GetGeGLUTmpBufferFactorSize

Function

To perform GeGLU computation in the kernel, developers need to reserve or allocate the temporary space. The relationship between the maximum temporary space (maxTmpBuffer) and the space occupied by the input (inputSize × typeSize) is as follows:

maxTmpBuffer = maxLiveNodeCount * inputSize * typeSize + extraBuf

maxLiveNodeCount indicates how many times the maximum temporary space is the space occupied by the input. extraBuf indicates the size of the extra temporary space.

This API is used to obtain maxLiveNodeCount and extraBuf. When the fixed space size is used, the maximum number of elements that can be computed by the operator at a time can be calculated based on maxLiveNodeCount and extraBuf.

The following is an example:

The GeGLU API needs to be called for operator implementation. Developers need to reserve space of the currBuff size and use the GetGeGLUTmpBufferFactorSize API to obtain the output values of maxLiveNodeCount and extraBuf and calculate the maximum number of elements calculated by the GeGLU operator at a time.

currentShapeSize = (currBuff - extraBuf) / maxLiveNodeCount / typeSize

Prototype

1

void GetGeGLUTmpBufferFactorSize(const uint32_t typeSize, uint32_t& maxLiveNodeCount, uint32_t& extraBuf)

Parameters

Returns

None

Restrictions

If currentShapeSize × typeSize < 256B is obtained based on maxLiveNodeCount and extraBuf, currentShapeSize should be rounded up based on the value of 256B/typeSize.

Example

1
2
3
4
5

uint32_t typeSize = sizeof(half);
uint32_t maxLiveNodeCount = 0;
uint32_t extraBuf = 0;

AscendC::GetGeGLUTmpBufferFactorSize(typeSize, maxLiveNodeCount, extraBuf);