[InstCombine] Failure to recognise a split i128 extension

[RKSimon]

Split off from #76524

https://alive2.llvm.org/ce/z/4SwMjF

----------------------------------------
define i128 @src(i32 noundef %x) {
  %coerce.sroa.0.0.extract.trunc = sext i32 noundef %x to i64
  %ashr = ashr i32 noundef %x, 31
  %coerce.sroa.2.0.extract.trunc = sext i32 %ashr to i64
  %x.sroa.2.0.insert.ext.i = zext i64 %coerce.sroa.2.0.extract.trunc to i128
  %x.sroa.2.0.insert.shift.i = shl nuw i128 %x.sroa.2.0.insert.ext.i, 64
  %x.sroa.0.0.insert.ext.i = zext i64 %coerce.sroa.0.0.extract.trunc to i128
  %x.sroa.0.0.insert.insert.i = or disjoint i128 %x.sroa.2.0.insert.shift.i, %x.sroa.0.0.insert.ext.i
  ret i128 %x.sroa.0.0.insert.insert.i
}
=>
define i128 @tgt(i32 noundef %x) {
  %xx = sext i32 noundef %x to i128
  ret i128 %xx
}
Transformation seems to be correct!

[RKSimon]

https://gcc.godbolt.org/z/Kd3znEovE

inline __int128 foo(__int128 i) {
    return i;
}
void bar(int x, __int128 *out) {
    *out = foo(x);
}

define dso_local void @bar(int, __int128*)(i32 noundef %x, ptr noundef writeonly captures(none) initializes((0, 16)) %out) local_unnamed_addr {
entry:
  %coerce.sroa.0.0.extract.trunc = sext i32 %x to i64
  %0 = ashr i32 %x, 31
  %coerce.sroa.2.0.extract.trunc = sext i32 %0 to i64
  %coerce1.sroa.2.0.insert.ext = zext i64 %coerce.sroa.2.0.extract.trunc to i128
  %coerce1.sroa.2.0.insert.shift = shl nuw i128 %coerce1.sroa.2.0.insert.ext, 64
  %coerce1.sroa.0.0.insert.ext = zext i64 %coerce.sroa.0.0.extract.trunc to i128
  %coerce1.sroa.0.0.insert.insert = or disjoint i128 %coerce1.sroa.2.0.insert.shift, %coerce1.sroa.0.0.insert.ext
  store i128 %coerce1.sroa.0.0.insert.insert, ptr %out, align 16
  ret void
}

[llvmbot]

Hi!

This issue may be a good introductory issue for people new to working on LLVM. If you would like to work on this issue, your first steps are:

1. Check that no other contributor has already been assigned to this issue. If you believe that no one is actually working on it despite an assignment, ping the person. After one week without a response, the assignee may be changed.
2. In the comments of this issue, request for it to be assigned to you, or just create a pull request after following the steps below. Mention this issue in the description of the pull request.
3. Fix the issue locally.
4. Run the test suite locally. Remember that the subdirectories under test/ create fine-grained testing targets, so you can e.g. use make check-clang-ast to only run Clang's AST tests.
5. Create a Git commit.
6. Run git clang-format HEAD~1 to format your changes.
7. Open a pull request to the upstream repository on GitHub. Detailed instructions can be found in GitHub's documentation. Mention this issue in the description of the pull request.

If you have any further questions about this issue, don't hesitate to ask via a comment in the thread below.

[llvmbot]

@llvm/issue-subscribers-good-first-issue

Author: Simon Pilgrim (RKSimon)

Split off from #76524

https://alive2.llvm.org/ce/z/4SwMjF

----------------------------------------
define i128 @<!-- -->src(i32 noundef %x) {
  %coerce.sroa.0.0.extract.trunc = sext i32 noundef %x to i64
  %ashr = ashr i32 noundef %x, 31
  %coerce.sroa.2.0.extract.trunc = sext i32 %ashr to i64
  %x.sroa.2.0.insert.ext.i = zext i64 %coerce.sroa.2.0.extract.trunc to i128
  %x.sroa.2.0.insert.shift.i = shl nuw i128 %x.sroa.2.0.insert.ext.i, 64
  %x.sroa.0.0.insert.ext.i = zext i64 %coerce.sroa.0.0.extract.trunc to i128
  %x.sroa.0.0.insert.insert.i = or disjoint i128 %x.sroa.2.0.insert.shift.i, %x.sroa.0.0.insert.ext.i
  ret i128 %x.sroa.0.0.insert.insert.i
}
=&gt;
define i128 @<!-- -->tgt(i32 noundef %x) {
  %xx = sext i32 noundef %x to i128
  ret i128 %xx
}
Transformation seems to be correct!

[bassiounix]

I want to try to solve this.

[dtcxzyw]

Please read https://llvm.org/docs/InstCombineContributorGuide.html before submitting a patch.
Good luck and have fun!

[RKSimon]

Thanks @bassiounix - the or(zext,shl(zext)) pattern is already handled inside matchOrConcat so you might want to look at adding patterns that match the sext/ashr sequence in there.

[RKSimon]

@bassiounix have you been able to take a look at this?

[bassiounix]

Yes, I'm working on it atm.
I'm still figuring out whether I would use the builder for a separate pattern for ashr/sext or combine it with the current pattern at matchOrConcat

[RKSimon]

If it helps, matchOrConcat shouldn't need to return an Instruction* - a Value* should be fine, which should simplify which builder you can use.

[bassiounix]

I've tried to rebuild the value of the UpperSrc with Builder.CreateSExt but did nothing to optimization tool
I tried to use LLVM_DEBUG and passed -debug to opt when trying the provided example and nothing is printed.

is this code even reachable in this case?
or is rebuilding UpperSrc with Builder to match the pattern the wrong way to solve it?

any hints?