[PATCH] D68766: [NFC][ArgPromo][Tests] Run update_test_checks on all ArgumentPromotion tests

Thu Oct 31 08:55:59 PDT 2019

jdoerfert added a comment.

In D68766#1728260 <https://reviews.llvm.org/D68766#1728260>, @lebedev.ri wrote:

> In D68766#1728230 <https://reviews.llvm.org/D68766#1728230>, @jdoerfert wrote:
>
> > run with ARGPROMOTION,ALL instead of ALL,ARGPROMOTION prefixes
>
>
> Uh, why, that's backwards? It will defeat the point of placing ALL first - then you won't have duplicate check lines for attributor.

I think, determined by browsing the diff, that this way the next diff (with the Attributor) is smaller. If you think I should reverse it again, I can do that as well.

================
Comment at: llvm/test/Transforms/ArgumentPromotion/musttail.ll:45
 define internal i32 @test2(%T* %p, i32 %p2) {
+; CHECK-LABEL: define internal i32 @test2(%T* %p, i32 %p2)
+; CHECK-NEXT:    [[A_GEP:%.*]] = getelementptr [[T:%.*]], %T* [[P:%.*]], i64 0, i32 3
----------------
jdoerfert wrote:
> lebedev.ri wrote:
> > does the script produce these verbose check-lines for define?
> > I think i usually saw it producing just the name.
> I took the original check lines where it made sense but I can also drop these in favor of the auto-generated ones, or I can allow the script to emit the arguments (which I think I preferred). I'll update the tests once I've done that.
The "new" script (see parent revisions) does generate verbose check lines for defines and arguments. The former always the latter on request.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D68766/new/

https://reviews.llvm.org/D68766