vec_relu

Description

Performs ReLU element-wise:

ReLU stands for rectified linear unit, and is the most used activation function in artificial neural networks.

Prototype

vec_relu(mask, dst, src, repeat_times, dst_rep_stride, src_rep_stride)

Parameters

For details, see Parameters.

dst has the same data type as src.

Atlas training products : Tensors of type float16

Atlas inference product AI Core: Tensors of type float16/float32/int32

Atlas inference product Vector Core: Tensors of type float16/float32/int32

Atlas A2 training products / Atlas A2 inference products : Tensors of type float16/float32/int32

Atlas 200I/500 A2 inference products : Tensors of type float16/float32/int32

Returns

None

Applicable Models

Atlas training products

Atlas inference product AI Core

Atlas inference product Vector Core

Atlas A2 training products / Atlas A2 inference products

Atlas 200I/500 A2 inference products

Restrictions

For details, see Restrictions.

Example

This example applies to a small amount of data that can be moved at a time, helping you understand the API functions. For more complex samples with a large amount of data, see Example.

from tbe import tik
tik_instance = tik.Tik()
src_gm = tik_instance.Tensor("float16", (128,), name="src_gm", scope=tik.scope_gm)
dst_gm = tik_instance.Tensor("float16", (128,), name="dst_gm", scope=tik.scope_gm)
src_ub = tik_instance.Tensor("float16", (128,), name="src_ub", scope=tik.scope_ubuf)
dst_ub = tik_instance.Tensor("float16", (128,), name="dst_ub", scope=tik.scope_ubuf)
# Move the user input from the Global Memory to the Unified Buffer.
tik_instance.data_move(src_ub, src_gm, 0, 1, 8, 0, 0)
tik_instance.vec_relu(128, dst_ub, src_ub, 1, 8, 8)
# Move the compute result from the Unified Buffer to the destination Global Memory.
tik_instance.data_move(dst_gm, dst_ub, 0, 1, 8, 0, 0)

tik_instance.BuildCCE(kernel_name="vec_relu", inputs=[src_gm], outputs=[dst_gm])

Result example:

Input (src_gm):
[ 6.938   -8.86    -0.2263   6.77     2.924    1.759    0.4253  -5.23
 -1.892   -3.049    4.      -9.49    -0.8145   1.974    7.793    2.13
 -3.799    1.292   -0.311   -6.883   -3.29     6.445    7.65     6.76
  8.96    -6.84     3.111   -6.984    7.773   -7.605   -1.563   -5.6
 -2.938    6.785   -1.157    2.373   -3.924   -1.134   -5.523    7.082
  0.5425   9.33     3.734   -7.004   -3.535   -6.35     2.137   -6.42
 -3.076    4.93    -8.234   -7.156   -9.96    -2.623   -2.625   -8.516
  0.88    -3.312   -9.23    -4.734   -0.834    1.154   -0.2268   6.79
  0.559   -4.3     -0.2212   0.02264 -2.775    3.691    8.13    -5.555
  8.766    0.1989  -4.473   -7.99    -5.81    -2.379   -8.64     9.85
  6.867    3.43    -5.176    8.89     5.55     4.586   -8.45     0.3813
  2.875    4.027   -8.96    -9.49    -3.764    4.688   -0.723    8.24
  4.67     4.016    5.266    9.47    -3.033    9.53     2.674    0.2131
  6.836    0.3386   9.95     4.73     5.87    -3.758   -9.45     2.574
 -8.914    9.49     7.42    -7.453    8.19     3.479   -0.0785   0.1791
 -7.098   -9.5      7.41     3.854   -7.57    -6.91     1.971    1.778  ]

Output (dst_gm):
[6.938   0.      0.      6.77    2.924   1.759   0.4253  0.      0.
 0.      4.      0.      0.      1.974   7.793   2.13    0.      1.292
 0.      0.      0.      6.445   7.65    6.76    8.96    0.      3.111
 0.      7.773   0.      0.      0.      0.      6.785   0.      2.373
 0.      0.      0.      7.082   0.5425  9.33    3.734   0.      0.
 0.      2.137   0.      0.      4.93    0.      0.      0.      0.
 0.      0.      0.88    0.      0.      0.      0.      1.154   0.
 6.79    0.559   0.      0.      0.02264 0.      3.691   8.13    0.
 8.766   0.1989  0.      0.      0.      0.      0.      9.85    6.867
 3.43    0.      8.89    5.55    4.586   0.      0.3813  2.875   4.027
 0.      0.      0.      4.688   0.      8.24    4.67    4.016   5.266
 9.47    0.      9.53    2.674   0.2131  6.836   0.3386  9.95    4.73
 5.87    0.      0.      2.574   0.      9.49    7.42    0.      8.19
 3.479   0.      0.1791  0.      0.      7.41    3.854   0.      0.
 1.971   1.778  ]

Parent topic: Single Source (Gather Mode)