For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Mojo function

pad_constant

def pad_constant[rank: Int, dtype: DType, padding_type: DType](output: Pointer[Scalar[dtype], _safe=False], output_shape: IndexList[rank], input: Pointer[Scalar[dtype], _safe=False], input_shape: IndexList[rank], paddings: Pointer[Scalar[padding_type], _safe=False], constant: Scalar[dtype], ctx: DeviceContext)

Fill output with values from input, and edges padded with constant based on paddings.

Example:

var input_shape = (X, Y, Z)
var paddings = [x0, x1, y0, y1, z0, z1]

out[x, y, z] =
  input[x - x0, y - y0, z - z0] if x ∈ [x0, x0 + X] &&
                                   y ∈ [y0, y0 + Y] &&
                                   z ∈ [z0, z0 + Z]
  else constant

Args:

output (Pointer[Scalar[dtype], _safe=False]): The output buffer.
output_shape (IndexList[rank]): The output shape.
input (Pointer[Scalar[dtype], _safe=False]): The input buffer.
input_shape (IndexList[rank]): The input shape.
paddings (Pointer[Scalar[padding_type], _safe=False]): Ordered (before, after) padding sizes for each axis.
constant (Scalar[dtype]): The constant to pad output with.
ctx (DeviceContext): Device context for participating GPU.