Make sure that the future fragment's storage pointer is properly aligned #39821
Conversation
| return reinterpret_cast<OpaqueValue *>( | ||
| reinterpret_cast<char *>(this) + storageOffset(resultType)); | ||
| reinterpret_cast<char *>(startAddrVal)); |
There was a problem hiding this comment.
Hmm. Normally this would be problematic because of the potential mismatch between an abstract offset calculation and an adjustment on an actual pointer. I guess this particular example might be fine because the overall allocation is known to be aligned at least as well as alignment, so in (mod K) arithmetic we get the same result? Is that true? If so, it's at least worth a comment.
There was a problem hiding this comment.
That is a good question. I had assume when writing this that we would be just calling swift_alloc which would have been ‘max alignment’ aligned. Reading this code: https://github.com/apple/swift/blob/1ab88ed090aadaceb295c22abd503b5d85017fd0/stdlib/public/Concurrency/Task.cpp#L534 I am not sure. There is a call to plain malloc which on Linux I don’t know what it returns. And there is that pointer that we get from some preallocated buffer.
There was a problem hiding this comment.
Hmm, if we're allocating tasks with just malloc, we might need to adjust that. We can do that in a separate PR, though.
|
I'm taking this over as #39829. I'll throw up a 5.5 PR as well. |
No description provided.