已合并
[Refactor] HIXL头文件整改 #438
ZhaoBaiwei创建于 5月7日
[Refactor] HIXL头文件整改 #438
已合并
共 37 个文件变更+153-325
| @@ -484,9 +484,7 @@ Software: metadef fedff4f | |||
| 484 | 484 | ||
| 485 | Official Repository: https://gitee.com/ascend/metadef | 485 | Official Repository: https://gitee.com/ascend/metadef |
| 486 | 486 | ||
| 487 | -Usage: " | 487 | +Usage: "tests/depends/error_manager/src/error_manager.h" |
| 488 | -src/llm_datadist/common/aligned_ptr.cc, | ||
| 489 | -tests/depends/error_manager/src/error_manager.h" | ||
| 490 | 488 | ||
| 491 | Copyright Notice: | 489 | Copyright Notice: |
| 492 | Copyright 2020 Huawei Technologies Co., Ltd. | 490 | Copyright 2020 Huawei Technologies Co., Ltd. |
| @@ -496,18 +494,6 @@ License: Apache License V2.0 | |||
| 496 | Please see above. | 494 | Please see above. |
| 497 | 495 | ||
| 498 | 496 | ||
| 499 | -Software: metadef 6e92218 | ||
| 500 | - | ||
| 501 | -Official Repository: https://gitee.com/ascend/metadef | ||
| 502 | - | ||
| 503 | -Usage: "src/llm_datadist/common/aligned_ptr.h" | ||
| 504 | - | ||
| 505 | -Copyright Notice: Copyright 2020 Huawei Technologies Co., Ltd. | ||
| 506 | - | ||
| 507 | -License: Apache License V2.0 | ||
| 508 | -Please see above. | ||
| 509 | - | ||
| 510 | - | ||
| 511 | Software: metadef d309580 | 497 | Software: metadef d309580 |
| 512 | 498 | ||
| 513 | Official Repository: https://gitee.com/ascend/metadef | 499 | Official Repository: https://gitee.com/ascend/metadef |
| @@ -8,47 +8,47 @@ | |||
| 8 | * See LICENSE in the root of the software repository for the full text of the License. | 8 | * See LICENSE in the root of the software repository for the full text of the License. |
| 9 | */ | 9 | */ |
| 10 | 10 | ||
| 11 | -#include "adxl_engine.h" | 11 | +#include "comm_engine.h" |
| 12 | 12 | ||
| 13 | 13 | ||
| 14 | namespace hixl { | 14 | namespace hixl { |
| 15 | -Status AdxlEngine::Initialize(const std::map<AscendString, AscendString> &options) { | 15 | +Status CommEngine::Initialize(const std::map<AscendString, AscendString> &options) { |
| 16 | return adxl_inner_engine_.Initialize(options); | 16 | return adxl_inner_engine_.Initialize(options); |
| 17 | } | 17 | } |
| 18 | 18 | ||
| 19 | -void AdxlEngine::Finalize() { | 19 | +void CommEngine::Finalize() { |
| 20 | adxl_inner_engine_.Finalize(); | 20 | adxl_inner_engine_.Finalize(); |
| 21 | } | 21 | } |
| 22 | 22 | ||
| 23 | -bool AdxlEngine::IsInitialized() const { | 23 | +bool CommEngine::IsInitialized() const { |
| 24 | return adxl_inner_engine_.IsInitialized(); | 24 | return adxl_inner_engine_.IsInitialized(); |
| 25 | } | 25 | } |
| 26 | 26 | ||
| 27 | -Status AdxlEngine::RegisterMem(const MemDesc &mem, MemType type, MemHandle &mem_handle) { | 27 | +Status CommEngine::RegisterMem(const MemDesc &mem, MemType type, MemHandle &mem_handle) { |
| 28 | adxl::MemDesc adxl_mem{mem.addr, mem.len}; | 28 | adxl::MemDesc adxl_mem{mem.addr, mem.len}; |
| 29 | adxl::MemType adxl_type = static_cast<adxl::MemType>(type); | 29 | adxl::MemType adxl_type = static_cast<adxl::MemType>(type); |
| 30 | return adxl_inner_engine_.RegisterMem(adxl_mem, adxl_type, mem_handle); | 30 | return adxl_inner_engine_.RegisterMem(adxl_mem, adxl_type, mem_handle); |
| 31 | } | 31 | } |
| 32 | 32 | ||
| 33 | -Status AdxlEngine::DeregisterMem(MemHandle mem_handle) { | 33 | +Status CommEngine::DeregisterMem(MemHandle mem_handle) { |
| 34 | return adxl_inner_engine_.DeregisterMem(mem_handle); | 34 | return adxl_inner_engine_.DeregisterMem(mem_handle); |
| 35 | } | 35 | } |
| 36 | 36 | ||
| 37 | -Status AdxlEngine::Connect(const AscendString &remote_engine, int32_t timeout_in_millis) { | 37 | +Status CommEngine::Connect(const AscendString &remote_engine, int32_t timeout_in_millis) { |
| 38 | return adxl_inner_engine_.Connect(remote_engine, timeout_in_millis); | 38 | return adxl_inner_engine_.Connect(remote_engine, timeout_in_millis); |
| 39 | } | 39 | } |
| 40 | 40 | ||
| 41 | -Status AdxlEngine::Disconnect(const AscendString &remote_engine, int32_t timeout_in_millis) { | 41 | +Status CommEngine::Disconnect(const AscendString &remote_engine, int32_t timeout_in_millis) { |
| 42 | return adxl_inner_engine_.Disconnect(remote_engine, timeout_in_millis); | 42 | return adxl_inner_engine_.Disconnect(remote_engine, timeout_in_millis); |
| 43 | } | 43 | } |
| 44 | 44 | ||
| 45 | -void AdxlEngine::Disconnect() { | 45 | +void CommEngine::Disconnect() { |
| 46 | adxl_inner_engine_.Disconnect(); | 46 | adxl_inner_engine_.Disconnect(); |
| 47 | } | 47 | } |
| 48 | 48 | ||
| 49 | -Status AdxlEngine::TransferSync(const AscendString &remote_engine, TransferOp operation, | 49 | +Status CommEngine::TransferSync(const AscendString &remote_engine, TransferOp operation, |
| 50 | const std::vector<TransferOpDesc> &op_descs, int32_t timeout_in_millis) { | 50 | const std::vector<TransferOpDesc> &op_descs, int32_t timeout_in_millis) { |
| 51 | - HIXL_CHK_BOOL_RET_STATUS(!op_descs.empty(), PARAM_INVALID, "[AdxlEngine] TransferSync failed, op_descs is empty."); | 51 | + HIXL_CHK_BOOL_RET_STATUS(!op_descs.empty(), PARAM_INVALID, "[CommEngine] TransferSync failed, op_descs is empty."); |
| 52 | adxl::TransferOp adxl_operation = static_cast<adxl::TransferOp>(operation); | 52 | adxl::TransferOp adxl_operation = static_cast<adxl::TransferOp>(operation); |
| 53 | std::vector<adxl::TransferOpDesc> adxl_op_descs; | 53 | std::vector<adxl::TransferOpDesc> adxl_op_descs; |
| 54 | for (const auto &op_desc : op_descs) { | 54 | for (const auto &op_desc : op_descs) { |
| @@ -57,10 +57,10 @@ Status AdxlEngine::TransferSync(const AscendString &remote_engine, TransferOp op | |||
| 57 | return adxl_inner_engine_.TransferSync(remote_engine, adxl_operation, adxl_op_descs, timeout_in_millis); | 57 | return adxl_inner_engine_.TransferSync(remote_engine, adxl_operation, adxl_op_descs, timeout_in_millis); |
| 58 | } | 58 | } |
| 59 | 59 | ||
| 60 | -Status AdxlEngine::TransferAsync(const AscendString &remote_engine, TransferOp operation, | 60 | +Status CommEngine::TransferAsync(const AscendString &remote_engine, TransferOp operation, |
| 61 | const std::vector<TransferOpDesc> &op_descs, const TransferArgs &optional_args, | 61 | const std::vector<TransferOpDesc> &op_descs, const TransferArgs &optional_args, |
| 62 | TransferReq &req) { | 62 | TransferReq &req) { |
| 63 | - HIXL_CHK_BOOL_RET_STATUS(!op_descs.empty(), PARAM_INVALID, "[AdxlEngine] TransferAsync failed, op_descs is empty."); | 63 | + HIXL_CHK_BOOL_RET_STATUS(!op_descs.empty(), PARAM_INVALID, "[CommEngine] TransferAsync failed, op_descs is empty."); |
| 64 | (void)optional_args; | 64 | (void)optional_args; |
| 65 | adxl::TransferOp adxl_operation = static_cast<adxl::TransferOp>(operation); | 65 | adxl::TransferOp adxl_operation = static_cast<adxl::TransferOp>(operation); |
| 66 | std::vector<adxl::TransferOpDesc> adxl_op_descs; | 66 | std::vector<adxl::TransferOpDesc> adxl_op_descs; |
| @@ -71,19 +71,19 @@ Status AdxlEngine::TransferAsync(const AscendString &remote_engine, TransferOp o | |||
| 71 | return adxl_inner_engine_.TransferAsync(remote_engine, adxl_operation, adxl_op_descs, adxl_optional_args, req); | 71 | return adxl_inner_engine_.TransferAsync(remote_engine, adxl_operation, adxl_op_descs, adxl_optional_args, req); |
| 72 | } | 72 | } |
| 73 | 73 | ||
| 74 | -Status AdxlEngine::GetTransferStatus(const TransferReq &req, TransferStatus &status) { | 74 | +Status CommEngine::GetTransferStatus(const TransferReq &req, TransferStatus &status) { |
| 75 | adxl::TransferStatus adxl_status; | 75 | adxl::TransferStatus adxl_status; |
| 76 | auto ret = adxl_inner_engine_.GetTransferStatus(req, adxl_status); | 76 | auto ret = adxl_inner_engine_.GetTransferStatus(req, adxl_status); |
| 77 | status = static_cast<hixl::TransferStatus>(adxl_status); | 77 | status = static_cast<hixl::TransferStatus>(adxl_status); |
| 78 | return ret; | 78 | return ret; |
| 79 | } | 79 | } |
| 80 | 80 | ||
| 81 | -Status AdxlEngine::SendNotify(const AscendString &remote_engine, const NotifyDesc ¬ify, int32_t timeout_in_millis) { | 81 | +Status CommEngine::SendNotify(const AscendString &remote_engine, const NotifyDesc ¬ify, int32_t timeout_in_millis) { |
| 82 | adxl::NotifyDesc adxl_notify{notify.name, notify.notify_msg}; | 82 | adxl::NotifyDesc adxl_notify{notify.name, notify.notify_msg}; |
| 83 | return adxl_inner_engine_.SendNotify(remote_engine, adxl_notify, timeout_in_millis); | 83 | return adxl_inner_engine_.SendNotify(remote_engine, adxl_notify, timeout_in_millis); |
| 84 | } | 84 | } |
| 85 | 85 | ||
| 86 | -Status AdxlEngine::GetNotifies(std::vector<NotifyDesc> ¬ifies) { | 86 | +Status CommEngine::GetNotifies(std::vector<NotifyDesc> ¬ifies) { |
| 87 | std::vector<adxl::NotifyDesc> adxl_notifies; | 87 | std::vector<adxl::NotifyDesc> adxl_notifies; |
| 88 | auto ret = adxl_inner_engine_.GetNotifies(adxl_notifies); | 88 | auto ret = adxl_inner_engine_.GetNotifies(adxl_notifies); |
| 89 | for (const auto &adxl_notify : adxl_notifies) { | 89 | for (const auto &adxl_notify : adxl_notifies) { |
| @@ -92,7 +92,7 @@ Status AdxlEngine::GetNotifies(std::vector<NotifyDesc> ¬ifies) { | |||
| 92 | return ret; | 92 | return ret; |
| 93 | } | 93 | } |
| 94 | 94 | ||
| 95 | -Status AdxlEngine::RegisterCallbackProcessor(int32_t msg_type, CallbackProcessor processor) { | 95 | +Status CommEngine::RegisterCallbackProcessor(int32_t msg_type, CallbackProcessor processor) { |
| 96 | return adxl_inner_engine_.RegisterCallbackProcessor(msg_type, processor); | 96 | return adxl_inner_engine_.RegisterCallbackProcessor(msg_type, processor); |
| 97 | } | 97 | } |
| 98 | } // namespace hixl | 98 | } // namespace hixl |
| @@ -17,11 +17,11 @@ | |||
| 17 | 17 | ||
| 18 | 18 | ||
| 19 | namespace hixl { | 19 | namespace hixl { |
| 20 | -class AdxlEngine : public Engine { | 20 | +class CommEngine : public Engine { |
| 21 | public: | 21 | public: |
| 22 | - explicit AdxlEngine(const AscendString &local_engine) : Engine(local_engine), adxl_inner_engine_(local_engine) {}; | 22 | + explicit CommEngine(const AscendString &local_engine) : Engine(local_engine), adxl_inner_engine_(local_engine) {}; |
| 23 | 23 | ||
| 24 | - ~AdxlEngine() override = default; | 24 | + ~CommEngine() override = default; |
| 25 | 25 | ||
| 26 | Status Initialize(const std::map<AscendString, AscendString> &options) override; | 26 | Status Initialize(const std::map<AscendString, AscendString> &options) override; |
| 27 | 27 | ||
| @@ -46,7 +46,7 @@ std::unique_ptr<Engine> EngineFactory::CreateEngine(const std::string local_engi | |||
| 46 | const auto hixl_it = options.find(hixl::OPTION_LOCAL_COMM_RES); | 46 | const auto hixl_it = options.find(hixl::OPTION_LOCAL_COMM_RES); |
| 47 | const auto adxl_it = options.find(adxl::OPTION_LOCAL_COMM_RES); | 47 | const auto adxl_it = options.find(adxl::OPTION_LOCAL_COMM_RES); |
| 48 | if ((hixl_it == options.end()) && (adxl_it == options.end())) { | 48 | if ((hixl_it == options.end()) && (adxl_it == options.end())) { |
| 49 | - return std::make_unique<AdxlEngine>(AscendString(local_engine.c_str())); | 49 | + return std::make_unique<CommEngine>(AscendString(local_engine.c_str())); |
| 50 | } | 50 | } |
| 51 | const auto &it = hixl_it == options.end() ? adxl_it : hixl_it; | 51 | const auto &it = hixl_it == options.end() ? adxl_it : hixl_it; |
| 52 | std::string local_comm_res = it->second.GetString(); | 52 | std::string local_comm_res = it->second.GetString(); |
| @@ -62,6 +62,6 @@ std::unique_ptr<Engine> EngineFactory::CreateEngine(const std::string local_engi | |||
| 62 | if (use_hixl) { | 62 | if (use_hixl) { |
| 63 | return std::make_unique<HixlEngine>(AscendString(local_engine.c_str())); | 63 | return std::make_unique<HixlEngine>(AscendString(local_engine.c_str())); |
| 64 | } | 64 | } |
| 65 | - return std::make_unique<AdxlEngine>(AscendString(local_engine.c_str())); | 65 | + return std::make_unique<CommEngine>(AscendString(local_engine.c_str())); |
| 66 | } | 66 | } |
| 67 | } // namespace hixl | 67 | } // namespace hixl |
| @@ -14,7 +14,7 @@ | |||
| 14 | 14 | ||
| 15 | 15 | ||
| 16 | 16 | ||
| 17 | -#include "adxl_engine.h" | 17 | +#include "comm_engine.h" |
| 18 | 18 | ||
| 19 | namespace hixl { | 19 | namespace hixl { |
| 20 | class EngineFactory { | 20 | class EngineFactory { |
| @@ -12,7 +12,7 @@ | |||
| 12 | 12 | ||
| 13 | 13 | ||
| 14 | 14 | ||
| 15 | -#include "adxl_engine.h" | 15 | +#include "comm_engine.h" |
| 16 | 16 | ||
| 17 | 17 | ||
| 18 | 18 | ||
| @@ -92,7 +92,7 @@ add_definitions(-D_GLIBCXX_USE_CXX11_ABI=0) | |||
| 92 | file(GLOB ADXL_SRC_LIST | 92 | file(GLOB ADXL_SRC_LIST |
| 93 | "${CMAKE_CURRENT_LIST_DIR}/adxl/*.cc" | 93 | "${CMAKE_CURRENT_LIST_DIR}/adxl/*.cc" |
| 94 | "${CMAKE_CURRENT_LIST_DIR}/llm_datadist_timer.cc" | 94 | "${CMAKE_CURRENT_LIST_DIR}/llm_datadist_timer.cc" |
| 95 | - "${CMAKE_CURRENT_LIST_DIR}/statistic_manager.cc" | 95 | + "${CMAKE_CURRENT_LIST_DIR}/comm_statistic_manager.cc" |
| 96 | "${CMAKE_CURRENT_LIST_DIR}/common/*.cc" | 96 | "${CMAKE_CURRENT_LIST_DIR}/common/*.cc" |
| 97 | "${CMAKE_CURRENT_LIST_DIR}/hccl/*.cc" | 97 | "${CMAKE_CURRENT_LIST_DIR}/hccl/*.cc" |
| 98 | "${CMAKE_CURRENT_LIST_DIR}/memory/*.cc" | 98 | "${CMAKE_CURRENT_LIST_DIR}/memory/*.cc" |
| @@ -16,7 +16,7 @@ | |||
| 16 | 16 | ||
| 17 | 17 | ||
| 18 | 18 | ||
| 19 | -#include "channel.h" | 19 | +#include "comm_channel.h" |
| 20 | 20 | ||
| 21 | 21 | ||
| 22 | namespace adxl { | 22 | namespace adxl { |
| @@ -427,7 +427,7 @@ Status ChannelManager::CreateChannel(const ChannelInfo &channel_info, ChannelPtr | |||
| 427 | "Channel already exists, channel_type = %d, channel id:%s", | 427 | "Channel already exists, channel_type = %d, channel id:%s", |
Z | |||
| 428 | static_cast<int32_t>(channel_info.channel_type), channel_info.channel_id.c_str()); | 428 | static_cast<int32_t>(channel_info.channel_type), channel_info.channel_id.c_str()); |
| 429 | } | 429 | } |
| 430 | - ChannelPtr channel = llm::MakeShared<Channel>(channel_info); | 430 | + ChannelPtr channel = llm::MakeShared<CommChannel>(channel_info); |
| 431 | ADXL_CHECK_NOTNULL(channel); | 431 | ADXL_CHECK_NOTNULL(channel); |
| 432 | ADXL_CHK_STATUS_RET(channel->Initialize(), "Failed to init channel"); | 432 | ADXL_CHK_STATUS_RET(channel->Initialize(), "Failed to init channel"); |
| 433 | channel->SetStreamPool(stream_pool_); | 433 | channel->SetStreamPool(stream_pool_); |
| @@ -18,7 +18,7 @@ | |||
| 18 | 18 | ||
| 19 | 19 | ||
| 20 | 20 | ||
| 21 | -#include "channel.h" | 21 | +#include "comm_channel.h" |
| 22 | 22 | ||
| 23 | 23 | ||
| 24 | 24 | ||
| @@ -16,7 +16,7 @@ | |||
| 16 | 16 | ||
| 17 | 17 | ||
| 18 | 18 | ||
| 19 | -#include "channel.h" | 19 | +#include "comm_channel.h" |
| 20 | 20 | ||
| 21 | 21 | ||
| 22 | 22 | ||
| @@ -8,7 +8,7 @@ | |||
| 8 | * See LICENSE in the root of the software repository for the full text of the License. | 8 | * See LICENSE in the root of the software repository for the full text of the License. |
| 9 | */ | 9 | */ |
| 10 | 10 | ||
| 11 | -#include "channel.h" | 11 | +#include "comm_channel.h" |
| 12 | 12 | ||
| 13 | 13 | ||
| 14 | 14 | ||
| @@ -36,9 +36,9 @@ uint64_t GetDurationUs(const std::chrono::steady_clock::time_point &start, | |||
| 36 | } | 36 | } |
| 37 | } // namespace | 37 | } // namespace |
| 38 | 38 | ||
| 39 | -int64_t Channel::timeout_in_millis_ = kHeartbeatTimeoutInMillis; | 39 | +int64_t CommChannel::timeout_in_millis_ = kHeartbeatTimeoutInMillis; |
| 40 | 40 | ||
| 41 | -Status Channel::Initialize() { | 41 | +Status CommChannel::Initialize() { |
| 42 | const auto hccl_start = std::chrono::steady_clock::now(); | 42 | const auto hccl_start = std::chrono::steady_clock::now(); |
| 43 | ADXL_CHK_STATUS_RET(InitializeHcclComm(), "Failed to initialize hccl comm"); | 43 | ADXL_CHK_STATUS_RET(InitializeHcclComm(), "Failed to initialize hccl comm"); |
| 44 | 44 | ||
| @@ -55,16 +55,16 @@ Status Channel::Initialize() { | |||
| 55 | return SUCCESS; | 55 | return SUCCESS; |
| 56 | } | 56 | } |
| 57 | 57 | ||
| 58 | -std::string Channel::GetChannelId() const { | 58 | +std::string CommChannel::GetChannelId() const { |
| 59 | return channel_info_.channel_id; | 59 | return channel_info_.channel_id; |
| 60 | } | 60 | } |
| 61 | 61 | ||
| 62 | -std::string Channel::GetStatisticChannelId() const { | 62 | +std::string CommChannel::GetStatisticChannelId() const { |
| 63 | return StatisticManager::GetStatisticChannelId(channel_info_.channel_id, | 63 | return StatisticManager::GetStatisticChannelId(channel_info_.channel_id, |
| 64 | channel_info_.channel_type == ChannelType::kClient); | 64 | channel_info_.channel_type == ChannelType::kClient); |
| 65 | } | 65 | } |
| 66 | 66 | ||
| 67 | -Status Channel::InitializeHcclComm() { | 67 | +Status CommChannel::InitializeHcclComm() { |
| 68 | LLMLOGI("HcclCommInitClusterInfoMemConfig begin, comm_name=%s, local rank_id=%u, rank_table=%s", | 68 | LLMLOGI("HcclCommInitClusterInfoMemConfig begin, comm_name=%s, local rank_id=%u, rank_table=%s", |
| 69 | channel_info_.comm_config.hcclCommName, channel_info_.local_rank_id, channel_info_.rank_table.c_str()); | 69 | channel_info_.comm_config.hcclCommName, channel_info_.local_rank_id, channel_info_.rank_table.c_str()); |
| 70 | std::lock_guard<std::mutex> lock(g_mutex_); | 70 | std::lock_guard<std::mutex> lock(g_mutex_); |
| @@ -78,7 +78,7 @@ Status Channel::InitializeHcclComm() { | |||
| 78 | return SUCCESS; | 78 | return SUCCESS; |
| 79 | } | 79 | } |
| 80 | 80 | ||
| 81 | -Status Channel::BindRegisteredMemory(std::vector<void *> &bind_handles) { | 81 | +Status CommChannel::BindRegisteredMemory(std::vector<void *> &bind_handles) { |
| 82 | const auto start = std::chrono::steady_clock::now(); | 82 | const auto start = std::chrono::steady_clock::now(); |
| 83 | for (const auto ®_handle_it : channel_info_.registered_mems) { | 83 | for (const auto ®_handle_it : channel_info_.registered_mems) { |
| 84 | auto reg_handle = reg_handle_it.first; | 84 | auto reg_handle = reg_handle_it.first; |
| @@ -91,7 +91,7 @@ Status Channel::BindRegisteredMemory(std::vector<void *> &bind_handles) { | |||
| 91 | return SUCCESS; | 91 | return SUCCESS; |
| 92 | } | 92 | } |
| 93 | 93 | ||
| 94 | -Status Channel::PrepareHcclComm(const std::chrono::steady_clock::time_point &hccl_start) { | 94 | +Status CommChannel::PrepareHcclComm(const std::chrono::steady_clock::time_point &hccl_start) { |
| 95 | const auto start = std::chrono::steady_clock::now(); | 95 | const auto start = std::chrono::steady_clock::now(); |
| 96 | HcclPrepareConfig prepareConfig{}; | 96 | HcclPrepareConfig prepareConfig{}; |
| 97 | ADXL_CHK_HCCL_RET( | 97 | ADXL_CHK_HCCL_RET( |
| @@ -104,14 +104,14 @@ Status Channel::PrepareHcclComm(const std::chrono::steady_clock::time_point &hcc | |||
| 104 | return SUCCESS; | 104 | return SUCCESS; |
| 105 | } | 105 | } |
| 106 | 106 | ||
| 107 | -void Channel::ClearNotifyMessages() { | 107 | +void CommChannel::ClearNotifyMessages() { |
| 108 | { | 108 | { |
| 109 | std::lock_guard<std::mutex> notify_lock(notify_message_mutex_); | 109 | std::lock_guard<std::mutex> notify_lock(notify_message_mutex_); |
| 110 | notify_messages_.clear(); | 110 | notify_messages_.clear(); |
| 111 | } | 111 | } |
| 112 | } | 112 | } |
| 113 | 113 | ||
| 114 | -Status Channel::Finalize() { | 114 | +Status CommChannel::Finalize() { |
| 115 | finalized_.store(true, std::memory_order_release); | 115 | finalized_.store(true, std::memory_order_release); |
| 116 | ADXL_CHK_STATUS_RET(ClearResources(), "Failed to clear channel resources."); | 116 | ADXL_CHK_STATUS_RET(ClearResources(), "Failed to clear channel resources."); |
| 117 | 117 | ||
| @@ -130,7 +130,7 @@ Status Channel::Finalize() { | |||
| 130 | return SUCCESS; | 130 | return SUCCESS; |
| 131 | } | 131 | } |
| 132 | 132 | ||
| 133 | -Status Channel::ClearResources() { | 133 | +Status CommChannel::ClearResources() { |
| 134 | std::lock_guard<std::mutex> lock(transfer_mutex_); | 134 | std::lock_guard<std::mutex> lock(transfer_mutex_); |
| 135 | auto ret = SUCCESS; | 135 | auto ret = SUCCESS; |
| 136 | for (const auto ®_handle_it : channel_info_.registered_mems) { | 136 | for (const auto ®_handle_it : channel_info_.registered_mems) { |
| @@ -167,11 +167,11 @@ Status Channel::ClearResources() { | |||
| 167 | return ret; | 167 | return ret; |
| 168 | } | 168 | } |
| 169 | 169 | ||
| 170 | -void Channel::SetStreamPool(StreamPool *stream_pool) { | 170 | +void CommChannel::SetStreamPool(StreamPool *stream_pool) { |
| 171 | stream_pool_ = stream_pool; | 171 | stream_pool_ = stream_pool; |
| 172 | } | 172 | } |
| 173 | 173 | ||
| 174 | -Status Channel::TransferAsync(TransferOp operation, const std::vector<TransferOpDesc> &op_descs, | 174 | +Status CommChannel::TransferAsync(TransferOp operation, const std::vector<TransferOpDesc> &op_descs, |
| 175 | const TransferArgs &optional_args, TransferReq &req) { | 175 | const TransferArgs &optional_args, TransferReq &req) { |
| 176 | (void)optional_args; | 176 | (void)optional_args; |
| 177 | aclrtStream stream = nullptr; | 177 | aclrtStream stream = nullptr; |
| @@ -200,7 +200,7 @@ Status Channel::TransferAsync(TransferOp operation, const std::vector<TransferOp | |||
| 200 | return SUCCESS; | 200 | return SUCCESS; |
| 201 | } | 201 | } |
| 202 | 202 | ||
| 203 | -Status Channel::GetTransferStatus(const TransferReq &req, TransferStatus &status) { | 203 | +Status CommChannel::GetTransferStatus(const TransferReq &req, TransferStatus &status) { |
| 204 | std::lock_guard<std::mutex> lock(transfer_reqs_mutex_); | 204 | std::lock_guard<std::mutex> lock(transfer_reqs_mutex_); |
| 205 | auto id = reinterpret_cast<uint64_t>(req); | 205 | auto id = reinterpret_cast<uint64_t>(req); |
| 206 | auto it = req_2_async_record_.find(id); | 206 | auto it = req_2_async_record_.find(id); |
| @@ -251,7 +251,7 @@ Status Channel::GetTransferStatus(const TransferReq &req, TransferStatus &status | |||
| 251 | return SUCCESS; | 251 | return SUCCESS; |
| 252 | } | 252 | } |
| 253 | 253 | ||
| 254 | -Status Channel::TransferAsync(TransferOp operation, const std::vector<TransferOpDesc> &op_descs, aclrtStream stream) { | 254 | +Status CommChannel::TransferAsync(TransferOp operation, const std::vector<TransferOpDesc> &op_descs, aclrtStream stream) { |
| 255 | ADXL_CHK_BOOL_RET_STATUS(channel_info_.comm != nullptr, FAILED, | 255 | ADXL_CHK_BOOL_RET_STATUS(channel_info_.comm != nullptr, FAILED, |
| 256 | "Channel comm is null, channel may have been finalized, channel_id:%s.", | 256 | "Channel comm is null, channel may have been finalized, channel_id:%s.", |
| 257 | channel_info_.channel_id.c_str()); | 257 | channel_info_.channel_id.c_str()); |
| @@ -273,7 +273,7 @@ Status Channel::TransferAsync(TransferOp operation, const std::vector<TransferOp | |||
| 273 | return SUCCESS; | 273 | return SUCCESS; |
| 274 | } | 274 | } |
| 275 | 275 | ||
| 276 | -Status Channel::TransferAsyncWithTimeout(TransferOp operation, const std::vector<TransferOpDesc> &op_descs, | 276 | +Status CommChannel::TransferAsyncWithTimeout(TransferOp operation, const std::vector<TransferOpDesc> &op_descs, |
| 277 | aclrtStream stream, uint64_t timeout) { | 277 | aclrtStream stream, uint64_t timeout) { |
| 278 | ADXL_CHK_BOOL_RET_STATUS(channel_info_.comm != nullptr, FAILED, | 278 | ADXL_CHK_BOOL_RET_STATUS(channel_info_.comm != nullptr, FAILED, |
| 279 | "Channel comm is null, channel may have been finalized, channel_id:%s.", | 279 | "Channel comm is null, channel may have been finalized, channel_id:%s.", |
| @@ -309,7 +309,7 @@ Status Channel::TransferAsyncWithTimeout(TransferOp operation, const std::vector | |||
| 309 | return SUCCESS; | 309 | return SUCCESS; |
| 310 | } | 310 | } |
| 311 | 311 | ||
| 312 | -Status Channel::TransferSync(TransferOp operation, const std::vector<TransferOpDesc> &op_descs, | 312 | +Status CommChannel::TransferSync(TransferOp operation, const std::vector<TransferOpDesc> &op_descs, |
| 313 | int32_t timeout_in_millis) { | 313 | int32_t timeout_in_millis) { |
| 314 | const auto start = std::chrono::steady_clock::now(); | 314 | const auto start = std::chrono::steady_clock::now(); |
| 315 | aclrtStream stream = nullptr; | 315 | aclrtStream stream = nullptr; |
| @@ -333,7 +333,7 @@ Status Channel::TransferSync(TransferOp operation, const std::vector<TransferOpD | |||
| 333 | return SUCCESS; | 333 | return SUCCESS; |
| 334 | } | 334 | } |
| 335 | 335 | ||
| 336 | -Status Channel::SetSocketNonBlocking(int32_t fd) { | 336 | +Status CommChannel::SetSocketNonBlocking(int32_t fd) { |
| 337 | std::lock_guard<std::mutex> lock(mutex_); | 337 | std::lock_guard<std::mutex> lock(mutex_); |
| 338 | int flags = fcntl(fd, F_GETFL, 0); | 338 | int flags = fcntl(fd, F_GETFL, 0); |
| 339 | ADXL_CHK_BOOL_RET_STATUS(flags != -1, FAILED, "Failed to get fd flags: %s", strerror(errno)); | 339 | ADXL_CHK_BOOL_RET_STATUS(flags != -1, FAILED, "Failed to get fd flags: %s", strerror(errno)); |
| @@ -347,13 +347,13 @@ Status Channel::SetSocketNonBlocking(int32_t fd) { | |||
| 347 | return SUCCESS; | 347 | return SUCCESS; |
| 348 | } | 348 | } |
| 349 | 349 | ||
| 350 | -void Channel::StopHeartbeat() { | 350 | +void CommChannel::StopHeartbeat() { |
| 351 | std::lock_guard<std::mutex> lock(mutex_); | 351 | std::lock_guard<std::mutex> lock(mutex_); |
| 352 | with_heartbeat_.store(false, std::memory_order_release); | 352 | with_heartbeat_.store(false, std::memory_order_release); |
| 353 | disconnect_flag_.store(true, std::memory_order_release); | 353 | disconnect_flag_.store(true, std::memory_order_release); |
| 354 | } | 354 | } |
| 355 | 355 | ||
| 356 | -Status Channel::CommWithFd(const std::function<Status(int32_t)> &func) { | 356 | +Status CommChannel::CommWithFd(const std::function<Status(int32_t)> &func) { |
| 357 | std::lock_guard<std::mutex> lock(mutex_); | 357 | std::lock_guard<std::mutex> lock(mutex_); |
| 358 | if (fd_ < 0) { | 358 | if (fd_ < 0) { |
| 359 | return FAILED; | 359 | return FAILED; |
| @@ -361,26 +361,26 @@ Status Channel::CommWithFd(const std::function<Status(int32_t)> &func) { | |||
| 361 | return func(fd_); | 361 | return func(fd_); |
| 362 | } | 362 | } |
| 363 | 363 | ||
| 364 | -Status Channel::SendControlMsg(const std::function<Status(int32_t)> &func) { | 364 | +Status CommChannel::SendControlMsg(const std::function<Status(int32_t)> &func) { |
| 365 | return CommWithFd(func); | 365 | return CommWithFd(func); |
| 366 | } | 366 | } |
| 367 | 367 | ||
| 368 | -Status Channel::SendHeartBeat(const std::function<Status(int32_t)> &func) { | 368 | +Status CommChannel::SendHeartBeat(const std::function<Status(int32_t)> &func) { |
| 369 | if (with_heartbeat_.load(std::memory_order_acquire)) { | 369 | if (with_heartbeat_.load(std::memory_order_acquire)) { |
| 370 | return CommWithFd(func); | 370 | return CommWithFd(func); |
| 371 | } | 371 | } |
| 372 | return SUCCESS; | 372 | return SUCCESS; |
| 373 | } | 373 | } |
| 374 | 374 | ||
| 375 | -void Channel::SetHeartbeatTimeout(int64_t timeout_in_millis) { | 375 | +void CommChannel::SetHeartbeatTimeout(int64_t timeout_in_millis) { |
| 376 | timeout_in_millis_ = timeout_in_millis; | 376 | timeout_in_millis_ = timeout_in_millis; |
| 377 | } | 377 | } |
| 378 | 378 | ||
| 379 | -void Channel::UpdateHeartbeatTime() { | 379 | +void CommChannel::UpdateHeartbeatTime() { |
| 380 | last_heartbeat_time_ = std::chrono::steady_clock::now(); | 380 | last_heartbeat_time_ = std::chrono::steady_clock::now(); |
| 381 | } | 381 | } |
| 382 | 382 | ||
| 383 | -bool Channel::IsHeartbeatTimeout() const { | 383 | +bool CommChannel::IsHeartbeatTimeout() const { |
| 384 | if (with_heartbeat_.load(std::memory_order_acquire)) { | 384 | if (with_heartbeat_.load(std::memory_order_acquire)) { |
| 385 | auto now = std::chrono::steady_clock::now(); | 385 | auto now = std::chrono::steady_clock::now(); |
| 386 | const auto cost = std::chrono::duration_cast<std::chrono::milliseconds>(now - last_heartbeat_time_).count(); | 386 | const auto cost = std::chrono::duration_cast<std::chrono::milliseconds>(now - last_heartbeat_time_).count(); |
| @@ -392,15 +392,15 @@ bool Channel::IsHeartbeatTimeout() const { | |||
| 392 | return false; | 392 | return false; |
| 393 | } | 393 | } |
| 394 | 394 | ||
| 395 | -StreamPool *Channel::GetStreamPool() { | 395 | +StreamPool *CommChannel::GetStreamPool() { |
| 396 | return stream_pool_; | 396 | return stream_pool_; |
| 397 | } | 397 | } |
| 398 | 398 | ||
| 399 | -std::mutex &Channel::GetTransferMutex() { | 399 | +std::mutex &CommChannel::GetTransferMutex() { |
| 400 | return transfer_mutex_; | 400 | return transfer_mutex_; |
| 401 | } | 401 | } |
| 402 | 402 | ||
| 403 | -void Channel::GetNotifyMessages(std::vector<NotifyDesc> ¬ifies) { | 403 | +void CommChannel::GetNotifyMessages(std::vector<NotifyDesc> ¬ifies) { |
| 404 | std::lock_guard<std::mutex> lock(notify_message_mutex_); | 404 | std::lock_guard<std::mutex> lock(notify_message_mutex_); |
| 405 | for (auto ¬ify_msg : notify_messages_) { | 405 | for (auto ¬ify_msg : notify_messages_) { |
| 406 | NotifyDesc notify; | 406 | NotifyDesc notify; |
| @@ -62,9 +62,9 @@ struct AsyncRecord { | |||
| 62 | 62 | ||
| 63 | enum class RecvState { WAITING_FOR_HEADER, WAITING_FOR_BODY }; | 63 | enum class RecvState { WAITING_FOR_HEADER, WAITING_FOR_BODY }; |
| 64 | 64 | ||
| 65 | -class Channel { | 65 | +class CommChannel { |
| 66 | public: | 66 | public: |
| 67 | - explicit Channel(ChannelInfo info) : channel_info_(std::move(info)){}; | 67 | + explicit CommChannel(ChannelInfo info) : channel_info_(std::move(info)){}; |
| 68 | Status Initialize(); | 68 | Status Initialize(); |
| 69 | Status Finalize(); | 69 | Status Finalize(); |
| 70 | std::string GetChannelId() const; | 70 | std::string GetChannelId() const; |
| @@ -155,7 +155,7 @@ class Channel { | |||
| 155 | std::unordered_map<uint64_t, AsyncRecord> req_2_async_record_; | 155 | std::unordered_map<uint64_t, AsyncRecord> req_2_async_record_; |
| 156 | StreamPool *stream_pool_ = nullptr; | 156 | StreamPool *stream_pool_ = nullptr; |
| 157 | }; | 157 | }; |
| 158 | -using ChannelPtr = std::shared_ptr<Channel>; | 158 | +using ChannelPtr = std::shared_ptr<CommChannel>; |
| 159 | } // namespace adxl | 159 | } // namespace adxl |
| 160 | 160 | ||
| 161 | 161 | ||
| @@ -11,7 +11,7 @@ | |||
| 11 | 11 | ||
| 12 | 12 | ||
| 13 | 13 | ||
| 14 | -#include "statistic_manager.h" | 14 | +#include "comm_statistic_manager.h" |
| 15 | 15 | ||
| 16 | 16 | ||
| 17 | 17 | ||
| @@ -9,13 +9,13 @@ | |||
| 9 | */ | 9 | */ |
| 10 | 10 | ||
| 11 | 11 | ||
| 12 | -#include "statistic_manager.h" | 12 | +#include "comm_statistic_manager.h" |
| 13 | namespace llm { | 13 | namespace llm { |
| 14 | -StatisticManager &StatisticManager::GetInstance() { | 14 | +CommStatisticManager &CommStatisticManager::GetInstance() { |
| 15 | - static StatisticManager instance; | 15 | + static CommStatisticManager instance; |
| 16 | return instance; | 16 | return instance; |
| 17 | } | 17 | } |
| 18 | -void StatisticManager::UpdateCost(const uint64_t cost, uint64_t &total_times, uint64_t &min_cost, uint64_t &max_cost, | 18 | +void CommStatisticManager::UpdateCost(const uint64_t cost, uint64_t &total_times, uint64_t &min_cost, uint64_t &max_cost, |
| 19 | uint64_t &total_cost) { | 19 | uint64_t &total_cost) { |
| 20 | total_times++; | 20 | total_times++; |
| 21 | total_cost += cost; | 21 | total_cost += cost; |
| @@ -23,7 +23,7 @@ void StatisticManager::UpdateCost(const uint64_t cost, uint64_t &total_times, ui | |||
| 23 | min_cost = (min_cost > cost) ? cost : min_cost; | 23 | min_cost = (min_cost > cost) ? cost : min_cost; |
| 24 | } | 24 | } |
| 25 | 25 | ||
| 26 | -void StatisticManager::UpdateCost(const uint64_t cost, std::atomic<uint64_t> &total_times, | 26 | +void CommStatisticManager::UpdateCost(const uint64_t cost, std::atomic<uint64_t> &total_times, |
| 27 | std::atomic<uint64_t> &min_cost, std::atomic<uint64_t> &max_cost, | 27 | std::atomic<uint64_t> &min_cost, std::atomic<uint64_t> &max_cost, |
| 28 | std::atomic<uint64_t> &total_cost) { | 28 | std::atomic<uint64_t> &total_cost) { |
| 29 | (void)total_times.fetch_add(1U); | 29 | (void)total_times.fetch_add(1U); |
| @@ -36,68 +36,68 @@ void StatisticManager::UpdateCost(const uint64_t cost, std::atomic<uint64_t> &to | |||
| 36 | } | 36 | } |
| 37 | } | 37 | } |
| 38 | 38 | ||
| 39 | -void StatisticManager::AddExchangeMemCost(const uint64_t cost) { | 39 | +void CommStatisticManager::AddExchangeMemCost(const uint64_t cost) { |
| 40 | UpdateCost(cost, link_statistic_info_.exchange_mem_times, link_statistic_info_.exchange_mem_min_cost, | 40 | UpdateCost(cost, link_statistic_info_.exchange_mem_times, link_statistic_info_.exchange_mem_min_cost, |
| 41 | link_statistic_info_.exchange_mem_max_cost, link_statistic_info_.exchange_mem_total_cost); | 41 | link_statistic_info_.exchange_mem_max_cost, link_statistic_info_.exchange_mem_total_cost); |
| 42 | } | 42 | } |
| 43 | 43 | ||
| 44 | -void StatisticManager::AddCommInitCost(const uint64_t cost) { | 44 | +void CommStatisticManager::AddCommInitCost(const uint64_t cost) { |
| 45 | UpdateCost(cost, link_statistic_info_.comm_init_times, link_statistic_info_.comm_init_min_cost, | 45 | UpdateCost(cost, link_statistic_info_.comm_init_times, link_statistic_info_.comm_init_min_cost, |
| 46 | link_statistic_info_.comm_init_max_cost, link_statistic_info_.comm_init_total_cost); | 46 | link_statistic_info_.comm_init_max_cost, link_statistic_info_.comm_init_total_cost); |
| 47 | } | 47 | } |
| 48 | 48 | ||
| 49 | -void StatisticManager::AddCommDestroyCost(const uint64_t cost) { | 49 | +void CommStatisticManager::AddCommDestroyCost(const uint64_t cost) { |
| 50 | UpdateCost(cost, link_statistic_info_.comm_destroy_times, link_statistic_info_.comm_destroy_min_cost, | 50 | UpdateCost(cost, link_statistic_info_.comm_destroy_times, link_statistic_info_.comm_destroy_min_cost, |
| 51 | link_statistic_info_.comm_destroy_max_cost, link_statistic_info_.comm_destroy_total_cost); | 51 | link_statistic_info_.comm_destroy_max_cost, link_statistic_info_.comm_destroy_total_cost); |
| 52 | } | 52 | } |
| 53 | 53 | ||
| 54 | -void StatisticManager::AddRegisterGlobalMemTimes() { | 54 | +void CommStatisticManager::AddRegisterGlobalMemTimes() { |
| 55 | link_statistic_info_.register_global_mem_times++; | 55 | link_statistic_info_.register_global_mem_times++; |
| 56 | } | 56 | } |
| 57 | 57 | ||
| 58 | -void StatisticManager::AddDeregisterGlobalMemTimes() { | 58 | +void CommStatisticManager::AddDeregisterGlobalMemTimes() { |
| 59 | link_statistic_info_.deregister_global_mem_times++; | 59 | link_statistic_info_.deregister_global_mem_times++; |
| 60 | } | 60 | } |
| 61 | 61 | ||
| 62 | -void StatisticManager::AddCommBindMemTimes() { | 62 | +void CommStatisticManager::AddCommBindMemTimes() { |
| 63 | link_statistic_info_.comm_bind_mem_times++; | 63 | link_statistic_info_.comm_bind_mem_times++; |
| 64 | } | 64 | } |
| 65 | 65 | ||
| 66 | -void StatisticManager::AddCommUnbindMemTimes() { | 66 | +void CommStatisticManager::AddCommUnbindMemTimes() { |
| 67 | link_statistic_info_.comm_unbind_mem_times++; | 67 | link_statistic_info_.comm_unbind_mem_times++; |
| 68 | } | 68 | } |
| 69 | 69 | ||
| 70 | -void StatisticManager::AddCommPrepareCost(const uint64_t cost) { | 70 | +void CommStatisticManager::AddCommPrepareCost(const uint64_t cost) { |
| 71 | UpdateCost(cost, link_statistic_info_.comm_prepare_times, link_statistic_info_.comm_prepare_min_cost, | 71 | UpdateCost(cost, link_statistic_info_.comm_prepare_times, link_statistic_info_.comm_prepare_min_cost, |
| 72 | link_statistic_info_.comm_prepare_max_cost, link_statistic_info_.comm_prepare_total_cost); | 72 | link_statistic_info_.comm_prepare_max_cost, link_statistic_info_.comm_prepare_total_cost); |
| 73 | } | 73 | } |
| 74 | 74 | ||
| 75 | -void StatisticManager::AddBatchPutCost(const uint64_t cost) { | 75 | +void CommStatisticManager::AddBatchPutCost(const uint64_t cost) { |
| 76 | send_statistic_info_.batch_put_times++; | 76 | send_statistic_info_.batch_put_times++; |
| 77 | send_statistic_info_.batch_put_total_cost += cost; | 77 | send_statistic_info_.batch_put_total_cost += cost; |
| 78 | } | 78 | } |
| 79 | 79 | ||
| 80 | -MemoryStatisticInfo &StatisticManager::GetMemoryStatisticInfo() { | 80 | +MemoryStatisticInfo &CommStatisticManager::GetMemoryStatisticInfo() { |
| 81 | return memory_statistic_info_; | 81 | return memory_statistic_info_; |
| 82 | } | 82 | } |
| 83 | -FuncStatisticInfo &StatisticManager::GetFuncStatisticInfo() { | 83 | +FuncStatisticInfo &CommStatisticManager::GetFuncStatisticInfo() { |
| 84 | return func_statistic_info_; | 84 | return func_statistic_info_; |
| 85 | } | 85 | } |
| 86 | 86 | ||
| 87 | -void StatisticManager::Dump() const{ | 87 | +void CommStatisticManager::Dump() const{ |
| 88 | DumpMemoryProfilingTrack(); | 88 | DumpMemoryProfilingTrack(); |
| 89 | DumpFuncProfilingTrack(); | 89 | DumpFuncProfilingTrack(); |
| 90 | DumpLinkProfilingTrack(); | 90 | DumpLinkProfilingTrack(); |
| 91 | DumpUnLinkProfilingTrack(); | 91 | DumpUnLinkProfilingTrack(); |
| 92 | } | 92 | } |
| 93 | 93 | ||
| 94 | -void StatisticManager::DumpMemoryProfilingTrack() const { | 94 | +void CommStatisticManager::DumpMemoryProfilingTrack() const { |
| 95 | LLMEVENT("Memory statistic info:alloc mem:%lu, free mem:%lu, alloc times:%lu, free times%:lu", | 95 | LLMEVENT("Memory statistic info:alloc mem:%lu, free mem:%lu, alloc times:%lu, free times%:lu", |
| 96 | memory_statistic_info_.alloc_mem, memory_statistic_info_.free_mem, memory_statistic_info_.alloc_times, | 96 | memory_statistic_info_.alloc_mem, memory_statistic_info_.free_mem, memory_statistic_info_.alloc_times, |
| 97 | memory_statistic_info_.free_times); | 97 | memory_statistic_info_.free_times); |
| 98 | } | 98 | } |
| 99 | 99 | ||
| 100 | -void StatisticManager::DumpFuncProfilingTrack() const { | 100 | +void CommStatisticManager::DumpFuncProfilingTrack() const { |
| 101 | const uint64_t link_func_avg_cost = | 101 | const uint64_t link_func_avg_cost = |
| 102 | func_statistic_info_.link_func_times == 0U | 102 | func_statistic_info_.link_func_times == 0U |
| 103 | ? 0U | 103 | ? 0U |
| @@ -143,7 +143,7 @@ void StatisticManager::DumpFuncProfilingTrack() const { | |||
| 143 | func_statistic_info_.transfer_func_min_cost.load(), transfer_func_avg_cost); | 143 | func_statistic_info_.transfer_func_min_cost.load(), transfer_func_avg_cost); |
| 144 | } | 144 | } |
| 145 | 145 | ||
| 146 | -void StatisticManager::DumpLinkProfilingTrack() const { | 146 | +void CommStatisticManager::DumpLinkProfilingTrack() const { |
| 147 | const uint64_t comm_init_avg_cost = | 147 | const uint64_t comm_init_avg_cost = |
| 148 | link_statistic_info_.comm_init_times == 0U | 148 | link_statistic_info_.comm_init_times == 0U |
| 149 | ? 0U | 149 | ? 0U |
| @@ -172,7 +172,7 @@ void StatisticManager::DumpLinkProfilingTrack() const { | |||
| 172 | ); | 172 | ); |
| 173 | } | 173 | } |
| 174 | 174 | ||
| 175 | -void StatisticManager::DumpUnLinkProfilingTrack() const { | 175 | +void CommStatisticManager::DumpUnLinkProfilingTrack() const { |
| 176 | const uint64_t comm_destroy_avg_cost = | 176 | const uint64_t comm_destroy_avg_cost = |
| 177 | link_statistic_info_.comm_destroy_times == 0U | 177 | link_statistic_info_.comm_destroy_times == 0U |
| 178 | ? 0U | 178 | ? 0U |
| @@ -187,7 +187,7 @@ void StatisticManager::DumpUnLinkProfilingTrack() const { | |||
| 187 | ); | 187 | ); |
| 188 | } | 188 | } |
| 189 | 189 | ||
| 190 | -void StatisticManager::Reset() { | 190 | +void CommStatisticManager::Reset() { |
| 191 | link_statistic_info_.Reset(); | 191 | link_statistic_info_.Reset(); |
| 192 | send_statistic_info_.Reset(); | 192 | send_statistic_info_.Reset(); |
| 193 | recv_statistic_info_.Reset(); | 193 | recv_statistic_info_.Reset(); |
| @@ -215,18 +215,18 @@ struct FuncStatisticInfo { | |||
| 215 | } | 215 | } |
| 216 | }; | 216 | }; |
| 217 | 217 | ||
| 218 | -class StatisticManager { | 218 | +class CommStatisticManager { |
| 219 | public: | 219 | public: |
| 220 | - static StatisticManager &GetInstance(); | 220 | + static CommStatisticManager &GetInstance(); |
| 221 | static void UpdateCost(const uint64_t cost, uint64_t &total_times, uint64_t &min_cost, uint64_t &max_cost, | 221 | static void UpdateCost(const uint64_t cost, uint64_t &total_times, uint64_t &min_cost, uint64_t &max_cost, |
| 222 | uint64_t &total_cost); | 222 | uint64_t &total_cost); |
| 223 | static void UpdateCost(const uint64_t cost, std::atomic<uint64_t> &total_times, std::atomic<uint64_t> &min_cost, | 223 | static void UpdateCost(const uint64_t cost, std::atomic<uint64_t> &total_times, std::atomic<uint64_t> &min_cost, |
| 224 | std::atomic<uint64_t> &max_cost, std::atomic<uint64_t> &total_cost); | 224 | std::atomic<uint64_t> &max_cost, std::atomic<uint64_t> &total_cost); |
| 225 | - ~StatisticManager() = default; | 225 | + ~CommStatisticManager() = default; |
| 226 | - StatisticManager(const StatisticManager &) = delete; | 226 | + CommStatisticManager(const CommStatisticManager &) = delete; |
| 227 | - StatisticManager(const StatisticManager &&) = delete; | 227 | + CommStatisticManager(const CommStatisticManager &&) = delete; |
| 228 | - StatisticManager &operator=(const StatisticManager &) = delete; | 228 | + CommStatisticManager &operator=(const CommStatisticManager &) = delete; |
| 229 | - StatisticManager &operator=(const StatisticManager &&) = delete; | 229 | + CommStatisticManager &operator=(const CommStatisticManager &&) = delete; |
| 230 | void Dump() const; | 230 | void Dump() const; |
| 231 | void Reset(); | 231 | void Reset(); |
| 232 | 232 | ||
| @@ -243,7 +243,7 @@ class StatisticManager { | |||
| 243 | FuncStatisticInfo &GetFuncStatisticInfo(); | 243 | FuncStatisticInfo &GetFuncStatisticInfo(); |
| 244 | 244 | ||
| 245 | private: | 245 | private: |
| 246 | - StatisticManager() = default; | 246 | + CommStatisticManager() = default; |
| 247 | void DumpMemoryProfilingTrack() const; | 247 | void DumpMemoryProfilingTrack() const; |
| 248 | void DumpFuncProfilingTrack() const; | 248 | void DumpFuncProfilingTrack() const; |
| 249 | void DumpLinkProfilingTrack() const; | 249 | void DumpLinkProfilingTrack() const; |
| @@ -1,52 +0,0 @@ | |||
| 1 | -/** | ||
| 2 | - * Copyright (c) 2025 Huawei Technologies Co., Ltd. | ||
| 3 | - * This program is free software, you can redistribute it and/or modify it under the terms and conditions of | ||
| 4 | - * CANN Open Software License Agreement Version 2.0 (the "License"). | ||
| 5 | - * Please refer to the License for details. You may not use this file except in compliance with the License. | ||
| 6 | - * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, | ||
| 7 | - * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. | ||
| 8 | - * See LICENSE in the root of the software repository for the full text of the License. | ||
| 9 | - | ||
| 10 | - * The code snippet comes from Huawei's open-source Ascend project. | ||
| 11 | - * Copyright 2020 Huawei Technologies Co., Ltd. | ||
| 12 | - * Licensed under the Apache License, Version 2.0 (the "License"); | ||
| 13 | - * You may obtain a copy of the License at | ||
| 14 | - * | ||
| 15 | - * http://www.apache.org/licenses/LICENSE-2.0 | ||
| 16 | - */ | ||
| 17 | - | ||
| 18 | - | ||
| 19 | - | ||
| 20 | - | ||
| 21 | - | ||
| 22 | - | ||
| 23 | -namespace llm { | ||
| 24 | -AlignedPtr::AlignedPtr(const size_t buffer_size, const size_t alignment) { | ||
| 25 | - size_t alloc_size = buffer_size; | ||
| 26 | - if (alignment > 0U) { | ||
| 27 | - alloc_size = buffer_size + alignment - 1U; | ||
| 28 | - } | ||
| 29 | - if ((buffer_size == 0U) || (alloc_size < buffer_size)) { | ||
| 30 | - LLMLOGW("[Allocate][Buffer] Allocate empty buffer or overflow, size=%zu, alloc_size=%zu", buffer_size, alloc_size); | ||
| 31 | - return; | ||
| 32 | - } | ||
| 33 | - | ||
| 34 | - base_ = | ||
| 35 | - std::unique_ptr<uint8_t[], AlignedPtr::Deleter>(new (std::nothrow) uint8_t[alloc_size], [](const uint8_t *ptr) { | ||
| 36 | - delete[] ptr; | ||
| 37 | - ptr = nullptr; | ||
| 38 | - }); | ||
| 39 | - if (base_ == nullptr) { | ||
| 40 | - LLMLOGW("[Allocate][Buffer] Allocate buffer failed, size=%zu", alloc_size); | ||
| 41 | - return; | ||
| 42 | - } | ||
| 43 | - | ||
| 44 | - if (alignment == 0U) { | ||
| 45 | - aligned_addr_ = base_.get(); | ||
| 46 | - } else { | ||
| 47 | - const size_t offset = alignment - 1U; | ||
| 48 | - aligned_addr_ = | ||
| 49 | - PtrToPtr<void, uint8_t>(ValueToPtr((PtrToValue(PtrToPtr<uint8_t, void>(base_.get())) + offset) & ~offset)); | ||
| 50 | - } | ||
| 51 | -} | ||
| 52 | -} // namespace llm | ||
| @@ -1,45 +0,0 @@ | |||
| 1 | -/** | ||
| 2 | - * Copyright (c) 2025 Huawei Technologies Co., Ltd. | ||
| 3 | - * This program is free software, you can redistribute it and/or modify it under the terms and conditions of | ||
| 4 | - * CANN Open Software License Agreement Version 2.0 (the "License"). | ||
| 5 | - * Please refer to the License for details. You may not use this file except in compliance with the License. | ||
| 6 | - * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, | ||
| 7 | - * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. | ||
| 8 | - * See LICENSE in the root of the software repository for the full text of the License. | ||
| 9 | - | ||
| 10 | - * The code snippet comes from Huawei's open-source Ascend project. | ||
| 11 | - * Copyright 2020 Huawei Technologies Co., Ltd. | ||
| 12 | - * Licensed under the Apache License, Version 2.0 (the "License"); | ||
| 13 | - * You may obtain a copy of the License at | ||
| 14 | - * | ||
| 15 | - * http://www.apache.org/licenses/LICENSE-2.0 | ||
| 16 | - */ | ||
| 17 | - | ||
| 18 | - | ||
| 19 | - | ||
| 20 | - | ||
| 21 | - | ||
| 22 | - | ||
| 23 | - | ||
| 24 | -namespace llm { | ||
| 25 | -class AlignedPtr { | ||
| 26 | - public: | ||
| 27 | - using Deleter = std::function<void(uint8_t *)>; | ||
| 28 | - using Allocator = std::function<void(std::unique_ptr<uint8_t[], Deleter> &base_addr)>; | ||
| 29 | - explicit AlignedPtr(const size_t buffer_size, const size_t alignment = 16U); | ||
| 30 | - AlignedPtr() = default; | ||
| 31 | - ~AlignedPtr() = default; | ||
| 32 | - AlignedPtr(const AlignedPtr &) = delete; | ||
| 33 | - AlignedPtr(AlignedPtr &&) = delete; | ||
| 34 | - AlignedPtr &operator=(const AlignedPtr &) = delete; | ||
| 35 | - AlignedPtr &operator=(AlignedPtr &&) = delete; | ||
| 36 | - | ||
| 37 | - const uint8_t *Get() const { return aligned_addr_; } | ||
| 38 | - uint8_t *MutableGet() { return aligned_addr_; } | ||
| 39 | - | ||
| 40 | - private: | ||
| 41 | - std::unique_ptr<uint8_t[], AlignedPtr::Deleter> base_ = nullptr; | ||
| 42 | - uint8_t *aligned_addr_ = nullptr; | ||
| 43 | -}; | ||
| 44 | -} // namespace llm | ||
| 45 | - | ||
| @@ -14,7 +14,7 @@ | |||
| 14 | 14 | ||
| 15 | 15 | ||
| 16 | 16 | ||
| 17 | -#include "statistic_manager.h" | 17 | +#include "comm_statistic_manager.h" |
| 18 | 18 | ||
| 19 | 19 | ||
| 20 | 20 | ||
| @@ -105,7 +105,7 @@ ge::Status D2DDataTransferJob::Process(bool &is_done) { | |||
| 105 | const auto cost = static_cast<uint64_t>( | 105 | const auto cost = static_cast<uint64_t>( |
| 106 | std::chrono::duration_cast<std::chrono::microseconds>(finished_time_point - timeout_point_).count()); | 106 | std::chrono::duration_cast<std::chrono::microseconds>(finished_time_point - timeout_point_).count()); |
| 107 | auto &send_statistic_info = comm_entity_->GetSendStatisticInfo(); | 107 | auto &send_statistic_info = comm_entity_->GetSendStatisticInfo(); |
| 108 | - StatisticManager::GetInstance().UpdateCost(cost, send_statistic_info.send_times, send_statistic_info.send_min_cost, | 108 | + CommStatisticManager::GetInstance().UpdateCost(cost, send_statistic_info.send_times, send_statistic_info.send_min_cost, |
| 109 | send_statistic_info.send_max_cost, send_statistic_info.send_total_cost); | 109 | send_statistic_info.send_max_cost, send_statistic_info.send_total_cost); |
| 110 | LLMLOGI("comm_entity:%s send all task of request finished", comm_entity_->GetDesc().c_str()); | 110 | LLMLOGI("comm_entity:%s send all task of request finished", comm_entity_->GetDesc().c_str()); |
| 111 | return ge::SUCCESS; | 111 | return ge::SUCCESS; |
| @@ -175,7 +175,7 @@ ge::Status LayerWiseTransferJob::SynchronizeTransferCacheWithRecord(const int32_ | |||
| 175 | const auto cost = | 175 | const auto cost = |
| 176 | static_cast<uint64_t>(std::chrono::duration_cast<std::chrono::microseconds>(finished - start).count()); | 176 | static_cast<uint64_t>(std::chrono::duration_cast<std::chrono::microseconds>(finished - start).count()); |
| 177 | auto &send_statistic_info = comm_entity_->GetSendStatisticInfo(stream_); | 177 | auto &send_statistic_info = comm_entity_->GetSendStatisticInfo(stream_); |
| 178 | - StatisticManager::GetInstance().UpdateCost(cost, send_statistic_info.send_times, send_statistic_info.send_min_cost, | 178 | + CommStatisticManager::GetInstance().UpdateCost(cost, send_statistic_info.send_times, send_statistic_info.send_min_cost, |
| 179 | send_statistic_info.send_max_cost, send_statistic_info.send_total_cost); | 179 | send_statistic_info.send_max_cost, send_statistic_info.send_total_cost); |
| 180 | LLMLOGI("comm_entity:%s send all task of request finished", comm_entity_->GetDesc().c_str()); | 180 | LLMLOGI("comm_entity:%s send all task of request finished", comm_entity_->GetDesc().c_str()); |
| 181 | return ge::SUCCESS; | 181 | return ge::SUCCESS; |
| @@ -12,7 +12,7 @@ | |||
| 12 | 12 | ||
| 13 | 13 | ||
| 14 | 14 | ||
| 15 | -#include "statistic_manager.h" | 15 | +#include "comm_statistic_manager.h" |
| 16 | 16 | ||
| 17 | 17 | ||
| 18 | 18 | ||
| @@ -127,7 +127,7 @@ HcclResult HcclAdapter::HcclExchangeMemDesc(HcclComm comm, uint32_t remote_rank, | |||
| 127 | const auto start = std::chrono::steady_clock::now(); | 127 | const auto start = std::chrono::steady_clock::now(); |
| 128 | auto ret = hccl_exchange_mem_desc_func_(comm, remote_rank, local, timeout, remote, actual_num); | 128 | auto ret = hccl_exchange_mem_desc_func_(comm, remote_rank, local, timeout, remote, actual_num); |
| 129 | const auto end = std::chrono::steady_clock::now(); | 129 | const auto end = std::chrono::steady_clock::now(); |
| 130 | - StatisticManager::GetInstance().AddExchangeMemCost( | 130 | + CommStatisticManager::GetInstance().AddExchangeMemCost( |
| 131 | std::chrono::duration_cast<std::chrono::microseconds>(end - start).count()); | 131 | std::chrono::duration_cast<std::chrono::microseconds>(end - start).count()); |
| 132 | return ret; | 132 | return ret; |
| 133 | } | 133 | } |
| @@ -137,7 +137,7 @@ HcclResult HcclAdapter::HcclCommInitClusterInfoMemConfig(const char *cluster, ui | |||
| 137 | const auto start = std::chrono::steady_clock::now(); | 137 | const auto start = std::chrono::steady_clock::now(); |
| 138 | auto ret = hccl_comm_init_cluster_info_mem_func_(cluster, rank, config, comm); | 138 | auto ret = hccl_comm_init_cluster_info_mem_func_(cluster, rank, config, comm); |
| 139 | const auto end = std::chrono::steady_clock::now(); | 139 | const auto end = std::chrono::steady_clock::now(); |
| 140 | - StatisticManager::GetInstance().AddCommInitCost( | 140 | + CommStatisticManager::GetInstance().AddCommInitCost( |
| 141 | std::chrono::duration_cast<std::chrono::microseconds>(end - start).count()); | 141 | std::chrono::duration_cast<std::chrono::microseconds>(end - start).count()); |
| 142 | return ret; | 142 | return ret; |
| 143 | } | 143 | } |
| @@ -176,7 +176,7 @@ HcclResult HcclAdapter::HcclCommDestroy(HcclComm comm) { | |||
| 176 | const auto start = std::chrono::steady_clock::now(); | 176 | const auto start = std::chrono::steady_clock::now(); |
| 177 | auto ret = hccl_comm_destroy_func_(comm); | 177 | auto ret = hccl_comm_destroy_func_(comm); |
| 178 | const auto end = std::chrono::steady_clock::now(); | 178 | const auto end = std::chrono::steady_clock::now(); |
| 179 | - StatisticManager::GetInstance().AddCommDestroyCost( | 179 | + CommStatisticManager::GetInstance().AddCommDestroyCost( |
| 180 | std::chrono::duration_cast<std::chrono::microseconds>(end - start).count()); | 180 | std::chrono::duration_cast<std::chrono::microseconds>(end - start).count()); |
| 181 | return ret; | 181 | return ret; |
| 182 | } | 182 | } |
| @@ -186,7 +186,7 @@ HcclResult HcclAdapter::HcclBatchPut(HcclComm comm, uint32_t remote_rank, HcclOn | |||
| 186 | const auto start = std::chrono::steady_clock::now(); | 186 | const auto start = std::chrono::steady_clock::now(); |
| 187 | auto ret = hccl_batch_put_func_(comm, remote_rank, desc, desc_num, stream); | 187 | auto ret = hccl_batch_put_func_(comm, remote_rank, desc, desc_num, stream); |
| 188 | const auto end = std::chrono::steady_clock::now(); | 188 | const auto end = std::chrono::steady_clock::now(); |
| 189 | - StatisticManager::GetInstance().AddBatchPutCost( | 189 | + CommStatisticManager::GetInstance().AddBatchPutCost( |
| 190 | std::chrono::duration_cast<std::chrono::microseconds>(end - start).count()); | 190 | std::chrono::duration_cast<std::chrono::microseconds>(end - start).count()); |
| 191 | return ret; | 191 | return ret; |
| 192 | } | 192 | } |
| @@ -212,7 +212,7 @@ HcclResult HcclAdapter::HcclRegisterGlobalMem(HcclMem *mem, void **mem_handle) c | |||
| 212 | auto ret = HCCL_E_NOT_SUPPORT; | 212 | auto ret = HCCL_E_NOT_SUPPORT; |
| 213 | if (hccl_register_global_mem_func_ != nullptr) { | 213 | if (hccl_register_global_mem_func_ != nullptr) { |
| 214 | ret = hccl_register_global_mem_func_(mem, mem_handle); | 214 | ret = hccl_register_global_mem_func_(mem, mem_handle); |
| 215 | - StatisticManager::GetInstance().AddRegisterGlobalMemTimes(); | 215 | + CommStatisticManager::GetInstance().AddRegisterGlobalMemTimes(); |
| 216 | } | 216 | } |
| 217 | return ret; | 217 | return ret; |
| 218 | } | 218 | } |
| @@ -221,7 +221,7 @@ HcclResult HcclAdapter::HcclDeregisterGlobalMem(void *mem_handle) const { | |||
| 221 | auto ret = HCCL_E_NOT_SUPPORT; | 221 | auto ret = HCCL_E_NOT_SUPPORT; |
| 222 | if (hccl_deregister_global_mem_func_ != nullptr) { | 222 | if (hccl_deregister_global_mem_func_ != nullptr) { |
| 223 | ret = hccl_deregister_global_mem_func_(mem_handle); | 223 | ret = hccl_deregister_global_mem_func_(mem_handle); |
| 224 | - StatisticManager::GetInstance().AddDeregisterGlobalMemTimes(); | 224 | + CommStatisticManager::GetInstance().AddDeregisterGlobalMemTimes(); |
| 225 | } | 225 | } |
| 226 | return ret; | 226 | return ret; |
| 227 | } | 227 | } |
| @@ -230,7 +230,7 @@ HcclResult HcclAdapter::HcclCommBindMem(HcclComm comm, void *mem_handle) const { | |||
| 230 | auto ret = HCCL_E_NOT_SUPPORT; | 230 | auto ret = HCCL_E_NOT_SUPPORT; |
| 231 | if (hccl_comm_bind_mem_func_ != nullptr) { | 231 | if (hccl_comm_bind_mem_func_ != nullptr) { |
| 232 | ret = hccl_comm_bind_mem_func_(comm, mem_handle); | 232 | ret = hccl_comm_bind_mem_func_(comm, mem_handle); |
| 233 | - StatisticManager::GetInstance().AddCommBindMemTimes(); | 233 | + CommStatisticManager::GetInstance().AddCommBindMemTimes(); |
| 234 | } | 234 | } |
| 235 | return ret; | 235 | return ret; |
| 236 | } | 236 | } |
| @@ -239,7 +239,7 @@ HcclResult HcclAdapter::HcclCommUnbindMem(HcclComm comm, void *mem_handle) const | |||
| 239 | auto ret = HCCL_E_NOT_SUPPORT; | 239 | auto ret = HCCL_E_NOT_SUPPORT; |
| 240 | if (hccl_comm_unbind_mem_func_ != nullptr) { | 240 | if (hccl_comm_unbind_mem_func_ != nullptr) { |
| 241 | ret = hccl_comm_unbind_mem_func_(comm, mem_handle); | 241 | ret = hccl_comm_unbind_mem_func_(comm, mem_handle); |
| 242 | - StatisticManager::GetInstance().AddCommUnbindMemTimes(); | 242 | + CommStatisticManager::GetInstance().AddCommUnbindMemTimes(); |
| 243 | } | 243 | } |
| 244 | return ret; | 244 | return ret; |
| 245 | } | 245 | } |
| @@ -250,7 +250,7 @@ HcclResult HcclAdapter::HcclCommPrepare(HcclComm comm, HcclPrepareConfig *prepar | |||
| 250 | const auto start = std::chrono::steady_clock::now(); | 250 | const auto start = std::chrono::steady_clock::now(); |
| 251 | ret = hccl_comm_prepare_func_(comm, prepare_config, timeout); | 251 | ret = hccl_comm_prepare_func_(comm, prepare_config, timeout); |
| 252 | const auto end = std::chrono::steady_clock::now(); | 252 | const auto end = std::chrono::steady_clock::now(); |
| 253 | - StatisticManager::GetInstance().AddCommPrepareCost( | 253 | + CommStatisticManager::GetInstance().AddCommPrepareCost( |
| 254 | std::chrono::duration_cast<std::chrono::microseconds>(end - start).count()); | 254 | std::chrono::duration_cast<std::chrono::microseconds>(end - start).count()); |
| 255 | } | 255 | } |
| 256 | return ret; | 256 | return ret; |
| @@ -500,7 +500,7 @@ ge::Status CommEntity::BatchPutAsync(std::vector<HcclOneSideOpDesc> &op_descs, a | |||
| 500 | const auto cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); | 500 | const auto cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); |
| 501 | LLMLOGI("HcclBatchPut success, num = %zu, cost = %ld us.", op_descs.size(), cost); | 501 | LLMLOGI("HcclBatchPut success, num = %zu, cost = %ld us.", op_descs.size(), cost); |
| 502 | auto &send_statistic_info = GetSendStatisticInfo(stream_to_use); | 502 | auto &send_statistic_info = GetSendStatisticInfo(stream_to_use); |
| 503 | - StatisticManager::GetInstance().UpdateCost( | 503 | + CommStatisticManager::GetInstance().UpdateCost( |
| 504 | cost, send_statistic_info.batch_put_times, send_statistic_info.batch_put_min_cost, | 504 | cost, send_statistic_info.batch_put_times, send_statistic_info.batch_put_min_cost, |
| 505 | send_statistic_info.batch_put_max_cost, send_statistic_info.batch_put_total_cost); | 505 | send_statistic_info.batch_put_max_cost, send_statistic_info.batch_put_total_cost); |
| 506 | send_statistic_info.send_total_num += op_descs.size(); | 506 | send_statistic_info.send_total_num += op_descs.size(); |
| @@ -531,7 +531,7 @@ ge::Status CommEntity::BatchGetAsync(std::vector<HcclOneSideOpDesc> &op_descs, a | |||
| 531 | const auto end = std::chrono::steady_clock::now(); | 531 | const auto end = std::chrono::steady_clock::now(); |
| 532 | const auto cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); | 532 | const auto cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); |
| 533 | LLMLOGI("HcclBatchGet success, num = %zu, cost = %ld us.", op_descs.size(), cost);; | 533 | LLMLOGI("HcclBatchGet success, num = %zu, cost = %ld us.", op_descs.size(), cost);; |
| 534 | - StatisticManager::UpdateCost( | 534 | + CommStatisticManager::UpdateCost( |
| 535 | cost, recv_statistic_info_.batch_get_times, recv_statistic_info_.batch_get_min_cost, | 535 | cost, recv_statistic_info_.batch_get_times, recv_statistic_info_.batch_get_min_cost, |
| 536 | recv_statistic_info_.batch_get_max_cost, recv_statistic_info_.batch_get_total_cost); | 536 | recv_statistic_info_.batch_get_max_cost, recv_statistic_info_.batch_get_total_cost); |
| 537 | recv_statistic_info_.get_total_num += op_descs.size(); | 537 | recv_statistic_info_.get_total_num += op_descs.size(); |
| @@ -567,13 +567,13 @@ ge::Status CommEntity::BatchTransfer(std::list<HcclOneSideOpDesc> &tasks, bool i | |||
| 567 | LLMLOGI("sync stream success, cost = %ld us.", cost); | 567 | LLMLOGI("sync stream success, cost = %ld us.", cost); |
| 568 | if (is_put) { | 568 | if (is_put) { |
| 569 | auto &send_statistic_info = GetSendStatisticInfo(stream_); | 569 | auto &send_statistic_info = GetSendStatisticInfo(stream_); |
| 570 | - StatisticManager::GetInstance().UpdateCost( | 570 | + CommStatisticManager::GetInstance().UpdateCost( |
| 571 | cost, send_statistic_info.batch_put_times, send_statistic_info.batch_put_min_cost, | 571 | cost, send_statistic_info.batch_put_times, send_statistic_info.batch_put_min_cost, |
| 572 | send_statistic_info.batch_put_max_cost, send_statistic_info.batch_put_total_cost); | 572 | send_statistic_info.batch_put_max_cost, send_statistic_info.batch_put_total_cost); |
| 573 | send_statistic_info.send_total_num += task_num; | 573 | send_statistic_info.send_total_num += task_num; |
| 574 | } else { | 574 | } else { |
| 575 | auto &recv_statistic_info = GetRecvStatisticInfo(); | 575 | auto &recv_statistic_info = GetRecvStatisticInfo(); |
| 576 | - StatisticManager::UpdateCost(cost, recv_statistic_info.pull_times, recv_statistic_info.pull_min_cost, | 576 | + CommStatisticManager::UpdateCost(cost, recv_statistic_info.pull_times, recv_statistic_info.pull_min_cost, |
| 577 | recv_statistic_info.pull_max_cost, recv_statistic_info.pull_total_cost); | 577 | recv_statistic_info.pull_max_cost, recv_statistic_info.pull_total_cost); |
| 578 | } | 578 | } |
| 579 | LLM_DISMISS_GUARD(stream); | 579 | LLM_DISMISS_GUARD(stream); |
| @@ -24,7 +24,7 @@ | |||
| 24 | 24 | ||
| 25 | 25 | ||
| 26 | 26 | ||
| 27 | -#include "statistic_manager.h" | 27 | +#include "comm_statistic_manager.h" |
| 28 | 28 | ||
| 29 | 29 | ||
| 30 | namespace llm { | 30 | namespace llm { |
| @@ -15,7 +15,6 @@ | |||
| 15 | 15 | ||
| 16 | 16 | ||
| 17 | 17 | ||
| 18 | - | ||
| 19 | 18 | ||
| 20 | namespace llm { | 19 | namespace llm { |
| 21 | using EntityPtr = std::shared_ptr<CommEntity>; | 20 | using EntityPtr = std::shared_ptr<CommEntity>; |
| @@ -45,7 +44,6 @@ class CommEntityManager { | |||
| 45 | bool need_handle_request_ = false; | 44 | bool need_handle_request_ = false; |
| 46 | std::thread cache_engine_thread_; | 45 | std::thread cache_engine_thread_; |
| 47 | std::unique_ptr<LlmMemPool> host_mem_pool_{}; | 46 | std::unique_ptr<LlmMemPool> host_mem_pool_{}; |
| 48 | - std::shared_ptr<AlignedPtr> host_buffer_; | ||
| 49 | aclrtContext aclrt_context_{}; | 47 | aclrtContext aclrt_context_{}; |
| 50 | std::atomic_uint64_t entity_id_gen_{1LU}; | 48 | std::atomic_uint64_t entity_id_gen_{1LU}; |
| 51 | std::atomic_bool mgr_high_priority_flag_{false}; | 49 | std::atomic_bool mgr_high_priority_flag_{false}; |
| @@ -11,7 +11,7 @@ | |||
| 11 | 11 | ||
| 12 | 12 | ||
| 13 | 13 | ||
| 14 | -#include "statistic_manager.h" | 14 | +#include "comm_statistic_manager.h" |
| 15 | 15 | ||
| 16 | 16 | ||
| 17 | 17 | ||
| @@ -48,7 +48,7 @@ ge::Status LLMDataDistV2::DoInnerInitialize(int32_t device_id, | |||
| 48 | 48 | ||
| 49 | LlmDatadistTimer::Instance().Init(); | 49 | LlmDatadistTimer::Instance().Init(); |
| 50 | statistic_timer_handle_ = LlmDatadistTimer::Instance().CreateTimer([this]() { | 50 | statistic_timer_handle_ = LlmDatadistTimer::Instance().CreateTimer([this]() { |
| 51 | - StatisticManager::GetInstance().Dump(); | 51 | + CommStatisticManager::GetInstance().Dump(); |
| 52 | comm_entity_manager_->Dump(); | 52 | comm_entity_manager_->Dump(); |
| 53 | }); | 53 | }); |
| 54 | constexpr uint32_t kStatisticTimerPeriod = 80U * 1000U; | 54 | constexpr uint32_t kStatisticTimerPeriod = 80U * 1000U; |
| @@ -111,8 +111,8 @@ void LLMDataDistV2::DoInnerFinalize() { | |||
| 111 | GlobalMemManager::GetInstance().Finalize(); | 111 | GlobalMemManager::GetInstance().Finalize(); |
| 112 | transfer_engine_->Finalize(); | 112 | transfer_engine_->Finalize(); |
| 113 | 113 | ||
| 114 | - StatisticManager::GetInstance().Dump(); | 114 | + CommStatisticManager::GetInstance().Dump(); |
| 115 | - StatisticManager::GetInstance().Reset(); | 115 | + CommStatisticManager::GetInstance().Reset(); |
| 116 | if (statistic_timer_handle_ != nullptr) { | 116 | if (statistic_timer_handle_ != nullptr) { |
| 117 | (void)LlmDatadistTimer::Instance().StopTimer(statistic_timer_handle_); | 117 | (void)LlmDatadistTimer::Instance().StopTimer(statistic_timer_handle_); |
| 118 | (void)LlmDatadistTimer::Instance().DeleteTimer(statistic_timer_handle_); | 118 | (void)LlmDatadistTimer::Instance().DeleteTimer(statistic_timer_handle_); |
| @@ -162,9 +162,9 @@ ge::Status LLMDataDistV2::Link(std::string &cluster_name, | |||
| 162 | hixl::TemporaryRtContext with_context(aclrt_context_); | 162 | hixl::TemporaryRtContext with_context(aclrt_context_); |
| 163 | LLM_CHK_STATUS_RET(transfer_engine_->Link(cluster_name, cluster2rank, rank_table, comm_id), "Link failed."); | 163 | LLM_CHK_STATUS_RET(transfer_engine_->Link(cluster_name, cluster2rank, rank_table, comm_id), "Link failed."); |
| 164 | const auto end = std::chrono::steady_clock::now(); | 164 | const auto end = std::chrono::steady_clock::now(); |
| 165 | - auto &func_statistic_info = StatisticManager::GetInstance().GetFuncStatisticInfo(); | 165 | + auto &func_statistic_info = CommStatisticManager::GetInstance().GetFuncStatisticInfo(); |
| 166 | const uint64_t cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); | 166 | const uint64_t cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); |
| 167 | - StatisticManager::UpdateCost(cost, func_statistic_info.link_func_times, func_statistic_info.link_func_min_cost, | 167 | + CommStatisticManager::UpdateCost(cost, func_statistic_info.link_func_times, func_statistic_info.link_func_min_cost, |
| 168 | func_statistic_info.link_func_max_cost, func_statistic_info.link_func_total_cost); | 168 | func_statistic_info.link_func_max_cost, func_statistic_info.link_func_total_cost); |
| 169 | return ge::SUCCESS; | 169 | return ge::SUCCESS; |
| 170 | } | 170 | } |
| @@ -176,9 +176,9 @@ ge::Status LLMDataDistV2::Unlink(uint64_t comm_id) { | |||
| 176 | hixl::TemporaryRtContext with_context(aclrt_context_); | 176 | hixl::TemporaryRtContext with_context(aclrt_context_); |
| 177 | LLM_CHK_STATUS_RET(transfer_engine_->Unlink(comm_id), "Unlink failed."); | 177 | LLM_CHK_STATUS_RET(transfer_engine_->Unlink(comm_id), "Unlink failed."); |
| 178 | const auto end = std::chrono::steady_clock::now(); | 178 | const auto end = std::chrono::steady_clock::now(); |
| 179 | - auto &func_statistic_info = StatisticManager::GetInstance().GetFuncStatisticInfo(); | 179 | + auto &func_statistic_info = CommStatisticManager::GetInstance().GetFuncStatisticInfo(); |
| 180 | const uint64_t cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); | 180 | const uint64_t cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); |
| 181 | - StatisticManager::UpdateCost(cost, func_statistic_info.unlink_func_times, func_statistic_info.unlink_func_min_cost, | 181 | + CommStatisticManager::UpdateCost(cost, func_statistic_info.unlink_func_times, func_statistic_info.unlink_func_min_cost, |
| 182 | func_statistic_info.unlink_func_max_cost, func_statistic_info.unlink_func_total_cost); | 182 | func_statistic_info.unlink_func_max_cost, func_statistic_info.unlink_func_total_cost); |
| 183 | return ge::SUCCESS; | 183 | return ge::SUCCESS; |
| 184 | } | 184 | } |
| @@ -196,7 +196,7 @@ ge::Status LLMDataDistV2::RegisterCache(const CacheDesc &cache_desc, Cache &cach | |||
| 196 | LLM_CHK_BOOL_RET_STATUS(is_initialized_.load(std::memory_order::memory_order_relaxed), ge::FAILED, | 196 | LLM_CHK_BOOL_RET_STATUS(is_initialized_.load(std::memory_order::memory_order_relaxed), ge::FAILED, |
| 197 | "Llm datadist of cluster:%lu is not initialized.", cluster_id_); | 197 | "Llm datadist of cluster:%lu is not initialized.", cluster_id_); |
| 198 | hixl::TemporaryRtContext with_context(aclrt_context_); | 198 | hixl::TemporaryRtContext with_context(aclrt_context_); |
| 199 | - auto &func_statistic_info = StatisticManager::GetInstance().GetFuncStatisticInfo(); | 199 | + auto &func_statistic_info = CommStatisticManager::GetInstance().GetFuncStatisticInfo(); |
| 200 | func_statistic_info.register_func_times++; | 200 | func_statistic_info.register_func_times++; |
| 201 | return data_cache_engine_->Register(cache_desc, cache_keys, cache); | 201 | return data_cache_engine_->Register(cache_desc, cache_keys, cache); |
| 202 | } | 202 | } |
| @@ -206,7 +206,7 @@ ge::Status LLMDataDistV2::AllocateCache(const CacheDesc &cache_desc, Cache &cach | |||
| 206 | LLM_CHK_BOOL_RET_STATUS(is_initialized_.load(std::memory_order::memory_order_relaxed), ge::FAILED, | 206 | LLM_CHK_BOOL_RET_STATUS(is_initialized_.load(std::memory_order::memory_order_relaxed), ge::FAILED, |
| 207 | "Llm datadist of cluster:%lu is not initialized.", cluster_id_); | 207 | "Llm datadist of cluster:%lu is not initialized.", cluster_id_); |
| 208 | hixl::TemporaryRtContext with_context(aclrt_context_); | 208 | hixl::TemporaryRtContext with_context(aclrt_context_); |
| 209 | - auto &mem_statistic_info = StatisticManager::GetInstance().GetMemoryStatisticInfo(); | 209 | + auto &mem_statistic_info = CommStatisticManager::GetInstance().GetMemoryStatisticInfo(); |
| 210 | mem_statistic_info.alloc_times++; | 210 | mem_statistic_info.alloc_times++; |
| 211 | return data_cache_engine_->Allocate(cache_desc, cache_keys, cache); | 211 | return data_cache_engine_->Allocate(cache_desc, cache_keys, cache); |
| 212 | } | 212 | } |
| @@ -215,7 +215,7 @@ ge::Status LLMDataDistV2::DeallocateCache(int64_t cache_id) { | |||
| 215 | LLM_CHK_BOOL_RET_STATUS(is_initialized_.load(std::memory_order::memory_order_relaxed), ge::FAILED, | 215 | LLM_CHK_BOOL_RET_STATUS(is_initialized_.load(std::memory_order::memory_order_relaxed), ge::FAILED, |
| 216 | "Llm datadist of cluster:%lu is not initialized.", cluster_id_); | 216 | "Llm datadist of cluster:%lu is not initialized.", cluster_id_); |
| 217 | hixl::TemporaryRtContext with_context(aclrt_context_); | 217 | hixl::TemporaryRtContext with_context(aclrt_context_); |
| 218 | - auto &mem_statistic_info = StatisticManager::GetInstance().GetMemoryStatisticInfo(); | 218 | + auto &mem_statistic_info = CommStatisticManager::GetInstance().GetMemoryStatisticInfo(); |
| 219 | mem_statistic_info.free_times++; | 219 | mem_statistic_info.free_times++; |
| 220 | return data_cache_engine_->Deallocate(cache_id); | 220 | return data_cache_engine_->Deallocate(cache_id); |
| 221 | } | 221 | } |
| @@ -232,9 +232,9 @@ ge::Status LLMDataDistV2::PullCache(int64_t cache_id, const CacheKey &cache_key, | |||
| 232 | "data can not be pulled from own cluster:%lu", cluster_id_); | 232 | "data can not be pulled from own cluster:%lu", cluster_id_); |
| 233 | LLM_CHK_STATUS_RET(data_cache_engine_->PullCache(cache_id, cache_key, pull_cache_param), "pull cache failed"); | 233 | LLM_CHK_STATUS_RET(data_cache_engine_->PullCache(cache_id, cache_key, pull_cache_param), "pull cache failed"); |
| 234 | const auto end = std::chrono::steady_clock::now(); | 234 | const auto end = std::chrono::steady_clock::now(); |
| 235 | - auto &func_statistic_info = StatisticManager::GetInstance().GetFuncStatisticInfo(); | 235 | + auto &func_statistic_info = CommStatisticManager::GetInstance().GetFuncStatisticInfo(); |
| 236 | const uint64_t cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); | 236 | const uint64_t cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); |
| 237 | - StatisticManager::UpdateCost(cost, func_statistic_info.pull_func_times, func_statistic_info.pull_func_min_cost, | 237 | + CommStatisticManager::UpdateCost(cost, func_statistic_info.pull_func_times, func_statistic_info.pull_func_min_cost, |
| 238 | func_statistic_info.pull_func_max_cost, func_statistic_info.pull_func_total_cost); | 238 | func_statistic_info.pull_func_max_cost, func_statistic_info.pull_func_total_cost); |
| 239 | return ge::SUCCESS; | 239 | return ge::SUCCESS; |
| 240 | } | 240 | } |
| @@ -262,9 +262,9 @@ ge::Status LLMDataDistV2::CopyCache(const CopyCacheParam ©_cache_param) { | |||
| 262 | hixl::TemporaryRtContext with_context(aclrt_context_); | 262 | hixl::TemporaryRtContext with_context(aclrt_context_); |
| 263 | LLM_CHK_STATUS_RET(data_cache_engine_->CopyCache(copy_cache_param), "copy cache failed"); | 263 | LLM_CHK_STATUS_RET(data_cache_engine_->CopyCache(copy_cache_param), "copy cache failed"); |
| 264 | const auto end = std::chrono::steady_clock::now(); | 264 | const auto end = std::chrono::steady_clock::now(); |
| 265 | - auto &func_statistic_info = StatisticManager::GetInstance().GetFuncStatisticInfo(); | 265 | + auto &func_statistic_info = CommStatisticManager::GetInstance().GetFuncStatisticInfo(); |
| 266 | const uint64_t cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); | 266 | const uint64_t cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); |
| 267 | - StatisticManager::UpdateCost(cost, func_statistic_info.copy_func_times, func_statistic_info.copy_func_min_cost, | 267 | + CommStatisticManager::UpdateCost(cost, func_statistic_info.copy_func_times, func_statistic_info.copy_func_min_cost, |
| 268 | func_statistic_info.copy_func_max_cost, func_statistic_info.copy_func_total_cost); | 268 | func_statistic_info.copy_func_max_cost, func_statistic_info.copy_func_total_cost); |
| 269 | return ge::SUCCESS; | 269 | return ge::SUCCESS; |
| 270 | } | 270 | } |
| @@ -292,9 +292,9 @@ ge::Status LLMDataDistV2::SwapBlocks(const Cache &src, const Cache &dst, const u | |||
| 292 | hixl::TemporaryRtContext with_context(aclrt_context_); | 292 | hixl::TemporaryRtContext with_context(aclrt_context_); |
| 293 | LLM_CHK_STATUS_RET(data_cache_engine_->SwapBlocks(src, dst, block_size, type, block_mapping), "swap blocks failed"); | 293 | LLM_CHK_STATUS_RET(data_cache_engine_->SwapBlocks(src, dst, block_size, type, block_mapping), "swap blocks failed"); |
| 294 | const auto end = std::chrono::steady_clock::now(); | 294 | const auto end = std::chrono::steady_clock::now(); |
| 295 | - auto &func_statistic_info = StatisticManager::GetInstance().GetFuncStatisticInfo(); | 295 | + auto &func_statistic_info = CommStatisticManager::GetInstance().GetFuncStatisticInfo(); |
| 296 | const uint64_t cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); | 296 | const uint64_t cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); |
| 297 | - StatisticManager::UpdateCost(cost, func_statistic_info.swap_func_times, func_statistic_info.swap_func_min_cost, | 297 | + CommStatisticManager::UpdateCost(cost, func_statistic_info.swap_func_times, func_statistic_info.swap_func_min_cost, |
| 298 | func_statistic_info.swap_func_max_cost, func_statistic_info.swap_func_total_cost); | 298 | func_statistic_info.swap_func_max_cost, func_statistic_info.swap_func_total_cost); |
| 299 | return ge::SUCCESS; | 299 | return ge::SUCCESS; |
| 300 | } | 300 | } |
| @@ -320,9 +320,9 @@ ge::Status LLMDataDistV2::TransferCache(const uint64_t task_id, const TransferCa | |||
| 320 | "task:%lu of cluster:%lu transfer cache of layer[%lu] failed", task_id, | 320 | "task:%lu of cluster:%lu transfer cache of layer[%lu] failed", task_id, |
| 321 | transfer_cache_config.cluster_id, transfer_cache_config.layer_index); | 321 | transfer_cache_config.cluster_id, transfer_cache_config.layer_index); |
| 322 | const auto end = std::chrono::steady_clock::now(); | 322 | const auto end = std::chrono::steady_clock::now(); |
| 323 | - auto &func_statistic_info = StatisticManager::GetInstance().GetFuncStatisticInfo(); | 323 | + auto &func_statistic_info = CommStatisticManager::GetInstance().GetFuncStatisticInfo(); |
| 324 | const uint64_t cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); | 324 | const uint64_t cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); |
| 325 | - StatisticManager::UpdateCost(cost, func_statistic_info.transfer_func_times, | 325 | + CommStatisticManager::UpdateCost(cost, func_statistic_info.transfer_func_times, |
| 326 | func_statistic_info.transfer_func_min_cost, func_statistic_info.transfer_func_max_cost, | 326 | func_statistic_info.transfer_func_min_cost, func_statistic_info.transfer_func_max_cost, |
| 327 | func_statistic_info.transfer_func_total_cost); | 327 | func_statistic_info.transfer_func_total_cost); |
| 328 | LLMLOGI("task:%lu of cluster:%lu transfer cache of layer[%lu] success", task_id, transfer_cache_config.cluster_id, | 328 | LLMLOGI("task:%lu of cluster:%lu transfer cache of layer[%lu] success", task_id, transfer_cache_config.cluster_id, |
| @@ -334,7 +334,7 @@ ge::Status LLMDataDistV2::UnregisterCache(int64_t cache_id) { | |||
| 334 | LLM_CHK_BOOL_RET_STATUS(is_initialized_.load(std::memory_order::memory_order_relaxed), ge::FAILED, | 334 | LLM_CHK_BOOL_RET_STATUS(is_initialized_.load(std::memory_order::memory_order_relaxed), ge::FAILED, |
| 335 | "Llm datadist of cluster:%lu is not initialized.", cluster_id_); | 335 | "Llm datadist of cluster:%lu is not initialized.", cluster_id_); |
| 336 | hixl::TemporaryRtContext with_context(aclrt_context_); | 336 | hixl::TemporaryRtContext with_context(aclrt_context_); |
| 337 | - auto &func_statistic_info = StatisticManager::GetInstance().GetFuncStatisticInfo(); | 337 | + auto &func_statistic_info = CommStatisticManager::GetInstance().GetFuncStatisticInfo(); |
| 338 | func_statistic_info.deregister_func_times++; | 338 | func_statistic_info.deregister_func_times++; |
| 339 | return data_cache_engine_->Unregister(cache_id); | 339 | return data_cache_engine_->Unregister(cache_id); |
| 340 | } | 340 | } |
| @@ -348,9 +348,9 @@ ge::Status LLMDataDistV2::LinkClusters(const std::vector<ClusterInfo> &clusters, | |||
| 348 | hixl::TemporaryRtContext with_context(aclrt_context_); | 348 | hixl::TemporaryRtContext with_context(aclrt_context_); |
| 349 | LLM_CHK_STATUS_RET(transfer_engine_->LinkClusters(clusters, rets, timeout), "Failed to link clusters."); | 349 | LLM_CHK_STATUS_RET(transfer_engine_->LinkClusters(clusters, rets, timeout), "Failed to link clusters."); |
| 350 | const auto end = std::chrono::steady_clock::now(); | 350 | const auto end = std::chrono::steady_clock::now(); |
| 351 | - auto &func_statistic_info = StatisticManager::GetInstance().GetFuncStatisticInfo(); | 351 | + auto &func_statistic_info = CommStatisticManager::GetInstance().GetFuncStatisticInfo(); |
| 352 | const uint64_t cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); | 352 | const uint64_t cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); |
| 353 | - StatisticManager::UpdateCost(cost, func_statistic_info.link_func_times, func_statistic_info.link_func_min_cost, | 353 | + CommStatisticManager::UpdateCost(cost, func_statistic_info.link_func_times, func_statistic_info.link_func_min_cost, |
| 354 | func_statistic_info.link_func_max_cost, func_statistic_info.link_func_total_cost); | 354 | func_statistic_info.link_func_max_cost, func_statistic_info.link_func_total_cost); |
| 355 | return ge::SUCCESS; | 355 | return ge::SUCCESS; |
| 356 | } | 356 | } |
| @@ -365,9 +365,9 @@ ge::Status LLMDataDistV2::UnlinkClusters(const std::vector<ClusterInfo> &cluster | |||
| 365 | hixl::TemporaryRtContext with_context(aclrt_context_); | 365 | hixl::TemporaryRtContext with_context(aclrt_context_); |
| 366 | LLM_CHK_STATUS_RET(transfer_engine_->UnlinkClusters(clusters, rets, timeout, force_flag), "Failed to unlink clusters."); | 366 | LLM_CHK_STATUS_RET(transfer_engine_->UnlinkClusters(clusters, rets, timeout, force_flag), "Failed to unlink clusters."); |
| 367 | const auto end = std::chrono::steady_clock::now(); | 367 | const auto end = std::chrono::steady_clock::now(); |
| 368 | - auto &func_statistic_info = StatisticManager::GetInstance().GetFuncStatisticInfo(); | 368 | + auto &func_statistic_info = CommStatisticManager::GetInstance().GetFuncStatisticInfo(); |
| 369 | const uint64_t cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); | 369 | const uint64_t cost = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count(); |
| 370 | - StatisticManager::UpdateCost(cost, func_statistic_info.unlink_func_times, func_statistic_info.unlink_func_min_cost, | 370 | + CommStatisticManager::UpdateCost(cost, func_statistic_info.unlink_func_times, func_statistic_info.unlink_func_min_cost, |
| 371 | func_statistic_info.unlink_func_max_cost, func_statistic_info.unlink_func_total_cost); | 371 | func_statistic_info.unlink_func_max_cost, func_statistic_info.unlink_func_total_cost); |
| 372 | return ge::SUCCESS; | 372 | return ge::SUCCESS; |
| 373 | } | 373 | } |
| @@ -46,7 +46,7 @@ file(GLOB HIXL_SRC_LIST | |||
| 46 | 46 | ||
| 47 | file(GLOB ADXL_SRC_LIST | 47 | file(GLOB ADXL_SRC_LIST |
| 48 | "${HIXL_CODE_DIR}/src/llm_datadist/api/adxl_engine_impl.cc" | 48 | "${HIXL_CODE_DIR}/src/llm_datadist/api/adxl_engine_impl.cc" |
| 49 | - "${HIXL_CODE_DIR}/src/llm_datadist/statistic_manager.cc" | 49 | + "${HIXL_CODE_DIR}/src/llm_datadist/comm_statistic_manager.cc" |
| 50 | "${HIXL_CODE_DIR}/src/llm_datadist/llm_datadist_timer.cc" | 50 | "${HIXL_CODE_DIR}/src/llm_datadist/llm_datadist_timer.cc" |
| 51 | "${HIXL_CODE_DIR}/src/llm_datadist/adxl/*.cc" | 51 | "${HIXL_CODE_DIR}/src/llm_datadist/adxl/*.cc" |
| 52 | "${HIXL_CODE_DIR}/src/llm_datadist/common/*.cc" | 52 | "${HIXL_CODE_DIR}/src/llm_datadist/common/*.cc" |
| @@ -21,7 +21,6 @@ set(LLM_DATADIST_TEST_FILES | |||
| 21 | llm_datadist_v2_api_unittest.cc | 21 | llm_datadist_v2_api_unittest.cc |
| 22 | llm_utils_v2_unittest.cc | 22 | llm_utils_v2_unittest.cc |
| 23 | data_cache_engine_unittest.cc | 23 | data_cache_engine_unittest.cc |
| 24 | - aligned_ptr_unittest.cc | ||
| 25 | hixl_api_unittest.cc | 24 | hixl_api_unittest.cc |
| 26 | adxl_engine_api_unittest.cc | 25 | adxl_engine_api_unittest.cc |
| 27 | segment_table_unittest.cc | 26 | segment_table_unittest.cc |
| @@ -24,7 +24,7 @@ | |||
| 24 | 24 | ||
| 25 | 25 | ||
| 26 | 26 | ||
| 27 | -#include "engine/adxl_engine.h" | 27 | +#include "engine/comm_engine.h" |
| 28 | 28 | ||
| 29 | 29 | ||
| 30 | 30 | ||
| @@ -151,7 +151,7 @@ TEST_F(AdxlEngineUTest, TestEngineFactoryFallbackToAdxlEngineWithoutLocalCommRes | |||
| 151 | 151 | ||
| 152 | auto engine = hixl::EngineFactory::CreateEngine("127.0.0.1", options); | 152 | auto engine = hixl::EngineFactory::CreateEngine("127.0.0.1", options); |
| 153 | ASSERT_NE(engine, nullptr); | 153 | ASSERT_NE(engine, nullptr); |
| 154 | - EXPECT_NE(dynamic_cast<hixl::AdxlEngine *>(engine.get()), nullptr); | 154 | + EXPECT_NE(dynamic_cast<hixl::CommEngine *>(engine.get()), nullptr); |
| 155 | EXPECT_EQ(dynamic_cast<hixl::HixlEngine *>(engine.get()), nullptr); | 155 | EXPECT_EQ(dynamic_cast<hixl::HixlEngine *>(engine.get()), nullptr); |
| 156 | } | 156 | } |
| 157 | 157 | ||
| @@ -159,7 +159,7 @@ TEST_F(AdxlEngineUTest, TestEngineFactoryUseHixlEngineWithLocalCommRes) { | |||
| 159 | auto engine = hixl::EngineFactory::CreateEngine("127.0.0.1", BuildHixlCsOptions("127.0.0.1")); | 159 | auto engine = hixl::EngineFactory::CreateEngine("127.0.0.1", BuildHixlCsOptions("127.0.0.1")); |
| 160 | ASSERT_NE(engine, nullptr); | 160 | ASSERT_NE(engine, nullptr); |
| 161 | EXPECT_NE(dynamic_cast<hixl::HixlEngine *>(engine.get()), nullptr); | 161 | EXPECT_NE(dynamic_cast<hixl::HixlEngine *>(engine.get()), nullptr); |
| 162 | - EXPECT_EQ(dynamic_cast<hixl::AdxlEngine *>(engine.get()), nullptr); | 162 | + EXPECT_EQ(dynamic_cast<hixl::CommEngine *>(engine.get()), nullptr); |
| 163 | } | 163 | } |
| 164 | 164 | ||
| 165 | TEST_F(AdxlEngineUTest, TestEngineFactoryUseHixlEngineWhenUboeNotFirstInProtocolDesc) { | 165 | TEST_F(AdxlEngineUTest, TestEngineFactoryUseHixlEngineWhenUboeNotFirstInProtocolDesc) { |
| @@ -173,7 +173,7 @@ TEST_F(AdxlEngineUTest, TestEngineFactoryUseHixlEngineWhenUboeNotFirstInProtocol | |||
| 173 | auto engine = hixl::EngineFactory::CreateEngine("127.0.0.1", options); | 173 | auto engine = hixl::EngineFactory::CreateEngine("127.0.0.1", options); |
| 174 | ASSERT_NE(engine, nullptr); | 174 | ASSERT_NE(engine, nullptr); |
| 175 | EXPECT_NE(dynamic_cast<hixl::HixlEngine *>(engine.get()), nullptr); | 175 | EXPECT_NE(dynamic_cast<hixl::HixlEngine *>(engine.get()), nullptr); |
| 176 | - EXPECT_EQ(dynamic_cast<hixl::AdxlEngine *>(engine.get()), nullptr); | 176 | + EXPECT_EQ(dynamic_cast<hixl::CommEngine *>(engine.get()), nullptr); |
| 177 | } | 177 | } |
| 178 | 178 | ||
| 179 | TEST_F(AdxlEngineUTest, TestEngineFactoryUseFabricMemEngineWhenFabricMemEnabled) { | 179 | TEST_F(AdxlEngineUTest, TestEngineFactoryUseFabricMemEngineWhenFabricMemEnabled) { |
| @@ -183,7 +183,7 @@ TEST_F(AdxlEngineUTest, TestEngineFactoryUseFabricMemEngineWhenFabricMemEnabled) | |||
| 183 | ASSERT_NE(engine, nullptr); | 183 | ASSERT_NE(engine, nullptr); |
| 184 | EXPECT_NE(dynamic_cast<hixl::FabricMemEngine *>(engine.get()), nullptr); | 184 | EXPECT_NE(dynamic_cast<hixl::FabricMemEngine *>(engine.get()), nullptr); |
| 185 | EXPECT_EQ(dynamic_cast<hixl::HixlEngine *>(engine.get()), nullptr); | 185 | EXPECT_EQ(dynamic_cast<hixl::HixlEngine *>(engine.get()), nullptr); |
| 186 | - EXPECT_EQ(dynamic_cast<hixl::AdxlEngine *>(engine.get()), nullptr); | 186 | + EXPECT_EQ(dynamic_cast<hixl::CommEngine *>(engine.get()), nullptr); |
| 187 | } | 187 | } |
| 188 | 188 | ||
| 189 | TEST_F(AdxlEngineUTest, TestAdxlEngine) { | 189 | TEST_F(AdxlEngineUTest, TestAdxlEngine) { |
| @@ -378,7 +378,7 @@ TEST_F(AdxlEngineUTest, TestDeregisterUnregisterMem) { | |||
| 378 | 378 | ||
| 379 | TEST_F(AdxlEngineUTest, TestHeartbeat) { | 379 | TEST_F(AdxlEngineUTest, TestHeartbeat) { |
| 380 | ChannelManager::SetHeartbeatWaitTime(10); // 10ms | 380 | ChannelManager::SetHeartbeatWaitTime(10); // 10ms |
| 381 | - Channel::SetHeartbeatTimeout(50); // 50ms | 381 | + CommChannel::SetHeartbeatTimeout(50); // 50ms |
| 382 | AdxlEngine engine1; | 382 | AdxlEngine engine1; |
| 383 | llm::AutoCommResRuntimeMock::SetDevice(0); | 383 | llm::AutoCommResRuntimeMock::SetDevice(0); |
| 384 | std::map<AscendString, AscendString> options1; | 384 | std::map<AscendString, AscendString> options1; |
| @@ -1,58 +0,0 @@ | |||
| 1 | -/** | ||
| 2 | - * Copyright (c) 2025 Huawei Technologies Co., Ltd. | ||
| 3 | - * This program is free software, you can redistribute it and/or modify it under the terms and conditions of | ||
| 4 | - * CANN Open Software License Agreement Version 2.0 (the "License"). | ||
| 5 | - * Please refer to the License for details. You may not use this file except in compliance with the License. | ||
| 6 | - * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, | ||
| 7 | - * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. | ||
| 8 | - * See LICENSE in the root of the software repository for the full text of the License. | ||
| 9 | - */ | ||
| 10 | - | ||
| 11 | - | ||
| 12 | - | ||
| 13 | - | ||
| 14 | - | ||
| 15 | -using namespace std; | ||
| 16 | -using namespace ::testing; | ||
| 17 | - | ||
| 18 | -namespace llm { | ||
| 19 | -class AlignedPtrTest : public ::testing::Test { | ||
| 20 | - protected: | ||
| 21 | - void SetUp() override {} | ||
| 22 | - // 在测试类中进行清理工作,如果需要的话 | ||
| 23 | - void TearDown() override {} | ||
| 24 | -}; | ||
| 25 | - | ||
| 26 | -TEST_F(AlignedPtrTest, NormalAlignment) { | ||
| 27 | - const size_t buffer_size = 1024; | ||
| 28 | - const size_t alignment = 64; | ||
| 29 | - AlignedPtr ptr(buffer_size, alignment); | ||
| 30 | - EXPECT_NE(ptr.Get(), nullptr); | ||
| 31 | - EXPECT_EQ(reinterpret_cast<uintptr_t>(ptr.Get()) % alignment, 0); | ||
| 32 | -} | ||
| 33 | - | ||
| 34 | -TEST_F(AlignedPtrTest, ZeroBufferSize) { | ||
| 35 | - AlignedPtr ptr(0, 16); | ||
| 36 | - EXPECT_EQ(ptr.Get(), nullptr); | ||
| 37 | -} | ||
| 38 | - | ||
| 39 | -TEST_F(AlignedPtrTest, ZeroAlignmentWithNonZeroSize) { | ||
| 40 | - const size_t buffer_size = 1024; | ||
| 41 | - AlignedPtr ptr(buffer_size, 0); | ||
| 42 | - EXPECT_NE(ptr.Get(), nullptr); | ||
| 43 | -} | ||
| 44 | - | ||
| 45 | -TEST_F(AlignedPtrTest, AllocationFailure) { | ||
| 46 | - const size_t huge_size = std::numeric_limits<size_t>::max(); | ||
| 47 | - AlignedPtr ptr(huge_size, 16); | ||
| 48 | - EXPECT_EQ(ptr.Get(), nullptr); | ||
| 49 | -} | ||
| 50 | - | ||
| 51 | -TEST_F(AlignedPtrTest, AlignmentLargerThanSize) { | ||
| 52 | - const size_t buffer_size = 16; | ||
| 53 | - const size_t alignment = 64; | ||
| 54 | - AlignedPtr ptr(buffer_size, alignment); | ||
| 55 | - EXPECT_NE(ptr.Get(), nullptr); | ||
| 56 | - EXPECT_EQ(reinterpret_cast<uintptr_t>(ptr.Get()) % alignment, 0); | ||
| 57 | -} | ||
| 58 | -} // namespace llm | ||
| @@ -17,7 +17,7 @@ | |||
| 17 | 17 | ||
| 18 | 18 | ||
| 19 | 19 | ||
| 20 | -#include "adxl/channel.h" | 20 | +#include "adxl/comm_channel.h" |
| 21 | 21 | ||
| 22 | 22 | ||
| 23 | namespace adxl { | 23 | namespace adxl { |
| @@ -46,7 +46,7 @@ ChannelPtr CreateChannel() { | |||
| 46 | channel_info.channel_id = kChannelId; | 46 | channel_info.channel_id = kChannelId; |
| 47 | channel_info.local_rank_id = 0; | 47 | channel_info.local_rank_id = 0; |
| 48 | channel_info.peer_rank_id = kPeerRankId; | 48 | channel_info.peer_rank_id = kPeerRankId; |
| 49 | - return std::make_shared<Channel>(channel_info); | 49 | + return std::make_shared<CommChannel>(channel_info); |
| 50 | } | 50 | } |
| 51 | } // namespace | 51 | } // namespace |
| 52 | 52 | ||
| @@ -28,7 +28,7 @@ class ChannelManagerUnitTest : public ::testing::Test { | |||
| 28 | ChannelInfo channel_info{}; | 28 | ChannelInfo channel_info{}; |
| 29 | channel_info.channel_type = ChannelType::kServer; | 29 | channel_info.channel_type = ChannelType::kServer; |
| 30 | channel_info.channel_id = kChannelId; | 30 | channel_info.channel_id = kChannelId; |
| 31 | - auto channel = std::make_shared<Channel>(channel_info); | 31 | + auto channel = std::make_shared<CommChannel>(channel_info); |
| 32 | ProtocolHeader header{magic, body_size}; | 32 | ProtocolHeader header{magic, body_size}; |
| 33 | channel->recv_buffer_.resize(sizeof(header)); | 33 | channel->recv_buffer_.resize(sizeof(header)); |
| 34 | memcpy_s(channel->recv_buffer_.data(), sizeof(header), &header, sizeof(header)); | 34 | memcpy_s(channel->recv_buffer_.data(), sizeof(header), &header, sizeof(header)); |
| @@ -79,7 +79,7 @@ TEST_F(ChannelManagerUnitTest, HandleNotifyMessage_WhenStorageLimitExceeded_Retu | |||
| 79 | ChannelInfo channel_info{}; | 79 | ChannelInfo channel_info{}; |
| 80 | channel_info.channel_type = ChannelType::kServer; | 80 | channel_info.channel_type = ChannelType::kServer; |
| 81 | channel_info.channel_id = kChannelId; | 81 | channel_info.channel_id = kChannelId; |
| 82 | - auto channel = std::make_shared<Channel>(channel_info); | 82 | + auto channel = std::make_shared<CommChannel>(channel_info); |
| 83 | 83 | ||
| 84 | // Fill notify messages to reach limit | 84 | // Fill notify messages to reach limit |
| 85 | for (size_t i = 0; i < kMaxNotifyStorageSize; ++i) { | 85 | for (size_t i = 0; i < kMaxNotifyStorageSize; ++i) { |
| @@ -117,7 +117,7 @@ TEST_F(ChannelManagerUnitTest, HandleRequestDisconnectMessage_WhenDisconnectCall | |||
| 117 | ChannelInfo channel_info{}; | 117 | ChannelInfo channel_info{}; |
| 118 | channel_info.channel_type = ChannelType::kServer; | 118 | channel_info.channel_type = ChannelType::kServer; |
| 119 | channel_info.channel_id = kChannelId; | 119 | channel_info.channel_id = kChannelId; |
| 120 | - auto channel = std::make_shared<Channel>(channel_info); | 120 | + auto channel = std::make_shared<CommChannel>(channel_info); |
| 121 | 121 | ||
| 122 | // Verify transfer count is 0 | 122 | // Verify transfer count is 0 |
| 123 | EXPECT_EQ(channel->GetTransferCount(), 0); | 123 | EXPECT_EQ(channel->GetTransferCount(), 0); |
| @@ -156,7 +156,7 @@ TEST_F(ChannelMsgHandlerUnitTest, ProcessServerEviction_WhenClientReturnsError_L | |||
| 156 | ChannelInfo channel_info{}; | 156 | ChannelInfo channel_info{}; |
| 157 | channel_info.channel_type = ChannelType::kServer; | 157 | channel_info.channel_type = ChannelType::kServer; |
| 158 | channel_info.channel_id = kRemoteEngine; | 158 | channel_info.channel_id = kRemoteEngine; |
| 159 | - auto channel = std::make_shared<Channel>(channel_info); | 159 | + auto channel = std::make_shared<CommChannel>(channel_info); |
| 160 | 160 | ||
| 161 | // Create a socket pair for communication | 161 | // Create a socket pair for communication |
| 162 | int socket_pair[2]; | 162 | int socket_pair[2]; |
| @@ -13,7 +13,7 @@ | |||
| 13 | 13 | ||
| 14 | 14 | ||
| 15 | 15 | ||
| 16 | -#include "adxl/channel.h" | 16 | +#include "adxl/comm_channel.h" |
| 17 | 17 | ||
| 18 | 18 | ||
| 19 | 19 | ||
| @@ -14,7 +14,7 @@ | |||
| 14 | 14 | ||
| 15 | 15 | ||
| 16 | 16 | ||
| 17 | -#include "adxl/channel.h" | 17 | +#include "adxl/comm_channel.h" |
| 18 | 18 | ||
| 19 | 19 | ||
| 20 | 20 | ||
| @@ -218,7 +218,7 @@ TEST_F(HixlUTest, TestDeregisterUnregisterMem) { | |||
| 218 | 218 | ||
| 219 | TEST_F(HixlUTest, TestHeartbeat) { | 219 | TEST_F(HixlUTest, TestHeartbeat) { |
| 220 | adxl::ChannelManager::SetHeartbeatWaitTime(10); // 10ms | 220 | adxl::ChannelManager::SetHeartbeatWaitTime(10); // 10ms |
| 221 | - adxl::Channel::SetHeartbeatTimeout(50); // 50ms | 221 | + adxl::CommChannel::SetHeartbeatTimeout(50); // 50ms |
| 222 | Hixl engine1; | 222 | Hixl engine1; |
| 223 | llm::AutoCommResRuntimeMock::SetDevice(0); | 223 | llm::AutoCommResRuntimeMock::SetDevice(0); |
| 224 | std::map<AscendString, AscendString> options1; | 224 | std::map<AscendString, AscendString> options1; |
| @@ -341,7 +341,7 @@ TEST_F(AdxlEngineSTest, TestAdxlEngineFabricMemWithStartAddress) { | |||
| 341 | 341 | ||
| 342 | TEST_F(AdxlEngineSTest, TestHeartbeat) { | 342 | TEST_F(AdxlEngineSTest, TestHeartbeat) { |
| 343 | ChannelManager::SetHeartbeatWaitTime(10); // 10ms | 343 | ChannelManager::SetHeartbeatWaitTime(10); // 10ms |
| 344 | - Channel::SetHeartbeatTimeout(50); // 50ms | 344 | + CommChannel::SetHeartbeatTimeout(50); // 50ms |
| 345 | AdxlEngine engine1; | 345 | AdxlEngine engine1; |
| 346 | llm::AutoCommResRuntimeMock::SetDevice(0); | 346 | llm::AutoCommResRuntimeMock::SetDevice(0); |
| 347 | std::map<AscendString, AscendString> options1; | 347 | std::map<AscendString, AscendString> options1; |
| @@ -375,7 +375,7 @@ TEST_F(HixlSTest, TestHixlDisableBufferPoolD2D) { | |||
| 375 | 375 | ||
| 376 | TEST_F(HixlSTest, TestHeartbeat) { | 376 | TEST_F(HixlSTest, TestHeartbeat) { |
| 377 | adxl::ChannelManager::SetHeartbeatWaitTime(10); // 10ms | 377 | adxl::ChannelManager::SetHeartbeatWaitTime(10); // 10ms |
| 378 | - adxl::Channel::SetHeartbeatTimeout(50); // 50ms | 378 | + adxl::CommChannel::SetHeartbeatTimeout(50); // 50ms |
| 379 | Hixl engine1; | 379 | Hixl engine1; |
| 380 | llm::AutoCommResRuntimeMock::SetDevice(0); | 380 | llm::AutoCommResRuntimeMock::SetDevice(0); |
| 381 | std::map<AscendString, AscendString> options1; | 381 | std::map<AscendString, AscendString> options1; |
| @@ -425,7 +425,7 @@ TEST_F(HixlSTest, TestHixlServerDown) { | |||
| 425 | 425 | ||
| 426 | TEST_F(HixlSTest, TestHixlAutoClearChannel) { | 426 | TEST_F(HixlSTest, TestHixlAutoClearChannel) { |
| 427 | adxl::ChannelManager::SetHeartbeatWaitTime(10); // 10ms | 427 | adxl::ChannelManager::SetHeartbeatWaitTime(10); // 10ms |
| 428 | - adxl::Channel::SetHeartbeatTimeout(50); // 50ms | 428 | + adxl::CommChannel::SetHeartbeatTimeout(50); // 50ms |
| 429 | llm::AutoCommResRuntimeMock::SetDevice(0); | 429 | llm::AutoCommResRuntimeMock::SetDevice(0); |
| 430 | Hixl engine1; | 430 | Hixl engine1; |
| 431 | std::map<AscendString, AscendString> options1; | 431 | std::map<AscendString, AscendString> options1; |
这种日志没必要改