[mlir][int-range] Limit xor int range inference to i1

[Hardcode84]

Fixes #82168

intrange::inferXor was incorrectly handling ranges for widths > i1 (see example in code). Limit it to i1 for now. For bigger widths it will return maxRange.

[llvmbot]

@llvm/pr-subscribers-mlir-arith

@llvm/pr-subscribers-mlir

Author: Ivan Butygin (Hardcode84)

Changes

Fixes #82168

intrange::inferXor was incorrectly handling ranges for widths > i1 (see example in code). Limit it to i1 for now. For bigger ranges it will return maxRange.

---

Full diff: https://github.com/llvm/llvm-project/pull/116968.diff

2 Files Affected:

• (modified) mlir/lib/Interfaces/Utils/InferIntRangeCommon.cpp (+14)
• (modified) mlir/test/Dialect/Arith/int-range-interface.mlir (+28-2)

diff --git a/mlir/lib/Interfaces/Utils/InferIntRangeCommon.cpp b/mlir/lib/Interfaces/Utils/InferIntRangeCommon.cpp
index 0b085b10b2b337..2c1276d577a55b 100644
--- a/mlir/lib/Interfaces/Utils/InferIntRangeCommon.cpp
+++ b/mlir/lib/Interfaces/Utils/InferIntRangeCommon.cpp
@@ -558,6 +558,20 @@ mlir::intrange::inferOr(ArrayRef<ConstantIntRanges> argRanges) {
 
 ConstantIntRanges
 mlir::intrange::inferXor(ArrayRef<ConstantIntRanges> argRanges) {
+  // TODO: The code below doesn't work for bitwidths > i1.
+  // For input ranges lhs=[2060639849, 2060639850], rhs=[2060639849, 2060639849]
+  // widenBitwiseBounds will produce:
+  // lhs:
+  // 2060639848  01111010110100101101111001101000
+  // 2060639851  01111010110100101101111001101011
+  // rhs:
+  // 2060639849  01111010110100101101111001101001
+  // 2060639849  01111010110100101101111001101001
+  // None of those combinations xor to 0, while intermediate values does.
+  unsigned width = argRanges[0].umin().getBitWidth();
+  if (width > 1)
+    return ConstantIntRanges::maxRange(width);
+
   auto [lhsZeros, lhsOnes] = widenBitwiseBounds(argRanges[0]);
   auto [rhsZeros, rhsOnes] = widenBitwiseBounds(argRanges[1]);
   auto xori = [](const APInt &a, const APInt &b) -> std::optional<APInt> {
diff --git a/mlir/test/Dialect/Arith/int-range-interface.mlir b/mlir/test/Dialect/Arith/int-range-interface.mlir
index afb0b4929bce70..4db846fa4656a3 100644
--- a/mlir/test/Dialect/Arith/int-range-interface.mlir
+++ b/mlir/test/Dialect/Arith/int-range-interface.mlir
@@ -454,9 +454,35 @@ func.func @ori(%arg0 : i128, %arg1 : i128) -> i1 {
     func.return %2 : i1
 }
 
+// CHECK-LABEL: func @xori_issue_82168
+// arith.cmpi was erroneously folded to %false, see Issue #82168.
+// CHECK: %[[R:.*]] = arith.cmpi eq, %{{.*}}, %{{.*}} : i64
+// CHECK: return %[[R]]
+func.func @xori_issue_82168() -> i1 {
+    %c0_i64 = arith.constant 0 : i64
+    %c2060639849_i64 = arith.constant 2060639849 : i64
+    %2 = test.with_bounds { umin = 2060639849 : i64, umax = 2060639850 : i64, smin = 2060639849 : i64, smax = 2060639850 : i64 } : i64
+    %3 = arith.xori %2, %c2060639849_i64 : i64
+    %4 = arith.cmpi eq, %3, %c0_i64 : i64
+    func.return %4 : i1
+}
+
+// CHECK-LABEL: func @xori_i1
+//   CHECK-DAG: %[[true:.*]] = arith.constant true
+//   CHECK-DAG: %[[false:.*]] = arith.constant false
+//       CHECK: return %[[true]], %[[false]]
+func.func @xori_i1() -> (i1, i1) {
+    %true = arith.constant true
+    %1 = test.with_bounds { umin = 0 : i1, umax = 0 : i1, smin = 0 : i1, smax = 0 : i1 } : i1
+    %2 = test.with_bounds { umin = 1 : i1, umax = 1 : i1, smin = 1 : i1, smax = 1 : i1 } : i1
+    %3 = arith.xori %1, %true : i1
+    %4 = arith.xori %2, %true : i1
+    func.return %3, %4 : i1, i1
+}
+
 // CHECK-LABEL: func @xori
-// CHECK: %[[false:.*]] = arith.constant false
-// CHECK: return %[[false]]
+// TODO: xor folding is temporarily disabled
+// CHECK-NOT: arith.constant false
 func.func @xori(%arg0 : i64, %arg1 : i64) -> i1 {
     %c0 = arith.constant 0 : i64
     %c7 = arith.constant 7 : i64

[krzysz00]

Approved to prevent miscompiles, will look into real implementation later