SoftmaxFlash Tiling

Function

Note: This API will be deprecated in the future. Do not use it for new development.

Obtains the tiling parameters required by SoftmaxFlash.

Prototype

  • APIs for obtaining the minimum and maximum temporary space required for kernel computation
    1
    uint32_t GetSoftMaxFlashMaxTmpSize(const ge::Shape& srcShape, const uint32_t dataTypeSize, const bool isUpdate, const bool isReuseSource)
    
    1
    uint32_t GetSoftMaxFlashMinTmpSize(const ge::Shape& srcShape, const uint32_t dataTypeSize, const bool isUpdate, const bool isReuseSource)
    
  • Tiling computation APIs
    • Computation API in the AscendC::optiling namespace
      1
      void SoftMaxFlashTilingFunc(const ge::Shape& srcShape, const uint32_t dataTypeSize, const uint32_t localWorkSpaceSize, optiling::SoftMaxTiling& softmaxFlashTiling, const bool isUpdate = false)
      
    • Computation API in the AscendC namespace
      1
      void SoftMaxFlashTilingFunc(const ge::Shape& srcShape, const uint32_t dataTypeSize, const uint32_t localWorkSpaceSize, AscendC::tiling::SoftMaxTiling& softmaxFlashTiling, const bool isUpdate = false)
      

Parameters

Table 1 SoftmaxFlash GetSoftMaxFlashMaxTmpSize/GetSoftMaxFlashMinTmpSize parameters

Parameter

Input/Output

Description

srcShape

Input

Shape of the input srcTensor.

dataTypeSize

Input

Data type size of maxTensor and sumTensor involved in computation, for example, half = 2.

isUpdate

Input

Whether to enable the refresh function. The value must be consistent with that of the SoftmaxFlash in the kernel. The default value is false.

isReuseSource

Input

The value must be the same as the API configuration in the kernel.

Table 2 SoftmaxFlash SoftMaxFlashTilingFunc parameters

Parameter

Input/Output

Description

srcShape

Input

Shape of the input srcTensor.

dataTypeSize

Input

Data type size of maxTensor and sumTensor involved in computation, for example, half = 2.

localWorkSpaceSize

Input

Size of the remaining space that can be used for SoftmaxFlash computation. The unit is byte. The value of localWorkSpaceSize must be greater than the minimum temporary space size required for computation by the GetSoftMaxFlashMinTmpSize API.

isUpdate

Input

Whether to enable the refresh function. The value must be consistent with that of the SoftmaxFlash in the kernel. The default value is false.

softmaxFlashTiling

Output

Tiling information required by the SoftmaxFlash APIs. The input parameters in the optiling::SoftMaxTiling and AscendC::tiling::SoftMaxTiling formats are supported.

Returns

GetSoftMaxFlashMaxTmpSize returns the maximum size (in bytes) of the temporary space required by SoftmaxFlash computation.

GetSoftMaxFlashMinTmpSize returns the minimum size (in bytes) of the temporary space required by SoftmaxFlash computation.

Restrictions

None