Merge branch 'develop' of github.com:ROCmSoftwarePlatform/composable_kernel...

Merge branch 'develop' of github.com:ROCmSoftwarePlatform/composable_kernel into barkocot/grouped-conv-weight-fp16-c1-k1

include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp

@@ -4,7 +4,8 @@
 #pragma once
 
 #include "ck/utility/common_header.hpp"
-#include "ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp"
+#include "ck/utility/loop_scheduler.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
 
 namespace ck {
 

include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_streamk.hpp

@@ -37,7 +37,8 @@ __global__ void
                                    index_t StrideC,
                                    typename GridwiseGemm::Block2CTileMap block_mapping)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     constexpr index_t shared_size = GridwiseGemm::GetSharedMemoryNumberOfByte();
 
     __shared__ uint8_t p_shared[shared_size];

library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp

@@ -101,6 +101,7 @@ using MultiplyAdd    = ck::tensor_operation::element_wise::MultiplyAdd;
 using ScaleAdd       = ck::tensor_operation::element_wise::ScaleAdd;
 using Gelu           = ck::tensor_operation::element_wise::Gelu;
 using Swish          = ck::tensor_operation::element_wise::Swish;
+using Add            = ck::tensor_operation::element_wise::Add;
 
 template <typename Activation>
 using Activation_Mul_Clamp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<Activation>;