Merge remote-tracking branch 'origin/develop' into contraction

include/ck/tensor_operation/gpu/device/device_gemm_xdl_cshuffle.hpp

@@ -65,8 +65,15 @@ template <typename ALayout,
           typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
           index_t CShuffleBlockTransferScalarPerVector_NPerBlock,
           LoopScheduler LoopSched = make_default_loop_scheduler()>
-struct DeviceGemm_Xdl_CShuffle
-    : public DeviceGemm<AElementwiseOperation, BElementwiseOperation, CElementwiseOperation>
+struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
+                                                   BLayout,
+                                                   CLayout,
+                                                   ADataType,
+                                                   BDataType,
+                                                   CDataType,
+                                                   AElementwiseOperation,
+                                                   BElementwiseOperation,
+                                                   CElementwiseOperation>
 {
     using DeviceOp = DeviceGemm_Xdl_CShuffle;
 
@@ -622,8 +629,7 @@ struct DeviceGemm_Xdl_CShuffle
                                                       index_t StrideC,
                                                       AElementwiseOperation a_element_op,
                                                       BElementwiseOperation b_element_op,
-                                                      CElementwiseOperation c_element_op,
-                                                      index_t /* KBatch */ = 1) override
+                                                      CElementwiseOperation c_element_op) override
     {
         return std::make_unique<Argument>(static_cast<const ADataType*>(p_a),
                                           static_cast<const BDataType*>(p_b),

include/ck/tensor_operation/gpu/device/device_gemm_xdl_splitk.hpp

@@ -10,7 +10,7 @@
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemm.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_splitk.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp"
 #include "ck/device_utility/device_prop.hpp"
@@ -56,8 +56,15 @@ template <typename ADataType,
           bool BBlockLdsAddExtraN,
           ck::index_t CThreadTransferSrcDstVectorDim,
           ck::index_t CThreadTransferDstScalarPerVector>
-struct DeviceGemmXdlSplitK
-    : public DeviceGemm<AElementwiseOperation, BElementwiseOperation, CElementwiseOperation>
+struct DeviceGemmXdlSplitK : public DeviceGemmSplitK<ALayout,
+                                                     BLayout,
+                                                     CLayout,
+                                                     ADataType,
+                                                     BDataType,
+                                                     CDataType,
+                                                     AElementwiseOperation,
+                                                     BElementwiseOperation,
+                                                     CElementwiseOperation>
 {
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};

include/ck/tensor_operation/gpu/element/element_wise_operation.hpp

@@ -159,7 +159,7 @@ struct Normalize
         using ck::math::sqrt;
 
         float variance = mean_square - (mean * mean);
-        y = ((x - mean) / sqrt(variance + static_cast<float>(epsilon_))) * gamma + beta;
+        y = ((x - mean) / sqrt(variance + type_convert<float>(epsilon_))) * gamma + beta;
     };
 
     template <>