Failure to recognise unnecessary sext for add+cmp

[RKSimon]

Pulled out #76524 - we fail to recognize that extension is unnecessary and that the trunc is nsw

define i8 @src(i16 noundef %x) {
entry:
  %conv = sext i16 noundef %x to i128
  %add = add i128 %conv, 128
  %or.cond.i = icmp ult i128 %add, 256
  assume i1 %or.cond.i
  %conv1 = trunc i16 noundef %x to i8
  ret i8 %conv1
}
=>
define i8 @tgt(i16 noundef %x) {
entry:
  %add = add i16 noundef %x, 128
  %or.cond.i = icmp ult i16 %add, 256
  assume i1 %or.cond.i
  %conv1 = trunc nsw i16 noundef %x to i8
  ret i8 %conv1
}
Transformation seems to be correct!

https://alive2.llvm.org/ce/z/epn8hd

[dtcxzyw]

It should be handled by SCCP(b1822ef). But currently it fails to infer nsw :(

[RKSimon]

This is just the sext+add+cmp -> add+cmp missing optimization:

----------------------------------------
define i1 @src(i16 noundef %x) {
#0:
  %conv = sext i16 noundef %x to i128
  %add = add i128 %conv, 128
  %or.cond.i = icmp ult i128 %add, 256
  ret i1 %or.cond.i
}
=>
define i1 @tgt(i16 noundef %x) {
#0:
  %add = add i16 noundef %x, 128
  %or.cond.i = icmp ult i16 %add, 256
  ret i1 %or.cond.i
}
Transformation seems to be correct!

[dtcxzyw]

This is just the sext+add+cmp -> add+cmp missing optimization:

---

define i1 @src(i16 noundef %x) {
#0:
%conv = sext i16 noundef %x to i128
%add = add i128 %conv, 128
%or.cond.i = icmp ult i128 %add, 256
ret i1 %or.cond.i
}
=>
define i1 @tgt(i16 noundef %x) {
#0:
%add = add i16 noundef %x, 128
%or.cond.i = icmp ult i16 %add, 256
ret i1 %or.cond.i
}
Transformation seems to be correct!

I see.
BTW, is it necessary for #76524 to infer trunc nsw after this optimization? If so, can you file another issue to track the nsw flag inference problem? I think it can be inferred in CVP as LVI is able to provide context-sensitive range information.

[RKSimon]

This is just the sext+add+cmp -> add+cmp missing optimization:

define i1 @src(i16 noundef %x) {
#0:
%conv = sext i16 noundef %x to i128
%add = add i128 %conv, 128
%or.cond.i = icmp ult i128 %add, 256
ret i1 %or.cond.i
}
=>
define i1 @tgt(i16 noundef %x) {
#0:
%add = add i16 noundef %x, 128
%or.cond.i = icmp ult i16 %add, 256
ret i1 %or.cond.i
}
Transformation seems to be correct!

I see. BTW, is it necessary for #76524 to infer trunc nsw after this optimization? If so, can you file another issue to track the nsw flag inference problem? I think it can be inferred in CVP as LVI is able to provide context-sensitive range information.

It looks like the final x86 assembly is good after #129363 - its just the IR fed to the backend still isn't ideal.

[dtcxzyw]

Generalized pattern: https://alive2.llvm.org/ce/z/JjLNvU

define i1 @src(i8 %x, i8 %y, i8 %z) {
entry:

  %conv = sext i8 %x to i32
  %sy = sext i8 %y to i32
  %sz = sext i8 %z to i32

  %diff = sub i32 %sy, %sz
  %cond = icmp ule i32 %diff, 128 ; INT_MIN
  call void @llvm.assume(i1 %cond)

  %add = add i32 %conv, %sy
  %cmp = icmp ult i32 %add, %sz
  ret i1 %cmp
}

define i1 @tgt(i8 %x, i8 %y, i8 %z) {
entry:
  %add = add i8 %x, %y
  %cmp = icmp ult i8 %add, %z
  ret i1 %cmp
}

[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)

Pulled out #76524 - we fail to recognize that extension is unnecessary and that the trunc is nsw

define i8 @<!-- -->src(i16 noundef %x) {
entry:
  %conv = sext i16 noundef %x to i128
  %add = add i128 %conv, 128
  %or.cond.i = icmp ult i128 %add, 256
  assume i1 %or.cond.i
  %conv1 = trunc i16 noundef %x to i8
  ret i8 %conv1
}
=&gt;
define i8 @<!-- -->tgt(i16 noundef %x) {
entry:
  %add = add i16 noundef %x, 128
  %or.cond.i = icmp ult i16 %add, 256
  assume i1 %or.cond.i
  %conv1 = trunc nsw i16 noundef %x to i8
  ret i8 %conv1
}
Transformation seems to be correct!

https://alive2.llvm.org/ce/z/epn8hd

[AbdAlRahmanGad]

Hi, Could you please assign this issue to me?

[AbdAlRahmanGad]

Hi @dtcxzyw @RKSimon ,

Could you please point me to code segment that I should change in order to fix this, and also the steps to reproduce?