[llvm] [LAA] Add stencil group merging to reduce runtime pointer checks (PR #187252)

[Igor Kirillov via llvm-commits]

================
@@ -737,6 +752,367 @@ void RuntimePointerChecking::groupChecks(
   }
 }
 
+/// Result of decomposing a SCEV expression into stencil offset form:
+///   Offset = Constant + sum(Coefficients[stride] * stride)
+/// where each stride is a loop-invariant SCEV expression.
+struct StencilDecomposition {
+  int64_t Constant = 0;
+  /// Map from loop-invariant stride SCEV to its integer coefficient.
+  SmallMapVector<const SCEV *, int64_t, 4> Coefficients;
+};
+
+/// Try to decompose \p Expr into a stencil offset function of loop-invariant
+/// strides: C + a1*s1 + a2*s2 + ...
+/// Relies on SCEV's canonical form: AddExpr operands are flattened,
+/// MulExpr has the constant operand first.
+/// Returns std::nullopt if the expression contains non-stencil terms.
+static std::optional<StencilDecomposition>
+decomposeStencilOffset(const SCEV *Expr, ScalarEvolution &SE, const Loop &L) {
+  StencilDecomposition D;
+
+  // Collect top-level additive terms.
+  SmallVector<const SCEV *, 4> Terms;
+  if (auto *Add = dyn_cast<SCEVAddExpr>(Expr))
+    Terms.append(Add->operands().begin(), Add->operands().end());
+  else
+    Terms.push_back(Expr);
+
+  for (const SCEV *Term : Terms) {
+    if (auto *C = dyn_cast<SCEVConstant>(Term)) {
+      D.Constant += C->getAPInt().getSExtValue();
+    } else if (auto *Mul = dyn_cast<SCEVMulExpr>(Term)) {
+      // SCEV canonical form: constant is operand 0 in a MulExpr.
+      if (Mul->getNumOperands() != 2)
+        return std::nullopt;
+      auto *C = dyn_cast<SCEVConstant>(Mul->getOperand(0));
+      if (!C || !SE.isLoopInvariant(Mul->getOperand(1), &L))
+        return std::nullopt;
+      D.Coefficients[Mul->getOperand(1)] += C->getAPInt().getSExtValue();
----------------
igogo-x86 wrote:

APInt has no signedness. We use `getSExtValue()` because real access patterns have small coefficients - signed or unsigned doesn't matter at small magnitudes, the bit pattern is the same. The only corner is a coefficient with bit pattern `0x8000000000000000`, but that's rather impossible in practice.

On the `mul i8 %v, 255` test: SCEV canonicalises it into `(sext i8 (-1 * %v) to i64)` - the i8 constant ends up inside the sext, never reaching getSExtValue. No realistic way to test.

https://github.com/llvm/llvm-project/pull/187252