Check vector size

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

@@ -101,18 +101,18 @@ struct DeviceNormalizationBwdGammaBetaImpl
                    (XSrcVectorDim == 1 && KThreadSliceSize % XSrcVectorSize == 0)),
                   "Invalid thread slice sizes and/or x vector sizes configuration, please check!");
 
-    static_assert(
-        ((MThreadSliceSize % DGammaDstVectorSize == 0) ||
-         (MThreadSliceSize % DBetaDstVectorSize == 0)),
-        "Invalid thread slice sizes and/or Gamma and beta vector sizes configuration, please "
-        "check!");
-
     static_assert(
         (MeanInvStdSrcVectorDim == 0 && MThreadSliceSize % MeanInvStdSrcVectorSize == 0) ||
             (MeanInvStdSrcVectorDim == 1 && KThreadSliceSize % MeanInvStdSrcVectorSize == 0),
         "Invalid thread slice sizes and/or mean and inverse std vector sizes configuration, please "
         "check!");
 
+    static_assert(
+        ((MThreadSliceSize % DGammaDstVectorSize == 0) ||
+         (MThreadSliceSize % DBetaDstVectorSize == 0)),
+        "Invalid thread slice sizes and/or Gamma and beta vector sizes configuration, please "
+        "check!");
+
     static constexpr index_t NumInvariantDim = Rank - NumReduceDim;
     static constexpr index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
     static constexpr index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize;

include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_bwd_gamma_beta.hpp

@@ -35,7 +35,7 @@ template <typename DYDataType,
           index_t DBetaDstVectorSize>
 struct GridwiseNormalizationBwdGammaBeta_mk_to_k
 {
-    // if we just check ThreadSliceSize  & VectorSize == 0, the performance may be poor
+    // if we just check ThreadSliceSize % VectorSize == 0, the performance may be poor (coalesce)
     static_assert(((DYSrcVectorDim == 0 && MThreadSliceSize == DYSrcVectorSize) ||
                    (DYSrcVectorDim == 1 && KThreadSliceSize == DYSrcVectorSize)),
                   "Invalid thread slice sizes and/or dy vector sizes configuration, please check!");
@@ -44,6 +44,14 @@ struct GridwiseNormalizationBwdGammaBeta_mk_to_k
                    (XSrcVectorDim == 1 && KThreadSliceSize == XSrcVectorSize)),
                   "Invalid thread slice sizes and/or x vector sizes configuration, please check!");
 
+    static_assert(
+        ((MeanInvStdSrcVectorDim == 0 && MThreadSliceSize == MeanInvStdSrcVectorSize) ||
+         (MeanInvStdSrcVectorDim == 1 && KThreadSliceSize == MeanInvStdSrcVectorSize)),
+        "Invalid thread slice sizes and/or mean/inv_std vector sizes configuration, please check!");
+
+    static_assert(MThreadSliceSize == DGammaDstVectorSize && MThreadSliceSize == DBetaDstVectorSize,
+                  "Invalid thread slice sizes and/or dx vector sizes configuration, please check!");
+
     using ThreadClusterLengths_M_K = Sequence<MThreadClusterSize, KThreadClusterSize>;
 
     using DYThreadBufferDimAccessOrder =