From 52c018b7456f38ab4ba6989cdea495ef2d4f4902 Mon Sep 17 00:00:00 2001
From: mtgu0705 <mtgu@amd.com>
Date: Thu, 13 Feb 2025 11:12:57 +0800
Subject: [PATCH] fix bug, function pass.

---
 .../block/blockwise_gemm_pipeline_xdlops_b_preshuffle_v1.hpp  | 4 ++--
 .../gpu/grid/gridwise_gemm_xdl_cshuffle_v3_b_preshuffle.hpp   | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_b_preshuffle_v1.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_b_preshuffle_v1.hpp
index 0e758bbbee..73c892382d 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_b_preshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_b_preshuffle_v1.hpp
@@ -365,10 +365,10 @@ struct BlockwiseGemmXdlops_pipeline_bpreshuffle_v1<BlockGemmPipelineScheduler::I
                     // B VGPR->VGPR dequant
                     b_thread_dequant_copy.Run(b_block_desc_n0_n1_k0_k1,
                                               b_block_origin_idx,
-                                              b_thread_bufs(mfma_reg_buf),
+                                              b_thread_bufs(local_read_buf),
                                               b_thread_desc_,
                                               make_tuple(I0, I0, I0, I0),
-                                              b_thread_dequant_bufs(mfma_reg_buf));
+                                              b_thread_dequant_bufs(local_read_buf));
 
                     HotLoopScheduler();
                     __builtin_amdgcn_sched_barrier(0);
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3_b_preshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3_b_preshuffle.hpp
index 1e4ceebeeb..5f15f24288 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3_b_preshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3_b_preshuffle.hpp
@@ -1620,7 +1620,7 @@ struct GridwiseGemm_xdl_cshuffle_v3_b_preshuffle
             decltype(b_block_desc_bk0_n_bk1),
             tensor_operation::element_wise::PassThrough,
             Sequence<Number<NXdlPerWave>{}, I1, Number<KRepeat>{}, Number<BK1Value>{}>,
-            Sequence<0, 1, 2, 3>,
+            Sequence<1, 2, 0, 3>,
             3,
             BK1Number>(b_element_op);