From 3edcd5fc23b4c9d3cdd31431da2ce398ec57c6c0 Mon Sep 17 00:00:00 2001
From: "Po-Yen, Chen" <PoYen.Chen@amd.com>
Date: Mon, 4 Jul 2022 22:54:38 +0800
Subject: [PATCH] Fix wrong template argument count of DeviceGemm<>

---
 .../device_gemm_xdl_producer_consumer_cshuffle.hpp   | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_xdl_producer_consumer_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_xdl_producer_consumer_cshuffle.hpp
index 7db1b20823..8d4ecf5599 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm_xdl_producer_consumer_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_xdl_producer_consumer_cshuffle.hpp
@@ -60,8 +60,16 @@ template <typename ALayout,
           index_t CShuffleNXdlPerWavePerShuffle,
           typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
           index_t CShuffleBlockTransferScalarPerVector_NPerBlock>
-struct DeviceGemm_Xdl_ProducerConsumer_CShuffle
-    : public DeviceGemm<AElementwiseOperation, BElementwiseOperation, CElementwiseOperation>
+struct DeviceGemm_Xdl_ProducerConsumer_CShuffle : public DeviceGemm<ALayout,
+                                                                    BLayout,
+                                                                    CLayout,
+                                                                    ADataType,
+                                                                    BDataType,
+                                                                    CDataType,
+                                                                    AElementwiseOperation,
+                                                                    BElementwiseOperation,
+                                                                    CElementwiseOperation>
+
 {
     using DeviceOp = DeviceGemm_Xdl_ProducerConsumer_CShuffle;