[PATCH] D105807: [X86] pr51000 in-register struct return tailcalling

[Nathan Sidwell via Phabricator via llvm-commits]

urnathan updated this revision to Diff 368329.
urnathan added a comment.

- Removed pr51000.cpp testcase
- Prevent tailcall on i686 from non-sret to sret
- Removed UB sibcall.ll test

AFAICT by the time we get to this point of code generation, in-register struct return has been flattened to explicit tuple types.  So this is mostly already handled by existing scalar/multiple-value sibcall machinery and testing.

The  StructReturnType enum's RegStructReturn enumerator is never checked -- so that bit of API could be cleaned up to just be a 'is-sret-pointer' boolean.  I also notice that {callIs,argsAre}StructReturn only look at the first result/parm, whereas formal parameter lowering iterates to the first one that has isSret set, and one of the sibcall tests passes a non-initial sret parameter.  That seems inconsistent (more UB?)


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D105807/new/

https://reviews.llvm.org/D105807

Files:
  llvm/lib/Target/X86/X86ISelLowering.cpp
  llvm/lib/Target/X86/X86ISelLowering.h
  llvm/test/CodeGen/X86/sibcall.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D105807.368329.patch
Type: text/x-patch
Size: 6689 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210824/d5628e97/attachment.bin>