[all-commits] [llvm/llvm-project] f5211d: [DebugInfo][GVNSink] Fix #77415: GVNSink fails to ...

[Shan Huang via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: f5211d79b9edf56c08143491ccde38d480b40ab8
      https://github.com/llvm/llvm-project/commit/f5211d79b9edf56c08143491ccde38d480b40ab8
  Author: Shan Huang <52285902006 at stu.ecnu.edu.cn>
  Date:   2024-05-20 (Mon, 20 May 2024)

  Changed paths:
    M llvm/lib/Transforms/Scalar/GVNSink.cpp
    A llvm/test/Transforms/GVNSink/sink-ignore-dbg-intrinsics.ll

  Log Message:
  -----------
  [DebugInfo][GVNSink] Fix #77415: GVNSink fails to optimize LLVM IR with debug info (#77602)

This PR fixes issue #77415 and is revised from PR #77419 .

PR #77419 breaks the newly added test in the same PR on windows, because
GVNSink is non-deterministic when sorting `BasicBlock*` pointers. This
is reflected in the failure report.

```
# | C:\src\llvm-project\llvm\test\Transforms\GVNSink\sink-ignore-dbg-intrinsics.ll:28:10: error: CHECK: expected string not found in input
# | ; CHECK: %a.sink = phi i32 [ %a, %if.then ], [ %b, %if.else ]
# |          ^
# | <stdin>:24:8: note: scanning from here
# | if.end: ; preds = %if.else, %if.then
# |        ^
# | <stdin>:25:2: note: possible intended match here
# |  %b.sink = phi i32 [ %b, %if.else ], [ %a, %if.then ]
# |  ^
# | 
# | Input file: <stdin>
# | Check file: C:\src\llvm-project\llvm\test\Transforms\GVNSink\sink-ignore-dbg-intrinsics.ll
```

According to the report, what the CheckFile wants to match is the
`%a.sink`, however there is `%b.sink`. But this mismatch does not mean
that this commit is wrong, since the occurrence of either `%a.sink` or
`%b.sink` is correct. The root cause of this test failure is the strict
check rule in the regression test committed.

So I refined the regression test with a more general check rule to only
detect whether there is an instruction with suffix `.sink` in the
`if.end` block. Hope this won't fail the test. If this PR still fails to
build, I will close this PR and try to find another right way to fix
this.



To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications