aclFormat

ACL_FORMAT_UNDEFINED = -1

Unknown data type (default)

ACL_FORMAT_NCHW = 0

NCHW format

ACL_FORMAT_NHWC = 1

NHWC format

ACL_FORMAT_ND = 2

ND: indicates that all formats are supported. Exercise cautions when using operators except for single-input operators that are processed inherently such as square and tanh.

ACL_FORMAT_NC1HWC0 = 3

5D data format. C0 is closely related to the micro-architecture, and the value is equal to the Cube Unit size, for example, 16. C1 is obtained by dividing the C dimension by C0, that is, C1 = C/C0. When the division is not exact, the last data segment is padded to C0.

ACL_FORMAT_FRACTAL_Z = 4

Format of the convolution weight

ACL_FORMAT_NC1HWC0_C04 = 12

5D data format. C0 has a fixed value of 4 while C1 = C/C0. When C is not exactly divisible by C0, the last data segment is padded to C0. It is not supported by the current version.

ACL_FORMAT_HWCN = 16

HWCN format.

ACL_FORMAT_NDHWC = 27

NDHWC format. The depth (D) dimension is required for 3D images.

ACL_FORMAT_FRACTAL_NZ = 29

Format for internal use only

ACL_FORMAT_NCDHW = 30

NCDHW format. The depth (D) dimension is required for 3D images.

ACL_FORMAT_NDC1HWC0 = 32

6D data format. Has an additional depth (D) dimension compared with NC1HWC0.

ACL_FRACTAL_Z_3D = 33

3D convolution weight format, which is required by operators such as Conv3D, MaxPool3D, and AvgPool3D.

ACL_FORMAT_NC = 35

2D format.

ACL_FORMAT_NCL = 47

3D format.