Weikai Tang

2_autograd_tutorial.ipynb 文件里，梯度部分 原文：反向传播 因为 out是一个纯量（scalar），out.backward() 等于out.backward(torch.tensor(1))。 运行时out.backward(torch.tensor(1))报错：RuntimeError: Expected is Floating Type (grads[i].type().scalarType()) to be true, but got false. 同时，scalar的翻译个人认为最好是标量。 改为：反向传播 因为 out是一个标量（scalar），out.backward() 等于out.backward(torch.tensor(1.))。

书中 10.4.3.2. 算法优化章节，提到“在前述5.3.3小节中，已经详细介绍了矩阵乘GEMM运算的优化”。请问该部分内容在哪里，谢谢！ 

Question about GCN/GIN forward propagation

4

Hi Yuke，I read your paper and code, of course excellent work and thanks for the repository. I have question about the cuda kernel. I can see you computing GCN forward...

Solve bank conflict

1

In my opinion, when loading data from global memory to shared memory(i.e. write shared memory) with vectorized access, because of the transposition, threads within a warp may write the same...