已合并
[fix]: clean up wording and typos #4864
Liang Xuheng创建于 8月26日
[fix]: clean up wording and typos #4864
已合并
共 97 个文件变更+299-289
| @@ -97,7 +97,7 @@ HcclResult RoceRegedMemMgr::UnregisterMemory(void* memHandle) | |||
| 97 | EXCEPTION_CATCH(resultPair = this->localRdmaRmaBufferMgr_->Del(tempKey), return HCCL_E_NOT_FOUND); | 97 | EXCEPTION_CATCH(resultPair = this->localRdmaRmaBufferMgr_->Del(tempKey), return HCCL_E_NOT_FOUND); |
| 98 | // 计数器大于1时,返回false,说明框架层有其它设备在使用这段内存,返回HCCL_E_AGAIN | 98 | // 计数器大于1时,返回false,说明框架层有其它设备在使用这段内存,返回HCCL_E_AGAIN |
| 99 | if (!resultPair) { | 99 | if (!resultPair) { |
| 100 | - HCCL_INFO("[RoceRegedMemMgr][[UnregisterMemory] Memory reference count is larger than 0" | 100 | + HCCL_INFO("[RoceRegedMemMgr][[UnregisterMemory] Memory reference count is larger than 0 " |
| 101 | "(used by other RemoteRank), do not deregister memory."); | 101 | "(used by other RemoteRank), do not deregister memory."); |
| 102 | return HCCL_SUCCESS; | 102 | return HCCL_SUCCESS; |
| 103 | } | 103 | } |
| @@ -247,7 +247,7 @@ HcclResult RoceRegedMemMgr::MemoryUnimport(const void* memDesc, uint32_t descLen | |||
| 247 | EXCEPTION_CATCH(resultPair = remoteRdmaRmaBufferMgrs_[endpointDesc]->Del(tempKey), return HCCL_E_NOT_FOUND); | 247 | EXCEPTION_CATCH(resultPair = remoteRdmaRmaBufferMgrs_[endpointDesc]->Del(tempKey), return HCCL_E_NOT_FOUND); |
| 248 | // 计数器大于1时,返回false,说明框架层有其它设备在使用这段内存,返回HCCL_E_AGAIN | 248 | // 计数器大于1时,返回false,说明框架层有其它设备在使用这段内存,返回HCCL_E_AGAIN |
| 249 | if (!resultPair) { | 249 | if (!resultPair) { |
| 250 | - HCCL_INFO("[RoceRegedMemMgr][[MemoryUnimport] Memory reference count is larger than 0" | 250 | + HCCL_INFO("[RoceRegedMemMgr][[MemoryUnimport] Memory reference count is larger than 0 " |
| 251 | "(used by other RemoteRank)."); | 251 | "(used by other RemoteRank)."); |
| 252 | return HCCL_E_AGAIN; | 252 | return HCCL_E_AGAIN; |
| 253 | } | 253 | } |
| @@ -102,7 +102,7 @@ HcclResult UbRegedMemMgr::UnregisterMemory(void* memHandle) | |||
| 102 | EXCEPTION_CATCH(resultPair = this->localUbRmaBufferMgr_->Del(tempKey), return HCCL_E_NOT_FOUND); | 102 | EXCEPTION_CATCH(resultPair = this->localUbRmaBufferMgr_->Del(tempKey), return HCCL_E_NOT_FOUND); |
| 103 | // 计数器大于1时,返回false,说明框架层有其它设备在使用这段内存,返回HCCL_E_AGAIN | 103 | // 计数器大于1时,返回false,说明框架层有其它设备在使用这段内存,返回HCCL_E_AGAIN |
| 104 | if (!resultPair) { | 104 | if (!resultPair) { |
| 105 | - HCCL_INFO("[UbRegedMemMgr][[UnregisterMemory] Memory reference count is larger than 0" | 105 | + HCCL_INFO("[UbRegedMemMgr][[UnregisterMemory] Memory reference count is larger than 0 " |
| 106 | "(used by other RemoteRank), do not deregister memory."); | 106 | "(used by other RemoteRank), do not deregister memory."); |
| 107 | return HCCL_SUCCESS; | 107 | return HCCL_SUCCESS; |
| 108 | } | 108 | } |
| @@ -364,7 +364,7 @@ HcommResult HcommCollectiveChannelCreate( | |||
| 364 | CHK_PTR_NULL(channelDescs); | 364 | CHK_PTR_NULL(channelDescs); |
| 365 | CHK_PTR_NULL(channels); | 365 | CHK_PTR_NULL(channels); |
| 366 | CHK_PRT_RET( | 366 | CHK_PRT_RET( |
| 367 | - (channelNum == 0), HCCL_ERROR("[%s]Invalid channelNum, channelNum[%u]", __func__, channelNum), HCCL_E_PARA); | 367 | + (channelNum == 0), HCCL_ERROR("[%s] Invalid channelNum, channelNum[%u]", __func__, channelNum), HCCL_E_PARA); |
| 368 | std::vector<HcommChannelDesc> channelDescFinals; | 368 | std::vector<HcommChannelDesc> channelDescFinals; |
| 369 | CHK_RET(static_cast<HcclResult>(NormalizeHcommChannelDescs(channelDescs, channelNum, channelDescFinals, engine))); | 369 | CHK_RET(static_cast<HcclResult>(NormalizeHcommChannelDescs(channelDescs, channelNum, channelDescFinals, engine))); |
| 370 | auto startut = std::chrono::steady_clock::now(); | 370 | auto startut = std::chrono::steady_clock::now(); |
| @@ -382,7 +382,8 @@ HcommResult HcommCollectiveChannelCreate( | |||
| 382 | HcommResult HcommChannelUpdateMemInfo(HcommMemHandle* memHandles, uint32_t memHandleNum, ChannelHandle channelHandle) | 382 | HcommResult HcommChannelUpdateMemInfo(HcommMemHandle* memHandles, uint32_t memHandleNum, ChannelHandle channelHandle) |
| 383 | { | 383 | { |
| 384 | CHK_PTR_NULL(memHandles); | 384 | CHK_PTR_NULL(memHandles); |
| 385 | - CHK_PRT_RET((memHandleNum == 0), HCCL_ERROR("[%s]Invalid memHandleNum, memHandleNum is 0.", __func__), HCCL_E_PARA); | 385 | + CHK_PRT_RET( |
| 386 | + (memHandleNum == 0), HCCL_ERROR("[%s] Invalid memHandleNum, memHandleNum is 0.", __func__), HCCL_E_PARA); | ||
| 386 | return ChannelProcess::ChannelUpdateMemInfo(memHandles, memHandleNum, channelHandle); | 387 | return ChannelProcess::ChannelUpdateMemInfo(memHandles, memHandleNum, channelHandle); |
| 387 | } | 388 | } |
| 388 | 389 | ||
| @@ -414,7 +415,7 @@ HcommResult HcommChannelCreate( | |||
| 414 | CHK_PTR_NULL(channelDescs); | 415 | CHK_PTR_NULL(channelDescs); |
| 415 | CHK_PTR_NULL(channels); | 416 | CHK_PTR_NULL(channels); |
| 416 | CHK_PRT_RET( | 417 | CHK_PRT_RET( |
| 417 | - (channelNum == 0), HCCL_ERROR("[%s]Invalid channelNum, channelNum[%u]", __func__, channelNum), HCCL_E_PARA); | 418 | + (channelNum == 0), HCCL_ERROR("[%s] Invalid channelNum, channelNum[%u]", __func__, channelNum), HCCL_E_PARA); |
| 418 | std::vector<HcommChannelDesc> channelDescFinals; | 419 | std::vector<HcommChannelDesc> channelDescFinals; |
| 419 | CHK_RET(static_cast<HcclResult>(NormalizeHcommChannelDescs(channelDescs, channelNum, channelDescFinals, engine))); | 420 | CHK_RET(static_cast<HcclResult>(NormalizeHcommChannelDescs(channelDescs, channelNum, channelDescFinals, engine))); |
| 420 | auto endpoint = GetEndpointMap().GetEndpoint(endpointHandle); | 421 | auto endpoint = GetEndpointMap().GetEndpoint(endpointHandle); |
| @@ -459,7 +460,7 @@ HcommResult HcommChannelGetStatus(const ChannelHandle* channelList, uint32_t lis | |||
| 459 | { | 460 | { |
| 460 | CHK_PTR_NULL(channelList); | 461 | CHK_PTR_NULL(channelList); |
| 461 | CHK_PTR_NULL(statusList); | 462 | CHK_PTR_NULL(statusList); |
| 462 | - CHK_PRT_RET((listNum == 0), HCCL_ERROR("[%s]Invalid listNum, listNum[%u]", __func__, listNum), HCCL_E_PARA); | 463 | + CHK_PRT_RET((listNum == 0), HCCL_ERROR("[%s] Invalid listNum, listNum[%u]", __func__, listNum), HCCL_E_PARA); |
| 463 | 464 | ||
| 464 | if (IS_PLUGIN_HANDLE(channelList[0])) { | 465 | if (IS_PLUGIN_HANDLE(channelList[0])) { |
| 465 | for (uint32_t i = 0; i < listNum; i++) { | 466 | for (uint32_t i = 0; i < listNum; i++) { |
| @@ -529,7 +530,7 @@ HcommResult HcommChannelDestroy(const ChannelHandle* channels, uint32_t channelN | |||
| 529 | { | 530 | { |
| 530 | CHK_PTR_NULL(channels); | 531 | CHK_PTR_NULL(channels); |
| 531 | CHK_PRT_RET( | 532 | CHK_PRT_RET( |
| 532 | - (channelNum == 0), HCCL_ERROR("[%s]Invalid channelNum, channelNum[%u]", __func__, channelNum), HCCL_E_PARA); | 533 | + (channelNum == 0), HCCL_ERROR("[%s] Invalid channelNum, channelNum[%u]", __func__, channelNum), HCCL_E_PARA); |
| 533 | if (IS_PLUGIN_HANDLE(channels[0])) { | 534 | if (IS_PLUGIN_HANDLE(channels[0])) { |
| 534 | for (uint32_t idx = 0; idx < channelNum; ++idx) { | 535 | for (uint32_t idx = 0; idx < channelNum; ++idx) { |
| 535 | auto* ch = CHANNEL_FROM_HANDLE(channels[idx]); | 536 | auto* ch = CHANNEL_FROM_HANDLE(channels[idx]); |
| @@ -622,7 +623,7 @@ HcommResult HcommChannelCreateWithConfig( | |||
| 622 | CHK_PTR_NULL(channelDescs); | 623 | CHK_PTR_NULL(channelDescs); |
| 623 | CHK_PTR_NULL(channels); | 624 | CHK_PTR_NULL(channels); |
| 624 | CHK_PRT_RET( | 625 | CHK_PRT_RET( |
| 625 | - (channelNum == 0), HCCL_ERROR("[%s]Invalid channelNum, channelNum[%u]", __func__, channelNum), HCCL_E_PARA); | 626 | + (channelNum == 0), HCCL_ERROR("[%s] Invalid channelNum, channelNum[%u]", __func__, channelNum), HCCL_E_PARA); |
| 626 | HCCL_INFO( | 627 | HCCL_INFO( |
| 627 | "[%s] START. endpointHandle[0x%llx], engine[%s], channelNum[%u], config[%p].", __func__, endpointHandle, | 628 | "[%s] START. endpointHandle[0x%llx], engine[%s], channelNum[%u], config[%p].", __func__, endpointHandle, |
| 628 | GetEnumToString(GetCommEngineStatusStrMap(), engine).c_str(), channelNum, config); | 629 | GetEnumToString(GetCommEngineStatusStrMap(), engine).c_str(), channelNum, config); |
| @@ -67,7 +67,7 @@ HcommResult HcommEngineCtxCopy(CommEngine engine, void* dstCtx, const void* srcC | |||
| 67 | CHK_SAFETY_FUNC_RET(memcpy_s(reinterpret_cast<uint8_t*>(dstCtx), size, srcCtx, size)); | 67 | CHK_SAFETY_FUNC_RET(memcpy_s(reinterpret_cast<uint8_t*>(dstCtx), size, srcCtx, size)); |
| 68 | } else { | 68 | } else { |
| 69 | HCCL_ERROR( | 69 | HCCL_ERROR( |
| 70 | - "[%s]copy engine ctx failed, Unsupported engine[%s]", __func__, | 70 | + "[%s] copy engine ctx failed, Unsupported engine[%s]", __func__, |
| 71 | GetEnumToString(GetCommEngineStatusStrMap(), engine).c_str()); | 71 | GetEnumToString(GetCommEngineStatusStrMap(), engine).c_str()); |
| 72 | return HCCL_E_PARA; | 72 | return HCCL_E_PARA; |
| 73 | } | 73 | } |
| @@ -61,7 +61,7 @@ HcclResult CcuJettyCtxMgrV2::CheckCtxGroupsByFeId(const uint32_t feId) | |||
| 61 | jettyNum < jettyGroupSize || jettyGroupSize == 0, // fe策略不够分1个jetty组,认为资源不足 | 61 | jettyNum < jettyGroupSize || jettyGroupSize == 0, // fe策略不够分1个jetty组,认为资源不足 |
| 62 | HCCL_WARNING( | 62 | HCCL_WARNING( |
| 63 | "[CcuJettyCtxMgrV2][%s] failed, jettyNum[%u] of feId[%u] is too small to " | 63 | "[CcuJettyCtxMgrV2][%s] failed, jettyNum[%u] of feId[%u] is too small to " |
| 64 | - "allocated a jetty group, groupSize[%u], devLogicId[%d], dieId[%u].", | 64 | + "allocate a jetty group, groupSize[%u], devLogicId[%d], dieId[%u].", |
| 65 | __func__, jettyNum, feId, jettyGroupSize, devLogicId_, dieId_), | 65 | __func__, jettyNum, feId, jettyGroupSize, devLogicId_, dieId_), |
| 66 | HcclResult::HCCL_E_UNAVAIL); | 66 | HcclResult::HCCL_E_UNAVAIL); |
| 67 | 67 | ||
| @@ -124,7 +124,7 @@ HcclResult CcuJettyCtxMgrV2::Alloc( | |||
| 124 | uint32_t allocSqSize = sqSize; | 124 | uint32_t allocSqSize = sqSize; |
| 125 | if (allocSqSize != CCU_V2_FIXED_SQ_SIZE) { | 125 | if (allocSqSize != CCU_V2_FIXED_SQ_SIZE) { |
| 126 | HCCL_RUN_WARNING( | 126 | HCCL_RUN_WARNING( |
| 127 | - "[CcuJettyCtxMgr][%s] failed, sqSize is not equal 32, " | 127 | + "[CcuJettyCtxMgr][%s] failed, sqSize is not equal to 32, " |
| 128 | "sqSize is [%u]", | 128 | "sqSize is [%u]", |
| 129 | __func__, allocSqSize); | 129 | __func__, allocSqSize); |
| 130 | allocSqSize = CCU_V2_FIXED_SQ_SIZE; | 130 | allocSqSize = CCU_V2_FIXED_SQ_SIZE; |
| @@ -859,7 +859,7 @@ bool CcuIsInited(const int32_t deviceLogicId) | |||
| 859 | CHK_PRT_RET( | 859 | CHK_PRT_RET( |
| 860 | (deviceLogicId < 0 || static_cast<u32>(deviceLogicId) >= MAX_MODULE_DEVICE_NUM), | 860 | (deviceLogicId < 0 || static_cast<u32>(deviceLogicId) >= MAX_MODULE_DEVICE_NUM), |
| 861 | HCCL_ERROR( | 861 | HCCL_ERROR( |
| 862 | - "[CcuIsInited]deviceLogicId[%d] error, MAX_MODULE_DEVICE_NUM[%u]", deviceLogicId, MAX_MODULE_DEVICE_NUM), | 862 | + "[CcuIsInited] deviceLogicId[%d] error, MAX_MODULE_DEVICE_NUM[%u]", deviceLogicId, MAX_MODULE_DEVICE_NUM), |
| 863 | false); | 863 | false); |
| 864 | 864 | ||
| 865 | if (!CheckCcuOpenSourceEnable()) { | 865 | if (!CheckCcuOpenSourceEnable()) { |
| @@ -887,7 +887,8 @@ HcclResult CcuSetTaskKill(const int32_t deviceLogicId) | |||
| 887 | CHK_PRT_RET( | 887 | CHK_PRT_RET( |
| 888 | (deviceLogicId < 0 || static_cast<u32>(deviceLogicId) >= MAX_MODULE_DEVICE_NUM), | 888 | (deviceLogicId < 0 || static_cast<u32>(deviceLogicId) >= MAX_MODULE_DEVICE_NUM), |
| 889 | HCCL_ERROR( | 889 | HCCL_ERROR( |
| 890 | - "[CcuSetTaskKill]deviceLogicId[%d] error, MAX_MODULE_DEVICE_NUM[%u]", deviceLogicId, MAX_MODULE_DEVICE_NUM), | 890 | + "[CcuSetTaskKill] deviceLogicId[%d] error, MAX_MODULE_DEVICE_NUM[%u]", deviceLogicId, |
| 891 | + MAX_MODULE_DEVICE_NUM), | ||
| 891 | HcclResult::HCCL_E_PARA); | 892 | HcclResult::HCCL_E_PARA); |
| 892 | HcclResult ret; | 893 | HcclResult ret; |
| 893 | EXCEPTION_HANDLE_BEGIN | 894 | EXCEPTION_HANDLE_BEGIN |
| @@ -904,7 +905,7 @@ HcclResult CcuSetTaskKillDone(const int32_t deviceLogicId) | |||
| 904 | CHK_PRT_RET( | 905 | CHK_PRT_RET( |
| 905 | (deviceLogicId < 0 || static_cast<u32>(deviceLogicId) >= MAX_MODULE_DEVICE_NUM), | 906 | (deviceLogicId < 0 || static_cast<u32>(deviceLogicId) >= MAX_MODULE_DEVICE_NUM), |
| 906 | HCCL_ERROR( | 907 | HCCL_ERROR( |
| 907 | - "[CcuSetTaskKillDone]deviceLogicId[%d] error, MAX_MODULE_DEVICE_NUM[%u]", deviceLogicId, | 908 | + "[CcuSetTaskKillDone] deviceLogicId[%d] error, MAX_MODULE_DEVICE_NUM[%u]", deviceLogicId, |
| 908 | MAX_MODULE_DEVICE_NUM), | 909 | MAX_MODULE_DEVICE_NUM), |
| 909 | HcclResult::HCCL_E_PARA); | 910 | HcclResult::HCCL_E_PARA); |
| 910 | HcclResult ret; | 911 | HcclResult ret; |
| @@ -922,7 +923,7 @@ HcclResult CcuCleanTaskKillState(const int32_t deviceLogicId) | |||
| 922 | CHK_PRT_RET( | 923 | CHK_PRT_RET( |
| 923 | (deviceLogicId < 0 || static_cast<u32>(deviceLogicId) >= MAX_MODULE_DEVICE_NUM), | 924 | (deviceLogicId < 0 || static_cast<u32>(deviceLogicId) >= MAX_MODULE_DEVICE_NUM), |
| 924 | HCCL_ERROR( | 925 | HCCL_ERROR( |
| 925 | - "[CcuCleanTaskKillState]deviceLogicId[%d] error, MAX_MODULE_DEVICE_NUM[%u]", deviceLogicId, | 926 | + "[CcuCleanTaskKillState] deviceLogicId[%d] error, MAX_MODULE_DEVICE_NUM[%u]", deviceLogicId, |
| 926 | MAX_MODULE_DEVICE_NUM), | 927 | MAX_MODULE_DEVICE_NUM), |
| 927 | HcclResult::HCCL_E_PARA); | 928 | HcclResult::HCCL_E_PARA); |
| 928 | HcclResult ret; | 929 | HcclResult ret; |
| @@ -940,7 +941,7 @@ HcclResult CcuCleanDieCkes(const int32_t deviceLogicId, const uint8_t dieId) | |||
| 940 | CHK_PRT_RET( | 941 | CHK_PRT_RET( |
| 941 | (deviceLogicId < 0 || static_cast<u32>(deviceLogicId) >= MAX_MODULE_DEVICE_NUM), | 942 | (deviceLogicId < 0 || static_cast<u32>(deviceLogicId) >= MAX_MODULE_DEVICE_NUM), |
| 942 | HCCL_ERROR( | 943 | HCCL_ERROR( |
| 943 | - "[CcuCleanDieCkes]deviceLogicId[%d] error, MAX_MODULE_DEVICE_NUM[%u]", deviceLogicId, | 944 | + "[CcuCleanDieCkes] deviceLogicId[%d] error, MAX_MODULE_DEVICE_NUM[%u]", deviceLogicId, |
| 944 | MAX_MODULE_DEVICE_NUM), | 945 | MAX_MODULE_DEVICE_NUM), |
| 945 | HcclResult::HCCL_E_PARA); | 946 | HcclResult::HCCL_E_PARA); |
| 946 | HcclResult ret; | 947 | HcclResult ret; |
| @@ -135,7 +135,7 @@ namespace CcuRep { | |||
| 135 | ccuProfilingInfoCache.name = kernelName.c_str(); | 135 | ccuProfilingInfoCache.name = kernelName.c_str(); |
| 136 | ccuProfilingInfoCache.dieId = defaultDieId; | 136 | ccuProfilingInfoCache.dieId = defaultDieId; |
| 137 | HCCL_DEBUG( | 137 | HCCL_DEBUG( |
| 138 | - "[%s]type[%d], name[%s], deafultDieId[0]", __func__, ccuProfilingInfoCache.type, | 138 | + "[%s]type[%d], name[%s], defaultDieId[%u]", __func__, ccuProfilingInfoCache.type, |
| 139 | ccuProfilingInfoCache.name.c_str(), ccuProfilingInfoCache.dieId); | 139 | ccuProfilingInfoCache.name.c_str(), ccuProfilingInfoCache.dieId); |
| 140 | profilingInfo.push_back(ccuProfilingInfoCache); | 140 | profilingInfo.push_back(ccuProfilingInfoCache); |
| 141 | } | 141 | } |
| @@ -212,7 +212,7 @@ HcclResult CcuJetty::Clean() | |||
| 212 | 212 | ||
| 213 | auto ret = RaCtxQpDestroy(jettyHandle); | 213 | auto ret = RaCtxQpDestroy(jettyHandle); |
| 214 | if (ret != 0) { | 214 | if (ret != 0) { |
| 215 | - HCCL_ERROR("[CcuJetty][%s] failed, jettyHanlde[0x%llx].", __func__, jettyHandle); | 215 | + HCCL_ERROR("[CcuJetty][%s] failed, jettyHandle[0x%llx].", __func__, jettyHandle); |
| 216 | return HcclResult::HCCL_E_NETWORK; | 216 | return HcclResult::HCCL_E_NETWORK; |
| 217 | } | 217 | } |
| 218 | } | 218 | } |
| @@ -23,7 +23,7 @@ HcclResult AivChannelHelper::FillDevEntities( | |||
| 23 | { | 23 | { |
| 24 | CHK_PTR_NULL(channelList); | 24 | CHK_PTR_NULL(channelList); |
| 25 | CHK_PTR_NULL(linkStatusList); | 25 | CHK_PTR_NULL(linkStatusList); |
| 26 | - CHK_PRT_RET((listNum == 0), HCCL_ERROR("[%s]Invalid listNum, listNum[%u]", __func__, listNum), HCCL_E_PARA); | 26 | + CHK_PRT_RET((listNum == 0), HCCL_ERROR("[%s] Invalid listNum, listNum[%u]", __func__, listNum), HCCL_E_PARA); |
| 27 | 27 | ||
| 28 | for (uint32_t i = 0; i < listNum; i++) { | 28 | for (uint32_t i = 0; i < listNum; i++) { |
| 29 | if (linkStatusList[i] != HCOMM_CHANNEL_STATUS_READY) { | 29 | if (linkStatusList[i] != HCOMM_CHANNEL_STATUS_READY) { |
| @@ -75,7 +75,7 @@ HcclResult AivChannelHelper::PreAllocChannels( | |||
| 75 | CHK_PTR_NULL(targetChannels); | 75 | CHK_PTR_NULL(targetChannels); |
| 76 | CHK_PTR_NULL(userChannels); | 76 | CHK_PTR_NULL(userChannels); |
| 77 | CHK_PRT_RET( | 77 | CHK_PRT_RET( |
| 78 | - (channelNum == 0), HCCL_ERROR("[%s]Invalid channelNum, channelNum[%u]", __func__, channelNum), HCCL_E_PARA); | 78 | + (channelNum == 0), HCCL_ERROR("[%s] Invalid channelNum, channelNum[%u]", __func__, channelNum), HCCL_E_PARA); |
| 79 | 79 | ||
| 80 | bool needD2HMap = false; | 80 | bool needD2HMap = false; |
| 81 | for (uint32_t i = 0; i < channelNum; i++) { | 81 | for (uint32_t i = 0; i < channelNum; i++) { |
| @@ -213,7 +213,7 @@ void BkfDcDisp(BkfDc *dc) | |||
| 213 | BKF_DISP_PRINTF(disp, "log(%#x/%#x)\n", BKF_MASK_ADDR(dc->argInit.log), BKF_MASK_ADDR(dc->log)); | 213 | BKF_DISP_PRINTF(disp, "log(%#x/%#x)\n", BKF_MASK_ADDR(dc->argInit.log), BKF_MASK_ADDR(dc->log)); |
| 214 | BKF_DISP_PRINTF(disp, "jobMng(%#x)/jobTypeId(%u)/jobPrio(%u)\n", BKF_MASK_ADDR(dc->argInit.jobMng), | 214 | BKF_DISP_PRINTF(disp, "jobMng(%#x)/jobTypeId(%u)/jobPrio(%u)\n", BKF_MASK_ADDR(dc->argInit.jobMng), |
| 215 | dc->argInit.jobTypeId, dc->argInit.jobPrio); | 215 | dc->argInit.jobTypeId, dc->argInit.jobPrio); |
| 216 | - BKF_DISP_PRINTF(disp, "slie_keyLen(%u)/keyCmp(%#x)/keyGetStrOrNull(%#x)/keyCodec(%#x)\n", | 216 | + BKF_DISP_PRINTF(disp, "slice_keyLen(%u)/keyCmp(%#x)/keyGetStrOrNull(%#x)/keyCodec(%#x)\n", |
| 217 | dc->argInit.sliceVTbl.keyLen, BKF_MASK_ADDR(dc->argInit.sliceVTbl.keyCmp), | 217 | dc->argInit.sliceVTbl.keyLen, BKF_MASK_ADDR(dc->argInit.sliceVTbl.keyCmp), |
| 218 | BKF_MASK_ADDR(dc->argInit.sliceVTbl.keyGetStrOrNull), BKF_MASK_ADDR(dc->argInit.sliceVTbl.keyCodec)); | 218 | BKF_MASK_ADDR(dc->argInit.sliceVTbl.keyGetStrOrNull), BKF_MASK_ADDR(dc->argInit.sliceVTbl.keyCodec)); |
| 219 | BKF_DISP_PRINTF(disp, "sysLogMng(%#x)\n", BKF_MASK_ADDR(dc->argInit.sysLogMng)); | 219 | BKF_DISP_PRINTF(disp, "sysLogMng(%#x)\n", BKF_MASK_ADDR(dc->argInit.sysLogMng)); |
Msrc/base_comm/resources/hccp/netco/idl/src/data_sync/processor/pub_sub/sub/conn/bkf_suber_conn_fsm.c+3-3
| @@ -286,7 +286,7 @@ uint32_t BkfSuberConnSndHello(BkfSuberConn *conn) | |||
| 286 | uint8_t *sendBuf = VOS_NULL; | 286 | uint8_t *sendBuf = VOS_NULL; |
| 287 | uint32_t ret = BkfSuberConnMallocSendBuf(conn, &sendBuf, BKF_CH_CLI_MALLOC_DATA_BUF_LEN_MAX); | 287 | uint32_t ret = BkfSuberConnMallocSendBuf(conn, &sendBuf, BKF_CH_CLI_MALLOC_DATA_BUF_LEN_MAX); |
| 288 | if (ret != BKF_OK) { | 288 | if (ret != BKF_OK) { |
| 289 | - BKF_LOG_ERROR(connMng->log, "SendHello mem alloc faild.url %s\n", | 289 | + BKF_LOG_ERROR(connMng->log, "SendHello mem alloc failed. url %s\n", |
| 290 | BkfUrlGetStr(&conn->puberUrl, urlStr, sizeof(urlStr))); | 290 | BkfUrlGetStr(&conn->puberUrl, urlStr, sizeof(urlStr))); |
| 291 | return BKF_ERR; | 291 | return BKF_ERR; |
| 292 | } | 292 | } |
| @@ -315,7 +315,7 @@ uint32_t BkfSuberConnSndHello(BkfSuberConn *conn) | |||
| 315 | 315 | ||
| 316 | if (ret != BKF_OK) { | 316 | if (ret != BKF_OK) { |
| 317 | BkfChCliFreeDataBuf(env->chCliMng, conn->connId, (void *)sendBuf); | 317 | BkfChCliFreeDataBuf(env->chCliMng, conn->connId, (void *)sendBuf); |
| 318 | - BKF_LOG_DEBUG(connMng->log, "Connect Send Hello encap fal.url %s\n", | 318 | + BKF_LOG_DEBUG(connMng->log, "Connect Send Hello encap fail. url %s\n", |
| 319 | BkfUrlGetStr(&conn->puberUrl, urlStr, sizeof(urlStr))); | 319 | BkfUrlGetStr(&conn->puberUrl, urlStr, sizeof(urlStr))); |
| 320 | return BKF_SUBER_SESS_SEND_BUF_NOT_ENOUGH; | 320 | return BKF_SUBER_SESS_SEND_BUF_NOT_ENOUGH; |
| 321 | } | 321 | } |
| @@ -576,4 +576,4 @@ void BkfSuberConnFsmUnInitFsmTmp(BkfSuberConnMng *connMng) | |||
| 576 | 576 | ||
| 577 | #ifdef __cplusplus | 577 | #ifdef __cplusplus |
| 578 | } | 578 | } |
| 579 | -#endif | 579 | +#endif |
| @@ -42,7 +42,7 @@ HCCP_ATTRI_VISI_DEF int RaSocketBatchConnectAsync(struct SocketConnectInfoT conn | |||
| 42 | 42 | ||
| 43 | phyId = socketHandle->rdevInfo.phyId; | 43 | phyId = socketHandle->rdevInfo.phyId; |
| 44 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 44 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 45 | - hccp_err("[batch_connect][ra_socket]phyId[%u]invalid, " | 45 | + hccp_err("[batch_connect][ra_socket]phyId[%u] invalid, " |
| 46 | "must in range of [0, %u)", | 46 | "must in range of [0, %u)", |
| 47 | phyId, RA_MAX_PHY_ID_NUM), | 47 | phyId, RA_MAX_PHY_ID_NUM), |
| 48 | ConverReturnCode(SOCKET_OP, -EINVAL)); | 48 | ConverReturnCode(SOCKET_OP, -EINVAL)); |
| @@ -93,7 +93,7 @@ HCCP_ATTRI_VISI_DEF int RaSocketListenStartAsync(struct SocketListenInfoT conn[] | |||
| 93 | 93 | ||
| 94 | phyId = socketHandle->rdevInfo.phyId; | 94 | phyId = socketHandle->rdevInfo.phyId; |
| 95 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 95 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 96 | - hccp_err("[listen_start][ra_socket]phyId[%u]invalid, " | 96 | + hccp_err("[listen_start][ra_socket]phyId[%u] invalid, " |
| 97 | "must in range of [0, %u)", | 97 | "must in range of [0, %u)", |
| 98 | phyId, RA_MAX_PHY_ID_NUM), | 98 | phyId, RA_MAX_PHY_ID_NUM), |
| 99 | ConverReturnCode(SOCKET_OP, -EINVAL)); | 99 | ConverReturnCode(SOCKET_OP, -EINVAL)); |
| @@ -133,7 +133,7 @@ HCCP_ATTRI_VISI_DEF int RaSocketListenStopAsync(struct SocketListenInfoT conn[], | |||
| 133 | 133 | ||
| 134 | phyId = socketHandle->rdevInfo.phyId; | 134 | phyId = socketHandle->rdevInfo.phyId; |
| 135 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 135 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 136 | - hccp_err("[listen_stop][ra_socket]phyId[%u]invalid, " | 136 | + hccp_err("[listen_stop][ra_socket]phyId[%u] invalid, " |
| 137 | "must in range of [0, %u)", | 137 | "must in range of [0, %u)", |
| 138 | phyId, RA_MAX_PHY_ID_NUM), | 138 | phyId, RA_MAX_PHY_ID_NUM), |
| 139 | ConverReturnCode(SOCKET_OP, -EINVAL)); | 139 | ConverReturnCode(SOCKET_OP, -EINVAL)); |
| @@ -173,7 +173,7 @@ HCCP_ATTRI_VISI_DEF int RaSocketBatchCloseAsync(struct SocketCloseInfoT conn[], | |||
| 173 | 173 | ||
| 174 | phyId = socketHandle->rdevInfo.phyId; | 174 | phyId = socketHandle->rdevInfo.phyId; |
| 175 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 175 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 176 | - hccp_err("[batch_close][ra_socket]phyId[%u]invalid, " | 176 | + hccp_err("[batch_close][ra_socket]phyId[%u] invalid, " |
| 177 | "must in range of [0, %u)", | 177 | "must in range of [0, %u)", |
| 178 | phyId, RA_MAX_PHY_ID_NUM), | 178 | phyId, RA_MAX_PHY_ID_NUM), |
| 179 | ConverReturnCode(SOCKET_OP, -EINVAL)); | 179 | ConverReturnCode(SOCKET_OP, -EINVAL)); |
| @@ -81,7 +81,7 @@ HCCP_ATTRI_VISI_DEF int RaGetDevEidInfoNum(struct RaInfo info, unsigned int *num | |||
| 81 | 81 | ||
| 82 | CHK_PRT_RETURN(num == NULL, hccp_err("[get][eid]num is NULL"), ConverReturnCode(RDMA_OP, -EINVAL)); | 82 | CHK_PRT_RETURN(num == NULL, hccp_err("[get][eid]num is NULL"), ConverReturnCode(RDMA_OP, -EINVAL)); |
| 83 | CHK_PRT_RETURN(info.phyId >= RA_MAX_PHY_ID_NUM, | 83 | CHK_PRT_RETURN(info.phyId >= RA_MAX_PHY_ID_NUM, |
| 84 | - hccp_err("[get][eid]phy_id(%u) must smaller than %u", info.phyId, RA_MAX_PHY_ID_NUM), | 84 | + hccp_err("[get][eid]phy_id(%u) must be smaller than %u", info.phyId, RA_MAX_PHY_ID_NUM), |
| 85 | ConverReturnCode(RDMA_OP, -EINVAL)); | 85 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 86 | 86 | ||
| 87 | hccp_run_info("Input parameters: phy_id[%u], nic_position:[%d]", info.phyId, info.mode); | 87 | hccp_run_info("Input parameters: phy_id[%u], nic_position:[%d]", info.phyId, info.mode); |
| @@ -110,7 +110,7 @@ HCCP_ATTRI_VISI_DEF int RaGetDevEidInfoList(struct RaInfo info, struct HccpDevEi | |||
| 110 | CHK_PRT_RETURN(infoList == NULL || num == NULL, hccp_err("[get][eid]info_list or num is NULL"), | 110 | CHK_PRT_RETURN(infoList == NULL || num == NULL, hccp_err("[get][eid]info_list or num is NULL"), |
| 111 | ConverReturnCode(RDMA_OP, -EINVAL)); | 111 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 112 | CHK_PRT_RETURN(info.phyId >= RA_MAX_PHY_ID_NUM, | 112 | CHK_PRT_RETURN(info.phyId >= RA_MAX_PHY_ID_NUM, |
| 113 | - hccp_err("[get][eid]phy_id(%u) must smaller than %u", info.phyId, RA_MAX_PHY_ID_NUM), | 113 | + hccp_err("[get][eid]phy_id(%u) must be smaller than %u", info.phyId, RA_MAX_PHY_ID_NUM), |
| 114 | ConverReturnCode(RDMA_OP, -EINVAL)); | 114 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 115 | 115 | ||
| 116 | hccp_run_info("Input parameters: phy_id[%u], nic_position:[%d]", info.phyId, info.mode); | 116 | hccp_run_info("Input parameters: phy_id[%u], nic_position:[%d]", info.phyId, info.mode); |
| @@ -160,7 +160,7 @@ HCCP_ATTRI_VISI_DEF int RaCtxInit(struct CtxInitCfg *cfg, struct CtxInitAttr *at | |||
| 160 | CHK_PRT_RETURN(cfg == NULL || attr == NULL || ctxHandle == NULL, | 160 | CHK_PRT_RETURN(cfg == NULL || attr == NULL || ctxHandle == NULL, |
| 161 | hccp_err("[init][ra_ctx]cfg or attr or ctx_handle is NULL"), ConverReturnCode(HCCP_INIT, -EINVAL)); | 161 | hccp_err("[init][ra_ctx]cfg or attr or ctx_handle is NULL"), ConverReturnCode(HCCP_INIT, -EINVAL)); |
| 162 | CHK_PRT_RETURN(attr->phyId >= RA_MAX_PHY_ID_NUM, | 162 | CHK_PRT_RETURN(attr->phyId >= RA_MAX_PHY_ID_NUM, |
| 163 | - hccp_err("[init][ra_ctx]phy_id(%u) must smaller than %u", attr->phyId, RA_MAX_PHY_ID_NUM), | 163 | + hccp_err("[init][ra_ctx]phy_id(%u) must be smaller than %u", attr->phyId, RA_MAX_PHY_ID_NUM), |
| 164 | ConverReturnCode(HCCP_INIT, -EINVAL)); | 164 | ConverReturnCode(HCCP_INIT, -EINVAL)); |
| 165 | 165 | ||
| 166 | ctxHandleTmp = calloc(1, sizeof(struct RaCtxHandle)); | 166 | ctxHandleTmp = calloc(1, sizeof(struct RaCtxHandle)); |
| @@ -213,8 +213,8 @@ HCCP_ATTRI_VISI_DEF int RaCtxGetAsyncEvents(void *ctxHandle, struct AsyncEvent e | |||
| 213 | hccp_err("[get][async_events]ctx_handle or events or num is NULL"), ConverReturnCode(RDMA_OP, -EINVAL)); | 213 | hccp_err("[get][async_events]ctx_handle or events or num is NULL"), ConverReturnCode(RDMA_OP, -EINVAL)); |
| 214 | 214 | ||
| 215 | CHK_PRT_RETURN(*num == 0 || *num > ASYNC_EVENT_MAX_NUM, | 215 | CHK_PRT_RETURN(*num == 0 || *num > ASYNC_EVENT_MAX_NUM, |
| 216 | - hccp_err("[get][async_events]num:%u must greater than 0" | 216 | + hccp_err("[get][async_events]num:%u must be greater than 0 " |
| 217 | - " and less or equal to %d", | 217 | + "and less than or equal to %d", |
| 218 | *num, ASYNC_EVENT_MAX_NUM), | 218 | *num, ASYNC_EVENT_MAX_NUM), |
| 219 | ConverReturnCode(RDMA_OP, -EINVAL)); | 219 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 220 | 220 | ||
| @@ -266,8 +266,8 @@ HCCP_ATTRI_VISI_DEF int RaGetEidByIp(void *ctxHandle, struct IpInfo ip[], union | |||
| 266 | hccp_err("[get][eid_by_ip]ctx_handle or ip or eid or num is NULL"), ConverReturnCode(RDMA_OP, -EINVAL)); | 266 | hccp_err("[get][eid_by_ip]ctx_handle or ip or eid or num is NULL"), ConverReturnCode(RDMA_OP, -EINVAL)); |
| 267 | 267 | ||
| 268 | CHK_PRT_RETURN(*num == 0 || *num > HCCP_EID_IP_QUERY_MAX_NUM, | 268 | CHK_PRT_RETURN(*num == 0 || *num > HCCP_EID_IP_QUERY_MAX_NUM, |
| 269 | - hccp_err("[get][eid_by_ip]num(%u) must greater than 0" | 269 | + hccp_err("[get][eid_by_ip]num(%u) must be greater than 0 " |
| 270 | - " and less or equal to %d", | 270 | + "and less than or equal to %d", |
| 271 | *num, HCCP_EID_IP_QUERY_MAX_NUM), | 271 | *num, HCCP_EID_IP_QUERY_MAX_NUM), |
| 272 | ConverReturnCode(RDMA_OP, -EINVAL)); | 272 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 273 | 273 | ||
| @@ -297,8 +297,8 @@ HCCP_ATTRI_VISI_DEF int RaGetIpByEid(void *ctxHandle, union HccpEid eid[], struc | |||
| 297 | hccp_err("[get][IpByEid]ctxHandle or eid or ip or num is NULL"), ConverReturnCode(RDMA_OP, -EINVAL)); | 297 | hccp_err("[get][IpByEid]ctxHandle or eid or ip or num is NULL"), ConverReturnCode(RDMA_OP, -EINVAL)); |
| 298 | 298 | ||
| 299 | CHK_PRT_RETURN(*num == 0 || *num > HCCP_EID_IP_QUERY_MAX_NUM, | 299 | CHK_PRT_RETURN(*num == 0 || *num > HCCP_EID_IP_QUERY_MAX_NUM, |
| 300 | - hccp_err("[get][IpByEid]num(%u) must greater than 0" | 300 | + hccp_err("[get][IpByEid]num(%u) must be greater than 0 " |
| 301 | - " and less or equal to %d", | 301 | + "and less than or equal to %d", |
| 302 | *num, HCCP_EID_IP_QUERY_MAX_NUM), | 302 | *num, HCCP_EID_IP_QUERY_MAX_NUM), |
| 303 | ConverReturnCode(RDMA_OP, -EINVAL)); | 303 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 304 | 304 | ||
| @@ -951,7 +951,7 @@ STATIC int RaCtxBatchSendWrCheck(struct RaCtxQpHandle *qpHandle, struct SendWrDa | |||
| 951 | if ((wrList[i].ub.opcode != RA_UB_OPC_WRITE && wrList[i].ub.opcode != RA_UB_OPC_WRITE_NOTIFY && | 951 | if ((wrList[i].ub.opcode != RA_UB_OPC_WRITE && wrList[i].ub.opcode != RA_UB_OPC_WRITE_NOTIFY && |
| 952 | wrList[i].ub.opcode != RA_UB_OPC_READ) && | 952 | wrList[i].ub.opcode != RA_UB_OPC_READ) && |
| 953 | wrList[i].ub.reduceInfo.reduceEn) { | 953 | wrList[i].ub.reduceInfo.reduceEn) { |
| 954 | - hccp_err("[send][ra_qp]wr[%u] opcode[%d] not supportreduce", i, wrList[i].ub.opcode); | 954 | + hccp_err("[send][ra_qp]wr[%u] opcode[%d] does not support reduce", i, wrList[i].ub.opcode); |
| 955 | return -EINVAL; | 955 | return -EINVAL; |
| 956 | } | 956 | } |
| 957 | } | 957 | } |
| @@ -1013,7 +1013,7 @@ HCCP_ATTRI_VISI_DEF int RaCtxGetAuxInfo(void *ctxHandle, struct HccpAuxInfoIn *i | |||
| 1013 | hccp_err("[get][aux_info]ctx_handle or in or out is NULL"), ConverReturnCode(RDMA_OP, -EINVAL)); | 1013 | hccp_err("[get][aux_info]ctx_handle or in or out is NULL"), ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1014 | 1014 | ||
| 1015 | CHK_PRT_RETURN(in->type < AUX_INFO_IN_TYPE_CQE || in->type >= AUX_INFO_IN_TYPE_MAX, | 1015 | CHK_PRT_RETURN(in->type < AUX_INFO_IN_TYPE_CQE || in->type >= AUX_INFO_IN_TYPE_MAX, |
| 1016 | - hccp_err("[get][aux_info]in->type(%d) must greater or equal to %d and less than %d", in->type, | 1016 | + hccp_err("[get][aux_info]in->type(%d) must be greater than or equal to %d and less than %d", in->type, |
| 1017 | AUX_INFO_IN_TYPE_CQE, AUX_INFO_IN_TYPE_MAX), | 1017 | AUX_INFO_IN_TYPE_CQE, AUX_INFO_IN_TYPE_MAX), |
| 1018 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1018 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1019 | 1019 | ||
| @@ -1039,8 +1039,8 @@ HCCP_ATTRI_VISI_DEF int RaCtxGetCrErrInfoList(void *ctxHandle, struct CrErrInfo | |||
| 1039 | hccp_err("[get][cr_err_info_list]ctx_handle or info_list or num is NULL"), ConverReturnCode(RDMA_OP, -EINVAL)); | 1039 | hccp_err("[get][cr_err_info_list]ctx_handle or info_list or num is NULL"), ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1040 | 1040 | ||
| 1041 | CHK_PRT_RETURN(*num == 0 || *num > CR_ERR_INFO_MAX_NUM, | 1041 | CHK_PRT_RETURN(*num == 0 || *num > CR_ERR_INFO_MAX_NUM, |
| 1042 | - hccp_err("[get][cr_err_info_list]num:%u must greater " | 1042 | + hccp_err("[get][cr_err_info_list]num:%u must be greater than 0 " |
| 1043 | - "than 0 and less or equal to %d", | 1043 | + "and less than or equal to %d", |
| 1044 | *num, CR_ERR_INFO_MAX_NUM), | 1044 | *num, CR_ERR_INFO_MAX_NUM), |
| 1045 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1045 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1046 | 1046 | ||
| @@ -1088,4 +1088,4 @@ HCCP_ATTRI_VISI_DEF int RaCtxGetJettyContext(void *qpHandle, uint8_t context[], | |||
| 1088 | ConverReturnCode(RDMA_OP, ret)); | 1088 | ConverReturnCode(RDMA_OP, ret)); |
| 1089 | 1089 | ||
| 1090 | return ConverReturnCode(RDMA_OP, ret); | 1090 | return ConverReturnCode(RDMA_OP, ret); |
| 1091 | -} | 1091 | +} |
| @@ -443,8 +443,8 @@ int RaRdevInitCheck(int mode, struct rdev rdevInfo, char localIp[], unsigned int | |||
| 443 | int ret; | 443 | int ret; |
| 444 | 444 | ||
| 445 | CHK_PRT_RETURN(rdevInfo.phyId >= RA_MAX_PHY_ID_NUM, | 445 | CHK_PRT_RETURN(rdevInfo.phyId >= RA_MAX_PHY_ID_NUM, |
| 446 | - hccp_err("[check][ra_rdev_init]phyId(%u) is invalid!" | 446 | + hccp_err("[check][ra_rdev_init]phyId(%u) is invalid! " |
| 447 | - "it must greater or equal to 0 and less than %d!", | 447 | + "it must be greater than or equal to 0 and less than %d!", |
| 448 | rdevInfo.phyId, RA_MAX_PHY_ID_NUM), | 448 | rdevInfo.phyId, RA_MAX_PHY_ID_NUM), |
| 449 | -EINVAL); | 449 | -EINVAL); |
| 450 | 450 | ||
| @@ -602,8 +602,8 @@ STATIC int RaRdevDeinitParaCheck(void *rdmaHandle, struct RaRdmaHandle **rdmaHan | |||
| 602 | *rdmaHandleTmp = (struct RaRdmaHandle *)rdmaHandle; | 602 | *rdmaHandleTmp = (struct RaRdmaHandle *)rdmaHandle; |
| 603 | rdevInfo = (*rdmaHandleTmp)->rdevInfo; | 603 | rdevInfo = (*rdmaHandleTmp)->rdevInfo; |
| 604 | CHK_PRT_RETURN(rdevInfo.phyId >= RA_MAX_PHY_ID_NUM, | 604 | CHK_PRT_RETURN(rdevInfo.phyId >= RA_MAX_PHY_ID_NUM, |
| 605 | - hccp_err("[deinit][ra_rdev]phyId(%u)" | 605 | + hccp_err("[deinit][ra_rdev]phyId(%u) " |
| 606 | - "must smaller than %u", | 606 | + "must be smaller than %u", |
| 607 | rdevInfo.phyId, RA_MAX_PHY_ID_NUM), | 607 | rdevInfo.phyId, RA_MAX_PHY_ID_NUM), |
| 608 | -EINVAL); | 608 | -EINVAL); |
| 609 | 609 | ||
| @@ -630,7 +630,7 @@ HCCP_ATTRI_VISI_DEF int RaRdevGetPortStatus(void *rdmaHandle, enum PortStatus *s | |||
| 630 | 630 | ||
| 631 | phyId = rdmaHandleTmp->rdevInfo.phyId; | 631 | phyId = rdmaHandleTmp->rdevInfo.phyId; |
| 632 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 632 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 633 | - hccp_err("[get][ra_port_status]phyId(%u) must smaller than %u", phyId, RA_MAX_PHY_ID_NUM), | 633 | + hccp_err("[get][ra_port_status]phyId(%u) must be smaller than %u", phyId, RA_MAX_PHY_ID_NUM), |
| 634 | ConverReturnCode(RDMA_OP, -EINVAL)); | 634 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 635 | 635 | ||
| 636 | CHK_PRT_RETURN(rdmaHandleTmp->rdmaOps == NULL || rdmaHandleTmp->rdmaOps->raRdevGetPortStatus == NULL, | 636 | CHK_PRT_RETURN(rdmaHandleTmp->rdmaOps == NULL || rdmaHandleTmp->rdmaOps->raRdevGetPortStatus == NULL, |
| @@ -658,7 +658,7 @@ HCCP_ATTRI_VISI_DEF int RaRdevDeinit(void *rdmaHandle, unsigned int notifyType) | |||
| 658 | 658 | ||
| 659 | ret = rdmaHandleTmp->rdmaOps->raRdevDeinit(rdmaHandleTmp, notifyType); | 659 | ret = rdmaHandleTmp->rdmaOps->raRdevDeinit(rdmaHandleTmp, notifyType); |
| 660 | if (ret) { | 660 | if (ret) { |
| 661 | - hccp_err("[deinit][ra_rdev]ra rdv deinit failed, ret(%d)", ret); | 661 | + hccp_err("[deinit][ra_rdev]ra rdev deinit failed, ret(%d)", ret); |
| 662 | goto free_rdma_handle; | 662 | goto free_rdma_handle; |
| 663 | } | 663 | } |
| 664 | 664 | ||
| @@ -707,7 +707,7 @@ HCCP_ATTRI_VISI_DEF int RaSocketBatchConnect(struct SocketConnectInfoT conn[], u | |||
| 707 | } | 707 | } |
| 708 | 708 | ||
| 709 | CHK_PRT_RETURN(socketHandle->rdevInfo.phyId >= RA_MAX_PHY_ID_NUM, | 709 | CHK_PRT_RETURN(socketHandle->rdevInfo.phyId >= RA_MAX_PHY_ID_NUM, |
| 710 | - hccp_err("[batch_connect][ra_socket]phyId(%u) must smaller than %u", socketHandle->rdevInfo.phyId, | 710 | + hccp_err("[batch_connect][ra_socket]phyId(%u) must be smaller than %u", socketHandle->rdevInfo.phyId, |
| 711 | RA_MAX_PHY_ID_NUM), | 711 | RA_MAX_PHY_ID_NUM), |
| 712 | ConverReturnCode(SOCKET_OP, -EINVAL)); | 712 | ConverReturnCode(SOCKET_OP, -EINVAL)); |
| 713 | 713 | ||
| @@ -755,7 +755,7 @@ HCCP_ATTRI_VISI_DEF int RaSocketBatchClose(struct SocketCloseInfoT conn[], unsig | |||
| 755 | } | 755 | } |
| 756 | phyId = socketHandle->rdevInfo.phyId; | 756 | phyId = socketHandle->rdevInfo.phyId; |
| 757 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 757 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 758 | - hccp_err("[batch_close][ra_socket]phyId(%u) must smaller than %u", phyId, RA_MAX_PHY_ID_NUM), | 758 | + hccp_err("[batch_close][ra_socket]phyId(%u) must be smaller than %u", phyId, RA_MAX_PHY_ID_NUM), |
| 759 | ConverReturnCode(SOCKET_OP, -EINVAL)); | 759 | ConverReturnCode(SOCKET_OP, -EINVAL)); |
| 760 | 760 | ||
| 761 | ret = RaInetPton(socketHandle->rdevInfo.family, socketHandle->rdevInfo.localIp, localIp, MAX_IP_LEN); | 761 | ret = RaInetPton(socketHandle->rdevInfo.family, socketHandle->rdevInfo.localIp, localIp, MAX_IP_LEN); |
| @@ -793,7 +793,7 @@ HCCP_ATTRI_VISI_DEF int RaSocketBatchAbort(struct SocketConnectInfoT conn[], uns | |||
| 793 | 793 | ||
| 794 | phyId = socketHandle->rdevInfo.phyId; | 794 | phyId = socketHandle->rdevInfo.phyId; |
| 795 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 795 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 796 | - hccp_err("[batch_abort][ra_socket]phyId(%u) must smaller than %u", phyId, RA_MAX_PHY_ID_NUM), | 796 | + hccp_err("[batch_abort][ra_socket]phyId(%u) must be smaller than %u", phyId, RA_MAX_PHY_ID_NUM), |
| 797 | ConverReturnCode(SOCKET_OP, -EINVAL)); | 797 | ConverReturnCode(SOCKET_OP, -EINVAL)); |
| 798 | 798 | ||
| 799 | ret = RaInetPton(socketHandle->rdevInfo.family, socketHandle->rdevInfo.localIp, localIp, MAX_IP_LEN); | 799 | ret = RaInetPton(socketHandle->rdevInfo.family, socketHandle->rdevInfo.localIp, localIp, MAX_IP_LEN); |
| @@ -835,7 +835,7 @@ HCCP_ATTRI_VISI_DEF int RaSocketListenStart(struct SocketListenInfoT conn[], uns | |||
| 835 | } | 835 | } |
| 836 | phyId = socketHandle->rdevInfo.phyId; | 836 | phyId = socketHandle->rdevInfo.phyId; |
| 837 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 837 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 838 | - hccp_err("[listen_start][ra_socket]phyId(%u) must smaller than %u", phyId, RA_MAX_PHY_ID_NUM), | 838 | + hccp_err("[listen_start][ra_socket]phyId(%u) must be smaller than %u", phyId, RA_MAX_PHY_ID_NUM), |
| 839 | ConverReturnCode(SOCKET_OP, -EINVAL)); | 839 | ConverReturnCode(SOCKET_OP, -EINVAL)); |
| 840 | 840 | ||
| 841 | ret = RaInetPton(socketHandle->rdevInfo.family, socketHandle->rdevInfo.localIp, localIp, MAX_IP_LEN); | 841 | ret = RaInetPton(socketHandle->rdevInfo.family, socketHandle->rdevInfo.localIp, localIp, MAX_IP_LEN); |
| @@ -872,7 +872,7 @@ HCCP_ATTRI_VISI_DEF int RaSocketListenStop(struct SocketListenInfoT conn[], unsi | |||
| 872 | 872 | ||
| 873 | phyId = socketHandle->rdevInfo.phyId; | 873 | phyId = socketHandle->rdevInfo.phyId; |
| 874 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 874 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 875 | - hccp_err("[listen_stop][ra_socket]phyId(%u) must smaller than %u", phyId, RA_MAX_PHY_ID_NUM), | 875 | + hccp_err("[listen_stop][ra_socket]phyId(%u) must be smaller than %u", phyId, RA_MAX_PHY_ID_NUM), |
| 876 | ConverReturnCode(SOCKET_OP, -EINVAL)); | 876 | ConverReturnCode(SOCKET_OP, -EINVAL)); |
| 877 | 877 | ||
| 878 | ret = RaInetPton(socketHandle->rdevInfo.family, socketHandle->rdevInfo.localIp, localIp, MAX_IP_LEN); | 878 | ret = RaInetPton(socketHandle->rdevInfo.family, socketHandle->rdevInfo.localIp, localIp, MAX_IP_LEN); |
| @@ -910,7 +910,7 @@ HCCP_ATTRI_VISI_DEF int RaGetSockets(unsigned int role, struct SocketInfoT conn[ | |||
| 910 | 910 | ||
| 911 | phyId = socketHandle->rdevInfo.phyId; | 911 | phyId = socketHandle->rdevInfo.phyId; |
| 912 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 912 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 913 | - hccp_err("[get][ra_socket]phyId(%u) must smaller than %u", phyId, RA_MAX_PHY_ID_NUM), | 913 | + hccp_err("[get][ra_socket]phyId(%u) must be smaller than %u", phyId, RA_MAX_PHY_ID_NUM), |
| 914 | ConverReturnCode(SOCKET_OP, -EINVAL)); | 914 | ConverReturnCode(SOCKET_OP, -EINVAL)); |
| 915 | 915 | ||
| 916 | ret = RaInetPton(socketHandle->rdevInfo.family, socketHandle->rdevInfo.localIp, localIp, MAX_IP_LEN); | 916 | ret = RaInetPton(socketHandle->rdevInfo.family, socketHandle->rdevInfo.localIp, localIp, MAX_IP_LEN); |
| @@ -952,7 +952,7 @@ HCCP_ATTRI_VISI_DEF int RaSocketRecv(const void *fdHandle, void *data, unsigned | |||
| 952 | } | 952 | } |
| 953 | phyId = socketHandleTmp->rdevInfo.phyId; | 953 | phyId = socketHandleTmp->rdevInfo.phyId; |
| 954 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 954 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 955 | - hccp_err("[recv][ra_socket]phyId(%u) must smaller than %u", phyId, RA_MAX_PHY_ID_NUM), | 955 | + hccp_err("[recv][ra_socket]phyId(%u) must be smaller than %u", phyId, RA_MAX_PHY_ID_NUM), |
| 956 | ConverReturnCode(SOCKET_OP, -EINVAL)); | 956 | ConverReturnCode(SOCKET_OP, -EINVAL)); |
| 957 | 957 | ||
| 958 | ret = socketHandleTmp->socketOps->raSocketRecv(phyId, fdHandle, data, size); | 958 | ret = socketHandleTmp->socketOps->raSocketRecv(phyId, fdHandle, data, size); |
| @@ -989,7 +989,7 @@ HCCP_ATTRI_VISI_DEF int RaSocketSend(const void *fdHandle, const void *data, uns | |||
| 989 | 989 | ||
| 990 | phyId = socketHandleTmp->rdevInfo.phyId; | 990 | phyId = socketHandleTmp->rdevInfo.phyId; |
| 991 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 991 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 992 | - hccp_err("[send][ra_socket]phyId(%u) must smaller than %u", phyId, RA_MAX_PHY_ID_NUM), | 992 | + hccp_err("[send][ra_socket]phyId(%u) must be smaller than %u", phyId, RA_MAX_PHY_ID_NUM), |
| 993 | ConverReturnCode(SOCKET_OP, -EINVAL)); | 993 | ConverReturnCode(SOCKET_OP, -EINVAL)); |
| 994 | 994 | ||
| 995 | ret = socketHandleTmp->socketOps->raSocketSend(phyId, fdHandle, data, size); | 995 | ret = socketHandleTmp->socketOps->raSocketSend(phyId, fdHandle, data, size); |
| @@ -1059,7 +1059,7 @@ HCCP_ATTRI_VISI_DEF int RaSetTcpRecvCallback(const void *socketHandle, const voi | |||
| 1059 | 1059 | ||
| 1060 | phyId = socketHandleTmp->rdevInfo.phyId; | 1060 | phyId = socketHandleTmp->rdevInfo.phyId; |
| 1061 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 1061 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 1062 | - hccp_err("[ra_socket]phyId(%u) must smaller than %u", phyId, RA_MAX_PHY_ID_NUM), | 1062 | + hccp_err("[ra_socket]phyId(%u) must be smaller than %u", phyId, RA_MAX_PHY_ID_NUM), |
| 1063 | ConverReturnCode(SOCKET_OP, -EINVAL)); | 1063 | ConverReturnCode(SOCKET_OP, -EINVAL)); |
| 1064 | 1064 | ||
| 1065 | RaPeerSetTcpRecvCallback(phyId, callback); | 1065 | RaPeerSetTcpRecvCallback(phyId, callback); |
| @@ -1082,8 +1082,8 @@ HCCP_ATTRI_VISI_DEF int RaGetTsqpDepth(void *rdevHandle, unsigned int *tempDepth | |||
| 1082 | 1082 | ||
| 1083 | phyId = rdmaHandleTmp->rdevInfo.phyId; | 1083 | phyId = rdmaHandleTmp->rdevInfo.phyId; |
| 1084 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 1084 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 1085 | - hccp_err("[get][ra_tsqp_depth]phyId(%u) is invalid! it must greater or equal to 0 and less than %d!", phyId, | 1085 | + hccp_err("[get][ra_tsqp_depth]phyId(%u) is invalid! it must be greater than or equal to 0 and less than %d!", |
| 1086 | - RA_MAX_PHY_ID_NUM), | 1086 | + phyId, RA_MAX_PHY_ID_NUM), |
| 1087 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1087 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1088 | 1088 | ||
| 1089 | hccp_info("Input parameters: phyId[%u], rdevIndex[%u]", phyId, rdmaHandleTmp->rdevIndex); | 1089 | hccp_info("Input parameters: phyId[%u], rdevIndex[%u]", phyId, rdmaHandleTmp->rdevIndex); |
| @@ -1105,14 +1105,14 @@ HCCP_ATTRI_VISI_DEF int RaSetTsqpDepth(void *rdevHandle, unsigned int tempDepth, | |||
| 1105 | CHK_PRT_RETURN(qpNum == NULL, hccp_err("[set][ra_tsqp_depth]qp_num is NULL"), ConverReturnCode(RDMA_OP, -EINVAL)); | 1105 | CHK_PRT_RETURN(qpNum == NULL, hccp_err("[set][ra_tsqp_depth]qp_num is NULL"), ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1106 | 1106 | ||
| 1107 | CHK_PRT_RETURN(tempDepth < RA_MIN_TEMPTH_DEPTH || tempDepth > RA_MAX_TEMPTH_DEPTH, | 1107 | CHK_PRT_RETURN(tempDepth < RA_MIN_TEMPTH_DEPTH || tempDepth > RA_MAX_TEMPTH_DEPTH, |
| 1108 | - hccp_err("[set][ra_tsqp_depth]param error! temp_depth(%u) can not smaller than %d or bigger than %d", tempDepth, | 1108 | + hccp_err("[set][ra_tsqp_depth]param error! temp_depth(%u) cannot be smaller than %d or bigger than %d", |
| 1109 | - RA_MIN_TEMPTH_DEPTH, RA_MAX_TEMPTH_DEPTH), | 1109 | + tempDepth, RA_MIN_TEMPTH_DEPTH, RA_MAX_TEMPTH_DEPTH), |
| 1110 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1110 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1111 | 1111 | ||
| 1112 | phyId = rdmaHandleTmp->rdevInfo.phyId; | 1112 | phyId = rdmaHandleTmp->rdevInfo.phyId; |
| 1113 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 1113 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 1114 | - hccp_err("[set][ra_tsqp_depth]phyId(%u) is invalid! it must greater or equal to 0 and less than %d!", phyId, | 1114 | + hccp_err("[set][ra_tsqp_depth]phyId(%u) is invalid! it must be greater than or equal to 0 and less than %d!", |
| 1115 | - RA_MAX_PHY_ID_NUM), | 1115 | + phyId, RA_MAX_PHY_ID_NUM), |
| 1116 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1116 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1117 | 1117 | ||
| 1118 | hccp_run_info("Input parameters: phyId[%u], rdevIndex[%u], tempDepth[%u]", phyId, rdmaHandleTmp->rdevIndex, | 1118 | hccp_run_info("Input parameters: phyId[%u], rdevIndex[%u], tempDepth[%u]", phyId, rdmaHandleTmp->rdevIndex, |
| @@ -1139,11 +1139,13 @@ HCCP_ATTRI_VISI_DEF int RaQpCreate(void *rdevHandle, int flag, int qpMode, void | |||
| 1139 | 1139 | ||
| 1140 | phyId = rdmaHandleTmp->rdevInfo.phyId; | 1140 | phyId = rdmaHandleTmp->rdevInfo.phyId; |
| 1141 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 1141 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 1142 | - hccp_err("[create][ra_qp]phyId(%u) must greater or equal to 0 and less than %d!", phyId, RA_MAX_PHY_ID_NUM), | 1142 | + hccp_err("[create][ra_qp]phyId(%u) must be greater than or equal to 0 and less than %d!", phyId, |
| 1143 | + RA_MAX_PHY_ID_NUM), | ||
| 1143 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1144 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1144 | 1145 | ||
| 1145 | CHK_PRT_RETURN(qpMode < 0 || qpMode >= RA_RS_ERR_QP_MODE, | 1146 | CHK_PRT_RETURN(qpMode < 0 || qpMode >= RA_RS_ERR_QP_MODE, |
| 1146 | - hccp_err("[create][ra_qp]QP mode(%d) must greater or equal to 0 and less than %d", qpMode, RA_RS_ERR_QP_MODE), | 1147 | + hccp_err("[create][ra_qp]QP mode(%d) must be greater than or equal to 0 and less than %d", qpMode, |
| 1148 | + RA_RS_ERR_QP_MODE), | ||
| 1147 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1149 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1148 | 1150 | ||
| 1149 | hccp_run_info("Input parameters: phyId[%u], flag[%d] qpMode [%d]", phyId, flag, qpMode); | 1151 | hccp_run_info("Input parameters: phyId[%u], flag[%d] qpMode [%d]", phyId, flag, qpMode); |
| @@ -1177,15 +1179,15 @@ HCCP_ATTRI_VISI_DEF int RaQpCreateWithAttrs(void *rdevHandle, struct QpExtAttrs | |||
| 1177 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1179 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1178 | 1180 | ||
| 1179 | CHK_PRT_RETURN(extAttrs->qpMode < 0 || extAttrs->qpMode >= RA_RS_ERR_QP_MODE, | 1181 | CHK_PRT_RETURN(extAttrs->qpMode < 0 || extAttrs->qpMode >= RA_RS_ERR_QP_MODE, |
| 1180 | - hccp_err("[create][ra_qp_with_attrs]QP mode[%d] must greater or equal to 0 and less than %d", extAttrs->qpMode, | 1182 | + hccp_err("[create][ra_qp_with_attrs]QP mode[%d] must be greater than or equal to 0 and less than %d", |
| 1181 | - RA_RS_ERR_QP_MODE), | 1183 | + extAttrs->qpMode, RA_RS_ERR_QP_MODE), |
| 1182 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1184 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1183 | // no need and disallow to set data_plane_flag, set it to default value 0 | 1185 | // no need and disallow to set data_plane_flag, set it to default value 0 |
| 1184 | extAttrs->dataPlaneFlag.value = 0; | 1186 | extAttrs->dataPlaneFlag.value = 0; |
| 1185 | 1187 | ||
| 1186 | phyId = rdmaHandleTmp->rdevInfo.phyId; | 1188 | phyId = rdmaHandleTmp->rdevInfo.phyId; |
| 1187 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 1189 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 1188 | - hccp_err("[create][ra_qp_with_attrs]phyId(%u) must greater or equal to 0 and less than %d!", phyId, | 1190 | + hccp_err("[create][ra_qp_with_attrs]phyId(%u) must be greater than or equal to 0 and less than %d!", phyId, |
| 1189 | RA_MAX_PHY_ID_NUM), | 1191 | RA_MAX_PHY_ID_NUM), |
| 1190 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1192 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1191 | 1193 | ||
| @@ -1229,13 +1231,14 @@ HCCP_ATTRI_VISI_DEF int RaAiQpCreate(void *rdevHandle, struct QpExtAttrs *attrs, | |||
| 1229 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1231 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1230 | 1232 | ||
| 1231 | CHK_PRT_RETURN(attrs->qpMode < 0 || attrs->qpMode >= RA_RS_ERR_QP_MODE, | 1233 | CHK_PRT_RETURN(attrs->qpMode < 0 || attrs->qpMode >= RA_RS_ERR_QP_MODE, |
| 1232 | - hccp_err("[create][ra_ai_qp]QP mode[%d] must greater or equal to 0 and less than %d", attrs->qpMode, | 1234 | + hccp_err("[create][ra_ai_qp]QP mode[%d] must be greater than or equal to 0 and less than %d", attrs->qpMode, |
| 1233 | RA_RS_ERR_QP_MODE), | 1235 | RA_RS_ERR_QP_MODE), |
| 1234 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1236 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1235 | 1237 | ||
| 1236 | phyId = rdmaHandleTmp->rdevInfo.phyId; | 1238 | phyId = rdmaHandleTmp->rdevInfo.phyId; |
| 1237 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 1239 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 1238 | - hccp_err("[create][ra_ai_qp]phyId(%u) must greater or equal to 0 and less than %d!", phyId, RA_MAX_PHY_ID_NUM), | 1240 | + hccp_err("[create][ra_ai_qp]phyId(%u) must be greater than or equal to 0 and less than %d!", phyId, |
| 1241 | + RA_MAX_PHY_ID_NUM), | ||
| 1239 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1242 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1240 | 1243 | ||
| 1241 | hccp_run_info("Input parameters: phyId[%u] qp_mode[%d] cq_attr{%d,%d,%d,%d} cqCstm[%d] " | 1244 | hccp_run_info("Input parameters: phyId[%u] qp_mode[%d] cq_attr{%d,%d,%d,%d} cqCstm[%d] " |
| @@ -1285,7 +1288,7 @@ HCCP_ATTRI_VISI_DEF int RaTypicalQpCreate(void *rdevHandle, int flag, int qpMode | |||
| 1285 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1288 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1286 | 1289 | ||
| 1287 | CHK_PRT_RETURN(qpMode < 0 || qpMode >= RA_RS_ERR_QP_MODE, | 1290 | CHK_PRT_RETURN(qpMode < 0 || qpMode >= RA_RS_ERR_QP_MODE, |
| 1288 | - hccp_err("[create][ra_typical_qp]QP mode(%d) must greater or equal to 0 and less than %d", qpMode, | 1291 | + hccp_err("[create][ra_typical_qp]QP mode(%d) must be greater than or equal to 0 and less than %d", qpMode, |
| 1289 | RA_RS_ERR_QP_MODE), | 1292 | RA_RS_ERR_QP_MODE), |
| 1290 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1293 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1291 | 1294 | ||
| @@ -1495,7 +1498,7 @@ HCCP_ATTRI_VISI_DEF int RaSendWrlistExt(void *qpHandle, struct SendWrlistDataExt | |||
| 1495 | int ret; | 1498 | int ret; |
| 1496 | 1499 | ||
| 1497 | CHK_PRT_RETURN(qpHandle == NULL || wr == NULL || opRsp == NULL || sendNum == 0 || completeNum == NULL, | 1500 | CHK_PRT_RETURN(qpHandle == NULL || wr == NULL || opRsp == NULL || sendNum == 0 || completeNum == NULL, |
| 1498 | - hccp_err("[send][ra_wrlist]qp_handle or wr or op_rsp or complete_num is NULL" | 1501 | + hccp_err("[send][ra_wrlist]qp_handle or wr or op_rsp or complete_num is NULL " |
| 1499 | "or send_num is zero, para error!"), | 1502 | "or send_num is zero, para error!"), |
| 1500 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1503 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1501 | 1504 | ||
| @@ -1597,7 +1600,7 @@ HCCP_ATTRI_VISI_DEF int RaSocketWhiteListAdd(void *socketHandle, struct SocketWl | |||
| 1597 | 1600 | ||
| 1598 | phyId = socketHandleTmp->rdevInfo.phyId; | 1601 | phyId = socketHandleTmp->rdevInfo.phyId; |
| 1599 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 1602 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 1600 | - hccp_err("[add][ra_socket_white_list]phyId(%u) must smaller than %u", phyId, RA_MAX_PHY_ID_NUM), | 1603 | + hccp_err("[add][ra_socket_white_list]phyId(%u) must be smaller than %u", phyId, RA_MAX_PHY_ID_NUM), |
| 1601 | ConverReturnCode(SOCKET_OP, -EINVAL)); | 1604 | ConverReturnCode(SOCKET_OP, -EINVAL)); |
| 1602 | 1605 | ||
| 1603 | hccp_info("Input parameters: phyId[%u], localIp[%s], num[%u]", phyId, localIp, num); | 1606 | hccp_info("Input parameters: phyId[%u], localIp[%s], num[%u]", phyId, localIp, num); |
| @@ -1631,7 +1634,7 @@ HCCP_ATTRI_VISI_DEF int RaSocketWhiteListDel(void *socketHandle, struct SocketWl | |||
| 1631 | 1634 | ||
| 1632 | phyId = socketHandleTmp->rdevInfo.phyId; | 1635 | phyId = socketHandleTmp->rdevInfo.phyId; |
| 1633 | if (phyId >= RA_MAX_PHY_ID_NUM) { | 1636 | if (phyId >= RA_MAX_PHY_ID_NUM) { |
| 1634 | - hccp_err("[del][ra_socket_white_list]phyId(%u) must smaller than %u", phyId, RA_MAX_PHY_ID_NUM); | 1637 | + hccp_err("[del][ra_socket_white_list]phyId(%u) must be smaller than %u", phyId, RA_MAX_PHY_ID_NUM); |
| 1635 | return ConverReturnCode(SOCKET_OP, -EINVAL); | 1638 | return ConverReturnCode(SOCKET_OP, -EINVAL); |
| 1636 | } | 1639 | } |
| 1637 | 1640 | ||
| @@ -1727,8 +1730,8 @@ HCCP_ATTRI_VISI_DEF int RaGetIfnum(struct RaGetIfattr *config, unsigned int *num | |||
| 1727 | ConverReturnCode(OTHERS, -EINVAL)); | 1730 | ConverReturnCode(OTHERS, -EINVAL)); |
| 1728 | 1731 | ||
| 1729 | CHK_PRT_RETURN(config->phyId >= RA_MAX_PHY_ID_NUM, | 1732 | CHK_PRT_RETURN(config->phyId >= RA_MAX_PHY_ID_NUM, |
| 1730 | - hccp_err("[get][ra_ifnum]phyId(%u) is invalid! it must greater or equal to 0 and less than %d!", config->phyId, | 1733 | + hccp_err("[get][ra_ifnum]phyId(%u) is invalid! it must be greater than or equal to 0 and less than %d!", |
| 1731 | - RA_MAX_PHY_ID_NUM), | 1734 | + config->phyId, RA_MAX_PHY_ID_NUM), |
| 1732 | ConverReturnCode(OTHERS, -EINVAL)); | 1735 | ConverReturnCode(OTHERS, -EINVAL)); |
| 1733 | 1736 | ||
| 1734 | hccp_run_info("Input parameters: phyId[%u], nicPosition:[%u], isAll:[%d]", config->phyId, config->nicPosition, | 1737 | hccp_run_info("Input parameters: phyId[%u], nicPosition:[%u], isAll:[%d]", config->phyId, config->nicPosition, |
| @@ -1775,11 +1778,12 @@ HCCP_ATTRI_VISI_DEF int RaGetIfaddrs(struct RaGetIfattr *config, struct Interfac | |||
| 1775 | hccp_err("[get][ra_ifaddrs]config or interface_infos or num is NULL"), ConverReturnCode(OTHERS, -EINVAL)); | 1778 | hccp_err("[get][ra_ifaddrs]config or interface_infos or num is NULL"), ConverReturnCode(OTHERS, -EINVAL)); |
| 1776 | 1779 | ||
| 1777 | CHK_PRT_RETURN(config->phyId >= RA_MAX_PHY_ID_NUM, | 1780 | CHK_PRT_RETURN(config->phyId >= RA_MAX_PHY_ID_NUM, |
| 1778 | - hccp_err("[get][ra_ifaddrs]phyId(%u) is invalid! it must greater or equal to 0 and less than %d!", | 1781 | + hccp_err("[get][ra_ifaddrs]phyId(%u) is invalid! it must be greater than or equal to 0 and less than %d!", |
| 1779 | config->phyId, RA_MAX_PHY_ID_NUM), | 1782 | config->phyId, RA_MAX_PHY_ID_NUM), |
| 1780 | ConverReturnCode(OTHERS, -EINVAL)); | 1783 | ConverReturnCode(OTHERS, -EINVAL)); |
| 1781 | 1784 | ||
| 1782 | - CHK_PRT_RETURN(*num == 0, hccp_err("[get][ra_ifaddrs]interface num(%u) is invalid! it must greater than 0", *num), | 1785 | + CHK_PRT_RETURN(*num == 0, |
| 1786 | + hccp_err("[get][ra_ifaddrs]interface num(%u) is invalid! it must be greater than 0", *num), | ||
| 1783 | ConverReturnCode(OTHERS, -EINVAL)); | 1787 | ConverReturnCode(OTHERS, -EINVAL)); |
| 1784 | 1788 | ||
| 1785 | hccp_run_info("Input parameters: phyId[%u], nicPosition:[%u], isAll[%d], interface num[%u]", config->phyId, | 1789 | hccp_run_info("Input parameters: phyId[%u], nicPosition:[%u], isAll[%d], interface num[%u]", config->phyId, |
| @@ -1811,7 +1815,7 @@ HCCP_ATTRI_VISI_DEF int RaGetInterfaceVersion(unsigned int phyId, unsigned int i | |||
| 1811 | int ret; | 1815 | int ret; |
| 1812 | 1816 | ||
| 1813 | if (interfaceVersion == NULL || phyId >= RA_MAX_PHY_ID_NUM || interfaceOpcode >= RA_RS_EXTER_OP_MAX_NUM) { | 1817 | if (interfaceVersion == NULL || phyId >= RA_MAX_PHY_ID_NUM || interfaceOpcode >= RA_RS_EXTER_OP_MAX_NUM) { |
| 1814 | - hccp_err("[get][ra_interface_version]para is invalid! interface_version is NULL or phyId(%u) is" | 1818 | + hccp_err("[get][ra_interface_version]para is invalid! interface_version is NULL or phyId(%u) is " |
| 1815 | "greater than [%u] or interface_opcode(%u) more than [%u]", | 1819 | "greater than [%u] or interface_opcode(%u) more than [%u]", |
| 1816 | phyId, RA_MAX_PHY_ID_NUM, interfaceOpcode, RA_RS_EXTER_OP_MAX_NUM); | 1820 | phyId, RA_MAX_PHY_ID_NUM, interfaceOpcode, RA_RS_EXTER_OP_MAX_NUM); |
| 1817 | return ConverReturnCode(OTHERS, -EINVAL); | 1821 | return ConverReturnCode(OTHERS, -EINVAL); |
| @@ -1910,14 +1914,14 @@ HCCP_ATTRI_VISI_DEF int RaCqCreate(void *rdevHandle, struct CqAttr *attr) | |||
| 1910 | 1914 | ||
| 1911 | phyId = rdmaHandleTmp->rdevInfo.phyId; | 1915 | phyId = rdmaHandleTmp->rdevInfo.phyId; |
| 1912 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 1916 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 1913 | - hccp_err("[create][ra_cq]phyId(%u) must less than %d!", phyId, RA_MAX_PHY_ID_NUM), | 1917 | + hccp_err("[create][ra_cq]phyId(%u) must be less than %d!", phyId, RA_MAX_PHY_ID_NUM), |
| 1914 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1918 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1915 | 1919 | ||
| 1916 | hccp_run_info("Input parameters: phyId[%u]", phyId); | 1920 | hccp_run_info("Input parameters: phyId[%u]", phyId); |
| 1917 | 1921 | ||
| 1918 | ret = rdmaHandleTmp->rdmaOps->raCqCreate(rdmaHandleTmp, attr); | 1922 | ret = rdmaHandleTmp->rdmaOps->raCqCreate(rdmaHandleTmp, attr); |
| 1919 | CHK_PRT_RETURN(ret != 0 || *attr->ibSendCq == NULL || *attr->ibRecvCq == NULL || *attr->qpContext == NULL, | 1923 | CHK_PRT_RETURN(ret != 0 || *attr->ibSendCq == NULL || *attr->ibRecvCq == NULL || *attr->qpContext == NULL, |
| 1920 | - hccp_err("[create][ra_cq]create cp failed, ret(%d) phyId(%u)", ret, phyId), ConverReturnCode(RDMA_OP, -EINVAL)); | 1924 | + hccp_err("[create][ra_cq]create cq failed, ret(%d) phyId(%u)", ret, phyId), ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1921 | 1925 | ||
| 1922 | return 0; | 1926 | return 0; |
| 1923 | } | 1927 | } |
| @@ -1935,14 +1939,14 @@ HCCP_ATTRI_VISI_DEF int RaCqDestroy(void *rdevHandle, struct CqAttr *attr) | |||
| 1935 | 1939 | ||
| 1936 | phyId = rdmaHandleTmp->rdevInfo.phyId; | 1940 | phyId = rdmaHandleTmp->rdevInfo.phyId; |
| 1937 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, | 1941 | CHK_PRT_RETURN(phyId >= RA_MAX_PHY_ID_NUM, |
| 1938 | - hccp_err("[destroy][ra_cq]phyId(%u) must less than %d!", phyId, RA_MAX_PHY_ID_NUM), | 1942 | + hccp_err("[destroy][ra_cq]phyId(%u) must be less than %d!", phyId, RA_MAX_PHY_ID_NUM), |
| 1939 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1943 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1940 | 1944 | ||
| 1941 | hccp_run_info("Input parameters: phyId[%u]", phyId); | 1945 | hccp_run_info("Input parameters: phyId[%u]", phyId); |
| 1942 | 1946 | ||
| 1943 | ret = rdmaHandleTmp->rdmaOps->raCqDestroy(rdmaHandleTmp, attr); | 1947 | ret = rdmaHandleTmp->rdmaOps->raCqDestroy(rdmaHandleTmp, attr); |
| 1944 | CHK_PRT_RETURN(ret != 0 || *attr->ibSendCq == NULL || *attr->ibRecvCq == NULL || *attr->qpContext == NULL, | 1948 | CHK_PRT_RETURN(ret != 0 || *attr->ibSendCq == NULL || *attr->ibRecvCq == NULL || *attr->qpContext == NULL, |
| 1945 | - hccp_err("[destroy][ra_cq]destroy cp failed, ret(%d) phyId(%u)", ret, phyId), | 1949 | + hccp_err("[destroy][ra_cq]destroy cq failed, ret(%d) phyId(%u)", ret, phyId), |
| 1946 | ConverReturnCode(RDMA_OP, -EINVAL)); | 1950 | ConverReturnCode(RDMA_OP, -EINVAL)); |
| 1947 | 1951 | ||
| 1948 | return 0; | 1952 | return 0; |
| @@ -1968,7 +1972,7 @@ HCCP_ATTRI_VISI_DEF int RaNormalQpCreate(void *rdevHandle, struct ibv_qp_init_at | |||
| 1968 | 1972 | ||
| 1969 | phyId = rdmaHandleTmp->rdevInfo.phyId; | 1973 | phyId = rdmaHandleTmp->rdevInfo.phyId; |
| 1970 | if (phyId >= RA_MAX_PHY_ID_NUM) { | 1974 | if (phyId >= RA_MAX_PHY_ID_NUM) { |
| 1971 | - hccp_err("[create][ra_normal_qp]phyId(%u) must less than %d!", phyId, RA_MAX_PHY_ID_NUM); | 1975 | + hccp_err("[create][ra_normal_qp]phyId(%u) must be less than %d!", phyId, RA_MAX_PHY_ID_NUM); |
| 1972 | return ConverReturnCode(RDMA_OP, -EINVAL); | 1976 | return ConverReturnCode(RDMA_OP, -EINVAL); |
| 1973 | } | 1977 | } |
| 1974 | 1978 | ||
| @@ -2122,7 +2126,7 @@ HCCP_ATTRI_VISI_DEF int RaGetCqeErrInfo(unsigned int phyId, struct CqeErrInfo *i | |||
| 2122 | } | 2126 | } |
| 2123 | 2127 | ||
| 2124 | if (phyId >= RA_MAX_PHY_ID_NUM) { | 2128 | if (phyId >= RA_MAX_PHY_ID_NUM) { |
| 2125 | - hccp_err("[ra_get_cqe_err_info]phyId(%u) must greater or equal to 0 and less than %d!", phyId, | 2129 | + hccp_err("[ra_get_cqe_err_info]phyId(%u) must be greater than or equal to 0 and less than %d!", phyId, |
| 2126 | RA_MAX_PHY_ID_NUM); | 2130 | RA_MAX_PHY_ID_NUM); |
| 2127 | return ConverReturnCode(RDMA_OP, -EINVAL); | 2131 | return ConverReturnCode(RDMA_OP, -EINVAL); |
| 2128 | } | 2132 | } |
| @@ -2373,7 +2377,8 @@ HCCP_ATTRI_VISI_DEF int RaTypicalQpModify(void *qpHandle, struct TypicalQp *loca | |||
| 2373 | 2377 | ||
| 2374 | phyId = raQpHandle->phyId; | 2378 | phyId = raQpHandle->phyId; |
| 2375 | if (phyId >= RA_MAX_PHY_ID_NUM) { | 2379 | if (phyId >= RA_MAX_PHY_ID_NUM) { |
| 2376 | - hccp_err("[modify][ra_qp]phyId(%u) must greater or equal to 0 and less than %d!", phyId, RA_MAX_PHY_ID_NUM); | 2380 | + hccp_err("[modify][ra_qp]phyId(%u) must be greater than or equal to 0 and less than %d!", phyId, |
| 2381 | + RA_MAX_PHY_ID_NUM); | ||
| 2377 | return ConverReturnCode(RDMA_OP, -EINVAL); | 2382 | return ConverReturnCode(RDMA_OP, -EINVAL); |
| 2378 | } | 2383 | } |
| 2379 | 2384 | ||
| @@ -2424,7 +2429,7 @@ HCCP_ATTRI_VISI_DEF int RaSocketGetVnicIpInfos(unsigned int phyId, enum IdType t | |||
| 2424 | } | 2429 | } |
| 2425 | 2430 | ||
| 2426 | if (phyId >= RA_MAX_PHY_ID_NUM) { | 2431 | if (phyId >= RA_MAX_PHY_ID_NUM) { |
| 2427 | - hccp_err("[get][vnic_ip]phyId(%u) is invalid! it must greater or equal to 0 and less than %d!", phyId, | 2432 | + hccp_err("[get][vnic_ip]phyId(%u) is invalid! it must be greater than or equal to 0 and less than %d!", phyId, |
| 2428 | RA_MAX_PHY_ID_NUM); | 2433 | RA_MAX_PHY_ID_NUM); |
| 2429 | return ConverReturnCode(OTHERS, -EINVAL); | 2434 | return ConverReturnCode(OTHERS, -EINVAL); |
| 2430 | } | 2435 | } |
| @@ -2475,7 +2480,7 @@ HCCP_ATTRI_VISI_DEF int RaQpBatchModify(void *rdmaHandle, void *qpHandle[], unsi | |||
| 2475 | raRdmaHandle = (struct RaRdmaHandle *)rdmaHandle; | 2480 | raRdmaHandle = (struct RaRdmaHandle *)rdmaHandle; |
| 2476 | phyId = raRdmaHandle->rdevInfo.phyId; | 2481 | phyId = raRdmaHandle->rdevInfo.phyId; |
| 2477 | if (phyId >= RA_MAX_PHY_ID_NUM || raRdmaHandle->rdmaOps == NULL || raRdmaHandle->rdmaOps->raQpBatchModify == NULL) { | 2482 | if (phyId >= RA_MAX_PHY_ID_NUM || raRdmaHandle->rdmaOps == NULL || raRdmaHandle->rdmaOps->raQpBatchModify == NULL) { |
| 2478 | - hccp_err("[modify][ra_qp]phyId(%u) must greater or equal to 0 and less than %d or ops is NULL or " | 2483 | + hccp_err("[modify][ra_qp]phyId(%u) must be greater than or equal to 0 and less than %d or ops is NULL or " |
| 2479 | "ra_rdma_handle->rdma_ops->ra_qp_batch_modify is NULL!", | 2484 | "ra_rdma_handle->rdma_ops->ra_qp_batch_modify is NULL!", |
| 2480 | phyId, RA_MAX_PHY_ID_NUM); | 2485 | phyId, RA_MAX_PHY_ID_NUM); |
| 2481 | return ConverReturnCode(RDMA_OP, -EINVAL); | 2486 | return ConverReturnCode(RDMA_OP, -EINVAL); |
| @@ -195,7 +195,7 @@ static int HdcSendRecvPktSend(struct drvHdcMsg *pMsgSnd, char *sendRcvBuf, unsig | |||
| 195 | CHK_PRT_RETURN(ret != 0, hccp_err("[send][hdc_send_recv_pkt]HDC send err ret(%d)", ret), ret); | 195 | CHK_PRT_RETURN(ret != 0, hccp_err("[send][hdc_send_recv_pkt]HDC send err ret(%d)", ret), ret); |
| 196 | 196 | ||
| 197 | ret = RA_HDC_OPS.reuseMsg(pMsgSnd); | 197 | ret = RA_HDC_OPS.reuseMsg(pMsgSnd); |
| 198 | - CHK_PRT_RETURN(ret != 0, hccp_err("[send][hdc_send_recv_pkt]HDC reuser msg err ret(%d)", ret), ret); | 198 | + CHK_PRT_RETURN(ret != 0, hccp_err("[send][hdc_send_recv_pkt]HDC reuse msg err ret(%d)", ret), ret); |
| 199 | 199 | ||
| 200 | *pMsgRcv = pMsgSnd; | 200 | *pMsgRcv = pMsgSnd; |
| 201 | return 0; | 201 | return 0; |
| @@ -369,7 +369,7 @@ STATIC int HdcSendRecvPktRecvCheck(int rcvBufLen, unsigned int outDataLen, struc | |||
| 369 | RA_HDC_OPS.freeMsg(pMsgRcv); | 369 | RA_HDC_OPS.freeMsg(pMsgRcv); |
| 370 | return -EPERM; | 370 | return -EPERM; |
| 371 | } | 371 | } |
| 372 | - hccp_err("[check][hdc_send_recv_pkt_recv]date len err out_data_len(%d) != rcv_buf_len(%d) ", outDataLen, | 372 | + hccp_err("[check][hdc_send_recv_pkt_recv]data len err out_data_len(%d) != rcv_buf_len(%d) ", outDataLen, |
| 373 | rcvBufLen); | 373 | rcvBufLen); |
| 374 | RA_HDC_OPS.freeMsg(pMsgRcv); | 374 | RA_HDC_OPS.freeMsg(pMsgRcv); |
| 375 | return -EPIPE; | 375 | return -EPIPE; |
| @@ -148,7 +148,7 @@ DL_ATTRI_VISI_DEF int RaHdcRdmaLiteApiInit(void) | |||
| 148 | ret = RaHdcOpenRdmaLiteSo(); | 148 | ret = RaHdcOpenRdmaLiteSo(); |
| 149 | if (ret) { | 149 | if (ret) { |
| 150 | pthread_mutex_unlock(&gRdmaLiteApiLock); | 150 | pthread_mutex_unlock(&gRdmaLiteApiLock); |
| 151 | - hccp_err("HccpDlopen[libascend_rdma_lite.so]" | 151 | + hccp_err("HccpDlopen[libascend_rdma_lite.so] " |
| 152 | "failed! ret=[%d][%s]. Please check rdma lite driver has been installed.", | 152 | "failed! ret=[%d][%s]. Please check rdma lite driver has been installed.", |
| 153 | ret, dlerror()); | 153 | ret, dlerror()); |
| 154 | return ret; | 154 | return ret; |
| @@ -249,7 +249,7 @@ STATIC int RsIbverbsApiInit(void) | |||
| 249 | 249 | ||
| 250 | ret = RsOpenIbverbsSo(); | 250 | ret = RsOpenIbverbsSo(); |
| 251 | CHK_PRT_RETURN(ret, | 251 | CHK_PRT_RETURN(ret, |
| 252 | - hccp_err("HccpDlopen[libibverbs.so or libibverbs.so.1] failed! ret=[%d]," | 252 | + hccp_err("HccpDlopen[libibverbs.so or libibverbs.so.1] failed! ret=[%d], " |
| 253 | "Please check network adapter driver has been installed", | 253 | "Please check network adapter driver has been installed", |
| 254 | ret), | 254 | ret), |
| 255 | ret); | 255 | ret); |
| @@ -429,7 +429,7 @@ STATIC int RsRoceUserApiInit(void) | |||
| 429 | 429 | ||
| 430 | ret = RsOpenRoceUserSo(&type); | 430 | ret = RsOpenRoceUserSo(&type); |
| 431 | CHK_PRT_RETURN(ret != 0, | 431 | CHK_PRT_RETURN(ret != 0, |
| 432 | - hccp_err("HccpDlopen[libhns-rdmav17.so or libhns-rdmav25.so or libhrn0-rdmav17.so]" | 432 | + hccp_err("HccpDlopen[libhns-rdmav17.so or libhns-rdmav25.so or libhrn0-rdmav17.so] " |
| 433 | "failed! ret=[%d]. Please check network adapter driver has been installed.", | 433 | "failed! ret=[%d]. Please check network adapter driver has been installed.", |
| 434 | ret), | 434 | ret), |
| 435 | ret); | 435 | ret); |
| @@ -576,7 +576,7 @@ uint16_t CcuTaskException::GetCcuCKEValue(int32_t deviceId, uint32_t dieId, uint | |||
| 576 | 576 | ||
| 577 | uint64_t ckeVal{0}; | 577 | uint64_t ckeVal{0}; |
| 578 | s32 sret = memcpy_s(&ckeVal, sizeof(ckeVal), outBuff.data.dataInfo.dataArray, inBuff.data.dataInfo.dataLen); | 578 | s32 sret = memcpy_s(&ckeVal, sizeof(ckeVal), outBuff.data.dataInfo.dataArray, inBuff.data.dataInfo.dataLen); |
| 579 | - CHK_PRT_RET(sret != EOK, HCCL_ERROR("[%s]memcpy failed. errorno[%d]:", __func__, sret), INVALID_U16); | 579 | + CHK_PRT_RET(sret != EOK, HCCL_ERROR("[%s]memcpy failed. errno[%d]", __func__, sret), INVALID_U16); |
| 580 | return static_cast<uint16_t>(ckeVal); | 580 | return static_cast<uint16_t>(ckeVal); |
| 581 | } | 581 | } |
| 582 | 582 | ||
| @@ -696,7 +696,7 @@ uint64_t CcuTaskException::GetCcuGSAValue(int32_t deviceId, uint32_t dieId, uint | |||
| 696 | DFX_INVALID_U64); | 696 | DFX_INVALID_U64); |
| 697 | 697 | ||
| 698 | auto sret = memcpy_s(&gsaVal, sizeof(gsaVal), outBuff.data.dataInfo.dataArray, inBuff.data.dataInfo.dataLen); | 698 | auto sret = memcpy_s(&gsaVal, sizeof(gsaVal), outBuff.data.dataInfo.dataArray, inBuff.data.dataInfo.dataLen); |
| 699 | - CHK_PRT_RET(sret != EOK, HCCL_ERROR("[%s]memcpy failed. errorno[%d]:", __func__, sret), DFX_INVALID_U64); | 699 | + CHK_PRT_RET(sret != EOK, HCCL_ERROR("[%s]memcpy failed. errno[%d]", __func__, sret), DFX_INVALID_U64); |
| 700 | return gsaVal; | 700 | return gsaVal; |
| 701 | } | 701 | } |
| 702 | 702 | ||
| @@ -724,7 +724,7 @@ uint64_t CcuTaskException::GetCcuXnValue(int32_t deviceId, uint32_t dieId, uint3 | |||
| 724 | DFX_INVALID_U64); | 724 | DFX_INVALID_U64); |
| 725 | 725 | ||
| 726 | auto sret = memcpy_s(&xnVal, sizeof(xnVal), outBuff.data.dataInfo.dataArray, inBuff.data.dataInfo.dataLen); | 726 | auto sret = memcpy_s(&xnVal, sizeof(xnVal), outBuff.data.dataInfo.dataArray, inBuff.data.dataInfo.dataLen); |
| 727 | - CHK_PRT_RET(sret != EOK, HCCL_ERROR("[%s]memcpy failed. errorno[%d]:", __func__, sret), DFX_INVALID_U64); | 727 | + CHK_PRT_RET(sret != EOK, HCCL_ERROR("[%s]memcpy failed. errno[%d]", __func__, sret), DFX_INVALID_U64); |
| 728 | return xnVal; | 728 | return xnVal; |
| 729 | } | 729 | } |
| 730 | 730 | ||
| @@ -750,7 +750,7 @@ void CcuTaskException::GenErrorInfoRemPostSem( | |||
| 750 | auto sret = memset_s( | 750 | auto sret = memset_s( |
| 751 | errorMsg.msg.waitSignal.channelId, sizeof(errorMsg.msg.waitSignal.channelId), 0xFF, | 751 | errorMsg.msg.waitSignal.channelId, sizeof(errorMsg.msg.waitSignal.channelId), 0xFF, |
| 752 | sizeof(errorMsg.msg.waitSignal.channelId)); | 752 | sizeof(errorMsg.msg.waitSignal.channelId)); |
| 753 | - CHK_PRT_RET(sret != EOK, HCCL_ERROR("[%s]memset_s failed. errorno[%d]:", __func__, sret), ); | 753 | + CHK_PRT_RET(sret != EOK, HCCL_ERROR("[%s]memset_s failed. errno[%d]", __func__, sret), ); |
| 754 | errorMsg.msg.waitSignal.channelId[0] = channelId; | 754 | errorMsg.msg.waitSignal.channelId[0] = channelId; |
| 755 | 755 | ||
| 756 | errorInfo.push_back(errorMsg); | 756 | errorInfo.push_back(errorMsg); |
| @@ -780,7 +780,7 @@ void CcuTaskException::GenErrorInfoRemWaitSem( | |||
| 780 | auto sret = memset_s( | 780 | auto sret = memset_s( |
| 781 | errorMsg.msg.waitSignal.channelId, sizeof(errorMsg.msg.waitSignal.channelId), 0xFF, | 781 | errorMsg.msg.waitSignal.channelId, sizeof(errorMsg.msg.waitSignal.channelId), 0xFF, |
| 782 | sizeof(errorMsg.msg.waitSignal.channelId)); | 782 | sizeof(errorMsg.msg.waitSignal.channelId)); |
| 783 | - CHK_PRT_RET(sret != EOK, HCCL_ERROR("[%s]memset_s failed. errorno[%d]:", __func__, sret), ); | 783 | + CHK_PRT_RET(sret != EOK, HCCL_ERROR("[%s]memset_s failed. errno[%d]", __func__, sret), ); |
| 784 | 784 | ||
| 785 | errorMsg.msg.waitSignal.channelId[0] = channelId; | 785 | errorMsg.msg.waitSignal.channelId[0] = channelId; |
| 786 | 786 | ||
| @@ -814,7 +814,7 @@ void CcuTaskException::GenErrorInfoRemPostVar( | |||
| 814 | auto sret = memset_s( | 814 | auto sret = memset_s( |
| 815 | errorMsg.msg.waitSignal.channelId, sizeof(errorMsg.msg.waitSignal.channelId), 0xFF, | 815 | errorMsg.msg.waitSignal.channelId, sizeof(errorMsg.msg.waitSignal.channelId), 0xFF, |
| 816 | sizeof(errorMsg.msg.waitSignal.channelId)); | 816 | sizeof(errorMsg.msg.waitSignal.channelId)); |
| 817 | - CHK_PRT_RET(sret != EOK, HCCL_ERROR("[%s]memset_s failed. errorno[%d]:", __func__, sret), ); | 817 | + CHK_PRT_RET(sret != EOK, HCCL_ERROR("[%s]memset_s failed. errno[%d]", __func__, sret), ); |
| 818 | 818 | ||
| 819 | errorMsg.msg.waitSignal.channelId[0] = channelId; | 819 | errorMsg.msg.waitSignal.channelId[0] = channelId; |
| 820 | errorMsg.msg.waitSignal.paramId = rmtXnId; | 820 | errorMsg.msg.waitSignal.paramId = rmtXnId; |
| @@ -1028,7 +1028,7 @@ void CcuTaskException::GenErrorInfoBufReduce( | |||
| 1028 | auto sret = memset_s( | 1028 | auto sret = memset_s( |
| 1029 | errorMsg.msg.bufReduce.bufIds, sizeof(errorMsg.msg.bufReduce.bufIds), 0xFF, | 1029 | errorMsg.msg.bufReduce.bufIds, sizeof(errorMsg.msg.bufReduce.bufIds), 0xFF, |
| 1030 | sizeof(errorMsg.msg.bufReduce.bufIds)); | 1030 | sizeof(errorMsg.msg.bufReduce.bufIds)); |
| 1031 | - CHK_PRT_RET(sret != EOK, HCCL_ERROR("[%s]memset failed. errorno[%d]:", __func__, sret), ); | 1031 | + CHK_PRT_RET(sret != EOK, HCCL_ERROR("[%s]memset failed. errno[%d]", __func__, sret), ); |
| 1032 | for (uint32_t i = 0; i < buffs.size() && i < BUF_REDUCE_ID_SIZE; ++i) { | 1032 | for (uint32_t i = 0; i < buffs.size() && i < BUF_REDUCE_ID_SIZE; ++i) { |
| 1033 | errorMsg.msg.bufReduce.bufIds[i] = GetMSIdPerDie(buffs[i].Id()); | 1033 | errorMsg.msg.bufReduce.bufIds[i] = GetMSIdPerDie(buffs[i].Id()); |
| 1034 | } | 1034 | } |
| @@ -92,7 +92,7 @@ void RankTableInfo::Check() | |||
| 92 | true, "EI0014", std::vector<std::string>({"value", "variable", "expect"}), | 92 | true, "EI0014", std::vector<std::string>({"value", "variable", "expect"}), |
| 93 | std::vector<std::string>( | 93 | std::vector<std::string>( |
| 94 | {std::to_string(rank.replacedLocalId), "replacedLocalId", | 94 | {std::to_string(rank.replacedLocalId), "replacedLocalId", |
| 95 | - "replacedLocalId equal to locaId[" + std::to_string(rank.localId) + "]"})); | 95 | + "replacedLocalId equal to localId[" + std::to_string(rank.localId) + "]"})); |
| 96 | THROW<InvalidParamsException>(StringFormat( | 96 | THROW<InvalidParamsException>(StringFormat( |
| 97 | "[Parse][ClusterInfo][RankTableInfo::Check] " | 97 | "[Parse][ClusterInfo][RankTableInfo::Check] " |
| 98 | "failed with replacedLocalId[%u] not equal to localId[%u].", | 98 | "failed with replacedLocalId[%u] not equal to localId[%u].", |
| @@ -136,7 +136,7 @@ void RankTableInfo::Check() | |||
| 136 | true, "EI0014", std::vector<std::string>({"value", "variable", "expect"}), | 136 | true, "EI0014", std::vector<std::string>({"value", "variable", "expect"}), |
| 137 | std::vector<std::string>( | 137 | std::vector<std::string>( |
| 138 | {std::to_string(recordedReplaceLocalId), "recordedReplacedLocalId", | 138 | {std::to_string(recordedReplaceLocalId), "recordedReplacedLocalId", |
| 139 | - "failed with configuring same local_id with replaced one simutaneously"})); | 139 | + "failed with configuring same local_id with replaced one simultaneously"})); |
| 140 | THROW<InvalidParamsException>(StringFormat( | 140 | THROW<InvalidParamsException>(StringFormat( |
| 141 | "[Parse][ClusterInfo][RankTableInfo::%s] failed with configuring " | 141 | "[Parse][ClusterInfo][RankTableInfo::%s] failed with configuring " |
| 142 | "same local_id[%u] with replaced one simutaneously", | 142 | "same local_id[%u] with replaced one simutaneously", |
| @@ -123,7 +123,7 @@ void EdgeInfo::DeserializeEndpoint(const nlohmann::json& edgeInfoJson) | |||
| 123 | } | 123 | } |
| 124 | } else { | 124 | } else { |
| 125 | if (edgeInfoJson.contains("local_b") || edgeInfoJson.contains("local_b_ports")) { | 125 | if (edgeInfoJson.contains("local_b") || edgeInfoJson.contains("local_b_ports")) { |
| 126 | - HCCL_WARNING("[EdgeInfo::%s] local_b and local_b_ports are not need when PEER2NET", __func__); | 126 | + HCCL_WARNING("[EdgeInfo::%s] local_b and local_b_ports are not needed when PEER2NET", __func__); |
| 127 | } | 127 | } |
| 128 | } | 128 | } |
| 129 | 129 | ||
| @@ -40,7 +40,7 @@ void RankInfoDispather::BroadcastRankTable( | |||
| 40 | PrepareResource(connectSockets, clusterInfo, failedAgentIdList, step); | 40 | PrepareResource(connectSockets, clusterInfo, failedAgentIdList, step); |
| 41 | ProcessSend(); | 41 | ProcessSend(); |
| 42 | HCCL_INFO( | 42 | HCCL_INFO( |
| 43 | - "[RankInfoDispather::%s] broadcast topoinfo success, rankNum[%u], threadNum[%u]", __func__, rankNum_, | 43 | + "[RankInfoDispatcher::%s] broadcast topoinfo success, rankNum[%u], threadNum[%u]", __func__, rankNum_, |
| 44 | threadNum_); | 44 | threadNum_); |
| 45 | } | 45 | } |
| 46 | 46 | ||
| @@ -168,7 +168,8 @@ HcclResult ChannelManager::PrepareHandleArray( | |||
| 168 | HcclResult ChannelManager::IsChannelExist(ChannelHandle channel) | 168 | HcclResult ChannelManager::IsChannelExist(ChannelHandle channel) |
| 169 | { | 169 | { |
| 170 | CHK_PRT_RET( | 170 | CHK_PRT_RET( |
| 171 | - (keyMap_.find(channel) == keyMap_.end()), HCCL_ERROR("[%s]ChannelHandle is not exist.", __func__), HCCL_E_PARA); | 171 | + (keyMap_.find(channel) == keyMap_.end()), HCCL_ERROR("[%s]ChannelHandle does not exist.", __func__), |
| 172 | + HCCL_E_PARA); | ||
| 172 | HCCL_INFO( | 173 | HCCL_INFO( |
| 173 | "[%s]ChannelHandle exist, ChannelHandle[%llu], channelKey[%s]", __func__, channel, keyMap_[channel].c_str()); | 174 | "[%s]ChannelHandle exist, ChannelHandle[%llu], channelKey[%s]", __func__, channel, keyMap_[channel].c_str()); |
| 174 | return HCCL_SUCCESS; | 175 | return HCCL_SUCCESS; |
| @@ -177,7 +178,8 @@ HcclResult ChannelManager::IsChannelExist(ChannelHandle channel) | |||
| 177 | HcclResult ChannelManager::UnregisterHandle(ChannelHandle channel) | 178 | HcclResult ChannelManager::UnregisterHandle(ChannelHandle channel) |
| 178 | { | 179 | { |
| 179 | CHK_PRT_RET( | 180 | CHK_PRT_RET( |
| 180 | - (keyMap_.find(channel) == keyMap_.end()), HCCL_ERROR("[%s]ChannelHandle is not exist.", __func__), HCCL_E_PARA); | 181 | + (keyMap_.find(channel) == keyMap_.end()), HCCL_ERROR("[%s]ChannelHandle does not exist.", __func__), |
| 182 | + HCCL_E_PARA); | ||
| 181 | 183 | ||
| 182 | channelHandleMap_.erase(keyMap_[channel]); | 184 | channelHandleMap_.erase(keyMap_[channel]); |
| 183 | keyMap_.erase(channel); | 185 | keyMap_.erase(channel); |
| @@ -208,7 +210,7 @@ HcclResult ChannelManager::GetHostChannel(ChannelHandle channel, ChannelHandle& | |||
| 208 | if (engineMap_[channel] == COMM_ENGINE_AICPU || engineMap_[channel] == COMM_ENGINE_AICPU_TS) { | 210 | if (engineMap_[channel] == COMM_ENGINE_AICPU || engineMap_[channel] == COMM_ENGINE_AICPU_TS) { |
| 209 | CHK_PRT_RET( | 211 | CHK_PRT_RET( |
| 210 | (channelD2HMap_.find(channel) == channelD2HMap_.end()), | 212 | (channelD2HMap_.find(channel) == channelD2HMap_.end()), |
| 211 | - HCCL_ERROR("[%s]device channel handle has not existed in channelD2HMap_.", __func__), HCCL_E_PARA); | 213 | + HCCL_ERROR("[%s]device channel handle does not exist in channelD2HMap_.", __func__), HCCL_E_PARA); |
| 212 | hostChannel = channelD2HMap_[channel]; | 214 | hostChannel = channelD2HMap_[channel]; |
| 213 | } else { | 215 | } else { |
| 214 | hostChannel = channel; | 216 | hostChannel = channel; |
| @@ -246,7 +248,7 @@ HcclResult ChannelManager::CheckNotifyOrQPMaxNum(u64& existNum, const u64& MaxNu | |||
| 246 | HcclResult ChannelManager::CreateWorkSpace(u64 size, DeviceMem& buffer) const | 248 | HcclResult ChannelManager::CreateWorkSpace(u64 size, DeviceMem& buffer) const |
| 247 | { | 249 | { |
| 248 | CHK_PRT_RET( | 250 | CHK_PRT_RET( |
| 249 | - size == 0, HCCL_INFO("[Create][WorkSpace]work space size is zero. not need to malloc memory"), HCCL_SUCCESS); | 251 | + size == 0, HCCL_INFO("[Create][WorkSpace]work space size is zero, no need to allocate memory"), HCCL_SUCCESS); |
| 250 | 252 | ||
| 251 | CHK_PRT_RET( | 253 | CHK_PRT_RET( |
| 252 | (size > ULONG_MAX), HCCL_ERROR("[Create][WorkSpace]work space size is greater than %llu", ULONG_MAX), | 254 | (size > ULONG_MAX), HCCL_ERROR("[Create][WorkSpace]work space size is greater than %llu", ULONG_MAX), |
| @@ -257,7 +259,7 @@ HcclResult ChannelManager::CreateWorkSpace(u64 size, DeviceMem& buffer) const | |||
| 257 | CHK_PRT_RET( | 259 | CHK_PRT_RET( |
| 258 | size > 0 && !buffer, | 260 | size > 0 && !buffer, |
| 259 | HCCL_ERROR( | 261 | HCCL_ERROR( |
| 260 | - "[Create][WorkSpace]Create work space size[%llu] fail," | 262 | + "[Create][WorkSpace]Create work space failed, size[%llu], " |
| 261 | "please check workspace size.", | 263 | "please check workspace size.", |
| 262 | size), | 264 | size), |
| 263 | HCCL_E_PTR); | 265 | HCCL_E_PTR); |
| @@ -269,7 +271,7 @@ HcclResult ChannelManager::AllocAndClearHostMem(u64 size, std::shared_ptr<HostMe | |||
| 269 | { | 271 | { |
| 270 | CHK_PRT_RET( | 272 | CHK_PRT_RET( |
| 271 | size == 0, | 273 | size == 0, |
| 272 | - HCCL_INFO("[ChannelManager][AllocAndClearHostMem] host memory size is zero. not need to malloc memory"), | 274 | + HCCL_INFO("[ChannelManager][AllocAndClearHostMem] host memory size is zero, no need to allocate memory"), |
| 273 | HCCL_SUCCESS); | 275 | HCCL_SUCCESS); |
| 274 | 276 | ||
| 275 | CHK_PRT_RET( | 277 | CHK_PRT_RET( |
| @@ -283,8 +285,8 @@ HcclResult ChannelManager::AllocAndClearHostMem(u64 size, std::shared_ptr<HostMe | |||
| 283 | CHK_PRT_RET( | 285 | CHK_PRT_RET( |
| 284 | size > 0 && !bufferPtr.get()->ptr(), | 286 | size > 0 && !bufferPtr.get()->ptr(), |
| 285 | HCCL_ERROR( | 287 | HCCL_ERROR( |
| 286 | - "[ChannelManager][AllocAndClearHostMem]host memory space size[%llu] fail," | 288 | + "[ChannelManager][AllocAndClearHostMem]alloc host memory failed, size[%llu], " |
| 287 | - "please check workspace size.", | 289 | + "please check host memory size.", |
| 288 | size), | 290 | size), |
| 289 | HCCL_E_PTR); | 291 | HCCL_E_PTR); |
| 290 | CHK_SAFETY_FUNC_RET(memset_s(bufferPtr.get()->ptr(), size, 0, size)); | 292 | CHK_SAFETY_FUNC_RET(memset_s(bufferPtr.get()->ptr(), size, 0, size)); |
| @@ -295,7 +297,7 @@ template <typename T> | |||
| 295 | HcclResult ChannelManager::CopyVectorToDeviceMem(const u64 len, DeviceMem& dstDeviceMem, const std::vector<T>& srcVec) | 297 | HcclResult ChannelManager::CopyVectorToDeviceMem(const u64 len, DeviceMem& dstDeviceMem, const std::vector<T>& srcVec) |
| 296 | { | 298 | { |
| 297 | CHK_PRT_RET( | 299 | CHK_PRT_RET( |
| 298 | - len == 0, HCCL_INFO("[ChannelManager][CopyVectorToDeviceMem] space size is zero. not need to malloc memory"), | 300 | + len == 0, HCCL_INFO("[ChannelManager][CopyVectorToDeviceMem] space size is zero, no need to allocate memory"), |
| 299 | HCCL_SUCCESS); | 301 | HCCL_SUCCESS); |
| 300 | 302 | ||
| 301 | CHK_PRT_RET( | 303 | CHK_PRT_RET( |
| @@ -560,7 +560,7 @@ HcclResult MultiDeterPipeline::PrepareTopoInfo(const SubCommInfo& level0CommInfo | |||
| 560 | "[%s] topoInfo: userRank[%u], intraRankId[%u], intraRankSize[%u], serverId[%u], interRankSize[%u]", __func__, | 560 | "[%s] topoInfo: userRank[%u], intraRankId[%u], intraRankSize[%u], serverId[%u], interRankSize[%u]", __func__, |
| 561 | userRank_, intraRankId_, intraRankSize_, serverId_, serverSize_); | 561 | userRank_, intraRankId_, intraRankSize_, serverId_, serverSize_); |
| 562 | HCCL_INFO( | 562 | HCCL_INFO( |
| 563 | - "[%s] topoInfo: severLinksNum[%zu], intraLinksNum[%zu]", __func__, serverLinks_.size(), intraLinks_.size()); | 563 | + "[%s] topoInfo: serverLinksNum[%zu], intraLinksNum[%zu]", __func__, serverLinks_.size(), intraLinks_.size()); |
| 564 | InitAlltoallRecvBlockIdxMap(); | 564 | InitAlltoallRecvBlockIdxMap(); |
| 565 | return HCCL_SUCCESS; | 565 | return HCCL_SUCCESS; |
| 566 | } | 566 | } |
| @@ -58,7 +58,7 @@ HcclResult AllGatherGraphPipeline::Prepare( | |||
| 58 | streamNotifyMain_ = notifyMain; | 58 | streamNotifyMain_ = notifyMain; |
| 59 | if (streamNotifyMain_.size() < intraRankSize_) { | 59 | if (streamNotifyMain_.size() < intraRankSize_) { |
| 60 | HCCL_ERROR( | 60 | HCCL_ERROR( |
| 61 | - "[AllGatherGraphPipeline][Prepare]rank[%u] streamNotifyMain_ size[%u] error, is smaller than," | 61 | + "[AllGatherGraphPipeline][Prepare]rank[%u] streamNotifyMain_ size[%u] is smaller than " |
| 62 | "intraRankSize_[%u]", | 62 | "intraRankSize_[%u]", |
| 63 | userRank_, streamNotifyMain_.size(), intraRankSize_); | 63 | userRank_, streamNotifyMain_.size(), intraRankSize_); |
| 64 | return HCCL_E_INTERNAL; | 64 | return HCCL_E_INTERNAL; |
| @@ -66,7 +66,7 @@ HcclResult AllGatherGraphPipeline::Prepare( | |||
| 66 | streamNotifySub_ = notifySub; | 66 | streamNotifySub_ = notifySub; |
| 67 | if (streamNotifySub_.size() < intraRankSize_) { | 67 | if (streamNotifySub_.size() < intraRankSize_) { |
| 68 | HCCL_ERROR( | 68 | HCCL_ERROR( |
| 69 | - "[AllGatherGraphPipeline][Prepare]rank[%u] streamNotifySub_ size[%u] error, is smaller than," | 69 | + "[AllGatherGraphPipeline][Prepare]rank[%u] streamNotifySub_ size[%u] is smaller than " |
| 70 | "intraRankSize_[%u]", | 70 | "intraRankSize_[%u]", |
| 71 | userRank_, streamNotifySub_.size(), intraRankSize_); | 71 | userRank_, streamNotifySub_.size(), intraRankSize_); |
| 72 | return HCCL_E_INTERNAL; | 72 | return HCCL_E_INTERNAL; |
| @@ -168,7 +168,7 @@ HcclResult AllGatherGraphPipeline::RunAsync() | |||
| 168 | UserMemType::OUTPUT_MEM, serverOffsetByte, static_cast<u8*>(dmaMem_[1].ptr()) + serverOffsetByte, | 168 | UserMemType::OUTPUT_MEM, serverOffsetByte, static_cast<u8*>(dmaMem_[1].ptr()) + serverOffsetByte, |
| 169 | memSliceSize, subStream_[0])); | 169 | memSliceSize, subStream_[0])); |
| 170 | HCCL_DEBUG( | 170 | HCCL_DEBUG( |
| 171 | - "[AllGatherGraphPipeline][RunAsync] local rank[%u] localOffset[%llu]tx with remoteRank[%u], " | 171 | + "[AllGatherGraphPipeline][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, memSliceSize); | 173 | userRank_, serverOffsetByte, nextInterRankId, serverOffsetByte, memSliceSize); |
| 174 | 174 | ||
| @@ -177,7 +177,7 @@ HcclResult AllGatherGraphPipeline::RunAsync() | |||
| 177 | static_cast<u8*>(dmaMem_[1].ptr()) + readRemoteOffsetByte, memSliceSize, | 177 | static_cast<u8*>(dmaMem_[1].ptr()) + readRemoteOffsetByte, memSliceSize, |
| 178 | subStream_[0])); // wait | 178 | subStream_[0])); // wait |
| 179 | HCCL_DEBUG( | 179 | HCCL_DEBUG( |
| 180 | - "[AllGatherGraphPipeline][RunAsync] read local rank[%u] localOffset[%llu]tx with remoteRank[%u], " | 180 | + "[AllGatherGraphPipeline][RunAsync] read local rank[%u] localOffset[%llu] tx with remoteRank[%u], " |
| 181 | "remoteOffset[%llu] with slice[%llu]", | 181 | "remoteOffset[%llu] with slice[%llu]", |
| 182 | userRank_, readRemoteOffsetByte, prevInterRankId, readRemoteOffsetByte, memSliceSize); | 182 | userRank_, readRemoteOffsetByte, prevInterRankId, readRemoteOffsetByte, memSliceSize); |
| 183 | 183 | ||
Msrc/legacy/ascend910/algorithm/base/alg_template/temp_all_gather/all_gather_halving_doubling.cc+2-2
| @@ -183,14 +183,14 @@ HcclResult AllGatherHalvingDoubling::RunAllGather(u32 rank, u32 stepNum, const s | |||
| 183 | 183 | ||
| 184 | // 本rank的发送侧的动作 | 184 | // 本rank的发送侧的动作 |
| 185 | HCCL_DEBUG( | 185 | HCCL_DEBUG( |
| 186 | - "Rank[%u] send to PeerRank[%u] in Round[%u], silce.offset[%llu], slice.size[%llu]", rank, peerRank, step, | 186 | + "Rank[%u] send to PeerRank[%u] in Round[%u], slice.offset[%llu], slice.size[%llu]", rank, peerRank, step, |
| 187 | txSlices_[step].offset, txSlices_[step].size); | 187 | txSlices_[step].offset, txSlices_[step].size); |
| 188 | ret = Tx(links[peerRank], txSlices_[step]); | 188 | ret = Tx(links[peerRank], txSlices_[step]); |
| 189 | CHK_PRT_RET( | 189 | CHK_PRT_RET( |
| 190 | ret != HCCL_SUCCESS, | 190 | ret != HCCL_SUCCESS, |
| 191 | HCCL_ERROR("[Run][AllGather]Rank[%u] end to PeerRank[%u] run tx failed ", rank, peerRank), ret); | 191 | HCCL_ERROR("[Run][AllGather]Rank[%u] end to PeerRank[%u] run tx failed ", rank, peerRank), ret); |
| 192 | HCCL_DEBUG( | 192 | HCCL_DEBUG( |
| 193 | - "Rank[%u]receive from PeerRank[%u] in Round[%u], silce.offset[%llu], slice.size[%llu]", rank, peerRank, | 193 | + "Rank[%u] receive from PeerRank[%u] in Round[%u], slice.offset[%llu], slice.size[%llu]", rank, peerRank, |
| 194 | step, rxSlices_[step].offset, rxSlices_[step].size); | 194 | step, rxSlices_[step].offset, rxSlices_[step].size); |
| 195 | // 本rank的接收侧的动作 | 195 | // 本rank的接收侧的动作 |
| 196 | ret = Rx(links[peerRank], rxSlices_[step]); | 196 | ret = Rx(links[peerRank], rxSlices_[step]); |
| @@ -51,7 +51,7 @@ HcclResult AllReduceAHC::RunInterAllReduce( | |||
| 51 | 51 | ||
| 52 | if (interLinksVector.size() != interSlicesVector.size()) { | 52 | if (interLinksVector.size() != interSlicesVector.size()) { |
| 53 | HCCL_ERROR( | 53 | HCCL_ERROR( |
| 54 | - "[AllReduceAHC][RunInterAllReduce]rank[%u] linksVector size[%llu] is no equal to slicesVector size [%u]", | 54 | + "[AllReduceAHC][RunInterAllReduce]rank[%u] linksVector size[%llu] is not equal to slicesVector size[%u]", |
| 55 | rank, interLinksVector.size(), interSlicesVector.size()); | 55 | rank, interLinksVector.size(), interSlicesVector.size()); |
| 56 | return HCCL_E_INTERNAL; | 56 | return HCCL_E_INTERNAL; |
| 57 | } | 57 | } |
Msrc/legacy/ascend910/algorithm/base/alg_template/temp_all_reduce/all_reduce_local_reduce_bcast.cc+4-2
| @@ -167,7 +167,9 @@ HcclResult AllReduceLocalReduceBcast::RunReduce(u32 rank, u32 rankSize, const st | |||
| 167 | } else { | 167 | } else { |
| 168 | ret = RunAllReduceBDReduceSend(rank, 0, links); | 168 | ret = RunAllReduceBDReduceSend(rank, 0, links); |
| 169 | } | 169 | } |
| 170 | - CHK_PRT_RET(ret != HCCL_SUCCESS, HCCL_ERROR("[AllReduceLocalReduceBcastReduce]rank[%u]failed", rank), ret); | 170 | + CHK_PRT_RET( |
| 171 | + ret != HCCL_SUCCESS, HCCL_ERROR("[AllReduceLocalReduceBcastReduce]rank[%u] RunAllReduceBDReduce failed", rank), | ||
| 172 | + ret); | ||
| 171 | return HCCL_SUCCESS; | 173 | return HCCL_SUCCESS; |
| 172 | } | 174 | } |
| 173 | 175 | ||
| @@ -181,7 +183,7 @@ HcclResult AllReduceLocalReduceBcast::RunBroadcast(u32 rank, u32 rankSize, const | |||
| 181 | } else { | 183 | } else { |
| 182 | ret = RunAllReduceBDMemcpyReceive(rank, 0, links); | 184 | ret = RunAllReduceBDMemcpyReceive(rank, 0, links); |
| 183 | } | 185 | } |
| 184 | - CHK_PRT_RET(ret != HCCL_SUCCESS, HCCL_ERROR("[AllReduceLocalReduceBcast]rank[%u]failed", rank), ret); | 186 | + CHK_PRT_RET(ret != HCCL_SUCCESS, HCCL_ERROR("[AllReduceLocalReduceBcast]rank[%u] RunBroadcast failed", rank), ret); |
| 185 | 187 | ||
| 186 | HCCL_INFO("AllReduceLocalReduceBcast RunBroadcast: rank[%u]", rank); | 188 | HCCL_INFO("AllReduceLocalReduceBcast RunBroadcast: rank[%u]", rank); |
| 187 | return HCCL_SUCCESS; | 189 | return HCCL_SUCCESS; |
| @@ -41,7 +41,7 @@ HcclResult AllReduceNB::RunAsync(const u32 rank, const u32 rankSize, const std:: | |||
| 41 | CHK_PRT_RET( | 41 | CHK_PRT_RET( |
| 42 | ret != HCCL_SUCCESS, | 42 | ret != HCCL_SUCCESS, |
| 43 | HCCL_ERROR( | 43 | HCCL_ERROR( |
| 44 | - "[AllReduceNB][RunAsync]rank[%u] count[%llu] failed in reducescater " | 44 | + "[AllReduceNB][RunAsync]rank[%u] count[%llu] failed in ReduceScatter " |
| 45 | "step", | 45 | "step", |
| 46 | rank, count_), | 46 | rank, count_), |
| 47 | ret); | 47 | ret); |
| @@ -84,7 +84,7 @@ AllReduceNB::RunAsyncStaged(const u32 rank, const u32 rankSize, const std::vecto | |||
| 84 | ret != HCCL_SUCCESS, | 84 | ret != HCCL_SUCCESS, |
| 85 | HCCL_ERROR( | 85 | HCCL_ERROR( |
| 86 | "[AllReduceNB][RunAsyncStaged]rank[%u] count[%llu] " | 86 | "[AllReduceNB][RunAsyncStaged]rank[%u] count[%llu] " |
| 87 | - "failed in reducescater step", | 87 | + "failed in ReduceScatter step", |
| 88 | rank, count_), | 88 | rank, count_), |
| 89 | ret); | 89 | ret); |
| 90 | break; | 90 | break; |
| @@ -100,7 +100,7 @@ AllReduceNB::RunAsyncStaged(const u32 rank, const u32 rankSize, const std::vecto | |||
| 100 | ret); | 100 | ret); |
| 101 | break; | 101 | break; |
| 102 | default: | 102 | default: |
| 103 | - HCCL_ERROR("[AllReduceNB][RunAsyncStaged]stage[%d]is not support", stage); | 103 | + HCCL_ERROR("[AllReduceNB][RunAsyncStaged]stage[%d] is not supported", stage); |
| 104 | return HCCL_E_NOT_SUPPORT; | 104 | return HCCL_E_NOT_SUPPORT; |
| 105 | } | 105 | } |
| 106 | HCCL_INFO("AllReduceNB RunAsyncStaged stage[%d] finished: rank[%u] ranksize[%u]", stage, rank, rankSize); | 106 | HCCL_INFO("AllReduceNB RunAsyncStaged stage[%d] finished: rank[%u] ranksize[%u]", stage, rank, rankSize); |
| @@ -57,7 +57,7 @@ HcclResult AllReduceNHR::RunAsync(const u32 rank, const u32 rankSize, const std: | |||
| 57 | CHK_PRT_RET( | 57 | CHK_PRT_RET( |
| 58 | ret != HCCL_SUCCESS, | 58 | ret != HCCL_SUCCESS, |
| 59 | HCCL_ERROR( | 59 | HCCL_ERROR( |
| 60 | - "[AllReduceNHR][RunAsync] rank[%u] count[%llu] failed in reducescater " | 60 | + "[AllReduceNHR][RunAsync] rank[%u] count[%llu] failed in ReduceScatter " |
| 61 | "step", | 61 | "step", |
| 62 | rank, count_), | 62 | rank, count_), |
| 63 | ret); | 63 | ret); |
| @@ -214,7 +214,7 @@ HcclResult AlltoAllVDirectFullMesh::UpdateCurrRankRecvInfo( | |||
| 214 | const u64 recvLen = 0; | 214 | const u64 recvLen = 0; |
| 215 | u64 userOutOffset = localSendRecvInfo.recvOffset[destRank]; | 215 | u64 userOutOffset = localSendRecvInfo.recvOffset[destRank]; |
| 216 | HCCL_DEBUG( | 216 | HCCL_DEBUG( |
| 217 | - "[AlltoAllVDirectFullMesh][UpdateCurrRankRecvInfo] usrRank[%u] recv from destRank [%u]" | 217 | + "[AlltoAllVDirectFullMesh][UpdateCurrRankRecvInfo] usrRank[%u] recv from destRank [%u] " |
| 218 | "recvStepIdx[%u] recvLen[%lu] userOutOffset[%llu] scratchOffset[%llu]", | 218 | "recvStepIdx[%u] recvLen[%lu] userOutOffset[%llu] scratchOffset[%llu]", |
| 219 | userRank_, destRank, recvStepIdx, recvLen, userOutOffset, scratchOffset); | 219 | userRank_, destRank, recvStepIdx, recvLen, userOutOffset, scratchOffset); |
| 220 | 220 | ||
| @@ -239,7 +239,7 @@ HcclResult AlltoAllVDirectFullMesh::UpdateCurrRankRecvInfo( | |||
| 239 | u64 recvLen = std::min(sdmaDataBlockSize_, currDataRemainLen); | 239 | u64 recvLen = std::min(sdmaDataBlockSize_, currDataRemainLen); |
| 240 | u64 userOutOffset = localSendRecvInfo.recvOffset[destRank] + dataOffset; | 240 | u64 userOutOffset = localSendRecvInfo.recvOffset[destRank] + dataOffset; |
| 241 | HCCL_DEBUG( | 241 | HCCL_DEBUG( |
| 242 | - "[AlltoAllVDirectFullMesh][UpdateCurrRankRecvInfo] usrRank[%u] recv from destRank [%u]" | 242 | + "[AlltoAllVDirectFullMesh][UpdateCurrRankRecvInfo] usrRank[%u] recv from destRank [%u] " |
| 243 | "recvStepIdx[%u] recvLen[%lu] userOutOffset[%llu] scratchOffset[%llu]", | 243 | "recvStepIdx[%u] recvLen[%lu] userOutOffset[%llu] scratchOffset[%llu]", |
| 244 | userRank_, destRank, recvStepIdx, recvLen, userOutOffset, scratchOffset); | 244 | userRank_, destRank, recvStepIdx, recvLen, userOutOffset, scratchOffset); |
| 245 | readInfo.push_back({recvLen, scratchOffset, userOutOffset}); | 245 | readInfo.push_back({recvLen, scratchOffset, userOutOffset}); |
| @@ -61,7 +61,7 @@ HcclResult AlltoAllVStagedMesh::RunAsync(const u32 rank, const u32 rankSize, con | |||
| 61 | CHK_SMART_PTR_NULL(subStreamsPtr_); | 61 | CHK_SMART_PTR_NULL(subStreamsPtr_); |
| 62 | 62 | ||
| 63 | CHK_PRT_RET( | 63 | CHK_PRT_RET( |
| 64 | - rankSize == 0, HCCL_ERROR("[AlltoAllVStagedMesh][Prepare] invilad rankSize[%u]", rankSize), HCCL_E_PARA); | 64 | + rankSize == 0, HCCL_ERROR("[AlltoAllVStagedMesh][Prepare] invalid rankSize[%u]", rankSize), HCCL_E_PARA); |
| 65 | 65 | ||
| 66 | CHK_PRT_RET( | 66 | CHK_PRT_RET( |
| 67 | rankSize != links.size(), | 67 | rankSize != links.size(), |
| @@ -81,7 +81,7 @@ HcclResult AlltoAllVStagedMesh::RunAsync(const u32 rank, const u32 rankSize, con | |||
| 81 | CHK_PRT_RET( | 81 | CHK_PRT_RET( |
| 82 | !sizeEqual, | 82 | !sizeEqual, |
| 83 | HCCL_ERROR( | 83 | HCCL_ERROR( |
| 84 | - "[AlltoAllVStagedMesh][RunAsync] invilad params: " | 84 | + "[AlltoAllVStagedMesh][RunAsync] invalid params: " |
| 85 | "sendAddrInfo size[%u] recvAddrInfo size[%u] rankSize[%u]", | 85 | "sendAddrInfo size[%u] recvAddrInfo size[%u] rankSize[%u]", |
| 86 | sendAddrInfo_.size(), recvAddrInfo_.size(), rankSize), | 86 | sendAddrInfo_.size(), recvAddrInfo_.size(), rankSize), |
| 87 | HCCL_E_PARA); | 87 | HCCL_E_PARA); |
| @@ -50,7 +50,7 @@ HcclResult AlltoAllVStagedPairwise::Prepare( | |||
| 50 | 50 | ||
| 51 | CHK_PRT_RET( | 51 | CHK_PRT_RET( |
| 52 | scratchInputMem.size() == 0, | 52 | scratchInputMem.size() == 0, |
| 53 | - HCCL_ERROR("[AlltoAllVStagedPairwise][Prepare] invilad scratchMemSize[%llu]", scratchInputMem.size()), | 53 | + HCCL_ERROR("[AlltoAllVStagedPairwise][Prepare] invalid scratchMemSize[%llu]", scratchInputMem.size()), |
| 54 | HCCL_E_PARA); | 54 | HCCL_E_PARA); |
| 55 | scratchInputMem_ = scratchInputMem; | 55 | scratchInputMem_ = scratchInputMem; |
| 56 | scratchOutputMem_ = scratchOutputMem; | 56 | scratchOutputMem_ = scratchOutputMem; |
| @@ -68,7 +68,7 @@ HcclResult AlltoAllVStagedPairwise::RunAsync(const u32 rank, const u32 rankSize, | |||
| 68 | CHK_PTR_NULL(mainStreamPtr_); | 68 | CHK_PTR_NULL(mainStreamPtr_); |
| 69 | 69 | ||
| 70 | CHK_PRT_RET( | 70 | CHK_PRT_RET( |
| 71 | - rankSize == 0, HCCL_ERROR("[AlltoAllVStagedPairwise][Prepare] invilad rankSize[%u]", rankSize), HCCL_E_PARA); | 71 | + rankSize == 0, HCCL_ERROR("[AlltoAllVStagedPairwise][Prepare] invalid rankSize[%u]", rankSize), HCCL_E_PARA); |
| 72 | 72 | ||
| 73 | CHK_PRT_RET( | 73 | CHK_PRT_RET( |
| 74 | rankSize != links.size(), | 74 | rankSize != links.size(), |
| @@ -81,7 +81,7 @@ HcclResult AlltoAllVStagedPairwise::RunAsync(const u32 rank, const u32 rankSize, | |||
| 81 | CHK_PRT_RET( | 81 | CHK_PRT_RET( |
| 82 | !sizeEqual, | 82 | !sizeEqual, |
| 83 | HCCL_ERROR( | 83 | HCCL_ERROR( |
| 84 | - "[AlltoAllVStagedPairwise][RunAsync] invilad params: " | 84 | + "[AlltoAllVStagedPairwise][RunAsync] invalid params: " |
| 85 | "sendAddrInfo size[%u] recvAddrInfo size[%u] rankSize[%u]", | 85 | "sendAddrInfo size[%u] recvAddrInfo size[%u] rankSize[%u]", |
| 86 | sendAddrInfo_.size(), recvAddrInfo_.size(), rankSize), | 86 | sendAddrInfo_.size(), recvAddrInfo_.size(), rankSize), |
| 87 | HCCL_E_PARA); | 87 | HCCL_E_PARA); |
| @@ -252,7 +252,7 @@ HcclResult AlltoAllVStagedPairwise::CheckPolicies( | |||
| 252 | CHK_PRT_RET( | 252 | CHK_PRT_RET( |
| 253 | times != policies.size(), | 253 | times != policies.size(), |
| 254 | HCCL_ERROR( | 254 | HCCL_ERROR( |
| 255 | - "[AlltoAllVStagedPairwise][CheckPolicies] invilad params: times[%llu] policies size[%u]", times, | 255 | + "[AlltoAllVStagedPairwise][CheckPolicies] invalid params: times[%llu] policies size[%u]", times, |
| 256 | policies.size()), | 256 | policies.size()), |
| 257 | HCCL_E_PARA); | 257 | HCCL_E_PARA); |
| 258 | 258 | ||
| @@ -264,7 +264,7 @@ HcclResult AlltoAllVStagedPairwise::CheckPolicies( | |||
| 264 | CHK_PRT_RET( | 264 | CHK_PRT_RET( |
| 265 | sum > scratchMemSize_, | 265 | sum > scratchMemSize_, |
| 266 | HCCL_ERROR( | 266 | HCCL_ERROR( |
| 267 | - "[AlltoAllVStagedPairwise][CheckPolicies] invilad params: curTime[%u] sum[%llu] scratchMemSize_[%u]", i, | 267 | + "[AlltoAllVStagedPairwise][CheckPolicies] invalid params: curTime[%u] sum[%llu] scratchMemSize_[%u]", i, |
| 268 | sum, scratchMemSize_), | 268 | sum, scratchMemSize_), |
| 269 | HCCL_E_PARA); | 269 | HCCL_E_PARA); |
| 270 | } | 270 | } |
| @@ -84,7 +84,7 @@ BroadcastStar::RunAsync(const u32 rank, const u32 rankSize, const std::vector<st | |||
| 84 | rank, root_), | 84 | rank, root_), |
| 85 | ret); | 85 | ret); |
| 86 | } | 86 | } |
| 87 | - HCCL_INFO("BroadBastStar finished: rank[%u]", rank); | 87 | + HCCL_INFO("BroadcastStar finished: rank[%u]", rank); |
| 88 | return HCCL_SUCCESS; | 88 | return HCCL_SUCCESS; |
| 89 | } | 89 | } |
| 90 | 90 | ||
Msrc/legacy/ascend910/algorithm/base/alg_template/temp_reduce_scatter/reduce_scatter_mesh_atomic.cc+1-1
| @@ -153,7 +153,7 @@ HcclResult ReduceScatterMeshAtomic::RunAsync(const u32 rank, const u32 rankSize, | |||
| 153 | ret != HCCL_SUCCESS, | 153 | ret != HCCL_SUCCESS, |
| 154 | HCCL_ERROR( | 154 | HCCL_ERROR( |
| 155 | "[ReduceScatterMeshAtomic][RunAsync]rank[%u] memcpy async from mem[%p] " | 155 | "[ReduceScatterMeshAtomic][RunAsync]rank[%u] memcpy async from mem[%p] " |
| 156 | - "to ouputmem[%p] failed", | 156 | + "to outputmem[%p] failed", |
| 157 | rank, src.ptr(), outputMem_.ptr()), | 157 | rank, src.ptr(), outputMem_.ptr()), |
| 158 | ret); | 158 | ret); |
| 159 | } | 159 | } |
| @@ -248,7 +248,7 @@ HcclResult ReduceScatterNHR::InlineReduceRxLastStep( | |||
| 248 | DeviceMem tmpMem = scratchMem_.range(outputSlices[rxSliceIdx].offset, outputSlices[rxSliceIdx].size); | 248 | DeviceMem tmpMem = scratchMem_.range(outputSlices[rxSliceIdx].offset, outputSlices[rxSliceIdx].size); |
| 249 | HCCL_DEBUG( | 249 | HCCL_DEBUG( |
| 250 | "[ReduceScatterNHR][RunReduceScatterNHR] final reduce rxSliceIdx[%u] will reduce with " | 250 | "[ReduceScatterNHR][RunReduceScatterNHR] final reduce rxSliceIdx[%u] will reduce with " |
| 251 | - "inputMem_ offset[%llu] to ouput_mem_ offset[%llu] size[%llu]", | 251 | + "inputMem_ offset[%llu] to output_mem_ offset[%llu] size[%llu]", |
| 252 | rxSliceIdx, inputSlices[rxSliceIdx].offset, outputSlices[rxSliceIdx].offset, outputSlices[rxSliceIdx].size); | 252 | rxSliceIdx, inputSlices[rxSliceIdx].offset, outputSlices[rxSliceIdx].offset, outputSlices[rxSliceIdx].size); |
| 253 | 253 | ||
| 254 | rxReduceMems.emplace_back( | 254 | rxReduceMems.emplace_back( |
| @@ -311,7 +311,7 @@ HcclResult ReduceScatterNHR::TbeReduceRxLastStep( | |||
| 311 | u64 dataCount = dstMem.size() / SIZE_TABLE[dataType_]; | 311 | u64 dataCount = dstMem.size() / SIZE_TABLE[dataType_]; |
| 312 | HCCL_DEBUG( | 312 | HCCL_DEBUG( |
| 313 | "[ReduceScatterNHR][RunReduceScatterNHR] final reduce rxSliceIdx[%u] will reduce with " | 313 | "[ReduceScatterNHR][RunReduceScatterNHR] final reduce rxSliceIdx[%u] will reduce with " |
| 314 | - "inputMem_ offset[%llu] to ouput_mem_ offset[%llu] size[%llu]", | 314 | + "inputMem_ offset[%llu] to output_mem_ offset[%llu] size[%llu]", |
| 315 | rxSliceIdx, inputSlices[rxSliceIdx].offset, outputSlices[rxSliceIdx].offset, outputSlices[rxSliceIdx].size); | 315 | rxSliceIdx, inputSlices[rxSliceIdx].offset, outputSlices[rxSliceIdx].offset, outputSlices[rxSliceIdx].size); |
| 316 | CHK_RET(HcclD2DMemcpyAsync( | 316 | CHK_RET(HcclD2DMemcpyAsync( |
| 317 | dispatcher_, tmpMem, srcMemRemote, stream_, linkLeft->GetRemoteRank(), // left的inputMem拷到本端的scratchMem | 317 | dispatcher_, tmpMem, srcMemRemote, stream_, linkLeft->GetRemoteRank(), // left的inputMem拷到本端的scratchMem |
| @@ -343,7 +343,7 @@ HcclResult ReduceScatterNHR::RunDestReducerLastStep( | |||
| 343 | DeviceMem tmpMem = scratchMem_.range(outputSlices[rxSliceIdx].offset, outputSlices[rxSliceIdx].size); | 343 | DeviceMem tmpMem = scratchMem_.range(outputSlices[rxSliceIdx].offset, outputSlices[rxSliceIdx].size); |
| 344 | HCCL_DEBUG( | 344 | HCCL_DEBUG( |
| 345 | "[ReduceScatterNHR][RunReduceScatterNHR] final reduce rxSliceIdx[%u] will reduce with " | 345 | "[ReduceScatterNHR][RunReduceScatterNHR] final reduce rxSliceIdx[%u] will reduce with " |
| 346 | - "inputMem_ offset[%llu] to ouput_mem_ offset[%llu] size[%llu]", | 346 | + "inputMem_ offset[%llu] to output_mem_ offset[%llu] size[%llu]", |
| 347 | rxSliceIdx, inputSlices[rxSliceIdx].offset, outputSlices[rxSliceIdx].offset, outputSlices[rxSliceIdx].size); | 347 | rxSliceIdx, inputSlices[rxSliceIdx].offset, outputSlices[rxSliceIdx].offset, outputSlices[rxSliceIdx].size); |
| 348 | 348 | ||
| 349 | rxReduceMems.emplace_back( | 349 | rxReduceMems.emplace_back( |
Msrc/legacy/ascend910/algorithm/base/alg_template/temp_reduce_scatter/reduce_scatter_pipeline.cc+2-2
| @@ -216,7 +216,7 @@ HcclResult ReduceScatterPipeline::Prepare( | |||
| 216 | streamNotifyMain_ = notifyMain; | 216 | streamNotifyMain_ = notifyMain; |
| 217 | if (streamNotifyMain_.size() < intraRankSize_) { | 217 | if (streamNotifyMain_.size() < intraRankSize_) { |
| 218 | HCCL_ERROR( | 218 | HCCL_ERROR( |
| 219 | - "[ReduceScatterPipeline][Prepare]rank[%u] streamNotifyMain_ size [%u] error, is smaller than," | 219 | + "[ReduceScatterPipeline][Prepare]rank[%u] streamNotifyMain_ size [%u] is smaller than " |
| 220 | "intraRankSize_[%u]", | 220 | "intraRankSize_[%u]", |
| 221 | rankId_, streamNotifyMain_.size(), intraRankSize_); | 221 | rankId_, streamNotifyMain_.size(), intraRankSize_); |
| 222 | return HCCL_E_INTERNAL; | 222 | return HCCL_E_INTERNAL; |
| @@ -224,7 +224,7 @@ HcclResult ReduceScatterPipeline::Prepare( | |||
| 224 | streamNotifySub_ = notifySub; | 224 | streamNotifySub_ = notifySub; |
| 225 | if (streamNotifySub_.size() < intraRankSize_) { | 225 | if (streamNotifySub_.size() < intraRankSize_) { |
| 226 | HCCL_ERROR( | 226 | HCCL_ERROR( |
| 227 | - "[ReduceScatterPipeline][Prepare]rank[%u] streamNotifySub_ size [%u] error, is smaller than," | 227 | + "[ReduceScatterPipeline][Prepare]rank[%u] streamNotifySub_ size [%u] is smaller than " |
| 228 | "intraRankSize_[%u]", | 228 | "intraRankSize_[%u]", |
| 229 | rankId_, streamNotifySub_.size(), intraRankSize_); | 229 | rankId_, streamNotifySub_.size(), intraRankSize_); |
| 230 | return HCCL_E_INTERNAL; | 230 | return HCCL_E_INTERNAL; |
| @@ -73,7 +73,7 @@ HcclResult ScatterMesh::RunAsync(const u32 rank, const u32 rankSize, const std:: | |||
| 73 | } | 73 | } |
| 74 | // rank存放scatter 结果的偏移 | 74 | // rank存放scatter 结果的偏移 |
| 75 | u64 scatterOffset = slices_[rank].offset; | 75 | u64 scatterOffset = slices_[rank].offset; |
| 76 | - HCCL_DEBUG("rank[%u] scatter_offset is [%llu] reslutsize[%llu]", rank, scatterOffset, slices_[rank].size); | 76 | + HCCL_DEBUG("rank[%u] scatter_offset is [%llu] resultSize[%llu]", rank, scatterOffset, slices_[rank].size); |
| 77 | 77 | ||
| 78 | // root rank向其他rank发送数据 | 78 | // root rank向其他rank发送数据 |
| 79 | if (rank == root_) { | 79 | if (rank == root_) { |
| @@ -72,7 +72,7 @@ bool IsSupportDirectFullmeshForAlltoallv( | |||
| 72 | HcclResult retCapture = GetStreamCaptureInfo(param.stream.ptr(), rtModel, isCapture); | 72 | HcclResult retCapture = GetStreamCaptureInfo(param.stream.ptr(), rtModel, isCapture); |
| 73 | CHK_PRT_CONT( | 73 | CHK_PRT_CONT( |
| 74 | retCapture != HCCL_SUCCESS, | 74 | retCapture != HCCL_SUCCESS, |
| 75 | - HCCL_ERROR("Get capture status error. return[%d], capture model", retCapture)); | 75 | + HCCL_ERROR("Get capture status error. return[%d], capture mode", retCapture)); |
| 76 | } | 76 | } |
| 77 | isHCCS = (userRankSize <= MAX_ALLTOALLV_DIRECT_FULLMESH_RANKSIZE | 77 | isHCCS = (userRankSize <= MAX_ALLTOALLV_DIRECT_FULLMESH_RANKSIZE |
| 78 | && serverNum <= MAX_ALLTOALLV_DIRECT_FULLMESH_SERVER_NUM && isCapture) | 78 | && serverNum <= MAX_ALLTOALLV_DIRECT_FULLMESH_SERVER_NUM && isCapture) |
| @@ -63,7 +63,7 @@ HcclResult CollAllGatherHccsSioExecutor::CalcLevel0CommInfo( | |||
| 63 | // 根据子通信索引设置isHccs的值 | 63 | // 根据子通信索引设置isHccs的值 |
| 64 | transportRequest.linkType = (subCommIndex == 0) ? TransportLinkType::SIO : TransportLinkType::HCCS; | 64 | transportRequest.linkType = (subCommIndex == 0) ? TransportLinkType::SIO : TransportLinkType::HCCS; |
| 65 | HCCL_INFO( | 65 | HCCL_INFO( |
| 66 | - "[CollAllGatherHccsSioExecutor][CalcLevel0CommInfo] set extral notifyNum[%u]", | 66 | + "[CollAllGatherHccsSioExecutor][CalcLevel0CommInfo] set extra notifyNum[%u]", |
| 67 | transportRequest.notifyNum); | 67 | transportRequest.notifyNum); |
| 68 | } | 68 | } |
| 69 | } | 69 | } |
| @@ -61,7 +61,7 @@ HcclResult AllGatherVMeshAivExecutor::CalNumBlocks( | |||
| 61 | HCCL_E_PARA); | 61 | HCCL_E_PARA); |
| 62 | 62 | ||
| 63 | HCCL_INFO( | 63 | HCCL_INFO( |
| 64 | - "[AllGatherVMeshAivExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommanded[%u]", numBlocks, | 64 | + "[AllGatherVMeshAivExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommended[%u]", numBlocks, |
| 65 | numBlocks_, bestNumBlocks); | 65 | numBlocks_, bestNumBlocks); |
| 66 | return HCCL_SUCCESS; | 66 | return HCCL_SUCCESS; |
| 67 | } | 67 | } |
| @@ -267,7 +267,7 @@ HcclResult CollAllReduceExecutor::RunLoopInner(OpParam& param, const ReduceType& | |||
| 267 | u64 curSize = execMem.count * unitSize; // 单位:字节 | 267 | u64 curSize = execMem.count * unitSize; // 单位:字节 |
| 268 | HCCL_DEBUG( | 268 | HCCL_DEBUG( |
| 269 | "[CollAllReduceExecutor][RunLoopInner]inputMem[%p][%llu], outputMem[%p][%llu], " | 269 | "[CollAllReduceExecutor][RunLoopInner]inputMem[%p][%llu], outputMem[%p][%llu], " |
| 270 | - "intputPtr[%p], outputPtr[%p], curCount[%llu], curSize[%llu]", | 270 | + "inputPtr[%p], outputPtr[%p], curCount[%llu], curSize[%llu]", |
| 271 | execMem.inputMem.ptr(), execMem.inputMem.size(), execMem.outputMem.ptr(), execMem.outputMem.size(), | 271 | execMem.inputMem.ptr(), execMem.inputMem.size(), execMem.outputMem.ptr(), execMem.outputMem.size(), |
| 272 | execMem.inputPtr, execMem.outputPtr, execMem.count, curSize); | 272 | execMem.inputPtr, execMem.outputPtr, execMem.count, curSize); |
| 273 | CHK_PRT_RET( | 273 | CHK_PRT_RET( |
| @@ -113,7 +113,7 @@ HcclResult CollAllReduceMeshAivFor91093Executor::CalNumBlocks( | |||
| 113 | HCCL_E_PARA); | 113 | HCCL_E_PARA); |
| 114 | 114 | ||
| 115 | HCCL_INFO( | 115 | HCCL_INFO( |
| 116 | - "[CollAllReduceMeshAivFor91093Executor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommanded[%u]", | 116 | + "[CollAllReduceMeshAivFor91093Executor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommended[%u]", |
| 117 | numBlocks, numBlocks_, bestNumBlocks); | 117 | numBlocks, numBlocks_, bestNumBlocks); |
| 118 | return HCCL_SUCCESS; | 118 | return HCCL_SUCCESS; |
| 119 | } | 119 | } |
| @@ -103,7 +103,7 @@ HcclResult CollAllReduceSmallCountAivRdmaExecutor::CalNumBlocks( | |||
| 103 | HCCL_E_PARA); | 103 | HCCL_E_PARA); |
| 104 | 104 | ||
| 105 | HCCL_INFO( | 105 | HCCL_INFO( |
| 106 | - "[CollAllReduceSmallCountAivRdmaExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommanded[%u]", | 106 | + "[CollAllReduceSmallCountAivRdmaExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommended[%u]", |
| 107 | numBlocks, numBlocks_, bestNumBlocks); | 107 | numBlocks, numBlocks_, bestNumBlocks); |
| 108 | return HCCL_SUCCESS; | 108 | return HCCL_SUCCESS; |
| 109 | } | 109 | } |
| @@ -86,7 +86,7 @@ HcclResult CollAlltoAllMeshAivSmallCountExecutor::CalNumBlocks( | |||
| 86 | } | 86 | } |
| 87 | 87 | ||
| 88 | HCCL_INFO( | 88 | HCCL_INFO( |
| 89 | - "[CollAlltoAllMeshAivSmallCountExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommanded[%u]", | 89 | + "[CollAlltoAllMeshAivSmallCountExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommended[%u]", |
| 90 | numBlocks, numBlocks_, bestNumBlocks); | 90 | numBlocks, numBlocks_, bestNumBlocks); |
| 91 | return HCCL_SUCCESS; | 91 | return HCCL_SUCCESS; |
| 92 | } | 92 | } |
| @@ -163,7 +163,7 @@ HcclResult CollBroadcastMeshAivFor91093Executor::Orchestrate(OpParam& param, Alg | |||
| 163 | CHK_PRT_RET( | 163 | CHK_PRT_RET( |
| 164 | ret != HCCL_SUCCESS, | 164 | ret != HCCL_SUCCESS, |
| 165 | HCCL_ERROR( | 165 | HCCL_ERROR( |
| 166 | - "[CollBroadcastMeshAivFor91093Executor][Orchestrate]errNo[0x%016llx] tag[%s] excutor kernel run failed", | 166 | + "[CollBroadcastMeshAivFor91093Executor][Orchestrate]errNo[0x%016llx] tag[%s] executor kernel run failed", |
| 167 | HCCL_ERROR_CODE(ret), param.tag.c_str()), | 167 | HCCL_ERROR_CODE(ret), param.tag.c_str()), |
| 168 | ret); | 168 | ret); |
| 169 | 169 | ||
| @@ -51,7 +51,7 @@ HcclResult CollNativeExecutorBase::CalcResRequest(const OpParam& param, AlgResou | |||
| 51 | 51 | ||
| 52 | CHK_RET(BuildResourceRequest(scratchMemSize, streamNum, notifyNum, aivBufferRequest, opTransport, resourceRequest)); | 52 | CHK_RET(BuildResourceRequest(scratchMemSize, streamNum, notifyNum, aivBufferRequest, opTransport, resourceRequest)); |
| 53 | HCCL_INFO( | 53 | HCCL_INFO( |
| 54 | - "streamNum[%u], notifyNum[%u], sctrachMemSize[%llu], aivBufferRequest[%llu]", resourceRequest.streamNum, | 54 | + "streamNum[%u], notifyNum[%u], scratchMemSize[%llu], aivBufferRequest[%llu]", resourceRequest.streamNum, |
| 55 | resourceRequest.notifyNum, resourceRequest.scratchMemSize, resourceRequest.aivBufferRequest); | 55 | resourceRequest.notifyNum, resourceRequest.scratchMemSize, resourceRequest.aivBufferRequest); |
| 56 | // 打印建链诉求 | 56 | // 打印建链诉求 |
| 57 | PrintTransportRequest(resourceRequest); | 57 | PrintTransportRequest(resourceRequest); |
| @@ -598,7 +598,7 @@ HcclResult CollNativeExecutorBase::SendRecvSignalOnLinks(OpParam& param, ExecMem | |||
| 598 | u64 size = std::min(execMem.inputMem.size(), HCCL_INPLACE_MEMCOPY_SIZE); // 传128K数据量占满所有端口 | 598 | u64 size = std::min(execMem.inputMem.size(), HCCL_INPLACE_MEMCOPY_SIZE); // 传128K数据量占满所有端口 |
| 599 | HCCL_INFO( | 599 | HCCL_INFO( |
| 600 | "[CollNativeExecutorBase][SendRecvSignalOnLinks]" | 600 | "[CollNativeExecutorBase][SendRecvSignalOnLinks]" |
| 601 | - "links[%zu] start memcopy start to memcopy data [%llu]B.", | 601 | + "links[%zu] start to memcopy data [%llu]B.", |
| 602 | i, size); | 602 | i, size); |
| 603 | CHK_RET(links[i]->TxAsync(UserMemType::INPUT_MEM, 0, execMem.inputMem.ptr(), size, param.stream)); | 603 | CHK_RET(links[i]->TxAsync(UserMemType::INPUT_MEM, 0, execMem.inputMem.ptr(), size, param.stream)); |
| 604 | CHK_RET(links[i]->RxAsync(UserMemType::INPUT_MEM, 0, execMem.inputMem.ptr(), size, param.stream)); | 604 | CHK_RET(links[i]->RxAsync(UserMemType::INPUT_MEM, 0, execMem.inputMem.ptr(), size, param.stream)); |
| @@ -106,7 +106,7 @@ HcclResult CollReduceRingFor91093Executor::KernelRun(const OpParam& param, ExecM | |||
| 106 | CHK_PRT_RET( | 106 | CHK_PRT_RET( |
| 107 | perDataSize == 0, | 107 | perDataSize == 0, |
| 108 | HCCL_ERROR( | 108 | HCCL_ERROR( |
| 109 | - "[CollReduceRingFor91093Executor][KernelRun]errNo[0x%01611x] datatype[%d] is invalid", | 109 | + "[CollReduceRingFor91093Executor][KernelRun]errNo[0x%016llx] datatype[%d] is invalid", |
| 110 | HCCL_ERROR_CODE(HCCL_E_PARA), param.DataDes.dataType), | 110 | HCCL_ERROR_CODE(HCCL_E_PARA), param.DataDes.dataType), |
| 111 | HCCL_E_PARA); | 111 | HCCL_E_PARA); |
| 112 | std::vector<Slice> dataSegsSlice; // 数据分成ranksize份,每份的起始偏移和大小 | 112 | std::vector<Slice> dataSegsSlice; // 数据分成ranksize份,每份的起始偏移和大小 |
| @@ -62,8 +62,7 @@ HcclResult CollReduceScatterFastDoubleRingFor91093Executor::DoubleRingReduceScat | |||
| 62 | ret != HCCL_SUCCESS, | 62 | ret != HCCL_SUCCESS, |
| 63 | HCCL_ERROR( | 63 | HCCL_ERROR( |
| 64 | "[CollReduceScatterFastDoubleRingFor91093Executor][DoubleRingReduceScatter] Double ring ReduceScatter " | 64 | "[CollReduceScatterFastDoubleRingFor91093Executor][DoubleRingReduceScatter] Double ring ReduceScatter " |
| 65 | - "failed" | 65 | + "failed, return[%d]", |
| 66 | - "failed,return[%d]", | ||
| 67 | ret), | 66 | ret), |
| 68 | ret); | 67 | ret); |
| 69 | u32 ringIndexOp = COMM_INDEX_0; | 68 | u32 ringIndexOp = COMM_INDEX_0; |
| @@ -76,8 +75,7 @@ HcclResult CollReduceScatterFastDoubleRingFor91093Executor::DoubleRingReduceScat | |||
| 76 | ret != HCCL_SUCCESS, | 75 | ret != HCCL_SUCCESS, |
| 77 | HCCL_ERROR( | 76 | HCCL_ERROR( |
| 78 | "[CollReduceScatterFastDoubleRingFor91093Executor][DoubleRingReduceScatter] Double ring ReduceScatter " | 77 | "[CollReduceScatterFastDoubleRingFor91093Executor][DoubleRingReduceScatter] Double ring ReduceScatter " |
| 79 | - "failed " | 78 | + "failed, return[%d]", |
| 80 | - "failed,return[%d]", | ||
| 81 | ret), | 79 | ret), |
| 82 | ret); | 80 | ret); |
| 83 | // 空拷贝用于后续操作附着 | 81 | // 空拷贝用于后续操作附着 |
| @@ -87,8 +85,7 @@ HcclResult CollReduceScatterFastDoubleRingFor91093Executor::DoubleRingReduceScat | |||
| 87 | ret != HCCL_SUCCESS, | 85 | ret != HCCL_SUCCESS, |
| 88 | HCCL_ERROR( | 86 | HCCL_ERROR( |
| 89 | "[CollReduceScatterFastDoubleRingFor91093Executor][DoubleRingReduceScatter] Double ring ReduceScatter " | 87 | "[CollReduceScatterFastDoubleRingFor91093Executor][DoubleRingReduceScatter] Double ring ReduceScatter " |
| 90 | - "failed " | 88 | + "failed, return[%d]", |
| 91 | - "failed,return[%d]", | ||
| 92 | ret), | 89 | ret), |
| 93 | ret); | 90 | ret); |
| 94 | 91 | ||
| @@ -90,7 +90,7 @@ HcclResult CollReduceScatterMeshAivSmallCountExecutor::CalNumBlocks( | |||
| 90 | 90 | ||
| 91 | HCCL_INFO( | 91 | HCCL_INFO( |
| 92 | "[CollReduceScatterMeshAivSmallCountExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], " | 92 | "[CollReduceScatterMeshAivSmallCountExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], " |
| 93 | - "recommanded[%u]", | 93 | + "recommended[%u]", |
| 94 | numBlocks, numBlocks_, bestNumBlocks); | 94 | numBlocks, numBlocks_, bestNumBlocks); |
| 95 | return HCCL_SUCCESS; | 95 | return HCCL_SUCCESS; |
| 96 | } | 96 | } |
| @@ -131,7 +131,7 @@ HcclResult CollReduceScatterMeshAivSmallCountExecutor::GetAivExecParam( | |||
| 131 | } | 131 | } |
| 132 | 132 | ||
| 133 | HCCL_INFO( | 133 | HCCL_INFO( |
| 134 | - "SPK, buffersIn [%p] [%p] [%p] [%p]" | 134 | + "SPK, buffersIn [%p] [%p] [%p] [%p] " |
| 135 | "buffersOut [%p] [%p] [%p] [%p]", | 135 | "buffersOut [%p] [%p] [%p] [%p]", |
| 136 | args.buffersIn[0], args.buffersIn[1], args.buffersIn[2], args.buffersIn[3], args.buffersOut[0], | 136 | args.buffersIn[0], args.buffersIn[1], args.buffersIn[2], args.buffersIn[3], args.buffersOut[0], |
| 137 | args.buffersOut[1], args.buffersOut[2], args.buffersOut[3]); | 137 | args.buffersOut[1], args.buffersOut[2], args.buffersOut[3]); |
| @@ -66,7 +66,7 @@ HcclResult CollReduceScatterVAIVBigCountExecutor::CalNumBlocks( | |||
| 66 | HCCL_E_PARA); | 66 | HCCL_E_PARA); |
| 67 | 67 | ||
| 68 | HCCL_INFO( | 68 | HCCL_INFO( |
| 69 | - "[CollReduceScatterVAIVBigCountExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommanded[%u]", | 69 | + "[CollReduceScatterVAIVBigCountExecutor][CalNumBlocks] numBlocks is set to [%u], limit[%u], recommended[%u]", |
| 70 | numBlocks, numBlocks_, bestNumBlocks); | 70 | numBlocks, numBlocks_, bestNumBlocks); |
| 71 | return HCCL_SUCCESS; | 71 | return HCCL_SUCCESS; |
| 72 | } | 72 | } |
Msrc/legacy/ascend910/algorithm/impl/coll_executor/coll_send_receive/coll_batch_send_recv_executor.cc+1-1
| @@ -146,7 +146,7 @@ HcclResult CollBatchSendRecvExecutor::ProcessSelfSendRecvTasks(Stream& stream) | |||
| 146 | sendToSelfDeque_.pop_front(); | 146 | sendToSelfDeque_.pop_front(); |
| 147 | recvFromSelfDeque_.pop_front(); | 147 | recvFromSelfDeque_.pop_front(); |
| 148 | } else { | 148 | } else { |
| 149 | - HCCL_ERROR("[HcclBatchSendRecv] Send task and recv task to self : count or dataType do not equal, please" | 149 | + HCCL_ERROR("[HcclBatchSendRecv] Send task and recv task to self : count or dataType do not equal, please " |
| 150 | "check the task list."); | 150 | "check the task list."); |
| 151 | return HCCL_E_PARA; | 151 | return HCCL_E_PARA; |
| 152 | } | 152 | } |
| @@ -319,7 +319,7 @@ void ProfilerManagerImpl::HandleGraphLaunchTask(struct TaskPara* taskPara) | |||
| 319 | bool isCapture = false; | 319 | bool isCapture = false; |
| 320 | HcclResult retCapture = GetStreamCaptureInfo(taskPara->stream, rtModel, isCapture); | 320 | HcclResult retCapture = GetStreamCaptureInfo(taskPara->stream, rtModel, isCapture); |
| 321 | CHK_PRT_CONT( | 321 | CHK_PRT_CONT( |
| 322 | - retCapture != HCCL_SUCCESS, HCCL_ERROR("Get capture status error. return[%d], capture model", retCapture)); | 322 | + retCapture != HCCL_SUCCESS, HCCL_ERROR("Get capture status error. return[%d], capture mode", retCapture)); |
| 323 | if (profilingManager.GetFftsLaunchApiState() || isCapture) { | 323 | if (profilingManager.GetFftsLaunchApiState() || isCapture) { |
| 324 | // 上报批量下发的ContextId信息 | 324 | // 上报批量下发的ContextId信息 |
| 325 | (void)profilingManager.CallMsprofReportContextIdInfo((taskPara->graphLaunch.ctxNum - 1)); | 325 | (void)profilingManager.CallMsprofReportContextIdInfo((taskPara->graphLaunch.ctxNum - 1)); |
| @@ -48,7 +48,7 @@ HcclResult AclgraphCallback::CleanCaptureRes(u64 modelId) | |||
| 48 | std::lock_guard<std::mutex> lock(resMutex_); | 48 | std::lock_guard<std::mutex> lock(resMutex_); |
| 49 | auto modelIt = captureResMap_.find(modelId); | 49 | auto modelIt = captureResMap_.find(modelId); |
| 50 | if (modelIt == captureResMap_.end()) { | 50 | if (modelIt == captureResMap_.end()) { |
| 51 | - HCCL_ERROR("[%s] modelID[%llu] is not record", __func__, modelId); | 51 | + HCCL_ERROR("[%s] modelID[%llu] is not recorded", __func__, modelId); |
| 52 | return HCCL_E_NOT_FOUND; | 52 | return HCCL_E_NOT_FOUND; |
| 53 | } | 53 | } |
| 54 | 54 | ||
| @@ -340,7 +340,7 @@ HcclResult OpRetryAgentRunning::ParseKfcErr(RetryContext* retryCtx, RetryState& | |||
| 340 | break; | 340 | break; |
| 341 | } | 341 | } |
| 342 | case KfcError::kSdma: { | 342 | case KfcError::kSdma: { |
| 343 | - HCCL_RUN_INFO("[OpRetry][Agent]Get ErrorCode[%d] rertryCnt[%u]", kfcError, retryCnt); | 343 | + HCCL_RUN_INFO("[OpRetry][Agent]Get ErrorCode[%d] retryCnt[%u]", kfcError, retryCnt); |
| 344 | nextState = RETRY_STATE_RESP_AICPU_ERR; | 344 | nextState = RETRY_STATE_RESP_AICPU_ERR; |
| 345 | break; | 345 | break; |
| 346 | } | 346 | } |
| @@ -400,7 +400,7 @@ HcclResult OpRetryAgentWaitCmd::ProcessEvent(RetryContext* retryCtx) | |||
| 400 | CHK_PRT_RET(elapsed > timeout, HCCL_ERROR("[OpRetry]WaitCommand timeout"), HCCL_E_TIMEOUT); | 400 | CHK_PRT_RET(elapsed > timeout, HCCL_ERROR("[OpRetry]WaitCommand timeout"), HCCL_E_TIMEOUT); |
| 401 | CHK_PRT_RET( | 401 | CHK_PRT_RET( |
| 402 | retryCtx->isAgentStateWaitResume_, | 402 | retryCtx->isAgentStateWaitResume_, |
| 403 | - HCCL_RUN_INFO("[OpRetry][Agent]switched state form wait cmd to wait resume"), HCCL_SUCCESS); | 403 | + HCCL_RUN_INFO("[OpRetry][Agent]switched state from wait cmd to wait resume"), HCCL_SUCCESS); |
| 404 | 404 | ||
| 405 | HcclResult ret = WaitCommandWithOpId(retryCtx->agentSocket_, commandinfo); | 405 | HcclResult ret = WaitCommandWithOpId(retryCtx->agentSocket_, commandinfo); |
| 406 | if (ret == HCCL_SUCCESS) { | 406 | if (ret == HCCL_SUCCESS) { |
| @@ -566,7 +566,7 @@ HcclResult OpRetryAgentPollAicpuStop::ProcessEvent(RetryContext* retryCtx) | |||
| 566 | while (true) { | 566 | while (true) { |
| 567 | CHK_PRT_RET( | 567 | CHK_PRT_RET( |
| 568 | retryCtx->isAgentStateWaitResume_, | 568 | retryCtx->isAgentStateWaitResume_, |
| 569 | - HCCL_RUN_INFO("[OpRetry][Agent]switched state form poll aicpu to wait resume"), HCCL_SUCCESS); | 569 | + HCCL_RUN_INFO("[OpRetry][Agent]switched state from poll aicpu to wait resume"), HCCL_SUCCESS); |
| 570 | // 读取aicpuCtx中的状态 | 570 | // 读取aicpuCtx中的状态 |
| 571 | KfcExecStatus& opInfo = retryCtx->localRetryInfo_.opInfo; | 571 | KfcExecStatus& opInfo = retryCtx->localRetryInfo_.opInfo; |
| 572 | CHK_RET(GetOpExecInfo(retryCtx->GetD2hPtr(), opInfo)); | 572 | CHK_RET(GetOpExecInfo(retryCtx->GetD2hPtr(), opInfo)); |
| @@ -721,7 +721,7 @@ HcclResult OpRetryAgentWaitChangeLinkInfo::ProcessEvent(RetryContext* retryCtx) | |||
| 721 | CHK_PRT_RET(elapsed > timeout, HCCL_ERROR("[OpRetry]WaitChangeLink timeout"), HCCL_E_TIMEOUT); | 721 | CHK_PRT_RET(elapsed > timeout, HCCL_ERROR("[OpRetry]WaitChangeLink timeout"), HCCL_E_TIMEOUT); |
| 722 | CHK_PRT_RET( | 722 | CHK_PRT_RET( |
| 723 | retryCtx->isAgentStateWaitResume_, | 723 | retryCtx->isAgentStateWaitResume_, |
| 724 | - HCCL_RUN_INFO("[OpRetry][Agent]switched state form wait change link to wait resume"), HCCL_SUCCESS); | 724 | + HCCL_RUN_INFO("[OpRetry][Agent]switched state from wait change link to wait resume"), HCCL_SUCCESS); |
| 725 | 725 | ||
| 726 | HcclResult ret = WaitChangeLink(retryCtx->agentSocket_, tmpRecvChangeLinkInfo); | 726 | HcclResult ret = WaitChangeLink(retryCtx->agentSocket_, tmpRecvChangeLinkInfo); |
| 727 | if (ret == HCCL_SUCCESS) { | 727 | if (ret == HCCL_SUCCESS) { |
| @@ -735,7 +735,7 @@ HcclResult OpRetryAgentWaitChangeLinkInfo::ProcessEvent(RetryContext* retryCtx) | |||
| 735 | += (std::to_string(retryCtx->localChangeLinkInfo_.remoteRankList[i]) + ":" | 735 | += (std::to_string(retryCtx->localChangeLinkInfo_.remoteRankList[i]) + ":" |
| 736 | + std::to_string(retryCtx->localChangeLinkInfo_.isUseDefaultPort[i]) + "; "); | 736 | + std::to_string(retryCtx->localChangeLinkInfo_.isUseDefaultPort[i]) + "; "); |
| 737 | } | 737 | } |
| 738 | - HCCL_RUN_INFO("[OpRetry][Agnet]changeLinkInfoStr:%s", changeLinkInfoStr.c_str()); | 738 | + HCCL_RUN_INFO("[OpRetry][Agent]changeLinkInfoStr:%s", changeLinkInfoStr.c_str()); |
| 739 | 739 | ||
| 740 | // 收到changelinkinfo后切换到RETRY_STATE_WAIT_CMD_RESUME_TRANSPORT状态等待接收resume transport命令 | 740 | // 收到changelinkinfo后切换到RETRY_STATE_WAIT_CMD_RESUME_TRANSPORT状态等待接收resume transport命令 |
| 741 | nextState = RETRY_STATE_WAIT_CMD_RESUME_TRANSPORT; | 741 | nextState = RETRY_STATE_WAIT_CMD_RESUME_TRANSPORT; |
| @@ -852,12 +852,12 @@ HcclResult SwitchNicAgentWaitCmd::ProcessEvent(RetryContext* retryCtx) | |||
| 852 | while (true) { | 852 | while (true) { |
| 853 | CHK_PRT_RET( | 853 | CHK_PRT_RET( |
| 854 | retryCtx->isAgentStateWaitResume_, | 854 | retryCtx->isAgentStateWaitResume_, |
| 855 | - HCCL_RUN_INFO("[OpRetry][Agent]switched state form switch nic to wait resume"), HCCL_SUCCESS); | 855 | + HCCL_RUN_INFO("[OpRetry][Agent]switched state from switch nic to wait resume"), HCCL_SUCCESS); |
| 856 | std::chrono::steady_clock::time_point curTime = std::chrono::steady_clock::now(); | 856 | std::chrono::steady_clock::time_point curTime = std::chrono::steady_clock::now(); |
| 857 | const auto elapsed = std::chrono::duration_cast<std::chrono::seconds>(curTime - startTime); | 857 | const auto elapsed = std::chrono::duration_cast<std::chrono::seconds>(curTime - startTime); |
| 858 | CHK_PRT_RET( | 858 | CHK_PRT_RET( |
| 859 | elapsed > timeout, | 859 | elapsed > timeout, |
| 860 | - HCCL_ERROR("[SwitchNic][Agent] timeout in getting cmd from server, waitime[%u s>%u s]", elapsed, timeout), | 860 | + HCCL_ERROR("[SwitchNic][Agent] timeout in getting cmd from server, waittime[%u s>%u s]", elapsed, timeout), |
| 861 | HCCL_E_TIMEOUT); | 861 | HCCL_E_TIMEOUT); |
| 862 | 862 | ||
| 863 | HcclResult ret = WaitCommand(retryCtx->agentSocket_, command); | 863 | HcclResult ret = WaitCommand(retryCtx->agentSocket_, command); |
| @@ -887,7 +887,7 @@ HcclResult SwitchNicAgentSendSwitchInfo::ChangeAicpuStatus(RetryContext* retryCt | |||
| 887 | while (true) { | 887 | while (true) { |
| 888 | CHK_PRT_RET( | 888 | CHK_PRT_RET( |
| 889 | retryCtx->isAgentStateWaitResume_, | 889 | retryCtx->isAgentStateWaitResume_, |
| 890 | - HCCL_RUN_INFO("[OpRetry][Agent]switched state form send swhitch Nic info to wait resume"), HCCL_SUCCESS); | 890 | + HCCL_RUN_INFO("[OpRetry][Agent]switched state from send switch nic info to wait resume"), HCCL_SUCCESS); |
| 891 | KfcExecStatus& opInfo = retryCtx->localRetryInfo_.opInfo; | 891 | KfcExecStatus& opInfo = retryCtx->localRetryInfo_.opInfo; |
| 892 | CHK_RET(GetOpExecInfo(retryCtx->GetD2hPtr(), opInfo)); | 892 | CHK_RET(GetOpExecInfo(retryCtx->GetD2hPtr(), opInfo)); |
| 893 | const KfcStatus& aicpuState = opInfo.execStatus.kfcStatus; | 893 | const KfcStatus& aicpuState = opInfo.execStatus.kfcStatus; |
Msrc/legacy/ascend910/framework/cluster_maintenance/recovery/operator_retry/opretry_connection.cc+2-2
| @@ -112,7 +112,7 @@ HcclResult OpRetryConnection::Init( | |||
| 112 | serverPort + OP_RETRY_CONN_PORT_MAX_RANGE); | 112 | serverPort + OP_RETRY_CONN_PORT_MAX_RANGE); |
| 113 | HCCL_ERROR("2. Somebody may have already listen on those ports, so we connect to wrong server " | 113 | HCCL_ERROR("2. Somebody may have already listen on those ports, so we connect to wrong server " |
| 114 | "and we will meet 'Recv unmatched ack' error"); | 114 | "and we will meet 'Recv unmatched ack' error"); |
| 115 | - HCCL_ERROR("You may can set system reserved port to avoid this error by:"); | 115 | + HCCL_ERROR("You can set system reserved port to avoid this error by:"); |
| 116 | HCCL_ERROR( | 116 | HCCL_ERROR( |
| 117 | "sysctl -w net.ipv4.ip_local_reserved_ports=%u-%u", serverPort, serverPort + OP_RETRY_CONN_PORT_MAX_RANGE); | 117 | "sysctl -w net.ipv4.ip_local_reserved_ports=%u-%u", serverPort, serverPort + OP_RETRY_CONN_PORT_MAX_RANGE); |
| 118 | return HCCL_E_INTERNAL; | 118 | return HCCL_E_INTERNAL; |
| @@ -437,7 +437,7 @@ HcclResult OpRetryConnection::Accept() | |||
| 437 | while (expectAcceptNum > 0) { | 437 | while (expectAcceptNum > 0) { |
| 438 | if (backgroudThreadStop_) { | 438 | if (backgroudThreadStop_) { |
| 439 | HCCL_ERROR( | 439 | HCCL_ERROR( |
| 440 | - "[OpRetryConnection][Accept] OpRetryConnection in acceptting but stop, rankSize_[%u] accept link[%u]", | 440 | + "[OpRetryConnection][Accept] OpRetryConnection in accepting but stop, rankSize_[%u] accept link[%u]", |
| 441 | rankSize_, rankSize_ - expectAcceptNum); | 441 | rankSize_, rankSize_ - expectAcceptNum); |
| 442 | return HCCL_E_INTERNAL; | 442 | return HCCL_E_INTERNAL; |
| 443 | } | 443 | } |
| @@ -512,7 +512,7 @@ HcclResult ParseEnvConfig(const EnvConfigParam& param, std::string& envValue, u3 | |||
| 512 | CHK_PRT_RET( | 512 | CHK_PRT_RET( |
| 513 | (ret != HCCL_SUCCESS || resultValue < param.minValue || resultValue > param.maxValue), | 513 | (ret != HCCL_SUCCESS || resultValue < param.minValue || resultValue > param.maxValue), |
| 514 | HCCL_ERROR( | 514 | HCCL_ERROR( |
| 515 | - "[Parse][%s] is invalid. except: [%u, %u], actual: [%u]", param.envName.c_str(), param.minValue, | 515 | + "[Parse][%s] is invalid. expect: [%u, %u], actual: [%u]", param.envName.c_str(), param.minValue, |
| 516 | param.maxValue, resultValue), | 516 | param.maxValue, resultValue), |
| 517 | HCCL_E_PARA); | 517 | HCCL_E_PARA); |
| 518 | 518 | ||
| @@ -591,7 +591,7 @@ HcclResult EnvConfig::ParseRDMATimeOut(std::pair<u32, u32>& rdmaTimeOutRange) | |||
| 591 | CHK_PRT_RET( | 591 | CHK_PRT_RET( |
| 592 | (ret != HCCL_SUCCESS || rdmaTimeOut < HCCL_RDMA_TIMEOUT_MIN || rdmaTimeOut > rdmaTimeOutMax), | 592 | (ret != HCCL_SUCCESS || rdmaTimeOut < HCCL_RDMA_TIMEOUT_MIN || rdmaTimeOut > rdmaTimeOutMax), |
| 593 | HCCL_ERROR( | 593 | HCCL_ERROR( |
| 594 | - "[Parse][RDMATimeOut]HCCL_RDMA_TIMEOUT[%s] is invalid. except: [%u, %u]", timeOutEnv.c_str(), | 594 | + "[Parse][RDMATimeOut]HCCL_RDMA_TIMEOUT[%s] is invalid. expect: [%u, %u]", timeOutEnv.c_str(), |
| 595 | HCCL_RDMA_TIMEOUT_MIN, rdmaTimeOutMax), | 595 | HCCL_RDMA_TIMEOUT_MIN, rdmaTimeOutMax), |
| 596 | HCCL_E_PARA); | 596 | HCCL_E_PARA); |
| 597 | 597 | ||
| @@ -630,7 +630,7 @@ HcclResult EnvConfig::ParseRDMARetryCnt() | |||
| 630 | CHK_PRT_RET( | 630 | CHK_PRT_RET( |
| 631 | (ret != HCCL_SUCCESS || rdmaRetryCnt < HCCL_RDMA_RETRY_CNT_MIN || rdmaRetryCnt > HCCL_RDMA_RETRY_CNT_MAX), | 631 | (ret != HCCL_SUCCESS || rdmaRetryCnt < HCCL_RDMA_RETRY_CNT_MIN || rdmaRetryCnt > HCCL_RDMA_RETRY_CNT_MAX), |
| 632 | HCCL_ERROR( | 632 | HCCL_ERROR( |
| 633 | - "[Parse][rdmaRetryCnt]HCCL_RDMA_RETRY_CNT[%s] is invalid. except: [%u, %u]", retryCntEnv.c_str(), | 633 | + "[Parse][rdmaRetryCnt]HCCL_RDMA_RETRY_CNT[%s] is invalid. expect: [%u, %u]", retryCntEnv.c_str(), |
| 634 | HCCL_RDMA_RETRY_CNT_MIN, HCCL_RDMA_RETRY_CNT_MAX), | 634 | HCCL_RDMA_RETRY_CNT_MIN, HCCL_RDMA_RETRY_CNT_MAX), |
| 635 | HCCL_E_PARA); | 635 | HCCL_E_PARA); |
| 636 | g_envConfig.rdmaRetryCnt = rdmaRetryCnt; | 636 | g_envConfig.rdmaRetryCnt = rdmaRetryCnt; |
| @@ -678,7 +678,7 @@ HcclResult ParseMonitor(std::string& taskMonitorInterval, s32& monitorTime) | |||
| 678 | g_envConfig.dfsTaskMonitorInterval = monitorTime; | 678 | g_envConfig.dfsTaskMonitorInterval = monitorTime; |
| 679 | } else { // 不在允许范围内报错 | 679 | } else { // 不在允许范围内报错 |
| 680 | HCCL_ERROR( | 680 | HCCL_ERROR( |
| 681 | - "[ParseDFSConfig] HCCL_DFS_CONFIG-task_monitor_interval[%d ms] is invalid, except: [0, %d ms]", monitorTime, | 681 | + "[ParseDFSConfig] HCCL_DFS_CONFIG-task_monitor_interval[%d ms] is invalid, expect: [0, %d ms]", monitorTime, |
| 682 | maxTimeInMs); | 682 | maxTimeInMs); |
| 683 | return HCCL_E_PARA; | 683 | return HCCL_E_PARA; |
| 684 | } | 684 | } |
| @@ -708,7 +708,7 @@ HcclResult ParseDFSConfig() | |||
| 708 | g_envConfig.enableClusterHeartBeat = true; | 708 | g_envConfig.enableClusterHeartBeat = true; |
| 709 | } else { | 709 | } else { |
| 710 | HCCL_RUN_WARNING( | 710 | HCCL_RUN_WARNING( |
| 711 | - "[HCCL_ENV][ParseDFSConfig] HCCL_DFS_CONFIG-cluster_heartbeat was configured to [%s], please configured to" | 711 | + "[HCCL_ENV][ParseDFSConfig] HCCL_DFS_CONFIG-cluster_heartbeat was configured to [%s], please configure to " |
| 712 | "'on' or 'off'", | 712 | "'on' or 'off'", |
| 713 | heartbeatSwitch.c_str()); | 713 | heartbeatSwitch.c_str()); |
| 714 | } | 714 | } |
| @@ -721,7 +721,7 @@ HcclResult ParseDFSConfig() | |||
| 721 | g_envConfig.opCounterEnable = true; | 721 | g_envConfig.opCounterEnable = true; |
| 722 | } else { | 722 | } else { |
| 723 | HCCL_RUN_WARNING( | 723 | HCCL_RUN_WARNING( |
| 724 | - "[HCCL_ENV][ParseDFSConfig] HCCL_DFS_CONFIG-stuck_detection was configured to [%s], please configured to" | 724 | + "[HCCL_ENV][ParseDFSConfig] HCCL_DFS_CONFIG-stuck_detection was configured to [%s], please configure to " |
| 725 | "'on' or 'off'", | 725 | "'on' or 'off'", |
| 726 | stuckDetectSwitch.c_str()); | 726 | stuckDetectSwitch.c_str()); |
| 727 | } | 727 | } |
| @@ -737,7 +737,7 @@ HcclResult ParseDFSConfig() | |||
| 737 | g_envConfig.inconsistentCheckSwitch = InconsistentCheckMode::FIRST; | 737 | g_envConfig.inconsistentCheckSwitch = InconsistentCheckMode::FIRST; |
| 738 | } else { | 738 | } else { |
| 739 | HCCL_RUN_WARNING( | 739 | HCCL_RUN_WARNING( |
| 740 | - "[ParseDFSConfig] HCCL_DFS_CONFIG-inconsistent_check was configured to [%s], please configured to" | 740 | + "[ParseDFSConfig] HCCL_DFS_CONFIG-inconsistent_check was configured to [%s], please configure to " |
| 741 | "'on' or 'off' or 'first'", | 741 | "'on' or 'off' or 'first'", |
| 742 | inconsistentCheckSwitch.c_str()); | 742 | inconsistentCheckSwitch.c_str()); |
| 743 | } | 743 | } |
| @@ -778,7 +778,7 @@ HcclResult ParseDFSConfig() | |||
| 778 | g_envConfig.dfsConnectionFaultDetectionTime = detctTime; | 778 | g_envConfig.dfsConnectionFaultDetectionTime = detctTime; |
| 779 | } else { // 不在允许范围内报错 | 779 | } else { // 不在允许范围内报错 |
| 780 | HCCL_ERROR( | 780 | HCCL_ERROR( |
| 781 | - "[ParseDFSConfig] HCCL_DFS_CONFIG-connection_fault_detection_time[%d] is invalid, except: [%d, %d]", | 781 | + "[ParseDFSConfig] HCCL_DFS_CONFIG-connection_fault_detection_time[%d] is invalid, expect: [%d, %d]", |
| 782 | detctTime, HCCL_MIN_CONNECT_FAULT_DETECTION_TIME, HCCL_MAX_LINK_TIME_OUT_S); | 782 | detctTime, HCCL_MIN_CONNECT_FAULT_DETECTION_TIME, HCCL_MAX_LINK_TIME_OUT_S); |
| 783 | return HCCL_E_PARA; | 783 | return HCCL_E_PARA; |
| 784 | } | 784 | } |
| @@ -152,7 +152,7 @@ HcclResult OpExeCounter::ClearOpCounterMem() | |||
| 152 | } | 152 | } |
| 153 | CHK_RET(hrtMemSet(headCountMem_, memSize_, memSize_)); | 153 | CHK_RET(hrtMemSet(headCountMem_, memSize_, memSize_)); |
| 154 | CHK_RET(hrtMemSet(tailCountMem_, memSize_, memSize_)); | 154 | CHK_RET(hrtMemSet(tailCountMem_, memSize_, memSize_)); |
| 155 | - HCCL_DEBUG("[OpExeCounter][ClearOpCounterMem] headCountMem or tailCountMem is to success set 0"); | 155 | + HCCL_DEBUG("[OpExeCounter][ClearOpCounterMem] headCountMem or tailCountMem is successfully set to 0"); |
| 156 | return HCCL_SUCCESS; | 156 | return HCCL_SUCCESS; |
| 157 | } | 157 | } |
| 158 | 158 | ||
| @@ -113,7 +113,7 @@ HcclResult TopoInfoExchangeAgent::Setup() | |||
| 113 | }); | 113 | }); |
| 114 | } | 114 | } |
| 115 | HCCL_ERROR( | 115 | HCCL_ERROR( |
| 116 | - "[TopoInfoExchangeAgent][Setup]VerifyCluseterInfo failed, g_broadcastStage[%d]", | 116 | + "[TopoInfoExchangeAgent][Setup]VerifyClusterInfo failed, g_broadcastStage[%d]", |
| 117 | g_broadcastStage.load()); | 117 | g_broadcastStage.load()); |
| 118 | } | 118 | } |
| 119 | 119 | ||
| @@ -649,7 +649,7 @@ HcclResult TopoInfoExchangeAgent::VerifyClusterInfo(RankTable_t& clusterInfo) | |||
| 649 | errormessage = "The number of ranks[" + std::to_string(localRankInfo_.rankSize) | 649 | errormessage = "The number of ranks[" + std::to_string(localRankInfo_.rankSize) |
| 650 | + "]passed by the communicator initialization interface does not match the number of ranks[" | 650 | + "]passed by the communicator initialization interface does not match the number of ranks[" |
| 651 | + std::to_string(clusterInfo.rankList.size()) | 651 | + std::to_string(clusterInfo.rankList.size()) |
| 652 | - + "]obtained during cluster information negotiction."; | 652 | + + "] obtained during cluster information negotiation."; |
| 653 | HCCL_ERROR( | 653 | HCCL_ERROR( |
| 654 | "[%s][%s]%s", LOG_KEYWORDS_INIT_GROUP.c_str(), LOG_KEYWORDS_RANKTABLE_DETECT.c_str(), errormessage.c_str()); | 654 | "[%s][%s]%s", LOG_KEYWORDS_INIT_GROUP.c_str(), LOG_KEYWORDS_RANKTABLE_DETECT.c_str(), errormessage.c_str()); |
| 655 | return HCCL_E_PARA; | 655 | return HCCL_E_PARA; |
| @@ -658,7 +658,7 @@ HcclResult TopoInfoExchangeAgent::VerifyClusterInfo(RankTable_t& clusterInfo) | |||
| 658 | if (clusterInfo.rankNum != localRankInfo_.rankSize) { | 658 | if (clusterInfo.rankNum != localRankInfo_.rankSize) { |
| 659 | errormessage = "The number of ranks[" + std::to_string(localRankInfo_.rankSize) | 659 | errormessage = "The number of ranks[" + std::to_string(localRankInfo_.rankSize) |
| 660 | + "]passed by the communicator initialization interface does not match the number of ranks[" | 660 | + "]passed by the communicator initialization interface does not match the number of ranks[" |
| 661 | - + std::to_string(clusterInfo.rankNum) + "] obtained during cluster information negotiction."; | 661 | + + std::to_string(clusterInfo.rankNum) + "] obtained during cluster information negotiation."; |
| 662 | HCCL_ERROR( | 662 | HCCL_ERROR( |
| 663 | "[%s][%s]%s", LOG_KEYWORDS_INIT_GROUP.c_str(), LOG_KEYWORDS_RANKTABLE_DETECT.c_str(), errormessage.c_str()); | 663 | "[%s][%s]%s", LOG_KEYWORDS_INIT_GROUP.c_str(), LOG_KEYWORDS_RANKTABLE_DETECT.c_str(), errormessage.c_str()); |
| 664 | return HCCL_E_PARA; | 664 | return HCCL_E_PARA; |
| @@ -206,7 +206,7 @@ HcclResult TopoinfoRanktableConcise::CheckNicDeployConsistence(RankTable_t& clus | |||
| 206 | (deploy == NICDeployment::NIC_DEPLOYMENT_HOST && it.hostIp.IsInvalid()), | 206 | (deploy == NICDeployment::NIC_DEPLOYMENT_HOST && it.hostIp.IsInvalid()), |
| 207 | HCCL_ERROR( | 207 | HCCL_ERROR( |
| 208 | "[Get][RanktableInfo]errNo" | 208 | "[Get][RanktableInfo]errNo" |
| 209 | - "[0x%016llx] hostIp config bettewn ranks is different.", | 209 | + "[0x%016llx] hostIp config between ranks is different.", |
| 210 | HCOM_ERROR_CODE(HCCL_E_PARA)), | 210 | HCOM_ERROR_CODE(HCCL_E_PARA)), |
| 211 | HCCL_E_PARA); | 211 | HCCL_E_PARA); |
| 212 | } | 212 | } |
| @@ -614,7 +614,7 @@ HcclResult TopoinfoRanktableConcise::GetSingleBackupDeviceIp( | |||
| 614 | } else { | 614 | } else { |
| 615 | HcclIpAddress invalidAddr; | 615 | HcclIpAddress invalidAddr; |
| 616 | rankinfo.deviceInfo.backupDeviceIp.push_back(invalidAddr); | 616 | rankinfo.deviceInfo.backupDeviceIp.push_back(invalidAddr); |
| 617 | - HCCL_WARNING("objIndex[%u],'bakcup_device_ip' is not set", objIndex); | 617 | + HCCL_WARNING("objIndex[%u], 'backup_device_ip' is not set", objIndex); |
| 618 | } | 618 | } |
| 619 | return HCCL_SUCCESS; | 619 | return HCCL_SUCCESS; |
| 620 | } | 620 | } |
| @@ -951,7 +951,8 @@ HcclResult TopoinfoRanktableConcise::GetSingleSuperPodSever( | |||
| 951 | CHK_PRT_BREAK( | 951 | CHK_PRT_BREAK( |
| 952 | ret != HCCL_SUCCESS, | 952 | ret != HCCL_SUCCESS, |
| 953 | HCCL_ERROR( | 953 | HCCL_ERROR( |
| 954 | - "[Get][SingleSuperPodSever]errNo[0x%016llx]server_id is not found or invalid", HCCL_ERROR_CODE(ret)), ); | 954 | + "[Get][SingleSuperPodSever]errNo[0x%016llx] server_id is not found or invalid", |
| 955 | + HCCL_ERROR_CODE(ret)), ); | ||
| 955 | 956 | ||
| 956 | // 将super_pod_list下的server_id在资源池中进行校验 | 957 | // 将super_pod_list下的server_id在资源池中进行校验 |
| 957 | ret = CheckUniqueAndInsertPool( | 958 | ret = CheckUniqueAndInsertPool( |
| @@ -1037,8 +1038,8 @@ HcclResult TopoinfoRanktableConcise::CheckSuperPodInfo(RankTable_t& clusterInfo) | |||
| 1037 | CHK_PRT_RET( | 1038 | CHK_PRT_RET( |
| 1038 | iter->second.find(rankInfo.superDeviceId) != iter->second.end(), | 1039 | iter->second.find(rankInfo.superDeviceId) != iter->second.end(), |
| 1039 | HCCL_ERROR( | 1040 | HCCL_ERROR( |
| 1040 | - "[Verify][SuperPodInfo]superDeviceId[0x%x] in superPod[%s]" | 1041 | + "[Verify][SuperPodInfo]superDeviceId[0x%x] in superPod[%s] " |
| 1041 | - "is already exist.", | 1042 | + "already exists.", |
| 1042 | rankInfo.superDeviceId, iter->first.c_str()), | 1043 | rankInfo.superDeviceId, iter->first.c_str()), |
| 1043 | HCCL_E_PARA); | 1044 | HCCL_E_PARA); |
| 1044 | } | 1045 | } |
| @@ -456,7 +456,7 @@ HcclResult hcclComm::InitBinHandle() | |||
| 456 | CHK_PRT_RET( | 456 | CHK_PRT_RET( |
| 457 | retCode != HCCL_SUCCESS, | 457 | retCode != HCCL_SUCCESS, |
| 458 | HCCL_ERROR( | 458 | HCCL_ERROR( |
| 459 | - "[InitCollComm]errNo[0x%016llx]load aicpu file fail, path[%s] optionType[%u]" | 459 | + "[InitCollComm]errNo[0x%016llx]load aicpu file fail, path[%s] optionType[%u] " |
| 460 | "cpuKernelMode[%u].", | 460 | "cpuKernelMode[%u].", |
| 461 | retCode, jsonPath.c_str(), ACL_RT_BINARY_LOAD_OPT_CPU_KERNEL_MODE, 0), | 461 | retCode, jsonPath.c_str(), ACL_RT_BINARY_LOAD_OPT_CPU_KERNEL_MODE, 0), |
| 462 | retCode); | 462 | retCode); |
| @@ -556,7 +556,7 @@ HcclResult hcclComm::GetCommStatus(HcclCommStatus& status) | |||
| 556 | } else if (devType_ == DevType::DEV_TYPE_910B && collComm_ != nullptr) { | 556 | } else if (devType_ == DevType::DEV_TYPE_910B && collComm_ != nullptr) { |
| 557 | status = collComm_->GetCommStatus(); | 557 | status = collComm_->GetCommStatus(); |
| 558 | } else { | 558 | } else { |
| 559 | - HCCL_ERROR("[%s] deviceType is not support", __func__); | 559 | + HCCL_ERROR("[%s] deviceType is not supported", __func__); |
| 560 | return HCCL_E_NOT_SUPPORT; | 560 | return HCCL_E_NOT_SUPPORT; |
| 561 | } | 561 | } |
| 562 | return HCCL_SUCCESS; | 562 | return HCCL_SUCCESS; |
| @@ -461,7 +461,7 @@ void ZeroCopyMemoryAgent::ParseReceivedRequests() | |||
| 461 | CHK_PRT_CONT( | 461 | CHK_PRT_CONT( |
| 462 | ParseReceivedRequest(req, remoteRank) != HCCL_SUCCESS, | 462 | ParseReceivedRequest(req, remoteRank) != HCCL_SUCCESS, |
| 463 | HCCL_ERROR( | 463 | HCCL_ERROR( |
| 464 | - "[ZeroCopyMemoryAgent][ParseReceivedRequest] failed prase requestType[%s] remote[%u]", | 464 | + "[ZeroCopyMemoryAgent][ParseReceivedRequest] failed to parse requestType[%s] remote[%u]", |
| 465 | GetReadableRequestType(*reinterpret_cast<RequestType*>(req.data())), kv.first)); | 465 | GetReadableRequestType(*reinterpret_cast<RequestType*>(req.data())), kv.first)); |
| 466 | recvMgr.praseIndex_++; | 466 | recvMgr.praseIndex_++; |
| 467 | if (recvMgr.praseIndex_ == ZERO_COPY_MEMORY_AGENT_RECV_QUEUE_SIZE) { | 467 | if (recvMgr.praseIndex_ == ZERO_COPY_MEMORY_AGENT_RECV_QUEUE_SIZE) { |
| @@ -111,7 +111,7 @@ HcclResult InterGroupSyncProcess(std::vector<AscCommServerInfo>& serverList, u32 | |||
| 111 | ++(server.retryCnt); | 111 | ++(server.retryCnt); |
| 112 | HCCL_INFO("Group sync(%u-%u) will be retried at seq num %u.", curGroupIdx, groupId, server.msgPos); | 112 | HCCL_INFO("Group sync(%u-%u) will be retried at seq num %u.", curGroupIdx, groupId, server.msgPos); |
| 113 | } else { | 113 | } else { |
| 114 | - HCCL_ERROR("Group sync(%u-%u) failed, handle id %d, error code %u.", groupId, handleId, ret); | 114 | + HCCL_ERROR("Group sync(%u-%u) failed, handle id %d, error code %u.", curGroupIdx, groupId, handleId, ret); |
| 115 | return ret; | 115 | return ret; |
| 116 | } | 116 | } |
| 117 | return HCCL_SUCCESS; | 117 | return HCCL_SUCCESS; |
| @@ -279,7 +279,7 @@ ANONYMOUS_NAMESPACE_END | |||
| 279 | 279 | ||
| 280 | void AicpuKfcBatchwriteProcess::FinishProcess() | 280 | void AicpuKfcBatchwriteProcess::FinishProcess() |
| 281 | { | 281 | { |
| 282 | - HCCL_INFO("master over task is finish."); | 282 | + HCCL_INFO("master task is finished."); |
| 283 | g_sharedCtx.taskFinishFlag.store(true, std::memory_order_release); | 283 | g_sharedCtx.taskFinishFlag.store(true, std::memory_order_release); |
| 284 | } | 284 | } |
| 285 | 285 | ||
| @@ -375,7 +375,7 @@ HcclResult AicpuKfcBatchwriteProcess::RunSlaveRpcServerForApi(AicpuComContext* c | |||
| 375 | #endif | 375 | #endif |
| 376 | 376 | ||
| 377 | if (g_sharedCtx.taskFinishFlag.load(std::memory_order_acquire)) { | 377 | if (g_sharedCtx.taskFinishFlag.load(std::memory_order_acquire)) { |
| 378 | - HCCL_INFO("task is finish, slave process exit"); | 378 | + HCCL_INFO("task is finished, slave process exit"); |
| 379 | break; | 379 | break; |
| 380 | } | 380 | } |
| 381 | u8 needSendTotalNum = 0; | 381 | u8 needSendTotalNum = 0; |
| @@ -297,7 +297,7 @@ static HcclResult groupLaunch() | |||
| 297 | CHK_RET(hcclStreamSynchronize(it)); | 297 | CHK_RET(hcclStreamSynchronize(it)); |
| 298 | } | 298 | } |
| 299 | } | 299 | } |
| 300 | - HCCL_INFO("groupLauch Done!"); | 300 | + HCCL_INFO("groupLaunch Done!"); |
| 301 | return HCCL_SUCCESS; | 301 | return HCCL_SUCCESS; |
| 302 | } | 302 | } |
| 303 | 303 | ||
| @@ -282,8 +282,8 @@ HcclResult HcomGetCommByGroup(const char* group, std::shared_ptr<hccl::hcclComm> | |||
| 282 | CHK_PRT_RET( | 282 | CHK_PRT_RET( |
| 283 | ret != HCCL_SUCCESS, | 283 | ret != HCCL_SUCCESS, |
| 284 | HCCL_WARNING( | 284 | HCCL_WARNING( |
| 285 | - "[Get][CommByGroup]errNo[0x%016llx] group[%s]" | 285 | + "[Get][CommByGroup]errNo[0x%016llx] group[%s] " |
| 286 | - "group is not exist", | 286 | + "does not exist", |
| 287 | HCOM_ERROR_CODE(HCCL_E_NOT_FOUND), strGroup.c_str()), | 287 | HCOM_ERROR_CODE(HCCL_E_NOT_FOUND), strGroup.c_str()), |
| 288 | HCCL_E_NOT_FOUND); | 288 | HCCL_E_NOT_FOUND); |
| 289 | } else { | 289 | } else { |
| @@ -539,7 +539,7 @@ HcclResult DestroyFlag(const char* group, bool flag) | |||
| 539 | auto iter = hcomInfo.hcomGroupMap.find(strGroup); | 539 | auto iter = hcomInfo.hcomGroupMap.find(strGroup); |
| 540 | if (iter == hcomInfo.hcomGroupMap.end()) { | 540 | if (iter == hcomInfo.hcomGroupMap.end()) { |
| 541 | HCCL_ERROR( | 541 | HCCL_ERROR( |
| 542 | - "[Get][CommByGroup]errNo[0x%016llx] group[%s] group is not exist", HCOM_ERROR_CODE(HCCL_E_NOT_FOUND), | 542 | + "[Get][CommByGroup]errNo[0x%016llx] group[%s] does not exist", HCOM_ERROR_CODE(HCCL_E_NOT_FOUND), |
| 543 | strGroup.c_str()); | 543 | strGroup.c_str()); |
| 544 | return HCCL_E_NOT_FOUND; // 不存在该服务器内相关dev的对应信息 | 544 | return HCCL_E_NOT_FOUND; // 不存在该服务器内相关dev的对应信息 |
| 545 | } | 545 | } |
| @@ -558,7 +558,7 @@ HcclResult QueryDestroyFlag(const char* group) | |||
| 558 | auto iter = hcomInfo.hcomGroupMap.find(strGroup); | 558 | auto iter = hcomInfo.hcomGroupMap.find(strGroup); |
| 559 | if (iter == hcomInfo.hcomGroupMap.end()) { | 559 | if (iter == hcomInfo.hcomGroupMap.end()) { |
| 560 | HCCL_WARNING( | 560 | HCCL_WARNING( |
| 561 | - "[Get][CommByGroup]errNo[0x%016llx] group[%s] group is not exist", HCOM_ERROR_CODE(HCCL_E_AGAIN), | 561 | + "[Get][CommByGroup]errNo[0x%016llx] group[%s] does not exist", HCOM_ERROR_CODE(HCCL_E_AGAIN), |
| 562 | strGroup.c_str()); | 562 | strGroup.c_str()); |
| 563 | return HCCL_E_AGAIN; // 不存在该服务器内相关dev的对应信息 | 563 | return HCCL_E_AGAIN; // 不存在该服务器内相关dev的对应信息 |
| 564 | } | 564 | } |
| @@ -603,7 +603,7 @@ HcclResult HcomDestroyGroupImpl(const std::string& group) | |||
| 603 | auto iter = hcomInfo.hcomGroupMap.find(group); | 603 | auto iter = hcomInfo.hcomGroupMap.find(group); |
| 604 | if (iter == hcomInfo.hcomGroupMap.end()) { | 604 | if (iter == hcomInfo.hcomGroupMap.end()) { |
| 605 | HCCL_ERROR( | 605 | HCCL_ERROR( |
| 606 | - "[Destroy][Group]errNo[0x%016llx] group[%s] is not exist", HCOM_ERROR_CODE(HCCL_E_PARA), group.c_str()); | 606 | + "[Destroy][Group]errNo[0x%016llx] group[%s] does not exist", HCOM_ERROR_CODE(HCCL_E_PARA), group.c_str()); |
| 607 | return HCCL_E_PARA; | 607 | return HCCL_E_PARA; |
| 608 | } | 608 | } |
| 609 | 609 | ||
| @@ -645,7 +645,8 @@ HcclResult HcomDestroyGroup(const char* group) | |||
| 645 | CHK_RET(HcomDestroyGroupImplV2(group)); | 645 | CHK_RET(HcomDestroyGroupImplV2(group)); |
| 646 | auto iter = hcomInfo.hcomGroupMap.find(group); | 646 | auto iter = hcomInfo.hcomGroupMap.find(group); |
| 647 | if (iter == hcomInfo.hcomGroupMap.end()) { | 647 | if (iter == hcomInfo.hcomGroupMap.end()) { |
| 648 | - HCCL_ERROR("[Destroy][Group]errNo[0x%016llx] group[%s] is not exist", HCOM_ERROR_CODE(HCCL_E_PARA), group); | 648 | + HCCL_ERROR( |
| 649 | + "[Destroy][Group]errNo[0x%016llx] group[%s] does not exist", HCOM_ERROR_CODE(HCCL_E_PARA), group); | ||
| 649 | return HCCL_E_PARA; | 650 | return HCCL_E_PARA; |
| 650 | } | 651 | } |
| 651 | hcomInfo.hcomGroupMap.erase(group); | 652 | hcomInfo.hcomGroupMap.erase(group); |
| @@ -707,7 +708,7 @@ HcclResult HcomStoreBackloggedGroup(const std::string& group, const std::vector< | |||
| 707 | HCCL_INFO("[device not set]hcom store group[%s]", group.c_str()); | 708 | HCCL_INFO("[device not set]hcom store group[%s]", group.c_str()); |
| 708 | std::unique_lock<std::mutex> groupParaLock(g_backloggedGroupLock); | 709 | std::unique_lock<std::mutex> groupParaLock(g_backloggedGroupLock); |
| 709 | if (g_backloggedGroup.find(group) != g_backloggedGroup.end()) { | 710 | if (g_backloggedGroup.find(group) != g_backloggedGroup.end()) { |
| 710 | - HCCL_INFO("[Store][BackloggedGroup]group[%s] is existed", group.c_str()); | 711 | + HCCL_INFO("[Store][BackloggedGroup]group[%s] already exists", group.c_str()); |
| 711 | if (g_backloggedGroup[group] == groupRanks) { | 712 | if (g_backloggedGroup[group] == groupRanks) { |
| 712 | HCCL_ERROR("[Store][BackloggedGroup]group[%s] has been created", group.c_str()); | 713 | HCCL_ERROR("[Store][BackloggedGroup]group[%s] has been created", group.c_str()); |
| 713 | return HCCL_E_PARA; | 714 | return HCCL_E_PARA; |
| @@ -731,7 +732,7 @@ HcclResult HcomStoreBackloggedGroup(const std::string& group, const std::vector< | |||
| 731 | 732 | ||
| 732 | std::unique_lock<std::mutex> lock(hcomInfo.backloggedGroupLock); | 733 | std::unique_lock<std::mutex> lock(hcomInfo.backloggedGroupLock); |
| 733 | if (hcomInfo.backloggedGroup.find(group) != hcomInfo.backloggedGroup.end()) { | 734 | if (hcomInfo.backloggedGroup.find(group) != hcomInfo.backloggedGroup.end()) { |
| 734 | - HCCL_ERROR("[Store][BackloggedGroup]group[%s] is existed", group.c_str()); | 735 | + HCCL_ERROR("[Store][BackloggedGroup]group[%s] already exists", group.c_str()); |
| 735 | return HCCL_E_PARA; | 736 | return HCCL_E_PARA; |
| 736 | } else { | 737 | } else { |
| 737 | hcomInfo.backloggedGroup.insert({group, groupRanks}); | 738 | hcomInfo.backloggedGroup.insert({group, groupRanks}); |
| @@ -747,10 +748,10 @@ HcclResult HcomDestroyBackloggedGroup(const std::string& group) | |||
| 747 | if (hcomInfo.backloggedGroup.find(group) == hcomInfo.backloggedGroup.end()) { | 748 | if (hcomInfo.backloggedGroup.find(group) == hcomInfo.backloggedGroup.end()) { |
| 748 | if (!hcomInfo.isHcomInit) { | 749 | if (!hcomInfo.isHcomInit) { |
| 749 | HCCL_WARNING( | 750 | HCCL_WARNING( |
| 750 | - "[Destroy][BackloggedGroup]group[%s] is not existed, and hcom has not been inited yet", group.c_str()); | 751 | + "[Destroy][BackloggedGroup]group[%s] does not exist, and hcom has not been inited yet", group.c_str()); |
| 751 | return HCCL_SUCCESS; | 752 | return HCCL_SUCCESS; |
| 752 | } else { | 753 | } else { |
| 753 | - HCCL_ERROR("[Destroy][BackloggedGroup]group[%s] is not existed", group.c_str()); | 754 | + HCCL_ERROR("[Destroy][BackloggedGroup]group[%s] does not exist", group.c_str()); |
| 754 | return HCCL_E_PARA; | 755 | return HCCL_E_PARA; |
| 755 | } | 756 | } |
| 756 | } else { | 757 | } else { |
| @@ -780,7 +781,7 @@ HcclResult HcomGetbackloggedByGroup(const char* group, std::vector<u32>& groupRa | |||
| 780 | if (iter == hcomInfo.backloggedGroup.end()) { | 781 | if (iter == hcomInfo.backloggedGroup.end()) { |
| 781 | groupSize = 0; | 782 | groupSize = 0; |
| 782 | HCCL_DEBUG( | 783 | HCCL_DEBUG( |
| 783 | - "[Get][CommByGroup]errNo[0x%016llx] group[%s] is not exist", HCOM_ERROR_CODE(HCCL_E_NOT_FOUND), group); | 784 | + "[Get][CommByGroup]errNo[0x%016llx] group[%s] does not exist", HCOM_ERROR_CODE(HCCL_E_NOT_FOUND), group); |
| 784 | return HCCL_SUCCESS; // 不存在该服务器内相关dev的对应信息 | 785 | return HCCL_SUCCESS; // 不存在该服务器内相关dev的对应信息 |
| 785 | } | 786 | } |
| 786 | groupRanks = iter->second; | 787 | groupRanks = iter->second; |
| @@ -800,7 +801,7 @@ HcclResult HcomQueryGroupRef(const char* group, u32& groupRef) | |||
| 800 | auto iter = hcomInfo.hcomGroupMap.find(strGroup); | 801 | auto iter = hcomInfo.hcomGroupMap.find(strGroup); |
| 801 | if (iter == hcomInfo.hcomGroupMap.end()) { | 802 | if (iter == hcomInfo.hcomGroupMap.end()) { |
| 802 | HCCL_WARNING( | 803 | HCCL_WARNING( |
| 803 | - "[Get][CommByGroup]errNo[0x%016llx] group[%s] group is not exist", HCOM_ERROR_CODE(HCCL_E_AGAIN), | 804 | + "[Get][CommByGroup]errNo[0x%016llx] group[%s] does not exist", HCOM_ERROR_CODE(HCCL_E_AGAIN), |
| 804 | strGroup.c_str()); | 805 | strGroup.c_str()); |
| 805 | return HCCL_E_AGAIN; // 不存在该服务器内相关dev的对应信息 | 806 | return HCCL_E_AGAIN; // 不存在该服务器内相关dev的对应信息 |
| 806 | } | 807 | } |
| @@ -956,7 +957,7 @@ HcclResult GetGroupRankInfo(const char* group, RankInfoType rankType, u32 inPara | |||
| 956 | CHK_PRT_RET( | 957 | CHK_PRT_RET( |
| 957 | (iter->second).groupRanks.empty(), | 958 | (iter->second).groupRanks.empty(), |
| 958 | HCCL_ERROR( | 959 | HCCL_ERROR( |
| 959 | - "[Get][GroupRankInfo]errNo[0x%016llx] group[%s]" | 960 | + "[Get][GroupRankInfo]errNo[0x%016llx] group[%s] " |
| 960 | "ranks is empty", | 961 | "ranks is empty", |
| 961 | HCOM_ERROR_CODE(HCCL_E_INTERNAL), group), | 962 | HCOM_ERROR_CODE(HCCL_E_INTERNAL), group), |
| 962 | HCCL_E_INTERNAL); | 963 | HCCL_E_INTERNAL); |
| @@ -1245,7 +1245,7 @@ HcclResult HcclCreateSubCommConfigInner( | |||
| 1245 | if (errorFlag) { | 1245 | if (errorFlag) { |
| 1246 | HCCL_ERROR( | 1246 | HCCL_ERROR( |
| 1247 | "[%s]Create sub communication failed, return[0x%016llx], " | 1247 | "[%s]Create sub communication failed, return[0x%016llx], " |
| 1248 | - "rankNum[%u], subCommRankId[%u], sub commm identifier[%s], server[%s], logicDevId[%d]", | 1248 | + "rankNum[%u], subCommRankId[%u], sub comm identifier[%s], server[%s], logicDevId[%d]", |
| 1249 | __func__, HCCL_ERROR_CODE(ret), rankNum, subCommRankId, commIdentifier.c_str(), | 1249 | __func__, HCCL_ERROR_CODE(ret), rankNum, subCommRankId, commIdentifier.c_str(), |
| 1250 | GetLocalServerId(subParams.serverId).c_str(), subParams.logicDevId); | 1250 | GetLocalServerId(subParams.serverId).c_str(), subParams.logicDevId); |
| 1251 | (void)HcclCommDestroy(pComm.get()); | 1251 | (void)HcclCommDestroy(pComm.get()); |
| @@ -1258,7 +1258,7 @@ HcclResult HcclCreateSubCommConfigInner( | |||
| 1258 | hcclNslbDp::GetInstance().SendTableFir(subCommRankId); | 1258 | hcclNslbDp::GetInstance().SendTableFir(subCommRankId); |
| 1259 | 1259 | ||
| 1260 | HCCL_RUN_INFO( | 1260 | HCCL_RUN_INFO( |
| 1261 | - "%s success, sub commm identifier[%s], rankNum[%u], rank[%u], server[%s], device[%d].", __func__, | 1261 | + "%s success, sub comm identifier[%s], rankNum[%u], rank[%u], server[%s], device[%d].", __func__, |
| 1262 | commIdentifier.c_str(), subRankTable.rankNum, subCommRankId, subParams.serverId.c_str(), subParams.logicDevId); | 1262 | commIdentifier.c_str(), subRankTable.rankNum, subCommRankId, subParams.serverId.c_str(), subParams.logicDevId); |
| 1263 | #endif | 1263 | #endif |
| 1264 | return HCCL_SUCCESS; | 1264 | return HCCL_SUCCESS; |
| @@ -1370,7 +1370,7 @@ HcclResult HcclCreateSubCommConfig( | |||
| 1370 | /* 关键状态记录 */ | 1370 | /* 关键状态记录 */ |
| 1371 | HCCL_RUN_INFO( | 1371 | HCCL_RUN_INFO( |
| 1372 | "[HCCL_TRACE]%s success, take time [%lld]us, " | 1372 | "[HCCL_TRACE]%s success, take time [%lld]us, " |
| 1373 | - "sub commm identifier[%s], rankNum[%u], rank[%u], deviceLogicId[%d]", | 1373 | + "sub comm identifier[%s], rankNum[%u], rank[%u], deviceLogicId[%d]", |
| 1374 | __func__, DURATION_US(TIME_NOW() - startut), commConfig.GetConfigCommName().c_str(), rankNum, subCommRankId, | 1374 | __func__, DURATION_US(TIME_NOW() - startut), commConfig.GetConfigCommName().c_str(), rankNum, subCommRankId, |
| 1375 | deviceLogicId); | 1375 | deviceLogicId); |
| 1376 | return HCCL_SUCCESS; | 1376 | return HCCL_SUCCESS; |
| @@ -287,7 +287,7 @@ HcclResult OpUnfoldCache::DumpSqeContent(const uint8_t* sqePtr, const uint8_t sq | |||
| 287 | memcpyAsyncSqePtr->dst_addr_low, memcpyAsyncSqePtr->dst_addr_high); | 287 | memcpyAsyncSqePtr->dst_addr_low, memcpyAsyncSqePtr->dst_addr_high); |
| 288 | PLF_CONFIG_DEBUG( | 288 | PLF_CONFIG_DEBUG( |
| 289 | PLF_TASK, | 289 | PLF_TASK, |
| 290 | - "[OpUnfoldCache][DumpSqeContent] linkType[%u] resvered[0][%u] resvered[1][%u] resvered[2][%u] " | 290 | + "[OpUnfoldCache][DumpSqeContent] linkType[%u] reserved[0][%u] reserved[1][%u] reserved[2][%u] " |
| 291 | "reslast[0][%u] reslast[1][%u] reslast[2][%u]", | 291 | "reslast[0][%u] reslast[1][%u] reslast[2][%u]", |
| 292 | memcpyAsyncSqePtr->linkType, memcpyAsyncSqePtr->resvered[0], memcpyAsyncSqePtr->resvered[1], | 292 | memcpyAsyncSqePtr->linkType, memcpyAsyncSqePtr->resvered[0], memcpyAsyncSqePtr->resvered[1], |
| 293 | memcpyAsyncSqePtr->resvered[2], memcpyAsyncSqePtr->reslast[0], memcpyAsyncSqePtr->reslast[1], | 293 | memcpyAsyncSqePtr->resvered[2], memcpyAsyncSqePtr->reslast[0], memcpyAsyncSqePtr->reslast[1], |
| @@ -731,7 +731,7 @@ HcclResult OpUnfoldCache::DumpSqeContent(const uint8_t* sqePtr, const uint8_t sq | |||
| 731 | placeholderSqePtr->res3); | 731 | placeholderSqePtr->res3); |
| 732 | PLF_CONFIG_DEBUG( | 732 | PLF_CONFIG_DEBUG( |
| 733 | PLF_TASK, | 733 | PLF_TASK, |
| 734 | - "[OpUnfoldCache][DumpSqeContent] u.cache_memcpy_record_task_info: length[%u]" | 734 | + "[OpUnfoldCache][DumpSqeContent] u.cache_memcpy_record_task_info: length[%u] " |
| 735 | "src_addr_low[0x%08x] src_addr_high[0x%08x] dst_addr_low[0x%08x] dst_addr_high[0x%08x]", | 735 | "src_addr_low[0x%08x] src_addr_high[0x%08x] dst_addr_low[0x%08x] dst_addr_high[0x%08x]", |
| 736 | placeholderSqePtr->u.cache_memcpy_record_task_info.length, | 736 | placeholderSqePtr->u.cache_memcpy_record_task_info.length, |
| 737 | placeholderSqePtr->u.cache_memcpy_record_task_info.src_addr_low, | 737 | placeholderSqePtr->u.cache_memcpy_record_task_info.src_addr_low, |
| @@ -740,7 +740,7 @@ HcclResult OpUnfoldCache::DumpSqeContent(const uint8_t* sqePtr, const uint8_t sq | |||
| 740 | placeholderSqePtr->u.cache_memcpy_record_task_info.dst_addr_high); | 740 | placeholderSqePtr->u.cache_memcpy_record_task_info.dst_addr_high); |
| 741 | PLF_CONFIG_DEBUG( | 741 | PLF_CONFIG_DEBUG( |
| 742 | PLF_TASK, | 742 | PLF_TASK, |
| 743 | - "[OpUnfoldCache][DumpSqeContent] u.cache_memcpy_record_task_info: opcode[%u]" | 743 | + "[OpUnfoldCache][DumpSqeContent] u.cache_memcpy_record_task_info: opcode[%u] " |
| 744 | "partid[%u] kernel_credit[%u] linkType[%u]", | 744 | "partid[%u] kernel_credit[%u] linkType[%u]", |
| 745 | placeholderSqePtr->u.cache_memcpy_record_task_info.opcode, | 745 | placeholderSqePtr->u.cache_memcpy_record_task_info.opcode, |
| 746 | placeholderSqePtr->u.cache_memcpy_record_task_info.partid, | 746 | placeholderSqePtr->u.cache_memcpy_record_task_info.partid, |
| @@ -662,7 +662,7 @@ HcclResult TransportHeterogEventRoce::ParseTagSrqes(const struct ibv_wc* wc, int | |||
| 662 | gQpnToTransportMap[wc[i].qp_num]->SaveEnvelope(envelopSummary); | 662 | gQpnToTransportMap[wc[i].qp_num]->SaveEnvelope(envelopSummary); |
| 663 | } else { | 663 | } else { |
| 664 | HCCL_ERROR( | 664 | HCCL_ERROR( |
| 665 | - "The transport is no exist, wrId[%llu] status[%u] opcode[%u] qpn[%u]", wc[i].wr_id, wc[i].status, | 665 | + "The transport does not exist, wrId[%llu] status[%u] opcode[%u] qpn[%u]", wc[i].wr_id, wc[i].status, |
| 666 | wc[i].opcode, wc[i].qp_num); | 666 | wc[i].opcode, wc[i].qp_num); |
| 667 | return HCCL_E_PTR; | 667 | return HCCL_E_PTR; |
| 668 | } | 668 | } |
| @@ -689,7 +689,7 @@ HcclResult TransportHeterogEventRoce::ParseDataSrqes(const struct ibv_wc* wc, in | |||
| 689 | gQpnToSqMaxWrMap[wc[i].qp_num]++; | 689 | gQpnToSqMaxWrMap[wc[i].qp_num]++; |
| 690 | } else { | 690 | } else { |
| 691 | HCCL_ERROR( | 691 | HCCL_ERROR( |
| 692 | - "The qpn is no exist, wrId[%llu] status[%u] opcode[%u] qpn[%u]", wc[i].wr_id, wc[i].status, | 692 | + "The qpn does not exist, wrId[%llu] status[%u] opcode[%u] qpn[%u]", wc[i].wr_id, wc[i].status, |
| 693 | wc[i].opcode, wc[i].qp_num); | 693 | wc[i].opcode, wc[i].qp_num); |
| 694 | return HCCL_E_PTR; | 694 | return HCCL_E_PTR; |
| 695 | } | 695 | } |
| @@ -206,7 +206,7 @@ HcclResult MemNameRepository::OpenIpcMem( | |||
| 206 | s32 sret = memcpy_s(memName.ipcName, HCCL_IPC_MEM_NAME_LEN, name, nameLen); | 206 | s32 sret = memcpy_s(memName.ipcName, HCCL_IPC_MEM_NAME_LEN, name, nameLen); |
| 207 | if (sret != EOK) { | 207 | if (sret != EOK) { |
| 208 | HCCL_ERROR( | 208 | HCCL_ERROR( |
| 209 | - "[Open][IpcMem]errNo[0x%016llx] In SecIpcName, memset_s failed. errorno[%d], params:" | 209 | + "[Open][IpcMem]errNo[0x%016llx] In SecIpcName, memset_s failed. errno[%d], params:" |
| 210 | "dest len[%u], src len[%u]", | 210 | "dest len[%u], src len[%u]", |
| 211 | HCCL_ERROR_CODE(HCCL_E_SYSCALL), sret, HCCL_IPC_MEM_NAME_LEN, nameLen); | 211 | HCCL_ERROR_CODE(HCCL_E_SYSCALL), sret, HCCL_IPC_MEM_NAME_LEN, nameLen); |
| 212 | return HCCL_E_SYSCALL; | 212 | return HCCL_E_SYSCALL; |
| @@ -280,7 +280,7 @@ void MemNameRepository::DestroyIpcMem(void* ptr, u64 size, bool isSioToHccs) | |||
| 280 | if (unavailable_) { | 280 | if (unavailable_) { |
| 281 | ClearMemNameRepositoryImpl(); | 281 | ClearMemNameRepositoryImpl(); |
| 282 | unavailable_ = false; | 282 | unavailable_ = false; |
| 283 | - HCCL_RUN_INFO("DestoryIpcMem unavailable_[%d]", unavailable_); | 283 | + HCCL_RUN_INFO("DestroyIpcMem unavailable_[%d]", unavailable_); |
| 284 | return; | 284 | return; |
| 285 | } | 285 | } |
| 286 | 286 | ||
| @@ -249,7 +249,7 @@ HcclResult TransportIpcMem::Write(const HcclBuf& remoteMem, const HcclBuf& local | |||
| 249 | CHK_PRT_RET( | 249 | CHK_PRT_RET( |
| 250 | (localMem.len == 0U) || (remoteMem.len == 0U), | 250 | (localMem.len == 0U) || (remoteMem.len == 0U), |
| 251 | HCCL_ERROR( | 251 | HCCL_ERROR( |
| 252 | - "[TransportIpcMem]localMem size[%llu] or remoteMem size[%llu]is invalid", localMem.len, remoteMem.len), | 252 | + "[TransportIpcMem]localMem size[%llu] or remoteMem size[%llu] is invalid", localMem.len, remoteMem.len), |
| 253 | HCCL_E_PARA); | 253 | HCCL_E_PARA); |
| 254 | CHK_PRT_RET(stream == nullptr, HCCL_ERROR("[TransportIpcMem]stream is invalid"), HCCL_E_PARA); | 254 | CHK_PRT_RET(stream == nullptr, HCCL_ERROR("[TransportIpcMem]stream is invalid"), HCCL_E_PARA); |
| 255 | 255 | ||
| @@ -267,7 +267,7 @@ HcclResult TransportIpcMem::Write(const RmaOpMem& remoteMem, const RmaOpMem& loc | |||
| 267 | CHK_PRT_RET( | 267 | CHK_PRT_RET( |
| 268 | (localMem.size == 0U) || (remoteMem.size == 0U), | 268 | (localMem.size == 0U) || (remoteMem.size == 0U), |
| 269 | HCCL_ERROR( | 269 | HCCL_ERROR( |
| 270 | - "[TransportIpcMem]localMem size[%llu] or remoteMem size[%llu]is invalid", localMem.size, remoteMem.size), | 270 | + "[TransportIpcMem]localMem size[%llu] or remoteMem size[%llu] is invalid", localMem.size, remoteMem.size), |
| 271 | HCCL_E_PARA); | 271 | HCCL_E_PARA); |
| 272 | CHK_PRT_RET(stream == nullptr, HCCL_ERROR("[TransportIpcMem]stream is invalid"), HCCL_E_PARA); | 272 | CHK_PRT_RET(stream == nullptr, HCCL_ERROR("[TransportIpcMem]stream is invalid"), HCCL_E_PARA); |
| 273 | 273 | ||
| @@ -285,7 +285,7 @@ HcclResult TransportIpcMem::Read(const HcclBuf& localMem, const HcclBuf& remoteM | |||
| 285 | CHK_PRT_RET( | 285 | CHK_PRT_RET( |
| 286 | (localMem.len == 0U) || (remoteMem.len == 0U), | 286 | (localMem.len == 0U) || (remoteMem.len == 0U), |
| 287 | HCCL_ERROR( | 287 | HCCL_ERROR( |
| 288 | - "[TransportIpcMem]localMem size[%llu] or remoteMem size[%llu]is invalid", localMem.len, remoteMem.len), | 288 | + "[TransportIpcMem]localMem size[%llu] or remoteMem size[%llu] is invalid", localMem.len, remoteMem.len), |
| 289 | HCCL_E_PARA); | 289 | HCCL_E_PARA); |
| 290 | CHK_PRT_RET(stream == nullptr, HCCL_ERROR("[TransportIpcMem]stream is invalid"), HCCL_E_PARA); | 290 | CHK_PRT_RET(stream == nullptr, HCCL_ERROR("[TransportIpcMem]stream is invalid"), HCCL_E_PARA); |
| 291 | 291 | ||
| @@ -303,7 +303,7 @@ HcclResult TransportIpcMem::Read(const RmaOpMem& localMem, const RmaOpMem& remot | |||
| 303 | CHK_PRT_RET( | 303 | CHK_PRT_RET( |
| 304 | (localMem.size == 0U) || (remoteMem.size == 0U), | 304 | (localMem.size == 0U) || (remoteMem.size == 0U), |
| 305 | HCCL_ERROR( | 305 | HCCL_ERROR( |
| 306 | - "[TransportIpcMem]localMem size[%llu] or remoteMem size[%llu]is invalid", localMem.size, remoteMem.size), | 306 | + "[TransportIpcMem]localMem size[%llu] or remoteMem size[%llu] is invalid", localMem.size, remoteMem.size), |
| 307 | HCCL_E_PARA); | 307 | HCCL_E_PARA); |
| 308 | CHK_PRT_RET(stream == nullptr, HCCL_ERROR("[TransportIpcMem]stream is invalid"), HCCL_E_PARA); | 308 | CHK_PRT_RET(stream == nullptr, HCCL_ERROR("[TransportIpcMem]stream is invalid"), HCCL_E_PARA); |
| 309 | 309 | ||
| @@ -329,7 +329,7 @@ HcclResult TransportIpcMem::GetMemInfo(u32& lkey, u32& rkey, HcclBuf& localMem, | |||
| 329 | CHK_PRT_RET( | 329 | CHK_PRT_RET( |
| 330 | (localMem.len == 0U) || (remoteMem.len == 0U), | 330 | (localMem.len == 0U) || (remoteMem.len == 0U), |
| 331 | HCCL_ERROR( | 331 | HCCL_ERROR( |
| 332 | - "[TransportIpcMem] localMem size[%llu] or remoteMem size[%llu]is invalid", localMem.len, remoteMem.len), | 332 | + "[TransportIpcMem] localMem size[%llu] or remoteMem size[%llu] is invalid", localMem.len, remoteMem.len), |
| 333 | HCCL_E_PARA); | 333 | HCCL_E_PARA); |
| 334 | 334 | ||
| 335 | RmaBufferSlice localRmaBufferSlice{}; | 335 | RmaBufferSlice localRmaBufferSlice{}; |
| @@ -471,7 +471,7 @@ HcclResult TransportBase::SendNotifyReadyMesg() | |||
| 471 | ret != HCCL_SUCCESS, | 471 | ret != HCCL_SUCCESS, |
| 472 | HCCL_ERROR( | 472 | HCCL_ERROR( |
| 473 | "[Send][IpcNotifyReadyMesg]errNo[0x%016llx]In send notify ready mesg, send read msg failed. remote " | 473 | "[Send][IpcNotifyReadyMesg]errNo[0x%016llx]In send notify ready mesg, send read msg failed. remote " |
| 474 | - "userrank[%u] notify locak rank[%u]", | 474 | + "userrank[%u] notify local rank[%u]", |
| 475 | HCCL_ERROR_CODE(ret), machinePara_.remoteUserrank, machinePara_.localUserrank), | 475 | HCCL_ERROR_CODE(ret), machinePara_.remoteUserrank, machinePara_.localUserrank), |
| 476 | ret); | 476 | ret); |
| 477 | 477 | ||
| @@ -520,7 +520,7 @@ HcclResult TransportBase::SendDeviceIpcNotifyReadyMesg() | |||
| 520 | ret != HCCL_SUCCESS, | 520 | ret != HCCL_SUCCESS, |
| 521 | HCCL_ERROR( | 521 | HCCL_ERROR( |
| 522 | "[Send][IpcNotifyReadyMesg]errNo[0x%016llx]In send notify ready mesg, send read msg failed. remote " | 522 | "[Send][IpcNotifyReadyMesg]errNo[0x%016llx]In send notify ready mesg, send read msg failed. remote " |
| 523 | - "userrank[%u] notify locak rank[%u]", | 523 | + "userrank[%u] notify local rank[%u]", |
| 524 | HCCL_ERROR_CODE(ret), machinePara_.remoteUserrank, machinePara_.localUserrank), | 524 | HCCL_ERROR_CODE(ret), machinePara_.remoteUserrank, machinePara_.localUserrank), |
| 525 | ret); | 525 | ret); |
| 526 | 526 | ||
| @@ -545,7 +545,7 @@ HcclResult TransportBase::SendDeviceIpcNotifyDoneMesg() | |||
| 545 | ret != HCCL_SUCCESS, | 545 | ret != HCCL_SUCCESS, |
| 546 | HCCL_ERROR( | 546 | HCCL_ERROR( |
| 547 | "[Send][IpcNotifyReadyMesg]errNo[0x%016llx]In send notify ready mesg, send read msg failed. remote " | 547 | "[Send][IpcNotifyReadyMesg]errNo[0x%016llx]In send notify ready mesg, send read msg failed. remote " |
| 548 | - "userrank[%u] notify locak rank[%u]", | 548 | + "userrank[%u] notify local rank[%u]", |
| 549 | HCCL_ERROR_CODE(ret), machinePara_.remoteUserrank, machinePara_.localUserrank), | 549 | HCCL_ERROR_CODE(ret), machinePara_.remoteUserrank, machinePara_.localUserrank), |
| 550 | ret); | 550 | ret); |
| 551 | 551 | ||
| @@ -114,7 +114,7 @@ public: | |||
| 114 | { | 114 | { |
| 115 | std::string temp = stream.str(); | 115 | std::string temp = stream.str(); |
| 116 | if (start >= temp.length()) { | 116 | if (start >= temp.length()) { |
| 117 | - HCCL_ERROR("[SplictStream]start[%llu] is bigger than stream length[%llu]", start, temp.length()); | 117 | + HCCL_ERROR("[SplitStream]start[%llu] is bigger than stream length[%llu]", start, temp.length()); |
| 118 | return ""; | 118 | return ""; |
| 119 | } | 119 | } |
| 120 | 120 | ||
| @@ -62,7 +62,7 @@ HcclResult SalStrToULong(const std::string str, int base, u32& val) | |||
| 62 | HCCL_ERROR("[Transform][StrToULong]stoull out of range, str[%s] base[%d] val[%u]", str.c_str(), base, val); | 62 | HCCL_ERROR("[Transform][StrToULong]stoull out of range, str[%s] base[%d] val[%u]", str.c_str(), base, val); |
| 63 | return HCCL_E_PARA; | 63 | return HCCL_E_PARA; |
| 64 | } catch (...) { | 64 | } catch (...) { |
| 65 | - HCCL_ERROR("[Transform][StrToULong]stoull catch errror, str[%s] base[%d] val[%u]", str.c_str(), base, val); | 65 | + HCCL_ERROR("[Transform][StrToULong]stoull catch error, str[%s] base[%d] val[%u]", str.c_str(), base, val); |
| 66 | return HCCL_E_PARA; | 66 | return HCCL_E_PARA; |
| 67 | } | 67 | } |
| 68 | return HCCL_SUCCESS; | 68 | return HCCL_SUCCESS; |
| @@ -74,7 +74,7 @@ u64 SalGetCurrentTimestamp() | |||
| 74 | struct timeval tv {}; | 74 | struct timeval tv {}; |
| 75 | int ret = gettimeofday(&tv, nullptr); | 75 | int ret = gettimeofday(&tv, nullptr); |
| 76 | if (ret != 0) { | 76 | if (ret != 0) { |
| 77 | - HCCL_ERROR("[Get][tCurrentTimestamp]get timestamp fail, return[%d].", ret); | 77 | + HCCL_ERROR("[Get][GetCurrentTimestamp]get timestamp fail, return[%d].", ret); |
| 78 | } | 78 | } |
| 79 | timestamp = tv.tv_sec * 1000000 + tv.tv_usec; // 1000000: 单位转换 秒 -> 微秒 | 79 | timestamp = tv.tv_sec * 1000000 + tv.tv_usec; // 1000000: 单位转换 秒 -> 微秒 |
| 80 | return timestamp; | 80 | return timestamp; |
| @@ -823,7 +823,7 @@ HcclResult CommunicatorImpl::AllocCollOpResource(const CollOpParams& opParams, v | |||
| 823 | } catch (...) { | 823 | } catch (...) { |
| 824 | SetCommStatus(CommStatus::COMM_READY); | 824 | SetCommStatus(CommStatus::COMM_READY); |
| 825 | HCCL_ERROR("AllocCollOpResource OperatorParams: %s", opParams.Describe().c_str()); | 825 | HCCL_ERROR("AllocCollOpResource OperatorParams: %s", opParams.Describe().c_str()); |
| 826 | - HCCL_ERROR("Unkown error occurs!"); | 826 | + HCCL_ERROR("Unknown error occurs!"); |
| 827 | return HcclResult::HCCL_E_INTERNAL; | 827 | return HcclResult::HCCL_E_INTERNAL; |
| 828 | } | 828 | } |
| 829 | return HcclResult::HCCL_SUCCESS; | 829 | return HcclResult::HCCL_SUCCESS; |
| @@ -916,8 +916,8 @@ void CommunicatorImpl::RegisterOffloadScratchBuffer( | |||
| 916 | if (scratchBuffer) { | 916 | if (scratchBuffer) { |
| 917 | offloadScrachBufferMap[opTag] = scratchBuffer; | 917 | offloadScrachBufferMap[opTag] = scratchBuffer; |
| 918 | HCCL_RUN_INFO( | 918 | HCCL_RUN_INFO( |
| 919 | - "[CommunicatorImpl] offloadScratchBuffer register, opTag[%s], offloadScrachBufferAddr[%llu], " | 919 | + "[CommunicatorImpl] offloadScratchBuffer register, opTag[%s], offloadScratchBufferAddr[%llu], " |
| 920 | - "offloadScrachBufferBufSize[%llu]M", | 920 | + "offloadScratchBufferBufSize[%llu]M", |
| 921 | opTag.c_str(), scratchBuffer->GetAddr(), scratchBuffer->GetSize() / HCCL_CCL_COMM_FIXED_CALC_BUFFER_SIZE); | 921 | opTag.c_str(), scratchBuffer->GetAddr(), scratchBuffer->GetSize() / HCCL_CCL_COMM_FIXED_CALC_BUFFER_SIZE); |
| 922 | } | 922 | } |
| 923 | } | 923 | } |
| @@ -1198,7 +1198,7 @@ void CommunicatorImpl::CovertToCurrentCollOperator( | |||
| 1198 | if (offloadScrachBufferMap.find(opTag) != offloadScrachBufferMap.end()) { | 1198 | if (offloadScrachBufferMap.find(opTag) != offloadScrachBufferMap.end()) { |
| 1199 | auto scratchMem = offloadScrachBufferMap[opTag]; | 1199 | auto scratchMem = offloadScrachBufferMap[opTag]; |
| 1200 | HCCL_INFO( | 1200 | HCCL_INFO( |
| 1201 | - "[CommunicatorImpl::CovertToCurrentCollOperator] offloadScrachBufferMap[%s] is [%s]", opTag.c_str(), | 1201 | + "[CommunicatorImpl::CovertToCurrentCollOperator] offloadScratchBufferMap[%s] is [%s]", opTag.c_str(), |
| 1202 | scratchMem->Describe().c_str()); | 1202 | scratchMem->Describe().c_str()); |
| 1203 | currentCollOperator->scratchMem = scratchMem; | 1203 | currentCollOperator->scratchMem = scratchMem; |
| 1204 | } | 1204 | } |
| @@ -3801,7 +3801,7 @@ HcclResult CommunicatorImpl::GetInstSizeListByNetLayer(uint32_t netLayer, uint32 | |||
| 3801 | auto ret = rankGraph->GetNetInstanceList(netLayer, instSizeVec, size); | 3801 | auto ret = rankGraph->GetNetInstanceList(netLayer, instSizeVec, size); |
| 3802 | if (ret != HCCL_SUCCESS) { | 3802 | if (ret != HCCL_SUCCESS) { |
| 3803 | HCCL_ERROR( | 3803 | HCCL_ERROR( |
| 3804 | - "[CommunicatorImpl::GetInstSizeListByNetLayer] Failed to get instSzie[%p] at netLayer[%u]", listSize, | 3804 | + "[CommunicatorImpl::GetInstSizeListByNetLayer] Failed to get instSize[%p] at netLayer[%u]", listSize, |
| 3805 | netLayer); | 3805 | netLayer); |
| 3806 | return ret; | 3806 | return ret; |
| 3807 | } | 3807 | } |
| @@ -4076,7 +4076,7 @@ HcclResult CommunicatorImpl::GetEndpointNum(uint32_t layer, uint32_t topoInstId, | |||
| 4076 | HcclResult ret = rankGraph->GetEndpointNum(layer, topoInstId, num); | 4076 | HcclResult ret = rankGraph->GetEndpointNum(layer, topoInstId, num); |
| 4077 | if (ret != HCCL_SUCCESS) { | 4077 | if (ret != HCCL_SUCCESS) { |
| 4078 | HCCL_ERROR( | 4078 | HCCL_ERROR( |
| 4079 | - "[CommunicatorImpl::GetEndpointNum] Faild to get endpoint num at netLayer [%u] with topoInstId[%u]", layer, | 4079 | + "[CommunicatorImpl::GetEndpointNum] Failed to get endpoint num at netLayer [%u] with topoInstId[%u]", layer, |
| 4080 | topoInstId); | 4080 | topoInstId); |
| 4081 | return ret; | 4081 | return ret; |
| 4082 | } | 4082 | } |
| @@ -4103,7 +4103,7 @@ HcclResult CommunicatorImpl::GetEndpointInfo( | |||
| 4103 | CHK_PTR_NULL(rankGraph); | 4103 | CHK_PTR_NULL(rankGraph); |
| 4104 | HcclResult ret = rankGraph->GetEndpointInfo(rankId, endPointDesc, endpointAttr, infoLen, info); | 4104 | HcclResult ret = rankGraph->GetEndpointInfo(rankId, endPointDesc, endpointAttr, infoLen, info); |
| 4105 | if (ret != HCCL_SUCCESS) { | 4105 | if (ret != HCCL_SUCCESS) { |
| 4106 | - HCCL_ERROR("[CommunicatorImpl::GetEndpointInfo] Faild to get endpoint info with rank [%u]", rankId); | 4106 | + HCCL_ERROR("[CommunicatorImpl::GetEndpointInfo] Failed to get endpoint info with rank [%u]", rankId); |
| 4107 | return ret; | 4107 | return ret; |
| 4108 | } | 4108 | } |
| 4109 | return HCCL_SUCCESS; | 4109 | return HCCL_SUCCESS; |
| @@ -4239,7 +4239,7 @@ HcclResult CommunicatorImpl::ClearOpResource(const std::string& opTag) | |||
| 4239 | CHK_RET(GetStreamManager().offload->ClearOpStream(opTag)); | 4239 | CHK_RET(GetStreamManager().offload->ClearOpStream(opTag)); |
| 4240 | // 清空workspaceMem资源 | 4240 | // 清空workspaceMem资源 |
| 4241 | offloadScrachBufferMap.erase(opTag); | 4241 | offloadScrachBufferMap.erase(opTag); |
| 4242 | - HCCL_RUN_INFO("[CommunicatorImpl][%s] offloadScrachBuffer free, opTag[%s]", __func__, opTag.c_str()); | 4242 | + HCCL_RUN_INFO("[CommunicatorImpl][%s] offloadScratchBuffer free, opTag[%s]", __func__, opTag.c_str()); |
| 4243 | // 清空input/output/scratch资源 | 4243 | // 清空input/output/scratch资源 |
| 4244 | CHK_RET(GetDataBufferManager().Deregister(opTag)); | 4244 | CHK_RET(GetDataBufferManager().Deregister(opTag)); |
| 4245 | CHK_RET(GetLocalRmaBufManager().Dereg(opTag)); | 4245 | CHK_RET(GetLocalRmaBufManager().Dereg(opTag)); |
| @@ -76,7 +76,7 @@ __attribute__((visibility("default"))) uint32_t RunDpuRpcSrvLaunch(const uint64_ | |||
| 76 | uint8_t newFlag = TASK_TERMINATE_RESPONSE; | 76 | uint8_t newFlag = TASK_TERMINATE_RESPONSE; |
| 77 | errno_t cpyRet = memcpy_s(static_cast<uint8_t*>(params->deviceMem), sizeof(newFlag), &newFlag, sizeof(newFlag)); | 77 | errno_t cpyRet = memcpy_s(static_cast<uint8_t*>(params->deviceMem), sizeof(newFlag), &newFlag, sizeof(newFlag)); |
| 78 | if (cpyRet != EOK) { | 78 | if (cpyRet != EOK) { |
| 79 | - HCCL_ERROR("set eixt flag failed: %d", cpyRet); | 79 | + HCCL_ERROR("set exit flag failed: %d", cpyRet); |
| 80 | return HCCL_E_INTERNAL; | 80 | return HCCL_E_INTERNAL; |
| 81 | } | 81 | } |
| 82 | } | 82 | } |
| @@ -381,7 +381,7 @@ bool ProfilingHandlerLite::IsProfSwitchOn(ProfilingLevel level) | |||
| 381 | bool ProfilingHandlerLite::IsL1fromOffToOn() | 381 | bool ProfilingHandlerLite::IsL1fromOffToOn() |
| 382 | { | 382 | { |
| 383 | if (((!GetProfL1State()) && IsProfSwitchOn(ProfilingLevel::L1))) { | 383 | if (((!GetProfL1State()) && IsProfSwitchOn(ProfilingLevel::L1))) { |
| 384 | - HCCL_INFO("Profiling L1 switch form off to on."); | 384 | + HCCL_INFO("Profiling L1 switch from off to on."); |
| 385 | return true; | 385 | return true; |
| 386 | } | 386 | } |
| 387 | return false; | 387 | return false; |
| @@ -455,8 +455,8 @@ void PrintPanicLogInfo(const uint8_t* panicLog) | |||
| 455 | HCCL_ERROR("get ccu dfx info fail, ccu dfx info not all correct"); | 455 | HCCL_ERROR("get ccu dfx info fail, ccu dfx info not all correct"); |
| 456 | } | 456 | } |
| 457 | HCCL_ERROR( | 457 | HCCL_ERROR( |
| 458 | - "CCU DFX INFO: SQE_RECV_CNT[%u] SQE_SEND_CNT[%u] MISSION_DFX[%u]" | 458 | + "CCU DFX INFO: SQE_RECV_CNT[%u] SQE_SEND_CNT[%u] MISSION_DFX[%u] " |
| 459 | - "TIF_SQE_CNT[%u] TIF_CQE_CNT[%u] CIF_SQE_CNT[%u] CIF_CQE_CNT[%u]" | 459 | + "TIF_SQE_CNT[%u] TIF_CQE_CNT[%u] CIF_SQE_CNT[%u] CIF_CQE_CNT[%u] " |
| 460 | "SQE_DROP_CNT[%u] SQE_ADDR_LEN_ERR_DROP_CNT[%u] ccumIsEnable[%u]", | 460 | "SQE_DROP_CNT[%u] SQE_ADDR_LEN_ERR_DROP_CNT[%u] ccumIsEnable[%u]", |
| 461 | info->ccum_sqe_recv_cnt, info->ccum_sqe_send_cnt, info->ccum_mission_dfx, info->ccum_tif_sqe_cnt, | 461 | info->ccum_sqe_recv_cnt, info->ccum_sqe_send_cnt, info->ccum_mission_dfx, info->ccum_tif_sqe_cnt, |
| 462 | info->ccum_tif_cqe_cnt, info->ccum_cif_sqe_cnt, info->ccum_cif_cqe_cnt, info->ccum_sqe_drop_cnt, | 462 | info->ccum_tif_cqe_cnt, info->ccum_cif_sqe_cnt, info->ccum_cif_cqe_cnt, info->ccum_sqe_drop_cnt, |
| @@ -806,7 +806,7 @@ void TaskExceptionHandler::PrintAicpuErrorMessage(rtExceptionInfo_t* exceptionIn | |||
| 806 | if (Hccl::g_commHadCallbackArrayV2[exceptionInfo->deviceid]) { | 806 | if (Hccl::g_commHadCallbackArrayV2[exceptionInfo->deviceid]) { |
| 807 | // 防止同一个device上出现通信主流和kernel流均出现task exception时runtime调用两次callback | 807 | // 防止同一个device上出现通信主流和kernel流均出现task exception时runtime调用两次callback |
| 808 | // HDC通道信息不是读清,防止aicpu task exception重复上报 | 808 | // HDC通道信息不是读清,防止aicpu task exception重复上报 |
| 809 | - HCCL_WARNING("aicpu error message been reported. deviceid[%u]", exceptionInfo->deviceid); | 809 | + HCCL_WARNING("aicpu error message has been reported. deviceid[%u]", exceptionInfo->deviceid); |
| 810 | return; | 810 | return; |
| 811 | } | 811 | } |
| 812 | lock.unlock(); | 812 | lock.unlock(); |
| @@ -863,7 +863,7 @@ void TaskExceptionHandler::PrintAicpuErrorMessage(rtExceptionInfo_t* exceptionIn | |||
| 863 | lock.lock(); | 863 | lock.lock(); |
| 864 | Hccl::g_commHadCallbackArrayV2[exceptionInfo->deviceid] = true; | 864 | Hccl::g_commHadCallbackArrayV2[exceptionInfo->deviceid] = true; |
| 865 | } else { | 865 | } else { |
| 866 | - HCCL_WARNING("PrintAicpuErrorMessage No Vaild errorMessage!"); | 866 | + HCCL_WARNING("PrintAicpuErrorMessage No Valid errorMessage!"); |
| 867 | } | 867 | } |
| 868 | } else { | 868 | } else { |
| 869 | HCCL_INFO("PrintAicpuErrorMessage streamId[%u] is not found.", exceptionInfo->streamid); | 869 | HCCL_INFO("PrintAicpuErrorMessage streamId[%u] is not found.", exceptionInfo->streamid); |
| @@ -895,7 +895,7 @@ void TaskExceptionHandler::PrintCcuErrorLog(const std::vector<CcuErrorInfo>& err | |||
| 895 | if (errorInfos.empty()) { | 895 | if (errorInfos.empty()) { |
| 896 | return; | 896 | return; |
| 897 | } | 897 | } |
| 898 | - HCCL_ERROR("[TaskExceptionHandler]Task run failed, ccu runtime information is: %s", __func__); | 898 | + HCCL_ERROR("[TaskExceptionHandler]Task run failed, ccu runtime information is:"); |
| 899 | for (const auto& errorInfo : errorInfos) { | 899 | for (const auto& errorInfo : errorInfos) { |
| 900 | HCCL_ERROR("[TaskExceptionHandler][%s]", GetCcuErrorMsgByType(errorInfo, taskInfo).c_str()); | 900 | HCCL_ERROR("[TaskExceptionHandler][%s]", GetCcuErrorMsgByType(errorInfo, taskInfo).c_str()); |
| 901 | } | 901 | } |
| @@ -393,7 +393,7 @@ UboeStatus RmaConnManager::GetUboeSocketStatus(Socket* socket, const LinkData& l | |||
| 393 | 393 | ||
| 394 | void RmaConnManager::WaitUboeSocketReady(Socket* socket, const LinkData& linkData) | 394 | void RmaConnManager::WaitUboeSocketReady(Socket* socket, const LinkData& linkData) |
| 395 | { | 395 | { |
| 396 | - HCCL_INFO("[RmaConnManager][%s] begain", __func__); | 396 | + HCCL_INFO("[RmaConnManager][%s] begin", __func__); |
| 397 | auto timeout = std::chrono::seconds(EnvConfig::GetInstance().GetSocketConfig().GetLinkTimeOut()); | 397 | auto timeout = std::chrono::seconds(EnvConfig::GetInstance().GetSocketConfig().GetLinkTimeOut()); |
| 398 | HcclUs startTime = std::chrono::steady_clock::now(); | 398 | HcclUs startTime = std::chrono::steady_clock::now(); |
| 399 | while (true) { | 399 | while (true) { |
| @@ -32,7 +32,7 @@ SocketHandle HostSocketHandleManager::Create(DevId devicePhyId, const IpAddress& | |||
| 32 | 32 | ||
| 33 | if (isDestroy) { | 33 | if (isDestroy) { |
| 34 | HCCL_WARNING( | 34 | HCCL_WARNING( |
| 35 | - "[HostSocketHandleManager::%s] devicePhyId[%u] HostSocketHandleManager has been destroy", __func__, | 35 | + "[HostSocketHandleManager::%s] devicePhyId[%u] HostSocketHandleManager has been destroyed", __func__, |
| 36 | devicePhyId); | 36 | devicePhyId); |
| 37 | return nullptr; | 37 | return nullptr; |
| 38 | } | 38 | } |
| @@ -76,19 +76,19 @@ SocketHandle HostSocketHandleManager::Get(DevId devicePhyId, const IpAddress& ho | |||
| 76 | 76 | ||
| 77 | if (isDestroy) { | 77 | if (isDestroy) { |
| 78 | HCCL_WARNING( | 78 | HCCL_WARNING( |
| 79 | - "[HostSocketHandleManager::%s] devicePhyId[%u] HostSocketHandleManager has been detroy", __func__, | 79 | + "[HostSocketHandleManager::%s] devicePhyId[%u] HostSocketHandleManager has been destroyed", __func__, |
| 80 | devicePhyId); | 80 | devicePhyId); |
| 81 | return nullptr; | 81 | return nullptr; |
| 82 | } | 82 | } |
| 83 | 83 | ||
| 84 | if (devicePhyId > hostSocketHandleMap.size() - 1) { | 84 | if (devicePhyId > hostSocketHandleMap.size() - 1) { |
| 85 | - HCCL_WARNING("HostSocketHandleManager for devicePhyId=%u dose not exist", devicePhyId); | 85 | + HCCL_WARNING("HostSocketHandleManager for devicePhyId=%u does not exist", devicePhyId); |
| 86 | return nullptr; | 86 | return nullptr; |
| 87 | } | 87 | } |
| 88 | 88 | ||
| 89 | auto tempiter = hostSocketHandleMap[devicePhyId].find(hostIp.GetIpStr()); | 89 | auto tempiter = hostSocketHandleMap[devicePhyId].find(hostIp.GetIpStr()); |
| 90 | if (tempiter == hostSocketHandleMap[devicePhyId].end()) { | 90 | if (tempiter == hostSocketHandleMap[devicePhyId].end()) { |
| 91 | - HCCL_WARNING("HostSocketHandleManager for IpAddress=%s dose not exist", hostIp.GetIpStr().c_str()); | 91 | + HCCL_WARNING("HostSocketHandleManager for IpAddress=%s does not exist", hostIp.GetIpStr().c_str()); |
| 92 | return nullptr; | 92 | return nullptr; |
| 93 | } | 93 | } |
| 94 | 94 | ||
| @@ -121,7 +121,7 @@ void HostSocketHandleManager::Destroy(DevId devicePhyId, const IpAddress& hostIp | |||
| 121 | 121 | ||
| 122 | if (isDestroy) { | 122 | if (isDestroy) { |
| 123 | HCCL_WARNING( | 123 | HCCL_WARNING( |
| 124 | - "[HostSocketHandleManager::%s] devicePhyId[%u] HostSocketHandleManager has been detroy", __func__, | 124 | + "[HostSocketHandleManager::%s] devicePhyId[%u] HostSocketHandleManager has been destroyed", __func__, |
| 125 | devicePhyId); | 125 | devicePhyId); |
| 126 | return; | 126 | return; |
| 127 | } | 127 | } |
| @@ -136,7 +136,7 @@ void HostSocketHandleManager::Destroy(DevId devicePhyId, const IpAddress& hostIp | |||
| 136 | CHK_PRT_THROW( | 136 | CHK_PRT_THROW( |
| 137 | hostSocketHandleMap[devicePhyId].count(hostIp.GetIpStr()) == 0, | 137 | hostSocketHandleMap[devicePhyId].count(hostIp.GetIpStr()) == 0, |
| 138 | HCCL_ERROR( | 138 | HCCL_ERROR( |
| 139 | - "[HostSocketHandleManager::%s] devicePhyId[%u] hostIp[%s] dose not exist", __func__, devicePhyId, | 139 | + "[HostSocketHandleManager::%s] devicePhyId[%u] hostIp[%s] does not exist", __func__, devicePhyId, |
| 140 | hostIp.GetIpStr().c_str()), | 140 | hostIp.GetIpStr().c_str()), |
| 141 | InvalidParamsException, "hostIp not exist"); | 141 | InvalidParamsException, "hostIp not exist"); |
| 142 | 142 | ||
| @@ -94,7 +94,7 @@ public: | |||
| 94 | if (portGroupSize != tgtPortGroupSize) { | 94 | if (portGroupSize != tgtPortGroupSize) { |
| 95 | HCCL_WARNING( | 95 | HCCL_WARNING( |
| 96 | "[LinkData][Constructor]srcConnIface.portGroupSize[%u] " | 96 | "[LinkData][Constructor]srcConnIface.portGroupSize[%u] " |
| 97 | - "is not euqal to targetConnIface.portGroupSize[%u]", | 97 | + "is not equal to targetConnIface.portGroupSize[%u]", |
| 98 | static_cast<u32>(portGroupSize), static_cast<u32>(tgtPortGroupSize)); | 98 | static_cast<u32>(portGroupSize), static_cast<u32>(tgtPortGroupSize)); |
| 99 | } | 99 | } |
| 100 | fullmesh = false; // 多链路场景,非fullmesh | 100 | fullmesh = false; // 多链路场景,非fullmesh |
| @@ -398,7 +398,7 @@ void CcuContextAllReduceMesh2DTwoShot::DoGroupBroadcast( | |||
| 398 | 398 | ||
| 399 | std::vector<uint64_t> CcuContextAllReduceMesh2DTwoShot::GeneArgs(const CcuTaskArg& arg) | 399 | std::vector<uint64_t> CcuContextAllReduceMesh2DTwoShot::GeneArgs(const CcuTaskArg& arg) |
| 400 | { | 400 | { |
| 401 | - HCCL_INFO("[CcuContextReduceScatterMesh2D] GeneArgs Starts"); | 401 | + HCCL_INFO("[CcuContextAllReduceMesh2DTwoShot] GeneArgs Starts"); |
| 402 | const CcuTaskArgAllReduceMesh2DTwoShot* taskArg = dynamic_cast<const CcuTaskArgAllReduceMesh2DTwoShot*>(&arg); | 402 | const CcuTaskArgAllReduceMesh2DTwoShot* taskArg = dynamic_cast<const CcuTaskArgAllReduceMesh2DTwoShot*>(&arg); |
| 403 | if (taskArg == nullptr) { | 403 | if (taskArg == nullptr) { |
| 404 | THROW<NullPtrException>(StringFormat("CcuContextAllReduceMesh2DTwoShot::taskArg ptr is null")); | 404 | THROW<NullPtrException>(StringFormat("CcuContextAllReduceMesh2DTwoShot::taskArg ptr is null")); |
| @@ -72,7 +72,7 @@ CcuContextAlltoAllMesh2D::CcuContextAlltoAllMesh2D( | |||
| 72 | anotherId = dimId[1 - axisId]; // 本rank在另一个轴上的Id | 72 | anotherId = dimId[1 - axisId]; // 本rank在另一个轴上的Id |
| 73 | anotherSize = dimSize[1 - axisId]; | 73 | anotherSize = dimSize[1 - axisId]; |
| 74 | HCCL_INFO( | 74 | HCCL_INFO( |
| 75 | - "[CcuContextAlltoAllMesh2D] RankId[%u], DimSize: D0[%u]--D1[%u], localId[%u], lcoalSize[%u]", rankId_, | 75 | + "[CcuContextAlltoAllMesh2D] RankId[%u], DimSize: D0[%u]--D1[%u], localId[%u], localSize[%u]", rankId_, |
| 76 | dimSize[0], dimSize[1], localId, localSize); | 76 | dimSize[0], dimSize[1], localId, localSize); |
| 77 | 77 | ||
| 78 | AllocGoResource(LOC_CPY_LOOP_NUM); // 只用8个loop做本地搬运,每个loop搬4K | 78 | AllocGoResource(LOC_CPY_LOOP_NUM); // 只用8个loop做本地搬运,每个loop搬4K |
| @@ -192,7 +192,7 @@ void CcuContextAlltoAllMesh2D::AxisSync(uint32_t signalIndex) | |||
| 192 | const uint32_t DIE_NUM = 2; // 2个die | 192 | const uint32_t DIE_NUM = 2; // 2个die |
| 193 | if (signalIndex > 1) { | 193 | if (signalIndex > 1) { |
| 194 | THROW<InvalidParamsException>( | 194 | THROW<InvalidParamsException>( |
| 195 | - StringFormat("[CcuContextAlltoAllMesh2D] Unexpected SignalInex[%u]", signalIndex)); | 195 | + StringFormat("[CcuContextAlltoAllMesh2D] Unexpected SignalIndex[%u]", signalIndex)); |
| 196 | } | 196 | } |
| 197 | LocalCtxPost(anotherAxisSignal, 1 << (axisId + signalIndex * DIE_NUM)); | 197 | LocalCtxPost(anotherAxisSignal, 1 << (axisId + signalIndex * DIE_NUM)); |
| 198 | LocalWait(localAxisSignal, 1 << (1 - axisId + signalIndex * DIE_NUM)); | 198 | LocalWait(localAxisSignal, 1 << (1 - axisId + signalIndex * DIE_NUM)); |
| @@ -78,7 +78,7 @@ void CcuContextBroadcastMesh1D::LoadAndExchangeData() | |||
| 78 | { | 78 | { |
| 79 | uint16_t selfBit = 1 << rankId_; | 79 | uint16_t selfBit = 1 << rankId_; |
| 80 | uint16_t allBit = ((1 << rankSize_) - 1) & (~(1 << rankId_)); | 80 | uint16_t allBit = ((1 << rankSize_) - 1) & (~(1 << rankId_)); |
| 81 | - HCCL_DEBUG("[CcuContextBroadcastMesh1D] BroadcastMesh1D LoadAndExchanageData: rankId[%u]", rankId_); | 81 | + HCCL_DEBUG("[CcuContextBroadcastMesh1D] BroadcastMesh1D LoadAndExchangeData: rankId[%u]", rankId_); |
| 82 | Load(input_); | 82 | Load(input_); |
| 83 | Load(output_[rankId_]); | 83 | Load(output_[rankId_]); |
| 84 | Load(token_[rankId_]); | 84 | Load(token_[rankId_]); |
| @@ -44,7 +44,7 @@ CcuContextReduceMesh2D::CcuContextReduceMesh2D( | |||
| 44 | localSize_ = dimSize_[axisId_]; // 本rank所在的行/列的总数 | 44 | localSize_ = dimSize_[axisId_]; // 本rank所在的行/列的总数 |
| 45 | 45 | ||
| 46 | HCCL_INFO( | 46 | HCCL_INFO( |
| 47 | - "[CcuContextReduceMesh2D] RankId[%u], DimSize0[%llu], DimSize1[%llu], localId[%llu], lcoalSize[%llu]", rankId_, | 47 | + "[CcuContextReduceMesh2D] RankId[%u], DimSize0[%llu], DimSize1[%llu], localId[%llu], localSize[%llu]", rankId_, |
| 48 | dimSize_[0], dimSize_[1], localId_, localSize_); | 48 | dimSize_[0], dimSize_[1], localId_, localSize_); |
| 49 | 49 | ||
| 50 | dataType_ = ctxArg->op_.dataType; | 50 | dataType_ = ctxArg->op_.dataType; |
| @@ -246,7 +246,7 @@ std::vector<uint64_t> CcuContextReduceMesh2D::GeneArgs(const CcuTaskArg& arg) | |||
| 246 | auto yAxisGoSize = CalGoSize(yAxisSize); | 246 | auto yAxisGoSize = CalGoSize(yAxisSize); |
| 247 | 247 | ||
| 248 | HCCL_INFO( | 248 | HCCL_INFO( |
| 249 | - "[CcuContextReduceMesh2D] ReduceMesh2D inputAddr [%llu] outputAddr [%llu] offset [%llu]" | 249 | + "[CcuContextReduceMesh2D] ReduceMesh2D inputAddr [%llu] outputAddr [%llu] offset [%llu], " |
| 250 | "xAxisSize [%llu] yAxisSize [%llu]", | 250 | "xAxisSize [%llu] yAxisSize [%llu]", |
| 251 | inputAddr, outputAddr, offset, xAxisSize, yAxisSize); | 251 | inputAddr, outputAddr, offset, xAxisSize, yAxisSize); |
| 252 | 252 | ||
| @@ -147,7 +147,7 @@ void CcuContextReduceNHR1DMem2mem::AxisSync(uint32_t signalIndex) | |||
| 147 | const uint32_t DIE_NUM = 2; | 147 | const uint32_t DIE_NUM = 2; |
| 148 | if (signalIndex > 1) { | 148 | if (signalIndex > 1) { |
| 149 | THROW<InvalidParamsException>( | 149 | THROW<InvalidParamsException>( |
| 150 | - StringFormat("[CcuContextReduceNHR1DMem2mem] Unexpected SignalInex[%u]", signalIndex)); | 150 | + StringFormat("[CcuContextReduceNHR1DMem2mem] Unexpected SignalIndex[%u]", signalIndex)); |
| 151 | } | 151 | } |
| 152 | LocalCtxPost(anotherAxisSignal_, 1 << (axisId_ + signalIndex * DIE_NUM)); | 152 | LocalCtxPost(anotherAxisSignal_, 1 << (axisId_ + signalIndex * DIE_NUM)); |
| 153 | LocalWait(localAxisSignal_, 1 << (1 - axisId_ + signalIndex * DIE_NUM)); | 153 | LocalWait(localAxisSignal_, 1 << (1 - axisId_ + signalIndex * DIE_NUM)); |
| @@ -273,7 +273,7 @@ HcclResult InsAllGatherParallelExecutor<AlgTopoMatch, InsAlgTemplate0, InsAlgTem | |||
| 273 | 273 | ||
| 274 | // 申请算法模板所需资源 | 274 | // 申请算法模板所需资源 |
| 275 | if (!(resReqIntra.queNum > 0 && resReqInter.queNum > 0)) { | 275 | if (!(resReqIntra.queNum > 0 && resReqInter.queNum > 0)) { |
| 276 | - HCCL_ERROR("Intra queNum and Inter queNum must larger than 0."); | 276 | + HCCL_ERROR("Intra queNum and Inter queNum must be larger than 0."); |
| 277 | return HcclResult::HCCL_E_INTERNAL; | 277 | return HcclResult::HCCL_E_INTERNAL; |
| 278 | } | 278 | } |
| 279 | u32 totalQueueNum = resReqIntra.queNum + resReqInter.queNum; | 279 | u32 totalQueueNum = resReqIntra.queNum + resReqInter.queNum; |
| @@ -313,7 +313,7 @@ HcclResult InsAllGatherParallelExecutor<AlgTopoMatch, InsAlgTemplate0, InsAlgTem | |||
| 313 | 313 | ||
| 314 | // 申请算法模板所需资源 | 314 | // 申请算法模板所需资源 |
| 315 | if (!(resReqIntra.queNum > 0 && resReqInter.queNum > 0)) { | 315 | if (!(resReqIntra.queNum > 0 && resReqInter.queNum > 0)) { |
| 316 | - HCCL_ERROR("[InsAllGatherParallelExecutor]resReqIntra.queNum and resReqInter.queNum must larger than 0."); | 316 | + HCCL_ERROR("[InsAllGatherParallelExecutor]resReqIntra.queNum and resReqInter.queNum must be larger than 0."); |
| 317 | return HcclResult::HCCL_E_INTERNAL; | 317 | return HcclResult::HCCL_E_INTERNAL; |
| 318 | } | 318 | } |
| 319 | u32 totalQueueNum = resReqIntra.queNum + resReqInter.queNum; | 319 | u32 totalQueueNum = resReqIntra.queNum + resReqInter.queNum; |
| @@ -279,7 +279,7 @@ HcclResult InsAlltoAllSoleExecutor<AlgTopoMatch, InsAlgTemplate>::OrchestrateOpb | |||
| 279 | tempFuncs.isForepart = true; // Usr Buff to CCL Buff required | 279 | tempFuncs.isForepart = true; // Usr Buff to CCL Buff required |
| 280 | tempFuncs.isBottom = true; // CCL Buff to Usr Buff required | 280 | tempFuncs.isBottom = true; // CCL Buff to Usr Buff required |
| 281 | CHK_RET(tempAlg.Run(tempFuncs, sliceInfoVec, buffInfo, tempResLinks_, requiredQue_)); | 281 | CHK_RET(tempAlg.Run(tempFuncs, sliceInfoVec, buffInfo, tempResLinks_, requiredQue_)); |
| 282 | - HCCL_INFO("[InsAlltoAllSoleExecutor][OrchestrateOpbase] Run templet success."); | 282 | + HCCL_INFO("[InsAlltoAllSoleExecutor][OrchestrateOpbase] Run template success."); |
| 283 | 283 | ||
| 284 | return HcclResult::HCCL_SUCCESS; | 284 | return HcclResult::HCCL_SUCCESS; |
| 285 | } | 285 | } |
| @@ -228,7 +228,7 @@ HcclResult InsScatterParallelExecutor<AlgTopoMatch, InsAlgTemplate0, InsAlgTempl | |||
| 228 | 228 | ||
| 229 | // 申请算法模板所需资源 | 229 | // 申请算法模板所需资源 |
| 230 | if (!(resReqIntra.queNum > 0 && resReqInter.queNum > 0)) { | 230 | if (!(resReqIntra.queNum > 0 && resReqInter.queNum > 0)) { |
| 231 | - HCCL_ERROR("[InsScatterParallelExecutor]resReqIntra.queNum and resReqInter.queNum must larger than 0."); | 231 | + HCCL_ERROR("[InsScatterParallelExecutor]resReqIntra.queNum and resReqInter.queNum must be larger than 0."); |
| 232 | return HcclResult::HCCL_E_INTERNAL; | 232 | return HcclResult::HCCL_E_INTERNAL; |
| 233 | } | 233 | } |
| 234 | u32 totalQueueNum = resReqIntra.queNum + resReqInter.queNum; | 234 | u32 totalQueueNum = resReqIntra.queNum + resReqInter.queNum; |
| @@ -268,7 +268,7 @@ HcclResult InsScatterParallelExecutor<AlgTopoMatch, InsAlgTemplate0, InsAlgTempl | |||
| 268 | 268 | ||
| 269 | // 申请算法模板所需资源 | 269 | // 申请算法模板所需资源 |
| 270 | if (!(resReqIntra.queNum > 0 && resReqInter.queNum > 0)) { | 270 | if (!(resReqIntra.queNum > 0 && resReqInter.queNum > 0)) { |
| 271 | - HCCL_ERROR("[InsScatterParallelExecutor]resReqIntra.queNum and resReqInter.queNum must larger than 0."); | 271 | + HCCL_ERROR("[InsScatterParallelExecutor]resReqIntra.queNum and resReqInter.queNum must be larger than 0."); |
| 272 | return HcclResult::HCCL_E_INTERNAL; | 272 | return HcclResult::HCCL_E_INTERNAL; |
| 273 | } | 273 | } |
| 274 | u32 totalQueueNum = resReqIntra.queNum + resReqInter.queNum; | 274 | u32 totalQueueNum = resReqIntra.queNum + resReqInter.queNum; |
| @@ -158,7 +158,7 @@ HcclResult CcuTempAllReduceMesh2DTwoShot::PrepareLinks(const ResLinks& tempLinks | |||
| 158 | } | 158 | } |
| 159 | } | 159 | } |
| 160 | HCCL_INFO( | 160 | HCCL_INFO( |
| 161 | - "[CcuTempAllReduceMesh2DTwoShot] PrepareLinks Eends. linksX Size[%u], linksY Size[%u]", linksX_.size(), | 161 | + "[CcuTempAllReduceMesh2DTwoShot] PrepareLinks Ends. linksX Size[%u], linksY Size[%u]", linksX_.size(), |
| 162 | linksY_.size()); | 162 | linksY_.size()); |
| 163 | return HcclResult::HCCL_SUCCESS; | 163 | return HcclResult::HCCL_SUCCESS; |
| 164 | } | 164 | } |
| @@ -77,7 +77,7 @@ HcclResult CcuTempAlltoAllVMesh1D::SetConcurrentSendRecvNum(const u32 concurrent | |||
| 77 | { | 77 | { |
| 78 | CHK_PRT_RET( | 78 | CHK_PRT_RET( |
| 79 | concurrentSendRecvNum == 0, | 79 | concurrentSendRecvNum == 0, |
| 80 | - HCCL_ERROR("[CcuTempAlltoAllVMesh1D][SetConcurrentSendRecvNum] concurrentSendRecvNum" | 80 | + HCCL_ERROR("[CcuTempAlltoAllVMesh1D][SetConcurrentSendRecvNum] concurrentSendRecvNum " |
| 81 | "should not be zero"), | 81 | "should not be zero"), |
| 82 | HcclResult::HCCL_E_PARA); | 82 | HcclResult::HCCL_E_PARA); |
| 83 | concurrentSendRecvNum_ = concurrentSendRecvNum; | 83 | concurrentSendRecvNum_ = concurrentSendRecvNum; |
| @@ -100,7 +100,7 @@ uint64_t CcuTempAlltoAllVMesh1D::CalcSendRecvNumSubStep() | |||
| 100 | = ((localSendRecvInfo_.recvLength[destRank] + UB_MAX_DATA_SIZE - 1) / UB_MAX_DATA_SIZE); | 100 | = ((localSendRecvInfo_.recvLength[destRank] + UB_MAX_DATA_SIZE - 1) / UB_MAX_DATA_SIZE); |
| 101 | recvNumSubStep_[destRank] = currRankRecvSubStep; | 101 | recvNumSubStep_[destRank] = currRankRecvSubStep; |
| 102 | HCCL_INFO( | 102 | HCCL_INFO( |
| 103 | - "[CcuTempAlltoAllVMesh1D][CalcNumSubStep] myRank [%d] currRankSendSubStep[%llu]" | 103 | + "[CcuTempAlltoAllVMesh1D][CalcNumSubStep] myRank [%d] currRankSendSubStep[%llu] " |
| 104 | "currRankRecvSubStep[%llu]", | 104 | "currRankRecvSubStep[%llu]", |
| 105 | myRank_, currRankSendSubStep, currRankRecvSubStep); | 105 | myRank_, currRankSendSubStep, currRankRecvSubStep); |
| 106 | numSubStep = std::max(numSubStep, std::max(currRankSendSubStep, currRankRecvSubStep)); | 106 | numSubStep = std::max(numSubStep, std::max(currRankSendSubStep, currRankRecvSubStep)); |
| @@ -123,7 +123,7 @@ HcclResult CcuTempAlltoAllVMesh1D::Run( | |||
| 123 | CHK_PRT_RET( | 123 | CHK_PRT_RET( |
| 124 | localSendRecvInfo_.sendLength[myRank_] != localSendRecvInfo_.recvLength[myRank_], | 124 | localSendRecvInfo_.sendLength[myRank_] != localSendRecvInfo_.recvLength[myRank_], |
| 125 | HCCL_ERROR( | 125 | HCCL_ERROR( |
| 126 | - "[CcuTempAlltoAllVMesh1D] rankSize = 1, sendLength[%llu] and recvLength[%llu]" | 126 | + "[CcuTempAlltoAllVMesh1D] rankSize = 1, sendLength[%llu] and recvLength[%llu] " |
| 127 | "should be equal.", | 127 | "should be equal.", |
| 128 | localSendRecvInfo_.sendLength[myRank_], localSendRecvInfo_.recvLength[myRank_]), | 128 | localSendRecvInfo_.sendLength[myRank_], localSendRecvInfo_.recvLength[myRank_]), |
| 129 | HcclResult::HCCL_E_PARA); | 129 | HcclResult::HCCL_E_PARA); |
| @@ -100,7 +100,7 @@ std::vector<std::tuple<QId, QId, u32>> InsTempReduceMesh2D::CreateNotifiesReques | |||
| 100 | 100 | ||
| 101 | HCCL_DEBUG( | 101 | HCCL_DEBUG( |
| 102 | "[InsTempReduceMesh2D] Create notifies request: " | 102 | "[InsTempReduceMesh2D] Create notifies request: " |
| 103 | - "totalNotifyReqNum[%u], ctrlNotfiyReqNum[%u], xNotifyReqNum[%u], yNotifyReqNum[%u]", | 103 | + "totalNotifyReqNum[%u], ctrlNotifyReqNum[%u], xNotifyReqNum[%u], yNotifyReqNum[%u]", |
| 104 | totalNotifyReqNum, ctrlNotfiyReqNum, xNotifyReqNum, yNotifyReqNum); | 104 | totalNotifyReqNum, ctrlNotfiyReqNum, xNotifyReqNum, yNotifyReqNum); |
| 105 | 105 | ||
| 106 | return notifyRequests; | 106 | return notifyRequests; |
| @@ -102,7 +102,7 @@ HcclResult InsTempReduceScatterNHR::PreCopy( | |||
| 102 | CHK_RET(MultiSliceLocalCopy(tempInsQues[0], tempFuncs.usrData.usrInSlices, tempFuncs.usrData.scratchInSlices)); | 102 | CHK_RET(MultiSliceLocalCopy(tempInsQues[0], tempFuncs.usrData.usrInSlices, tempFuncs.usrData.scratchInSlices)); |
| 103 | } else { | 103 | } else { |
| 104 | // 图模式或者单算子模式下非第一个算子,数据已经在 inbuff 上了,不需要拷贝 | 104 | // 图模式或者单算子模式下非第一个算子,数据已经在 inbuff 上了,不需要拷贝 |
| 105 | - HCCL_INFO("[InsTempReduceScatterNHR][PreCopy] not forpat and opbse, skip precopy"); | 105 | + HCCL_INFO("[InsTempReduceScatterNHR][PreCopy] not forepart and opbase, skip precopy"); |
| 106 | } | 106 | } |
| 107 | return HcclResult::HCCL_SUCCESS; | 107 | return HcclResult::HCCL_SUCCESS; |
| 108 | } | 108 | } |
| @@ -136,10 +136,10 @@ HcclResult InsTempReduceScatterNHR::PostCopy( | |||
| 136 | u64 srcOffset = sliceInfoVec[tempVirtRankMap_[myRank_]][0].offset + buffInfo_.inBuffBaseOff; | 136 | u64 srcOffset = sliceInfoVec[tempVirtRankMap_[myRank_]][0].offset + buffInfo_.inBuffBaseOff; |
| 137 | u64 dstOffset = buffInfo_.outBuffBaseOff; | 137 | u64 dstOffset = buffInfo_.outBuffBaseOff; |
| 138 | if (buffInfo_.inBuffType == buffInfo_.outBuffType && srcOffset == dstOffset) { | 138 | if (buffInfo_.inBuffType == buffInfo_.outBuffType && srcOffset == dstOffset) { |
| 139 | - HCCL_INFO( | 139 | + HCCL_INFO("[InsTempReduceScatterNHR][PostCopy] not forepart and opbase, inBuffType same as outBuffType, " |
| 140 | - "[InsTempReduceScatterNHR][PostCopy] not forpat and opbse, inBuffType same as outBuffType, skip copy"); | 140 | + "skip copy"); |
| 141 | } else { | 141 | } else { |
| 142 | - HCCL_INFO("[InsTempReduceScatterNHR][PostCopy] not forpat and opbse, copy from outBuff to userOut"); | 142 | + HCCL_INFO("[InsTempReduceScatterNHR][PostCopy] not forepart and opbase, copy from outBuff to userOut"); |
| 143 | DataSlice srcSlice = DataSlice(buffInfo_.inBuffType, srcOffset, size); | 143 | DataSlice srcSlice = DataSlice(buffInfo_.inBuffType, srcOffset, size); |
| 144 | DataSlice dstSlice = DataSlice(buffInfo_.outBuffType, dstOffset, size); | 144 | DataSlice dstSlice = DataSlice(buffInfo_.outBuffType, dstOffset, size); |
| 145 | CHK_RET(LocalCopy(tempInsQues[0], srcSlice, dstSlice)); | 145 | CHK_RET(LocalCopy(tempInsQues[0], srcSlice, dstSlice)); |
| @@ -66,7 +66,7 @@ UbMemoryTransport::UBTransportStatus UbMemoryTransport::GetStatus() | |||
| 66 | HCCL_ERROR("%s", e.what()); | 66 | HCCL_ERROR("%s", e.what()); |
| 67 | return UbMemoryTransport::UBTransportStatus::CONNECT_FAILED; | 67 | return UbMemoryTransport::UBTransportStatus::CONNECT_FAILED; |
| 68 | } catch (...) { | 68 | } catch (...) { |
| 69 | - HCCL_ERROR("Unknown error occured when StateMachine!"); | 69 | + HCCL_ERROR("Unknown error occurred when StateMachine!"); |
| 70 | return UbMemoryTransport::UBTransportStatus::CONNECT_FAILED; | 70 | return UbMemoryTransport::UBTransportStatus::CONNECT_FAILED; |
| 71 | } | 71 | } |
| 72 | return status; | 72 | return status; |
| @@ -218,7 +218,7 @@ HcclResult CcuResSpecifications::GetXnBaseAddr(const uint8_t dieId, uint64_t& xn | |||
| 218 | if (ccuResAddr > UINT64_MAX - ccuXnOffset) { | 218 | if (ccuResAddr > UINT64_MAX - ccuXnOffset) { |
| 219 | HCCL_ERROR( | 219 | HCCL_ERROR( |
| 220 | "[CcuResSpecifications][%s] failed, CCU resource base address[%llu] is " | 220 | "[CcuResSpecifications][%s] failed, CCU resource base address[%llu] is " |
| 221 | - "greater then expected, ccu xn offset[%llu], their sum will exceeds the range " | 221 | + "greater than expected, ccu xn offset[%llu], their sum will exceed the range " |
| 222 | "of uint64_t.", | 222 | "of uint64_t.", |
| 223 | __func__, ccuResAddr, ccuXnOffset); | 223 | __func__, ccuResAddr, ccuXnOffset); |
| 224 | } | 224 | } |
| @@ -1025,7 +1025,7 @@ void HrtRaSocketGetVnicIpInfos(u32 phyId, DeviceIdType deviceIdType, u32 deviceI | |||
| 1025 | 1025 | ||
| 1026 | static u32 HrtGetIfNum(struct RaGetIfattr& config) | 1026 | static u32 HrtGetIfNum(struct RaGetIfattr& config) |
| 1027 | { | 1027 | { |
| 1028 | - HCCL_INFO("[HrtGetIfNum] Input params: phyId=%u, nicPosistion=%u", config.phyId, config.nicPosition); | 1028 | + HCCL_INFO("[HrtGetIfNum] Input params: phyId=%u, nicPosition=%u", config.phyId, config.nicPosition); |
| 1029 | 1029 | ||
| 1030 | u32 num = 0; | 1030 | u32 num = 0; |
| 1031 | s32 ret = RaGetIfnum(&config, &num); | 1031 | s32 ret = RaGetIfnum(&config, &num); |
| @@ -3147,7 +3147,7 @@ HcclResult HrtGetCcuMemInfo( | |||
| 3147 | 3147 | ||
| 3148 | HcclResult HrtRaGetEidByIp(RdmaHandle handle, const vector<IpAddress>& ipV4AddrList, vector<IpAddress>& eidAddrList) | 3148 | HcclResult HrtRaGetEidByIp(RdmaHandle handle, const vector<IpAddress>& ipV4AddrList, vector<IpAddress>& eidAddrList) |
| 3149 | { | 3149 | { |
| 3150 | - HCCL_INFO("[HrtRaGetEidByIp] begain, ipV4AddrList size=%zu", ipV4AddrList.size()); | 3150 | + HCCL_INFO("[HrtRaGetEidByIp] begin, ipV4AddrList size=%zu", ipV4AddrList.size()); |
| 3151 | size_t ipV4AddrListSize = ipV4AddrList.size(); | 3151 | size_t ipV4AddrListSize = ipV4AddrList.size(); |
| 3152 | unsigned int num = ipV4AddrListSize; | 3152 | unsigned int num = ipV4AddrListSize; |
| 3153 | IpInfo ipInfoList[num] = {}; | 3153 | IpInfo ipInfoList[num] = {}; |
| @@ -3244,7 +3244,7 @@ bool HrtRaSupportsGetTpAttr(u32 phyId) | |||
| 3244 | HcclResult | 3244 | HcclResult |
| 3245 | HrtRaSetTpAttrAsync(RdmaHandle handle, uint64_t tpHandle, uint32_t attrBitmap, TpAttr& attr, RequestHandle& reqHandle) | 3245 | HrtRaSetTpAttrAsync(RdmaHandle handle, uint64_t tpHandle, uint32_t attrBitmap, TpAttr& attr, RequestHandle& reqHandle) |
| 3246 | { | 3246 | { |
| 3247 | - HCCL_INFO("[HrtRaSetTpAttrAsync] begain, reqHandle[%llu]", reqHandle); | 3247 | + HCCL_INFO("[HrtRaSetTpAttrAsync] begin, reqHandle[%llu]", reqHandle); |
| 3248 | void* raReqHandle = nullptr; | 3248 | void* raReqHandle = nullptr; |
| 3249 | s32 ret = RaSetTpAttrAsync(handle, tpHandle, attrBitmap, &attr, &raReqHandle); | 3249 | s32 ret = RaSetTpAttrAsync(handle, tpHandle, attrBitmap, &attr, &raReqHandle); |
| 3250 | if (ret != 0) { | 3250 | if (ret != 0) { |
| @@ -3260,7 +3260,7 @@ HrtRaSetTpAttrAsync(RdmaHandle handle, uint64_t tpHandle, uint32_t attrBitmap, T | |||
| 3260 | HcclResult HrtRaGetTpAttrAsync( | 3260 | HcclResult HrtRaGetTpAttrAsync( |
| 3261 | u32 phyId, RdmaHandle handle, uint64_t tpHandle, uint32_t& attrBitmap, TpAttr& attr, RequestHandle& reqHandle) | 3261 | u32 phyId, RdmaHandle handle, uint64_t tpHandle, uint32_t& attrBitmap, TpAttr& attr, RequestHandle& reqHandle) |
| 3262 | { | 3262 | { |
| 3263 | - HCCL_INFO("[HrtRaGetTpAttrAsync] begain, reqHandle[%llu]", reqHandle); | 3263 | + HCCL_INFO("[HrtRaGetTpAttrAsync] begin, reqHandle[%llu]", reqHandle); |
| 3264 | if (!HrtRaSupportsGetTpAttr(phyId)) { | 3264 | if (!HrtRaSupportsGetTpAttr(phyId)) { |
| 3265 | HCCL_ERROR( | 3265 | HCCL_ERROR( |
| 3266 | "this package does not support RaGetTpAttrAsync for device, please change new package, phyId=%u.", phyId); | 3266 | "this package does not support RaGetTpAttrAsync for device, please change new package, phyId=%u.", phyId); |
| @@ -459,7 +459,7 @@ void* HrtMalloc(u64 size, aclrtMemType_t memType) | |||
| 459 | aclrtMallocAttribute attrs{.attr = ACL_RT_MEM_ATTR_MODULE_ID, .value = moduleIdValue}; | 459 | aclrtMallocAttribute attrs{.attr = ACL_RT_MEM_ATTR_MODULE_ID, .value = moduleIdValue}; |
| 460 | aclrtMallocConfig cfg{.attrs = &attrs, .numAttrs = 1}; | 460 | aclrtMallocConfig cfg{.attrs = &attrs, .numAttrs = 1}; |
| 461 | ret = aclrtMallocWithCfg(&devPtr, size, static_cast<aclrtMemMallocPolicy>(memType), &cfg); | 461 | ret = aclrtMallocWithCfg(&devPtr, size, static_cast<aclrtMemMallocPolicy>(memType), &cfg); |
| 462 | - HCCL_INFO("[HrtMalloc] ret[%d] size[%llu], memType[%d], devPtr[%p], moudleId: HCCL.", ret, size, memType, devPtr); | 462 | + HCCL_INFO("[HrtMalloc] ret[%d] size[%llu], memType[%d], devPtr[%p], moduleId: HCCL.", ret, size, memType, devPtr); |
| 463 | if (ret == ACL_ERROR_RT_MEMORY_ALLOCATION) { | 463 | if (ret == ACL_ERROR_RT_MEMORY_ALLOCATION) { |
| 464 | RPT_INPUT_ERR( | 464 | RPT_INPUT_ERR( |
| 465 | true, "EI0007", std::vector<std::string>({"resource_type", "resource_info"}), | 465 | true, "EI0007", std::vector<std::string>({"resource_type", "resource_info"}), |
| @@ -693,7 +693,7 @@ void* HrtMallocHost(u64 size) | |||
| 693 | aclrtMallocConfig cfg{.attrs = &attrs, .numAttrs = 1}; | 693 | aclrtMallocConfig cfg{.attrs = &attrs, .numAttrs = 1}; |
| 694 | aclError ret = aclrtMallocHostWithCfg(&hostPtr, size, &cfg); | 694 | aclError ret = aclrtMallocHostWithCfg(&hostPtr, size, &cfg); |
| 695 | HCCL_INFO( | 695 | HCCL_INFO( |
| 696 | - "Call aclrtMallocHostWithCfg. return value[%d], para: hostPtr[%p], size[%llu], moudleId: HCCL.", ret, hostPtr, | 696 | + "Call aclrtMallocHostWithCfg. return value[%d], para: hostPtr[%p], size[%llu], moduleId: HCCL.", ret, hostPtr, |
| 697 | size); | 697 | size); |
| 698 | if (ret != ACL_SUCCESS) { | 698 | if (ret != ACL_SUCCESS) { |
| 699 | RPT_INPUT_ERR( | 699 | RPT_INPUT_ERR( |
| @@ -986,7 +986,7 @@ void DevUbConnection::ProcessSlices( | |||
| 986 | } | 986 | } |
| 987 | 987 | ||
| 988 | HCCL_INFO( | 988 | HCCL_INFO( |
| 989 | - "[DevUbConnection::%s] end, locBufSize[%u], sliceNUm[%u], sliceSize[%u], lastSliceSize[%u]", __func__, | 989 | + "[DevUbConnection::%s] end, locBufSize[%u], sliceNum[%u], sliceSize[%u], lastSliceSize[%u]", __func__, |
| 990 | locBufSize, sliceNum, sliceSize, lastSliceSize); | 990 | locBufSize, sliceNum, sliceSize, lastSliceSize); |
| 991 | } | 991 | } |
| 992 | 992 | ||
【代码风格】函数名称打印是用[%s] + __func__还是直接打印,需要统一吗