Ping Zhu

oneflow.where dtype check error

4

## Code to reproduce bug ``` import oneflow as flow x = flow.randn(5, 5) y = flow.where(x > 0, x, 0.0) ``` Output: ``` RuntimeError: Check failed: (x_dtype) == (GetDataType::value)...

register reduce_sum_like backward function and add testcase

2

- 问题来源： https://github.com/Oneflow-Inc/oneflow/issues/8831

Add double grad for broadcast_matmul_grad_b op

5

- 添加 broadcast_matmul_grad_b 二阶导算子 - 矩阵乘法 a * b - 对 b 的梯度计算由 broadcast_matmul_grad_b 计算，需要另外实现二阶导算子， - 对 a 的梯度计算由 matmul 计算，已闭包，不需要另外实现二阶导算子（broadcast 时还需要 reduce_sum_like 的一阶导 https://github.com/Oneflow-Inc/oneflow/issues/8831 ）

- 添加 div 二阶导算子 - broadcast 除法需要 reduce_sum_like 可导，部分修改来自 PR https://github.com/Oneflow-Inc/oneflow/pull/8895

## Summary Op `reduce_sum_like` miss backward function ## Code to reproduce bug ``` import oneflow as flow a = flow.randn(4, 2, 3, requires_grad=True) b = flow.randn(2, 3, requires_grad=True) c =...

## Summary `broadcast_like` op throw std::bad_alloc when x.dim > like.dim ## Code to reproduce bug ``` import oneflow as flow a = flow.randn(1, 1) b = flow.randn(2, 2, 2) c...

[Code ](https://github.com/BBuf/how-to-optim-algorithm-in-cuda/commit/90e431347c22339df1c614eb69deebf8dce264db) fixed [README](https://github.com/BBuf/how-to-optim-algorithm-in-cuda/blob/master/reduce/README.md) not fix