lint

Author: hgt312

python/tvm/runtime/ndarray.py

@@ -147,7 +147,9 @@ def copyfrom(self, source_array):
                     source_array.shape, shape
                 )
             )
-        source_array = np.ascontiguousarray(source_array, dtype="uint16" if dtype == "bfloat16" else dtype)
+        source_array = np.ascontiguousarray(
+            source_array, dtype="uint16" if dtype == "bfloat16" else dtype
+        )
         assert source_array.flags["C_CONTIGUOUS"]
         data = source_array.ctypes.data_as(ctypes.c_void_p)
         nbytes = ctypes.c_size_t(source_array.size * source_array.dtype.itemsize)

src/target/source/codegen_cuda.cc

@@ -378,8 +378,8 @@ void CodeGenCUDA::PrintVecElemStore(const std::string& vec, DataType t, int i,
     stream << "((half2*)(&(" << vec << "." << access[i / 2] << ")))->" << access[i % 2] << " = "
            << value << ";\n";
   } else if (t.is_bfloat16()) {
-    stream << "((nv_bfloat162*)(&(" << vec << "." << access[i / 2] << ")))->" << access[i % 2] << " = "
-           << value << ";\n";
+    stream << "((nv_bfloat162*)(&(" << vec << "." << access[i / 2] << ")))->" << access[i % 2]
+           << " = " << value << ";\n";
   } else {
     stream << vec << "." << access[i] << " = " << value << ";\n";
   }

tests/python/unittest/test_target_codegen_cuda.py

@@ -95,9 +95,9 @@ def check_cuda(n, lanes):
         xo, xi = s[B].split(B.op.axis[0], factor=num_thread)
         s[B].bind(xo, bx)
         s[B].bind(xi, tx)
-        with tvm.transform.PassContext(disabled_pass=["tir.BF16Promote",
-                                                      "tir.BF16CastElimination",
-                                                      "tir.BF16TypeLowering"]):
+        with tvm.transform.PassContext(
+            disabled_pass=["tir.BF16Promote", "tir.BF16CastElimination", "tir.BF16TypeLowering"]
+        ):
             fun = tvm.build(s, [A, B], "cuda")
         ctx = tvm.gpu(0)
         np_a = np.random.uniform(size=(n, lanes)).astype("float32")