[Bug]: RuntimeError: A leaf tensor that requires grad is being used in an inplace operator, InplaceIndexPut, which is forbidden!

### Checklist - [x] 1. 我已经搜索过相关问题，但没有得到预期的帮助. (https://gitee.com/mindspore/mindspore/issues) - [x] 2. 最新版本中该错误尚未修复. - [x] 3. 请注意，如果您提交的Bug描述缺少相应的环境信息和最小可复现的demo，我们将很难复现和解决该问题，从而降低收到反馈的可能性，甚至该问题将被关闭. ### 🐞 问题详细描述 对x求导，但是校验了y的梯度计算，具体代码： ```python import mindspore as ms from mindspore import mint def test(x, y, idx): y[idx] = 0 return (x * y).mean() x = mint.ones((2, 3)) y = mint.ones((2, 3)) idx = ms.Tensor([[0, 1], [1, 1]]) grad_func = ms.value_and_grad(test, (0,)) z, grad = grad_func(x, y, idx) print(z, grad.shape) ``` 报错： ``` Traceback (most recent call last): File "/cache/zhaoting/test_rcp/test_inplace.py", line 15, in <module> z, grad = grad_func(x, y, idx) File "/home/ma-user/miniconda3/envs/zt/lib/python3.9/site-packages/mindspore/ops/composite/base.py", line 634, in after_grad return grad_(fn_, weights, grad_position)(*args, **kwargs) File "/home/ma-user/miniconda3/envs/zt/lib/python3.9/site-packages/mindspore/common/api.py", line 188, in wrapper results = fn(*arg, **kwargs) File "/home/ma-user/miniconda3/envs/zt/lib/python3.9/site-packages/mindspore/ops/composite/base.py", line 610, in after_grad run_args, res = self._pynative_forward_run(fn, grad_, weights, *args, **kwargs) File "/home/ma-user/miniconda3/envs/zt/lib/python3.9/site-packages/mindspore/ops/composite/base.py", line 672, in _pynative_forward_run _pynative_executor.end_graph(fn, outputs, *args, **kwargs) File "/home/ma-user/miniconda3/envs/zt/lib/python3.9/site-packages/mindspore/common/api.py", line 1731, in end_graph self._executor.end_graph(obj, output, *args, *(kwargs.values())) RuntimeError: A leaf tensor that requires grad is being used in an inplace operator, InplaceIndexPut, which is forbidden! ``` ### 详细的环境信息描述 MindSpore master CANN 8.0.0 ### 其他辅助信息 相同代码torch没有报错： ```python import torch def test(x, y, idx): y[idx] = 0 return (x * y).mean() x = torch.ones((2, 3)) y = torch.ones((2, 3)) idx = torch.tensor([[0, 1], [1, 1]]) x.requires_grad…