[DA] use NSW arithmetic

This patch fixes #51512

DA uses SCEVs to represent linear constraints on memory accesses.
addToCoefficient is only called from the "Constraint Propagation engine":
https://github.com/llvm/llvm-project/blob/main/llvm/lib/Analysis/DependenceAnalysis.cpp#L3877

The propagation engine works on MIV subscripts.
MIV subscripts are pairs of *linear* memory accesses that vary in multiple loops.
At this point all memory accesses are linear, non-linear have been filtered out.

When addToCoefficient creates an AddRec with flags SCEV::FlagAnyWrap,
classifyPair returns llvm::DependenceInfo::Subscript::NonLinear.
https://github.com/llvm/llvm-project/blob/main/llvm/lib/Analysis/DependenceAnalysis.cpp#L3882

Having a non-linear effect this late in the dependence test is surprising, and
the test fails with llvm_unreachable("bad subscript classification").

Instead of building non-linear functions, this patch modifies the flags to
SCEV::FlagNSW that has the meaning of linear functions that do not wrap around.

Author: sebpop

llvm/lib/Analysis/DependenceAnalysis.cpp

@@ -3115,15 +3115,17 @@ const SCEV *DependenceInfo::zeroCoefficient(const SCEV *Expr,
 // coefficient corresponding to the specified TargetLoop.
 // For example, given a*i + b*j + c*k, adding 1 to the coefficient
 // corresponding to the j loop would yield a*i + (b+1)*j + c*k.
+
+// When Expr is invariant in TargetLoop, i.e., the stride in TargetLoop is 0,
+// addToCoefficient returns an AddRec {Expr, +, Value}_TargetLoop.
+// NoWrapFlags is set to FlagNSW for the evolution function to be linear.
 const SCEV *DependenceInfo::addToCoefficient(const SCEV *Expr,
                                              const Loop *TargetLoop,
                                              const SCEV *Value) const {
   const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(Expr);
-  if (!AddRec) // create a new addRec
-    return SE->getAddRecExpr(Expr,
-                             Value,
-                             TargetLoop,
-                             SCEV::FlagAnyWrap); // Worst case, with no info.
+  if (!AddRec)
+    return SE->getAddRecExpr(Expr, Value, TargetLoop, SCEV::FlagNSW);
+
   if (AddRec->getLoop() == TargetLoop) {
     const SCEV *Sum = SE->getAddExpr(AddRec->getStepRecurrence(*SE), Value);
     if (Sum->isZero())
@@ -3134,7 +3136,7 @@ const SCEV *DependenceInfo::addToCoefficient(const SCEV *Expr,
                              AddRec->getNoWrapFlags());
   }
   if (SE->isLoopInvariant(AddRec, TargetLoop))
-    return SE->getAddRecExpr(AddRec, Value, TargetLoop, SCEV::FlagAnyWrap);
+    return SE->getAddRecExpr(AddRec, Value, TargetLoop, SCEV::FlagNSW);
   return SE->getAddRecExpr(
       addToCoefficient(AddRec->getStart(), TargetLoop, Value),
       AddRec->getStepRecurrence(*SE), AddRec->getLoop(),

llvm/test/Analysis/DependenceAnalysis/PR51512.ll

@@ -0,0 +1,44 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --version 5
+; RUN: opt < %s -disable-output "-passes=print<da>" -aa-pipeline=basic-aa 2>&1 \
+; RUN: | FileCheck %s
+
+; Check that the testcase does not crash the compiler.
+; See https://github.com/llvm/llvm-project/issues/51512 for details.
+
+define void @foo() {
+; CHECK-LABEL: 'foo'
+; CHECK-NEXT:  Src: store i32 42, ptr %getelementptr, align 1 --> Dst: store i32 42, ptr %getelementptr, align 1
+; CHECK-NEXT:    da analyze - consistent output [0 S]!
+; CHECK-NEXT:  Src: store i32 42, ptr %getelementptr, align 1 --> Dst: store i32 0, ptr %getelementptr5, align 1
+; CHECK-NEXT:    da analyze - output [0 *|<]!
+; CHECK-NEXT:  Src: store i32 0, ptr %getelementptr5, align 1 --> Dst: store i32 0, ptr %getelementptr5, align 1
+; CHECK-NEXT:    da analyze - none!
+;
+bb:
+  %alloca = alloca [2 x [5 x i32]], align 1
+  br label %bb1
+
+bb1:                                              ; preds = %bb7, %bb
+  %phi = phi i32 [ 0, %bb ], [ %add8, %bb7 ]
+  %trunc = trunc i32 %phi to i16
+  %add = add i16 %trunc, 3
+  %getelementptr = getelementptr inbounds [2 x [5 x i32]], ptr %alloca, i16 0, i16 %trunc, i16 %add
+  br label %bb2
+
+bb2:                                              ; preds = %bb2, %bb1
+  %phi3 = phi i32 [ 0, %bb1 ], [ %add6, %bb2 ]
+  store i32 42, ptr %getelementptr, align 1
+  %trunc4 = trunc i32 %phi3 to i16
+  %getelementptr5 = getelementptr inbounds [2 x [5 x i32]], ptr %alloca, i16 0, i16 %trunc4, i16 %add
+  store i32 0, ptr %getelementptr5, align 1
+  %add6 = add nuw nsw i32 %phi3, 1
+  br i1 false, label %bb2, label %bb7
+
+bb7:                                              ; preds = %bb2
+  %add8 = add nuw nsw i32 %phi, 1
+  %icmp = icmp eq i32 %phi, 0
+  br i1 %icmp, label %bb1, label %bb9
+
+bb9:                                              ; preds = %bb7
+  ret void
+}