SetSplitRange
Function
Sets the maximum and minimum values of baseM, baseN, and baseK. Currently, tiling does not support this function.
Prototype
1 | int32_t SetSplitRange(int32_t maxBaseM = -1, int32_t maxBaseN = -1, int32_t maxBaseK = -1, int32_t minBaseM = -1, int32_t minBaseN = -1, int32_t minBaseK = -1) |
Parameters
Parameter |
Input/Output |
Description |
|---|---|---|
maxBaseM |
Input |
The maximum value of baseM. The default value is -1. The value -1 indicates that the maximum value of baseM is not set. Instead, it is calculated by the tiling function. |
maxBaseN |
Input |
The maximum value of baseN. The default value is -1. The value -1 indicates that the maximum value of baseN is not set. Instead, it is calculated by the tiling function. |
maxBaseK |
Input |
The maximum value of baseK. The default value is -1. The value -1 indicates that the maximum value of baseK is not set. Instead, it is calculated by the tiling function. |
minBaseM |
Input |
The minimum value of baseM. The default value is -1. The value -1 indicates that the minimum value of baseM is not set. Instead, it is calculated by the tiling function. |
minBaseN |
Input |
The minimum value of baseN. The default value is -1. The value -1 indicates that the minimum value of baseN is not set. Instead, it is calculated by the tiling function. |
minBaseK |
Input |
The minimum value of baseK. The default value is -1. The value -1 indicates that the minimum value of baseK is not set. Instead, it is calculated by the tiling function. |
Returns
-1: setting failed; 0: setting succeeded.
Restrictions
If baseM, baseN, or baseK is not aligned with C0_size, the value will be aligned with C0_size during tiling computation. C0_size of the half/bfloat16_t data type is 16, that of the float data type is 8, that of the int8_t data type is 32, and that of the int4b_t data type is 64.