habana_frameworks.mediapipe.fn.Add

Class:
  • habana_frameworks.mediapipe.fn.Add(**kwargs)

Define graph call:
  • __call__(input1, input2)

Parameter:
  • input1 - First input tensor to operator. Supported dimensions: minimum = 1, maximum = 5. Supported data types: INT16, INT32, FLOAT16, BFLOAT16, FLOAT32.

  • input2 - Second input tensor to operator. Supported dimensions: minimum = 1, maximum = 5. Supported data types: INT16, INT32, FLOAT16, BFLOAT16, FLOAT32.

Description:

The resulting tensor is formed from the summation of the two operands element-wise. This operator performs element-wise addition and supports Broadcasting. Computes output as: output = (input1 + input2), element-wise.

Supported backend:
  • HPU, CPU

Keyword Arguments

kwargs

Description

dtype

Output data type.

  • Type: habana_frameworks.mediapipe.media_types.dtype

  • Default: FLOAT32

  • Optional: yes

  • Supported data types:

    • INT32

    • BFLOAT16

    • FLOAT32

Note

  1. All input/output tensors must be of the same data type and must have the same dimensionality except in broadcast support where dimensionality can be different.

  2. This operator is agnostic to the data layout.

Example: Add Operator

The following code snippet shows usage of Add operator:

from habana_frameworks.mediapipe import fn
from habana_frameworks.mediapipe.mediapipe import MediaPipe
from habana_frameworks.mediapipe.media_types import dtype as dt
import numpy as np
import os

# Create MediaPipe derived class


class myMediaPipe(MediaPipe):
    def __init__(self, device, queue_depth, batch_size, num_threads, op_device, dir):
        super(myMediaPipe, self).__init__(
            device,
            queue_depth,
            batch_size,
            num_threads,
            self.__class__.__name__)

        self.inp1 = fn.ReadNumpyDatasetFromDir(num_outputs=1,
                                              shuffle=False,
                                              dir=dir,
                                              pattern="inp_x_*.npy",
                                              dense=True,
                                              dtype=dt.FLOAT32,
                                              device="cpu")

        self.inp2 = fn.ReadNumpyDatasetFromDir(num_outputs=1,
                                              shuffle=False,
                                              dir=dir,
                                              pattern="inp_y_*.npy",
                                              dense=True,
                                              dtype=dt.FLOAT32,
                                              device="cpu")
        self.add = fn.Add(dtype=dt.FLOAT32,
                          device=op_device)

    def definegraph(self):
        inp0 = self.inp1()
        inp1 = self.inp2()
        out = self.add(inp0, inp1)
        return out, inp0, inp1


def run(device, op_device):
    batch_size = 2
    queue_depth = 2
    num_threads = 1
    base_dir = os.environ['DATASET_DIR']
    dir = base_dir+"/npy_data/fp32/"

    # Create MediaPipe object
    pipe = myMediaPipe(device, queue_depth, batch_size,
                      num_threads, op_device, dir)

    # Build MediaPipe
    pipe.build()

    # Initialize MediaPipe iterator
    pipe.iter_init()

    # Run MediaPipe
    out, inp1, inp2 = pipe.run()

    def as_cpu(tensor):
        if (callable(getattr(tensor, "as_cpu", None))):
            tensor = tensor.as_cpu()
        return tensor

    # Copy data to host from device as numpy array
    out = as_cpu(out).as_nparray()
    inp1 = as_cpu(inp1).as_nparray()
    inp2 = as_cpu(inp2).as_nparray()

    del pipe

    print("\ninp1 tensor shape:", inp1.shape)
    print("inp1 tensor dtype:", inp1.dtype)
    print("inp1 tensor data:\n", inp1)

    print("\ninp2 tensor shape:", inp2.shape)
    print("inp2 tensor dtype:", inp2.dtype)
    print("inp2 tensor data:\n", inp2)

    print("\nout tensor shape:", out.shape)
    print("out tensor dtype:", out.dtype)
    print("out tensor data:\n", out)

    return inp1, inp2, out


def compare_ref(inp1, inp2, out):
    ref = inp1 + inp2
    if np.array_equal(ref, out) == False:
        raise ValueError(f"Mismatch w.r.t ref")


if __name__ == "__main__":
    dev_opdev = {'cpu': ['cpu'],
                'mixed': ['hpu'],
                'legacy': ['hpu']}
    for dev in dev_opdev.keys():
        for op_dev in dev_opdev[dev]:
            inp1, inp2, out = run(dev, op_dev)
            compare_ref(inp1, inp2, out)

The following is the output of Add operator:

inp1 tensor shape: (2, 3, 2, 3)
inp1 tensor dtype: float32
inp1 tensor data:
[[[[182. 227. 113.]
  [175. 128. 253.]]

  [[ 58. 140. 136.]
  [ 86.  80. 111.]]

  [[175. 196. 178.]
  [ 20. 163. 108.]]]


[[[186. 254.  96.]
  [180.  64. 132.]]

  [[149.  50. 117.]
  [213.   6. 111.]]

  [[ 77.  11. 160.]
  [129. 102. 154.]]]]

inp2 tensor shape: (2, 3, 2, 3)
inp2 tensor dtype: float32
inp2 tensor data:
[[[[ 56. 168.  82.]
  [157.  42. 155.]]

  [[ 62. 235. 238.]
  [ 94. 125. 192.]]

  [[125. 162.   1.]
  [206.  77. 123.]]]


[[[138. 196. 246.]
  [137. 203.   7.]]

  [[217. 194.  11.]
  [167. 218. 226.]]

  [[ 68. 160. 254.]
  [243.  93.  70.]]]]

out tensor shape: (2, 3, 2, 3)
out tensor dtype: float32
out tensor data:
[[[[238. 395. 195.]
  [332. 170. 408.]]

  [[120. 375. 374.]
  [180. 205. 303.]]

  [[300. 358. 179.]
  [226. 240. 231.]]]


[[[324. 450. 342.]
  [317. 267. 139.]]

  [[366. 244. 128.]
  [380. 224. 337.]]

  [[145. 171. 414.]
  [372. 195. 224.]]]]