Unify the alignment to be 8 for Q/K/V Lds decriptors

include/ck_tile/ops/fmha/pipeline/block_fmha_pipeline_qx_ks_vs_custom_policy.hpp

@@ -192,7 +192,8 @@ struct BlockFmhaPipelineQXCustomPolicy</* QLoadOnce = */ false>
     template <typename Problem>
     CK_TILE_HOST_DEVICE static constexpr auto MakeQLdsBlockDescriptor()
     {
-        using QDataType = remove_cvref_t<typename Problem::QDataType>;
+        using QDataType                 = remove_cvref_t<typename Problem::QDataType>;
+        constexpr index_t MaxVectorSize = 16 / sizeof(typename Problem::QDataType);
 
         constexpr index_t kMPerBlock = Problem::BlockFmhaShape::kM0;
         constexpr index_t kKPerBlock = Problem::BlockFmhaShape::kK0;
@@ -201,7 +202,7 @@ struct BlockFmhaPipelineQXCustomPolicy</* QLoadOnce = */ false>
         constexpr auto q_lds_block_desc_0 = make_naive_tensor_descriptor(
             make_tuple(number<kKPerBlock / kKPack>{}, number<kMPerBlock>{}, number<kKPack>{}),
             make_tuple(number<(kMPerBlock + 1) * kKPack>{}, number<kKPack>{}, number<1>{}),
-            number<8>{},
+            number<MaxVectorSize>{},
             number<1>{});
 
         constexpr auto q_lds_block_desc = transform_tensor_descriptor(
@@ -415,6 +416,9 @@ struct BlockFmhaPipelineQXKSVSCustomPolicy : BlockFmhaPipelineQXCustomPolicy<QLo
     template <typename Problem>
     CK_TILE_HOST_DEVICE static constexpr auto MakeKLdsBlockDescriptor()
     {
+        using KDataType                 = remove_cvref_t<typename Problem::KDataType>;
+        constexpr index_t MaxVectorSize = 16 / sizeof(typename Problem::KDataType);
+
         constexpr index_t NumKLdsBuffers = GetNumKLdsBuffers<Problem>();
         constexpr index_t kNPerBlock     = Problem::BlockFmhaShape::kN0;
         constexpr index_t kKPerBlock     = Problem::BlockFmhaShape::kK0;
@@ -429,7 +433,7 @@ struct BlockFmhaPipelineQXKSVSCustomPolicy : BlockFmhaPipelineQXCustomPolicy<QLo
                                                     number<(kNPerBlock + 1) * kKPack>{},
                                                     number<kKPack>{},
                                                     number<1>{}),
-                                         number<8>{},
+                                         number<MaxVectorSize>{},
                                          number<1>{});
 
         constexpr auto k_lds_block_desc = transform_tensor_descriptor(
@@ -447,7 +451,9 @@ struct BlockFmhaPipelineQXKSVSCustomPolicy : BlockFmhaPipelineQXCustomPolicy<QLo
     template <typename Problem>
     CK_TILE_HOST_DEVICE static constexpr auto MakeVLdsBlockDescriptor()
     {
-        using VDataType                = remove_cvref_t<typename Problem::VDataType>;
+        using VDataType                 = remove_cvref_t<typename Problem::VDataType>;
+        constexpr index_t MaxVectorSize = 16 / sizeof(typename Problem::VDataType);
+
         constexpr index_t Banks        = 32; // TODO: need change based on arch
         constexpr index_t PixelsPerRow = Banks * 4 / sizeof(VDataType);
         constexpr index_t kKPack       = GetSmemKPackV<Problem>();
@@ -471,7 +477,7 @@ struct BlockFmhaPipelineQXKSVSCustomPolicy : BlockFmhaPipelineQXCustomPolicy<QLo
                        number<PixelsPerRow + kKPack>{},
                        number<kKPack>{},
                        number<1>{}),
-            number<kKPack>{},
+            number<MaxVectorSize>{},
             number<1>{});
 
         constexpr auto v_lds_block_desc = transform_tensor_descriptor(