COMPMID-1060 LSTM FP32 NEON

Change-Id: I0bdf874e61917903c26f713ec41a7ffc29e07233
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/143892
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

6 files changed, 1107 insertions, 132 deletions

diff --git a/arm_compute/runtime/CL/functions/CLLSTMLayer.h b/arm_compute/runtime/CL/functions/CLLSTMLayer.h
index 6896a97e31..72e41a7aca 100644
--- a/arm_compute/runtime/CL/functions/CLLSTMLayer.h
+++ b/arm_compute/runtime/CL/functions/CLLSTMLayer.h
@@ -39,6 +39,7 @@
 #include "arm_compute/runtime/CL/functions/CLGEMM.h"
 #include "arm_compute/runtime/CL/functions/CLWidthConcatenateLayer.h"
 #include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/common/LSTMParams.h"
 
 #include <memory>
 
@@ -46,138 +47,6 @@ namespace arm_compute
 {
 class ICLTensor;
 
-template <typename T>
-class LSTMParams
-{
-public:
-    /** Constructor */
-    LSTMParams()
-        : _input_to_input_weights(nullptr), _recurrent_to_input_weights(nullptr), _cell_to_input_weights(nullptr), _input_gate_bias(nullptr), _cell_to_forget_weights(nullptr),
-          _cell_to_output_weights(nullptr), _projection_weights(nullptr), _projection_bias(nullptr), _has_peephole_opt(false), _has_projection(false), _has_cifg_opt(true)
-    {
-    }
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    LSTMParams(const LSTMParams &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    LSTMParams &operator=(const LSTMParams &) = delete;
-    /** Default destructor */
-    ~LSTMParams() = default;
-    /** Set CIFG tensor parameters.
-     *
-     * @param[in] input_to_input_weights     2D weights tensor with dimensions [input_size, num_units]. Data types supported: F16/F32.
-     * @param[in] recurrent_to_input_weights 2D weights tensor with dimensions [output_size, num_units]. Data type supported: Same as @p input_to_input_weights.
-     * @param[in] cell_to_input_weights      1D weights tensor with dimensions [num_units]. Can be nullptr. Data type supported: Same as @p input_to_input_weights.
-     * @param[in] input_gate_bias            1D weights tensor with dimensions [num_units]. Data type supported: Same as @p input_to_input_weights
-     *
-     * @return Reference to this LSTMParams object
-     */
-    LSTMParams &set_cifg_params(const T *input_to_input_weights, const T *recurrent_to_input_weights, const T *cell_to_input_weights, const T *input_gate_bias)
-    {
-        _input_to_input_weights     = input_to_input_weights;
-        _recurrent_to_input_weights = recurrent_to_input_weights;
-        _cell_to_input_weights      = cell_to_input_weights;
-        _input_gate_bias            = input_gate_bias;
-        _has_cifg_opt               = false;
-        return *this;
-    }
-    /** Set projection tensor parameters.
-     *
-     * @param[in] projection_weights 2D weights tensor with dimensions [output_size, num_units]. Data type supported: Data types supported: F16/F32.
-     * @param[in] projection_bias    1D weights tensor with dimensions [output_size]. Data type supported: Same as @p projection_weights.
-     *
-     * @return Reference to this LSTMParams object
-     */
-    LSTMParams &set_projection_params(const T *projection_weights, const T *projection_bias)
-    {
-        _projection_weights = projection_weights;
-        _projection_bias    = projection_bias;
-        _has_projection     = true;
-        return *this;
-    }
-    /** Set peephole tensor parameters.
-     *
-     * @param[in] cell_to_forget_weights 1D weights tensor with dimensions [num_units]. Data type supported: Data types supported: F16/F32.
-     * @param[in] cell_to_output_weights 1D weights tensor with dimensions [num_units]. Data type supported: Same as @p cell_to_input_weights.
-     *
-     * @return Reference to this LSTMParams object
-     */
-    LSTMParams &set_peephole_params(const T *cell_to_forget_weights, const T *cell_to_output_weights)
-    {
-        _cell_to_forget_weights = cell_to_forget_weights;
-        _cell_to_output_weights = cell_to_output_weights;
-        _has_peephole_opt       = true;
-        return *this;
-    }
-
-    const T *input_to_input_weights() const
-    {
-        return _input_to_input_weights;
-    }
-
-    const T *recurrent_to_input_weights() const
-    {
-        return _recurrent_to_input_weights;
-    }
-
-    const T *cell_to_input_weights() const
-    {
-        return _cell_to_input_weights;
-    }
-
-    const T *input_gate_bias() const
-    {
-        return _input_gate_bias;
-    }
-
-    const T *cell_to_forget_weights() const
-    {
-        return _cell_to_forget_weights;
-    }
-
-    const T *cell_to_output_weights() const
-    {
-        return _cell_to_output_weights;
-    }
-
-    const T *projection_weights() const
-    {
-        return _projection_weights;
-    }
-
-    const T *projection_bias() const
-    {
-        return _projection_bias;
-    }
-
-    bool has_peephole_opt() const
-    {
-        return _has_peephole_opt;
-    }
-
-    bool has_projection() const
-    {
-        return _has_projection;
-    }
-
-    bool has_cifg_opt() const
-    {
-        return _has_cifg_opt;
-    }
-
-private:
-    const T *_input_to_input_weights;
-    const T *_recurrent_to_input_weights;
-    const T *_cell_to_input_weights;
-    const T *_input_gate_bias;
-    const T *_cell_to_forget_weights;
-    const T *_cell_to_output_weights;
-    const T *_projection_weights;
-    const T *_projection_bias;
-    bool     _has_peephole_opt;
-    bool     _has_projection;
-    bool     _has_cifg_opt;
-};
-
 /** This function performs a single time step in a Long Short-Term Memory (LSTM) layer.
  *
  */
diff --git a/arm_compute/runtime/NEON/NEFunctions.h b/arm_compute/runtime/NEON/NEFunctions.h
index e6d7114384..dabd78c256 100644
--- a/arm_compute/runtime/NEON/NEFunctions.h
+++ b/arm_compute/runtime/NEON/NEFunctions.h
@@ -82,6 +82,7 @@
 #include "arm_compute/runtime/NEON/functions/NEIm2Col.h"
 #include "arm_compute/runtime/NEON/functions/NEIntegralImage.h"
 #include "arm_compute/runtime/NEON/functions/NEL2NormalizeLayer.h"
+#include "arm_compute/runtime/NEON/functions/NELSTMLayer.h"
 #include "arm_compute/runtime/NEON/functions/NELaplacianPyramid.h"
 #include "arm_compute/runtime/NEON/functions/NELaplacianReconstruct.h"
 #include "arm_compute/runtime/NEON/functions/NELocallyConnectedLayer.h"
diff --git a/arm_compute/runtime/NEON/functions/NELSTMLayer.h b/arm_compute/runtime/NEON/functions/NELSTMLayer.h
new file mode 100644
index 0000000000..9c4ab2b068
--- /dev/null
+++ b/arm_compute/runtime/NEON/functions/NELSTMLayer.h
@@ -0,0 +1,210 @@
+/*
+ * Copyright (c) 2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_NELSTMLAYER_H__
+#define __ARM_COMPUTE_NELSTMLAYER_H__
+
+#include "arm_compute/core/NEON/kernels/NEActivationLayerKernel.h"
+#include "arm_compute/core/NEON/kernels/NEArithmeticAdditionKernel.h"
+#include "arm_compute/core/NEON/kernels/NEArithmeticSubtractionKernel.h"
+#include "arm_compute/core/NEON/kernels/NECopyKernel.h"
+#include "arm_compute/core/NEON/kernels/NEPixelWiseMultiplicationKernel.h"
+
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/NEON/INESimpleFunction.h"
+#include "arm_compute/runtime/NEON/functions/NEArithmeticAddition.h"
+#include "arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h"
+#include "arm_compute/runtime/NEON/functions/NEGEMM.h"
+#include "arm_compute/runtime/NEON/functions/NEWidthConcatenateLayer.h"
+#include "arm_compute/runtime/common/LSTMParams.h"
+
+namespace arm_compute
+{
+// Forward declarations
+class ITensor;
+
+/** Basic function to run @ref NELSTMLayer */
+class NELSTMLayer : public IFunction
+{
+public:
+    /** Default constructor */
+    NELSTMLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    /** Initialize function's tensors.
+     *
+     * @param[in]  input                       Source tensor. Input is a 2D tensor with dimensions [input_size, batch_size]. Data types supported: F16/F32.
+     * @param[in]  input_to_forget_weights     2D weights tensor with dimensions [input_size, num_units]. Data type supported: Same as @p input.
+     * @param[in]  input_to_cell_weights       2D weights tensor with dimensions [input_size, num_units]. Data type supported: Same as @p input.
+     * @param[in]  input_to_output_weights     2D weights tensor with dimensions [input_size, num_units]. Data type supported: Same as @p input.
+     * @param[in]  recurrent_to_forget_weights 2D weights tensor with dimensions [output_size, num_units]. Data type supported: Same as @p input.
+     * @param[in]  recurrent_to_cell_weights   2D weights tensor with dimensions [output_size, num_units]. Data type supported: Same as @p input.
+     * @param[in]  recurrent_to_output_weights 2D weights tensor with dimensions [output_size, num_units]. Data type supported: Same as @p input.
+     * @param[in]  forget_gate_bias            1D weights tensor with dimensions [num_units]. Data type supported: Same as @p input.
+     * @param[in]  cell_bias                   1D weights tensor with dimensions [num_units]. Data type supported: Same as @p input.
+     * @param[in]  output_gate_bias            1D weights tensor with dimensions [num_units]. Data type supported: Same as @p input.
+     * @param[in]  output_state_in             2D weights tensor with dimensions [output_size, batch_size]. Data type supported: Same as @p input.
+     * @param[in]  cell_state_in               2D tensor with dimensions [num_units, batch_size]. Data type supported: Same as @p input.
+     * @param[out] scratch_buffer              2D tensor with dimensions [num_units * 4, batch_size] with CIFG or [num_units * 3, batch_size] without CIGF. Data type supported: Same as @p input.
+     * @param[out] output_state_out            2D weights tensor with dimensions [output_size, batch_size]. Data type supported: Same as @p input.
+     * @param[out] cell_state_out              2D tensor with dimensions [num_units, batch_size]. Data type supported: Same as @p input.
+     * @param[out] output                      Destination tensor. Output is a 2D tensor with dimensions [output_size, batch_size].
+     *                                         Data types supported: Same as @p input.
+     * @param[in]  lstm_params                 (Optional) Weights tensors used in peephole optimization:
+     *                                         input_to_input_weights       2D weights tensor with dimensions [input_size, num_units]. Data type supported: Same as @p input.
+     *                                         recurrent_to_input_weights   2D weights tensor with dimensions [output_size, num_units]. Data type supported: Same as @p input.
+     *                                         cell_to_input_weights        1D weights tensor with dimensions [num_units]. Can be nullptr. Data type supported: Same as @p input.
+     *                                         cell_to_forget_weights       1D weights tensor with dimensions [num_units]. Data type supported: Same as @p input.
+     *                                         cell_to_output_weights       1D weights tensor with dimensions [num_units]. Data type supported: Same as @p input.
+     *                                         input_gate_bias              1D weights tensor with dimensions [num_units]. Data type supported: Same as @p input
+     *                                         projection_weights           2D weights tensor with dimensions [output_size, num_units]. Data type supported: Same as @p input.
+     *                                         projection_bias              1D weights tensor with dimensions [output_size]. Data type supported: Same as @p input.
+     * @param[in]  activation_info             Contains activation information described in @ref ActivationLayerInfo.
+     * @param[in]  cell_threshold              The clipping threshold for the cell state, such that values are bound within [-cell_clip, cell_clip]. If set to 0.0 then clipping is disabled.
+     * @param[in]  projection_threshold        The clipping threshold for the output from the projection layer, such that values are bound within [-proj_clip, proj_clip]. If set to 0.0 then clipping is disabled.
+     */
+    void configure(const ITensor *input,
+                   const ITensor *input_to_forget_weights, const ITensor *input_to_cell_weights, const ITensor *input_to_output_weights,
+                   const ITensor *recurrent_to_forget_weights, const ITensor *recurrent_to_cell_weights, const ITensor *recurrent_to_output_weights,
+                   const ITensor *forget_gate_bias, const ITensor *cell_bias, const ITensor *output_gate_bias,
+                   const ITensor *output_state_in, const ITensor *cell_state_in,
+                   ITensor *scratch_buffer, ITensor *output_state_out, ITensor *cell_state_out, ITensor *output,
+                   const LSTMParams<ITensor> &lstm_params, const ActivationLayerInfo &activation_info, float cell_threshold = 0.f, float projection_threshold = 0.f);
+
+    /** Static function to check if given info will lead to a valid configuration of @ref NELSTMLayer
+     *
+     * @param[in] input                       Source tensor. Input is a 2D tensor with dimensions [input_size, batch_size]. Data types supported: F16/F32.
+     * @param[in] input_to_forget_weights     2D weights tensor with dimensions [input_size, num_units]. Data type supported: Same as @p input.
+     * @param[in] input_to_cell_weights       2D weights tensor with dimensions [input_size, num_units]. Data type supported: Same as @p input.
+     * @param[in] input_to_output_weights     2D weights tensor with dimensions [input_size, num_units]. Data type supported: Same as @p input.
+     * @param[in] recurrent_to_forget_weights 2D weights tensor with dimensions [output_size, num_units]. Data type supported: Same as @p input.
+     * @param[in] recurrent_to_cell_weights   2D weights tensor with dimensions [output_size, num_units]. Data type supported: Same as @p input.
+     * @param[in] recurrent_to_output_weights 2D weights tensor with dimensions [output_size, num_units]. Data type supported: Same as @p input.
+     * @param[in] forget_gate_bias            1D weights tensor with dimensions [num_units]. Data type supported: Same as @p input.
+     * @param[in] cell_bias                   1D weights tensor with dimensions [num_units]. Data type supported: Same as @p input.
+     * @param[in] output_gate_bias            1D weights tensor with dimensions [num_units]. Data type supported: Same as @p input.
+     * @param[in] output_state_in             2D weights tensor with dimensions [output_size, batch_size]. Data type supported: Same as @p input.
+     * @param[in] cell_state_in               2D tensor with dimensions [num_units, batch_size]. Data type supported: Same as @p input.
+     * @param[in] scratch_buffer              2D tensor with dimensions [num_units * 4, batch_size] with CIFG or [num_units * 3, batch_size] without CIGF. Data type supported: Same as @p input.
+     * @param[in] output_state_out            2D weights tensor with dimensions [output_size, batch_size]. Data type supported: Same as @p input.
+     * @param[in] cell_state_out              2D tensor with dimensions [num_units, batch_size]. Data type supported: Same as @p input.
+     * @param[in] output                      Destination tensor. Output is a 2D tensor with dimensions [output_size, batch_size].
+     *                                        Data types supported: Same as @p input.
+     * @param[in] lstm_params                 (Optional) Weights tensors used in peephole optimization:
+     *                                        input_to_input_weights       2D weights tensor with dimensions [input_size, num_units]. Data type supported: Same as @p input.
+     *                                        recurrent_to_input_weights   2D weights tensor with dimensions [output_size, num_units]. Data type supported: Same as @p input.
+     *                                        cell_to_input_weights        1D weights tensor with dimensions [num_units]. Can be nullptr. Data type supported: Same as @p input.
+     *                                        cell_to_forget_weights       1D weights tensor with dimensions [num_units]. Data type supported: Same as @p input.
+     *                                        cell_to_output_weights       1D weights tensor with dimensions [num_units]. Data type supported: Same as @p input.
+     *                                        input_gate_bias              1D weights tensor with dimensions [num_units]. Data type supported: Same as @p input
+     *                                        projection_weights           2D weights tensor with dimensions [output_size, num_units]. Data type supported: Same as @p input.
+     *                                        projection_bias              1D weights tensor with dimensions [output_size]. Data type supported: Same as @p input.
+     * @param[in] activation_info             Contains activation information described in @ref ActivationLayerInfo.
+     * @param[in] cell_threshold              The clipping threshold for the cell state, such that values are bound within [-cell_clip, cell_clip]. If set to 0.0 then clipping is disabled.
+     * @param[in] projection_threshold        The clipping threshold for the output from the projection layer, such that values are bound within [-proj_clip, proj_clip]. If set to 0.0 then clipping is disabled.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input,
+                           const ITensorInfo *input_to_forget_weights, const ITensorInfo *input_to_cell_weights, const ITensorInfo *input_to_output_weights,
+                           const ITensorInfo *recurrent_to_forget_weights, const ITensorInfo *recurrent_to_cell_weights, const ITensorInfo *recurrent_to_output_weights,
+                           const ITensorInfo *forget_gate_bias, const ITensorInfo *cell_bias, const ITensorInfo *output_gate_bias,
+                           const ITensorInfo *output_state_in, const ITensorInfo *cell_state_in,
+                           const ITensorInfo *scratch_buffer, const ITensorInfo *output_state_out, const ITensorInfo *cell_state_out, const ITensorInfo *output,
+                           const LSTMParams<ITensorInfo> &lstm_params, const ActivationLayerInfo &activation_info, float cell_threshold = 0.f, float projection_threshold = 0.f);
+
+    // Inherited methods overridden:
+    void run() override;
+
+private:
+    MemoryGroup                     _memory_group;
+    NEFullyConnectedLayer           _fully_connected_input_gate;
+    NEGEMM                          _gemm_input_gate;
+    NETransposeKernel               _transpose_input_gate;
+    NEArithmeticAdditionKernel      _accum_input_gate1;
+    NEArithmeticAddition            _accum_input_gate2;
+    NEArithmeticSubtractionKernel   _subtract_input_gate;
+    NEPixelWiseMultiplicationKernel _pixelwise_mul_input_gate;
+    NEActivationLayerKernel         _activation_input_gate;
+    NEFullyConnectedLayer           _fully_connected_forget_gate;
+    NEGEMM                          _gemm_forget_gate;
+    NETransposeKernel               _transpose_forget_gate;
+    NEArithmeticAdditionKernel      _accum_forget_gate1;
+    NEArithmeticAddition            _accum_forget_gate2;
+    NEPixelWiseMultiplicationKernel _pixelwise_mul_forget_gate;
+    NEActivationLayerKernel         _activation_forget_gate;
+    NEFullyConnectedLayer           _fully_connected_cell_state;
+    NEGEMM                          _gemm_cell_state1;
+    NEGEMM                          _gemm_cell_state2;
+    NETransposeKernel               _transpose_cell_state;
+    NEArithmeticAdditionKernel      _accum_cell_state1;
+    NEArithmeticAdditionKernel      _accum_cell_state2;
+    NEPixelWiseMultiplicationKernel _pixelwise_mul_cell_state1;
+    NEActivationLayerKernel         _activation_cell_state;
+    NEActivationLayerKernel         _cell_clip;
+    NEPixelWiseMultiplicationKernel _pixelwise_mul_cell_state2;
+    NEFullyConnectedLayer           _fully_connected_output;
+    NEGEMM                          _gemm_output;
+    NEPixelWiseMultiplicationKernel _pixelwise_mul_output_state1;
+    NETransposeKernel               _transpose_output;
+    NEArithmeticAdditionKernel      _accum_output1;
+    NEArithmeticAddition            _accum_output2;
+    NEActivationLayerKernel         _activation_output;
+    NEActivationLayerKernel         _activation_output_state;
+    NEPixelWiseMultiplicationKernel _pixelwise_mul_output_state2;
+    NEFullyConnectedLayer           _fully_connected_output_state;
+    NEGEMM                          _gemm_output_state;
+    NEArithmeticAdditionKernel      _accum_output_state;
+    NEActivationLayerKernel         _projection_clip;
+    NECopyKernel                    _copy_cell_state;
+    NECopyKernel                    _copy_output;
+    NEWidthConcatenateLayer         _concat_scratch_buffer;
+    Tensor                          _input_gate_out1;
+    Tensor                          _input_gate_out2;
+    Tensor                          _input_gate_out3;
+    Tensor                          _input_gate_out4;
+    Tensor                          _input_gate_out5;
+    Tensor                          _forget_gate_out1;
+    Tensor                          _forget_gate_out2;
+    Tensor                          _forget_gate_out3;
+    Tensor                          _forget_gate_out4;
+    Tensor                          _forget_gate_out5;
+    Tensor                          _cell_state_out1;
+    Tensor                          _cell_state_out2;
+    Tensor                          _cell_state_out3;
+    Tensor                          _cell_state_out4;
+    Tensor                          _cell_state_out5;
+    Tensor                          _output1;
+    Tensor                          _output2;
+    Tensor                          _output3;
+    Tensor                          _output4;
+    Tensor                          _output5;
+    Tensor                          _cell_state_activation;
+    Tensor                          _output_state1;
+    Tensor                          _ones;
+    bool                            _run_peephole_opt;
+    bool                            _run_cifg_opt;
+    bool                            _perform_cell_clipping;
+    bool                            _has_projection_weights;
+    bool                            _perform_projection_clipping;
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_NELSTMLAYER_H__ */
diff --git a/arm_compute/runtime/common/LSTMParams.h b/arm_compute/runtime/common/LSTMParams.h
new file mode 100644
index 0000000000..5b33e2e937
--- /dev/null
+++ b/arm_compute/runtime/common/LSTMParams.h
@@ -0,0 +1,169 @@
+/*
+ * Copyright (c) 2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_LSTMPARAMS_H__
+#define __ARM_COMPUTE_LSTMPARAMS_H__
+
+#include "arm_compute/core/IPyramid.h"
+#include "arm_compute/core/PyramidInfo.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/Tensor.h"
+
+#include <cstddef>
+#include <memory>
+
+namespace arm_compute
+{
+template <typename T>
+class LSTMParams
+{
+public:
+    /** Constructor */
+    LSTMParams()
+        : _input_to_input_weights(nullptr), _recurrent_to_input_weights(nullptr), _cell_to_input_weights(nullptr), _input_gate_bias(nullptr), _cell_to_forget_weights(nullptr),
+          _cell_to_output_weights(nullptr), _projection_weights(nullptr), _projection_bias(nullptr), _has_peephole_opt(false), _has_projection(false), _has_cifg_opt(true)
+    {
+    }
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    LSTMParams(const LSTMParams &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    LSTMParams &operator=(const LSTMParams &) = delete;
+    /** Default destructor */
+    ~LSTMParams() = default;
+    /** Set CIFG tensor parameters.
+     *
+     * @param[in] input_to_input_weights     2D weights tensor with dimensions [input_size, num_units]. Data types supported: F16/F32.
+     * @param[in] recurrent_to_input_weights 2D weights tensor with dimensions [output_size, num_units]. Data type supported: Same as @p input_to_input_weights.
+     * @param[in] cell_to_input_weights      1D weights tensor with dimensions [num_units]. Can be nullptr. Data type supported: Same as @p input_to_input_weights.
+     * @param[in] input_gate_bias            1D weights tensor with dimensions [num_units]. Data type supported: Same as @p input_to_input_weights
+     *
+     * @return Reference to this LSTMParams object
+     */
+    LSTMParams &set_cifg_params(const T *input_to_input_weights, const T *recurrent_to_input_weights, const T *cell_to_input_weights, const T *input_gate_bias)
+    {
+        _input_to_input_weights     = input_to_input_weights;
+        _recurrent_to_input_weights = recurrent_to_input_weights;
+        _cell_to_input_weights      = cell_to_input_weights;
+        _input_gate_bias            = input_gate_bias;
+        _has_cifg_opt               = false;
+        return *this;
+    }
+    /** Set projection tensor parameters.
+     *
+     * @param[in] projection_weights 2D weights tensor with dimensions [output_size, num_units]. Data type supported: Data types supported: F16/F32.
+     * @param[in] projection_bias    1D weights tensor with dimensions [output_size]. Data type supported: Same as @p projection_weights.
+     *
+     * @return Reference to this LSTMParams object
+     */
+    LSTMParams &set_projection_params(const T *projection_weights, const T *projection_bias)
+    {
+        _projection_weights = projection_weights;
+        _projection_bias    = projection_bias;
+        _has_projection     = true;
+        return *this;
+    }
+    /** Set peephole tensor parameters.
+     *
+     * @param[in] cell_to_forget_weights 1D weights tensor with dimensions [num_units]. Data type supported: Data types supported: F16/F32.
+     * @param[in] cell_to_output_weights 1D weights tensor with dimensions [num_units]. Data type supported: Same as @p cell_to_input_weights.
+     *
+     * @return Reference to this LSTMParams object
+     */
+    LSTMParams &set_peephole_params(const T *cell_to_forget_weights, const T *cell_to_output_weights)
+    {
+        _cell_to_forget_weights = cell_to_forget_weights;
+        _cell_to_output_weights = cell_to_output_weights;
+        _has_peephole_opt       = true;
+        return *this;
+    }
+
+    const T *input_to_input_weights() const
+    {
+        return _input_to_input_weights;
+    }
+
+    const T *recurrent_to_input_weights() const
+    {
+        return _recurrent_to_input_weights;
+    }
+
+    const T *cell_to_input_weights() const
+    {
+        return _cell_to_input_weights;
+    }
+
+    const T *input_gate_bias() const
+    {
+        return _input_gate_bias;
+    }
+
+    const T *cell_to_forget_weights() const
+    {
+        return _cell_to_forget_weights;
+    }
+
+    const T *cell_to_output_weights() const
+    {
+        return _cell_to_output_weights;
+    }
+
+    const T *projection_weights() const
+    {
+        return _projection_weights;
+    }
+
+    const T *projection_bias() const
+    {
+        return _projection_bias;
+    }
+
+    bool has_peephole_opt() const
+    {
+        return _has_peephole_opt;
+    }
+
+    bool has_projection() const
+    {
+        return _has_projection;
+    }
+
+    bool has_cifg_opt() const
+    {
+        return _has_cifg_opt;
+    }
+
+private:
+    const T *_input_to_input_weights;
+    const T *_recurrent_to_input_weights;
+    const T *_cell_to_input_weights;
+    const T *_input_gate_bias;
+    const T *_cell_to_forget_weights;
+    const T *_cell_to_output_weights;
+    const T *_projection_weights;
+    const T *_projection_bias;
+    bool     _has_peephole_opt;
+    bool     _has_projection;
+    bool     _has_cifg_opt;
+};
+}
+#endif /*__ARM_COMPUTE_LSTMPARAMS_H__ */
diff --git a/src/runtime/NEON/functions/NELSTMLayer.cpp b/src/runtime/NEON/functions/NELSTMLayer.cpp
new file mode 100644
index 0000000000..934761a8ef
--- /dev/null
+++ b/src/runtime/NEON/functions/NELSTMLayer.cpp
@@ -0,0 +1,545 @@
+/*
+ * Copyright (c) 2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/runtime/NEON/functions/NELSTMLayer.h"
+
+#include "arm_compute/core/PixelValue.h"
+#include "arm_compute/core/Utils.h"
+#include "arm_compute/core/Validate.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+#include "arm_compute/core/utils/quantization/AsymmHelpers.h"
+#include "arm_compute/runtime/common/LSTMParams.h"
+
+#include <cmath>
+#include <memory>
+#include <tuple>
+
+using namespace arm_compute;
+using namespace arm_compute::misc::shape_calculator;
+
+NELSTMLayer::NELSTMLayer(std::shared_ptr<IMemoryManager> memory_manager)
+    : _memory_group(std::move(memory_manager)), _fully_connected_input_gate(), _gemm_input_gate(), _transpose_input_gate(), _accum_input_gate1(), _accum_input_gate2(), _subtract_input_gate(),
+      _pixelwise_mul_input_gate(), _activation_input_gate(), _fully_connected_forget_gate(), _gemm_forget_gate(), _transpose_forget_gate(), _accum_forget_gate1(), _accum_forget_gate2(),
+      _pixelwise_mul_forget_gate(), _activation_forget_gate(), _fully_connected_cell_state(), _gemm_cell_state1(), _gemm_cell_state2(), _transpose_cell_state(), _accum_cell_state1(), _accum_cell_state2(),
+      _pixelwise_mul_cell_state1(), _activation_cell_state(), _cell_clip(), _pixelwise_mul_cell_state2(), _fully_connected_output(), _gemm_output(), _pixelwise_mul_output_state1(), _transpose_output(),
+      _accum_output1(), _accum_output2(), _activation_output(), _activation_output_state(), _pixelwise_mul_output_state2(), _fully_connected_output_state(), _gemm_output_state(), _accum_output_state(),
+      _projection_clip(), _copy_cell_state(), _copy_output(), _concat_scratch_buffer(), _input_gate_out1(), _input_gate_out2(), _input_gate_out3(), _input_gate_out4(), _input_gate_out5(),
+      _forget_gate_out1(), _forget_gate_out2(), _forget_gate_out3(), _forget_gate_out4(), _forget_gate_out5(), _cell_state_out1(), _cell_state_out2(), _cell_state_out3(), _cell_state_out4(),
+      _cell_state_out5(), _output1(), _output2(), _output3(), _output4(), _output5(), _cell_state_activation(), _output_state1(), _ones(), _run_peephole_opt(false), _run_cifg_opt(false),
+      _perform_cell_clipping(false), _has_projection_weights(false), _perform_projection_clipping(false)
+{
+}
+
+void NELSTMLayer::configure(const ITensor *input,
+                            const ITensor *input_to_forget_weights, const ITensor *input_to_cell_weights, const ITensor *input_to_output_weights,
+                            const ITensor *recurrent_to_forget_weights, const ITensor *recurrent_to_cell_weights, const ITensor *recurrent_to_output_weights,
+                            const ITensor *forget_gate_bias, const ITensor *cell_bias, const ITensor *output_gate_bias,
+                            const ITensor *output_state_in, const ITensor *cell_state_in,
+                            ITensor *scratch_buffer, ITensor *output_state_out, ITensor *cell_state_out, ITensor *output,
+                            const LSTMParams<ITensor> &lstm_params, const ActivationLayerInfo &activation_info, float cell_threshold, float projection_threshold)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input,
+                                 input_to_forget_weights, input_to_cell_weights, input_to_output_weights,
+                                 recurrent_to_forget_weights, recurrent_to_cell_weights, recurrent_to_output_weights,
+                                 forget_gate_bias, cell_bias, output_gate_bias,
+                                 output_state_in, cell_state_in,
+                                 scratch_buffer, output_state_out, cell_state_out, output);
+
+    // Set lstm parameters
+    LSTMParams<ITensorInfo> lstm_params_info;
+    if(lstm_params.has_peephole_opt())
+    {
+        lstm_params_info.set_peephole_params(lstm_params.cell_to_forget_weights()->info(), lstm_params.cell_to_output_weights()->info());
+    }
+    if(lstm_params.has_projection())
+    {
+        lstm_params_info.set_projection_params(lstm_params.projection_weights()->info(),
+                                               lstm_params.projection_bias() != nullptr ? lstm_params.projection_bias()->info() : nullptr);
+    }
+    if(!lstm_params.has_cifg_opt())
+    {
+        const ITensorInfo *cell_to_input_weights_info = (lstm_params.has_peephole_opt()) ? lstm_params.cell_to_input_weights()->info() : nullptr;
+        lstm_params_info.set_cifg_params(lstm_params.input_to_input_weights()->info(), lstm_params.recurrent_to_input_weights()->info(),
+                                         cell_to_input_weights_info, lstm_params.input_gate_bias()->info());
+    }
+
+    // Validate
+    ARM_COMPUTE_ERROR_THROW_ON(NELSTMLayer::validate(input->info(), input_to_forget_weights->info(),
+                                                     input_to_cell_weights->info(), input_to_output_weights->info(),
+                                                     recurrent_to_forget_weights->info(), recurrent_to_cell_weights->info(), recurrent_to_output_weights->info(),
+                                                     forget_gate_bias->info(), cell_bias->info(), output_gate_bias->info(),
+                                                     output_state_in->info(), cell_state_in->info(),
+                                                     scratch_buffer->info(), output_state_out->info(), cell_state_out->info(), output->info(),
+                                                     lstm_params_info, activation_info, cell_threshold, projection_threshold));
+
+    const TensorShape cell_state_shape = cell_state_in->info()->tensor_shape();
+
+    // Configure block that calculates the forget gate
+    // forget_gate = Activation(input * input_to_forget_weights + output_state_in * recurrent_to_forget_weights + PixelWiseMul(cell_state, cell_to_forget_weights) + forget_gate_bias)
+    TensorShape forget_gate1_shape = compute_transposed_shape(*recurrent_to_output_weights->info());
+    _forget_gate_out1.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+    _forget_gate_out2.allocator()->init(TensorInfo(forget_gate1_shape, 1, input->info()->data_type()));
+    _forget_gate_out3.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+    _forget_gate_out5.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+
+    _memory_group.manage(&_forget_gate_out1);
+    _fully_connected_forget_gate.configure(input, input_to_forget_weights, forget_gate_bias, &_forget_gate_out1);
+    _memory_group.manage(&_forget_gate_out2);
+    _transpose_forget_gate.configure(recurrent_to_forget_weights, &_forget_gate_out2);
+    _memory_group.manage(&_forget_gate_out3);
+    _gemm_forget_gate.configure(output_state_in, &_forget_gate_out2, nullptr, &_forget_gate_out3, 1.f, 0.f);
+    _forget_gate_out2.allocator()->allocate();
+    _memory_group.manage(&_forget_gate_out5);
+    _accum_forget_gate1.configure(&_forget_gate_out1, &_forget_gate_out3, &_forget_gate_out5, ConvertPolicy::SATURATE);
+    Tensor *forget_gate_out = &_forget_gate_out5;
+
+    if(lstm_params.has_peephole_opt())
+    {
+        _forget_gate_out4.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+
+        _run_peephole_opt = true;
+        _memory_group.manage(&_forget_gate_out4);
+        _pixelwise_mul_forget_gate.configure(cell_state_in, lstm_params.cell_to_forget_weights(), &_forget_gate_out4, 1, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO);
+        _accum_forget_gate2.configure(&_forget_gate_out5, &_forget_gate_out4, &_forget_gate_out3, ConvertPolicy::SATURATE);
+        _forget_gate_out4.allocator()->allocate();
+        _forget_gate_out5.allocator()->allocate();
+        forget_gate_out = &_forget_gate_out3;
+    }
+    else
+    {
+        _forget_gate_out3.allocator()->allocate();
+    }
+    _activation_forget_gate.configure(forget_gate_out, &_forget_gate_out1, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC));
+    forget_gate_out->allocator()->allocate();
+
+    // Configure block that calculates the input gate
+    // input_gate = Activation(input * input_to_input_weights + output_state * recurrent_to_input_weights + PixelWiseMul(cell_state, cell_to_input_weights) + input_gate_bias), without CIFG
+    // input_gate = 1 - forget_gate, with CIFG
+    _input_gate_out1.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+    if(lstm_params.has_cifg_opt())
+    {
+        _memory_group.manage(&_input_gate_out1);
+        _ones.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+        _subtract_input_gate.configure(&_ones, &_forget_gate_out1, &_input_gate_out1, ConvertPolicy::SATURATE);
+        _ones.allocator()->allocate();
+        _run_cifg_opt = true;
+    }
+    else
+    {
+        TensorShape input_gate_shape = compute_transposed_shape(*recurrent_to_output_weights->info());
+
+        _input_gate_out2.allocator()->init(TensorInfo(input_gate_shape, 1, input->info()->data_type()));
+        _input_gate_out3.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+        _input_gate_out4.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+        _input_gate_out5.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+
+        _memory_group.manage(&_input_gate_out1);
+        _fully_connected_input_gate.configure(input, lstm_params.input_to_input_weights(), lstm_params.input_gate_bias(), &_input_gate_out1);
+        _memory_group.manage(&_input_gate_out2);
+        _transpose_input_gate.configure(lstm_params.recurrent_to_input_weights(), &_input_gate_out2);
+        _memory_group.manage(&_input_gate_out3);
+        _gemm_input_gate.configure(output_state_in, &_input_gate_out2, nullptr, &_input_gate_out3, 1.f, 0.f);
+        _input_gate_out2.allocator()->allocate();
+        _memory_group.manage(&_input_gate_out4);
+        _accum_input_gate1.configure(&_input_gate_out1, &_input_gate_out3, &_input_gate_out4, ConvertPolicy::SATURATE);
+        if(_run_peephole_opt)
+        {
+            _memory_group.manage(&_input_gate_out5);
+            _pixelwise_mul_input_gate.configure(cell_state_in, lstm_params.cell_to_input_weights(), &_input_gate_out5, 1, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO);
+            _accum_input_gate2.configure(&_input_gate_out4, &_input_gate_out5, &_input_gate_out1, ConvertPolicy::SATURATE);
+            _input_gate_out5.allocator()->allocate();
+        }
+        _input_gate_out3.allocator()->allocate();
+        _input_gate_out4.allocator()->allocate();
+        _activation_input_gate.configure(&_input_gate_out1, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC));
+    }
+
+    // Configure block that calculates the cell state
+    // cell_state = Clip((PixelwiseMul(input_gate, Activation(input * input_to_cell_weights + output_state_in * recurrent_to_cell_weights + cell_bias)) + PixelwiseMul(forget_gate, cell_state)), cell_threshold)
+    TensorShape cell_state1_shape = compute_transposed_shape(*recurrent_to_output_weights->info());
+    _cell_state_out1.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+    _cell_state_out2.allocator()->init(TensorInfo(cell_state1_shape, 1, input->info()->data_type()));
+    _cell_state_out3.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+    _cell_state_out4.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+    _cell_state_out5.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+
+    _memory_group.manage(&_cell_state_out1);
+    _fully_connected_cell_state.configure(input, input_to_cell_weights, cell_bias, &_cell_state_out1);
+    _memory_group.manage(&_cell_state_out2);
+    _transpose_cell_state.configure(recurrent_to_cell_weights, &_cell_state_out2);
+    _memory_group.manage(&_cell_state_out3);
+    _gemm_cell_state1.configure(output_state_in, &_cell_state_out2, nullptr, &_cell_state_out3, 1.f, 0.f);
+    _cell_state_out2.allocator()->allocate();
+    _memory_group.manage(&_cell_state_out4);
+    _accum_cell_state1.configure(&_cell_state_out1, &_cell_state_out3, &_cell_state_out4, ConvertPolicy::SATURATE);
+    _activation_cell_state.configure(&_cell_state_out4, nullptr, activation_info);
+    _memory_group.manage(&_cell_state_out5);
+    _pixelwise_mul_cell_state1.configure(&_cell_state_out4, &_input_gate_out1, &_cell_state_out5, 1, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO);
+    _input_gate_out1.allocator()->allocate();
+    _cell_state_out4.allocator()->allocate();
+    _pixelwise_mul_cell_state2.configure(&_forget_gate_out1, cell_state_in, &_cell_state_out3, 1, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO);
+    _forget_gate_out1.allocator()->allocate();
+    _accum_cell_state2.configure(&_cell_state_out5, &_cell_state_out3, &_cell_state_out1, ConvertPolicy::SATURATE);
+    _cell_state_out3.allocator()->allocate();
+    _cell_state_out5.allocator()->allocate();
+    // Perform clipping
+    if(cell_threshold != 0.f)
+    {
+        _perform_cell_clipping = true;
+        _cell_clip.configure(&_cell_state_out1, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, -cell_threshold, cell_threshold));
+    }
+
+    // Configure block that calculates the output
+    // output_state_out = Activation(input * input_to_output_weights + output_state_in * recurrent_to_output_weights + PixelWiseMul(cell_state, cell_to_output_weights) + output_gate_bias)
+    TensorShape output1_shape = compute_transposed_shape(*recurrent_to_output_weights->info());
+    _output1.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+    _output2.allocator()->init(TensorInfo(output1_shape, 1, input->info()->data_type()));
+    _output3.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+    _output5.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+
+    _memory_group.manage(&_output1);
+    _fully_connected_output.configure(input, input_to_output_weights, output_gate_bias, &_output1);
+    _memory_group.manage(&_output2);
+    _transpose_output.configure(recurrent_to_output_weights, &_output2);
+    _memory_group.manage(&_output3);
+    _gemm_output.configure(output_state_in, &_output2, nullptr, &_output3, 1.f, 0.f);
+    _output2.allocator()->allocate();
+    _memory_group.manage(&_output5);
+    _accum_output1.configure(&_output1, &_output3, &_output5, ConvertPolicy::SATURATE);
+    _output3.allocator()->allocate();
+    Tensor *output_gate_out = &_output5;
+    if(lstm_params.has_peephole_opt())
+    {
+        _output4.allocator()->init(TensorInfo(_cell_state_out1.info()->tensor_shape(), 1, input->info()->data_type()));
+
+        _memory_group.manage(&_output4);
+        _pixelwise_mul_output_state1.configure(&_cell_state_out1, lstm_params.cell_to_output_weights(), &_output4, 1, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO);
+        _accum_output2.configure(&_output5, &_output4, &_output1, ConvertPolicy::SATURATE);
+        _output5.allocator()->allocate();
+        output_gate_out = &_output1;
+
+        // Allocate intermediate buffers
+        _output4.allocator()->allocate();
+    }
+    else
+    {
+        _output1.allocator()->allocate();
+    }
+    _activation_output.configure(output_gate_out, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC));
+    output_gate_out->allocator()->allocate();
+
+    // Configure block that calculates the output state
+    /** lstm_res = PixelwiseMul(output, Activation(cell_state))
+     *
+     *                      -- Clip(lstm_res * projection_weights + projection_bias, projection_threshold) , if there is a projection
+     *                     /
+     *  output_state =  --
+     *                     \
+     *                      -- lstm_res , otherwise
+     */
+    ITensor *output_state_out_tmp = lstm_params.has_projection() ? &_output_state1 : output_state_out;
+    _cell_state_activation.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+    _output_state1.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
+
+    _memory_group.manage(&_cell_state_activation);
+    _activation_output_state.configure(&_cell_state_out1, &_cell_state_activation, activation_info);
+    _pixelwise_mul_output_state2.configure(&_cell_state_activation, output_gate_out, output_state_out_tmp, 1, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO);
+    _cell_state_activation.allocator()->allocate();
+
+    if(lstm_params.has_projection())
+    {
+        _has_projection_weights = true;
+        _fully_connected_output_state.configure(output_state_out_tmp, lstm_params.projection_weights(), lstm_params.projection_bias(), output_state_out);
+        _output_state1.allocator()->allocate();
+        // Perform clipping
+        if(projection_threshold != 0.f)
+        {
+            _perform_projection_clipping = true;
+            _projection_clip.configure(output_state_out, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, -projection_threshold, projection_threshold));
+        }
+    }
+
+    // Copy cell state and output
+    _copy_cell_state.configure(&_cell_state_out1, cell_state_out);
+    _cell_state_out1.allocator()->allocate();
+    _copy_output.configure(output_state_out, output);
+
+    // Vector for holding the tensors to store in scratch buffer
+    std::vector<ITensor *> scratch_inputs;
+    if(lstm_params.has_cifg_opt())
+    {
+        scratch_inputs.emplace_back(&_input_gate_out1);
+    }
+    scratch_inputs.emplace_back(&_cell_state_out1);
+    scratch_inputs.emplace_back(forget_gate_out);
+    scratch_inputs.emplace_back(output_gate_out);
+    _concat_scratch_buffer.configure(scratch_inputs, scratch_buffer);
+}
+
+Status NELSTMLayer::validate(const ITensorInfo *input,
+                             const ITensorInfo *input_to_forget_weights, const ITensorInfo *input_to_cell_weights, const ITensorInfo *input_to_output_weights,
+                             const ITensorInfo *recurrent_to_forget_weights, const ITensorInfo *recurrent_to_cell_weights, const ITensorInfo *recurrent_to_output_weights,
+                             const ITensorInfo *forget_gate_bias, const ITensorInfo *cell_bias, const ITensorInfo *output_gate_bias,
+                             const ITensorInfo *output_state_in, const ITensorInfo *cell_state_in,
+                             const ITensorInfo *scratch_buffer, const ITensorInfo *output_state_out, const ITensorInfo *cell_state_out, const ITensorInfo *output,
+                             const LSTMParams<ITensorInfo> &lstm_params, const ActivationLayerInfo &activation_info, float cell_threshold, float projection_threshold)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input,
+                                        input_to_forget_weights, input_to_cell_weights, input_to_output_weights,
+                                        recurrent_to_forget_weights, recurrent_to_cell_weights, recurrent_to_output_weights,
+                                        forget_gate_bias, cell_bias, output_gate_bias,
+                                        output_state_in, cell_state_in,
+                                        scratch_buffer, output_state_out, cell_state_out, output);
+
+    // Check data types
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input,
+                                                       input_to_forget_weights, input_to_cell_weights, input_to_output_weights,
+                                                       recurrent_to_forget_weights, recurrent_to_cell_weights, recurrent_to_output_weights,
+                                                       forget_gate_bias, cell_bias, output_gate_bias,
+                                                       output_state_in, cell_state_in,
+                                                       scratch_buffer, output_state_out, cell_state_out, output);
+
+    // Check dimensions
+    ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 2);
+    ARM_COMPUTE_RETURN_ERROR_ON(input_to_forget_weights->num_dimensions() > 2);
+    ARM_COMPUTE_RETURN_ERROR_ON(input_to_cell_weights->num_dimensions() > 2);
+    ARM_COMPUTE_RETURN_ERROR_ON(input_to_output_weights->num_dimensions() > 2);
+    ARM_COMPUTE_RETURN_ERROR_ON(recurrent_to_forget_weights->num_dimensions() > 2);
+    ARM_COMPUTE_RETURN_ERROR_ON(recurrent_to_cell_weights->num_dimensions() > 2);
+    ARM_COMPUTE_RETURN_ERROR_ON(recurrent_to_output_weights->num_dimensions() > 2);
+    ARM_COMPUTE_RETURN_ERROR_ON(forget_gate_bias->num_dimensions() > 1);
+    ARM_COMPUTE_RETURN_ERROR_ON(cell_bias->num_dimensions() > 1);
+    ARM_COMPUTE_RETURN_ERROR_ON(output_gate_bias->num_dimensions() > 1);
+    ARM_COMPUTE_RETURN_ERROR_ON(output_state_in->num_dimensions() > 2);
+    ARM_COMPUTE_RETURN_ERROR_ON(cell_state_in->num_dimensions() > 2);
+    ARM_COMPUTE_RETURN_ERROR_ON(scratch_buffer->num_dimensions() > 2);
+    ARM_COMPUTE_RETURN_ERROR_ON(output_state_out->num_dimensions() > 2);
+    ARM_COMPUTE_RETURN_ERROR_ON(cell_state_out->num_dimensions() > 2);
+    ARM_COMPUTE_RETURN_ERROR_ON(output->num_dimensions() > 2);
+    ARM_COMPUTE_RETURN_ERROR_ON(cell_bias->dimension(0) * 4 != scratch_buffer->dimension(0)
+                                && cell_bias->dimension(0) * 3 != scratch_buffer->dimension(0));
+
+    const unsigned int num_batches = input->dimension(1);
+    const unsigned int num_cells   = input_to_output_weights->dimension(1);
+
+    // Check peephole optimization
+    if(lstm_params.has_peephole_opt())
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(lstm_params.cell_to_output_weights(), lstm_params.cell_to_forget_weights());
+        ARM_COMPUTE_RETURN_ERROR_ON(lstm_params.cell_to_forget_weights()->num_dimensions() > 1);
+        ARM_COMPUTE_RETURN_ERROR_ON(lstm_params.cell_to_output_weights()->num_dimensions() > 1);
+    }
+
+    TensorShape      units_out_transposed_shape = compute_transposed_shape(*recurrent_to_output_weights);
+    TensorShape      num_units_transposed_shape = compute_transposed_shape(*forget_gate_bias);
+    const TensorInfo units_out_transposed_info  = TensorInfo(units_out_transposed_shape, 1, input->data_type());
+    const TensorInfo num_units_transposed_info  = TensorInfo(num_units_transposed_shape, 1, input->data_type());
+
+    TensorInfo input_gate      = TensorInfo(TensorShape(num_cells, num_batches), 1, input->data_type());
+    TensorInfo forget_gate     = TensorInfo(TensorShape(num_cells, num_batches), 1, input->data_type());
+    TensorInfo output_gate_tmp = TensorInfo(TensorShape(num_cells, num_batches), 1, input->data_type());
+    TensorInfo cell_state_tmp  = TensorInfo(TensorShape(num_cells, num_batches), 1, input->data_type());
+
+    // Validate forget gate
+    ARM_COMPUTE_RETURN_ON_ERROR(NEFullyConnectedLayer::validate(input, input_to_forget_weights, forget_gate_bias, &forget_gate));
+    ARM_COMPUTE_RETURN_ON_ERROR(NEGEMM::validate(output_state_in, &units_out_transposed_info, nullptr, &forget_gate, 1.f, 0.f, GEMMInfo()));
+    ARM_COMPUTE_RETURN_ON_ERROR(NEArithmeticAdditionKernel::validate(&forget_gate, &forget_gate, &forget_gate, ConvertPolicy::SATURATE));
+    if(lstm_params.has_peephole_opt())
+    {
+        ARM_COMPUTE_RETURN_ON_ERROR(NEPixelWiseMultiplicationKernel::validate(cell_state_in, lstm_params.cell_to_forget_weights(), &forget_gate, 1, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO));
+        ARM_COMPUTE_RETURN_ON_ERROR(NEArithmeticAddition::validate(&forget_gate, &forget_gate, &forget_gate, ConvertPolicy::SATURATE));
+    }
+    ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayerKernel::validate(&forget_gate, &forget_gate, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC)));
+
+    // Validate input gate
+    if(!lstm_params.has_cifg_opt())
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(lstm_params.input_to_input_weights(),
+                                            lstm_params.recurrent_to_input_weights(),
+                                            lstm_params.input_gate_bias());
+        ARM_COMPUTE_RETURN_ERROR_ON(lstm_params.input_to_input_weights()->num_dimensions() > 2);
+        ARM_COMPUTE_RETURN_ERROR_ON(lstm_params.recurrent_to_input_weights()->num_dimensions() > 2);
+        ARM_COMPUTE_RETURN_ERROR_ON(lstm_params.input_gate_bias()->num_dimensions() > 1);
+
+        ARM_COMPUTE_RETURN_ON_ERROR(NEFullyConnectedLayer::validate(input, lstm_params.input_to_input_weights(), lstm_params.input_gate_bias(), &input_gate));
+        ARM_COMPUTE_RETURN_ON_ERROR(NEGEMM::validate(output_state_in, &units_out_transposed_info, nullptr, &input_gate, 1.f, 0.f, GEMMInfo()));
+        ARM_COMPUTE_RETURN_ON_ERROR(NEArithmeticAddition::validate(&input_gate, &input_gate, &input_gate, ConvertPolicy::SATURATE));
+        if(lstm_params.has_peephole_opt())
+        {
+            ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(lstm_params.cell_to_input_weights());
+            ARM_COMPUTE_RETURN_ERROR_ON(lstm_params.cell_to_input_weights()->num_dimensions() > 1);
+            ARM_COMPUTE_RETURN_ON_ERROR(NEPixelWiseMultiplicationKernel::validate(cell_state_in, lstm_params.cell_to_input_weights(), &input_gate, 1, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO));
+            ARM_COMPUTE_RETURN_ON_ERROR(NEArithmeticAddition::validate(&input_gate, &input_gate, &input_gate, ConvertPolicy::SATURATE));
+        }
+        ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayerKernel::validate(&input_gate, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC)));
+    }
+    else
+    {
+        ARM_COMPUTE_RETURN_ON_ERROR(NEArithmeticSubtractionKernel::validate(&forget_gate, &forget_gate, &forget_gate, ConvertPolicy::SATURATE));
+    }
+
+    // Validate cell state
+    ARM_COMPUTE_RETURN_ON_ERROR(NEFullyConnectedLayer::validate(input, input_to_cell_weights, cell_bias, &cell_state_tmp));
+    ARM_COMPUTE_RETURN_ON_ERROR(NEGEMM::validate(output_state_in, &units_out_transposed_info, nullptr, &cell_state_tmp, 1.f, 0.f, GEMMInfo()));
+    ARM_COMPUTE_RETURN_ON_ERROR(NEArithmeticAddition::validate(&cell_state_tmp, &cell_state_tmp, &cell_state_tmp, ConvertPolicy::SATURATE));
+    ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayerKernel::validate(&cell_state_tmp, nullptr, activation_info));
+    ARM_COMPUTE_RETURN_ON_ERROR(NEPixelWiseMultiplicationKernel::validate(&cell_state_tmp, &input_gate, &cell_state_tmp, 1, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO));
+    ARM_COMPUTE_RETURN_ON_ERROR(NEPixelWiseMultiplicationKernel::validate(&cell_state_tmp, &forget_gate, &cell_state_tmp, 1, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO));
+    ARM_COMPUTE_RETURN_ON_ERROR(NEArithmeticAddition::validate(&cell_state_tmp, &cell_state_tmp, &cell_state_tmp, ConvertPolicy::SATURATE));
+    if(cell_threshold != 0.f)
+    {
+        ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayerKernel::validate(&cell_state_tmp, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, -cell_threshold,
+                                                                                                                    cell_threshold)));
+    }
+
+    // Validate output gate tmp
+    ARM_COMPUTE_RETURN_ON_ERROR(NEFullyConnectedLayer::validate(input, input_to_output_weights, output_gate_bias, &output_gate_tmp));
+    ARM_COMPUTE_RETURN_ON_ERROR(NEGEMM::validate(output_state_in, &units_out_transposed_info, nullptr, &output_gate_tmp, 1.f, 0.f, GEMMInfo()));
+    ARM_COMPUTE_RETURN_ON_ERROR(NEArithmeticAddition::validate(&output_gate_tmp, &output_gate_tmp, &output_gate_tmp, ConvertPolicy::SATURATE));
+    if(lstm_params.has_peephole_opt())
+    {
+        ARM_COMPUTE_RETURN_ON_ERROR(NEPixelWiseMultiplicationKernel::validate(&cell_state_tmp, lstm_params.cell_to_output_weights(), &output_gate_tmp, 1, ConvertPolicy::SATURATE,
+                                                                              RoundingPolicy::TO_ZERO));
+        ARM_COMPUTE_RETURN_ON_ERROR(NEArithmeticAddition::validate(&output_gate_tmp, &output_gate_tmp, &output_gate_tmp, ConvertPolicy::SATURATE));
+    }
+    ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayerKernel::validate(&output_gate_tmp, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC)));
+
+    // Validate output state
+    ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayerKernel::validate(&cell_state_tmp, &cell_state_tmp, activation_info));
+    ARM_COMPUTE_RETURN_ON_ERROR(NEPixelWiseMultiplicationKernel::validate(&cell_state_tmp, &output_gate_tmp, &output_gate_tmp, 1, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO));
+    if(lstm_params.has_projection())
+    {
+        ARM_COMPUTE_RETURN_ON_ERROR(NEFullyConnectedLayer::validate(&output_gate_tmp, lstm_params.projection_weights(), lstm_params.projection_bias(), output_state_out));
+        if(projection_threshold != 0.f)
+        {
+            ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayerKernel::validate(output_state_out, output_state_out,
+                                                                          ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, -projection_threshold, projection_threshold)));
+        }
+    }
+
+    // Validate copy kernel
+    ARM_COMPUTE_RETURN_ON_ERROR(NECopyKernel::validate(&cell_state_tmp, cell_state_out));
+    ARM_COMPUTE_RETURN_ON_ERROR(NECopyKernel::validate(output_state_out, output));
+
+    // Validate scratch concatenation
+    std::vector<ITensorInfo *> inputs_vector_info_raw;
+    if(lstm_params.has_cifg_opt())
+    {
+        inputs_vector_info_raw.push_back(&input_gate);
+    }
+    inputs_vector_info_raw.push_back(&cell_state_tmp);
+    inputs_vector_info_raw.push_back(&forget_gate);
+    inputs_vector_info_raw.push_back(&output_gate_tmp);
+
+    ARM_COMPUTE_RETURN_ON_ERROR(NEWidthConcatenateLayer::validate(inputs_vector_info_raw, scratch_buffer));
+    return Status{};
+}
+
+void NELSTMLayer::run()
+{
+    _memory_group.acquire();
+
+    _fully_connected_forget_gate.run();
+    NEScheduler::get().schedule(&_transpose_forget_gate, Window::DimY);
+    _gemm_forget_gate.run();
+    NEScheduler::get().schedule(&_accum_forget_gate1, Window::DimY);
+
+    if(_run_peephole_opt)
+    {
+        NEScheduler::get().schedule(&_pixelwise_mul_forget_gate, Window::DimY);
+        _accum_forget_gate2.run();
+    }
+    NEScheduler::get().schedule(&_activation_forget_gate, Window::DimY);
+
+    if(_run_cifg_opt)
+    {
+        if(_ones.info()->data_type() == DataType::F16)
+        {
+            std::fill_n(reinterpret_cast<half *>(_ones.buffer()), _ones.info()->total_size() / _ones.info()->element_size(), 1);
+        }
+        else
+        {
+            std::fill_n(reinterpret_cast<float *>(_ones.buffer()), _ones.info()->total_size() / _ones.info()->element_size(), 1);
+        }
+        NEScheduler::get().schedule(&_subtract_input_gate, Window::DimY);
+    }
+    else
+    {
+        _fully_connected_input_gate.run();
+        NEScheduler::get().schedule(&_transpose_input_gate, Window::DimY);
+        _gemm_input_gate.run();
+        NEScheduler::get().schedule(&_accum_input_gate1, Window::DimY);
+        if(_run_peephole_opt)
+        {
+            NEScheduler::get().schedule(&_pixelwise_mul_input_gate, Window::DimY);
+            _accum_input_gate2.run();
+        }
+        NEScheduler::get().schedule(&_activation_input_gate, Window::DimY);
+    }
+
+    _fully_connected_cell_state.run();
+    NEScheduler::get().schedule(&_transpose_cell_state, Window::DimY);
+    _gemm_cell_state1.run();
+    NEScheduler::get().schedule(&_accum_cell_state1, Window::DimY);
+    NEScheduler::get().schedule(&_activation_cell_state, Window::DimY);
+    NEScheduler::get().schedule(&_pixelwise_mul_cell_state1, Window::DimY);
+    NEScheduler::get().schedule(&_pixelwise_mul_cell_state2, Window::DimY);
+    NEScheduler::get().schedule(&_accum_cell_state2, Window::DimY);
+
+    if(_perform_cell_clipping)
+    {
+        NEScheduler::get().schedule(&_cell_clip, Window::DimY);
+    }
+
+    _fully_connected_output.run();
+    NEScheduler::get().schedule(&_transpose_output, Window::DimY);
+    _gemm_output.run();
+    NEScheduler::get().schedule(&_accum_output1, Window::DimY);
+
+    if(_run_peephole_opt)
+    {
+        NEScheduler::get().schedule(&_pixelwise_mul_output_state1, Window::DimY);
+        _accum_output2.run();
+    }
+    NEScheduler::get().schedule(&_activation_output, Window::DimY);
+
+    NEScheduler::get().schedule(&_activation_output_state, Window::DimY);
+    NEScheduler::get().schedule(&_pixelwise_mul_output_state2, Window::DimY);
+
+    if(_has_projection_weights)
+    {
+        _fully_connected_output_state.run();
+        if(_perform_projection_clipping)
+        {
+            NEScheduler::get().schedule(&_projection_clip, Window::DimY);
+        }
+    }
+
+    NEScheduler::get().schedule(&_copy_cell_state, Window::DimY);
+    NEScheduler::get().schedule(&_copy_output, Window::DimY);
+
+    _concat_scratch_buffer.run();
+
+    _memory_group.release();
+}
\ No newline at end of file
diff --git a/tests/validation/NEON/LSTMLayer.cpp b/tests/validation/NEON/LSTMLayer.cpp
new file mode 100644
index 0000000000..3693d4f20d
--- /dev/null
+++ b/tests/validation/NEON/LSTMLayer.cpp
@@ -0,0 +1,181 @@
+/*
+ * Copyright (c) 2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/runtime/NEON/functions/NELSTMLayer.h"
+#include "tests/NEON/Accessor.h"
+#include "tests/PaddingCalculator.h"
+#include "tests/datasets/LSTMLayerDataset.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Macros.h"
+#include "tests/framework/datasets/Datasets.h"
+#include "tests/validation/Validation.h"
+#include "tests/validation/fixtures/LSTMLayerFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace
+{
+RelativeTolerance<float> tolerance_f32(0.001f);
+RelativeTolerance<half>  tolerance_f16(half(0.1));
+} // namespace
+
+TEST_SUITE(NEON)
+TEST_SUITE(LSTMLayer)
+
+// *INDENT-OFF*
+// clang-format off
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip(zip(zip(zip(
+               framework::dataset::make("InputInfo", { TensorInfo(TensorShape(8U, 2U), 1, DataType::U8),      // Wrong data type
+                                                       TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::F32), // Wrong input size
+                                                       TensorInfo(TensorShape(8U, 2U), 1, DataType::F32),     // Wrong input weights size
+                                                       TensorInfo(TensorShape(8U, 2U), 1, DataType::F32),     // Wrong recurrent weights size
+                                                       TensorInfo(TensorShape(8U, 2U), 1, DataType::F32),     // Wrong cell bias size
+                                                       TensorInfo(TensorShape(8U, 2U), 1, DataType::F32),     // Wrong cell state size
+                                                       TensorInfo(TensorShape(8U, 2U), 1, DataType::F32),     // Wrong output size
+                                                       TensorInfo(TensorShape(8U, 2U), 1, DataType::F32),     // Wrong scratch size
+               }),
+               framework::dataset::make("InputWeightsInfo", { TensorInfo(TensorShape(8U, 16U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(8U, 16U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(27U, 11U, 2U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(8U, 16U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(8U, 16U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(8U, 16U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(8U, 16U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(8U, 16U), 1, DataType::F32),
+               })),
+               framework::dataset::make("RecurrentWeightsInfo", { TensorInfo(TensorShape(16U, 16U), 1, DataType::F32),
+                                                                  TensorInfo(TensorShape(16U, 16U), 1, DataType::F32),
+                                                                  TensorInfo(TensorShape(16U, 16U), 1, DataType::F32),
+                                                                  TensorInfo(TensorShape(25U, 11U, 2U), 1, DataType::F32),
+                                                                  TensorInfo(TensorShape(16U, 16U), 1, DataType::F32),
+                                                                  TensorInfo(TensorShape(16U, 16U), 1, DataType::F32),
+                                                                  TensorInfo(TensorShape(16U, 16U), 1, DataType::F32),
+                                                                  TensorInfo(TensorShape(16U, 16U), 1, DataType::F32),
+               })),
+               framework::dataset::make("CellBiasInfo", { TensorInfo(TensorShape(16U), 1, DataType::F32),
+                                                      TensorInfo(TensorShape(16U), 1, DataType::F32),
+                                                      TensorInfo(TensorShape(16U), 1, DataType::F32),
+                                                      TensorInfo(TensorShape(16U), 1, DataType::F32),
+                                                      TensorInfo(TensorShape(30U), 1, DataType::F32),
+                                                      TensorInfo(TensorShape(16U), 1, DataType::F32),
+                                                      TensorInfo(TensorShape(16U), 1, DataType::F32),
+                                                      TensorInfo(TensorShape(16U), 1, DataType::F32),
+               })),
+               framework::dataset::make("ProjectionBiasInfo", { TensorInfo(TensorShape(16U), 1, DataType::F32),
+                                                      TensorInfo(TensorShape(16U), 1, DataType::F32),
+                                                      TensorInfo(TensorShape(16U), 1, DataType::F32),
+                                                      TensorInfo(TensorShape(16U), 1, DataType::F32),
+                                                      TensorInfo(TensorShape(16U), 1, DataType::F32),
+                                                      TensorInfo(TensorShape(16U), 1, DataType::F32),
+                                                      TensorInfo(TensorShape(16U), 1, DataType::F32),
+                                                      TensorInfo(TensorShape(16U), 1, DataType::F32),
+               })),
+               framework::dataset::make("CellStateInfo", { TensorInfo(TensorShape(16U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(16U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(16U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(16U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(16U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(11U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(16U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(16U, 2U), 1, DataType::F32),
+               })),
+               framework::dataset::make("OutputInfo", { TensorInfo(TensorShape(16U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(16U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(16U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(16U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(16U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(16U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(11U, 13U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(16U, 2U), 1, DataType::F32),
+               })),
+               framework::dataset::make("ScratchInfo", { TensorInfo(TensorShape(64U, 2U), 1, DataType::F32),
+                                                             TensorInfo(TensorShape(64U, 2U), 1, DataType::F32),
+                                                             TensorInfo(TensorShape(64U, 2U), 1, DataType::F32),
+                                                             TensorInfo(TensorShape(64U, 2U), 1, DataType::F32),
+                                                             TensorInfo(TensorShape(64U, 2U), 1, DataType::F32),
+                                                             TensorInfo(TensorShape(64U, 2U), 1, DataType::F32),
+                                                             TensorInfo(TensorShape(64U, 2U), 1, DataType::F32),
+                                                             TensorInfo(TensorShape(12U, 2U), 1, DataType::F32),
+               })),
+               framework::dataset::make("ActivationInfo", { ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU),
+                                                            ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU),
+                                                            ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU),
+                                                            ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU),
+                                                            ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU),
+                                                            ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU),
+                                                            ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU),
+                                                            ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU),
+               })),
+               framework::dataset::make("Expected", { false, false, false, false, false, false, false, false })),
+               input_info, input_weights_info, recurrent_weights_info, cell_bias_info, projection_bias_info, cell_state_info, output_info, scratch_info, info, expected)
+{
+    LSTMParams<ITensorInfo> lstm_params_info;
+    lstm_params_info.set_peephole_params(&cell_bias_info, &cell_bias_info)
+                    .set_projection_params(&recurrent_weights_info, &projection_bias_info)
+                    .set_cifg_params(&input_weights_info, &recurrent_weights_info, &cell_bias_info, &cell_bias_info);
+
+    ARM_COMPUTE_EXPECT(bool(NELSTMLayer::validate(&input_info.clone()->set_is_resizable(false), &input_weights_info.clone()->set_is_resizable(false), &input_weights_info.clone()->set_is_resizable(false),
+                                                  &input_weights_info.clone()->set_is_resizable(false), &recurrent_weights_info.clone()->set_is_resizable(false), &recurrent_weights_info.clone()->set_is_resizable(false),
+                                                  &recurrent_weights_info.clone()->set_is_resizable(false), &cell_bias_info.clone()->set_is_resizable(false), &cell_bias_info.clone()->set_is_resizable(false),
+                                                  &cell_bias_info.clone()->set_is_resizable(false),
+                                                  &output_info.clone()->set_is_resizable(false), &cell_state_info.clone()->set_is_resizable(false),
+                                                  &scratch_info.clone()->set_is_resizable(false), &output_info.clone()->set_is_resizable(false), &cell_state_info.clone()->set_is_resizable(false), &output_info.clone()->set_is_resizable(false),
+                                                  lstm_params_info, info, 0.05, 0.9)) == expected, framework::LogLevel::ERRORS);
+}
+// clang-format on
+// *INDENT-ON*
+
+template <typename T>
+using NELSTMLayerFixture = LSTMLayerValidationFixture<Tensor, Accessor, NELSTMLayer, LSTMParams<ITensor>, T>;
+
+TEST_SUITE(FP32)
+FIXTURE_DATA_TEST_CASE(RunSmall, NELSTMLayerFixture<float>, framework::DatasetMode::ALL, combine(combine(combine(datasets::SmallLSTMLayerDataset(), framework::dataset::make("DataType",
+                                                                                                                 DataType::F32)),
+                                                                                                         framework::dataset::make("ProjectionOpt", { true, false })),
+                                                                                                 framework::dataset::make("PeepholeOpt", { true, false })))
+{
+    // Validate output
+    validate(Accessor(_target), _reference, tolerance_f32);
+}
+TEST_SUITE_END() // FP32
+
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+TEST_SUITE(FP16)
+FIXTURE_DATA_TEST_CASE(RunSmall, NELSTMLayerFixture<half>, framework::DatasetMode::ALL, combine(combine(combine(datasets::SmallLSTMLayerDataset(), framework::dataset::make("DataType", DataType::F16)),
+                                                                                                        framework::dataset::make("ProjectionOpt", { true, false })),
+                                                                                                framework::dataset::make("PeepholeOpt", { true, false })))
+{
+    // Validate output
+    validate(Accessor(_target), _reference, tolerance_f16);
+}
+TEST_SUITE_END() // FP16
+#endif           /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
+TEST_SUITE_END() // LSTMLayer
+TEST_SUITE_END() // NEON
+} // namespace validation
+} // namespace test
+} // namespace arm_compute