[GlobalISel] Add G_ABS computeKnownBits

[r41k0u]

The code is taken from SelectionDAG::computeKnownBits.
This ticks off ABS from #150515

[github-actions]

Thank you for submitting a Pull Request (PR) to the LLVM Project!

This PR will be automatically labeled and the relevant teams will be notified.

If you wish to, you can add reviewers by using the "Reviewers" section on this page.

If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using @ followed by their GitHub username.

If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers.

If you have further questions, they may be answered by the LLVM GitHub User Guide.

You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu
@llvm/pr-subscribers-backend-aarch64

@llvm/pr-subscribers-llvm-globalisel

Author: Pragyansh Chaturvedi (r41k0u)

Changes

The code is taken from SelectionDAG::computeKnownBits.
This ticks off ABS from #150515

---

Full diff: https://github.com/llvm/llvm-project/pull/154413.diff

1 Files Affected:

• (modified) llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp (+8)

diff --git a/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp b/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
index 974fc40de6222..df1b325fa5baf 100644
--- a/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp
@@ -697,6 +697,14 @@ void GISelValueTracking::computeKnownBitsImpl(Register R, KnownBits &Known,
     }
     break;
   }
+  case TargetOpcode::G_ABS: {
+    Register SrcReg = MI.getOperand(1).getReg();
+    computeKnownBitsImpl(SrcReg, Known, DemandedElts, Depth + 1);
+    Known = Known.abs();
+    Known.Zero.setHighBits(computeNumSignBits(SrcReg, DemandedElts, Depth + 1) -
+                           1);
+    break;
+  }
   }
 
   LLVM_DEBUG(dumpResult(MI, Known, Depth));

[r41k0u]

I ran the tests and found no new regressions. I am not sure how to add tests for this (as I saw in other PRs attached to the issue). I came across update_givaluetracking_test_checks.py as the test files in other PRs seem generated. If they are required, please point me towards a test writing guide and I'll add them to this PR as well.

[arsenm]

I ran the tests and found no new regressions. I am not sure how to add tests for this (as I saw in other PRs attached to the issue). I came across update_givaluetracking_test_checks.py as the test files in other PRs seem generated.

The output is generated, the input is not. That's a new style to directly check the output. The older style is to write the simplest code that will optimize or not based on the analysis

If they are required, please point me towards a test writing guide and I'll add them to this PR as well.

https://llvm.org/docs/TestingGuide.html#generating-assertions-in-regression-tests

You're probably best off copy-pasting one of the tests from the recent GIValueTracking commits as a starting point

[r41k0u]

Hi @davemgreen, I've added a couple of tests for this. Please lmk if more of them are needed.
I also goofed up by not creating a new branch for this PR, so once this is approved I'll rebase this against main.

[r41k0u]

I came across another issue while working on this, I will add a fix for that (and I must add a sign extension test case as well).

[r41k0u]

I have added more tests and a fix for the cache assertion. If you think I should move it to GISelValueTrackingPrinterPass::run, something like this:

assert(VTA.ComputeKnownBitsCache.empty() && "Cache should have been cleared");
KnownBits Known = VTA.getKnownBits(Reg);
unsigned SignedBits = VTA.computeNumSignBits(Reg);
ComputeKnownBitsCache.clear();

instead of what I have done right now, please let me know and I'll change it.

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[davemgreen]

Thanks - It looks like there is another one that goes computeNumSignBitsImpl -> getValidMinimumShiftAmount -> getValidShiftAmountRange -> getKnownBits.

I'm wondering if it makes sense to keep the cache, or whether it is doing more harm than good. I tried checking the compile time and it wasn't very useful (although the hit rate was relatively good). I will try and put up a patch that removes it and see what people think.

[r41k0u]

The cache assertion in getKnownBits looks like a temporary thing until the function is fleshed out (although you'd know more about this). And since it is being added to computeNumSignBits as well, I wonder if that function can benefit from the cache somehow as well.

[davemgreen]

I put up #157352 to see what people think of removing it.

[r41k0u]

Thanks. This is much more concise now. I have rebased over main and removed the cache related changes

[davemgreen]

Thank you, LGTM.

The title could do with an update now, otherwise are you happy for this to be submitted?

[r41k0u]

Yes, edited the title.

[davemgreen]

Thanks - it looks like there might be some AMD tests that need to be updated again?

[r41k0u]

My bad, I was only checking on some arches. Did a full build for all tests, and have updated the AMDGPU ones

[davemgreen]

Thanks. LGTM. Looks like it makes a nice improvement.

[github-actions]

@r41k0u Congratulations on having your first Pull Request (PR) merged into the LLVM Project!

Your changes will be combined with recent changes from other authors, then tested by our build bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR.

Please check whether problems have been caused by your change specifically, as the builds can include changes from many authors. It is not uncommon for your change to be included in a build that fails due to someone else's changes, or infrastructure issues.

How to do this, and the rest of the post-merge process, is covered in detail here.

If your change does cause a problem, it may be reverted, or you can revert it yourself. This is a normal part of LLVM development. You can fix your changes and open a new PR to merge them again.

If you don't get any reports, no action is required from you. Your changes are working as expected, well done!