader(model_header_, *partition_table, partition_datas, model); } if (ret == SUCCESS) { - GELOGI("Save model success without encrypt."); + GELOGD("Save model success without encrypt."); } return ret; #else diff --git a/ge/common/kernel_store.cc b/ge/common/kernel_store.cc index e339b30c..0fad096a 100755 --- a/ge/common/kernel_store.cc +++ b/ge/common/kernel_store.cc @@ -51,7 +51,7 @@ bool KernelStore::Build() { kernel_head.name_len = static_cast(kernel->GetName().length()); kernel_head.bin_len = static_cast(kernel->GetBinDataSize()); - GELOGI("get kernel bin name %s, addr %p, size %u", + GELOGD("get kernel bin name %s, addr %p, size %u", kernel->GetName().c_str(), kernel->GetBinData(), kernel->GetBinDataSize()); mem_ret = memcpy_s(next_buffer, remain_len, &kernel_head, sizeof(kernel_head)); GE_CHK_BOOL_EXEC_NOLOG(mem_ret == EOK, return false); @@ -95,7 +95,7 @@ bool KernelStore::Load(const uint8_t *data, const size_t &len) { std::string name(next_buffer, kernel_head->name_len); next_buffer += kernel_head->name_len; - GELOGI("Load kernel from om:%s,%u,%u", name.c_str(), kernel_head->name_len, kernel_head->bin_len); + GELOGD("Load kernel from om:%s,%u,%u", name.c_str(), kernel_head->name_len, kernel_head->bin_len); std::vector kernel_bin(next_buffer, next_buffer + kernel_head->bin_len); KernelBinPtr teb_kernel_ptr = ge::MakeShared(name, std::move(kernel_bin)); if (teb_kernel_ptr != nullptr) { diff --git a/ge/common/model_parser/base.cc b/ge/common/model_parser/base.cc index 25cc6645..5d050a2b 100644 --- a/ge/common/model_parser/base.cc +++ b/ge/common/model_parser/base.cc @@ -17,7 +17,6 @@ #include "common/model_parser/base.h" #include "common/helper/model_helper.h" #include -#include #include #include #include @@ -107,7 +106,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelParserBase::ParseMo model_data = data; model_len = file_header->length; - GELOGI("Model_len is %u, model_file_head_len is %zu.", model_len, sizeof(ModelFileHeader)); + GELOGD("Model_len is %u, model_file_head_len is %zu.", model_len, sizeof(ModelFileHeader)); } else { GELOGE(ACL_ERROR_GE_EXEC_MODEL_NOT_SUPPORT_ENCRYPTION, "Invalid model. ModelEncryptType not supported."); res = ACL_ERROR_GE_EXEC_MODEL_NOT_SUPPORT_ENCRYPTION; diff --git a/ge/common/model_saver.cc b/ge/common/model_saver.cc index fb1cd0a7..b8442948 100755 --- a/ge/common/model_saver.cc +++ b/ge/common/model_saver.cc @@ -16,9 +16,7 @@ #include "common/model_saver.h" -#include #include -#include #include #include #include @@ -51,14 +49,14 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelSaver::SaveJsonToFi return FAILED; } - char real_path[PATH_MAX] = {0}; - GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(strlen(file_path) >= PATH_MAX, return FAILED, "file path is too long!"); - GE_IF_BOOL_EXEC(realpath(file_path, real_path) == nullptr, + char real_path[MMPA_MAX_PATH] = {0}; + GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(strlen(file_path) >= MMPA_MAX_PATH, return FAILED, "file path is too long!"); + GE_IF_BOOL_EXEC(mmRealPath(file_path, real_path, MMPA_MAX_PATH) != EN_OK, GELOGI("File %s does not exit, it will be created.", file_path)); // Open file - mode_t mode = S_IRUSR | S_IWUSR; - int32_t fd = mmOpen2(real_path, O_RDWR | O_CREAT | O_TRUNC, mode); + mmMode_t mode = M_IRUSR | M_IWUSR; + int32_t fd = mmOpen2(real_path, M_RDWR | M_CREAT | O_TRUNC, mode); if (fd == EN_ERROR || fd == EN_INVALID_PARAM) { ErrorManager::GetInstance().ATCReportErrMessage("E19001", {"file", "errmsg"}, {file_path, strerror(errno)}); GELOGE(FAILED, "Open file[%s] failed. %s", file_path, strerror(errno)); @@ -72,7 +70,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelSaver::SaveJsonToFi ErrorManager::GetInstance().ATCReportErrMessage( "E19004", {"file", "errmsg"}, {file_path, strerror(errno)}); // Need to both print the error info of mmWrite and mmClose, so return ret after mmClose - GELOGE(FAILED, "Write to file failed. errno = %d, %s", mmpa_ret, strerror(errno)); + GELOGE(FAILED, "Write to file failed. errno = %ld, %s", mmpa_ret, strerror(errno)); ret = FAILED; } // Close file diff --git a/ge/common/profiling/profiling_manager.cc b/ge/common/profiling/profiling_manager.cc index 6e01ee87..2f0f061f 100644 --- a/ge/common/profiling/profiling_manager.cc +++ b/ge/common/profiling/profiling_manager.cc @@ -214,8 +214,12 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY ge::Status ProfilingManager::Pa FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY ge::Status ProfilingManager::InitFromOptions(const Options &options) { #ifdef DAVINCI_SUPPORT_PROFILING // enable profiling support two ways: env and front end - const char *profiling_mode = std::getenv("PROFILING_MODE"); - const char *prof_options = std::getenv("PROFILING_OPTIONS"); + char profiling_mode_temp[MMPA_MAX_PATH] = { 0x00 }; + char prof_options_temp[MMPA_MAX_PATH] = { 0x00 }; + (void)mmGetEnv("PROFILING_MODE", profiling_mode_temp, MMPA_MAX_PATH); + (void)mmGetEnv("PROFILING_OPTIONS", prof_options_temp, MMPA_MAX_PATH ); + const char *profiling_mode = profiling_mode_temp; + const char *prof_options = prof_options_temp; if ((profiling_mode == nullptr) || (strcmp("true", profiling_mode) != 0) || (prof_options == nullptr)) { is_load_profiling_ = false; is_execute_profiling_ = false; @@ -554,7 +558,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY void ProfilingManager::ReportPr GELOGE(rt_ret, "runtime get logic_device_id failed, current logic_device_id:%d", logic_device_id); return; } - GELOGI("current logic_device_id:%d", logic_device_id); + GELOGD("current logic_device_id:%d", logic_device_id); if (check_device) { auto ret = std::find(device_id_.begin(), device_id_.end(), logic_device_id); if (ret == device_id_.end()) { @@ -562,11 +566,11 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY void ProfilingManager::ReportPr return; } } - GELOGI("start ProfilingTaskDescInfo."); + GELOGD("start ProfilingTaskDescInfo."); ProfilingTaskDescInfo(model_id, task_desc_info, logic_device_id); - GELOGI("start ProfilingGraphDescInfo."); + GELOGD("start ProfilingGraphDescInfo."); ProfilingGraphDescInfo(model_id, compute_graph_desc_info, logic_device_id); - GELOGI("Report profiling data for GE end."); + GELOGD("Report profiling data for GE end."); #endif } @@ -855,7 +859,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ProfilingManager::ProfSt for (int32_t i = 0; i < device_num; i++) { device_id_ptr[i] = static_cast(device_list[i]); } - GELOGI("Runtime config param: 0x%llx, device num: %d.", module, device_num); + GELOGD("Runtime config param: 0x%llx, device num: %d.", module, device_num); rtError_t rt_ret = rtProfilerStart(module, device_num, device_id_ptr.get()); if (rt_ret != RT_ERROR_NONE) { @@ -874,7 +878,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ProfilingManager::ProfSt GELOGW("Prof start: load model module is invalid."); } UpdateDeviceIdModuleMap(kProfStart, module, device_list); - GELOGI("Prof start profiling success."); + GELOGD("Prof start profiling success."); #endif return SUCCESS; } @@ -897,7 +901,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ProfilingManager::ProfSt for (int32_t i = 0; i < device_num; i++) { device_id_ptr[i] = static_cast(device_list[i]); } - GELOGI("Prof stop: runtime config param: 0x%llx, device num: %d", module, device_num); + GELOGD("Prof stop: runtime config param: 0x%llx, device num: %d", module, device_num); rtError_t rt_ret = rtProfilerStop(module, device_num, device_id_ptr.get()); if (rt_ret != RT_ERROR_NONE) { GELOGE(FAILED, "Prof stop: runtime profiler config proc failed."); @@ -917,7 +921,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ProfilingManager::ProfSt GELOGW("Prof stop: load model module is invalid."); } UpdateDeviceIdModuleMap(kProfStop, module, device_list); - GELOGI("Prof stop profiling success."); + GELOGD("Prof stop profiling success."); #endif return SUCCESS; } @@ -959,14 +963,14 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY bool ProfilingManager::Profilin if (rt_ret != RT_ERROR_NONE) { GELOGE(rt_ret, "Runtime get logic_device_id failed, current logic_device_id:%d", logic_device_id); } - GELOGI("Current logic_device_id:%d", logic_device_id); + GELOGD("Current logic_device_id:%d", logic_device_id); bool execute_model_prof_on = false; auto iter = std::find(device_id_.begin(), device_id_.end(), logic_device_id); if (iter != device_id_.end()) { execute_model_prof_on = true; } - GELOGI("Flag is_execute_profiling: %d, execute_model_prof_on: %d", is_execute_profiling_, execute_model_prof_on); + GELOGD("Flag is_execute_profiling: %d, execute_model_prof_on: %d", is_execute_profiling_, execute_model_prof_on); return is_execute_profiling_ || execute_model_prof_on; } diff --git a/ge/common/properties_manager.h b/ge/common/properties_manager.h index 634113a8..b4c5aad1 100644 --- a/ge/common/properties_manager.h +++ b/ge/common/properties_manager.h @@ -25,13 +25,14 @@ #include "common/dump/dump_properties.h" #include "graph/op_desc.h" +#include "common/ge_compiler_options.h" namespace ge { // Configuration property management -static const char *SYSMODE __attribute__((unused)) = "FMK_SYSMODE"; -static const char *USE_FUSION __attribute__((unused)) = "FMK_USE_FUSION"; -static const char *TIMESTAT_ENABLE __attribute__((unused)) = "DAVINCI_TIMESTAT_ENABLE"; -static const char *ANNDROID_DEBUG __attribute__((unused)) = "ANNDROID_DEBUG"; +static const char *SYSMODE GE_ATTRIBUTE_UNUSED = "FMK_SYSMODE"; +static const char *USE_FUSION GE_ATTRIBUTE_UNUSED = "FMK_USE_FUSION"; +static const char *TIMESTAT_ENABLE GE_ATTRIBUTE_UNUSED = "DAVINCI_TIMESTAT_ENABLE"; +static const char *ANNDROID_DEBUG GE_ATTRIBUTE_UNUSED = "ANNDROID_DEBUG"; class PropertiesManager { public: diff --git a/ge/common/util.cc b/ge/common/util.cc index 2ddb4b2c..3c5954db 100644 --- a/ge/common/util.cc +++ b/ge/common/util.cc @@ -16,11 +16,12 @@ #include "framework/common/util.h" -#include #include - +#ifdef __GNUC__ #include -#include +#else +#include +#endif #include #include #include @@ -208,29 +209,30 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY bool ReadBytesFromBinaryFile(co FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY int CreateDirectory(const std::string &directory_path) { GE_CHK_BOOL_EXEC(!directory_path.empty(), return -1, "directory path is empty."); auto dir_path_len = directory_path.length(); - if (dir_path_len >= PATH_MAX) { - ErrorManager::GetInstance().ATCReportErrMessage("E19002", {"filepath", "size"}, - {directory_path, std::to_string(PATH_MAX)}); - GELOGW("Path[%s] len is too long, it must be less than %d", directory_path.c_str(), PATH_MAX); + if (dir_path_len >= MMPA_MAX_PATH) { + ErrorManager::GetInstance().ATCReportErrMessage( + "E19002", {"filepath", "size"}, {directory_path, std::to_string(MMPA_MAX_PATH)}); + GELOGW("Path[%s] len is too long, it must be less than %d", directory_path.c_str(), MMPA_MAX_PATH); return -1; } - char tmp_dir_path[PATH_MAX] = {0}; + char tmp_dir_path[MMPA_MAX_PATH] = {0}; for (size_t i = 0; i < dir_path_len; i++) { tmp_dir_path[i] = directory_path[i]; if ((tmp_dir_path[i] == '\\') || (tmp_dir_path[i] == '/')) { - if (access(tmp_dir_path, F_OK) != 0) { - int32_t ret = mmMkdir(tmp_dir_path, S_IRUSR | S_IWUSR | S_IXUSR); // 700 + if (mmAccess2(tmp_dir_path, M_F_OK) != EN_OK) { + int32_t ret = mmMkdir(tmp_dir_path, M_IRUSR | M_IWUSR | M_IXUSR); // 700 if (ret != 0) { if (errno != EEXIST) { ErrorManager::GetInstance().ATCReportErrMessage("E19006", {"path"}, {directory_path}); - GELOGW("Can not create directory %s. Make sure the directory exists and writable.", directory_path.c_str()); + GELOGW("Can not create directory %s. Make sure the directory exists and writable.", + directory_path.c_str()); return ret; } } } } } - int32_t ret = mmMkdir(const_cast(directory_path.c_str()), S_IRUSR | S_IWUSR | S_IXUSR); // 700 + int32_t ret = mmMkdir(const_cast(directory_path.c_str()), M_IRUSR | M_IWUSR | M_IXUSR); // 700 if (ret != 0) { if (errno != EEXIST) { ErrorManager::GetInstance().ATCReportErrMessage("E19006", {"path"}, {directory_path}); @@ -305,9 +307,9 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY bool ReadProtoFromMem(const cha } FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY uint64_t GetCurrentTimestamp() { - struct timeval tv {}; - int ret = gettimeofday(&tv, nullptr); - GE_LOGE_IF(ret != 0, "Func gettimeofday may failed: ret=%d", ret); + mmTimeval tv {}; + int ret = mmGetTimeOfDay(&tv, nullptr); + GE_LOGE_IF(ret != EN_OK, "Func gettimeofday may failed: ret=%d", ret); auto total_use_time = tv.tv_usec + tv.tv_sec * 1000000; // 1000000: seconds to microseconds return static_cast(total_use_time); } @@ -347,16 +349,15 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY bool CheckInt64MulOverflow(int6 FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY std::string RealPath(const char *path) { GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(path == nullptr, return "", "path pointer is NULL."); - GE_CHK_BOOL_TRUE_EXEC_WITH_LOG( - strlen(path) >= PATH_MAX, - ErrorManager::GetInstance().ATCReportErrMessage("E19002", {"filepath", "size"}, {path, std::to_string(PATH_MAX)}); - return "", "Path[%s] len is too long, it must be less than %d", path, PATH_MAX); + GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(strlen(path) >= MMPA_MAX_PATH, + ErrorManager::GetInstance().ATCReportErrMessage("E19002", {"filepath", "size"}, {path, std::to_string(MMPA_MAX_PATH)}); + return "", "Path[%s] len is too long, it must be less than %d", path, MMPA_MAX_PATH); // Nullptr is returned when the path does not exist or there is no permission // Return absolute path when path is accessible std::string res; - char resolved_path[PATH_MAX] = {0}; - if (realpath(path, resolved_path) != nullptr) { + char resolved_path[MMPA_MAX_PATH] = {0}; + if (mmRealPath(path, resolved_path, MMPA_MAX_PATH) == EN_OK) { res = resolved_path; } @@ -383,7 +384,11 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY bool CheckInputPathValid(const // A regular matching expression to verify the validity of the input file path // Path section: Support upper and lower case letters, numbers dots(.) chinese and underscores // File name section: Support upper and lower case letters, numbers, underscores chinese and dots(.) - std::string mode = "^[\u4e00-\u9fa5A-Za-z0-9./_-]+$"; +#ifdef __GNUC__ + std::string mode = "^[\u4e00-\u9fa5A-Za-z0-9./_-]+$"; +#else + std::string mode = "^[a-zA-Z]:([\\\\/][^\\s\\\\/:*?<>\"|][^\\\\/:*?<>\"|]*)*([/\\\\][^\\s\\\\/:*?<>\"|])?$"; +#endif GE_CHK_BOOL_TRUE_EXEC_WITH_LOG( !ValidateStr(real_path, mode), @@ -392,7 +397,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY bool CheckInputPathValid(const return false, "Invalid value for %s[%s], %s.", atc_param.c_str(), real_path.c_str(), kPathValidReason); // The absolute path points to a file that is not readable - if (access(real_path.c_str(), R_OK) != 0) { + if (mmAccess2(real_path.c_str(), M_R_OK) != EN_OK) { ErrorManager::GetInstance().ATCReportErrMessage("E19003", {"file", "errmsg"}, {file_path.c_str(), strerror(errno)}); GELOGW("Read file[%s] failed, errmsg[%s]", file_path.c_str(), strerror(errno)); return false; @@ -410,15 +415,19 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY bool CheckOutputPathValid(const return false; } - GE_CHK_BOOL_TRUE_EXEC_WITH_LOG( - strlen(file_path.c_str()) >= PATH_MAX, ErrorManager::GetInstance().ATCReportErrMessage( - "E19002", {"filepath", "size"}, {file_path, std::to_string(PATH_MAX)}); - return "", "Path[%s] len is too long, it must be less than %d", file_path.c_str(), PATH_MAX); + GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(strlen(file_path.c_str()) >= MMPA_MAX_PATH, + ErrorManager::GetInstance().ATCReportErrMessage( + "E19002", {"filepath", "size"}, {file_path, std::to_string(MMPA_MAX_PATH)}); + return "", "Path[%s] len is too long, it must be less than %d", file_path.c_str(), MMPA_MAX_PATH); // A regular matching expression to verify the validity of the input file path // Path section: Support upper and lower case letters, numbers dots(.) chinese and underscores // File name section: Support upper and lower case letters, numbers, underscores chinese and dots(.) - std::string mode = "^[\u4e00-\u9fa5A-Za-z0-9./_-]+$"; +#ifdef __GNUC__ + std::string mode = "^[\u4e00-\u9fa5A-Za-z0-9./_-]+$"; +#else + std::string mode = "^[a-zA-Z]:([\\\\/][^\\s\\\\/:*?<>\"|][^\\\\/:*?<>\"|]*)*([/\\\\][^\\s\\\\/:*?<>\"|])?$"; +#endif GE_CHK_BOOL_TRUE_EXEC_WITH_LOG( !ValidateStr(file_path, mode), @@ -430,7 +439,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY bool CheckOutputPathValid(const // Can get absolute path (file exists) if (!real_path.empty()) { // File is not readable or writable - if (access(real_path.c_str(), W_OK | F_OK) != 0) { + if (mmAccess2(real_path.c_str(), M_W_OK | M_F_OK) != EN_OK) { ErrorManager::GetInstance().ATCReportErrMessage("E19004", {"file", "errmsg"}, {real_path, strerror(errno)}); GELOGW("Write file[%s] failed, errmsg[%s]", real_path.c_str(), strerror(errno)); return false; @@ -461,6 +470,7 @@ FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY bool CheckOutputPathValid(const } FMK_FUNC_HOST_VISIBILITY bool ValidateStr(const std::string &str, const std::string &mode) { +#ifdef __GNUC__ char ebuff[kMaxBuffSize]; regex_t reg; int cflags = REG_EXTENDED | REG_NOSUB; @@ -482,6 +492,23 @@ FMK_FUNC_HOST_VISIBILITY bool ValidateStr(const std::string &str, const std::str regfree(®); return true; +#else + std::wstring wstr(str.begin(), str.end()); + std::wstring wmode(mode.begin(), mode.end()); + std::wsmatch match; + bool res = false; + + try { + std::wregex reg(wmode, std::regex::icase); + // Matching string part + res = regex_match(wstr, match, reg); + res = regex_search(str, std::regex("[`!@#$%^&*()|{}';',<>?]")); + } catch (std::exception &ex) { + GELOGW("The directory %s is invalid, error: %s.", str.c_str(), ex.what()); + return false; + } + return !(res) && (str.size() == match.str().size()); +#endif } FMK_FUNC_HOST_VISIBILITY bool IsValidFile(const char *file_path) { diff --git a/ge/engine_manager/dnnengine_manager.cc b/ge/engine_manager/dnnengine_manager.cc index f8d58208..b23993b6 100644 --- a/ge/engine_manager/dnnengine_manager.cc +++ b/ge/engine_manager/dnnengine_manager.cc @@ -433,7 +433,7 @@ Status DNNEngineManager::ReadJsonFile(const std::string &file_path, JsonHandle h return FAILED; } const char *file = file_path.data(); - if ((access(file, F_OK)) == -1) { + if ((mmAccess2(file, M_F_OK)) != EN_OK) { if (engines_map_.size() != 0) { GELOGE(FAILED, "The json file %s is not exist, %s", file_path.c_str(), strerror(errno)); return FAILED; diff --git a/ge/executor/ge_executor.cc b/ge/executor/ge_executor.cc index 86b4de54..7ceb9b05 100755 --- a/ge/executor/ge_executor.cc +++ b/ge/executor/ge_executor.cc @@ -588,7 +588,7 @@ Status GeExecutor::LoadModel(uint32_t &model_id, const ModelData &model_data, } Status GeExecutor::UnloadModel(uint32_t model_id) { - GELOGI("unload model %u begin.", model_id); + GELOGD("unload model %u begin.", model_id); if (!isInit_) { GELOGE(ACL_ERROR_GE_EXEC_NOT_INIT, "GeExecutor has not been initialized!"); return ACL_ERROR_GE_EXEC_NOT_INIT; @@ -630,7 +630,6 @@ Status GeExecutor::RunModel(const ge::RunModelData &input_data, ge::RunModelData // Get input and output descriptor Status GeExecutor::GetModelDescInfo(uint32_t model_id, std::vector &input_desc, std::vector &output_desc, bool new_model_desc) { - GELOGI("get model desc info begin."); if (!isInit_) { GELOGE(ACL_ERROR_GE_EXEC_NOT_INIT, "GeExecutor has not been initialized!"); return ACL_ERROR_GE_EXEC_NOT_INIT; @@ -665,7 +664,6 @@ Status GeExecutor::GetModelDescInfo(uint32_t model_id, std::vector> &batch_info, int32_t &dynamic_type) { - GELOGI("Begin to get dynamic batch info."); if (!isInit_) { GELOGE(ACL_ERROR_GE_EXEC_NOT_INIT, "GeExecutor has not been initialized!"); return ACL_ERROR_GE_EXEC_NOT_INIT; @@ -690,8 +687,6 @@ Status GeExecutor::GetDynamicBatchInfo(uint32_t model_id, std::vector &user_designate_shape_order) { - GELOGI("Begin to get user designate shape info."); if (!isInit_) { GELOGE(ACL_ERROR_GE_EXEC_NOT_INIT, "GeExecutor has not been initialized!"); return ACL_ERROR_GE_EXEC_NOT_INIT; @@ -739,7 +733,6 @@ Status GeExecutor::GetUserDesignateShapeOrder(uint32_t model_id, vector return ret; } - GELOGI("Get user designate shape order succ."); return SUCCESS; } @@ -782,7 +775,6 @@ Status GeExecutor::GetAippType(uint32_t model_id, uint32_t index, InputAippType } Status GeExecutor::GetModelAttr(uint32_t model_id, std::vector &dynamic_output_shape_info) { - GELOGI("Begin to get dynamic batch output shape info"); if (!isInit_) { GELOGE(ACL_ERROR_GE_EXEC_NOT_INIT, "not inited yet!"); return ACL_ERROR_GE_EXEC_NOT_INIT; @@ -792,8 +784,6 @@ Status GeExecutor::GetModelAttr(uint32_t model_id, std::vector &dyn GELOGE(ret, "Get dynamic batch output shape info failed."); return ret; } - - GELOGI("Get dynamic batch output shape info succ."); return SUCCESS; } @@ -835,8 +825,6 @@ Status GeExecutor::GetModelDescInfoForZeroCopy(uint32_t model_id, std::vectorGetName().c_str(), ret); + GELOGW("Failed to compute host cpu op. node = %s", op_desc->GetName().c_str()); return FAILED; } op.BreakConnect(); diff --git a/ge/ge_local_engine/module.mk b/ge/ge_local_engine/module.mk index a3c571d9..c7a18721 100755 --- a/ge/ge_local_engine/module.mk +++ b/ge/ge_local_engine/module.mk @@ -88,6 +88,25 @@ LOCAL_C_INCLUDES := $(local_lib_inc_path) include ${BUILD_HOST_SHARED_LIBRARY} +#compiler for device libge_local_opskernel_builder.so +include $(CLEAR_VARS) +LOCAL_MODULE := libge_local_opskernel_builder +LOCAL_CFLAGS += -Werror +LOCAL_CFLAGS += -std=c++11 -Dgoogle=ascend_private +LOCAL_LDFLAGS := + +LOCAL_STATIC_LIBRARIES := +LOCAL_SHARED_LIBRARIES := libascend_protobuf \ + libc_sec \ + libslog \ + libregister \ + libgraph + +LOCAL_SRC_FILES := $(ops_kernel_builder_src_files) + +LOCAL_C_INCLUDES := $(local_lib_inc_path) + +include ${BUILD_SHARED_LIBRARY} #compiler for libge_local_opskernel_builder.so in atc include $(CLEAR_VARS) diff --git a/ge/ge_local_engine/ops_kernel_store/ge_local_ops_kernel_builder.cc b/ge/ge_local_engine/ops_kernel_store/ge_local_ops_kernel_builder.cc index 9496d0fc..602dc35f 100644 --- a/ge/ge_local_engine/ops_kernel_store/ge_local_ops_kernel_builder.cc +++ b/ge/ge_local_engine/ops_kernel_store/ge_local_ops_kernel_builder.cc @@ -111,7 +111,7 @@ Status GeLocalOpsKernelBuilder::CalcOpRunningParam(Node &ge_node) { TypeUtils::DataTypeToSerialString(data_type).c_str(), output_mem_size); return FAILED; } - GELOGI( + GELOGD( "Calc op[%s:%s] out[%zu] mem size is %ld," " format=%s, data_type=%s.", node_name.c_str(), node_type.c_str(), i, output_mem_size, TypeUtils::FormatToSerialString(format).c_str(), @@ -174,7 +174,7 @@ Status GeLocalOpsKernelBuilder::GenerateTask(const Node &node, RunContext &conte GELOGE(ret, "Node:%s(%s) op run failed.", name.c_str(), type.c_str()); return ret; } - GELOGI("Ge local generate task for node:%s(%s) end, tasks.size()=%zu.", name.c_str(), type.c_str(), tasks.size()); + GELOGD("Ge local generate task for node:%s(%s) end, tasks.size()=%zu.", name.c_str(), type.c_str(), tasks.size()); return ret; } } // namespace ge_local diff --git a/ge/ge_local_engine/ops_kernel_store/op/no_op.cc b/ge/ge_local_engine/ops_kernel_store/op/no_op.cc index 51c65ce0..c2104693 100755 --- a/ge/ge_local_engine/ops_kernel_store/op/no_op.cc +++ b/ge/ge_local_engine/ops_kernel_store/op/no_op.cc @@ -24,7 +24,7 @@ namespace ge_local { NoOp::NoOp(const Node &node, RunContext &run_context) : Op(node, run_context) {} Status NoOp::Run() { - GELOGI("Node:%s type is %s, no need generate task.", name_.c_str(), type_.c_str()); + GELOGD("Node:%s type is %s, no need generate task.", name_.c_str(), type_.c_str()); // Do nothing return SUCCESS; } diff --git a/ge/ge_runner.mk b/ge/ge_runner.mk index c9be823f..a2679ed1 100644 --- a/ge/ge_runner.mk +++ b/ge/ge_runner.mk @@ -372,12 +372,12 @@ LOCAL_SRC_FILES += $(LIBCLIENT_LOCAL_SRC_FILES) LOCAL_STATIC_LIBRARIES := libge_memory \ libadump_server \ libmsprofiler \ + libmmpa \ LOCAL_SHARED_LIBRARIES := \ libc_sec \ libascend_protobuf \ libslog \ - libmmpa \ libgraph \ libregister \ libge_common \ diff --git a/ge/ge_runtime/task/hccl_task.cc b/ge/ge_runtime/task/hccl_task.cc index 3d5f8504..4d76e492 100644 --- a/ge/ge_runtime/task/hccl_task.cc +++ b/ge/ge_runtime/task/hccl_task.cc @@ -15,83 +15,56 @@ */ #include "ge_runtime/task/hccl_task.h" +#include #include "ge_runtime/task/task_factory.h" #include "common/opskernel/ops_kernel_info_store.h" #include "common/opskernel/ge_task_info.h" namespace ge { namespace model_runner { +std::map>>> + HcclTask::model_stream_mapping_; +std::mutex HcclTask::model_stream_mapping_mutex_; + HcclTask::HcclTask(const ModelContext &model_context, const std::shared_ptr &task_info) : TaskRepeater(model_context, task_info), task_info_(task_info), stream_(nullptr), + workspace_mem_(nullptr), rt_model_handle_(nullptr), priority_(0), - slave_stream_list_(), - hcom_bind_model_(nullptr), - hcom_unbind_model_(nullptr), - hcom_distribute_task_(nullptr) { + secondary_stream_list_() { if (task_info_ == nullptr) { GELOGW("task_info_ is null!"); } - hcom_bind_model_ = task_info->hcom_bind_model(); - hcom_unbind_model_ = task_info->hcom_unbind_model(); - priority_ = model_context.priority(); rt_model_handle_ = model_context.rt_model_handle(); auto stream_list = model_context.stream_list(); - if (hcom_bind_model_ != nullptr) { - if (rt_model_handle_list_.insert(rt_model_handle_).second) { - for (auto stream : stream_list) { - (void)hcom_bind_model_(rt_model_handle_, stream); - } - } - } - if (stream_list.size() == 1) { stream_ = stream_list[0]; } else if (stream_list.size() > task_info->stream_id()) { stream_ = stream_list[task_info->stream_id()]; } else { - GELOGW("index: %u >= stream_list.size(): %zu.", task_info->stream_id(), stream_list.size()); + GELOGW("Index: %u >= stream_list.size(): %zu.", task_info->stream_id(), stream_list.size()); } } HcclTask::~HcclTask() { - for (size_t i = 0; i < slave_stream_list_.size(); ++i) { - rtError_t rt_ret = rtModelUnbindStream(rt_model_handle_, slave_stream_list_[i]); + if (workspace_mem_ != nullptr) { + rtError_t rt_ret = rtFree(workspace_mem_); if (rt_ret != RT_ERROR_NONE) { - GELOGE(RT_FAILED, "Unbind stream from model failed! Index: %zu", i); - } - } - - for (size_t i = 0; i < slave_stream_list_.size(); ++i) { - rtError_t rt_ret = rtStreamDestroy(slave_stream_list_[i]); - if (rt_ret != RT_ERROR_NONE) { - GELOGE(RT_FAILED, "Destroy stream failed! Index: %zu", i); - } - } - - if (hcom_unbind_model_ != nullptr) { - if (rt_model_handle_list_.find(rt_model_handle_) != rt_model_handle_list_.end()) { - (void)hcom_unbind_model_(rt_model_handle_); - (void)rt_model_handle_list_.erase(rt_model_handle_); + GELOGE(RT_FAILED, "rtFree workspace_mem_ failed! ret: 0x%X.", rt_ret); } + workspace_mem_ = nullptr; } } bool HcclTask::Distribute() { - // No ops kernel info store - hcom_distribute_task_ = task_info_->hcom_distribute_task(); - if (hcom_distribute_task_ != nullptr) { - return hcom_distribute_task_(task_info_, stream_); - } - // Ops kernel info store // Get privateDef and opsKernelStorePtr - GELOGI("get custom info in modelTaskDef"); + GELOGI("Get custom info in modelTaskDef"); void *ops_kernel_store = task_info_->ops_kernel_store(); OpsKernelInfoStore *ops_kernel_info_store = reinterpret_cast(ops_kernel_store); if (ops_kernel_store == nullptr) { @@ -101,25 +74,15 @@ bool HcclTask::Distribute() { char *private_def = reinterpret_cast(const_cast(task_info_->private_def().data())); auto private_def_len = static_cast(task_info_->private_def().size()); - GELOGI("the first address of the custom info, privateDef=%p", private_def); - - GELOGI("hcclStreamNum =%ld", task_info_->hccl_stream_num()); - for (int64_t i = 0; i < task_info_->hccl_stream_num(); ++i) { - rtStream_t stream = nullptr; - rtError_t rt_ret = rtStreamCreateWithFlags(&stream, priority_, RT_STREAM_PERSISTENT | RT_STREAM_FORCE_COPY); - if (rt_ret != RT_ERROR_NONE) { - GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret); - return false; - } + GELOGI("The first address of the custom info, privateDef=%p", private_def); + SetSecondaryStream(); - rt_ret = rtModelBindStream(rt_model_handle_, stream, RT_HEAD_STREAM); + if (task_info_->workspace_size() > 0) { + rtError_t rt_ret = rtMalloc(&workspace_mem_, task_info_->workspace_size(), RT_MEMORYINFO_HBM); if (rt_ret != RT_ERROR_NONE) { GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret); return false; } - - GELOGI("hccl_stream addr is=%p", stream); - slave_stream_list_.push_back(stream); } GELOGI("HcclTaskInfo Distribute Start. begin to call function LoadTask in hccl."); @@ -128,17 +91,22 @@ bool HcclTask::Distribute() { ge_task.type = static_cast(RT_MODEL_TASK_HCCL); ge_task.stream = stream_; + ge_task.kernelHcclInfo = std::vector(1); ge_task.kernelHcclInfo[0].hccl_type = task_info_->hccl_type(); ge_task.kernelHcclInfo[0].inputDataAddr = task_info_->input_data_addr(); ge_task.kernelHcclInfo[0].outputDataAddr = task_info_->output_data_addr(); - ge_task.kernelHcclInfo[0].workSpaceAddr = task_info_->workspace_addr(); + ge_task.kernelHcclInfo[0].workSpaceAddr = workspace_mem_; ge_task.kernelHcclInfo[0].workSpaceMemSize = task_info_->workspace_size(); ge_task.kernelHcclInfo[0].count = task_info_->count(); ge_task.kernelHcclInfo[0].dataType = static_cast(task_info_->data_type()); ge_task.kernelHcclInfo[0].opType = static_cast(task_info_->op_type()); ge_task.kernelHcclInfo[0].rootId = task_info_->root_id(); - ge_task.kernelHcclInfo[0].hcclStreamList = slave_stream_list_; + std::vector secondary_stream_list; + std::transform(secondary_stream_list_.begin(), secondary_stream_list_.end(), + std::back_inserter(secondary_stream_list), + [](const std::shared_ptr &stream) -> rtStream_t { return stream->GetStream(); }); + ge_task.kernelHcclInfo[0].hcclStreamList = secondary_stream_list; ge_task.privateDef = private_def; ge_task.privateDefLen = private_def_len; @@ -151,10 +119,152 @@ bool HcclTask::Distribute() { return false; } - GELOGI("call function LoadTask end."); + GELOGI("Call function LoadTask end."); return true; } +bool HcclTask::SetSecondaryStream() { + const uint32_t master_stream_id = task_info_->stream_id(); + const int64_t hccl_secondary_stream_num = task_info_->hccl_stream_num(); + Status ret; + std::lock_guard lock(model_stream_mapping_mutex_); + if (model_stream_mapping_.find(rt_model_handle_) == model_stream_mapping_.end()) { + GELOGI("Need to create map for rt_model_handle_:%p with new mainstream %ld.", rt_model_handle_, master_stream_id); + ret = CreateStream(hccl_secondary_stream_num, master_stream_id); + if (!ret) { + GELOGE(RT_FAILED, "Create hccl stream failed."); + return false; + } + return true; + } + + std::map>> &master_secondary_stream_map = + model_stream_mapping_.at(rt_model_handle_); + auto iter = master_secondary_stream_map.find(master_stream_id); + if (iter != master_secondary_stream_map.end()) { + std::vector> &secondary_stream_vec = iter->second; + auto lock_weak_ptr = [&secondary_stream_vec, this](int64_t index) -> bool { + auto stream = secondary_stream_vec[index].lock(); + if (stream == nullptr) { + rtStream_t new_stream = nullptr; + bool ret = CreateStream(rt_model_handle_, &new_stream); + if (!ret) { + GELOGE(FAILED, "CreateStream failed."); + return false; + } + stream = std::make_shared(rt_model_handle_, new_stream); + if (stream == nullptr) { + GELOGE(FAILED, "MakeShared failed."); + return false; + } + secondary_stream_vec[index] = stream; + } + secondary_stream_list_.push_back(stream); + return true; + }; + + if (static_cast(hccl_secondary_stream_num) <= secondary_stream_vec.size()) { + GELOGI("Number of secondary stream is enough to be reused."); + for (int64_t i = 0; i < hccl_secondary_stream_num; ++i) { + if (!lock_weak_ptr(i)) { + GELOGE(FAILED, "Lock weak ptr failed."); + return false; + } + } + } else { + GELOGI("Need to reuse secondary stream and create new secondary stream."); + size_t created_stream_num = secondary_stream_vec.size(); + for (size_t i = 0; i < secondary_stream_vec.size(); ++i) { + if (!lock_weak_ptr(i)) { + GELOGE(FAILED, "Lock weak ptr failed."); + return false; + } + } + ret = CreateStream(hccl_secondary_stream_num - created_stream_num, master_stream_id); + if (ret != SUCCESS) { + GELOGE(RT_FAILED, "Create hccl stream failed."); + return false; + } + } + GELOGI("Initialize hccl secondary stream success, hccl_secondary_stream_num =%ld", hccl_secondary_stream_num); + } else { + GELOGI("Need to create secondary stream for %s with new mainstream %ld.", task_info_->op_name().c_str(), + master_stream_id); + ret = CreateStream(hccl_secondary_stream_num, master_stream_id); + if (!ret) { + GELOGE(RT_FAILED, "Create hccl stream failed."); + return false; + } + } + return true; +} + +bool HcclTask::CreateStream(int64_t stream_num, int64_t master_stream_id) { + GELOGI("Start to create %ld hccl secondary stream.", stream_num); + for (int64_t i = 0; i < stream_num; ++i) { + rtStream_t stream = nullptr; + bool ret = CreateStream(rt_model_handle_, &stream); + if (!ret) { + GELOGE(FAILED, "CreateStream failed."); + return false; + } + + GELOGD("hccl_stream addr is=%p", stream); + auto shared_stream = std::make_shared(rt_model_handle_, stream); + if (shared_stream == nullptr) { + GELOGE(FAILED, "MakeShared failed."); + return false; + } + SaveHcclSecondaryStream(master_stream_id, shared_stream); + secondary_stream_list_.push_back(shared_stream); + } + GELOGI("CreateStream success."); + return true; +} + +bool HcclTask::CreateStream(rtModel_t model, rtStream_t *stream) const { + if (stream == nullptr) { + GELOGE(FAILED, "Output param stream is null."); + return false; + } + + rtError_t rt_ret = rtStreamCreateWithFlags(stream, priority_, RT_STREAM_PERSISTENT | RT_STREAM_FORCE_COPY); + if (rt_ret != RT_ERROR_NONE) { + GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret); + return false; + } + // Create secondary stream, inactive by default, activated by hccl + rt_ret = rtModelBindStream(model, *stream, RT_MODEL_WAIT_ACTIVE_STREAM); + if (rt_ret != RT_ERROR_NONE) { + GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret); + return false; + } + return true; +} + +void HcclTask::SaveHcclSecondaryStream(int64_t master_stream_id, const std::shared_ptr &stream) { + if (model_stream_mapping_.find(rt_model_handle_) == model_stream_mapping_.end()) { + model_stream_mapping_.emplace(rt_model_handle_, std::map>>()); + } + std::map>> &master_secondary_stream_map = + model_stream_mapping_.at(rt_model_handle_); + master_secondary_stream_map[master_stream_id].emplace_back(stream); +} + +HcclTask::StreamGuard::~StreamGuard() { + rtError_t rt_ret = rtModelUnbindStream(model_, stream_); + if (rt_ret != RT_ERROR_NONE) { + GELOGE(RT_FAILED, "Unbind stream from model failed!"); + return; + } + + rt_ret = rtStreamDestroy(stream_); + if (rt_ret != RT_ERROR_NONE) { + GELOGE(RT_FAILED, "Destroy stream failed!"); + return; + } +} + REGISTER_TASK(TaskInfoType::HCCL, HcclTask, HcclTaskInfo); } // namespace model_runner } // namespace ge diff --git a/ge/ge_runtime/task/hccl_task.h b/ge/ge_runtime/task/hccl_task.h index 1649a8bd..11d88944 100644 --- a/ge/ge_runtime/task/hccl_task.h +++ b/ge/ge_runtime/task/hccl_task.h @@ -19,7 +19,9 @@ #include #include +#include #include +#include #include "ge_runtime/task/task.h" namespace ge { @@ -33,18 +35,34 @@ class HcclTask : public TaskRepeater { bool Distribute() override; private: + class StreamGuard; + bool SetSecondaryStream(); + bool CreateStream(int64_t stream_num, int64_t master_stream_id); + bool CreateStream(rtModel_t model, rtStream_t *stream) const; + void SaveHcclSecondaryStream(int64_t master_stream_id, const std::shared_ptr &stream); + std::shared_ptr task_info_; void *stream_; + void *workspace_mem_; rtModel_t rt_model_handle_; int32_t priority_; - std::vector slave_stream_list_; - std::function hcom_bind_model_; - std::function hcom_unbind_model_; - std::function, void *)> hcom_distribute_task_; - static std::set rt_model_handle_list_; + std::vector> secondary_stream_list_; + + // map>> + static std::map>>> model_stream_mapping_; + static std::mutex model_stream_mapping_mutex_; }; -std::set HcclTask::rt_model_handle_list_{}; +class HcclTask::StreamGuard { + public: + StreamGuard(rtModel_t model, rtStream_t stream) : model_(model), stream_(stream) {} + ~StreamGuard(); + rtStream_t GetStream() const { return stream_; } + + private: + rtModel_t model_; + rtStream_t stream_; +}; } // namespace model_runner } // namespace ge diff --git a/ge/generator/ge_generator.cc b/ge/generator/ge_generator.cc index 85f10ecc..a097bd61 100644 --- a/ge/generator/ge_generator.cc +++ b/ge/generator/ge_generator.cc @@ -200,7 +200,6 @@ static Status AddOutputs(const ComputeGraphPtr &graph, const NodePtr &node, cons } static void GetOpsProtoPath(string &opsproto_path) { - GELOGI("Start to get ops proto path schedule."); const char *path_env = std::getenv("ASCEND_OPP_PATH"); if (path_env != nullptr) { string path = path_env; @@ -383,7 +382,6 @@ bool GeGenerator::Impl::ParseVersion(const std::string &line, std::string &versi } version = temp.substr(pos + flag.size()); - GELOGI("Version=%s", version.c_str()); return true; } @@ -425,7 +423,6 @@ bool GeGenerator::Impl::SetAtcVersionInfo(AttrHolder &obj) { path_base = path_base.substr(0, path_base.rfind('/') + 1); std::string version_path = path_base + "version.info"; - GELOGI("version_path is %s", version_path.c_str()); std::string version; if (!GetVersionFromPath(version_path, version)) { GELOGW("Get atc version information failed!"); @@ -436,7 +433,6 @@ bool GeGenerator::Impl::SetAtcVersionInfo(AttrHolder &obj) { GELOGW("Ge model set atc version failed!"); return false; } - GELOGI("Ge model set atc version information success."); return true; } @@ -449,7 +445,6 @@ bool GeGenerator::Impl::SetOppVersionInfo(AttrHolder &obj) { } std::string version_path = path_env; version_path += "/version.info"; - GELOGI("version_path is %s", version_path.c_str()); std::string version; if (!GetVersionFromPath(version_path, version)) { GELOGW("Get opp version information failed!"); @@ -460,7 +455,6 @@ bool GeGenerator::Impl::SetOppVersionInfo(AttrHolder &obj) { GELOGW("Ge model set opp version failed!"); return false; } - GELOGI("Ge Model set opp version information success."); return true; } @@ -469,7 +463,7 @@ Status GeGenerator::GenerateModel(const Graph &graph, const string &file_name_pr rtContext_t ctx = nullptr; auto rt = rtCtxGetCurrent(&ctx); if (rt != RT_ERROR_NONE) { - GELOGW("Current ctx is null."); + GELOGD("Current ctx is null."); ctx = nullptr; } @@ -524,7 +518,6 @@ Status GeGenerator::GenerateModel(const Graph &graph, const string &file_name_pr (void)rtCtxSetCurrent(ctx); } - GELOGI("GenerateOfflineModel success."); return SUCCESS; } @@ -713,7 +706,6 @@ Status GeGenerator::Impl::BuildModel(const Graph &graph, const vector return GE_GENERATOR_GRAPH_MANAGER_ADD_GRAPH_FAILED; } - GELOGI("Model inputs size is %zu", inputs.size()); graph_manager_.SetOptionsRunGraphFlag(false); static std::atomic atomic_session_id(0); diff --git a/ge/graph/build/graph_builder.cc b/ge/graph/build/graph_builder.cc index 1da84991..e52f84ed 100644 --- a/ge/graph/build/graph_builder.cc +++ b/ge/graph/build/graph_builder.cc @@ -102,7 +102,6 @@ void GraphBuilder::SetOptions(const ge::GraphManagerOptions &options) { } Status GraphBuilder::CalcOpParam(const ge::ComputeGraphPtr &graph) { - GELOGI("Begin to calculate op running param."); GE_CHECK_NOTNULL(graph); auto instance_ptr = ge::GELib::GetInstance(); if (instance_ptr == nullptr || !instance_ptr->InitFlag()) { @@ -140,7 +139,6 @@ Status GraphBuilder::CalcOpParam(const ge::ComputeGraphPtr &graph) { auto parent_node = graph->GetParentNode(); if (parent_node == nullptr) { - GELOGI("Graph[%s] do not have parent node, no need update parent node output size.", graph->GetName().c_str()); return SUCCESS; } @@ -189,7 +187,6 @@ Status GraphBuilder::UpdateParentNodeOutputSize(const ge::ComputeGraphPtr &graph Status GraphBuilder::Build(ComputeGraphPtr &comp_graph, std::vector &subgraph_ptr_list, GeRootModelPtr &ge_root_model_ptr, uint64_t session_id) { - GELOGI("Start to build model."); if (comp_graph == nullptr) { GELOGE(GE_GRAPH_PARAM_NULLPTR, "Graph build comp_graph is null."); return GE_GRAPH_PARAM_NULLPTR; @@ -267,7 +264,7 @@ Status GraphBuilder::BuildForKnownShapeGraph(ComputeGraphPtr &comp_graph, std::v } GE_CHK_STATUS_RET(builder.SaveDataToModel(*model_ptr, *ge_model_ptr), "Graph[%s] builder SaveDataToModel() return fail.", comp_graph->GetName().c_str()); - GELOGI("Success to build graph[%s] model.", comp_graph->GetName().c_str()); + GELOGD("Success to build graph[%s] model.", comp_graph->GetName().c_str()); GE_TIMESTAMP_END(BuildSubgraph, "GraphBuilder::Build"); return SUCCESS; } @@ -306,7 +303,7 @@ Status GraphBuilder::BuildForUnknownShapeGraph(ComputeGraphPtr &comp_graph, GeMo } GE_CHK_STATUS_RET(builder.SaveDataToModel(*model_ptr, *ge_model_ptr), "Graph[%s] builder SaveDataToModel() return fail.", comp_graph->GetName().c_str()); - GELOGI("Success to build graph[%s] model.", comp_graph->GetName().c_str()); + GELOGD("Success to build graph[%s] model.", comp_graph->GetName().c_str()); return SUCCESS; } @@ -542,7 +539,6 @@ Status GraphBuilder::CalcDynShapeRootGraphDataSize(const ge::OpDescPtr &op_desc) } Status GraphBuilder::SecondPartition(ge::ComputeGraphPtr &comp_graph, vector &subgraph_ptr_list) { - GELOGI("[SecondPartition] second partition."); GE_TIMESTAMP_START(GraphPartition2); auto ret = graph_partitioner_.Partition(comp_graph, GraphPartitioner::kSecondPartitioning); if (ret != SUCCESS) { diff --git a/ge/graph/build/label_allocator.cc b/ge/graph/build/label_allocator.cc index 51f31003..28d0e084 100644 --- a/ge/graph/build/label_allocator.cc +++ b/ge/graph/build/label_allocator.cc @@ -33,7 +33,7 @@ Status LabelAllocator::AssignFunctionalLabels() { } // Add label task for sub graph. - GELOGI("AssignFunctionalLabels start: %s.", compute_graph_->GetName().c_str()); + GELOGD("AssignFunctionalLabels start: %s.", compute_graph_->GetName().c_str()); std::set functional_nodes; for (auto graph : compute_graph_->GetAllSubgraphs()) { if (!CollectFunctionalNode(graph, functional_nodes)) { diff --git a/ge/graph/build/logical_stream_allocator.cc b/ge/graph/build/logical_stream_allocator.cc index df65f0a9..8ea7fe71 100644 --- a/ge/graph/build/logical_stream_allocator.cc +++ b/ge/graph/build/logical_stream_allocator.cc @@ -597,10 +597,10 @@ Status LogicalStreamAllocator::DoAssign(const ComputeGraphPtr &graph, const Grap return status; } - GELOGI("Subgraphs of graph %s:", graph->GetName().c_str()); + GELOGD("Subgraphs of graph %s:", graph->GetName().c_str()); for (const auto &subgraph : subgraphs) { if (subgraph != nullptr) { - GELOGI("subgraph: %s", subgraph->name.c_str()); + GELOGD("subgraph: %s", subgraph->name.c_str()); } } @@ -664,9 +664,9 @@ Status LogicalStreamAllocator::RunPasses(const ComputeGraphPtr &graph, const vec Status status = pass->Run(graph, subgraphs, context_); if (status == SUCCESS) { - GELOGI("Stream pass %s return SUCCESS.", pass->GetName().c_str()); + GELOGD("Stream pass %s return SUCCESS.", pass->GetName().c_str()); } else if (status == NOT_CHANGED) { - GELOGI("Stream pass %s return NOT_CHANGED.", pass->GetName().c_str()); + GELOGD("Stream pass %s return NOT_CHANGED.", pass->GetName().c_str()); } else { GELOGE(status, "Stream pass %s failed.", pass->GetName().c_str()); return status; diff --git a/ge/graph/build/memory/binary_block_mem_assigner.cc b/ge/graph/build/memory/binary_block_mem_assigner.cc index 61dd3462..ecd2488c 100644 --- a/ge/graph/build/memory/binary_block_mem_assigner.cc +++ b/ge/graph/build/memory/binary_block_mem_assigner.cc @@ -76,7 +76,7 @@ Status BinaryBlockMemAssigner::GetMemoryRanges(vector &range_ceils) { auto range_number = static_cast( ceil(log(all_memory_size.back() / static_cast(all_memory_size.front())) / log(kLogBase))); range_number = (range_number == 0) ? 1 : range_number; - GELOGI("Range number: %zu", range_number); + GELOGD("Range number: %zu", range_number); vector> ranges(range_number); GE_CHK_BOOL_EXEC((range_number != 0), return PARAM_INVALID, "range_number can't be 0."); @@ -114,7 +114,7 @@ Status BinaryBlockMemAssigner::GetMemoryRanges(vector &range_ceils) { range_ceils.push_back(range.back()); } } - GELOGI("Range ceils: %s", ToString(range_ceils).c_str()); + GELOGD("Range ceils: %s", ToString(range_ceils).c_str()); return SUCCESS; } diff --git a/ge/graph/build/memory/block_mem_assigner.cc b/ge/graph/build/memory/block_mem_assigner.cc index ecf253e1..00f47573 100755 --- a/ge/graph/build/memory/block_mem_assigner.cc +++ b/ge/graph/build/memory/block_mem_assigner.cc @@ -455,12 +455,11 @@ void BlockMemAssigner::GetOutAndWorkSpaceMem(vector &all_memory_size) { GetNodeWorkSpaceSize(n, temp); all_memory_size.insert(all_memory_size.end(), temp.begin(), temp.end()); } - GELOGI("The last atomic_addr_clean node id: %ld", atomic_addr_clean_id_); for (const auto &pair : symbol_size_) { all_memory_size.emplace_back(pair.second); } sort(all_memory_size.begin(), all_memory_size.end()); - GELOGI("All memory size: %s", ToString(all_memory_size).c_str()); + GELOGD("All memory size: %s", ToString(all_memory_size).c_str()); for (auto iter = all_memory_size.begin(); iter != all_memory_size.end();) { if (*iter == 0) { @@ -495,7 +494,7 @@ size_t GetBlockSize(size_t size, const vector &ranges) { bool IsDirectOutputNode(const NodePtr &node, int idx) { if ((node != nullptr) && (node->GetOpDesc() != nullptr) && (node->GetOpDesc()->GetType() == NETOUTPUT)) { - GELOGI("This is netoutput node, the input node mem can not be reused"); + GELOGD("This is netoutput node, the input node mem can not be reused"); return true; } return false; @@ -1102,7 +1101,7 @@ bool IsAtomicOutputMemory(const ge::NodePtr &node, uint32_t output_index, bool i if (static_cast(index) == output_index) { if (node->GetOwnerComputeGraph() != nullptr) { string graph_name = node->GetOwnerComputeGraph()->GetName(); - GELOGD("[IMAS]Atomic no assign %s name[%s] output[%d] streamid[%ld].", graph_name.c_str(), + GELOGD("[IMAS]Atomic no assign %s name[%s] output[%ld] streamid[%ld].", graph_name.c_str(), op_desc->GetName().c_str(), index, op_desc->GetStreamId()); } return true; @@ -1219,7 +1218,7 @@ Status BlockMemAssigner::AssignOutputMemoryWithReuse(const NodePtr &node, vector int64_t stream_id = op_desc->GetStreamId(); vector memorys_type; bool has_mem_type_attr = ge::AttrUtils::GetListInt(op_desc, ATTR_NAME_OUTPUT_MEM_TYPE_LIST, memorys_type); - GELOGI("Assign memory node[%s], output size[%d], output memory type size[%d]", op_desc->GetName().c_str(), + GELOGD("Assign memory node[%s], output size[%zu], output memory type size[%zu]", op_desc->GetName().c_str(), op_desc->GetOutputsSize(), memorys_type.size()); if (has_mem_type_attr && (memorys_type.size() != op_desc->GetOutputsSize())) { GELOGE(INTERNAL_ERROR, "fusion: node[%s], output memory size err[outputsize:%zu, memorysize:%zu]", @@ -1257,7 +1256,7 @@ Status BlockMemAssigner::AssignOutputMemoryWithReuse(const NodePtr &node, vector // fusion: other type's size not means malloc HBM memory bool l1_flag = has_mem_type_attr && memorys_type[i] == RT_MEMORY_L1; if (l1_flag) { - GELOGI("fusion: node[%s], output[%s], output memory type [%d]", + GELOGI("fusion: node[%s], output[%s], output memory type [%ld]", op_desc->GetName().c_str(), op_desc->GetOutputNameByIndex(i).c_str(), memorys_type[i]); size = 0; } @@ -1311,7 +1310,7 @@ Status BlockMemAssigner::AssignOutputMemoryWithReuse(const NodePtr &node, vector /// void BlockMemAssigner::AssignMemoryWithReuse(vector &ranges) { (void)ge::GetContext().GetOption(OPTION_EXEC_DISABLE_REUSED_MEMORY, ge_disable_reuse_mem_env_); - GEEVENT("Reuse memory %s", ge_disable_reuse_mem_env_ == "1" ? "close" : "open"); + GELOGD("Reuse memory %s", ge_disable_reuse_mem_env_ == "1" ? "close" : "open"); string op_no_reuse_mem_str; const char *op_no_reuse_mem = std::getenv(OP_NO_REUSE_MEM); GE_IF_BOOL_EXEC(op_no_reuse_mem != nullptr, op_no_reuse_mem_str = string(op_no_reuse_mem); @@ -1337,7 +1336,7 @@ void BlockMemAssigner::AssignMemoryWithReuse(vector &ranges) { vector workspace_reuse_flag; GE_IF_BOOL_EXEC(!ge::AttrUtils::GetListBool(node_op_desc, kAttrNameWorkspaceReuseFlag, workspace_reuse_flag), GELOGD("OP %s get workspace_reuse_flag attr failed", node_op_desc->GetName().c_str())); - GELOGI("Assign memory node[%s], size [temp:%zu, memory type size:%zu]", node_op_desc->GetName().c_str(), + GELOGD("Assign memory node[%s], size [temp:%zu, memory type size:%zu]", node_op_desc->GetName().c_str(), temp.size(), tvm_workspace_memory_type.size()); if (has_tvm_workspace_mem_type_attr && (temp.size() != tvm_workspace_memory_type.size())) { @@ -1350,7 +1349,7 @@ void BlockMemAssigner::AssignMemoryWithReuse(vector &ranges) { bool workspace_skip_flag = false; if (has_tvm_workspace_mem_type_attr && tvm_workspace_memory_type[i] == RT_MEMORY_L1) { GELOGI( - "fusion: node[%s]workspace index[%d] is not hbm type, add to zero_memory_list, workspace memory type [%ld]", + "fusion: node[%s]workspace index[%zu] is not hbm type, add to zero_memory_list, workspace memory type [%ld]", node_op_desc->GetName().c_str(), i, tvm_workspace_memory_type[i]); workspace_skip_flag = true; } @@ -1628,7 +1627,7 @@ void BlockMemAssigner::ResizeMemoryBlocks() { memory_block->SetTailOffset(p2p_mem_offset_ - 1); } } - GELOGI("mem_offset_ exclude zero_copy_memory is %zu, p2p_mem_offset_ exclude zero_copy_memory is %zu.", + GELOGD("mem_offset_ exclude zero_copy_memory is %zu, p2p_mem_offset_ exclude zero_copy_memory is %zu.", mem_offset_, p2p_mem_offset_); } diff --git a/ge/graph/build/memory/graph_mem_assigner.cc b/ge/graph/build/memory/graph_mem_assigner.cc index 7d35596c..ad0235d5 100755 --- a/ge/graph/build/memory/graph_mem_assigner.cc +++ b/ge/graph/build/memory/graph_mem_assigner.cc @@ -117,7 +117,7 @@ Status GraphMemoryAssigner::AssignMemory() { return ge::FAILED; } int64_t var_size_assign = ge::VarManager::Instance(session_id)->GetVarMemSize(RT_MEMORY_HBM) - var_size_before_assign; - GELOGI("GraphMemoryAssigner::AssignMemory variable size = %ld", var_size_assign); + GELOGD("GraphMemoryAssigner::AssignMemory variable size = %ld", var_size_assign); mem_assigner_ = std::move(mem_assigner); @@ -296,7 +296,6 @@ Status GraphMemoryAssigner::AssignZeroCopyMemory(map &mem_offse mem_offset[RT_MEMORY_HBM] += memory_block->Size(); memory_block->SetTailOffset(mem_offset[RT_MEMORY_HBM] - 1); } - GELOGI("mem_offset_ include zero_copy_memory is %zu.", mem_offset[RT_MEMORY_HBM]); // set offset for zero copy nodes priority_assigner->SetOpMemOffset(true); @@ -309,14 +308,13 @@ Status GraphMemoryAssigner::AssignZeroCopyMemory(map &mem_offse } iter->second.mem_offset_ = mem_offset[RT_MEMORY_HBM]; - GELOGI("max_mem_offset:%zu, mem_offset:%zu, zero_mem_copy_size:%zu.", mem_offset[RT_MEMORY_HBM], mem_offset_tmp, + GELOGD("max_mem_offset:%zu, mem_offset:%zu, zero_mem_copy_size:%zu.", mem_offset[RT_MEMORY_HBM], mem_offset_tmp, zero_mem_copy_size); return SUCCESS; } Status GraphMemoryAssigner::ReAssignContinuousMemory(bool is_loop_graph) { - GELOGI("Begin to reassign continuous memory"); Status ret; for (auto &node : compute_graph_->GetAllNodes()) { // Get the continuous input type of the node, default is false @@ -387,7 +385,7 @@ Status GraphMemoryAssigner::ReAssignContinuousMemory(bool is_loop_graph) { } } for (auto pair : memory_offset_) { - GELOGI("After reassign continuous memory, memory type = %ld, memoffset = %zu.", pair.first, + GELOGD("After reassign continuous memory, memory type = %ld, memoffset = %zu.", pair.first, pair.second.mem_offset_); } return ge::SUCCESS; @@ -456,7 +454,7 @@ Status GraphMemoryAssigner::AssignContinuousInputMemory(const ge::NodePtr &node, output_list.at(peer_out_data_anchor->GetIdx()) - continuous_mem_start + tensor_desc_size + MEM_ALIGN_SIZE; } GELOGI( - "[IMAS]Check Continuous input : Set %s name[%s] output[%d] offset to [%zu] stream_id[%ld] size[%zu] " + "[IMAS]Check Continuous input : Set %s name[%s] output[%d] offset to [%ld] stream_id[%ld] size[%u] " "real_size[%u].", node->GetOwnerComputeGraph()->GetName().c_str(), peer_op_desc->GetName().c_str(), peer_out_data_anchor->GetIdx(), output_list.at(peer_out_data_anchor->GetIdx()), peer_op_desc->GetStreamId(), @@ -834,7 +832,6 @@ Status GraphMemoryAssigner::ReAssignVirtualNodesMemory(map nodes_mem_offset_list; for (auto &i_map : mem_reuse_nodes_map) { @@ -1507,7 +1504,7 @@ ge::Status GraphMemoryAssigner::UpdateOpInputOffset(const NodePtr &node, vector< GE_CHK_STATUS(TensorUtils::GetDataOffset(tensor_desc, input_offset)); } - GELOGI("%s node[%s] input[%d] is set from node[%s] out index[%lu] offset[%ld]", + GELOGD("%s node[%s] input[%d] is set from node[%s] out index[%lu] offset[%ld]", has_mem_type_attr == true ? "Fusion" : "", tmp_op_desc->GetName().c_str(), valid_input_index, diff --git a/ge/graph/build/memory/hybrid_mem_assigner.cc b/ge/graph/build/memory/hybrid_mem_assigner.cc index 8a36ad62..462e190a 100755 --- a/ge/graph/build/memory/hybrid_mem_assigner.cc +++ b/ge/graph/build/memory/hybrid_mem_assigner.cc @@ -62,9 +62,9 @@ Status HybridMemAssigner::Assign() { std::unique_ptr priority_assigner; - GELOGI("Binary-block memory size:%zu, max-block memory size:%zu", bin_mem_size, max_mem_size); + GELOGD("Binary-block memory size:%zu, max-block memory size:%zu", bin_mem_size, max_mem_size); if (bin_mem_size <= max_mem_size) { - GELOGI("Use binary-block memory assigner method"); + GELOGD("Use binary-block memory assigner method"); priority_assigner = std::move(binary_assigner); } else { GELOGI("Use max-block memory assigner method"); diff --git a/ge/graph/build/model_builder.cc b/ge/graph/build/model_builder.cc index e8657a49..d7039cfb 100755 --- a/ge/graph/build/model_builder.cc +++ b/ge/graph/build/model_builder.cc @@ -189,7 +189,6 @@ void ModelBuilder::SetInputIsConst(const ge::NodePtr &n) { GE_IF_BOOL_EXEC(peer_out_anchor == nullptr, continue); const auto &src_node = peer_out_anchor->GetOwnerNode(); if (!NodeUtils::GetConstOpType(src_node, const_type)) { - GELOGI("Node %s:%zu, sorce node: %s Not Const", n->GetName().c_str(), index, src_node->GetName().c_str()); continue; } @@ -232,7 +231,6 @@ Status ModelBuilder::AdjustConstWeightSize(const ge::NodePtr &node, size_t &mem_ Status ModelBuilder::SetInputOutputDesc() { Status ret; - GELOGI("Start to SetInputOutputDesc."); for (const ge::NodePtr &n : compute_graph_->GetNodes(compute_graph_->GetGraphUnknownFlag())) { auto node_op_desc = n->GetOpDesc(); @@ -245,7 +243,6 @@ Status ModelBuilder::SetInputOutputDesc() { // final graph. if ((GetLocalOmgContext().format == domi::DOMI_TENSOR_ND) && (!node_op_desc->HasAttr("_is_single_op")) && ((node_op_desc->GetType() == DATA_TYPE) || (node_op_desc->GetType() == NETOUTPUT))) { - GELOGI("The node [%s] format should be set ND.", node_op_desc->GetName().c_str()); auto inputDescsPtr = node_op_desc->GetAllInputsDescPtr(); auto outputDescsPtr = node_op_desc->GetAllOutputsDescPtr(); ge::Format format = ge::FORMAT_ND; @@ -290,7 +287,7 @@ void ModelBuilder::AddNodeInputProperty() { vector src_index_list; for (const auto &in_data_anchor : node->GetAllInDataAnchors()) { auto peer_out_anchor = in_data_anchor->GetPeerOutAnchor(); - GE_IF_BOOL_EXEC(peer_out_anchor == nullptr, GELOGW("peer_out_anchor is nullptr!"); continue); + GE_IF_BOOL_EXEC(peer_out_anchor == nullptr, continue); GE_IF_BOOL_EXEC(node_op_desc->HasAttr(MERGE_PRENODE_FLAG), continue); ge::NodePtr src_node = peer_out_anchor->GetOwnerNode(); @@ -347,7 +344,6 @@ void ModelBuilder::AddNodeInputProperty() { } Status ModelBuilder::AdjustInputTensorFlag() { - GELOGI("Start to AdjustInputTensorFlag."); for (const ge::NodePtr &n : compute_graph_->GetNodes(compute_graph_->GetGraphUnknownFlag())) { if ((n->GetType() == DATA_TYPE) || (n->GetType() == AIPP_DATA_TYPE)) { GELOGD("Data node: %s.", n->GetName().c_str()); @@ -441,7 +437,6 @@ Status ModelBuilder::BuildModelDef(ge::Model &model) { return FAILED); const DumpProperties &dump_properties = PropertiesManager::Instance().GetDumpProperties(session_id_); bool is_op_debug = dump_properties.IsOpDebugOpen(); - GELOGI("Get op debug:%d", is_op_debug); if (is_op_debug) { if (!ge::AttrUtils::SetBool(&model, ATTR_OP_DEBUG_FLAG, is_op_debug)) { GELOGE(FAILED, "SetBool of ATTR_OP_DEBUG_FLAG failed."); @@ -608,7 +603,6 @@ Status ModelBuilder::SaveDataToModel(ge::Model &model, ge::GeModel &ge_model) { } tbe_name_set.insert(tbe_kernel->GetName()); tbe_kernel_store_.AddTBEKernel(tbe_kernel); - GELOGI("Add tbe kernel bin %s", tbe_kernel->GetName().c_str()); } for (const ge::NodePtr &n : compute_graph_->GetNodes(compute_graph_->GetGraphUnknownFlag())) { @@ -678,7 +672,6 @@ Status ModelBuilder::PreBuildModel() { GELOGE(FAILED, "Graph_ is not valid."); return FAILED; } - GELOGI("BuildModel begin."); GE_CHK_STATUS_RET(SetInputOutputDesc(), "SetInputOutputDesc Failed!"); diff --git a/ge/graph/build/run_context.cc b/ge/graph/build/run_context.cc index 3aa4d3de..50094cf3 100644 --- a/ge/graph/build/run_context.cc +++ b/ge/graph/build/run_context.cc @@ -140,7 +140,7 @@ void RunContextUtil::DestroyRtModelResources() noexcept { Status RunContextUtil::CreateRunContext(Model &model, const ComputeGraphPtr &graph, Buffer &buffer, const uint64_t session_id) { - GELOGI("Begin to Create RunContext, session_id = %lu", session_id); + GELOGD("Begin to Create RunContext, session_id = %lu", session_id); // check params if (graph == nullptr) { GELOGE(PARAM_INVALID, "CreateRunContext param graph is null. session_id=%lu", session_id); @@ -152,21 +152,21 @@ Status RunContextUtil::CreateRunContext(Model &model, const ComputeGraphPtr &gra GELOGE(INTERNAL_ERROR, "Get stream_num attr from model_def failed. session_id=%lu", session_id); return INTERNAL_ERROR; } - GELOGI("Stream_num = %u", stream_num); + GELOGD("Stream_num = %u", stream_num); uint32_t event_num = 0; if (!AttrUtils::GetInt(&model, ATTR_MODEL_EVENT_NUM, event_num)) { GELOGE(INTERNAL_ERROR, "Get event_num attr from model failed. session_id=%lu", session_id); return INTERNAL_ERROR; } - GELOGI("Event_num = %u", event_num); + GELOGD("Event_num = %u", event_num); uint32_t label_num = 0; if (!AttrUtils::GetInt(&model, ATTR_MODEL_LABEL_NUM, label_num)) { GELOGE(INTERNAL_ERROR, "Get label_num attr from model failed. session_id=%lu", session_id); return INTERNAL_ERROR; } - GELOGI("Label_num = %u", label_num); + GELOGD("Label_num = %u", label_num); Status ret = CreateRtModelResources(stream_num, event_num, label_num); if (ret != SUCCESS) { @@ -198,11 +198,11 @@ Status RunContextUtil::CreateRunContext(Model &model, const ComputeGraphPtr &gra void RunContextUtil::PrintMemInfo() { for (auto iter : mem_type_to_data_mem_base_) { - GELOGI("CreateRunContext: memory type = %ld, data memory base = %p", iter.first, iter.second); + GELOGD("CreateRunContext: memory type = %ld, data memory base = %p", iter.first, iter.second); } for (auto iter : mem_type_to_data_mem_size_) { - GELOGI("CreateRunContext: memory type = %ld, data memory size = %lu", iter.first, iter.second); + GELOGD("CreateRunContext: memory type = %ld, data memory size = %lu", iter.first, iter.second); } } diff --git a/ge/graph/build/stream_allocator.cc b/ge/graph/build/stream_allocator.cc index 3aba8fd1..2f32eb73 100644 --- a/ge/graph/build/stream_allocator.cc +++ b/ge/graph/build/stream_allocator.cc @@ -67,11 +67,10 @@ StreamAllocator::StreamAllocator(ComputeGraphPtr whole_graph, const Graph2SubGra } enable_single_stream_ = (single_stream_str == kTrueStr) ? true : false; - GELOGI("Enable single stream: %s.", enable_single_stream_ ? kTrueStr : kFalseStr); + GELOGD("Enable single stream: %s.", enable_single_stream_ ? kTrueStr : kFalseStr); } Status StreamAllocator::AssignLogicalStreams(const std::map &max_parallel_num, bool hcom_parallel) { - GELOGI("Assign logical streams start."); GE_CHECK_NOTNULL(whole_graph_); GE_DUMP(whole_graph_, "BeforeAssignedLogicalStreams"); @@ -92,15 +91,12 @@ Status StreamAllocator::AssignLogicalStreams(const std::map &m return status; } GE_DUMP(whole_graph_, "AfterAssignedLogicalStreams"); - GELOGI("Assign logical streams success."); - return SUCCESS; } // After allocating the logical stream in the graph, refresh the stream in the // graph and insert the synchronization node. Status StreamAllocator::RefreshRealStream(int64_t &stream_num, int64_t &event_num) { - GELOGI("RefreshRealStream start."); GE_CHECK_NOTNULL(whole_graph_); GE_DUMP(whole_graph_, "BeforeRefreshRealStream"); @@ -174,8 +170,7 @@ Status StreamAllocator::RefreshRealStream(int64_t &stream_num, int64_t &event_nu GELOGI("None of nodes need to assign stream, stream num is 0, it will cause error, so change it to 1"); stream_num_ = 1; } - GELOGI("stream num: %ld, event num: %u.", stream_num_, event_num_); - GELOGI("RefreshRealStream successfully."); + GELOGD("stream num: %ld, event num: %u.", stream_num_, event_num_); stream_num = stream_num_; event_num = static_cast(event_num_); @@ -1241,7 +1236,7 @@ void StreamAllocator::DumpEvents() { for (const auto &one_pair : after_refresh_stream_nodes) { int64_t stream_id = one_pair.first; - GELOGI("After RefreshRealStream: stream %ld.", stream_id); + GELOGD("After RefreshRealStream: stream %ld.", stream_id); for (const auto &node : one_pair.second) { string send_event_str; @@ -1273,7 +1268,7 @@ Status StreamAllocator::GetMaxStreamAndTask(bool huge_stream, uint32_t &max_stre GELOGE(FAILED, "Get max stream and task count by rts failed."); return FAILED; } - GELOGI("Allowed max stream count: %u, max task count per stream: %u.", max_stream_count, max_task_count); + GELOGD("Allowed max stream count: %u, max task count per stream: %u.", max_stream_count, max_task_count); return SUCCESS; } diff --git a/ge/graph/build/stream_graph_optimizer.cc b/ge/graph/build/stream_graph_optimizer.cc index 21625a1e..582c080b 100644 --- a/ge/graph/build/stream_graph_optimizer.cc +++ b/ge/graph/build/stream_graph_optimizer.cc @@ -30,7 +30,7 @@ StreamGraphOptimizer::~StreamGraphOptimizer() {} void StreamGraphOptimizer::RefreshNodeId(const ComputeGraphPtr &comp_graph, Graph2SubGraphInfoList &subgraph_map) { size_t node_size = comp_graph->GetAllNodesSize(); - GELOGI("Refresh placeholder and end nodeId start from node num: %zu", node_size); + GELOGD("Refresh placeholder and end nodeId start from node num: %zu", node_size); for (const auto &subgraph_pair : subgraph_map) { for (const auto &subgraph_info : subgraph_pair.second) { ComputeGraphPtr subgraph = subgraph_info->GetSubGraph(); @@ -74,8 +74,6 @@ bool StreamGraphOptimizer::IsSameStreamId(const ComputeGraphPtr &comp_graph) { Status StreamGraphOptimizer::OptimizeStreamedSubGraph(const ComputeGraphPtr &comp_graph, Graph2SubGraphInfoList &subgraph_map, struct RunContext &run_context) { - GELOGI("Optimize streamed subgraph start."); - RefreshNodeId(comp_graph, subgraph_map); std::shared_ptr instance = ge::GELib::GetInstance(); @@ -86,7 +84,7 @@ Status StreamGraphOptimizer::OptimizeStreamedSubGraph(const ComputeGraphPtr &com ComputeGraphPtr subgraph = subgraph_info->GetSubGraph(); GE_CHECK_NOTNULL(subgraph); - GELOGI("Optimize subgraph %s", subgraph->GetName().c_str()); + GELOGD("Optimize subgraph %s", subgraph->GetName().c_str()); std::string engine_name = subgraph_info->GetEngineName(); @@ -128,7 +126,7 @@ Status StreamGraphOptimizer::OptimizeStreamedSubGraph(const ComputeGraphPtr &com subgraph->GetName().c_str(), engine_name.c_str(), graph_optimizers.size(), ret); return ret; } - GELOGI( + GELOGD( "[optimizeStreamedSubGraph]: optimize streamed subgraph success, subgraph: %s, engine_name: %s, graph " "Optimizer num: %zu!", subgraph->GetName().c_str(), engine_name.c_str(), graph_optimizers.size()); @@ -137,7 +135,7 @@ Status StreamGraphOptimizer::OptimizeStreamedSubGraph(const ComputeGraphPtr &com } } - GELOGI("Optimize streamed subgraph success."); + GELOGD("Optimize streamed subgraph success."); return SUCCESS; } } // namespace ge diff --git a/ge/graph/build/task_generator.cc b/ge/graph/build/task_generator.cc index 9246fa6e..41607f1f 100755 --- a/ge/graph/build/task_generator.cc +++ b/ge/graph/build/task_generator.cc @@ -68,7 +68,7 @@ TaskGenerator::TaskGenerator(uint8_t *var_mem_base, uint64_t var_mem_size) { TaskGenerator::~TaskGenerator() {} Status TaskGenerator::GetTaskInfo(Model &model, ComputeGraphPtr &graph, uint64_t session_id, RunContext &run_context) { - GELOGI("Begin to Get TaskInfo. session_id=%lu", session_id); + GELOGD("Begin to Get TaskInfo. session_id=%lu", session_id); // Check params if (graph == nullptr) { GELOGE(PARAM_INVALID, "GetTaskInfo param graph is null. session_id=%lu", session_id); @@ -120,7 +120,7 @@ Status TaskGenerator::GetTaskInfo(Model &model, ComputeGraphPtr &graph, uint64_t return ret; } - GELOGI("Get TaskInfo success. session_id=%lu", session_id); + GELOGD("Get TaskInfo success. session_id=%lu", session_id); return SUCCESS; } @@ -232,7 +232,7 @@ Status TaskGenerator::SaveFusionNodes(map> &fusion } } } - GELOGI("Fusion: get fusion group numbers [%zu].", fusion_nodes.size()); + GELOGD("Fusion: get fusion group numbers [%zu].", fusion_nodes.size()); return SUCCESS; } @@ -575,7 +575,7 @@ Status TaskGenerator::MarkFirstAndLastOps(const vector &ops, bool is_ continuous_op_lists.back().emplace_back(op_desc); } } - GELOGI("Number of continuous node lists is %zu.", continuous_op_lists.size()); + GELOGD("Number of continuous node lists is %zu.", continuous_op_lists.size()); for (const auto &continuous_ops : continuous_op_lists) { map> first_and_last_ops; @@ -846,13 +846,12 @@ Status TaskGenerator::GetFpBpIndex(const ComputeGraphPtr &graph, ProfilingPoint Status TaskGenerator::FindProfilingTaskIndex(const ComputeGraphPtr &graph, ProfilingPoint &profiling_point, vector &all_reduce_nodes) const { - GELOGI("Start FindProfilingTaskIndex."); GE_CHECK_NOTNULL(graph); const char *profiling_mode = std::getenv(kProfilingMode); bool is_profiling = (profiling_mode != nullptr) || ProfilingManager::Instance().ProfilingOn() || ProfilingManager::Instance().ProfilingTrainingTraceOn(); if (!is_profiling) { - GELOGW("Profiling is not open."); + GELOGD("Profiling is not open."); return SUCCESS; } diff --git a/ge/graph/execute/graph_execute.cc b/ge/graph/execute/graph_execute.cc index 6907e7a6..97e2fd1b 100755 --- a/ge/graph/execute/graph_execute.cc +++ b/ge/graph/execute/graph_execute.cc @@ -115,7 +115,7 @@ Status GraphExecutor::FreeInOutBuffer() { malloc_flag_ = false; return SUCCESS; } else { - GELOGI("[GraphManager] not malloc buffer."); + GELOGD("[GraphManager] not malloc buffer."); return SUCCESS; } } diff --git a/ge/graph/load/graph_loader.cc b/ge/graph/load/graph_loader.cc index c7a572d4..2eeecc0f 100755 --- a/ge/graph/load/graph_loader.cc +++ b/ge/graph/load/graph_loader.cc @@ -286,7 +286,7 @@ Status GraphLoader::ExecuteModel(uint32_t model_id, rtStream_t stream, bool asyn return ret; } - GELOGI("Execute model success, model_id:%u.", model_id); + GELOGD("Execute model success, model_id:%u.", model_id); return SUCCESS; } diff --git a/ge/graph/load/new_model_manager/cpu_queue_schedule.cc b/ge/graph/load/new_model_manager/cpu_queue_schedule.cc index 7f406985..430321bd 100644 --- a/ge/graph/load/new_model_manager/cpu_queue_schedule.cc +++ b/ge/graph/load/new_model_manager/cpu_queue_schedule.cc @@ -131,7 +131,7 @@ Status CpuTaskZeroCopy::Init(std::vector &mbuf_list, std::map(reinterpret_cast(virtual_args_addr.second.at(i)))); + dst_addrs.push_back(static_cast(reinterpret_cast(virtual_args_addr.second.at(i)))); } } index++; diff --git a/ge/graph/load/new_model_manager/data_dumper.cc b/ge/graph/load/new_model_manager/data_dumper.cc index c6100129..b4b31941 100644 --- a/ge/graph/load/new_model_manager/data_dumper.cc +++ b/ge/graph/load/new_model_manager/data_dumper.cc @@ -159,7 +159,6 @@ void DataDumper::SetLoopAddr(void *global_step, void *loop_per_iter, void *loop_ } void DataDumper::SaveDumpInput(const std::shared_ptr &node) { - GELOGI("Start to save data %s message", node->GetName().c_str()); if (node != nullptr) { auto input_op_desc = node->GetOpDesc(); if (input_op_desc == nullptr) { @@ -180,7 +179,6 @@ void DataDumper::SaveDumpInput(const std::shared_ptr &node) { {op_desc->GetName(), {input_op_desc, dst_in_data_anchor->GetIdx(), out_data_anchor->GetIdx()}}); } } - GELOGI("Save data message successfully"); } } @@ -218,7 +216,7 @@ void DataDumper::SaveDumpOpInfo(const RuntimeParam &model_param, const OpDescPtr GELOGW("Get input size failed"); return; } - GELOGI("Save dump op info, the input size is %ld", input_size); + GELOGD("Save dump op info, the input size is %ld", input_size); op_desc_info.input_size.emplace_back(input_size); } for (size_t j = 0; j < op->GetOutputsSize(); ++j) { @@ -234,7 +232,7 @@ void DataDumper::SaveDumpOpInfo(const RuntimeParam &model_param, const OpDescPtr GELOGW("Get input size failed"); return; } - GELOGI("Save dump op info, the output size is %ld", output_size); + GELOGD("Save dump op info, the output size is %ld", output_size); op_desc_info.output_size.emplace_back(output_size); } op_desc_info.input_addrs = ModelUtils::GetInputDataAddrs(model_param, op); @@ -301,22 +299,16 @@ static void SetOpMappingLoopAddr(uintptr_t step_id, uintptr_t loop_per_iter, uin if (step_id != 0) { GELOGI("step_id exists."); op_mapping_info.set_step_id_addr(static_cast(step_id)); - } else { - GELOGI("step_id is null."); } if (loop_per_iter != 0) { GELOGI("loop_per_iter exists."); op_mapping_info.set_iterations_per_loop_addr(static_cast(loop_per_iter)); - } else { - GELOGI("loop_per_iter is null."); } if (loop_cond != 0) { GELOGI("loop_cond exists."); op_mapping_info.set_loop_cond_addr(static_cast(loop_cond)); - } else { - GELOGI("loop_cond is null."); } } @@ -672,7 +664,7 @@ Status DataDumper::LoadDumpInfo() { PrintCheckLog(dump_list_key); if (op_list_.empty()) { - GELOGW("op_list_ is empty"); + GELOGD("op_list_ is empty"); } aicpu::dump::OpMappingInfo op_mapping_info; @@ -684,8 +676,6 @@ Status DataDumper::LoadDumpInfo() { op_mapping_info.set_flag(kAicpuLoadFlag); op_mapping_info.set_dump_step(dump_properties_.GetDumpStep()); SetOpMappingLoopAddr(global_step_, loop_per_iter_, loop_cond_, op_mapping_info); - GELOGI("Dump step is %s and dump path is %s dump model is %s in load dump info", - dump_properties_.GetDumpStep().c_str(), dump_path.c_str(), dump_list_key.c_str()); auto ret = BuildTaskInfo(op_mapping_info); if (ret != SUCCESS) { GELOGE(ret, "Build task info failed"); @@ -812,7 +802,6 @@ void DataDumper::SetOpDebugIdToAicpu(uint32_t task_id, uint32_t stream_id, void Status DataDumper::UnloadDumpInfo() { if (!load_flag_) { - GELOGI("No need to UnloadDumpInfo."); load_flag_ = false; return SUCCESS; } @@ -838,7 +827,6 @@ Status DataDumper::UnloadDumpInfo() { void DataDumper::PrintCheckLog(string &dump_list_key) { std::set model_list = dump_properties_.GetAllDumpModel(); if (model_list.empty()) { - GELOGI("No model need dump."); return; } diff --git a/ge/graph/load/new_model_manager/davinci_model.cc b/ge/graph/load/new_model_manager/davinci_model.cc index 10e1c8b3..0344fd7e 100755 --- a/ge/graph/load/new_model_manager/davinci_model.cc +++ b/ge/graph/load/new_model_manager/davinci_model.cc @@ -17,11 +17,7 @@ #include "graph/load/new_model_manager/davinci_model.h" #include -#include #include -#include -#include -#include #include #include #include @@ -206,7 +202,6 @@ DavinciModel::~DavinciModel() { OpDebugUnRegister(); - GELOGI("do ReleaseTask"); ReleaseTask(); CleanTbeHandle(); @@ -337,7 +332,6 @@ Status DavinciModel::InitModelMem(void *dev_ptr, size_t mem_size, void *weight_p GELOGI("[IMAS]InitModelMem graph_%u MallocMemory type[W] memaddr[%p] mem_size[%zu]", runtime_param_.graph_id, weights_mem_base_, weights_size); GE_CHK_RT_RET(rtMemcpy(weights_mem_base_, weights_size, weights.GetData(), weights_size, RT_MEMCPY_HOST_TO_DEVICE)); - GELOGI("copy weights data to device"); } GE_CHK_STATUS_RET(InitVariableMem(), "Init variable memory failed."); @@ -488,7 +482,7 @@ Status DavinciModel::SetTSDevice() { int64_t value = 0; bool ret = ge::AttrUtils::GetInt(ge_model_, ATTR_MODEL_CORE_TYPE, value); uint32_t core_type = ret ? static_cast(value) : 0; - GELOGI("SetTSDevice: %u", core_type); + GELOGD("SetTSDevice: %u", core_type); rtError_t rt_ret = rtSetTSDevice(core_type); if (rt_ret != RT_ERROR_NONE) { GELOGE(RT_FAILED, "SetTSDevice failed, ret: 0x%X", rt_ret); @@ -543,7 +537,6 @@ Status DavinciModel::OpDebugRegister() { } void DavinciModel::OpDebugUnRegister() { - GELOGI("OpDebugUnRegister, is_op_debug_reg_ = %d", is_op_debug_reg_); if (is_op_debug_reg_) { debug_reg_mutex_.unlock(); rtError_t rt_ret = RT_ERROR_NONE; @@ -648,7 +641,7 @@ Status DavinciModel::Init(void *dev_ptr, size_t mem_size, void *weight_ptr, size GE_CHK_STATUS_RET(TransVarDataUtils::CopyVarData(compute_graph, session_id_, device_id_), "copy var data failed."); GE_TIMESTAMP_START(InitModelMem); - GELOGI("Known node is %d", known_node_); + GELOGD("Known node is %d", known_node_); if (!known_node_) { GE_CHK_STATUS_RET_NOLOG(InitModelMem(dev_ptr, mem_size, weight_ptr, weight_size)); data_inputer_ = new (std::nothrow) DataInputer(); @@ -708,7 +701,6 @@ Status DavinciModel::Init(void *dev_ptr, size_t mem_size, void *weight_ptr, size } Shrink(); - GELOGI("Davinci model init success."); return ret; } @@ -968,18 +960,21 @@ Status DavinciModel::InitDataOp(const NodePtr &node, uint32_t &data_op_index, ma const vector output_size_list = ModelUtils::GetOutputSize(op_desc); const vector virtual_addr_list = ModelUtils::GetOutputDataAddrs(runtime_param_, op_desc); const vector output_offset_list = op_desc->GetOutputOffset(); - if (output_offset_list.size() != virtual_addr_list.size()) { - GELOGE(PARAM_INVALID, "virtual_addr size:%zu should be equal to offset size:%zu.", virtual_addr_list.size(), - output_offset_list.size()); + if (output_size_list.empty() || virtual_addr_list.empty() || (output_size_list.size() != virtual_addr_list.size()) || + (output_offset_list.size() != virtual_addr_list.size())) { + GELOGE(PARAM_INVALID, "Data[%s] init failed: output size is %zu, virtual_addr size is %zu, offset size is %zu.", + op_desc->GetName().c_str(), output_size_list.size(), virtual_addr_list.size(), output_offset_list.size()); return PARAM_INVALID; } auto data_index = data_op_index; if (AttrUtils::GetInt(op_desc, ATTR_NAME_INDEX, data_index)) { - GELOGI("ge_train: get new index %u, old %u", data_index, data_op_index); + GELOGD("ge_train: get new index %u, old %u", data_index, data_op_index); } bool fusion_flag = false; ZeroCopyOffset zero_copy_offset; - Status ret = zero_copy_offset.InitInputDataInfo(output_size_list, virtual_addr_list, op_desc, fusion_flag); + int64_t data_size = output_size_list[kDataIndex]; + void *virtual_addr = virtual_addr_list[kDataIndex]; + Status ret = zero_copy_offset.InitInputDataInfo(data_size, virtual_addr, op_desc, fusion_flag); if (ret != SUCCESS) { GELOGE(PARAM_INVALID, "InitDataInfo of input_info %s failed.", op_desc->GetName().c_str()); return PARAM_INVALID; @@ -996,7 +991,6 @@ Status DavinciModel::InitDataOp(const NodePtr &node, uint32_t &data_op_index, ma new_input_outside_addrs_[addr] = zero_copy_offset; } - GELOGI("SetInputOutsideAddr success."); data_op_index++; if (InitInputZeroCopyNodes(node) != SUCCESS) { GELOGE(PARAM_INVALID, "Input zero copy nodes init failed!"); @@ -1131,7 +1125,6 @@ Status DavinciModel::InitNetOutput(const NodePtr &node) { DisableZeroCopy(real_addr); real_virtual_addrs_.insert(real_addr); } - GELOGI("SetOutputOutsideAddr success."); } GE_IF_BOOL_EXEC(InitOutputZeroCopyNodes(node) != SUCCESS, @@ -1147,8 +1140,6 @@ Status DavinciModel::InitNetOutput(const NodePtr &node) { GE_IF_BOOL_EXEC(GetGearAndRealOutShapeInfo(input_count, op_desc) != SUCCESS, GELOGE(PARAM_INVALID, "Failed to get gear and real out shape info."); return PARAM_INVALID;); } - - GELOGI("DavinciModel::InitNetoutput success."); return SUCCESS; } @@ -1459,7 +1450,7 @@ Status DavinciModel::InitLabelSet(const OpDescPtr &op_desc) { return INTERNAL_ERROR; } if (label_index >= LabelNum()) { - GELOGE(INTERNAL_ERROR, "InitLabelSet: label index: %u >= label size: %zu.", label_index, LabelNum()); + GELOGE(INTERNAL_ERROR, "InitLabelSet: label index: %u >= label size: %u.", label_index, LabelNum()); return INTERNAL_ERROR; } if (label_id_indication_.count(label_index) > 0) { @@ -1968,7 +1959,6 @@ void DavinciModel::GetModelAttr(std::vector &dynamic_output_shape_i if (op->GetType() != NETOUTPUT) { continue; } - GELOGI("Start to get dynamic output dims attr"); if (!AttrUtils::GetListStr(op, ATTR_NAME_DYNAMIC_OUTPUT_DIMS, dynamic_output_shape_info)) { GELOGD("Can not get dynamic output dims attr"); } @@ -2124,7 +2114,7 @@ void DavinciModel::CreateOutput(uint32_t index, OpDescPtr &op_desc, InputOutputD } Status DavinciModel::GetOutputDescInfo(vector &output_desc, std::vector &formats) { - GELOGI("Output node size: %zu", output_op_list_.size()); + GELOGD("Output node size: %zu", output_op_list_.size()); for (size_t i = 0; i < output_op_list_.size(); i++) { auto &op_desc = output_op_list_[i]; uint32_t out_size = static_cast(op_desc->GetInputsSize()); @@ -2187,7 +2177,7 @@ Status DavinciModel::CopyInputData(const InputData &input_data, bool device_data void *mem_addr = data.second.GetBasicAddr(); void *data_buf_addr = reinterpret_cast(reinterpret_cast(data_buf.data)); uint64_t data_buf_length = data_buf.length; - GELOGI("[IMAS]CopyPlainData memcpy graph_%lu type[F] input[%lu] dst[%p] src[%p] mem_size[%lu] datasize[%lu]", + GELOGI("[IMAS]CopyPlainData memcpy graph_%u type[F] input[%u] dst[%p] src[%p] mem_size[%lu] datasize[%lu]", runtime_param_.graph_id, data.first, mem_addr, data_buf_addr, data_size, data_buf_length); GE_CHK_RT_RET(rtMemcpy(mem_addr, data_size, data_buf_addr, data_buf_length, kind)); } @@ -2235,8 +2225,6 @@ Status DavinciModel::SinkModelProfile() { Msprof::Engine::Reporter *reporter = PluginImpl::GetPluginReporter(); GE_IF_BOOL_EXEC(reporter == nullptr, GELOGI("Profiling report is nullptr!"); return SUCCESS); - GELOGI("Start collect model load profiling data."); - Msprof::Engine::ReporterData reporter_data{}; // report model data tag name std::string tag_name; @@ -2294,7 +2282,6 @@ Status DavinciModel::SinkModelProfile() { uint32_t op_num = fusion_op_info->original_op_names.size(); uint32_t task_id = task->GetTaskID(); if (op_num > 0) { - GELOGI("task.id = %u, opNum = %u", task_id, op_num); op_id_map.insert(std::make_pair(fusion_op_info->op_index, task_id)); } } @@ -2552,21 +2539,23 @@ Status DavinciModel::CopyOutputData(uint32_t data_id, OutputData &output_data, r if (is_dynamic_) { GELOGI("No need to check output data size."); } else if (buffer.length < mem_size) { - GELOGE(FAILED, "Tensor data size=%lu, buffer size=%u", mem_size, buffer.length); + GELOGE(FAILED, "Tensor data size=%lu, buffer size=%lu", mem_size, buffer.length); return FAILED; } else if (buffer.length > mem_size) { - GELOGW("Tensor data size=%lu, buffer size=%u", mem_size, buffer.length); + GELOGW("Tensor data size=%lu, buffer size=%lu", mem_size, buffer.length); } int64_t data_size = output.second.GetDataSize(); if (is_online_infer_dynamic_) { - auto gear_and_real_out_size_info = merge_nodes_gear_and_real_out_size_info_[idx]; - data_size = gear_and_real_out_size_info[cur_dynamic_dims_]; + if (merge_nodes_gear_and_real_out_size_info_.find(idx) != merge_nodes_gear_and_real_out_size_info_.end()) { + auto gear_and_real_out_size_info = merge_nodes_gear_and_real_out_size_info_[idx]; + data_size = gear_and_real_out_size_info[cur_dynamic_dims_]; + } } uint64_t buffer_length = buffer.length; void *buffer_addr = reinterpret_cast(reinterpret_cast(buffer.data)); - GELOGI("[IMAS]CopyPlainData memcpy graph_%u type[F] output[%u] memaddr[%p] mem_size[%ld] datasize[%u]", + GELOGI("[IMAS]CopyPlainData memcpy graph_%u type[F] output[%u] memaddr[%p] mem_size[%lu] datasize[%lu]", runtime_param_.graph_id, output.first, output.second.GetBasicAddr(), data_size, buffer_length); GE_CHK_RT_RET(rtMemcpy(buffer_addr, buffer_length, output.second.GetBasicAddr(), data_size, kind)); idx++; @@ -2598,11 +2587,13 @@ Status DavinciModel::GenOutputTensorInfo(const OpDescPtr &op_desc, uint32_t data return ret); std::vector output_shape = input_desc->GetShape().GetDims(); if (is_online_infer_dynamic_) { - auto gear_and_real_out_size_info = merge_nodes_gear_and_real_out_size_info_[i]; - size = gear_and_real_out_size_info[cur_dynamic_dims_]; - auto gear_and_real_out_shape_info = merge_nodes_gear_and_real_out_shape_info_[i]; - output_shape = gear_and_real_out_shape_info[cur_dynamic_dims_]; - is_dynamic_ = true; + if (merge_nodes_gear_and_real_out_size_info_.find(i) != merge_nodes_gear_and_real_out_size_info_.end()) { + auto gear_and_real_out_size_info = merge_nodes_gear_and_real_out_size_info_[i]; + size = gear_and_real_out_size_info[cur_dynamic_dims_]; + auto gear_and_real_out_shape_info = merge_nodes_gear_and_real_out_shape_info_[i]; + output_shape = gear_and_real_out_shape_info[cur_dynamic_dims_]; + is_dynamic_ = true; + } } GELOGI("Output size is %ld, output shape is %s.", size, formats::JoinToString(output_shape).c_str()); out_buffer_size_vec.push_back(size); @@ -2759,16 +2750,6 @@ void *DavinciModel::Run(DavinciModel *model) { InputData current_data = data_wrapper->GetInput(); GELOGI("Model thread Run begin, model id:%u, data index:%u.", model_id, current_data.index); - if (model->is_online_infer_dynamic_ && !model->is_getnext_sink_dynamic_) { - model->cur_dynamic_dims_.clear(); - GE_IF_BOOL_EXEC(current_data.blobs.empty(), break); - auto shape_data_buffer_data = current_data.blobs.back().data; - auto shape_data_buffer_length = current_data.blobs.back().length; - model->cur_dynamic_dims_.assign(reinterpret_cast(shape_data_buffer_data), - reinterpret_cast(shape_data_buffer_data) + - shape_data_buffer_length / sizeof(int64_t)); - GELOGD("Data: cur dynamic dims is %s", formats::JoinToString(model->cur_dynamic_dims_).c_str()); - } GE_TIMESTAMP_START(Model_SyncVarData); ret = model->SyncVarData(); GE_CHK_BOOL_TRUE_EXEC_WITH_LOG( @@ -2785,6 +2766,18 @@ void *DavinciModel::Run(DavinciModel *model) { ret != SUCCESS, (void)model->ReturnResult(current_data.index, false, false, data_wrapper->GetOutput()); CsaInteract::GetInstance().StoreInternalErrorCode(ret, ERROR_MODULE_FMK, JOBSUBSTATE_GRAPH_EXEC); continue, "Copy input data to model failed."); // [No need to check value] + if (model->is_online_infer_dynamic_ && !model->is_getnext_sink_dynamic_) { + model->cur_dynamic_dims_.clear(); + GE_IF_BOOL_EXEC(current_data.blobs.empty(), break); + auto shape_data_buffer_data = current_data.blobs.back().data; + auto shape_data_buffer_length = current_data.blobs.back().length; + model->cur_dynamic_dims_.assign(reinterpret_cast(shape_data_buffer_data), + reinterpret_cast(shape_data_buffer_data) + + shape_data_buffer_length / sizeof(int64_t)); + GELOGD("Data: cur dynamic dims is %s", formats::JoinToString(model->cur_dynamic_dims_).c_str()); + delete[] (int64_t *)current_data.blobs.back().data; + current_data.blobs.pop_back(); + } GE_IF_BOOL_EXEC(ProfilingManager::Instance().ProfilingModelExecuteOn(), model->SetProfileTime(MODEL_PRE_PROC_END)); GE_IF_BOOL_EXEC(ProfilingManager::Instance().ProfilingModelExecuteOn(), model->SetProfileTime(MODEL_INFER_START)); if (ProfilingManager::Instance().ProfilingOpTraceOn()) { @@ -2982,7 +2975,7 @@ void DavinciModel::UnbindTaskSinkStream() { Status DavinciModel::CreateKnownZeroCopyMap(const vector &inputs, const vector &outputs) { GELOGI("DavinciModel::CreateKnownZeroCopyMap in."); if (inputs.size() > data_op_list_.size()) { - GELOGE(FAILED, "input data addr %u should less than input op number %u.", inputs.size(), data_op_list_.size()); + GELOGE(FAILED, "input data addr %zu should less than input op number %zu.", inputs.size(), data_op_list_.size()); return FAILED; } // remove zero copy addr in last iteration @@ -2991,16 +2984,16 @@ Status DavinciModel::CreateKnownZeroCopyMap(const vector &inputs, const for (size_t i = 0; i < inputs.size(); ++i) { const vector addr_list = ModelUtils::GetOutputDataAddrs(runtime_param_, data_op_list_[i]); knonw_input_data_info_[addr_list[kDataIndex]] = inputs[i]; - GELOGI("DavinciModel::CreateKnownZeroCopyMap input %d,v addr %p,p addr %p .", i, addr_list[kDataIndex], inputs[i]); + GELOGI("DavinciModel::CreateKnownZeroCopyMap input %zu,v addr %p,p addr %p .", i, addr_list[kDataIndex], inputs[i]); } if (output_op_list_.size() < kOutputNum) { - GELOGW("output op num in graph is %u.", output_op_list_.size()); + GELOGW("output op num in graph is %zu.", output_op_list_.size()); return SUCCESS; } const vector addr_list = ModelUtils::GetInputDataAddrs(runtime_param_, output_op_list_[kDataIndex]); for (size_t i = 0; i < addr_list.size() && i < outputs.size(); ++i) { knonw_output_data_info_[addr_list[i]] = outputs[i]; - GELOGI("DavinciModel::CreateKnownZeroCopyMap output %d,v addr %p,p addr %p .", i, addr_list[i], outputs[i]); + GELOGI("DavinciModel::CreateKnownZeroCopyMap output %zu,v addr %p,p addr %p .", i, addr_list[i], outputs[i]); } GELOGI("DavinciModel::CreateKnownZeroCopyMap success."); return SUCCESS; @@ -3010,13 +3003,13 @@ Status DavinciModel::UpdateKnownZeroCopyAddr() { for (size_t i = 0; i < total_io_addrs_.size(); ++i) { auto it_in = knonw_input_data_info_.find(total_io_addrs_[i]); if (it_in != knonw_input_data_info_.end()) { - GELOGI("DavinciModel::UpdateKnownZeroCopyAddr input %d,v addr %p,p addr %p .", i, total_io_addrs_[i], + GELOGI("DavinciModel::UpdateKnownZeroCopyAddr input %zu,v addr %p,p addr %p .", i, total_io_addrs_[i], knonw_input_data_info_.at(total_io_addrs_[i])); total_io_addrs_[i] = knonw_input_data_info_.at(total_io_addrs_[i]); } auto it_out = knonw_output_data_info_.find(total_io_addrs_[i]); if (it_out != knonw_output_data_info_.end()) { - GELOGI("DavinciModel::UpdateKnownZeroCopyAddr output %d,v addr %p,p addr %p .", i, total_io_addrs_[i], + GELOGI("DavinciModel::UpdateKnownZeroCopyAddr output %zu,v addr %p,p addr %p .", i, total_io_addrs_[i], knonw_output_data_info_.at(total_io_addrs_[i])); total_io_addrs_[i] = knonw_output_data_info_.at(total_io_addrs_[i]); } @@ -3037,7 +3030,7 @@ Status DavinciModel::UpdateKnownNodeArgs(const vector &inputs, const vec if (task != nullptr) { Status ret = task->UpdateArgs(); if (ret != SUCCESS) { - GELOGE(FAILED, "task %d created by davinci model is nullptr.", task_index); + GELOGE(FAILED, "task %zu created by davinci model is nullptr.", task_index); return FAILED; } } @@ -3066,7 +3059,7 @@ Status DavinciModel::UpdateKnownNodeArgs(const vector &inputs, const vec } Status DavinciModel::InitTaskInfo(domi::ModelTaskDef &model_task_def) { - GELOGI("InitTaskInfo in, task size %zu", model_task_def.task().size()); + GELOGI("InitTaskInfo in, task size %d", model_task_def.task().size()); task_list_.resize(model_task_def.task_size()); for (int i = 0; i < model_task_def.task_size(); ++i) { // dynamic shape will create task_list_ before @@ -3142,14 +3135,14 @@ Status DavinciModel::DistributeTask() { task_desc_info_.clear(); bool flag = GetL1FusionEnableOption(); - char *skt_enable_env = std::getenv("SKT_ENABLE"); - int64_t env_flag = (skt_enable_env != nullptr) ? std::strtol(skt_enable_env, nullptr, kDecimal) : 0; + char skt_enable_env[MMPA_MAX_PATH] = { 0x00 }; + INT32 res = mmGetEnv("SKT_ENABLE", skt_enable_env, MMPA_MAX_PATH); + int64_t env_flag = (res == EN_OK) ? std::strtol(skt_enable_env, nullptr, kDecimal) : 0; if (env_flag != 0) { flag = true; } const auto &model_task_def = ge_model_->GetModelTaskDefPtr(); - GELOGI("there are %zu task need to save.", task_list_.size()); for (size_t task_index = 0; task_index < task_list_.size(); ++task_index) { auto &task = task_list_.at(task_index); GE_CHK_STATUS_RET(task->Distribute(), "Task[%zu] distribute fail", task_index); @@ -3331,7 +3324,7 @@ bool DavinciModel::CheckInputAndModelSize(const int64_t &input_size, const int64 if (input_size > op_size) { GELOGW( - "Input size [%u] is bigger than om size need [%u], " + "Input size [%ld] is bigger than om size need [%ld], " "MAY cause inference result ERROR, please check model input", input_size, op_size); } @@ -3413,7 +3406,7 @@ Status DavinciModel::UpdateIoTaskArgs(const std::map & for (const auto &data : data_info) { if (data.first >= blobs.size()) { // check data index. - GELOGE(FAILED, "Verify %s data num failed: can not find No.%zu data, because user only feeds %zu", + GELOGE(FAILED, "Verify %s data num failed: can not find No.%u data, because user only feeds %zu", input_or_output.c_str(), data.first, blobs.size()); return FAILED; } @@ -3522,7 +3515,7 @@ Status DavinciModel::InitConstant(const OpDescPtr &op_desc) { GeTensor *tensor = const_cast(v_weights[0].get()); GE_IF_BOOL_EXEC(static_cast(v_output_size[0]) < tensor->GetData().size(), - GELOGE(PARAM_INVALID, "output size:%u less than weight data size:%zu", v_output_size[0], + GELOGE(PARAM_INVALID, "output size:%ld less than weight data size:%zu", v_output_size[0], tensor->GetData().size()); return PARAM_INVALID;); @@ -3546,12 +3539,12 @@ Status DavinciModel::InitConstant(const OpDescPtr &op_desc) { uint64_t offset = static_cast(elem_num * kBytes); uint64_t hbm_raw_data_base_addr = - reinterpret_cast(reinterpret_cast(v_output_addr[0])) + offset; + static_cast(reinterpret_cast(v_output_addr[0])) + offset; for (int64_t i = elem_num - 1; i >= 0; --i) { buff[i] = hbm_raw_data_base_addr + (buff[i] - buff[0]); } } - GELOGI("[IMAS]InitConstant memcpy graph_%u type[V] name[%s] output[%d] memaddr[%p] mem_size[%u] datasize[%zu]", + GELOGI("[IMAS]InitConstant memcpy graph_%u type[V] name[%s] output[%d] memaddr[%p] mem_size[%lu] datasize[%zu]", runtime_param_.graph_id, op_desc->GetName().c_str(), 0, v_output_addr[0], v_output_size[0], tensor->GetData().size()); GE_CHK_RT_RET(rtMemcpy(v_output_addr[0], v_output_size[0], tensor->GetData().data(), tensor->GetData().size(), @@ -3582,12 +3575,12 @@ Status DavinciModel::InitTbeHandle(const OpDescPtr &op_desc) { if (rtQueryFunctionRegistered(bin_file_key) != RT_ERROR_NONE) { void *bin_handle = nullptr; if (!kernel_store.FindTBEHandle(bin_file_key, bin_handle)) { - GELOGI("TBE: can't find the kernel_name[%s] in HandleMap", bin_file_key); + GELOGD("TBE: can't find the kernel_name[%s] in HandleMap", bin_file_key); rtDevBinary_t binary; std::string json_string; GE_IF_BOOL_EXEC(AttrUtils::GetStr(op_desc, TVM_ATTR_NAME_MAGIC, json_string), - GELOGI("Get original type of session_graph_id.")); + GELOGD("Get original type of session_graph_id.")); if (json_string == "RT_DEV_BINARY_MAGIC_ELF_AICPU") { binary.magic = RT_DEV_BINARY_MAGIC_ELF_AICPU; } else if (json_string == "RT_DEV_BINARY_MAGIC_ELF") { @@ -3603,13 +3596,13 @@ Status DavinciModel::InitTbeHandle(const OpDescPtr &op_desc) { binary.data = tbe_kernel->GetBinData(); binary.length = tbe_kernel->GetBinDataSize(); - GELOGI("TBE: binary.length: %lu", binary.length); + GELOGD("TBE: binary.length: %lu", binary.length); GE_CHK_RT_RET(rtDevBinaryRegister(&binary, &bin_handle)); std::string meta_data; GE_IF_BOOL_EXEC(AttrUtils::GetStr(op_desc, TVM_ATTR_NAME_METADATA, meta_data), GELOGI("Get original type of json_string")); - GELOGI("TBE: meta data: %s", meta_data.empty() ? "null" : meta_data.c_str()); + GELOGD("TBE: meta data: %s", meta_data.empty() ? "null" : meta_data.c_str()); GE_IF_BOOL_EXEC(!meta_data.empty(), GE_CHK_RT_RET(rtMetadataRegister(bin_handle, meta_data.c_str()))); kernel_store.StoreTBEHandle(bin_file_key, bin_handle, tbe_kernel); @@ -3620,8 +3613,7 @@ Status DavinciModel::InitTbeHandle(const OpDescPtr &op_desc) { std::string kernel_name; GE_IF_BOOL_EXEC(AttrUtils::GetStr(op_desc, op_desc->GetName() + "_kernelname", kernel_name), - GELOGI("Get original type of kernel_name")); - GELOGI("TBE: binfile_key=%s, kernel_name=%s", bin_file_key, kernel_name.c_str()); + GELOGD("Get original type of kernel_name")); GE_CHK_RT_RET(rtFunctionRegister(bin_handle, bin_file_key, bin_file_key, kernel_name.c_str(), 0)); used_tbe_handle_map_[bin_file_key] = 1; // Init used num to 1. return SUCCESS; @@ -3816,7 +3808,7 @@ Status DavinciModel::InitModelStream(rtStream_t stream) { Status DavinciModel::NnExecute(rtStream_t stream, bool async_mode, const InputData &input_data, OutputData &output_data) { is_async_mode_ = async_mode; - GELOGI("Model Run begin, model id:%u, data index:%u, flag:%d.", model_id_, input_data.index, is_async_mode_); + GELOGD("Model Run begin, model id:%u, data index:%u, flag:%d.", model_id_, input_data.index, is_async_mode_); GE_CHK_STATUS_RET(InitModelStream(stream), "Init model stream failed."); is_dynamic_ = input_data.is_dynamic_batch; if (!is_dynamic_) { @@ -3828,7 +3820,7 @@ Status DavinciModel::NnExecute(rtStream_t stream, bool async_mode, const InputDa GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(ret != SUCCESS, return ret, "Copy input data to model failed. model id: %u", model_id_); - GELOGI("current_data.index=%u", input_data.index); + GELOGD("current_data.index=%u", input_data.index); GE_IF_BOOL_EXEC(ProfilingManager::Instance().ProfilingModelExecuteOn(), SetProfileTime(MODEL_PRE_PROC_END)); if (!task_list_.empty()) { @@ -3837,7 +3829,7 @@ Status DavinciModel::NnExecute(rtStream_t stream, bool async_mode, const InputDa rtError_t rt_ret = rtModelExecute(rt_model_handle_, rt_model_stream_, 0); GE_CHK_RT_EXEC(rt_ret, return RT_ERROR_TO_GE_STATUS(rt_ret)); GE_IF_BOOL_EXEC(ProfilingManager::Instance().ProfilingModelExecuteOn(), SetProfileTime(MODEL_INFER_END)); - GELOGI("rtModelExecute end"); + GELOGD("rtModelExecute end"); } if (!is_async_mode_) { @@ -3849,7 +3841,7 @@ Status DavinciModel::NnExecute(rtStream_t stream, bool async_mode, const InputDa // report model time data GE_IF_BOOL_EXEC(ProfilingManager::Instance().ProfilingModelExecuteOn(), (void)SinkTimeProfile(input_data)); - GELOGI("Model run end, model id:%u", model_id_); + GELOGD("Model run end, model id:%u", model_id_); return SUCCESS; } @@ -3906,7 +3898,9 @@ Status DavinciModel::InitEntryTask() { uint8_t *DavinciModel::MallocFeatureMapMem(size_t data_size) { uint8_t *mem_base = nullptr; const string purpose("feature map,used for op input and output."); - if (std::getenv(kEnvGeuseStaticMemory) != nullptr) { + char ge_static_mem_env[MMPA_MAX_PATH] = { 0x00 }; + INT32 res = mmGetEnv(kEnvGeuseStaticMemory, ge_static_mem_env, MMPA_MAX_PATH); + if (res == EN_OK) { data_size = static_cast(VarManager::Instance(session_id_)->GetGraphMemoryMaxSize()); string memory_key = std::to_string(0) + "_f"; mem_base = MemManager::Instance(RT_MEMORY_HBM)->MallocMemory(purpose, memory_key, data_size, GetDeviceId()); @@ -3936,7 +3930,9 @@ uint8_t *DavinciModel::MallocP2PMem(size_t p2p_data_size) { uint8_t *DavinciModel::MallocWeightsMem(size_t weights_size) { uint8_t *weights_mem_base = nullptr; const string purpose("weights memory in inference network."); - if (std::getenv(kEnvGeuseStaticMemory) != nullptr) { + char ge_static_mem_env[MMPA_MAX_PATH] = { 0x00 }; + INT32 res = mmGetEnv(kEnvGeuseStaticMemory, ge_static_mem_env, MMPA_MAX_PATH); + if (res == EN_OK) { string weight_memory_key = std::to_string(0) + "_w"; weights_mem_base = MemManager::Instance(RT_MEMORY_HBM)->MallocMemory(purpose, weight_memory_key, weights_size, GetDeviceId()); @@ -3947,7 +3943,9 @@ uint8_t *DavinciModel::MallocWeightsMem(size_t weights_size) { } void DavinciModel::FreeFeatureMapMem() { - if (std::getenv(kEnvGeuseStaticMemory) != nullptr && is_inner_mem_base_) { + char ge_static_mem_env[MMPA_MAX_PATH] = { 0x00 }; + INT32 res = mmGetEnv(kEnvGeuseStaticMemory, ge_static_mem_env, MMPA_MAX_PATH); + if (res == EN_OK && is_inner_mem_base_) { string weight_memory_key = std::to_string(0) + "_f"; if (MemManager::Instance(RT_MEMORY_HBM)->GetMemoryAddr(weight_memory_key) != nullptr) { GE_CHK_STATUS(MemManager::Instance(RT_MEMORY_HBM)->FreeMemory(weight_memory_key, GetDeviceId()), @@ -3979,7 +3977,9 @@ void DavinciModel::FreeP2PMem() { } void DavinciModel::FreeWeightsMem() { - if (std::getenv(kEnvGeuseStaticMemory) != nullptr) { + char ge_static_mem_env[MMPA_MAX_PATH] = { 0x00 }; + INT32 res = mmGetEnv(kEnvGeuseStaticMemory, ge_static_mem_env, MMPA_MAX_PATH); + if (res == EN_OK) { string memory_key = std::to_string(0) + "_w"; if (MemManager::Instance(RT_MEMORY_HBM)->GetMemoryAddr(memory_key) != nullptr) { GE_CHK_STATUS(MemManager::Instance(RT_MEMORY_HBM)->FreeMemory(memory_key, GetDeviceId()), @@ -3995,7 +3995,6 @@ void DavinciModel::FreeWeightsMem() { } Status DavinciModel::TransAllVarData(ComputeGraphPtr &graph, uint32_t graph_id) { - GELOGI("TransAllVarData start: session_id:%lu, graph_id: %u.", session_id_, graph_id); rtContext_t ctx = nullptr; rtError_t rt_ret = rtCtxGetCurrent(&ctx); if (rt_ret != RT_ERROR_NONE) { @@ -4016,13 +4015,10 @@ Status DavinciModel::TransAllVarData(ComputeGraphPtr &graph, uint32_t graph_id) GE_CHK_STATUS_RET_NOLOG( TransVarDataUtils::TransAllVarData(variable_node_list, session_id_, ctx, graph_id, kThreadNum)); - - GELOGI("TransAllVarData success."); return SUCCESS; } void DavinciModel::SetDataDumperArgs(const ComputeGraphPtr &compute_graph) { - GELOGI("set data dumper args, name: %s, id: %u.", name_.c_str(), model_id_); data_dumper_.SetModelName(name_); data_dumper_.SetModelId(model_id_); data_dumper_.SetOmName(om_name_); @@ -4048,15 +4044,13 @@ void DavinciModel::SetDataDumperArgs(const ComputeGraphPtr &compute_graph) { } return v_output_addr[0]; } - GELOGW("op is null."); + GELOGD("op is null."); return nullptr; }; data_dumper_.SetLoopAddr(get_var_addr(GetVariableOp(NODE_NAME_GLOBAL_STEP), runtime_param_), get_var_addr(GetVariableOp(NODE_NAME_FLOWCTRL_LOOP_PER_ITER), runtime_param_), get_var_addr(GetVariableOp(NODE_NAME_FLOWCTRL_LOOP_COND), runtime_param_)); - - GELOGI("SetDataDumperArgs end."); } uint32_t DavinciModel::GetFlowctrlIndex(uint32_t op_index) { @@ -4075,7 +4069,6 @@ void DavinciModel::SaveHcclFollowStream(int64_t main_stream_id, rtStream_t strea } Status DavinciModel::GetComputeGraphInfo(vector &graph_desc_info) { - GELOGI("GetComputeGraphInfo start."); auto &all_op_desc = data_dumper_.GetAllOpDescInfo(); for (auto &op_desc : all_op_desc) { ComputeGraphDescInfo compute_graph_info; @@ -4095,7 +4088,6 @@ Status DavinciModel::GetComputeGraphInfo(vector &graph_des graph_desc_info.emplace_back(compute_graph_info); } - GELOGI("GetComputeGraphInfo end."); return SUCCESS; } @@ -4160,7 +4152,7 @@ Status DavinciModel::GetAllAippInputOutputDims(uint32_t index, std::vector inputs; if (AttrUtils::GetListStr(data_op, ATTR_NAME_AIPP_INPUTS, inputs) && !inputs.empty()) { - GELOGI("GetAllAippInputOutputDims: Data: %s has %u related aippInfo.", data_op->GetName().c_str(), inputs.size()); + GELOGI("GetAllAippInputOutputDims: Data: %s has %zu related aippInfo.", data_op->GetName().c_str(), inputs.size()); for (auto it : inputs) { InputOutputDims input_info; ParseAIPPInfo(it, input_info); @@ -4171,7 +4163,7 @@ Status DavinciModel::GetAllAippInputOutputDims(uint32_t index, std::vectorGetInputDescPtr(kDataIndex)), data_input_size); GELOGD( - "GetAllAippInputOutputDims related Data[%d]: tensor_name is %s, dim_num is %u, tensor_size: %zu, format: " + "GetAllAippInputOutputDims related Data[%d]: tensor_name is %s, dim_num is %zu, tensor_size: %zu, format: " "%s, data_type: %s, shape: %s .", index, data_op->GetName().c_str(), data_input_desc->GetShape().GetDimNum(), data_input_size, TypeUtils::FormatToSerialString(data_input_desc->GetFormat()).c_str(), diff --git a/ge/graph/load/new_model_manager/model_manager.cc b/ge/graph/load/new_model_manager/model_manager.cc index e51a205d..446c512c 100755 --- a/ge/graph/load/new_model_manager/model_manager.cc +++ b/ge/graph/load/new_model_manager/model_manager.cc @@ -202,7 +202,6 @@ void ModelManager::DestroyAicpuSession(uint64_t session_id) { } ge::Status ModelManager::DestroyAicpuSessionForInfer(uint32_t model_id) { - GELOGI("Destroy aicpu session for infer, model id is %u.", model_id); std::lock_guard lock(map_mutex_); auto it = model_map_.find(model_id); if (it == model_map_.end()) { @@ -210,7 +209,6 @@ ge::Status ModelManager::DestroyAicpuSessionForInfer(uint32_t model_id) { return GE_EXEC_MODEL_ID_INVALID; } uint64_t session_id = it->second->GetSessionId(); - GELOGI("Destroy aicpu session for infer, session id is %lu.", session_id); DestroyAicpuSession(session_id); return SUCCESS; } @@ -407,10 +405,6 @@ Status ModelManager::Unload(uint32_t model_id) { } std::lock_guard lock(exeception_infos_mutex_); exception_infos_.clear(); - for (auto addr : shape_data_addrs_[model_id]) { - delete[] addr; - } - shape_data_addrs_.erase(model_id); return SUCCESS; } @@ -475,6 +469,19 @@ Status ModelManager::GetCurDynamicDims(const vector> &user_real_ } } GELOGD("Cur dynamic dims is %s.", formats::JoinToString(cur_dynamic_dims).c_str()); + bool cur_dynamic_dims_valid = false; + std::vector shape_strs = ge::StringUtils::Split(GetLocalOmgContext().dynamic_dims, ';'); + for (auto dynamic_dim : shape_strs) { + if (dynamic_dim == formats::JoinToString(cur_dynamic_dims)) { + cur_dynamic_dims_valid = true; + break; + } + } + if (!cur_dynamic_dims_valid) { + GELOGE(INTERNAL_ERROR, "Cur dynamic dims is %s, not exist in options.", + formats::JoinToString(cur_dynamic_dims).c_str()); + return INTERNAL_ERROR; + } return SUCCESS; } @@ -517,7 +524,6 @@ Status ModelManager::DataInputTensor(uint32_t model_id, const std::vector(data.data)); } } @@ -1019,8 +1025,8 @@ Status ModelManager::GetAippType(uint32_t model_id, uint32_t index, InputAippTyp Status ModelManager::GenSessionId(uint64_t &session_id) { std::lock_guard lock(session_id_create_mutex_); - struct timeval tv; - if (gettimeofday(&tv, nullptr) != 0) { + mmTimeval tv; + if (mmGetTimeOfDay(&tv, nullptr) != 0) { GELOGE(INTERNAL_ERROR, "Failed to get current time."); return INTERNAL_ERROR; } @@ -1037,8 +1043,8 @@ Status ModelManager::GenSessionId(uint64_t &session_id) { Status ModelManager::LoadModelOffline(uint32_t &model_id, const ModelData &model, shared_ptr listener, void *dev_ptr, size_t mem_size, void *weight_ptr, size_t weight_size) { - GE_CHK_BOOL_RET_STATUS(model.key.empty() || access(model.key.c_str(), F_OK) == 0, - ACL_ERROR_GE_EXEC_MODEL_KEY_PATH_INVALID, + GE_CHK_BOOL_RET_STATUS(model.key.empty() || mmAccess2(model.key.c_str(), M_F_OK) == EN_OK, + ACL_ERROR_GE_EXEC_MODEL_KEY_PATH_INVALID, "input key file path %s is invalid, %s", model.key.c_str(), strerror(errno)); GenModelId(&model_id); @@ -1123,7 +1129,7 @@ Status ModelManager::LoadModelOffline(uint32_t &model_id, const ModelData &model Status ModelManager::LoadModelWithQ(uint32_t &model_id, const ModelData &model_data, const std::vector &input_queue_ids, const std::vector &output_queue_ids) { - GE_CHK_BOOL_RET_STATUS(model_data.key.empty() || access(model_data.key.c_str(), F_OK) == 0, + GE_CHK_BOOL_RET_STATUS(model_data.key.empty() || mmAccess2(model_data.key.c_str(), M_F_OK) == EN_OK, ACL_ERROR_GE_EXEC_MODEL_KEY_PATH_INVALID, "input key file path %s is not valid, %s", model_data.key.c_str(), strerror(errno)); @@ -1205,7 +1211,7 @@ Status ModelManager::ExecuteModel(uint32_t model_id, rtStream_t stream, bool asy Status status = davinci_model->NnExecute(stream, async_mode, input_data, output_data); if (status == SUCCESS) { - GELOGI("Execute model %u success.", model_id); + GELOGD("Execute model %u success.", model_id); } return status; @@ -1262,7 +1268,6 @@ Status ModelManager::LoadCustAicpuSo(const OpDescPtr &op_desc, const string &so_ } Status ModelManager::LaunchKernelCustAicpuSo(const string &kernel_name) { - GELOGI("LaunchCustAucpuSo in, kernel name %s", kernel_name.c_str()); std::lock_guard lock(cust_aicpu_mutex_); if (cust_aicpu_so_.size() == 0) return SUCCESS; // get current context diff --git a/ge/graph/load/new_model_manager/model_manager.h b/ge/graph/load/new_model_manager/model_manager.h index 1179de43..b2bb4564 100755 --- a/ge/graph/load/new_model_manager/model_manager.h +++ b/ge/graph/load/new_model_manager/model_manager.h @@ -18,7 +18,6 @@ #define GE_GRAPH_LOAD_NEW_MODEL_MANAGER_MODEL_MANAGER_H_ #include -#include #include #include #include @@ -364,7 +363,6 @@ class FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY ModelManager { std::map> cust_aicpu_so_; static DumpProperties dump_properties_; - std::map> shape_data_addrs_; }; } // namespace ge diff --git a/ge/graph/load/new_model_manager/model_utils.cc b/ge/graph/load/new_model_manager/model_utils.cc index 7ebdad8a..34fb7ff3 100755 --- a/ge/graph/load/new_model_manager/model_utils.cc +++ b/ge/graph/load/new_model_manager/model_utils.cc @@ -337,9 +337,7 @@ vector ModelUtils::GetInputDataAddrs(const RuntimeParam &model_param, Co continue; } - GE_IF_BOOL_EXEC(non_const_index >= v_input_offset.size(), - GELOGW("offsets=%zu, inputs=%zu, index=%zu.", v_input_offset.size(), inputs_size, non_const_index); - break); + GE_IF_BOOL_EXEC(non_const_index >= v_input_offset.size(), break); int64_t input_offset = v_input_offset[non_const_index]; non_const_index++; @@ -356,7 +354,7 @@ vector ModelUtils::GetInputDataAddrs(const RuntimeParam &model_param, Co // feature maps void *mem_addr = nullptr; if (has_mem_type_attr && v_memory_type[i] == RT_MEMORY_L1) { // fusion - mem_addr = reinterpret_cast(reinterpret_cast(input_offset)); + mem_addr = reinterpret_cast(static_cast(input_offset)); v_input_data_addr.push_back(mem_addr); } else if (has_mem_type_attr && v_memory_type[i] == RT_MEMORY_TS_4G) { int64_t tensor_size = 0; @@ -424,7 +422,7 @@ vector ModelUtils::GetOutputDataAddrs(const RuntimeParam &model_param, C // feature maps void *mem_addr = nullptr; if (has_mem_type_attr && v_memory_type[i] == RT_MEMORY_L1) { // fusion - mem_addr = reinterpret_cast(reinterpret_cast(v_output_offset[i])); + mem_addr = reinterpret_cast(static_cast(v_output_offset[i])); v_output_data_addr.push_back(mem_addr); } else if (has_mem_type_attr && v_memory_type[i] == RT_MEMORY_TS_4G) { const GeTensorDescPtr tensor_desc = op_desc->MutableOutputDesc(i); @@ -500,7 +498,7 @@ vector ModelUtils::GetWorkspaceDataAddrs(const RuntimeParam &model_param continue; } if (has_mem_type_attr && v_memory_type[i] == RT_MEMORY_L1) { - v_workspace_data_addr.push_back(reinterpret_cast(reinterpret_cast(v_workspace_offset[i]))); + v_workspace_data_addr.push_back(reinterpret_cast(static_cast(v_workspace_offset[i]))); GELOGI("[IMAS]GetWorkspaceDataAddrs graph_%u type[L1] name[%s], mem_addr[workspace index %zu]:0x%lx", model_param.graph_id, op_desc->GetName().c_str(), i, v_workspace_offset[i]); } else if (v_workspace_bytes[i] == 0) { diff --git a/ge/graph/load/new_model_manager/task_info/kernel_ex_task_info.cc b/ge/graph/load/new_model_manager/task_info/kernel_ex_task_info.cc index 2a295915..e5574e47 100644 --- a/ge/graph/load/new_model_manager/task_info/kernel_ex_task_info.cc +++ b/ge/graph/load/new_model_manager/task_info/kernel_ex_task_info.cc @@ -149,7 +149,7 @@ Status KernelExTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davin return FAILED; } - uint64_t workspace_base_addr = reinterpret_cast(reinterpret_cast(workspace_data_addrs[0])); + uint64_t workspace_base_addr = static_cast(reinterpret_cast(workspace_data_addrs[0])); const vector input_addrs = ModelUtils::GetInputDataAddrs(rts_param, op_desc); const vector output_addrs = ModelUtils::GetOutputDataAddrs(rts_param, op_desc); vector io_addrs; @@ -287,7 +287,7 @@ Status KernelExTaskInfo::CopyTaskInfo(const domi::KernelExDef &kernel_def, const } if (workspace_data_sizes[0] < static_cast(kernel_def.task_info_size())) { - GELOGE(FAILED, "Node:%s workspace size is %zu, task info size is %zu.", op_desc->GetName().c_str(), + GELOGE(FAILED, "Node:%s workspace size is %ld, task info size is %d.", op_desc->GetName().c_str(), workspace_data_sizes[0], kernel_def.task_info_size()); return FAILED; } diff --git a/ge/graph/load/new_model_manager/task_info/kernel_task_info.cc b/ge/graph/load/new_model_manager/task_info/kernel_task_info.cc index b6b218cf..04607c02 100755 --- a/ge/graph/load/new_model_manager/task_info/kernel_task_info.cc +++ b/ge/graph/load/new_model_manager/task_info/kernel_task_info.cc @@ -31,6 +31,7 @@ #include "runtime/kernel.h" #include "super_kernel/super_kernel.h" #include "super_kernel/super_kernel_factory.h" +#include "cce/aicpu_engine_struct.h" namespace { const uint8_t kL2LoadToDdr = 1; @@ -73,7 +74,8 @@ Status KernelTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davinci GELOGD("node[%s] is_n_batch_spilt %d", op_desc_->GetName().c_str(), is_n_batch_spilt_); (void)AttrUtils::GetInt(*op_desc_, ATTR_NAME_FUSION_GROUP_KEY, group_key_); has_group_key_ = (group_key_ != kInvalidGroupKey); - GELOGD("node[%s] has_group_key_ %ld, group key is [%ld]", op_desc_->GetName().c_str(), has_group_key_, group_key_); + GELOGD("node[%s] has_group_key_ %d, group key is [%ld]", op_desc_->GetName().c_str(), has_group_key_, group_key_); + // fusion_op_info vector original_op_names; bool result = AttrUtils::GetListStr(op_desc_, ge::ATTR_NAME_DATA_DUMP_ORIGIN_OP_NAMES, original_op_names); @@ -176,7 +178,7 @@ void KernelTaskInfo::UpdateTaskId() { } task_id_ = task_id; stream_id_ = stream_id; - GELOGI("UpdateTaskId:UpdateTaskId [%u], stream id [%u]:", task_id, stream_id); + GELOGD("UpdateTaskId:UpdateTaskId [%u], stream id [%u]:", task_id, stream_id); } } @@ -216,7 +218,7 @@ Status KernelTaskInfo::SuperKernelLaunch() { rtError_t rt_ret; auto &skt_kernel_list = skt_info_.kernel_list; auto &skt_arg_list = skt_info_.arg_list; - GELOGI("SuperKernelLaunch: Skt_kernel_list size[%d] skt_arg_list[%d]", skt_kernel_list.size(), skt_arg_list.size()); + GELOGI("SuperKernelLaunch: Skt_kernel_list size[%zu] skt_arg_list[%zu]", skt_kernel_list.size(), skt_arg_list.size()); if (skt_kernel_list.size() == kSKTSingleSize && skt_arg_list.size() == kSKTSingleSize) { rt_ret = rtKernelLaunchWithFlag(skt_info_.kernel_list[0], static_cast(skt_info_.last_block_dim), skt_info_.arg_list[0], skt_info_.last_args_size, @@ -367,8 +369,9 @@ Status KernelTaskInfo::Distribute() { GELOGI("Known node %s args addr %p, offset %u.", op_desc_->GetName().c_str(), args_, args_offset_); } rtError_t rt_ret = RT_ERROR_NONE; - char *skt_enable_env = getenv("SKT_ENABLE"); - int64_t env_flag = (skt_enable_env != nullptr) ? strtol(skt_enable_env, nullptr, 10) : 0; + char skt_enable_env[MMPA_MAX_PATH] = { 0x00 }; + INT32 res = mmGetEnv("SKT_ENABLE", skt_enable_env, MMPA_MAX_PATH); + int64_t env_flag = (res == EN_OK) ? strtol(skt_enable_env, nullptr, 10) : 0; bool call_skt = ((env_flag != 0) || is_l1_fusion_enable_); if (kernel_type_ == cce::ccKernelType::AI_CPU || kernel_type_ == cce::ccKernelType::CUST_AI_CPU) { GELOGI("distribute task info kernel_type %d, flag %d", kernel_type_, dump_flag_); @@ -379,7 +382,7 @@ Status KernelTaskInfo::Distribute() { call_save_dump_ = true; } else { /* default: not skt launch */ - GELOGI( + GELOGD( "KernelTaskInfo Distribute Start, sktenable:%d taskid:%u sktid:%u last_sktid:%u stubfunc_name:%s " "stubfunc:%p blockdim:%u stream:%p", call_skt, task_id_, skt_id_, skt_info_.last_task_id, stub_func_name_.c_str(), stub_func_, block_dim_, stream_); @@ -406,7 +409,7 @@ Status KernelTaskInfo::Distribute() { } // set for task_id_ UpdateTaskId(); - GELOGI( + GELOGD( "KernelTaskInfo Distribute Success. sktenable:%d taskid:%d sktid:%d stubfunc_name:%s stubfunc:%p " "blockdim:%d stream:%p", call_skt, task_id_, skt_id_, stub_func_name_.c_str(), stub_func_, block_dim_, stream_); @@ -747,15 +750,15 @@ Status KernelTaskInfo::InitAICPUCustomTask(uint32_t op_index, const domi::Kernel } } *(reinterpret_cast(args + ctx_.argsOffset[0])) = - reinterpret_cast(reinterpret_cast(custom_info_.input_descs)); // arg 0 + static_cast(reinterpret_cast(custom_info_.input_descs)); // arg 0 *(reinterpret_cast(args + ctx_.argsOffset[1])) = - reinterpret_cast(reinterpret_cast(custom_info_.input_addrs)); // arg 1 + static_cast(reinterpret_cast(custom_info_.input_addrs)); // arg 1 *(reinterpret_cast(args + ctx_.argsOffset[2])) = - reinterpret_cast(reinterpret_cast(custom_info_.output_descs)); // arg 2 + static_cast(reinterpret_cast(custom_info_.output_descs)); // arg 2 *(reinterpret_cast(args + ctx_.argsOffset[3])) = - reinterpret_cast(reinterpret_cast(custom_info_.output_addrs)); // arg 3 + static_cast(reinterpret_cast(custom_info_.output_addrs)); // arg 3 *(reinterpret_cast(args + ctx_.argsOffset[4])) = - reinterpret_cast(reinterpret_cast(custom_info_.attr_handle)); // arg 4 + static_cast(reinterpret_cast(custom_info_.attr_handle)); // arg 4 rt_ret = rtMalloc(&args_, args_size_, RT_MEMORY_HBM); if (rt_ret != RT_ERROR_NONE) { @@ -913,7 +916,7 @@ Status KernelTaskInfo::InitAicpuTask(uint32_t op_index, const domi::KernelDef &k op_desc_->GetType().c_str(), ext_info.size(), aicpu_ext_info_addr_); aicpu_param_head->extInfoAddr = reinterpret_cast(aicpu_ext_info_addr_); - aicpu_param_head->extInfoLength = reinterpret_cast(ext_info.size()); + aicpu_param_head->extInfoLength = static_cast(ext_info.size()); // malloc device memory for args rtError_t rt_ret = rtMalloc(static_cast(&args_), args_size_, RT_MEMORY_HBM); @@ -956,12 +959,40 @@ Status KernelTaskInfo::InitAicpuTaskExtInfo(const std::string &ext_info) { if (ext_info.empty()) { return SUCCESS; } + + std::unique_ptr copy_ext_info; + copy_ext_info.reset(new(std::nothrow)uint8_t[ext_info.size()]); + GE_CHECK_NOTNULL(copy_ext_info); + auto sec_ret = memcpy_s(copy_ext_info.get(), ext_info.size(), ext_info.c_str(), ext_info.size()); + if (sec_ret != EOK) { + GELOGE(FAILED, "memcpy failed, ret: %d", sec_ret); + return FAILED; + } + + auto ext_info_data = copy_ext_info.get(); + size_t offset = 0; + while (offset + sizeof(aicpu::FWKAdapter::ExtInfo) <= ext_info.size()) { + auto aicpu_ext_info = reinterpret_cast(ext_info_data + offset); + GELOGD("Ext infoType=%d, infoLen=%u.", aicpu_ext_info->infoType, aicpu_ext_info->infoLen); + if (aicpu_ext_info->infoType == aicpu::FWKAdapter::FWK_ADPT_EXT_SESSION_INFO) { + GE_CHK_BOOL_RET_STATUS(aicpu_ext_info->infoLen == sizeof(SessionInfo), PARAM_INVALID, + "Parse ext session info failed as infoLen must be %zu but %u.", + sizeof(SessionInfo), aicpu_ext_info->infoLen); + SessionInfo *session_info = reinterpret_cast(aicpu_ext_info->infoMsg); + session_info->sessionId = davinci_model_->GetSessionId(); + session_info->sessFlag = true; + GELOGD("Update aicpu_task ext_info session_info session_id is %lu", session_info->sessionId); + } + offset += sizeof(aicpu::FWKAdapter::ExtInfo); + offset += aicpu_ext_info->infoLen; + } + auto rt_ret = rtMalloc(&aicpu_ext_info_addr_, ext_info.size(), RT_MEMORY_HBM); if (rt_ret != RT_ERROR_NONE) { GELOGE(RT_FAILED, "rtMalloc ext_info error: 0x%X, size=%zu", rt_ret, ext_info.size()); return RT_ERROR_TO_GE_STATUS(rt_ret); } - rt_ret = rtMemcpy(aicpu_ext_info_addr_, ext_info.size(), ext_info.c_str(), ext_info.size(), RT_MEMCPY_HOST_TO_DEVICE); + rt_ret = rtMemcpy(aicpu_ext_info_addr_, ext_info.size(), ext_info_data, ext_info.size(), RT_MEMCPY_HOST_TO_DEVICE); if (rt_ret != RT_ERROR_NONE) { GELOGE(RT_FAILED, "rtMemcpy ext_info error: 0x%X, size=%zu", rt_ret, ext_info.size()); return RT_ERROR_TO_GE_STATUS(rt_ret); @@ -1122,18 +1153,24 @@ Status KernelTaskInfo::CceUpdateKernelArgs(const domi::KernelContext &context, u } GELOGI("FileName:%s, Path:%s.", file_name.c_str(), canonicalPath.c_str()); - auto handle = dlopen(canonicalPath.c_str(), RTLD_NOW | RTLD_GLOBAL); + auto handle = mmDlopen(canonicalPath.c_str(), MMPA_RTLD_NOW | MMPA_RTLD_GLOBAL); + const char *error = ""; if (handle == nullptr) { - GELOGE(GE_PLGMGR_SO_NOT_EXIST, "Failed in dlopen %s! ", dlerror()); + error = mmDlerror(); + GE_IF_BOOL_EXEC(error == nullptr, error = ""); + GELOGE(GE_PLGMGR_SO_NOT_EXIST, "Failed in dlopen %s! ", error); return FAILED; } cce::ccStatus_t cc_ret; + std::string update_kernel_args = "ccUpdateKernelArgs"; auto cceUpdateKernelArgs = (cce::ccStatus_t(*)(cce::ccOpContext &, uint64_t, uint64_t, uint64_t, void *, uint64_t, - void *))dlsym(handle, "ccUpdateKernelArgs"); + void *))mmDlsym(handle, const_cast(update_kernel_args.c_str())); if (cceUpdateKernelArgs == nullptr) { GELOGE(FAILED, "Failed to invoke function ccUpdateKernelArgs"); - if (dlclose(handle) != 0) { - GELOGW("Failed to close handle %s", dlerror()); + if (mmDlclose(handle) != 0) { + error = mmDlerror(); + GE_IF_BOOL_EXEC(error == nullptr, error = ""); + GELOGW("Failed to close handle %s", error); } return FAILED; } else { @@ -1146,8 +1183,10 @@ Status KernelTaskInfo::CceUpdateKernelArgs(const domi::KernelContext &context, u const_cast(kernel_def.args().data()), args_size_, sm_contrl); } } - if (dlclose(handle) != 0) { - GELOGW("Failed to close handle %s", dlerror()); + if (mmDlclose(handle) != 0) { + error = mmDlerror(); + GE_IF_BOOL_EXEC(error == nullptr, error = ""); + GELOGW("Failed to close handle %s", error); return FAILED; } if (cc_ret != cce::CC_STATUS_SUCCESS) { @@ -1188,7 +1227,7 @@ Status KernelTaskInfo::SetFlowtable(std::string &flowtable, const domi::KernelDe *(reinterpret_cast( args + (reinterpret_cast(const_cast(context.args_offset().data())))[0])) = - reinterpret_cast(reinterpret_cast(flowtable_)); + static_cast(reinterpret_cast(flowtable_)); } return SUCCESS; } diff --git a/ge/graph/load/new_model_manager/task_info/super_kernel/super_kernel.cc b/ge/graph/load/new_model_manager/task_info/super_kernel/super_kernel.cc index 09ed7458..63f29f84 100644 --- a/ge/graph/load/new_model_manager/task_info/super_kernel/super_kernel.cc +++ b/ge/graph/load/new_model_manager/task_info/super_kernel/super_kernel.cc @@ -23,7 +23,7 @@ Status SuperKernel::Launch(rtStream_t stream, uint32_t dump_flag) { const void *func_stub_ = this->GetFuncStub(); const void *args[] = {this->GetNavTablePtr(), - reinterpret_cast(reinterpret_cast(this->GetNavTableSize()))}; + reinterpret_cast(static_cast(this->GetNavTableSize()))}; rtError_t rt_ret = rtMalloc((void **)&(device_args_addr_), sizeof(args), RT_MEMORY_HBM); GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "rtMalloc failied. error: 0x%X", rt_ret); return diff --git a/ge/graph/load/new_model_manager/task_info/super_kernel/super_kernel_factory.cc b/ge/graph/load/new_model_manager/task_info/super_kernel/super_kernel_factory.cc index d237d56c..69f7b159 100644 --- a/ge/graph/load/new_model_manager/task_info/super_kernel/super_kernel_factory.cc +++ b/ge/graph/load/new_model_manager/task_info/super_kernel/super_kernel_factory.cc @@ -27,7 +27,7 @@ SuperKernelFactory &SuperKernelFactory::GetInstance() { Status SuperKernelFactory::Init() { if (!is_init_) { std::string skt_bin = "libcce_aicore.so"; - handle_ = dlopen(skt_bin.c_str(), RTLD_NOW | RTLD_GLOBAL); + handle_ = mmDlopen(skt_bin.c_str(), MMPA_RTLD_NOW | MMPA_RTLD_GLOBAL); if (handle_ == nullptr) { GELOGE(FAILED, "SKT: open skt lib failed, please check LD_LIBRARY_PATH."); } @@ -85,8 +85,10 @@ Status SuperKernelFactory::FuseKernels(const std::vector &stub_func_list "equal to 2"); return FAILED; } - GELOGI("SKT: superkernel start fuse, superkernel size %d.", stub_func_list.size()); - uint64_t nav_table[2 * stub_func_list.size()]; + GELOGI("SKT: superkernel start fuse, superkernel size %zu.", stub_func_list.size()); + const size_t nav_table_len = 2 * stub_func_list.size(); + std::unique_ptr nav_table(new(std::nothrow) uint64_t[nav_table_len]); + GE_CHECK_NOTNULL(nav_table); uint64_t nav_table_size = 2 * stub_func_list.size() * sizeof(int64_t); rtError_t rt_ret; @@ -99,16 +101,16 @@ Status SuperKernelFactory::FuseKernels(const std::vector &stub_func_list GELOGD("SKT: fuseKernels subFunc %p, device func address %p", stub_func_list[i], sub_device_func); // store two uint64_t address // address divided by 4 because of 32bits encoding, call offset will *4 when calculating - nav_table[i * 2] = reinterpret_cast(reinterpret_cast(sub_device_func)) / 4; + nav_table[i * 2] = static_cast(reinterpret_cast(sub_device_func)) / 4; GELOGD("SKT: CALL offet %lu", nav_table[i * 2]); - nav_table[i * 2 + 1] = reinterpret_cast(reinterpret_cast(args_addr_list[i])); + nav_table[i * 2 + 1] = static_cast(reinterpret_cast(args_addr_list[i])); GELOGD("SKT: fuseKernels args base address %lu", nav_table[i * 2 + 1]); } rt_ret = rtMalloc((void **)&hbm_nav_table_addr, nav_table_size, RT_MEMORY_HBM); GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "rtMalloc failed. error: 0x%X", rt_ret); return RT_ERROR_TO_GE_STATUS(rt_ret);) rt_ret = - rtMemcpy((void *)hbm_nav_table_addr, nav_table_size, (void *)nav_table, nav_table_size, RT_MEMCPY_HOST_TO_DEVICE); + rtMemcpy((void *)hbm_nav_table_addr, nav_table_size, (void *)nav_table.get(), nav_table_size, RT_MEMCPY_HOST_TO_DEVICE); GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "rtMemcpy failed. error: 0x%X", rt_ret); GE_CHK_RT(rtFree(hbm_nav_table_addr)); return RT_ERROR_TO_GE_STATUS(rt_ret);) // Create the necessary metadata for the super kernel diff --git a/ge/graph/load/new_model_manager/task_info/super_kernel/super_kernel_factory.h b/ge/graph/load/new_model_manager/task_info/super_kernel/super_kernel_factory.h index efd61ef7..aeb5b49b 100644 --- a/ge/graph/load/new_model_manager/task_info/super_kernel/super_kernel_factory.h +++ b/ge/graph/load/new_model_manager/task_info/super_kernel/super_kernel_factory.h @@ -34,8 +34,10 @@ class SuperKernelFactory { ~SuperKernelFactory() { if (handle_ != nullptr) { GELOGI("SKT: SKT LIB PATH release."); - if (dlclose(handle_) != 0) { - GELOGW("failed to close handle, message: %s", dlerror()); + if (mmDlclose(handle_) != 0) { + const char *error = mmDlerror(); + GE_IF_BOOL_EXEC(error == nullptr, error = ""); + GELOGW("failed to close handle, message: %s", error); } } }; diff --git a/ge/graph/load/new_model_manager/zero_copy_offset.cc b/ge/graph/load/new_model_manager/zero_copy_offset.cc index e93a7250..970b292c 100644 --- a/ge/graph/load/new_model_manager/zero_copy_offset.cc +++ b/ge/graph/load/new_model_manager/zero_copy_offset.cc @@ -30,49 +30,37 @@ ZeroCopyOffset::ZeroCopyOffset() {} ZeroCopyOffset::~ZeroCopyOffset() {} -Status ZeroCopyOffset::InitInputDataInfo(const vector &output_size_list, - const vector &virtual_addr_list, const OpDescPtr &op_desc, +Status ZeroCopyOffset::InitInputDataInfo(int64_t output_size, void *virtual_addr, const OpDescPtr &op_desc, bool &fusion_flag) { GELOGI("[ZCPY] Start to InitInputDataInfo of %s, total_data_size is %ld, virtual_addr is %p", - op_desc->GetName().c_str(), output_size_list[kDataIndex], virtual_addr_list[kDataIndex]); - if (output_size_list.empty() || virtual_addr_list.empty() || (output_size_list.size() != virtual_addr_list.size())) { - GELOGE(PARAM_INVALID, "Data[%s] init failed: Output size is %zu, Output addr is %zu", op_desc->GetName().c_str(), - output_size_list.size(), virtual_addr_list.size()); - return PARAM_INVALID; - } - - basic_addr_ = virtual_addr_list[kDataIndex]; + op_desc->GetName().c_str(), output_size, virtual_addr); + basic_addr_ = virtual_addr; (void)ge::AttrUtils::GetListInt(op_desc, ATTR_ZERO_COPY_BASIC_OFFSET, zero_copy_basic_offset_); (void)ge::AttrUtils::GetListInt(op_desc, ATTR_ZERO_COPY_RELATIVE_OFFSET, zero_copy_relative_offset_); GE_CHK_BOOL_EXEC(zero_copy_basic_offset_.size() == zero_copy_relative_offset_.size(), return PARAM_INVALID, "basic_offset_size should be equal to relative_offset_size"); - GELOGI("[ZCPY] zero_copy_basic_offset size is %zu", zero_copy_basic_offset_.size()); + GELOGD("[ZCPY] zero_copy_basic_offset size is %zu", zero_copy_basic_offset_.size()); int64_t virtual_addr_offset = op_desc->GetOutputOffset().at(kDataIndex); - GELOGI("virtual_addr_offset is %ld.", virtual_addr_offset); IsL2Fusion(zero_copy_basic_offset_, virtual_addr_offset, fusion_flag); uint32_t out_count = 0; - data_size_ = output_size_list[kDataIndex]; + data_size_ = output_size; if (!fusion_flag) { - GELOGI("[ZCPY] %s not set l2_fusion.", op_desc->GetName().c_str()); out_count++; - data_info_.emplace_back(output_size_list[kDataIndex], virtual_addr_list[kDataIndex]); + data_info_.emplace_back(output_size, virtual_addr); relative_offset_.emplace_back(0); - GELOGI("[ZCPY] %s size is %ld, virtual_addr is %p.", op_desc->GetName().c_str(), output_size_list[kDataIndex], - virtual_addr_list[kDataIndex]); + GELOGD("[ZCPY] %s size is %ld, virtual_addr is %p.", op_desc->GetName().c_str(), output_size, virtual_addr); } else { GELOGI("[ZCPY] set l2_fusion for %s.", op_desc->GetName().c_str()); for (size_t index = 0; index < zero_copy_basic_offset_.size(); ++index) { if (zero_copy_basic_offset_.at(index) == virtual_addr_offset) { out_count++; - uint64_t out_offset = - reinterpret_cast(virtual_addr_list[kDataIndex]) + zero_copy_relative_offset_.at(index); - int64_t real_data_size = ModelUtils::GetOutputSize(op_desc).at(kDataIndex); - data_info_.emplace_back(real_data_size, reinterpret_cast(reinterpret_cast(out_offset))); + uint64_t out_offset = reinterpret_cast(virtual_addr) + zero_copy_relative_offset_.at(index); + data_info_.emplace_back(output_size, reinterpret_cast(static_cast(out_offset))); relative_offset_.emplace_back(zero_copy_relative_offset_.at(index)); GELOGI("[ZCPY] virtual_addr: %p has been l2-fusion to %lu, need copy data_size is %ld.", basic_addr_, - out_offset, real_data_size); + out_offset, output_size); } } } @@ -83,7 +71,6 @@ Status ZeroCopyOffset::InitInputDataInfo(const vector &output_size_list Status ZeroCopyOffset::InitOutputDataInfo(const vector &input_size_list, const vector &virtual_addr_list, const OpDescPtr &op_desc, const size_t &idx, bool &fusion_flag) { - GELOGI("[ZCPY] Start to InitOutputDataInfo of %s.", op_desc->GetName().c_str()); int64_t size = input_size_list[idx]; auto tensor_desc = op_desc->GetInputDescPtr(idx); GE_CHECK_NOTNULL(tensor_desc); @@ -92,7 +79,7 @@ Status ZeroCopyOffset::InitOutputDataInfo(const vector &input_size_list return FAILED; } - GELOGI("Tensor data size: GetSize=%ld, GetTensorSizeInBytes=%ld", input_size_list[idx], size); + GELOGD("Tensor data size: GetSize=%ld, GetTensorSizeInBytes=%ld", input_size_list[idx], size); basic_addr_ = virtual_addr_list[idx]; (void)ge::AttrUtils::GetListInt(op_desc, ATTR_ZERO_COPY_BASIC_OFFSET, zero_copy_basic_offset_); @@ -100,13 +87,11 @@ Status ZeroCopyOffset::InitOutputDataInfo(const vector &input_size_list GE_CHK_BOOL_EXEC(zero_copy_basic_offset_.size() == zero_copy_relative_offset_.size(), return PARAM_INVALID, "basic_offset_size should be equal to relative_offset_size"); int64_t virtual_addr_offset = op_desc->GetInputOffset().at(idx); - GELOGI("virtual_addr_offset is %ld.", virtual_addr_offset); IsL2Fusion(zero_copy_basic_offset_, virtual_addr_offset, fusion_flag); uint32_t in_count = 0; data_size_ = size; if (!fusion_flag) { - GELOGI("[ZCPY] %s not set l2-fusion.", op_desc->GetName().c_str()); in_count++; data_info_.emplace_back(size, virtual_addr_list[idx]); // op_desc not set l2fusion when fusion_flag is false @@ -119,7 +104,7 @@ Status ZeroCopyOffset::InitOutputDataInfo(const vector &input_size_list in_count++; uint64_t in_offset = reinterpret_cast(virtual_addr_list[idx]) + zero_copy_relative_offset_.at(index); int64_t real_data_size = ModelUtils::GetInputSize(op_desc).at(idx); - data_info_.emplace_back(real_data_size, reinterpret_cast(reinterpret_cast(in_offset))); + data_info_.emplace_back(real_data_size, reinterpret_cast(static_cast(in_offset))); relative_offset_.emplace_back(zero_copy_relative_offset_.at(index)); GELOGI("[ZCPY] virtual_addr: %p has been l2-fusion from %lu, need copy data_size is %ld.", basic_addr_, in_offset, real_data_size); @@ -142,10 +127,8 @@ void ZeroCopyOffset::IsL2Fusion(const vector &fusion_basic_addrs, const void ZeroCopyOffset::SetInputOutsideAddrs(const vector &output_offset_list, void *addr, const size_t &index, bool fusion_flag, std::set &real_virtual_addrs) { - GELOGI("[ZCPY] Start to SetInputOutsideAddrs for virtual_addr %p.", addr); uint32_t out_count = 0; if (!fusion_flag) { - GELOGI("[ZCPY] not set l2-fusion for virtual_adr %p.", addr); out_count++; std::map> addr_mapping; addr_mapping[addr] = {}; @@ -175,7 +158,6 @@ void ZeroCopyOffset::SetOutputOutsideAddrs(const int64_t &input_offset, const bo GELOGI("[ZCPY] Start to SetOutputOutsideAddrs for virtual_addr %p.", addr); uint32_t out_count = 0; if (!fusion_flag) { - GELOGI("[ZCPY] not set l2-fusion for virtual_addr %p.", addr); out_count++; std::map> addr_mapping; addr_mapping[addr] = {}; @@ -209,7 +191,7 @@ bool ZeroCopyOffset::SetOutsideAddrsValue(ZeroCopyTask &zero_copy_task, void *ou GE_CHK_STATUS(zero_copy_task.SetTaskArgsOffset(addr_val, offset), "Input args invalid."); void *args_val = static_cast(args) + offset; args_addrs->second.push_back(args_val); - GELOGI("[ZCPY] set copy input: virtual_addr: 0x%lx, task_addr: %p, args: %p, offset: %zu.", addr_val, args_val, + GELOGD("[ZCPY] set copy input: virtual_addr: 0x%lx, task_addr: %p, args: %p, offset: %zu.", addr_val, args_val, args, offset); set_batch_label_flag = true; } diff --git a/ge/graph/load/new_model_manager/zero_copy_offset.h b/ge/graph/load/new_model_manager/zero_copy_offset.h index c662032b..025d1b14 100644 --- a/ge/graph/load/new_model_manager/zero_copy_offset.h +++ b/ge/graph/load/new_model_manager/zero_copy_offset.h @@ -42,8 +42,7 @@ class ZeroCopyOffset { ZeroCopyOffset(); ~ZeroCopyOffset(); - Status InitInputDataInfo(const vector &output_size_list, const vector &virtual_addr_list, - const OpDescPtr &op_desc, bool &fusion_flag); + Status InitInputDataInfo(int64_t output_size, void *virtual_addr, const OpDescPtr &op_desc, bool &fusion_flag); void SetInputOutsideAddrs(const vector &output_offset_list, void *addr, const size_t &index, bool fusion_flag, std::set &real_virtual_addrs); diff --git a/ge/graph/load/new_model_manager/zero_copy_task.cc b/ge/graph/load/new_model_manager/zero_copy_task.cc index 2079034e..9b42d563 100755 --- a/ge/graph/load/new_model_manager/zero_copy_task.cc +++ b/ge/graph/load/new_model_manager/zero_copy_task.cc @@ -19,6 +19,7 @@ #include "framework/common/debug/ge_log.h" #include "framework/common/util.h" #include "graph/load/new_model_manager/model_utils.h" +#include "common/ge_compiler_options.h" namespace ge { const char *const kDefaultBatchLable = "Batch_default"; @@ -48,7 +49,7 @@ Status ZeroCopyTask::SetTaskArgsOffset(uintptr_t addr, size_t offset) { it->second.insert(offset); } - GELOGI("[ZCPY] %s set task, virtual_addr: 0x%lx, args_addr: %p, size: %zu, offset: %zu", name_.c_str(), addr, + GELOGD("[ZCPY] %s set task, virtual_addr: 0x%lx, args_addr: %p, size: %zu, offset: %zu", name_.c_str(), addr, args_addr_, args_size_, offset); return SUCCESS; } @@ -157,7 +158,7 @@ Status ZeroCopyTask::DistributeParam(bool async_mode, rtStream_t stream) { rt_err = rtMemcpyAsync(args_addr_, args_size_, args_info_.data(), args_info_.size(), RT_MEMCPY_HOST_TO_DEVICE_EX, stream); } else { - __builtin_prefetch(args_addr_); + GE_BUILTIN_PREFETCH(args_addr_); rt_err = rtMemcpy(args_addr_, args_size_, args_info_.data(), args_info_.size(), RT_MEMCPY_HOST_TO_DEVICE); } @@ -166,7 +167,7 @@ Status ZeroCopyTask::DistributeParam(bool async_mode, rtStream_t stream) { return RT_ERROR_TO_GE_STATUS(rt_err); } - GELOGI("[ZCPY] %s refresh task args success, args_addr: %p, size: %zu, args_info_: %p, length: %zu", name_.c_str(), + GELOGD("[ZCPY] %s refresh task args success, args_addr: %p, size: %zu, args_info_: %p, length: %zu", name_.c_str(), args_addr_, args_size_, args_info_.data(), args_info_.size()); return SUCCESS; } diff --git a/ge/graph/manager/graph_manager.cc b/ge/graph/manager/graph_manager.cc index f00c477f..5e6f8577 100755 --- a/ge/graph/manager/graph_manager.cc +++ b/ge/graph/manager/graph_manager.cc @@ -363,7 +363,7 @@ Status GraphManager::AddGraph(const GraphId &graph_id, const Graph &graph, for (auto &subgraph : compute_graph->GetAllSubgraphs()) { (void)AttrUtils::SetStr(*subgraph, ATTR_NAME_SESSION_GRAPH_ID, session_graph_id); } - GELOGW("Get graph session_graph_id attr failed, set session id to default value: [0]"); + GELOGD("Get graph session_graph_id attr failed, set session id to default value: [0]"); } GraphNodePtr graph_node = MakeShared(graph_id); @@ -396,8 +396,6 @@ Status GraphManager::AddGraph(const GraphId &graph_id, const Graph &graph, stages.builder.SetOptions(options_); var_acc_ctrl_.AddGraph(graph_id, compute_graph); - - GELOGI("[GraphManager] add graph success, graph_id = %u.", graph_id); return SUCCESS; } @@ -435,7 +433,7 @@ Status GraphManager::AddGraphWithCopy(const GraphId &graph_id, const Graph &grap for (auto &subgraph : new_compute_graph->GetAllSubgraphs()) { (void)AttrUtils::SetStr(*subgraph, ATTR_NAME_SESSION_GRAPH_ID, session_graph_id); } - GELOGW("Get graph session_graph_id attr failed, set session id to default value: [0]"); + GELOGD("Get graph session_graph_id attr failed, set session id to default value: [0]"); } GraphNodePtr graph_node = MakeShared(graph_id); @@ -468,8 +466,6 @@ Status GraphManager::AddGraphWithCopy(const GraphId &graph_id, const Graph &grap stages.builder.SetOptions(options_); var_acc_ctrl_.AddGraph(graph_id, new_compute_graph); - - GELOGI("[GraphManager] add graph success, graph_id = %u.", graph_id); return SUCCESS; } @@ -546,7 +542,7 @@ Status GraphManager::OptimizeSubGraphWithMultiThreads(ComputeGraphPtr compute_gr const auto &root_subgraph_list = sub_graph_map[compute_graph]; std::string op_compile_strategy; (void)AttrUtils::GetStr(compute_graph, ATTR_NAME_OP_COMPILE_STRATEGY, op_compile_strategy); - GELOGI("OptimizeSubGraphWithMultiThreads Process op_compile_strategy:%s", op_compile_strategy.c_str()); + GELOGD("OptimizeSubGraphWithMultiThreads Process op_compile_strategy:%s", op_compile_strategy.c_str()); for (const auto &subgraph : root_subgraph_list) { if (!op_compile_strategy.empty()) { (void) AttrUtils::SetStr(subgraph->GetSubGraph(), ATTR_NAME_OP_COMPILE_STRATEGY, op_compile_strategy); @@ -576,7 +572,7 @@ Status GraphManager::OptimizeSubGraphWithMultiThreads(ComputeGraphPtr compute_gr vector_future.emplace_back(std::move(f)); } } - GELOGI("All sub graph num is %zu", vector_future.size()); + GELOGD("All sub graph num is %zu", vector_future.size()); for (size_t i = 0; i < vector_future.size(); ++i) { Status ret_status = vector_future[i].get(); if (ret_status != SUCCESS) { @@ -700,7 +696,7 @@ Status GraphManager::SetSubgraph(uint64_t session_id, ComputeGraphPtr compute_gr /// Multiply optimize subgraph: /// 1. run lx buffer while build_mode is normal and buffer_optimize is empty or "off_optimize"; /// 2. run lx fusion or buffer according build_mode and build_step in fe. - GELOGI("Directly optimize subgraph with build mode:%s, and step:%s, buffer_optimize:%s.", + GELOGD("Directly optimize subgraph with build mode:%s, and step:%s, buffer_optimize:%s.", options_.build_mode.c_str(), options_.build_step.c_str(), buffer_optimize.c_str()); @@ -747,7 +743,7 @@ Status GraphManager::PreRunOptimizeOriginalGraph(const GraphNodePtr &graph_node, GE_CHK_STATUS_RET(graph_pass.Run(compute_graph)); GE_CHK_STATUS_RET(stages.optimizer.IdentifyReference(compute_graph), "Identify reference failed."); - GELOGI("PreRun:PreRunOptimizeOriginalGraph success."); + GELOGD("PreRun:PreRunOptimizeOriginalGraph success."); return SUCCESS; } @@ -762,10 +758,10 @@ Status GraphManager::PreRunOptimizeSubGraph(const GraphNodePtr &graph_node, if (options_.build_mode == BUILD_MODE_TUNING && options_.build_step == BUILD_STEP_AFTER_UB_MATCH) { std::string tuning_path; (void) GetContext().GetOption(TUNING_PATH, tuning_path); - GELOGI("Dump path:%s.", tuning_path.c_str()); + GELOGD("Dump path:%s.", tuning_path.c_str()); GraphUtils::DumpGEGraph(compute_graph, "", true, tuning_path); } - GELOGI("PreRun:PreRunOptimizeSubGraph success."); + GELOGD("PreRun:PreRunOptimizeSubGraph success."); return SUCCESS; } @@ -785,12 +781,12 @@ Status GraphManager::PreRunAfterOptimizeSubGraph(const GraphNodePtr &graph_node, } GM_RUN_AND_DUMP_PERF("Build", Build, graph_node, compute_graph, ge_root_model, session_id); - GELOGI("PreRun:PreRunAfterOptimizeSubGraph success."); + GELOGD("PreRun:PreRunAfterOptimizeSubGraph success."); return SUCCESS; } Status GraphManager::SetRtContext(rtContext_t rt_context, rtCtxMode_t mode, uint64_t session_id, uint32_t graph_id) { - GELOGI("set rt_context, session id: %lu, graph id: %u, mode %d, device id:%u.", session_id, graph_id, + GELOGD("set rt_context, session id: %lu, graph id: %u, mode %d, device id:%u.", session_id, graph_id, static_cast(mode), ge::GetContext().DeviceId()); rtError_t rt_ret = rtCtxCreate(&rt_context, mode, ge::GetContext().DeviceId()); @@ -1251,7 +1247,7 @@ Status GraphManager::BuildGraphForUnregisteredOp(const GraphId &graph_id, const Status GraphManager::BuildGraph(const GraphId &graph_id, const std::vector &inputs, GeRootModelPtr &ge_root_model, uint64_t session_id, bool async) { - GELOGI("[BuildGraph] start to build graph, graph_id=%u.", graph_id); + GELOGD("[BuildGraph] start to build graph, graph_id=%u.", graph_id); if (inputs.empty()) { GELOGW("[BuildGraph] BuildGraph warning: empty GeTensor inputs"); } @@ -1531,7 +1527,6 @@ Status GraphManager::ParseOptions(const std::map &opti // Set save_original_model flag (ge.save_original_model) ParseOption(options, SAVE_ORIGINAL_MODEL, options_.save_original_model); - GELOGI("Set save original model flag %s", options_.save_original_model.c_str()); // Original model file name ParseOption(options, ORIGINAL_MODEL_FILE, options_.original_model_file); @@ -1540,16 +1535,6 @@ Status GraphManager::ParseOptions(const std::map &opti ParseOption(options, DYNAMIC_NODE_TYPE, options_.dynamic_node_type); GELOGD("Dynamic dims params: input shape is %s, dynamic dims is %s, dynamic node type is %d.", options_.input_shape.c_str(), options_.dynamic_dims.c_str(), options_.dynamic_node_type); - if ((!options_.input_shape.empty() && options_.dynamic_dims.empty()) || - (options_.input_shape.empty() && !options_.dynamic_dims.empty())) { - GELOGE(GRAPH_PARAM_INVALID, "Should set input shape and dynamic dims at the same time"); - return GRAPH_PARAM_INVALID; - } - if ((!options_.input_shape.empty() && options_.dynamic_node_type == kInvalidDynaimcDimsType) || - (!options_.dynamic_dims.empty() && options_.dynamic_node_type == kInvalidDynaimcDimsType)) { - GELOGE(GRAPH_PARAM_INVALID, "Should set valid dynamic node type"); - return GRAPH_PARAM_INVALID; - } // Set Build model and step ParseOption(options, BUILD_MODE, options_.build_mode); @@ -2252,7 +2237,7 @@ Status GraphManager::OptimizeStage1(ge::ComputeGraphPtr &compute_graph) { } Status GraphManager::OptimizeStage2(ge::ComputeGraphPtr &compute_graph) { - GELOGI("Start optimize after merge sub graph."); + GELOGD("Start optimize after merge sub graph."); PassManager after_merge_passes; GE_CHK_STATUS_RET(after_merge_passes.AddPass("OptimizeStage2::AfterMergePasses::LinkGenMaskNodesPass", @@ -2502,7 +2487,7 @@ Status GraphManager::ProcessSubGraphWithMultiThreads(GraphManager *graph_manager ComputeGraphPtr compute_graph_tmp = sub_graph_info_ptr->GetSubGraph(); const std::string &engine_name = sub_graph_info_ptr->GetEngineName(); - GELOGI("ProcessSubGraphWithMultiThreads start, graph name is %s, engine_name is %s, thread id is %lu", + GELOGD("ProcessSubGraphWithMultiThreads start, graph name is %s, engine_name is %s, thread id is %lu", compute_graph_tmp != nullptr ? compute_graph_tmp->GetName().c_str() : "", engine_name.c_str(), pthread_self()); GE_DUMP(compute_graph_tmp, "OptimizeSubGraphBefore"); @@ -2514,11 +2499,11 @@ Status GraphManager::ProcessSubGraphWithMultiThreads(GraphManager *graph_manager GELOGE(ret, "SubGraph optimize Failed %s", engine_name.c_str()); return ret; } else { - GELOGI("SubGraph optimize success %s", engine_name.c_str()); + GELOGD("SubGraph optimize success %s", engine_name.c_str()); } GE_DUMP(compute_graph_tmp, "OptimizeSubGraphAfter"); sub_graph_info_ptr->SetSubGraph(compute_graph_tmp); - GELOGI("ProcessSubGraphWithMultiThreads end, graph name is %s, engine_name is %s, thread id is %lu", + GELOGD("ProcessSubGraphWithMultiThreads end, graph name is %s, engine_name is %s, thread id is %lu", compute_graph_tmp != nullptr ? compute_graph_tmp->GetName().c_str() : "", engine_name.c_str(), pthread_self()); } else { @@ -2849,13 +2834,15 @@ void GraphManager::RunThread(GraphManager *graph_manager) { if (args.graph_node->graph_run_async_listener_ != nullptr) { args.graph_node->graph_run_async_listener_->SetCallback(args.callback); } + Status ret; // parse inputs.dims to vector> dynamic_dims - if (graph_manager->ParseInputsDims(args.input_tensor) != SUCCESS) { - GELOGE(PARAM_INVALID, "Parse input dims failed."); + ret = graph_manager->ParseInputsDims(args.input_tensor); + if (ret != SUCCESS) { + ReturnError(graph_manager, args.callback, ret, "ParseInputsDims failed, thread exit."); + args.graph_node->Unlock(); return; } - Status ret; if (!args.graph_node->GetLoadFlag()) { ret = graph_manager->LoadGraphAsync(args.ge_root_model, args.graph_node); if (ret != SUCCESS || args.ge_root_model == nullptr) { @@ -2880,12 +2867,12 @@ void GraphManager::RunThread(GraphManager *graph_manager) { ret = graph_manager->graph_executor_.ExecuteGraphAsync(args.graph_id, args.graph_node->GetGeRootModel(), args.input_tensor); args.graph_node->SetRunFlag(false); - args.graph_node->Unlock(); if (ret != SUCCESS) { - GELOGE(ret, "[GraphManager] Run graph async failed, graph_id=%u.", args.graph_id); - StopQueue(graph_manager); + ReturnError(graph_manager, args.callback, ret, "ExecuteGraphAsync failed, thread exit."); + args.graph_node->Unlock(); return; } + args.graph_node->Unlock(); GELOGI("[GraphManager] Run graph async success, graph_id=%u.", args.graph_id); } } diff --git a/ge/graph/manager/graph_var_manager.cc b/ge/graph/manager/graph_var_manager.cc index 0b9f1ff9..be7d4eb2 100755 --- a/ge/graph/manager/graph_var_manager.cc +++ b/ge/graph/manager/graph_var_manager.cc @@ -92,13 +92,13 @@ ge::Status VarResource::SaveVarAddr(const std::string &var_name, const ge::GeTen GELOGD("VarResource::SaveVarAddr, var_key = %s", var_key.c_str()); if (var_addr_mgr_map_.count(var_key) == 0) { uint64_t logic_address = VarManager::Instance(session_id_)->GetVarMemLogicBase() + - reinterpret_cast(reinterpret_cast(address)); + static_cast(reinterpret_cast(address)); GELOGI("SaveVarAddr node_name %s, tensor_desc format %s, type %s.", var_name.c_str(), TypeUtils::FormatToSerialString(tensor_desc.GetFormat()).c_str(), TypeUtils::DataTypeToSerialString(tensor_desc.GetDataType()).c_str()); VarAddrMgr var_addr_mgr; - var_addr_mgr.address = reinterpret_cast(reinterpret_cast(logic_address)); - var_addr_mgr.offset = reinterpret_cast(reinterpret_cast(address)); + var_addr_mgr.address = reinterpret_cast(static_cast(logic_address)); + var_addr_mgr.offset = static_cast(reinterpret_cast(address)); var_addr_mgr.tensor_desc = tensor_desc; var_addr_mgr.memory_type = memory_type; var_addr_mgr_map_[var_key] = var_addr_mgr; @@ -510,7 +510,7 @@ ge::Status VarManager::AssignVarMem(const std::string &var_name, const ge::GeTen } result = var_resource_->SaveVarAddr( - var_name, tensor_desc, reinterpret_cast(reinterpret_cast(mem_offset)), memory_type); + var_name, tensor_desc, reinterpret_cast(static_cast(mem_offset)), memory_type); if (result != SUCCESS) { GELOGE(ge::INTERNAL_ERROR, "AssignVarMem by offset failed."); return ge::INTERNAL_ERROR; @@ -527,7 +527,7 @@ ge::Status VarManager::AssignVarMem(const std::string &var_name, const ge::GeTen result = var_resource_->GetCurVarDesc(var_name, cur_tensor_desc); if (result != SUCCESS) { var_resource_->SetVarAddr(var_name, tensor_desc, - reinterpret_cast(reinterpret_cast(mem_offset)), memory_type); + reinterpret_cast(static_cast(mem_offset)), memory_type); return SUCCESS; } @@ -542,7 +542,7 @@ ge::Status VarManager::AssignVarMem(const std::string &var_name, const ge::GeTen ge::TypeUtils::FormatToSerialString(cur_tensor_desc.GetFormat()).c_str(), cur_tensor_desc.GetShape().GetDims().size()); var_resource_->SetVarAddr(var_name, tensor_desc, - reinterpret_cast(reinterpret_cast(mem_offset)), memory_type); + reinterpret_cast(static_cast(mem_offset)), memory_type); } return SUCCESS; @@ -642,7 +642,7 @@ ge::Status VarManager::SyncBroadCastData2Var(uint32_t graph_id, const std::strin bool VarManager::IsVarAddr(const int64_t &offset) { std::lock_guard lock(mutex_); if (var_resource_ == nullptr) { - GELOGW("VarManager has not been init."); + GELOGD("VarManager has not been init."); return false; } return var_resource_->IsVarAddr(offset); diff --git a/ge/graph/manager/trans_var_data_utils.cc b/ge/graph/manager/trans_var_data_utils.cc index d6865716..cd992d3d 100644 --- a/ge/graph/manager/trans_var_data_utils.cc +++ b/ge/graph/manager/trans_var_data_utils.cc @@ -374,7 +374,7 @@ Status TransVarDataUtils::SyncVarData2BroadCast(const string &var_name, const ge GE_MAKE_GUARD_RTMEM(src_host_addr); GE_CHK_STATUS_RET(SyncTensorToHost(var_name, src_tensor_desc, &src_host_addr, src_addr_size, session_id)); - GELOGI("src_addr_size: %u, dst_addr_size: %u", src_addr_size, dst_addr_size); + GELOGI("src_addr_size: %ld, dst_addr_size: %ld", src_addr_size, dst_addr_size); GE_CHK_BOOL_RET_STATUS(src_addr_size == dst_addr_size, FAILED, "var data size is not equal broadcast "); GE_CHK_RT_RET(rtMemcpy(dst_addr, dst_addr_size, src_host_addr, src_addr_size, RT_MEMCPY_HOST_TO_DEVICE)); @@ -403,7 +403,7 @@ Status TransVarDataUtils::SyncTensorToHost(const string &var_name, const ge::GeT GE_CHK_STATUS_RET(VarManager::Instance(session_id)->GetVarAddr(var_name, src_tensor_desc, &src_addr)); uint8_t *mem_addr = src_addr - - static_cast(reinterpret_cast(VarManager::Instance(session_id)->GetVarMemLogicBase())) + + static_cast(static_cast(VarManager::Instance(session_id)->GetVarMemLogicBase())) + static_cast( reinterpret_cast(VarManager::Instance(session_id)->GetVarMemoryBase(RT_MEMORY_HBM))); GE_CHK_RT_RET(rtMallocHost(reinterpret_cast(host_addr), src_tensor_size)); @@ -420,7 +420,7 @@ Status TransVarDataUtils::SyncTensorToDevice(const string &var_name, const uint8 GE_CHK_STATUS_RET(VarManager::Instance(session_id)->GetVarAddr(var_name, dst_tensor_desc, &dst_addr)); uint8_t *mem_addr = dst_addr - - static_cast(reinterpret_cast(VarManager::Instance(session_id)->GetVarMemLogicBase())) + + static_cast(static_cast(VarManager::Instance(session_id)->GetVarMemLogicBase())) + static_cast( reinterpret_cast(VarManager::Instance(session_id)->GetVarMemoryBase(RT_MEMORY_HBM))); GE_CHK_RT_RET(rtMemcpy(mem_addr, addr_size, host_addr, addr_size, RT_MEMCPY_HOST_TO_DEVICE)); @@ -501,7 +501,7 @@ Status TransVarDataUtils::TransAllVarData(const vector &variable_nodes, } Status TransVarDataUtils::CopyVarData(const ComputeGraphPtr &compute_graph, uint64_t session_id, uint32_t device_id) { - GELOGI("CopyVarData start: session_id:%lu.", session_id); + GELOGD("CopyVarData start: session_id:%lu.", session_id); if (compute_graph == nullptr) { GELOGE(FAILED, "compute_graph is nullptr"); return FAILED; diff --git a/ge/graph/manager/util/debug.cc b/ge/graph/manager/util/debug.cc index b2ef1c92..45c070c6 100644 --- a/ge/graph/manager/util/debug.cc +++ b/ge/graph/manager/util/debug.cc @@ -32,7 +32,7 @@ Debug::~Debug() = default; void Debug::DumpProto(const Message &proto, const char *file) { std::string file_path = RealPath(file); - int fd = open(file_path.c_str(), O_WRONLY | O_CREAT | O_TRUNC, S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH); + int fd = mmOpen2(file_path.c_str(), M_WRONLY | M_CREAT | O_TRUNC, M_IRUSR | M_IWUSR | M_UMASK_GRPREAD | M_UMASK_OTHREAD); if (fd == -1) { GELOGW("Write %s failed", file_path.c_str()); return; @@ -40,7 +40,7 @@ void Debug::DumpProto(const Message &proto, const char *file) { auto output = ge::MakeShared(fd); if (output == nullptr) { GELOGW("create output failed."); - if (close(fd) != 0) { + if (mmClose(fd) != 0) { GELOGW("close fd failed."); } return; @@ -49,7 +49,7 @@ void Debug::DumpProto(const Message &proto, const char *file) { if (!ret) { GELOGW("dump proto failed."); } - if (close(fd) != 0) { + if (mmClose(fd) != 0) { GELOGW("close fd failed."); } } diff --git a/ge/graph/manager/util/debug.h b/ge/graph/manager/util/debug.h index 9cedb680..e1b13caf 100755 --- a/ge/graph/manager/util/debug.h +++ b/ge/graph/manager/util/debug.h @@ -17,7 +17,6 @@ #ifndef GE_GRAPH_MANAGER_UTIL_DEBUG_H_ #define GE_GRAPH_MANAGER_UTIL_DEBUG_H_ -#include #include #include #include @@ -25,7 +24,6 @@ #include #include #include -#include #include #include #include diff --git a/ge/graph/optimize/graph_optimize.cc b/ge/graph/optimize/graph_optimize.cc index 931d529b..c94408de 100644 --- a/ge/graph/optimize/graph_optimize.cc +++ b/ge/graph/optimize/graph_optimize.cc @@ -58,8 +58,7 @@ void AddNodeInputProperty(ComputeGraphPtr &compute_graph) { for (auto &in_data_anchor : node->GetAllInDataAnchors()) { auto peer_out_anchor = in_data_anchor->GetPeerOutAnchor(); - GE_IF_BOOL_EXEC( - peer_out_anchor == nullptr, GELOGW("peer_out_anchor is nullptr! node: %s", node->GetName().c_str()); continue); + GE_IF_BOOL_EXEC(peer_out_anchor == nullptr, continue); ge::NodePtr src_node = peer_out_anchor->GetOwnerNode(); src_index_list = node_op_desc->GetSrcIndex(); @@ -242,11 +241,11 @@ Status GraphOptimize::OptimizeGraphBeforeBuildForRts(ComputeGraphPtr &compute_gr } auto graph_optimizer = instance_ptr->OpsKernelManagerObj().GetAllGraphOptimizerObjsByPriority(); - GELOGI("optimize by opskernel in graph optimize before build phase. num of graph_optimizer is %zu.", + GELOGD("optimize by opskernel in graph optimize before build phase. num of graph_optimizer is %zu.", graph_optimizer.size()); Status ret = SUCCESS; string exclude_core_Type = (core_type_ == kVectorCore) ? kAicoreEngine : kVectorEngine; - GELOGI("[OptimizeGraphBeforeBuildForRts]: engine type will exclude: %s, core_type_: %s", + GELOGD("[OptimizeGraphBeforeBuildForRts]: engine type will exclude: %s, core_type_: %s", exclude_core_Type.c_str(), core_type_.c_str()); if (graph_optimizer.size() != 0) { for (auto iter = graph_optimizer.begin(); iter != graph_optimizer.end(); ++iter) { diff --git a/ge/graph/optimize/optimizer/allreduce_fusion_pass.cc b/ge/graph/optimize/optimizer/allreduce_fusion_pass.cc deleted file mode 100644 index 34c3a0de..00000000 --- a/ge/graph/optimize/optimizer/allreduce_fusion_pass.cc +++ /dev/null @@ -1,397 +0,0 @@ -/** - * Copyright 2020 Huawei Technologies Co., Ltd - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * 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. - */ - -#include "graph/optimize/optimizer/allreduce_fusion_pass.h" -#include -#include "common/debug/log.h" -#include "framework/common/debug/ge_log.h" -#include "common/types.h" -#include "common/util.h" -#include "graph/anchor.h" -#include "graph/node.h" -#include "graph/op_desc.h" -#include "graph/utils/attr_utils.h" -#include "graph/utils/graph_utils.h" -#include "graph/utils/tensor_utils.h" -#include "graph/debug/ge_attr_define.h" -#include "hccl/base.h" -#include "hccl/hcom.h" - -namespace ge { -Status AllReducePass::Run(ge::ComputeGraphPtr graph) { - GELOGI("FusionAllReducePass: start"); - std::vector fusionOps; - std::vector inputGradientSize; - std::vector inputGradientTime; - - static const float inputGradientSizeTemp = 0.0; - static const float inputGradientTimeTemp = 0.0; - - // Get all nodes - for (auto nodePtr : graph->GetDirectNode()) { - GE_IF_BOOL_EXEC(nullptr == nodePtr, GELOGW("FusionAllReducePass: null node exists"); continue;); - - ge::OpDescPtr opDescPtr = nodePtr->GetOpDesc(); - GE_IF_BOOL_EXEC(nullptr == opDescPtr, - GELOGW("FusionAllReducePass: desc of node %s is null", nodePtr->GetName().c_str()); - continue;) - GE_IF_BOOL_EXEC(HCOMALLREDUCE == opDescPtr->GetType(), - // the op is allreduce and fusion > 0, then run fusion - std::int64_t hcom_fusion = 1; - GE_IF_BOOL_EXEC(!ge::AttrUtils::GetInt(opDescPtr, HCOM_ATTR_FUSION, hcom_fusion), - GELOGW("FusionAllReducePass: not get hcom_fusion from opDescPtr " - "by HCOM_ATTR_FUSION")); - GELOGI("after GetInt, hcom_fusion is :%ld", hcom_fusion); GE_IF_BOOL_EXEC( - hcom_fusion > 0, fusionOps.push_back(nodePtr); inputGradientSize.push_back(inputGradientSizeTemp); - inputGradientTime.push_back(inputGradientTimeTemp);)) - } - // The number of allredecue operator must be more than 1 - GE_IF_BOOL_EXEC(1 >= fusionOps.size(), GELOGW("FusionAllReducePass NOT_CHANGED: the graph has " - "%lu allreduce operator", - fusionOps.size()); - return NOT_CHANGED;); - - string group = "group"; - u32 gradientNum = fusionOps.size(); - string model_name_str = graph->GetName(); - const char *model_name = model_name_str.c_str(); - model_feature modelFeature{model_name, gradientNum, inputGradientSize.data(), inputGradientTime.data()}; - - u32 segmentNum = 0; - u32 segmentIndex[HCCL_MAX_SEGMENT_NUM] = {}; - - // Call HCCL function: hcom_gradient_segment - GELOGI("FusionAllReducePass: invoking hcom_get_split_strategy"); - GE_IF_BOOL_EXEC(HCCL_SUCCESS != hcom_get_split_strategy(group.c_str(), &modelFeature, HCCL_MAX_SEGMENT_NUM, - &segmentNum, segmentIndex), - GELOGE(FAILED, "FusionAllReducePass FAILED: the graph has %lu allreduce operator", fusionOps.size()); - return FAILED;) - GELOGI("FusionAllReducePass: invoke hcom_get_split_strategy successfully"); - - // check whether segmentNum is legal or not - GE_IF_BOOL_EXEC((HCCL_MAX_SEGMENT_NUM < segmentNum || 1 > segmentNum || segmentNum > gradientNum), - GELOGE(FAILED, - "FusionAllReducePass FAILED: illegal segmentNum=%u, " - "HCCL_MAX_SEGMENT_NUM=%u, gradientNum=%u", - segmentNum, HCCL_MAX_SEGMENT_NUM, gradientNum); - return FAILED;); - - // check whether segmentIndex is legal or not - GE_IF_BOOL_EXEC((segmentIndex[segmentNum - 1] != gradientNum - 1), - GELOGE(FAILED, - "FusionAllReducePass FAILED: illegal segmentIndex[0]=%u, " - "segmentIndex[segmentNum-1]=%u, gradientNum=%u", - segmentIndex[0], segmentIndex[(segmentNum)-1], gradientNum); - return FAILED;); - - for (uint32_t i = 0; i < segmentNum - 1; i++) { - GE_IF_BOOL_EXEC(segmentIndex[i] >= segmentIndex[i + 1], GELOGE(FAILED, - "FusionAllReducePass FAILED: illegal " - "segmentIndex[%u]=%u, segmentIndex[%u]=%u", - i, segmentIndex[i], i + 1, segmentIndex[i + 1]); - return FAILED;); - } - - // check whether fusion is needed or not - GE_IF_BOOL_EXEC( - segmentNum == gradientNum, - GELOGE(NOT_CHANGED, "FusionAllReducePass NOT_CHANGED: segmentNum=%u, gradientNum=%u", segmentNum, gradientNum); - return NOT_CHANGED;) - - std::unordered_set anchorPtrSet; - std::vector fusionOpPeerOutDataAnchor; - std::vector fusionOpPeerOutDataToInControl; - std::vector fusionOpPeerOutControlAnchor; - std::vector> fusionOpPeerInDataAnchor; - std::vector> fusionOpPeerInControlFromOutData; - std::vector fusionOpPeerInControlAnchor; - ge::OutControlAnchorPtr previousNewAllreduceOutControlAnchor = nullptr; - - // Traversing the segmentNum - uint32_t start = 0; - uint32_t end = 0; - for (uint32_t segmentIdx = 0; segmentIdx < segmentNum; segmentIdx++) { - end = segmentIndex[segmentIdx]; - GE_IF_BOOL_EXEC(end - start < 1, - GELOGI("FusionAllReducePass: segmentIndex[%u]=%u", segmentIdx, segmentIndex[segmentIdx]); - start = end + 1; continue;); - - ge::OpDescPtr originDescPtr = fusionOps[start]->GetOpDesc(); - GE_CHECK_NOTNULL(originDescPtr); - ge::OpDescPtr newAllreduceDesc = AttrUtils::CloneOpDesc(originDescPtr); - GE_CHECK_NOTNULL(newAllreduceDesc); - - // Cleat buffer - anchorPtrSet.clear(); - fusionOpPeerOutDataAnchor.clear(); - fusionOpPeerOutDataToInControl.clear(); - fusionOpPeerOutControlAnchor.clear(); - fusionOpPeerInDataAnchor.clear(); - fusionOpPeerInControlFromOutData.clear(); - fusionOpPeerInControlAnchor.clear(); - - // Traversing the Allreduce operators of each group - int outDataAnchorIndex = 0; - GE_CHK_STATUS_RET(GetPeerOutDataToInData(anchorPtrSet, fusionOpPeerOutDataAnchor, fusionOps[start]), - "Get peer outDataAnchor to inDataAnchor failed"); - - GE_CHK_STATUS_RET(GetPeerInAnchorToOutData(anchorPtrSet, fusionOpPeerInDataAnchor, fusionOpPeerInControlFromOutData, - fusionOps[start]), - "Get peer inDataAnchor and inControlAnchor to outDataAnchor failed"); - - GE_CHK_STATUS_RET(GetPeerOutDataToInControl(anchorPtrSet, fusionOpPeerOutDataToInControl, fusionOps[start]), - "Get peer outDataAnchor to inControlAnchor failed"); - GE_CHK_STATUS_RET(GetPeerOutControlToInControl(anchorPtrSet, fusionOpPeerOutControlAnchor, fusionOps[start]), - "Get peer outControlAnchor to inControlAnchor failed"); - GE_CHK_STATUS_RET(GetPeerInControlFromOutControl(anchorPtrSet, fusionOpPeerInControlAnchor, fusionOps[start]), - "Get peer outControlAnchor from inControlAnchor failed"); - GE_CHK_STATUS_RET(graph->RemoveNode(fusionOps[start]), "FusionAllReducePass FAILED: remove node %s\n.", - fusionOps[start]->GetName().c_str()); - - for (uint32_t idx = start + 1; idx <= end; idx++) { - GE_CHK_STATUS_RET( - GetPeerOutDataToInData(anchorPtrSet, fusionOpPeerOutDataAnchor, fusionOps[idx], newAllreduceDesc), - "Get peer outDataAnchor to inDataAnchor failed"); - GE_CHK_STATUS_RET(GetPeerOutDataToInControl(anchorPtrSet, fusionOpPeerOutDataToInControl, fusionOps[idx]), - "Get peer outDataAnchor to inControlAnchor failed"); - GE_CHK_STATUS_RET(GetPeerOutControlToInControl(anchorPtrSet, fusionOpPeerOutControlAnchor, fusionOps[idx]), - "Get peer outControlAnchor to inControlAnchor failed"); - GE_CHK_STATUS_RET( - GetPeerAnchorFromOutData(anchorPtrSet, fusionOpPeerInDataAnchor, fusionOpPeerInControlFromOutData, - fusionOps[idx], newAllreduceDesc, outDataAnchorIndex), - "Get peerAnchor from outDataAnchor failed"); - GE_CHK_STATUS_RET(GetPeerInControlFromOutControl(anchorPtrSet, fusionOpPeerInControlAnchor, fusionOps[idx]), - "Get peer outControlAnchor from inControlAnchor failed"); - - // Delete the node - GE_CHK_STATUS_RET(graph->RemoveNode(fusionOps[idx]), "FusionAllReducePass FAILED: remove node %s\n.", - fusionOps[idx]->GetName().c_str()); - } - - NodePtr newAllReducePtr = graph->AddNode(newAllreduceDesc); - GE_CHECK_NOTNULL(newAllReducePtr); - // Link the inputDataAnchor - for (uint32_t i = 0; i < fusionOpPeerOutDataAnchor.size(); i++) { - GE_CHK_STATUS_RET( - GraphUtils::AddEdge(fusionOpPeerOutDataAnchor[i], newAllReducePtr->GetInDataAnchor(static_cast(i))), - "FusionAllReducePass FAILED: add input data edge failed"); - } - - // Link the inputControlAnchor - for (uint32_t i = 0; i < fusionOpPeerOutControlAnchor.size(); i++) { - GE_CHK_STATUS_RET(GraphUtils::AddEdge(fusionOpPeerOutControlAnchor[i], newAllReducePtr->GetInControlAnchor()), - "FusionAllReducePass FAILED: add input control edge failed"); - } - - for (uint32_t i = 0; i < fusionOpPeerOutDataToInControl.size(); i++) { - GE_CHK_STATUS_RET(GraphUtils::AddEdge(fusionOpPeerOutDataToInControl[i], newAllReducePtr->GetInControlAnchor()), - "FusionAllReducePass FAILED: add edge from out data to incontrol " - "failed"); - } - - // Link the outputDataAnchor - for (uint32_t i = 0; i < fusionOpPeerInDataAnchor.size(); i++) { - auto peerInDataAnchor = fusionOpPeerInDataAnchor[i].second; - GE_CHK_STATUS_RET( - GraphUtils::AddEdge(newAllReducePtr->GetOutDataAnchor(fusionOpPeerInDataAnchor[i].first), peerInDataAnchor), - "FusionAllReducePass FAILED: add output data edge failed"); - } - for (uint32_t i = 0; i < fusionOpPeerInControlFromOutData.size(); i++) { - auto peerInControlAnchor = fusionOpPeerInControlFromOutData[i].second; - GE_CHK_STATUS_RET( - GraphUtils::AddEdge(newAllReducePtr->GetOutDataAnchor(fusionOpPeerInControlFromOutData[i].first), - peerInControlAnchor), - "FusionAllReducePass FAILED: add edge from out data to in control " - "failed"); - } - - // Link the outputControlAnchor - for (uint32_t i = 0; i < fusionOpPeerInControlAnchor.size(); i++) { - GE_CHK_STATUS_RET(GraphUtils::AddEdge(newAllReducePtr->GetOutControlAnchor(), fusionOpPeerInControlAnchor[i]), - "FusionAllReducePass FAILED: add output control edge failed"); - } - - // Link the newAllreduce - if (segmentIdx > 0 && previousNewAllreduceOutControlAnchor != nullptr) { - GE_CHK_STATUS_RET( - GraphUtils::AddEdge(previousNewAllreduceOutControlAnchor, newAllReducePtr->GetInControlAnchor()), - "FusionAllReducePass FAILED: add input previous control edge failed"); - } - - previousNewAllreduceOutControlAnchor = newAllReducePtr->GetOutControlAnchor(); - start = end + 1; - } - - return SUCCESS; -} - -Status AllReducePass::GetPeerOutDataToInData(std::unordered_set &anchorSet, - vector &peerOutDataAnchorVec, - ge::NodePtr &srcNodePtr) { - for (auto inDataAnchor : srcNodePtr->GetAllInDataAnchors()) { - GE_IF_BOOL_EXEC(inDataAnchor == nullptr, continue;); - OutDataAnchorPtr peerOutDataAnchor = inDataAnchor->GetPeerOutAnchor(); - GE_IF_BOOL_EXEC(peerOutDataAnchor == nullptr, continue;); - if (anchorSet.count(peerOutDataAnchor.get()) == 0) { - peerOutDataAnchorVec.push_back(peerOutDataAnchor); - anchorSet.insert(peerOutDataAnchor.get()); - GE_CHK_STATUS_RET(GraphUtils::RemoveEdge(peerOutDataAnchor, inDataAnchor)); - } - } - return SUCCESS; -} - -Status AllReducePass::GetPeerInAnchorToOutData( - std::unordered_set &anchorSet, std::vector> &fusionOpPeerInDataAnchor, - std::vector> &fusionOpPeerInControlFromOutData, ge::NodePtr &srcNodePtr) { - for (auto outDataAnchor : srcNodePtr->GetAllOutDataAnchors()) { - GE_IF_BOOL_EXEC(outDataAnchor == nullptr, continue;); - for (auto peerInDataAnchor : outDataAnchor->GetPeerInDataAnchors()) { - GE_IF_BOOL_EXEC(peerInDataAnchor == nullptr, continue;); - if (anchorSet.count(peerInDataAnchor.get()) == 0) { - std::pair pairPeerInDataAnchor; - pairPeerInDataAnchor.first = 0; - pairPeerInDataAnchor.second = peerInDataAnchor; - fusionOpPeerInDataAnchor.push_back(pairPeerInDataAnchor); - anchorSet.insert(peerInDataAnchor.get()); - GE_CHK_STATUS_RET(GraphUtils::RemoveEdge(outDataAnchor, peerInDataAnchor)); - } - } - - for (auto peerInControlAnchorFromData : outDataAnchor->GetPeerInControlAnchors()) { - GE_IF_BOOL_EXEC(peerInControlAnchorFromData == nullptr, continue;); - if (anchorSet.count(peerInControlAnchorFromData.get()) == 0) { - std::pair pairPeerInControlAnchorFromData; - pairPeerInControlAnchorFromData.first = 0; - pairPeerInControlAnchorFromData.second = peerInControlAnchorFromData; - fusionOpPeerInControlFromOutData.push_back(pairPeerInControlAnchorFromData); - anchorSet.insert(peerInControlAnchorFromData.get()); - GE_CHK_STATUS_RET(GraphUtils::RemoveEdge(outDataAnchor, peerInControlAnchorFromData)); - } - } - } - return SUCCESS; -} - -Status AllReducePass::GetPeerOutDataToInData(std::unordered_set &anchorSet, - vector &peerOutDataAnchorVec, - ge::NodePtr &srcNodePtr, ge::OpDescPtr &dstOpDescPtr) { - for (auto inDataAnchor : srcNodePtr->GetAllInDataAnchors()) { - GE_IF_BOOL_EXEC(inDataAnchor == nullptr, continue;); - OutDataAnchorPtr peerOutDataAnchor = inDataAnchor->GetPeerOutAnchor(); - GE_IF_BOOL_EXEC(peerOutDataAnchor == nullptr, continue;); - if (anchorSet.count(peerOutDataAnchor.get()) == 0) { - peerOutDataAnchorVec.push_back(peerOutDataAnchor); - anchorSet.insert(peerOutDataAnchor.get()); - if (dstOpDescPtr->AddInputDesc(inDataAnchor->GetOwnerNode()->GetOpDesc()->GetInputDesc(inDataAnchor->GetIdx())) != - ge::GRAPH_SUCCESS) { - GELOGW("GetPeerOutDataToInData: AddInputDesc failed"); - } - GE_CHK_STATUS_RET(GraphUtils::RemoveEdge(peerOutDataAnchor, inDataAnchor)); - } - } - return SUCCESS; -} - -Status AllReducePass::GetPeerOutDataToInControl(std::unordered_set &anchorSet, - vector &peerOutDataToInControlVec, - ge::NodePtr &srcNodePtr) { - InControlAnchorPtr inControlAnchor = srcNodePtr->GetInControlAnchor(); - GE_CHECK_NOTNULL(inControlAnchor); - for (auto peerOutDataToInControl : inControlAnchor->GetPeerOutDataAnchors()) { - GE_IF_BOOL_EXEC(peerOutDataToInControl == nullptr, continue;); - if (anchorSet.count(peerOutDataToInControl.get()) == 0) { - peerOutDataToInControlVec.push_back(peerOutDataToInControl); - anchorSet.insert(peerOutDataToInControl.get()); - GE_CHK_STATUS_RET(GraphUtils::RemoveEdge(peerOutDataToInControl, inControlAnchor)); - } - } - return SUCCESS; -} - -Status AllReducePass::GetPeerOutControlToInControl(std::unordered_set &anchorSet, - vector &peerOutControlToInControlVec, - ge::NodePtr &srcNodePtr) { - InControlAnchorPtr inControlAnchor = srcNodePtr->GetInControlAnchor(); - GE_CHECK_NOTNULL(inControlAnchor); - for (auto peerOutControlAnchor : inControlAnchor->GetPeerOutControlAnchors()) { - GE_IF_BOOL_EXEC(peerOutControlAnchor == nullptr, continue;); - if (anchorSet.count(peerOutControlAnchor.get()) == 0) { - peerOutControlToInControlVec.push_back(peerOutControlAnchor); - anchorSet.insert(peerOutControlAnchor.get()); - GE_CHK_STATUS_RET(GraphUtils::RemoveEdge(peerOutControlAnchor, inControlAnchor)); - } - } - return SUCCESS; -} - -Status AllReducePass::GetPeerAnchorFromOutData( - std::unordered_set &anchorSet, vector> &peerInDataFromOutDataVec, - vector> &peerInControlFromOutDataVec, ge::NodePtr &srcNodePtr, - ge::OpDescPtr &dstOpDescPtr, int &index) { - for (auto outDataAnchor : srcNodePtr->GetAllOutDataAnchors()) { - GE_IF_BOOL_EXEC(outDataAnchor == nullptr, continue;) - if (outDataAnchor->GetPeerInDataAnchors().size() > 0 || outDataAnchor->GetPeerInControlAnchors().size() > 0) { - if (dstOpDescPtr->AddOutputDesc( - outDataAnchor->GetOwnerNode()->GetOpDesc()->GetOutputDesc(outDataAnchor->GetIdx())) != ge::GRAPH_SUCCESS) { - GELOGW("GetPeerAnchorFromOutData: AddOutputDesc failed"); - } - index++; - } - - for (auto peerInDataAnchor : outDataAnchor->GetPeerInDataAnchors()) { - GE_IF_BOOL_EXEC(peerInDataAnchor == nullptr, continue;) - if (anchorSet.count(peerInDataAnchor.get()) == 0) { - std::pair pairPeerInDataAnchor; - pairPeerInDataAnchor.first = index; - pairPeerInDataAnchor.second = peerInDataAnchor; - peerInDataFromOutDataVec.push_back(pairPeerInDataAnchor); - anchorSet.insert(peerInDataAnchor.get()); - GE_CHK_STATUS_RET(GraphUtils::RemoveEdge(outDataAnchor, peerInDataAnchor)) - } - } - - for (auto peerInControlAnchorFromData : outDataAnchor->GetPeerInControlAnchors()) { - GE_IF_BOOL_EXEC(peerInControlAnchorFromData == nullptr, continue;) - if (anchorSet.count(peerInControlAnchorFromData.get()) == 0) { - std::pair pairPeerInControlAnchorFromData; - pairPeerInControlAnchorFromData.first = index; - pairPeerInControlAnchorFromData.second = peerInControlAnchorFromData; - peerInControlFromOutDataVec.push_back(pairPeerInControlAnchorFromData); - anchorSet.insert(peerInControlAnchorFromData.get()); - GE_CHK_STATUS_RET(GraphUtils::RemoveEdge(outDataAnchor, peerInControlAnchorFromData)) - } - } - } - return SUCCESS; -} - -Status AllReducePass::GetPeerInControlFromOutControl(std::unordered_set &anchorSet, - vector &peerInControlFromOutControlVec, - ge::NodePtr &srcNodePtr) { - OutControlAnchorPtr outControlAnchor = srcNodePtr->GetOutControlAnchor(); - GE_CHECK_NOTNULL(outControlAnchor); - for (auto peerInControlAnchor : outControlAnchor->GetPeerInControlAnchors()) { - GE_IF_BOOL_EXEC(peerInControlAnchor == nullptr, continue;) - if (anchorSet.count(peerInControlAnchor.get()) == 0) { - peerInControlFromOutControlVec.push_back(peerInControlAnchor); - anchorSet.insert(peerInControlAnchor.get()); - GE_CHK_STATUS_RET(GraphUtils::RemoveEdge(outControlAnchor, peerInControlAnchor)) - } - } - return SUCCESS; -} -} // namespace ge diff --git a/ge/graph/optimize/optimizer/allreduce_fusion_pass.h b/ge/graph/optimize/optimizer/allreduce_fusion_pass.h deleted file mode 100644 index 8b2168d9..00000000 --- a/ge/graph/optimize/optimizer/allreduce_fusion_pass.h +++ /dev/null @@ -1,56 +0,0 @@ -/** - * Copyright 2020 Huawei Technologies Co., Ltd - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * 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. - */ - -#ifndef GE_GRAPH_OPTIMIZE_OPTIMIZER_ALLREDUCE_FUSION_PASS_H_ -#define GE_GRAPH_OPTIMIZE_OPTIMIZER_ALLREDUCE_FUSION_PASS_H_ - -#include -#include -#include -#include "inc/graph_pass.h" - -namespace ge { -// -class AllReducePass : public GraphPass { - public: - Status Run(ge::ComputeGraphPtr graph) override; - - private: - Status GetPeerOutDataToInData(std::unordered_set &anchorSet, - vector &peerOutDataAnchorVec, ge::NodePtr &srcNodePtr, - ge::OpDescPtr &dstOpDescPtr); - Status GetPeerOutDataToInControl(std::unordered_set &anchorSet, - vector &peerOutDataToInControlVec, ge::NodePtr &srcNodePtr); - Status GetPeerOutControlToInControl(std::unordered_set &anchorSet, - vector &peerOutControlToInControlVec, - ge::NodePtr &srcNodePtr); - Status GetPeerAnchorFromOutData(std::unordered_set &anchorSet, - vector> &peerInDataFromOutDataVec, - vector> &peerInControlFromOutDataVec, - ge::NodePtr &srcNodePtr, ge::OpDescPtr &dstOpDescPtr, int &index); - Status GetPeerInControlFromOutControl(std::unordered_set &anchorSet, - vector &peerInControlFromOutControlVec, - ge::NodePtr &srcNodePtr); - Status GetPeerOutDataToInData(std::unordered_set &anchorSet, - std::vector &peerOutDataAnchorVec, - ge::NodePtr &srcNodePtr); - Status GetPeerInAnchorToOutData(std::unordered_set &anchorSet, - std::vector> &fusionOpPeerInDataAnchor, - std::vector>&fusionOpPeerInControlFromOutData, - ge::NodePtr &srcNodePtr); -}; -} // namespace ge -#endif // GE_GRAPH_OPTIMIZE_OPTIMIZER_ALLREDUCE_FUSION_PASS_H_ diff --git a/ge/graph/partition/engine_place.cc b/ge/graph/partition/engine_place.cc index cdf29e56..f4ebbdca 100755 --- a/ge/graph/partition/engine_place.cc +++ b/ge/graph/partition/engine_place.cc @@ -50,7 +50,7 @@ Status EnginePlacer::Check() const { Status EnginePlacer::Run() { std::lock_guard lock(check_support_cost_mutex); - GELOGI("Engine placer starts."); + GELOGD("Engine placer starts."); if (Check() != SUCCESS) { return FAILED; } @@ -101,7 +101,7 @@ Status EnginePlacer::Run() { for (auto &it : ge::GELib::GetInstance()->DNNEngineManagerObj().GetCheckSupportCost()) { GEEVENT("The time cost of %s::CheckSupported is [%lu] micro second.", it.first.c_str(), it.second); } - GELOGI("Engine placer ends."); + GELOGD("Engine placer ends."); return is_check_support_success ? SUCCESS : FAILED; } diff --git a/ge/graph/partition/graph_partition.cc b/ge/graph/partition/graph_partition.cc index a918bdb6..9e053020 100755 --- a/ge/graph/partition/graph_partition.cc +++ b/ge/graph/partition/graph_partition.cc @@ -223,7 +223,7 @@ Status ge::GraphPartitioner::MergeSubGraph(ge::ComputeGraphPtr &output_merged_co GELOGE(GE_GRAPH_UNSUPPORTED, "Cannot call merging in partition mode"); return FAILED; } - GELOGI("Graph merge starts."); + GELOGD("Graph merge starts."); // check input param for (const auto &it : sub_graph_list) { if (it == nullptr) { @@ -261,7 +261,7 @@ Status ge::GraphPartitioner::MergeSubGraph(ge::ComputeGraphPtr &output_merged_co return FAILED; } GE_TIMESTAMP_END(MergeSubGraphEnginePlacerRun, "GraphPartitioner::MergeGraphEnginePlacerRun"); - GELOGI("Graph merge ends."); + GELOGD("Graph merge ends."); return SUCCESS; } @@ -581,7 +581,7 @@ Status ge::GraphPartitioner::Initialize(ge::ComputeGraphPtr compute_graph) { new_cluster->engine_name_.c_str(), new_cluster->index_, new_cluster->stream_label_.c_str()); temp_index++; } - GELOGI("Initialize ends."); + GELOGD("Initialize ends."); return SUCCESS; } @@ -754,11 +754,11 @@ void ge::GraphPartitioner::MarkClusters() { } } } - GELOGI("MarkClusters ends."); + GELOGD("MarkClusters ends."); } Status ge::GraphPartitioner::SplitSubGraphs(ge::ComputeGraphPtr compute_graph) { - GELOGI("SplitSubGraphs starts."); + GELOGD("SplitSubGraphs starts."); if (compute_graph == nullptr) { GELOGE(FAILED, "parameter ptr is null."); return FAILED; @@ -823,7 +823,7 @@ Status ge::GraphPartitioner::SplitSubGraphs(ge::ComputeGraphPtr compute_graph) { } } } - GELOGI("SplitSubGraphs ends."); + GELOGD("SplitSubGraphs ends."); return SUCCESS; } diff --git a/ge/graph/passes/atomic_addr_clean_pass.cc b/ge/graph/passes/atomic_addr_clean_pass.cc index 690dee27..60742eb1 100755 --- a/ge/graph/passes/atomic_addr_clean_pass.cc +++ b/ge/graph/passes/atomic_addr_clean_pass.cc @@ -46,7 +46,7 @@ Status AtomicAddrCleanPass::Run(ComputeGraphPtr graph) { } } if (atomic_node_vec.empty()) { - GELOGI("There is no atomic node. Ignore atomicAddrClean pass."); + GELOGD("There is no atomic node. Ignore atomicAddrClean pass."); return SUCCESS; } diff --git a/ge/graph/passes/cond_remove_pass.cc b/ge/graph/passes/cond_remove_pass.cc index 8df7f8ac..bf2e1170 100644 --- a/ge/graph/passes/cond_remove_pass.cc +++ b/ge/graph/passes/cond_remove_pass.cc @@ -332,7 +332,7 @@ Status CondRemovePass::GetCondInfo(const NodePtr &node, ComputeGraphPtr &graph, return FAILED; } } else { - GELOGI("no need cond_remove_pass for node %s.", node->GetName().c_str()); + GELOGD("no need cond_remove_pass for node %s.", node->GetName().c_str()); return NOT_CHANGED; } diff --git a/ge/graph/passes/mark_agnostic_pass.cc b/ge/graph/passes/mark_agnostic_pass.cc index 00a3dad9..4fdc8e1b 100644 --- a/ge/graph/passes/mark_agnostic_pass.cc +++ b/ge/graph/passes/mark_agnostic_pass.cc @@ -16,6 +16,7 @@ #include "graph/passes/mark_agnostic_pass.h" #include "graph/utils/node_utils.h" +#include "graph/utils/tensor_utils.h" namespace ge { Status MarkAgnosticPass::Run(ComputeGraphPtr graph) { @@ -47,6 +48,16 @@ Status MarkAgnosticPass::Run(ComputeGraphPtr graph) { } if (node_type == MERGE) { GELOGD("Mark format agnostic and continuous for merge node %s", node->GetName().c_str()); + auto in_nodes = node->GetInAllNodes(); + vector input_nodes(in_nodes.begin(), in_nodes.end()); + /// Enter-----------+ + /// +-> Merge + /// NextIteration---+ + if (input_nodes.size() == 2) { + if (input_nodes[0]->GetType() == ENTER && input_nodes[1]->GetType() == NEXTITERATION) { + continue; + } + } const OpDescPtr op_desc = node->GetOpDesc(); const GeTensorDescPtr op_tensor = op_desc->MutableOutputDesc(0); if (op_tensor == nullptr) { diff --git a/ge/graph/passes/memcpy_addr_async_pass.cc b/ge/graph/passes/memcpy_addr_async_pass.cc index 75437157..b930f7cb 100755 --- a/ge/graph/passes/memcpy_addr_async_pass.cc +++ b/ge/graph/passes/memcpy_addr_async_pass.cc @@ -278,7 +278,7 @@ Status MemcpyAddrAsyncPass::InsertMemcpyAddrAsyncNode(const OutDataAnchorPtr &ou } Status MemcpyAddrAsyncPass::InsertMemAddrAsyncNodeBeforeNetoutput(const ComputeGraphPtr &graph, const NodePtr &node) { - GELOGI("Start AddMemcpyAddrAsyncNode for %s.", node->GetName().c_str()); + GELOGD("Start AddMemcpyAddrAsyncNode for %s.", node->GetName().c_str()); for (const auto &in_data_anchor : node->GetAllInDataAnchors()) { auto in_node = NodeUtils::GetInDataNodeByIndex(*node, in_data_anchor->GetIdx()); GE_CHECK_NOTNULL(in_node); diff --git a/ge/graph/passes/multi_batch_pass.cc b/ge/graph/passes/multi_batch_pass.cc index 8a38f83a..74f7e30e 100644 --- a/ge/graph/passes/multi_batch_pass.cc +++ b/ge/graph/passes/multi_batch_pass.cc @@ -33,7 +33,7 @@ Status MultiBatchPass::Run(ComputeGraphPtr graph) { OutDataAnchorPtr pred_value = nullptr; Status ret = FindPredValue(graph, pred_value); if (ret == NOT_CHANGED) { - GELOGI("SwitchN node not exist, graph not changed."); + GELOGD("SwitchN node not exist, graph not changed."); return SUCCESS; } if (ret != SUCCESS) { @@ -158,7 +158,7 @@ Status MultiBatchPass::FindPredValue(const ComputeGraphPtr &graph, OutDataAnchor } if (switch_n_nodes_.empty()) { - GELOGI("SwitchN node not exist."); + GELOGD("SwitchN node not exist."); return NOT_CHANGED; } diff --git a/ge/graph/passes/set_input_output_offset_pass.cc b/ge/graph/passes/set_input_output_offset_pass.cc index beac831c..ec41d6be 100644 --- a/ge/graph/passes/set_input_output_offset_pass.cc +++ b/ge/graph/passes/set_input_output_offset_pass.cc @@ -128,7 +128,7 @@ Status SetInputOutputOffsetPass::SetInputOffsetForHcom(const ge::NodePtr &node, } Status SetInputOutputOffsetPass::SetInputOffset(const NodePtr &node, const vector &connect_input) { - GELOGI("Start to SetInputOffset for %s.", node->GetName().c_str()); + GELOGD("Start to SetInputOffset for %s.", node->GetName().c_str()); std::vector memory_type; auto op_desc = node->GetOpDesc(); (void)ge::AttrUtils::GetListInt(op_desc, ATTR_NAME_INPUT_MEM_TYPE_LIST, memory_type); @@ -241,7 +241,7 @@ Status SetInputOutputOffsetPass::SetOutputOffsetForHcom(const NodePtr &node, con } Status SetInputOutputOffsetPass::SetOutputOffset(const NodePtr &node, const vector &connect_output) { - GELOGI("Start SetOutputOffset of %s.", node->GetName().c_str()); + GELOGD("Start SetOutputOffset of %s.", node->GetName().c_str()); bool attr_no_task = false; bool get_attr_no_task = ge::AttrUtils::GetBool(node->GetOpDesc(), ATTR_NAME_NOTASK, attr_no_task); if (get_attr_no_task && attr_no_task) { diff --git a/ge/graph/preprocess/graph_preprocess.cc b/ge/graph/preprocess/graph_preprocess.cc index 58718c9f..b899ee83 100644 --- a/ge/graph/preprocess/graph_preprocess.cc +++ b/ge/graph/preprocess/graph_preprocess.cc @@ -117,6 +117,7 @@ #include "graph/passes/variable_op_pass.h" #include "graph/passes/variable_prepare_op_pass.h" #include "graph/passes/variable_ref_delete_op_pass.h" +#include "graph/passes/mark_agnostic_pass.h" namespace ge { @@ -1700,6 +1701,7 @@ Status GraphPrepare::PrepareOptimize() { try { (void)original_graph_passes.AddPass("PrepareOptimize::ShapeOperateOpRemovePass", new ShapeOperateOpRemovePass); (void)original_graph_passes.AddPass("PrepareOptimize::ReplaceTransShapePass", new ReplaceTransShapePass); + (void)original_graph_passes.AddPass("PrepareOptimize::MarkAgnosticPass" , new MarkAgnosticPass); } catch (std::bad_alloc &e) { GELOGE(INTERNAL_ERROR, "Add pass failed, bad memory allocation occurs."); return INTERNAL_ERROR; diff --git a/ge/graph/preprocess/multi_batch_copy_graph.cc b/ge/graph/preprocess/multi_batch_copy_graph.cc index c3647b93..9ab74d70 100644 --- a/ge/graph/preprocess/multi_batch_copy_graph.cc +++ b/ge/graph/preprocess/multi_batch_copy_graph.cc @@ -1571,6 +1571,10 @@ void GetDynamicShapeByMerge(const ComputeGraphPtr &graph, const NodePtr &node, // Connect NetOutput directly void GetDirectOutputShape(const ComputeGraphPtr &graph, const NodePtr &node, const set &dynamic_output_index, vector &dynamic_output_dims) { + if (!GetLocalOmgContext().dynamic_node_type.empty()) { + GELOGD("No need to get directly shape info of %s when train.", node->GetName().c_str()); + return; + } GELOGD("Try get directly shape info, Graph: %s, Node: %s", graph->GetName().c_str(), node->GetName().c_str()); const auto &netoutput_desc = node->GetOpDesc(); const auto &inputnode_to_netoutput = node->GetInAllNodes(); @@ -1578,9 +1582,6 @@ void GetDirectOutputShape(const ComputeGraphPtr &graph, const NodePtr &node, if (dynamic_output_index.count(i) > 0) { continue; } - if (inputnode_to_netoutput.at(i)->GetType() == GETDYNAMICDIMS) { - continue; - } auto tensor_desc = netoutput_desc->GetInputDesc(i); auto shape = tensor_desc.GetShape().ToString(); diff --git a/ge/graph/preprocess/multi_batch_options.cc b/ge/graph/preprocess/multi_batch_options.cc index c168e077..f33c2983 100644 --- a/ge/graph/preprocess/multi_batch_options.cc +++ b/ge/graph/preprocess/multi_batch_options.cc @@ -84,8 +84,10 @@ Status DistinguishGetNextAndData(ComputeGraphPtr &graph, vector &data_n if (op_desc->GetType() == DATA && op_desc->GetName() != kShapeDataName) { if (op_desc->GetName().find(kSubstrOfGetNextNosinkName) == string::npos) { data_nodes.emplace_back(input_node); + GELOGD("Name of data node is %s.", op_desc->GetName().c_str()); } else { getnext_nosink_nodes.emplace_back(input_node); + GELOGD("Name of getnext nosink is %s.", op_desc->GetName().c_str()); } } if (IsGetNextType(input_node)) { @@ -111,6 +113,8 @@ Status CheckSequenceOfData(ComputeGraphPtr &graph, const vector &data_n GE_CHECK_NOTNULL(data_node->GetOpDesc()); auto output_shape = data_node->GetOpDesc()->GetOutputDesc(0).GetShape().GetDims(); auto dynamic_dims = GetLocalOmgContext().user_input_dims.at(i).second; + GELOGD("The %zu data node is %s, node shape is %s, dynamic dim is %s.", i, data_node->GetName().c_str(), + formats::JoinToString(output_shape).c_str(), formats::JoinToString(dynamic_dims).c_str()); if (output_shape.empty() && dynamic_dims.size() == 1 && dynamic_dims.at(0) == 0) { GELOGI("No need to check sequence for constant."); continue; @@ -151,6 +155,8 @@ Status CheckSequenceOfGetnext(ComputeGraphPtr &graph, const vector &get for (size_t i = 0; i < data_count; ++i) { auto output_shape = data_node->GetOpDesc()->GetOutputDesc(i).GetShape().GetDims(); auto dynamic_dims = GetLocalOmgContext().user_input_dims.at(i).second; + GELOGD("The %zu getnext node is %s, node shape is %s, dynamic dim is %s.", i, data_node->GetName().c_str(), + formats::JoinToString(output_shape).c_str(), formats::JoinToString(dynamic_dims).c_str()); if (output_shape.empty() && dynamic_dims.size() == 1 && dynamic_dims.at(0) == 0) { GELOGI("No need to check sequence for constant."); continue; diff --git a/ge/host_cpu_engine/module.mk b/ge/host_cpu_engine/module.mk index e6cfce80..861ab19b 100644 --- a/ge/host_cpu_engine/module.mk +++ b/ge/host_cpu_engine/module.mk @@ -80,6 +80,26 @@ LOCAL_C_INCLUDES := $(local_lib_inc_path) include ${BUILD_HOST_SHARED_LIBRARY} +#compiler for device ops kernel builder +include $(CLEAR_VARS) +LOCAL_MODULE := libhost_cpu_opskernel_builder +LOCAL_CFLAGS += -Werror +LOCAL_CFLAGS += -std=c++11 -Dgoogle=ascend_private +LOCAL_LDFLAGS := + +LOCAL_STATIC_LIBRARIES := +LOCAL_SHARED_LIBRARIES := libascend_protobuf \ + libc_sec \ + libslog \ + libgraph \ + libregister \ + +LOCAL_SRC_FILES := ops_kernel_store/host_cpu_ops_kernel_builder.cc + +LOCAL_C_INCLUDES := $(local_lib_inc_path) + +include ${BUILD_SHARED_LIBRARY} + #compiler for host static lib include $(CLEAR_VARS) LOCAL_MODULE := libhost_cpu_opskernel_builder diff --git a/ge/hybrid/executor/hybrid_execution_context.h b/ge/hybrid/executor/hybrid_execution_context.h index 05ed1157..0910d2c7 100644 --- a/ge/hybrid/executor/hybrid_execution_context.h +++ b/ge/hybrid/executor/hybrid_execution_context.h @@ -57,9 +57,9 @@ struct GraphExecutionContext { do { \ if ((context != nullptr) && (context)->profiler != nullptr) { \ if (node_name != nullptr) { \ - context->profiler->RecordEvent(evt_type, "tid:%lu [%s] [%s] " fmt, GetTid(), node_name, category, ##__VA_ARGS__);\ + context->profiler->RecordEvent(evt_type, "tid:%lu [%s] [%s] " fmt, GeLog::GetTid(), node_name, category, ##__VA_ARGS__);\ } else { \ - context->profiler->RecordEvent(evt_type, "tid:%lu [%s] " fmt, GetTid(), category, ##__VA_ARGS__); \ + context->profiler->RecordEvent(evt_type, "tid:%lu [%s] " fmt, GeLog::GetTid(), category, ##__VA_ARGS__); \ }\ } \ } while (0) diff --git a/ge/hybrid/node_executor/aicpu/aicpu_ext_info.cc b/ge/hybrid/node_executor/aicpu/aicpu_ext_info.cc index 3974e29b..d7837144 100644 --- a/ge/hybrid/node_executor/aicpu/aicpu_ext_info.cc +++ b/ge/hybrid/node_executor/aicpu/aicpu_ext_info.cc @@ -57,6 +57,9 @@ Status AicpuExtInfoHandler::Parse(const std::string &ext_info) { case aicpu::FWKAdapter::FWK_ADPT_EXT_OUTPUT_SHAPE: GE_CHK_STATUS_RET(ParseExtOutputShape(aicpu_ext_info), "Parse ext output shape failed."); break; + case aicpu::FWKAdapter::FWK_ADPT_EXT_SESSION_INFO: + GE_CHK_STATUS_RET(ParseExtSessionInfo(aicpu_ext_info), "Parse ext session info failed."); + break; default: GELOGD("Node[%s] ignore infoType=%d, infoLen=%u.", node_name_.c_str(), aicpu_ext_info->infoType, aicpu_ext_info->infoLen); @@ -123,6 +126,39 @@ Status AicpuExtInfoHandler::ParseExtOutputShape(AicpuExtInfo *aicpu_ext_info) { return SUCCESS; } +Status AicpuExtInfoHandler::ParseExtSessionInfo(AicpuExtInfo *aicpu_ext_info) { + GE_CHK_BOOL_RET_STATUS(aicpu_ext_info->infoLen == sizeof(AicpuSessionInfo), PARAM_INVALID, + "Node[%s] parse ext session info failed as infoLen must be %zu but %u.", + node_name_.c_str(), sizeof(SessionInfo), aicpu_ext_info->infoLen); + + session_info_ = reinterpret_cast(aicpu_ext_info->infoMsg); + GELOGI("Node[%s] parse session info success infoLen=%u.", node_name_.c_str(), aicpu_ext_info->infoLen); + return SUCCESS; +} + +Status AicpuExtInfoHandler::UpdateSessionInfo(uint64_t session_id, uint64_t kernel_id, bool sess_flag) { + if (session_info_ == nullptr) { + GELOGD("There is no session info in ext_info, no need update."); + return SUCCESS; + } + + session_info_->sessionId = session_id; + session_info_->kernelId = kernel_id; + session_info_->sessFlag = sess_flag; + return SUCCESS; +} + +Status AicpuExtInfoHandler::UpdateSessionInfoSessionId(uint64_t session_id) { + if (session_info_ == nullptr) { + GELOGD("There is no session info in ext_info, no need update."); + return SUCCESS; + } + + session_info_->sessionId = session_id; + session_info_->sessFlag = true; + return SUCCESS; +} + Status AicpuExtInfoHandler::UpdateInputShapeAndType(uint32_t input_index, const GeTensorDesc &input_desc) { GE_CHECK_LE(input_index, input_num_); const auto &shape = input_desc.GetShape(); diff --git a/ge/hybrid/node_executor/aicpu/aicpu_ext_info.h b/ge/hybrid/node_executor/aicpu/aicpu_ext_info.h index 9c867cdc..2defba8f 100644 --- a/ge/hybrid/node_executor/aicpu/aicpu_ext_info.h +++ b/ge/hybrid/node_executor/aicpu/aicpu_ext_info.h @@ -19,6 +19,7 @@ #include "external/ge/ge_api_error_codes.h" #include "cce/fwk_adpt_struct.h" +#include "cce/aicpu_engine_struct.h" #include "graph/op_desc.h" #include "graph/ge_tensor.h" @@ -26,6 +27,7 @@ namespace ge { namespace hybrid { using AicpuShapeAndType = aicpu::FWKAdapter::ShapeAndType; using AicpuExtInfo = aicpu::FWKAdapter::ExtInfo; +using AicpuSessionInfo = SessionInfo; class AicpuExtInfoHandler { public: @@ -51,6 +53,10 @@ class AicpuExtInfoHandler { Status UpdateOutputShapeAndType(uint32_t output_index, const GeTensorDesc &output_desc); + Status UpdateSessionInfo(uint64_t session_id, uint64_t kernel_id, bool sess_flag); + + Status UpdateSessionInfoSessionId(uint64_t session_id); + Status GetOutputShapeAndType(uint32_t output_index, GeShape &shape, DataType &data_type); private: @@ -58,6 +64,7 @@ class AicpuExtInfoHandler { Status ParseExtShapeType(AicpuExtInfo *aicpu_ext_info); Status ParseExtInputShape(AicpuExtInfo *aicpu_ext_info); Status ParseExtOutputShape(AicpuExtInfo *aicpu_ext_info); + Status ParseExtSessionInfo(AicpuExtInfo *aicpu_ext_info); static Status UpdateShapeAndType(const GeShape &shape, DataType data_type, @@ -72,6 +79,7 @@ class AicpuExtInfoHandler { const uint32_t input_num_; const uint32_t output_num_; UnknowShapeOpType unknown_type_; + AicpuSessionInfo *session_info_ = nullptr; std::unique_ptr ext_info_; size_t ext_info_len_ = 0; diff --git a/ge/hybrid/node_executor/aicpu/aicpu_node_executor.cc b/ge/hybrid/node_executor/aicpu/aicpu_node_executor.cc index 61af55dd..d921c57c 100755 --- a/ge/hybrid/node_executor/aicpu/aicpu_node_executor.cc +++ b/ge/hybrid/node_executor/aicpu/aicpu_node_executor.cc @@ -40,29 +40,36 @@ Status AicpuNodeTaskBase::AllocTensorBuffer(size_t size, std::unique_ptris_dynamic) { - // dynamic node must have ext info - GE_CHK_STATUS_RET(aicpu_ext_handle_.Parse(kernel_ext_info), - "Node[%s] parse kernel ext info failed, kernel_ext_info_size=%zu.", - node_name_.c_str(), kernel_ext_info.size()); - } - - // if no ext info no need copy to device. +Status AicpuNodeTaskBase::InitExtInfo(const std::string &kernel_ext_info, int64_t session_id) { if (kernel_ext_info.empty()) { - GELOGI("Node[%s] kernel_ext_info is empty, no need copy to device, is_dynamic=%s.", - node_name_.c_str(), node_item_->is_dynamic ? "true" : "false"); - return SUCCESS; + if (node_item_->is_dynamic) { + // dynamic node must have ext info + GELOGE(PARAM_INVALID, "Node[%s] parse ext info failed as ext info is empty.", node_name_.c_str()); + return PARAM_INVALID; + } else { + // if no ext info no need copy to device. + GELOGI("Node[%s] kernel_ext_info is empty, no need copy to device, is_dynamic=%s.", + node_name_.c_str(), node_item_->is_dynamic ? "true" : "false"); + return SUCCESS; + } } + GE_CHK_STATUS_RET(aicpu_ext_handle_.Parse(kernel_ext_info), + "Node[%s] parse kernel ext info failed, kernel_ext_info_size=%zu.", + node_name_.c_str(), kernel_ext_info.size()); + GELOGD("To update aicpu_task ext_info session_info session_id to %lu", session_id); + GE_CHK_STATUS_RET(aicpu_ext_handle_.UpdateSessionInfoSessionId(session_id), + "UpdateSessionInfoSessionId failed."); + // copy task args buf - GE_CHK_STATUS_RET(AllocTensorBuffer(kernel_ext_info.size(), ext_info_addr_dev_), + GE_CHK_STATUS_RET(AllocTensorBuffer(aicpu_ext_handle_.GetExtInfoLen(), ext_info_addr_dev_), "Node[%s] alloc kernel_ext_info buf failed, size=%zu", - node_name_.c_str(), kernel_ext_info.size()); + node_name_.c_str(), aicpu_ext_handle_.GetExtInfoLen()); // copy default ext info to device GE_CHK_RT_RET(rtMemcpy(ext_info_addr_dev_->GetData(), ext_info_addr_dev_->GetSize(), - kernel_ext_info.data(), kernel_ext_info.size(), RT_MEMCPY_HOST_TO_DEVICE)); + aicpu_ext_handle_.GetExtInfo(), aicpu_ext_handle_.GetExtInfoLen(), + RT_MEMCPY_HOST_TO_DEVICE)); return SUCCESS; } @@ -290,7 +297,8 @@ Status AicpuTfNodeTask::Init(const HybridModel &model) { node_name_.c_str(), kernel_ext_info.size(), kernel_ext_info_size); // init ext info - GE_CHK_STATUS_RET(InitExtInfo(kernel_ext_info), "Node[%s] init ext info failed.", node_name_.c_str()); + uint64_t ext_session_id = model.GetSessionId(); + GE_CHK_STATUS_RET(InitExtInfo(kernel_ext_info, ext_session_id), "Node[%s] init ext info failed.", node_name_.c_str()); GE_CHK_STATUS_RET(InitForDependComputeTask(), "Node[%s] init for depend compute task failed.", node_name_.c_str()); // build fwk_op_kernel. @@ -679,7 +687,8 @@ Status AicpuNodeTask::Init(const HybridModel &model) { "Node[%s] task def kernel_ext_info.size=%zu, but kernel_ext_info_size=%u.", node_name.c_str(), kernel_ext_info.size(), kernel_ext_info_size); - GE_CHK_STATUS_RET(InitExtInfo(kernel_ext_info), "Node[%s] init ext info failed.", node_name.c_str()); + uint64_t ext_session_id = model.GetSessionId(); + GE_CHK_STATUS_RET(InitExtInfo(kernel_ext_info, ext_session_id), "Node[%s] init ext info failed.", node_name.c_str()); if (ext_info_addr_dev_ == nullptr) { aicpu_param_head->extInfoLength = 0; diff --git a/ge/hybrid/node_executor/aicpu/aicpu_node_executor.h b/ge/hybrid/node_executor/aicpu/aicpu_node_executor.h index 7caabd66..b984cc86 100644 --- a/ge/hybrid/node_executor/aicpu/aicpu_node_executor.h +++ b/ge/hybrid/node_executor/aicpu/aicpu_node_executor.h @@ -43,7 +43,7 @@ class AicpuNodeTaskBase : public NodeTask { Status ExecuteAsync(TaskContext &context, std::function done_callback) override; protected: - virtual Status InitExtInfo(const std::string &kernel_ext_info); + virtual Status InitExtInfo(const std::string &kernel_ext_info, int64_t session_id); virtual Status UpdateExtInfo(); diff --git a/ge/init/gelib.cc b/ge/init/gelib.cc index 8a5cb610..306a804a 100755 --- a/ge/init/gelib.cc +++ b/ge/init/gelib.cc @@ -110,7 +110,7 @@ Status GELib::InnerInitialize(const map &options) { Status initSystemStatus = SystemInitialize(options); GE_TIMESTAMP_END(SystemInitialize, "InnerInitialize::SystemInitialize"); if (initSystemStatus != SUCCESS) { - GELOGE(initSystemStatus); + GELOGE(initSystemStatus, "GE system initial failed."); RollbackInit(); return initSystemStatus; } @@ -120,7 +120,7 @@ Status GELib::InnerInitialize(const map &options) { Status initEmStatus = engineManager_.Initialize(options); GE_TIMESTAMP_END(EngineInitialize, "InnerInitialize::EngineInitialize"); if (initEmStatus != SUCCESS) { - GELOGE(initEmStatus); + GELOGE(initEmStatus, "GE engine manager initial failed."); RollbackInit(); return initEmStatus; } @@ -130,7 +130,7 @@ Status GELib::InnerInitialize(const map &options) { Status initOpsStatus = opsManager_.Initialize(options); GE_TIMESTAMP_END(OpsManagerInitialize, "InnerInitialize::OpsManagerInitialize"); if (initOpsStatus != SUCCESS) { - GELOGE(initOpsStatus); + GELOGE(initOpsStatus, "GE ops manager initial failed."); RollbackInit(); return initOpsStatus; } @@ -140,7 +140,7 @@ Status GELib::InnerInitialize(const map &options) { Status initOpsBuilderStatus = OpsKernelBuilderManager::Instance().Initialize(options); GE_TIMESTAMP_END(OpsKernelBuilderManagerInitialize, "InnerInitialize::OpsKernelBuilderManager"); if (initOpsBuilderStatus != SUCCESS) { - GELOGE(initOpsBuilderStatus); + GELOGE(initOpsBuilderStatus, "GE ops builder manager initial failed."); RollbackInit(); return initOpsBuilderStatus; } @@ -150,7 +150,7 @@ Status GELib::InnerInitialize(const map &options) { Status initSmStatus = sessionManager_.Initialize(options); GE_TIMESTAMP_END(SessionManagerInitialize, "InnerInitialize::SessionManagerInitialize"); if (initSmStatus != SUCCESS) { - GELOGE(initSmStatus); + GELOGE(initSmStatus, "GE session manager initial failed."); RollbackInit(); return initSmStatus; } diff --git a/ge/ir_build/atc_ir_common.cc b/ge/ir_build/atc_ir_common.cc index f0450f0f..eaff928b 100755 --- a/ge/ir_build/atc_ir_common.cc +++ b/ge/ir_build/atc_ir_common.cc @@ -504,7 +504,7 @@ void PrintOptionMap(std::map &options, std::string tip for (auto iter = options.begin(); iter != options.end(); iter++) { std::string key = iter->first; std::string option_name = iter->second; - GELOGI("%s set successfully, option_key=%s, option_value=%s", tips.c_str(), key.c_str(), option_name.c_str()); + GELOGD("%s set successfully, option_key=%s, option_value=%s", tips.c_str(), key.c_str(), option_name.c_str()); } } diff --git a/ge/ir_build/ge_ir_build.cc b/ge/ir_build/ge_ir_build.cc index ccf3b24e..53f758a4 100644 --- a/ge/ir_build/ge_ir_build.cc +++ b/ge/ir_build/ge_ir_build.cc @@ -340,7 +340,7 @@ void Impl::SetRtSocVersion() { if (rt_ret != RT_ERROR_NONE) { GELOGW("Set soc version %s failed. ret:0x%X", soc_version, rt_ret); } - GELOGI("Set soc version %s success.", soc_version); + GELOGD("Set soc version %s success.", soc_version); } } @@ -359,25 +359,25 @@ graphStatus Impl::CreateInputsForIRBuild(const ge::Graph &graph, vectorGetType() == DATA) { (void)AttrUtils::SetInt(op, ATTR_NAME_INDEX, index++); - GELOGI("Data op inputDesc size: %zu", op->GetAllInputsDesc().size()); + GELOGD("Data op inputDesc size: %zu", op->GetAllInputsDesc().size()); ge::GeTensorDesc tensor = op->GetInputDesc(0); string data_op_name = op->GetName(); - GELOGI("Data op name: %s", data_op_name.c_str()); + GELOGD("Data op name: %s", data_op_name.c_str()); ge::GeShape data_shape; auto iter = omg_context_.input_dims.find(data_op_name); if (iter != omg_context_.input_dims.end()) { data_shape = ge::GeShape(iter->second); - GELOGI("Data op get shape from Context."); + GELOGD("Data op get shape from Context."); } else { data_shape = tensor.GetShape(); - GELOGI("Data op get shape from InputDesc in ge ir graph."); + GELOGD("Data op get shape from InputDesc in ge ir graph."); } // If user point input format, do work for all data ops; else do according to tensor_desc auto data_format = omg_context_.format != domi::DOMI_TENSOR_ND ? ge::TypeUtils::DomiFormatToFormat(omg_context_.format) : tensor.GetFormat(); ge::DataType data_type = tensor.GetDataType(); string data_type_str = ge::TypeUtils::DataTypeToSerialString(data_type); - GELOGI("Data op get data type:%s from InputDesc in ge ir graph.", data_type_str.c_str()); + GELOGD("Data op get data type:%s from InputDesc in ge ir graph.", data_type_str.c_str()); ge::GeTensor inputTensor; ge::GeTensorDesc desc(data_shape, ge::Format(data_format), data_type); diff --git a/ge/offline/CMakeLists.txt b/ge/offline/CMakeLists.txt index e8e91327..49af37c0 100644 --- a/ge/offline/CMakeLists.txt +++ b/ge/offline/CMakeLists.txt @@ -69,7 +69,7 @@ target_link_libraries(atc PRIVATE json runtime_compile slog - mmpa + static_mmpa -lrt -ldl ) diff --git a/ge/omm/csa_interact.cc b/ge/omm/csa_interact.cc index dd3f6240..1599af94 100644 --- a/ge/omm/csa_interact.cc +++ b/ge/omm/csa_interact.cc @@ -52,9 +52,11 @@ void CsaInteract::Init(int32_t dev_index, int64_t job_id) { if (!is_init_) { dev_index_ = dev_index; job_id_ = job_id; - char *file_dir_env = std::getenv(FMK_STATUS_FILE_DIR_ENV); + + char file_dir_env[MMPA_MAX_PATH] = { 0x00 }; + INT32 res = mmGetEnv(FMK_STATUS_FILE_DIR_ENV, file_dir_env, MMPA_MAX_PATH); string csa_path_prefix; - if (file_dir_env != nullptr) { + if (res == EN_OK) { csa_path_prefix = file_dir_env; } if (!csa_path_prefix.empty()) { @@ -186,21 +188,21 @@ Status CsaInteract::WriteHcomDetection(const std::string &content) { /// Status CsaInteract::WriteFile(const std::string &file_name, const std::string &content) { // if file path is not exist, then make path - INT32 flags = O_WRONLY | O_TRUNC | O_CREAT; - int32_t fd = mmOpen2(file_name.c_str(), flags, M_IRUSR | M_IWUSR | S_IRGRP); + INT32 flags = M_WRONLY | O_TRUNC | M_CREAT; + int32_t fd = mmOpen2(file_name.c_str(), flags, M_IRUSR | M_IWUSR | M_UMASK_GRPREAD); if (fd == EN_ERROR) { if (MakePath(file_name) != SUCCESS) { GELOGE(INTERNAL_ERROR, "csainteract create file path fail, errno is %d", errno); return INTERNAL_ERROR; } - fd = mmOpen2(file_name.c_str(), flags, M_IRUSR | M_IWUSR | S_IRGRP); + fd = mmOpen2(file_name.c_str(), flags, M_IRUSR | M_IWUSR | M_UMASK_GRPREAD); if (fd == EN_ERROR) { GELOGE(INTERNAL_ERROR, "open file fail, errno is %d", errno); return INTERNAL_ERROR; } } - ssize_t ret = write(fd, content.c_str(), content.length()); + mmSsize_t ret = mmWrite(fd, (void *)content.c_str(), content.length()); if (ret == EN_ERROR) { GELOGE(INTERNAL_ERROR, "write file fail, errno is %d", errno); ret = mmClose(fd); @@ -239,7 +241,7 @@ Status CsaInteract::MakePath(const std::string &file_name) { while (found != std::string::npos) { std::string pre_path = file_path.substr(0, found + 1); if (mmAccess(pre_path.c_str()) != EN_OK) { - if (mmMkdir(pre_path.c_str(), S_IRWXU) != EN_OK) { + if (mmMkdir(pre_path.c_str(), M_IRWXU) != EN_OK) { GELOGE(INTERNAL_ERROR, "csainteract mkdir fail, errno is %d", errno); return INTERNAL_ERROR; } diff --git a/ge/opskernel_manager/ops_kernel_manager.cc b/ge/opskernel_manager/ops_kernel_manager.cc index 12894292..8134a463 100644 --- a/ge/opskernel_manager/ops_kernel_manager.cc +++ b/ge/opskernel_manager/ops_kernel_manager.cc @@ -85,7 +85,7 @@ Status OpsKernelManager::Initialize(const map &options_const) { initialize_ = options; Status rst0 = plugin_manager_.InvokeAll &, Status>(kInitialize, initialize_); if (rst0 == FAILED) { - GELOGE(GE_OPS_GET_NO_VALID_SO); + GELOGE(GE_OPS_GET_NO_VALID_SO, "There is invalid so about OpsKernelInfoStore."); return GE_OPS_GET_NO_VALID_SO; } Status rst1 = @@ -391,7 +391,7 @@ void OpsKernelManager::GetGraphOptimizerByEngine(const std::string &engine_name, continue; } if (attrs.engineName == engine_name) { - GELOGI("GetGraphOptimizerByEngine GraphOptimzer name: %s, engineName: %s", (it.first).c_str(), + GELOGD("GetGraphOptimizerByEngine GraphOptimzer name: %s, engineName: %s", (it.first).c_str(), attrs.engineName.c_str()); graph_optimizer.push_back(it.second); } diff --git a/ge/session/session_manager.cc b/ge/session/session_manager.cc index 81efb080..69a62923 100755 --- a/ge/session/session_manager.cc +++ b/ge/session/session_manager.cc @@ -61,7 +61,7 @@ Status SessionManager::SetRtContext(SessionId session_id, rtContext_t rt_context Status SessionManager::CreateSession(const std::map &options, SessionId &session_id) { if (!init_flag_) { - GELOGE(GE_SESSION_MANAGER_NOT_INIT); + GELOGE(GE_SESSION_MANAGER_NOT_INIT, "Session manager is not initialized."); return GE_SESSION_MANAGER_NOT_INIT; } SessionId next_session_id = 0; @@ -92,7 +92,7 @@ Status SessionManager::CreateSession(const std::map &o Status SessionManager::DestroySession(SessionId session_id) { if (!init_flag_) { - GELOGE(GE_SESSION_MANAGER_NOT_INIT); + GELOGE(GE_SESSION_MANAGER_NOT_INIT, "Session manager is not initialized."); return GE_SESSION_MANAGER_NOT_INIT; } std::lock_guard lock(mutex_); @@ -119,7 +119,7 @@ Status SessionManager::DestroySession(SessionId session_id) { Status SessionManager::GetVariable(SessionId session_id, const std::string &name, Tensor &val) { if (!init_flag_) { - GELOGE(GE_SESSION_MANAGER_NOT_INIT); + GELOGE(GE_SESSION_MANAGER_NOT_INIT, "Session manager is not initialized."); return GE_SESSION_MANAGER_NOT_INIT; } SessionPtr innerSession = nullptr; @@ -143,7 +143,7 @@ Status SessionManager::AddGraph(SessionId session_id, uint32_t graph_id, const G Status SessionManager::AddGraph(SessionId session_id, uint32_t graph_id, const Graph &graph, const std::map &options) { if (!init_flag_) { - GELOGE(GE_SESSION_MANAGER_NOT_INIT); + GELOGE(GE_SESSION_MANAGER_NOT_INIT, "Session manager is not initialized."); return GE_SESSION_MANAGER_NOT_INIT; } SessionPtr innerSession = nullptr; @@ -173,7 +173,7 @@ Status SessionManager::AddGraph(SessionId session_id, uint32_t graph_id, const G Status SessionManager::AddGraphWithCopy(SessionId session_id, uint32_t graph_id, const Graph &graph, const std::map &options) { if (!init_flag_) { - GELOGE(GE_SESSION_MANAGER_NOT_INIT); + GELOGE(GE_SESSION_MANAGER_NOT_INIT, "Session manager is not initialized."); return GE_SESSION_MANAGER_NOT_INIT; } SessionPtr innerSession = nullptr; @@ -203,7 +203,7 @@ Status SessionManager::AddGraphWithCopy(SessionId session_id, uint32_t graph_id, Status SessionManager::RunGraph(SessionId session_id, uint32_t graph_id, const std::vector &inputs, std::vector &outputs) { if (!init_flag_) { - GELOGE(GE_SESSION_MANAGER_NOT_INIT); + GELOGE(GE_SESSION_MANAGER_NOT_INIT, "Session manager is not initialized."); return GE_SESSION_MANAGER_NOT_INIT; } SessionPtr innerSession = nullptr; @@ -221,7 +221,7 @@ Status SessionManager::RunGraph(SessionId session_id, uint32_t graph_id, const s Status SessionManager::RemoveGraph(SessionId session_id, uint32_t graph_id) { if (!init_flag_) { - GELOGE(GE_SESSION_MANAGER_NOT_INIT); + GELOGE(GE_SESSION_MANAGER_NOT_INIT, "Session manager is not initialized."); return GE_SESSION_MANAGER_NOT_INIT; } SessionPtr innerSession = nullptr; @@ -239,7 +239,7 @@ Status SessionManager::RemoveGraph(SessionId session_id, uint32_t graph_id) { bool SessionManager::HasSession(SessionId session_id) { if (!init_flag_) { - GELOGE(GE_SESSION_MANAGER_NOT_INIT); + GELOGE(GE_SESSION_MANAGER_NOT_INIT, "Session manager is not initialized."); return false; } return session_manager_map_.find(session_id) != session_manager_map_.end(); @@ -247,7 +247,7 @@ bool SessionManager::HasSession(SessionId session_id) { Status SessionManager::GetNextSessionId(SessionId &next_session_id) { if (!init_flag_) { - GELOGE(GE_SESSION_MANAGER_NOT_INIT); + GELOGE(GE_SESSION_MANAGER_NOT_INIT, "Session manager is not initialized."); return GE_SESSION_MANAGER_NOT_INIT; } static SessionId session_id = 0; @@ -260,7 +260,7 @@ Status SessionManager::RegisterCallBackFunc( SessionId session_id, const std::string &key, const std::function &)> &callback) { if (!init_flag_) { - GELOGE(GE_SESSION_MANAGER_NOT_INIT); + GELOGE(GE_SESSION_MANAGER_NOT_INIT, "Session manager is not initialized."); return GE_SESSION_MANAGER_NOT_INIT; } SessionPtr innerSession = nullptr; @@ -278,7 +278,7 @@ Status SessionManager::RegisterCallBackFunc( Status SessionManager::BuildGraph(SessionId session_id, uint32_t graph_id, const std::vector &inputs) { if (!init_flag_) { - GELOGE(GE_SESSION_MANAGER_NOT_INIT); + GELOGE(GE_SESSION_MANAGER_NOT_INIT, "Session manager is not initialized."); return GE_SESSION_MANAGER_NOT_INIT; } SessionPtr innerSession = nullptr; @@ -297,7 +297,7 @@ Status SessionManager::BuildGraph(SessionId session_id, uint32_t graph_id, const Status SessionManager::RunGraphAsync(SessionId session_id, uint32_t graph_id, const std::vector &inputs, RunAsyncCallback callback) { if (!init_flag_) { - GELOGE(GE_SESSION_MANAGER_NOT_INIT); + GELOGE(GE_SESSION_MANAGER_NOT_INIT, "Session manager is not initialized."); return GE_SESSION_MANAGER_NOT_INIT; } SessionPtr innerSession = nullptr; @@ -317,7 +317,7 @@ Status SessionManager::GetVariables(SessionId session_id, const std::vector &var_values) { // step 0: init session manager if (!init_flag_) { - GELOGE(GE_SESSION_MANAGER_NOT_INIT); + GELOGE(GE_SESSION_MANAGER_NOT_INIT, "Session manager is not initialized."); return GE_SESSION_MANAGER_NOT_INIT; } SessionPtr innerSession = nullptr; @@ -383,7 +383,7 @@ Status SessionManager::GetVariables(SessionId session_id, const std::vectorDestroyAicpuSession(aicpu_session_id_); } Status SingleOp::ValidateArgs(const std::vector &inputs, const std::vector &outputs) { @@ -59,7 +57,7 @@ Status SingleOp::ValidateArgs(const std::vector &inputs, const std:: for (size_t i = 0; i < num_inputs; ++i) { // preventing from read out of bound size_t aligned_size = GetAlignedSize(inputs[i].length); - GELOGI("Input [%zu], aligned_size:%zu, inputs.length:%lu, input_sizes_:%lu", + GELOGI("Input [%zu], aligned_size:%zu, inputs.length:%lu, input_sizes_:%zu", i, aligned_size, inputs[i].length, input_sizes_[i]); if (aligned_size < input_sizes_[i]) { GELOGE(ACL_ERROR_GE_PARAM_INVALID, "Input size mismatch. index = %zu, model expect %zu," @@ -77,7 +75,7 @@ Status SingleOp::ValidateArgs(const std::vector &inputs, const std:: for (size_t i = 0; i < num_outputs; ++i) { // preventing from write out of bound size_t aligned_size = GetAlignedSize(outputs[i].length); - GELOGI("Output [%zu], aligned_size:%zu, outputs.length:%lu, output_sizes_:%lu", + GELOGI("Output [%zu], aligned_size:%zu, outputs.length:%lu, output_sizes_:%zu", i, aligned_size, outputs[i].length, output_sizes_[i]); if (aligned_size < output_sizes_[i]) { GELOGE(ACL_ERROR_GE_PARAM_INVALID, "Output size mismatch. index = %zu, model expect %zu," @@ -143,7 +141,7 @@ Status SingleOp::UpdateArgs(const std::vector &inputs, const std::ve GE_CHECK_NOTNULL(task_io_addr); auto io_addr = reinterpret_cast(const_cast(task_io_addr)); for (size_t i = 0; i < io_addr_num; ++i) { - io_addr[i] = reinterpret_cast(args_[i]); + io_addr[i] = static_cast(args_[i]); } } else { GELOGW("Only TF_kernel aicpu and aicpu_CC are supported, but got %u", task->GetOpTaskType()); @@ -180,17 +178,11 @@ void SingleOp::SetStream(rtStream_t stream) { stream_ = stream; } -void SingleOp::SetSessionID(uint64_t session_id) { - aicpu_session_id_ = session_id; -} - DynamicSingleOp::DynamicSingleOp(uintptr_t resource_id, std::mutex *stream_mutex, rtStream_t stream) : resource_id_(resource_id), stream_mutex_(stream_mutex), stream_(stream) { } DynamicSingleOp::~DynamicSingleOp() { - GELOGI("DynamicSingleOp destory sessionId = %lu", aicpu_session_id_); - ModelManager::GetInstance()->DestroyAicpuSession(aicpu_session_id_); } Status DynamicSingleOp::ValidateParams(const vector &input_desc, @@ -299,8 +291,4 @@ Status DynamicSingleOp::ExecuteAsync(const vector &input_desc, return ACL_ERROR_GE_OP_TASK_TYPE_INVALID; } } - -void DynamicSingleOp::SetSessionID(uint64_t session_id) { - aicpu_session_id_ = session_id; -} } // namespace ge diff --git a/ge/single_op/single_op.h b/ge/single_op/single_op.h index bd671017..14ef8ce1 100755 --- a/ge/single_op/single_op.h +++ b/ge/single_op/single_op.h @@ -37,7 +37,6 @@ class SingleOp { Status ExecuteAsync(const std::vector &inputs, const std::vector &outputs); void SetStream(rtStream_t stream); - void SetSessionID(uint64_t session_id); private: Status ValidateArgs(const std::vector &inputs, const std::vector &outputs); @@ -52,7 +51,6 @@ class SingleOp { std::vector output_addr_list_; std::vector output_sizes_; std::vector args_; - uint64_t aicpu_session_id_ = 0; std::vector tasks_; std::vector> arg_table_; @@ -66,7 +64,6 @@ class DynamicSingleOp { const std::vector &inputs, std::vector &output_desc, std::vector &outputs); - void SetSessionID(uint64_t session_id); private: friend class SingleOpModel; @@ -89,7 +86,6 @@ class DynamicSingleOp { rtStream_t stream_ = nullptr; size_t num_inputs_ = 0; size_t num_outputs_ = 0; - uint64_t aicpu_session_id_ = 0; }; } // namespace ge #endif // GE_SINGLE_OP_SINGLE_OP_H_ diff --git a/ge/single_op/single_op_model.cc b/ge/single_op/single_op_model.cc index 65eb58be..49968f4f 100755 --- a/ge/single_op/single_op_model.cc +++ b/ge/single_op/single_op_model.cc @@ -32,7 +32,7 @@ #include "task/aicpu_kernel_task_builder.h" #include "task/tbe_task_builder.h" -static std::atomic aicpu_sessionid(0); +static std::atomic aicpu_kernel_id(0); using domi::TaskDef; using std::unique_ptr; @@ -252,7 +252,9 @@ Status SingleOpModel::BuildTaskList(SingleOp &single_op) { } else if (kernel_type == cce::ccKernelType::AI_CPU || kernel_type == cce::ccKernelType::CUST_AI_CPU) { GELOGD("Building AICPU_CC task"); OpTask *task = nullptr; - auto ret = BuildCpuKernelTask(task_def.kernel(), &task); + uint64_t singleop_kernel_id = aicpu_kernel_id++; + GELOGI("Build singleOp CCTask, kernel_id = %lu", singleop_kernel_id); + auto ret = BuildCpuKernelTask(task_def.kernel(), &task, singleop_kernel_id); if (ret != SUCCESS) { return ret; } @@ -265,14 +267,13 @@ Status SingleOpModel::BuildTaskList(SingleOp &single_op) { GELOGD("Building AICPU_TF task"); AiCpuTask *aicpu_task = nullptr; bool depend_compute_flag = false; - uint64_t singleop_sessionid = aicpu_sessionid++; - GELOGI("Build singleOp, sessionId = %lu", singleop_sessionid); - auto ret = BuildKernelExTask(task_def.kernel_ex(), &aicpu_task, false, depend_compute_flag, singleop_sessionid); + uint64_t singleop_kernel_id = aicpu_kernel_id++; + GELOGI("Build singleOp TfTask, kernel_id = %lu", singleop_kernel_id); + auto ret = BuildKernelExTask(task_def.kernel_ex(), &aicpu_task, false, depend_compute_flag, singleop_kernel_id); if (ret != SUCCESS) { return ret; } single_op.tasks_.emplace_back(aicpu_task); - single_op.SetSessionID(singleop_sessionid); } else { // skip GELOGD("Skip task type: %d", static_cast(task_type)); @@ -329,7 +330,7 @@ Status SingleOpModel::BuildKernelTask(const domi::KernelDef &kernel_def, TbeOpTa } Status SingleOpModel::BuildKernelExTask(const domi::KernelExDef &kernel_def, AiCpuTask **task, - bool dynamic_flag, bool& depend_compute_flag, uint64_t session_id) { + bool dynamic_flag, bool& depend_compute_flag, uint64_t kernel_id) { auto iter = op_list_.find(kernel_def.op_index()); if (iter == op_list_.end()) { GELOGE(ACL_ERROR_GE_INTERNAL_ERROR, "op desc not found. op index = %u", kernel_def.op_index()); @@ -342,7 +343,7 @@ Status SingleOpModel::BuildKernelExTask(const domi::KernelExDef &kernel_def, AiC return ACL_ERROR_GE_MEMORY_ALLOCATION; } auto builder = AiCpuTaskBuilder(iter->second->GetOpDesc(), kernel_def); - auto ret = builder.BuildTask(*aicpu_task, model_params_, dynamic_flag, session_id); + auto ret = builder.BuildTask(*aicpu_task, model_params_, dynamic_flag, kernel_id); if (ret != SUCCESS) { GELOGE(ret, "build aicpu_TF op task failed"); return ret; @@ -353,7 +354,7 @@ Status SingleOpModel::BuildKernelExTask(const domi::KernelExDef &kernel_def, AiC return SUCCESS; } -Status SingleOpModel::BuildCpuKernelTask(const domi::KernelDef &kernel_def, OpTask **task) { +Status SingleOpModel::BuildCpuKernelTask(const domi::KernelDef &kernel_def, OpTask **task, uint64_t kernel_id) { const auto &context = kernel_def.context(); auto iter = op_list_.find(context.op_index()); if (iter == op_list_.end()) { @@ -367,7 +368,7 @@ Status SingleOpModel::BuildCpuKernelTask(const domi::KernelDef &kernel_def, OpTa } auto builder = AiCpuCCTaskBuilder(iter->second->GetOpDesc(), kernel_def); - auto ret = builder.BuildTask(*aicpucc_task); + auto ret = builder.BuildTask(*aicpucc_task, kernel_id); if (ret != SUCCESS) { GELOGE(ret, "build aicpu_CC op task failed"); return ret; @@ -396,7 +397,9 @@ Status SingleOpModel::BuildModelTaskKernel(const TaskDef &task_def, DynamicSingl } else if (kernel_type == cce::ccKernelType::AI_CPU || kernel_type == cce::ccKernelType::CUST_AI_CPU) { GELOGD("Building AICPU_CC task"); OpTask *task = nullptr; - GE_CHK_STATUS_RET_NOLOG(BuildCpuKernelTask(task_def.kernel(), &task)); + uint64_t dynamic_singleop_kernel_id = aicpu_kernel_id++; + GELOGI("Build dynamic singleOp CCTask, kernel_id = %lu", dynamic_singleop_kernel_id); + GE_CHK_STATUS_RET_NOLOG(BuildCpuKernelTask(task_def.kernel(), &task, dynamic_singleop_kernel_id)); single_op.op_task_.reset(task); } else { GELOGE(ACL_ERROR_GE_OP_KERNEL_TYPE_INVALID, @@ -430,10 +433,10 @@ Status SingleOpModel::BuildTaskListForDynamicOp(DynamicSingleOp &single_op) { GELOGD("Building AICPU_TF task"); AiCpuTask *aicpu_task = nullptr; bool depend_compute_flag = false; - uint64_t dynamic_singleop_sessionid = aicpu_sessionid++; - GELOGI("Build dynamic singleOp, sessionId = %lu", dynamic_singleop_sessionid); + uint64_t dynamic_singleop_kernel_id = aicpu_kernel_id++; + GELOGI("Build dynamic singleOp TfTask, kernel_id = %lu", dynamic_singleop_kernel_id); GE_CHK_STATUS_RET_NOLOG(BuildKernelExTask(task_def.kernel_ex(), &aicpu_task, true, - depend_compute_flag, dynamic_singleop_sessionid)); + depend_compute_flag, dynamic_singleop_kernel_id)); if (depend_compute_flag) { if (i >= tasks.size() - 1) { GELOGE(ACL_ERROR_GE_PARAM_INVALID, "The copy task of the fourth operator was not found."); @@ -444,7 +447,6 @@ Status SingleOpModel::BuildTaskListForDynamicOp(DynamicSingleOp &single_op) { GE_CHK_STATUS_RET_NOLOG(aicpu_task->SetMemCopyTask(copy_task_def.kernel_ex())); } single_op.op_task_.reset(aicpu_task); - single_op.SetSessionID(dynamic_singleop_sessionid); } else { // skip GELOGD("Skip task type: %d", static_cast(task_type)); diff --git a/ge/single_op/single_op_model.h b/ge/single_op/single_op_model.h index 2e6b37dc..50aeb7ab 100755 --- a/ge/single_op/single_op_model.h +++ b/ge/single_op/single_op_model.h @@ -69,8 +69,8 @@ class SingleOpModel { Status BuildTaskListForDynamicOp(DynamicSingleOp &dynamic_single_op); Status BuildKernelTask(const domi::KernelDef &kernel_def, TbeOpTask **task); Status BuildKernelExTask(const domi::KernelExDef &kernel_def, AiCpuTask **task, - bool dynamic_flag, bool& depend_compute_flag, uint64_t session_id); - Status BuildCpuKernelTask(const domi::KernelDef &kernel_def, OpTask **task); + bool dynamic_flag, bool& depend_compute_flag, uint64_t kernel_id); + Status BuildCpuKernelTask(const domi::KernelDef &kernel_def, OpTask **task, uint64_t kernel_id); Status BuildModelTaskKernel(const domi::TaskDef &task_def, DynamicSingleOp &single_op); static void ParseOpModelParams(ModelHelper &model_helper, SingleOpModelParam ¶m); diff --git a/ge/single_op/task/aicpu_kernel_task_builder.cc b/ge/single_op/task/aicpu_kernel_task_builder.cc index 8e140a6f..26f6a166 100755 --- a/ge/single_op/task/aicpu_kernel_task_builder.cc +++ b/ge/single_op/task/aicpu_kernel_task_builder.cc @@ -46,7 +46,7 @@ Status AiCpuCCTaskBuilder::SetKernelArgs(AiCpuCCTask &task) { return SUCCESS; } -Status AiCpuCCTaskBuilder::BuildTask(AiCpuCCTask &task) { +Status AiCpuCCTaskBuilder::BuildTask(AiCpuCCTask &task, uint64_t kernel_id) { auto ret = SetKernelArgs(task); if (ret != SUCCESS) { return ret; @@ -76,7 +76,7 @@ Status AiCpuCCTaskBuilder::BuildTask(AiCpuCCTask &task) { "task def kernel_ext_info.size=%zu, but kernel_ext_info_size=%u.", kernel_ext_info.size(), kernel_ext_info_size); - ret = task.SetExtInfoAndType(kernel_ext_info); + ret = task.SetExtInfoAndType(kernel_ext_info, kernel_id); if (ret != SUCCESS) { GELOGE(ret, "Init ext info failed."); return ret; diff --git a/ge/single_op/task/aicpu_kernel_task_builder.h b/ge/single_op/task/aicpu_kernel_task_builder.h index f9ca0530..e77e3c10 100755 --- a/ge/single_op/task/aicpu_kernel_task_builder.h +++ b/ge/single_op/task/aicpu_kernel_task_builder.h @@ -30,7 +30,7 @@ class AiCpuCCTaskBuilder { explicit AiCpuCCTaskBuilder(const OpDescPtr &op_desc, const domi::KernelDef &kernel_def); ~AiCpuCCTaskBuilder() = default; - Status BuildTask(AiCpuCCTask &task); + Status BuildTask(AiCpuCCTask &task, uint64_t kernel_id); private: Status SetKernelArgs(AiCpuCCTask &task); diff --git a/ge/single_op/task/aicpu_task_builder.cc b/ge/single_op/task/aicpu_task_builder.cc index 468fee86..d91bba17 100755 --- a/ge/single_op/task/aicpu_task_builder.cc +++ b/ge/single_op/task/aicpu_task_builder.cc @@ -111,7 +111,7 @@ namespace ge { } Status AiCpuTaskBuilder::BuildTask(ge::AiCpuTask &task, const SingleOpModelParam ¶m, - bool dynamic_flag, uint64_t session_id) { + bool dynamic_flag, uint64_t kernel_id) { GE_CHK_STATUS_RET_NOLOG(InitWorkspaceAndIO(&task.io_addr_, &task.workspace_addr_, param, dynamic_flag)); STR_FWK_OP_KERNEL fwk_op_kernel = {0}; @@ -130,7 +130,7 @@ namespace ge { GE_CHK_BOOL_RET_STATUS(kernel_ext_info.size() == kernel_ext_info_size, FAILED, "task def kernel_ext_info.size=%zu, but kernel_ext_info_size=%u.", kernel_ext_info.size(), kernel_ext_info_size); - GE_CHK_STATUS_RET(task.SetExtInfoAndType(kernel_ext_info), "Init ext info failed."); + GE_CHK_STATUS_RET(task.SetExtInfoAndType(kernel_ext_info, kernel_id), "Init ext info failed."); if (task.ext_info_addr_dev_ != nullptr) { fwk_op_kernel.fwkKernelBase.fwk_kernel.extInfoAddr = reinterpret_cast(task.ext_info_addr_dev_); @@ -138,13 +138,9 @@ namespace ge { } GE_CHK_STATUS_RET(task.InitForSummaryAndCopy(), "AiCpuTask init for summary and copy task failed."); - // Create session - fwk_op_kernel.fwkKernelBase.fwk_kernel.sessionID = session_id; - GELOGI("Begin to CreateAicpuSession, session id: %lu", session_id); - GE_CHECK_NOTNULL(ModelManager::GetInstance()); - GE_IF_BOOL_EXEC(ModelManager::GetInstance()->CreateAicpuSession(session_id) != SUCCESS, - GELOGE(ACL_ERROR_GE_INTERNAL_ERROR, "CreateAicpuSession error. session id: %lu", session_id); - return ACL_ERROR_GE_INTERNAL_ERROR;) + fwk_op_kernel.fwkKernelBase.fwk_kernel.sessionID = ULLONG_MAX; + fwk_op_kernel.fwkKernelBase.fwk_kernel.kernelID = kernel_id; + fwk_op_kernel.fwkKernelBase.fwk_kernel.opType = aicpu::FWKAdapter::FWKOperateType::FWK_ADPT_KERNEL_RUN_NO_SESS; ret = SetKernelArgs(&task.args_, fwk_op_kernel); if (ret != SUCCESS) { return ret; diff --git a/ge/single_op/task/aicpu_task_builder.h b/ge/single_op/task/aicpu_task_builder.h index 6dcd7a0f..4669e118 100755 --- a/ge/single_op/task/aicpu_task_builder.h +++ b/ge/single_op/task/aicpu_task_builder.h @@ -29,7 +29,7 @@ namespace ge { AiCpuTaskBuilder(const OpDescPtr &op_desc, const domi::KernelExDef &kernel_def); ~AiCpuTaskBuilder() = default; - Status BuildTask(AiCpuTask &task, const SingleOpModelParam ¶m, bool dynamic_flag, uint64_t session_id); + Status BuildTask(AiCpuTask &task, const SingleOpModelParam ¶m, bool dynamic_flag, uint64_t kernel_id); private: static Status SetKernelArgs(void **args, STR_FWK_OP_KERNEL &kernel); diff --git a/ge/single_op/task/op_task.cc b/ge/single_op/task/op_task.cc index 30d4d311..1b4b23ff 100755 --- a/ge/single_op/task/op_task.cc +++ b/ge/single_op/task/op_task.cc @@ -245,7 +245,7 @@ AiCpuBaseTask::~AiCpuBaseTask() { } } -Status AiCpuBaseTask::SetExtInfoAndType(const std::string &kernel_ext_info) { +Status AiCpuBaseTask::SetExtInfoAndType(const std::string &kernel_ext_info, uint64_t kernel_id) { if (kernel_ext_info.empty()) { GELOGI("Kernel_ext_info is empty, no need copy to device."); return SUCCESS; @@ -268,9 +268,13 @@ Status AiCpuBaseTask::SetExtInfoAndType(const std::string &kernel_ext_info) { return ret; } - GE_CHK_RT_RET(rtMalloc(&ext_info_addr_dev_, kernel_ext_info.size(), RT_MEMORY_HBM)); - GE_CHK_RT_RET(rtMemcpy(ext_info_addr_dev_, kernel_ext_info.size(), - kernel_ext_info.data(), kernel_ext_info.size(), RT_MEMCPY_HOST_TO_DEVICE)); + GE_CHK_STATUS_RET(aicpu_ext_handle_->UpdateSessionInfo(ULLONG_MAX, kernel_id, false), + "UpdateSessionInfo failed."); + + GE_CHK_RT_RET(rtMalloc(&ext_info_addr_dev_, aicpu_ext_handle_->GetExtInfoLen(), RT_MEMORY_HBM)); + GE_CHK_RT_RET(rtMemcpy(ext_info_addr_dev_, aicpu_ext_handle_->GetExtInfoLen(), + aicpu_ext_handle_->GetExtInfo(), aicpu_ext_handle_->GetExtInfoLen(), + RT_MEMCPY_HOST_TO_DEVICE)); return SUCCESS; } diff --git a/ge/single_op/task/op_task.h b/ge/single_op/task/op_task.h index 1b4d9c02..bf8316f6 100644 --- a/ge/single_op/task/op_task.h +++ b/ge/single_op/task/op_task.h @@ -132,7 +132,7 @@ class AiCpuBaseTask : public OpTask { const UnknowShapeOpType GetUnknownType() const { return unknown_type_; } protected: - Status SetExtInfoAndType(const std::string &kernel_ext_info); + Status SetExtInfoAndType(const std::string &kernel_ext_info, uint64_t kernel_id); Status UpdateExtInfo(const std::vector &input_desc, std::vector &output_desc, diff --git a/inc/external/ge/ge_ir_build.h b/inc/external/ge/ge_ir_build.h index e6401093..7ac54781 100644 --- a/inc/external/ge/ge_ir_build.h +++ b/inc/external/ge/ge_ir_build.h @@ -27,12 +27,11 @@ namespace { #define IR_MAJOR_VERSION (int(1)) #define IR_MINOR_VERSION (int(0)) #define IR_PATCH_VERSION (int(0)) -} +} // namespace -namespace ge{ +namespace ge { -struct ModelBufferData -{ +struct ModelBufferData { std::shared_ptr data = nullptr; uint64_t length; }; @@ -64,7 +63,8 @@ void aclgrphBuildFinalize(); * @retval GRAPH_SUCCESS The function is successfully executed. * @retval OtherValues Failure */ -graphStatus aclgrphBuildModel(const ge::Graph &graph, const std::map &build_options, ModelBufferData& model); +graphStatus aclgrphBuildModel(const ge::Graph &graph, const std::map &build_options, + ModelBufferData &model); /** * @ingroup AscendCL @@ -75,7 +75,7 @@ graphStatus aclgrphBuildModel(const ge::Graph &graph, const std::map -#include -#include #include "framework/common/ge_inner_error_codes.h" #include "toolchain/slog.h" +#ifdef __GNUC__ +#include +#include +#else +#include "mmpa/mmpa_api.h" +#endif + +#ifdef __cplusplus +extern "C" { +#endif #define GE_MODULE_NAME static_cast(GE) // trace status of log enum TraceStatus { TRACE_INIT = 0, TRACE_RUNNING, TRACE_WAITING, TRACE_STOP }; -#define GELOGE(ERROR_CODE, ...) GE_LOG_ERROR(GE_MODULE_NAME, ERROR_CODE, __VA_ARGS__) -#define GELOGW(...) GE_LOG_WARN(GE_MODULE_NAME, __VA_ARGS__) -#define GELOGI(...) GE_LOG_INFO(GE_MODULE_NAME, __VA_ARGS__) -#define GELOGD(...) GE_LOG_DEBUG(GE_MODULE_NAME, __VA_ARGS__) -#define GEEVENT(...) GE_LOG_EVENT(GE_MODULE_NAME, __VA_ARGS__) -#define GELOGO(...) GE_LOG_OPLOG(GE_MODULE_NAME, __VA_ARGS__) -#define GELOGT(VALUE, ...) GE_LOG_TRACE(GE_MODULE_NAME, VALUE, __VA_ARGS__) +class GeLog { +public: +#ifdef __GNUC__ +static pid_t GetTid() { + thread_local static pid_t tid = syscall(__NR_gettid); + return tid; +} +#else +static int GetTid() { + thread_local static int tid = static_cast(GetCurrentThreadId()); + return tid; +} +#endif +}; inline bool IsLogEnable(int module_name, int log_level) { int32_t enable = CheckLogLevel(module_name, log_level); @@ -46,23 +61,41 @@ inline bool IsLogEnable(int module_name, int log_level) { return false; } -inline pid_t GetTid() { - thread_local static pid_t tid = syscall(__NR_gettid); - return tid; -} +#define GELOGE(ERROR_CODE, fmt, ...) \ + dlog_error(GE_MODULE_NAME, "%lu %s: ErrorNo: %d(%s) " fmt, GeLog::GetTid(), __FUNCTION__, ERROR_CODE, \ + ((GE_GET_ERRORNO_STR(ERROR_CODE)).c_str()), ##__VA_ARGS__) +#define GELOGW(fmt, ...) \ + if (IsLogEnable(GE_MODULE_NAME, DLOG_WARN)) dlog_warn(GE_MODULE_NAME, "%lu %s:" fmt, GeLog::GetTid(), __FUNCTION__, ##__VA_ARGS__) +#define GELOGI(fmt, ...) \ + if (IsLogEnable(GE_MODULE_NAME, DLOG_INFO)) dlog_info(GE_MODULE_NAME, "%lu %s:" fmt, GeLog::GetTid(), __FUNCTION__, ##__VA_ARGS__) +#define GELOGD(fmt, ...) \ + if (IsLogEnable(GE_MODULE_NAME, DLOG_DEBUG)) dlog_debug(GE_MODULE_NAME, "%lu %s:" fmt, GeLog::GetTid(), __FUNCTION__, ##__VA_ARGS__) +#define GEEVENT(fmt, ...) dlog_event(GE_MODULE_NAME, "%lu %s:" fmt, GeLog::GetTid(), __FUNCTION__, ##__VA_ARGS__) +#define GELOGO(fmt, ...) \ + Dlog(GE_MODULE_NAME, DLOG_OPLOG, "%lu %s:" fmt, GeLog::GetTid(), __FUNCTION__, ##__VA_ARGS__) +#define GELOGT(VALUE, fmt, ...) \ + do { \ + TraceStatus stat = VALUE; \ + const char *const TraceStatStr[] = {"INIT", "RUNNING", "WAITING", "STOP"}; \ + int idx = static_cast(stat); \ + char *k = const_cast("status"); \ + char *v = const_cast(TraceStatStr[idx]); \ + KeyValue kv = {k, v}; \ + DlogWithKV(static_cast(GE_MODULE_NAME), DLOG_TRACE, &kv, 1, "%lu %s:" fmt, GeLog::GetTid(), __FUNCTION__, ##__VA_ARGS__); \ + } while (0) #define GE_LOG_ERROR(MOD_NAME, ERROR_CODE, fmt, ...) \ - dlog_error(MOD_NAME, "%lu %s: ErrorNo: %d(%s) " fmt, GetTid(), __FUNCTION__, ERROR_CODE, \ + dlog_error(MOD_NAME, "%lu %s: ErrorNo: %d(%s) " fmt, GeLog::GetTid(), __FUNCTION__, ERROR_CODE, \ ((GE_GET_ERRORNO_STR(ERROR_CODE)).c_str()), ##__VA_ARGS__) #define GE_LOG_WARN(MOD_NAME, fmt, ...) \ - if (IsLogEnable(MOD_NAME, DLOG_WARN)) dlog_warn(MOD_NAME, "%lu %s:" fmt, GetTid(), __FUNCTION__, ##__VA_ARGS__) + if (IsLogEnable(MOD_NAME, DLOG_WARN)) dlog_warn(MOD_NAME, "%lu %s:" fmt, GeLog::GetTid(), __FUNCTION__, ##__VA_ARGS__) #define GE_LOG_INFO(MOD_NAME, fmt, ...) \ - if (IsLogEnable(MOD_NAME, DLOG_INFO)) dlog_info(MOD_NAME, "%lu %s:" fmt, GetTid(), __FUNCTION__, ##__VA_ARGS__) + if (IsLogEnable(MOD_NAME, DLOG_INFO)) dlog_info(MOD_NAME, "%lu %s:" fmt, GeLog::GetTid(), __FUNCTION__, ##__VA_ARGS__) #define GE_LOG_DEBUG(MOD_NAME, fmt, ...) \ - if (IsLogEnable(MOD_NAME, DLOG_DEBUG)) dlog_debug(MOD_NAME, "%lu %s:" fmt, GetTid(), __FUNCTION__, ##__VA_ARGS__) -#define GE_LOG_EVENT(MOD_NAME, fmt, ...) dlog_event(MOD_NAME, "%lu %s:" fmt, GetTid(), __FUNCTION__, ##__VA_ARGS__) + if (IsLogEnable(MOD_NAME, DLOG_DEBUG)) dlog_debug(MOD_NAME, "%lu %s:" fmt, GeLog::GetTid(), __FUNCTION__, ##__VA_ARGS__) +#define GE_LOG_EVENT(MOD_NAME, fmt, ...) dlog_event(MOD_NAME, "%lu %s:" fmt, GeLog::GetTid(), __FUNCTION__, ##__VA_ARGS__) #define GE_LOG_OPLOG(MOD_NAME, fmt, ...) \ - Dlog(MOD_NAME, DLOG_OPLOG, "%lu %s:" fmt, GetTid(), __FUNCTION__, ##__VA_ARGS__) + Dlog(MOD_NAME, DLOG_OPLOG, "%lu %s:" fmt, GeLog::GetTid(), __FUNCTION__, ##__VA_ARGS__) #define GE_LOG_TRACE(MOD_NAME, value, fmt, ...) \ do { \ @@ -72,7 +105,7 @@ inline pid_t GetTid() { char *k = const_cast("status"); \ char *v = const_cast(TraceStatStr[idx]); \ KeyValue kv = {k, v}; \ - DlogWithKV(static_cast(MOD_NAME), DLOG_TRACE, &kv, 1, "%lu %s:" fmt, GetTid(), __FUNCTION__, ##__VA_ARGS__); \ + DlogWithKV(static_cast(MOD_NAME), DLOG_TRACE, &kv, 1, "%lu %s:" fmt, GeLog::GetTid(), __FUNCTION__, ##__VA_ARGS__); \ } while (0) // print memory when it is greater than 1KB. @@ -82,4 +115,7 @@ inline pid_t GetTid() { GELOGI("MallocMemory, func=%s, size=%zu, purpose=%s", (#FUNC), static_cast(SIZE), (PURPOSE)); \ } \ } while (0); +#ifdef __cplusplus +} +#endif #endif // INC_FRAMEWORK_COMMON_DEBUG_GE_LOG_H_ diff --git a/inc/framework/common/ge_compiler_options.h b/inc/framework/common/ge_compiler_options.h new file mode 100644 index 00000000..5c947346 --- /dev/null +++ b/inc/framework/common/ge_compiler_options.h @@ -0,0 +1,32 @@ +/** + * Copyright 2020 Huawei Technologies Co., Ltd + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * 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. + */ + +#ifndef INC_FRAMEWORK_COMMON_GE_COMPILER_OPTIONS_H_ +#define INC_FRAMEWORK_COMMON_GE_COMPILER_OPTIONS_H_ + +namespace ge { +#ifdef __GNUC__ +#define GE_ATTRIBUTE_UNUSED __attribute__((unused)) +#define GE_FUNCTION_IDENTIFIER __PRETTY_FUNCTION__ +#define GE_BUILTIN_PREFETCH(args_addr) __builtin_prefetch(args_addr) +#else +#define GE_ATTRIBUTE_UNUSED +#define GE_FUNCTION_IDENTIFIER __FUNCSIG__ +#define GE_BUILTIN_PREFETCH(args_addr) +#endif +} // namespace ge + +#endif // INC_FRAMEWORK_COMMON_GE_COMPILER_OPTIONS_H_ \ No newline at end of file diff --git a/inc/framework/common/string_util.h b/inc/framework/common/string_util.h index 47e80e75..de19807c 100644 --- a/inc/framework/common/string_util.h +++ b/inc/framework/common/string_util.h @@ -17,6 +17,7 @@ #ifndef INC_FRAMEWORK_COMMON_STRING_UTIL_H_ #define INC_FRAMEWORK_COMMON_STRING_UTIL_H_ +#include #include #include diff --git a/inc/framework/common/types.h b/inc/framework/common/types.h index 2a4f3ca3..22e85e0b 100644 --- a/inc/framework/common/types.h +++ b/inc/framework/common/types.h @@ -18,7 +18,6 @@ #define INC_FRAMEWORK_COMMON_TYPES_H_ #include -#include #include #include #include @@ -529,7 +528,7 @@ REGISTER_OPTYPE_DECLARE(HVDWAIT, "HorovodWait"); // aicpu op for online_infer dynamic_dims REGISTER_OPTYPE_DECLARE(GETDYNAMICDIMS, "GetDynamicDims"); -enum InputMode { INPUT = 0, CONST }; +enum InputMode { INPUT = 0, CONST_INPUT}; // Definition of the processing status enum of the process module enum ModelProcessState { diff --git a/inc/framework/ge_runtime/task_info.h b/inc/framework/ge_runtime/task_info.h index e36c4333..f59c6454 100644 --- a/inc/framework/ge_runtime/task_info.h +++ b/inc/framework/ge_runtime/task_info.h @@ -18,7 +18,6 @@ #define INC_FRAMEWORK_GE_RUNTIME_TASK_INFO_H_ #include -#include #include #include #include @@ -219,9 +218,9 @@ class LabelSwitchTaskInfo : public TaskInfo { label_list_(label_list), cond_(cond) {} ~LabelSwitchTaskInfo() override {} - uint32_t label_size() { return label_size_; }; - const std::vector &label_list() { return label_list_; }; - void *cond() { return cond_; }; + uint32_t label_size() const { return label_size_; } + const std::vector &label_list() const { return label_list_; } + void *cond() const { return cond_; } private: uint32_t label_size_; @@ -236,7 +235,7 @@ class EventTaskInfo : public TaskInfo { protected: EventTaskInfo(const std::string &op_name, uint32_t stream_id, TaskInfoType type, uint32_t event_id) : TaskInfo(op_name, stream_id, type, false), event_id_(event_id) {} - virtual ~EventTaskInfo() override {} + ~EventTaskInfo() override {} uint32_t event_id_; }; @@ -272,16 +271,13 @@ class FusionEndTaskInfo : public TaskInfo { class HcclTaskInfo : public TaskInfo { public: HcclTaskInfo(const std::string &op_name, uint32_t stream_id, const std::string hccl_type, void *input_data_addr, - void *output_data_addr, void *workspace_addr, int64_t workspace_size, int64_t hccl_stream_num, + void *output_data_addr, int64_t workspace_size, int64_t hccl_stream_num, const std::vector &private_def, void *ops_kernel_store, int32_t count, int64_t root_id, - int64_t op_type, int64_t data_type, const std::string &group, - std::function hcom_bind_model, std::function hcom_unbind_model, - std::function, void *)> hcom_distribute_task, bool dump_flag) + int64_t op_type, int64_t data_type, const std::string &group, bool dump_flag) : TaskInfo(op_name, stream_id, TaskInfoType::HCCL, dump_flag), hccl_type_(hccl_type), input_data_addr_(input_data_addr), output_data_addr_(output_data_addr), - workspace_addr_(workspace_addr), workspace_size_(workspace_size), hccl_stream_num_(hccl_stream_num), private_def_(private_def), @@ -290,16 +286,12 @@ class HcclTaskInfo : public TaskInfo { root_id_(root_id), op_type_(op_type), data_type_(data_type), - group_(group), - hcom_bind_model_(hcom_bind_model), - hcom_unbind_model_(hcom_unbind_model), - hcom_distribute_task_(hcom_distribute_task) {} + group_(group) {} ~HcclTaskInfo() override {} const std::string &hccl_type() const { return hccl_type_; } void *input_data_addr() const { return input_data_addr_; } void *output_data_addr() const { return output_data_addr_; } - void *workspace_addr() const { return workspace_addr_; } int64_t workspace_size() const { return workspace_size_; } int64_t hccl_stream_num() const { return hccl_stream_num_; } const std::vector &private_def() const { return private_def_; } @@ -309,17 +301,11 @@ class HcclTaskInfo : public TaskInfo { int64_t op_type() const { return op_type_; } int64_t data_type() const { return data_type_; } const std::string &group() const { return group_; } - std::function hcom_bind_model() const { return hcom_bind_model_; } - std::function hcom_unbind_model() const { return hcom_unbind_model_; } - std::function, void *)> hcom_distribute_task() const { - return hcom_distribute_task_; - } private: std::string hccl_type_; void *input_data_addr_; void *output_data_addr_; - void *workspace_addr_; int64_t workspace_size_; int64_t hccl_stream_num_; std::vector private_def_; @@ -329,9 +315,6 @@ class HcclTaskInfo : public TaskInfo { int64_t op_type_; int64_t data_type_; std::string group_; - std::function hcom_bind_model_; - std::function hcom_unbind_model_; - std::function, void *)> hcom_distribute_task_; }; class ProfilerTraceTaskInfo : public TaskInfo { diff --git a/metadef b/metadef index 1b09ed04..57e72aac 160000 --- a/metadef +++ b/metadef @@ -1 +1 @@ -Subproject commit 1b09ed04b6dd22d1aed1bee92fd42736c0fafc65 +Subproject commit 57e72aac24a35e40799e342fdacca362a66395c4 diff --git a/parser b/parser index 3d49906d..bb6424dc 160000 --- a/parser +++ b/parser @@ -1 +1 @@ -Subproject commit 3d49906d119b1cc01f4256d7992759ce9f3dcfcd +Subproject commit bb6424dc6d9252a3ac70650cde2f547761237681 diff --git a/third_party/fwkacllib/inc/cce/aicpu_engine_struct.h b/third_party/fwkacllib/inc/cce/aicpu_engine_struct.h index d61c981d..a5f43be9 100644 --- a/third_party/fwkacllib/inc/cce/aicpu_engine_struct.h +++ b/third_party/fwkacllib/inc/cce/aicpu_engine_struct.h @@ -40,6 +40,12 @@ typedef struct { } fwkKernelBase; } __attribute__((packed)) STR_FWK_OP_KERNEL; +struct SessionInfo { + uint64_t sessionId; + uint64_t kernelId; + bool sessFlag; +} __attribute__((packed)); + #ifdef __cplusplus } #endif diff --git a/third_party/fwkacllib/inc/cce/fwk_adpt_struct.h b/third_party/fwkacllib/inc/cce/fwk_adpt_struct.h index 957117cc..79d94023 100644 --- a/third_party/fwkacllib/inc/cce/fwk_adpt_struct.h +++ b/third_party/fwkacllib/inc/cce/fwk_adpt_struct.h @@ -48,7 +48,8 @@ enum FWKOperateType { FWK_ADPT_KERNEL_RUN, FWK_ADPT_KERNEL_DESTROY, FWK_ADPT_SESSION_DESTROY, - FWK_ADPT_SINGLE_OP_RUN + FWK_ADPT_SINGLE_OP_RUN, + FWK_ADPT_KERNEL_RUN_NO_SESS, }; // Extend Info type for task @@ -58,6 +59,7 @@ enum FWKTaskExtInfoType { FWK_ADPT_EXT_OUTPUT_SHAPE, FWK_ADPT_EXT_UPDATE_ADDR, FWK_ADPT_EXT_OP_NAME, + FWK_ADPT_EXT_SESSION_INFO, FWK_ADPT_EXT_INVALID }; diff --git a/third_party/fwkacllib/inc/hccl/base.h b/third_party/fwkacllib/inc/hccl/base.h index 94253bf4..8194097e 100644 --- a/third_party/fwkacllib/inc/hccl/base.h +++ b/third_party/fwkacllib/inc/hccl/base.h @@ -68,8 +68,10 @@ struct MemRegisterAddr { u64 addr; u64 length; }; - -const u32 HCCL_MAX_MEM_REGISTER_NUM = 8; // The max number of memory register address. +/* + * @brief The max number of memory register address for remote access. + */ +const u32 HCCL_MAX_MEM_REGISTER_NUM = 32; enum GradSplitForceMode { FORCE_NONE, /**< no force */ diff --git a/third_party/fwkacllib/inc/mmpa/mmpa_api.h b/third_party/fwkacllib/inc/mmpa/mmpa_api.h index f5b80269..38a689ee 100644 --- a/third_party/fwkacllib/inc/mmpa/mmpa_api.h +++ b/third_party/fwkacllib/inc/mmpa/mmpa_api.h @@ -26,6 +26,12 @@ #define _GNU_SOURCE #endif +#ifdef FUNC_VISIBILITY +#define MMPA_FUNC_VISIBILITY __attribute__((visibility("default"))) +#else +#define MMPA_FUNC_VISIBILITY +#endif + #include #include #include @@ -86,6 +92,13 @@ #if(OS_TYPE == WIN) //lint !e553 + +#ifdef FUNC_VISIBILITY +#define MMPA_FUNC_VISIBILITY _declspec(dllexport) +#else +#define MMPA_FUNC_VISIBILITY +#endif + #include #include #include "Windows.h" diff --git a/third_party/fwkacllib/inc/mmpa/sub_inc/mmpa_linux.h b/third_party/fwkacllib/inc/mmpa/sub_inc/mmpa_linux.h index aced4968..ea51f497 100644 --- a/third_party/fwkacllib/inc/mmpa/sub_inc/mmpa_linux.h +++ b/third_party/fwkacllib/inc/mmpa/sub_inc/mmpa_linux.h @@ -44,12 +44,16 @@ typedef VOID *mmExitCode; typedef key_t mmKey_t; typedef int mmMsgid; typedef struct dirent mmDirent; +typedef struct dirent mmDirent2; typedef struct shmid_ds mmshmId_ds; typedef int (*mmFilter)(const mmDirent *entry); +typedef int (*mmFilter2)(const mmDirent2 *entry); typedef int (*mmSort)(const mmDirent **a, const mmDirent **b); +typedef int (*mmSort2)(const mmDirent2 **a, const mmDirent2 **b); typedef size_t mmSize_t; typedef off_t mmOfft_t; typedef pid_t mmPid_t; +typedef long MM_LONG; typedef VOID *(*userProcFunc)(VOID *pulArg); @@ -113,6 +117,7 @@ typedef struct { } mmIoctlBuf; typedef int mmAtomicType; +typedef int mmAtomicType64; typedef enum { pollTypeRead = 1, // pipe read @@ -151,8 +156,8 @@ typedef struct { } mmTimeval; typedef struct { - LONG tv_sec; - LONG tv_nsec; + MM_LONG tv_sec; + MM_LONG tv_nsec; } mmTimespec; typedef struct { @@ -272,6 +277,9 @@ typedef struct { #define M_R_OK R_OK #define M_W_OK W_OK +#define MM_DT_DIR DT_DIR +#define MM_DT_REG DT_REG + #define MMPA_STDIN STDIN_FILENO #define MMPA_STDOUT STDOUT_FILENO #define MMPA_STDERR STDERR_FILENO @@ -283,182 +291,195 @@ typedef struct { #define MMPA_DL_EXT_NAME ".so" -extern INT32 mmCreateTask(mmThread *threadHandle, mmUserBlock_t *funcBlock); -extern INT32 mmJoinTask(mmThread *threadHandle); -extern INT32 mmMutexInit(mmMutex_t *mutex); -extern INT32 mmMutexLock(mmMutex_t *mutex); -extern INT32 mmMutexTryLock(mmMutex_t *mutex); -extern INT32 mmMutexUnLock(mmMutex_t *mutex); -extern INT32 mmMutexDestroy(mmMutex_t *mutex); -extern INT32 mmCondInit(mmCond *cond); -extern INT32 mmCondLockInit(mmMutexFC *mutex); -extern INT32 mmCondLock(mmMutexFC *mutex); -extern INT32 mmCondUnLock(mmMutexFC *mutex); -extern INT32 mmCondLockDestroy(mmMutexFC *mutex); -extern INT32 mmRWLockInit(mmRWLock_t *rwLock); -extern INT32 mmRWLockRDLock(mmRWLock_t *rwLock); -extern INT32 mmRWLockTryRDLock(mmRWLock_t *rwLock); -extern INT32 mmRWLockWRLock(mmRWLock_t *rwLock); -extern INT32 mmRWLockTryWRLock(mmRWLock_t *rwLock); -extern INT32 mmRDLockUnLock(mmRWLock_t *rwLock); -extern INT32 mmWRLockUnLock(mmRWLock_t *rwLock); -extern INT32 mmRWLockDestroy(mmRWLock_t *rwLock); -extern INT32 mmCondWait(mmCond *cond, mmMutexFC *mutex); -extern INT32 mmCondTimedWait(mmCond *cond, mmMutexFC *mutex, UINT32 milliSecond); -extern INT32 mmCondNotify(mmCond *cond); -extern INT32 mmCondNotifyAll(mmCond *cond); -extern INT32 mmCondDestroy(mmCond *cond); -extern INT32 mmGetPid(); -extern INT32 mmGetTid(); -extern INT32 mmGetPidHandle(mmProcess *processHandle); -extern INT32 mmGetLocalTime(mmSystemTime_t *sysTime); -extern INT32 mmGetSystemTime(mmSystemTime_t *sysTime); - -extern INT32 mmSemInit(mmSem_t *sem, UINT32 value); -extern INT32 mmSemWait(mmSem_t *sem); -extern INT32 mmSemPost(mmSem_t *sem); -extern INT32 mmSemDestroy(mmSem_t *sem); -extern INT32 mmOpen(const CHAR *pathName, INT32 flags); -extern INT32 mmOpen2(const CHAR *pathName, INT32 flags, MODE mode); -extern FILE *mmPopen(CHAR *command, CHAR *type); -extern INT32 mmClose(INT32 fd); -extern INT32 mmPclose(FILE *stream); -extern mmSsize_t mmWrite(INT32 fd, VOID *buf, UINT32 bufLen); -extern mmSsize_t mmRead(INT32 fd, VOID *buf, UINT32 bufLen); -extern mmSockHandle mmSocket(INT32 sockFamily, INT32 type, INT32 protocol); -extern INT32 mmBind(mmSockHandle sockFd, mmSockAddr *addr, mmSocklen_t addrLen); -extern INT32 mmListen(mmSockHandle sockFd, INT32 backLog); -extern mmSockHandle mmAccept(mmSockHandle sockFd, mmSockAddr *addr, mmSocklen_t *addrLen); -extern INT32 mmConnect(mmSockHandle sockFd, mmSockAddr *addr, mmSocklen_t addrLen); -extern INT32 mmCloseSocket(mmSockHandle sockFd); -extern mmSsize_t mmSocketSend(mmSockHandle sockFd, VOID *sendBuf, INT32 sendLen, INT32 sendFlag); -extern mmSsize_t mmSocketRecv(mmSockHandle sockFd, VOID *recvBuf, INT32 recvLen, INT32 recvFlag); -extern INT32 mmSocketSendTo(mmSockHandle sockFd, +MMPA_FUNC_VISIBILITY INT32 mmCreateTask(mmThread *threadHandle, mmUserBlock_t *funcBlock); +MMPA_FUNC_VISIBILITY INT32 mmJoinTask(mmThread *threadHandle); +MMPA_FUNC_VISIBILITY INT32 mmMutexInit(mmMutex_t *mutex); +MMPA_FUNC_VISIBILITY INT32 mmMutexLock(mmMutex_t *mutex); +MMPA_FUNC_VISIBILITY INT32 mmMutexTryLock(mmMutex_t *mutex); +MMPA_FUNC_VISIBILITY INT32 mmMutexUnLock(mmMutex_t *mutex); +MMPA_FUNC_VISIBILITY INT32 mmMutexDestroy(mmMutex_t *mutex); +MMPA_FUNC_VISIBILITY INT32 mmCondInit(mmCond *cond); +MMPA_FUNC_VISIBILITY INT32 mmCondLockInit(mmMutexFC *mutex); +MMPA_FUNC_VISIBILITY INT32 mmCondLock(mmMutexFC *mutex); +MMPA_FUNC_VISIBILITY INT32 mmCondUnLock(mmMutexFC *mutex); +MMPA_FUNC_VISIBILITY INT32 mmCondLockDestroy(mmMutexFC *mutex); +MMPA_FUNC_VISIBILITY INT32 mmRWLockInit(mmRWLock_t *rwLock); +MMPA_FUNC_VISIBILITY INT32 mmRWLockRDLock(mmRWLock_t *rwLock); +MMPA_FUNC_VISIBILITY INT32 mmRWLockTryRDLock(mmRWLock_t *rwLock); +MMPA_FUNC_VISIBILITY INT32 mmRWLockWRLock(mmRWLock_t *rwLock); +MMPA_FUNC_VISIBILITY INT32 mmRWLockTryWRLock(mmRWLock_t *rwLock); +MMPA_FUNC_VISIBILITY INT32 mmRDLockUnLock(mmRWLock_t *rwLock); +MMPA_FUNC_VISIBILITY INT32 mmWRLockUnLock(mmRWLock_t *rwLock); +MMPA_FUNC_VISIBILITY INT32 mmRWLockDestroy(mmRWLock_t *rwLock); +MMPA_FUNC_VISIBILITY INT32 mmCondWait(mmCond *cond, mmMutexFC *mutex); +MMPA_FUNC_VISIBILITY INT32 mmCondTimedWait(mmCond *cond, mmMutexFC *mutex, UINT32 milliSecond); +MMPA_FUNC_VISIBILITY INT32 mmCondNotify(mmCond *cond); +MMPA_FUNC_VISIBILITY INT32 mmCondNotifyAll(mmCond *cond); +MMPA_FUNC_VISIBILITY INT32 mmCondDestroy(mmCond *cond); +MMPA_FUNC_VISIBILITY INT32 mmGetPid(); +MMPA_FUNC_VISIBILITY INT32 mmGetTid(); +MMPA_FUNC_VISIBILITY INT32 mmGetPidHandle(mmProcess *processHandle); +MMPA_FUNC_VISIBILITY INT32 mmGetLocalTime(mmSystemTime_t *sysTime); +MMPA_FUNC_VISIBILITY INT32 mmGetSystemTime(mmSystemTime_t *sysTime); + +MMPA_FUNC_VISIBILITY INT32 mmSemInit(mmSem_t *sem, UINT32 value); +MMPA_FUNC_VISIBILITY INT32 mmSemWait(mmSem_t *sem); +MMPA_FUNC_VISIBILITY INT32 mmSemPost(mmSem_t *sem); +MMPA_FUNC_VISIBILITY INT32 mmSemDestroy(mmSem_t *sem); +MMPA_FUNC_VISIBILITY INT32 mmOpen(const CHAR *pathName, INT32 flags); +MMPA_FUNC_VISIBILITY INT32 mmOpen2(const CHAR *pathName, INT32 flags, MODE mode); +MMPA_FUNC_VISIBILITY FILE *mmPopen(CHAR *command, CHAR *type); +MMPA_FUNC_VISIBILITY INT32 mmClose(INT32 fd); +MMPA_FUNC_VISIBILITY INT32 mmPclose(FILE *stream); +MMPA_FUNC_VISIBILITY mmSsize_t mmWrite(INT32 fd, VOID *buf, UINT32 bufLen); +MMPA_FUNC_VISIBILITY mmSsize_t mmRead(INT32 fd, VOID *buf, UINT32 bufLen); +MMPA_FUNC_VISIBILITY mmSockHandle mmSocket(INT32 sockFamily, INT32 type, INT32 protocol); +MMPA_FUNC_VISIBILITY INT32 mmBind(mmSockHandle sockFd, mmSockAddr *addr, mmSocklen_t addrLen); +MMPA_FUNC_VISIBILITY INT32 mmListen(mmSockHandle sockFd, INT32 backLog); +MMPA_FUNC_VISIBILITY mmSockHandle mmAccept(mmSockHandle sockFd, mmSockAddr *addr, mmSocklen_t *addrLen); +MMPA_FUNC_VISIBILITY INT32 mmConnect(mmSockHandle sockFd, mmSockAddr *addr, mmSocklen_t addrLen); +MMPA_FUNC_VISIBILITY INT32 mmCloseSocket(mmSockHandle sockFd); +MMPA_FUNC_VISIBILITY mmSsize_t mmSocketSend(mmSockHandle sockFd, VOID *sendBuf, INT32 sendLen, INT32 sendFlag); +MMPA_FUNC_VISIBILITY mmSsize_t mmSocketRecv(mmSockHandle sockFd, VOID *recvBuf, INT32 recvLen, INT32 recvFlag); +MMPA_FUNC_VISIBILITY INT32 mmSocketSendTo(mmSockHandle sockFd, VOID *sendMsg, INT32 sendLen, UINT32 sendFlag, const mmSockAddr* addr, INT32 tolen); -extern mmSsize_t mmSocketRecvFrom(mmSockHandle sockFd, +MMPA_FUNC_VISIBILITY mmSsize_t mmSocketRecvFrom(mmSockHandle sockFd, VOID *recvBuf, mmSize recvLen, UINT32 recvFlag, mmSockAddr* addr, mmSocklen_t *FromLen); -extern INT32 mmSAStartup(); -extern INT32 mmSACleanup(); -extern VOID *mmDlopen(const CHAR *fileName, INT32 mode); -extern INT32 mmDladdr(VOID *addr, mmDlInfo *info); -extern VOID *mmDlsym(VOID *handle, CHAR *funcName); -extern INT32 mmDlclose(VOID *handle); -extern CHAR *mmDlerror(); -extern INT32 mmCreateAndSetTimer(mmTimer *timerHandle, mmUserBlock_t *timerBlock, UINT milliSecond, UINT period); -extern INT32 mmDeleteTimer(mmTimer timerHandle); -extern INT32 mmStatGet(const CHAR *path, mmStat_t *buffer); -extern INT32 mmStat64Get(const CHAR *path, mmStat64_t *buffer); -extern INT32 mmFStatGet(INT32 fd, mmStat_t *buffer); -extern INT32 mmMkdir(const CHAR *pathName, mmMode_t mode); -extern INT32 mmSleep(UINT32 milliSecond); - -extern INT32 mmCreateTaskWithAttr(mmThread *threadHandle, mmUserBlock_t *funcBlock); -extern INT32 mmGetProcessPrio(mmProcess pid); -extern INT32 mmSetProcessPrio(mmProcess pid, INT32 processPrio); -extern INT32 mmGetThreadPrio(mmThread *threadHandle); -extern INT32 mmSetThreadPrio(mmThread *threadHandle, INT32 threadPrio); -extern INT32 mmAccess(const CHAR *pathName); -extern INT32 mmAccess2(const CHAR *pathName, INT32 mode); -extern INT32 mmRmdir(const CHAR *pathName); - -extern INT32 mmIoctl(mmProcess fd, INT32 ioctlCode, mmIoctlBuf *bufPtr); -extern INT32 mmSemTimedWait(mmSem_t *sem, INT32 timeout); -extern mmSsize_t mmWritev(mmProcess fd, mmIovSegment *iov, INT32 iovcnt); -extern VOID mmMb(); -extern INT32 mmInetAton(const CHAR *addrStr, mmInAddr *addr); - -extern mmProcess mmOpenFile(const CHAR *fileName, UINT32 access, mmCreateFlag fileFlag); -extern mmSsize_t mmReadFile(mmProcess fileId, VOID *buffer, INT32 len); -extern mmSsize_t mmWriteFile(mmProcess fileId, VOID *buffer, INT32 len); -extern INT32 mmCloseFile(mmProcess fileId); - -extern mmAtomicType mmSetData(mmAtomicType *ptr, mmAtomicType value); -extern mmAtomicType mmValueInc(mmAtomicType *ptr, mmAtomicType value); -extern mmAtomicType mmValueSub(mmAtomicType *ptr, mmAtomicType value); -extern INT32 mmCreateTaskWithDetach(mmThread *threadHandle, mmUserBlock_t *funcBlock); +MMPA_FUNC_VISIBILITY INT32 mmSAStartup(); +MMPA_FUNC_VISIBILITY INT32 mmSACleanup(); +MMPA_FUNC_VISIBILITY VOID *mmDlopen(const CHAR *fileName, INT32 mode); +MMPA_FUNC_VISIBILITY INT32 mmDladdr(VOID *addr, mmDlInfo *info); +MMPA_FUNC_VISIBILITY VOID *mmDlsym(VOID *handle, const CHAR *funcName); +MMPA_FUNC_VISIBILITY INT32 mmDlclose(VOID *handle); +MMPA_FUNC_VISIBILITY CHAR *mmDlerror(); +MMPA_FUNC_VISIBILITY INT32 mmCreateAndSetTimer(mmTimer *timerHandle, mmUserBlock_t *timerBlock, UINT milliSecond, UINT period); +MMPA_FUNC_VISIBILITY INT32 mmDeleteTimer(mmTimer timerHandle); +MMPA_FUNC_VISIBILITY INT32 mmStatGet(const CHAR *path, mmStat_t *buffer); +MMPA_FUNC_VISIBILITY INT32 mmStat64Get(const CHAR *path, mmStat64_t *buffer); +MMPA_FUNC_VISIBILITY INT32 mmFStatGet(INT32 fd, mmStat_t *buffer); +MMPA_FUNC_VISIBILITY INT32 mmMkdir(const CHAR *pathName, mmMode_t mode); +MMPA_FUNC_VISIBILITY INT32 mmSleep(UINT32 milliSecond); + +MMPA_FUNC_VISIBILITY INT32 mmCreateTaskWithAttr(mmThread *threadHandle, mmUserBlock_t *funcBlock); +MMPA_FUNC_VISIBILITY INT32 mmGetProcessPrio(mmProcess pid); +MMPA_FUNC_VISIBILITY INT32 mmSetProcessPrio(mmProcess pid, INT32 processPrio); +MMPA_FUNC_VISIBILITY INT32 mmGetThreadPrio(mmThread *threadHandle); +MMPA_FUNC_VISIBILITY INT32 mmSetThreadPrio(mmThread *threadHandle, INT32 threadPrio); +MMPA_FUNC_VISIBILITY INT32 mmAccess(const CHAR *pathName); +MMPA_FUNC_VISIBILITY INT32 mmAccess2(const CHAR *pathName, INT32 mode); +MMPA_FUNC_VISIBILITY INT32 mmRmdir(const CHAR *pathName); + +MMPA_FUNC_VISIBILITY INT32 mmIoctl(mmProcess fd, INT32 ioctlCode, mmIoctlBuf *bufPtr); +MMPA_FUNC_VISIBILITY INT32 mmSemTimedWait(mmSem_t *sem, INT32 timeout); +MMPA_FUNC_VISIBILITY mmSsize_t mmWritev(mmProcess fd, mmIovSegment *iov, INT32 iovcnt); +MMPA_FUNC_VISIBILITY VOID mmMb(); +MMPA_FUNC_VISIBILITY INT32 mmInetAton(const CHAR *addrStr, mmInAddr *addr); + +MMPA_FUNC_VISIBILITY mmProcess mmOpenFile(const CHAR *fileName, UINT32 access, mmCreateFlag fileFlag); +MMPA_FUNC_VISIBILITY mmSsize_t mmReadFile(mmProcess fileId, VOID *buffer, INT32 len); +MMPA_FUNC_VISIBILITY mmSsize_t mmWriteFile(mmProcess fileId, VOID *buffer, INT32 len); +MMPA_FUNC_VISIBILITY INT32 mmCloseFile(mmProcess fileId); + +MMPA_FUNC_VISIBILITY mmAtomicType mmSetData(mmAtomicType *ptr, mmAtomicType value); +MMPA_FUNC_VISIBILITY mmAtomicType mmValueInc(mmAtomicType *ptr, mmAtomicType value); +MMPA_FUNC_VISIBILITY mmAtomicType mmValueSub(mmAtomicType *ptr, mmAtomicType value); +MMPA_FUNC_VISIBILITY mmAtomicType64 mmSetData64(mmAtomicType64 *ptr, mmAtomicType64 value); +MMPA_FUNC_VISIBILITY mmAtomicType64 mmValueInc64(mmAtomicType64 *ptr, mmAtomicType64 value); +MMPA_FUNC_VISIBILITY mmAtomicType64 mmValueSub64(mmAtomicType64 *ptr, mmAtomicType64 value); +MMPA_FUNC_VISIBILITY INT32 mmCreateTaskWithDetach(mmThread *threadHandle, mmUserBlock_t *funcBlock); // The following 3 interfaces are to be deleted -extern INT32 mmCreateNamedPipe(mmPipeHandle pipe[], CHAR *pipeName[], INT32 waitMode); -extern INT32 mmOpenNamePipe(mmPipeHandle pipe[], CHAR *pipeName[], INT32 waitMode); -extern VOID mmCloseNamedPipe(mmPipeHandle namedPipe[]); +MMPA_FUNC_VISIBILITY INT32 mmCreateNamedPipe(mmPipeHandle pipe[], CHAR *pipeName[], INT32 waitMode); +MMPA_FUNC_VISIBILITY INT32 mmOpenNamePipe(mmPipeHandle pipe[], CHAR *pipeName[], INT32 waitMode); +MMPA_FUNC_VISIBILITY VOID mmCloseNamedPipe(mmPipeHandle namedPipe[]); -extern INT32 mmCreatePipe(mmPipeHandle pipe[], CHAR *pipeName[], UINT32 pipeCount, INT32 waitMode); -extern INT32 mmOpenPipe(mmPipeHandle pipe[], CHAR *pipeName[], UINT32 pipeCount, INT32 waitMode); -extern VOID mmClosePipe(mmPipeHandle pipe[], UINT32 pipeCount); +MMPA_FUNC_VISIBILITY INT32 mmCreatePipe(mmPipeHandle pipe[], CHAR *pipeName[], UINT32 pipeCount, INT32 waitMode); +MMPA_FUNC_VISIBILITY INT32 mmOpenPipe(mmPipeHandle pipe[], CHAR *pipeName[], UINT32 pipeCount, INT32 waitMode); +MMPA_FUNC_VISIBILITY VOID mmClosePipe(mmPipeHandle pipe[], UINT32 pipeCount); // Poll related interface -extern mmCompletionHandle mmCreateCompletionPort(); -extern VOID mmCloseCompletionPort(mmCompletionHandle handle); -extern INT32 mmPoll(mmPollfd *fds, INT32 fdCount, INT32 timeout, mmCompletionHandle handleIOCP, pmmPollData polledData, - mmPollBack pollBack); -extern INT32 mmGetErrorCode(); -extern CHAR *mmGetErrorFormatMessage(mmErrorMsg errnum, CHAR *buf, mmSize size); -extern INT32 mmGetTimeOfDay(mmTimeval *timeVal, mmTimezone *timeZone); -extern mmTimespec mmGetTickCount(); -extern INT32 mmGetRealPath(CHAR *path, CHAR *realPath); -extern INT32 mmRealPath(const CHAR *path, CHAR *realPath, INT32 realPathLen); +MMPA_FUNC_VISIBILITY mmCompletionHandle mmCreateCompletionPort(); +MMPA_FUNC_VISIBILITY VOID mmCloseCompletionPort(mmCompletionHandle handle); +MMPA_FUNC_VISIBILITY INT32 mmPoll(mmPollfd *fds, INT32 fdCount, INT32 timeout, mmCompletionHandle handleIOCP, + pmmPollData polledData, mmPollBack pollBack); +MMPA_FUNC_VISIBILITY INT32 mmGetErrorCode(); +MMPA_FUNC_VISIBILITY CHAR *mmGetErrorFormatMessage(mmErrorMsg errnum, CHAR *buf, mmSize size); +MMPA_FUNC_VISIBILITY INT32 mmGetTimeOfDay(mmTimeval *timeVal, mmTimezone *timeZone); +MMPA_FUNC_VISIBILITY mmTimespec mmGetTickCount(); +MMPA_FUNC_VISIBILITY INT32 mmGetRealPath(CHAR *path, CHAR *realPath); +MMPA_FUNC_VISIBILITY INT32 mmRealPath(const CHAR *path, CHAR *realPath, INT32 realPathLen); -extern INT32 mmDup2(INT32 oldFd, INT32 newFd); +MMPA_FUNC_VISIBILITY INT32 mmDup2(INT32 oldFd, INT32 newFd); -extern INT32 mmDup(INT32 fd); +MMPA_FUNC_VISIBILITY INT32 mmDup(INT32 fd); -extern INT32 mmUnlink(const CHAR *filename); +MMPA_FUNC_VISIBILITY INT32 mmUnlink(const CHAR *filename); -extern INT32 mmChmod(const CHAR *filename, INT32 mode); +MMPA_FUNC_VISIBILITY INT32 mmChmod(const CHAR *filename, INT32 mode); -extern INT32 mmFileno(FILE *stream); +MMPA_FUNC_VISIBILITY INT32 mmFileno(FILE *stream); -extern INT32 mmScandir(const CHAR *path, mmDirent ***entryList, mmFilter filterFunc, mmSort sort); +MMPA_FUNC_VISIBILITY INT32 mmScandir(const CHAR *path, mmDirent ***entryList, mmFilter filterFunc, mmSort sort); +MMPA_FUNC_VISIBILITY INT32 mmScandir2(const CHAR *path, mmDirent2 ***entryList, mmFilter2 filterFunc, mmSort2 sort); -extern VOID mmScandirFree(mmDirent **entryList, INT32 count); +MMPA_FUNC_VISIBILITY VOID mmScandirFree(mmDirent **entryList, INT32 count); +MMPA_FUNC_VISIBILITY VOID mmScandirFree2(mmDirent2 **entryList, INT32 count); -extern mmMsgid mmMsgCreate(mmKey_t key, INT32 msgFlag); +MMPA_FUNC_VISIBILITY mmMsgid mmMsgCreate(mmKey_t key, INT32 msgFlag); -extern mmMsgid mmMsgOpen(mmKey_t key, INT32 msgFlag); +MMPA_FUNC_VISIBILITY mmMsgid mmMsgOpen(mmKey_t key, INT32 msgFlag); -extern INT32 mmMsgSnd(mmMsgid msqid, VOID *buf, INT32 bufLen, INT32 msgFlag); +MMPA_FUNC_VISIBILITY INT32 mmMsgSnd(mmMsgid msqid, VOID *buf, INT32 bufLen, INT32 msgFlag); -extern INT32 mmMsgRcv(mmMsgid msqid, VOID *buf, INT32 bufLen, INT32 msgFlag); +MMPA_FUNC_VISIBILITY INT32 mmMsgRcv(mmMsgid msqid, VOID *buf, INT32 bufLen, INT32 msgFlag); -extern INT32 mmMsgClose(mmMsgid msqid); +MMPA_FUNC_VISIBILITY INT32 mmMsgClose(mmMsgid msqid); -extern INT32 mmLocalTimeR(const time_t *timep, struct tm *result); +MMPA_FUNC_VISIBILITY INT32 mmLocalTimeR(const time_t *timep, struct tm *result); -extern INT32 mmGetOpt(INT32 argc, char *const *argv, const char *opts); -extern INT32 mmGetOptLong(INT32 argc, char *const *argv, const char *opts, const mmStructOption *longOpts, - INT32 *longIndex); +MMPA_FUNC_VISIBILITY INT32 mmGetOptErr(); +MMPA_FUNC_VISIBILITY VOID mmSetOptErr(INT32 mmOptErr); +MMPA_FUNC_VISIBILITY INT32 mmGetOptInd(); +MMPA_FUNC_VISIBILITY VOID mmSetOptInd(INT32 mmOptInd); +MMPA_FUNC_VISIBILITY INT32 mmGetOptOpt(); +MMPA_FUNC_VISIBILITY VOID mmSetOpOpt(INT32 mmOptOpt); +MMPA_FUNC_VISIBILITY CHAR *mmGetOptArg(); +MMPA_FUNC_VISIBILITY VOID mmSetOptArg(CHAR *mmOptArg); +MMPA_FUNC_VISIBILITY INT32 mmGetOpt(INT32 argc, char *const *argv, const char *opts); +MMPA_FUNC_VISIBILITY INT32 mmGetOptLong(INT32 argc, char *const *argv, const char *opts, const mmStructOption *longOpts, + INT32 *longIndex); -extern LONG mmLseek(INT32 fd, INT64 offset, INT32 seekFlag); -extern INT32 mmFtruncate(mmProcess fd, UINT32 length); +MMPA_FUNC_VISIBILITY LONG mmLseek(INT32 fd, INT64 offset, INT32 seekFlag); +MMPA_FUNC_VISIBILITY INT32 mmFtruncate(mmProcess fd, UINT32 length); -extern INT32 mmTlsCreate(mmThreadKey *key, VOID (*destructor)(VOID *)); -extern INT32 mmTlsSet(mmThreadKey key, const VOID *value); -extern VOID *mmTlsGet(mmThreadKey key); -extern INT32 mmTlsDelete(mmThreadKey key); -extern INT32 mmGetOsType(); +MMPA_FUNC_VISIBILITY INT32 mmTlsCreate(mmThreadKey *key, VOID (*destructor)(VOID *)); +MMPA_FUNC_VISIBILITY INT32 mmTlsSet(mmThreadKey key, const VOID *value); +MMPA_FUNC_VISIBILITY VOID *mmTlsGet(mmThreadKey key); +MMPA_FUNC_VISIBILITY INT32 mmTlsDelete(mmThreadKey key); +MMPA_FUNC_VISIBILITY INT32 mmGetOsType(); -extern INT32 mmFsync(mmProcess fd); -extern INT32 mmFsync2(INT32 fd); -extern INT32 mmChdir(const CHAR *path); -extern INT32 mmUmask(INT32 pmode); -extern INT32 mmThreadKill(mmThread id); -extern INT32 mmWaitPid(mmProcess pid, INT32 *status, INT32 options); +MMPA_FUNC_VISIBILITY INT32 mmFsync(mmProcess fd); +MMPA_FUNC_VISIBILITY INT32 mmFsync2(INT32 fd); +MMPA_FUNC_VISIBILITY INT32 mmChdir(const CHAR *path); +MMPA_FUNC_VISIBILITY INT32 mmUmask(INT32 pmode); +MMPA_FUNC_VISIBILITY INT32 mmThreadKill(mmThread id); +MMPA_FUNC_VISIBILITY INT32 mmWaitPid(mmProcess pid, INT32 *status, INT32 options); -extern INT32 mmGetCwd(CHAR *buffer, INT32 maxLen); -extern INT32 mmGetEnv(const CHAR *name, CHAR *value, UINT32 len); -extern INT32 mmSetEnv(const CHAR *name, const CHAR *value, INT32 overwrite); -extern CHAR *mmStrTokR(CHAR *str, const CHAR *delim, CHAR **saveptr); -extern CHAR *mmDirName(CHAR *path); -extern CHAR *mmBaseName(CHAR *path); -extern INT32 mmGetDiskFreeSpace(const char *path, mmDiskSize *diskSize); +MMPA_FUNC_VISIBILITY INT32 mmGetCwd(CHAR *buffer, INT32 maxLen); +MMPA_FUNC_VISIBILITY INT32 mmGetEnv(const CHAR *name, CHAR *value, UINT32 len); +MMPA_FUNC_VISIBILITY INT32 mmSetEnv(const CHAR *name, const CHAR *value, INT32 overwrite); +MMPA_FUNC_VISIBILITY CHAR *mmStrTokR(CHAR *str, const CHAR *delim, CHAR **saveptr); +MMPA_FUNC_VISIBILITY CHAR *mmDirName(CHAR *path); +MMPA_FUNC_VISIBILITY CHAR *mmBaseName(CHAR *path); +MMPA_FUNC_VISIBILITY INT32 mmGetDiskFreeSpace(const char *path, mmDiskSize *diskSize); /* * Function: set the thread name created by mmcreatetask @@ -467,7 +488,7 @@ extern INT32 mmGetDiskFreeSpace(const char *path, mmDiskSize *diskSize); * The input parameter error returns EN_INVALID_PARAM, the execution success returns EN_OK, and the * execution failure returns EN_ERROR */ -extern INT32 mmSetThreadName(mmThread *threadHandle, const CHAR *name); +MMPA_FUNC_VISIBILITY INT32 mmSetThreadName(mmThread *threadHandle, const CHAR *name); /* * Function: get thread name @@ -477,37 +498,38 @@ extern INT32 mmSetThreadName(mmThread *threadHandle, const CHAR *name); * The input parameter error returns EN_INVALID_PARAM, the execution success returns EN_OK, and the * execution failure returns EN_ERROR */ -extern INT32 mmGetThreadName(mmThread *threadHandle, CHAR *name, INT32 size); +MMPA_FUNC_VISIBILITY INT32 mmGetThreadName(mmThread *threadHandle, CHAR *name, INT32 size); /* * Function:Set the thread name of the currently executing thread - call inside the thread body * Input:name:Thread name to be set * The input parameter error returns EN_INVALID_PARAM, the execution success returns EN_OK, and the * execution failure returns EN_ERROR */ -extern INT32 mmSetCurrentThreadName(const CHAR *name); +MMPA_FUNC_VISIBILITY INT32 mmSetCurrentThreadName(const CHAR *name); /* * Function:Get the thread name of the currently executing thread - in body call * Input:name:The name of the thread to get, and the cache is allocated by the user，size>=MMPA_THREADNAME_SIZE * The input parameter error returns EN_INVALID_PARAM, the execution success returns EN_OK, and the * execution failure returns EN_ERROR */ -extern INT32 mmGetCurrentThreadName(CHAR *name, INT32 size); -extern INT32 mmGetFileSize(const CHAR *fileName, ULONGLONG *length); -extern INT32 mmIsDir(const CHAR *fileName); -extern INT32 mmGetOsName(CHAR *name, INT32 nameSize); -extern INT32 mmGetOsVersion(CHAR *versionInfo, INT32 versionLength); -extern INT32 mmGetMac(mmMacInfo **list, INT32 *count); -extern INT32 mmGetMacFree(mmMacInfo *list, INT32 count); -extern INT32 mmGetCpuInfo(mmCpuDesc **cpuInfo, INT32 *count); -extern INT32 mmCpuInfoFree(mmCpuDesc *cpuInfo, INT32 count); -extern INT32 mmCreateProcess(const CHAR *fileName, const mmArgvEnv *env, const char *stdoutRedirectFile, mmProcess *id); - -extern INT32 mmCreateTaskWithThreadAttr(mmThread *threadHandle, const mmUserBlock_t *funcBlock, - const mmThreadAttr *threadAttr); -extern mmFileHandle mmShmOpen(const CHAR *name, INT32 oflag, mmMode_t mode); -extern INT32 mmShmUnlink(const CHAR *name); -extern VOID *mmMmap(mmFd_t fd, mmSize_t size, mmOfft_t offset, mmFd_t *extra, INT32 prot, INT32 flags); -extern INT32 mmMunMap(VOID *data, mmSize_t size, mmFd_t *extra); +MMPA_FUNC_VISIBILITY INT32 mmGetCurrentThreadName(CHAR *name, INT32 size); +MMPA_FUNC_VISIBILITY INT32 mmGetFileSize(const CHAR *fileName, ULONGLONG *length); +MMPA_FUNC_VISIBILITY INT32 mmIsDir(const CHAR *fileName); +MMPA_FUNC_VISIBILITY INT32 mmGetOsName(CHAR *name, INT32 nameSize); +MMPA_FUNC_VISIBILITY INT32 mmGetOsVersion(CHAR *versionInfo, INT32 versionLength); +MMPA_FUNC_VISIBILITY INT32 mmGetMac(mmMacInfo **list, INT32 *count); +MMPA_FUNC_VISIBILITY INT32 mmGetMacFree(mmMacInfo *list, INT32 count); +MMPA_FUNC_VISIBILITY INT32 mmGetCpuInfo(mmCpuDesc **cpuInfo, INT32 *count); +MMPA_FUNC_VISIBILITY INT32 mmCpuInfoFree(mmCpuDesc *cpuInfo, INT32 count); +MMPA_FUNC_VISIBILITY INT32 mmCreateProcess(const CHAR *fileName, const mmArgvEnv *env, const char *stdoutRedirectFile, + mmProcess *id); + +MMPA_FUNC_VISIBILITY INT32 mmCreateTaskWithThreadAttr(mmThread *threadHandle, const mmUserBlock_t *funcBlock, + const mmThreadAttr *threadAttr); +MMPA_FUNC_VISIBILITY mmFileHandle mmShmOpen(const CHAR *name, INT32 oflag, mmMode_t mode); +MMPA_FUNC_VISIBILITY INT32 mmShmUnlink(const CHAR *name); +MMPA_FUNC_VISIBILITY VOID *mmMmap(mmFd_t fd, mmSize_t size, mmOfft_t offset, mmFd_t *extra, INT32 prot, INT32 flags); +MMPA_FUNC_VISIBILITY INT32 mmMunMap(VOID *data, mmSize_t size, mmFd_t *extra); #define MMPA_DLL_API #ifdef __cplusplus diff --git a/third_party/fwkacllib/inc/mmpa/sub_inc/mmpa_win.h b/third_party/fwkacllib/inc/mmpa/sub_inc/mmpa_win.h index 6adc1426..5db6bbf8 100644 --- a/third_party/fwkacllib/inc/mmpa/sub_inc/mmpa_win.h +++ b/third_party/fwkacllib/inc/mmpa/sub_inc/mmpa_win.h @@ -30,11 +30,6 @@ extern "C" { #define MMPA_MACINFO_DEFAULT_SIZE 18 #define MMPA_CPUDESC_DEFAULT_SIZE 64 -MMPA_DLL_API extern char *optarg; -MMPA_DLL_API extern int opterr; -MMPA_DLL_API extern int optind; -MMPA_DLL_API extern int optopt; - #pragma section(".CRT$XCU", long, read) #pragma section(".CRT$XPU", long, read) @@ -65,6 +60,7 @@ typedef struct sockaddr mmSockAddr; typedef int mmSocklen_t; typedef int mmSemTimeout_t; typedef long mmAtomicType; +typedef long long mmAtomicType64; typedef DWORD mmExitCode; typedef DWORD mmErrorMsg; typedef int mmKey_t; @@ -76,6 +72,7 @@ typedef INT32 mmSsize_t; typedef int mmSize; // size typedef size_t mmSize_t; typedef VOID mmshmId_ds; +typedef long long MM_LONG; typedef enum { DT_DIR = FILE_ATTRIBUTE_DIRECTORY, @@ -86,8 +83,15 @@ typedef struct { char d_name[MAX_PATH]; // file name } mmDirent; +typedef struct { + unsigned long d_type; + char d_name[MAX_PATH]; // file name +} mmDirent2; + typedef int (*mmFilter)(const mmDirent *entry); +typedef int (*mmFilter2)(const mmDirent2 *entry); typedef int (*mmSort)(const mmDirent **a, const mmDirent **b); +typedef int (*mmSort2)(const mmDirent2 **a, const mmDirent2 **b); typedef struct { VOID *sendBuf; @@ -162,8 +166,8 @@ typedef struct { } mmTimezone; typedef struct { - LONG tv_sec; - LONG tv_nsec; + MM_LONG tv_sec; + MM_LONG tv_nsec; } mmTimespec; typedef mmTimerHandle mmTimer; @@ -283,12 +287,13 @@ typedef VOID (*mmPf)(VOID); #define DT_UNKNOWN 0 #define DT_FIFO 1 #define DT_CHR 2 -#define DT_DIR 4 #define DT_BLK 6 #define DT_REG 8 #define DT_LNK 10 #define DT_SOCK 12 #define DT_WHT 14 +#define MM_DT_DIR 16 +#define MM_DT_REG 32 #define mmConstructor(x) __declspec(allocate(".CRT$XCU")) mmPf con = x #define mmDestructor(x) __declspec(allocate(".CRT$XPU")) mmPf de = x @@ -327,175 +332,188 @@ typedef VOID (*mmPf)(VOID); #define __attribute__(v) -_declspec(dllexport) INT32 mmCreateTask(mmThread *threadHandle, mmUserBlock_t *funcBlock); -_declspec(dllexport) INT32 mmJoinTask(mmThread *threadHandle); -_declspec(dllexport) INT32 mmMutexInit(mmMutex_t *mutex); -_declspec(dllexport) INT32 mmMutexLock(mmMutex_t *mutex); -_declspec(dllexport) INT32 mmMutexTryLock(mmMutex_t *mutex); -_declspec(dllexport) INT32 mmMutexUnLock(mmMutex_t *mutex); -_declspec(dllexport) INT32 mmMutexDestroy(mmMutex_t *mutex); -_declspec(dllexport) INT32 mmCondInit(mmCond *cond); -_declspec(dllexport) INT32 mmCondLockInit(mmMutexFC *mutex); -_declspec(dllexport) INT32 mmCondLock(mmMutexFC *mutex); -_declspec(dllexport) INT32 mmCondUnLock(mmMutexFC *mutex); -_declspec(dllexport) INT32 mmCondLockDestroy(mmMutexFC *mutex); -_declspec(dllexport) INT32 mmRWLockInit(mmRWLock_t *rwLock); -_declspec(dllexport) INT32 mmRWLockRDLock(mmRWLock_t *rwLock); -_declspec(dllexport) INT32 mmRWLockTryRDLock(mmRWLock_t *rwLock); -_declspec(dllexport) INT32 mmRWLockWRLock(mmRWLock_t *rwLock); -_declspec(dllexport) INT32 mmRWLockTryWRLock(mmRWLock_t *rwLock); -_declspec(dllexport) INT32 mmRDLockUnLock(mmRWLock_t *rwLock); -_declspec(dllexport) INT32 mmWRLockUnLock(mmRWLock_t *rwLock); -_declspec(dllexport) INT32 mmRWLockDestroy(mmRWLock_t *rwLock); -_declspec(dllexport) INT32 mmCondWait(mmCond *cond, mmMutexFC *mutex); -_declspec(dllexport) INT32 mmCondTimedWait(mmCond *cond, mmMutexFC *mutex, UINT32 milliSecond); - -_declspec(dllexport) INT32 mmCondNotify(mmCond *cond); -_declspec(dllexport) INT32 mmCondNotifyAll(mmCond *cond); -_declspec(dllexport) INT32 mmCondDestroy(mmCond *cond); -_declspec(dllexport) INT32 mmGetPid(VOID); -_declspec(dllexport) INT32 mmGetTid(VOID); -_declspec(dllexport) INT32 mmGetPidHandle(mmProcess *processHandle); -_declspec(dllexport) INT32 mmGetLocalTime(mmSystemTime_t *sysTime); -_declspec(dllexport) INT32 mmGetSystemTime(mmSystemTime_t *sysTime); -_declspec(dllexport) INT32 mmSemInit(mmSem_t *sem, UINT32 value); -_declspec(dllexport) INT32 mmSemWait(mmSem_t *sem); -_declspec(dllexport) INT32 mmSemPost(mmSem_t *sem); -_declspec(dllexport) INT32 mmSemDestroy(mmSem_t *sem); -_declspec(dllexport) INT32 mmOpen(const CHAR *pathName, INT32 flags); -_declspec(dllexport) INT32 mmOpen2(const CHAR *pathName, INT32 flags, MODE mode); -_declspec(dllexport) FILE *mmPopen(CHAR *command, CHAR *type); -_declspec(dllexport) INT32 mmClose(INT32 fd); -_declspec(dllexport) INT32 mmPclose(FILE *stream); -_declspec(dllexport) mmSsize_t mmWrite(INT32 fd, VOID *buf, UINT32 bufLen); -_declspec(dllexport) mmSsize_t mmRead(INT32 fd, VOID *buf, UINT32 bufLen); -_declspec(dllexport) mmSockHandle mmSocket(INT32 sockFamily, INT32 type, INT32 protocol); -_declspec(dllexport) INT32 mmBind(mmSockHandle sockFd, mmSockAddr *addr, mmSocklen_t addrLen); -_declspec(dllexport) INT32 mmListen(mmSockHandle sockFd, INT32 backLog); -_declspec(dllexport) mmSockHandle mmAccept(mmSockHandle sockFd, mmSockAddr *addr, mmSocklen_t *addrLen); -_declspec(dllexport) INT32 mmConnect(mmSockHandle sockFd, mmSockAddr *addr, mmSocklen_t addrLen); -_declspec(dllexport) INT32 mmCloseSocket(mmSockHandle sockFd); -_declspec(dllexport) mmSsize_t mmSocketRecv(mmSockHandle sockFd, VOID *recvBuf, INT32 recvLen, INT32 recvFlag); -_declspec(dllexport) mmSsize_t mmSocketSend(mmSockHandle sockFd, VOID *sendBuf, INT32 sendLen, INT32 sendFlag); -_declspec(dllexport) INT32 mmSocketSendTo(mmSockHandle sockFd, +MMPA_FUNC_VISIBILITY INT32 mmCreateTask(mmThread *threadHandle, mmUserBlock_t *funcBlock); +MMPA_FUNC_VISIBILITY INT32 mmJoinTask(mmThread *threadHandle); +MMPA_FUNC_VISIBILITY INT32 mmMutexInit(mmMutex_t *mutex); +MMPA_FUNC_VISIBILITY INT32 mmMutexLock(mmMutex_t *mutex); +MMPA_FUNC_VISIBILITY INT32 mmMutexTryLock(mmMutex_t *mutex); +MMPA_FUNC_VISIBILITY INT32 mmMutexUnLock(mmMutex_t *mutex); +MMPA_FUNC_VISIBILITY INT32 mmMutexDestroy(mmMutex_t *mutex); +MMPA_FUNC_VISIBILITY INT32 mmCondInit(mmCond *cond); +MMPA_FUNC_VISIBILITY INT32 mmCondLockInit(mmMutexFC *mutex); +MMPA_FUNC_VISIBILITY INT32 mmCondLock(mmMutexFC *mutex); +MMPA_FUNC_VISIBILITY INT32 mmCondUnLock(mmMutexFC *mutex); +MMPA_FUNC_VISIBILITY INT32 mmCondLockDestroy(mmMutexFC *mutex); +MMPA_FUNC_VISIBILITY INT32 mmRWLockInit(mmRWLock_t *rwLock); +MMPA_FUNC_VISIBILITY INT32 mmRWLockRDLock(mmRWLock_t *rwLock); +MMPA_FUNC_VISIBILITY INT32 mmRWLockTryRDLock(mmRWLock_t *rwLock); +MMPA_FUNC_VISIBILITY INT32 mmRWLockWRLock(mmRWLock_t *rwLock); +MMPA_FUNC_VISIBILITY INT32 mmRWLockTryWRLock(mmRWLock_t *rwLock); +MMPA_FUNC_VISIBILITY INT32 mmRDLockUnLock(mmRWLock_t *rwLock); +MMPA_FUNC_VISIBILITY INT32 mmWRLockUnLock(mmRWLock_t *rwLock); +MMPA_FUNC_VISIBILITY INT32 mmRWLockDestroy(mmRWLock_t *rwLock); +MMPA_FUNC_VISIBILITY INT32 mmCondWait(mmCond *cond, mmMutexFC *mutex); +MMPA_FUNC_VISIBILITY INT32 mmCondTimedWait(mmCond *cond, mmMutexFC *mutex, UINT32 milliSecond); + +MMPA_FUNC_VISIBILITY INT32 mmCondNotify(mmCond *cond); +MMPA_FUNC_VISIBILITY INT32 mmCondNotifyAll(mmCond *cond); +MMPA_FUNC_VISIBILITY INT32 mmCondDestroy(mmCond *cond); +MMPA_FUNC_VISIBILITY INT32 mmGetPid(VOID); +MMPA_FUNC_VISIBILITY INT32 mmGetTid(VOID); +MMPA_FUNC_VISIBILITY INT32 mmGetPidHandle(mmProcess *processHandle); +MMPA_FUNC_VISIBILITY INT32 mmGetLocalTime(mmSystemTime_t *sysTime); +MMPA_FUNC_VISIBILITY INT32 mmGetSystemTime(mmSystemTime_t *sysTime); +MMPA_FUNC_VISIBILITY INT32 mmSemInit(mmSem_t *sem, UINT32 value); +MMPA_FUNC_VISIBILITY INT32 mmSemWait(mmSem_t *sem); +MMPA_FUNC_VISIBILITY INT32 mmSemPost(mmSem_t *sem); +MMPA_FUNC_VISIBILITY INT32 mmSemDestroy(mmSem_t *sem); +MMPA_FUNC_VISIBILITY INT32 mmOpen(const CHAR *pathName, INT32 flags); +MMPA_FUNC_VISIBILITY INT32 mmOpen2(const CHAR *pathName, INT32 flags, MODE mode); +MMPA_FUNC_VISIBILITY FILE *mmPopen(CHAR *command, CHAR *type); +MMPA_FUNC_VISIBILITY INT32 mmClose(INT32 fd); +MMPA_FUNC_VISIBILITY INT32 mmPclose(FILE *stream); +MMPA_FUNC_VISIBILITY mmSsize_t mmWrite(INT32 fd, VOID *buf, UINT32 bufLen); +MMPA_FUNC_VISIBILITY mmSsize_t mmRead(INT32 fd, VOID *buf, UINT32 bufLen); +MMPA_FUNC_VISIBILITY mmSockHandle mmSocket(INT32 sockFamily, INT32 type, INT32 protocol); +MMPA_FUNC_VISIBILITY INT32 mmBind(mmSockHandle sockFd, mmSockAddr *addr, mmSocklen_t addrLen); +MMPA_FUNC_VISIBILITY INT32 mmListen(mmSockHandle sockFd, INT32 backLog); +MMPA_FUNC_VISIBILITY mmSockHandle mmAccept(mmSockHandle sockFd, mmSockAddr *addr, mmSocklen_t *addrLen); +MMPA_FUNC_VISIBILITY INT32 mmConnect(mmSockHandle sockFd, mmSockAddr *addr, mmSocklen_t addrLen); +MMPA_FUNC_VISIBILITY INT32 mmCloseSocket(mmSockHandle sockFd); +MMPA_FUNC_VISIBILITY mmSsize_t mmSocketRecv(mmSockHandle sockFd, VOID *recvBuf, INT32 recvLen, INT32 recvFlag); +MMPA_FUNC_VISIBILITY mmSsize_t mmSocketSend(mmSockHandle sockFd, VOID *sendBuf, INT32 sendLen, INT32 sendFlag); +MMPA_FUNC_VISIBILITY INT32 mmSocketSendTo(mmSockHandle sockFd, VOID *sendMsg, INT32 sendLen, UINT32 sendFlag, const mmSockAddr* addr, INT32 tolen); -_declspec(dllexport) mmSsize_t mmSocketRecvFrom(mmSockHandle sockFd, +MMPA_FUNC_VISIBILITY mmSsize_t mmSocketRecvFrom(mmSockHandle sockFd, VOID *recvBuf, mmSize recvLen, UINT32 recvFlag, mmSockAddr* addr, mmSocklen_t *FromLen); -_declspec(dllexport) INT32 mmSAStartup(VOID); -_declspec(dllexport) INT32 mmSACleanup(VOID); -_declspec(dllexport) VOID *mmDlopen(const CHAR *fileName, INT mode); -_declspec(dllexport) INT32 mmDladdr(VOID *addr, mmDlInfo *info); -_declspec(dllexport) VOID *mmDlsym(VOID *handle, CHAR *fileName); -_declspec(dllexport) INT32 mmDlclose(VOID *handle); -_declspec(dllexport) CHAR *mmDlerror(VOID); -_declspec(dllexport) INT32 +MMPA_FUNC_VISIBILITY INT32 mmSAStartup(VOID); +MMPA_FUNC_VISIBILITY INT32 mmSACleanup(VOID); +MMPA_FUNC_VISIBILITY VOID *mmDlopen(const CHAR *fileName, INT mode); +MMPA_FUNC_VISIBILITY INT32 mmDladdr(VOID *addr, mmDlInfo *info); +MMPA_FUNC_VISIBILITY VOID *mmDlsym(VOID *handle, const CHAR *fileName); +MMPA_FUNC_VISIBILITY INT32 mmDlclose(VOID *handle); +MMPA_FUNC_VISIBILITY CHAR *mmDlerror(VOID); +MMPA_FUNC_VISIBILITY INT32 mmCreateAndSetTimer(mmTimer *timerHandle, mmUserBlock_t *timerBlock, UINT milliSecond, UINT period); -_declspec(dllexport) INT32 mmDeleteTimer(mmTimer timerHandle); -_declspec(dllexport) INT32 mmStatGet(const CHAR *path, mmStat_t *buffer); -_declspec(dllexport) INT32 mmStat64Get(const CHAR *path, mmStat64_t *buffer); -_declspec(dllexport) INT32 mmFStatGet(INT32 fd, mmStat_t *buffer); -_declspec(dllexport) INT32 mmMkdir(const CHAR *pathName, mmMode_t mode); -_declspec(dllexport) INT32 mmSleep(UINT32 milliSecond); -_declspec(dllexport) INT32 mmCreateTaskWithAttr(mmThread *threadHandle, mmUserBlock_t *funcBlock); -_declspec(dllexport) INT32 mmGetProcessPrio(mmProcess pid); -_declspec(dllexport) INT32 mmSetProcessPrio(mmProcess pid, INT32 processPrio); -_declspec(dllexport) INT32 mmGetThreadPrio(mmThread *threadHandle); -_declspec(dllexport) INT32 mmSetThreadPrio(mmThread *threadHandle, INT32 threadPrio); -_declspec(dllexport) INT32 mmAccess(const CHAR *pathName); -_declspec(dllexport) INT32 mmAccess2(const CHAR *pathName, INT32 mode); -_declspec(dllexport) INT32 mmRmdir(const CHAR *pathName); - -_declspec(dllexport) INT32 mmIoctl(mmProcess fd, INT32 ioctlCode, mmIoctlBuf *bufPtr); -_declspec(dllexport) INT32 mmSemTimedWait(mmSem_t *sem, INT32 timeout); -_declspec(dllexport) mmSsize_t mmWritev(mmSockHandle fd, mmIovSegment *iov, INT32 iovcnt); -_declspec(dllexport) VOID mmMb(); -_declspec(dllexport) INT32 mmInetAton(const CHAR *addrStr, mmInAddr *addr); - -_declspec(dllexport) mmProcess mmOpenFile(const CHAR *fileName, UINT32 access, mmCreateFlag fileFlag); -_declspec(dllexport) mmSsize_t mmReadFile(mmProcess fileId, VOID *buffer, INT32 len); -_declspec(dllexport) mmSsize_t mmWriteFile(mmProcess fileId, VOID *buffer, INT32 len); -_declspec(dllexport) INT32 mmCloseFile(mmProcess fileId); - -_declspec(dllexport) mmAtomicType mmSetData(mmAtomicType *ptr, mmAtomicType value); -_declspec(dllexport) mmAtomicType mmValueInc(mmAtomicType *ptr, mmAtomicType value); -_declspec(dllexport) mmAtomicType mmValueSub(mmAtomicType *ptr, mmAtomicType value); -_declspec(dllexport) INT32 mmCreateTaskWithDetach(mmThread *threadHandle, mmUserBlock_t *funcBlock); - -_declspec(dllexport) INT32 mmCreateNamedPipe(mmPipeHandle pipe[], CHAR *pipeName[], INT32 waitMode); -_declspec(dllexport) INT32 mmOpenNamePipe(mmPipeHandle pipe[], CHAR *pipeName[], INT32 waitMode); -_declspec(dllexport) VOID mmCloseNamedPipe(mmPipeHandle namedPipe[]); - -_declspec(dllexport) INT32 mmCreatePipe(mmPipeHandle pipe[], CHAR *pipeName[], UINT32 pipeCount, INT32 waitMode); -_declspec(dllexport) INT32 mmOpenPipe(mmPipeHandle pipe[], CHAR *pipeName[], UINT32 pipeCount, INT32 waitMode); -_declspec(dllexport) VOID mmClosePipe(mmPipeHandle pipe[], UINT32 pipeCount); - -_declspec(dllexport) mmCompletionHandle mmCreateCompletionPort(); -_declspec(dllexport) VOID mmCloseCompletionPort(mmCompletionHandle handle); -_declspec(dllexport) INT32 mmPoll(mmPollfd *fds, INT32 fdCount, INT32 timeout, mmCompletionHandle handleIOCP, +MMPA_FUNC_VISIBILITY INT32 mmDeleteTimer(mmTimer timerHandle); +MMPA_FUNC_VISIBILITY INT32 mmStatGet(const CHAR *path, mmStat_t *buffer); +MMPA_FUNC_VISIBILITY INT32 mmStat64Get(const CHAR *path, mmStat64_t *buffer); +MMPA_FUNC_VISIBILITY INT32 mmFStatGet(INT32 fd, mmStat_t *buffer); +MMPA_FUNC_VISIBILITY INT32 mmMkdir(const CHAR *pathName, mmMode_t mode); +MMPA_FUNC_VISIBILITY INT32 mmSleep(UINT32 milliSecond); +MMPA_FUNC_VISIBILITY INT32 mmCreateTaskWithAttr(mmThread *threadHandle, mmUserBlock_t *funcBlock); +MMPA_FUNC_VISIBILITY INT32 mmGetProcessPrio(mmProcess pid); +MMPA_FUNC_VISIBILITY INT32 mmSetProcessPrio(mmProcess pid, INT32 processPrio); +MMPA_FUNC_VISIBILITY INT32 mmGetThreadPrio(mmThread *threadHandle); +MMPA_FUNC_VISIBILITY INT32 mmSetThreadPrio(mmThread *threadHandle, INT32 threadPrio); +MMPA_FUNC_VISIBILITY INT32 mmAccess(const CHAR *pathName); +MMPA_FUNC_VISIBILITY INT32 mmAccess2(const CHAR *pathName, INT32 mode); +MMPA_FUNC_VISIBILITY INT32 mmRmdir(const CHAR *pathName); + +MMPA_FUNC_VISIBILITY INT32 mmIoctl(mmProcess fd, INT32 ioctlCode, mmIoctlBuf *bufPtr); +MMPA_FUNC_VISIBILITY INT32 mmSemTimedWait(mmSem_t *sem, INT32 timeout); +MMPA_FUNC_VISIBILITY mmSsize_t mmWritev(mmSockHandle fd, mmIovSegment *iov, INT32 iovcnt); +MMPA_FUNC_VISIBILITY VOID mmMb(); +MMPA_FUNC_VISIBILITY INT32 mmInetAton(const CHAR *addrStr, mmInAddr *addr); + +MMPA_FUNC_VISIBILITY mmProcess mmOpenFile(const CHAR *fileName, UINT32 access, mmCreateFlag fileFlag); +MMPA_FUNC_VISIBILITY mmSsize_t mmReadFile(mmProcess fileId, VOID *buffer, INT32 len); +MMPA_FUNC_VISIBILITY mmSsize_t mmWriteFile(mmProcess fileId, VOID *buffer, INT32 len); +MMPA_FUNC_VISIBILITY INT32 mmCloseFile(mmProcess fileId); + +MMPA_FUNC_VISIBILITY mmAtomicType mmSetData(mmAtomicType *ptr, mmAtomicType value); +MMPA_FUNC_VISIBILITY mmAtomicType mmValueInc(mmAtomicType *ptr, mmAtomicType value); +MMPA_FUNC_VISIBILITY mmAtomicType mmValueSub(mmAtomicType *ptr, mmAtomicType value); +MMPA_FUNC_VISIBILITY mmAtomicType64 mmSetData64(mmAtomicType64 *ptr, mmAtomicType64 value); +MMPA_FUNC_VISIBILITY mmAtomicType64 mmValueInc64(mmAtomicType64 *ptr, mmAtomicType64 value); +MMPA_FUNC_VISIBILITY mmAtomicType64 mmValueSub64(mmAtomicType64 *ptr, mmAtomicType64 value); +MMPA_FUNC_VISIBILITY INT32 mmCreateTaskWithDetach(mmThread *threadHandle, mmUserBlock_t *funcBlock); + +MMPA_FUNC_VISIBILITY INT32 mmCreateNamedPipe(mmPipeHandle pipe[], CHAR *pipeName[], INT32 waitMode); +MMPA_FUNC_VISIBILITY INT32 mmOpenNamePipe(mmPipeHandle pipe[], CHAR *pipeName[], INT32 waitMode); +MMPA_FUNC_VISIBILITY VOID mmCloseNamedPipe(mmPipeHandle namedPipe[]); + +MMPA_FUNC_VISIBILITY INT32 mmCreatePipe(mmPipeHandle pipe[], CHAR *pipeName[], UINT32 pipeCount, INT32 waitMode); +MMPA_FUNC_VISIBILITY INT32 mmOpenPipe(mmPipeHandle pipe[], CHAR *pipeName[], UINT32 pipeCount, INT32 waitMode); +MMPA_FUNC_VISIBILITY VOID mmClosePipe(mmPipeHandle pipe[], UINT32 pipeCount); + +MMPA_FUNC_VISIBILITY mmCompletionHandle mmCreateCompletionPort(); +MMPA_FUNC_VISIBILITY VOID mmCloseCompletionPort(mmCompletionHandle handle); +MMPA_FUNC_VISIBILITY INT32 mmPoll(mmPollfd *fds, INT32 fdCount, INT32 timeout, mmCompletionHandle handleIOCP, pmmPollData polledData, mmPollBack pollBack); -_declspec(dllexport) INT32 mmGetErrorCode(); -_declspec(dllexport) CHAR *mmGetErrorFormatMessage(mmErrorMsg errnum, CHAR *buf, mmSize size); -_declspec(dllexport) INT32 mmGetTimeOfDay(mmTimeval *timeVal, mmTimezone *timeZone); -_declspec(dllexport) mmTimespec mmGetTickCount(); -_declspec(dllexport) INT32 mmGetRealPath(CHAR *path, CHAR *realPath); - -_declspec(dllexport) INT32 mmRealPath(const CHAR *path, CHAR *realPath, INT32 realPathLen); - -_declspec(dllexport) INT32 mmDup2(INT32 oldFd, INT32 newFd); -_declspec(dllexport) INT32 mmDup(INT32 fd); -_declspec(dllexport) INT32 mmUnlink(const CHAR *filename); -_declspec(dllexport) INT32 mmChmod(const CHAR *filename, INT32 mode); -_declspec(dllexport) INT32 mmFileno(FILE *stream); -_declspec(dllexport) INT32 mmScandir(const CHAR *path, mmDirent ***entryList, mmFilter filterFunc, mmSort sort); -_declspec(dllexport) VOID mmScandirFree(mmDirent **entryList, INT32 count); - -_declspec(dllexport) mmMsgid mmMsgCreate(mmKey_t key, INT32 msgFlag); -_declspec(dllexport) mmMsgid mmMsgOpen(mmKey_t key, INT32 msgFlag); -_declspec(dllexport) INT32 mmMsgRcv(mmMsgid msqid, VOID *buf, INT32 bufLen, INT32 msgFlag); -_declspec(dllexport) INT32 mmMsgSnd(mmMsgid msqid, VOID *buf, INT32 bufLen, INT32 msgFlag); - -_declspec(dllexport) INT32 mmMsgClose(mmMsgid msqid); - -_declspec(dllexport) INT32 mmLocalTimeR(const time_t *timep, struct tm *result); -_declspec(dllexport) INT32 mmGetOpt(INT32 argc, char *const *argv, const char *opts); -_declspec(dllexport) INT32 +MMPA_FUNC_VISIBILITY INT32 mmGetErrorCode(); +MMPA_FUNC_VISIBILITY CHAR *mmGetErrorFormatMessage(mmErrorMsg errnum, CHAR *buf, mmSize size); +MMPA_FUNC_VISIBILITY INT32 mmGetTimeOfDay(mmTimeval *timeVal, mmTimezone *timeZone); +MMPA_FUNC_VISIBILITY mmTimespec mmGetTickCount(); +MMPA_FUNC_VISIBILITY INT32 mmGetRealPath(CHAR *path, CHAR *realPath); + +MMPA_FUNC_VISIBILITY INT32 mmRealPath(const CHAR *path, CHAR *realPath, INT32 realPathLen); + +MMPA_FUNC_VISIBILITY INT32 mmDup2(INT32 oldFd, INT32 newFd); +MMPA_FUNC_VISIBILITY INT32 mmDup(INT32 fd); +MMPA_FUNC_VISIBILITY INT32 mmUnlink(const CHAR *filename); +MMPA_FUNC_VISIBILITY INT32 mmChmod(const CHAR *filename, INT32 mode); +MMPA_FUNC_VISIBILITY INT32 mmFileno(FILE *stream); +MMPA_FUNC_VISIBILITY INT32 mmScandir(const CHAR *path, mmDirent ***entryList, mmFilter filterFunc, mmSort sort); +MMPA_FUNC_VISIBILITY INT32 mmScandir2(const CHAR *path, mmDirent2 ***entryList, mmFilter2 filterFunc, mmSort2 sort); +MMPA_FUNC_VISIBILITY VOID mmScandirFree(mmDirent **entryList, INT32 count); +MMPA_FUNC_VISIBILITY VOID mmScandirFree2(mmDirent2 **entryList, INT32 count); + +MMPA_FUNC_VISIBILITY mmMsgid mmMsgCreate(mmKey_t key, INT32 msgFlag); +MMPA_FUNC_VISIBILITY mmMsgid mmMsgOpen(mmKey_t key, INT32 msgFlag); +MMPA_FUNC_VISIBILITY INT32 mmMsgRcv(mmMsgid msqid, VOID *buf, INT32 bufLen, INT32 msgFlag); +MMPA_FUNC_VISIBILITY INT32 mmMsgSnd(mmMsgid msqid, VOID *buf, INT32 bufLen, INT32 msgFlag); + +MMPA_FUNC_VISIBILITY INT32 mmMsgClose(mmMsgid msqid); + +MMPA_FUNC_VISIBILITY INT32 mmLocalTimeR(const time_t *timep, struct tm *result); +MMPA_FUNC_VISIBILITY INT32 mmGetOptErr(); +MMPA_FUNC_VISIBILITY VOID mmSetOptErr(INT32 mmOptErr); +MMPA_FUNC_VISIBILITY INT32 mmGetOptInd(); +MMPA_FUNC_VISIBILITY VOID mmSetOptInd(INT32 mmOptInd); +MMPA_FUNC_VISIBILITY INT32 mmGetOptOpt(); +MMPA_FUNC_VISIBILITY VOID mmSetOpOpt(INT32 mmOptOpt); +MMPA_FUNC_VISIBILITY CHAR *mmGetOptArg(); +MMPA_FUNC_VISIBILITY VOID mmSetOptArg(CHAR *mmOptArg); +MMPA_FUNC_VISIBILITY INT32 mmGetOpt(INT32 argc, char *const *argv, const char *opts); +MMPA_FUNC_VISIBILITY INT32 mmGetOptLong(INT32 argc, CHAR *const *argv, const CHAR *opts, const mmStructOption *longopts, INT32 *longindex); -_declspec(dllexport) LONG mmLseek(INT32 fd, INT64 offset, INT32 seekFlag); -_declspec(dllexport) INT32 mmFtruncate(mmProcess fd, UINT32 length); +MMPA_FUNC_VISIBILITY LONG mmLseek(INT32 fd, INT64 offset, INT32 seekFlag); +MMPA_FUNC_VISIBILITY INT32 mmFtruncate(mmProcess fd, UINT32 length); -_declspec(dllexport) INT32 mmTlsCreate(mmThreadKey *key, VOID (*destructor)(VOID *)); -_declspec(dllexport) INT32 mmTlsSet(mmThreadKey key, const VOID *value); -_declspec(dllexport) VOID *mmTlsGet(mmThreadKey key); -_declspec(dllexport) INT32 mmTlsDelete(mmThreadKey key); -_declspec(dllexport) INT32 mmGetOsType(); +MMPA_FUNC_VISIBILITY INT32 mmTlsCreate(mmThreadKey *key, VOID (*destructor)(VOID *)); +MMPA_FUNC_VISIBILITY INT32 mmTlsSet(mmThreadKey key, const VOID *value); +MMPA_FUNC_VISIBILITY VOID *mmTlsGet(mmThreadKey key); +MMPA_FUNC_VISIBILITY INT32 mmTlsDelete(mmThreadKey key); +MMPA_FUNC_VISIBILITY INT32 mmGetOsType(); -_declspec(dllexport) INT32 mmFsync(mmProcess fd); -_declspec(dllexport) INT32 mmFsync2(INT32 fd); -_declspec(dllexport) INT32 mmChdir(const CHAR *path); -_declspec(dllexport) INT32 mmUmask(INT32 pmode); -_declspec(dllexport) INT32 mmWaitPid(mmProcess pid, INT32 *status, INT32 options); +MMPA_FUNC_VISIBILITY INT32 mmFsync(mmProcess fd); +MMPA_FUNC_VISIBILITY INT32 mmFsync2(INT32 fd); +MMPA_FUNC_VISIBILITY INT32 mmChdir(const CHAR *path); +MMPA_FUNC_VISIBILITY INT32 mmUmask(INT32 pmode); +MMPA_FUNC_VISIBILITY INT32 mmWaitPid(mmProcess pid, INT32 *status, INT32 options); -_declspec(dllexport) INT32 mmGetCwd(CHAR *buffer, INT32 maxLen); -_declspec(dllexport) CHAR *mmStrTokR(CHAR *str, const CHAR *delim, CHAR **saveptr); +MMPA_FUNC_VISIBILITY INT32 mmGetCwd(CHAR *buffer, INT32 maxLen); +MMPA_FUNC_VISIBILITY CHAR *mmStrTokR(CHAR *str, const CHAR *delim, CHAR **saveptr); -_declspec(dllexport) INT32 mmGetEnv(const CHAR *name, CHAR *value, UINT32 len); -_declspec(dllexport) INT32 mmSetEnv(const CHAR *name, const CHAR *value, INT32 overwrite); -_declspec(dllexport) CHAR *mmDirName(CHAR *path); -_declspec(dllexport) CHAR *mmBaseName(CHAR *path); -_declspec(dllexport) INT32 mmGetDiskFreeSpace(const char *path, mmDiskSize *diskSize); +MMPA_FUNC_VISIBILITY INT32 mmGetEnv(const CHAR *name, CHAR *value, UINT32 len); +MMPA_FUNC_VISIBILITY INT32 mmSetEnv(const CHAR *name, const CHAR *value, INT32 overwrite); +MMPA_FUNC_VISIBILITY CHAR *mmDirName(CHAR *path); +MMPA_FUNC_VISIBILITY CHAR *mmBaseName(CHAR *path); +MMPA_FUNC_VISIBILITY INT32 mmGetDiskFreeSpace(const char *path, mmDiskSize *diskSize); -_declspec(dllexport) INT32 mmSetThreadName(mmThread *threadHandle, const CHAR *name); -_declspec(dllexport) INT32 mmGetThreadName(mmThread *threadHandle, CHAR *name, INT32 size); +MMPA_FUNC_VISIBILITY INT32 mmSetThreadName(mmThread *threadHandle, const CHAR *name); +MMPA_FUNC_VISIBILITY INT32 mmGetThreadName(mmThread *threadHandle, CHAR *name, INT32 size); /* * Function: set the thread name of the currently executing thread - internal call of thread, which is not supported @@ -504,7 +522,7 @@ _declspec(dllexport) INT32 mmGetThreadName(mmThread *threadHandle, CHAR *name, I * The input parameter error returns EN_INVALID_PARAM, the execution success returns EN_OK, and the * execution failure returns EN_ERROR */ -_declspec(dllexport) INT32 mmSetCurrentThreadName(const CHAR *name); +MMPA_FUNC_VISIBILITY INT32 mmSetCurrentThreadName(const CHAR *name); /* * Function: Get the thread name of the currently executing thread - thread body call, not supported under windows, null @@ -513,25 +531,25 @@ _declspec(dllexport) INT32 mmSetCurrentThreadName(const CHAR *name); * The input parameter error returns EN_INVALID_PARAM, the execution success returns * EN_OK, and the execution failure returns EN_ERROR */ -_declspec(dllexport) INT32 mmGetCurrentThreadName(CHAR *name, INT32 size); - -_declspec(dllexport) INT32 mmGetFileSize(const CHAR *fileName, ULONGLONG *length); -_declspec(dllexport) INT32 mmIsDir(const CHAR *fileName); -_declspec(dllexport) INT32 mmGetOsName(CHAR *name, INT32 nameSize); -_declspec(dllexport) INT32 mmGetOsVersion(CHAR *versionInfo, INT32 versionLength); -_declspec(dllexport) INT32 mmGetMac(mmMacInfo **list, INT32 *count); -_declspec(dllexport) INT32 mmGetMacFree(mmMacInfo *list, INT32 count); -_declspec(dllexport) INT32 mmGetCpuInfo(mmCpuDesc **cpuInfo, INT32 *count); -_declspec(dllexport) INT32 mmCpuInfoFree(mmCpuDesc *cpuInfo, INT32 count); -_declspec(dllexport) INT32 +MMPA_FUNC_VISIBILITY INT32 mmGetCurrentThreadName(CHAR *name, INT32 size); + +MMPA_FUNC_VISIBILITY INT32 mmGetFileSize(const CHAR *fileName, ULONGLONG *length); +MMPA_FUNC_VISIBILITY INT32 mmIsDir(const CHAR *fileName); +MMPA_FUNC_VISIBILITY INT32 mmGetOsName(CHAR *name, INT32 nameSize); +MMPA_FUNC_VISIBILITY INT32 mmGetOsVersion(CHAR *versionInfo, INT32 versionLength); +MMPA_FUNC_VISIBILITY INT32 mmGetMac(mmMacInfo **list, INT32 *count); +MMPA_FUNC_VISIBILITY INT32 mmGetMacFree(mmMacInfo *list, INT32 count); +MMPA_FUNC_VISIBILITY INT32 mmGetCpuInfo(mmCpuDesc **cpuInfo, INT32 *count); +MMPA_FUNC_VISIBILITY INT32 mmCpuInfoFree(mmCpuDesc *cpuInfo, INT32 count); +MMPA_FUNC_VISIBILITY INT32 mmCreateProcess(const CHAR *fileName, const mmArgvEnv *env, const char *stdoutRedirectFile, mmProcess *id); -_declspec(dllexport) INT32 +MMPA_FUNC_VISIBILITY INT32 mmCreateTaskWithThreadAttr(mmThread *threadHandle, const mmUserBlock_t *funcBlock, const mmThreadAttr *threadAttr); -_declspec(dllexport) mmFileHandle mmShmOpen(const CHAR *name, INT32 oflag, mmMode_t mode); -_declspec(dllexport) INT32 mmShmUnlink(const CHAR *name); -_declspec(dllexport) VOID *mmMmap(mmFd_t fd, mmSize_t size, mmOfft_t offset, mmFd_t *extra, INT32 prot, INT32 flags); -_declspec(dllexport) INT32 mmMunMap(VOID *data, mmSize_t size, mmFd_t *extra); +MMPA_FUNC_VISIBILITY mmFileHandle mmShmOpen(const CHAR *name, INT32 oflag, mmMode_t mode); +MMPA_FUNC_VISIBILITY INT32 mmShmUnlink(const CHAR *name); +MMPA_FUNC_VISIBILITY VOID *mmMmap(mmFd_t fd, mmSize_t size, mmOfft_t offset, mmFd_t *extra, INT32 prot, INT32 flags); +MMPA_FUNC_VISIBILITY INT32 mmMunMap(VOID *data, mmSize_t size, mmFd_t *extra); #ifdef __cplusplus #if __cplusplus } diff --git a/third_party/fwkacllib/inc/register/host_cpu_context.h b/third_party/fwkacllib/inc/register/host_cpu_context.h deleted file mode 100644 index 4d6d5855..00000000 --- a/third_party/fwkacllib/inc/register/host_cpu_context.h +++ /dev/null @@ -1,39 +0,0 @@ -/** - * Copyright 2019-2020 Huawei Technologies Co., Ltd - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * 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. - */ - -#ifndef INC_REGISTER_HOST_CPU_CONTEXT_H_ -#define INC_REGISTER_HOST_CPU_CONTEXT_H_ - -#include "external/ge/ge_api_error_codes.h" -#include "register/register_types.h" - -namespace ge { -class HostCpuContext { - public: - HostCpuContext() = default; - ~HostCpuContext() = default; - private: - class Impl; - Impl *impl_; -}; -} // namespace ge - -extern "C" { -// Unified definition for registering host_cpu_kernel_wrapper when so is opened -FMK_FUNC_HOST_VISIBILITY ge::Status Initialize(const ge::HostCpuContext &ctx); -} - -#endif //INC_REGISTER_HOST_CPU_CONTEXT_H_ diff --git a/third_party/fwkacllib/inc/register/op_kernel_registry.h b/third_party/fwkacllib/inc/register/op_kernel_registry.h deleted file mode 100644 index 5fed8960..00000000 --- a/third_party/fwkacllib/inc/register/op_kernel_registry.h +++ /dev/null @@ -1,49 +0,0 @@ -/** - * Copyright 2019-2020 Huawei Technologies Co., Ltd - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * 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. - */ - -#ifndef INC_REGISTER_OP_KERNEL_REGISTRY_H_ -#define INC_REGISTER_OP_KERNEL_REGISTRY_H_ -#include -#include -#include "register/register_types.h" -#include "register.h" - -namespace ge { -class FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY OpKernelRegistry { - public: - using CreateFn = HostCpuOp* (*)(); - ~OpKernelRegistry(); - - static OpKernelRegistry& GetInstance() { - static OpKernelRegistry instance; - return instance; - } - - bool IsRegistered(const std::string &op_type); - - void RegisterHostCpuOp(const std::string &op_type, CreateFn create_fn); - - std::unique_ptr CreateHostCpuOp(const std::string &op_type); - - private: - OpKernelRegistry(); - class OpKernelRegistryImpl; - /*lint -e148*/ - std::unique_ptr impl_; -}; -} // namespace ge - -#endif // INC_REGISTER_OP_KERNEL_REGISTRY_H_ diff --git a/third_party/fwkacllib/inc/register/op_registry.h b/third_party/fwkacllib/inc/register/op_registry.h deleted file mode 100644 index 3feea0df..00000000 --- a/third_party/fwkacllib/inc/register/op_registry.h +++ /dev/null @@ -1,90 +0,0 @@ -/** - * Copyright 2019-2020 Huawei Technologies Co., Ltd - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * 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. - */ - -#ifndef INC_REGISTER_OP_REGISTRY_H_ -#define INC_REGISTER_OP_REGISTRY_H_ - -#include -#include -#include -#include -#include - -#include "register/register.h" - -namespace domi { -enum RemoveInputType { - OMG_MOVE_TYPE_DTYPE = 0, - OMG_MOVE_TYPE_VALUE, - OMG_MOVE_TYPE_SHAPE, - OMG_MOVE_TYPE_FORMAT, - OMG_MOVE_TYPE_AXIS, - OMG_MOVE_TYPE_SCALAR_VALUE, - OMG_REMOVE_TYPE_WITH_COND = 1000, - OMG_REMOVE_INPUT_WITH_ORIGINAL_TYPE, - OMG_INPUT_REORDER, -}; - -struct RemoveInputConfigure { - int inputIdx = INT_MAX; - std::string attrName; - RemoveInputType moveType; - bool attrValue = false; - std::string originalType; - std::vector input_order; -}; - -class FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY OpRegistry { - public: - static OpRegistry *Instance(); - - std::vector registrationDatas; - - bool Register(const OpRegistrationData ®_data); - - domi::ImplyType GetImplyType(const std::string &op_type); - - void GetOpTypeByImplyType(std::vector &vec_op_type, const domi::ImplyType &imply_type); - - domi::ParseParamFunc GetParseParamFunc(const std::string &op_type, const std::string &ori_type); - - domi::ParseParamByOpFunc GetParseParamByOperatorFunc(const std::string &ori_type); - - domi::FusionParseParamFunc GetFusionParseParamFunc(const std::string &op_type, const std::string &ori_type); - - domi::FusionParseParamByOpFunc GetFusionParseParamByOpFunc(const std::string &op_type, - const std::string &ori_type); - - domi::ParseSubgraphFunc GetParseSubgraphPostFunc(const std::string &op_type); - - domi::ImplyType GetImplyTypeByOriOpType(const std::string &ori_optype); - - const std::vector &GetRemoveInputConfigure(const std::string &ori_optype) const; - - bool GetOmTypeByOriOpType(const std::string &ori_optype, std::string &om_type); - - private: - std::unordered_map op_run_mode_map_; - std::unordered_map op_parse_params_fn_map_; - std::unordered_map parse_params_by_op_func_map_; - std::unordered_map fusion_op_parse_params_fn_map_; - std::unordered_map fusion_parse_params_by_op_fn_map_; - std::unordered_map op_types_to_parse_subgraph_post_func_; - std::unordered_map> remove_input_configure_map_; - std::unordered_map origin_type_to_om_type_; -}; -} // namespace domi -#endif // INC_REGISTER_OP_REGISTRY_H_ diff --git a/third_party/fwkacllib/inc/register/op_tiling.h b/third_party/fwkacllib/inc/register/op_tiling.h deleted file mode 100644 index bcd4cd5e..00000000 --- a/third_party/fwkacllib/inc/register/op_tiling.h +++ /dev/null @@ -1,135 +0,0 @@ -/** - * Copyright 2019-2020 Huawei Technologies Co., Ltd - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * 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. - */ - -#ifndef INC_OP_TILING_H_ -#define INC_OP_TILING_H_ - -#include "external/register/register_types.h" -#include "external/graph/tensor.h" -#include "graph/debug/ge_attr_define.h" -#include "graph/node.h" - -#include -#include -#include -#include -#include -#include -#include -#include "graph/node.h" - -#define REGISTER_OP_TILING_FUNC(optype, opfunc) \ - REGISTER_OP_TILING_FUNC_UNIQ_HELPER(optype, opfunc, __COUNTER__) - -#define REGISTER_OP_TILING_FUNC_UNIQ_HELPER(optype, opfunc, counter) \ - REGISTER_OP_TILING_FUNC_UNIQ(optype, opfunc, counter) - -#define REGISTER_OP_TILING_FUNC_UNIQ(optype, opfunc, counter) \ - static OpTilingInterf g_##optype##TilingInterf##counter(#optype, opfunc) - -namespace optiling { - -enum TensorArgType { - TA_NONE, - TA_SINGLE, - TA_LIST, -}; - - -using ByteBuffer = std::stringstream; - -struct TeOpTensor { - std::vector shape; - std::vector ori_shape; - std::string format; - std::string ori_format; - std::string dtype; - std::map attrs; -}; - - -struct TeOpTensorArg { - TensorArgType arg_type; - std::vector tensor; -}; - -struct OpRunInfo { - uint32_t block_dim; - std::vector workspaces; - ByteBuffer tiling_data; -}; - - -using TeOpAttrArgs = std::vector; -using TeConstTensorData = std::tuple; - -struct TeOpParas { - std::vector inputs; - std::vector outputs; - std::map const_inputs; - TeOpAttrArgs attrs; -}; - - -using OpTilingFunc = std::function; - -using OpTilingFuncPtr = bool(*)(const std::string&, const TeOpParas&, const nlohmann::json& , OpRunInfo&); - -class FMK_FUNC_HOST_VISIBILITY OpTilingInterf -{ -public: - OpTilingInterf(std::string op_type, OpTilingFunc func); - ~OpTilingInterf() = default; - static std::map &RegisteredOpInterf(); - static std::string OpTilingUuid; -}; - - -template -ByteBuffer& ByteBufferPut(ByteBuffer &buf, const T &value) -{ - buf.write(reinterpret_cast(&value), sizeof(value)); - buf.flush(); - return buf; -} - -template -ByteBuffer& ByteBufferGet(ByteBuffer &buf, T &value) -{ - buf.read(reinterpret_cast(&value), sizeof(value)); - return buf; -} - -inline size_t ByteBufferGetAll(ByteBuffer &buf, char *dest, size_t dest_len) -{ - size_t nread = 0; - size_t rn = 0; - do { - rn = buf.readsome(dest + nread, dest_len - nread); - nread += rn; - } while (rn > 0 && dest_len > nread); - - return nread; -} - - -extern "C" ge::graphStatus OpParaCalculate(const ge::Node &node, OpRunInfo &run_info); -extern "C" ge::graphStatus OpAtomicCalculate(const ge::Node &node, OpRunInfo &run_info); - -} - -#endif // INC_OP_TILING_H_ diff --git a/third_party/fwkacllib/inc/register/ops_kernel_builder_registry.h b/third_party/fwkacllib/inc/register/ops_kernel_builder_registry.h deleted file mode 100644 index 96ac931b..00000000 --- a/third_party/fwkacllib/inc/register/ops_kernel_builder_registry.h +++ /dev/null @@ -1,62 +0,0 @@ -/** - * Copyright 2019-2020 Huawei Technologies Co., Ltd - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * 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. - */ - -#ifndef INC_REGISTER_OPS_KERNEL_BUILDER_REGISTRY_H_ -#define INC_REGISTER_OPS_KERNEL_BUILDER_REGISTRY_H_ - -#include -#include "register/register_types.h" -#include "common/opskernel/ops_kernel_builder.h" - -namespace ge { -using OpsKernelBuilderPtr = std::shared_ptr; - -class FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY OpsKernelBuilderRegistry { - public: - static OpsKernelBuilderRegistry &GetInstance(); - - void Register(const std::string &lib_name, const OpsKernelBuilderPtr &instance); - - void UnregisterAll(); - - const std::map &GetAll() const; - - private: - std::map kernel_builders_; -}; - -class FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY OpsKernelBuilderRegistrar { - public: - using CreateFn = OpsKernelBuilder *(*)(); - OpsKernelBuilderRegistrar(const std::string &kernel_lib_name, CreateFn fn); - ~OpsKernelBuilderRegistrar() = default; -}; - -#define REGISTER_OPS_KERNEL_BUILDER(kernel_lib_name, builder) \ - REGISTER_OPS_KERNEL_BUILDER_UNIQ_HELPER(__COUNTER__, kernel_lib_name, builder) - -#define REGISTER_OPS_KERNEL_BUILDER_UNIQ_HELPER(ctr, kernel_lib_name, builder) \ - REGISTER_OPS_KERNEL_BUILDER_UNIQ(ctr, kernel_lib_name, builder) - -#define REGISTER_OPS_KERNEL_BUILDER_UNIQ(ctr, kernel_lib_name, builder) \ - static ::ge::OpsKernelBuilderRegistrar register_op_kernel_builder_##ctr \ - __attribute__((unused)) = \ - ::ge::OpsKernelBuilderRegistrar(kernel_lib_name, []()->::ge::OpsKernelBuilder* { \ - return new (std::nothrow) builder(); \ - }) -} // namespace ge - -#endif // INC_REGISTER_OPS_KERNEL_BUILDER_REGISTRY_H_ diff --git a/third_party/fwkacllib/inc/register/register.h b/third_party/fwkacllib/inc/register/register.h deleted file mode 100644 index d98edaa4..00000000 --- a/third_party/fwkacllib/inc/register/register.h +++ /dev/null @@ -1,54 +0,0 @@ -/** - * Copyright 2019-2020 Huawei Technologies Co., Ltd - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * 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. - */ - -#ifndef INC_REGISTER_REGISTRY_H_ -#define INC_REGISTER_REGISTRY_H_ - -#include "external/register/register.h" -#include "external/ge/ge_api_error_codes.h" - -namespace ge { -class FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY HostCpuOp { - public: - HostCpuOp() = default; - virtual ~HostCpuOp() = default; - - virtual graphStatus Compute(Operator &op, - const std::map &inputs, - std::map &outputs) = 0; -}; - -class FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY HostCpuOpRegistrar { - public: - HostCpuOpRegistrar(const char *op_type, HostCpuOp *(*create_fn)()); - ~HostCpuOpRegistrar() = default; -}; - -#define REGISTER_HOST_CPU_OP_BUILDER(name, op) \ - REGISTER_HOST_CPU_OP_BUILDER_UNIQ_HELPER(__COUNTER__, name, op) - -#define REGISTER_HOST_CPU_OP_BUILDER_UNIQ_HELPER(ctr, name, op) \ - REGISTER_HOST_CPU_OP_BUILDER_UNIQ(ctr, name, op) - -#define REGISTER_HOST_CPU_OP_BUILDER_UNIQ(ctr, name, op) \ - static ::ge::HostCpuOpRegistrar register_host_cpu_op##ctr \ - __attribute__((unused)) = \ - ::ge::HostCpuOpRegistrar(name, []()->::ge::HostCpuOp* { \ - return new (std::nothrow) op(); \ - }) -} // namespace ge - -#endif //INC_REGISTER_REGISTRY_H_ diff --git a/third_party/fwkacllib/inc/register/register_format_transfer.h b/third_party/fwkacllib/inc/register/register_format_transfer.h deleted file mode 100644 index 72da94fc..00000000 --- a/third_party/fwkacllib/inc/register/register_format_transfer.h +++ /dev/null @@ -1,79 +0,0 @@ -/** - * Copyright 2019-2020 Huawei Technologies Co., Ltd - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * 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. - */ - -#ifndef INC_REGISTER_REGISTER_FORMAT_TRANSFER_H_ -#define INC_REGISTER_REGISTER_FORMAT_TRANSFER_H_ - -#include -#include -#include - -#include "external/graph/types.h" -#include "ge/ge_api_error_codes.h" - -namespace ge { -namespace formats { -struct TransArgs { - const uint8_t *data; - Format src_format; - Format dst_format; - // For scenes that need to supplement the shape, for example, 5D to 4D - // It is not possible to convert the format normally if you only get the src_shape, - // and must get the shape before you mend the shape. - // So the parameters here need to be passed in both src_shape and dst_shape - std::vector src_shape; - std::vector dst_shape; - DataType src_data_type; -}; - -struct TransResult { - std::shared_ptr data; - // data length in bytes - size_t length; -}; - -class FormatTransfer { - public: - virtual ~FormatTransfer() = default; - virtual Status TransFormat(const TransArgs &args, TransResult &result) = 0; - virtual Status TransShape(Format src_format, const std::vector &src_shape, DataType data_type, - Format dst_format, std::vector &dst_shape) = 0; -}; - -using FormatTransferBuilder = std::function()>; - -class FormatTransferRegister { - public: - FormatTransferRegister(FormatTransferBuilder builder, Format src, Format dst); - ~FormatTransferRegister() = default; -}; - -#define REGISTER_FORMAT_TRANSFER(TransferClass, format1, format2) \ - namespace { \ - FormatTransferRegister format_transfer_register_##TransferClass##format1##format2( \ - []() { return std::make_shared(); }, format1, format2); \ - } - -/// Build a formattransfer according to 'args' -/// @param args -/// @param result -/// @return -std::shared_ptr BuildFormatTransfer(const TransArgs &args); - -bool FormatTransferExists(const TransArgs &args); -} // namespace formats -} // namespace ge -#endif // INC_REGISTER_REGISTER_FORMAT_TRANSFER_H_ \ No newline at end of file diff --git a/third_party/fwkacllib/inc/runtime/base.h b/third_party/fwkacllib/inc/runtime/base.h index fcb7a627..4e735438 100644 --- a/third_party/fwkacllib/inc/runtime/base.h +++ b/third_party/fwkacllib/inc/runtime/base.h @@ -391,6 +391,8 @@ typedef void (*rtErrorCallback)(rtExceptionType); typedef void (*rtTaskFailCallback)(rtExceptionInfo *exceptionInfo); +typedef void (*rtDeviceStateCallback)(uint32_t devId, bool isOpen); + /** * @ingroup dvrt_base * @brief stream handle. @@ -419,7 +421,7 @@ RTS_API rtError_t rtSetProfDirEx(const char *profDir, const char *address, const * @ingroup profiling_base * @brief init profiler object. */ -RTS_API rtError_t rtProfilerInit(const char *profdir, const char *address, const char *job_ctx); +RTS_API rtError_t rtProfilerInit(const char *profDir, const char *address, const char *jobCtx); /** * @ingroup profiling_base @@ -475,6 +477,16 @@ RTS_API rtError_t rtSetTaskFailCallback(rtTaskFailCallback callback); /** * @ingroup dvrt_base + * @brief register callback for deviceid + * @param [in] uniName unique register name, can't be null + * @param [in] callback Device state callback function + * @param [out] NA + * @return RT_ERROR_NONE for ok + */ +RTS_API rtError_t rtRegDeviceStateCallback(const char *regName, rtDeviceStateCallback callback); + +/** + * @ingroup dvrt_base * @brief notify handle. */ typedef void *rtNotify_t; @@ -593,7 +605,7 @@ RTS_API rtError_t rtLabelCreateEx(rtLabel_t *label, rtStream_t stream); * @return RT_ERROR_NONE for ok * @return RT_ERROR_INVALID_VALUE for input null ptr */ -RTS_API rtError_t rtGetTaskIdAndStreamID(uint32_t *taskid, uint32_t *streamid); +RTS_API rtError_t rtGetTaskIdAndStreamID(uint32_t *taskId, uint32_t *streamId); #if defined(__cplusplus) && !defined(COMPILE_OMG_PACKAGE) } diff --git a/third_party/fwkacllib/inc/runtime/context.h b/third_party/fwkacllib/inc/runtime/context.h index ffcf7e4b..4be49a8c 100644 --- a/third_party/fwkacllib/inc/runtime/context.h +++ b/third_party/fwkacllib/inc/runtime/context.h @@ -139,7 +139,7 @@ RTS_API rtError_t rtSetGroup(int32_t groupId); * @param [in] groupid count * @return RT_ERROR_NONE for ok, errno for failed */ -RTS_API rtError_t rtGetGroupInfo(int32_t groupId, rtGroupInfo_t* groupInfo, uint32_t count); +RTS_API rtError_t rtGetGroupInfo(int32_t groupId, rtGroupInfo_t *groupInfo, uint32_t count); /** * @ingroup diff --git a/third_party/fwkacllib/inc/runtime/dev.h b/third_party/fwkacllib/inc/runtime/dev.h index f3f2937c..b378e3b0 100644 --- a/third_party/fwkacllib/inc/runtime/dev.h +++ b/third_party/fwkacllib/inc/runtime/dev.h @@ -295,7 +295,7 @@ RTS_API rtError_t rtGetRunMode(rtRunMode *mode); * @return RT_ERROR_NONE for ok * @return RT_ERROR_DRV_ERR for can not get aicpu deploy */ -RTS_API rtError_t rtGetAicpuDeploy(rtAicpuDeployType_t *deplyType); +RTS_API rtError_t rtGetAicpuDeploy(rtAicpuDeployType_t *deployType); /** * @ingroup dvrt_dev @@ -309,7 +309,7 @@ RTS_API rtError_t rtSetSocVersion(const char *version); * @brief get chipType * @return RT_ERROR_NONE for ok */ -rtError_t rtGetSocVersion(char *version, const uint32_t maxLen); +RTS_API rtError_t rtGetSocVersion(char *version, const uint32_t maxLen); /** * @ingroup dvrt_dev diff --git a/third_party/fwkacllib/inc/runtime/event.h b/third_party/fwkacllib/inc/runtime/event.h index c8a227e8..41e611ea 100644 --- a/third_party/fwkacllib/inc/runtime/event.h +++ b/third_party/fwkacllib/inc/runtime/event.h @@ -59,6 +59,16 @@ RTS_API rtError_t rtEventDestroy(rtEvent_t event); /** * @ingroup dvrt_event + * @brief get event id + * @param [in] event_ event to be get + * @param [in|out] event_id event_id id + * @return RT_ERROR_NONE for ok + * @return RT_ERROR_INVALID_VALUE for error input + */ +RTS_API rtError_t rtGetEventID(rtEvent_t event, uint32_t *eventId); + +/** + * @ingroup dvrt_event * @brief event record * @param [int] event event to record * @param [int] stream stream handle @@ -116,13 +126,13 @@ RTS_API rtError_t rtEventGetTimeStamp(uint64_t *time, rtEvent_t event); /** * @ingroup dvrt_event * @brief name an event - * @param [in] event_ event to be named + * @param [in] event event to be named * @param [in] name identification name * @return RT_ERROR_NONE for ok * @return RT_ERROR_INVALID_VALUE for error input of event, name * @return RT_ERROR_DRV_ERR for driver error */ -RTS_API rtError_t rtNameEvent(rtEvent_t event_, const char *name); +RTS_API rtError_t rtNameEvent(rtEvent_t event, const char *name); /** * @ingroup dvrt_event @@ -132,7 +142,7 @@ RTS_API rtError_t rtNameEvent(rtEvent_t event_, const char *name); * @return RT_ERROR_NONE for ok * @return RT_ERROR_INVALID_VALUE for error input */ -RTS_API rtError_t rtNotifyCreate(int32_t device_id, rtNotify_t *notify_); +RTS_API rtError_t rtNotifyCreate(int32_t deviceId, rtNotify_t *notify); /** * @ingroup dvrt_event @@ -142,7 +152,7 @@ RTS_API rtError_t rtNotifyCreate(int32_t device_id, rtNotify_t *notify_); * @return RT_ERROR_INVALID_VALUE for error input * @return RT_ERROR_DRV_ERR for driver error */ -RTS_API rtError_t rtNotifyDestroy(rtNotify_t notify_); +RTS_API rtError_t rtNotifyDestroy(rtNotify_t notify); /** * @ingroup dvrt_event @@ -153,7 +163,7 @@ RTS_API rtError_t rtNotifyDestroy(rtNotify_t notify_); * @return RT_ERROR_INVALID_VALUE for error input * @return RT_ERROR_STREAM_CONTEXT for stream is not in current ctx */ -RTS_API rtError_t rtNotifyRecord(rtNotify_t notify_, rtStream_t stream_); +RTS_API rtError_t rtNotifyRecord(rtNotify_t notify, rtStream_t stream); /** * @ingroup dvrt_event @@ -164,7 +174,7 @@ RTS_API rtError_t rtNotifyRecord(rtNotify_t notify_, rtStream_t stream_); * @return RT_ERROR_INVALID_VALUE for error input * @return RT_ERROR_STREAM_CONTEXT for stream is not in current ctx */ -RTS_API rtError_t rtNotifyWait(rtNotify_t notify_, rtStream_t stream_); +RTS_API rtError_t rtNotifyWait(rtNotify_t notify, rtStream_t stream); /** * @ingroup dvrt_event @@ -174,7 +184,7 @@ RTS_API rtError_t rtNotifyWait(rtNotify_t notify_, rtStream_t stream_); * @return RT_ERROR_NONE for ok * @return RT_ERROR_INVALID_VALUE for error input */ -RTS_API rtError_t rtNameNotify(rtNotify_t notify_, const char *name); +RTS_API rtError_t rtNameNotify(rtNotify_t notify, const char *name); /** * @ingroup dvrt_event @@ -184,7 +194,7 @@ RTS_API rtError_t rtNameNotify(rtNotify_t notify_, const char *name); * @return RT_ERROR_NONE for ok * @return RT_ERROR_INVALID_VALUE for error input */ -RTS_API rtError_t rtGetNotifyID(rtNotify_t notify_, uint32_t *notify_id); +RTS_API rtError_t rtGetNotifyID(rtNotify_t notify, uint32_t *notifyId); /** * @ingroup dvrt_event @@ -199,7 +209,7 @@ RTS_API rtError_t rtIpcSetNotifyName(rtNotify_t notify, char *name, uint32_t len /** * @ingroup dvrt_event * @brief Open IPC notify - * @param [in] notify notify to be opened + * @param [out] notify the opened notify * @param [in] name identification name * @return RT_ERROR_NONE for ok * @return RT_ERROR_INVALID_VALUE for error input diff --git a/third_party/fwkacllib/inc/runtime/kernel.h b/third_party/fwkacllib/inc/runtime/kernel.h index de806a9e..5f519442 100644 --- a/third_party/fwkacllib/inc/runtime/kernel.h +++ b/third_party/fwkacllib/inc/runtime/kernel.h @@ -170,6 +170,18 @@ typedef void (*rtCallback_t)(void *fnData); #define RT_DEV_BINARY_MAGIC_ELF_AIVEC 0x41415246 /** + * @ingroup rt_kernel + * @brief magic number of elf binary for aicube + */ +#define RT_DEV_BINARY_MAGIC_ELF_AICUBE 0x41415247 + +/** + * @ingroup rt_kernel + * @brief magic number of elf binary for aivector + */ +#define RT_DEV_BINARY_MAGIC_ELF_AIVECTOR 0x41415248 + +/** * @ingroup rt_kernel_flags * @brief kernel op bit flags */ @@ -285,7 +297,7 @@ RTS_API rtError_t rtQueryFunctionRegistered(const char *stubName); * @return RT_ERROR_INVALID_VALUE for error input */ RTS_API rtError_t rtKernelConfigDump(uint32_t kind, uint32_t dumpSizePerBlock, uint32_t blockDim, void **dumpBaseAddr, - rtStream_t stream_); + rtStream_t stream); /** * @ingroup rt_kernel @@ -297,7 +309,7 @@ RTS_API rtError_t rtKernelConfigDump(uint32_t kind, uint32_t dumpSizePerBlock, u * @param [in] smDesc shared memory description * @param [in] stream associated stream * @return RT_ERROR_NONE for ok - * @return RT_ERROR_INVALID_VALUE for error input + * @return RT_ERROR_INVALID_VALUE for error input */ RTS_API rtError_t rtKernelLaunch(const void *stubFunc, uint32_t blockDim, void *args, uint32_t argsSize, rtSmDesc_t *smDesc, rtStream_t stream); @@ -313,7 +325,7 @@ RTS_API rtError_t rtKernelLaunch(const void *stubFunc, uint32_t blockDim, void * * @param [in] stream associated stream * @param [in] flag dump flag * @return RT_ERROR_NONE for ok - * @return RT_ERROR_INVALID_VALUE for error input + * @return RT_ERROR_INVALID_VALUE for error input */ RTS_API rtError_t rtKernelLaunchWithFlag(const void *stubFunc, uint32_t blockDim, void *args, uint32_t argsSize, rtSmDesc_t *smDesc, rtStream_t stream, uint32_t flags); @@ -375,7 +387,7 @@ typedef void *rtModel_t; * @return RT_ERROR_NONE for ok * @return RT_ERROR_INVALID_VALUE for error input */ - RTS_API rtError_t rtDumpAddrSet(rtModel_t model, void *addr , uint32_t dumpSize, uint32_t flag); + RTS_API rtError_t rtDumpAddrSet(rtModel_t model, void *addr, uint32_t dumpSize, uint32_t flag); /** * @ingroup rt_kernel @@ -444,7 +456,7 @@ RTS_API rtError_t rtKernelConfigTransArg(const void *ptr, uint64_t size, uint32_ * @brief start fusion kernels. * @param [in] stream stream for fusion kernels * @return RT_ERROR_NONE for ok - * @return RT_ERROR_INVALID_VALUE for error input + * @return RT_ERROR_INVALID_VALUE for error input */ RTS_API rtError_t rtKernelFusionStart(rtStream_t stream); @@ -453,7 +465,7 @@ RTS_API rtError_t rtKernelFusionStart(rtStream_t stream); * @brief end fusion kernels. * @param [in] stream stream for fusion kernels * @return RT_ERROR_NONE for ok - * @return RT_ERROR_INVALID_VALUE for error input + * @return RT_ERROR_INVALID_VALUE for error input */ RTS_API rtError_t rtKernelFusionEnd(rtStream_t stream); @@ -462,7 +474,7 @@ RTS_API rtError_t rtKernelFusionEnd(rtStream_t stream); * @brief set kernelinfo callback * @param [in] callback * @return RT_ERROR_NONE for ok - * @return RT_ERROR_INVALID_VALUE for error input + * @return RT_ERROR_INVALID_VALUE for error input */ RTS_API rtError_t rtSetKernelReportCallback(rtKernelReportCallback callBack); @@ -472,7 +484,7 @@ RTS_API rtError_t rtSetKernelReportCallback(rtKernelReportCallback callBack); * @param [in] threadId thread id for stream * @param [in] stream stream for subscribe * @return RT_ERROR_NONE for ok - * @return RT_ERROR_INVALID_VALUE for error input + * @return RT_ERROR_INVALID_VALUE for error input */ RTS_API rtError_t rtSubscribeReport(uint64_t threadId, rtStream_t stream); @@ -483,7 +495,7 @@ RTS_API rtError_t rtSubscribeReport(uint64_t threadId, rtStream_t stream); * @param [in] fnData user data * @param [in] stream subscribed stream * @return RT_ERROR_NONE for ok - * @return RT_ERROR_INVALID_VALUE for error input + * @return RT_ERROR_INVALID_VALUE for error input */ RTS_API rtError_t rtCallbackLaunch(rtCallback_t callBackFunc, void *fnData, rtStream_t stream, bool isBlock); @@ -492,7 +504,7 @@ RTS_API rtError_t rtCallbackLaunch(rtCallback_t callBackFunc, void *fnData, rtSt * @brief process callback report. * @param [in] timeout if timeout=-1, while(1); else timeout * @return RT_ERROR_NONE for ok - * @return RT_ERROR_INVALID_VALUE for error input + * @return RT_ERROR_INVALID_VALUE for error input */ RTS_API rtError_t rtProcessReport(int32_t timeout); @@ -502,7 +514,7 @@ RTS_API rtError_t rtProcessReport(int32_t timeout); * @param [in] threadId thread id for stream * @param [in] stream stream for subscribe * @return RT_ERROR_NONE for ok - * @return RT_ERROR_INVALID_VALUE for error input + * @return RT_ERROR_INVALID_VALUE for error input */ RTS_API rtError_t rtUnSubscribeReport(uint64_t threadId, rtStream_t stream); @@ -510,7 +522,7 @@ RTS_API rtError_t rtUnSubscribeReport(uint64_t threadId, rtStream_t stream); * @ingroup profiling_base * @brief start online prof. * @return RT_ERROR_NONE for ok - * @return RT_ERROR_INVALID_VALUE for error input + * @return RT_ERROR_INVALID_VALUE for error input */ RTS_API rtError_t rtStartOnlineProf(rtStream_t stream, uint32_t sampleNum); @@ -518,7 +530,7 @@ RTS_API rtError_t rtStartOnlineProf(rtStream_t stream, uint32_t sampleNum); * @ingroup profiling_base * @brief stop online prof. * @return RT_ERROR_NONE for ok - * @return RT_ERROR_INVALID_VALUE for error input + * @return RT_ERROR_INVALID_VALUE for error input */ RTS_API rtError_t rtStopOnlineProf(rtStream_t stream); @@ -526,9 +538,26 @@ RTS_API rtError_t rtStopOnlineProf(rtStream_t stream); * @ingroup profiling_base * @brief get online prof. * @return RT_ERROR_NONE for ok - * @return RT_ERROR_INVALID_VALUE for error input + * @return RT_ERROR_INVALID_VALUE for error input */ RTS_API rtError_t rtGetOnlineProfData(rtStream_t stream, rtProfDataInfo_t *pProfData, uint32_t profDataNum); + +/** + * @ingroup profiling_base + * @brief start mdc profiler. + * @return RT_ERROR_NONE for ok + * @return RT_ERROR_INVALID_VALUE for error input + */ +RTS_API rtError_t rtStartMDCProfiler(void **addr, uint32_t length); + +/** + * @ingroup profiling_base + * @brief stop mdc profiler. + * @return RT_ERROR_NONE for ok + * @return RT_ERROR_INVALID_VALUE for error input + */ +RTS_API rtError_t rtStopMDCProfiler(void *addr); + #if defined(__cplusplus) && !defined(COMPILE_OMG_PACKAGE) } #endif diff --git a/third_party/fwkacllib/inc/runtime/mem.h b/third_party/fwkacllib/inc/runtime/mem.h index 22a00ac6..e65d8604 100644 --- a/third_party/fwkacllib/inc/runtime/mem.h +++ b/third_party/fwkacllib/inc/runtime/mem.h @@ -475,7 +475,7 @@ RTS_API rtError_t rtIpcSetMemoryName(const void *ptr, uint64_t byteCount, char * * @return RT_ERROR_INVALID_VALUE for error input * @return RT_ERROR_DRV_ERR for driver error */ -rtError_t rtIpcDestroyMemoryName(const char *name); +RTS_API rtError_t rtIpcDestroyMemoryName(const char *name); /** * @ingroup dvrt_mem @@ -503,13 +503,13 @@ RTS_API rtError_t rtIpcCloseMemory(const void *ptr); * @ingroup dvrt_mem * @brief HCCL Async memory cpy * @param [in] index sq index - * @param [in] wqe_index moudle index + * @param [in] wqeIndex moudle index * @param [in] stream asynchronized task stream * @return RT_ERROR_NONE for ok * @return RT_ERROR_INVALID_VALUE for error input * @return RT_ERROR_DRV_ERR for driver error */ -RTS_API rtError_t rtRDMASend(uint32_t index, uint32_t wqe_index, rtStream_t stream); +RTS_API rtError_t rtRDMASend(uint32_t index, uint32_t wqeIndex, rtStream_t stream); /** * @ingroup dvrt_mem diff --git a/third_party/fwkacllib/inc/runtime/rt_model.h b/third_party/fwkacllib/inc/runtime/rt_model.h index 78bb2321..b72b142d 100644 --- a/third_party/fwkacllib/inc/runtime/rt_model.h +++ b/third_party/fwkacllib/inc/runtime/rt_model.h @@ -438,7 +438,8 @@ RTS_API rtError_t rtModelGetId(rtModel_t model, uint32_t *modelId); * @return RT_ERROR_NONE for ok * @return RT_ERROR_INVALID_VALUE for error input */ -rtError_t rtDebugRegister(rtModel_t model, uint32_t flag, const void *addr, uint32_t *streamId, uint32_t *taskId); +RTS_API rtError_t rtDebugRegister(rtModel_t model, uint32_t flag, const void *addr, + uint32_t *streamId, uint32_t *taskId); /* * @ingroup rt_model diff --git a/third_party/fwkacllib/inc/runtime/stream.h b/third_party/fwkacllib/inc/runtime/stream.h index 81810a21..388fd3c2 100644 --- a/third_party/fwkacllib/inc/runtime/stream.h +++ b/third_party/fwkacllib/inc/runtime/stream.h @@ -126,17 +126,17 @@ RTS_API rtError_t rtGetStreamId(rtStream_t stream, int32_t *streamId); * @return RT_ERROR_NONE for complete * @return RT_ERROR_INVALID_VALUE for error input */ -RTS_API rtError_t rtGetMaxStreamAndTask(uint32_t streamType, uint32_t *MaxStrCount, uint32_t *MaxTaskCount); +RTS_API rtError_t rtGetMaxStreamAndTask(uint32_t streamType, uint32_t *maxStrCount, uint32_t *maxTaskCount); /** * @ingroup dvrt_stream * @brief Name a stream - * @param [in] stream_ stream to be named + * @param [in] stream stream to be named * @param [in] name identification name * @return RT_ERROR_NONE for complete * @return RT_ERROR_INVALID_VALUE for error input */ -RTS_API rtError_t rtNameStream(rtStream_t stream_, const char *name); +RTS_API rtError_t rtNameStream(rtStream_t stream, const char *name); /** * @ingroup dvrt_stream @@ -144,12 +144,12 @@ RTS_API rtError_t rtNameStream(rtStream_t stream_, const char *name); * @param [in] ptr Determine the address where the value of the true and false branches is located * @param [in] condition switch condition * @param [in] value switch value - * @param [in] true_stream Stream that needs to be activated when the value is non-zero + * @param [in] trueStream Stream that needs to be activated when the value is non-zero * @param [in] stream input stream to init task * @return RT_ERROR_NONE for complete * @return RT_ERROR_INVALID_VALUE for error input */ -RTS_API rtError_t rtStreamSwitch(void *ptr, rtCondition_t condition, int64_t value, rtStream_t true_stream, +RTS_API rtError_t rtStreamSwitch(void *ptr, rtCondition_t condition, int64_t value, rtStream_t trueStream, rtStream_t stream); /** @@ -162,18 +162,18 @@ RTS_API rtError_t rtStreamSwitch(void *ptr, rtCondition_t condition, int64_t val * @param [in] dataType data type of target value * @return RT_ERROR_NONE for complete */ -RTS_API rtError_t rtStreamSwitchEx(void *ptr, rtCondition_t condition, void *value_ptr, rtStream_t true_stream, +RTS_API rtError_t rtStreamSwitchEx(void *ptr, rtCondition_t condition, void *valuePtr, rtStream_t trueStream, rtStream_t stream, rtSwitchDataType_t dataType); /** * @ingroup dvrt_stream * @brief Active a stream - * @param [in] active_stream stream to be activated + * @param [in] activeStream stream to be activated * @param [in] stream input stream to init task * @return RT_ERROR_NONE for complete * @return RT_ERROR_INVALID_VALUE for error input */ -RTS_API rtError_t rtStreamActive(rtStream_t active_stream, rtStream_t stream); +RTS_API rtError_t rtStreamActive(rtStream_t activeStream, rtStream_t stream); /** * @brief execute extensible stream case switch task diff --git a/third_party/fwkacllib/inc/tdt/index_transform.h b/third_party/fwkacllib/inc/tdt/index_transform.h new file mode 100644 index 00000000..a5af2c83 --- /dev/null +++ b/third_party/fwkacllib/inc/tdt/index_transform.h @@ -0,0 +1,29 @@ +/** +* @file index_transform.h +* +* Copyright (C) Huawei Technologies Co., Ltd. 2018-2019. All Rights Reserved. +* +* This program is used to get logical device id by phy device id. +*/ + +#ifndef INC_TDT_INDEX_TRANSFORM_H +#define INC_TDT_INDEX_TRANSFORM_H + +#include "stdint.h" +/** +* @ingroup IndexTransform +* @brief get logical device id by phy device id. +* +* @par Function get logical device id by phy device id. +* +* @param phyId [IN] physical device id +* @param logicalId [OUT] logical device id +* @retval 0 Success +* @retval OtherValues Fail +* +* @par Dependency +* @li libruntime.so: Library to which the interface belongs. +*/ + +int32_t IndexTransform(const uint32_t phyId, uint32_t &logicId); +#endif diff --git a/third_party/fwkacllib/inc/tdt/status.h b/third_party/fwkacllib/inc/tdt/status.h index a42f38a5..d30564b8 100644 --- a/third_party/fwkacllib/inc/tdt/status.h +++ b/third_party/fwkacllib/inc/tdt/status.h @@ -193,6 +193,7 @@ enum { TDT_HDC_SRV_TYPE_ERROR_CODE, TDT_TSD_CLT_OPEN_FAILED_CODE, TDT_TSD_CLT_CLOSE_FAILED_CODE, + TDT_TSD_CLT_UPDATE_PROFILING_FAILED_CODE, TDT_TSD_CLT_INTERFACE_NOT_SUPPORT_CODE, TDT_SUPERVISOR_ILLEGAL_HEARTBEAT_TIME_CODE, TDT_SUPERVISOR_INOTIFY_READ_SIZE_ERROR_CODE, @@ -697,6 +698,8 @@ TDT_DEF_ERROR_CODE(MODID_HDC_SERVER, TDT_ERROR, TDT_BIND_CPUCORE_FAILED, "thread TDT_DEF_ERROR_CODE(MODID_HDC_SERVER, TDT_ERROR, TDT_HDC_SRV_CLOSED, "hdc server has been closed"); TDT_DEF_ERROR_CODE(MODID_TSD_CLIENT, TDT_ERROR, TDT_TSD_CLT_OPEN_FAILED, "tsd client open failed"); TDT_DEF_ERROR_CODE(MODID_TSD_CLIENT, TDT_ERROR, TDT_TSD_CLT_CLOSE_FAILED, "tsd client close failed"); +TDT_DEF_ERROR_CODE(MODID_TSD_CLIENT, TDT_ERROR, TDT_TSD_CLT_UPDATE_PROFILING_FAILED, + "tsd client update profiling failed"); TDT_DEF_ERROR_CODE(MODID_TSD_CLIENT, TDT_ERROR, TDT_TSD_CLT_INTERFACE_NOT_SUPPORT, "tsd client func not support"); TDT_DEF_ERROR_CODE(MODID_TDT_PREFETCH, TDT_ERROR, TDT_PREFETCH_FILELIST_NOT_EXIST, "tdt filelist open failed"); TDT_DEF_ERROR_CODE(MODID_TDT_PREFETCH, TDT_ERROR, TDT_PREFETCH_SAMPLE_FILE_NOT_FOUND, "tdt sample file is empty"); diff --git a/third_party/fwkacllib/inc/tdt/tsd_client.h b/third_party/fwkacllib/inc/tdt/tsd_client.h index 7886488e..6066a12e 100644 --- a/third_party/fwkacllib/inc/tdt/tsd_client.h +++ b/third_party/fwkacllib/inc/tdt/tsd_client.h @@ -49,7 +49,7 @@ extern "C" { * @li tsd_client.h: Header file where the interface declaration is located. * @li data_common.h: Header file where 'TDT_StatusT' defined */ -TDT_StatusT TsdOpen(const uint32_t phyDeviceId, const uint32_t rankSize); +TDT_LIB_EXPORT TDT_StatusT TsdOpen(const uint32_t phyDeviceId, const uint32_t rankSize); /** * @ingroup Close @@ -67,7 +67,25 @@ TDT_StatusT TsdOpen(const uint32_t phyDeviceId, const uint32_t rankSize); * @li tsd_client.h: Header file where the interface declaration is located. * @li data_common.h: Header file where 'TDT_StatusT' defined */ -TDT_StatusT TsdClose(const uint32_t phyDeviceId); +TDT_LIB_EXPORT TDT_StatusT TsdClose(const uint32_t phyDeviceId); + +/** +* @ingroup UpdateProfilingMode +* @brief notify TSDClient update profiling mode +* +* @par Function +* notify TSDClient update profiling mode +* +* @param NA +* @retval TDT_OK Success +* @retval OtherValues Failure +* +* @par Dependency +* @li libtsdclient.so: Library to which the interface belongs. +* @li tsd_client.h: Header file where the interface declaration is located. +* @li data_common.h: Header file where 'TDT_StatusT' defined +*/ +TDT_LIB_EXPORT TDT_StatusT UpdateProfilingMode(const uint32_t phyDeviceId, const uint32_t flag); /** * @ingroup CreateCmdParameterObj diff --git a/third_party/fwkacllib/inc/toolchain/slog.h b/third_party/fwkacllib/inc/toolchain/slog.h index 2cb00a05..bce58f32 100644 --- a/third_party/fwkacllib/inc/toolchain/slog.h +++ b/third_party/fwkacllib/inc/toolchain/slog.h @@ -25,10 +25,20 @@ extern "C" { #define LINUX 0 #endif // LINUX +#ifndef WIN +#define WIN 1 +#endif + #ifndef OS_TYPE #define OS_TYPE 0 #endif // OS_TYPE +#if (OS_TYPE == LINUX) +#define DLL_EXPORT __attribute__((visibility("default"))) +#else +#define DLL_EXPORT _declspec(dllexport) +#endif + /** * @ingroup slog * @@ -180,12 +190,11 @@ enum { INVLID_MOUDLE_ID }; -#if (OS_TYPE == LINUX) /** * @ingroup slog * @brief External log interface, which called by modules */ -extern void dlog_init(void); +DLL_EXPORT void dlog_init(void); /** * @ingroup slog @@ -195,7 +204,7 @@ extern void dlog_init(void); * @param [out]enableEvent: 1: enable; 0: disable * @return: module level(0: debug, 1: info, 2: warning, 3: error, 4: null output) */ -extern int dlog_getlevel(int moduleId, int *enableEvent); +DLL_EXPORT int dlog_getlevel(int moduleId, int *enableEvent); /** * @ingroup slog @@ -206,7 +215,7 @@ extern int dlog_getlevel(int moduleId, int *enableEvent); * @param [in]enableEvent: 1: enable; 0: disable, others:invalid * @return: 0: SUCCEED, others: FAILED */ -extern int dlog_setlevel(int moduleId, int level, int enableEvent); +DLL_EXPORT int dlog_setlevel(int moduleId, int level, int enableEvent); /** * @ingroup slog @@ -217,7 +226,7 @@ extern int dlog_setlevel(int moduleId, int level, int enableEvent); * @param [in]logLevel: eg: DLOG_EVENT/DLOG_ERROR/DLOG_WARN/DLOG_INFO/DLOG_DEBUG * @return: 1:enable, 0:disable */ -extern int CheckLogLevel(int moduleId, int logLevel); +DLL_EXPORT int CheckLogLevel(int moduleId, int logLevel); /** * @ingroup slog @@ -343,7 +352,7 @@ extern int CheckLogLevel(int moduleId, int logLevel); * @ingroup slog * @brief DlogFlush: flush log buffer to file */ -void DlogFlush(void); +DLL_EXPORT void DlogFlush(void); /** * @ingroup slog @@ -357,11 +366,6 @@ void DlogEventInner(int moduleId, const char *fmt, ...); void DlogInner(int moduleId, int level, const char *fmt, ...); void DlogWithKVInner(int moduleId, int level, KeyValue *pstKVArray, int kvNum, const char *fmt, ...); -#else -_declspec(dllexport) void dlog_init(void); -_declspec(dllexport) int dlog_getlevel(int moduleId, int *enableEvent); -#endif // OS_TYPE - #ifdef __cplusplus } #endif // __cplusplus diff --git a/third_party/fwkacllib/inc/toolchain/tuning_tool/tune_api.h b/third_party/fwkacllib/inc/toolchain/tuning_tool/tune_api.h new file mode 100644 index 00000000..12b6aa1e --- /dev/null +++ b/third_party/fwkacllib/inc/toolchain/tuning_tool/tune_api.h @@ -0,0 +1,72 @@ +/** + * @file tune_api.h + * + * Copyright (c) Huawei Technologies Co., Ltd. 2020-2020. All rights reserved.\n + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n + * 描述：mstune调优接口头文件 + */ +/** @defgroup mstune mstune调优接口 */ +#ifndef TUNE_API_H +#define TUNE_API_H +#include +#include +#include +#include "graph/graph.h" +#include "ge/ge_api.h" + +/** + * @ingroup mstune + * + * mstune status + */ +enum MsTuneStatus { + MSTUNE_SUCCESS, /** tune success */ + MSTUNE_FAILED, /** tune failed */ +}; + +// Option key: for train options sets +const std::string MSTUNE_SELF_KEY = "mstune"; +const std::string MSTUNE_GEINIT_KEY = "initialize"; +const std::string MSTUNE_GESESS_KEY = "session"; + +/** + * @ingroup mstune + * @par 描述: 命令行调优 + * + * @attention 无 + * @param option [IN] 调优参数 + * @param msg [OUT] 调优异常下返回信息 + * @retval #MSTUNE_SUCCESS 执行成功 + * @retval #MSTUNE_FAILED 执行失败 + * @par 依赖: + * @li tune_api.cpp：该接口所属的开发包。 + * @li tune_api.h：该接口声明所在的头文件。 + * @see 无 + * @since + */ +MsTuneStatus MsTuning(const std::map &option, std::string &msg); + +/** + * @ingroup mstune + * @par 描述: 梯度调优 + * + * @attention 无 + * @param tuningGraph [IN] 调优图 + * @param dependGraph [IN] 调优依赖图 + * @param session [IN] ge连接会话 + * @param option [IN] 参数集. 包含调优参数及ge参数 + * @retval #MSTUNE_SUCCESS 执行成功 + * @retval #MSTUNE_FAILED 执行失败 + * @par 依赖: + * @li tune_api.cpp：该接口所属的开发包。 + * @li tune_api.h：该接口声明所在的头文件。 + * @see 无 + * @since + */ +extern "C" MsTuneStatus MsTrainTuning(ge::Graph &tuningGraph, std::vector &dependGraph, + ge::Session *session, const std::map> &option); + +#endif diff --git a/third_party/prebuild/aarch64/libmmpa.a b/third_party/prebuild/aarch64/libmmpa.a new file mode 100755 index 00000000..d7c29e2b Binary files /dev/null and b/third_party/prebuild/aarch64/libmmpa.a differ diff --git a/third_party/prebuild/x86_64/libmmpa.a b/third_party/prebuild/x86_64/libmmpa.a new file mode 100755 index 00000000..bec195ad Binary files /dev/null and b/third_party/prebuild/x86_64/libmmpa.a differ