diff --git a/Lib/test/test_capi/test_opt.py b/Lib/test/test_capi/test_opt.py
index 25b2c393e6773de..3b7dd49a3c1cdfa 100644
--- a/Lib/test/test_capi/test_opt.py
+++ b/Lib/test/test_capi/test_opt.py
@@ -5582,8 +5582,7 @@ def testfunc(*args):
         uops = get_opnames(ex)
 
         self.assertIn("_BINARY_OP_SUBSCR_INIT_CALL", uops)
-        # _POP_TOP_NOP is a sign the optimizer ran and didn't hit contradiction.
-        self.assertGreaterEqual(count_ops(ex, "_POP_TOP_NOP"), 1)
+        self.assertNotIn("_POP_TOP_NOP", uops)
 
     def test_load_attr_property_frame(self):
         class B:
@@ -6215,6 +6214,39 @@ def __exit__(self, e, v, t): ...
                 f1()
         """), PYTHON_JIT="1")
 
+    def test_peephole_cancels_adjacent_push_pop(self):
+        def negated_constant(n):
+            x = 0
+            for i in range(n):
+                a = 1
+                result = -a
+                if result < 0:
+                    x += 1
+            return x
+
+        res, ex = self._run_with_optimizer(negated_constant, TIER2_THRESHOLD)
+        self.assertEqual(res, TIER2_THRESHOLD)
+        self.assertIsNotNone(ex)
+        uops = get_opnames(ex)
+        self.assertNotIn("_SWAP", uops)
+        self.assertNotIn("_SWAP_2", uops)
+
+    def test_peephole_removes_redundant_rrot(self):
+        def add(a, b):
+            return a + b
+
+        def testfunc(n):
+            x = 0
+            for _ in range(n):
+                x += add(1, 2)
+            return x
+
+        res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
+        self.assertEqual(res, 3 * TIER2_THRESHOLD)
+        self.assertIsNotNone(ex)
+        uops = get_opnames(ex)
+        self.assertNotIn("_RROT_3", uops)
+
 def global_identity(x):
     return x
 
diff --git a/Python/optimizer_analysis.c b/Python/optimizer_analysis.c
index e726dc0e6fd1114..375ea2ea19432fa 100644
--- a/Python/optimizer_analysis.c
+++ b/Python/optimizer_analysis.c
@@ -719,6 +719,80 @@ const uint16_t op_without_pop[MAX_UOP_ID + 1] = {
     [_POP_TOP_UNICODE] = _NOP,
 };
 
+static int
+previous_non_skip_uop(_PyUOpInstruction *buffer, int pc)
+{
+    while (pc >= 0 && op_skip[buffer[pc].opcode]) {
+        pc--;
+    }
+    return pc;
+}
+
+/* Remove redundant stack shuffles left by constant-folding rewrites:
+ *     push push _SWAP(2) pop
+ *     push push push _RROT_3 pop pop
+ * In both forms, only the push closest to the shuffle survives. */
+static bool
+remove_folded_stack_shuffle(_PyUOpInstruction *buffer, int pc, int arity)
+{
+    int pops[2];
+    int idx = pc;
+    for (int i = 0; i < arity - 1; i++) {
+        if (idx < 0 || !op_without_pop[buffer[idx].opcode]) {
+            return false;
+        }
+        pops[i] = idx;
+        idx = previous_non_skip_uop(buffer, idx - 1);
+    }
+    int shuf = idx;
+    if (shuf < 0) {
+        return false;
+    }
+    uint16_t shuf_op = buffer[shuf].opcode;
+    bool is_shuffle =
+        (arity == 2 && shuf_op == _SWAP && buffer[shuf].oparg == 2) ||
+        (arity == 3 && shuf_op == _RROT_3);
+    if (!is_shuffle) {
+        return false;
+    }
+
+    int dead_pushes[2];
+    for (int i = 0; i < arity; i++) {
+        idx = previous_non_skip_uop(buffer, idx - 1);
+        if (idx < 0) {
+            return false;
+        }
+        uint16_t push_op = buffer[idx].opcode;
+        if (push_op == _COPY || !op_without_push[push_op]) {
+            return false;
+        }
+        if (i > 0) {
+            dead_pushes[i - 1] = idx;
+        }
+    }
+
+    for (int i = 0; i < arity - 1; i++) {
+        buffer[dead_pushes[i]].opcode = _NOP;
+    }
+    buffer[shuf].opcode = _NOP;
+    for (int i = 0; i < arity - 1; i++) {
+        buffer[pops[i]].opcode = _NOP;
+    }
+    return true;
+}
+
+static bool
+remove_adjacent_push_pop(_PyUOpInstruction *buffer, int pc)
+{
+    int last = previous_non_skip_uop(buffer, pc - 1);
+    if (last < 0 || !op_without_push[buffer[last].opcode]) {
+        return false;
+    }
+    buffer[last].opcode = _NOP;
+    buffer[pc].opcode = _NOP;
+    return true;
+}
+
 
 static int
 remove_unneeded_uops(_PyUOpInstruction *buffer, int buffer_size)
@@ -755,21 +829,14 @@ remove_unneeded_uops(_PyUOpInstruction *buffer, int buffer_size)
                 // ...becomes:
                 //     _NOP + _NOP + _POP_TOP + _NOP + _NOP
                 while (op_without_pop[opcode]) {
-                    _PyUOpInstruction *last = &buffer[pc - 1];
-                    while (op_skip[last->opcode]) {
-                        last--;
-                    }
-                    if (op_without_push[last->opcode] && op_without_pop[opcode]) {
-                        last->opcode = op_without_push[last->opcode];
-                        opcode = buffer[pc].opcode = op_without_pop[opcode];
-                        if (op_without_pop[last->opcode]) {
-                            opcode = last->opcode;
-                            pc = (int)(last - buffer);
-                        }
-                    }
-                    else {
-                        break;
+                    if (remove_folded_stack_shuffle(buffer, pc, 2) ||
+                        remove_folded_stack_shuffle(buffer, pc, 3) ||
+                        remove_adjacent_push_pop(buffer, pc))
+                    {
+                        opcode = buffer[pc].opcode;
+                        continue;
                     }
+                    break;
                 }
                 /* _PUSH_FRAME doesn't escape or error, but it
                  * does need the IP for the return address */