[PATCH] D13455: [safestack] Fast access to the unsafe stack pointer on AArch64/Android.

[Evgeniy Stepanov via llvm-commits]

eugenis added a comment.

Moving all of this into SafeStackPass would work, and would be much simpler, but it looks like llvm generally tries to avoid target-specific code in IR passes. I'm not really opposed to that, just want to see what people think.

Another option would be to implement x86.thread.pointer intrinsic (tied to __builtin_thread_pointer) and use that in safe stack pass. It would have to include an optimization that wraps x86.thread.pointer + offset into a single %gs:offset load (straightforward codegen would do *(%gs:0)+offset).


Repository:
  rL LLVM

http://reviews.llvm.org/D13455