ConcatD

English | 简体中文

Product Support Status

Product	Supported
Ascend 950PR&950DT products	√
Atlas A3 products	√
Atlas A2 products	√
Atlas 200I/500 A2 inference products	√
Atlas inference products	√
Atlas training products	√
Kirin X90 processor series products	√
Kirin 9030 processor series products	√

Function Description

• Operator function: concatenates all tensors in tensors along dimension dim. All dimensions except the dimension corresponding to dim must be the same.

Parameter Description

Parameter Name	Input/Output/Attribute	Description	Data Type	Data Format
x	Input	List of tensors to be concatenated.	FLOAT, FLOAT16, INT32, INT64, INT16, INT8, UINT8, UINT16, UINT32, UINT64, BOOL, BFLOAT16, DOUBLE, COMPLEX64, HIFLOAT8, FLOAT8_E5M2, FLOAT8_E4M3FN, FLOAT8_E8M0, FLOAT4_E1M2, FLOAT4_E2M1	ND
concat_dim	Attribute	Dimension along which tensors are concatenated.	INT	-
N	Optional attribute	Number of tensors to be concatenated.	INT	-
y	Output	Output tensor.	FLOAT, FLOAT16, INT32, INT64, INT16, INT8, UINT8, UINT16, UINT32, UINT64, BOOL, BFLOAT16, DOUBLE, COMPLEX64, HIFLOAT8, FLOAT8_E5M2, FLOAT8_E4M3FN, FLOAT8_E8M0, FLOAT4_E1M2, FLOAT4_E2M1	ND

• Atlas training products, Atlas inference products, and Atlas 200I/500 A2 inference products: BFLOAT16 is not supported.
• Atlas A2 products and Atlas A3 products: UINT16, UINT32, UINT64, FLOAT4_E1M2, and FLOAT4_E2M1 are not supported.
• Kirin X90/9030 processor series products: The data types FLOAT, FLOAT16, INT32, INT64, INT16, INT8, UINT8, UINT16, UINT32, and UINT64 are supported; 1 to 32 non-empty ND inputs are supported, the input rank is known at compile time and is 1 to 8, and each dimension must be resolved to a positive number before tiling at runtime.

Constraint Description

• The data types and data formats of elements in the x list must be within the supported range.
• Data type inference can be performed on the x list.
• The inferred data type must be convertible to the specified output type of y.
• The shapes of non-concatenated dimensions must be the same.
• dim must be within the dimension range of x.
• When the input data type is FLOAT4_E1M2 or FLOAT4_E2M1, the product of the sizes of each input tensor along the concat axis and the dimensions after it must be an even number.

Invocation Description

Invocation Method	Sample Code	Description
aclnn API	test_aclnn_cat	Invokes the ConcatD operator through the aclnnCat API.