diff --git a/sim/kelvin_vector_instructions.cc b/sim/kelvin_vector_instructions.cc
index 94c9724..5470b24 100644
--- a/sim/kelvin_vector_instructions.cc
+++ b/sim/kelvin_vector_instructions.cc
@@ -267,8 +267,12 @@
       inst, scalar, strip_mine,
       std::function<typename std::make_unsigned<T>::type(T, T)>(
           [](T vs1, T vs2) -> typename std::make_unsigned<T>::type {
-            T result = vs1 > vs2 ? vs1 - vs2 : vs2 - vs1;
-            return static_cast<typename std::make_unsigned<T>::type>(result);
+            using UT = typename std::make_unsigned<T>::type;
+            // Cast to unsigned type before the operation to avoid undefined
+            // overflow behavior in intx_t.
+            UT uvs1 = static_cast<UT>(vs1);
+            UT uvs2 = static_cast<UT>(vs2);
+            return vs1 > vs2 ? uvs1 - uvs2 : uvs2 - uvs1;
           }));
 }
 template void KelvinVAbsd<int8_t>(bool, bool, Instruction *);
diff --git a/sim/test/kelvin_vector_instructions_test.cc b/sim/test/kelvin_vector_instructions_test.cc
index ee6ccfb..6aede99 100644
--- a/sim/test/kelvin_vector_instructions_test.cc
+++ b/sim/test/kelvin_vector_instructions_test.cc
@@ -450,7 +450,9 @@
 template <typename Vd, typename Vs1, typename Vs2>
 struct VAbsdOp {
   static Vd Op(Vs1 vs1, Vs2 vs2) {
-    Vs1 result = vs1 > vs2 ? vs1 - vs2 : vs2 - vs1;
+    int64_t vs1_ext = static_cast<int64_t>(vs1);
+    int64_t vs2_ext = static_cast<int64_t>(vs2);
+    auto result = vs1_ext > vs2_ext ? vs1_ext - vs2_ext : vs2_ext - vs1_ext;
     return static_cast<Vd>(result);
   }
   static void KelvinOp(bool scalar, bool strip_mine, Instruction *inst) {
@@ -458,8 +460,7 @@
   }
 };
 
-// TODO(b/291683818): Check the implementation and re-enable the logic.
-TEST_F(KelvinVectorInstructionsTest, DISABLED_VAbsd) {
+TEST_F(KelvinVectorInstructionsTest, VAbsd) {
   KelvinVectorBinaryOpHelper<VAbsdOp, uint8_t, int8_t, int8_t, uint16_t,
                              int16_t, int16_t, uint32_t, int32_t, int32_t>(
       "VAbsd");