merge upstream

include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp

@@ -500,6 +500,7 @@ struct DeviceGroupedGemm_Xdl : public DeviceGroupedGemm<ALayout,
 
             for(std::size_t i = 0; i < arg.gemm_desc_kernel_arg_.size(); i++)
             {
+#if DEBUG_LOG
                 std::cout << "group: " << i << " arg.a_grid_desc_ak0_m_ak1_{"
                           << arg.gemm_desc_kernel_arg_[i].a_grid_desc_ak0_m_ak1_.GetLength(I0)
                           << ", "
@@ -520,6 +521,7 @@ struct DeviceGroupedGemm_Xdl : public DeviceGroupedGemm<ALayout,
                           << arg.gemm_desc_kernel_arg_[i].e_grid_desc_m_n_.GetLength(I0) << ", "
                           << arg.gemm_desc_kernel_arg_[i].e_grid_desc_m_n_.GetLength(I1) << "}"
                           << std::endl;
+#endif
 
                 if(!GridwiseGemm::CheckValidity(arg.gemm_desc_kernel_arg_[i].a_grid_desc_m_k_,
                                                 arg.gemm_desc_kernel_arg_[i].b_grid_desc_n_k_,

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

@@ -95,6 +95,8 @@ struct Scale
         y = scale_ * x;
     };
 
+    __host__ __device__ auto Value() const { return scale_; }
+
     float scale_;
 };
 

include/ck/utility/math_v2.hpp

@@ -3,7 +3,9 @@
 
 #pragma once
 
+#ifndef __HIP_DEVICE_COMPILE__
 #include <cmath>
+#endif
 
 #include "ck/utility/data_type.hpp"
 #include "ck/utility/type.hpp"