[LAA] Improve the output remark for LoopVectorize

[Allen]

Summary

Don't report 'Use #pragma loop distribute(enable) to allow loop
distribution' when we already add #pragma clang loop distribute(enable)

Fixes https://github.com/llvm/llvm-project/issues/64637

[Allen]

replace the test under the llvm/test/Transforms/LoopVectorize

[Allen]

adjust some comment in test file

[Allen]

ping

[fhahn]

llvm/test/Transforms/LoopVectorize/pr64637.ll

2

As the change is in LoopAccessAnalysis only, can this just be a test running LoopAccessAnalysis (as in llvm/test/Analysis/LoopAccessAnalysis/)?

[Allen]

address comment, replace the case under llvm/test/Analysis/LoopAccessAnalysis and run with LoopAccessAnalysis only

[Allen]

llvm/test/Transforms/LoopVectorize/pr64637.ll

2

Done, thanks for your suggestion

[Allen]

this was re-submitted as a github PR: https://github.com/llvm/llvm-project/pull/65832

llvm/lib/Analysis/LoopAccessAnalysis.cpp

   });
   if (Found == Deps->end())
     return;
   MemoryDepChecker::Dependence Dep = *Found;
 
   LLVM_DEBUG(dbgs() << "LAA: unsafe dependent memory operations in loop\n");
 
   // Emit remark for first unsafe dependence
+  bool HasForcedDistribution = false;
+  std::optional<const MDOperand *> Value =
+      findStringMetadataForLoop(TheLoop, "llvm.loop.distribute.enable");
+  if (Value) {
+    const MDOperand *Op = *Value;
+    assert(Op && mdconst::hasa<ConstantInt>(*Op) && "invalid metadata");
+    HasForcedDistribution = mdconst::extract<ConstantInt>(*Op)->getZExtValue();
+  }
+
+  const std::string Info =
+      HasForcedDistribution
+          ? "unsafe dependent memory operations in loop."
+          : "unsafe dependent memory operations in loop. Use "
+            "#pragma loop distribute(enable) to allow loop distribution "
+            "to attempt to isolate the offending operations into a separate "
+            "loop";
   OptimizationRemarkAnalysis &R =
-      recordAnalysis("UnsafeDep", Dep.getDestination(*this))
-      << "unsafe dependent memory operations in loop. Use "
-         "#pragma loop distribute(enable) to allow loop distribution "
-         "to attempt to isolate the offending operations into a separate "
-         "loop";
+      recordAnalysis("UnsafeDep", Dep.getDestination(*this)) << Info;
 
   switch (Dep.Type) {
   case MemoryDepChecker::Dependence::NoDep:
   case MemoryDepChecker::Dependence::Forward:
   case MemoryDepChecker::Dependence::BackwardVectorizable:
     llvm_unreachable("Unexpected dependence");
   case MemoryDepChecker::Dependence::Backward:
     R << "\nBackward loop carried data dependence.";

llvm/test/Transforms/LoopVectorize/pr64637.ll

+; RUN: opt -passes=loop-vectorize -S -pass-remarks-analysis=loop-vectorize -disable-output < %s 2>&1 | FileCheck %s --check-prefix=ANALYSIS
+
+; Test that LoopVectorize is not inhibited by the presence of a[i+1]
+; after the store to w[i].
+;
+; Testcase derived from the following C:
+;
+; #define M  100
+; void foo (int *restrict y, int *restrict x, int *restrict indices, int n)
+; {
+;   int k = 3;
+;   #pragma clang loop distribute(enable)
+;   for (int i = 0; i < n; i++) {
+;     y[i + k * M] = y[i + k* M] + 1;
+;     y[i + k * (M+1)] = indices[i] + 2;
+;   }
+; }
+
+define void @foo(ptr noalias nocapture noundef %y, ptr noalias nocapture noundef readnone %x, ptr noalias nocapture noundef readonly %indices, i32 noundef %n) {
+; ANALYSIS: loop not vectorized: unsafe dependent memory operations in loop.
+; ANALYSIS: Backward loop carried data dependence that prevents store-to-load forwarding.
+entry:
+  %cmp22 = icmp sgt i32 %n, 0
+  br i1 %cmp22, label %for.body.preheader, label %for.cond.cleanup
+
+for.body.preheader:                               ; preds = %entry
+  %wide.trip.count = zext i32 %n to i64
+  br label %for.body
+
+for.cond.cleanup.loopexit:                        ; preds = %for.body
+  br label %for.cond.cleanup
+
+for.cond.cleanup:                                 ; preds = %for.cond.cleanup.loopexit, %entry
+  ret void
+
+for.body:                                         ; preds = %for.body, %for.body.preheader
+  %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]
+  %0 = add nuw nsw i64 %indvars.iv, 300
+  %arrayidx = getelementptr inbounds i32, ptr %y, i64 %0
+  %1 = load i32, ptr %arrayidx, align 4
+  %add1 = add nsw i32 %1, 1
+  store i32 %add1, ptr %arrayidx, align 4
+  %arrayidx7 = getelementptr inbounds i32, ptr %indices, i64 %indvars.iv
+  %2 = load i32, ptr %arrayidx7, align 4
+  %add8 = add nsw i32 %2, 2
+  %3 = add nuw nsw i64 %indvars.iv, 303
+  %arrayidx12 = getelementptr inbounds i32, ptr %y, i64 %3
+  store i32 %add8, ptr %arrayidx12, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count
+  br i1 %exitcond.not, label %for.cond.cleanup.loopexit, label %for.body, !llvm.loop !0
+}
+
+!0 = distinct !{!0, !1}
+!1 = !{!"llvm.loop.distribute.enable", i1 true}