Fix buffer overrun bug in CPU upsample op #1501
Conversation
| @@ -342,7 +342,7 @@ Status Upsample<T>::BaseCompute(OpKernelContext* context, const std::vector<floa | |||
| Tensor* Y = context->Output(0, Y_dims); | |||
|
|
|||
| if (no_scale) { | |||
| memcpy(Y->MutableDataRaw(), X->DataRaw(), Y->Size() * sizeof(T)); | |||
| memcpy(Y->MutableDataRaw(), X->DataRaw(), Y->Size()); | |||
There was a problem hiding this comment.
Should we rename Tensor::Size to Tensor::SizeInBytes in a separate PR?
There was a problem hiding this comment.
Should we rename Tensor::Size to Tensor::SizeInBytes in a separate PR?
It comes with a cost. You know, it's a source code level break change. Many partners are using this class.
There was a problem hiding this comment.
Thanks for the fixing. Didn't know Size() returns byte size.
There was a problem hiding this comment.
I'd prefer making the names more explicit at the cost of possibly breaking some ongoing code. Otherwise, someone else would make the same mistake later and it's more expensive to debug and fix then.
There was a problem hiding this comment.
Agree. Let me make the change in a separate PR.
Description:
Fix buffer overrun bug in CPU upsample op. Introduced in #1484
Motivation and Context
Why is this change required? What problem does it solve?
A buffer overrun bug may cause application crash. (wrote more bytes than the buffer has)
If it fixes an open issue, please link to the issue here.