Optimize CPU mul layer on quantized data

Resolves : [COMPMID-5461]

Signed-off-by: Omar Al Khatib <omar.alkhatib@arm.com>
Change-Id: I89b99d267c32b00ef44f9bb6e7c714dfe4a0d29d
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8420
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

4 files changed, 345 insertions, 34 deletions

diff --git a/src/core/NEON/wrapper/intrinsics/shr.h b/src/core/NEON/wrapper/intrinsics/shr.h
index d740091464..e41e9b8b31 100644
--- a/src/core/NEON/wrapper/intrinsics/shr.h
+++ b/src/core/NEON/wrapper/intrinsics/shr.h
@@ -25,21 +25,19 @@
 #ifndef ARM_COMPUTE_WRAPPER_SHR_H
 #define ARM_COMPUTE_WRAPPER_SHR_H
 
-#include <type_traits>
 #include <arm_neon.h>
+#include <type_traits>
 
 namespace arm_compute
 {
 namespace wrapper
 {
-
 #define VQRSHRN_IMPL(half_vtype, vtype, prefix, postfix) \
     template <int b>                                     \
     inline half_vtype vqrshrn(const vtype &a)            \
     {                                                    \
         return prefix##_##postfix(a, b);                 \
     }
-
 VQRSHRN_IMPL(int8x8_t, int16x8_t, vqrshrn_n, s16)
 VQRSHRN_IMPL(uint8x8_t, uint16x8_t, vqrshrn_n, u16)
 VQRSHRN_IMPL(int16x4_t, int32x4_t, vqrshrn_n, s32)
@@ -77,20 +75,38 @@ VQRSHRN_SCALAR_IMPL(uint32_t, uint64_t, vqrshrnd_n, u64)
     {                                                                                                            \
         return prefix_signed##_##postfix(a, b);                                                                  \
     }                                                                                                            \
-                                                                                                                 \
+    \
     template <int b, typename T>                                                                                 \
     inline typename std::enable_if<std::is_integral<T>::value && !std::is_signed<T>::value, u##half_vtype>::type \
     vqrshrn_ex(const vtype &a)                                                                                   \
     {                                                                                                            \
         return prefix_unsigned##_##postfix(a, b);                                                                \
     }
-
 VQRSHRN_EX_IMPL(int8x8_t, int16x8_t, vqrshrn_n, vqrshrun_n, s16)
 VQRSHRN_EX_IMPL(int16x4_t, int32x4_t, vqrshrn_n, vqrshrun_n, s32)
 VQRSHRN_EX_IMPL(int32x2_t, int64x2_t, vqrshrn_n, vqrshrun_n, s64)
-
 #undef VQRSHRN_EX_IMPL
 
+#define VSHR_IMPL(vtype, prefix, postfix) \
+    template <int b>                      \
+    inline vtype vshr_n(const vtype &a)   \
+    {                                     \
+        return prefix##_##postfix(a, b);  \
+    }
+VSHR_IMPL(uint8x8_t, vshr_n, u8)
+VSHR_IMPL(int8x8_t, vshr_n, s8)
+#undef VSHR_IMPL
+
+#define VSHRQ_IMPL(vtype, prefix, postfix) \
+    template <int b>                       \
+    inline vtype vshrq_n(const vtype &a)   \
+    {                                      \
+        return prefix##_##postfix(a, b);   \
+    }
+VSHRQ_IMPL(uint32x4_t, vshrq_n, u32)
+VSHRQ_IMPL(int32x4_t, vshrq_n, s32)
+#undef VSHRQ_IMPL
+
 #ifdef __aarch64__
 #define VQRSHRN_EX_SCALAR_IMPL(half_vtype, vtype, prefix_signed, prefix_unsigned, postfix)                       \
     template <int b, typename T>                                                                                 \
diff --git a/src/core/NEON/wrapper/intrinsics/store.h b/src/core/NEON/wrapper/intrinsics/store.h
index 6dda432ea9..ce1b9a554e 100644
--- a/src/core/NEON/wrapper/intrinsics/store.h
+++ b/src/core/NEON/wrapper/intrinsics/store.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2020, 2022 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -44,8 +44,6 @@ VSTORE_IMPL(uint16_t, uint16x4_t, vst1, u16)
 VSTORE_IMPL(int16_t, int16x4_t, vst1, s16)
 VSTORE_IMPL(uint32_t, uint32x2_t, vst1, u32)
 VSTORE_IMPL(int32_t, int32x2_t, vst1, s32)
-//VSTORE_IMPL(uint64_t, 1, vst1, u64)
-//VSTORE_IMPL(int64_t, 1, vst1, s64)
 VSTORE_IMPL(float, float32x2_t, vst1, f32)
 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
 VSTORE_IMPL(float16_t, float16x4_t, vst1, f16)
@@ -57,8 +55,6 @@ VSTORE_IMPL(uint16_t, uint16x8_t, vst1q, u16)
 VSTORE_IMPL(int16_t, int16x8_t, vst1q, s16)
 VSTORE_IMPL(uint32_t, uint32x4_t, vst1q, u32)
 VSTORE_IMPL(int32_t, int32x4_t, vst1q, s32)
-//VSTORE_IMPL(uint64_t, 2, vst1q, u64)
-//VSTORE_IMPL(int64_t, 2, vst1q, s64)
 VSTORE_IMPL(float, float32x4_t, vst1q, f32)
 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
 VSTORE_IMPL(float16_t, float16x8_t, vst1q, f16)
diff --git a/src/core/NEON/wrapper/intrinsics/sub.h b/src/core/NEON/wrapper/intrinsics/sub.h
index 475986d0f6..20436714ef 100644
--- a/src/core/NEON/wrapper/intrinsics/sub.h
+++ b/src/core/NEON/wrapper/intrinsics/sub.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2020, 2022 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -98,6 +98,21 @@ VQSUB_IMPL(float16x8_t, float16x8_t, vsubq, f16)
 #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
 #undef VQSUB_IMPL
 
+#define VSUBL_IMPL(rtype, vtype, prefix, postfix)      \
+    inline rtype vsubl(const vtype &a, const vtype &b) \
+    {                                                  \
+        return prefix##_##postfix(a, b);               \
+    }
+
+VSUBL_IMPL(int16x8_t, int8x8_t, vsubl, s8)
+VSUBL_IMPL(int32x4_t, int16x4_t, vsubl, s16)
+VSUBL_IMPL(int64x2_t, int32x2_t, vsubl, s32)
+VSUBL_IMPL(uint16x8_t, uint8x8_t, vsubl, u8)
+VSUBL_IMPL(uint32x4_t, uint16x4_t, vsubl, u16)
+VSUBL_IMPL(uint64x2_t, uint32x2_t, vsubl, u32)
+
+#undef VSUB_IMPL
+
 } // namespace wrapper
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_WRAPPER_SUB_H */
diff --git a/src/cpu/kernels/CpuMulKernel.cpp b/src/cpu/kernels/CpuMulKernel.cpp
index 2f04bf9f26..487954b889 100644
--- a/src/cpu/kernels/CpuMulKernel.cpp
+++ b/src/cpu/kernels/CpuMulKernel.cpp
@@ -241,7 +241,8 @@ void mul_saturate_quantized_8(const ITensor *src1, const ITensor *src2, ITensor
         Iterator input2(src2, input2_win);
         Iterator dst(out, win);
 
-        execute_window_loop(win, [&](const Coordinates &)
+        execute_window_loop(
+            win, [&](const Coordinates &)
         {
             const auto input1_ptr = reinterpret_cast<const T *>(input1.ptr());
             const auto input2_ptr = reinterpret_cast<const T *>(input2.ptr());
@@ -290,6 +291,262 @@ void mul_saturate_quantized_8(const ITensor *src1, const ITensor *src2, ITensor
     }
 }
 
+bool mul_q8_neon_fixedpoint_possible(const ITensorInfo *src0, const ITensorInfo *src1, const ITensorInfo *dst, float scale)
+{
+    const auto iq0 = src0->quantization_info().uniform();
+    const auto iq1 = src1->quantization_info().uniform();
+    const auto oq  = dst->quantization_info().uniform();
+
+    const auto multiplier = ((iq0.scale * iq1.scale) / oq.scale) * scale;
+
+    if(multiplier < -8191.f || multiplier > 8191.f)
+    {
+        //The multiplier cannot be stored as a 14.18 signed fixed-point number
+        return false;
+    }
+
+    const auto offset_out = float(oq.offset);
+
+    const auto max_result = multiplier * (256) * (256) + offset_out;
+
+    if(max_result > 8191.f)
+    {
+        //It might not be possible to store the result as a 14.18 signed fixed-point number.
+        return false;
+    }
+
+    return true;
+}
+
+template <typename ScalarType>
+void mul_q8_neon_fixedpoint(const ITensor *src0, const ITensor *src1, ITensor *dst, const Window &window, float scale)
+{
+    const auto in0_info = src0->info();
+    const auto in1_info = src1->info();
+
+    const auto &in0_shape = in0_info->tensor_shape();
+    const auto &in1_shape = in1_info->tensor_shape();
+
+    // Create input windows.
+    Window in0_win = window.broadcast_if_dimension_le_one(in0_shape);
+    Window in1_win = window.broadcast_if_dimension_le_one(in1_shape);
+
+    // Clear the x dimension on the execution window as we process the whole row each iteration.
+    Window win = window;
+    win.set(Window::DimX, Window::Dimension(0, 1, 1));
+
+    constexpr int window_step_x         = 16;
+    const auto    window_start_x        = window.x().start();
+    const auto    window_end_x          = window.x().end();
+    const auto    is_broadcast_across_x = in0_shape.x() != in1_shape.x();
+
+    const auto iq0_info = in0_info->quantization_info().uniform();
+    const auto iq1_info = in1_info->quantization_info().uniform();
+    const auto oq_info  = dst->info()->quantization_info().uniform();
+
+    const auto in0_offset = iq0_info.offset;
+    const auto in1_offset = iq1_info.offset;
+    const auto out_offset = oq_info.offset;
+    const auto multiplier = ((iq0_info.scale * iq1_info.scale) / oq_info.scale) * scale;
+
+    constexpr int32_t two_pwr18i = 262144;
+    constexpr float   two_pwr18f = 262144.f;
+
+    const auto in0_offset_16p0  = static_cast<int16_t>(in0_offset);
+    const auto in1_offset_16p0  = static_cast<int16_t>(in1_offset);
+    const auto out_offset_14p18 = static_cast<int32_t>(out_offset * two_pwr18i);
+    const auto multiplier_14p18 = static_cast<int32_t>(multiplier * two_pwr18f);
+
+    if(is_broadcast_across_x)
+    {
+        // Prefix: a = non-broadcast, b = broadcast.
+
+        const auto is_broadcast_input_1 = in1_win.x().step() == 0;
+        auto       a_win                = is_broadcast_input_1 ? in0_win : in1_win;
+        auto       b_win                = is_broadcast_input_1 ? in1_win : in0_win;
+        const auto a_tensor             = is_broadcast_input_1 ? src0 : src1;
+        const auto b_tensor             = is_broadcast_input_1 ? src1 : src0;
+
+        const auto a_offset_16p0 = is_broadcast_input_1 ? in0_offset_16p0 : in1_offset_16p0;
+        const auto b_offset_16p0 = is_broadcast_input_1 ? in1_offset : in0_offset;
+#ifndef __aarch64__
+        const auto a_offset = is_broadcast_input_1 ? in0_offset : in1_offset;
+        const auto b_offset = is_broadcast_input_1 ? in1_offset : in0_offset;
+#endif //__aarch64__
+        const auto a_voffset_16p0 = wrapper::vdup_n(a_offset_16p0, wrapper::traits::vector_64_tag());
+
+        // Clear the x dimension on the execution window as we process the whole row each iteration.
+        a_win.set(Window::DimX, Window::Dimension(0, 1, 1));
+
+        Iterator a_input_it(a_tensor, a_win);
+        Iterator b_input_it(b_tensor, b_win);
+        Iterator out_it(dst, win);
+
+        execute_window_loop(
+            win, [&](const Coordinates &)
+        {
+            const auto a_ptr   = reinterpret_cast<const ScalarType *>(a_input_it.ptr());
+            const auto b_ptr   = reinterpret_cast<const ScalarType *>(b_input_it.ptr());
+            const auto out_ptr = reinterpret_cast<ScalarType *>(out_it.ptr());
+
+            const auto b_val            = *b_ptr;
+            const auto b_offseted_32p0  = static_cast<int32_t>(b_val - b_offset_16p0);
+            const auto b_voffseted_32p0 = wrapper::vdup_n(b_offseted_32p0, wrapper::traits::vector_128_tag());
+
+            const auto vmultiplier_14p18 = wrapper::vdup_n(multiplier_14p18, wrapper::traits::vector_128_tag());
+            const auto voffsetout_14p18  = wrapper::vdup_n(out_offset_14p18, wrapper::traits::vector_128_tag());
+
+            int x = window_start_x;
+
+            for(; x <= (window_end_x - window_step_x); x += window_step_x)
+            {
+                // Load the inputs.
+                const auto a_vin_8p0 = wrapper::vloadq(a_ptr + x);
+
+                // Widen the non-broadcast elements to signed 16-bit regardless of the input signedness.
+                const auto a_vin_16p0_0 = wrapper::vreinterpret(wrapper::vmovl(wrapper::vgetlow(a_vin_8p0)));
+                const auto a_vin_16p0_1 = wrapper::vreinterpret(wrapper::vmovl(wrapper::vgethigh(a_vin_8p0)));
+
+                const auto voffseted_32p0_00 = wrapper::vsubl(wrapper::vgetlow(a_vin_16p0_0), a_voffset_16p0);
+                const auto voffseted_32p0_01 = wrapper::vsubl(wrapper::vgethigh(a_vin_16p0_0), a_voffset_16p0);
+                const auto voffseted_32p0_10 = wrapper::vsubl(wrapper::vgetlow(a_vin_16p0_1), a_voffset_16p0);
+                const auto voffseted_32p0_11 = wrapper::vsubl(wrapper::vgethigh(a_vin_16p0_1), a_voffset_16p0);
+
+                const auto vinnermul_32p0_00 = wrapper::vmul(voffseted_32p0_00, b_voffseted_32p0);
+                const auto vinnermul_32p0_01 = wrapper::vmul(voffseted_32p0_01, b_voffseted_32p0);
+                const auto vinnermul_32p0_10 = wrapper::vmul(voffseted_32p0_10, b_voffseted_32p0);
+                const auto vinnermul_32p0_11 = wrapper::vmul(voffseted_32p0_11, b_voffseted_32p0);
+
+                const auto vout_14p18_00 = wrapper::vmla(voffsetout_14p18, vinnermul_32p0_00, vmultiplier_14p18);
+                const auto vout_14p18_01 = wrapper::vmla(voffsetout_14p18, vinnermul_32p0_01, vmultiplier_14p18);
+                const auto vout_14p18_10 = wrapper::vmla(voffsetout_14p18, vinnermul_32p0_10, vmultiplier_14p18);
+                const auto vout_14p18_11 = wrapper::vmla(voffsetout_14p18, vinnermul_32p0_11, vmultiplier_14p18);
+
+                // These shift rights are to revert the multiplication by twopwr18. Hard limit of a maximum shift by 8 requires multiple shift instructions to achieve this.
+                const auto vout_15p1_00 = wrapper::vqrshrn_ex<8, ScalarType>(wrapper::vshrq_n<8>(vout_14p18_00));
+                const auto vout_15p1_01 = wrapper::vqrshrn_ex<8, ScalarType>(wrapper::vshrq_n<8>(vout_14p18_01));
+                const auto vout_15p1_10 = wrapper::vqrshrn_ex<8, ScalarType>(wrapper::vshrq_n<8>(vout_14p18_10));
+                const auto vout_15p1_11 = wrapper::vqrshrn_ex<8, ScalarType>(wrapper::vshrq_n<8>(vout_14p18_11));
+
+                const auto vout_15p1_0 = wrapper::vcombine(
+                                             vout_15p1_00,
+                                             vout_15p1_01);
+
+                const auto vout_15p1_1 = wrapper::vcombine(
+                                             vout_15p1_10,
+                                             vout_15p1_11);
+                const auto out_ptr = reinterpret_cast<ScalarType *>(out_it.ptr());
+
+                const auto vout_8p0 = wrapper::vcombine(
+                                          wrapper::vqrshrn<2>(vout_15p1_0),
+                                          wrapper::vqrshrn<2>(vout_15p1_1));
+                wrapper::vstore(out_ptr + x, vout_8p0);
+            }
+
+            //Process the left-over elements.
+            for(; x < window_end_x; ++x)
+            {
+#ifdef __aarch64__
+                out_ptr[x] = wrapper::vqrshrn<2>(wrapper::vqrshrn<8>(wrapper::vqrshrn_ex<8, ScalarType>((multiplier_14p18 * (int32_t(a_ptr[x]) - a_offset_16p0) * (int32_t(
+                                                                                                             b_val) - b_offset_16p0)) + out_offset_14p18)));
+#else  //__aarch64__
+                out_ptr[x] = utility::clamp<int32_t, ScalarType>(support::cpp11::lround(multiplier * ((float(a_ptr[x]) - a_offset) * (float(b_val) - b_offset)) + float(out_offset)));
+#endif //__aarch64__
+            }
+        },
+        a_input_it, b_input_it, out_it);
+    }
+    else
+    {
+        const auto voffset0_16p0     = wrapper::vdup_n(in0_offset_16p0, wrapper::traits::vector_64_tag());
+        const auto voffset1_16p0     = wrapper::vdup_n(in1_offset_16p0, wrapper::traits::vector_64_tag());
+        const auto voffsetout_14p18  = wrapper::vdup_n(out_offset_14p18, wrapper::traits::vector_128_tag());
+        const auto vmultiplier_14p18 = wrapper::vdup_n(multiplier_14p18, wrapper::traits::vector_128_tag());
+
+        // Clear the x dimension on the execution window as we process the whole row each iteration.
+        in0_win.set(Window::DimX, Window::Dimension(0, 1, 1));
+        in1_win.set(Window::DimX, Window::Dimension(0, 1, 1));
+
+        Iterator in0_it(src0, in0_win);
+        Iterator in1_it(src1, in1_win);
+        Iterator out_it(dst, win);
+
+        execute_window_loop(
+            win, [&](const Coordinates &)
+        {
+            const auto in0_ptr = reinterpret_cast<const ScalarType *>(in0_it.ptr());
+            const auto in1_ptr = reinterpret_cast<const ScalarType *>(in1_it.ptr());
+            const auto out_ptr = reinterpret_cast<ScalarType *>(out_it.ptr());
+
+            int x = window_start_x;
+
+            for(; x <= (window_end_x - window_step_x); x += window_step_x)
+            {
+                // Load the inputs.
+                const auto vin0_8p0 = wrapper::vloadq(in0_ptr + x);
+                const auto vin1_8p0 = wrapper::vloadq(in1_ptr + x);
+
+                // Widen the input elements to signed 16-bit regardless of the input signedness.
+                const auto vin0_16p0_0 = wrapper::vreinterpret(wrapper::vmovl(wrapper::vgetlow(vin0_8p0)));
+                const auto vin0_16p0_1 = wrapper::vreinterpret(wrapper::vmovl(wrapper::vgethigh(vin0_8p0)));
+                const auto vin1_16p0_0 = wrapper::vreinterpret(wrapper::vmovl(wrapper::vgetlow(vin1_8p0)));
+                const auto vin1_16p0_1 = wrapper::vreinterpret(wrapper::vmovl(wrapper::vgethigh(vin1_8p0)));
+
+                const auto voffseted0_32p0_00 = wrapper::vsubl(wrapper::vgetlow(vin0_16p0_0), voffset0_16p0);
+                const auto voffseted0_32p0_01 = wrapper::vsubl(wrapper::vgethigh(vin0_16p0_0), voffset0_16p0);
+                const auto voffseted0_32p0_10 = wrapper::vsubl(wrapper::vgetlow(vin0_16p0_1), voffset0_16p0);
+                const auto voffseted0_32p0_11 = wrapper::vsubl(wrapper::vgethigh(vin0_16p0_1), voffset0_16p0);
+
+                const auto voffseted1_32p0_00 = wrapper::vsubl(wrapper::vgetlow(vin1_16p0_0), voffset1_16p0);
+                const auto voffseted1_32p0_01 = wrapper::vsubl(wrapper::vgethigh(vin1_16p0_0), voffset1_16p0);
+                const auto voffseted1_32p0_10 = wrapper::vsubl(wrapper::vgetlow(vin1_16p0_1), voffset1_16p0);
+                const auto voffseted1_32p0_11 = wrapper::vsubl(wrapper::vgethigh(vin1_16p0_1), voffset1_16p0);
+
+                const auto vinnermul_32p0_00 = wrapper::vmul(voffseted0_32p0_00, voffseted1_32p0_00);
+                const auto vinnermul_32p0_01 = wrapper::vmul(voffseted0_32p0_01, voffseted1_32p0_01);
+                const auto vinnermul_32p0_10 = wrapper::vmul(voffseted0_32p0_10, voffseted1_32p0_10);
+                const auto vinnermul_32p0_11 = wrapper::vmul(voffseted0_32p0_11, voffseted1_32p0_11);
+
+                const auto vout_14p18_00 = wrapper::vmla(voffsetout_14p18, vinnermul_32p0_00, vmultiplier_14p18);
+                const auto vout_14p18_01 = wrapper::vmla(voffsetout_14p18, vinnermul_32p0_01, vmultiplier_14p18);
+                const auto vout_14p18_10 = wrapper::vmla(voffsetout_14p18, vinnermul_32p0_10, vmultiplier_14p18);
+                const auto vout_14p18_11 = wrapper::vmla(voffsetout_14p18, vinnermul_32p0_11, vmultiplier_14p18);
+
+                // These shift rights are to revert the multiplication by twopwr18. Hard limit of a maximum shift by 8 requires multiple shift instructions to achieve this.
+                const auto vout_14p2_00 = wrapper::vqrshrn_ex<8, ScalarType>(wrapper::vshrq_n<8>(vout_14p18_00));
+                const auto vout_14p2_01 = wrapper::vqrshrn_ex<8, ScalarType>(wrapper::vshrq_n<8>(vout_14p18_01));
+                const auto vout_14p2_10 = wrapper::vqrshrn_ex<8, ScalarType>(wrapper::vshrq_n<8>(vout_14p18_10));
+                const auto vout_14p2_11 = wrapper::vqrshrn_ex<8, ScalarType>(wrapper::vshrq_n<8>(vout_14p18_11));
+
+                const auto vout_14p2_0 = wrapper::vcombine(
+                                             vout_14p2_00,
+                                             vout_14p2_01);
+
+                const auto vout_14p2_1 = wrapper::vcombine(
+                                             vout_14p2_10,
+                                             vout_14p2_11);
+
+                const auto vout_8p0 = wrapper::vcombine(
+                                          wrapper::vqrshrn<2>(vout_14p2_0),
+                                          wrapper::vqrshrn<2>(vout_14p2_1));
+                wrapper::vstore(out_ptr + x, vout_8p0);
+            }
+
+            //Process the left-over elements.
+            for(; x < window_end_x; ++x)
+            {
+#ifdef __aarch64__
+                out_ptr[x] = wrapper::vqrshrn<2>(wrapper::vqrshrn<8>(wrapper::vqrshrn_ex<8, ScalarType>((multiplier_14p18 * (int32_t(in0_ptr[x]) - in0_offset_16p0) * (int32_t(
+                                                                                                             in1_ptr[x]) - in1_offset_16p0)) + out_offset_14p18)));
+#else  //__aarch64__
+                out_ptr[x] = utility::clamp<int32_t, ScalarType>(support::cpp11::lround(multiplier * ((float(in0_ptr[x]) - in0_offset) * (float(in1_ptr[x]) - in1_offset)) + float(out_offset)));
+#endif //__aarch64__
+            }
+        },
+        in0_it, in1_it, out_it);
+    }
+}
+
 void mul_saturate_QSYMM16_QSYMM16_QSYMM16(const ITensor *src1, const ITensor *src2, ITensor *out, const Window &window, float scale)
 {
     const UniformQuantizationInfo input1_qua_info = src1->info()->quantization_info().uniform();
@@ -316,7 +573,8 @@ void mul_saturate_QSYMM16_QSYMM16_QSYMM16(const ITensor *src1, const ITensor *sr
 
     const UniformQuantizationInfo tmp_qua_info = { output_qua_info.scale / scale, output_qua_info.offset };
 
-    execute_window_loop(win, [&](const Coordinates &)
+    execute_window_loop(
+        win, [&](const Coordinates &)
     {
         const auto input1_ptr = reinterpret_cast<const qsymm16_t *>(input1.ptr());
         const auto input2_ptr = reinterpret_cast<const qsymm16_t *>(input2.ptr());
@@ -397,7 +655,8 @@ void mul_QSYMM16_QSYMM16_S32(const ITensor *src1, const ITensor *src2, ITensor *
     const auto window_start_x = static_cast<int>(window.x().start());
     const auto window_end_x   = static_cast<int>(window.x().end());
 
-    execute_window_loop(win, [&](const Coordinates &)
+    execute_window_loop(
+        win, [&](const Coordinates &)
     {
         const auto input1_ptr = reinterpret_cast<const qsymm16_t *>(input1.ptr());
         const auto input2_ptr = reinterpret_cast<const qsymm16_t *>(input2.ptr());
@@ -488,7 +747,8 @@ void mul_U8_U8_U8(const ITensor *src1, const ITensor *src2, ITensor *out, const
     const auto window_start_x = static_cast<int>(window.x().start());
     const auto window_end_x   = static_cast<int>(window.x().end());
 
-    execute_window_loop(win, [&](const Coordinates &)
+    execute_window_loop(
+        win, [&](const Coordinates &)
     {
         const auto input1_ptr = reinterpret_cast<const uint8_t *>(input1.ptr());
         const auto input2_ptr = reinterpret_cast<const uint8_t *>(input2.ptr());
@@ -653,7 +913,8 @@ void mul_S16_S16_S16(const ITensor *src1, const ITensor *src2, ITensor *out, con
     const auto window_start_x = static_cast<int>(window.x().start());
     const auto window_end_x   = static_cast<int>(window.x().end());
 
-    execute_window_loop(win, [&](const Coordinates &)
+    execute_window_loop(
+        win, [&](const Coordinates &)
     {
         const auto input1_ptr = reinterpret_cast<const int16_t *>(input1.ptr());
         const auto input2_ptr = reinterpret_cast<const int16_t *>(input2.ptr());
@@ -716,7 +977,7 @@ void mul_S16_S16_S16(const ITensor *src1, const ITensor *src2, ITensor *out, con
     input1, input2, dst);
 }
 
-template <bool   is_sat>
+template <bool is_sat>
 inline int32x4_t mul_S32_S32_S32_n_loop(const int32x4_t &src1, const int32x4_t &src2, int n)
 {
     const int32x2_t input1_1 = vget_low_s32(src1);
@@ -756,7 +1017,7 @@ inline int32x4_t mul_S32_S32_S32_n_loop(const int32x4_t &src1, const int32x4_t &
     }
 }
 
-template <bool     is_sat>
+template <bool is_sat>
 inline int32x4x2_t mul_S32_S32_S32_n_k(const int32x4x2_t &src1, const int32x4x2_t &src2, int n)
 {
     const int32x4x2_t result =
@@ -803,7 +1064,8 @@ void mul_S32_S32_S32(const ITensor *src1, const ITensor *src2, ITensor *out, con
         Iterator non_broadcast_input(non_broadcast_tensor, non_broadcast_win);
         Iterator dst(out, win);
 
-        execute_window_loop(win, [&](const Coordinates &)
+        execute_window_loop(
+            win, [&](const Coordinates &)
         {
             const auto non_broadcast_input_ptr = reinterpret_cast<const int32_t *>(non_broadcast_input.ptr());
             const auto output_ptr              = reinterpret_cast<int32_t *>(dst.ptr());
@@ -868,7 +1130,8 @@ void mul_S32_S32_S32(const ITensor *src1, const ITensor *src2, ITensor *out, con
         Iterator input2(src2, input2_win);
         Iterator dst(out, win);
 
-        execute_window_loop(win, [&](const Coordinates &)
+        execute_window_loop(
+            win, [&](const Coordinates &)
         {
             const auto input1_ptr = reinterpret_cast<const int32_t *>(input1.ptr());
             const auto input2_ptr = reinterpret_cast<const int32_t *>(input2.ptr());
@@ -955,7 +1218,8 @@ void mul_F32_F32_F32(const ITensor *src1, const ITensor *src2, ITensor *out, con
         Iterator non_broadcast_input(non_broadcast_tensor, non_broadcast_win);
         Iterator dst(out, win);
 
-        execute_window_loop(win, [&](const Coordinates &)
+        execute_window_loop(
+            win, [&](const Coordinates &)
         {
             const auto non_broadcast_input_ptr = reinterpret_cast<const float *>(non_broadcast_input.ptr());
             const auto output_ptr              = reinterpret_cast<float *>(dst.ptr());
@@ -992,7 +1256,8 @@ void mul_F32_F32_F32(const ITensor *src1, const ITensor *src2, ITensor *out, con
         Iterator input2(src2, input2_win);
         Iterator dst(out, win);
 
-        execute_window_loop(win, [&](const Coordinates &)
+        execute_window_loop(
+            win, [&](const Coordinates &)
         {
             const auto input1_ptr = reinterpret_cast<const float *>(input1.ptr());
             const auto input2_ptr = reinterpret_cast<const float *>(input2.ptr());
@@ -1053,7 +1318,8 @@ void c_mul_F32_F32_F32_n(const ITensor *src1, const ITensor *src2, ITensor *out,
         Iterator non_broadcast_input(non_broadcast_tensor, non_broadcast_win);
         Iterator dst(out, win);
 
-        execute_window_loop(win, [&](const Coordinates &)
+        execute_window_loop(
+            win, [&](const Coordinates &)
         {
             const auto non_broadcast_input_ptr = reinterpret_cast<const float *>(non_broadcast_input.ptr());
             const auto output_ptr              = reinterpret_cast<float *>(dst.ptr());
@@ -1106,7 +1372,8 @@ void c_mul_F32_F32_F32_n(const ITensor *src1, const ITensor *src2, ITensor *out,
         Iterator input2(src2, input2_win);
         Iterator dst(out, win);
 
-        execute_window_loop(win, [&](const Coordinates &)
+        execute_window_loop(
+            win, [&](const Coordinates &)
         {
             const auto input1_ptr = reinterpret_cast<const float *>(input1.ptr());
             const auto input2_ptr = reinterpret_cast<const float *>(input2.ptr());
@@ -1180,12 +1447,13 @@ void mul_F16_F16_F16(const ITensor *src1, const ITensor *src2, ITensor *out, con
         Iterator broadcast_input(broadcast_tensor, broadcast_win);
         Iterator non_broadcast_input(non_broadcast_tensor, non_broadcast_win);
         Iterator dst(out, win);
-        execute_window_loop(win, [&](const Coordinates &)
+        execute_window_loop(
+            win, [&](const Coordinates &)
         {
             const auto          non_broadcast_input_ptr = reinterpret_cast<const float16_t *>(non_broadcast_input.ptr());
             const auto          output_ptr              = reinterpret_cast<float16_t *>(dst.ptr());
             const auto          broadcast_value         = *reinterpret_cast<const float16_t *>(broadcast_input.ptr());
-            const float16x8x2_t broadcast_value_vec =
+            const float16x8x2_t broadcast_value_vec     =
             {
                 {
                     vdupq_n_f16(broadcast_value),
@@ -1230,7 +1498,8 @@ void mul_F16_F16_F16(const ITensor *src1, const ITensor *src2, ITensor *out, con
         Iterator input1(src1, input1_win);
         Iterator input2(src2, input2_win);
         Iterator dst(out, win);
-        execute_window_loop(win, [&](const Coordinates &)
+        execute_window_loop(
+            win, [&](const Coordinates &)
         {
             const auto input1_ptr = reinterpret_cast<const float16_t *>(input1.ptr());
             const auto input2_ptr = reinterpret_cast<const float16_t *>(input2.ptr());
@@ -1254,7 +1523,7 @@ void mul_F16_F16_F16(const ITensor *src1, const ITensor *src2, ITensor *out, con
                     }
                 };
                 const float16x8_t   scale_vec = vdupq_n_f16(scale);
-                const float16x8x2_t result =
+                const float16x8x2_t result    =
                 {
                     {
                         vmulq_f16(vmulq_f16(ta1.val[0], ta2.val[0]), scale_vec),
@@ -1298,7 +1567,8 @@ void mul_U8_U8_S16(const ITensor *src1, const ITensor *src2, ITensor *out, const
     const auto window_start_x = static_cast<int>(window.x().start());
     const auto window_end_x   = static_cast<int>(window.x().end());
 
-    execute_window_loop(win, [&](const Coordinates &)
+    execute_window_loop(
+        win, [&](const Coordinates &)
     {
         const auto input1_ptr = reinterpret_cast<const uint8_t *>(input1.ptr());
         const auto input2_ptr = reinterpret_cast<const uint8_t *>(input2.ptr());
@@ -1396,7 +1666,8 @@ void mul_S16_U8_S16(const ITensor *src1, const ITensor *src2, ITensor *out, cons
     const auto window_start_x = static_cast<int>(window.x().start());
     const auto window_end_x   = static_cast<int>(window.x().end());
 
-    execute_window_loop(win, [&](const Coordinates &)
+    execute_window_loop(
+        win, [&](const Coordinates &)
     {
         const auto input1_ptr = reinterpret_cast<const int16_t *>(input1.ptr());
         const auto input2_ptr = reinterpret_cast<const uint8_t *>(input2.ptr());
@@ -1520,14 +1791,27 @@ void CpuMulKernel::configure(ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *
         case DataType::QASYMM8:
             if(dt_input2 == DataType::QASYMM8 && dt_output == DataType::QASYMM8)
             {
-                _func_quantized = &mul_saturate_quantized_8<uint8_t>;
+                if(mul_q8_neon_fixedpoint_possible(src1, src2, dst, scale))
+                {
+                    _func_quantized = &mul_q8_neon_fixedpoint<uint8_t>;
+                }
+                else
+                {
+                    _func_quantized = &mul_saturate_quantized_8<uint8_t>;
+                }
             }
             break;
         case DataType::QASYMM8_SIGNED:
             if(dt_input2 == DataType::QASYMM8_SIGNED)
             {
-                _func_quantized = &mul_saturate_quantized_8<int8_t>;
-                ;
+                if(mul_q8_neon_fixedpoint_possible(src1, src2, dst, scale))
+                {
+                    _func_quantized = &mul_q8_neon_fixedpoint<int8_t>;
+                }
+                else
+                {
+                    _func_quantized = &mul_saturate_quantized_8<int8_t>;
+                }
             }
             break;
         case DataType::QSYMM16: