diff --git a/example/ck_tile/01_fmha/fmha_fwd.cpp b/example/ck_tile/01_fmha/fmha_fwd.cpp
index f2e966597c..01ec88b945 100644
--- a/example/ck_tile/01_fmha/fmha_fwd.cpp
+++ b/example/ck_tile/01_fmha/fmha_fwd.cpp
@@ -1106,9 +1106,7 @@ bool run(const ck_tile::ArgParser& arg_parser)
             ck_tile::reference_batched_rotary_position_embedding<RoPEComputeDataType>(
                 q_host_ref, rotary_cos_host, rotary_sin_host, is_rotary_interleaved, q_host_ref_ro);
 
-            #if 0
             q_host_ref.ForEach([&](auto& self, auto i) { self(i) = q_host_ref_ro(i); });
-            #endif
         }
         #if 0
         HOST_DEBUG_STMTS {
diff --git a/include/ck_tile/ops/fmha/kernel/fmha_fwd_appendkv_kernel.hpp b/include/ck_tile/ops/fmha/kernel/fmha_fwd_appendkv_kernel.hpp
index 7c3447ae2e..4a19116a0e 100644
--- a/include/ck_tile/ops/fmha/kernel/fmha_fwd_appendkv_kernel.hpp
+++ b/include/ck_tile/ops/fmha/kernel/fmha_fwd_appendkv_kernel.hpp
@@ -494,7 +494,7 @@ struct FmhaFwdAppendKVKernel
                 const auto rotary_cos_dram_native =
                     make_naive_tensor_view<address_space_enum::global>(
                         reinterpret_cast<const KDataType*>(kargs.rotary_cos_ptr),
-                        make_tuple(kargs.seqlen_k + kargs.seqlen_knew, kargs.rotary_dim / 2),
+                        make_tuple(kargs.seqlen_q, kargs.rotary_dim / 2),
                         make_tuple(kargs.rotary_dim / 2, 1),
                         number<8>{},
                         number<1>{});
@@ -519,7 +519,7 @@ struct FmhaFwdAppendKVKernel
                 const auto rotary_sin_dram_native =
                     make_naive_tensor_view<address_space_enum::global>(
                         reinterpret_cast<const KDataType*>(kargs.rotary_sin_ptr),
-                        make_tuple(kargs.seqlen_k + kargs.seqlen_knew, kargs.rotary_dim / 2),
+                        make_tuple(kargs.seqlen_q, kargs.rotary_dim / 2),
                         make_tuple(kargs.rotary_dim / 2, 1),
                         number<8>{},
                         number<1>{});
diff --git a/include/ck_tile/ops/fmha/pipeline/block_fmha_fwd_appendkv_pipeline.hpp b/include/ck_tile/ops/fmha/pipeline/block_fmha_fwd_appendkv_pipeline.hpp
index a6e012df1e..efd13ed503 100644
--- a/include/ck_tile/ops/fmha/pipeline/block_fmha_fwd_appendkv_pipeline.hpp
+++ b/include/ck_tile/ops/fmha/pipeline/block_fmha_fwd_appendkv_pipeline.hpp
@@ -108,10 +108,6 @@ struct BlockFmhaFwdAppendKVPipeline
     {
 #if defined(ENABLE_DEVICE_DEBUG_STMTS)
         auto* const ksmem = reinterpret_cast<KDataType*>(smem_ptr);
-        if(threadIdx.x == 0)
-        {
-            printf("\n");
-        }
 #endif
 
         auto print_tile = [&](const auto& tile, index_t num_display_rows = -1) {
@@ -140,6 +136,7 @@ struct BlockFmhaFwdAppendKVPipeline
 
             DEVICE_DEBUG_STMTS
             {
+                printf("\n");
                 for(int row = 0;
                     row < (0 < num_display_rows ? std::min(num_display_rows, num_rows) : num_rows);
                     ++row)
@@ -195,7 +192,7 @@ struct BlockFmhaFwdAppendKVPipeline
                                                         rotary_dim,
                                                         thread_end);
             }
-            print_tile(knew_tile, 2);
+            // print_tile(knew_tile, 2);
             store_tile(k_dram_block_window, knew_tile);
 
             auto vnew_window = make_tile_window(