fix output dtype for paddle.sum

[GuoxiaWang]

PR types

Bug fixes

PR changes

APIs

Describe

support bool dtype for paddle.sum

Test Code

import numpy as np
import torch
import paddle

for dtype in ['bool', 'float32', 'float64', 'int32', 'int64']:
    print('='*20)
    np_arr = np.ones((12, 3), dtype=dtype)
    print('input:')
    print(np_arr, np_arr.dtype)

    print('sum of numpy:')
    np_sum = np_arr.sum(0)
    print(np_sum, np_sum.dtype)

    pd_arr = paddle.to_tensor(np_arr)
    print('sum of paddle:')
    pd_sum = pd_arr.sum(0)
    print(pd_sum, pd_sum.dtype)

    pt_arr = torch.tensor(np_arr)
    print('sum of torch:')
    pt_sum = pt_arr.sum(0)
    print(pt_sum, pt_sum.dtype)

Outputs

====================
input:
[[ True  True  True]
 [ True  True  True]
 [ True  True  True]
 [ True  True  True]
 [ True  True  True]
 [ True  True  True]
 [ True  True  True]
 [ True  True  True]
 [ True  True  True]
 [ True  True  True]
 [ True  True  True]
 [ True  True  True]] bool
sum of numpy:
[12 12 12] int64
sum of paddle:
Tensor(shape=[3], dtype=int64, place=CUDAPlace(0), stop_gradient=True,
       [12, 12, 12]) paddle.int64
sum of torch:
tensor([12, 12, 12]) torch.int64
====================
input:
[[1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]] float32
sum of numpy:
[12. 12. 12.] float32
sum of paddle:
Tensor(shape=[3], dtype=float32, place=CUDAPlace(0), stop_gradient=True,
       [12., 12., 12.]) paddle.float32
sum of torch:
tensor([12., 12., 12.]) torch.float32
====================
input:
[[1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]] float64
sum of numpy:
[12. 12. 12.] float64
sum of paddle:
Tensor(shape=[3], dtype=float64, place=CUDAPlace(0), stop_gradient=True,
       [12., 12., 12.]) paddle.float64
sum of torch:
tensor([12., 12., 12.], dtype=torch.float64) torch.float64
====================
input:
[[1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]] int32
sum of numpy:
[12 12 12] int64
sum of paddle:
Tensor(shape=[3], dtype=int64, place=CUDAPlace(0), stop_gradient=True,
       [12, 12, 12]) paddle.int64
sum of torch:
tensor([12, 12, 12]) torch.int64
====================
input:
[[1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]
 [1 1 1]] int64
sum of numpy:
[12 12 12] int64
sum of paddle:
Tensor(shape=[3], dtype=int64, place=CUDAPlace(0), stop_gradient=True,
       [12, 12, 12]) paddle.int64
sum of torch:
tensor([12, 12, 12]) torch.int64

[paddle-bot-old]

Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

[sandyhouse]

LGTM

[zhiqiu]

LGTM for core.ops