| @@ -88,6 +88,10 @@ def log_info(message): | |||
| 88 | get_console_logger().info(message) | 88 | get_console_logger().info(message) |
| 89 | 89 | ||
| 90 | 90 | ||
| 91 | +def log_warning(message): | ||
| 92 | + get_console_logger().warning(message) | ||
| 93 | + | ||
| 94 | + | ||
| 91 | def log_error(message): | 95 | def log_error(message): |
| 92 | get_console_logger().error(message) | 96 | get_console_logger().error(message) |
| 93 | 97 | ||
| @@ -103,7 +103,7 @@ class ParallelCommandExecutor: | |||
| 103 | # 将输出放入队列(带进程索引) | 103 | # 将输出放入队列(带进程索引) |
| 104 | self.output_queue.put((index, line.strip())) | 104 | self.output_queue.put((index, line.strip())) |
| 105 | except Exception as e: | 105 | except Exception as e: |
| 106 | - self.output_queue.put((index, f"错误捕获输出: {str(e)}")) | 106 | + self.output_queue.put((index, f"error captured output: {str(e)}")) |
| 107 | finally: | 107 | finally: |
| 108 | # 标记进程完成 | 108 | # 标记进程完成 |
| 109 | self.output_queue.put((index, None)) | 109 | self.output_queue.put((index, None)) |
| @@ -223,14 +223,14 @@ class GitSync: | |||
| 223 | 223 | ||
| 224 | remote_commit = self.get_remote_commit(remote_repo) | 224 | remote_commit = self.get_remote_commit(remote_repo) |
| 225 | if not remote_commit: | 225 | if not remote_commit: |
| 226 | - common.log_info("[warning] remote repo is empty, doing full sync") | 226 | + common.log_warning("[warning] remote repo is empty, doing full sync") |
| 227 | self.full_sync(local_repo, remote_repo) | 227 | self.full_sync(local_repo, remote_repo) |
| 228 | return | 228 | return |
| 229 | remote_commit = self.validate_commit(remote_commit) | 229 | remote_commit = self.validate_commit(remote_commit) |
| 230 | 230 | ||
| 231 | changed_files, deleted_files = self.get_changed_files(local_repo, remote_commit) | 231 | changed_files, deleted_files = self.get_changed_files(local_repo, remote_commit) |
| 232 | if not changed_files and not deleted_files: | 232 | if not changed_files and not deleted_files: |
| 233 | - common.log_info("[warning] no change detected, doing nothing") | 233 | + common.log_warning("[warning] no change detected, doing nothing") |
| 234 | return | 234 | return |
| 235 | 235 | ||
| 236 | self.log_changed_files(changed_files, deleted_files) | 236 | self.log_changed_files(changed_files, deleted_files) |
| @@ -320,7 +320,7 @@ class GitSync: | |||
| 320 | self.execute_command(f"mkdir -p {self.quote_remote_path(self.remote_path_join(remote_repo, '.git'))}") | 320 | self.execute_command(f"mkdir -p {self.quote_remote_path(self.remote_path_join(remote_repo, '.git'))}") |
| 321 | self.create_bundle(local_repo, local_bundle_path, remote_commit) | 321 | self.create_bundle(local_repo, local_bundle_path, remote_commit) |
| 322 | if not self.bundle_has_data(local_repo, local_bundle_path): | 322 | if not self.bundle_has_data(local_repo, local_bundle_path): |
| 323 | - common.log_info("[warning] bundle is empty, doing nothing") | 323 | + common.log_warning("[warning] bundle is empty, doing nothing") |
| 324 | os.remove(local_bundle_path) | 324 | os.remove(local_bundle_path) |
| 325 | return | 325 | return |
| 326 | 326 | ||
| @@ -416,7 +416,7 @@ if __name__ == "__main__": | |||
| 416 | # 使用git bundle进行高效同步 | 416 | # 使用git bundle进行高效同步 |
| 417 | sync.sync_with_bundle(local_repo, remote_repo) | 417 | sync.sync_with_bundle(local_repo, remote_repo) |
| 418 | except KeyboardInterrupt: | 418 | except KeyboardInterrupt: |
| 419 | - common.log_info("[warning] interrupted by user") | 419 | + common.log_warning("[warning] interrupted by user") |
| 420 | except paramiko.ssh_exception.NoValidConnectionsError: | 420 | except paramiko.ssh_exception.NoValidConnectionsError: |
| 421 | common.log_error("[error] no valid connections") | 421 | common.log_error("[error] no valid connections") |
| 422 | except (paramiko.SSHException, OSError, RuntimeError, ValueError) as err: | 422 | except (paramiko.SSHException, OSError, RuntimeError, ValueError) as err: |
| @@ -538,7 +538,7 @@ int main(int argc, char** argv) | |||
| 538 | if (!lat_file.is_open() || !pass_rate_file.is_open() || !bad_link_file.is_open() | 538 | if (!lat_file.is_open() || !pass_rate_file.is_open() || !bad_link_file.is_open() |
| 539 | || !bad_link_candidate_file.is_open() | 539 | || !bad_link_candidate_file.is_open() |
| 540 | || (enable_l2_pingpong && (!l2_latency_file.is_open() || !l2_passrate_file.is_open()))) { | 540 | || (enable_l2_pingpong && (!l2_latency_file.is_open() || !l2_passrate_file.is_open()))) { |
| 541 | - std::cerr << "无法打开输出文件!" << std::endl; | 541 | + std::cerr << "failed to open output file!" << std::endl; |
[LOW] 输出文件打开失败日志未带文件路径,6 个文件无法定位是哪个打开失败 本次把中文文案改为英文( 建议修复:带上路径,例如:
或拆开逐文件判定各自打印路径(link_lat_path 等变量在函数前文已存在,直接可用)。第 541 行:failed to open output file! 另(PR 完备性):PR 描述「关联的Issue」为 NA,仓内 CONTRIBUTING.md 第 22 条要求所有 PR 必须关联 Issue;本批 296 项整改建议补建跟踪 Issue 并关联,便于记录整改范围与剩余分批计划。 ![]() ![]() | |||
| 542 | return -1; | 542 | return -1; |
| 543 | } | 543 | } |
| 544 | lat_file << std::fixed << std::setprecision(3); | 544 | lat_file << std::fixed << std::setprecision(3); |
| @@ -309,7 +309,7 @@ void ProbeHelper::mesh_full_trace() | |||
| 309 | for (auto& this_trace : trace_ports_res) { | 309 | for (auto& this_trace : trace_ports_res) { |
| 310 | int len = this_trace.size(); | 310 | int len = this_trace.size(); |
| 311 | if ((len != level_trace_len[link_level]) && (len + 1 != level_trace_len[link_level])) { | 311 | if ((len != level_trace_len[link_level]) && (len + 1 != level_trace_len[link_level])) { |
| 312 | - std::cout << "[probe_topo][Warning] there are path len error!"; | 312 | + std::cout << "[probe_topo][Warning] path length error detected!"; |
| 313 | this_trace.resize(level_trace_len[link_level]); | 313 | this_trace.resize(level_trace_len[link_level]); |
| 314 | } else if (len % 2 == 0) { | 314 | } else if (len % 2 == 0) { |
| 315 | mesh.add_kid_link( | 315 | mesh.add_kid_link( |
| @@ -445,7 +445,7 @@ void ProbeHelper::addition_tracert_for_pinglist() | |||
| 445 | for (auto& this_trace : trace_ports_res) { | 445 | for (auto& this_trace : trace_ports_res) { |
| 446 | int len = this_trace.size(); | 446 | int len = this_trace.size(); |
| 447 | if ((len != level_trace_len[link_level]) && (len + 1 != level_trace_len[link_level])) { | 447 | if ((len != level_trace_len[link_level]) && (len + 1 != level_trace_len[link_level])) { |
| 448 | - std::cout << "[probe_topo][Warning] there are path len error!"; | 448 | + std::cout << "[probe_topo][Warning] path length error detected!"; |
| 449 | this_trace.resize(level_trace_len[link_level]); | 449 | this_trace.resize(level_trace_len[link_level]); |
| 450 | } else if (len % 2 == 0) { | 450 | } else if (len % 2 == 0) { |
| 451 | mesh.add_kid_link( | 451 | mesh.add_kid_link( |
| @@ -466,7 +466,7 @@ void ProbeHelper::gen_id() | |||
| 466 | { | 466 | { |
| 467 | if (Mesh::get_unfinished_link() > 0) { | 467 | if (Mesh::get_unfinished_link() > 0) { |
| 468 | // throw std::runtime_error("[probe_topo] there are unfinished link"); | 468 | // throw std::runtime_error("[probe_topo] there are unfinished link"); |
| 469 | - std::cout << "[probe_topo][Warning] there are unfinished link" << std::endl; | 469 | + std::cout << "[probe_topo][Warning] there are unfinished links" << std::endl; |
| 470 | } | 470 | } |
| 471 | auto& tracert_res_id = meshTopo.tracert_res_id; | 471 | auto& tracert_res_id = meshTopo.tracert_res_id; |
| 472 | /*生成tracert_res_id(探测结果从ip_list转化为id_list)*/ | 472 | /*生成tracert_res_id(探测结果从ip_list转化为id_list)*/ |
| @@ -54,7 +54,7 @@ HcclResult DlUrmaFunction::DlUrmaFunctionInit() | |||
| 54 | handle_ == nullptr, | 54 | handle_ == nullptr, |
| 55 | HCCL_ERROR( | 55 | HCCL_ERROR( |
| 56 | "dlopen [%s] failed, %s", "liburma.so.0", | 56 | "dlopen [%s] failed, %s", "liburma.so.0", |
| 57 | - (errMsg == nullptr) ? "please check the file exist or permission denied." : errMsg), | 57 | + (errMsg == nullptr) ? "Please verify that the file exists and that you have permission." : errMsg), |
| 58 | HCCL_E_OPEN_FILE_FAILURE); | 58 | HCCL_E_OPEN_FILE_FAILURE); |
| 59 | } | 59 | } |
| 60 | 60 | ||
| @@ -98,7 +98,7 @@ int32_t HcommLocalReduceOnThread( | |||
| 98 | CHK_PRT_RET( | 98 | CHK_PRT_RET( |
| 99 | (IsSupportReduce(dataType, reduceOp) == false), | 99 | (IsSupportReduce(dataType, reduceOp) == false), |
| 100 | HCCL_ERROR( | 100 | HCCL_ERROR( |
| 101 | - "[HcommLocalReduceOnThread]Not support reduce, " | 101 | + "[HcommLocalReduceOnThread] Not support reduce, " |
| 102 | "dst[%p], src[%p], count[%llu], dataType[%d], reduceOp[%d]", | 102 | "dst[%p], src[%p], count[%llu], dataType[%d], reduceOp[%d]", |
| 103 | dst, src, count, dataType, reduceOp), | 103 | dst, src, count, dataType, reduceOp), |
| 104 | HCCL_E_PARA); | 104 | HCCL_E_PARA); |
| @@ -351,7 +351,7 @@ int32_t HcommWriteReduceOnThread( | |||
| 351 | CHK_PRT_RET( | 351 | CHK_PRT_RET( |
| 352 | (IsSupportReduce(dataType, reduceOp) == false), | 352 | (IsSupportReduce(dataType, reduceOp) == false), |
| 353 | HCCL_ERROR( | 353 | HCCL_ERROR( |
| 354 | - "[HcommWriteReduceOnThread]Not support reduce, " | 354 | + "[HcommWriteReduceOnThread] Not support reduce, " |
| 355 | "dst[%p], src[%p], count[%llu], dataType[%d], reduceOp[%d]", | 355 | "dst[%p], src[%p], count[%llu], dataType[%d], reduceOp[%d]", |
| 356 | dst, src, count, dataType, reduceOp), | 356 | dst, src, count, dataType, reduceOp), |
| 357 | HCCL_E_PARA); | 357 | HCCL_E_PARA); |
| @@ -391,7 +391,7 @@ HcclResult CommWriteReduceWithNotify( | |||
| 391 | CHK_PRT_RET( | 391 | CHK_PRT_RET( |
| 392 | (IsSupportReduce(dataType, reduceOp) == false), | 392 | (IsSupportReduce(dataType, reduceOp) == false), |
| 393 | HCCL_ERROR( | 393 | HCCL_ERROR( |
| 394 | - "[CommWriteReduceWithNotify]Not support reduce, " | 394 | + "[CommWriteReduceWithNotify] Not support reduce, " |
| 395 | "dst[%p], src[%p], count[%llu], dataType[%d], reduceOp[%d]", | 395 | "dst[%p], src[%p], count[%llu], dataType[%d], reduceOp[%d]", |
| 396 | dst, src, count, dataType, reduceOp), | 396 | dst, src, count, dataType, reduceOp), |
| 397 | HCCL_E_PARA); | 397 | HCCL_E_PARA); |
| @@ -546,7 +546,7 @@ int32_t HcommReadReduceOnThread( | |||
| 546 | CHK_PRT_RET( | 546 | CHK_PRT_RET( |
| 547 | (IsSupportReduce(dataType, reduceOp) == false), | 547 | (IsSupportReduce(dataType, reduceOp) == false), |
| 548 | HCCL_ERROR( | 548 | HCCL_ERROR( |
| 549 | - "[HcommReadReduceOnThread]Not support reduce, " | 549 | + "[HcommReadReduceOnThread] Not support reduce, " |
| 550 | "dst[%p], src[%p], count[%llu], dataType[%d], reduceOp[%d]", | 550 | "dst[%p], src[%p], count[%llu], dataType[%d], reduceOp[%d]", |
| 551 | dst, src, count, dataType, reduceOp), | 551 | dst, src, count, dataType, reduceOp), |
| 552 | HCCL_E_PARA); | 552 | HCCL_E_PARA); |
| @@ -152,7 +152,7 @@ static HcclResult GetSqeBuffVa(const uint64_t ccuResBaseVa, const uint32_t jetty | |||
| 152 | if (UINT32_MAX / CCU_SQ_BUFFER_SIZE < jettyCtxId) { | 152 | if (UINT32_MAX / CCU_SQ_BUFFER_SIZE < jettyCtxId) { |
| 153 | HCCL_ERROR( | 153 | HCCL_ERROR( |
| 154 | "[CcuJettyCtxMgr][%s] jetty context id[%u] is greater " | 154 | "[CcuJettyCtxMgr][%s] jetty context id[%u] is greater " |
| 155 | - "than expected, CCU_SQ_UBFFER_SIZE[%u], their product will exceed the " | 155 | + "than expected, CCU_SQ_BUFFER_SIZE[%u], their product will exceed the " |
| 156 | "range of uint32_t.", | 156 | "range of uint32_t.", |
| 157 | __func__, jettyCtxId, CCU_SQ_BUFFER_SIZE); | 157 | __func__, jettyCtxId, CCU_SQ_BUFFER_SIZE); |
| 158 | return HcclResult::HCCL_E_INTERNAL; | 158 | return HcclResult::HCCL_E_INTERNAL; |
| @@ -215,7 +215,7 @@ namespace CcuRep { | |||
| 215 | } | 215 | } |
| 216 | if (ccuRepBufReduce->GetCount() > CCU_REDUCE_MAX_MS || ccuRepBufReduce->GetMem().size() > CCU_REDUCE_MAX_MS) { | 216 | if (ccuRepBufReduce->GetCount() > CCU_REDUCE_MAX_MS || ccuRepBufReduce->GetMem().size() > CCU_REDUCE_MAX_MS) { |
| 217 | HCCL_ERROR( | 217 | HCCL_ERROR( |
| 218 | - "count[%u] and mem size[%zu] must less than %u", ccuRepBufReduce->GetCount(), | 218 | + "count[%u] and mem size[%zu] must be less than %u", ccuRepBufReduce->GetCount(), |
[MEDIUM] must less than 语法错误整改遗漏(与本 PR 已改的 ccu_ins_generator_v1 同构校验) 本 PR 把 第 40 行: 两处是同一个 CCU BUF_REDUCE 表示的参数校验(translator 侧与 data 侧各一份),同 PR 只改一侧会造成同类告警残留。 建议修复:
顺带建议同批统一该异常文案风格(当前 translator 侧带 ![]() ![]() | |||
| 219 | ccuRepBufReduce->GetMem().size(), CCU_REDUCE_MAX_MS); | 219 | ccuRepBufReduce->GetMem().size(), CCU_REDUCE_MAX_MS); |
| 220 | return HCCL_E_PARA; | 220 | return HCCL_E_PARA; |
| 221 | } | 221 | } |
| @@ -367,7 +367,7 @@ HcclResult AicpuLaunchMgr::NotifyKernelLaunchFree( | |||
| 367 | 367 | ||
| 368 | CHK_RET(LaunchNotifyKernel(opParam, binCustomHandle)); | 368 | CHK_RET(LaunchNotifyKernel(opParam, binCustomHandle)); |
| 369 | HCCL_RUN_INFO( | 369 | HCCL_RUN_INFO( |
| 370 | - "[AicpuLaunchMgr][%s] notify free kernalLaunch success, commid[%s], notifyNum[%u]", __func__, commId.c_str(), | 370 | + "[AicpuLaunchMgr][%s] notify free kernelLaunch success, commid[%s], notifyNum[%u]", __func__, commId.c_str(), |
| 371 | aicpuNotifys.size()); | 371 | aicpuNotifys.size()); |
| 372 | return HCCL_SUCCESS; | 372 | return HCCL_SUCCESS; |
| 373 | } | 373 | } |
| @@ -79,7 +79,7 @@ HcclResult AicpuTsChannelHelper::LaunchKernel( | |||
| 79 | const HcommChannelDesc* channelDescs, aclrtBinHandle binHandle) | 79 | const HcommChannelDesc* channelDescs, aclrtBinHandle binHandle) |
| 80 | { | 80 | { |
| 81 | CHK_PTR_NULL(channelList); | 81 | CHK_PTR_NULL(channelList); |
| 82 | - CHK_PRT_RET((listNum == 0), HCCL_ERROR("[%s]Invalid listNum, listNum[%u]", __func__, listNum), HCCL_E_PARA); | 82 | + CHK_PRT_RET((listNum == 0), HCCL_ERROR("[%s] Invalid listNum, listNum[%u]", __func__, listNum), HCCL_E_PARA); |
| 83 | 83 | ||
| 84 | // 过滤出未就绪的子集,避免重复下 kernel 导致 device 侧 channel 对象泄漏 | 84 | // 过滤出未就绪的子集,避免重复下 kernel 导致 device 侧 channel 对象泄漏 |
| 85 | std::vector<ChannelHandle> subHostHandles; | 85 | std::vector<ChannelHandle> subHostHandles; |
| @@ -157,7 +157,7 @@ HcclResult AicpuTsChannelHelper::PreAllocChannels( | |||
| 157 | CHK_PTR_NULL(targetChannels); | 157 | CHK_PTR_NULL(targetChannels); |
| 158 | CHK_PTR_NULL(userChannels); | 158 | CHK_PTR_NULL(userChannels); |
| 159 | CHK_PRT_RET( | 159 | CHK_PRT_RET( |
| 160 | - (channelNum == 0), HCCL_ERROR("[%s]Invalid channelNum, channelNum[%u]", __func__, channelNum), HCCL_E_PARA); | 160 | + (channelNum == 0), HCCL_ERROR("[%s] Invalid channelNum, channelNum[%u]", __func__, channelNum), HCCL_E_PARA); |
| 161 | 161 | ||
| 162 | for (uint32_t i = 0; i < channelNum; i++) { | 162 | for (uint32_t i = 0; i < channelNum; i++) { |
| 163 | auto* channel = reinterpret_cast<Channel*>(targetChannels[i]); | 163 | auto* channel = reinterpret_cast<Channel*>(targetChannels[i]); |
| @@ -33,7 +33,7 @@ HcclResult DevRdmaConnectionV2::Init() | |||
| 33 | { | 33 | { |
| 34 | if (rdmaConnStatus_ != RdmaConnStatus::CLOSED) { | 34 | if (rdmaConnStatus_ != RdmaConnStatus::CLOSED) { |
| 35 | HCCL_INFO( | 35 | HCCL_INFO( |
| 36 | - "[DevRdmaConnectionV2][%s] status[%s] is not need init.", __func__, rdmaConnStatus_.Describe().c_str()); | 36 | + "[DevRdmaConnectionV2][%s] status[%s] does not need init.", __func__, rdmaConnStatus_.Describe().c_str()); |
| 37 | return HCCL_SUCCESS; | 37 | return HCCL_SUCCESS; |
| 38 | } | 38 | } |
| 39 | 39 | ||
| @@ -168,7 +168,7 @@ HcclResult DevRdmaConnectionV2::GetDmaMode() | |||
| 168 | HcclResult DevRdmaConnectionV2::CreateQp() | 168 | HcclResult DevRdmaConnectionV2::CreateQp() |
| 169 | { | 169 | { |
| 170 | if (socket_->GetStatus() != Hccl::SocketStatus::OK) { | 170 | if (socket_->GetStatus() != Hccl::SocketStatus::OK) { |
| 171 | - HCCL_WARNING("[DevRdmaConnectionV2::CreateQp] socket status is not ok, please"); | 171 | + HCCL_WARNING("[DevRdmaConnectionV2::CreateQp] socket status is not ok, please check socket connection."); |
| 172 | return HCCL_E_AGAIN; | 172 | return HCCL_E_AGAIN; |
| 173 | } | 173 | } |
| 174 | 174 | ||
| @@ -280,7 +280,7 @@ HcclResult DevRdmaConnectionV2::ModifyQp() | |||
| 280 | } | 280 | } |
| 281 | 281 | ||
| 282 | if (!rmtQpAttr_.IsValid()) { | 282 | if (!rmtQpAttr_.IsValid()) { |
| 283 | - HCCL_ERROR("[DevRdmaConnectionV2][%s] romate Qp Attr is empty, exchange qp attr first", __func__); | 283 | + HCCL_ERROR("[DevRdmaConnectionV2][%s] remote Qp Attr is empty, exchange qp attr first", __func__); |
| 284 | return HCCL_E_INTERNAL; | 284 | return HCCL_E_INTERNAL; |
| 285 | } | 285 | } |
| 286 | 286 | ||
| @@ -72,7 +72,7 @@ std::string HostRdmaConnection::Describe() const | |||
| 72 | HcclResult HostRdmaConnection::CreateQp() | 72 | HcclResult HostRdmaConnection::CreateQp() |
| 73 | { | 73 | { |
| 74 | if (socket_->GetStatus() != Hccl::SocketStatus::OK) { | 74 | if (socket_->GetStatus() != Hccl::SocketStatus::OK) { |
| 75 | - HCCL_WARNING("[HostRdmaConnection::CreateQp] socket status is not ok, please"); | 75 | + HCCL_WARNING("[HostRdmaConnection::CreateQp] socket status is not ok, please check socket connection."); |
| 76 | return HCCL_E_AGAIN; | 76 | return HCCL_E_AGAIN; |
| 77 | } | 77 | } |
| 78 | 78 | ||
| @@ -221,7 +221,7 @@ HcclResult HostRdmaConnection::ModifyQp() | |||
| 221 | } | 221 | } |
| 222 | 222 | ||
| 223 | if (!rmtQpAttr_.IsValid()) { | 223 | if (!rmtQpAttr_.IsValid()) { |
| 224 | - HCCL_ERROR("[HostRdmaConnection][%s] romate Qp Attr is empty, exchange qp attr first", __func__); | 224 | + HCCL_ERROR("[HostRdmaConnection][%s] remote Qp Attr is empty, exchange qp attr first", __func__); |
| 225 | return HCCL_E_INTERNAL; | 225 | return HCCL_E_INTERNAL; |
| 226 | } | 226 | } |
| 227 | 227 | ||
| @@ -51,7 +51,7 @@ void BkfDcItorUninit(BkfDc *dc) | |||
| 51 | if (dc == VOS_NULL) { | 51 | if (dc == VOS_NULL) { |
| 52 | return; | 52 | return; |
| 53 | } | 53 | } |
| 54 | - BKF_LOG_DEBUG(BKF_LOG_HND, "2unit\n"); | 54 | + BKF_LOG_DEBUG(BKF_LOG_HND, "dc itor uninit done\n"); |
| 55 | 55 | ||
| 56 | return; | 56 | return; |
| 57 | } | 57 | } |
| @@ -574,7 +574,7 @@ STATIC BkfDispFunc *BkfDispProcParseDispFunc(BkfDisp *disp, char **inStr, uint8_ | |||
| 574 | dispFunc = BkfDispFindFunc(disp, funcName); | 574 | dispFunc = BkfDispFindFunc(disp, funcName); |
| 575 | if ((dispFunc == VOS_NULL) || !BKF_DISP_FUNC_MAY_USE(disp, dispFunc)) { | 575 | if ((dispFunc == VOS_NULL) || !BKF_DISP_FUNC_MAY_USE(disp, dispFunc)) { |
| 576 | BKF_DISP_INC_CNT(disp, BKF_DISP_CNT_PROC_NO_VALID_FUNC_FIND); | 576 | BKF_DISP_INC_CNT(disp, BKF_DISP_CNT_PROC_NO_VALID_FUNC_FIND); |
| 577 | - BKF_DISP_PRINTF(disp, "[%s(^)], no valid func find, BkfDispHelp for help\n", funcName); | 577 | + BKF_DISP_PRINTF(disp, "[%s(^)], no valid func found, BkfDispHelp for help\n", funcName); |
| 578 | return VOS_NULL; | 578 | return VOS_NULL; |
| 579 | } | 579 | } |
| 580 | 580 | ||
| @@ -1659,7 +1659,7 @@ void BkfDispFuncObjNameFunc(BkfDisp *disp, char *funcName) | |||
| 1659 | /* func */ | 1659 | /* func */ |
| 1660 | dispFunc = BkfDispFindFunc(disp, funcName); | 1660 | dispFunc = BkfDispFindFunc(disp, funcName); |
| 1661 | if ((dispFunc == VOS_NULL) || !BKF_DISP_FUNC_MAY_USE(disp, dispFunc)) { | 1661 | if ((dispFunc == VOS_NULL) || !BKF_DISP_FUNC_MAY_USE(disp, dispFunc)) { |
| 1662 | - BKF_DISP_PRINTF(disp, "%s, func not find\n", funcName); | 1662 | + BKF_DISP_PRINTF(disp, "%s, func not found\n", funcName); |
| 1663 | return; | 1663 | return; |
| 1664 | } | 1664 | } |
| 1665 | 1665 | ||
| @@ -233,7 +233,7 @@ STATIC void BkfChCliLetcpDisconn(BkfChCli *ch, BkfChCliConnId *connId) | |||
| 233 | BKF_RETURNvoid_IF(ch == VOS_NULL); | 233 | BKF_RETURNvoid_IF(ch == VOS_NULL); |
| 234 | BKF_RETURNvoid_IF(connId == VOS_NULL); | 234 | BKF_RETURNvoid_IF(connId == VOS_NULL); |
| 235 | BKF_RETURNvoid_IF(connId->ch == VOS_NULL); | 235 | BKF_RETURNvoid_IF(connId->ch == VOS_NULL); |
| 236 | - BKF_LOG_INFO(BKF_LOG_HND, "ch(%#x)/serUrl(%s)/localUrl(%s), fd %d, conndId %#x\n", BKF_MASK_ADDR(ch), | 236 | + BKF_LOG_INFO(BKF_LOG_HND, "ch(%#x)/serUrl(%s)/localUrl(%s), fd %d, connId %#x\n", BKF_MASK_ADDR(ch), |
| 237 | BkfUrlGetStr(&connId->keyUrlSer, buf, sizeof(buf)), BkfUrlGetStr(&connId->keyUrlCli, buf2, sizeof(buf2)), | 237 | BkfUrlGetStr(&connId->keyUrlSer, buf, sizeof(buf)), BkfUrlGetStr(&connId->keyUrlCli, buf2, sizeof(buf2)), |
| 238 | connId->connFd, BKF_MASK_ADDR(connId)); | 238 | connId->connFd, BKF_MASK_ADDR(connId)); |
| 239 | 239 | ||
| @@ -281,7 +281,7 @@ STATIC int32_t BkfChCliLetcpRead(BkfChCliConnId *connId, void *dataBuf, int32_t | |||
| 281 | uint8_t buf[BKF_1K / 8]; | 281 | uint8_t buf[BKF_1K / 8]; |
| 282 | uint8_t buf2[BKF_1K / 8]; | 282 | uint8_t buf2[BKF_1K / 8]; |
| 283 | 283 | ||
| 284 | - BKF_LOG_INFO(BKF_LOG_HND, "LetcpRead serUrl(%s)/localUrl(%s), conndId %#x\n", | 284 | + BKF_LOG_INFO(BKF_LOG_HND, "LetcpRead serUrl(%s)/localUrl(%s), connId %#x\n", |
| 285 | BkfUrlGetStr(&connId->keyUrlSer, buf, sizeof(buf)), BkfUrlGetStr(&connId->keyUrlCli, buf2, sizeof(buf2)), | 285 | BkfUrlGetStr(&connId->keyUrlSer, buf, sizeof(buf)), BkfUrlGetStr(&connId->keyUrlCli, buf2, sizeof(buf2)), |
| 286 | BKF_MASK_ADDR(connId)); | 286 | BKF_MASK_ADDR(connId)); |
| 287 | int32_t ret = recv(connId->connFd, dataBuf, bufLen, 0); | 287 | int32_t ret = recv(connId->connFd, dataBuf, bufLen, 0); |
| @@ -299,7 +299,7 @@ STATIC int32_t BkfChCliLetcpSend(BkfChCliConnId *connId, void *dataBuf, int32_t | |||
| 299 | uint8_t buf[BKF_1K / 8]; | 299 | uint8_t buf[BKF_1K / 8]; |
| 300 | uint8_t buf2[BKF_1K / 8]; | 300 | uint8_t buf2[BKF_1K / 8]; |
| 301 | 301 | ||
| 302 | - BKF_LOG_INFO(BKF_LOG_HND, "LetcpSend serUrl(%s)/localUrl(%s), conndId %#x\n", | 302 | + BKF_LOG_INFO(BKF_LOG_HND, "LetcpSend serUrl(%s)/localUrl(%s), connId %#x\n", |
| 303 | BkfUrlGetStr(&connId->keyUrlSer, buf, sizeof(buf)), BkfUrlGetStr(&connId->keyUrlCli, buf2, sizeof(buf2)), | 303 | BkfUrlGetStr(&connId->keyUrlSer, buf, sizeof(buf)), BkfUrlGetStr(&connId->keyUrlCli, buf2, sizeof(buf2)), |
| 304 | BKF_MASK_ADDR(connId)); | 304 | BKF_MASK_ADDR(connId)); |
| 305 | int32_t ret = send(connId->connFd, dataBuf, dataLen, 0); | 305 | int32_t ret = send(connId->connFd, dataBuf, dataLen, 0); |
| @@ -614,7 +614,7 @@ STATIC uint32_t BkfChCliLetcpConnToSer(BkfChCliConnId *connId) | |||
| 614 | uint8_t buf[BKF_1K / 8]; | 614 | uint8_t buf[BKF_1K / 8]; |
| 615 | uint8_t buf2[BKF_1K / 8]; | 615 | uint8_t buf2[BKF_1K / 8]; |
| 616 | 616 | ||
| 617 | - BKF_LOG_INFO(BKF_LOG_HND, "ConnToSer serUrl(%s)/localUrl(%s), conndId %#x\n", | 617 | + BKF_LOG_INFO(BKF_LOG_HND, "ConnToSer serUrl(%s)/localUrl(%s), connId %#x\n", |
| 618 | BkfUrlGetStr(&connId->keyUrlSer, buf, sizeof(buf)), BkfUrlGetStr(&connId->keyUrlCli, buf2, sizeof(buf2)), | 618 | BkfUrlGetStr(&connId->keyUrlSer, buf, sizeof(buf)), BkfUrlGetStr(&connId->keyUrlCli, buf2, sizeof(buf2)), |
| 619 | BKF_MASK_ADDR(connId)); | 619 | BKF_MASK_ADDR(connId)); |
| 620 | 620 | ||
| @@ -37,7 +37,7 @@ int RaRsTlvInitV1(char *inBuf, char *outBuf, int *outLen, int *opResult, int rcv | |||
| 37 | (void)outBuf; | 37 | (void)outBuf; |
| 38 | (void)outLen; | 38 | (void)outLen; |
| 39 | (void)rcvBufLen; | 39 | (void)rcvBufLen; |
| 40 | - hccp_warn("Tlv init is not support in this version."); | 40 | + hccp_warn("Tlv init is not supported in this version."); |
| 41 | *opResult = -ENOTSUPP; | 41 | *opResult = -ENOTSUPP; |
| 42 | return 0; | 42 | return 0; |
| 43 | } | 43 | } |
| @@ -26,8 +26,8 @@ HCCP_ATTRI_VISI_DEF int RaGetEidByIpAsync(void *ctxHandle, struct IpInfo ip[], u | |||
| 26 | ConverReturnCode(RDMA_OP, -EINVAL)); | 26 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 27 | 27 | ||
| 28 | CHK_PRT_RETURN(*num == 0 || *num > HCCP_EID_IP_QUERY_MAX_NUM, | 28 | CHK_PRT_RETURN(*num == 0 || *num > HCCP_EID_IP_QUERY_MAX_NUM, |
| 29 | - hccp_err("[get][eid_by_ip]num(%u) must greater than 0" | 29 | + hccp_err("[get][eid_by_ip]num(%u) must be greater than 0" |
| 30 | - " and less or equal to %d", | 30 | + " and less than or equal to %d", |
| 31 | *num, HCCP_EID_IP_QUERY_MAX_NUM), | 31 | *num, HCCP_EID_IP_QUERY_MAX_NUM), |
| 32 | ConverReturnCode(RDMA_OP, -EINVAL)); | 32 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 33 | 33 | ||
| @@ -55,8 +55,8 @@ HCCP_ATTRI_VISI_DEF int RaGetIpByEidAsync(void *ctxHandle, union HccpEid eid[], | |||
| 55 | ConverReturnCode(RDMA_OP, -EINVAL)); | 55 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 56 | 56 | ||
| 57 | CHK_PRT_RETURN(*num == 0 || *num > HCCP_EID_IP_QUERY_MAX_NUM, | 57 | CHK_PRT_RETURN(*num == 0 || *num > HCCP_EID_IP_QUERY_MAX_NUM, |
| 58 | - hccp_err("[get][IpByEid]num(%u) must greater than 0" | 58 | + hccp_err("[get][IpByEid]num(%u) must be greater than 0" |
| 59 | - " and less or equal to %d", | 59 | + " and less than or equal to %d", |
| 60 | *num, HCCP_EID_IP_QUERY_MAX_NUM), | 60 | *num, HCCP_EID_IP_QUERY_MAX_NUM), |
| 61 | ConverReturnCode(RDMA_OP, -EINVAL)); | 61 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 62 | 62 | ||
| @@ -76,7 +76,7 @@ HCCP_ATTRI_VISI_DEF int RaIsLastUsed(int insId) | |||
| 76 | HCCP_ATTRI_VISI_DEF int RaRdevGetHandle(unsigned int phyId, void **rdmaHandle) | 76 | HCCP_ATTRI_VISI_DEF int RaRdevGetHandle(unsigned int phyId, void **rdmaHandle) |
| 77 | { | 77 | { |
| 78 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 78 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 79 | - hccp_err("[get][ra_rdev]phyId(%u) must smaller than %u", phyId, RA_MAX_PHY_ID_NUM), -EINVAL); | 79 | + hccp_err("[get][ra_rdev]phyId(%u) must be smaller than %u", phyId, RA_MAX_PHY_ID_NUM), -EINVAL); |
| 80 | CHK_PRT_RETURN(rdmaHandle == NULL, hccp_err("[get][ra_rdev]rdma_handle is NULL, phyId(%u)", phyId), -EINVAL); | 80 | CHK_PRT_RETURN(rdmaHandle == NULL, hccp_err("[get][ra_rdev]rdma_handle is NULL, phyId(%u)", phyId), -EINVAL); |
| 81 | CHK_PRT_RETURN(gRaRdevHandle[phyId] == NULL, hccp_run_info("[get][ra_rdev]handle is NULL, phyId(%u)", phyId), | 81 | CHK_PRT_RETURN(gRaRdevHandle[phyId] == NULL, hccp_run_info("[get][ra_rdev]handle is NULL, phyId(%u)", phyId), |
| 82 | -ENODEV); | 82 | -ENODEV); |
| @@ -88,7 +88,7 @@ HCCP_ATTRI_VISI_DEF int RaRdevGetHandle(unsigned int phyId, void **rdmaHandle) | |||
| 88 | void RaRdevSetHandle(unsigned int phyId, void *rdmaHandle) | 88 | void RaRdevSetHandle(unsigned int phyId, void *rdmaHandle) |
| 89 | { | 89 | { |
| 90 | if (phyId >= RA_MAX_PHY_ID_NUM) { | 90 | if (phyId >= RA_MAX_PHY_ID_NUM) { |
| 91 | - hccp_warn("[set][ra_rdev]phyId(%u) must smaller than %u", phyId, RA_MAX_PHY_ID_NUM); | 91 | + hccp_warn("[set][ra_rdev]phyId(%u) must be smaller than %u", phyId, RA_MAX_PHY_ID_NUM); |
| 92 | return; | 92 | return; |
| 93 | } | 93 | } |
| 94 | 94 | ||
| @@ -153,7 +153,7 @@ HCCP_ATTRI_VISI_DEF int RaInit(struct RaInitConfig *config) | |||
| 153 | 153 | ||
| 154 | phyId = config->phyId; | 154 | phyId = config->phyId; |
| 155 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 155 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 156 | - hccp_err("[init][ra]phyId(%u) is invalid! it must greater or " | 156 | + hccp_err("[init][ra]phyId(%u) is invalid! it must be greater than or " |
| 157 | "equal to 0 and less than %d!", | 157 | "equal to 0 and less than %d!", |
| 158 | phyId, RA_MAX_PHY_ID_NUM), | 158 | phyId, RA_MAX_PHY_ID_NUM), |
| 159 | ConverReturnCode(HCCP_INIT, -EINVAL)); | 159 | ConverReturnCode(HCCP_INIT, -EINVAL)); |
| @@ -217,7 +217,7 @@ HCCP_ATTRI_VISI_DEF int RaDeinit(struct RaInitConfig *config) | |||
| 217 | 217 | ||
| 218 | phyId = config->phyId; | 218 | phyId = config->phyId; |
| 219 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 219 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 220 | - hccp_err("[deinit][ra]phyId(%u) is invalid! it must greater or equal to 0 and less than %d!", phyId, | 220 | + hccp_err("[deinit][ra]phyId(%u) is invalid! it must be greater than or equal to 0 and less than %d!", phyId, |
| 221 | RA_MAX_PHY_ID_NUM), | 221 | RA_MAX_PHY_ID_NUM), |
| 222 | ConverReturnCode(HCCP_INIT, -EINVAL)); | 222 | ConverReturnCode(HCCP_INIT, -EINVAL)); |
| 223 | 223 | ||
| @@ -34,7 +34,7 @@ STATIC int RaUdevInitCheck(unsigned int phyId, void *pingHandle) | |||
| 34 | { | 34 | { |
| 35 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 35 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 36 | hccp_err("[check][ra_ping_init]phyId(%u) is invalid! " | 36 | hccp_err("[check][ra_ping_init]phyId(%u) is invalid! " |
| 37 | - "it must greater or equal to 0 and less than %d!", | 37 | + "it must be greater than or equal to 0 and less than %d!", |
| 38 | phyId, RA_MAX_PHY_ID_NUM), | 38 | phyId, RA_MAX_PHY_ID_NUM), |
| 39 | -EINVAL); | 39 | -EINVAL); |
| 40 | CHK_PRT_RETURN(pingHandle == NULL, hccp_err("[check][ra_ping_init]phyId(%u) ping_handle is null!", phyId), -EINVAL); | 40 | CHK_PRT_RETURN(pingHandle == NULL, hccp_err("[check][ra_ping_init]phyId(%u) ping_handle is null!", phyId), -EINVAL); |
| @@ -83,7 +83,7 @@ STATIC int RaPingInitGetHandle(struct PingInitAttr *initAttr, struct PingInitInf | |||
| 83 | pingHandle->bufferSize = initAttr->bufferSize; | 83 | pingHandle->bufferSize = initAttr->bufferSize; |
| 84 | 84 | ||
| 85 | ret = pthread_mutex_init(&pingHandle->mutex, NULL); | 85 | ret = pthread_mutex_init(&pingHandle->mutex, NULL); |
| 86 | - CHK_PRT_RETURN(ret, hccp_err("[init][ra_ping]init mutext failed, ret:%d", ret), ret); | 86 | + CHK_PRT_RETURN(ret, hccp_err("[init][ra_ping]init mutex failed, ret:%d", ret), ret); |
| 87 | 87 | ||
| 88 | return 0; | 88 | return 0; |
| 89 | } | 89 | } |
| @@ -138,7 +138,7 @@ HCCP_ATTRI_VISI_DEF int RaPingTargetAdd(void *pingHandle, struct PingTargetInfo | |||
| 138 | 138 | ||
| 139 | phyId = pingHandleTmp->phyId; | 139 | phyId = pingHandleTmp->phyId; |
| 140 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 140 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 141 | - hccp_err("[add][ra_ping]phyId(%u) must less than %d!", phyId, RA_MAX_PHY_ID_NUM), | 141 | + hccp_err("[add][ra_ping]phyId(%u) must be less than %d!", phyId, RA_MAX_PHY_ID_NUM), |
| 142 | ConverReturnCode(RDMA_OP, -EINVAL)); | 142 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 143 | CHK_PRT_RETURN(pingHandleTmp->targetCnt + num < num, | 143 | CHK_PRT_RETURN(pingHandleTmp->targetCnt + num < num, |
| 144 | hccp_err("[add][ra_ping]pingHandleTmp->targetCnt + num is out of range"), ConverReturnCode(RDMA_OP, -EINVAL)); | 144 | hccp_err("[add][ra_ping]pingHandleTmp->targetCnt + num is out of range"), ConverReturnCode(RDMA_OP, -EINVAL)); |
| @@ -185,7 +185,7 @@ HCCP_ATTRI_VISI_DEF int RaPingTaskStart(void *pingHandle, struct PingTaskAttr *a | |||
| 185 | 185 | ||
| 186 | phyId = pingHandleTmp->phyId; | 186 | phyId = pingHandleTmp->phyId; |
| 187 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 187 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 188 | - hccp_err("[start][ra_ping]phyId(%u) must less than %d!", phyId, RA_MAX_PHY_ID_NUM), | 188 | + hccp_err("[start][ra_ping]phyId(%u) must be less than %d!", phyId, RA_MAX_PHY_ID_NUM), |
| 189 | ConverReturnCode(RDMA_OP, -EINVAL)); | 189 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 190 | 190 | ||
| 191 | RA_PTHREAD_MUTEX_LOCK(&pingHandleTmp->mutex); | 191 | RA_PTHREAD_MUTEX_LOCK(&pingHandleTmp->mutex); |
| @@ -240,7 +240,7 @@ HCCP_ATTRI_VISI_DEF int RaPingGetResults(void *pingHandle, struct PingTargetResu | |||
| 240 | 240 | ||
| 241 | phyId = pingHandleTmp->phyId; | 241 | phyId = pingHandleTmp->phyId; |
| 242 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 242 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 243 | - hccp_err("[get][ra_ping]phyId(%u) must less than %d!", phyId, RA_MAX_PHY_ID_NUM), | 243 | + hccp_err("[get][ra_ping]phyId(%u) must be less than %d!", phyId, RA_MAX_PHY_ID_NUM), |
| 244 | ConverReturnCode(RDMA_OP, -EINVAL)); | 244 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 245 | 245 | ||
| 246 | // num invalid, bigger than target exist | 246 | // num invalid, bigger than target exist |
| @@ -292,7 +292,7 @@ HCCP_ATTRI_VISI_DEF int RaPingTargetDel(void *pingHandle, struct PingTargetCommI | |||
| 292 | 292 | ||
| 293 | phyId = pingHandleTmp->phyId; | 293 | phyId = pingHandleTmp->phyId; |
| 294 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 294 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 295 | - hccp_err("[del][ra_ping]phyId(%u) must less than %d!", phyId, RA_MAX_PHY_ID_NUM), | 295 | + hccp_err("[del][ra_ping]phyId(%u) must be less than %d!", phyId, RA_MAX_PHY_ID_NUM), |
| 296 | ConverReturnCode(RDMA_OP, -EINVAL)); | 296 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 297 | 297 | ||
| 298 | ret = pingHandleTmp->pingOps->raPingTargetDel(pingHandleTmp, target, num); | 298 | ret = pingHandleTmp->pingOps->raPingTargetDel(pingHandleTmp, target, num); |
| @@ -326,7 +326,7 @@ HCCP_ATTRI_VISI_DEF int RaPingTaskStop(void *pingHandle) | |||
| 326 | 326 | ||
| 327 | phyId = pingHandleTmp->phyId; | 327 | phyId = pingHandleTmp->phyId; |
| 328 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 328 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 329 | - hccp_err("[stop][ra_ping]phyId(%u) must less than %d!", phyId, RA_MAX_PHY_ID_NUM), | 329 | + hccp_err("[stop][ra_ping]phyId(%u) must be less than %d!", phyId, RA_MAX_PHY_ID_NUM), |
| 330 | ConverReturnCode(RDMA_OP, -EINVAL)); | 330 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 331 | 331 | ||
| 332 | // no task to stop | 332 | // no task to stop |
| @@ -366,7 +366,7 @@ STATIC int RaPingDeinitParaCheck(struct RaPingHandle *pingHandle) | |||
| 366 | 366 | ||
| 367 | phyId = pingHandle->phyId; | 367 | phyId = pingHandle->phyId; |
| 368 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 368 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 369 | - hccp_err("[deinit][ra_ping]phyId(%u) must smaller than %u", phyId, RA_MAX_PHY_ID_NUM), -EINVAL); | 369 | + hccp_err("[deinit][ra_ping]phyId(%u) must be smaller than %u", phyId, RA_MAX_PHY_ID_NUM), -EINVAL); |
| 370 | 370 | ||
| 371 | devInfo = pingHandle->dev; | 371 | devInfo = pingHandle->dev; |
| 372 | if (pingHandle->protocol == PROTOCOL_RDMA) { | 372 | if (pingHandle->protocol == PROTOCOL_RDMA) { |
| @@ -182,7 +182,7 @@ int RaHdcSocketListenStop(unsigned int phyId, struct SocketListenInfoT conn[], u | |||
| 182 | ret = RaGetSocketListenInfo(conn, num, socketListenData.txData.conn, MAX_SOCKET_NUM); | 182 | ret = RaGetSocketListenInfo(conn, num, socketListenData.txData.conn, MAX_SOCKET_NUM); |
| 183 | CHK_PRT_RETURN(ret, | 183 | CHK_PRT_RETURN(ret, |
| 184 | hccp_err("[listen_stop][ra_hdc_socket]ra_hdc_socket_listen_stop memcpy_s failed, ret(%d)" | 184 | hccp_err("[listen_stop][ra_hdc_socket]ra_hdc_socket_listen_stop memcpy_s failed, ret(%d)" |
| 185 | - "phyId(%u).", | 185 | + " phyId(%u).", |
| 186 | ret, phyId), | 186 | ret, phyId), |
| 187 | -EINVAL); | 187 | -EINVAL); |
| 188 | 188 | ||
| @@ -397,7 +397,7 @@ int RaHdcGetSockets(unsigned int phyId, unsigned int role, struct SocketInfoT co | |||
| 397 | socketInfoData = (union OpSocketInfoData *)calloc(sizeof(union OpSocketInfoData), sizeof(char)); | 397 | socketInfoData = (union OpSocketInfoData *)calloc(sizeof(union OpSocketInfoData), sizeof(char)); |
| 398 | CHK_PRT_RETURN(socketInfoData == NULL, | 398 | CHK_PRT_RETURN(socketInfoData == NULL, |
| 399 | hccp_err("[get][ra_hdc_sockets]socket info data" | 399 | hccp_err("[get][ra_hdc_sockets]socket info data" |
| 400 | - "calloc failed phyId(%u)", | 400 | + " calloc failed phyId(%u)", |
| 401 | phyId), | 401 | phyId), |
| 402 | -ENOMEM); | 402 | -ENOMEM); |
| 403 | socketInfoData->txData.num = num; | 403 | socketInfoData->txData.num = num; |
| @@ -23,7 +23,7 @@ | |||
| 23 | do { \ | 23 | do { \ |
| 24 | ret_ = close(fd); \ | 24 | ret_ = close(fd); \ |
| 25 | if (ret_ < 0) { \ | 25 | if (ret_ < 0) { \ |
| 26 | - hccp_warn("close filedscp[%d] unsuccessful, errno:%d", fd, errno); \ | 26 | + hccp_warn("close fd[%d] unsuccessful, errno:%d", fd, errno); \ |
| 27 | } \ | 27 | } \ |
| 28 | } while ((ret_ < 0) && (errno == EINTR)); \ | 28 | } while ((ret_ < 0) && (errno == EINTR)); \ |
| 29 | fd = -1; \ | 29 | fd = -1; \ |
| @@ -248,7 +248,7 @@ int RsUbGetDevCb(struct rs_cb *rscb, unsigned int devIndex, struct RsUbDevCb **d | |||
| 248 | } | 248 | } |
| 249 | 249 | ||
| 250 | *devCb = NULL; | 250 | *devCb = NULL; |
| 251 | - hccp_err("dev_cb for devIndex:0x%x do not available!", devIndex); | 251 | + hccp_err("dev_cb for devIndex:0x%x is not available!", devIndex); |
| 252 | return -ENODEV; | 252 | return -ENODEV; |
| 253 | } | 253 | } |
| 254 | 254 | ||
| @@ -521,7 +521,7 @@ STATIC int RsUbGetJfcCb(struct RsUbDevCb *devCb, unsigned long long addr, struct | |||
| 521 | } | 521 | } |
| 522 | 522 | ||
| 523 | *tempJfcCb = NULL; | 523 | *tempJfcCb = NULL; |
| 524 | - hccp_err("jfc_cb for jfc_addr:0x%llx do not available!", addr); | 524 | + hccp_err("jfc_cb for jfc_addr:0x%llx is not available!", addr); |
| 525 | 525 | ||
| 526 | return -ENODEV; | 526 | return -ENODEV; |
| 527 | } | 527 | } |
| @@ -537,11 +537,11 @@ STATIC int RsUbFreeJfcCb(struct RsUbDevCb *devCb, struct RsCtxJfcCb *jfcCb) | |||
| 537 | if (jfcCb->jfcType == JFC_MODE_STARS_POLL || jfcCb->jfcType == JFC_MODE_CCU_POLL || | 537 | if (jfcCb->jfcType == JFC_MODE_STARS_POLL || jfcCb->jfcType == JFC_MODE_CCU_POLL || |
| 538 | jfcCb->jfcType == JFC_MODE_USER_CTL_NORMAL) { | 538 | jfcCb->jfcType == JFC_MODE_USER_CTL_NORMAL) { |
| 539 | (void)RsUbDeleteJfcExt(devCb, jfcCb); | 539 | (void)RsUbDeleteJfcExt(devCb, jfcCb); |
| 540 | - hccp_info("[deinit][rs_jfc]destroy success, dev jfcCnt:%u", devCb->jfcCnt); | 540 | + hccp_info("[deinit][rs_jfc]destroy success by RsUbDeleteJfcExt, dev jfcCnt:%u", devCb->jfcCnt); |
| 541 | } else if (jfcCb->jfcType == JFC_MODE_NORMAL) { | 541 | } else if (jfcCb->jfcType == JFC_MODE_NORMAL) { |
| 542 | urmaJfc = (urma_jfc_t *)(uintptr_t)jfcCb->jfcAddr; | 542 | urmaJfc = (urma_jfc_t *)(uintptr_t)jfcCb->jfcAddr; |
| 543 | (void)RsUrmaDeleteJfc(urmaJfc); | 543 | (void)RsUrmaDeleteJfc(urmaJfc); |
| 544 | - hccp_info("[deinit][rs_jfc]destroy success, dev jfcCnt:%u", devCb->jfcCnt); | 544 | + hccp_info("[deinit][rs_jfc]destroy success by RsUrmaDeleteJfc, dev jfcCnt:%u", devCb->jfcCnt); |
| 545 | } else { | 545 | } else { |
| 546 | hccp_err("jfc_type:%d is invalid, not support!", jfcCb->jfcType); | 546 | hccp_err("jfc_type:%d is invalid, not support!", jfcCb->jfcType); |
| 547 | ret = -EINVAL; | 547 | ret = -EINVAL; |
| @@ -580,7 +580,7 @@ STATIC void RsUbFreeJfcCbList(struct RsUbDevCb *devCb, struct RsListHead *jfcLis | |||
| 580 | 580 | ||
| 581 | RS_PTHREAD_MUTEX_LOCK(&devCb->mutex); | 581 | RS_PTHREAD_MUTEX_LOCK(&devCb->mutex); |
| 582 | if (!RsListEmpty(jfcList)) { | 582 | if (!RsListEmpty(jfcList)) { |
| 583 | - hccp_warn("jfc list do not empty!"); | 583 | + hccp_warn("jfc list is not empty!"); |
| 584 | RS_LIST_GET_HEAD_ENTRY(jfcCurr, jfcNext, jfcList, list, struct RsCtxJfcCb); | 584 | RS_LIST_GET_HEAD_ENTRY(jfcCurr, jfcNext, jfcList, list, struct RsCtxJfcCb); |
| 585 | for (; (&jfcCurr->list) != jfcList; | 585 | for (; (&jfcCurr->list) != jfcList; |
| 586 | jfcCurr = jfcNext, jfcNext = list_entry(jfcNext->list.next, struct RsCtxJfcCb, list)) { | 586 | jfcCurr = jfcNext, jfcNext = list_entry(jfcNext->list.next, struct RsCtxJfcCb, list)) { |
| @@ -616,7 +616,7 @@ STATIC void RsUbFreeSegCbList(struct RsUbDevCb *devCb, struct RsListHead *lsegLi | |||
| 616 | 616 | ||
| 617 | RS_PTHREAD_MUTEX_LOCK(&devCb->mutex); | 617 | RS_PTHREAD_MUTEX_LOCK(&devCb->mutex); |
| 618 | if (!RsListEmpty(lsegList)) { | 618 | if (!RsListEmpty(lsegList)) { |
| 619 | - hccp_warn("lseg list do not empty!"); | 619 | + hccp_warn("lseg list is not empty!"); |
| 620 | RS_LIST_GET_HEAD_ENTRY(segCurr, segNext, lsegList, list, struct RsSegCb); | 620 | RS_LIST_GET_HEAD_ENTRY(segCurr, segNext, lsegList, list, struct RsSegCb); |
| 621 | for (; (&segCurr->list) != lsegList; | 621 | for (; (&segCurr->list) != lsegList; |
| 622 | segCurr = segNext, segNext = list_entry(segNext->list.next, struct RsSegCb, list)) { | 622 | segCurr = segNext, segNext = list_entry(segNext->list.next, struct RsSegCb, list)) { |
| @@ -630,7 +630,7 @@ STATIC void RsUbFreeSegCbList(struct RsUbDevCb *devCb, struct RsListHead *lsegLi | |||
| 630 | 630 | ||
| 631 | RS_PTHREAD_MUTEX_LOCK(&devCb->mutex); | 631 | RS_PTHREAD_MUTEX_LOCK(&devCb->mutex); |
| 632 | if (!RsListEmpty(rsegList)) { | 632 | if (!RsListEmpty(rsegList)) { |
| 633 | - hccp_warn("rseg list do not empty!"); | 633 | + hccp_warn("rseg list is not empty!"); |
| 634 | RS_LIST_GET_HEAD_ENTRY(segCurr, segNext, rsegList, list, struct RsSegCb); | 634 | RS_LIST_GET_HEAD_ENTRY(segCurr, segNext, rsegList, list, struct RsSegCb); |
| 635 | for (; (&segCurr->list) != rsegList; | 635 | for (; (&segCurr->list) != rsegList; |
| 636 | segCurr = segNext, segNext = list_entry(segNext->list.next, struct RsSegCb, list)) { | 636 | segCurr = segNext, segNext = list_entry(segNext->list.next, struct RsSegCb, list)) { |
| @@ -701,7 +701,7 @@ STATIC void RsUbUnbindJettyCbList(struct RsUbDevCb *devCb, struct RsListHead *je | |||
| 701 | int ret; | 701 | int ret; |
| 702 | 702 | ||
| 703 | if (!RsListEmpty(jettyList)) { | 703 | if (!RsListEmpty(jettyList)) { |
| 704 | - hccp_warn("jetty list do not empty! start to unbind"); | 704 | + hccp_warn("jetty list is not empty! start to unbind"); |
| 705 | RS_LIST_GET_HEAD_ENTRY(jettyCurr, jettyNext, jettyList, list, struct RsCtxJettyCb); | 705 | RS_LIST_GET_HEAD_ENTRY(jettyCurr, jettyNext, jettyList, list, struct RsCtxJettyCb); |
| 706 | for (; (&jettyCurr->list) != jettyList; | 706 | for (; (&jettyCurr->list) != jettyList; |
| 707 | jettyCurr = jettyNext, jettyNext = list_entry(jettyNext->list.next, struct RsCtxJettyCb, list)) { | 707 | jettyCurr = jettyNext, jettyNext = list_entry(jettyNext->list.next, struct RsCtxJettyCb, list)) { |
| @@ -727,7 +727,7 @@ STATIC void RsUbUnimportJettyCbList(struct RsUbDevCb *devCb, struct RsListHead * | |||
| 727 | int ret; | 727 | int ret; |
| 728 | 728 | ||
| 729 | if (!RsListEmpty(rjettyList)) { | 729 | if (!RsListEmpty(rjettyList)) { |
| 730 | - hccp_warn("rjetty list do not empty! start to unimport"); | 730 | + hccp_warn("rjetty list is not empty! start to unimport"); |
| 731 | RS_LIST_GET_HEAD_ENTRY(remJettyCurr, remJettyNext, rjettyList, list, struct RsCtxRemJettyCb); | 731 | RS_LIST_GET_HEAD_ENTRY(remJettyCurr, remJettyNext, rjettyList, list, struct RsCtxRemJettyCb); |
| 732 | for (; (&remJettyCurr->list) != rjettyList; | 732 | for (; (&remJettyCurr->list) != rjettyList; |
| 733 | remJettyCurr = remJettyNext, | 733 | remJettyCurr = remJettyNext, |
| @@ -932,7 +932,7 @@ STATIC int RsUbGetJfceCb(struct RsUbDevCb *devCb, unsigned long long addr, struc | |||
| 932 | } | 932 | } |
| 933 | 933 | ||
| 934 | *tempJfceCb = NULL; | 934 | *tempJfceCb = NULL; |
| 935 | - hccp_err("jfce_cb for jfce_addr:0x%llx do not available!", addr); | 935 | + hccp_err("jfce_cb for jfce_addr:0x%llx is not available!", addr); |
| 936 | 936 | ||
| 937 | return -ENODEV; | 937 | return -ENODEV; |
| 938 | } | 938 | } |
| @@ -995,7 +995,7 @@ STATIC void RsUbFreeJfceCbList(struct RsUbDevCb *devCb, struct RsListHead *jfceL | |||
| 995 | 995 | ||
| 996 | RS_PTHREAD_MUTEX_LOCK(&devCb->mutex); | 996 | RS_PTHREAD_MUTEX_LOCK(&devCb->mutex); |
| 997 | if (!RsListEmpty(jfceList)) { | 997 | if (!RsListEmpty(jfceList)) { |
| 998 | - hccp_warn("jfce list do not empty!"); | 998 | + hccp_warn("jfce list is not empty!"); |
| 999 | RS_LIST_GET_HEAD_ENTRY(jfceCurr, jfceNext, jfceList, list, struct RsCtxJfceCb); | 999 | RS_LIST_GET_HEAD_ENTRY(jfceCurr, jfceNext, jfceList, list, struct RsCtxJfceCb); |
| 1000 | for (; (&jfceCurr->list) != jfceList; | 1000 | for (; (&jfceCurr->list) != jfceList; |
| 1001 | jfceCurr = jfceNext, jfceNext = list_entry(jfceNext->list.next, struct RsCtxJfceCb, list)) { | 1001 | jfceCurr = jfceNext, jfceNext = list_entry(jfceNext->list.next, struct RsCtxJfceCb, list)) { |
| @@ -1055,7 +1055,7 @@ STATIC int RsUbGetTokenIdCb(struct RsUbDevCb *devCb, unsigned long long addr, st | |||
| 1055 | } | 1055 | } |
| 1056 | 1056 | ||
| 1057 | *tempTokenIdCb = NULL; | 1057 | *tempTokenIdCb = NULL; |
| 1058 | - hccp_err("token_id_cb for token_id addr:0x%llx do not available! devIndex:0x%x", addr, devCb->index); | 1058 | + hccp_err("token_id_cb for token_id addr:0x%llx is not available! devIndex:0x%x", addr, devCb->index); |
| 1059 | 1059 | ||
| 1060 | return -ENODEV; | 1060 | return -ENODEV; |
| 1061 | } | 1061 | } |
| @@ -1109,7 +1109,7 @@ STATIC void RsUbFreeTokenIdCbList(struct RsUbDevCb *devCb, struct RsListHead *to | |||
| 1109 | 1109 | ||
| 1110 | RS_PTHREAD_MUTEX_LOCK(&devCb->mutex); | 1110 | RS_PTHREAD_MUTEX_LOCK(&devCb->mutex); |
| 1111 | if (!RsListEmpty(tokenIdList)) { | 1111 | if (!RsListEmpty(tokenIdList)) { |
| 1112 | - hccp_warn("token_id list do not empty!"); | 1112 | + hccp_warn("token_id list is not empty!"); |
| 1113 | RS_LIST_GET_HEAD_ENTRY(tokenIdCurr, tokenIdNext, tokenIdList, list, struct RsTokenIdCb); | 1113 | RS_LIST_GET_HEAD_ENTRY(tokenIdCurr, tokenIdNext, tokenIdList, list, struct RsTokenIdCb); |
| 1114 | for (; (&tokenIdCurr->list) != tokenIdList; | 1114 | for (; (&tokenIdCurr->list) != tokenIdList; |
| 1115 | tokenIdCurr = tokenIdNext, tokenIdNext = list_entry(tokenIdNext->list.next, struct RsTokenIdCb, list)) { | 1115 | tokenIdCurr = tokenIdNext, tokenIdNext = list_entry(tokenIdNext->list.next, struct RsTokenIdCb, list)) { |
| @@ -1137,7 +1137,7 @@ STATIC void RsUbFreeAsyncEventCbList(struct RsUbDevCb *devCb, struct RsListHead | |||
| 1137 | RS_PTHREAD_MUTEX_LOCK(&devCb->mutex); | 1137 | RS_PTHREAD_MUTEX_LOCK(&devCb->mutex); |
| 1138 | (void)RsEpollCtl(devCb->rscb->connCb.epollfd, EPOLL_CTL_DEL, devCb->urmaCtx->async_fd, EPOLLIN | EPOLLRDHUP); | 1138 | (void)RsEpollCtl(devCb->rscb->connCb.epollfd, EPOLL_CTL_DEL, devCb->urmaCtx->async_fd, EPOLLIN | EPOLLRDHUP); |
| 1139 | if (!RsListEmpty(asyncEventList)) { | 1139 | if (!RsListEmpty(asyncEventList)) { |
| 1140 | - hccp_run_warn("async_event list do not empty!"); | 1140 | + hccp_run_warn("async_event list is not empty!"); |
| 1141 | RS_LIST_GET_HEAD_ENTRY(asyncEventCurr, asyncEventNext, asyncEventList, list, struct RsCtxAsyncEventCb); | 1141 | RS_LIST_GET_HEAD_ENTRY(asyncEventCurr, asyncEventNext, asyncEventList, list, struct RsCtxAsyncEventCb); |
| 1142 | for (; (&asyncEventCurr->list) != asyncEventList; | 1142 | for (; (&asyncEventCurr->list) != asyncEventList; |
| 1143 | asyncEventCurr = asyncEventNext, | 1143 | asyncEventCurr = asyncEventNext, |
| @@ -1875,7 +1875,7 @@ STATIC int RsUbGetRemJettyCb(struct RsUbDevCb *devCb, urma_jetty_id_t *remJettyI | |||
| 1875 | } | 1875 | } |
| 1876 | 1876 | ||
| 1877 | *tempJettyCb = NULL; | 1877 | *tempJettyCb = NULL; |
| 1878 | - hccp_warn("devIndex:0x%x remJettyId:%u do not available!", devCb->index, remJettyId->id); | 1878 | + hccp_warn("devIndex:0x%x remJettyId:%u is not available!", devCb->index, remJettyId->id); |
| 1879 | return -ENODEV; | 1879 | return -ENODEV; |
| 1880 | } | 1880 | } |
| 1881 | 1881 | ||
| @@ -2008,7 +2008,7 @@ STATIC int RsUbGetRemJettyCbDeprecated(struct RsUbDevCb *devCb, unsigned int rem | |||
| 2008 | } | 2008 | } |
| 2009 | 2009 | ||
| 2010 | *tempJettyCb = NULL; | 2010 | *tempJettyCb = NULL; |
| 2011 | - hccp_err("rjetty_cb for rem_jetty %u do not available!", remJettyId); | 2011 | + hccp_err("rjetty_cb for rem_jetty %u is not available!", remJettyId); |
| 2012 | 2012 | ||
| 2013 | return -ENODEV; | 2013 | return -ENODEV; |
| 2014 | } | 2014 | } |
| @@ -182,7 +182,7 @@ int RsGetRdevCb(struct rs_cb *rsCb, unsigned int rdevIndex, struct RsRdevCb **rd | |||
| 182 | } | 182 | } |
| 183 | 183 | ||
| 184 | *rdevCb = NULL; | 184 | *rdevCb = NULL; |
| 185 | - hccp_err("rdev_cb for rdev_index[%u] do not available!", rdevIndex); | 185 | + hccp_err("rdev_cb for rdev_index[%u] is not available!", rdevIndex); |
| 186 | 186 | ||
| 187 | return -ENODEV; | 187 | return -ENODEV; |
| 188 | } | 188 | } |
| @@ -1115,7 +1115,7 @@ STATIC void RsDestroyQpList(unsigned int phyId, unsigned int rdevIndex, struct R | |||
| 1115 | int ret; | 1115 | int ret; |
| 1116 | 1116 | ||
| 1117 | if (!RsListEmpty(&rdevCb->qpList)) { | 1117 | if (!RsListEmpty(&rdevCb->qpList)) { |
| 1118 | - hccp_warn("qp list do not empty!"); | 1118 | + hccp_warn("qp list is not empty!"); |
| 1119 | RS_LIST_GET_HEAD_ENTRY(qpCb, qpCb2, &rdevCb->qpList, list, struct RsQpCb); | 1119 | RS_LIST_GET_HEAD_ENTRY(qpCb, qpCb2, &rdevCb->qpList, list, struct RsQpCb); |
| 1120 | for (; (&qpCb->list) != &rdevCb->qpList; | 1120 | for (; (&qpCb->list) != &rdevCb->qpList; |
| 1121 | qpCb = qpCb2, qpCb2 = list_entry(qpCb2->list.next, struct RsQpCb, list)) { | 1121 | qpCb = qpCb2, qpCb2 = list_entry(qpCb2->list.next, struct RsQpCb, list)) { |
| @@ -1362,7 +1362,7 @@ STATIC void RsFreeAcceptOneNode(struct rs_cb *rscb, struct RsAcceptInfo *accept) | |||
| 1362 | 1362 | ||
| 1363 | if (rscb->sslEnable == RS_SSL_ENABLE) { | 1363 | if (rscb->sslEnable == RS_SSL_ENABLE) { |
| 1364 | if (accept->ssl == NULL) { | 1364 | if (accept->ssl == NULL) { |
| 1365 | - hccp_warn("[Server] accept->ssl is NULL, it maybe has not establish tls link"); | 1365 | + hccp_warn("[Server] accept->ssl is NULL, it may not have established tls link"); |
| 1366 | } else { | 1366 | } else { |
| 1367 | ssl_adp_shutdown(accept->ssl); | 1367 | ssl_adp_shutdown(accept->ssl); |
| 1368 | ssl_adp_free(accept->ssl); | 1368 | ssl_adp_free(accept->ssl); |
| @@ -1386,7 +1386,7 @@ STATIC void RsFreeAccpetList(struct rs_cb *rscb) | |||
| 1386 | struct RsAcceptInfo *accept2 = NULL; | 1386 | struct RsAcceptInfo *accept2 = NULL; |
| 1387 | 1387 | ||
| 1388 | if (!RsListEmpty(&rscb->connCb.serverAcceptList)) { | 1388 | if (!RsListEmpty(&rscb->connCb.serverAcceptList)) { |
| 1389 | - hccp_warn("Server accept list do not empty!"); | 1389 | + hccp_warn("Server accept list is not empty!"); |
| 1390 | RS_LIST_GET_HEAD_ENTRY(accept, accept2, &rscb->connCb.serverAcceptList, list, struct RsAcceptInfo); | 1390 | RS_LIST_GET_HEAD_ENTRY(accept, accept2, &rscb->connCb.serverAcceptList, list, struct RsAcceptInfo); |
| 1391 | for (; (&accept->list) != &rscb->connCb.serverAcceptList; | 1391 | for (; (&accept->list) != &rscb->connCb.serverAcceptList; |
| 1392 | accept = accept2, accept2 = list_entry(accept2->list.next, struct RsAcceptInfo, list)) { | 1392 | accept = accept2, accept2 = list_entry(accept2->list.next, struct RsAcceptInfo, list)) { |
| @@ -1427,7 +1427,7 @@ STATIC void RsFreeConnOneNode(struct rs_cb *rscb, struct RsConnInfo *conn) | |||
| 1427 | 1427 | ||
| 1428 | if (rscb->sslEnable == RS_SSL_ENABLE) { | 1428 | if (rscb->sslEnable == RS_SSL_ENABLE) { |
| 1429 | if (conn->ssl == NULL) { | 1429 | if (conn->ssl == NULL) { |
| 1430 | - hccp_warn("[Client] conn->ssl is NULL, it maybe has not establish tls link"); | 1430 | + hccp_warn("[Client] conn->ssl is NULL, it may not have established tls link"); |
| 1431 | } else { | 1431 | } else { |
| 1432 | ssl_adp_shutdown(conn->ssl); | 1432 | ssl_adp_shutdown(conn->ssl); |
| 1433 | ssl_adp_free(conn->ssl); | 1433 | ssl_adp_free(conn->ssl); |
| @@ -1453,7 +1453,7 @@ STATIC void RsFreeClientConnList(struct rs_cb *rscb) | |||
| 1453 | struct RsConnInfo *conn2 = NULL; | 1453 | struct RsConnInfo *conn2 = NULL; |
| 1454 | 1454 | ||
| 1455 | if (!RsListEmpty(&rscb->connCb.clientConnList)) { | 1455 | if (!RsListEmpty(&rscb->connCb.clientConnList)) { |
| 1456 | - hccp_warn("Client conn node do not empty!"); | 1456 | + hccp_warn("Client conn node is not empty!"); |
| 1457 | RS_LIST_GET_HEAD_ENTRY(conn, conn2, &rscb->connCb.clientConnList, list, struct RsConnInfo); | 1457 | RS_LIST_GET_HEAD_ENTRY(conn, conn2, &rscb->connCb.clientConnList, list, struct RsConnInfo); |
| 1458 | for (; (&conn->list) != &rscb->connCb.clientConnList; | 1458 | for (; (&conn->list) != &rscb->connCb.clientConnList; |
| 1459 | conn = conn2, conn2 = list_entry(conn2->list.next, struct RsConnInfo, list)) { | 1459 | conn = conn2, conn2 = list_entry(conn2->list.next, struct RsConnInfo, list)) { |
| @@ -1475,7 +1475,7 @@ STATIC void RsFreeDesignatedClientConnNode(struct rs_cb *rscb, struct RsIpAddrIn | |||
| 1475 | for (; (&conn->list) != &rscb->connCb.clientConnList; | 1475 | for (; (&conn->list) != &rscb->connCb.clientConnList; |
| 1476 | conn = conn2, conn2 = list_entry(conn2->list.next, struct RsConnInfo, list)) { | 1476 | conn = conn2, conn2 = list_entry(conn2->list.next, struct RsConnInfo, list)) { |
| 1477 | if (!RsCompareIpAddr(&conn->clientIp, localIp)) { | 1477 | if (!RsCompareIpAddr(&conn->clientIp, localIp)) { |
| 1478 | - hccp_warn("Client conn node for IP[%s] do not empty!", localIp->readAddr); | 1478 | + hccp_warn("Client conn node for IP[%s] is not empty!", localIp->readAddr); |
| 1479 | RsFreeConnOneNode(rscb, conn); | 1479 | RsFreeConnOneNode(rscb, conn); |
| 1480 | conn = NULL; | 1480 | conn = NULL; |
| 1481 | } | 1481 | } |
| @@ -1491,7 +1491,7 @@ STATIC void RsFreeServerConnList(struct rs_cb *rscb) | |||
| 1491 | struct RsConnInfo *conn2 = NULL; | 1491 | struct RsConnInfo *conn2 = NULL; |
| 1492 | 1492 | ||
| 1493 | if (!RsListEmpty(&rscb->connCb.serverConnList)) { | 1493 | if (!RsListEmpty(&rscb->connCb.serverConnList)) { |
| 1494 | - hccp_warn("Server conn node do not empty!"); | 1494 | + hccp_warn("Server conn node is not empty!"); |
| 1495 | RS_LIST_GET_HEAD_ENTRY(conn, conn2, &rscb->connCb.serverConnList, list, struct RsConnInfo); | 1495 | RS_LIST_GET_HEAD_ENTRY(conn, conn2, &rscb->connCb.serverConnList, list, struct RsConnInfo); |
| 1496 | for (; (&conn->list) != &rscb->connCb.serverConnList; | 1496 | for (; (&conn->list) != &rscb->connCb.serverConnList; |
| 1497 | conn = conn2, conn2 = list_entry(conn2->list.next, struct RsConnInfo, list)) { | 1497 | conn = conn2, conn2 = list_entry(conn2->list.next, struct RsConnInfo, list)) { |
| @@ -1513,7 +1513,7 @@ STATIC void RsFreeDesignatedServerConnNode(struct rs_cb *rscb, struct RsIpAddrIn | |||
| 1513 | for (; (&conn->list) != &rscb->connCb.serverConnList; | 1513 | for (; (&conn->list) != &rscb->connCb.serverConnList; |
| 1514 | conn = conn2, conn2 = list_entry(conn2->list.next, struct RsConnInfo, list)) { | 1514 | conn = conn2, conn2 = list_entry(conn2->list.next, struct RsConnInfo, list)) { |
| 1515 | if (!RsCompareIpAddr(&conn->serverIp, localIp)) { | 1515 | if (!RsCompareIpAddr(&conn->serverIp, localIp)) { |
| 1516 | - hccp_warn("Server conn node for IP[%s] do not empty!", localIp->readAddr); | 1516 | + hccp_warn("Server conn node for IP[%s] is not empty!", localIp->readAddr); |
| 1517 | RsFreeConnOneNode(rscb, conn); | 1517 | RsFreeConnOneNode(rscb, conn); |
| 1518 | conn = NULL; | 1518 | conn = NULL; |
| 1519 | } | 1519 | } |
| @@ -1553,7 +1553,7 @@ STATIC void RsFreeListenList(struct rs_cb *rscb) | |||
| 1553 | struct RsListenInfo *listen2 = NULL; | 1553 | struct RsListenInfo *listen2 = NULL; |
| 1554 | 1554 | ||
| 1555 | if (!RsListEmpty(&rscb->connCb.listenList)) { | 1555 | if (!RsListEmpty(&rscb->connCb.listenList)) { |
| 1556 | - hccp_warn("Server listen node do not empty!"); | 1556 | + hccp_warn("Server listen node is not empty!"); |
| 1557 | RS_LIST_GET_HEAD_ENTRY(listen, listen2, &rscb->connCb.listenList, list, struct RsListenInfo); | 1557 | RS_LIST_GET_HEAD_ENTRY(listen, listen2, &rscb->connCb.listenList, list, struct RsListenInfo); |
| 1558 | for (; (&listen->list) != &rscb->connCb.listenList; | 1558 | for (; (&listen->list) != &rscb->connCb.listenList; |
| 1559 | listen = listen2, listen2 = list_entry(listen2->list.next, struct RsListenInfo, list)) { | 1559 | listen = listen2, listen2 = list_entry(listen2->list.next, struct RsListenInfo, list)) { |
| @@ -1623,7 +1623,7 @@ STATIC void RsFreeWhiteList(struct rs_cb *rscb) | |||
| 1623 | struct RsWhiteList *wlist2 = NULL; | 1623 | struct RsWhiteList *wlist2 = NULL; |
| 1624 | 1624 | ||
| 1625 | if (!RsListEmpty(&rscb->connCb.whiteList)) { | 1625 | if (!RsListEmpty(&rscb->connCb.whiteList)) { |
| 1626 | - hccp_warn("Server white list do not empty!"); | 1626 | + hccp_warn("Server white list is not empty!"); |
| 1627 | RS_LIST_GET_HEAD_ENTRY(wlist, wlist2, &rscb->connCb.whiteList, list, struct RsWhiteList); | 1627 | RS_LIST_GET_HEAD_ENTRY(wlist, wlist2, &rscb->connCb.whiteList, list, struct RsWhiteList); |
| 1628 | for (; (&wlist->list) != &rscb->connCb.whiteList; | 1628 | for (; (&wlist->list) != &rscb->connCb.whiteList; |
| 1629 | wlist = wlist2, wlist2 = list_entry(wlist2->list.next, struct RsWhiteList, list)) { | 1629 | wlist = wlist2, wlist2 = list_entry(wlist2->list.next, struct RsWhiteList, list)) { |
| @@ -1762,7 +1762,7 @@ STATIC void RsFreeHeterogTcpFdList(struct rs_cb *rsCb) | |||
| 1762 | struct RsHeterogTcpFdInfo *fdNode1 = NULL; | 1762 | struct RsHeterogTcpFdInfo *fdNode1 = NULL; |
| 1763 | 1763 | ||
| 1764 | if (!RsListEmpty(&rsCb->heterogTcpFdList)) { | 1764 | if (!RsListEmpty(&rsCb->heterogTcpFdList)) { |
| 1765 | - hccp_warn("heterog_tcp_fd_list do not empty!"); | 1765 | + hccp_warn("heterog_tcp_fd_list is not empty!"); |
| 1766 | RS_LIST_GET_HEAD_ENTRY(fdNode, fdNode1, &rsCb->heterogTcpFdList, list, struct RsHeterogTcpFdInfo); | 1766 | RS_LIST_GET_HEAD_ENTRY(fdNode, fdNode1, &rsCb->heterogTcpFdList, list, struct RsHeterogTcpFdInfo); |
| 1767 | for (; (&fdNode->list) != &rsCb->heterogTcpFdList; | 1767 | for (; (&fdNode->list) != &rsCb->heterogTcpFdList; |
| 1768 | fdNode = fdNode1, fdNode1 = list_entry(fdNode1->list.next, struct RsHeterogTcpFdInfo, list)) { | 1768 | fdNode = fdNode1, fdNode1 = list_entry(fdNode1->list.next, struct RsHeterogTcpFdInfo, list)) { |
| @@ -323,7 +323,7 @@ HcclResult HccpUbCreateJettyAsync( | |||
| 323 | if (ret != 0 || !raReqHandle) { | 323 | if (ret != 0 || !raReqHandle) { |
| 324 | HCCL_ERROR( | 324 | HCCL_ERROR( |
| 325 | "[%s] failed, call interface error[%d], raReqHandle[%p], " | 325 | "[%s] failed, call interface error[%d], raReqHandle[%p], " |
| 326 | - "ctxHanlde[%p].", | 326 | + "ctxHandle[%p].", |
| 327 | __func__, ret, raReqHandle, ctxhandle); | 327 | __func__, ret, raReqHandle, ctxhandle); |
| 328 | return HcclResult::HCCL_E_NETWORK; | 328 | return HcclResult::HCCL_E_NETWORK; |
| 329 | } | 329 | } |
| @@ -85,8 +85,8 @@ HcclResult HcclCommMemReg(HcclComm comm, const char* memTag, const CommMem* mem, | |||
| 85 | attr.value = 0; | 85 | attr.value = 0; |
| 86 | HcclResult ret = hcclComm->GetIndependentOp().GetCommMemMgr().CommRegMem(memTagStr, hcclMem, attr, memHandle); | 86 | HcclResult ret = hcclComm->GetIndependentOp().GetCommMemMgr().CommRegMem(memTagStr, hcclMem, attr, memHandle); |
| 87 | CHK_PRT_RET( | 87 | CHK_PRT_RET( |
| 88 | - ret != HCCL_SUCCESS, HCCL_ERROR("[HcclCommMemReg]legcy Bind failed. memTag[%s], ret[%d]", memTag, ret), ret); | 88 | + ret != HCCL_SUCCESS, HCCL_ERROR("[HcclCommMemReg]legacy Bind failed. memTag[%s], ret[%d]", memTag, ret), ret); |
| 89 | - HCCL_INFO("[HcclCommMemReg]legcy success: raw handle[%p]", *memHandle); | 89 | + HCCL_INFO("[HcclCommMemReg]legacy success: raw handle[%p]", *memHandle); |
| 90 | 90 | ||
| 91 | return HCCL_SUCCESS; | 91 | return HCCL_SUCCESS; |
| 92 | } | 92 | } |
| @@ -200,7 +200,7 @@ bool CheckPortGroupSize(u32 netLayer, NetInstance::Link& srcLink, NetInstance::L | |||
| 200 | auto localRankId = std::dynamic_pointer_cast<NetInstance::Peer>(srcPeer)->GetRankId(); | 200 | auto localRankId = std::dynamic_pointer_cast<NetInstance::Peer>(srcPeer)->GetRankId(); |
| 201 | auto remoteRankId = std::dynamic_pointer_cast<NetInstance::Peer>(targetPeer)->GetRankId(); | 201 | auto remoteRankId = std::dynamic_pointer_cast<NetInstance::Peer>(targetPeer)->GetRankId(); |
| 202 | HCCL_WARNING( | 202 | HCCL_WARNING( |
| 203 | - "[GetPaths][CheckPortGroupSize] portGroupSize is not equal => src[%u], target[%u]." | 203 | + "[GetPaths][CheckPortGroupSize] portGroupSize is not equal => src[%u], target[%u]. " |
| 204 | "LocatedInfo: NetLayer[%u], localRank[%u], rmtRank[%u], localAddr[%s], rmtAddr[%s]", | 204 | "LocatedInfo: NetLayer[%u], localRank[%u], rmtRank[%u], localAddr[%s], rmtAddr[%s]", |
| 205 | srcPortGroupSize, tgtPortGroupSize, netLayer, localRankId, remoteRankId, localAddr.Describe().c_str(), | 205 | srcPortGroupSize, tgtPortGroupSize, netLayer, localRankId, remoteRankId, localAddr.Describe().c_str(), |
| 206 | remoteAddr.Describe().c_str()); | 206 | remoteAddr.Describe().c_str()); |
| @@ -179,7 +179,7 @@ bool FindHostIPByNicClass( | |||
| 179 | return true; | 179 | return true; |
| 180 | } | 180 | } |
| 181 | 181 | ||
| 182 | - HCCL_WARNING("[%s] find hostIp by nic class[normal_docket_lo] fail.", __func__); | 182 | + HCCL_WARNING("[%s] find hostIp by nic class[normal_docker_lo] fail.", __func__); |
| 183 | return false; | 183 | return false; |
| 184 | } | 184 | } |
| 185 | 185 | ||
| @@ -416,8 +416,8 @@ void RankInfoDetectService::DisplayConnectingStatus(u32 totalSockets, u32 waitSo | |||
| 416 | } | 416 | } |
| 417 | } | 417 | } |
| 418 | 418 | ||
| 419 | - std::string infoStr = "succ sockets is [" + std::to_string((totalSockets - waitSockets)) + "], waiting sockets is [" | 419 | + std::string infoStr = "succ sockets are [" + std::to_string((totalSockets - waitSockets)) |
| 420 | - + std::to_string(waitSockets) + "], wait sockets rankid: "; | 420 | + + "], waiting sockets is [" + std::to_string(waitSockets) + "], wait sockets rankid: "; |
| 421 | for (u32 index = 0; index < unsocketinfos.size(); index++) { | 421 | for (u32 index = 0; index < unsocketinfos.size(); index++) { |
| 422 | if (index == (unsocketinfos.size() - 1)) { | 422 | if (index == (unsocketinfos.size() - 1)) { |
| 423 | infoStr += "[" + unsocketinfos[index] + "]"; | 423 | infoStr += "[" + unsocketinfos[index] + "]"; |
| @@ -58,7 +58,7 @@ namespace { | |||
| 58 | CHK_PRT_RET( | 58 | CHK_PRT_RET( |
| 59 | sRet != EOK, | 59 | sRet != EOK, |
| 60 | HCCL_ERROR( | 60 | HCCL_ERROR( |
| 61 | - "[%s] memcpy root info fail. errorno[%d] params:destMaxSize[%u], count[%u]", __func__, sRet, | 61 | + "[%s] memcpy root info fail. errno[%d] params:destMaxSize[%u], count[%u]", __func__, sRet, |
| 62 | HCCL_ROOT_INFO_BYTES, rootHandleLen), | 62 | HCCL_ROOT_INFO_BYTES, rootHandleLen), |
| 63 | HCCL_E_MEMORY); | 63 | HCCL_E_MEMORY); |
| 64 | 64 | ||
| @@ -806,7 +806,10 @@ HcclResult ExecutorBase::PrepareSliceData( | |||
| 806 | temp.offset = totalSize - residueSize + piplineOffset; | 806 | temp.offset = totalSize - residueSize + piplineOffset; |
| 807 | i++; | 807 | i++; |
| 808 | CHK_PRT_RET( | 808 | CHK_PRT_RET( |
| 809 | - (sliceSize <= 0), HCCL_ERROR("[Prepare][SliceData]data_slice_prepare sliceSize[%llu]", sliceSize), | 809 | + (sliceSize <= 0), |
| 810 | + HCCL_ERROR( | ||
| 811 | + "[Prepare][SliceData]data_slice_prepare sliceSize[%llu] is invalid, should be greater than 0", | ||
| 812 | + sliceSize), | ||
| 810 | HCCL_E_PARA); | 813 | HCCL_E_PARA); |
| 811 | residueSize -= sliceSize; | 814 | residueSize -= sliceSize; |
| 812 | dataSlice.push_back(temp); | 815 | dataSlice.push_back(temp); |
| @@ -877,7 +880,7 @@ ExecutorBase::CalcBinaryBlockHalvingDoubleLinkReleation(u32 rank, u32 rankSize, | |||
| 877 | CalcBinaryBlockParams( | 880 | CalcBinaryBlockParams( |
| 878 | rank, rankSize, stepsInBlock, lowerBlockSize, myBlockSize, rankInMyBlock, myBlockOffset, higherBlockSize); | 881 | rank, rankSize, stepsInBlock, lowerBlockSize, myBlockSize, rankInMyBlock, myBlockOffset, higherBlockSize); |
| 879 | if (lowerBlockSize == 0) { | 882 | if (lowerBlockSize == 0) { |
| 880 | - HCCL_ERROR("[Calc][BinaryBlockHalvingDoubleLinkReleation]lowerBlockSize size is zero."); | 883 | + HCCL_ERROR("[Calc][BinaryBlockHalvingDoubleLinkRelation]lowerBlockSize size is zero."); |
| 881 | return HCCL_E_PARA; | 884 | return HCCL_E_PARA; |
| 882 | } | 885 | } |
| 883 | for (u32 i = 0; i < rankSize; i++) { | 886 | for (u32 i = 0; i < rankSize; i++) { |
| @@ -44,7 +44,7 @@ HcclResult Reducer::run( | |||
| 44 | 44 | ||
| 45 | u64 dataBytes = remoteRcvTemp.size(); | 45 | u64 dataBytes = remoteRcvTemp.size(); |
| 46 | HCCL_DEBUG( | 46 | HCCL_DEBUG( |
| 47 | - "localSrc[%p] localDst[%p] remoteRcvtmep[%p] offset[%llu]", localSrc.ptr(), localDst.ptr(), remoteRcvTemp.ptr(), | 47 | + "localSrc[%p] localDst[%p] remoteRcvTemp[%p] offset[%llu]", localSrc.ptr(), localDst.ptr(), remoteRcvTemp.ptr(), |
| 48 | remoteMemOffset); | 48 | remoteMemOffset); |
| 49 | 49 | ||
| 50 | // server 内 reduce 并且 reduceAttribute_ 也支持,走该分支 | 50 | // server 内 reduce 并且 reduceAttribute_ 也支持,走该分支 |
| @@ -55,7 +55,7 @@ HcclResult AllGatherVPipeline::Prepare( | |||
| 55 | streamNotifyMain_ = notifyMain; | 55 | streamNotifyMain_ = notifyMain; |
| 56 | if (streamNotifyMain_.size() < intraRankSize_) { | 56 | if (streamNotifyMain_.size() < intraRankSize_) { |
| 57 | HCCL_ERROR( | 57 | HCCL_ERROR( |
| 58 | - "[AllGatherVPipeline][Prepare]rank[%u] streamNotifyMain_ size[%u] error, is smaller than," | 58 | + "[AllGatherVPipeline][Prepare]rank[%u] streamNotifyMain_ size[%u] is smaller than " |
| 59 | "intraRankSize_[%u]", | 59 | "intraRankSize_[%u]", |
| 60 | userRank_, streamNotifyMain_.size(), intraRankSize_); | 60 | userRank_, streamNotifyMain_.size(), intraRankSize_); |
| 61 | return HCCL_E_INTERNAL; | 61 | return HCCL_E_INTERNAL; |
| @@ -63,7 +63,7 @@ HcclResult AllGatherVPipeline::Prepare( | |||
| 63 | streamNotifySub_ = notifySub; | 63 | streamNotifySub_ = notifySub; |
| 64 | if (streamNotifySub_.size() < intraRankSize_) { | 64 | if (streamNotifySub_.size() < intraRankSize_) { |
| 65 | HCCL_ERROR( | 65 | HCCL_ERROR( |
| 66 | - "[AllGatherVPipeline][Prepare]rank[%u] streamNotifySub_ size[%u] error, is smaller than, " | 66 | + "[AllGatherVPipeline][Prepare]rank[%u] streamNotifySub_ size[%u] is smaller than " |
| 67 | "intraRankSize_[%u]", | 67 | "intraRankSize_[%u]", |
| 68 | userRank_, streamNotifySub_.size(), intraRankSize_); | 68 | userRank_, streamNotifySub_.size(), intraRankSize_); |
| 69 | return HCCL_E_INTERNAL; | 69 | return HCCL_E_INTERNAL; |
| @@ -168,7 +168,7 @@ HcclResult AllGatherVPipeline::RunAsync() | |||
| 168 | static_cast<u8*>(dmaMem_[srcDMAMemSliceId].ptr()) + serverOffsetByte, | 168 | static_cast<u8*>(dmaMem_[srcDMAMemSliceId].ptr()) + serverOffsetByte, |
| 169 | userMemSlice_[serverRankOffset].size, subStream_[0])); | 169 | userMemSlice_[serverRankOffset].size, subStream_[0])); |
| 170 | HCCL_DEBUG( | 170 | HCCL_DEBUG( |
| 171 | - "[AllGatherVPipeline][RunAsync] local rank[%u] localOffset[%llu]tx with remoteRank[%u]," | 171 | + "[AllGatherVPipeline][RunAsync] local rank[%u] localOffset[%llu] tx with remoteRank[%u], " |
| 172 | "remoteOffset[%llu] with slice[%llu]", | 172 | "remoteOffset[%llu] with slice[%llu]", |
| 173 | userRank_, serverOffsetByte, nextInterRankId, serverOffsetByte, userMemSlice_[serverRankOffset].size); | 173 | userRank_, serverOffsetByte, nextInterRankId, serverOffsetByte, userMemSlice_[serverRankOffset].size); |
| 174 | // 对于RDM RxAsync,内存属性入参无效 RDMA::Wait | 174 | // 对于RDM RxAsync,内存属性入参无效 RDMA::Wait |
| @@ -178,7 +178,7 @@ HcclResult AllGatherVPipeline::RunAsync() | |||
| 178 | static_cast<u8*>(dmaMem_[dstDMAMemSliceId].ptr()) + readRemoteOffsetByte, | 178 | static_cast<u8*>(dmaMem_[dstDMAMemSliceId].ptr()) + readRemoteOffsetByte, |
| 179 | userMemSlice_[readRemoteOffset].size, subStream_[0])); // wait | 179 | userMemSlice_[readRemoteOffset].size, subStream_[0])); // wait |
| 180 | HCCL_DEBUG( | 180 | HCCL_DEBUG( |
| 181 | - "[AllGatherVPipeline][RunAsync]read local rank[%u] localOffset[%llu]tx with remoteRank[%u]," | 181 | + "[AllGatherVPipeline][RunAsync]read local rank[%u] localOffset[%llu] tx with remoteRank[%u], " |
| 182 | "remoteOffset[%llu] with slice[%llu]", | 182 | "remoteOffset[%llu] with slice[%llu]", |
| 183 | userRank_, readRemoteOffsetByte, readRemoteOffset, readRemoteOffsetByte, | 183 | userRank_, readRemoteOffsetByte, readRemoteOffset, readRemoteOffsetByte, |
| 184 | userMemSlice_[readRemoteOffset].size); | 184 | userMemSlice_[readRemoteOffset].size); |
| @@ -60,7 +60,7 @@ HcclResult AllReduceAHCBroke::RunInterAllReduce( | |||
| 60 | 60 | ||
| 61 | if (interLinksVector.size() != interSlicesVector.size()) { | 61 | if (interLinksVector.size() != interSlicesVector.size()) { |
| 62 | HCCL_ERROR( | 62 | HCCL_ERROR( |
| 63 | - "[AllReduceAHCBroke][RunInterAllReduceBrokeType]rank[%u] linksVector size[%llu] is no equal to " | 63 | + "[AllReduceAHCBroke][RunInterAllReduceBrokeType]rank[%u] linksVector size[%llu] is not equal to " |
| 64 | "slicesVector size [%u]", | 64 | "slicesVector size [%u]", |
| 65 | rank, interLinksVector.size(), interSlicesVector.size()); | 65 | rank, interLinksVector.size(), interSlicesVector.size()); |
| 66 | return HCCL_E_INTERNAL; | 66 | return HCCL_E_INTERNAL; |
| @@ -153,7 +153,7 @@ HcclResult AllReduceChunkMesh::RunAsync(const u32 rank, const u32 rankSize, cons | |||
| 153 | CHK_PRT_RET( | 153 | CHK_PRT_RET( |
| 154 | ret != HCCL_SUCCESS, | 154 | ret != HCCL_SUCCESS, |
| 155 | HCCL_ERROR( | 155 | HCCL_ERROR( |
| 156 | - "[AllReduceRing][RunAsync]rank[%u] count[%llu] failed in reducescater " | 156 | + "[AllReduceRing][RunAsync]rank[%u] count[%llu] failed in ReduceScatter " |
| 157 | "step", | 157 | "step", |
| 158 | rank, count_), | 158 | rank, count_), |
| 159 | ret); | 159 | ret); |
| @@ -331,7 +331,7 @@ HcclResult AllReduceChunkMesh::RunAllGather(u32 rank, u32 rankSize, const std::v | |||
| 331 | 331 | ||
| 332 | CHK_RET(links[dstRank]->TxDataSignal(subStream)); | 332 | CHK_RET(links[dstRank]->TxDataSignal(subStream)); |
| 333 | CHK_RET(links[dstRank]->RxDataSignal(subStream)); | 333 | CHK_RET(links[dstRank]->RxDataSignal(subStream)); |
| 334 | - HCCL_DEBUG("[AllReduceChunkMesh]round %u success"); | 334 | + HCCL_DEBUG("[AllReduceChunkMesh]round %u success", round); |
| 335 | } | 335 | } |
| 336 | 336 | ||
| 337 | CHK_RET(HcclD2DMemcpyAsync(dispatcher_, emptyDst, emptySrc, stream_)); | 337 | CHK_RET(HcclD2DMemcpyAsync(dispatcher_, emptyDst, emptySrc, stream_)); |
| @@ -122,7 +122,7 @@ AllReduceNHRV1::RunAsyncStaged(const u32 rank, const u32 rankSize, const std::ve | |||
| 122 | ret); | 122 | ret); |
| 123 | break; | 123 | break; |
| 124 | default: | 124 | default: |
| 125 | - HCCL_ERROR("[AllReduceNHRV1][RunAsyncStaged]stage[%d]is not support", stage); | 125 | + HCCL_ERROR("[AllReduceNHRV1][RunAsyncStaged]stage[%d] is not supported", stage); |
| 126 | return HCCL_E_NOT_SUPPORT; | 126 | return HCCL_E_NOT_SUPPORT; |
| 127 | } | 127 | } |
| 128 | HCCL_INFO("AllReduceNHRV1 RunAsyncStaged stage[%d] finished: rank[%u] ranksize[%u]", stage, rank, rankSize); | 128 | HCCL_INFO("AllReduceNHRV1 RunAsyncStaged stage[%d] finished: rank[%u] ranksize[%u]", stage, rank, rankSize); |
| @@ -73,7 +73,7 @@ HcclResult AllReduceRecursiveHalvingDoubling::RunAsyncStaged( | |||
| 73 | CHK_RET(GatherInPartOne(rank, links)); | 73 | CHK_RET(GatherInPartOne(rank, links)); |
| 74 | break; | 74 | break; |
| 75 | default: | 75 | default: |
| 76 | - HCCL_ERROR("[AllReduceRecursiveHalvingDoubling][RunAsyncStaged]stage[%d]is not support", stage); | 76 | + HCCL_ERROR("[AllReduceRecursiveHalvingDoubling][RunAsyncStaged]stage[%d] is not supported", stage); |
| 77 | return HCCL_E_NOT_SUPPORT; | 77 | return HCCL_E_NOT_SUPPORT; |
| 78 | } | 78 | } |
| 79 | HCCL_INFO( | 79 | HCCL_INFO( |
| @@ -146,13 +146,13 @@ HcclResult AllReduceRecursiveHalvingDoubling::ReduceInPartOne(u32 rank, const st | |||
| 146 | HcclResult ret = link->TxAck(stream_); | 146 | HcclResult ret = link->TxAck(stream_); |
| 147 | CHK_PRT_RET( | 147 | CHK_PRT_RET( |
| 148 | ret != HCCL_SUCCESS, | 148 | ret != HCCL_SUCCESS, |
| 149 | - HCCL_ERROR("[Reduce][InPartOneToEven]rank[%u] tx ack from peerank[%u] failed", rank, peerRank), ret); | 149 | + HCCL_ERROR("[Reduce][InPartOneToEven]rank[%u] tx ack from peerRank[%u] failed", rank, peerRank), ret); |
| 150 | ret = link->RxAck(stream_); | 150 | ret = link->RxAck(stream_); |
| 151 | CHK_PRT_RET( | 151 | CHK_PRT_RET( |
| 152 | ret != HCCL_SUCCESS, | 152 | ret != HCCL_SUCCESS, |
| 153 | - HCCL_ERROR("[Reduce][InPartOneToEven]rank[%u] rx ack from peerank[%u] failed", rank, peerRank), ret); | 153 | + HCCL_ERROR("[Reduce][InPartOneToEven]rank[%u] rx ack from peerRank[%u] failed", rank, peerRank), ret); |
| 154 | // 接收数据到本端的 output | 154 | // 接收数据到本端的 output |
| 155 | - HCCL_DEBUG("send mem[%p] size[%llu] to peerank[%u]", outputMem_.ptr(), outputMem_.size(), peerRank); | 155 | + HCCL_DEBUG("send mem[%p] size[%llu] to peerRank[%u]", outputMem_.ptr(), outputMem_.size(), peerRank); |
| 156 | ret = link->TxAsync(UserMemType::INPUT_MEM, baseOffset_, outputMem_.ptr(), 0, stream_); | 156 | ret = link->TxAsync(UserMemType::INPUT_MEM, baseOffset_, outputMem_.ptr(), 0, stream_); |
| 157 | CHK_PRT_RET( | 157 | CHK_PRT_RET( |
| 158 | ret != HCCL_SUCCESS, | 158 | ret != HCCL_SUCCESS, |
| @@ -174,18 +174,18 @@ HcclResult AllReduceRecursiveHalvingDoubling::ReduceInPartOne(u32 rank, const st | |||
| 174 | HcclResult ret = link->TxAck(stream_); | 174 | HcclResult ret = link->TxAck(stream_); |
| 175 | CHK_PRT_RET( | 175 | CHK_PRT_RET( |
| 176 | ret != HCCL_SUCCESS, | 176 | ret != HCCL_SUCCESS, |
| 177 | - HCCL_ERROR("[Reduce][InPartOneToEven]rank[%u] tx ack from peerank[%u] failed", rank, peerRank), ret); | 177 | + HCCL_ERROR("[Reduce][InPartOneToEven]rank[%u] tx ack from peerRank[%u] failed", rank, peerRank), ret); |
[MEDIUM] peerank 拼写修正跨文件不彻底:temp_reduce/reduce_recursive_hd.cc 仍有 8 处 本 PR 把 建议修复:同批统一(以 L107 为例):
建议用 ![]() ![]() | |||
| 178 | ret = link->RxAck(stream_); | 178 | ret = link->RxAck(stream_); |
| 179 | CHK_PRT_RET( | 179 | CHK_PRT_RET( |
| 180 | ret != HCCL_SUCCESS, | 180 | ret != HCCL_SUCCESS, |
| 181 | - HCCL_ERROR("[Reduce][InPartOneToEven]rank[%u] rx ack from peerank[%u] failed", rank, peerRank), ret); | 181 | + HCCL_ERROR("[Reduce][InPartOneToEven]rank[%u] rx ack from peerRank[%u] failed", rank, peerRank), ret); |
| 182 | // 发送到对端的output | 182 | // 发送到对端的output |
| 183 | HCCL_DEBUG( | 183 | HCCL_DEBUG( |
| 184 | "rank[%u] sends inputMem[%p] to PeerRank[%u] Offset[%llu], Size[%llu]", rank, inputMem_.ptr(), peerRank, | 184 | "rank[%u] sends inputMem[%p] to PeerRank[%u] Offset[%llu], Size[%llu]", rank, inputMem_.ptr(), peerRank, |
| 185 | baseOffset_, inputMem_.size()); | 185 | baseOffset_, inputMem_.size()); |
| 186 | ret = senderInfo_->run(link, baseOffset_, inputMem_, stream_); | 186 | ret = senderInfo_->run(link, baseOffset_, inputMem_, stream_); |
| 187 | CHK_PRT_RET( | 187 | CHK_PRT_RET( |
| 188 | - ret != HCCL_SUCCESS, HCCL_ERROR("[Reduce][InPartOne]tx sync to peerank[%u] failed", peerRank), ret); | 188 | + ret != HCCL_SUCCESS, HCCL_ERROR("[Reduce][InPartOne]tx sync to peerRank[%u] failed", peerRank), ret); |
| 189 | ret = link->RxAsync(UserMemType::OUTPUT_MEM, baseOffset_, inputMem_.ptr(), 0, stream_); | 189 | ret = link->RxAsync(UserMemType::OUTPUT_MEM, baseOffset_, inputMem_.ptr(), 0, stream_); |
| 190 | CHK_PRT_RET( | 190 | CHK_PRT_RET( |
| 191 | ret != HCCL_SUCCESS, | 191 | ret != HCCL_SUCCESS, |
| @@ -284,13 +284,13 @@ HcclResult AllReduceRecursiveHalvingDoubling::GatherInPartOne(u32 rank, const st | |||
| 284 | HcclResult ret = links[peerRank]->TxAck(stream_); | 284 | HcclResult ret = links[peerRank]->TxAck(stream_); |
| 285 | CHK_PRT_RET( | 285 | CHK_PRT_RET( |
| 286 | ret != HCCL_SUCCESS, | 286 | ret != HCCL_SUCCESS, |
| 287 | - HCCL_ERROR("[Gather][InPartOneToEven]rank[%u] tx ack from peerank[%u] failed", rank, peerRank), ret); | 287 | + HCCL_ERROR("[Gather][InPartOneToEven]rank[%u] tx ack from peerRank[%u] failed", rank, peerRank), ret); |
| 288 | ret = links[peerRank]->RxAck(stream_); | 288 | ret = links[peerRank]->RxAck(stream_); |
| 289 | CHK_PRT_RET( | 289 | CHK_PRT_RET( |
| 290 | ret != HCCL_SUCCESS, | 290 | ret != HCCL_SUCCESS, |
| 291 | - HCCL_ERROR("[Gather][InPartOneToEven]rank[%u] rx ack from peerank[%u] failed", rank, peerRank), ret); | 291 | + HCCL_ERROR("[Gather][InPartOneToEven]rank[%u] rx ack from peerRank[%u] failed", rank, peerRank), ret); |
[建议] 同文件 本 hunk 已改 L287/L291 等处的
请一并改为 补充:同目录
![]() ![]() | |||
| 292 | HCCL_DEBUG( | 292 | HCCL_DEBUG( |
| 293 | - "rank[%u] outputMem[%p] sends to peerrank[%u] outputmem, offset[%llu], size[%llu]", rank, | 293 | + "rank[%u] outputMem[%p] sends to peerRank[%u] outputMem, offset[%llu], size[%llu]", rank, |
| 294 | outputMem_.ptr(), peerRank, baseOffset_, outputMem_.size()); | 294 | outputMem_.ptr(), peerRank, baseOffset_, outputMem_.size()); |
| 295 | ret = ExecuteTxSync( | 295 | ret = ExecuteTxSync( |
| 296 | links[peerRank], UserMemType::OUTPUT_MEM, baseOffset_, outputMem_.ptr(), outputMem_.size(), stream_); | 296 | links[peerRank], UserMemType::OUTPUT_MEM, baseOffset_, outputMem_.ptr(), outputMem_.size(), stream_); |
| @@ -313,11 +313,11 @@ HcclResult AllReduceRecursiveHalvingDoubling::GatherInPartOne(u32 rank, const st | |||
| 313 | HcclResult ret = links[peerRank]->TxAck(stream_); | 313 | HcclResult ret = links[peerRank]->TxAck(stream_); |
| 314 | CHK_PRT_RET( | 314 | CHK_PRT_RET( |
| 315 | ret != HCCL_SUCCESS, | 315 | ret != HCCL_SUCCESS, |
| 316 | - HCCL_ERROR("[Gather][InPartOneToEven]rank[%u] tx ack from peerank[%u] failed", rank, peerRank), ret); | 316 | + HCCL_ERROR("[Gather][InPartOneToEven]rank[%u] tx ack from peerRank[%u] failed", rank, peerRank), ret); |
| 317 | ret = links[peerRank]->RxAck(stream_); | 317 | ret = links[peerRank]->RxAck(stream_); |
| 318 | CHK_PRT_RET( | 318 | CHK_PRT_RET( |
| 319 | ret != HCCL_SUCCESS, | 319 | ret != HCCL_SUCCESS, |
| 320 | - HCCL_ERROR("[Gather][InPartOneToEven]rank[%u] rx ack from peerank[%u] failed", rank, peerRank), ret); | 320 | + HCCL_ERROR("[Gather][InPartOneToEven]rank[%u] rx ack from peerRank[%u] failed", rank, peerRank), ret); |
| 321 | // 等待对端可以接收数据 | 321 | // 等待对端可以接收数据 |
| 322 | HCCL_DEBUG( | 322 | HCCL_DEBUG( |
| 323 | "rank[%u] outputMem[%p] receive from PeerRank[%u] outputMem, Offset[%llu], " | 323 | "rank[%u] outputMem[%p] receive from PeerRank[%u] outputMem, Offset[%llu], " |
| @@ -149,7 +149,7 @@ HcclResult AllReduceReduceBcast::RunReduce(u32 rank, u32 rankSize, const std::ve | |||
| 149 | } else { | 149 | } else { |
| 150 | ret = RunAllReduceBDReduceSend(rank, 0, links); | 150 | ret = RunAllReduceBDReduceSend(rank, 0, links); |
| 151 | } | 151 | } |
| 152 | - CHK_PRT_RET(ret != HCCL_SUCCESS, HCCL_ERROR("[AllReduceReduceBcastReduce]rank[%u]failed", rank), ret); | 152 | + CHK_PRT_RET(ret != HCCL_SUCCESS, HCCL_ERROR("[AllReduceReduceBcastReduce]rank[%u] failed", rank), ret); |
| 153 | return HCCL_SUCCESS; | 153 | return HCCL_SUCCESS; |
| 154 | } | 154 | } |
| 155 | 155 | ||
| @@ -177,7 +177,7 @@ HcclResult AllReduceReduceBcast::RunBroadcast(u32 rank, u32 rankSize, const std: | |||
| 177 | } else { | 177 | } else { |
| 178 | ret = RunAllReduceBDMemcpyReceive(rank, 0, links); | 178 | ret = RunAllReduceBDMemcpyReceive(rank, 0, links); |
| 179 | } | 179 | } |
| 180 | - CHK_PRT_RET(ret != HCCL_SUCCESS, HCCL_ERROR("[AllReduceReduceBcast]rank[%u]failed", rank), ret); | 180 | + CHK_PRT_RET(ret != HCCL_SUCCESS, HCCL_ERROR("[AllReduceReduceBcast]rank[%u] failed", rank), ret); |
| 181 | 181 | ||
| 182 | HCCL_INFO("AllReduceReduceBcast RunBroadcast: rank[%u]", rank); | 182 | HCCL_INFO("AllReduceReduceBcast RunBroadcast: rank[%u]", rank); |
| 183 | return HCCL_SUCCESS; | 183 | return HCCL_SUCCESS; |
| @@ -42,7 +42,7 @@ HcclResult AllReduceRing::RunAsync(const u32 rank, const u32 rankSize, const std | |||
| 42 | CHK_PRT_RET( | 42 | CHK_PRT_RET( |
| 43 | ret != HCCL_SUCCESS, | 43 | ret != HCCL_SUCCESS, |
| 44 | HCCL_ERROR( | 44 | HCCL_ERROR( |
| 45 | - "[AllReduceRing][RunAsync]rank[%u] count[%llu] failed in reducescater " | 45 | + "[AllReduceRing][RunAsync]rank[%u] count[%llu] failed in ReduceScatter " |
| 46 | "step", | 46 | "step", |
| 47 | rank, count_), | 47 | rank, count_), |
| 48 | ret); | 48 | ret); |
| @@ -86,7 +86,7 @@ AllReduceRing::RunAsyncStaged(const u32 rank, const u32 rankSize, const std::vec | |||
| 86 | ret != HCCL_SUCCESS, | 86 | ret != HCCL_SUCCESS, |
| 87 | HCCL_ERROR( | 87 | HCCL_ERROR( |
| 88 | "[AllReduceRing][RunAsyncStaged]rank[%u] count[%llu] " | 88 | "[AllReduceRing][RunAsyncStaged]rank[%u] count[%llu] " |
| 89 | - "failed in reducescater step", | 89 | + "failed in ReduceScatter step", |
| 90 | rank, count_), | 90 | rank, count_), |
| 91 | ret); | 91 | ret); |
| 92 | break; | 92 | break; |
| @@ -102,7 +102,7 @@ AllReduceRing::RunAsyncStaged(const u32 rank, const u32 rankSize, const std::vec | |||
| 102 | ret); | 102 | ret); |
| 103 | break; | 103 | break; |
| 104 | default: | 104 | default: |
| 105 | - HCCL_ERROR("[AllReduceRing][RunAsyncStaged]stage[%d]is not support", stage); | 105 | + HCCL_ERROR("[AllReduceRing][RunAsyncStaged]stage[%d] is not supported", stage); |
| 106 | return HCCL_E_NOT_SUPPORT; | 106 | return HCCL_E_NOT_SUPPORT; |
| 107 | } | 107 | } |
| 108 | HCCL_INFO("AllReduceRing RunAsyncStaged stage[%d] finished: rank[%u] ranksize[%u]", stage, rank, rankSize); | 108 | HCCL_INFO("AllReduceRing RunAsyncStaged stage[%d] finished: rank[%u] ranksize[%u]", stage, rank, rankSize); |
| @@ -201,7 +201,7 @@ HcclResult GatherMesh::AddMainSteamSubStreamSyncPre(u32 rank, u32 rankSize) | |||
| 201 | meshStreams_[streamIndex], dispatcher_, (*meshSignalAux_)[streamIndex], profilerInput_.stage)); | 201 | meshStreams_[streamIndex], dispatcher_, (*meshSignalAux_)[streamIndex], profilerInput_.stage)); |
| 202 | 202 | ||
| 203 | HCCL_DEBUG( | 203 | HCCL_DEBUG( |
| 204 | - "rank[%u] siganl_aux index[%u] signal record signalaux[%p] ", rank, streamIndex, | 204 | + "rank[%u] signal_aux index[%u] signal record signalaux[%p] ", rank, streamIndex, |
| 205 | (*meshSignalAux_)[streamIndex]->ptr()); | 205 | (*meshSignalAux_)[streamIndex]->ptr()); |
| 206 | CHK_RET(LocalNotify::Post(stream_, dispatcher_, (*meshSignalAux_)[streamIndex], profilerInput_.stage)); | 206 | CHK_RET(LocalNotify::Post(stream_, dispatcher_, (*meshSignalAux_)[streamIndex], profilerInput_.stage)); |
| 207 | } | 207 | } |
| @@ -101,13 +101,13 @@ HcclResult ReduceRecursiveHalvingDoubling::ReduceInPartOne(u32 rank, const std:: | |||
| 101 | 101 | ||
| 102 | HcclResult ret = links[peerRank]->TxAck(stream_); | 102 | HcclResult ret = links[peerRank]->TxAck(stream_); |
| 103 | CHK_PRT_RET( | 103 | CHK_PRT_RET( |
| 104 | - ret != HCCL_SUCCESS, HCCL_ERROR("[Reduce][InPartOne]tx ack to peerrank[%u] failed", peerRank), ret); | 104 | + ret != HCCL_SUCCESS, HCCL_ERROR("[Reduce][InPartOne]tx ack to peerRank[%u] failed", peerRank), ret); |
| 105 | ret = links[peerRank]->RxAck(stream_); | 105 | ret = links[peerRank]->RxAck(stream_); |
| 106 | CHK_PRT_RET( | 106 | CHK_PRT_RET( |
| 107 | - ret != HCCL_SUCCESS, HCCL_ERROR("[Reduce][InPartOne]rx ack from peerank[%u] failed", peerRank), ret); | 107 | + ret != HCCL_SUCCESS, HCCL_ERROR("[Reduce][InPartOne]rx ack from peerRank[%u] failed", peerRank), ret); |
| 108 | 108 | ||
| 109 | // 接收数据到本端的 output | 109 | // 接收数据到本端的 output |
| 110 | - HCCL_DEBUG("send mem[%p] size[%llu] to peerank[%u]", outputMem_.ptr(), outputMem_.size(), peerRank); | 110 | + HCCL_DEBUG("send mem[%p] size[%llu] to peerRank[%u]", outputMem_.ptr(), outputMem_.size(), peerRank); |
| 111 | ret = links[peerRank]->TxAsync(UserMemType::INPUT_MEM, baseOffset_, outputMem_.ptr(), 0, stream_); | 111 | ret = links[peerRank]->TxAsync(UserMemType::INPUT_MEM, baseOffset_, outputMem_.ptr(), 0, stream_); |
| 112 | CHK_PRT_RET( | 112 | CHK_PRT_RET( |
| 113 | ret != HCCL_SUCCESS, | 113 | ret != HCCL_SUCCESS, |
| @@ -128,17 +128,17 @@ HcclResult ReduceRecursiveHalvingDoubling::ReduceInPartOne(u32 rank, const std:: | |||
| 128 | CHK_SMART_PTR_NULL(links[peerRank]); | 128 | CHK_SMART_PTR_NULL(links[peerRank]); |
| 129 | HcclResult ret = links[peerRank]->TxAck(stream_); | 129 | HcclResult ret = links[peerRank]->TxAck(stream_); |
| 130 | CHK_PRT_RET( | 130 | CHK_PRT_RET( |
| 131 | - ret != HCCL_SUCCESS, HCCL_ERROR("[Reduce][InPartOne]tx ack to peerrank[%u] failed", peerRank), ret); | 131 | + ret != HCCL_SUCCESS, HCCL_ERROR("[Reduce][InPartOne]tx ack to peerRank[%u] failed", peerRank), ret); |
| 132 | ret = links[peerRank]->RxAck(stream_); | 132 | ret = links[peerRank]->RxAck(stream_); |
| 133 | CHK_PRT_RET( | 133 | CHK_PRT_RET( |
| 134 | - ret != HCCL_SUCCESS, HCCL_ERROR("[Reduce][InPartOne]rx ack from peerank[%u] failed", peerRank), ret); | 134 | + ret != HCCL_SUCCESS, HCCL_ERROR("[Reduce][InPartOne]rx ack from peerRank[%u] failed", peerRank), ret); |
| 135 | // 发送到对端的output | 135 | // 发送到对端的output |
| 136 | HCCL_DEBUG( | 136 | HCCL_DEBUG( |
| 137 | "rank[%u] sends inputMem[%p] to PeerRank[%u] Offset[%llu], Size[%llu]", rank, inputMem_.ptr(), peerRank, | 137 | "rank[%u] sends inputMem[%p] to PeerRank[%u] Offset[%llu], Size[%llu]", rank, inputMem_.ptr(), peerRank, |
| 138 | baseOffset_, inputMem_.size()); | 138 | baseOffset_, inputMem_.size()); |
| 139 | ret = senderInfo_->run(links[peerRank], baseOffset_, inputMem_, stream_); | 139 | ret = senderInfo_->run(links[peerRank], baseOffset_, inputMem_, stream_); |
| 140 | CHK_PRT_RET( | 140 | CHK_PRT_RET( |
| 141 | - ret != HCCL_SUCCESS, HCCL_ERROR("[Reduce][InPartOne]tx sync to peerank[%u] failed", peerRank), ret); | 141 | + ret != HCCL_SUCCESS, HCCL_ERROR("[Reduce][InPartOne]tx sync to peerRank[%u] failed", peerRank), ret); |
| 142 | ret = links[peerRank]->RxAsync(UserMemType::OUTPUT_MEM, baseOffset_, inputMem_.ptr(), 0, stream_); | 142 | ret = links[peerRank]->RxAsync(UserMemType::OUTPUT_MEM, baseOffset_, inputMem_.ptr(), 0, stream_); |
| 143 | CHK_PRT_RET( | 143 | CHK_PRT_RET( |
| 144 | ret != HCCL_SUCCESS, | 144 | ret != HCCL_SUCCESS, |
| @@ -290,11 +290,11 @@ HcclResult ReduceRecursiveHalvingDoubling::GatherInBlock(u32 rank, u32 rankSize, | |||
| 290 | HcclResult ret = subLinks[peerRank]->TxAck(stream_); | 290 | HcclResult ret = subLinks[peerRank]->TxAck(stream_); |
| 291 | CHK_PRT_RET( | 291 | CHK_PRT_RET( |
| 292 | ret != HCCL_SUCCESS, | 292 | ret != HCCL_SUCCESS, |
| 293 | - HCCL_ERROR("[Gather][InBlock]rank[%u] tx ack from peerank[%u] failed", rank, peerRank), ret); | 293 | + HCCL_ERROR("[Gather][InBlock]rank[%u] tx ack from peerRank[%u] failed", rank, peerRank), ret); |
| 294 | ret = subLinks[peerRank]->RxAck(stream_); | 294 | ret = subLinks[peerRank]->RxAck(stream_); |
| 295 | CHK_PRT_RET( | 295 | CHK_PRT_RET( |
| 296 | ret != HCCL_SUCCESS, | 296 | ret != HCCL_SUCCESS, |
| 297 | - HCCL_ERROR("[Gather][InBlock]rank[%u] rx ack from peerank[%u] failed", rank, peerRank), ret); | 297 | + HCCL_ERROR("[Gather][InBlock]rank[%u] rx ack from peerRank[%u] failed", rank, peerRank), ret); |
| 298 | 298 | ||
| 299 | // 等待对端可以接收数据 | 299 | // 等待对端可以接收数据 |
| 300 | HCCL_DEBUG( | 300 | HCCL_DEBUG( |
| @@ -315,13 +315,13 @@ HcclResult ReduceRecursiveHalvingDoubling::GatherInBlock(u32 rank, u32 rankSize, | |||
| 315 | HcclResult ret = subLinks[peerRank]->TxAck(stream_); | 315 | HcclResult ret = subLinks[peerRank]->TxAck(stream_); |
| 316 | CHK_PRT_RET( | 316 | CHK_PRT_RET( |
| 317 | ret != HCCL_SUCCESS, | 317 | ret != HCCL_SUCCESS, |
| 318 | - HCCL_ERROR("[Gather][InBlock]rank[%u] tx ack from peerank[%u] failed", rank, peerRank), ret); | 318 | + HCCL_ERROR("[Gather][InBlock]rank[%u] tx ack from peerRank[%u] failed", rank, peerRank), ret); |
| 319 | ret = subLinks[peerRank]->RxAck(stream_); | 319 | ret = subLinks[peerRank]->RxAck(stream_); |
| 320 | CHK_PRT_RET( | 320 | CHK_PRT_RET( |
| 321 | ret != HCCL_SUCCESS, | 321 | ret != HCCL_SUCCESS, |
| 322 | - HCCL_ERROR("[Gather][InBlock]rank[%u] rx ack from peerank[%u] failed", rank, peerRank), ret); | 322 | + HCCL_ERROR("[Gather][InBlock]rank[%u] rx ack from peerRank[%u] failed", rank, peerRank), ret); |
| 323 | HCCL_DEBUG( | 323 | HCCL_DEBUG( |
| 324 | - "rank[%u] outputMem[%p] sends to peerrank[%u] outputmem, offset[%llu], " | 324 | + "rank[%u] outputMem[%p] sends to peerRank[%u] outputmem, offset[%llu], " |
| 325 | "size[%llu]", | 325 | "size[%llu]", |
| 326 | rank, outputMem_.ptr(), peerRank, baseOffset_ + txSlices_[step].offset, txSlices_[step].size); | 326 | rank, outputMem_.ptr(), peerRank, baseOffset_ + txSlices_[step].offset, txSlices_[step].size); |
| 327 | ret = ExecuteTxSync( | 327 | ret = ExecuteTxSync( |
| @@ -187,7 +187,7 @@ HcclResult ReduceScatterGraphPipeline::Prepare( | |||
| 187 | streamNotifyMain_ = notifyMain; | 187 | streamNotifyMain_ = notifyMain; |
| 188 | if (streamNotifyMain_.size() < intraRankSize_) { | 188 | if (streamNotifyMain_.size() < intraRankSize_) { |
| 189 | HCCL_ERROR( | 189 | HCCL_ERROR( |
| 190 | - "[ReduceScatterGraphPipeline][Prepare]rank[%u] streamNotifyMain_ size [%u] error, is smaller than," | 190 | + "[ReduceScatterGraphPipeline][Prepare]rank[%u] streamNotifyMain_ size [%u] error, smaller than " |
| 191 | "intraRankSize_[%u]", | 191 | "intraRankSize_[%u]", |
| 192 | rankId_, streamNotifyMain_.size(), intraRankSize_); | 192 | rankId_, streamNotifyMain_.size(), intraRankSize_); |
| 193 | return HCCL_E_INTERNAL; | 193 | return HCCL_E_INTERNAL; |
| @@ -195,7 +195,7 @@ HcclResult ReduceScatterGraphPipeline::Prepare( | |||
| 195 | streamNotifySub_ = notifySub; | 195 | streamNotifySub_ = notifySub; |
| 196 | if (streamNotifySub_.size() < intraRankSize_) { | 196 | if (streamNotifySub_.size() < intraRankSize_) { |
| 197 | HCCL_ERROR( | 197 | HCCL_ERROR( |
| 198 | - "[ReduceScatterGraphPipeline][Prepare]rank[%u] streamNotifySub_ size [%u] error, is smaller than," | 198 | + "[ReduceScatterGraphPipeline][Prepare]rank[%u] streamNotifySub_ size [%u] error, smaller than " |
| 199 | "intraRankSize_[%u]", | 199 | "intraRankSize_[%u]", |
| 200 | rankId_, streamNotifySub_.size(), intraRankSize_); | 200 | rankId_, streamNotifySub_.size(), intraRankSize_); |
| 201 | return HCCL_E_INTERNAL; | 201 | return HCCL_E_INTERNAL; |
| @@ -274,7 +274,7 @@ HcclResult ReduceScatterNB::RunDestReducerNB( | |||
| 274 | DeviceMem srcMem = inputMem_.range(inputSlices[rxIndex].offset, inputSlices[rxIndex].size); | 274 | DeviceMem srcMem = inputMem_.range(inputSlices[rxIndex].offset, inputSlices[rxIndex].size); |
| 275 | DeviceMem scratchMem = scratchMem_.range(outputSlices[rxIndex].offset, outputSlices[rxIndex].size); | 275 | DeviceMem scratchMem = scratchMem_.range(outputSlices[rxIndex].offset, outputSlices[rxIndex].size); |
| 276 | HCCL_DEBUG( | 276 | HCCL_DEBUG( |
| 277 | - "final reduce rxSliceIdx[%u] will reduce with inputMem_ offset[%llu] to ouput_mem_ " | 277 | + "final reduce rxSliceIdx[%u] will reduce with inputMem_ offset[%llu] to output_mem_ " |
| 278 | "offset[%llu] size[%llu]", | 278 | "offset[%llu] size[%llu]", |
| 279 | rxIndex, inputSlices[rxIndex].offset, outputSlices[rxIndex].offset, outputSlices[rxIndex].size); | 279 | rxIndex, inputSlices[rxIndex].offset, outputSlices[rxIndex].offset, outputSlices[rxIndex].size); |
| 280 | 280 | ||
| @@ -310,14 +310,14 @@ HcclResult ReduceScatterNHRV1::RunLastCopyStep(const u32 rank, const std::vector | |||
| 310 | ret = links[peerRank]->TxAck(stream_); | 310 | ret = links[peerRank]->TxAck(stream_); |
| 311 | CHK_PRT_RET( | 311 | CHK_PRT_RET( |
| 312 | ret != HCCL_SUCCESS, | 312 | ret != HCCL_SUCCESS, |
| 313 | - HCCL_ERROR("[ReduceScatterNHRV1][RunLastCopyStep]rank[%u] tx ack from peerank[%u] failed", rank, peerRank), | 313 | + HCCL_ERROR("[ReduceScatterNHRV1][RunLastCopyStep]rank[%u] tx ack from peerRank[%u] failed", rank, peerRank), |
| 314 | ret); | 314 | ret); |
| 315 | 315 | ||
| 316 | // RxAck | 316 | // RxAck |
| 317 | ret = links[peerRank]->RxAck(stream_); | 317 | ret = links[peerRank]->RxAck(stream_); |
| 318 | CHK_PRT_RET( | 318 | CHK_PRT_RET( |
| 319 | ret != HCCL_SUCCESS, | 319 | ret != HCCL_SUCCESS, |
| 320 | - HCCL_ERROR("[ReduceScatterNHRV1][RunLastCopyStep]rank[%u] rx ack from peerank[%u] failed", rank, peerRank), | 320 | + HCCL_ERROR("[ReduceScatterNHRV1][RunLastCopyStep]rank[%u] rx ack from peerRank[%u] failed", rank, peerRank), |
| 321 | ret); | 321 | ret); |
| 322 | 322 | ||
| 323 | if (hIndex == info.GetRowSize() - 1) { // 发数据 | 323 | if (hIndex == info.GetRowSize() - 1) { // 发数据 |
| @@ -74,7 +74,7 @@ HcclResult ReduceScatterRecursiveHalvingDoubling::RunAsync( | |||
| 74 | part1Size_, blockSize_, rankSize), | 74 | part1Size_, blockSize_, rankSize), |
| 75 | ret); | 75 | ret); |
| 76 | 76 | ||
| 77 | - HCCL_DEBUG("rank[%u] calculate par1size[%u] blocksize[%u] ranksize[%u]", rank, part1Size_, blockSize_, rankSize); | 77 | + HCCL_DEBUG("rank[%u] calculate part1size[%u] blocksize[%u] ranksize[%u]", rank, part1Size_, blockSize_, rankSize); |
| 78 | 78 | ||
| 79 | CHK_RET(ReduceInPartOne(rank, links)); | 79 | CHK_RET(ReduceInPartOne(rank, links)); |
| 80 | 80 | ||
| @@ -245,9 +245,9 @@ ReduceScatterRecursiveHalvingDoubling::ScatterInPartOne(u32 rank, u32 rankSize, | |||
| 245 | ret = ExecuteTxSync( | 245 | ret = ExecuteTxSync( |
| 246 | links[peerRank], UserMemType::INPUT_MEM, offset + baseOffset_, srcAddr, bytesPerSlice, stream_); | 246 | links[peerRank], UserMemType::INPUT_MEM, offset + baseOffset_, srcAddr, bytesPerSlice, stream_); |
| 247 | CHK_PRT_RET( | 247 | CHK_PRT_RET( |
| 248 | - ret != HCCL_SUCCESS, HCCL_ERROR("[Scatter][InPartOne]tx sync to peerank[%u] failed", peerRank), ret); | 248 | + ret != HCCL_SUCCESS, HCCL_ERROR("[Scatter][InPartOne]tx sync to peerRank[%u] failed", peerRank), ret); |
| 249 | ret = links[peerRank]->TxWaitDone(stream_); | 249 | ret = links[peerRank]->TxWaitDone(stream_); |
| 250 | - CHK_PRT_RET(ret != HCCL_SUCCESS, HCCL_ERROR("[Reduce][InPartOne]TxWaitDone failed"), ret); | 250 | + CHK_PRT_RET(ret != HCCL_SUCCESS, HCCL_ERROR("[Scatter][InPartOne]TxWaitDone failed"), ret); |
[建议] 残留 本文件改了 L250 的
建议改为 ![]() ![]() | |||
| 251 | } | 251 | } |
| 252 | } else if (rank < part1Size_ && rank % 2 == 1) { // 模2计算奇偶性,奇数rank接收偶数rank发过来下半份的数据 | 252 | } else if (rank < part1Size_ && rank % 2 == 1) { // 模2计算奇偶性,奇数rank接收偶数rank发过来下半份的数据 |
| 253 | u32 peerRank = rank - 1; | 253 | u32 peerRank = rank - 1; |
| @@ -269,9 +269,9 @@ ReduceScatterRecursiveHalvingDoubling::ScatterInPartOne(u32 rank, u32 rankSize, | |||
| 269 | ret = ExecuteRxSync( | 269 | ret = ExecuteRxSync( |
| 270 | links[peerRank], UserMemType::INPUT_MEM, offset + baseOffset_, dstAddr, bytesPerSlice, stream_); | 270 | links[peerRank], UserMemType::INPUT_MEM, offset + baseOffset_, dstAddr, bytesPerSlice, stream_); |
| 271 | CHK_PRT_RET( | 271 | CHK_PRT_RET( |
| 272 | - ret != HCCL_SUCCESS, HCCL_ERROR("[Scatter][InPartOne]rx sync from peerank[%u] failed", peerRank), ret); | 272 | + ret != HCCL_SUCCESS, HCCL_ERROR("[Scatter][InPartOne]rx sync from peerRank[%u] failed", peerRank), ret); |
| 273 | ret = links[peerRank]->RxWaitDone(stream_); | 273 | ret = links[peerRank]->RxWaitDone(stream_); |
| 274 | - CHK_PRT_RET(ret != HCCL_SUCCESS, HCCL_ERROR("[Reduce][InPartOne]RxWaitDone failed"), ret); | 274 | + CHK_PRT_RET(ret != HCCL_SUCCESS, HCCL_ERROR("[Scatter][InPartOne]RxWaitDone failed"), ret); |
| 275 | } | 275 | } |
| 276 | } | 276 | } |
| 277 | 277 | ||
| @@ -80,7 +80,7 @@ HcclResult ScatterRing::RunScatterOnEndRank() | |||
| 80 | CHK_RET(linkLeft_->TxAck(stream_)); | 80 | CHK_RET(linkLeft_->TxAck(stream_)); |
| 81 | 81 | ||
| 82 | dst = outputMem_.range(scatterOffset, scatterResult); | 82 | dst = outputMem_.range(scatterOffset, scatterResult); |
| 83 | - HCCL_DEBUG("last rank[%u] rx data ouputoffset[%llu] size[%llu]", interRank_, scatterOffset, scatterResult); | 83 | + HCCL_DEBUG("last rank[%u] rx data outputoffset[%llu] size[%llu]", interRank_, scatterOffset, scatterResult); |
| 84 | HcclResult ret | 84 | HcclResult ret |
| 85 | = linkLeft_->RxAsync(UserMemType::OUTPUT_MEM, scatterOffset + baseOffset_, dst.ptr(), scatterResult, stream_); | 85 | = linkLeft_->RxAsync(UserMemType::OUTPUT_MEM, scatterOffset + baseOffset_, dst.ptr(), scatterResult, stream_); |
| 86 | CHK_PRT_RET( | 86 | CHK_PRT_RET( |
| @@ -659,7 +659,7 @@ HcclResult CommBase::CreateDestLink( | |||
| 659 | MachinePara machinePara; | 659 | MachinePara machinePara; |
| 660 | CHK_RET(SetMachinePara(machineType, serverId, dstRank, sockets, machinePara)); | 660 | CHK_RET(SetMachinePara(machineType, serverId, dstRank, sockets, machinePara)); |
| 661 | HCCL_INFO( | 661 | HCCL_INFO( |
| 662 | - "[creakLink para]rank[%u]-localUserrank[%u]-localIpAddr[%s], linkMode[%d] " | 662 | + "[createLink para]rank[%u]-localUserrank[%u]-localIpAddr[%s], linkMode[%d] " |
| 663 | "dst_rank[%u]-remoteUserrank[%u]-remote_ip_addr[%s], machineType[%d], serverId[%s], nicDeploy[%d] ", | 663 | "dst_rank[%u]-remoteUserrank[%u]-remote_ip_addr[%s], machineType[%d], serverId[%s], nicDeploy[%d] ", |
| 664 | rank_, paraVector_[rank_].worldRank, paraVector_[rank_].serverId.c_str(), machinePara.linkMode, dstRank, | 664 | rank_, paraVector_[rank_].worldRank, paraVector_[rank_].serverId.c_str(), machinePara.linkMode, dstRank, |
| 665 | paraVector_[dstRank].worldRank, paraVector_[dstRank].serverId.c_str(), machinePara.machineType, | 665 | paraVector_[dstRank].worldRank, paraVector_[dstRank].serverId.c_str(), machinePara.machineType, |
| @@ -677,14 +677,14 @@ HcclResult CommBase::CreateDestLink( | |||
| 677 | HCCL_ERROR("%s", err_str.c_str()); | 677 | HCCL_ERROR("%s", err_str.c_str()); |
| 678 | } | 678 | } |
| 679 | const std::string CREATE_LINK_ERR | 679 | const std::string CREATE_LINK_ERR |
| 680 | - = "[Create][DestLink]Create Dest error! creakLink para:rank[" + std::to_string(rank_) + "]-localUserrank[" | 680 | + = "[Create][DestLink]Create Dest error! createLink para:rank[" + std::to_string(rank_) + "]-localUserrank[" |
| 681 | + std::to_string(paraVector_[rank_].worldRank) + "]-localIpAddr[" + paraVector_[rank_].serverId.c_str() | 681 | + std::to_string(paraVector_[rank_].worldRank) + "]-localIpAddr[" + paraVector_[rank_].serverId.c_str() |
| 682 | + "], dst_rank[" + std::to_string(dstRank) + "]-remoteUserrank[" | 682 | + "], dst_rank[" + std::to_string(dstRank) + "]-remoteUserrank[" |
| 683 | + std::to_string(paraVector_[dstRank].worldRank) + "]-remote_ip_addr[" | 683 | + std::to_string(paraVector_[dstRank].worldRank) + "]-remote_ip_addr[" |
| 684 | + paraVector_[dstRank].serverId.c_str() + "]"; | 684 | + paraVector_[dstRank].serverId.c_str() + "]"; |
| 685 | 685 | ||
| 686 | HCCL_ERROR( | 686 | HCCL_ERROR( |
| 687 | - "[Create][DestLink]Transport init error! creakLink para:rank[%u]-localUserrank[%u]-localIpAddr[%s], " | 687 | + "[Create][DestLink]Transport init error! createLink para:rank[%u]-localUserrank[%u]-localIpAddr[%s], " |
| 688 | "dst_rank[%u]-remoteUserrank[%u]-remote_ip_addr[%s], machineType[%d], serverId[%s], linkMode[%d], " | 688 | "dst_rank[%u]-remoteUserrank[%u]-remote_ip_addr[%s], machineType[%d], serverId[%s], linkMode[%d], " |
| 689 | "shmDev_[%u], tag[%s]", | 689 | "shmDev_[%u], tag[%s]", |
| 690 | rank_, paraVector_[rank_].worldRank, paraVector_[rank_].serverId.c_str(), dstRank, | 690 | rank_, paraVector_[rank_].worldRank, paraVector_[rank_].serverId.c_str(), dstRank, |
| @@ -693,7 +693,7 @@ HcclResult CommBase::CreateDestLink( | |||
| 693 | return ret; | 693 | return ret; |
| 694 | } | 694 | } |
| 695 | HCCL_INFO( | 695 | HCCL_INFO( |
| 696 | - "[creakLink success]:rank[%u]-localUserrank[%u]-localIpAddr[%s], " | 696 | + "[createLink success]:rank[%u]-localUserrank[%u]-localIpAddr[%s], " |
| 697 | "dst_rank[%u]-remoteUserrank[%u]-remote_ip_addr[%s], transportType_[%d], tag[%s]", | 697 | "dst_rank[%u]-remoteUserrank[%u]-remote_ip_addr[%s], transportType_[%d], tag[%s]", |
| 698 | rank_, paraVector_[rank_].worldRank, paraVector_[rank_].serverId.c_str(), dstRank, | 698 | rank_, paraVector_[rank_].worldRank, paraVector_[rank_].serverId.c_str(), dstRank, |
| 699 | paraVector_[dstRank].worldRank, paraVector_[dstRank].serverId.c_str(), transportType_[dstRank], | 699 | paraVector_[dstRank].worldRank, paraVector_[dstRank].serverId.c_str(), transportType_[dstRank], |
| @@ -714,7 +714,7 @@ HcclResult CommBase::TransportInit(const u32 dstRank, MachinePara& machinePara) | |||
| 714 | { | 714 | { |
| 715 | CHK_PRT_RET( | 715 | CHK_PRT_RET( |
| 716 | dstRank >= transportInfo_.size(), | 716 | dstRank >= transportInfo_.size(), |
| 717 | - HCCL_ERROR("[TransportQuerry] Transport[%u] is invalid, should init before query it.", dstRank), HCCL_E_PARA); | 717 | + HCCL_ERROR("[TransportQuery] Transport[%u] is invalid, should init before query it.", dstRank), HCCL_E_PARA); |
| 718 | // 实例化TransportBase | 718 | // 实例化TransportBase |
| 719 | CHK_RET(SetTransportType(dstRank)); | 719 | CHK_RET(SetTransportType(dstRank)); |
| 720 | TransportPara para{}; | 720 | TransportPara para{}; |
| @@ -904,7 +904,7 @@ HcclResult CommBase::BuildQuerry(u32& status) | |||
| 904 | } | 904 | } |
| 905 | } | 905 | } |
| 906 | CHK_RET(GetBuildStatus(status)); | 906 | CHK_RET(GetBuildStatus(status)); |
| 907 | - HCCL_DEBUG("BuildQuerry: %u", status); | 907 | + HCCL_DEBUG("BuildQuery: %u", status); |
| 908 | return HCCL_SUCCESS; | 908 | return HCCL_SUCCESS; |
| 909 | } | 909 | } |
| 910 | 910 | ||
| @@ -944,7 +944,7 @@ HcclResult CommBase::TransportBuildAsync( | |||
| 944 | { | 944 | { |
| 945 | CHK_PRT_RET( | 945 | CHK_PRT_RET( |
| 946 | dstRank >= transportInfo_.size(), | 946 | dstRank >= transportInfo_.size(), |
| 947 | - HCCL_ERROR("[TransportQuerry] Transport[%u] is invalid, should init before query it.", dstRank), HCCL_E_PARA); | 947 | + HCCL_ERROR("[TransportQuery] Transport[%u] is invalid, should init before query it.", dstRank), HCCL_E_PARA); |
| 948 | MachinePara machinePara; | 948 | MachinePara machinePara; |
| 949 | CHK_RET(SetMachinePara(machineType, serverId, dstRank, sockets, machinePara)); | 949 | CHK_RET(SetMachinePara(machineType, serverId, dstRank, sockets, machinePara)); |
| 950 | // 实例化TransportBase | 950 | // 实例化TransportBase |
| @@ -971,7 +971,7 @@ HcclResult CommBase::TransportBuildQuerry(u32 dstRank, u32& status) | |||
| 971 | { | 971 | { |
| 972 | CHK_PRT_RET( | 972 | CHK_PRT_RET( |
| 973 | dstRank >= transportInfo_.size(), | 973 | dstRank >= transportInfo_.size(), |
| 974 | - HCCL_ERROR("[TransportQuerry] Transport[%u] is invalid, should init before query it.", dstRank), HCCL_E_PARA); | 974 | + HCCL_ERROR("[TransportQuery] Transport[%u] is invalid, should init before query it.", dstRank), HCCL_E_PARA); |
| 975 | if (transportInfo_[dstRank]) { | 975 | if (transportInfo_[dstRank]) { |
| 976 | CHK_RET(transportInfo_[dstRank]->ConnectQuerry(status)); | 976 | CHK_RET(transportInfo_[dstRank]->ConnectQuerry(status)); |
| 977 | if (status == HETEROG_P2P_SUCCESS && checkStatus_[dstRank] == false) { | 977 | if (status == HETEROG_P2P_SUCCESS && checkStatus_[dstRank] == false) { |
| @@ -981,7 +981,7 @@ HcclResult CommBase::TransportBuildQuerry(u32 dstRank, u32& status) | |||
| 981 | } else { | 981 | } else { |
| 982 | status = HETEROG_P2P_WAIT; | 982 | status = HETEROG_P2P_WAIT; |
| 983 | } | 983 | } |
| 984 | - HCCL_DEBUG("TransportBuildQuerry[%u] %u", dstRank, status); | 984 | + HCCL_DEBUG("TransportBuildQuery[%u] %u", dstRank, status); |
| 985 | return HCCL_SUCCESS; | 985 | return HCCL_SUCCESS; |
| 986 | } | 986 | } |
| 987 | 987 | ||
| @@ -58,7 +58,7 @@ HcclResult CollAllGatherAivRdmaExecutor::CalNumBlocks( | |||
| 58 | HCCL_E_PARA); | 58 | HCCL_E_PARA); |
| 59 | 59 | ||
| 60 | HCCL_INFO( | 60 | HCCL_INFO( |
| 61 | - "[CollAllGatherAivRdmaExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommanded[%u]", numBlocks, | 61 | + "[CollAllGatherAivRdmaExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommended[%u]", numBlocks, |
| 62 | numBlocks_, bestNumBlocks); | 62 | numBlocks_, bestNumBlocks); |
| 63 | return HCCL_SUCCESS; | 63 | return HCCL_SUCCESS; |
| 64 | } | 64 | } |
| @@ -93,7 +93,7 @@ HcclResult CollAllGatherMeshAivExecutor::CalNumBlocks( | |||
| 93 | } | 93 | } |
| 94 | 94 | ||
| 95 | HCCL_INFO( | 95 | HCCL_INFO( |
| 96 | - "[CollAllGatherMeshAivExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommanded[%u]", numBlocks, | 96 | + "[CollAllGatherMeshAivExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommended[%u]", numBlocks, |
| 97 | numBlocks_, bestNumBlocks); | 97 | numBlocks_, bestNumBlocks); |
| 98 | return HCCL_SUCCESS; | 98 | return HCCL_SUCCESS; |
| 99 | } | 99 | } |
| @@ -375,7 +375,7 @@ HcclResult CollAllGatherRingFor91093Executor::KernelRun(const OpParam& param, Ex | |||
| 375 | 375 | ||
| 376 | CHK_RET(RunTemplate(level2AGExecutor, level2CommInfo)); | 376 | CHK_RET(RunTemplate(level2AGExecutor, level2CommInfo)); |
| 377 | HCCL_INFO( | 377 | HCCL_INFO( |
| 378 | - "AllGather ring [superpod] level2 AllGather run successtopoType_[%u], agv[%u]", topoType_, isAllGatherV_); | 378 | + "AllGather ring [superpod] level2 AllGather run success, topoType_[%u], agv[%u]", topoType_, isAllGatherV_); |
| 379 | } | 379 | } |
| 380 | if (level1RankSize > 1) { | 380 | if (level1RankSize > 1) { |
| 381 | // 计算slice, 不同超节点相同slice | 381 | // 计算slice, 不同超节点相同slice |
| @@ -427,7 +427,7 @@ HcclResult CollAllGatherRingFor91093Executor::KernelRun(const OpParam& param, Ex | |||
| 427 | 427 | ||
| 428 | CHK_RET(RunTemplate(level1AGExecutor, logicalLevel1CommInfo_)); | 428 | CHK_RET(RunTemplate(level1AGExecutor, logicalLevel1CommInfo_)); |
| 429 | HCCL_INFO( | 429 | HCCL_INFO( |
| 430 | - "AllGather ring [superpod] level1 AllGather run successtopoType_[%u], agv[%u]", topoType_, isAllGatherV_); | 430 | + "AllGather ring [superpod] level1 AllGather run success, topoType_[%u], agv[%u]", topoType_, isAllGatherV_); |
| 431 | } | 431 | } |
| 432 | // 节点内做AllGather ring | 432 | // 节点内做AllGather ring |
| 433 | std::vector<std::vector<Slice>> multRingsSlice; | 433 | std::vector<std::vector<Slice>> multRingsSlice; |
| @@ -469,7 +469,7 @@ HcclResult CollAllGatherRingFor91093Executor::Getlevel1CommRank(SubCommInfo& lev | |||
| 469 | return HCCL_SUCCESS; | 469 | return HCCL_SUCCESS; |
| 470 | } | 470 | } |
| 471 | if (CheckCommSize(COMM_LEVEL2, COMM_INDEX_0 + 1) != HCCL_SUCCESS) { | 471 | if (CheckCommSize(COMM_LEVEL2, COMM_INDEX_0 + 1) != HCCL_SUCCESS) { |
| 472 | - HCCL_INFO("[nslbdp] Getlevel1CommRank size not match."); | 472 | + HCCL_INFO("[AllGatherRingFor91093] Getlevel1CommRank size not match."); |
| 473 | return HCCL_E_UNAVAIL; | 473 | return HCCL_E_UNAVAIL; |
| 474 | } | 474 | } |
| 475 | level1CommInfo = GetSubCommInfo(COMM_LEVEL2, COMM_INDEX_0); | 475 | level1CommInfo = GetSubCommInfo(COMM_LEVEL2, COMM_INDEX_0); |
| @@ -480,7 +480,7 @@ HcclResult CollAllGatherRingFor91093Executor::Getlevel1CommRank(SubCommInfo& lev | |||
| 480 | HcclResult | 480 | HcclResult |
| 481 | CollAllGatherRingFor91093Executor::SelectTempAlg(std::unique_ptr<AlgTemplateBase>& level1TempAlg, u32 level1RankSize) | 481 | CollAllGatherRingFor91093Executor::SelectTempAlg(std::unique_ptr<AlgTemplateBase>& level1TempAlg, u32 level1RankSize) |
| 482 | { | 482 | { |
| 483 | - HCCL_INFO("[nslbdp] Entry SelectTempAlg, level1RankSize = [%u].", level1RankSize); | 483 | + HCCL_INFO("[AllGatherRingFor91093] Entry SelectTempAlg, level1RankSize = [%u].", level1RankSize); |
| 484 | bool isSelectAHC | 484 | bool isSelectAHC |
| 485 | = (algType_.algoLevel1 == AlgTypeLevel1::ALG_LEVEL1_AHC | 485 | = (algType_.algoLevel1 == AlgTypeLevel1::ALG_LEVEL1_AHC |
| 486 | || algType_.algoLevel1 == AlgTypeLevel1::ALG_LEVEL1_AHC_BROKE); | 486 | || algType_.algoLevel1 == AlgTypeLevel1::ALG_LEVEL1_AHC_BROKE); |
| @@ -62,7 +62,7 @@ HcclResult CollAllGatherVMeshAivSmallCountExecutor::CalNumBlocks( | |||
| 62 | HCCL_E_PARA); | 62 | HCCL_E_PARA); |
| 63 | 63 | ||
| 64 | HCCL_INFO( | 64 | HCCL_INFO( |
| 65 | - "[CollAllGatherVMeshAivSmallCountExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommanded[%u]", | 65 | + "[CollAllGatherVMeshAivSmallCountExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommended[%u]", |
| 66 | numBlocks, numBlocks_, bestNumBlocks); | 66 | numBlocks, numBlocks_, bestNumBlocks); |
| 67 | return HCCL_SUCCESS; | 67 | return HCCL_SUCCESS; |
| 68 | } | 68 | } |
| @@ -155,7 +155,7 @@ HcclResult CollAllReduceMeshAivFor91093Executor::GetAivExecParam( | |||
| 155 | args.reduceOp = param.reduceType; | 155 | args.reduceOp = param.reduceType; |
| 156 | 156 | ||
| 157 | HCCL_INFO( | 157 | HCCL_INFO( |
| 158 | - "SPK [CollAllReduceMeshAivFor91093Executor][GetAivExecParam], rank[%llu], rankSize[%llu], " | 158 | + "[CollAllReduceMeshAivFor91093Executor][GetAivExecParam], rank[%llu], rankSize[%llu], " |
| 159 | "len[%llu],datatype[%llu], op[%llu]", | 159 | "len[%llu],datatype[%llu], op[%llu]", |
| 160 | args.rank, args.rankSize, args.len, args.dataType, args.reduceOp); | 160 | args.rank, args.rankSize, args.len, args.dataType, args.reduceOp); |
| 161 | 161 | ||
| @@ -72,7 +72,7 @@ HcclResult CollAllReduceMeshAivSmallCountExecutor::CalNumBlocks( | |||
| 72 | HCCL_E_PARA); | 72 | HCCL_E_PARA); |
| 73 | 73 | ||
| 74 | HCCL_INFO( | 74 | HCCL_INFO( |
| 75 | - "[CollAllReduceMeshAivSmallCountExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommanded[%u]", | 75 | + "[CollAllReduceMeshAivSmallCountExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommended[%u]", |
| 76 | numBlocks, numBlocks_, bestNumBlocks); | 76 | numBlocks, numBlocks_, bestNumBlocks); |
| 77 | return HCCL_SUCCESS; | 77 | return HCCL_SUCCESS; |
| 78 | } | 78 | } |
| @@ -113,7 +113,7 @@ HcclResult CollAllReduceMeshAivSmallCountExecutor::GetAivExecParam( | |||
| 113 | } | 113 | } |
| 114 | 114 | ||
| 115 | HCCL_INFO( | 115 | HCCL_INFO( |
| 116 | - "SPK, buffersIn [%p] [%p] [%p] [%p]" | 116 | + "[CollAllReduceMeshAivSmallCountExecutor][GetAivExecParam] buffersIn [%p] [%p] [%p] [%p] " |
| 117 | "buffersOut [%p] [%p] [%p] [%p]", | 117 | "buffersOut [%p] [%p] [%p] [%p]", |
| 118 | args.buffersIn[0], args.buffersIn[1], args.buffersIn[2], args.buffersIn[3], args.buffersOut[0], | 118 | args.buffersIn[0], args.buffersIn[1], args.buffersIn[2], args.buffersIn[3], args.buffersOut[0], |
| 119 | args.buffersOut[1], args.buffersOut[2], args.buffersOut[3]); | 119 | args.buffersOut[1], args.buffersOut[2], args.buffersOut[3]); |
| @@ -125,7 +125,7 @@ HcclResult CollAllReduceMeshAivSmallCountExecutor::GetAivExecParam( | |||
| 125 | args.reduceOp = param.reduceType; | 125 | args.reduceOp = param.reduceType; |
| 126 | args.devType = static_cast<u32>(topoAttr_.deviceType); | 126 | args.devType = static_cast<u32>(topoAttr_.deviceType); |
| 127 | HCCL_INFO( | 127 | HCCL_INFO( |
| 128 | - "SPK [CollAllReduceMeshAivSmallCountExecutor][GetAivExecParam], rank[%llu], rankSize[%llu], " | 128 | + "[CollAllReduceMeshAivSmallCountExecutor][GetAivExecParam], rank[%llu], rankSize[%llu], " |
| 129 | "len[%llu],datatype[%llu], op[%llu]", | 129 | "len[%llu],datatype[%llu], op[%llu]", |
| 130 | args.rank, args.rankSize, args.len, args.dataType, args.reduceOp); | 130 | args.rank, args.rankSize, args.len, args.dataType, args.reduceOp); |
| 131 | 131 | ||
| @@ -264,7 +264,7 @@ HcclResult CollAllReduceMeshOpbaseMidCountDeterministicExecutor::RunLoopInner( | |||
| 264 | const u64 curSize = execMem.count * unitSize; | 264 | const u64 curSize = execMem.count * unitSize; |
| 265 | HCCL_DEBUG( | 265 | HCCL_DEBUG( |
| 266 | "[%s]inputMem[%p][%llu], outputMem[%p][%llu], " | 266 | "[%s]inputMem[%p][%llu], outputMem[%p][%llu], " |
| 267 | - "intputPtr[%p], outputPtr[%p], curCount[%llu], curSize[%llu]", | 267 | + "inputPtr[%p], outputPtr[%p], curCount[%llu], curSize[%llu]", |
| 268 | __func__, execMem.inputMem.ptr(), execMem.inputMem.size(), execMem.outputMem.ptr(), execMem.outputMem.size(), | 268 | __func__, execMem.inputMem.ptr(), execMem.inputMem.size(), execMem.outputMem.ptr(), execMem.outputMem.size(), |
| 269 | execMem.inputPtr, execMem.outputPtr, execMem.count, curSize); | 269 | execMem.inputPtr, execMem.outputPtr, execMem.count, curSize); |
| 270 | CHK_PRT_RET((execMem.count == 0), HCCL_ERROR("[%s]In OP_BASE curCount is zero.", __func__), HCCL_E_PARA); | 270 | CHK_PRT_RET((execMem.count == 0), HCCL_ERROR("[%s]In OP_BASE curCount is zero.", __func__), HCCL_E_PARA); |
| @@ -111,7 +111,7 @@ HcclResult CollAlltoAllDirectFullmeshAIVExecutor::CalNumBlocks( | |||
| 111 | HCCL_E_PARA); | 111 | HCCL_E_PARA); |
| 112 | 112 | ||
| 113 | HCCL_INFO( | 113 | HCCL_INFO( |
| 114 | - "[CollAlltoAllDirectFullmeshAIVExecutor][%s] numBlocks is set to [%u], limit[%u], recommanded[%u]", __func__, | 114 | + "[CollAlltoAllDirectFullmeshAIVExecutor][%s] numBlocks is set to [%u], limit[%u], recommended[%u]", __func__, |
| 115 | numBlocks, numBlocks_, numBlocks); | 115 | numBlocks, numBlocks_, numBlocks); |
| 116 | return HCCL_SUCCESS; | 116 | return HCCL_SUCCESS; |
| 117 | } | 117 | } |
| @@ -125,7 +125,7 @@ HcclResult CollAlltoAllMeshAivFor91093Executor::GetAivExecParam( | |||
| 125 | args.reduceOp = param.reduceType; | 125 | args.reduceOp = param.reduceType; |
| 126 | args.devType = static_cast<u32>(topoAttr_.deviceType); | 126 | args.devType = static_cast<u32>(topoAttr_.deviceType); |
| 127 | HCCL_INFO( | 127 | HCCL_INFO( |
| 128 | - "SPK [CollAlltoAllMeshAivFor91093Executor][GetAivExecParam], rank[%llu], rankSize[%llu], " | 128 | + "[CollAlltoAllMeshAivFor91093Executor][GetAivExecParam], rank[%llu], rankSize[%llu], " |
| 129 | "len[%llu],datatype[%llu], op[%llu]", | 129 | "len[%llu],datatype[%llu], op[%llu]", |
| 130 | args.rank, args.rankSize, args.len, args.dataType, args.reduceOp); | 130 | args.rank, args.rankSize, args.len, args.dataType, args.reduceOp); |
| 131 | 131 | ||
| @@ -108,7 +108,7 @@ HcclResult CollRunAlltoAllStagedAivRdmaExecutor::CalNumBlocks( | |||
| 108 | HCCL_E_PARA); | 108 | HCCL_E_PARA); |
| 109 | 109 | ||
| 110 | HCCL_INFO( | 110 | HCCL_INFO( |
| 111 | - "[CollRunAlltoAllStagedAivRdmaExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommanded[%u]", | 111 | + "[CollRunAlltoAllStagedAivRdmaExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommended[%u]", |
| 112 | numBlocks, numBlocks_, bestNumBlocks); | 112 | numBlocks, numBlocks_, bestNumBlocks); |
| 113 | return HCCL_SUCCESS; | 113 | return HCCL_SUCCESS; |
| 114 | } | 114 | } |
| @@ -52,7 +52,7 @@ HcclResult CollRunAlltoAllDirectFullmesh::Orchestrate(OpParam& param, AlgResourc | |||
| 52 | 52 | ||
| 53 | HcclResult CollRunAlltoAllDirectFullmesh::GetAdjInfo(AlgResourceResponse& algRes, AdjInfo& adjInfo) | 53 | HcclResult CollRunAlltoAllDirectFullmesh::GetAdjInfo(AlgResourceResponse& algRes, AdjInfo& adjInfo) |
| 54 | { | 54 | { |
| 55 | - HCCL_INFO("[GetAdjInfo-nslbdp] GetAdjInfo."); | 55 | + HCCL_INFO("[CollRunAlltoAllDirectFullmesh] GetAdjInfo."); |
| 56 | algResResp_ = &algRes; | 56 | algResResp_ = &algRes; |
| 57 | SubCommInfo levelCommInfo = {}; | 57 | SubCommInfo levelCommInfo = {}; |
| 58 | AdjInfo nslbAdjInfo = {}; | 58 | AdjInfo nslbAdjInfo = {}; |
| @@ -63,7 +63,7 @@ HcclResult CollRunAlltoAllDirectFullmesh::GetAdjInfo(AlgResourceResponse& algRes | |||
| 63 | 63 | ||
| 64 | std::unique_ptr<AlgTemplateBase> levelTempAlg; | 64 | std::unique_ptr<AlgTemplateBase> levelTempAlg; |
| 65 | 65 | ||
| 66 | - HCCL_INFO("[GetAdjInfo-nslbdp] SelectTempAlg."); | 66 | + HCCL_INFO("[CollRunAlltoAllDirectFullmesh] SelectTempAlg."); |
| 67 | levelTempAlg = AlgTemplateRegistry::Instance().GetAlgTemplate( | 67 | levelTempAlg = AlgTemplateRegistry::Instance().GetAlgTemplate( |
| 68 | TemplateType::TEMPLATE_ALL_2_ALL_V_DIRECT_FULL_MESH, dispatcher_); | 68 | TemplateType::TEMPLATE_ALL_2_ALL_V_DIRECT_FULL_MESH, dispatcher_); |
| 69 | CHK_SMART_PTR_NULL(levelTempAlg); | 69 | CHK_SMART_PTR_NULL(levelTempAlg); |
| @@ -71,7 +71,7 @@ HcclResult CollRunAlltoAllDirectFullmesh::GetAdjInfo(AlgResourceResponse& algRes | |||
| 71 | CHK_RET(GetLocalSDMAGroupInfo(topoAttr_.userRank, devNumInlocalPod, rankIdxInPod)); | 71 | CHK_RET(GetLocalSDMAGroupInfo(topoAttr_.userRank, devNumInlocalPod, rankIdxInPod)); |
| 72 | 72 | ||
| 73 | if (devNumInlocalPod == INVALID_VALUE_RANKSIZE) { | 73 | if (devNumInlocalPod == INVALID_VALUE_RANKSIZE) { |
| 74 | - HCCL_INFO("[GetAdjInfo-nslbdp] devNumInlocalPod == INVALID_VALUE_RANKSIZE."); | 74 | + HCCL_INFO("[CollRunAlltoAllDirectFullmesh] devNumInlocalPod == INVALID_VALUE_RANKSIZE."); |
| 75 | return HCCL_SUCCESS; | 75 | return HCCL_SUCCESS; |
| 76 | } | 76 | } |
| 77 | 77 | ||
| @@ -79,7 +79,7 @@ HcclResult CollRunAlltoAllDirectFullmesh::GetAdjInfo(AlgResourceResponse& algRes | |||
| 79 | CHK_RET(levelTempAlg->GetNslbAdjInfo(localRank, localRankSize, levelCommInfo.links, nslbAdjInfo)); | 79 | CHK_RET(levelTempAlg->GetNslbAdjInfo(localRank, localRankSize, levelCommInfo.links, nslbAdjInfo)); |
| 80 | 80 | ||
| 81 | adjInfo.dstRankNum = nslbAdjInfo.dstRankNum; | 81 | adjInfo.dstRankNum = nslbAdjInfo.dstRankNum; |
| 82 | - HCCL_INFO("[GetAdjInfo-nslbdp] adjInfo.dstRankNum[%u].", adjInfo.dstRankNum); | 82 | + HCCL_INFO("[CollRunAlltoAllDirectFullmesh] adjInfo.dstRankNum[%u].", adjInfo.dstRankNum); |
| 83 | 83 | ||
| 84 | for (size_t i = 0; i < nslbAdjInfo.nsAdjInfo.size(); i++) { | 84 | for (size_t i = 0; i < nslbAdjInfo.nsAdjInfo.size(); i++) { |
| 85 | NslbDpAdjInfo dpAdjInfo = {}; | 85 | NslbDpAdjInfo dpAdjInfo = {}; |
| @@ -88,8 +88,8 @@ HcclResult CollRunAlltoAllDirectFullmesh::GetAdjInfo(AlgResourceResponse& algRes | |||
| 88 | dpAdjInfo.rev = 0; | 88 | dpAdjInfo.rev = 0; |
| 89 | adjInfo.nsAdjInfo.push_back(dpAdjInfo); | 89 | adjInfo.nsAdjInfo.push_back(dpAdjInfo); |
| 90 | HCCL_INFO( | 90 | HCCL_INFO( |
| 91 | - "[nslbdp]GetAdjInfo dstLocalRankId[%u], phaseId[%u].", nslbAdjInfo.nsAdjInfo[i].dstLocalRankId, | 91 | + "[CollRunAlltoAllDirectFullmesh][GetAdjInfo] dstLocalRankId[%u], phaseId[%u].", |
| 92 | - nslbAdjInfo.nsAdjInfo[i].phaseId); | 92 | + nslbAdjInfo.nsAdjInfo[i].dstLocalRankId, nslbAdjInfo.nsAdjInfo[i].phaseId); |
| 93 | } | 93 | } |
| 94 | return HCCL_SUCCESS; | 94 | return HCCL_SUCCESS; |
| 95 | } | 95 | } |
| @@ -420,7 +420,7 @@ HcclResult CollRunAlltoAllDirectFullmesh::KernelRun(const OpParam& param, ExecMe | |||
| 420 | HcclResult CollRunAlltoAllDirectFullmesh::Getlevel1CommRank(SubCommInfo& level1CommInfo) | 420 | HcclResult CollRunAlltoAllDirectFullmesh::Getlevel1CommRank(SubCommInfo& level1CommInfo) |
| 421 | { | 421 | { |
| 422 | HCCL_INFO( | 422 | HCCL_INFO( |
| 423 | - "[GetAdjInfo-nslbdp] Getlevel1CommRank userRank[%u]--userRankSize[%u].", topoAttr_.userRank, | 423 | + "[CollRunAlltoAllDirectFullmesh] Getlevel1CommRank userRank[%u]--userRankSize[%u].", topoAttr_.userRank, |
| 424 | topoAttr_.userRankSize); | 424 | topoAttr_.userRankSize); |
| 425 | level1CommInfo.localRank = topoAttr_.userRank; | 425 | level1CommInfo.localRank = topoAttr_.userRank; |
| 426 | level1CommInfo.localRankSize = topoAttr_.userRankSize; | 426 | level1CommInfo.localRankSize = topoAttr_.userRankSize; |
| @@ -431,7 +431,7 @@ HcclResult | |||
| 431 | CollRunAlltoAllDirectFullmesh::SelectTempAlg(std::unique_ptr<AlgTemplateBase>& level1TempAlg, u32 level1RankSize) | 431 | CollRunAlltoAllDirectFullmesh::SelectTempAlg(std::unique_ptr<AlgTemplateBase>& level1TempAlg, u32 level1RankSize) |
| 432 | { | 432 | { |
| 433 | (void)level1RankSize; | 433 | (void)level1RankSize; |
| 434 | - HCCL_INFO("[GetAdjInfo-nslbdp] SelectTempAlg."); | 434 | + HCCL_INFO("[CollRunAlltoAllDirectFullmesh] SelectTempAlg."); |
| 435 | level1TempAlg = AlgTemplateRegistry::Instance().GetAlgTemplate( | 435 | level1TempAlg = AlgTemplateRegistry::Instance().GetAlgTemplate( |
| 436 | TemplateType::TEMPLATE_ALL_2_ALL_V_DIRECT_FULL_MESH, dispatcher_); | 436 | TemplateType::TEMPLATE_ALL_2_ALL_V_DIRECT_FULL_MESH, dispatcher_); |
| 437 | CHK_SMART_PTR_NULL(level1TempAlg); | 437 | CHK_SMART_PTR_NULL(level1TempAlg); |
| @@ -441,7 +441,7 @@ CollRunAlltoAllDirectFullmesh::SelectTempAlg(std::unique_ptr<AlgTemplateBase>& l | |||
| 441 | 441 | ||
| 442 | HcclResult CollRunAlltoAllDirectFullmesh::GetDevNumInlocalPod(u32& devNumInlocalPod) | 442 | HcclResult CollRunAlltoAllDirectFullmesh::GetDevNumInlocalPod(u32& devNumInlocalPod) |
| 443 | { | 443 | { |
| 444 | - HCCL_INFO("[GetAdjInfo-nslbdp] GetDevNumInlocalPod."); | 444 | + HCCL_INFO("[CollRunAlltoAllDirectFullmesh] GetDevNumInlocalPod."); |
| 445 | // 获取当前超节点内总卡数 | 445 | // 获取当前超节点内总卡数 |
| 446 | u32 rankIdxInPod = INVALID_VALUE_RANKID; | 446 | u32 rankIdxInPod = INVALID_VALUE_RANKID; |
| 447 | CHK_RET(GetLocalSDMAGroupInfo(topoAttr_.userRank, devNumInlocalPod, rankIdxInPod)); | 447 | CHK_RET(GetLocalSDMAGroupInfo(topoAttr_.userRank, devNumInlocalPod, rankIdxInPod)); |
| @@ -168,7 +168,7 @@ HcclResult CollBroadcastMidCountFor91093Executor::RunLoopInner(OpParam& param, E | |||
| 168 | 168 | ||
| 169 | HCCL_DEBUG( | 169 | HCCL_DEBUG( |
| 170 | "[CollBroadcastMidCountFor91093Executor][RunLoopInner]inputMem[%p], outputMem[%p]" | 170 | "[CollBroadcastMidCountFor91093Executor][RunLoopInner]inputMem[%p], outputMem[%p]" |
| 171 | - "intputPtr[%p], curCount[%llu], curSize[%llu]", | 171 | + "inputPtr[%p], curCount[%llu], curSize[%llu]", |
| 172 | execMem.inputMem.ptr(), execMem.outputMem.ptr(), execMem.inputPtr, execMem.count, curSize); | 172 | execMem.inputMem.ptr(), execMem.outputMem.ptr(), execMem.inputPtr, execMem.count, curSize); |
| 173 | CHK_PRT_RET( | 173 | CHK_PRT_RET( |
| 174 | (execMem.count == 0), | 174 | (execMem.count == 0), |
| @@ -80,7 +80,7 @@ HcclResult CollBroadCastMix::KernelRun(const OpParam& param, ExecMem& execMem) | |||
| 80 | CHK_PRT_RET( | 80 | CHK_PRT_RET( |
| 81 | perDataSize == 0, | 81 | perDataSize == 0, |
| 82 | HCCL_ERROR( | 82 | HCCL_ERROR( |
| 83 | - "[CollBroadCastMix][KernelRun]errNo[0x%01611x] datatype[%d] is invalid", HCCL_ERROR_CODE(HCCL_E_PARA), | 83 | + "[CollBroadCastMix][KernelRun]errNo[0x%016llx] datatype[%d] is invalid", HCCL_ERROR_CODE(HCCL_E_PARA), |
| 84 | param.DataDes.dataType), | 84 | param.DataDes.dataType), |
| 85 | HCCL_E_PARA); | 85 | HCCL_E_PARA); |
| 86 | 86 | ||
| @@ -110,7 +110,8 @@ HcclResult CollBroadCastMix::KernelRun(const OpParam& param, ExecMem& execMem) | |||
| 110 | CHK_PRT_RET( | 110 | CHK_PRT_RET( |
| 111 | mulRingSlice.size() != ringNum, | 111 | mulRingSlice.size() != ringNum, |
| 112 | HCCL_ERROR( | 112 | HCCL_ERROR( |
| 113 | - "[CollBroadCastMix][KernelRun] ringNum[%u] !=mulRingSlice size[%zu]", ringNum, mulRingSlice.size()), | 113 | + "[CollBroadCastMix][KernelRun] ringNum[%u] != mulRingSlice size[%zu]", ringNum, |
| 114 | + mulRingSlice.size()), | ||
| 114 | HCCL_E_INTERNAL); | 115 | HCCL_E_INTERNAL); |
| 115 | } else { | 116 | } else { |
| 116 | mulRingSlice.push_back(dataSegsSlice); // 应该offset全为0,而大小和dataSegsSlice中一样,里面的offset不使用 | 117 | mulRingSlice.push_back(dataSegsSlice); // 应该offset全为0,而大小和dataSegsSlice中一样,里面的offset不使用 |
| @@ -94,7 +94,7 @@ HcclResult CollReduceScatterAivRdmaExecutor::CalNumBlocks( | |||
| 94 | HCCL_E_PARA); | 94 | HCCL_E_PARA); |
| 95 | 95 | ||
| 96 | HCCL_INFO( | 96 | HCCL_INFO( |
| 97 | - "[CollReduceScatterAivRdmaExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommanded[%u]", | 97 | + "[CollReduceScatterAivRdmaExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommended[%u]", |
| 98 | numBlocks, numBlocks_, bestNumBlocks); | 98 | numBlocks, numBlocks_, bestNumBlocks); |
| 99 | return HCCL_SUCCESS; | 99 | return HCCL_SUCCESS; |
| 100 | } | 100 | } |
| @@ -95,7 +95,7 @@ HcclResult CollReduceScatterMeshAivExecutor::CalNumBlocks( | |||
| 95 | } | 95 | } |
| 96 | 96 | ||
| 97 | HCCL_INFO( | 97 | HCCL_INFO( |
| 98 | - "[CollReduceScatterMeshAivExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommanded[%u]", | 98 | + "[CollReduceScatterMeshAivExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommended[%u]", |
| 99 | numBlocks, numBlocks_, bestNumBlocks); | 99 | numBlocks, numBlocks_, bestNumBlocks); |
| 100 | return HCCL_SUCCESS; | 100 | return HCCL_SUCCESS; |
| 101 | } | 101 | } |
| @@ -135,7 +135,7 @@ HcclResult CollReduceScatterMeshAivExecutor::GetAivExecParam( | |||
| 135 | } | 135 | } |
| 136 | } | 136 | } |
| 137 | HCCL_INFO( | 137 | HCCL_INFO( |
| 138 | - "SPK, buffersIn [%p] [%p] [%p] [%p]" | 138 | + "[CollReduceScatterMeshAivExecutor][GetAivExecParam], buffersIn [%p] [%p] [%p] [%p] " |
| 139 | "buffersOut [%p] [%p] [%p] [%p]", | 139 | "buffersOut [%p] [%p] [%p] [%p]", |
| 140 | args.buffersIn[0], args.buffersIn[1], args.buffersIn[2], args.buffersIn[3], args.buffersOut[0], | 140 | args.buffersIn[0], args.buffersIn[1], args.buffersIn[2], args.buffersIn[3], args.buffersOut[0], |
| 141 | args.buffersOut[1], args.buffersOut[2], args.buffersOut[3]); | 141 | args.buffersOut[1], args.buffersOut[2], args.buffersOut[3]); |
| @@ -154,7 +154,7 @@ HcclResult CollReduceScatterMeshAivExecutor::GetAivExecParam( | |||
| 154 | HCCL_ERROR_CODE(ret), param.tag.c_str()), | 154 | HCCL_ERROR_CODE(ret), param.tag.c_str()), |
| 155 | ret); | 155 | ret); |
| 156 | HCCL_INFO( | 156 | HCCL_INFO( |
| 157 | - "SPK [CollReduceScatterMeshAivExecutor][GetAivExecParam], rank[%llu], rankSize[%llu], " | 157 | + "[CollReduceScatterMeshAivExecutor][GetAivExecParam], rank[%llu], rankSize[%llu], " |
| 158 | "len[%llu],datatype[%llu], op[%llu], devType[%u]", | 158 | "len[%llu],datatype[%llu], op[%llu], devType[%u]", |
| 159 | args.rank, args.rankSize, args.len, args.dataType, args.reduceOp, args.devType); | 159 | args.rank, args.rankSize, args.len, args.dataType, args.reduceOp, args.devType); |
| 160 | 160 | ||
| @@ -64,7 +64,7 @@ HcclResult CollReduceScatterVMeshAivSmallCountExecutor::CalNumBlocks( | |||
| 64 | 64 | ||
| 65 | HCCL_INFO( | 65 | HCCL_INFO( |
| 66 | "[CollReduceScatterVMeshAivSmallCountExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], " | 66 | "[CollReduceScatterVMeshAivSmallCountExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], " |
| 67 | - "recommanded[%u]", | 67 | + "recommended[%u]", |
| 68 | numBlocks, numBlocks_, bestNumBlocks); | 68 | numBlocks, numBlocks_, bestNumBlocks); |
| 69 | return HCCL_SUCCESS; | 69 | return HCCL_SUCCESS; |
| 70 | } | 70 | } |
| @@ -306,7 +306,7 @@ HcclResult CollBatchSendRecvExecutor::GetAdjInfo(AlgResourceResponse& algRes, Ad | |||
| 306 | CHK_RET(level1TempAlg->GetNslbAdjInfo(localRank, localRankSize, level1CommInfo.links, nslbAdjInfo)); | 306 | CHK_RET(level1TempAlg->GetNslbAdjInfo(localRank, localRankSize, level1CommInfo.links, nslbAdjInfo)); |
| 307 | 307 | ||
| 308 | adjInfo.dstRankNum = nslbAdjInfo.dstRankNum; | 308 | adjInfo.dstRankNum = nslbAdjInfo.dstRankNum; |
| 309 | - HCCL_INFO("[nslbdp] adjInfo.dstRankNum[%u].", adjInfo.dstRankNum); | 309 | + HCCL_INFO("[CollBatchSendRecvExecutor][GetAdjInfo] adjInfo.dstRankNum[%u].", adjInfo.dstRankNum); |
| 310 | 310 | ||
| 311 | for (size_t i = 0; i < nslbAdjInfo.nsAdjInfo.size(); i++) { | 311 | for (size_t i = 0; i < nslbAdjInfo.nsAdjInfo.size(); i++) { |
| 312 | NslbDpAdjInfo dpAdjInfo = {}; | 312 | NslbDpAdjInfo dpAdjInfo = {}; |
| @@ -315,8 +315,8 @@ HcclResult CollBatchSendRecvExecutor::GetAdjInfo(AlgResourceResponse& algRes, Ad | |||
| 315 | dpAdjInfo.rev = 0; | 315 | dpAdjInfo.rev = 0; |
| 316 | adjInfo.nsAdjInfo.push_back(dpAdjInfo); | 316 | adjInfo.nsAdjInfo.push_back(dpAdjInfo); |
| 317 | HCCL_INFO( | 317 | HCCL_INFO( |
| 318 | - "[nslbdp]GetAdjInfo dstLocalRankId[%u], phaseId[%u].", nslbAdjInfo.nsAdjInfo[i].dstLocalRankId, | 318 | + "[CollBatchSendRecvExecutor][GetAdjInfo] dstLocalRankId[%u], phaseId[%u].", |
| 319 | - nslbAdjInfo.nsAdjInfo[i].phaseId); | 319 | + nslbAdjInfo.nsAdjInfo[i].dstLocalRankId, nslbAdjInfo.nsAdjInfo[i].phaseId); |
| 320 | } | 320 | } |
| 321 | return HCCL_SUCCESS; | 321 | return HCCL_SUCCESS; |
| 322 | } | 322 | } |
| @@ -764,7 +764,7 @@ HcclResult hcclImpl::CreateMutiStreamRes( | |||
| 764 | std::unique_lock<std::mutex> mutiStreamLock(tagStreamInfoLock_); | 764 | std::unique_lock<std::mutex> mutiStreamLock(tagStreamInfoLock_); |
| 765 | CHK_PRT_RET( | 765 | CHK_PRT_RET( |
| 766 | tagStreamInfo_.find(tag) != tagStreamInfo_.end(), | 766 | tagStreamInfo_.find(tag) != tagStreamInfo_.end(), |
| 767 | - HCCL_DEBUG("[Create][MutiStreamRes]tag[%s] is already exit, do nothing", tag.c_str()), HCCL_SUCCESS); | 767 | + HCCL_DEBUG("[Create][MutiStreamRes]tag[%s] already exists, do nothing", tag.c_str()), HCCL_SUCCESS); |
| 768 | 768 | ||
| 769 | level1StreamInfo_t streamInfo; | 769 | level1StreamInfo_t streamInfo; |
| 770 | CHK_RET(CreateMutiStreamRes(tag, stream, streamInfo, algType, false, isBatchSendRecv, ringNum)); | 770 | CHK_RET(CreateMutiStreamRes(tag, stream, streamInfo, algType, false, isBatchSendRecv, ringNum)); |
| @@ -147,7 +147,7 @@ HcclResult HcclSocketManager::ConstructWhiteList( | |||
| 147 | wlistInfo.connLimit = GetConnLimit(socketType); | 147 | wlistInfo.connLimit = GetConnLimit(socketType); |
| 148 | s32 sRet = memcpy_s(&wlistInfo.tag[0], sizeof(wlistInfo.tag), tag.c_str(), tag.size() + 1); | 148 | s32 sRet = memcpy_s(&wlistInfo.tag[0], sizeof(wlistInfo.tag), tag.c_str(), tag.size() + 1); |
| 149 | if (sRet != EOK) { | 149 | if (sRet != EOK) { |
| 150 | - HCCL_ERROR("[Construct][WhiteList]memory copy failed. errorno[%d]", sRet); | 150 | + HCCL_ERROR("[Construct][WhiteList]memory copy failed. errno[%d]", sRet); |
| 151 | return HCCL_E_MEMORY; | 151 | return HCCL_E_MEMORY; |
| 152 | } | 152 | } |
| 153 | 153 | ||
| @@ -402,7 +402,7 @@ HcclResult HcclSocketManager::CreateSockets( | |||
| 402 | } | 402 | } |
| 403 | 403 | ||
| 404 | HCCL_INFO( | 404 | HCCL_INFO( |
| 405 | - "[Create][Sockets]client socket map size %u, server socket map size %u", serverSocketsMap.size(), | 405 | + "[Create][Sockets]client socket map size %u, server socket map size %u", clientSocketsMap.size(), |
| 406 | serverSocketsMap.size()); | 406 | serverSocketsMap.size()); |
| 407 | 407 | ||
| 408 | // 需要等待连接建立成功时,则会阻塞 | 408 | // 需要等待连接建立成功时,则会阻塞 |
| @@ -539,7 +539,7 @@ HcclResult TopoMatcher::GetLocalServerRankSize(const u32 userRank, u32& devNumIn | |||
| 539 | HcclResult TopoMatcher::SetDeterministicConfig(const u8 deterministic) | 539 | HcclResult TopoMatcher::SetDeterministicConfig(const u8 deterministic) |
| 540 | { | 540 | { |
| 541 | if (deterministic > DETERMINISTIC_STRICT) { | 541 | if (deterministic > DETERMINISTIC_STRICT) { |
| 542 | - HCCL_ERROR("[SetDeterministicConfig] deterministic should be 0, 1 or 2."); | 542 | + HCCL_ERROR("[SetDeterministicConfig] deterministic[%u] is invalid, should be 0, 1 or 2.", deterministic); |
| 543 | return HCCL_E_PARA; | 543 | return HCCL_E_PARA; |
| 544 | } | 544 | } |
| 545 | HCCL_INFO("[SetDeterministicConfig]deterministic is set to [%d]", deterministic); | 545 | HCCL_INFO("[SetDeterministicConfig]deterministic is set to [%d]", deterministic); |
| @@ -31,7 +31,8 @@ HcclResult CalcCrc::HcclCalcCrc(const char* data, u64 length, u32& crcValue) | |||
| 31 | 31 | ||
| 32 | CHK_PRT_RET( | 32 | CHK_PRT_RET( |
| 33 | length <= 0 || length > STRING_MAX_LENGTH, | 33 | length <= 0 || length > STRING_MAX_LENGTH, |
| 34 | - HCCL_ERROR("[Calc][StringCrc]String length[%llu] is empty or over than %d bytes.", length, STRING_MAX_LENGTH), | 34 | + HCCL_ERROR( |
| 35 | + "[Calc][StringCrc]String length[%llu] is empty or greater than %d bytes.", length, STRING_MAX_LENGTH), | ||
| 35 | HCCL_E_PARA); | 36 | HCCL_E_PARA); |
| 36 | HCCL_DEBUG("data[%s], length[%llu]", data, length); | 37 | HCCL_DEBUG("data[%s], length[%llu]", data, length); |
| 37 | 38 | ||
| @@ -34,7 +34,7 @@ public: | |||
| 34 | } | 34 | } |
| 35 | capacity_ = capacity; | 35 | capacity_ = capacity; |
| 36 | if (capacity_ <= 0) { | 36 | if (capacity_ <= 0) { |
| 37 | - HCCL_ERROR("[RingBuffer] capacity[%u] must greater than 0", capacity_); | 37 | + HCCL_ERROR("[RingBuffer] capacity[%u] must be greater than 0", capacity_); |
[MEDIUM] errorno→errno 修正只改了 3 处,全仓 src/ 仍残留 125 处/29 个文件 本 PR 把 建议修复:以本文件为例:
建议二选一:要么本 PR 把 ![]() ![]() | |||
| 38 | return HCCL_E_PARA; | 38 | return HCCL_E_PARA; |
| 39 | } | 39 | } |
| 40 | 40 | ||
| @@ -628,7 +628,7 @@ HcclResult OpRetryBase::InitChangeLinkInfo(RetryContext* retryCtx, bool incre, b | |||
| 628 | 628 | ||
| 629 | retryCtx->lastLinkPortStatus_.insert({remoteRank, true}); | 629 | retryCtx->lastLinkPortStatus_.insert({remoteRank, true}); |
| 630 | 630 | ||
| 631 | - HCCL_RUN_INFO("[OpRetry][Agnet]init changeLinkInfoStr add remoteRank[%u]", remoteRank); | 631 | + HCCL_RUN_INFO("[OpRetry][Agent]init changeLinkInfoStr add remoteRank[%u]", remoteRank); |
| 632 | } | 632 | } |
| 633 | } | 633 | } |
| 634 | } else { | 634 | } else { |
| @@ -646,7 +646,7 @@ HcclResult OpRetryBase::InitChangeLinkInfo(RetryContext* retryCtx, bool incre, b | |||
| 646 | + std::to_string(retryCtx->localChangeLinkInfo_.isUseDefaultPort[i]) + "; "); | 646 | + std::to_string(retryCtx->localChangeLinkInfo_.isUseDefaultPort[i]) + "; "); |
| 647 | retryCtx->lastLinkPortStatus_.insert({retryCtx->localChangeLinkInfo_.remoteRankList[i], true}); | 647 | retryCtx->lastLinkPortStatus_.insert({retryCtx->localChangeLinkInfo_.remoteRankList[i], true}); |
| 648 | } | 648 | } |
| 649 | - HCCL_RUN_INFO("[OpRetry][Agnet]init changeLinkInfoStr:%s", changeLinkInfoStr.c_str()); | 649 | + HCCL_RUN_INFO("[OpRetry][Agent]init changeLinkInfoStr:%s", changeLinkInfoStr.c_str()); |
| 650 | } | 650 | } |
| 651 | 651 | ||
| 652 | return HCCL_SUCCESS; | 652 | return HCCL_SUCCESS; |
| @@ -790,11 +790,11 @@ OpRetryBase::RecvActiveSwitchInfo(std::shared_ptr<HcclSocket> socket, const u32 | |||
| 790 | CHK_PRT_RET( | 790 | CHK_PRT_RET( |
| 791 | elapsed > timeout, | 791 | elapsed > timeout, |
| 792 | HCCL_ERROR( | 792 | HCCL_ERROR( |
| 793 | - "[SwitchNic][Server] timeout in recv agent ActiveSwitchInfo, waitime[%u s>%u s]", elapsed, timeout), | 793 | + "[SwitchNic][Server] timeout in recv agent ActiveSwitchInfo, waittime[%u s>%u s]", elapsed, timeout), |
| 794 | HCCL_E_TIMEOUT); | 794 | HCCL_E_TIMEOUT); |
| 795 | ret = WaitActiveSwitchInfo(socket, switchInfo); | 795 | ret = WaitActiveSwitchInfo(socket, switchInfo); |
| 796 | if (ret == HCCL_SUCCESS) { | 796 | if (ret == HCCL_SUCCESS) { |
| 797 | - HCCL_INFO("[SwitchNic][Server] recv ActiveSwitchInfo form rank[%u]", rankId); | 797 | + HCCL_INFO("[SwitchNic][Server] recv ActiveSwitchInfo from rank[%u]", rankId); |
| 798 | break; | 798 | break; |
| 799 | } else if (ret == HCCL_E_AGAIN) { | 799 | } else if (ret == HCCL_E_AGAIN) { |
| 800 | // 未收到数据,发送一个保活数据给agent | 800 | // 未收到数据,发送一个保活数据给agent |
| @@ -185,7 +185,7 @@ HcclResult OpRetryServerHandleError::ProcessEvent(RetryContext* retryCtx) | |||
| 185 | while (true) { | 185 | while (true) { |
| 186 | CHK_PRT_RET( | 186 | CHK_PRT_RET( |
| 187 | retryCtx->isServerStateWaitResume_, | 187 | retryCtx->isServerStateWaitResume_, |
| 188 | - HCCL_RUN_INFO("[OpRetry][Server]switched state form wait handle error to wait resume"), HCCL_SUCCESS); | 188 | + HCCL_RUN_INFO("[OpRetry][Server]switched state from wait handle error to wait resume"), HCCL_SUCCESS); |
| 189 | // 判断是否超时 | 189 | // 判断是否超时 |
| 190 | std::chrono::steady_clock::time_point curTime = std::chrono::steady_clock::now(); | 190 | std::chrono::steady_clock::time_point curTime = std::chrono::steady_clock::now(); |
| 191 | const auto elapsed = std::chrono::duration_cast<std::chrono::seconds>(curTime - startTime); | 191 | const auto elapsed = std::chrono::duration_cast<std::chrono::seconds>(curTime - startTime); |
| @@ -446,7 +446,7 @@ HcclResult OpRetryServerWaitResp::ProcessEvent(RetryContext* retryCtx) | |||
| 446 | while (recvVaild.size() < retryCtx->needRetryServerRanks_.size()) { | 446 | while (recvVaild.size() < retryCtx->needRetryServerRanks_.size()) { |
| 447 | CHK_PRT_RET( | 447 | CHK_PRT_RET( |
| 448 | retryCtx->isServerStateWaitResume_, | 448 | retryCtx->isServerStateWaitResume_, |
| 449 | - HCCL_RUN_INFO("[OpRetry][Server]switched state form wait resp to wait resume"), HCCL_SUCCESS); | 449 | + HCCL_RUN_INFO("[OpRetry][Server]switched state from wait resp to wait resume"), HCCL_SUCCESS); |
| 450 | std::chrono::steady_clock::time_point curTime = std::chrono::steady_clock::now(); | 450 | std::chrono::steady_clock::time_point curTime = std::chrono::steady_clock::now(); |
| 451 | const auto elapsed = std::chrono::duration_cast<std::chrono::seconds>(curTime - startTime); | 451 | const auto elapsed = std::chrono::duration_cast<std::chrono::seconds>(curTime - startTime); |
| 452 | CHK_PRT_RET(elapsed > timeout, HCCL_ERROR("[OpRetry][Server]OpRetryServerWaitResp timeout"), HCCL_E_TIMEOUT); | 452 | CHK_PRT_RET(elapsed > timeout, HCCL_ERROR("[OpRetry][Server]OpRetryServerWaitResp timeout"), HCCL_E_TIMEOUT); |
| @@ -541,7 +541,7 @@ HcclResult OpRetryServerWaitLinkInfo::ProcessEvent(RetryContext* retryCtx) | |||
| 541 | while (recvVaild.size() < retryCtx->needRetryServerRanks_.size()) { | 541 | while (recvVaild.size() < retryCtx->needRetryServerRanks_.size()) { |
| 542 | CHK_PRT_RET( | 542 | CHK_PRT_RET( |
| 543 | retryCtx->isServerStateWaitResume_, | 543 | retryCtx->isServerStateWaitResume_, |
| 544 | - HCCL_RUN_INFO("[OpRetry][Server]switched state form wait link to wait resume"), HCCL_SUCCESS); | 544 | + HCCL_RUN_INFO("[OpRetry][Server]switched state from wait link to wait resume"), HCCL_SUCCESS); |
| 545 | std::chrono::steady_clock::time_point curTime = std::chrono::steady_clock::now(); | 545 | std::chrono::steady_clock::time_point curTime = std::chrono::steady_clock::now(); |
| 546 | const auto elapsed = std::chrono::duration_cast<std::chrono::seconds>(curTime - startTime); | 546 | const auto elapsed = std::chrono::duration_cast<std::chrono::seconds>(curTime - startTime); |
| 547 | CHK_PRT_RET( | 547 | CHK_PRT_RET( |
| @@ -747,12 +747,12 @@ HcclResult SwitchNicServerCheckAllSwitchRanks::CollectSingleAgentActiveSwitchInf | |||
| 747 | while (true) { | 747 | while (true) { |
| 748 | CHK_PRT_RET( | 748 | CHK_PRT_RET( |
| 749 | retryCtx->isServerStateWaitResume_, | 749 | retryCtx->isServerStateWaitResume_, |
| 750 | - HCCL_RUN_INFO("[OpRetry][Server]switched state form check switch Nic to wait resume"), HCCL_SUCCESS); | 750 | + HCCL_RUN_INFO("[OpRetry][Server]switched state from check switch Nic to wait resume"), HCCL_SUCCESS); |
| 751 | std::chrono::steady_clock::time_point curTime = std::chrono::steady_clock::now(); | 751 | std::chrono::steady_clock::time_point curTime = std::chrono::steady_clock::now(); |
| 752 | const auto elapsed = std::chrono::duration_cast<std::chrono::seconds>(curTime - startTime); | 752 | const auto elapsed = std::chrono::duration_cast<std::chrono::seconds>(curTime - startTime); |
| 753 | CHK_PRT_RET( | 753 | CHK_PRT_RET( |
| 754 | elapsed > timeout, | 754 | elapsed > timeout, |
| 755 | - HCCL_ERROR("[SwitchNic][Server] timeout in recv agent RetryInfo, waitime[%u s>%u s]", elapsed, timeout), | 755 | + HCCL_ERROR("[SwitchNic][Server] timeout in recv agent RetryInfo, waittime[%u s>%u s]", elapsed, timeout), |
| 756 | HCCL_E_TIMEOUT); | 756 | HCCL_E_TIMEOUT); |
| 757 | ret = WaitResponse(agentInfo.socket, agentInfo.retryInfo); | 757 | ret = WaitResponse(agentInfo.socket, agentInfo.retryInfo); |
| 758 | if (ret == HCCL_SUCCESS) { // 成功接收到数据 | 758 | if (ret == HCCL_SUCCESS) { // 成功接收到数据 |
| @@ -762,7 +762,7 @@ HcclResult SwitchNicServerCheckAllSwitchRanks::CollectSingleAgentActiveSwitchInf | |||
| 762 | if (ret != HCCL_SUCCESS) { | 762 | if (ret != HCCL_SUCCESS) { |
| 763 | return ret; | 763 | return ret; |
| 764 | } | 764 | } |
| 765 | - HCCL_INFO("[SwitchNic][server] recv ActiveSwitchInfo form rank[%u] while collecting", rankId); | 765 | + HCCL_INFO("[SwitchNic][server] recv ActiveSwitchInfo from rank[%u] while collecting", rankId); |
| 766 | retryCtx->switchInfoMap_[rankId] = agentInfo.switchInfo; | 766 | retryCtx->switchInfoMap_[rankId] = agentInfo.switchInfo; |
| 767 | return HCCL_SUCCESS; | 767 | return HCCL_SUCCESS; |
| 768 | } | 768 | } |
| @@ -463,8 +463,8 @@ HcclResult TopoInfoExchangeServer::DisplayConnectingStatus( | |||
| 463 | } | 463 | } |
| 464 | } | 464 | } |
| 465 | 465 | ||
| 466 | - std::string infoStr = "succ sockets is [" + std::to_string((totalSockets - waitSockets)) | 466 | + std::string infoStr = "succ sockets are [" + std::to_string((totalSockets - waitSockets)) |
[MEDIUM] succ sockets is→are 单复数修正只改了 legacy 姊妹实现,本文件同构文案未改 本 PR 把 第 426 行: 建议修复:与 topoinfo_exchange_server.cc 的改法对齐:
第 466 行:succ sockets are ![]() ![]() | |||
| 467 | - + "], waiting sockets is [" + std::to_string(waitSockets) + "], wait sockets rankid: "; | 467 | + + "], waiting sockets are [" + std::to_string(waitSockets) + "], wait sockets rankid: "; |
| 468 | for (u32 index = 0; index < unsocketinfos.size(); index++) { | 468 | for (u32 index = 0; index < unsocketinfos.size(); index++) { |
| 469 | if (index == (unsocketinfos.size() - 1)) { | 469 | if (index == (unsocketinfos.size() - 1)) { |
| 470 | infoStr += "[" + unsocketinfos[index] + "]"; | 470 | infoStr += "[" + unsocketinfos[index] + "]"; |
| @@ -475,8 +475,8 @@ HcclResult TopoInfoExchangeServer::DisplayConnectingStatus( | |||
| 475 | 475 | ||
| 476 | HCCL_RUN_INFO("[HCCL_TRACE] %s", infoStr.c_str()); | 476 | HCCL_RUN_INFO("[HCCL_TRACE] %s", infoStr.c_str()); |
| 477 | } else { | 477 | } else { |
| 478 | - std::string infoStr = "succ sockets is [" + std::to_string(totalSockets - waitSockets) | 478 | + std::string infoStr = "succ sockets are [" + std::to_string(totalSockets - waitSockets) |
| 479 | - + "], waiting sockets is [" + std::to_string(waitSockets) + "]"; | 479 | + + "], waiting sockets are [" + std::to_string(waitSockets) + "]"; |
| 480 | HCCL_RUN_INFO("[HCCL_TRACE] %s , isByMasterInfo[%d]", infoStr.c_str(), isByMasterInfo_); | 480 | HCCL_RUN_INFO("[HCCL_TRACE] %s , isByMasterInfo[%d]", infoStr.c_str(), isByMasterInfo_); |
| 481 | } | 481 | } |
| 482 | 482 | ||
| @@ -492,7 +492,7 @@ HcclResult TopoInfoExchangeServer::GetRemoteFdAndRankSize( | |||
| 492 | auto iter = connectSockets.find(agentID); | 492 | auto iter = connectSockets.find(agentID); |
| 493 | CHK_PRT_RET( | 493 | CHK_PRT_RET( |
| 494 | iter != connectSockets.end(), | 494 | iter != connectSockets.end(), |
| 495 | - HCCL_ERROR("[Get][Connection]GetConnection failed. agnet[%s] has been connected.", agentID.c_str()), | 495 | + HCCL_ERROR("[Get][Connection]GetConnection failed. agent[%s] has been connected.", agentID.c_str()), |
| 496 | HCCL_E_INTERNAL); | 496 | HCCL_E_INTERNAL); |
| 497 | connectSockets.insert({agentID, socket}); | 497 | connectSockets.insert({agentID, socket}); |
| 498 | 498 | ||
| @@ -645,8 +645,8 @@ HcclResult TopoInfoExchangeServer::VerifyRemoteRankNum(u32& previousRankNum, u32 | |||
| 645 | CHK_PRT_RET( | 645 | CHK_PRT_RET( |
| 646 | (remoteRankNum != previousRankNum), | 646 | (remoteRankNum != previousRankNum), |
| 647 | HCCL_ERROR( | 647 | HCCL_ERROR( |
| 648 | - "[Verify][RemoteRankNum]VerifyRemoteRankNum failed. remoteRankNum[%u] is difference " | 648 | + "[Verify][RemoteRankNum]VerifyRemoteRankNum failed. remoteRankNum[%u] is different " |
| 649 | - "with others[%u].", | 649 | + "from others[%u].", |
| 650 | remoteRankNum, previousRankNum), | 650 | remoteRankNum, previousRankNum), |
| 651 | HCCL_E_INTERNAL); | 651 | HCCL_E_INTERNAL); |
| 652 | } | 652 | } |
| @@ -192,7 +192,7 @@ HcclResult TopoinfoRanktableHeterog::GetRanktableInfo(RankTable_t& clusterInfo) | |||
| 192 | CHK_RET(GetJsonProperty(fileContent_, "collective_id", collective_id, false)); | 192 | CHK_RET(GetJsonProperty(fileContent_, "collective_id", collective_id, false)); |
| 193 | if (collective_id.length() > COLLECTIVEID_MAX_LEN) { | 193 | if (collective_id.length() > COLLECTIVEID_MAX_LEN) { |
| 194 | HCCL_ERROR( | 194 | HCCL_ERROR( |
| 195 | - "[Get][RanktableInfo]errNo[0x%016llx] collectiveId length is over than %d bytes.", | 195 | + "[Get][RanktableInfo]errNo[0x%016llx] collectiveId length is greater than %d bytes.", |
| 196 | HCOM_ERROR_CODE(HCCL_E_PARA), collective_id.length()); | 196 | HCOM_ERROR_CODE(HCCL_E_PARA), collective_id.length()); |
| 197 | return HCCL_E_PARA; | 197 | return HCCL_E_PARA; |
| 198 | } | 198 | } |
| @@ -240,7 +240,7 @@ HcclResult TopoinfoRanktableHeterog::CheckNicDeployConsistence(RankTable_t& clus | |||
| 240 | || (tmpNicDeply == NICDeployment::NIC_DEPLOYMENT_HOST && it.hostIp.IsInvalid()), | 240 | || (tmpNicDeply == NICDeployment::NIC_DEPLOYMENT_HOST && it.hostIp.IsInvalid()), |
| 241 | HCCL_ERROR( | 241 | HCCL_ERROR( |
| 242 | "[Get][RanktableInfo] errNo[0x%016llx] " | 242 | "[Get][RanktableInfo] errNo[0x%016llx] " |
| 243 | - "hostIp config bettewn ranks is different.", | 243 | + "hostIp config between ranks is different.", |
| 244 | HCOM_ERROR_CODE(HCCL_E_PARA)), | 244 | HCOM_ERROR_CODE(HCCL_E_PARA)), |
| 245 | HCCL_E_PARA); | 245 | HCCL_E_PARA); |
| 246 | } | 246 | } |
| @@ -529,7 +529,7 @@ HcclResult TopoInfoRanktableParser::CheckUniqueAndInsertPool( | |||
| 529 | if (srvIt != uniqueInfoCheckPool_[static_cast<u32>(type)].end()) { | 529 | if (srvIt != uniqueInfoCheckPool_[static_cast<u32>(type)].end()) { |
| 530 | RPT_INPUT_ERR( | 530 | RPT_INPUT_ERR( |
| 531 | true, "EI0014", std::vector<std::string>({"value", "variable", "expect"}), | 531 | true, "EI0014", std::vector<std::string>({"value", "variable", "expect"}), |
| 532 | - std::vector<std::string>({value, " \"IP\" is repeate", "please check the ranktable"})); | 532 | + std::vector<std::string>({value, " \"IP\" is repeated", "please check the ranktable"})); |
| 533 | HCCL_ERROR( | 533 | HCCL_ERROR( |
| 534 | "[%s][%s]errNo[0x%016llx] [%s]:[%s] is already exist", LOG_KEYWORDS_INIT_GROUP.c_str(), | 534 | "[%s][%s]errNo[0x%016llx] [%s]:[%s] is already exist", LOG_KEYWORDS_INIT_GROUP.c_str(), |
| 535 | LOG_KEYWORDS_RANKTABLE_CHECK.c_str(), HCOM_ERROR_CODE(HCCL_E_PARA), strUniqueInfoType.c_str(), | 535 | LOG_KEYWORDS_RANKTABLE_CHECK.c_str(), HCOM_ERROR_CODE(HCCL_E_PARA), strUniqueInfoType.c_str(), |
| @@ -51,7 +51,7 @@ HcclResult PreemptPortManager::ListenPreempt( | |||
| 51 | const std::shared_ptr<HcclSocket>& listenSocket, const std::vector<HcclSocketPortRange>& portRange, u32& usePort) | 51 | const std::shared_ptr<HcclSocket>& listenSocket, const std::vector<HcclSocketPortRange>& portRange, u32& usePort) |
| 52 | { | 52 | { |
| 53 | CHK_PRT_RET( | 53 | CHK_PRT_RET( |
| 54 | - !initialized, HCCL_ERROR("[PreemptPortManager][ListenPreempt] preempt port manager has already been release."), | 54 | + !initialized, HCCL_ERROR("[PreemptPortManager][ListenPreempt] preempt port manager has already been released."), |
| 55 | HCCL_E_INTERNAL); | 55 | HCCL_E_INTERNAL); |
| 56 | 56 | ||
| 57 | CHK_SMART_PTR_NULL(listenSocket); | 57 | CHK_SMART_PTR_NULL(listenSocket); |
| @@ -71,7 +71,7 @@ HcclResult PreemptPortManager::ListenPreempt( | |||
| 71 | HcclResult PreemptPortManager::Release(const std::shared_ptr<HcclSocket>& listenSocket) | 71 | HcclResult PreemptPortManager::Release(const std::shared_ptr<HcclSocket>& listenSocket) |
| 72 | { | 72 | { |
| 73 | CHK_PRT_RET( | 73 | CHK_PRT_RET( |
| 74 | - !initialized, HCCL_RUN_WARNING("[PreemptPortManager][Release] preempt port manager has already been release."), | 74 | + !initialized, HCCL_RUN_WARNING("[PreemptPortManager][Release] preempt port manager has already been released."), |
| 75 | HCCL_SUCCESS); | 75 | HCCL_SUCCESS); |
| 76 | 76 | ||
| 77 | CHK_SMART_PTR_NULL(listenSocket); | 77 | CHK_SMART_PTR_NULL(listenSocket); |
| @@ -131,7 +131,7 @@ HcclResult PreemptPortManager::PreemptPortInRange( | |||
| 131 | ret != HCCL_E_UNAVAIL, | 131 | ret != HCCL_E_UNAVAIL, |
| 132 | HCCL_ERROR( | 132 | HCCL_ERROR( |
| 133 | "[PreemptPortManager][PreemptPortInRange] attempt to listen on port[%u] for ip[%s] fail." | 133 | "[PreemptPortManager][PreemptPortInRange] attempt to listen on port[%u] for ip[%s] fail." |
| 134 | - " some unexpected error occurs, errNo[0x%016llx]. attemptation is stopped.", | 134 | + " some unexpected error occurs, errNo[0x%016llx]. attempt is stopped.", |
| 135 | port, ipAddr.c_str(), ret), | 135 | port, ipAddr.c_str(), ret), |
| 136 | ret); | 136 | ret); |
| 137 | // 当前端口已被占用,尝试抢占下一个端口 | 137 | // 当前端口已被占用,尝试抢占下一个端口 |
| @@ -154,7 +154,7 @@ HcclResult PreemptPortManager::PreemptPortInRange( | |||
| 154 | HCCL_ERROR("[PreemptPortManager][PreemptPortInRange] All ports in socket port range are bound already. " | 154 | HCCL_ERROR("[PreemptPortManager][PreemptPortInRange] All ports in socket port range are bound already. " |
| 155 | "no available port to listen. Please check the ports status, or change the port range to listen on."); | 155 | "no available port to listen. Please check the ports status, or change the port range to listen on."); |
| 156 | HCCL_ERROR("NOTICE: Users need to make sure ports in HCCL_HOST_SOCKET_PORT_RANGE and HCCL_NPU_SOCKET_PORT_RANGE " | 156 | HCCL_ERROR("NOTICE: Users need to make sure ports in HCCL_HOST_SOCKET_PORT_RANGE and HCCL_NPU_SOCKET_PORT_RANGE " |
| 157 | - "are available for HCCL. Please double check whether the port are used by others unexpected process. " | 157 | + "are available for HCCL. Please double check whether the ports are used by other unexpected processes. " |
| 158 | "The port ranges size should also be enough when running multi-process HCCL."); | 158 | "The port ranges size should also be enough when running multi-process HCCL."); |
| 159 | HCCL_ERROR("NOTICE: The host/npu port range size is not suggested to be smaller than the process number" | 159 | HCCL_ERROR("NOTICE: The host/npu port range size is not suggested to be smaller than the process number" |
| 160 | " on current rank."); | 160 | " on current rank."); |
| @@ -184,7 +184,7 @@ HcclResult PreemptPortManager::ReleasePreempt( | |||
| 184 | ref.Count() <= 0, | 184 | ref.Count() <= 0, |
| 185 | HCCL_ERROR( | 185 | HCCL_ERROR( |
| 186 | "[PreemptPortManager][ReleasePreempt] ref[%u], ip[%s] port[%u] has already been released. " | 186 | "[PreemptPortManager][ReleasePreempt] ref[%u], ip[%s] port[%u] has already been released. " |
| 187 | - "Please do not dulplicate release.", | 187 | + "Please do not duplicate release.", |
[LOW] dulplicate 拼写修正遗漏(本 PR 已修 910 侧同构 preempt_port_manager.cc) 本 PR 把 第 169 行: 建议修复:
第 187 行:do not duplicate release ![]() ![]() | |||
| 188 | ref.Count(), ipAddr.c_str(), port), | 188 | ref.Count(), ipAddr.c_str(), port), |
| 189 | HCCL_E_INTERNAL); | 189 | HCCL_E_INTERNAL); |
| 190 | 190 | ||
| @@ -204,7 +204,7 @@ HcclResult AicpuAllgather::GenRingTask( | |||
| 204 | HcclResult ret = HCCL_SUCCESS; | 204 | HcclResult ret = HCCL_SUCCESS; |
| 205 | if (step == 1U) { // 首轮send->winIn | 205 | if (step == 1U) { // 首轮send->winIn |
| 206 | ret = AicpuDispatcher::CopyData(streamId, sndAddr, winIn, gatherSize, dataType, HCCL_REDUCE_RESERVED, rankId_); | 206 | ret = AicpuDispatcher::CopyData(streamId, sndAddr, winIn, gatherSize, dataType, HCCL_REDUCE_RESERVED, rankId_); |
| 207 | - CHK_PRT_RET(ret != HCCL_SUCCESS, HCCL_ERROR("Turn %u step %u send to clock winIn failed", turn_, step), ret); | 207 | + CHK_PRT_RET(ret != HCCL_SUCCESS, HCCL_ERROR("Turn %u step %u send to block winIn failed", turn_, step), ret); |
| 208 | } | 208 | } |
| 209 | // 片间同步 notify后卡 wait前卡 | 209 | // 片间同步 notify后卡 wait前卡 |
| 210 | ret = AicpuDispatcher::SignalRecord(streamId, postRankId, AicpuDispatcher::IPC, AicpuDispatcher::PRE_SYNC); | 210 | ret = AicpuDispatcher::SignalRecord(streamId, postRankId, AicpuDispatcher::IPC, AicpuDispatcher::PRE_SYNC); |
| @@ -39,7 +39,7 @@ HcclResult AicpuReduceScatter::RunAlgorithm( | |||
| 39 | } | 39 | } |
| 40 | // dataCount 为tile的输入的数据量(scatter前) | 40 | // dataCount 为tile的输入的数据量(scatter前) |
| 41 | if (dataCount % rankNum_ != 0) { // 每个tile数据量必须能均分至每张卡 | 41 | if (dataCount % rankNum_ != 0) { // 每个tile数据量必须能均分至每张卡 |
| 42 | - HCCL_ERROR("Reduce scatter dataCount %lu max be multiple of rankNum_.", dataCount); | 42 | + HCCL_ERROR("Reduce scatter dataCount %lu must be a multiple of rankNum %u.", dataCount, rankNum_); |
| 43 | return HCCL_E_NOT_SUPPORT; | 43 | return HCCL_E_NOT_SUPPORT; |
| 44 | } | 44 | } |
| 45 | 45 | ||
| @@ -326,7 +326,7 @@ HcclResult AicpuReduceScatter::GenRingTask( | |||
| 326 | HcclResult ret = HCCL_SUCCESS; | 326 | HcclResult ret = HCCL_SUCCESS; |
| 327 | if (step == 1U) { // 首轮send->winIn winOut | 327 | if (step == 1U) { // 首轮send->winIn winOut |
| 328 | ret = AicpuDispatcher::CopyData(streamId, sndAddr, winIn, curSize, dataType, HCCL_REDUCE_RESERVED, rankId_); | 328 | ret = AicpuDispatcher::CopyData(streamId, sndAddr, winIn, curSize, dataType, HCCL_REDUCE_RESERVED, rankId_); |
| 329 | - CHK_PRT_RET(ret != HCCL_SUCCESS, HCCL_ERROR("Turn %u step %u send to clock winIn failed", turn_, step), ret); | 329 | + CHK_PRT_RET(ret != HCCL_SUCCESS, HCCL_ERROR("Turn %u step %u send to block winIn failed", turn_, step), ret); |
| 330 | 330 | ||
| 331 | ret = AicpuDispatcher::CopyData( | 331 | ret = AicpuDispatcher::CopyData( |
| 332 | streamId, sndAddr + scatterSize, winOut, curSize, dataType, HCCL_REDUCE_RESERVED, rankId_); | 332 | streamId, sndAddr + scatterSize, winOut, curSize, dataType, HCCL_REDUCE_RESERVED, rankId_); |
| @@ -511,7 +511,7 @@ bool CheckNsCommand(hccl::HcclCommAicpu* comm) | |||
| 511 | return false; | 511 | return false; |
| 512 | } | 512 | } |
| 513 | comm->SetNsStopLaunchStatus(true); | 513 | comm->SetNsStopLaunchStatus(true); |
| 514 | - HCCL_WARNING("N second stop Launch for recv stop launch cmd."); | 514 | + HCCL_WARNING("N second stop launch for receiving stop launch cmd."); |
| 515 | return true; | 515 | return true; |
| 516 | } | 516 | } |
| 517 | 517 | ||
| @@ -791,7 +791,8 @@ bool SelectAlgName(const std::string& algConfig, u32 topoType, std::string& algN | |||
| 791 | algName = res->second; | 791 | algName = res->second; |
| 792 | return true; | 792 | return true; |
| 793 | } | 793 | } |
| 794 | - HCCL_ERROR("[AicpuHcclProcess][%s] algo_name is not exist, algConfig %s is no.", __func__, algConfig.c_str()); | 794 | + HCCL_ERROR( |
| 795 | + "[AicpuHcclProcess][%s] algo_name does not exist, algConfig %s is invalid.", __func__, algConfig.c_str()); | ||
| 795 | return false; | 796 | return false; |
| 796 | } | 797 | } |
| 797 | 798 | ||
| @@ -800,7 +801,7 @@ bool SplitHcclAlgoGetLevel1Res(std::string& algoConfig, std::string& algos) | |||
| 800 | std::string remainAlgoConfig; | 801 | std::string remainAlgoConfig; |
| 801 | std::size_t found = algoConfig.find(";"); | 802 | std::size_t found = algoConfig.find(";"); |
| 802 | if ((found == 0) || (found == (algoConfig.length() - 1)) || (found == std::string::npos)) { | 803 | if ((found == 0) || (found == (algoConfig.length() - 1)) || (found == std::string::npos)) { |
| 803 | - HCCL_INFO("algoConfig %s thereis no level1 algo config", algoConfig.c_str()); | 804 | + HCCL_INFO("algoConfig %s: there is no level1 algo config", algoConfig.c_str()); |
| 804 | return true; | 805 | return true; |
| 805 | } | 806 | } |
| 806 | remainAlgoConfig = algoConfig.substr(found + 1); | 807 | remainAlgoConfig = algoConfig.substr(found + 1); |
| @@ -154,7 +154,7 @@ HcclMsg (*AicpuKfcRpcServerV2::GetMsgWorkSpace())[HCCL_MSG_CNT] | |||
| 154 | uint64_t AicpuKfcRpcServerV2::GetFinishAddr(int32_t idx) const | 154 | uint64_t AicpuKfcRpcServerV2::GetFinishAddr(int32_t idx) const |
| 155 | { | 155 | { |
| 156 | if (idx >= static_cast<int32_t>(HCCL_MSG_CNT) || hcclMsgArea_ == nullptr) { | 156 | if (idx >= static_cast<int32_t>(HCCL_MSG_CNT) || hcclMsgArea_ == nullptr) { |
| 157 | - HCCL_ERROR("idx %d exceed max or msg area is not initialized.", idx); | 157 | + HCCL_ERROR("idx %d exceeds max idx %u or msg area is not initialized.", idx, HCCL_MSG_CNT); |
| 158 | return 0; | 158 | return 0; |
| 159 | } | 159 | } |
| 160 | return reinterpret_cast<uint64_t>(&(hcclMsgArea_->commMsg.singleMsg.finishedTurnCnt[idx].cnt)); | 160 | return reinterpret_cast<uint64_t>(&(hcclMsgArea_->commMsg.singleMsg.finishedTurnCnt[idx].cnt)); |
| @@ -163,7 +163,7 @@ uint64_t AicpuKfcRpcServerV2::GetFinishAddr(int32_t idx) const | |||
| 163 | uint64_t AicpuKfcRpcServerV2::GetCommitareaAddr(int32_t idx) const | 163 | uint64_t AicpuKfcRpcServerV2::GetCommitareaAddr(int32_t idx) const |
| 164 | { | 164 | { |
| 165 | if (idx >= static_cast<int32_t>(HCCL_MSG_CNT) || hcclMsgArea_ == nullptr) { | 165 | if (idx >= static_cast<int32_t>(HCCL_MSG_CNT) || hcclMsgArea_ == nullptr) { |
| 166 | - HCCL_ERROR("idx %d exceed max or hcclMsgArea_ is not initialized.", idx); | 166 | + HCCL_ERROR("idx %d exceeds max idx %u or hcclMsgArea_ is not initialized.", idx, HCCL_MSG_CNT); |
| 167 | return 0; | 167 | return 0; |
| 168 | } | 168 | } |
| 169 | return reinterpret_cast<uint64_t>(&(hcclMsgArea_->commMsg.singleMsg.commitTurnCnt[idx].cnt)); | 169 | return reinterpret_cast<uint64_t>(&(hcclMsgArea_->commMsg.singleMsg.commitTurnCnt[idx].cnt)); |
| @@ -25,14 +25,14 @@ | |||
| 25 | namespace dfx_tracer { | 25 | namespace dfx_tracer { |
| 26 | void ExecutorTracer::BackGroundDfx(void* info) | 26 | void ExecutorTracer::BackGroundDfx(void* info) |
| 27 | { | 27 | { |
| 28 | - HCCL_RUN_INFO("Start to back ground."); | 28 | + HCCL_RUN_INFO("Start to background."); |
| 29 | // 外部保证info有效 | 29 | // 外部保证info有效 |
| 30 | auto ctx = static_cast<AicpuComContext*>(info); | 30 | auto ctx = static_cast<AicpuComContext*>(info); |
| 31 | hccl::HcclCommAicpu::ResetErrMsgReport(); // 业务重新拉起的场景,重置ErrMesg上报标记位 | 31 | hccl::HcclCommAicpu::ResetErrMsgReport(); // 业务重新拉起的场景,重置ErrMesg上报标记位 |
| 32 | while (true) { | 32 | while (true) { |
| 33 | // 停止背景线程 | 33 | // 停止背景线程 |
| 34 | if (ctx->dfxExtendInfo.commandToBackGroud == CommandToBackGroud::kStop) { | 34 | if (ctx->dfxExtendInfo.commandToBackGroud == CommandToBackGroud::kStop) { |
| 35 | - HCCL_INFO("Back ground thread returned."); | 35 | + HCCL_INFO("Background thread returned."); |
| 36 | break; | 36 | break; |
| 37 | } | 37 | } |
| 38 | HandleDestroyComm(ctx); | 38 | HandleDestroyComm(ctx); |
| @@ -40,7 +40,7 @@ void ExecutorTracer::BackGroundDfx(void* info) | |||
| 40 | bool isNotStop = false; | 40 | bool isNotStop = false; |
| 41 | StopBackGround(ctx, isNotStop); | 41 | StopBackGround(ctx, isNotStop); |
| 42 | if (!isNotStop) { | 42 | if (!isNotStop) { |
| 43 | - HCCL_RUN_INFO("stop backGround Thread"); | 43 | + HCCL_RUN_INFO("stop background thread"); |
| 44 | break; | 44 | break; |
| 45 | } | 45 | } |
| 46 | HandleReportStatusInComm(); | 46 | HandleReportStatusInComm(); |
| @@ -135,7 +135,7 @@ void ExecutorTracer::StopBackGroundDfx(void* info) | |||
| 135 | // 外部保证info有效 | 135 | // 外部保证info有效 |
| 136 | auto ctx = static_cast<AicpuComContext*>(info); | 136 | auto ctx = static_cast<AicpuComContext*>(info); |
| 137 | ctx->dfxExtendInfo.commandToBackGroud = CommandToBackGroud::kStop; | 137 | ctx->dfxExtendInfo.commandToBackGroud = CommandToBackGroud::kStop; |
| 138 | - HCCL_INFO("Stop back ground thread."); | 138 | + HCCL_INFO("Stop background thread."); |
| 139 | } | 139 | } |
| 140 | 140 | ||
| 141 | // handle StopLaunch Command | 141 | // handle StopLaunch Command |
| @@ -158,7 +158,7 @@ void ExecutorTracer::StopLaunchCommandHandle(AicpuComContext* const ctx) | |||
| 158 | hcclAicpu->SetCommRecoveryFlag(true); | 158 | hcclAicpu->SetCommRecoveryFlag(true); |
| 159 | hcclAicpu->SetNsStopLaunchStatus(true); | 159 | hcclAicpu->SetNsStopLaunchStatus(true); |
| 160 | HCCL_RUN_INFO( | 160 | HCCL_RUN_INFO( |
| 161 | - "[NsRecovery][backGround]group[%s] send in aicpu environment", | 161 | + "[NsRecovery][background]group[%s] send in aicpu environment", |
| 162 | hcclAicpu->GetGroupName().c_str()); | 162 | hcclAicpu->GetGroupName().c_str()); |
| 163 | } | 163 | } |
| 164 | } | 164 | } |
| @@ -207,7 +207,7 @@ bool hcclNslbDp::CheckAhcSupport(u8 algType, std::string identifier) | |||
| 207 | 207 | ||
| 208 | s32 ret = strncpy_s(commDesc, COMM_DESC_MAX_LENGTH, identifier.c_str(), identifier.size()); | 208 | s32 ret = strncpy_s(commDesc, COMM_DESC_MAX_LENGTH, identifier.c_str(), identifier.size()); |
| 209 | if (ret != EOK) { | 209 | if (ret != EOK) { |
| 210 | - HCCL_INFO("strncpy_s commDesc fail"); | 210 | + HCCL_WARNING("strncpy_s commDesc fail"); |
| 211 | return true; | 211 | return true; |
| 212 | } | 212 | } |
| 213 | commDesc[COMM_DESC_MAX_LENGTH - 1] = '\0'; | 213 | commDesc[COMM_DESC_MAX_LENGTH - 1] = '\0'; |
| @@ -642,7 +642,7 @@ HcclResult hcclNslbDp::SetNslbDpRootRank(HcclCMDType opType, u32 rootRank, std:: | |||
| 642 | s32 ret | 642 | s32 ret |
| 643 | = strncpy_s(hcclNslbDpRootRankVal_.commDesc, COMM_DESC_MAX_LENGTH, identifier.c_str(), identifier.size()); | 643 | = strncpy_s(hcclNslbDpRootRankVal_.commDesc, COMM_DESC_MAX_LENGTH, identifier.c_str(), identifier.size()); |
| 644 | if (ret != EOK) { | 644 | if (ret != EOK) { |
| 645 | - HCCL_INFO("strncpy_s hcclNslbDpRootRankVal_.commDesc fail"); | 645 | + HCCL_WARNING("strncpy_s hcclNslbDpRootRankVal_.commDesc fail"); |
| 646 | return HCCL_SUCCESS; | 646 | return HCCL_SUCCESS; |
| 647 | } | 647 | } |
| 648 | hcclNslbDpRootRankVal_.commDesc[COMM_DESC_MAX_LENGTH - 1] = '\0'; | 648 | hcclNslbDpRootRankVal_.commDesc[COMM_DESC_MAX_LENGTH - 1] = '\0'; |
| @@ -1003,7 +1003,7 @@ bool hcclNslbDp::FillAlgInfoAdjInfo(NslbDpAlgorithmInfo& algorithmInfo, const Ad | |||
| 1003 | 1003 | ||
| 1004 | if (nslbAdjInfo.nsAdjInfo.size() == 0) { | 1004 | if (nslbAdjInfo.nsAdjInfo.size() == 0) { |
| 1005 | algorithmInfo.dstRankNum = 0; | 1005 | algorithmInfo.dstRankNum = 0; |
| 1006 | - HCCL_INFO("[NSLB-DP] get nsAdjInfo fail dstRankNum:[%u]", algorithmInfo.dstRankNum); | 1006 | + HCCL_WARNING("[NSLB-DP] get nsAdjInfo fail dstRankNum:[%u]", algorithmInfo.dstRankNum); |
| 1007 | return false; | 1007 | return false; |
| 1008 | } | 1008 | } |
| 1009 | 1009 | ||
| @@ -1037,7 +1037,7 @@ void hcclNslbDp::fullcommDescInitTime(std::string identifier, NslbDpOperatorInfo | |||
| 1037 | /* 获取通信域唯一标识 */ | 1037 | /* 获取通信域唯一标识 */ |
| 1038 | s32 ret = strncpy_s(OperatorInfo.commDesc, COMM_DESC_MAX_LENGTH, identifier.c_str(), identifier.size()); | 1038 | s32 ret = strncpy_s(OperatorInfo.commDesc, COMM_DESC_MAX_LENGTH, identifier.c_str(), identifier.size()); |
| 1039 | if (ret != EOK) { | 1039 | if (ret != EOK) { |
| 1040 | - HCCL_INFO("[NSLB-DP] strncpy_s OperatorInfo.commDesc fail"); | 1040 | + HCCL_WARNING("[NSLB-DP] strncpy_s OperatorInfo.commDesc fail"); |
| 1041 | return; | 1041 | return; |
| 1042 | } | 1042 | } |
| 1043 | HCCL_DEBUG("[NSLB-DP-OPER] fullcommDescInitTime commDesc[%s] .", identifier.c_str()); | 1043 | HCCL_DEBUG("[NSLB-DP-OPER] fullcommDescInitTime commDesc[%s] .", identifier.c_str()); |
| @@ -1097,7 +1097,7 @@ HcclResult hcclNslbDp::GenerateOpAndAdjTable( | |||
| 1097 | 1097 | ||
| 1098 | // 去除不存在的通信域信息 | 1098 | // 去除不存在的通信域信息 |
| 1099 | if (CheckCommDescExit(OperatorInfo) == false) { | 1099 | if (CheckCommDescExit(OperatorInfo) == false) { |
| 1100 | - HCCL_INFO("[NSLB-DP-OPER] CheckCommDesc not exit "); | 1100 | + HCCL_INFO("[NSLB-DP-OPER] CheckCommDesc not exist "); |
| 1101 | return HCCL_SUCCESS; | 1101 | return HCCL_SUCCESS; |
| 1102 | } | 1102 | } |
| 1103 | 1103 | ||
| @@ -1240,7 +1240,7 @@ void hcclNslbDp::fullCommConfigInfo(NslbDpCommConfigInfo& tab_f, NslbDpCommConfi | |||
| 1240 | tab_f.taskId = cominfo.taskId; | 1240 | tab_f.taskId = cominfo.taskId; |
| 1241 | s32 sRet = memcpy_s(tab_f.commDesc, sizeof(tab_f.commDesc), cominfo.commDesc, COMM_DESC_MAX_LENGTH); | 1241 | s32 sRet = memcpy_s(tab_f.commDesc, sizeof(tab_f.commDesc), cominfo.commDesc, COMM_DESC_MAX_LENGTH); |
| 1242 | if (sRet != EOK) { | 1242 | if (sRet != EOK) { |
| 1243 | - HCCL_INFO("memcpy_s commDesc fail"); | 1243 | + HCCL_WARNING("memcpy_s commDesc fail"); |
| 1244 | } | 1244 | } |
| 1245 | tab_f.commInitTime = cominfo.commInitTime; | 1245 | tab_f.commInitTime = cominfo.commInitTime; |
| 1246 | 1246 | ||
| @@ -1253,7 +1253,7 @@ void hcclNslbDp::fullCommConfigInfo(NslbDpCommConfigInfo& tab_f, NslbDpCommConfi | |||
| 1253 | } | 1253 | } |
| 1254 | sRet = memcpy_s(tab_f.commMd5Sum, sizeof(tab_f.commMd5Sum), cominfo.commMd5Sum, sizeof(cominfo.commMd5Sum)); | 1254 | sRet = memcpy_s(tab_f.commMd5Sum, sizeof(tab_f.commMd5Sum), cominfo.commMd5Sum, sizeof(cominfo.commMd5Sum)); |
| 1255 | if (sRet != EOK) { | 1255 | if (sRet != EOK) { |
| 1256 | - HCCL_INFO("memcpy_s commmd5 fail"); | 1256 | + HCCL_WARNING("memcpy_s commmd5 fail"); |
| 1257 | } | 1257 | } |
| 1258 | } | 1258 | } |
| 1259 | 1259 | ||
| @@ -1329,7 +1329,7 @@ HcclResult hcclNslbDp::SendRankTable(NslbDpCommConfigInfo tab_f) | |||
| 1329 | HCCL_INFO("[NSLB-DP] SendRankTable tlvData.len:[%u] success.", datlen); | 1329 | HCCL_INFO("[NSLB-DP] SendRankTable tlvData.len:[%u] success.", datlen); |
| 1330 | 1330 | ||
| 1331 | if (nslbdp_handle_ == nullptr) { | 1331 | if (nslbdp_handle_ == nullptr) { |
| 1332 | - HCCL_INFO("[NSLB-DP] ndlbdp nslbdp_handle_ error SendRankTable."); | 1332 | + HCCL_WARNING("[NSLB-DP] nslbdp nslbdp_handle_ error SendRankTable."); |
| 1333 | return HCCL_SUCCESS; | 1333 | return HCCL_SUCCESS; |
| 1334 | } | 1334 | } |
| 1335 | 1335 | ||
| @@ -1451,7 +1451,7 @@ HcclResult hcclNslbDp::SendOpAndAdjTable() | |||
| 1451 | tab_f.maskLen = hcclNslbDpOperatorVal_[i].maskLen; | 1451 | tab_f.maskLen = hcclNslbDpOperatorVal_[i].maskLen; |
| 1452 | SendRankTableOpAndAdj(tab_f); | 1452 | SendRankTableOpAndAdj(tab_f); |
| 1453 | hcclNslbDpOperatorVal_[i].sedFlag = 1; | 1453 | hcclNslbDpOperatorVal_[i].sedFlag = 1; |
| 1454 | - HCCL_INFO("[NSLB-DP] try to sen RankTableOpAndAdj times:[%u].", i); | 1454 | + HCCL_INFO("[NSLB-DP] try to send RankTableOpAndAdj times:[%u].", i); |
| 1455 | } | 1455 | } |
| 1456 | HCCL_INFO("[NSLB-DP] SendOpAndAdjTable end."); | 1456 | HCCL_INFO("[NSLB-DP] SendOpAndAdjTable end."); |
| 1457 | return HCCL_SUCCESS; | 1457 | return HCCL_SUCCESS; |
| @@ -1570,7 +1570,7 @@ HcclResult hcclNslbDp::SendAlgorithmInfoTable() | |||
| 1570 | tab_f.commMd5Sum, sizeof(tab_f.commMd5Sum), hcclNslbDpAlgorithmInfo_[i].commMd5Sum, | 1570 | tab_f.commMd5Sum, sizeof(tab_f.commMd5Sum), hcclNslbDpAlgorithmInfo_[i].commMd5Sum, |
| 1571 | sizeof(hcclNslbDpAlgorithmInfo_[i].commMd5Sum)); | 1571 | sizeof(hcclNslbDpAlgorithmInfo_[i].commMd5Sum)); |
| 1572 | if (sRet != EOK) { | 1572 | if (sRet != EOK) { |
| 1573 | - HCCL_INFO("memcpy_s commDesc fail"); | 1573 | + HCCL_WARNING("memcpy_s commDesc fail"); |
| 1574 | } | 1574 | } |
| 1575 | tab_f.srcLocalRankId = hcclNslbDpAlgorithmInfo_[i].srcLocalRankId; | 1575 | tab_f.srcLocalRankId = hcclNslbDpAlgorithmInfo_[i].srcLocalRankId; |
| 1576 | tab_f.rootRank = hcclNslbDpAlgorithmInfo_[i].rootRank; | 1576 | tab_f.rootRank = hcclNslbDpAlgorithmInfo_[i].rootRank; |
| @@ -1584,7 +1584,7 @@ HcclResult hcclNslbDp::SendAlgorithmInfoTable() | |||
| 1584 | tab_f.AdjInfo = hcclNslbDpAlgorithmInfo_[i].AdjInfo; | 1584 | tab_f.AdjInfo = hcclNslbDpAlgorithmInfo_[i].AdjInfo; |
| 1585 | 1585 | ||
| 1586 | SendRankTableAlgorithmInfo(tab_f); | 1586 | SendRankTableAlgorithmInfo(tab_f); |
| 1587 | - HCCL_INFO("[NSLB-DP] try to sen AlgorithmInfoTable times:[%u].", i); | 1587 | + HCCL_INFO("[NSLB-DP] try to send AlgorithmInfoTable times:[%u].", i); |
| 1588 | 1588 | ||
| 1589 | hcclNslbDpAlgorithmInfo_[i].sedFlag = 1; | 1589 | hcclNslbDpAlgorithmInfo_[i].sedFlag = 1; |
| 1590 | } | 1590 | } |
| @@ -1856,7 +1856,7 @@ HcclResult hcclNslbDp::SendGlobalDisRankTable() | |||
| 1856 | s32 sRet = memcpy_s( | 1856 | s32 sRet = memcpy_s( |
| 1857 | tab_f.rev, sizeof(tab_f.rev), hcclNslbDpGlobalDisRankVal_.rev, sizeof(hcclNslbDpGlobalDisRankVal_.rev)); | 1857 | tab_f.rev, sizeof(tab_f.rev), hcclNslbDpGlobalDisRankVal_.rev, sizeof(hcclNslbDpGlobalDisRankVal_.rev)); |
| 1858 | if (sRet != EOK) { | 1858 | if (sRet != EOK) { |
| 1859 | - HCCL_INFO("memcpy_s rev info fail"); | 1859 | + HCCL_WARNING("memcpy_s rev info fail"); |
| 1860 | } | 1860 | } |
| 1861 | tab_f.rankTotalNum = hcclNslbDpGlobalDisRankVal_.rankTotalNum; | 1861 | tab_f.rankTotalNum = hcclNslbDpGlobalDisRankVal_.rankTotalNum; |
| 1862 | 1862 | ||
| @@ -1954,7 +1954,7 @@ HcclResult hcclNslbDp::SendRootRankTable() | |||
| 1954 | tab_f.commDesc, sizeof(tab_f.commDesc), hcclNslbDpRootRankVal_.commDesc, | 1954 | tab_f.commDesc, sizeof(tab_f.commDesc), hcclNslbDpRootRankVal_.commDesc, |
| 1955 | sizeof(hcclNslbDpRootRankVal_.commDesc)); | 1955 | sizeof(hcclNslbDpRootRankVal_.commDesc)); |
| 1956 | if (sRet != EOK) { | 1956 | if (sRet != EOK) { |
| 1957 | - HCCL_INFO("memcpy_s commDesc info fail"); | 1957 | + HCCL_WARNING("memcpy_s commDesc info fail"); |
| 1958 | } | 1958 | } |
| 1959 | tab_f.commInitTime = hcclNslbDpRootRankVal_.commInitTime; | 1959 | tab_f.commInitTime = hcclNslbDpRootRankVal_.commInitTime; |
| 1960 | tab_f.oper = hcclNslbDpRootRankVal_.oper; | 1960 | tab_f.oper = hcclNslbDpRootRankVal_.oper; |
| @@ -440,7 +440,7 @@ HcclResult HccdImplPml::AtomicInitSet() | |||
| 440 | initializedFlag_.test_and_set(), | 440 | initializedFlag_.test_and_set(), |
| 441 | HCCL_ERROR( | 441 | HCCL_ERROR( |
| 442 | "[HcclImplBase][AtomicInitSet]errNo[0x%016llx] instance " | 442 | "[HcclImplBase][AtomicInitSet]errNo[0x%016llx] instance " |
| 443 | - "already been initialized", | 443 | + "has already been initialized", |
| 444 | HCCL_ERROR_CODE(HCCL_E_INTERNAL)), | 444 | HCCL_ERROR_CODE(HCCL_E_INTERNAL)), |
| 445 | HCCL_E_INTERNAL); | 445 | HCCL_E_INTERNAL); |
| 446 | return HCCL_SUCCESS; | 446 | return HCCL_SUCCESS; |
| @@ -676,7 +676,7 @@ HcclResult HrtRaSendWrlist( | |||
| 676 | CHK_PRT_RET( | 676 | CHK_PRT_RET( |
| 677 | bTimeout, | 677 | bTimeout, |
| 678 | HCCL_ERROR( | 678 | HCCL_ERROR( |
| 679 | - "[Send][RaWrList]errNo[0x%016llx] ra send wrlsit async timeout[%d s]. " | 679 | + "[Send][RaWrList]errNo[0x%016llx] ra send wrlist async timeout[%d s]. " |
| 680 | "return[%d], params: send_wrAddr[%p], opRspAddr[%p]", | 680 | "return[%d], params: send_wrAddr[%p], opRspAddr[%p]", |
| 681 | HCCL_ERROR_CODE(HCCL_E_ROCE_TRANSFER), timeout, ret, wr, opRsp), | 681 | HCCL_ERROR_CODE(HCCL_E_ROCE_TRANSFER), timeout, ret, wr, opRsp), |
| 682 | HCCL_E_ROCE_TRANSFER); | 682 | HCCL_E_ROCE_TRANSFER); |
| @@ -805,7 +805,7 @@ HcclResult HrtRaSendNormalWrlist( | |||
| 805 | CHK_PRT_RET( | 805 | CHK_PRT_RET( |
| 806 | bTimeout, | 806 | bTimeout, |
| 807 | HCCL_ERROR( | 807 | HCCL_ERROR( |
| 808 | - "[Send][HrtRaSendNormalWrlist]errNo[0x%016llx] ra send wrlsit async timeout[%d s]. " | 808 | + "[Send][HrtRaSendNormalWrlist]errNo[0x%016llx] ra send wrlist async timeout[%d s]. " |
| 809 | "return[%d], params: send_wrAddr[%p], opRspAddr[%p]", | 809 | "return[%d], params: send_wrAddr[%p], opRspAddr[%p]", |
| 810 | HCCL_ERROR_CODE(HCCL_E_ROCE_TRANSFER), timeout, ret, wr, opRsp), | 810 | HCCL_ERROR_CODE(HCCL_E_ROCE_TRANSFER), timeout, ret, wr, opRsp), |
| 811 | HCCL_E_ROCE_TRANSFER); | 811 | HCCL_E_ROCE_TRANSFER); |
| @@ -1652,7 +1652,7 @@ HcclResult hrtRaSocketNonBlockSendHeterog(const FdHandle fdHandle, const void* d | |||
| 1652 | } else if (ret == SOCK_EAGAIN) { | 1652 | } else if (ret == SOCK_EAGAIN) { |
| 1653 | return HCCL_E_AGAIN; | 1653 | return HCCL_E_AGAIN; |
| 1654 | } else { | 1654 | } else { |
| 1655 | - HCCL_RUN_INFO( | 1655 | + HCCL_RUN_WARNING( |
| 1656 | "[hrtRaSocketNonBlockSend]ra socket send failed, data[%p], size[%llu Byte], " | 1656 | "[hrtRaSocketNonBlockSend]ra socket send failed, data[%p], size[%llu Byte], " |
| 1657 | "sent[%llu Byte], ret[%d]", | 1657 | "sent[%llu Byte], ret[%d]", |
| 1658 | data, size, *sentSize, ret); | 1658 | data, size, *sentSize, ret); |
| @@ -1773,7 +1773,7 @@ HcclResult hrtRaSocketNonBlockRecvHeterog(const FdHandle fdHandle, void* data, u | |||
| 1773 | } else if (ret == SOCK_EAGAIN) { | 1773 | } else if (ret == SOCK_EAGAIN) { |
| 1774 | return HCCL_E_AGAIN; | 1774 | return HCCL_E_AGAIN; |
| 1775 | } else { | 1775 | } else { |
| 1776 | - HCCL_RUN_INFO( | 1776 | + HCCL_RUN_WARNING( |
| 1777 | "[hrtRaSocketNonBlockRecv]ra socket recv failed, data[%p], size[%llu Byte], " | 1777 | "[hrtRaSocketNonBlockRecv]ra socket recv failed, data[%p], size[%llu Byte], " |
| 1778 | "recv[%llu Byte], ret[%d], errno[%d][%s]", | 1778 | "recv[%llu Byte], ret[%d], errno[%d][%s]", |
| 1779 | data, size, recvSize, ret, errno, strerror(errno)); | 1779 | data, size, recvSize, ret, errno, strerror(errno)); |
| @@ -2618,7 +2618,7 @@ HcclResult DestroyQpWithSharedCq(const QpInfo& info, s32 qpAppend) | |||
| 2618 | 2618 | ||
| 2619 | unique_lock<mutex> lock(g_qpRecordsMutex); | 2619 | unique_lock<mutex> lock(g_qpRecordsMutex); |
| 2620 | if (g_qpRecords[label].empty()) { | 2620 | if (g_qpRecords[label].empty()) { |
| 2621 | - HCCL_ERROR("qp label[%s] no exist.", label.c_str()); | 2621 | + HCCL_ERROR("qp label[%s] does not exist.", label.c_str()); |
| 2622 | return HCCL_E_PARA; | 2622 | return HCCL_E_PARA; |
| 2623 | } else { | 2623 | } else { |
| 2624 | for (auto itCq = g_qpRecords[label].begin(); itCq != g_qpRecords[label].end(); itCq++) { | 2624 | for (auto itCq = g_qpRecords[label].begin(); itCq != g_qpRecords[label].end(); itCq++) { |
| @@ -2643,11 +2643,11 @@ HcclResult DestroyQpWithSharedCq(const QpInfo& info, s32 qpAppend) | |||
| 2643 | return HCCL_SUCCESS; | 2643 | return HCCL_SUCCESS; |
| 2644 | } | 2644 | } |
| 2645 | } | 2645 | } |
| 2646 | - HCCL_ERROR("DestroyQp: the qp is no exist"); | 2646 | + HCCL_ERROR("DestroyQp: the qp does not exist"); |
| 2647 | return HCCL_E_PARA; | 2647 | return HCCL_E_PARA; |
| 2648 | } | 2648 | } |
| 2649 | } | 2649 | } |
| 2650 | - HCCL_ERROR("DestroyQp: the cq is no exist"); | 2650 | + HCCL_ERROR("DestroyQp: the cq does not exist"); |
| 2651 | return HCCL_E_PARA; | 2651 | return HCCL_E_PARA; |
| 2652 | } | 2652 | } |
| 2653 | } | 2653 | } |
| @@ -2849,7 +2849,7 @@ HcclResult hrtRaDestroyCompChannel(RdmaHandle rdmaHandle, void* compChannel) | |||
| 2849 | CHK_PRT_RET( | 2849 | CHK_PRT_RET( |
| 2850 | ret != 0, | 2850 | ret != 0, |
| 2851 | HCCL_ERROR( | 2851 | HCCL_ERROR( |
| 2852 | - "[Destroy][CompChannel]errNo[0x%016llx] ra destroy normal qp fail. " | 2852 | + "[Destroy][CompChannel]errNo[0x%016llx] ra destroy comp channel fail. " |
| 2853 | "return[%d], params: rdmaHandle[%p], compChannel[%p]", | 2853 | "return[%d], params: rdmaHandle[%p], compChannel[%p]", |
| 2854 | HCCL_ERROR_CODE(HCCL_E_NETWORK), ret, rdmaHandle, compChannel), | 2854 | HCCL_ERROR_CODE(HCCL_E_NETWORK), ret, rdmaHandle, compChannel), |
| 2855 | HCCL_E_NETWORK); | 2855 | HCCL_E_NETWORK); |
| @@ -2967,7 +2967,7 @@ HcclResult hrtRaDestroySrq(RdmaHandle rdmaHandle, SrqInfo& srqInfo) | |||
| 2967 | CHK_PRT_RET( | 2967 | CHK_PRT_RET( |
| 2968 | ret != 0, | 2968 | ret != 0, |
| 2969 | HCCL_ERROR( | 2969 | HCCL_ERROR( |
| 2970 | - "[Destroy][Srq]errNo[0x%016llx] ra destroy normal qp fail. " | 2970 | + "[Destroy][Srq]errNo[0x%016llx] ra destroy srq fail. " |
| 2971 | "return[%d], params: rdmaHandle[%p]", | 2971 | "return[%d], params: rdmaHandle[%p]", |
| 2972 | HCCL_ERROR_CODE(HCCL_E_NETWORK), ret, rdmaHandle), | 2972 | HCCL_ERROR_CODE(HCCL_E_NETWORK), ret, rdmaHandle), |
| 2973 | HCCL_E_NETWORK); | 2973 | HCCL_E_NETWORK); |
| @@ -3162,7 +3162,7 @@ HcclResult hrtRaRecvWrlist(QpHandle handle, struct RecvWrlistData* wr, unsigned | |||
| 3162 | CHK_PRT_RET( | 3162 | CHK_PRT_RET( |
| 3163 | bTimeout, | 3163 | bTimeout, |
| 3164 | HCCL_ERROR( | 3164 | HCCL_ERROR( |
| 3165 | - "[Recv][RaWrList]errNo[0x%016llx] ra Recv wrlsit async timeout[%d s]. " | 3165 | + "[Recv][RaWrList]errNo[0x%016llx] ra Recv wrlist async timeout[%d s]. " |
| 3166 | "return[%d], params: send_wrAddr[%p]", | 3166 | "return[%d], params: send_wrAddr[%p]", |
| 3167 | HCCL_ERROR_CODE(HCCL_E_ROCE_TRANSFER), timeout, ret, wr), | 3167 | HCCL_ERROR_CODE(HCCL_E_ROCE_TRANSFER), timeout, ret, wr), |
| 3168 | HCCL_E_ROCE_TRANSFER); | 3168 | HCCL_E_ROCE_TRANSFER); |
| @@ -3236,7 +3236,7 @@ HcclResult hrtRaGetSocketVnicIpInfos(u32 phyId, enum IdType type, vector<u32> de | |||
| 3236 | CHK_PRT_RET( | 3236 | CHK_PRT_RET( |
| 3237 | ret != 0, | 3237 | ret != 0, |
| 3238 | HCCL_ERROR( | 3238 | HCCL_ERROR( |
| 3239 | - "[hrtRaGetSocketVnicIpInfo]errNo[0x%016llx] ra get VnicIpfail. ret[%d]", | 3239 | + "[hrtRaGetSocketVnicIpInfo]errNo[0x%016llx] ra get VnicIp fail. ret[%d]", |
| 3240 | HCCL_ERROR_CODE(HCCL_E_TCP_CONNECT), ret), | 3240 | HCCL_ERROR_CODE(HCCL_E_TCP_CONNECT), ret), |
| 3241 | HCCL_E_TCP_CONNECT); | 3241 | HCCL_E_TCP_CONNECT); |
| 3242 | 3242 | ||
| @@ -551,8 +551,8 @@ HcclResult IsAllDigit(const char* strNum) | |||
| 551 | for (; index < nLength; index++) { | 551 | for (; index < nLength; index++) { |
| 552 | if (!isdigit(strNum[index])) { | 552 | if (!isdigit(strNum[index])) { |
| 553 | HCCL_ERROR( | 553 | HCCL_ERROR( |
| 554 | - "[Check][Isdigit]errNo[0x%016llx] In judge all digit, check isdigit failed." | 554 | + "[Check][Isdigit]errNo[0x%016llx] In judge all digit, check isdigit failed. " |
| 555 | - "ensure that the number is an integer. strNum[%u] is [%d](Dec)", | 555 | + "Ensure that the number is an integer. strNum[%u] is [%d](Dec)", |
| 556 | HCCL_ERROR_CODE(HCCL_E_PARA), index, strNum[index]); | 556 | HCCL_ERROR_CODE(HCCL_E_PARA), index, strNum[index]); |
| 557 | return HCCL_E_PARA; | 557 | return HCCL_E_PARA; |
| 558 | } | 558 | } |
| @@ -358,7 +358,7 @@ inline HcclResult RpingUbAttrInit( | |||
| 358 | initAttr.mode = NETWORK_OFFLINE; // net work mode 枚举值 | 358 | initAttr.mode = NETWORK_OFFLINE; // net work mode 枚举值 |
| 359 | initAttr.ub.phyId = deviceId; | 359 | initAttr.ub.phyId = deviceId; |
| 360 | if (eidmap.find(ipAddr.GetEid()) == eidmap.end()) { | 360 | if (eidmap.find(ipAddr.GetEid()) == eidmap.end()) { |
| 361 | - HCCL_ERROR("eidmap don't have input Eid,Input Eid %s", ipAddr.GetEid().Describe().c_str()); | 361 | + HCCL_ERROR("eidmap does not have the input Eid, input Eid: %s", ipAddr.GetEid().Describe().c_str()); |
| 362 | return HCCL_E_NOT_FOUND; | 362 | return HCCL_E_NOT_FOUND; |
| 363 | } | 363 | } |
| 364 | initAttr.dev.ub.eidIndex = eidmap.at(ipAddr.GetEid()); // 从eid_list获取eidIndex | 364 | initAttr.dev.ub.eidIndex = eidmap.at(ipAddr.GetEid()); // 从eid_list获取eidIndex |
| @@ -884,13 +884,13 @@ HcclResult PingMesh::HccnRpingInit( | |||
| 884 | if (ret != HCCL_SUCCESS) { | 884 | if (ret != HCCL_SUCCESS) { |
| 885 | status = RpingInitState::HCCL_TSD_NEED_CLOSE; | 885 | status = RpingInitState::HCCL_TSD_NEED_CLOSE; |
| 886 | HCCL_ERROR( | 886 | HCCL_ERROR( |
| 887 | - "[HccnRpingInit]call ra_get_dev_eid_map failed, devideId[%u], error code =%d.", deviceId, ret); | 887 | + "[HccnRpingInit]call ra_get_dev_eid_map failed, deviceId[%u], error code =%d.", deviceId, ret); |
| 888 | break; | 888 | break; |
| 889 | } | 889 | } |
| 890 | ret = RpingUbAttrInit(devicePhyId_, ipAddr, port, nodeNum, bufferSize, sl, tc, initAttr, eidmap); | 890 | ret = RpingUbAttrInit(devicePhyId_, ipAddr, port, nodeNum, bufferSize, sl, tc, initAttr, eidmap); |
| 891 | if (ret != HCCL_SUCCESS) { | 891 | if (ret != HCCL_SUCCESS) { |
| 892 | status = RpingInitState::HCCL_TSD_NEED_CLOSE; | 892 | status = RpingInitState::HCCL_TSD_NEED_CLOSE; |
| 893 | - HCCL_ERROR("[HccnRpingInit]RpingUbAttrInit failed, devideId[%u], error code =%d.", deviceId, ret); | 893 | + HCCL_ERROR("[HccnRpingInit]RpingUbAttrInit failed, deviceId[%u], error code =%d.", deviceId, ret); |
| 894 | break; | 894 | break; |
| 895 | } | 895 | } |
| 896 | } | 896 | } |
| @@ -163,7 +163,7 @@ HcclResult HDCommunicate::Get(u32 offset, u32 length, u8* value) | |||
| 163 | CHK_PTR_NULL(value); | 163 | CHK_PTR_NULL(value); |
| 164 | CHK_PRT_RET( | 164 | CHK_PRT_RET( |
| 165 | (offset + length > buffLen_), | 165 | (offset + length > buffLen_), |
| 166 | - HCCL_ERROR("[HDCommunicate][Get]Invalid length, offset=%u, length=%u, befferLen=%u", offset, length, buffLen_), | 166 | + HCCL_ERROR("[HDCommunicate][Get]Invalid length, offset=%u, length=%u, bufferLen=%u", offset, length, buffLen_), |
| 167 | HCCL_E_PARA); | 167 | HCCL_E_PARA); |
| 168 | std::shared_lock<std::shared_mutex> lock(lock_); | 168 | std::shared_lock<std::shared_mutex> lock(lock_); |
| 169 | HcclResult ret = Read(offset, length, value); | 169 | HcclResult ret = Read(offset, length, value); |
| @@ -930,7 +930,7 @@ HcclResult TransportHeterogRoce::GetNetworkResource() | |||
| 930 | CHK_RET(NetworkManager::GetInstance(index_).GetRaResourceInfo(raResourceInfo)); | 930 | CHK_RET(NetworkManager::GetInstance(index_).GetRaResourceInfo(raResourceInfo)); |
| 931 | auto it = raResourceInfo.nicSocketMap.find(selfIp_); | 931 | auto it = raResourceInfo.nicSocketMap.find(selfIp_); |
| 932 | if (it == raResourceInfo.nicSocketMap.end()) { | 932 | if (it == raResourceInfo.nicSocketMap.end()) { |
| 933 | - HCCL_ERROR("[TransportHeterogRoce][Init]nic socket handle did not found"); | 933 | + HCCL_ERROR("[TransportHeterogRoce][Init]nic socket handle was not found"); |
| 934 | return HCCL_E_PARA; | 934 | return HCCL_E_PARA; |
| 935 | } | 935 | } |
| 936 | nicSocketHandle_ = it->second.nicSocketHandle; | 936 | nicSocketHandle_ = it->second.nicSocketHandle; |
| @@ -1325,7 +1325,7 @@ HcclResult TransportHeterogRoce::PsRdmaDbSend(uint32_t dbindex, uint64_t dbinfo, | |||
| 1325 | ret != RT_ERROR_NONE, | 1325 | ret != RT_ERROR_NONE, |
| 1326 | HCCL_ERROR( | 1326 | HCCL_ERROR( |
| 1327 | "[rtRDMADBSend]errNo[0x%016llx] rt rdma send fail, " | 1327 | "[rtRDMADBSend]errNo[0x%016llx] rt rdma send fail, " |
| 1328 | - "return[%d]. para: dbindex[%u]dbinfo[%llu].", | 1328 | + "return[%d]. para: dbindex[%u] dbinfo[%llu].", |
| 1329 | HCCL_ERROR_CODE(HCCL_E_RUNTIME), ret, dbindex, dbinfo), | 1329 | HCCL_ERROR_CODE(HCCL_E_RUNTIME), ret, dbindex, dbinfo), |
| 1330 | HCCL_E_RUNTIME); | 1330 | HCCL_E_RUNTIME); |
| 1331 | if (deviceLogicId_ == HOST_DEVICE_ID) { | 1331 | if (deviceLogicId_ == HOST_DEVICE_ID) { |
| @@ -123,7 +123,7 @@ TransportMem::DoExchangeMemDesc(const RmaMemDescs& localMemDescs, RmaMemDescs& r | |||
| 123 | CHK_RET(SendLocalMemDesc(localMemDescs)); | 123 | CHK_RET(SendLocalMemDesc(localMemDescs)); |
| 124 | CHK_RET(ReceiveRemoteMemDesc(remoteMemDescs, actualNumOfRemote)); | 124 | CHK_RET(ReceiveRemoteMemDesc(remoteMemDescs, actualNumOfRemote)); |
| 125 | } | 125 | } |
| 126 | - HCCL_INFO("[HcclOneSidedConn][ExchangeMemDesc]get actualNumOfRemotee[%u]", actualNumOfRemote); | 126 | + HCCL_INFO("[HcclOneSidedConn][ExchangeMemDesc]get actualNumOfRemote[%u]", actualNumOfRemote); |
| 127 | // 校验remoteDescs中的remoteRankId和conn对象中保存的localRankId是否一样 | 127 | // 校验remoteDescs中的remoteRankId和conn对象中保存的localRankId是否一样 |
| 128 | for (u32 i = 0; i < actualNumOfRemote; i++) { | 128 | for (u32 i = 0; i < actualNumOfRemote; i++) { |
| 129 | CHK_PTR_NULL((remoteMemDescs.array) + i); | 129 | CHK_PTR_NULL((remoteMemDescs.array) + i); |
| @@ -286,7 +286,7 @@ HcclResult TransportRoceMem::SetSocket(const std::shared_ptr<HcclSocket>& socket | |||
| 286 | { | 286 | { |
| 287 | CHK_SMART_PTR_NULL(socket); | 287 | CHK_SMART_PTR_NULL(socket); |
| 288 | if (socket->GetStatus() != HcclSocketStatus::SOCKET_OK) { | 288 | if (socket->GetStatus() != HcclSocketStatus::SOCKET_OK) { |
| 289 | - HCCL_ERROR("sockets does not connected"); | 289 | + HCCL_ERROR("sockets is not connected"); |
| 290 | return HCCL_E_PARA; | 290 | return HCCL_E_PARA; |
| 291 | } | 291 | } |
| 292 | socket_ = socket; | 292 | socket_ = socket; |
| @@ -423,7 +423,7 @@ HcclResult TransportRoceMem::Write(const HcclBuf& remoteMem, const HcclBuf& loca | |||
| 423 | CHK_PRT_RET( | 423 | CHK_PRT_RET( |
| 424 | (localMem.len == 0U) || (remoteMem.len == 0U), | 424 | (localMem.len == 0U) || (remoteMem.len == 0U), |
| 425 | HCCL_ERROR( | 425 | HCCL_ERROR( |
| 426 | - "[TransportRoceMem]localMem size[%llu] or remoteMem size[%llu]is invalid", localMem.len, remoteMem.len), | 426 | + "[TransportRoceMem]localMem size[%llu] or remoteMem size[%llu] is invalid", localMem.len, remoteMem.len), |
| 427 | HCCL_E_PARA); | 427 | HCCL_E_PARA); |
| 428 | CHK_PRT_RET(stream == nullptr, HCCL_ERROR("[TransportRoceMem]stream is invalid"), HCCL_E_PARA); | 428 | CHK_PRT_RET(stream == nullptr, HCCL_ERROR("[TransportRoceMem]stream is invalid"), HCCL_E_PARA); |
| 429 | 429 | ||
| @@ -443,7 +443,7 @@ HcclResult TransportRoceMem::Write(const RmaOpMem& remoteMem, const RmaOpMem& lo | |||
| 443 | CHK_PRT_RET( | 443 | CHK_PRT_RET( |
| 444 | (localMem.size == 0U) || (remoteMem.size == 0U), | 444 | (localMem.size == 0U) || (remoteMem.size == 0U), |
| 445 | HCCL_ERROR( | 445 | HCCL_ERROR( |
| 446 | - "[TransportRoceMem]localMem size[%llu] or remoteMem size[%llu]is invalid", localMem.size, remoteMem.size), | 446 | + "[TransportRoceMem]localMem size[%llu] or remoteMem size[%llu] is invalid", localMem.size, remoteMem.size), |
| 447 | HCCL_E_PARA); | 447 | HCCL_E_PARA); |
| 448 | CHK_PRT_RET(stream == nullptr, HCCL_ERROR("[TransportRoceMem]stream is invalid"), HCCL_E_PARA); | 448 | CHK_PRT_RET(stream == nullptr, HCCL_ERROR("[TransportRoceMem]stream is invalid"), HCCL_E_PARA); |
| 449 | 449 | ||
| @@ -463,7 +463,7 @@ HcclResult TransportRoceMem::Read(const HcclBuf& localMem, const HcclBuf& remote | |||
| 463 | CHK_PRT_RET( | 463 | CHK_PRT_RET( |
| 464 | (localMem.len == 0U) || (remoteMem.len == 0U), | 464 | (localMem.len == 0U) || (remoteMem.len == 0U), |
| 465 | HCCL_ERROR( | 465 | HCCL_ERROR( |
| 466 | - "[TransportRoceMem]localMem size[%llu] or remoteMem size[%llu]is invalid", localMem.len, remoteMem.len), | 466 | + "[TransportRoceMem]localMem size[%llu] or remoteMem size[%llu] is invalid", localMem.len, remoteMem.len), |
| 467 | HCCL_E_PARA); | 467 | HCCL_E_PARA); |
| 468 | CHK_PRT_RET(stream == nullptr, HCCL_ERROR("[TransportRoceMem]stream is invalid"), HCCL_E_PARA); | 468 | CHK_PRT_RET(stream == nullptr, HCCL_ERROR("[TransportRoceMem]stream is invalid"), HCCL_E_PARA); |
| 469 | 469 | ||
| @@ -483,7 +483,7 @@ HcclResult TransportRoceMem::Read(const RmaOpMem& localMem, const RmaOpMem& remo | |||
| 483 | CHK_PRT_RET( | 483 | CHK_PRT_RET( |
| 484 | (localMem.size == 0U) || (remoteMem.size == 0U), | 484 | (localMem.size == 0U) || (remoteMem.size == 0U), |
| 485 | HCCL_ERROR( | 485 | HCCL_ERROR( |
| 486 | - "[TransportRoceMem]localMem size[%llu] or remoteMem size[%llu]is invalid", localMem.size, remoteMem.size), | 486 | + "[TransportRoceMem]localMem size[%llu] or remoteMem size[%llu] is invalid", localMem.size, remoteMem.size), |
| 487 | HCCL_E_PARA); | 487 | HCCL_E_PARA); |
| 488 | CHK_PRT_RET(stream == nullptr, HCCL_ERROR("[TransportRoceMem]stream is invalid"), HCCL_E_PARA); | 488 | CHK_PRT_RET(stream == nullptr, HCCL_ERROR("[TransportRoceMem]stream is invalid"), HCCL_E_PARA); |
| 489 | 489 | ||
| @@ -541,7 +541,7 @@ HcclResult TransportRoceMem::GetMemInfo(u32& lkey, u32& rkey, HcclBuf& localMem, | |||
| 541 | CHK_PRT_RET( | 541 | CHK_PRT_RET( |
| 542 | (localMem.len == 0U) || (remoteMem.len == 0U), | 542 | (localMem.len == 0U) || (remoteMem.len == 0U), |
| 543 | HCCL_ERROR( | 543 | HCCL_ERROR( |
| 544 | - "[TransportRoceMem] localMem size[%llu] or remoteMem size[%llu]is invalid", localMem.len, remoteMem.len), | 544 | + "[TransportRoceMem] localMem size[%llu] or remoteMem size[%llu] is invalid", localMem.len, remoteMem.len), |
| 545 | HCCL_E_PARA); | 545 | HCCL_E_PARA); |
| 546 | 546 | ||
| 547 | RmaBufferSlice localRmaBufferSlice{}; | 547 | RmaBufferSlice localRmaBufferSlice{}; |
| @@ -821,7 +821,7 @@ HcclResult TransportRoceMem::RdmaDbSend(u32 dbindex, u64 dbinfo, const struct Se | |||
| 821 | ret != HCCL_SUCCESS, | 821 | ret != HCCL_SUCCESS, |
| 822 | HCCL_ERROR( | 822 | HCCL_ERROR( |
| 823 | "[RdmaDbSend]errNo[0x%016llx] rdma db send fail, " | 823 | "[RdmaDbSend]errNo[0x%016llx] rdma db send fail, " |
| 824 | - "return[%d]. para: dbindex[%u]dbinfo[%llu].", | 824 | + "return[%d]. para: dbindex[%u] dbinfo[%llu].", |
| 825 | HCCL_ERROR_CODE(HCCL_E_INTERNAL), ret, dbindex, dbinfo), | 825 | HCCL_ERROR_CODE(HCCL_E_INTERNAL), ret, dbindex, dbinfo), |
| 826 | HCCL_E_INTERNAL); | 826 | HCCL_E_INTERNAL); |
| 827 | return HCCL_SUCCESS; | 827 | return HCCL_SUCCESS; |
| @@ -702,7 +702,11 @@ HcclResult HcclBatchPutMRByAscendQP( | |||
| 702 | CHK_RET(CheckSendRecvLinkInfo(sendRecvLinkInfo)); | 702 | CHK_RET(CheckSendRecvLinkInfo(sendRecvLinkInfo)); |
| 703 | CHK_PRT_RET( | 703 | CHK_PRT_RET( |
| 704 | sendRecvLinkInfo->wqePerDoorbell == 0 || sendRecvLinkInfo->wqePerDoorbell > MAX_WQE_PER_DOORBELL, | 704 | sendRecvLinkInfo->wqePerDoorbell == 0 || sendRecvLinkInfo->wqePerDoorbell > MAX_WQE_PER_DOORBELL, |
| 705 | - HCCL_ERROR("[HcclBatchPutMRByAscendQP] The value of wqePerDoorbell is exceed 300 or equal to 0."), HCCL_E_PARA); | 705 | + HCCL_ERROR( |
| 706 | + "[HcclBatchPutMRByAscendQP] wqePerDoorbell[%u] is invalid, should be greater than 0 and not exceed " | ||
| 707 | + "MAX_WQE_PER_DOORBELL[%u].", | ||
| 708 | + sendRecvLinkInfo->wqePerDoorbell, MAX_WQE_PER_DOORBELL), | ||
| 709 | + HCCL_E_PARA); | ||
| 706 | QpHandle qpHandle; | 710 | QpHandle qpHandle; |
| 707 | CHK_RET(TypicalQpManager::GetInstance().GetQpHandleByQpn(sendRecvLinkInfo->localQPinfo->qpn, qpHandle)); | 711 | CHK_RET(TypicalQpManager::GetInstance().GetQpHandleByQpn(sendRecvLinkInfo->localQPinfo->qpn, qpHandle)); |
| 708 | 712 | ||
| @@ -792,7 +796,10 @@ HcclResult HcclOneSideBatchPutByAscendQP( | |||
| 792 | CHK_RET(CheckSendLinkInfo(sendlinkInfo)); | 796 | CHK_RET(CheckSendLinkInfo(sendlinkInfo)); |
| 793 | CHK_PRT_RET( | 797 | CHK_PRT_RET( |
| 794 | sendlinkInfo->wqePerDoorbell == 0 || sendlinkInfo->wqePerDoorbell > MAX_WQE_PER_DOORBELL, | 798 | sendlinkInfo->wqePerDoorbell == 0 || sendlinkInfo->wqePerDoorbell > MAX_WQE_PER_DOORBELL, |
| 795 | - HCCL_ERROR("[HcclOneSideBatchPutByAscendQP] The value of wqePerDoorbell is exceed 300 or equal to 0."), | 799 | + HCCL_ERROR( |
| 800 | + "[HcclOneSideBatchPutByAscendQP] wqePerDoorbell[%u] is invalid, should be greater than 0 and not exceed " | ||
| 801 | + "MAX_WQE_PER_DOORBELL[%u].", | ||
| 802 | + sendlinkInfo->wqePerDoorbell, MAX_WQE_PER_DOORBELL), | ||
| 796 | HCCL_E_PARA); | 803 | HCCL_E_PARA); |
| 797 | QpHandle qpHandle; | 804 | QpHandle qpHandle; |
| 798 | CHK_RET(TypicalQpManager::GetInstance().GetQpHandleByQpn(sendlinkInfo->localQPinfo->qpn, qpHandle)); | 805 | CHK_RET(TypicalQpManager::GetInstance().GetQpHandleByQpn(sendlinkInfo->localQPinfo->qpn, qpHandle)); |
| @@ -30,7 +30,7 @@ static bool CreateCcuJettys(CcuCommunicator& ccuComm, const std::vector<LinkData | |||
| 30 | createStatus = false; // 预留处理资源不足回退情况,当前不支持回退 | 30 | createStatus = false; // 预留处理资源不足回退情况,当前不支持回退 |
| 31 | HCCL_WARNING( | 31 | HCCL_WARNING( |
| 32 | "[CcuInsPreprocessor][%s] create ccu jettys failed, " | 32 | "[CcuInsPreprocessor][%s] create ccu jettys failed, " |
| 33 | - "ccu resource is unavaialble, please check.", | 33 | + "ccu resource is unavailable, please check.", |
| 34 | __func__); | 34 | __func__); |
| 35 | return false; | 35 | return false; |
| 36 | } | 36 | } |
| @@ -58,7 +58,7 @@ static bool CreateCcuTransports( | |||
| 58 | createStatus = false; // 预留处理资源不足回退情况,当前不支持回退 | 58 | createStatus = false; // 预留处理资源不足回退情况,当前不支持回退 |
| 59 | HCCL_WARNING( | 59 | HCCL_WARNING( |
| 60 | "[CcuInsPreprocessor][%s] create ccu transports failed, " | 60 | "[CcuInsPreprocessor][%s] create ccu transports failed, " |
| 61 | - "ccu resource is unavaialble, please check.", | 61 | + "ccu resource is unavailable, please check.", |
| 62 | __func__); | 62 | __func__); |
| 63 | return false; | 63 | return false; |
| 64 | } | 64 | } |
| @@ -160,7 +160,7 @@ void CcuInsPreprocessor::CreateCcuCtxGroup( | |||
| 160 | bool CcuInsPreprocessor::CheckCtxTransportStatus(bool resAllocSuccess) | 160 | bool CcuInsPreprocessor::CheckCtxTransportStatus(bool resAllocSuccess) |
| 161 | { | 161 | { |
| 162 | if (!resAllocSuccess) { | 162 | if (!resAllocSuccess) { |
| 163 | - HCCL_INFO( | 163 | + HCCL_WARNING( |
| 164 | "[CcuInsPreprocessor::%s] CreateCcuCtx alloc local resource fail, ccuCtxGroups" | 164 | "[CcuInsPreprocessor::%s] CreateCcuCtx alloc local resource fail, ccuCtxGroups" |
| 165 | " size[%zu], resPackIdxs size[%zu], ctxSignatures size[%zu], insPtrs size[%zu]", | 165 | " size[%zu], resPackIdxs size[%zu], ctxSignatures size[%zu], insPtrs size[%zu]", |
| 166 | __func__, ccuCtxGroups.size(), resPackIdxs.size(), ctxSignatures.size(), insPtrs.size()); | 166 | __func__, ccuCtxGroups.size(), resPackIdxs.size(), ctxSignatures.size(), insPtrs.size()); |
| @@ -218,7 +218,7 @@ void CcuInsPreprocessor::InsPreprocess(InsIterator& insIter, u32 resPackIndex, b | |||
| 218 | if (res != HcclResult::HCCL_E_UNAVAIL) { | 218 | if (res != HcclResult::HCCL_E_UNAVAIL) { |
| 219 | THROW<InternalException>("[CcuCtxMgr::AllocRes]AllocRes failed, unexpected error, please check."); | 219 | THROW<InternalException>("[CcuCtxMgr::AllocRes]AllocRes failed, unexpected error, please check."); |
| 220 | } | 220 | } |
| 221 | - HCCL_INFO("[CcuInsPreprocessor::%s] AllocRes failed, ret[%u]", __func__, res); | 221 | + HCCL_WARNING("[CcuInsPreprocessor::%s] AllocRes failed, ret[%u]", __func__, res); |
| 222 | resAllocSuccess = false; | 222 | resAllocSuccess = false; |
| 223 | } | 223 | } |
| 224 | } else { | 224 | } else { |
| @@ -244,7 +244,7 @@ void CcuInsPreprocessor::PrepareCcuCtx(std::shared_ptr<InsQueue>& insQueue, bool | |||
| 244 | } | 244 | } |
| 245 | InsPreprocess(ins, resPackIndex, isMc2); | 245 | InsPreprocess(ins, resPackIndex, isMc2); |
| 246 | if (needHandShake && !resAllocSuccess) { | 246 | if (needHandShake && !resAllocSuccess) { |
| 247 | - HCCL_INFO( | 247 | + HCCL_WARNING( |
| 248 | "[CcuInsPreprocessor::%s] slave insQueue ins alloc local resource fail, " | 248 | "[CcuInsPreprocessor::%s] slave insQueue ins alloc local resource fail, " |
| 249 | "resPackIndex[%u], ins[%s].", | 249 | "resPackIndex[%u], ins[%s].", |
| 250 | __func__, resPackIndex, ins->Describe().c_str()); | 250 | __func__, resPackIndex, ins->Describe().c_str()); |
| @@ -264,7 +264,7 @@ void CcuInsPreprocessor::PrepareCcuCtx(std::shared_ptr<InsQueue>& insQueue, bool | |||
| 264 | } | 264 | } |
| 265 | InsPreprocess(ins, resPackIndex, isMc2); | 265 | InsPreprocess(ins, resPackIndex, isMc2); |
| 266 | if (needHandShake && !resAllocSuccess) { | 266 | if (needHandShake && !resAllocSuccess) { |
| 267 | - HCCL_INFO( | 267 | + HCCL_WARNING( |
| 268 | "[CcuInsPreprocessor::%s] master insQueue ins alloc local resource fail, " | 268 | "[CcuInsPreprocessor::%s] master insQueue ins alloc local resource fail, " |
| 269 | "resPackIndex[%u], ins[%s].", | 269 | "resPackIndex[%u], ins[%s].", |
| 270 | __func__, resPackIndex, ins->Describe().c_str()); | 270 | __func__, resPackIndex, ins->Describe().c_str()); |
| @@ -57,7 +57,7 @@ HcclResult CcuJettyMgr::PrepareCreate(const std::vector<LinkData>& links) | |||
| 57 | ret == HcclResult::HCCL_E_UNAVAIL, | 57 | ret == HcclResult::HCCL_E_UNAVAIL, |
| 58 | HCCL_WARNING( | 58 | HCCL_WARNING( |
| 59 | "[CcuJettyMgr][%s] failed to alloc ccu channels, ccu resources " | 59 | "[CcuJettyMgr][%s] failed to alloc ccu channels, ccu resources " |
| 60 | - "are unavaialble, locAddr[%s], devLogicId[%d].", | 60 | + "are unavailable, locAddr[%s], devLogicId[%d].", |
| 61 | __func__, locAddr.Describe().c_str(), devLogicId_), | 61 | __func__, locAddr.Describe().c_str(), devLogicId_), |
| 62 | ret); | 62 | ret); |
| 63 | CHK_RET(ret); | 63 | CHK_RET(ret); |
| @@ -100,7 +100,7 @@ HcclResult CcuJettyMgr::GetAvailableBatch(const BatchKey& batchKey, ResourceBatc | |||
| 100 | ret == HcclResult::HCCL_E_UNAVAIL, | 100 | ret == HcclResult::HCCL_E_UNAVAIL, |
| 101 | HCCL_WARNING( | 101 | HCCL_WARNING( |
| 102 | "[CcuJettyMgr][%s] failed to alloc ccu channels, ccu resources " | 102 | "[CcuJettyMgr][%s] failed to alloc ccu channels, ccu resources " |
| 103 | - "are unavaialble, locAddr[%s], sqSize[%u], devLogicId[%d].", | 103 | + "are unavailable, locAddr[%s], sqSize[%u], devLogicId[%d].", |
| 104 | __func__, batchKey.Describe().c_str(), sqSize, devLogicId_), | 104 | __func__, batchKey.Describe().c_str(), sqSize, devLogicId_), |
| 105 | ret); | 105 | ret); |
| 106 | // 其他错误直接返回错误码 | 106 | // 其他错误直接返回错误码 |
| @@ -108,7 +108,7 @@ HcclResult CcuJettyMgr::GetAvailableBatch(const BatchKey& batchKey, ResourceBatc | |||
| 108 | ret != HCCL_SUCCESS, | 108 | ret != HCCL_SUCCESS, |
| 109 | HCCL_ERROR( | 109 | HCCL_ERROR( |
| 110 | "[CcuJettyMgr][%s] failed to alloc ccu channels, ccu resources " | 110 | "[CcuJettyMgr][%s] failed to alloc ccu channels, ccu resources " |
| 111 | - "are unavaialble, locAddr[%s], sqSize[%u], devLogicId[%d].", | 111 | + "are unavailable, locAddr[%s], sqSize[%u], devLogicId[%d].", |
| 112 | __func__, batchKey.Describe().c_str(), sqSize, devLogicId_), | 112 | __func__, batchKey.Describe().c_str(), sqSize, devLogicId_), |
| 113 | ret); | 113 | ret); |
| 114 | // 如果新增资源保存失败,手动释放避免泄露 | 114 | // 如果新增资源保存失败,手动释放避免泄露 |
| @@ -818,13 +818,13 @@ std::vector<CcuTaskParam> Mc2Compont::GetAlgoCcuTaskInfo(InsExeQue::ExtInsExeEnt | |||
| 818 | std::vector<CcuTaskParam> ccuTaskParam{}; | 818 | std::vector<CcuTaskParam> ccuTaskParam{}; |
| 819 | auto serverItor = ccuServerMap.find(execId); | 819 | auto serverItor = ccuServerMap.find(execId); |
| 820 | if (serverItor == ccuServerMap.end()) { | 820 | if (serverItor == ccuServerMap.end()) { |
| 821 | - HCCL_INFO("[Mc2Compont]Failed to find ccuServer by executeId[%llu]", execId); | 821 | + HCCL_WARNING("[Mc2Compont]Failed to find ccuServer by executeId[%llu]", execId); |
| 822 | return ccuTaskParam; | 822 | return ccuTaskParam; |
| 823 | } | 823 | } |
| 824 | for (uint64_t algoSign : serverItor->second) { | 824 | for (uint64_t algoSign : serverItor->second) { |
| 825 | auto algoTemplateItor = algoTemplateMap.find(algoSign); | 825 | auto algoTemplateItor = algoTemplateMap.find(algoSign); |
| 826 | if (algoTemplateItor == algoTemplateMap.end()) { | 826 | if (algoTemplateItor == algoTemplateMap.end()) { |
| 827 | - HCCL_INFO("[Mc2Compont]Failed to find ccuTaskParam by algoSign[%llu]", algoSign); | 827 | + HCCL_WARNING("[Mc2Compont]Failed to find ccuTaskParam by algoSign[%llu]", algoSign); |
| 828 | continue; | 828 | continue; |
| 829 | } | 829 | } |
| 830 | for (const auto& taskParam : algoTemplateItor->second) { | 830 | for (const auto& taskParam : algoTemplateItor->second) { |
| @@ -116,7 +116,7 @@ int AicpuUtils::GetException( | |||
| 116 | } | 116 | } |
| 117 | uint32_t reportNum = recvInfo->report_cqe_num; | 117 | uint32_t reportNum = recvInfo->report_cqe_num; |
| 118 | if (reportNum > MAX_REPORT_CNT) { | 118 | if (reportNum > MAX_REPORT_CNT) { |
| 119 | - HCCL_WARNING("[%s]report cqe num %u should not big than %u", typeStr, reportNum, MAX_REPORT_CNT); | 119 | + HCCL_WARNING("[%s]report cqe num %u should not be bigger than %u", typeStr, reportNum, MAX_REPORT_CNT); |
| 120 | return 1; | 120 | return 1; |
| 121 | } | 121 | } |
| 122 | 122 | ||
| @@ -191,7 +191,8 @@ void TaskExceptionFunc::Call() | |||
| 191 | uint32_t reportNum = recvInfo->report_cqe_num; | 191 | uint32_t reportNum = recvInfo->report_cqe_num; |
| 192 | if (reportNum > MAX_REPORT_CNT) { | 192 | if (reportNum > MAX_REPORT_CNT) { |
| 193 | HCCL_ERROR( | 193 | HCCL_ERROR( |
| 194 | - "[TaskExceptionFunc]report cqe num %u should not big than %u", reportNum, MAX_REPORT_CNT); | 194 | + "[TaskExceptionFunc]report cqe num %u should not be greater than %u", reportNum, |
| 195 | + MAX_REPORT_CNT); | ||
| 195 | continue; | 196 | continue; |
| 196 | } | 197 | } |
| 197 | for (uint32_t idx = 0U; idx < reportNum; ++idx) { | 198 | for (uint32_t idx = 0U; idx < reportNum; ++idx) { |
| @@ -67,7 +67,7 @@ HcclResult SnapShotParser::ParseSnapshotToLocalBuff(void* snapshotBuf, uint32_t | |||
| 67 | return HCCL_E_INTERNAL; | 67 | return HCCL_E_INTERNAL; |
| 68 | }; | 68 | }; |
| 69 | HCCL_INFO( | 69 | HCCL_INFO( |
| 70 | - "[%s] end, napshotBuf is:subGroup num[%u], comm groupName[%s]", __func__, localBuff.groupNum, | 70 | + "[%s] end, snapshotBuf is:subGroup num[%u], comm groupName[%s]", __func__, localBuff.groupNum, |
| 71 | localBuff.snapshot.groupName); | 71 | localBuff.snapshot.groupName); |
| 72 | return HCCL_SUCCESS; | 72 | return HCCL_SUCCESS; |
| 73 | } | 73 | } |
| @@ -405,14 +405,14 @@ HcclResult SnapShotParser::DeserializeParamsInfo(BinaryStream& binaryStream, Hcc | |||
| 405 | 405 | ||
| 406 | HcclResult SnapShotParser::DeserializeRankTableInfo(BinaryStream& binaryStream, RankTableInfo& rankTableInfo) const | 406 | HcclResult SnapShotParser::DeserializeRankTableInfo(BinaryStream& binaryStream, RankTableInfo& rankTableInfo) const |
| 407 | { | 407 | { |
| 408 | - HCCL_INFO("[%s]Snapshot recovering: Start to dserialized rankTable info.", __func__); | 408 | + HCCL_INFO("[%s]Snapshot recovering: Start to deserialize rankTable info.", __func__); |
| 409 | rankTableInfo = RankTableInfo(binaryStream); | 409 | rankTableInfo = RankTableInfo(binaryStream); |
| 410 | return HCCL_SUCCESS; | 410 | return HCCL_SUCCESS; |
| 411 | } | 411 | } |
| 412 | 412 | ||
| 413 | HcclResult SnapShotParser::DeserializeTopoInfo(BinaryStream& binaryStream, TopoInfo& topoInfo) const | 413 | HcclResult SnapShotParser::DeserializeTopoInfo(BinaryStream& binaryStream, TopoInfo& topoInfo) const |
| 414 | { | 414 | { |
| 415 | - HCCL_INFO("[%s]Snapshot recovering: Start to dserialized topo info.", __func__); | 415 | + HCCL_INFO("[%s]Snapshot recovering: Start to deserialize topo info.", __func__); |
| 416 | topoInfo = TopoInfo(binaryStream); | 416 | topoInfo = TopoInfo(binaryStream); |
| 417 | return HCCL_SUCCESS; | 417 | return HCCL_SUCCESS; |
| 418 | } | 418 | } |
| @@ -421,7 +421,7 @@ void SnapShotParser::SerializeSubCommInfo( | |||
| 421 | const CommParams& commParams, const HcclCommConfig& subConfig, const std::vector<u32>& rankId, | 421 | const CommParams& commParams, const HcclCommConfig& subConfig, const std::vector<u32>& rankId, |
| 422 | BinaryStream& binStream) const | 422 | BinaryStream& binStream) const |
| 423 | { | 423 | { |
| 424 | - HCCL_INFO("[%s]Snapshot saving: Start to serial subComm static info.", __func__); | 424 | + HCCL_INFO("[%s]Snapshot saving: Start to serialize subComm static info.", __func__); |
| 425 | SerializeSubCommParamsInfo(commParams, binStream); | 425 | SerializeSubCommParamsInfo(commParams, binStream); |
| 426 | SerializeSubCommConfigInfo(subConfig, binStream); | 426 | SerializeSubCommConfigInfo(subConfig, binStream); |
| 427 | SerializeRankIds(rankId, binStream); | 427 | SerializeRankIds(rankId, binStream); |
| @@ -471,7 +471,7 @@ HcclResult SnapShotParser::DeserializeSubCommInfo(BinaryStream& stream, SubSnaps | |||
| 471 | 471 | ||
| 472 | HcclResult SnapShotParser::DeserializeSubCommConfigInfo(BinaryStream& binaryStream, HcclCommConfig& subConfig) const | 472 | HcclResult SnapShotParser::DeserializeSubCommConfigInfo(BinaryStream& binaryStream, HcclCommConfig& subConfig) const |
| 473 | { | 473 | { |
| 474 | - HCCL_INFO("[%s]Snapshot recovering: Start to deserial sub commConfig info.", __func__); | 474 | + HCCL_INFO("[%s]Snapshot recovering: Start to deserialize sub commConfig info.", __func__); |
| 475 | binaryStream >> subConfig.reserved >> subConfig.hcclBufferSize >> subConfig.hcclDeterministic | 475 | binaryStream >> subConfig.reserved >> subConfig.hcclBufferSize >> subConfig.hcclDeterministic |
| 476 | >> subConfig.hcclCommName >> subConfig.hcclUdi; | 476 | >> subConfig.hcclCommName >> subConfig.hcclUdi; |
| 477 | HCCL_INFO( | 477 | HCCL_INFO( |
| @@ -55,7 +55,7 @@ void PreemptPortManager::ListenPreempt( | |||
| 55 | { | 55 | { |
| 56 | CHK_PRT_RET( | 56 | CHK_PRT_RET( |
| 57 | !initialized, | 57 | !initialized, |
| 58 | - HCCL_ERROR("[PreemptPortManager::%s] preempt port manager has already been release.", __func__), ); | 58 | + HCCL_ERROR("[PreemptPortManager::%s] preempt port manager has already been released.", __func__), ); |
| 59 | 59 | ||
| 60 | CHK_SMART_PTR_RET_NULL(listenSocket); | 60 | CHK_SMART_PTR_RET_NULL(listenSocket); |
| 61 | NicType nicType = listenSocket->GetNicType(); | 61 | NicType nicType = listenSocket->GetNicType(); |
| @@ -69,7 +69,7 @@ void PreemptPortManager::Release(const std::shared_ptr<Socket>& listenSocket) | |||
| 69 | { | 69 | { |
| 70 | CHK_PRT_RET( | 70 | CHK_PRT_RET( |
| 71 | !initialized, | 71 | !initialized, |
| 72 | - HCCL_WARNING("[PreemptPortManager::%s] preempt port manager has already been release.", __func__), ); | 72 | + HCCL_WARNING("[PreemptPortManager::%s] preempt port manager has already been released.", __func__), ); |
| 73 | 73 | ||
| 74 | CHK_SMART_PTR_RET_NULL(listenSocket); | 74 | CHK_SMART_PTR_RET_NULL(listenSocket); |
| 75 | NicType nicType = listenSocket->GetNicType(); | 75 | NicType nicType = listenSocket->GetNicType(); |
| @@ -136,7 +136,7 @@ void PreemptPortManager::PreemptPortInRange( | |||
| 136 | = nicType == NicType::HOST_NIC_TYPE ? "HCCL_HOST_SOCKET_PORT_RANGE" : "HCCL_NPU_SOCKET_PORT_RANGE"; | 136 | = nicType == NicType::HOST_NIC_TYPE ? "HCCL_HOST_SOCKET_PORT_RANGE" : "HCCL_NPU_SOCKET_PORT_RANGE"; |
| 137 | HCCL_ERROR( | 137 | HCCL_ERROR( |
| 138 | "NOTICE: Users need to make sure ports in %s are available for HCCL." | 138 | "NOTICE: Users need to make sure ports in %s are available for HCCL." |
| 139 | - "Please double check whether the port are used by others unexpected process. " | 139 | + "Please double check whether the ports are used by other unexpected processes. " |
| 140 | "The port ranges size should also be enough when running multi-process HCCL.", | 140 | "The port ranges size should also be enough when running multi-process HCCL.", |
| 141 | envName.c_str()); | 141 | envName.c_str()); |
| 142 | HCCL_ERROR("NOTICE: The host port range size is not suggested to be smaller than the process number" | 142 | HCCL_ERROR("NOTICE: The host port range size is not suggested to be smaller than the process number" |
| @@ -166,7 +166,7 @@ void PreemptPortManager::ReleasePreempt( | |||
| 166 | HCCL_ERROR( | 166 | HCCL_ERROR( |
| 167 | "[PreemptPortManager::%s] ref[%d], ip[%s] port[%u] has already been released.", __func__, ref.Count(), | 167 | "[PreemptPortManager::%s] ref[%d], ip[%s] port[%u] has already been released.", __func__, ref.Count(), |
| 168 | ipAddr.c_str(), port), | 168 | ipAddr.c_str(), port), |
| 169 | - InvalidParamsException, "socket port dulplicate release"); | 169 | + InvalidParamsException, "socket port duplicate release"); |
| 170 | 170 | ||
| 171 | // 释放绑定端口的Socket | 171 | // 释放绑定端口的Socket |
| 172 | listenSocket->StopListen(); | 172 | listenSocket->StopListen(); |
| @@ -62,7 +62,7 @@ CcuContextAllGatherMesh2D::CcuContextAllGatherMesh2D( | |||
| 62 | localAxisSignalName_ = "CcuContextAllGatherMesh2DAxisSync_" + std::to_string(axisId_); | 62 | localAxisSignalName_ = "CcuContextAllGatherMesh2DAxisSync_" + std::to_string(axisId_); |
| 63 | anotherAxisSignalName_ = "CcuContextAllGatherMesh2DAxisSync_" + std::to_string(1 - axisId_); | 63 | anotherAxisSignalName_ = "CcuContextAllGatherMesh2DAxisSync_" + std::to_string(1 - axisId_); |
| 64 | HCCL_INFO( | 64 | HCCL_INFO( |
| 65 | - "[CcuContextAllGatherMesh2D] RankId[%u], DimSize: D0[%u]--D1[%u], localId[%u], lcoalSize[%u]", rankId_, | 65 | + "[CcuContextAllGatherMesh2D] RankId[%u], DimSize: D0[%u]--D1[%u], localId[%u], localSize[%u]", rankId_, |
| 66 | dimSize_[0], dimSize_[1], localId_, localSize_); | 66 | dimSize_[0], dimSize_[1], localId_, localSize_); |
| 67 | } | 67 | } |
| 68 | 68 | ||
| @@ -196,7 +196,7 @@ void CcuContextAllGatherMesh2D::FirstStep() | |||
| 196 | 196 | ||
| 197 | void CcuContextAllGatherMesh2D::SecondStep() | 197 | void CcuContextAllGatherMesh2D::SecondStep() |
| 198 | { | 198 | { |
| 199 | - HCCL_INFO("[CcuContextAllGatherMesh2D] secodeStep run begins"); | 199 | + HCCL_INFO("[CcuContextAllGatherMesh2D] secondStep run begins"); |
| 200 | uint64_t anotherSize = dimSize_[1 - axisId_]; | 200 | uint64_t anotherSize = dimSize_[1 - axisId_]; |
| 201 | 201 | ||
| 202 | CcuRep::Memory src = CreateMemory(); | 202 | CcuRep::Memory src = CreateMemory(); |
| @@ -319,7 +319,7 @@ std::vector<uint64_t> CcuContextAllGatherNHR1D::GeneArgs(const CcuTaskArg& arg) | |||
| 319 | 319 | ||
| 320 | HCCL_INFO( | 320 | HCCL_INFO( |
| 321 | "[CcuContextAllGatherNHR1D] TaskArgs: inputAddr[%llu], outputAddr[%llu], " | 321 | "[CcuContextAllGatherNHR1D] TaskArgs: inputAddr[%llu], outputAddr[%llu], " |
| 322 | - "die0Size[%llu], die1Size[%llu], repeatNum[%llu]" | 322 | + "die0Size[%llu], die1Size[%llu], repeatNum[%llu] " |
| 323 | "inputSliceStride[%llu], outputSliceStride[%llu], inputRepeatStride[%llu], outputRepeatStride[%llu]", | 323 | "inputSliceStride[%llu], outputSliceStride[%llu], inputRepeatStride[%llu], outputRepeatStride[%llu]", |
| 324 | inputAddr, outputAddr, die0Size, die1Size, repeatNum, inputSliceStride, outputSliceStride, inputRepeatStride, | 324 | inputAddr, outputAddr, die0Size, die1Size, repeatNum, inputSliceStride, outputSliceStride, inputRepeatStride, |
| 325 | outputRepeatStride); | 325 | outputRepeatStride); |


[建议] 已改用
log_warning后,请去掉文案里的[warning]前缀(级别由 logger 表达即可)。本文件同类写法共 4 处(约 L226/L233/L323/L419),建议一并去掉,例如:
common.log_warning("remote repo is empty, doing full sync")而不是
log_warning("[warning] ...")。