Apply clang-format on repository

Code is formatted as per a revised clang format configuration
file(not part of this delivery). Version 14.0.6 is used.

Exclusion List:
- files with .cl extension
- files that are not strictly C/C++ (e.g. Android.bp, Sconscript ...)
And the following directories
- compute_kernel_writer/validation/
- tests/
- include/
- src/core/NEON/kernels/convolution/
- src/core/NEON/kernels/arm_gemm/
- src/core/NEON/kernels/arm_conv/
- data/

There will be a follow up for formatting of .cl files and the
files under tests/ and compute_kernel_writer/validation/.

Signed-off-by: Felix Thomasmathibalan <felixjohnny.thomasmathibalan@arm.com>
Change-Id: Ib7eb1fcf4e7537b9feaefcfc15098a804a3fde0a
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/10391
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>

7 files changed, 575 insertions, 532 deletions

diff --git a/src/cpu/kernels/activation/generic/neon/fp16.cpp b/src/cpu/kernels/activation/generic/neon/fp16.cpp
index e51b5b3423..ddc6dc24cd 100644
--- a/src/cpu/kernels/activation/generic/neon/fp16.cpp
+++ b/src/cpu/kernels/activation/generic/neon/fp16.cpp
@@ -31,7 +31,7 @@ namespace cpu
 {
 namespace
 {
-constexpr ActFpImplParams Fp16Params = { static_cast<float16_t>(1e-7), 8 };
+constexpr ActFpImplParams Fp16Params = {static_cast<float16_t>(1e-7), 8};
 } // namespace
 
 void neon_fp16_activation(const ITensor *src, ITensor *dst, const ActivationLayerInfo &act_info, const Window &window)
@@ -40,4 +40,4 @@ void neon_fp16_activation(const ITensor *src, ITensor *dst, const ActivationLaye
 }
 } // namespace cpu
 } // namespace arm_compute
-#endif /* defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) && defined(ENABLE_FP16_KERNELS) */
\ No newline at end of file
+#endif /* defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) && defined(ENABLE_FP16_KERNELS) */
diff --git a/src/cpu/kernels/activation/generic/neon/fp32.cpp b/src/cpu/kernels/activation/generic/neon/fp32.cpp
index 2a3b8a0bfd..e558f8c73e 100644
--- a/src/cpu/kernels/activation/generic/neon/fp32.cpp
+++ b/src/cpu/kernels/activation/generic/neon/fp32.cpp
@@ -29,7 +29,7 @@ namespace cpu
 {
 namespace
 {
-constexpr ActFpImplParams Fp32Params = { static_cast<float>(1e-24), 4 };
+constexpr ActFpImplParams Fp32Params = {static_cast<float>(1e-24), 4};
 } // namespace
 void neon_fp32_activation(const ITensor *src, ITensor *dst, const ActivationLayerInfo &act_info, const Window &window)
 {
diff --git a/src/cpu/kernels/activation/generic/neon/impl.h b/src/cpu/kernels/activation/generic/neon/impl.h
index 05885d8476..afeb6f7f3d 100644
--- a/src/cpu/kernels/activation/generic/neon/impl.h
+++ b/src/cpu/kernels/activation/generic/neon/impl.h
@@ -24,6 +24,7 @@
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/Window.h"
 #include "arm_compute/function_info/ActivationLayerInfo.h"
+
 #include "src/core/NEON/wrapper/wrapper.h"
 namespace arm_compute
 {
@@ -56,10 +57,14 @@ inline float16x8_t mask_float_vector(const float16x8_t &in, const uint16x8_t &ma
 #endif /* __aarch64__ */
 
 template <typename T, const ActFpImplParams &P>
-void fp_neon_activation_impl(const ITensor *src, ITensor *dst, const ActivationLayerInfo &act_info, const Window &window)
+void fp_neon_activation_impl(const ITensor             *src,
+                             ITensor                   *dst,
+                             const ActivationLayerInfo &act_info,
+                             const Window              &window)
 {
     /** SIMD vector tag type. */
-    using ExactTagType                                           = typename arm_compute::wrapper::traits::neon_bitvector_tag_t<T, wrapper::traits::BitWidth::W128>;
+    using ExactTagType =
+        typename arm_compute::wrapper::traits::neon_bitvector_tag_t<T, wrapper::traits::BitWidth::W128>;
     constexpr int                                 window_step_x  = P.step_x;
     const auto                                    window_start_x = static_cast<int>(window.x().start());
     const auto                                    window_end_x   = static_cast<int>(window.x().end());
@@ -72,12 +77,12 @@ void fp_neon_activation_impl(const ITensor *src, ITensor *dst, const ActivationL
     // to prevent NAN values caused by zeros in inputs to SQRT.
     // In case of aarh64, we call vsqrt directly, so we don't use delta.
 #ifndef __aarch64__
-    const auto delta = wrapper::vdup_n(static_cast<T>(P.delta), ExactTagType {});
+    const auto delta = wrapper::vdup_n(static_cast<T>(P.delta), ExactTagType{});
 #else  /* #ifndef __aarch64__ */
-    const auto const_inv_2      = wrapper::vdup_n(static_cast<T>(0.5f), ExactTagType {});
+    const auto const_inv_2      = wrapper::vdup_n(static_cast<T>(0.5f), ExactTagType{});
     const auto const_inv_sqrt_2 = wrapper::vdup_n(static_cast<T>(0.70710678118f), ExactTagType{});
 #endif /* __aarch64__ */
-    const auto      const_1           = wrapper::vdup_n(static_cast<T>(1.f), ExactTagType {});
+    const auto      const_1           = wrapper::vdup_n(static_cast<T>(1.f), ExactTagType{});
     const auto      const_0           = wrapper::vdup_n(static_cast<T>(0.f), ExactTagType{});
     const auto      const_6           = wrapper::vdup_n(static_cast<T>(6.f), ExactTagType{});
     const auto      const_3           = wrapper::vdup_n(static_cast<T>(3.f), ExactTagType{});
@@ -88,143 +93,154 @@ void fp_neon_activation_impl(const ITensor *src, ITensor *dst, const ActivationL
     const auto      vb                = wrapper::vdup_n(static_cast<T>(act_info.b()), ExactTagType{});
     const auto      a                 = static_cast<T>(act_info.a());
     const auto      b                 = static_cast<T>(act_info.b());
-    execute_window_loop(win_collapsed, [&](const Coordinates &)
-    {
-        const auto input_ptr  = reinterpret_cast<const T *>(input.ptr());
-        const auto output_ptr = reinterpret_cast<T *>(output.ptr());
-        wrapper::traits::neon_bitvector_t<T, wrapper::traits::BitWidth::W128> tmp;
-        // Compute S elements per iteration
-        int x = window_start_x;
-        for(; x <= (window_end_x - window_step_x); x += window_step_x)
+    execute_window_loop(
+        win_collapsed,
+        [&](const Coordinates &)
         {
-            const auto vin = wrapper::vloadq(input_ptr + x);
-            switch(act)
+            const auto input_ptr  = reinterpret_cast<const T *>(input.ptr());
+            const auto output_ptr = reinterpret_cast<T *>(output.ptr());
+            wrapper::traits::neon_bitvector_t<T, wrapper::traits::BitWidth::W128> tmp;
+            // Compute S elements per iteration
+            int x = window_start_x;
+            for (; x <= (window_end_x - window_step_x); x += window_step_x)
             {
-                case ActivationLayerInfo::ActivationFunction::ABS:
-                    tmp = wrapper::vabs(vin);
-                    break;
-                case ActivationLayerInfo::ActivationFunction::LINEAR:
-                    tmp = wrapper::vmla(vb, va, vin);
-                    break;
-                case ActivationLayerInfo::ActivationFunction::LOGISTIC:
-                    tmp = wrapper::vinv(wrapper::vadd(const_1, wrapper::vexpq(wrapper::vneg(vin))));
-                    break;
-                case ActivationLayerInfo::ActivationFunction::RELU:
-                    tmp = wrapper::vmax(const_0, vin);
-                    break;
-                case ActivationLayerInfo::ActivationFunction::BOUNDED_RELU:
-                    tmp = wrapper::vmin(va, wrapper::vmax(const_0, vin));
-                    break;
-                case ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU:
-                    tmp = wrapper::vmin(va, wrapper::vmax(vb, vin));
-                    break;
-                case ActivationLayerInfo::ActivationFunction::LEAKY_RELU:
-                    tmp = wrapper::vbsl(wrapper::vcgt(vin, const_0), vin, wrapper::vmul(va, vin));
-                    break;
-                case ActivationLayerInfo::ActivationFunction::SOFT_RELU:
-                    tmp = wrapper::vbsl(wrapper::vcgt(vin, vsoft_relu_thresh), vin, wrapper::vlog(wrapper::vadd(const_1, wrapper::vexpq(vin))));
-                    break;
-                case ActivationLayerInfo::ActivationFunction::ELU:
-                    tmp = wrapper::vbsl(wrapper::vcge(vin, const_0), vin, wrapper::vmul(va, wrapper::vsub(wrapper::vexpq(vin), const_1)));
-                    break;
-                case ActivationLayerInfo::ActivationFunction::SQRT:
+                const auto vin = wrapper::vloadq(input_ptr + x);
+                switch (act)
+                {
+                    case ActivationLayerInfo::ActivationFunction::ABS:
+                        tmp = wrapper::vabs(vin);
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::LINEAR:
+                        tmp = wrapper::vmla(vb, va, vin);
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::LOGISTIC:
+                        tmp = wrapper::vinv(wrapper::vadd(const_1, wrapper::vexpq(wrapper::vneg(vin))));
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::RELU:
+                        tmp = wrapper::vmax(const_0, vin);
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::BOUNDED_RELU:
+                        tmp = wrapper::vmin(va, wrapper::vmax(const_0, vin));
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU:
+                        tmp = wrapper::vmin(va, wrapper::vmax(vb, vin));
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::LEAKY_RELU:
+                        tmp = wrapper::vbsl(wrapper::vcgt(vin, const_0), vin, wrapper::vmul(va, vin));
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::SOFT_RELU:
+                        tmp = wrapper::vbsl(wrapper::vcgt(vin, vsoft_relu_thresh), vin,
+                                            wrapper::vlog(wrapper::vadd(const_1, wrapper::vexpq(vin))));
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::ELU:
+                        tmp = wrapper::vbsl(wrapper::vcge(vin, const_0), vin,
+                                            wrapper::vmul(va, wrapper::vsub(wrapper::vexpq(vin), const_1)));
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::SQRT:
 #ifdef __aarch64__
-                    tmp = wrapper::vsqrt(vin);
+                        tmp = wrapper::vsqrt(vin);
 #else  /* __aarch64__ */
                     {
                         const auto bitmask = wrapper::vceq(vin, wrapper::vdup_n(0.f, ExactTagType{}));
-                        tmp                 = wrapper::vinv(wrapper::vinvsqrt(wrapper::vadd(vin, mask_float_vector(delta, bitmask))));
-                        tmp                 = mask_float_vector(tmp, wrapper::vnot(bitmask));
+                        tmp = wrapper::vinv(wrapper::vinvsqrt(wrapper::vadd(vin, mask_float_vector(delta, bitmask))));
+                        tmp = mask_float_vector(tmp, wrapper::vnot(bitmask));
                     }
 #endif /* __aarch64__ */
-                    break;
-                case ActivationLayerInfo::ActivationFunction::SQUARE:
-                    tmp = wrapper::vmul(vin, vin);
-                    break;
-                case ActivationLayerInfo::ActivationFunction::TANH:
-                    tmp = wrapper::vmul(va, wrapper::vtanh(wrapper::vmul(vb, vin)));
-                    break;
-                case ActivationLayerInfo::ActivationFunction::IDENTITY:
-                    tmp = vin;
-                    break;
-                case ActivationLayerInfo::ActivationFunction::HARD_SWISH:
-                    tmp = wrapper::vmul(vin, wrapper::vmul(const_inv_6, wrapper::vmin(const_6, wrapper::vmax(const_0, wrapper::vadd(vin, const_3)))));
-                    break;
-                case ActivationLayerInfo::ActivationFunction::SWISH:
-                    tmp = wrapper::vmul(vin, wrapper::vinv(wrapper::vadd(const_1, wrapper::vexpq(wrapper::vneg(wrapper::vmul(va, vin))))));
-                    break;
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::SQUARE:
+                        tmp = wrapper::vmul(vin, vin);
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::TANH:
+                        tmp = wrapper::vmul(va, wrapper::vtanh(wrapper::vmul(vb, vin)));
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::IDENTITY:
+                        tmp = vin;
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::HARD_SWISH:
+                        tmp = wrapper::vmul(
+                            vin,
+                            wrapper::vmul(const_inv_6,
+                                          wrapper::vmin(const_6, wrapper::vmax(const_0, wrapper::vadd(vin, const_3)))));
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::SWISH:
+                        tmp = wrapper::vmul(vin, wrapper::vinv(wrapper::vadd(
+                                                     const_1, wrapper::vexpq(wrapper::vneg(wrapper::vmul(va, vin))))));
+                        break;
 #ifdef __aarch64__
-                case ActivationLayerInfo::ActivationFunction::GELU:
-                    tmp = wrapper::vmul(vin, wrapper::vmul(const_inv_2, wrapper::vadd(const_1, wrapper::verf(wrapper::vmul(vin, const_inv_sqrt_2)))));
-                    break;
+                    case ActivationLayerInfo::ActivationFunction::GELU:
+                        tmp = wrapper::vmul(
+                            vin,
+                            wrapper::vmul(const_inv_2,
+                                          wrapper::vadd(const_1, wrapper::verf(wrapper::vmul(vin, const_inv_sqrt_2)))));
+                        break;
 #endif /* __aarch64__ */
-                default:
-                    ARM_COMPUTE_ERROR("Unsupported activation function");
+                    default:
+                        ARM_COMPUTE_ERROR("Unsupported activation function");
+                }
+                wrapper::vstore(output_ptr + x, tmp);
             }
-            wrapper::vstore(output_ptr + x, tmp);
-        }
-        // Compute left-over elements
-        for(; x < window_end_x; ++x)
-        {
-            const T in = *(reinterpret_cast<const T *>(input_ptr + x));
-            T       tmp;
-            switch(act)
+            // Compute left-over elements
+            for (; x < window_end_x; ++x)
             {
-                case ActivationLayerInfo::ActivationFunction::ABS:
-                    tmp = std::abs(in);
-                    break;
-                case ActivationLayerInfo::ActivationFunction::LINEAR:
-                    tmp = a * in + b;
-                    break;
-                case ActivationLayerInfo::ActivationFunction::LOGISTIC:
-                    tmp = static_cast<T>(1) / (static_cast<T>(1) + std::exp(-in));
-                    break;
-                case ActivationLayerInfo::ActivationFunction::RELU:
-                    tmp = std::max<T>(static_cast<T>(0), in);
-                    break;
-                case ActivationLayerInfo::ActivationFunction::BOUNDED_RELU:
-                    tmp = std::min<T>(a, std::max(static_cast<T>(0), in));
-                    break;
-                case ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU:
-                    tmp = std::min<T>(a, std::max<T>(b, in));
-                    break;
-                case ActivationLayerInfo::ActivationFunction::LEAKY_RELU:
-                    tmp = (in > 0) ? in : a * in;
-                    break;
-                case ActivationLayerInfo::ActivationFunction::SOFT_RELU:
-                    tmp = (in > soft_relu_thresh) ? in : std::log(static_cast<T>(1) + std::exp(in));
-                    break;
-                case ActivationLayerInfo::ActivationFunction::ELU:
-                    tmp = (in >= 0) ? in : a * (std::exp(in) - 1);
-                    break;
-                case ActivationLayerInfo::ActivationFunction::SQRT:
-                    tmp = std::sqrt(in);
-                    break;
-                case ActivationLayerInfo::ActivationFunction::SQUARE:
-                    tmp = in * in;
-                    break;
-                case ActivationLayerInfo::ActivationFunction::TANH:
-                    tmp = a * std::tanh(b * in);
-                    break;
-                case ActivationLayerInfo::ActivationFunction::IDENTITY:
-                    tmp = in;
-                    break;
-                case ActivationLayerInfo::ActivationFunction::HARD_SWISH:
-                    tmp = in * ((std::min(std::max((in + 3), 0.0f), 6.0f)) * 0.166666667f);
-                    break;
-                case ActivationLayerInfo::ActivationFunction::SWISH:
-                    tmp = in / (static_cast<T>(1) + std::exp(-a * in));
-                    break;
-                case ActivationLayerInfo::ActivationFunction::GELU:
-                    tmp = in * static_cast<T>(0.5f * (1.0f + erff(static_cast<float>(in) / 1.41421356237f)));
-                    break;
-                default:
-                    ARM_COMPUTE_ERROR("Unsupported activation function");
+                const T in = *(reinterpret_cast<const T *>(input_ptr + x));
+                T       tmp;
+                switch (act)
+                {
+                    case ActivationLayerInfo::ActivationFunction::ABS:
+                        tmp = std::abs(in);
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::LINEAR:
+                        tmp = a * in + b;
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::LOGISTIC:
+                        tmp = static_cast<T>(1) / (static_cast<T>(1) + std::exp(-in));
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::RELU:
+                        tmp = std::max<T>(static_cast<T>(0), in);
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::BOUNDED_RELU:
+                        tmp = std::min<T>(a, std::max(static_cast<T>(0), in));
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU:
+                        tmp = std::min<T>(a, std::max<T>(b, in));
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::LEAKY_RELU:
+                        tmp = (in > 0) ? in : a * in;
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::SOFT_RELU:
+                        tmp = (in > soft_relu_thresh) ? in : std::log(static_cast<T>(1) + std::exp(in));
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::ELU:
+                        tmp = (in >= 0) ? in : a * (std::exp(in) - 1);
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::SQRT:
+                        tmp = std::sqrt(in);
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::SQUARE:
+                        tmp = in * in;
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::TANH:
+                        tmp = a * std::tanh(b * in);
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::IDENTITY:
+                        tmp = in;
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::HARD_SWISH:
+                        tmp = in * ((std::min(std::max((in + 3), 0.0f), 6.0f)) * 0.166666667f);
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::SWISH:
+                        tmp = in / (static_cast<T>(1) + std::exp(-a * in));
+                        break;
+                    case ActivationLayerInfo::ActivationFunction::GELU:
+                        tmp = in * static_cast<T>(0.5f * (1.0f + erff(static_cast<float>(in) / 1.41421356237f)));
+                        break;
+                    default:
+                        ARM_COMPUTE_ERROR("Unsupported activation function");
+                }
+                *(output_ptr + x) = tmp;
             }
-            *(output_ptr + x) = tmp;
-        }
-    },
-    input, output);
+        },
+        input, output);
 }
 } // namespace cpu
 } // namespace arm_compute
diff --git a/src/cpu/kernels/activation/generic/neon/lut.cpp b/src/cpu/kernels/activation/generic/neon/lut.cpp
index c973e964e4..f289c80d4b 100644
--- a/src/cpu/kernels/activation/generic/neon/lut.cpp
+++ b/src/cpu/kernels/activation/generic/neon/lut.cpp
@@ -24,6 +24,7 @@
 
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/function_info/ActivationLayerInfo.h"
+
 #include "src/cpu/kernels/lut/list.h"
 
 namespace arm_compute
@@ -33,19 +34,22 @@ namespace cpu
 #ifdef __aarch64__
 void neon_q8_activation_lut(const ITensor *src, ITensor *dst, const ActivationLayerInfo &act_info, const Window &window)
 {
-    ARM_COMPUTE_ERROR_ON(src->info()->data_type() != DataType::QASYMM8 && src->info()->data_type() != DataType::QASYMM8_SIGNED);
+    ARM_COMPUTE_ERROR_ON(src->info()->data_type() != DataType::QASYMM8 &&
+                         src->info()->data_type() != DataType::QASYMM8_SIGNED);
     const auto window_end_x  = window.x().end();
     Window     win_collapsed = window.collapse_if_possible(window, Window::DimZ);
     win_collapsed.set(Window::DimX, Window::Dimension(0, 1, 1));
     Iterator input(src, win_collapsed);
     Iterator output(dst, win_collapsed);
-    execute_window_loop(win_collapsed, [&](const Coordinates &)
-    {
-        const auto input_ptr  = reinterpret_cast<const uint8_t *>(input.ptr());
-        auto       output_ptr = reinterpret_cast<uint8_t *>(output.ptr());
-        lut_u8_neon(act_info.lut().data(), 1u, window_end_x, &input_ptr, &output_ptr);
-    },
-    input, output);
+    execute_window_loop(
+        win_collapsed,
+        [&](const Coordinates &)
+        {
+            const auto input_ptr  = reinterpret_cast<const uint8_t *>(input.ptr());
+            auto       output_ptr = reinterpret_cast<uint8_t *>(output.ptr());
+            lut_u8_neon(act_info.lut().data(), 1u, window_end_x, &input_ptr, &output_ptr);
+        },
+        input, output);
 }
 #endif // __aarch64__
 } // namespace cpu
diff --git a/src/cpu/kernels/activation/generic/neon/qasymm8.cpp b/src/cpu/kernels/activation/generic/neon/qasymm8.cpp
index e7c146e46f..1451301ea2 100644
--- a/src/cpu/kernels/activation/generic/neon/qasymm8.cpp
+++ b/src/cpu/kernels/activation/generic/neon/qasymm8.cpp
@@ -25,6 +25,7 @@
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/Window.h"
 #include "arm_compute/function_info/ActivationLayerInfo.h"
+
 #include "src/core/NEON/NEAsymm.h"
 #include "src/core/NEON/NEMath.h"
 #include "src/core/NEON/wrapper/wrapper.h"
@@ -38,7 +39,10 @@ namespace arm_compute
 {
 namespace cpu
 {
-void neon_qasymm8_activation(const ITensor *src, ITensor *dst, const ActivationLayerInfo &act_info, const Window &window)
+void neon_qasymm8_activation(const ITensor             *src,
+                             ITensor                   *dst,
+                             const ActivationLayerInfo &act_info,
+                             const Window              &window)
 {
     constexpr int                                 window_step_x  = 16;
     const auto                                    window_start_x = static_cast<int>(window.x().start());
@@ -85,206 +89,222 @@ void neon_qasymm8_activation(const ITensor *src, ITensor *dst, const ActivationL
     float32x4_t vs = vdupq_n_f32(s);
     float32x4_t vo = vdupq_n_f32(o);
 
-    execute_window_loop(win_collapsed, [&](const Coordinates &)
-    {
-        const auto input_ptr  = reinterpret_cast<const qasymm8_t *>(input.ptr());
-        const auto output_ptr = reinterpret_cast<qasymm8_t *>(output.ptr());
+    execute_window_loop(
+        win_collapsed,
+        [&](const Coordinates &)
+        {
+            const auto input_ptr  = reinterpret_cast<const qasymm8_t *>(input.ptr());
+            const auto output_ptr = reinterpret_cast<qasymm8_t *>(output.ptr());
 
-        wrapper::traits::neon_bitvector_t<qasymm8_t, wrapper::traits::BitWidth::W128> tmp;
+            wrapper::traits::neon_bitvector_t<qasymm8_t, wrapper::traits::BitWidth::W128> tmp;
 
-        // Compute S elements per iteration
-        int x = window_start_x;
-        for(; x <= (window_end_x - window_step_x); x += window_step_x)
-        {
-            const auto vin = wrapper::vloadq(input_ptr + x);
-            if(act == ActivationLayerInfo::ActivationFunction::RELU)
-            {
-                // Perform activation
-                tmp = vmaxq_u8(vconst_0, vin);
-                // Re-quantize to new output space
-                tmp = vmlaq_qasymm8<RoundingPolicy::TO_NEAREST_UP>(tmp, vs, vo);
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::BOUNDED_RELU)
-            {
-                // Perform activation
-                tmp = vminq_u8(va, vmaxq_u8(vconst_0, vin));
-                // Re-quantize to new output space
-                tmp = vmlaq_qasymm8<RoundingPolicy::TO_NEAREST_UP>(tmp, vs, vo);
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU)
+            // Compute S elements per iteration
+            int x = window_start_x;
+            for (; x <= (window_end_x - window_step_x); x += window_step_x)
             {
-                // Perform activation
-                tmp = vminq_u8(va, vmaxq_u8(vb, vin));
-                // Re-quantize to new output space
-                tmp = vmlaq_qasymm8<RoundingPolicy::TO_NEAREST_UP>(tmp, vs, vo);
-            }
+                const auto vin = wrapper::vloadq(input_ptr + x);
+                if (act == ActivationLayerInfo::ActivationFunction::RELU)
+                {
+                    // Perform activation
+                    tmp = vmaxq_u8(vconst_0, vin);
+                    // Re-quantize to new output space
+                    tmp = vmlaq_qasymm8<RoundingPolicy::TO_NEAREST_UP>(tmp, vs, vo);
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::BOUNDED_RELU)
+                {
+                    // Perform activation
+                    tmp = vminq_u8(va, vmaxq_u8(vconst_0, vin));
+                    // Re-quantize to new output space
+                    tmp = vmlaq_qasymm8<RoundingPolicy::TO_NEAREST_UP>(tmp, vs, vo);
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU)
+                {
+                    // Perform activation
+                    tmp = vminq_u8(va, vmaxq_u8(vb, vin));
+                    // Re-quantize to new output space
+                    tmp = vmlaq_qasymm8<RoundingPolicy::TO_NEAREST_UP>(tmp, vs, vo);
+                }
 #ifndef __aarch64__ // LUT-based implementation is used for aarch64 instead.
-            else if(act == ActivationLayerInfo::ActivationFunction::LOGISTIC)
-            {
-                // De-quantize
-                const auto vin_deq = vdequantize(vin, qi_in);
-                // Perform activation
-                const float32x4x4_t tmp_dep =
+                else if (act == ActivationLayerInfo::ActivationFunction::LOGISTIC)
                 {
-                    {
+                    // De-quantize
+                    const auto vin_deq = vdequantize(vin, qi_in);
+                    // Perform activation
+                    const float32x4x4_t tmp_dep = {{
                         wrapper::vdiv(vconst_1, wrapper::vadd(vconst_1, wrapper::vexpq(wrapper::vneg(vin_deq.val[0])))),
                         wrapper::vdiv(vconst_1, wrapper::vadd(vconst_1, wrapper::vexpq(wrapper::vneg(vin_deq.val[1])))),
                         wrapper::vdiv(vconst_1, wrapper::vadd(vconst_1, wrapper::vexpq(wrapper::vneg(vin_deq.val[2])))),
                         wrapper::vdiv(vconst_1, wrapper::vadd(vconst_1, wrapper::vexpq(wrapper::vneg(vin_deq.val[3])))),
-                    }
-                };
-                // Re-quantize to new output space
-                tmp = vquantize(tmp_dep, qi_out);
-            }
+                    }};
+                    // Re-quantize to new output space
+                    tmp = vquantize(tmp_dep, qi_out);
+                }
 #endif // __aarch64__
-            else if(act == ActivationLayerInfo::ActivationFunction::TANH)
-            {
-                // De-quantize
-                const auto vin_deq = vdequantize(vin, qi_in);
-                // Perform activation
-                const float32x4x4_t tmp_dep =
+                else if (act == ActivationLayerInfo::ActivationFunction::TANH)
                 {
-                    {
+                    // De-quantize
+                    const auto vin_deq = vdequantize(vin, qi_in);
+                    // Perform activation
+                    const float32x4x4_t tmp_dep = {{
                         wrapper::vmul(va_f32, wrapper::vtanh(wrapper::vmul(vin_deq.val[0], vb_f32))),
                         wrapper::vmul(va_f32, wrapper::vtanh(wrapper::vmul(vin_deq.val[1], vb_f32))),
                         wrapper::vmul(va_f32, wrapper::vtanh(wrapper::vmul(vin_deq.val[2], vb_f32))),
                         wrapper::vmul(va_f32, wrapper::vtanh(wrapper::vmul(vin_deq.val[3], vb_f32))),
-                    }
-                };
-                // Re-quantize to new output space
-                tmp = vquantize(tmp_dep, qi_out);
-            }
+                    }};
+                    // Re-quantize to new output space
+                    tmp = vquantize(tmp_dep, qi_out);
+                }
 #ifndef __aarch64__ // LUT-based implementation is used for aarch64 instead.
-            else if(act == ActivationLayerInfo::ActivationFunction::HARD_SWISH)
-            {
-                // De-quantize
-                const auto vin_deq = vdequantize(vin, qi_in);
-                // Perform activation
-                const float32x4x4_t tmp_dep =
+                else if (act == ActivationLayerInfo::ActivationFunction::HARD_SWISH)
                 {
-                    {
-                        wrapper::vmul(vin_deq.val[0], wrapper::vmul(const_inv_6_f32, wrapper::vmin(const_6_f32, wrapper::vmax(const_0_f32, wrapper::vadd(vin_deq.val[0], const_3_f32))))),
-                        wrapper::vmul(vin_deq.val[1], wrapper::vmul(const_inv_6_f32, wrapper::vmin(const_6_f32, wrapper::vmax(const_0_f32, wrapper::vadd(vin_deq.val[1], const_3_f32))))),
-                        wrapper::vmul(vin_deq.val[2], wrapper::vmul(const_inv_6_f32, wrapper::vmin(const_6_f32, wrapper::vmax(const_0_f32, wrapper::vadd(vin_deq.val[2], const_3_f32))))),
-                        wrapper::vmul(vin_deq.val[3], wrapper::vmul(const_inv_6_f32, wrapper::vmin(const_6_f32, wrapper::vmax(const_0_f32, wrapper::vadd(vin_deq.val[3], const_3_f32))))),
-                    }
-                };
-                // Re-quantize to new output space
-                tmp = vquantize(tmp_dep, qi_out);
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::LEAKY_RELU)
-            {
-                const auto vin_deq = vdequantize(vin, qi_in);
-
-                const uint32x4x4_t pos_mask =
+                    // De-quantize
+                    const auto vin_deq = vdequantize(vin, qi_in);
+                    // Perform activation
+                    const float32x4x4_t tmp_dep = {{
+                        wrapper::vmul(
+                            vin_deq.val[0],
+                            wrapper::vmul(
+                                const_inv_6_f32,
+                                wrapper::vmin(const_6_f32,
+                                              wrapper::vmax(const_0_f32, wrapper::vadd(vin_deq.val[0], const_3_f32))))),
+                        wrapper::vmul(
+                            vin_deq.val[1],
+                            wrapper::vmul(
+                                const_inv_6_f32,
+                                wrapper::vmin(const_6_f32,
+                                              wrapper::vmax(const_0_f32, wrapper::vadd(vin_deq.val[1], const_3_f32))))),
+                        wrapper::vmul(
+                            vin_deq.val[2],
+                            wrapper::vmul(
+                                const_inv_6_f32,
+                                wrapper::vmin(const_6_f32,
+                                              wrapper::vmax(const_0_f32, wrapper::vadd(vin_deq.val[2], const_3_f32))))),
+                        wrapper::vmul(
+                            vin_deq.val[3],
+                            wrapper::vmul(
+                                const_inv_6_f32,
+                                wrapper::vmin(const_6_f32,
+                                              wrapper::vmax(const_0_f32, wrapper::vadd(vin_deq.val[3], const_3_f32))))),
+                    }};
+                    // Re-quantize to new output space
+                    tmp = vquantize(tmp_dep, qi_out);
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::LEAKY_RELU)
                 {
-                    {
+                    const auto vin_deq = vdequantize(vin, qi_in);
+
+                    const uint32x4x4_t pos_mask = {{
                         wrapper::vcgt(vin_deq.val[0], vconst_0_f32),
                         wrapper::vcgt(vin_deq.val[1], vconst_0_f32),
                         wrapper::vcgt(vin_deq.val[2], vconst_0_f32),
                         wrapper::vcgt(vin_deq.val[3], vconst_0_f32),
-                    }
-                };
+                    }};
 
-                const float32x4x4_t tmp_dep =
-                {
-                    {
+                    const float32x4x4_t tmp_dep = {{
                         wrapper::vbsl(pos_mask.val[0], vin_deq.val[0], wrapper::vmul(va_f32, vin_deq.val[0])),
                         wrapper::vbsl(pos_mask.val[1], vin_deq.val[1], wrapper::vmul(va_f32, vin_deq.val[1])),
                         wrapper::vbsl(pos_mask.val[2], vin_deq.val[2], wrapper::vmul(va_f32, vin_deq.val[2])),
                         wrapper::vbsl(pos_mask.val[3], vin_deq.val[3], wrapper::vmul(va_f32, vin_deq.val[3])),
-                    }
-                };
+                    }};
 
-                tmp = vquantize(tmp_dep, qi_out);
-            }
+                    tmp = vquantize(tmp_dep, qi_out);
+                }
 #else  // #ifndef __aarch64__
-            else if (act == ActivationLayerInfo::ActivationFunction::GELU)
-            {
-                const auto vin_deq = vdequantize(vin, qi_in);
-                // Perform activation
-                const float32x4x4_t tmp_dep =
+                else if (act == ActivationLayerInfo::ActivationFunction::GELU)
                 {
-                    {
-                        wrapper::vmul(vin_deq.val[0], wrapper::vmul(const_inv_2, wrapper::vadd(vconst_1, wrapper::verf(wrapper::vmul(vin_deq.val[0], const_inv_sqrt_2))))),
-                        wrapper::vmul(vin_deq.val[1], wrapper::vmul(const_inv_2, wrapper::vadd(vconst_1, wrapper::verf(wrapper::vmul(vin_deq.val[1], const_inv_sqrt_2))))),
-                        wrapper::vmul(vin_deq.val[2], wrapper::vmul(const_inv_2, wrapper::vadd(vconst_1, wrapper::verf(wrapper::vmul(vin_deq.val[2], const_inv_sqrt_2))))),
-                        wrapper::vmul(vin_deq.val[3], wrapper::vmul(const_inv_2, wrapper::vadd(vconst_1, wrapper::verf(wrapper::vmul(vin_deq.val[3], const_inv_sqrt_2))))),
-                    }
-                };
-                // Re-quantize to new output space
-                tmp = vquantize(tmp_dep, qi_out);
-            }
+                    const auto vin_deq = vdequantize(vin, qi_in);
+                    // Perform activation
+                    const float32x4x4_t tmp_dep = {{
+                        wrapper::vmul(vin_deq.val[0],
+                                      wrapper::vmul(const_inv_2,
+                                                    wrapper::vadd(vconst_1, wrapper::verf(wrapper::vmul(
+                                                                                vin_deq.val[0], const_inv_sqrt_2))))),
+                        wrapper::vmul(vin_deq.val[1],
+                                      wrapper::vmul(const_inv_2,
+                                                    wrapper::vadd(vconst_1, wrapper::verf(wrapper::vmul(
+                                                                                vin_deq.val[1], const_inv_sqrt_2))))),
+                        wrapper::vmul(vin_deq.val[2],
+                                      wrapper::vmul(const_inv_2,
+                                                    wrapper::vadd(vconst_1, wrapper::verf(wrapper::vmul(
+                                                                                vin_deq.val[2], const_inv_sqrt_2))))),
+                        wrapper::vmul(vin_deq.val[3],
+                                      wrapper::vmul(const_inv_2,
+                                                    wrapper::vadd(vconst_1, wrapper::verf(wrapper::vmul(
+                                                                                vin_deq.val[3], const_inv_sqrt_2))))),
+                    }};
+                    // Re-quantize to new output space
+                    tmp = vquantize(tmp_dep, qi_out);
+                }
 #endif // __aarch64__
-            else
-            {
-                ARM_COMPUTE_ERROR("Unsupported activation function");
+                else
+                {
+                    ARM_COMPUTE_ERROR("Unsupported activation function");
+                }
+                wrapper::vstore(output_ptr + x, tmp);
             }
-            wrapper::vstore(output_ptr + x, tmp);
-        }
 
-        // Compute left-over elements
-        for(; x < window_end_x; ++x)
-        {
-            qasymm8_t in  = *(reinterpret_cast<const qasymm8_t *>(input_ptr + x));
-            qasymm8_t tmp = 0;
-            if(act == ActivationLayerInfo::ActivationFunction::RELU)
-            {
-                tmp = std::max(const_0, in);
-                tmp = utility::clamp<int32_t, qasymm8_t>(support::cpp11::lround(tmp * s + o));
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::BOUNDED_RELU)
+            // Compute left-over elements
+            for (; x < window_end_x; ++x)
             {
-                tmp = std::min(a, std::max(const_0, in));
-                tmp = utility::clamp<int32_t, qasymm8_t>(support::cpp11::lround(tmp * s + o));
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU)
-            {
-                tmp = std::min(a, std::max(b, in));
-                tmp = utility::clamp<int32_t, qasymm8_t>(support::cpp11::lround(tmp * s + o));
-            }
+                qasymm8_t in  = *(reinterpret_cast<const qasymm8_t *>(input_ptr + x));
+                qasymm8_t tmp = 0;
+                if (act == ActivationLayerInfo::ActivationFunction::RELU)
+                {
+                    tmp = std::max(const_0, in);
+                    tmp = utility::clamp<int32_t, qasymm8_t>(support::cpp11::lround(tmp * s + o));
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::BOUNDED_RELU)
+                {
+                    tmp = std::min(a, std::max(const_0, in));
+                    tmp = utility::clamp<int32_t, qasymm8_t>(support::cpp11::lround(tmp * s + o));
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU)
+                {
+                    tmp = std::min(a, std::max(b, in));
+                    tmp = utility::clamp<int32_t, qasymm8_t>(support::cpp11::lround(tmp * s + o));
+                }
 #ifndef __aarch64__ // LUT-based implementation is used for aarch64 instead.
-            else if(act == ActivationLayerInfo::ActivationFunction::LOGISTIC)
-            {
-                float tmp_f = dequantize_qasymm8(in, qi_in);
-                tmp_f       = 1.f / (1.f + std::exp(-tmp_f));
-                tmp         = quantize_qasymm8(tmp_f, qi_out);
-            }
+                else if (act == ActivationLayerInfo::ActivationFunction::LOGISTIC)
+                {
+                    float tmp_f = dequantize_qasymm8(in, qi_in);
+                    tmp_f       = 1.f / (1.f + std::exp(-tmp_f));
+                    tmp         = quantize_qasymm8(tmp_f, qi_out);
+                }
 #endif // __aarch64__
-            else if(act == ActivationLayerInfo::ActivationFunction::TANH)
-            {
-                float tmp_f = dequantize_qasymm8(in, qi_in);
-                tmp_f       = a_f32 * std::tanh(b_f32 * tmp_f);
-                tmp         = quantize_qasymm8(tmp_f, qi_out);
-            }
+                else if (act == ActivationLayerInfo::ActivationFunction::TANH)
+                {
+                    float tmp_f = dequantize_qasymm8(in, qi_in);
+                    tmp_f       = a_f32 * std::tanh(b_f32 * tmp_f);
+                    tmp         = quantize_qasymm8(tmp_f, qi_out);
+                }
 #ifndef __aarch64__ // LUT-based implementation is used for aarch64 instead.
-            else if(act == ActivationLayerInfo::ActivationFunction::HARD_SWISH)
-            {
-                float tmp_f = dequantize_qasymm8(in, qi_in);
-                tmp_f       = tmp_f * ((std::min(std::max((tmp_f + 3), 0.0f), 6.0f)) * 0.166666667f);
-                tmp         = quantize_qasymm8(tmp_f, qi_out);
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::LEAKY_RELU)
-            {
-                float tmp_f = dequantize_qasymm8(in, qi_in);
-                tmp_f       = tmp_f > 0 ? tmp_f : tmp_f * a_f32;
-                tmp         = quantize_qasymm8(tmp_f, qi_out);
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::GELU)
-            {
-                float tmp_f = dequantize_qasymm8(in, qi_in);
-                tmp         = tmp_f * 0.5f * (1.0f + std::erff(in / 1.41421356237f));
-                tmp         = quantize_qasymm8(tmp_f, qi_out);
-            }
+                else if (act == ActivationLayerInfo::ActivationFunction::HARD_SWISH)
+                {
+                    float tmp_f = dequantize_qasymm8(in, qi_in);
+                    tmp_f       = tmp_f * ((std::min(std::max((tmp_f + 3), 0.0f), 6.0f)) * 0.166666667f);
+                    tmp         = quantize_qasymm8(tmp_f, qi_out);
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::LEAKY_RELU)
+                {
+                    float tmp_f = dequantize_qasymm8(in, qi_in);
+                    tmp_f       = tmp_f > 0 ? tmp_f : tmp_f * a_f32;
+                    tmp         = quantize_qasymm8(tmp_f, qi_out);
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::GELU)
+                {
+                    float tmp_f = dequantize_qasymm8(in, qi_in);
+                    tmp         = tmp_f * 0.5f * (1.0f + std::erff(in / 1.41421356237f));
+                    tmp         = quantize_qasymm8(tmp_f, qi_out);
+                }
 #endif // __aarch64__
-            else
-            {
-                ARM_COMPUTE_ERROR("Unsupported activation function");
+                else
+                {
+                    ARM_COMPUTE_ERROR("Unsupported activation function");
+                }
+                *(output_ptr + x) = tmp;
             }
-            *(output_ptr + x) = tmp;
-        }
-    },
-    input, output);
+        },
+        input, output);
 }
 } // namespace cpu
 } // namespace arm_compute
diff --git a/src/cpu/kernels/activation/generic/neon/qasymm8_signed.cpp b/src/cpu/kernels/activation/generic/neon/qasymm8_signed.cpp
index 52c396459b..a2f588245a 100644
--- a/src/cpu/kernels/activation/generic/neon/qasymm8_signed.cpp
+++ b/src/cpu/kernels/activation/generic/neon/qasymm8_signed.cpp
@@ -24,6 +24,7 @@
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/Window.h"
 #include "arm_compute/function_info/ActivationLayerInfo.h"
+
 #include "src/core/NEON/NEAsymm.h"
 #include "src/core/NEON/NEMath.h"
 #include "src/core/NEON/wrapper/wrapper.h"
@@ -36,7 +37,10 @@ namespace arm_compute
 {
 namespace cpu
 {
-void neon_qasymm8_signed_activation(const ITensor *src, ITensor *dst, const ActivationLayerInfo &act_info, const Window &window)
+void neon_qasymm8_signed_activation(const ITensor             *src,
+                                    ITensor                   *dst,
+                                    const ActivationLayerInfo &act_info,
+                                    const Window              &window)
 {
     constexpr int                                 window_step_x  = 16;
     const auto                                    window_start_x = static_cast<int>(window.x().start());
@@ -76,191 +80,195 @@ void neon_qasymm8_signed_activation(const ITensor *src, ITensor *dst, const Acti
     float32x4_t vs = vdupq_n_f32(s);
     float32x4_t vo = vdupq_n_f32(o);
 
-    execute_window_loop(win_collapsed, [&](const Coordinates &)
-    {
-        const auto input_ptr  = reinterpret_cast<const qasymm8_signed_t *>(input.ptr());
-        const auto output_ptr = reinterpret_cast<qasymm8_signed_t *>(output.ptr());
+    execute_window_loop(
+        win_collapsed,
+        [&](const Coordinates &)
+        {
+            const auto input_ptr  = reinterpret_cast<const qasymm8_signed_t *>(input.ptr());
+            const auto output_ptr = reinterpret_cast<qasymm8_signed_t *>(output.ptr());
 
-        wrapper::traits::neon_bitvector_t<qasymm8_signed_t, wrapper::traits::BitWidth::W128> tmp;
+            wrapper::traits::neon_bitvector_t<qasymm8_signed_t, wrapper::traits::BitWidth::W128> tmp;
 
-        // Compute S elements per iteration
-        int x = window_start_x;
-        for(; x <= (window_end_x - window_step_x); x += window_step_x)
-        {
-            const auto vin = wrapper::vloadq(input_ptr + x);
-            if(act == ActivationLayerInfo::ActivationFunction::RELU)
-            {
-                // Perform activation
-                tmp = vmaxq_s8(vconst_0, vin);
-                // Re-quantize to new output space
-                tmp = vmlaq_qasymm8_signed<RoundingPolicy::TO_NEAREST_UP>(tmp, vs, vo);
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::BOUNDED_RELU)
-            {
-                // Perform activation
-                tmp = vminq_s8(va, vmaxq_s8(vconst_0, vin));
-                // Re-quantize to new output space
-                tmp = vmlaq_qasymm8_signed<RoundingPolicy::TO_NEAREST_UP>(tmp, vs, vo);
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU)
+            // Compute S elements per iteration
+            int x = window_start_x;
+            for (; x <= (window_end_x - window_step_x); x += window_step_x)
             {
-                // Perform activation
-                tmp = vminq_s8(va, vmaxq_s8(vb, vin));
-                // Re-quantize to new output space
-                tmp = vmlaq_qasymm8_signed<RoundingPolicy::TO_NEAREST_UP>(tmp, vs, vo);
-            }
+                const auto vin = wrapper::vloadq(input_ptr + x);
+                if (act == ActivationLayerInfo::ActivationFunction::RELU)
+                {
+                    // Perform activation
+                    tmp = vmaxq_s8(vconst_0, vin);
+                    // Re-quantize to new output space
+                    tmp = vmlaq_qasymm8_signed<RoundingPolicy::TO_NEAREST_UP>(tmp, vs, vo);
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::BOUNDED_RELU)
+                {
+                    // Perform activation
+                    tmp = vminq_s8(va, vmaxq_s8(vconst_0, vin));
+                    // Re-quantize to new output space
+                    tmp = vmlaq_qasymm8_signed<RoundingPolicy::TO_NEAREST_UP>(tmp, vs, vo);
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU)
+                {
+                    // Perform activation
+                    tmp = vminq_s8(va, vmaxq_s8(vb, vin));
+                    // Re-quantize to new output space
+                    tmp = vmlaq_qasymm8_signed<RoundingPolicy::TO_NEAREST_UP>(tmp, vs, vo);
+                }
 #ifndef __aarch64__ // LUT-based implementation is used for aarch64 instead.
-            else if(act == ActivationLayerInfo::ActivationFunction::LOGISTIC)
-            {
-                // De-quantize
-                const auto vin_deq = vdequantize(vin, qi_in);
-                // Perform activation
-                const float32x4x4_t tmp_dep =
+                else if (act == ActivationLayerInfo::ActivationFunction::LOGISTIC)
                 {
-                    {
+                    // De-quantize
+                    const auto vin_deq = vdequantize(vin, qi_in);
+                    // Perform activation
+                    const float32x4x4_t tmp_dep = {{
                         wrapper::vdiv(vconst_1, wrapper::vadd(vconst_1, wrapper::vexpq(wrapper::vneg(vin_deq.val[0])))),
                         wrapper::vdiv(vconst_1, wrapper::vadd(vconst_1, wrapper::vexpq(wrapper::vneg(vin_deq.val[1])))),
                         wrapper::vdiv(vconst_1, wrapper::vadd(vconst_1, wrapper::vexpq(wrapper::vneg(vin_deq.val[2])))),
                         wrapper::vdiv(vconst_1, wrapper::vadd(vconst_1, wrapper::vexpq(wrapper::vneg(vin_deq.val[3])))),
-                    }
-                };
-                // Re-quantize to new output space
-                tmp = vquantize_signed(tmp_dep, qi_out);
-            }
+                    }};
+                    // Re-quantize to new output space
+                    tmp = vquantize_signed(tmp_dep, qi_out);
+                }
 #endif // __aarch64__
-            else if(act == ActivationLayerInfo::ActivationFunction::TANH)
-            {
-                // De-quantize
-                const auto vin_deq = vdequantize(vin, qi_in);
-                // Perform activation
-                const float32x4x4_t tmp_dep =
+                else if (act == ActivationLayerInfo::ActivationFunction::TANH)
                 {
-                    {
+                    // De-quantize
+                    const auto vin_deq = vdequantize(vin, qi_in);
+                    // Perform activation
+                    const float32x4x4_t tmp_dep = {{
                         wrapper::vmul(va_f32, wrapper::vtanh(wrapper::vmul(vin_deq.val[0], vb_f32))),
                         wrapper::vmul(va_f32, wrapper::vtanh(wrapper::vmul(vin_deq.val[1], vb_f32))),
                         wrapper::vmul(va_f32, wrapper::vtanh(wrapper::vmul(vin_deq.val[2], vb_f32))),
                         wrapper::vmul(va_f32, wrapper::vtanh(wrapper::vmul(vin_deq.val[3], vb_f32))),
-                    }
-                };
-                // Re-quantize to new output space
-                tmp = vquantize_signed(tmp_dep, qi_out);
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::HARD_SWISH)
-            {
-                // De-quantize
-                const auto vin_deq = vdequantize(vin, qi_in);
-                // Perform activation
-                const float32x4x4_t tmp_dep =
+                    }};
+                    // Re-quantize to new output space
+                    tmp = vquantize_signed(tmp_dep, qi_out);
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::HARD_SWISH)
                 {
-                    {
-                        wrapper::vmul(vin_deq.val[0], wrapper::vmul(const_inv_6_f32, wrapper::vmin(const_6_f32, wrapper::vmax(const_0_f32, wrapper::vadd(vin_deq.val[0], const_3_f32))))),
-                        wrapper::vmul(vin_deq.val[1], wrapper::vmul(const_inv_6_f32, wrapper::vmin(const_6_f32, wrapper::vmax(const_0_f32, wrapper::vadd(vin_deq.val[1], const_3_f32))))),
-                        wrapper::vmul(vin_deq.val[2], wrapper::vmul(const_inv_6_f32, wrapper::vmin(const_6_f32, wrapper::vmax(const_0_f32, wrapper::vadd(vin_deq.val[2], const_3_f32))))),
-                        wrapper::vmul(vin_deq.val[3], wrapper::vmul(const_inv_6_f32, wrapper::vmin(const_6_f32, wrapper::vmax(const_0_f32, wrapper::vadd(vin_deq.val[3], const_3_f32))))),
-                    }
-                };
-                // Re-quantize to new output space
-                tmp = vquantize_signed(tmp_dep, qi_out);
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::LEAKY_RELU)
-            {
-                const auto vin_deq = vdequantize(vin, qi_in);
+                    // De-quantize
+                    const auto vin_deq = vdequantize(vin, qi_in);
+                    // Perform activation
+                    const float32x4x4_t tmp_dep = {{
+                        wrapper::vmul(
+                            vin_deq.val[0],
+                            wrapper::vmul(
+                                const_inv_6_f32,
+                                wrapper::vmin(const_6_f32,
+                                              wrapper::vmax(const_0_f32, wrapper::vadd(vin_deq.val[0], const_3_f32))))),
+                        wrapper::vmul(
+                            vin_deq.val[1],
+                            wrapper::vmul(
+                                const_inv_6_f32,
+                                wrapper::vmin(const_6_f32,
+                                              wrapper::vmax(const_0_f32, wrapper::vadd(vin_deq.val[1], const_3_f32))))),
+                        wrapper::vmul(
+                            vin_deq.val[2],
+                            wrapper::vmul(
+                                const_inv_6_f32,
+                                wrapper::vmin(const_6_f32,
+                                              wrapper::vmax(const_0_f32, wrapper::vadd(vin_deq.val[2], const_3_f32))))),
+                        wrapper::vmul(
+                            vin_deq.val[3],
+                            wrapper::vmul(
+                                const_inv_6_f32,
+                                wrapper::vmin(const_6_f32,
+                                              wrapper::vmax(const_0_f32, wrapper::vadd(vin_deq.val[3], const_3_f32))))),
+                    }};
+                    // Re-quantize to new output space
+                    tmp = vquantize_signed(tmp_dep, qi_out);
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::LEAKY_RELU)
+                {
+                    const auto vin_deq = vdequantize(vin, qi_in);
 
 #ifdef __aarch64__
-                const uint32x4x4_t pos_mask =
-                {
-                    {
+                    const uint32x4x4_t pos_mask = {{
                         wrapper::vcgtz(vin_deq.val[0]),
                         wrapper::vcgtz(vin_deq.val[1]),
                         wrapper::vcgtz(vin_deq.val[2]),
                         wrapper::vcgtz(vin_deq.val[3]),
-                    }
-                };
+                    }};
 #else  // __aarch64__
-                const uint32x4x4_t pos_mask =
-                {
-                    {
+                    const uint32x4x4_t pos_mask = {{
                         wrapper::vcgt(vin_deq.val[0], vconst_0_f32),
                         wrapper::vcgt(vin_deq.val[1], vconst_0_f32),
                         wrapper::vcgt(vin_deq.val[2], vconst_0_f32),
                         wrapper::vcgt(vin_deq.val[3], vconst_0_f32),
-                    }
-                };
+                    }};
 #endif // __aarch64__
 
-                const float32x4x4_t tmp_dep =
-                {
-                    {
+                    const float32x4x4_t tmp_dep = {{
                         wrapper::vbsl(pos_mask.val[0], vin_deq.val[0], wrapper::vmul(va_f32, vin_deq.val[0])),
                         wrapper::vbsl(pos_mask.val[1], vin_deq.val[1], wrapper::vmul(va_f32, vin_deq.val[1])),
                         wrapper::vbsl(pos_mask.val[2], vin_deq.val[2], wrapper::vmul(va_f32, vin_deq.val[2])),
                         wrapper::vbsl(pos_mask.val[3], vin_deq.val[3], wrapper::vmul(va_f32, vin_deq.val[3])),
-                    }
-                };
+                    }};
 
-                tmp = vquantize_signed(tmp_dep, qi_out);
-            }
-            else
-            {
-                ARM_COMPUTE_ERROR("Unsupported activation function");
+                    tmp = vquantize_signed(tmp_dep, qi_out);
+                }
+                else
+                {
+                    ARM_COMPUTE_ERROR("Unsupported activation function");
+                }
+                wrapper::vstore(output_ptr + x, tmp);
             }
-            wrapper::vstore(output_ptr + x, tmp);
-        }
 
-        // Compute left-over elements
-        for(; x < window_end_x; ++x)
-        {
-            qasymm8_signed_t in  = *(reinterpret_cast<const qasymm8_signed_t *>(input_ptr + x));
-            qasymm8_signed_t tmp = 0;
-            if(act == ActivationLayerInfo::ActivationFunction::RELU)
-            {
-                tmp = std::max(const_0, in);
-                tmp = utility::clamp<int32_t, qasymm8_signed_t>(support::cpp11::lround(tmp * s + o));
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::BOUNDED_RELU)
-            {
-                tmp = std::min(a, std::max(const_0, in));
-                tmp = utility::clamp<int32_t, qasymm8_signed_t>(support::cpp11::lround(tmp * s + o));
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU)
+            // Compute left-over elements
+            for (; x < window_end_x; ++x)
             {
-                tmp = std::min(a, std::max(b, in));
-                tmp = utility::clamp<int32_t, qasymm8_signed_t>(support::cpp11::lround(tmp * s + o));
-            }
+                qasymm8_signed_t in  = *(reinterpret_cast<const qasymm8_signed_t *>(input_ptr + x));
+                qasymm8_signed_t tmp = 0;
+                if (act == ActivationLayerInfo::ActivationFunction::RELU)
+                {
+                    tmp = std::max(const_0, in);
+                    tmp = utility::clamp<int32_t, qasymm8_signed_t>(support::cpp11::lround(tmp * s + o));
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::BOUNDED_RELU)
+                {
+                    tmp = std::min(a, std::max(const_0, in));
+                    tmp = utility::clamp<int32_t, qasymm8_signed_t>(support::cpp11::lround(tmp * s + o));
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU)
+                {
+                    tmp = std::min(a, std::max(b, in));
+                    tmp = utility::clamp<int32_t, qasymm8_signed_t>(support::cpp11::lround(tmp * s + o));
+                }
 #ifndef __aarch64__ // LUT-based implementation is used for aarch64 instead.
-            else if(act == ActivationLayerInfo::ActivationFunction::LOGISTIC)
-            {
-                float tmp_f = dequantize_qasymm8_signed(in, qi_in);
-                tmp_f       = 1.f / (1.f + std::exp(-tmp_f));
-                tmp         = quantize_qasymm8_signed(tmp_f, qi_out);
-            }
+                else if (act == ActivationLayerInfo::ActivationFunction::LOGISTIC)
+                {
+                    float tmp_f = dequantize_qasymm8_signed(in, qi_in);
+                    tmp_f       = 1.f / (1.f + std::exp(-tmp_f));
+                    tmp         = quantize_qasymm8_signed(tmp_f, qi_out);
+                }
 #endif // __aarch64__
-            else if(act == ActivationLayerInfo::ActivationFunction::TANH)
-            {
-                float tmp_f = dequantize_qasymm8_signed(in, qi_in);
-                tmp_f       = a_f32 * std::tanh(b_f32 * tmp_f);
-                tmp         = quantize_qasymm8_signed(tmp_f, qi_out);
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::HARD_SWISH)
-            {
-                float tmp_f = dequantize_qasymm8_signed(in, qi_in);
-                tmp_f       = tmp_f * ((std::min(std::max((tmp_f + 3), 0.0f), 6.0f)) * 0.166666667f);
-                tmp         = quantize_qasymm8_signed(tmp_f, qi_out);
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::LEAKY_RELU)
-            {
-                float tmp_f = dequantize_qasymm8_signed(in, qi_in);
-                tmp_f       = tmp_f > 0 ? tmp_f : tmp_f * a_f32;
-                tmp         = quantize_qasymm8_signed(tmp_f, qi_out);
-            }
-            else
-            {
-                ARM_COMPUTE_ERROR("Unsupported activation function");
+                else if (act == ActivationLayerInfo::ActivationFunction::TANH)
+                {
+                    float tmp_f = dequantize_qasymm8_signed(in, qi_in);
+                    tmp_f       = a_f32 * std::tanh(b_f32 * tmp_f);
+                    tmp         = quantize_qasymm8_signed(tmp_f, qi_out);
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::HARD_SWISH)
+                {
+                    float tmp_f = dequantize_qasymm8_signed(in, qi_in);
+                    tmp_f       = tmp_f * ((std::min(std::max((tmp_f + 3), 0.0f), 6.0f)) * 0.166666667f);
+                    tmp         = quantize_qasymm8_signed(tmp_f, qi_out);
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::LEAKY_RELU)
+                {
+                    float tmp_f = dequantize_qasymm8_signed(in, qi_in);
+                    tmp_f       = tmp_f > 0 ? tmp_f : tmp_f * a_f32;
+                    tmp         = quantize_qasymm8_signed(tmp_f, qi_out);
+                }
+                else
+                {
+                    ARM_COMPUTE_ERROR("Unsupported activation function");
+                }
+                *(output_ptr + x) = tmp;
             }
-            *(output_ptr + x) = tmp;
-        }
-    },
-    input, output);
+        },
+        input, output);
 }
 } // namespace cpu
 } // namespace arm_compute
diff --git a/src/cpu/kernels/activation/generic/neon/qsymm16.cpp b/src/cpu/kernels/activation/generic/neon/qsymm16.cpp
index 2aea6cba3c..891646ea00 100644
--- a/src/cpu/kernels/activation/generic/neon/qsymm16.cpp
+++ b/src/cpu/kernels/activation/generic/neon/qsymm16.cpp
@@ -21,11 +21,12 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
+#include "arm_compute/core/experimental/Types.h"
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/ITensorPack.h"
 #include "arm_compute/core/Window.h"
-#include "arm_compute/core/experimental/Types.h"
 #include "arm_compute/function_info/ActivationLayerInfo.h"
+
 #include "src/core/NEON/NEMath.h"
 #include "src/core/NEON/NESymm.h"
 #include "src/core/NEON/wrapper/wrapper.h"
@@ -38,7 +39,10 @@ namespace arm_compute
 {
 namespace cpu
 {
-void neon_qsymm16_activation(const ITensor *src, ITensor *dst, const ActivationLayerInfo &act_info, const Window &window)
+void neon_qsymm16_activation(const ITensor             *src,
+                             ITensor                   *dst,
+                             const ActivationLayerInfo &act_info,
+                             const Window              &window)
 {
     constexpr int                                 window_step_x  = 8;
     const auto                                    window_start_x = static_cast<int>(window.x().start());
@@ -59,103 +63,94 @@ void neon_qsymm16_activation(const ITensor *src, ITensor *dst, const ActivationL
     const float                   a_f32    = act_info.a();
     const float                   b_f32    = act_info.b();
 
-    execute_window_loop(win_collapsed, [&](const Coordinates &)
-    {
-        const auto input_ptr  = reinterpret_cast<const qsymm16_t *>(input.ptr());
-        const auto output_ptr = reinterpret_cast<qsymm16_t *>(output.ptr());
+    execute_window_loop(
+        win_collapsed,
+        [&](const Coordinates &)
+        {
+            const auto input_ptr  = reinterpret_cast<const qsymm16_t *>(input.ptr());
+            const auto output_ptr = reinterpret_cast<qsymm16_t *>(output.ptr());
 
-        wrapper::traits::neon_bitvector_t<qsymm16_t, wrapper::traits::BitWidth::W128> tmp;
-        ARM_COMPUTE_UNUSED(tmp);
+            wrapper::traits::neon_bitvector_t<qsymm16_t, wrapper::traits::BitWidth::W128> tmp;
+            ARM_COMPUTE_UNUSED(tmp);
 
-        // Compute S elements per iteration
-        int x = window_start_x;
-        for(; x <= (window_end_x - window_step_x); x += window_step_x)
-        {
-            const auto vin = wrapper::vloadq(input_ptr + x);
-            if(act == ActivationLayerInfo::ActivationFunction::LOGISTIC)
+            // Compute S elements per iteration
+            int x = window_start_x;
+            for (; x <= (window_end_x - window_step_x); x += window_step_x)
             {
-                // De-quantize
-                const auto vin_deq = vdequantize_int16(vin, qi_in.scale);
-                // Perform activation
-                const float32x4x2_t tmp_dep =
+                const auto vin = wrapper::vloadq(input_ptr + x);
+                if (act == ActivationLayerInfo::ActivationFunction::LOGISTIC)
                 {
-                    {
+                    // De-quantize
+                    const auto vin_deq = vdequantize_int16(vin, qi_in.scale);
+                    // Perform activation
+                    const float32x4x2_t tmp_dep = {{
                         wrapper::vdiv(vconst_1, wrapper::vadd(vconst_1, wrapper::vexpq(wrapper::vneg(vin_deq.val[0])))),
                         wrapper::vdiv(vconst_1, wrapper::vadd(vconst_1, wrapper::vexpq(wrapper::vneg(vin_deq.val[1])))),
-                    }
-                };
-                // Re-quantize to new output space
-                tmp = vquantize_int16(tmp_dep, qi_out.scale);
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::TANH)
-            {
-                // De-quantize
-                const auto vin_deq = vdequantize_int16(vin, qi_in.scale);
-                // Perform activation
-                const float32x4x2_t tmp_dep =
+                    }};
+                    // Re-quantize to new output space
+                    tmp = vquantize_int16(tmp_dep, qi_out.scale);
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::TANH)
                 {
-                    {
+                    // De-quantize
+                    const auto vin_deq = vdequantize_int16(vin, qi_in.scale);
+                    // Perform activation
+                    const float32x4x2_t tmp_dep = {{
                         wrapper::vmul(va_f32, wrapper::vtanh(wrapper::vmul(vin_deq.val[0], vb_f32))),
                         wrapper::vmul(va_f32, wrapper::vtanh(wrapper::vmul(vin_deq.val[1], vb_f32))),
-                    }
-                };
-                // Re-quantize to new output space
-                tmp = vquantize_int16(tmp_dep, qi_out.scale);
-            }
+                    }};
+                    // Re-quantize to new output space
+                    tmp = vquantize_int16(tmp_dep, qi_out.scale);
+                }
 
-            else if(act == ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU)
-            {
-                // De-quantize
-                const auto vin_deq = vdequantize_int16(vin, qi_in.scale);
-                // Perform activation
-                const float32x4x2_t tmp_dep =
+                else if (act == ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU)
                 {
-                    {
-                        wrapper::vmin(va_f32, wrapper::vmax(vb_f32, vin_deq.val[0])),
-                        wrapper::vmin(va_f32, wrapper::vmax(vb_f32, vin_deq.val[1]))
-                    }
-                };
-                // Re-quantize to new output space
-                tmp = vquantize_int16(tmp_dep, qi_out.scale);
-            }
-            else
-            {
-                ARM_COMPUTE_ERROR("Unsupported activation function");
+                    // De-quantize
+                    const auto vin_deq = vdequantize_int16(vin, qi_in.scale);
+                    // Perform activation
+                    const float32x4x2_t tmp_dep = {{wrapper::vmin(va_f32, wrapper::vmax(vb_f32, vin_deq.val[0])),
+                                                    wrapper::vmin(va_f32, wrapper::vmax(vb_f32, vin_deq.val[1]))}};
+                    // Re-quantize to new output space
+                    tmp = vquantize_int16(tmp_dep, qi_out.scale);
+                }
+                else
+                {
+                    ARM_COMPUTE_ERROR("Unsupported activation function");
+                }
+                wrapper::vstore(output_ptr + x, tmp);
             }
-            wrapper::vstore(output_ptr + x, tmp);
-        }
 
-        // Compute left-over elements
-        for(; x < window_end_x; ++x)
-        {
-            qsymm16_t in  = *(reinterpret_cast<const qsymm16_t *>(input_ptr + x));
-            qsymm16_t tmp = 0;
-            if(act == ActivationLayerInfo::ActivationFunction::LOGISTIC)
-            {
-                float tmp_f = dequantize_qsymm16(in, qi_in.scale);
-                tmp_f       = 1.f / (1.f + std::exp(-tmp_f));
-                tmp         = quantize_qsymm16(tmp_f, qi_out);
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::TANH)
+            // Compute left-over elements
+            for (; x < window_end_x; ++x)
             {
-                float tmp_f = dequantize_qsymm16(in, qi_in.scale);
-                tmp_f       = a_f32 * std::tanh(b_f32 * tmp_f);
-                tmp         = quantize_qsymm16(tmp_f, qi_out);
-            }
-            else if(act == ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU)
-            {
-                float tmp_f = dequantize_qsymm16(in, qi_in.scale);
-                tmp_f       = std::min<float>(a_f32, std::max<float>(b_f32, tmp_f));
-                tmp         = quantize_qsymm16(tmp_f, qi_out);
-            }
-            else
-            {
-                ARM_COMPUTE_ERROR("Unsupported activation function");
+                qsymm16_t in  = *(reinterpret_cast<const qsymm16_t *>(input_ptr + x));
+                qsymm16_t tmp = 0;
+                if (act == ActivationLayerInfo::ActivationFunction::LOGISTIC)
+                {
+                    float tmp_f = dequantize_qsymm16(in, qi_in.scale);
+                    tmp_f       = 1.f / (1.f + std::exp(-tmp_f));
+                    tmp         = quantize_qsymm16(tmp_f, qi_out);
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::TANH)
+                {
+                    float tmp_f = dequantize_qsymm16(in, qi_in.scale);
+                    tmp_f       = a_f32 * std::tanh(b_f32 * tmp_f);
+                    tmp         = quantize_qsymm16(tmp_f, qi_out);
+                }
+                else if (act == ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU)
+                {
+                    float tmp_f = dequantize_qsymm16(in, qi_in.scale);
+                    tmp_f       = std::min<float>(a_f32, std::max<float>(b_f32, tmp_f));
+                    tmp         = quantize_qsymm16(tmp_f, qi_out);
+                }
+                else
+                {
+                    ARM_COMPUTE_ERROR("Unsupported activation function");
+                }
+                *(output_ptr + x) = tmp;
             }
-            *(output_ptr + x) = tmp;
-        }
-    },
-    input, output);
+        },
+        input, output);
 }
 } // namespace cpu
 } // namespace arm_compute