[AArch64] can use XAR for vector rotates where possible

[k-arrows]

Here is the code from gcc testsuite.
https://godbolt.org/z/fejTchexj

typedef char __attribute__ ((vector_size (16))) v16qi;
typedef unsigned short __attribute__ ((vector_size (16))) v8hi;
typedef unsigned int __attribute__ ((vector_size (16))) v4si;
typedef unsigned long long __attribute__ ((vector_size (16))) v2di;
typedef char __attribute__ ((vector_size (8))) v8qi;
typedef unsigned short __attribute__ ((vector_size (8))) v4hi;
typedef unsigned int __attribute__ ((vector_size (8))) v2si;

v2di
G1 (v2di r) {
    return (r >> 39) | (r << 25);
}

v4si
G2 (v4si r) {
    return (r >> 23) | (r << 9);
}

v8hi
G3 (v8hi r) {
    return (r >> 5) | (r << 11);
}

v16qi
G4 (v16qi r)
{
  return (r << 2) | (r >> 6);
}

v2si
G5 (v2si r) {
    return (r >> 22) | (r << 10);
}

v4hi
G6 (v4hi r) {
    return (r >> 7) | (r << 9);
}

v8qi
G7 (v8qi r)
{
  return (r << 3) | (r >> 5);
}

[llvmbot]

@llvm/issue-subscribers-backend-aarch64

Author: None (k-arrows)

Here is the code from gcc testsuite. https://godbolt.org/z/fejTchexj ```c typedef char __attribute__ ((vector_size (16))) v16qi; typedef unsigned short __attribute__ ((vector_size (16))) v8hi; typedef unsigned int __attribute__ ((vector_size (16))) v4si; typedef unsigned long long __attribute__ ((vector_size (16))) v2di; typedef char __attribute__ ((vector_size (8))) v8qi; typedef unsigned short __attribute__ ((vector_size (8))) v4hi; typedef unsigned int __attribute__ ((vector_size (8))) v2si;

v2di
G1 (v2di r) {
return (r >> 39) | (r << 25);
}

v4si
G2 (v4si r) {
return (r >> 23) | (r << 9);
}

v8hi
G3 (v8hi r) {
return (r >> 5) | (r << 11);
}

v16qi
G4 (v16qi r)
{
return (r << 2) | (r >> 6);
}

v2si
G5 (v2si r) {
return (r >> 22) | (r << 10);
}

v4hi
G6 (v4hi r) {
return (r >> 7) | (r << 9);
}

v8qi
G7 (v8qi r)
{
return (r << 3) | (r >> 5);
}

</details>

[davemgreen]

AArch64DAGToDAGISel::trySelectXAR could be updated where the xor is not present, generating a mov 0 instead.

[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: None (k-arrows)

Here is the code from gcc testsuite. https://godbolt.org/z/fejTchexj ```c typedef char __attribute__ ((vector_size (16))) v16qi; typedef unsigned short __attribute__ ((vector_size (16))) v8hi; typedef unsigned int __attribute__ ((vector_size (16))) v4si; typedef unsigned long long __attribute__ ((vector_size (16))) v2di; typedef char __attribute__ ((vector_size (8))) v8qi; typedef unsigned short __attribute__ ((vector_size (8))) v4hi; typedef unsigned int __attribute__ ((vector_size (8))) v2si;

v2di
G1 (v2di r) {
return (r >> 39) | (r << 25);
}

v4si
G2 (v4si r) {
return (r >> 23) | (r << 9);
}

v8hi
G3 (v8hi r) {
return (r >> 5) | (r << 11);
}

v16qi
G4 (v16qi r)
{
return (r << 2) | (r >> 6);
}

v2si
G5 (v2si r) {
return (r >> 22) | (r << 10);
}

v4hi
G6 (v4hi r) {
return (r >> 7) | (r << 9);
}

v8qi
G7 (v8qi r)
{
return (r << 3) | (r >> 5);
}

</details>

[gxyd]

Hi, I'm new to LLVM community, do you think if I can try fixing this issue? ✋

[davemgreen]

Sure thing - let us know if you run into any issues.

[Rajveer100]

@davemgreen
Consider this example:

$$X = 0 0 1 0 0 0 1 0 0 1$$ $$Y = 0 0 1 1 1 1 0 0 0 1$$ $$V = XOR(X, Y) = 0 0 0 1 1 1 1 0 0 0$$ $$BITS = 10$$ $$IMM = 7$$ $$ROTR(V, IMM) = 1 1 1 1 0 0 0 0 0 0$$ $$N0 = SHL(V, BITS-IMM) = 1 1 1 1 0 0 0 0 0 0$$ $$N1 = SRL(V, IMM) = 0 0 0 0 0 0 0 0 0 0$$ $$OR(N0, N1) = 1 1 1 1 0 0 0 0 0 0$$

[davemgreen]

Can you explain what you mean?

[Rajveer100]

I was facing an assertion error so was just making sure I understood the transformation correctly.

I presume the OR result should be equal to the ROTR result?

More info in the changes I made.

[gxyd]

I'm sorry, I was a little busy with my work priorities, seems like someone has already sent a patch fixing the issue?

[gxyd]

@davemgreen , kindly let me know if there is a guideline about a timeline within which I've to complete an issue, otherwise the issue can be taken up by someone else in llvm project, even though the issue is assigned to me?

Also, let me know if I now should be looking at some other issue instead maybe?

[davemgreen]

@gxyd Sorry, I'm not sure I'm afraid. The only details I found were the brief details in https://www.llvm.org/docs/Contributing.html#bug-fixes. As you might imagine, we get some people sign up for first time issues only to disappear into the void never to be heard from again. The usual process is that if someone else is interested they ping the issue and checks if it is still being working on, or if the person has become busy with other things. If you want something else to look into, perhaps #119921 is a good one that requires some new tablegen patterns.

@Rajveer100 I wasn't sure what the output meant exactly, perhaps explain more about what you are trying to prove if you still need to. I'm pretty sure the transform is OK. It looks like your patch is a good start, I'll comment there.

[gxyd]

@gxyd Sorry, I'm not sure I'm afraid. The only details I found were the brief details in https://www.llvm.org/docs/Contributing.html#bug-fixes. As you might imagine, we get some people sign up for first time issues only to disappear into the void never to be heard from again. The usual process is that if someone else is interested they ping the issue and checks if it is still being working on, or if the person has become busy with other things. If you want something else to look into, perhaps #119921 is a good one that requires some new tablegen patterns.

Understood, really appreciate the response. Then I'll un-assign myself from this issue, and once I've more bandwidth I'll work more actively.

[davemgreen]

I created #139229 as a followup if you are interested in making more of the fixed-width types work too. Thanks