passthrough add bfp16 to fp16 convert

include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp

@@ -63,6 +63,12 @@ struct PassThrough
         y = type_convert<bhalf_t>(x);
     }
 
+    template <>
+    __host__ __device__ void operator()<half_t, bhalf_t>(half_t& y, const bhalf_t& x) const
+    {
+        y = type_convert<half_t>(x);
+    }
+
     template <>
     __host__ __device__ void operator()<int8_t, int8_t>(int8_t& y, const int8_t& x) const
     {

include/ck/tensor_operation/gpu/grid/gridwise_batched_mha_bwd_xdl_cshuffle_qloop_ydotygrad.hpp

@@ -126,6 +126,7 @@ struct GridwiseBatchedMultiheadAttentionBackward_YDotYGrad
 
         static_assert(ThreadClusterLength_O * ThreadSliceLength_O == BlockSliceLength_O_, "");
         static_assert(ThreadClusterLength_M * ThreadSliceLength_M == BlockSliceLength_M_, "");
+        static_assert(SrcScalarPerVector % YSrcScalarPerVector == 0, "");
 
         using SrcBufType = StaticBuffer<AddressSpaceEnum::Vgpr,
                                         FloatD,