SetFlag/WaitFlag(ISASI)

Applicability

Product	Supported
Atlas 350 Accelerator Card	√
Atlas A3 training product/Atlas A3 inference product	√
Atlas A2 training product/Atlas A2 inference product	√
Atlas 200I/500 A2 inference product	x
Atlas inference product AI Core	√
Atlas inference product Vector Core	x
Atlas training product	√

Function Usage

Synchronizes different pipelines in the same core. This synchronization operation needs to be inserted between different pipeline instructions with data dependency.

Prototype

template <HardEvent event>
__aicore__ inline void SetFlag(int32_t eventID)
template <HardEvent event>
__aicore__ inline void WaitFlag(int32_t eventID)

Parameters

**Table 1** Parameters
Parameter	Input/Output	Description
event	Input	Template parameter. Synchronization event of the HardEvent data type. For details, see the following description of synchronization types.
eventID	Input	Event ID. The data type is int32_t. The definition is as follows: In TPipe- and TQue-based programming scenarios, the event ID needs to be obtained by using AllocEventID (AllocEventID) or FetchEventID (FetchEventID). Atlas training product: The value ranges from 0 to 3. Atlas inference product AI Core: The value ranges from 0 to 7. Atlas A2 training product/Atlas A2 inference product: The value ranges from 0 to 7. Atlas A3 training product/Atlas A3 inference product: The value ranges from 0 to 7. Atlas 350 Accelerator Card: The value ranges from 0 to 7.

The synchronization types are described as follows:

enum class HardEvent : uint8_t {
   // Name (source pipeline_target pipeline). For example, MTE2_V indicates that PIPE_MTE2 is the source pipeline and PIPE_V is the target pipeline. Synchronization is performed between PIPE_MTE2 and PIPE_V, and PIPE_V waits for PIPE_MTE2.
    MTE2_MTE1
    MTE1_MTE2
    MTE1_M
    M_MTE1
    MTE2_V
    V_MTE2
    MTE3_V
    V_MTE3
    M_V
    V_M
    V_V
    MTE3_MTE1
    MTE1_MTE3
    MTE1_V
    MTE2_M
    M_MTE2
    V_MTE1
    M_FIX
    FIX_M
    MTE3_MTE2
    MTE2_MTE3
    S_V
    V_S
    S_MTE2
    MTE2_S
    S_MTE3
    MTE3_S
    MTE2_FIX
    FIX_MTE2
    FIX_S
    M_S
    FIX_MTE3
}

Returns

None

Constraints

SetFlag and WaitFlag must be used in pairs.
In TPipe- and TQue-based programming scenarios, you are not advised to specify the event ID when using SetFlag and WaitFlag. Otherwise, the event ID may conflict with the framework synchronization event, causing suspension.
In static tensor programming scenarios, the event types and event IDs are managed by developers. However, note that event IDs 6 and 7 cannot be used (as they may conflict with internal event IDs, leading to undefined behavior).

Example

DataCopy can be executed only after SetValue is executed. In this case, you need to insert the synchronization operation between PIPE_S and PIPE_MTE3.

AscendC::GlobalTensor<half> dstGlobal;
AscendC::LocalTensor<half> dstLocal;
dstLocal.SetValue(0, 0);
uint32_t dataSize = 512;
// In TPipe- and TQue-based programming scenarios, the event ID needs to be obtained by using AllocEventID or FetchEventID.
int32_t eventIDSToMTE3 = static_cast<int32_t>(GetTPipePtr()->FetchEventID(AscendC::HardEvent::S_MTE3));
AscendC::SetFlag<AscendC::HardEvent::S_MTE3>(eventIDSToMTE3);
AscendC::WaitFlag<AscendC::HardEvent::S_MTE3>(eventIDSToMTE3);
// In static tensor programming scenarios, the event ID is managed by developers.
// AscendC::SetFlag<AscendC::HardEvent::S_MTE3>(EVENT_ID0);
// AscendC::WaitFlag<AscendC::HardEvent::S_MTE3>(EVENT_ID0);
AscendC::DataCopy(dstGlobal, dstLocal, dataSize);

Parent topic: Intra-Core Synchronization