Merge branch 'develop' of...

Merge branch 'develop' of https://github.com/ROCmSoftwarePlatform/composable_kernel into navi3x_mD_batchedGEMM_GroupConvFwd

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

@@ -434,7 +434,7 @@ struct GridwiseElementwiseLayernormWelfordVariance_mk_to_mk
             });
 
             static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
-                auto divisor = 1 / __builtin_amdgcn_sqrtf(var_thread_buf(iM) + epsilon);
+                auto divisor = 1 / ck::math::sqrt(var_thread_buf(iM) + epsilon);
                 static_for<0, XThreadBufferNumber, 1>{}([&](auto iK0) {
                     static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
                         constexpr auto offset_m_k =

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

@@ -319,7 +319,7 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
             });
 
             static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
-                auto divisor = 1 / __builtin_amdgcn_sqrtf(var_thread_buf(iM) + epsilon);
+                auto divisor = 1 / ck::math::sqrt(var_thread_buf(iM) + epsilon);
                 static_for<0, XThreadBufferNumber, 1>{}([&](auto iK0) {
                     static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
                         constexpr auto offset_m_k =

include/ck/utility/synchronization.hpp

@@ -18,6 +18,7 @@ __device__ void block_sync_lds()
     __syncthreads();
 #endif
 }
+
 __device__ void s_nop()
 {
 #if 1

library/include/ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp

@@ -90,10 +90,13 @@ struct ReferenceLayernorm : public device::BaseOperator
 
             for(int m = 0; m < M; ++m)
             {
+                AccDataType divisor =
+                    static_cast<AccDataType>(1) / ck::math::sqrt(var(m) + arg.epsilon_);
+
                 for(int n = 0; n < N; ++n)
                 {
                     auto x_val = ck::type_convert<AccDataType>(arg.x_m_n_(m, n));
-                    auto y_val = (x_val - mean(m)) / sqrt(var(m) + arg.epsilon_);
+                    auto y_val = (x_val - mean(m)) * divisor;
                     y_val      = (y_val * arg.gamma_n_(n)) + arg.beta_n_(n);
                     arg.acc_elementwise_op_(y_val, y_val);
                     arg.y_m_n_(m, n) = ck::type_convert<YDataType>(y_val);

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

@@ -90,6 +90,7 @@ using Bilinear       = ck::tensor_operation::element_wise::Bilinear;
 using AddAddFastGelu = ck::tensor_operation::element_wise::AddAddFastGelu;
 using AddFastGelu    = ck::tensor_operation::element_wise::AddFastGelu;
 using FastGelu       = ck::tensor_operation::element_wise::FastGelu;
+using AddMultiply    = ck::tensor_operation::element_wise::AddMultiply;
 
 template <typename Activation>
 using Activation_Mul_Clamp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<Activation>;