[0-size Tensor No.37] Add 0-size Tensor support for index_put

[co63oc]

PR Category

Execute Infrastructure

PR Types

Improvements

Description

index_put 输出维度和输入维度相同，反向 value 参数填充0, x numel为0不填充
torch 测试反向

args = []
args.append(paddle.randn([10, 0]))
args.append([paddle.to_tensor([-4, -9, -7, -3, -8, -8,  0, -1, -6,  2], dtype=paddle.int64)])
args.append(paddle.randn([1, 1]))
args.append(False)
args[2].stop_gradient=False
args[0].stop_gradient=False
result = paddle.index_put(*args)
result.sum().backward()
print(args[2].grad)

Tensor(shape=[1, 1], dtype=float32, place=Place(gpu:0), stop_gradient=False,
       [[0.]])

PaddleAPITest 测试numpy error，其他通过

[paddle-bot]

你的PR提交成功，感谢你对开源项目的贡献!
请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.