COMPMID-1419: Make NEGEMMAssemblyDispatch dynamically typed instead of templated

This makes it easier to integrate in GEMMLowpMatrixMultiplyCore

Change-Id: Ibf80803f016a2e6a24d943ffafb50b48f04ec545
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/140868
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Jenkins <bsgcomp@arm.com>

28 files changed, 47 insertions, 12 deletions

diff --git a/src/core/NEON/kernels/NEActivationLayerKernel.cpp b/src/core/NEON/kernels/NEActivationLayerKernel.cpp
index bdc93ed1b8..1dad531a40 100644
--- a/src/core/NEON/kernels/NEActivationLayerKernel.cpp
+++ b/src/core/NEON/kernels/NEActivationLayerKernel.cpp
@@ -23,6 +23,7 @@
  */
 #include "arm_compute/core/NEON/kernels/NEActivationLayerKernel.h"
 
+#include "arm_compute/core/CPP/Validate.h"
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/ITensor.h"
 #include "arm_compute/core/NEON/NEAsymm.h"
@@ -44,7 +45,7 @@ namespace
 {
 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output)
 {
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input);
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::QASYMM8, DataType::F16, DataType::F32);
 
     // Checks performed when output is configured
diff --git a/src/core/NEON/kernels/NEArithmeticAdditionKernel.cpp b/src/core/NEON/kernels/NEArithmeticAdditionKernel.cpp
index f8e2b6d73e..a6102b159f 100644
--- a/src/core/NEON/kernels/NEArithmeticAdditionKernel.cpp
+++ b/src/core/NEON/kernels/NEArithmeticAdditionKernel.cpp
@@ -23,6 +23,7 @@
  */
 #include "arm_compute/core/NEON/kernels/NEArithmeticAdditionKernel.h"
 
+#include "arm_compute/core/CPP/Validate.h"
 #include "arm_compute/core/Error.h"
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/IAccessWindow.h"
@@ -330,6 +331,7 @@ Status validate_arguments(const ITensorInfo &input1, const ITensorInfo &input2,
 {
     ARM_COMPUTE_UNUSED(policy);
 
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(&input1);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input1, 1, DataType::U8, DataType::S16, DataType::F16, DataType::F32);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input2, 1, DataType::U8, DataType::S16, DataType::F16, DataType::F32);
 
diff --git a/src/core/NEON/kernels/NEArithmeticSubtractionKernel.cpp b/src/core/NEON/kernels/NEArithmeticSubtractionKernel.cpp
index 5a162e3b2c..3c76548b0a 100644
--- a/src/core/NEON/kernels/NEArithmeticSubtractionKernel.cpp
+++ b/src/core/NEON/kernels/NEArithmeticSubtractionKernel.cpp
@@ -23,6 +23,7 @@
  */
 #include "arm_compute/core/NEON/kernels/NEArithmeticSubtractionKernel.h"
 
+#include "arm_compute/core/CPP/Validate.h"
 #include "arm_compute/core/Error.h"
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/ITensor.h"
@@ -320,6 +321,7 @@ void sub_saturate_U8_U8_S16(const ITensor *in1, const ITensor *in2, ITensor *out
 inline Status validate_arguments(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ConvertPolicy policy)
 {
     ARM_COMPUTE_UNUSED(policy);
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input1);
     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input1, input2, output);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input1, 1, DataType::U8, DataType::S16, DataType::F16, DataType::F32);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input2, 1, DataType::U8, DataType::S16, DataType::F16, DataType::F32);
diff --git a/src/core/NEON/kernels/NEBatchNormalizationLayerKernel.cpp b/src/core/NEON/kernels/NEBatchNormalizationLayerKernel.cpp
index abfaa0cd26..ac1fc393c4 100644
--- a/src/core/NEON/kernels/NEBatchNormalizationLayerKernel.cpp
+++ b/src/core/NEON/kernels/NEBatchNormalizationLayerKernel.cpp
@@ -23,6 +23,7 @@
  */
 #include "arm_compute/core/NEON/kernels/NEBatchNormalizationLayerKernel.h"
 
+#include "arm_compute/core/CPP/Validate.h"
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/NEON/NEFixedPoint.h"
 #include "arm_compute/core/NEON/NEMath.h"
@@ -43,6 +44,7 @@ validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const IT
                    const ITensorInfo *beta, const ITensorInfo *gamma, float epsilon, ActivationLayerInfo act_info)
 {
     ARM_COMPUTE_UNUSED(epsilon);
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F16,
                                                          DataType::F32);
 
diff --git a/src/core/NEON/kernels/NECol2ImKernel.cpp b/src/core/NEON/kernels/NECol2ImKernel.cpp
index d09d174e4f..b9c7a9ac3b 100644
--- a/src/core/NEON/kernels/NECol2ImKernel.cpp
+++ b/src/core/NEON/kernels/NECol2ImKernel.cpp
@@ -50,6 +50,7 @@ TensorShape get_output_shape(const ITensorInfo *input, const Size2D &convolved_d
 
 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const Size2D &convolved_dims)
 {
+    //Note: ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input) is not needed here as this kernel doesn't use NEON FP16 instructions.
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::S8, DataType::QASYMM8,
                                                          DataType::U16, DataType::S16,
                                                          DataType::U32, DataType::S32,
diff --git a/src/core/NEON/kernels/NEConvertFullyConnectedWeightsKernel.cpp b/src/core/NEON/kernels/NEConvertFullyConnectedWeightsKernel.cpp
index 198565b1d5..b6d166d30e 100644
--- a/src/core/NEON/kernels/NEConvertFullyConnectedWeightsKernel.cpp
+++ b/src/core/NEON/kernels/NEConvertFullyConnectedWeightsKernel.cpp
@@ -66,6 +66,7 @@ void NEConvertFullyConnectedWeightsKernel::configure(const ITensor *input, ITens
 Status NEConvertFullyConnectedWeightsKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const TensorShape &original_input_shape,
                                                       DataLayout data_layout)
 {
+    //Note: ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input) is not needed here as this kernel doesn't use NEON FP16 instructions.
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1,
                                                          DataType::U8, DataType::S8, DataType::QASYMM8,
                                                          DataType::U16, DataType::S16,
diff --git a/src/core/NEON/kernels/NEDepthConcatenateLayerKernel.cpp b/src/core/NEON/kernels/NEDepthConcatenateLayerKernel.cpp
index 1b937b5be8..8c875cdb2d 100644
--- a/src/core/NEON/kernels/NEDepthConcatenateLayerKernel.cpp
+++ b/src/core/NEON/kernels/NEDepthConcatenateLayerKernel.cpp
@@ -95,6 +95,7 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, unsi
 Status validate_arguments(const ITensorInfo *input, unsigned int depth_offset, const ITensorInfo *output)
 {
     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
+    //Note: ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input) is not needed here as this kernel doesn't use NEON FP16 instructions.
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F16, DataType::F32);
     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
 
diff --git a/src/core/NEON/kernels/NEDepthwiseIm2ColKernel.cpp b/src/core/NEON/kernels/NEDepthwiseIm2ColKernel.cpp
index 91b29cdf03..92ee8d5809 100644
--- a/src/core/NEON/kernels/NEDepthwiseIm2ColKernel.cpp
+++ b/src/core/NEON/kernels/NEDepthwiseIm2ColKernel.cpp
@@ -42,6 +42,7 @@ namespace
 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, unsigned int depth_multiplier)
 {
     ARM_COMPUTE_UNUSED(conv_info);
+    //Note: ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input) is not needed here as this kernel doesn't use NEON FP16 instructions.
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F16, DataType::F32);
     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
     ARM_COMPUTE_RETURN_ERROR_ON(is_data_type_quantized_asymmetric(input->data_type()) && has_bias);
diff --git a/src/core/NEON/kernels/NEDepthwiseVectorToTensorKernel.cpp b/src/core/NEON/kernels/NEDepthwiseVectorToTensorKernel.cpp
index fe141bef56..2d17c237a3 100644
--- a/src/core/NEON/kernels/NEDepthwiseVectorToTensorKernel.cpp
+++ b/src/core/NEON/kernels/NEDepthwiseVectorToTensorKernel.cpp
@@ -24,6 +24,7 @@
 #include "arm_compute/core/NEON/kernels/NEDepthwiseVectorToTensorKernel.h"
 
 #include "arm_compute/core/AccessWindowTranspose.h"
+#include "arm_compute/core/CPP/Validate.h"
 #include "arm_compute/core/Coordinates.h"
 #include "arm_compute/core/Error.h"
 #include "arm_compute/core/Helpers.h"
@@ -43,6 +44,7 @@ namespace
 {
 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, size_t conv_w, size_t conv_h)
 {
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::S32, DataType::F16, DataType::F32);
 
     if(output->total_size() != 0)
diff --git a/src/core/NEON/kernels/NEDepthwiseWeightsReshapeKernel.cpp b/src/core/NEON/kernels/NEDepthwiseWeightsReshapeKernel.cpp
index 2c7a379c25..22a2cf8f2d 100644
--- a/src/core/NEON/kernels/NEDepthwiseWeightsReshapeKernel.cpp
+++ b/src/core/NEON/kernels/NEDepthwiseWeightsReshapeKernel.cpp
@@ -80,6 +80,7 @@ void weights_reshape(const ITensor *input, const ITensor *bias, ITensor *output,
 
 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *biases)
 {
+    //Note: ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input) is not needed here as this kernel doesn't use NEON FP16 instructions.
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F16, DataType::F32);
     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
     ARM_COMPUTE_RETURN_ERROR_ON(is_data_type_quantized_asymmetric(input->data_type()) && (biases != nullptr));
diff --git a/src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp b/src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp
index 54a046846a..59244c876c 100644
--- a/src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp
+++ b/src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp
@@ -25,6 +25,7 @@
 #include "arm_compute/core/NEON/kernels/detail/NEDirectConvolutionDetail.h"
 
 #include "arm_compute/core/AccessWindowStatic.h"
+#include "arm_compute/core/CPP/Validate.h"
 #include "arm_compute/core/Error.h"
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/IAccessWindow.h"
@@ -972,6 +973,7 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *weights,
 {
     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, weights, output);
     ARM_COMPUTE_RETURN_ERROR_ON(input->data_layout() == DataLayout::UNKNOWN);
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input);
     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, weights);
 
diff --git a/src/core/NEON/kernels/NEDirectConvolutionLayerOutputStageKernel.cpp b/src/core/NEON/kernels/NEDirectConvolutionLayerOutputStageKernel.cpp
index 6a373de1c3..eefbd98dd8 100644
--- a/src/core/NEON/kernels/NEDirectConvolutionLayerOutputStageKernel.cpp
+++ b/src/core/NEON/kernels/NEDirectConvolutionLayerOutputStageKernel.cpp
@@ -24,6 +24,7 @@
 #include "arm_compute/core/NEON/kernels/NEDirectConvolutionLayerOutputStageKernel.h"
 
 #include "arm_compute/core/AccessWindowStatic.h"
+#include "arm_compute/core/CPP/Validate.h"
 #include "arm_compute/core/Error.h"
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/ITensor.h"
@@ -43,7 +44,7 @@ namespace
 {
 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output)
 {
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input);
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input);
     ARM_COMPUTE_RETURN_ERROR_ON(input->data_layout() == DataLayout::UNKNOWN);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8,
                                                          DataType::F16,
diff --git a/src/core/NEON/kernels/NEFillBorderKernel.cpp b/src/core/NEON/kernels/NEFillBorderKernel.cpp
index 3d08cafa93..aef4d4865a 100644
--- a/src/core/NEON/kernels/NEFillBorderKernel.cpp
+++ b/src/core/NEON/kernels/NEFillBorderKernel.cpp
@@ -105,6 +105,7 @@ NEFillBorderKernel::NEFillBorderKernel()
 
 void NEFillBorderKernel::configure(ITensor *tensor, BorderSize border_size, BorderMode border_mode, const PixelValue &constant_border_value)
 {
+    //Note: ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input) is not needed here as this kernel doesn't use NEON FP16 instructions.
     ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(tensor, 1, DataType::U8, DataType::QASYMM8,
                                                   DataType::U16, DataType::S16,
                                                   DataType::U32, DataType::S32,
diff --git a/src/core/NEON/kernels/NEGEMMInterleave4x4Kernel.cpp b/src/core/NEON/kernels/NEGEMMInterleave4x4Kernel.cpp
index 6519a39b9c..5483602786 100644
--- a/src/core/NEON/kernels/NEGEMMInterleave4x4Kernel.cpp
+++ b/src/core/NEON/kernels/NEGEMMInterleave4x4Kernel.cpp
@@ -44,6 +44,7 @@ namespace
 {
 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output)
 {
+    //Note: ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input) is not needed here as this kernel doesn't use NEON FP16 instructions.
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::U8, DataType::S8,
                                                          DataType::U16, DataType::S16, DataType::U32, DataType::S32,
                                                          DataType::F16, DataType::F32);
diff --git a/src/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.cpp b/src/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.cpp
index 421a6f0ef9..42353ed0eb 100644
--- a/src/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.cpp
+++ b/src/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.cpp
@@ -24,6 +24,7 @@
 #include "arm_compute/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.h"
 
 #include "arm_compute/core/AccessWindowStatic.h"
+#include "arm_compute/core/CPP/Validate.h"
 #include "arm_compute/core/Error.h"
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/ITensor.h"
@@ -43,6 +44,7 @@ namespace
 {
 inline Status validate_arguments(const ITensorInfo *accum, const ITensorInfo *biases)
 {
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(accum);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(accum, 1, DataType::F16, DataType::F32);
     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(biases, accum);
     ARM_COMPUTE_RETURN_ERROR_ON(biases->num_dimensions() > 1);
diff --git a/src/core/NEON/kernels/NEGEMMMatrixAdditionKernel.cpp b/src/core/NEON/kernels/NEGEMMMatrixAdditionKernel.cpp
index d02504329a..cd6aa553db 100644
--- a/src/core/NEON/kernels/NEGEMMMatrixAdditionKernel.cpp
+++ b/src/core/NEON/kernels/NEGEMMMatrixAdditionKernel.cpp
@@ -23,6 +23,7 @@
  */
 #include "arm_compute/core/NEON/kernels/NEGEMMMatrixAdditionKernel.h"
 
+#include "arm_compute/core/CPP/Validate.h"
 #include "arm_compute/core/Error.h"
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/NEON/NEFixedPoint.h"
@@ -100,6 +101,7 @@ NEGEMMMatrixAdditionKernel::NEGEMMMatrixAdditionKernel()
 
 void NEGEMMMatrixAdditionKernel::configure(const ITensor *input, ITensor *output, float beta)
 {
+    ARM_COMPUTE_ERROR_ON_CPU_F16_UNSUPPORTED(input);
     ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F16, DataType::F32);
     ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::F16, DataType::F32);
     ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
diff --git a/src/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.cpp b/src/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.cpp
index 196398a2de..0ca24748af 100644
--- a/src/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.cpp
+++ b/src/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.cpp
@@ -25,6 +25,7 @@
 
 #include "arm_compute/core/AccessWindowStatic.h"
 #include "arm_compute/core/AccessWindowTranspose.h"
+#include "arm_compute/core/CPP/Validate.h"
 #include "arm_compute/core/Error.h"
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/IAccessWindow.h"
@@ -810,6 +811,7 @@ inline Status validate_arguments(const ITensorInfo *input0, const ITensorInfo *i
 {
     ARM_COMPUTE_UNUSED(alpha);
 
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input0);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input0, 1, DataType::F16, DataType::F32);
     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input0, input1, output);
 
diff --git a/src/core/NEON/kernels/NEGEMMTranspose1xWKernel.cpp b/src/core/NEON/kernels/NEGEMMTranspose1xWKernel.cpp
index 4517f46139..2e14e7a8c0 100644
--- a/src/core/NEON/kernels/NEGEMMTranspose1xWKernel.cpp
+++ b/src/core/NEON/kernels/NEGEMMTranspose1xWKernel.cpp
@@ -54,6 +54,7 @@ TensorShape get_output_shape(const ITensorInfo *input)
 
 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output)
 {
+    //Note: ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input) is not needed here as this kernel doesn't use NEON FP16 instructions.
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::U8, DataType::S8,
                                                          DataType::U16, DataType::S16, DataType::U32, DataType::S32,
                                                          DataType::F16, DataType::F32);
diff --git a/src/core/NEON/kernels/NEIm2ColKernel.cpp b/src/core/NEON/kernels/NEIm2ColKernel.cpp
index f03bc49ed3..16525ac22e 100644
--- a/src/core/NEON/kernels/NEIm2ColKernel.cpp
+++ b/src/core/NEON/kernels/NEIm2ColKernel.cpp
@@ -23,6 +23,7 @@
  */
 #include "arm_compute/core/NEON/kernels/NEIm2ColKernel.h"
 
+#include "arm_compute/core/CPP/Validate.h"
 #include "arm_compute/core/Error.h"
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/ITensor.h"
@@ -46,6 +47,7 @@ namespace
 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info,
                           bool has_bias, bool is_fully_connected, bool is_flatten, const Size2D &dilation)
 {
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F16, DataType::F32);
     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
     ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::QASYMM8 && has_bias);
diff --git a/src/core/NEON/kernels/NELocallyConnectedMatrixMultiplyKernel.cpp b/src/core/NEON/kernels/NELocallyConnectedMatrixMultiplyKernel.cpp
index 099626d259..4d3ec46e34 100644
--- a/src/core/NEON/kernels/NELocallyConnectedMatrixMultiplyKernel.cpp
+++ b/src/core/NEON/kernels/NELocallyConnectedMatrixMultiplyKernel.cpp
@@ -24,6 +24,7 @@
 #include "arm_compute/core/NEON/kernels/NELocallyConnectedMatrixMultiplyKernel.h"
 
 #include "arm_compute/core/AccessWindowTranspose.h"
+#include "arm_compute/core/CPP/Validate.h"
 #include "arm_compute/core/Error.h"
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/IAccessWindow.h"
@@ -305,7 +306,7 @@ void vector_matrix_multiply_f32(const ITensor *input0, const ITensor *input1, IT
 
 Status validate_arguments(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output)
 {
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input0, input1, output);
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input0);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input0, 1, DataType::F16, DataType::F32);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input1, 1, DataType::F16, DataType::F32);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::F16, DataType::F32);
diff --git a/src/core/NEON/kernels/NENormalizationLayerKernel.cpp b/src/core/NEON/kernels/NENormalizationLayerKernel.cpp
index 253a93f196..cb1996f33e 100644
--- a/src/core/NEON/kernels/NENormalizationLayerKernel.cpp
+++ b/src/core/NEON/kernels/NENormalizationLayerKernel.cpp
@@ -23,6 +23,7 @@
  */
 #include "arm_compute/core/NEON/kernels/NENormalizationLayerKernel.h"
 
+#include "arm_compute/core/CPP/Validate.h"
 #include "arm_compute/core/Error.h"
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/NEON/NEFixedPoint.h"
@@ -39,6 +40,7 @@ namespace
 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *input_squared, const ITensorInfo *output, const NormalizationLayerInfo &norm_info)
 {
     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, input_squared, output);
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input);
     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_squared);
diff --git a/src/core/NEON/kernels/NEPermuteKernel.cpp b/src/core/NEON/kernels/NEPermuteKernel.cpp
index e9bc8effc6..8d3fd88329 100644
--- a/src/core/NEON/kernels/NEPermuteKernel.cpp
+++ b/src/core/NEON/kernels/NEPermuteKernel.cpp
@@ -45,6 +45,7 @@ namespace
 {
 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const PermutationVector &perm)
 {
+    //Note: ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input) is not needed here as this kernel doesn't use NEON FP16 instructions.
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::S8, DataType::QASYMM8,
                                                          DataType::U16, DataType::S16,
                                                          DataType::U32, DataType::S32,
diff --git a/src/core/NEON/kernels/NEPixelWiseMultiplicationKernel.cpp b/src/core/NEON/kernels/NEPixelWiseMultiplicationKernel.cpp
index 0ec7e823a1..a4f51436b4 100644
--- a/src/core/NEON/kernels/NEPixelWiseMultiplicationKernel.cpp
+++ b/src/core/NEON/kernels/NEPixelWiseMultiplicationKernel.cpp
@@ -23,6 +23,7 @@
  */
 #include "arm_compute/core/NEON/kernels/NEPixelWiseMultiplicationKernel.h"
 
+#include "arm_compute/core/CPP/Validate.h"
 #include "arm_compute/core/Error.h"
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/IAccessWindow.h"
@@ -61,6 +62,7 @@ inline Status validate_arguments(const ITensorInfo *input1, const ITensorInfo *i
     ARM_COMPUTE_UNUSED(overflow_policy);
     ARM_COMPUTE_UNUSED(rounding_policy);
 
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input1);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input1, 1, DataType::U8, DataType::S16, DataType::F16, DataType::F32);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input2, 1, DataType::U8, DataType::S16, DataType::F16, DataType::F32);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::U8, DataType::S16, DataType::F16, DataType::F32);
@@ -607,4 +609,4 @@ BorderSize NEPixelWiseMultiplicationKernel::border_size() const
     const unsigned int replicateSize = _output->info()->dimension(0) - std::min(_input1->info()->dimension(0), _input2->info()->dimension(0));
     const unsigned int border        = std::min<unsigned int>(num_elems_processed_per_iteration - 1U, replicateSize);
     return BorderSize(0, border, 0, 0);
-}
\ No newline at end of file
+}
diff --git a/src/core/NEON/kernels/NEPoolingLayerKernel.cpp b/src/core/NEON/kernels/NEPoolingLayerKernel.cpp
index e586b72d30..2ca6090674 100644
--- a/src/core/NEON/kernels/NEPoolingLayerKernel.cpp
+++ b/src/core/NEON/kernels/NEPoolingLayerKernel.cpp
@@ -24,6 +24,7 @@
 #include "arm_compute/core/NEON/kernels/NEPoolingLayerKernel.h"
 
 #include "arm_compute/core/AccessWindowStatic.h"
+#include "arm_compute/core/CPP/Validate.h"
 #include "arm_compute/core/Error.h"
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/ITensor.h"
@@ -139,6 +140,7 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, c
     std::tie(pool_stride_x, pool_stride_y) = pad_stride_info.stride();
     static const std::set<int> supported_pool_sizes = { 2, 3 };
 
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F16, DataType::F32);
     ARM_COMPUTE_RETURN_ERROR_ON(pool_type == PoolingType::L2 && is_data_type_quantized(input->data_type()));
 
diff --git a/src/core/NEON/kernels/NEReshapeLayerKernel.cpp b/src/core/NEON/kernels/NEReshapeLayerKernel.cpp
index d6f470445f..a8a7440270 100644
--- a/src/core/NEON/kernels/NEReshapeLayerKernel.cpp
+++ b/src/core/NEON/kernels/NEReshapeLayerKernel.cpp
@@ -24,6 +24,7 @@
 #include "arm_compute/core/NEON/kernels/NEReshapeLayerKernel.h"
 
 #include "arm_compute/core/AccessWindowStatic.h"
+#include "arm_compute/core/CPP/Validate.h"
 #include "arm_compute/core/Error.h"
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/IAccessWindow.h"
@@ -59,6 +60,7 @@ inline void reshape_tensor(const Window &window, const ITensor *input, ITensor *
 
 void NEReshapeLayerKernel::configure(const ITensor *input, ITensor *output)
 {
+    ARM_COMPUTE_ERROR_ON_CPU_F16_UNSUPPORTED(input);
     ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::S8, DataType::QASYMM8, DataType::U16, DataType::S16,
                                                   DataType::U32, DataType::S32, DataType::F16, DataType::F32);
     ARM_COMPUTE_ERROR_ON_NULLPTR(output);
diff --git a/src/core/NEON/kernels/NESoftmaxLayerKernel.cpp b/src/core/NEON/kernels/NESoftmaxLayerKernel.cpp
index 9946f002de..4041b623b1 100644
--- a/src/core/NEON/kernels/NESoftmaxLayerKernel.cpp
+++ b/src/core/NEON/kernels/NESoftmaxLayerKernel.cpp
@@ -24,6 +24,7 @@
 #include "arm_compute/core/NEON/kernels/NESoftmaxLayerKernel.h"
 
 #include "arm_compute/core/AccessWindowStatic.h"
+#include "arm_compute/core/CPP/Validate.h"
 #include "arm_compute/core/Error.h"
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/ITensor.h"
@@ -320,11 +321,8 @@ namespace
 {
 Status validate_arguments_logits_1d_max(const ITensorInfo &input, const ITensorInfo &output)
 {
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(&input);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input, 1, DataType::QASYMM8, DataType::F16, DataType::F32);
-#else  /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input, 1, DataType::QASYMM8, DataType::F32);
-#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
 
     // Validate in case of configured output
     if(output.total_size() != 0)
@@ -486,11 +484,8 @@ Status validate_arguments_logits_softmax(const ITensorInfo &input, const ITensor
 {
     ARM_COMPUTE_UNUSED(beta);
     // Check input
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(&input);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input, 1, DataType::QASYMM8, DataType::F16, DataType::F32);
-#else  /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input, 1, DataType::QASYMM8, DataType::F32);
-#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
 
     const bool is_quantized_asymmetric = is_data_type_quantized_asymmetric(input.data_type());
 
diff --git a/src/core/NEON/kernels/NETransposeKernel.cpp b/src/core/NEON/kernels/NETransposeKernel.cpp
index 2630159561..32a5acd2f4 100644
--- a/src/core/NEON/kernels/NETransposeKernel.cpp
+++ b/src/core/NEON/kernels/NETransposeKernel.cpp
@@ -74,6 +74,7 @@ unsigned int num_elems_processed(size_t element_size)
 
 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output)
 {
+    //Note: ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input) is not needed here as this kernel doesn't use NEON FP16 instructions.
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::S8, DataType::QASYMM8, DataType::U16, DataType::S16, DataType::U32, DataType::S32,
                                                          DataType::F16,
                                                          DataType::F32);
diff --git a/src/core/NEON/kernels/NEWeightsReshapeKernel.cpp b/src/core/NEON/kernels/NEWeightsReshapeKernel.cpp
index f398409b26..2c9ad923aa 100644
--- a/src/core/NEON/kernels/NEWeightsReshapeKernel.cpp
+++ b/src/core/NEON/kernels/NEWeightsReshapeKernel.cpp
@@ -105,6 +105,7 @@ TensorShape get_output_shape(const ITensorInfo *input, bool has_bias)
 
 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *biases, const ITensorInfo *output)
 {
+    //Note: ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input) is not needed here as this kernel doesn't use NEON FP16 instructions.
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F16, DataType::F32);
     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output);