[LoopPeel] Check for onlyAccessesInaccessibleMemory instead of llvm.assume in peelToTurnInvariantLoadsDereferenceable.

[topperc]

onlyAccessesInaccessibleMemory can't alias with a load. This allows us to ignore more intrinsics than llvm.assume.

Follow up from #171547

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

Author: Craig Topper (topperc)

Changes

onlyAccessesInaccessibleMemory can't alias with a load. This allows us to ignore more intrinsics than llvm.assume.

Follow up from #171547

---

Full diff: https://github.com/llvm/llvm-project/pull/171910.diff

1 Files Affected:

• (modified) llvm/lib/Transforms/Utils/LoopPeel.cpp (+3-3)

diff --git a/llvm/lib/Transforms/Utils/LoopPeel.cpp b/llvm/lib/Transforms/Utils/LoopPeel.cpp
index 635b90020f625..960ec9d4c7d6e 100644
--- a/llvm/lib/Transforms/Utils/LoopPeel.cpp
+++ b/llvm/lib/Transforms/Utils/LoopPeel.cpp
@@ -447,10 +447,10 @@ static unsigned peelToTurnInvariantLoadsDereferenceable(Loop &L,
   const DataLayout &DL = L.getHeader()->getDataLayout();
   for (BasicBlock *BB : L.blocks()) {
     for (Instruction &I : *BB) {
-      // Don't consider llvm.assume as writing to memory.
+      // Calls that only access inaccessible memory can never alias with loads.
       if (I.mayWriteToMemory() &&
-          !(isa<IntrinsicInst>(I) &&
-            cast<IntrinsicInst>(I).getIntrinsicID() == Intrinsic::assume))
+          !(isa<CallBase>(I) &&
+            cast<CallBase>(I).onlyAccessesInaccessibleMemory()))
         return 0;
 
       if (LoadUsers.contains(&I))

[nikic]

LGTM

[fhahn]

LGTM, thanks.

Can you add a test with a different call that is supported now?

[topperc]

LGTM, thanks.

Can you add a test with a different call that is supported now?

Sure. Do you have a good suggestion?

[topperc]

LGTM, thanks.
Can you add a test with a different call that is supported now?

Sure. Do you have a good suggestion?

I used llvm.sideeffect.

[llvm-ci]

LLVM Buildbot has detected a new failure on builder ppc64le-mlir-rhel-clang running on ppc64le-mlir-rhel-test while building llvm at step 6 "test-build-check-mlir-build-only-check-mlir".

Full details are available at: https://lab.llvm.org/buildbot/#/builders/129/builds/34928

Here is the relevant piece of the build log for the reference

Step 6 (test-build-check-mlir-build-only-check-mlir) failure: 1200 seconds without output running [b'ninja', b'check-mlir'], attempting to kill
...
PASS: MLIR :: mlir-tblgen/op-decl-and-defs.td (3747 of 3758)
PASS: MLIR-Unit :: Interfaces/./MLIRInterfacesTests/13/22 (3748 of 3758)
PASS: MLIR :: mlir-tblgen/llvm-intrinsics.td (3749 of 3758)
PASS: MLIR :: mlir-translate/split-markers.mlir (3750 of 3758)
PASS: MLIR :: mlir-reduce/dce-test.mlir (3751 of 3758)
PASS: MLIR :: Pass/pipeline-parsing.mlir (3752 of 3758)
PASS: MLIR :: mlir-tblgen/cpp-class-comments.td (3753 of 3758)
PASS: MLIR :: mlir-runner/simple.mlir (3754 of 3758)
PASS: MLIR :: Pass/pipeline-options-parsing.mlir (3755 of 3758)
PASS: MLIR :: mlir-tblgen/op-error.td (3756 of 3758)
command timed out: 1200 seconds without output running [b'ninja', b'check-mlir'], attempting to kill
process killed by signal 9
program finished with exit code -1
elapsedTime=1921.823751