This is an archive of the discontinued LLVM Phabricator instance.

[ADCE] Fix non-deterministic behaviour due to iterating over a pointer set
ClosedPublic

Authored by aemerson on Jul 16 2019, 12:42 AM.

Details

Summary

The following IR run through opt -adce 50 times will show a difference, but writing a test for this particular kind of behavior isn't possible AFAIK.

Original patch by Yann Laigle-Chapuy

Diff Detail

Repository
rL LLVM

Event Timeline

aemerson created this revision.Jul 16 2019, 12:42 AM
Herald added a project: Restricted Project. · View Herald TranscriptJul 16 2019, 12:42 AM

Test IR:

target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-apple-macosx10.10.0"

; Function Attrs: noinline nounwind ssp uwtable
define void @ham() local_unnamed_addr #0 {
bb:
  br i1 false, label %bb1, label %bb22

bb1:                                              ; preds = %bb
  br i1 undef, label %bb2, label %bb20

bb2:                                              ; preds = %bb1
  br label %bb5

bb5:                                              ; preds = %bb16, %bb2
  br i1 undef, label %bb6, label %bb17

bb6:                                              ; preds = %bb5
  br i1 undef, label %bb7, label %bb16

bb7:                                              ; preds = %bb6
  br i1 undef, label %bb9, label %bb8

bb8:                                              ; preds = %bb7
  br i1 undef, label %bb9, label %bb10

bb9:                                              ; preds = %bb8, %bb7
  br label %bb13

bb10:                                             ; preds = %bb8
  br label %bb12

bb12:                                             ; preds = %bb10
  br label %bb13

bb13:                                             ; preds = %bb12, %bb9
  br label %bb14

bb14:                                             ; preds = %bb13
  br label %bb15

bb15:                                             ; preds = %bb14
  br label %bb16

bb16:                                             ; preds = %bb15, %bb6
  br label %bb5

bb17:                                             ; preds = %bb5
  br label %bb19

bb19:                                             ; preds = %bb17
  br label %bb21

bb20:                                             ; preds = %bb1
  br label %bb21

bb21:                                             ; preds = %bb20, %bb19
  br label %bb22

bb22:                                             ; preds = %bb21, %bb
  ret void
}

attributes #0 = { noinline nounwind ssp uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="core2" "target-features"="+cx16,+fxsr,+mmx,+sse,+sse2,+sse3,+ssse3,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }

!llvm.module.flags = !{!0}

!0 = !{i32 7, !"PIC Level", i32 2}
fhahn accepted this revision.Jul 16 2019, 3:54 AM

LGTM, thanks. We iterate over BlocksWithDeadTerminators later on, which can cause edges to BBs to be removed a different order.

I think you should add the test IR test, just pass and check -preserve-ll-uselistorder. There should be an upstream bot setting LLVM_REVERSE_ITERATION, which should result in a different use-list ordering without this patch.

This revision is now accepted and ready to land.Jul 16 2019, 3:54 AM
This revision was automatically updated to reflect the committed changes.