fix comments

include/ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_permute.hpp

@@ -87,9 +87,6 @@ template <index_t NumDimG,
           MaskingSpecialization MaskingSpec>
 struct DeviceBatchedMultiheadAttentionForward : public BaseOperator
 {
-    static constexpr index_t NumAcc0Bias = 1;
-    static constexpr index_t NumAcc1Bias = 0;
-
     virtual std::unique_ptr<BaseArgument> MakeArgumentPointer(
         const void* p_a,
         const void* p_b0,

include/ck/tensor_operation/gpu/device/impl/device_grouped_mha_fwd_xdl_cshuffle_v2.hpp

@@ -698,7 +698,7 @@ struct DeviceGroupedMultiheadAttentionForward_Xdl_CShuffle_V2
             {
                 const auto p_a_grid  = static_cast<const ADataType*>(p_a_vec[i]);
                 const auto p_b_grid  = static_cast<const BDataType*>(p_b_vec[i]);
-                const auto p_d0_grid = p_acc0_biases_vec.size()
+                const auto p_d0_grid = p_acc0_biases_vec.size() > 0
                                            ? static_cast<const D0DataType*>(p_acc0_biases_vec[i])
                                            : nullptr;
                 const auto p_b1_grid  = static_cast<const B1DataType*>(p_b1_vec[i]);

include/ck/tensor_operation/gpu/grid/gridwise_batched_mha_fwd_xdl_cshuffle_v2.hpp

@@ -1288,7 +1288,7 @@ struct GridwiseBatchedMultiheadAttentionForward_Xdl_CShuffle_V2
             block_sync_lds(); // wait for lds read in gemm0 blockwise gemm
 
             // add bias
-            if constexpr(!std::is_void<D0DataType>::value)
+            if constexpr(!is_same<D0DataType, void>::value)
             {
                 const auto d0_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
                     p_d0_grid, d0_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5.GetElementSpaceSize());