[Mlir-commits] [mlir] [MLIR][SideEffects][MemoryEffects] Modified LICM to be more aggressive when checking movability of ops with MemWrite effects (PR #155344)

[Mehdi Amini]

================
@@ -317,14 +319,53 @@ bool mlir::wouldOpBeTriviallyDead(Operation *op) {
   return wouldOpBeTriviallyDeadImpl(op);
 }
 
+std::optional<bool> mlir::isZeroTrip(mlir::LoopLikeOpInterface &loop) {
+  auto lbs = loop.getLoopLowerBounds();
+  auto ubs = loop.getLoopUpperBounds();
+  auto steps = loop.getLoopSteps();
+
+  if (!lbs || !ubs || !steps)
+    return std::nullopt;
+
+  if (lbs->size() != ubs->size() || ubs->size() != steps->size())
+    return std::nullopt;
+
+  for (size_t i = 0; i < steps->size(); ++i) {
+    auto lb = getConstantIntValue((*lbs)[i]);
+    auto ub = getConstantIntValue((*ubs)[i]);
+    auto st = getConstantIntValue((*steps)[i]);
+
+    if (!lb || !ub || !st)
+      return std::nullopt; // non-constant -> unknown
+
+    if (*st >= 0 && *lb >= *ub)
+      return true;
+    if (*st < 0 && *lb <= *ub)
+      return true;
----------------
joker-eph wrote:

I'm not sure if this is logic is flawless: for example it's not clear to me that this accounts for the signed/unsigned comparison flag on scf.for.


https://github.com/llvm/llvm-project/pull/155344