【 Paddle Tensor 规范化第二期 】解决隐式广播不能够处理0-size的问题

[aquagull]

PR Category

User Experience

PR Types

Bug fixes

Description

先前的隐式广播是不能够处理输入存在维度为0的情况：

修改后:

测试代码

import paddle
import pdb
def test_paddle_add_broadcast(shape1, shape2):
    x = paddle.randn(shape1, dtype='float32') if shape1 != [0] and 0 not in shape1 else paddle.to_tensor([], dtype='float32').reshape(shape1)
    y = paddle.randn(shape2, dtype='float32') if shape2 != [0] and 0 not in shape2 else paddle.to_tensor([], dtype='float32').reshape(shape2)

    try:
        # pdb.set_trace()
        result = x + y
        print(f"Paddle Add: Shapes: {shape1} + {shape2}, Result shape: {result.shape}")

    except Exception as e:
        print(f"Paddle Add/Pow Add Error: Shapes: {shape1} + {shape2}, Error: {e}")

shapes = [
    ([0], [0]),
    ([0, 0], [0]),
    ([0], [0, 0]),
    ([0, 3], [3]),
    ([3], [0, 3]),
    ([0, 3], [0, 3]),
    ([2, 0], [0]),
    ([0], [2, 0]),
]


for shape1, shape2 in shapes:
    test_paddle_add_broadcast(shape1, shape2)

[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.

[MrXnneHang]

mark,感觉很有意思.

[HydrogenSulfate]

添加一个关于0-size的广播单测，x、y的形状包含0-size、0-Dim、非0-size，长度可以在0~7，与np.broadcast_shapes的结果进行对比即可。

[gongshaotian]

LGTM for BroadcastKernel

[HydrogenSulfate]

LGTM

[gongshaotian]

初步看 GetBroadcastDimsArray 的修改可能会对下列算子产生影响 MatrixRank、Atan2、CompareRaw、Complex、SendUERecv、SendUV、BroadcastOp，单测CI没有报错，有可能是因为这些算子单测没有覆盖0 size的情况

[aquagull]

初步看 GetBroadcastDimsArray 的修改可能会对下列算子产生影响 MatrixRank、Atan2、CompareRaw、Complex、SendUERecv、SendUV、BroadcastOp，单测CI没有报错，有可能是因为这些算子单测没有覆盖0 size的情况

确实是如此🤔，后续需要检查这些算子吗，还是说会放到添加单测的那个开源任务去