add issupportargument check to light version

include/ck/tensor_operation/gpu/device/impl/device_grouped_mha_bwd_xdl_cshuffle_qloop_light_v1.hpp

@@ -1335,6 +1335,13 @@ struct DeviceGroupedMultiheadAttentionBackward_Qloop_Xdl_CShuffle_Light_V1
                 return false;
             }
 
+            // saving dQ data with atomic_add instruction, so KzRaw must be a multiple of 2
+            if(KzRaw % 2 != 0)
+            {
+                std::cout << "K_q must be a multiple of 2" << std::endl;
+                return false;
+            }
+
             // Check vector load/store requirement
             const auto a_stride_lowest = ABlockTransferSrcVectorDim == 2
                                              ? device_arg.a_mz_kz_strides_[1]

include/ck/tensor_operation/gpu/device/impl/device_grouped_mha_bwd_xdl_cshuffle_qloop_light_v2.hpp

@@ -1407,6 +1407,13 @@ struct DeviceGroupedMultiheadAttentionBackward_Qloop_Xdl_CShuffle_Light_V2
                 return false;
             }
 
+            // saving dQ data with atomic_add instruction, so KzRaw must be a multiple of 2
+            if(KzRaw % 2 != 0)
+            {
+                std::cout << "K_q must be a multiple of 2" << std::endl;
+                return false;
+            }
+
             // Check vector load/store requirement
             const auto a_stride_lowest = ABlockTransferSrcVectorDim == 2
                                              ? device_arg.a_mz_kz_strides_[1]