cast_to

Description

Casts a data type, specifically, from data type of data into target dtype. The supported cast types are listed as follows.

**Table 1** Supported cast types
Source Data Type	Target Data Type	Rounding Mode	Description
float32	float16	N/A	Supported by all SoC types
float32	int8	Round down	Supported by all SoC types
float32	uint8	Round down	Supported by all SoC types
float16	float32	N/A	Supported by all SoC types
float16	int8	Round down	Supported by all SoC types
float16	uint8	Round down	Supported by all SoC types
float16	int32	Round down	Supported by all SoC types
int8	float16	N/A	Supported by all SoC types
int8	uint8	Round down	Supported by all SoC types
int32	float16	N/A	Supported by all SoC types
int32	int8	Round down	Supported by all SoC types
int32	uint8	Round down	Supported by all SoC types
uint8	int8	Round down	Supported by: Atlas inference product
uint8	float16	N/A	Supported by: Atlas inference product
int16	int8	Round down	Supported by: Atlas inference product
int16	uint8	Round down	Supported by: Atlas inference product
int16	float16	N/A	Supported by: Atlas inference product
float32	int32	Round down	Supported by: Atlas inference product
int32	float32	N/A	Supported by: Atlas inference product
int32	int64	Round down	Supported by: Atlas inference product
int64	int32	Round down	Supported by: Atlas inference product
int64	float16	N/A	Supported by: Atlas training product Atlas inference product Atlas 200I/500 A2 inference product Atlas A2 training product/Atlas A2 inference product Atlas A3 training product/Atlas A3 inference product
float16	complex32	N/A	When a real number is converted to a complex number, the imaginary part is 0 by default. Supported by: Atlas A2 training product/Atlas A2 inference product Atlas A3 training product/Atlas A3 inference product
float32	complex64	N/A	When a real number is converted to a complex number, the imaginary part is 0 by default. Supported by: Atlas A2 training product/Atlas A2 inference product Atlas A3 training product/Atlas A3 inference product

Prototype

cast_to(data, dtype, f1628IntegerFlag=True)

Parameters

data: a tvm.tensor for the input tensor.
dtype: a string for the target data type.
f1628IntegerFlag: defaults to True. If the decimal part of data is 0, set f1628IntegerFlag to True. Otherwise, set f1628IntegerFlag to False.

Returns

res_tensor: a tvm.tensor for the result tensor

Restrictions

None

Applicability

Atlas training product

Atlas inference product

Atlas 200I/500 A2 inference product

Atlas A2 training product/Atlas A2 inference product

Atlas A3 training product/Atlas A3 inference product

Example

from tbe import tvm
from tbe import dsl
shape = (1024,1024)
input_dtype = "float16"
data = tvm.placeholder(shape, name="data", dtype=input_dtype)
res = dsl.cast_to(data,"float32")

Parent topic: Cast Compute APIs