* Copyright (c) 2025 Huawei Technologies Co., Ltd.
* This program is free software, you can redistribute it and/or modify it under the terms and conditions of
* CANN Open Software License Agreement Version 2.0 (the "License").
* Please refer to the License for details. You may not use this file except in compliance with the License.
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
* See LICENSE in the root of the software repository for the full text of the License.
*/
#include "base.hpp"
#include "dev_info_manage.h"
#include "stars_base.hpp"
#include "stars_base_cond_isa_define.hpp"
#include "task_base.hpp"
namespace cce {
namespace runtime {
REGISTER_PLATFORM_LIB_INFO(CHIP_CLOUD, "libruntime_v100.so");
static const std::unordered_set<RtOptionalFeatureType> CHIP_CLOUD_FEATURE{
RtOptionalFeatureType::RT_FEATURE_DEVICE_SPM_POOL,
RtOptionalFeatureType::RT_FEATURE_IPC_MEMORY,
RtOptionalFeatureType::RT_FEATURE_IPC_NOTIFY,
RtOptionalFeatureType::RT_FEATURE_STREAM_GET_AVILIABLE_NUM,
RtOptionalFeatureType::RT_FEATURE_DRIVER_NUMA_TS_MEM_CTRL,
RtOptionalFeatureType::RT_FEATURE_PROFILING_ONLINE,
RtOptionalFeatureType::RT_FEATURE_TASK_FLIP,
RtOptionalFeatureType::RT_FEATURE_DFX_ARGS_DOT_ALLOC_ERROR_LOG,
RtOptionalFeatureType::RT_FEATURE_DEVICE_AICORE_NUM_BY_LEVEL,
RtOptionalFeatureType::RT_FEATURE_TASK_RDMA,
RtOptionalFeatureType::RT_FEATURE_TS_SCHED_AICPU_MODE,
RtOptionalFeatureType::RT_FEATURE_STREAM_RECLAIM_ASYNC,
RtOptionalFeatureType::RT_FEATURE_MEM_MIX_HBM_AND_DDR,
RtOptionalFeatureType::RT_FEATURE_TASK_TRY_RECYCLE_DISABLE_HWTS,
RtOptionalFeatureType::RT_FEATURE_NOTIFY_WAIT_TIMEOUT,
RtOptionalFeatureType::RT_FEATURE_DEVICE_VIRTUAL_MODE_CHECK,
RtOptionalFeatureType::RT_FEATURE_DEVICE_VISIBLE,
RtOptionalFeatureType::RT_FEATURE_TASK_MODEL_EXECUTE_COPY_ONCE,
RtOptionalFeatureType::RT_FEATURE_DEVICE_HAS_TS_DEAMON,
RtOptionalFeatureType::RT_FEATURE_DFX_HWTS_LOG,
RtOptionalFeatureType::RT_FEATURE_DEVICE_P2P,
RtOptionalFeatureType::RT_FEATURE_PROFILING_AICPU,
RtOptionalFeatureType::RT_FEATURE_TASK_RECYCLE_THREAD,
RtOptionalFeatureType::RT_FEATURE_DEVICE_MESSAGE_QUEUE,
RtOptionalFeatureType::RT_FEATURE_DEVICE_STREAM_CTRL,
RtOptionalFeatureType::RT_FEATURE_DRIVER_DFX_RECORD_MODULE_ID,
RtOptionalFeatureType::RT_FEATURE_TASK_LABEL_GOTO,
RtOptionalFeatureType::RT_FEATURE_DEVICE_FAILURE_MODE_SET,
RtOptionalFeatureType::RT_FEATURE_DFX_TS_GET_DEVICE_MSG,
RtOptionalFeatureType::RT_FEATURE_STREAM_HUGE_DEPTH,
RtOptionalFeatureType::RT_FEATURE_STREAM_ABORT,
RtOptionalFeatureType::RT_FEATURE_MODEL_ABORT,
RtOptionalFeatureType::RT_FEATURE_MODEL_ABORT_USE_DEFAULT_STREAM,
RtOptionalFeatureType::RT_FEATURE_DFX_FLOAT_OVERFLOW_DEBUG,
RtOptionalFeatureType::RT_FEATURE_DEVICE_GET_RESOURCE_NUM_DYNAMIC,
RtOptionalFeatureType::RT_FEATURE_DEVICE_SET_RUNTIME_CAPABLITY,
RtOptionalFeatureType::RT_FEATURE_STREAM_DOT_SET_MODE,
RtOptionalFeatureType::RT_FEATURE_TASK_MEMORY_COPY_DOT_HOST,
RtOptionalFeatureType::RT_FEATURE_MEM_HOST_REGISTER,
RtOptionalFeatureType::RT_FEATURE_KERNEL_ELF_NO_FUNC_TYPE,
RtOptionalFeatureType::RT_FEATURE_KERNEL_ELF_AICPU_MACHINE,
RtOptionalFeatureType::RT_FEATURE_TASK_MEMCPY_D2D_BY_OFFSET,
RtOptionalFeatureType::RT_FEATURE_STREAM_ATTR_FAILURE_MODE,
RtOptionalFeatureType::RT_FEATURE_DEVICE_C2C_SYNC,
RtOptionalFeatureType::RT_FEATURE_KERNEL_BINARY_LOAD_DOT_WITHOUT_TILING,
RtOptionalFeatureType::RT_FEATURE_KERNEL_BINARY_GET_FUNCTION_DOT_BY_NAME,
RtOptionalFeatureType::RT_FEATURE_MODEL_ID_FOR_AICPU,
RtOptionalFeatureType::RT_FEATURE_NOTIFY_WAIT,
RtOptionalFeatureType::RT_FEATURE_KERNEL_UMA_SUPER_ARGS_ALLOC,
RtOptionalFeatureType::RT_FEATURE_TASK_DAVINCI_WITH_ARGS_SIZE,
RtOptionalFeatureType::RT_FEATURE_MEM_DESTROY_ADDR_BATCH};
REGISTER_CHIP_FEATURE_SET(CHIP_CLOUD, CHIP_CLOUD_FEATURE);
static constexpr uint32_t CQE_DEPTH = 1024U;
static constexpr uint32_t DEFAULT_CQE_SIZE = 12U;
static constexpr uint32_t RT_HWTS_910A_MAX_BLOCK_DIM_UINT32 = 32U;
static constexpr uint32_t RT_HWTS_910A_DEFAULT_BLOCK_DIM_UINT32 =
1U;
static constexpr double RT_HWTS_910A_TASK_KERNEL_CREDIT_SCALE_US = 68719476.736;
static const DevProperties CHIP_CLOUD_PROPERTIES = {
.engineType = "HWTS",
.isStars = false,
.isStarsV2 = false,
.pthreadStackSize = PTHREAD_STACK_SIZE,
.eventWaitTimeout = EventWaitTimeoutType::SET_OP_WAIT_TIMEOUT_CONFIG,
.tsCount = 1U,
.sqTailOffset = static_cast<uint16_t>(STARS_SIMPLE_SQ_TAIL_OFFSET),
.maxGroupId = MAX_GROUP_ID,
.reduceOverflow = ReduceOverflowType::REDUCE_OVERFLOW_TS_VERSION_REDUCE_V2_ID,
.supportOverflowMode = OVERFLOW_MODE_SATURATION,
.sdmaReduceKind = 0x7C00U,
.starsResourceAddrCalculateMethod = StarsResourceAddrCalculateMethod::STARS_RESOURCE_ADDR_CALCULATE_STATIC,
.eventBase = RT_STARS_BASE_ADDR_520000000 + STARS_EVENT_BASE_ADDR,
.starsBaseMethod = StarsBaseMethod::STARS_BASE_CALCULATE_STATIC,
.fsmSelBase = RT_STARS_BASE_ADDR_520000000 + STARS_SIMPLE_RTSQ_FSM_SEL_REG,
.notifyBase = RT_STARS_BASE_ADDR_520000000 + STARS_NOTIFY_BASE_ADDR,
.starsNotifyTableSize = STARS_NOTIFY_NUM_OF_SINGLE_TABLE_128,
.ipcNotifyNumMask = 0x7FUL,
.mc2FeatureFlag = 0U,
.stackPhyBase = RT_SCALAR_BUFFER_SIZE_32K_75,
.maxCustomerStackSize = DEFAULT_CUSTOM_STACK_SIZE_MAX,
.ioDieNum = 0U,
.defaultTaskRatio = NORMAL_TASK_RATION,
.aicNum = RT_AICORE_NUM_25,
.aivNum = RT_AIVECTOR_NUM_50,
.aicNumPerDie = 0U,
.aivNumPerDie = 0U,
.ringbufSize = DEVICE_RINGBUFFER_SIZE,
.hugeManagedFlag = TS_PAGE_HUGE_ALIGNED | TS_WITH_HBM,
.memInfoType = RT_MEM_INFO_TYPE_HBM_SIZE,
.taskPrefetchCount = static_cast<uint8_t>(PREFETCH_CNT_8),
.maxAllocHugeStreamNum = DEFAULT,
.reduceAicNum = false,
.timeoutUpdateMethod = TimeoutUpdateMethod::DEFAULT_METHOD,
.hugePolicyFlag = DEFAULT,
.memInfoMapType = DEFAULT,
.resAllocRange = DEFAULT,
.supportSnapshot = SupportSnapshot::SUPPORT_ON_HIGH_TS_VERSION,
.MaxKernelCredit = RT_HWTS_910A_MAX_BLOCK_DIM_UINT32,
.eventTimestampFreq = RT_DEFAULT_TIMESTAMP_FREQ,
.taskEngineType = EngineCreateType::DIRECT_HWTS_ENGINE,
.CmoSqeVersion = SqeVersion::CMO_SQE_VERSION_V1,
.DefaultKernelCredit = RT_HWTS_910A_DEFAULT_BLOCK_DIM_UINT32,
.starsDefaultKernelCredit = RT_STARS_DEFAULT_KERNEL_CREDIT,
.KernelCreditScale = RT_HWTS_910A_TASK_KERNEL_CREDIT_SCALE_US,
.isSupportInitFuncCallPara = false,
.rtsqVirtualAddr = {0ULL, 0ULL, 0ULL, 0ULL},
.rtsqFsmStateAddrCalMethod = RtsqFsmStateAddrCalMethod::NOT_SUPPORT,
.starsBaseAddrMethod = StarsBaseAddrMethod::NOT_SUPPORT,
.rtSqEnableAddrCalMethod = RtSqEnableAddrCalMethod::NOT_SUPPORT,
.isSupportDvppAccelerator = false,
.isSupportDevVA2PA = false,
.cmoAddrInfoType = CmoAddrInfoType::CMO_ADDR_INFO_TYPE_DAFAULT,
.memcpyAsyncTaskD2DQos = UINT32_MAX,
.isMemcpyAsyncTaskSqeType = true,
.sqeDepth = UINT32_MAX,
.cqeSize = DEFAULT_CQE_SIZE,
.cqeDepth = CQE_DEPTH,
.cqeWaitTimeout = RT_REPORT_STARS_TIMEOUT_TIME,
.getRtsqPosition = UINT16_MAX,
.creditStartValue = UINT16_MAX,
.argsItemSize = MULTI_GRAPH_ARG_ENTRY_SIZE_OTH,
.argInitCountSize = DEFAULT_INIT_CNT_OTH,
.argsAllocatorSize = TINY_INIT_CNT_DEFAULT,
.superArgAllocatorSize = SUPER_ARG_AllOC_SIZE_128,
.maxArgAllocatorSize = ARG_MAX_ENTRY_INIT_NUM,
.handleAllocatorSize = HANDLE_ALLOCATOR_SIZE_1024,
.kernelInfoAllocatorSize = KERNEL_INFO_ALLOC_SIZE,
.isNeedlogErrorLevel = true,
.opExecuteTimeout = false,
.dcacheLockMixType = DcacheLockMixType::DCACHE_LOCK_MIX_TYPE_FROM_DEFAULT,
.taskPoolSizeFromRtsqDepth = false,
.taskPoolSize = 2048U,
.getTsMemTypeMethod = GetTsMemTypeMethod::GET_TS_MEM_TYPE_JUDGE_FROM_DRIVER,
.enabledTSNum = ENABLED_TS_NUM_1,
.deviceSatStatus = DeviceSatStatus::DEVICE_SAT_STATUS_OTHER,
.ringBufferMemCopyCycleTimes = 1U,
.tsOverflowHandling = TsOverflowHandling::TS_OVER_FLOW_HANDING_FROM_MEM,
.getCapabilityMethod = GetCapabilityMethod::GET_CAPABILITY_BY_BOTH_FEATURE_AND_DRIVER_CHECK,
.memcpyDescriptorSize = MEMCPY_DESC_SIZE_V1,
.cmoDDRStructInfoSize = 0U,
.streamWaitEventTimeout = StreamWaitEventTimeout::NOT_SUPPORT,
.timelineEventId = TIMELINE_EVENT_ID_DEFAULT,
.deviceSatStatusImpl = DeviceSatStatusImpl::DEVICE_SAT_STATUS_CONTEXT_LEVEL,
.allocManagedFlag = AllocManagedFlag::ALLOC_MANAGED_MEM_SET_ALIGN_SIZE,
.sdmaCopyMethod = SdmaCopyMethod::SDMA_COPY_BY_MEM_SYNC_ADAPTER,
.eventPoolSize = 0U,
.rtsqShamt = 0x1FFU,
.supportCreateTaskRes = SupportCreateTaskRes::CREATE_TASK_RES_SUPPORT,
.physicalMemTypePolicy = PhysicalMemTypePolicy::DEFAULT,
.aicNumForCoreStack = RT_AICORE_NUM_25,
.engineWaitCompletionTImeout = 0UL,
.reportWaitTimeout = RT_REPORT_TIMEOUT_TIME,
.hostAtomicCapabilities = {},
.p2pAtomicCapabilities = {},
.maxPersistTaskNum = 60000U,
.maxSupportTaskNum = 4182016U,
.stubEventCount = 65536U,
.maxReportTimeoutCnt = 36,
.rtcqDepth = 0U,
.baseAicpuStreamId = 1024U,
.expandStreamRsvTaskNum = 0U,
.expandStreamSqDepthAdapt = 0U,
.expandStreamAdditionalSqeNum = 0U,
.rsvAicpuStreamNum = 0U,
.maxPhysicalStreamNum = 2016U,
.maxAllocStreamNum = 2016U,
.rtsqDepth = 4096U,
.maxTaskNumPerStream = 2042U,
.maxTaskNumPerHugeStream = 0U,
.rtsqReservedTaskNum = 0U,
.simtWarpSize = 0U,
.simtMaxThreadPerVectorCore = 0U,
.simtUbufPerVectorCore = 0U,
.simtMaxGridDimX = 0U,
.simtMaxGridDimY = 0U,
.simtMaxGridDimZ = 0U,
.simtMaxBlockPerGrid = 0U,
.simtMaxThreadsPerBlock = 0U,
.simtMaxBlockDimX = 0U,
.simtMaxBlockDimY = 0U,
.simtMaxBlockDimZ = 0U,
.cvArchType = DeviceCvArchType::CV_ARCH_INTERGRATION,
.npuArch = 0,
.sqDisableStatPollingCycleNum = SQ_DISABLE_POLLING_CYCLE_COMMON_CNT,
.swapBufferBaseAddr = 0UL,
.swapBufferUpdateRegOffset = 0UL,
.sqSwapShift = 0U,
.swapBufferProfCfgOffset = 0U,
};
REGISTER_DEV_PROPERTIES(CHIP_CLOUD, CHIP_CLOUD_PROPERTIES);
}
}