4 files changed, 43 insertions, 6 deletions
diff --git a/src/cpu/kernels/CpuActivationKernel.cpp b/src/cpu/kernels/CpuActivationKernel.cpp
index 61efcb2dd6..f1e485883c 100644
--- a/src/cpu/kernels/CpuActivationKernel.cpp
+++ b/src/cpu/kernels/CpuActivationKernel.cpp
@@ -182,10 +182,16 @@ std::pair<Status, Window> validate_and_configure_window(const ITensorInfo *src,
 
 void CpuActivationKernel::configure(const ITensorInfo *src, ITensorInfo *dst, ActivationLayerInfo activation_info)
 {
+    ARM_COMPUTE_UNUSED(dst);
     ARM_COMPUTE_ERROR_ON_NULLPTR(src);
     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst, activation_info));
 
     const auto uk = CpuActivationKernel::get_implementation(ActivationDataTypeISASelectorData{ src->data_type(), CPUInfo::get().get_isa(), activation_info.activation() });
+    if(dst != nullptr)
+    {
+        // dst auto inizialitation if not yet initialized
+        auto_init_if_empty(*dst, *src->clone());
+    }
 
     ARM_COMPUTE_ERROR_ON_NULLPTR(uk);
 
@@ -200,10 +206,20 @@ void CpuActivationKernel::configure(const ITensorInfo *src, ITensorInfo *dst, Ac
 #endif // __aarch64__
     _act_info = activation_info;
 
-    // Configure kernel window
-    auto win_config = validate_and_configure_window(src, dst);
-    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
-    ICPPKernel::configure(win_config.second);
+    Window win;
+
+    if(src->data_layout() != DataLayout::NHWC)
+    {
+        // Use squashed window
+        std::tie(win, _split_dimension) = calculate_squashed_or_max_window(*src);
+        ICPPKernel::configure(win);
+    }
+    else
+    {
+        // Configure kernel window
+        win = calculate_max_window(*src, Steps());
+        ICPPKernel::configure(win);
+    }
 }
 
 Status CpuActivationKernel::validate(const ITensorInfo *src, const ITensorInfo *dst, const ActivationLayerInfo &act_info)
diff --git a/src/cpu/kernels/CpuActivationKernel.h b/src/cpu/kernels/CpuActivationKernel.h
index d856a9357f..fe2d783059 100644
--- a/src/cpu/kernels/CpuActivationKernel.h
+++ b/src/cpu/kernels/CpuActivationKernel.h
@@ -73,6 +73,15 @@ public:
     void run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info) override;
     const char *name() const override;
 
+    /** Get the preferred dimension in which the scheduler splits the work into multiple jobs.
+     *
+     * @return The split dimension hint.
+     */
+    size_t get_split_dimension_hint() const
+    {
+        return _split_dimension;
+    }
+
     struct ActivationKernel
     {
         const char                                *name;
@@ -85,6 +94,7 @@ public:
 private:
     ActivationLayerInfo _act_info{};
     ActivationKernelPtr _run_method{ nullptr };
+    size_t              _split_dimension{ Window::DimY };
     std::string         _name{};
 };
 } // namespace kernels
diff --git a/src/cpu/operators/CpuActivation.cpp b/src/cpu/operators/CpuActivation.cpp
index 3945fa59a5..197e9850b9 100644
--- a/src/cpu/operators/CpuActivation.cpp
+++ b/src/cpu/operators/CpuActivation.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021 Arm Limited.
+ * Copyright (c) 2021-2022 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -23,6 +23,7 @@
  */
 #include "src/cpu/operators/CpuActivation.h"
 
+#include "arm_compute/runtime/NEON/NEScheduler.h"
 #include "src/common/IOperator.h"
 #include "src/common/utils/LegacySupport.h"
 #include "src/common/utils/Log.h"
@@ -46,6 +47,13 @@ Status CpuActivation::validate(const ITensorInfo *input, const ITensorInfo *outp
     return kernels::CpuActivationKernel::validate(input, output, activation_info);
 }
 
+void CpuActivation::run(ITensorPack &tensors)
+{
+    ARM_COMPUTE_ERROR_ON_MSG(tensors.empty(), "No inputs provided");
+    auto split_dimension = static_cast<kernels::CpuActivationKernel *>(_kernel.get())->get_split_dimension_hint();
+    NEScheduler::get().schedule_op(_kernel.get(), split_dimension, _kernel->window(), tensors);
+}
+
 std::tuple<IOperator *, StatusCode> CpuContext::create_activation(const AclTensorDescriptor &src, const AclTensorDescriptor &dst, const AclActivationDescriptor &act, bool is_validate)
 {
     TensorInfo src_info = detail::convert_to_legacy_tensor_info(src);
diff --git a/src/cpu/operators/CpuActivation.h b/src/cpu/operators/CpuActivation.h
index 9b97c9d24f..f1807d5e47 100644
--- a/src/cpu/operators/CpuActivation.h
+++ b/src/cpu/operators/CpuActivation.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021 Arm Limited.
+ * Copyright (c) 2021-2022 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -48,6 +48,9 @@ public:
      * @return a status
      */
     static Status validate(const ITensorInfo *input, const ITensorInfo *output, const ActivationLayerInfo &act_info);
+
+    // Inherited methods overridden:
+    void run(ITensorPack &tensors) override;
 };
 } // namespace cpu
 } // namespace arm_compute