Tile

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	√

Function Description

API function: replicates the input tensor along each dimension for the number of times specified in repeats. Example: Assume that the input tensor is [[a,b],[c,d],[e,f]], that is, the shape is [3,2], and repeats is (2,4). Then the generated tensor has the shape [6,8], and the values are as follows:

>>> x = torch.tensor([[a,b],[c,d],[e,f]])
>>> x.repeat(2,4)
tensor([[a,b,a,b,a,b,a,b],
        [c,d,c,d,c,d,c,d],
        [e,f,e,f,e,f,e,f],
        [a,b,a,b,a,b,a,b],
        [c,d,c,d,c,d,c,d],
        [e,f,e,f,e,f,e,f],
        ])

When repeats is (2,4,2), that is, the number of elements in repeats is greater than the number of dimensions in the tensor, the output tensor is equivalent to the following operations: first, expand the shape of the input tensor to the number of dimensions equal to the number of elements in repeats: [1,3,2]; then, expand the tensor along each dimension according to the value in repeats, that is, the output tensor has the shape [2,12,4]. The result is as follows:

>>> x.repeat(2,4,2)
tensor([[[a,b,a,b],
         [c,d,c,d],
         [e,f,e,f],
         [a,b,a,b],
         [c,d,c,d],
         [e,f,e,f],
         [a,b,a,b],
         [c,d,c,d],
         [e,f,e,f],
         [a,b,a,b],
         [c,d,c,d],
         [e,f,e,f]],

        [[a,b,a,b],
         [c,d,c,d],
         [e,f,e,f],
         [a,b,a,b],
         [c,d,c,d],
         [e,f,e,f],
         [a,b,a,b],
         [c,d,c,d],
         [e,f,e,f],
         [a,b,a,b],
         [c,d,c,d],
         [e,f,e,f]]])

The following conditions must be met during computation: The number of parameters in repeats cannot be less than the number of dimensions of the input tensor. The values in repeats must be greater than or equal to 0.

Parameter Description

Parameter Name	Input/Output	Description	Usage Instructions	Data Type	Data Format	Dimension (shape)	Non-contiguous Tensor
self	Input	-	-	FLOAT, DOUBLE, FLOAT16, COMPLEX64, COMPLEX128, UINT8, INT8, INT16, INT32, INT64, UINT16, UINT32, UINT64, BOOL, BFLOAT16, HIFLOAT8, FLOAT8_E5M2, FLOAT8_E4M3FN	ND	≤8	√
repeats	Input	-	Indicates the number of times to repeat the input tensor along each dimension. The number of parameters is not greater than 8. Currently, the scenario of repeating more than 4 dimensions at the same time is not supported. For details about the constraints, see Constraint Description.	INT64	-	-	-
out	Output	-	-	Same as self	ND	≤8	√

Constraint Description

• Deterministic computation:
  • By default, the implementation is deterministic.

Invocation Example

Invocation Method	Sample Code	Description
aclnn API	test_aclnn_repeat	Invokes the Tile operator through the aclnnRepeat API.
Graph-mode invocation	test_geir_tile	Invokes the Tile operator by building a graph through the operator IR.