[MTHREADS]: enable resolve_neg op (#1234)

Author: lingfengqiu

benchmark/test_special_perf.py

@@ -30,7 +30,10 @@ def topk_input_fn(shape, dtype, device):
 
 def resolve_neg_input_fn(shape, dtype, device):
     x = torch.randn(size=shape, dtype=dtype, device=device)
-    yield x.conj().imag,
+    if vendor_name == "mthreads":
+        yield x.conj(),
+    else:
+        yield x.conj().imag,
 
 
 def resolve_conj_input_fn(shape, dtype, device):
@@ -61,8 +64,6 @@ def resolve_conj_input_fn(shape, dtype, device):
     ],
 )
 def test_special_operations_benchmark(op_name, torch_op, dtypes, input_fn):
-    if vendor_name == "mthreads" and op_name in ["resolve_neg"]:
-        pytest.skip("Torch not supported complex")
     bench = GenericBenchmarkExcluse1D(
         input_fn=input_fn, op_name=op_name, dtypes=dtypes, torch_op=torch_op
     )