Support py-boolean in indexing

[zoooo0820]

PR types

New features

PR changes

APIs

Description

Pcard-66985

Support single py-boolean in indexing. The tensor will firstly promote the corresponding dimension, and then the Bool value will be treated as advanced-index.

>>> x = paddle.randn((4,3,2))
>>> x
Tensor(shape=[4, 3, 2], dtype=float32, place=Place(cpu), stop_gradient=True,
       [[[ 0.47239113,  0.56600982],
         [-2.01862693,  1.10056722],
         [-0.65087581,  0.76055008]],

        [[ 0.11856326,  1.35000455],
         [ 0.32557386,  0.86446381],
         [-0.76793766, -0.04744434]],

        [[-0.66354942,  2.03688121],
         [ 1.28386986, -0.19520277],
         [-1.12073815,  1.21294630]],

        [[-1.17965138, -0.35202494],
         [ 1.22328508,  1.34666431],
         [-0.49775603,  0.60246438]]])
>>> x[False]
Tensor(shape=[0, 4, 3, 2], dtype=float32, place=Place(cpu), stop_gradient=True,
       [])

>>> x[True]
Tensor(shape=[1, 4, 3, 2], dtype=float32, place=Place(cpu), stop_gradient=True,
       [[[[ 0.47239113,  0.56600982],
          [-2.01862693,  1.10056722],
          [-0.65087581,  0.76055008]],

         [[ 0.11856326,  1.35000455],
          [ 0.32557386,  0.86446381],
          [-0.76793766, -0.04744434]],

         [[-0.66354942,  2.03688121],
          [ 1.28386986, -0.19520277],
          [-1.12073815,  1.21294630]],

         [[-1.17965138, -0.35202494],
          [ 1.22328508,  1.34666431],
          [-0.49775603,  0.60246438]]]])

[jeff41404]

LGTM