Documentation

aclFormat

typedef enum {
    ACL_FORMAT_UNDEFINED = -1,
    ACL_FORMAT_NCHW = 0,
    ACL_FORMAT_NHWC = 1,
    ACL_FORMAT_ND = 2,
    ACL_FORMAT_NC1HWC0 = 3,
    ACL_FORMAT_FRACTAL_Z = 4,
    ACL_FORMAT_NC1HWC0_C04 = 12,
    ACL_FORMAT_HWCN = 16,
    ACL_FORMAT_NDHWC = 27,
    ACL_FORMAT_FRACTAL_NZ = 29,
    ACL_FORMAT_NCDHW = 30,
    ACL_FORMAT_NDC1HWC0 = 32,
    ACL_FRACTAL_Z_3D = 33,
    ACL_FORMAT_NC = 35,
    ACL_FORMAT_NCL = 47,
} aclFormat;
  • UNDEFINED (default): unknown format
  • NCHW: a 4D format.
  • NHWC: a 4D format.
  • ND: any format, applicable to operators that take singular inputs, such as Square and Tanh.
  • NC1HWC0: a 5D 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.
  • FRACTAL_Z: the format of the convolution weight.
  • NC1HWC0_C04: a 5D 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.
  • HWCN: a 4D format.
  • NDHWC: The depth (D) dimension is required for 3D images.
  • FRACTAL_NZ: an internal fractal format, which is not used currently.
  • NCDHW: The depth (D) dimension is required for 3D images.
  • NDC1HWC0: a 6D format. Has an additional depth (D) dimension compared with NC1HWC0.
  • FRACTAL_Z_3D: a 3D convolution weight format, which is required by operators such as Conv3D, MaxPool3D, and AvgPool3D.
  • NC: a 2D format.
  • NCL: a 3D format.

The available dimensions of a feature image are the batch size (N), height (H), width (W), channels (C), depth (D), and length (L).