missed fold, failed to kill redundant shufflevector

[zhengyang92]

https://godbolt.org/z/T4sTh8W87
https://alive2.llvm.org/ce/z/fh2VmQ

define <2 x float> @src(<2 x float> %0, i8 %1) {
if.else433:
  %2 = shufflevector <2 x float> %0, <2 x float> poison, <2 x i32> <i32 1, i32 poison>
  %3 = fneg <2 x float> %0
  %4 = shufflevector <2 x float> %2, <2 x float> %3, <2 x i32> <i32 0, i32 3>
  ret <2 x float> %4
}

define <2 x float> @tgt(<2 x float> %0, i8 %1) {
if.else433:
  %2 = fneg <2 x float> %0
  %sv = shufflevector <2 x float> %0, <2 x float> %2, <2 x i32> <i32 1, i32 3>
  ret <2 x float> %sv
}

@dtcxzyw @regehr @RKSimon

[dtcxzyw]

Similar issue: #84645

It may be a SLP issue. @zhengyang92 Could you please provide the IR before SLP?

cc @alexey-bataev

[RKSimon]

Instcombine already does something similar for binops in InstCombinerImpl::foldVectorBinop, so we probably need to try to add an InstCombinerImpl::foldVectorUnaryop equivalent

[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. In the comments of the issue, request for it to be assigned to you.
2. Fix the issue locally.
3. 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.
4. Create a Git commit.
5. Run git clang-format HEAD~1 to format your changes.
6. Open a pull request to the upstream repository on GitHub. Detailed instructions can be found in GitHub's documentation.

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: Zhengyang Liu (zhengyang92)

https://godbolt.org/z/T4sTh8W87 https://alive2.llvm.org/ce/z/fh2VmQ

define &lt;2 x float&gt; @<!-- -->src(&lt;2 x float&gt; %0, i8 %1) {
if.else433:
  %2 = shufflevector &lt;2 x float&gt; %0, &lt;2 x float&gt; poison, &lt;2 x i32&gt; &lt;i32 1, i32 poison&gt;
  %3 = fneg &lt;2 x float&gt; %0
  %4 = shufflevector &lt;2 x float&gt; %2, &lt;2 x float&gt; %3, &lt;2 x i32&gt; &lt;i32 0, i32 3&gt;
  ret &lt;2 x float&gt; %4
}

define &lt;2 x float&gt; @<!-- -->tgt(&lt;2 x float&gt; %0, i8 %1) {
if.else433:
  %2 = fneg &lt;2 x float&gt; %0
  %sv = shufflevector &lt;2 x float&gt; %0, &lt;2 x float&gt; %2, &lt;2 x i32&gt; &lt;i32 1, i32 3&gt;
  ret &lt;2 x float&gt; %sv
}

@dtcxzyw @regehr @RKSimon

[aniplcc]

Hi, I'm new to instcombine. Am I right for thinking this implies the following:

1. implement foldVectorUnaryop(UnaryOperator &Inst).
2. in /llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp, for visitFNeg, add the new foldVectorUnaryop( call.
   I'm interested in implementing this.

[dtcxzyw]

%2 = shufflevector <2 x float> %0, <2 x float> poison, <2 x i32> <i32 1, i32 poison>
%4 = shufflevector <2 x float> %2, <2 x float> %3, <2 x i32> <i32 0, i32 3>
%sv = shufflevector <2 x float> %0, <2 x float> %2, <2 x i32> <i32 1, i32 3>

@RKSimon Is it safe to change the mask?

[dbelow02]

Hey there! I'm new to llvm and I'd like for this task to be assigned to me.

[zxc12523]

@RKSimon Could you explain more on this topic? I don't understand the relation of this optimization with InstCombinerImpl::foldVectorUnaryop, or does this optimization related to

llvm-project/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp

Line 2454 in ed1cfff

static Instruction *foldShuffleOfUnaryOps(ShuffleVectorInst &Shuf,

[RKSimon]

Sorry - you're better off with somebody who understands all the InstCombine folds, there might be something already present that almost does this that I'm not aware of

[madhur13490]

Is anybody still looking at this?