use from_blob to avoid memcpy #4118
Conversation
|
Good to see that |
|
Nice, thanks for doing this and handling the deleter! |
|
|
||
| return outTensor; | ||
| auto out_tensor = | ||
| torch::from_blob(jpegBuf, {(long)jpegSize}, ::free, options); |
There was a problem hiding this comment.
This line is raising the following clang-tidy warning in our internal CI:
There's an unchecked dereference of an object that is nullable. Make sure you null-check the object before you call methods on it, pass it as non-nullable parameter, or dereference it. If you are sure the the object cannot be nullptr, make it explicit with CHECK_NOTNULL() or assert(). jpegBuf is nullable.
Looks like jpegBuf is set above in jpeg_mem_dest, but I couldn't find any relevant docs about this function to know what happens if the allocation fails.
There was a problem hiding this comment.
I think if we have errors the setjmp that we do in https://github.com/pytorch/vision/pull/4118/files#diff-dfb505641c632421ae5cc17a4398fe0e3a47c436f62825e88ba2b0aebf497d5eR37 will kick in and the check that jpegBuf != nullptr, so maybe this clang-tidy error is a false-positive.
Still let's keep an eye on this
Summary: Reviewed By: NicolasHug Differential Revision: D29516853 fbshipit-source-id: 88d8c3962204071883317d1b2eb383900d90d430 Co-authored-by: Nicolas Hug <nicolashug@fb.com> Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>
No description provided.