[PATCH] D92254: [VE] Clean check routines of branch types

[Kazushi Marukawa via Phabricator via llvm-commits]

kaz7 planned changes to this revision.
kaz7 added a comment.

Thank you for reviewing.



================
Comment at: llvm/lib/Target/VE/VEInstrInfo.cpp:97
+// For example, br.l.t and br.l.  We don't use branch relative word/double/
+// float alwasy isntructions.
 static bool isUncondBranchOpcode(int Opc) {
----------------
simoll wrote:
> typo alwasy
Thanks.


================
Comment at: llvm/lib/Target/VE/VEInstrInfo.cpp:102
+#define BRKIND(NAME) (Opc == NAME##a || Opc == NAME##a_nt || Opc == NAME##a_t)
+  assert(!BRKIND(BRCFW) && !BRKIND(BRCFD) && !BRKIND(BRCFS) &&
+         "Branch relative word/double/float always instructions should not be "
----------------
simoll wrote:
> I am lacking context here but is this really the right place for this assertion? Does "should not be used" mean that these instructions are invalid/incorrect or are they just not recommendable/inefficient?
Those instructions are not lowered here from LLVM-IR because we don't use such instructions in our lowering patterns.  Those are acceptable by meaning of HW instructions but not acceptable by meaning of lowering consitency.  And, writing all possible instruction here is definitely not a good idea.  Therefore, I write "should not be used" as invalid.  I'll write such comments in the souce.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D92254/new/

https://reviews.llvm.org/D92254